Bordered Heegaard Floer homology

Robert Lipshitz
Peter S. Ozsvth
Dylan P. Thurston
Abstract. We construct Heegaard Floer theory for 3-manifolds with connected boundary.
The theory associates to an oriented, parametrized two-manifold a dierential graded alge-
bra. For a three-manifold with parametrized boundary, the invariant comes in two dierent
versions, one of which (type D) is a module over the algebra and the other of which (type
A) is an /

module. Both are well-dened up to chain homotopy equivalence. For a de-

composition of a 3-manifold into two pieces, the /

tensor product of the type D module

of one piece and the type A module from the other piece is

HF of the glued manifold.
As a special case of the construction, we specialize to the case of three-manifolds with
torus boundary. This case can be used to give another proof of the surgery exact triangle
for

HF. We relate the bordered Floer homology of a three-manifold with torus boundary
with the knot Floer homology of a lling.
a
r
X
i
v
:
0
8
1
0
.
0
6
8
7
v
4


[
m
a
t
h
.
G
T
]


1

J
u
n

2
0
1
1
Department of Mathematics, Columbia University, New York, NY 10027
E-mail address: lipshitz@math.columbia.edu
Department of Mathematics, Columbia University, New York, NY 10027
E-mail address: petero@math.columbia.edu
Dept. of Math., Barnard College, Columbia University, New York, NY
10027
E-mail address: dthurston@barnard.edu
RL was supported by an NSF Mathematical Sciences Postdoctoral Research Fellowship,
NSF grant number DMS-0905796 and a Sloan Research Fellowship.
PSO was supported by NSF grant number DMS-0505811.
DPT was supported by NSF Grant DMS-1008049 and a Sloan Research Fellowship.
Contents
Chapter 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1. Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2. The bordered Floer homology package . . . . . . . . . . . . . . . . . . . . . 2
1.3. On gradings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4. The case of three-manifolds with torus boundary . . . . . . . . . . . . . . . 5
1.5. Previous work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6. Further developments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.7. Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Chapter 2. /

structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1. /

algebras and modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.2. /

tensor products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3. Type 1 structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4. Another model for the /

tensor product . . . . . . . . . . . . . . . . . . . 21
2.5. Gradings by non-commutative groups . . . . . . . . . . . . . . . . . . . . . . 24
Chapter 3. The algebra associated to a pointed matched circle . . . . . . . . . . . . 29
3.1. The strands algebra /(:. /) . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2. Matched circles and their algebras . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3. Gradings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Chapter 4. Bordered Heegaard diagrams . . . . . . . . . . . . . . . . . . . . . . . . 45
4.1. Bordered Heegaard diagrams: denition, existence, and uniqueness . . . . . 45
4.2. Examples of bordered Heegaard diagrams . . . . . . . . . . . . . . . . . . . 51
4.3. Generators, homology classes and spin
c
structures . . . . . . . . . . . . . . . 51
4.4. Admissibility criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.5. Closed diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Chapter 5. Moduli spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.1. Overview of the moduli spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2. Holomorphic curves in [0. 1] 1 . . . . . . . . . . . . . . . . . . . . . . 61
5.3. Holomorphic curves in 2 1 [0. 1] 1 . . . . . . . . . . . . . . . . . . . . 66
5.4. Compactications via holomorphic combs . . . . . . . . . . . . . . . . . . . 69
5.5. Gluing results for holomorphic combs . . . . . . . . . . . . . . . . . . . . . . 75
5.6. Degenerations of holomorphic curves . . . . . . . . . . . . . . . . . . . . . . 80
5.7. More on expected dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Chapter 6. Type 1 modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.1. Denition of the type 1 module . . . . . . . . . . . . . . . . . . . . . . . . . 99
iii
iv CONTENTS
6.2.
2
= 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.3. Invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.4. Twisted coecients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
Chapter 7. Type modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
7.1. Denition of the type module . . . . . . . . . . . . . . . . . . . . . . . . . 133
7.2. Compatibility with algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7.3. Invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
7.4. Twisted coecients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
Chapter 8. Pairing theorem via nice diagrams . . . . . . . . . . . . . . . . . . . . . 145
Chapter 9. Pairing theorem via time dilation . . . . . . . . . . . . . . . . . . . . . 149
9.1. Moduli of matched pairs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
9.2. Dilating time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
9.3. Dilating to innity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
9.4. Completion of the proof of the pairing theorem . . . . . . . . . . . . . . . . 169
9.5. A twisted pairing theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
9.6. An example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
Chapter 10. Gradings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
10.1. Algebra review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
10.2. Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
10.3. Type structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
10.4. Type 1 structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
10.5. Rened gradings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
10.6. Tensor product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
Chapter 11. Bordered manifolds with torus boundary . . . . . . . . . . . . . . . . . 191
11.1. Torus algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
11.2. Surgery exact triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
11.3. Preliminaries on knot Floer homology . . . . . . . . . . . . . . . . . . . . . 195
11.4. From

CFD to HFK

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
11.5. From CFK

to

CFD: Statement of results . . . . . . . . . . . . . . . . . . 202
11.6. Generalized coecient maps and boundary degenerations . . . . . . . . . . 206
11.7. From CFK

to

CFD: Basis-free version . . . . . . . . . . . . . . . . . . . . 209
11.8. Proof of Theorem 11.27 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
11.9. Cables revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
Appendix A. Bimodules and change of framing . . . . . . . . . . . . . . . . . . . . 235
A.1. Statement of results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
A.2. Sketch of the construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
A.3. Computations for 3-manifolds with torus boundary . . . . . . . . . . . . . . 240
A.4. From HFK to

CFD for arbitrary integral framings . . . . . . . . . . . . . . 245
Index of Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
Index of Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
List of Figures
1.1 Example of cutting a Heegaard diagram . . . . . . . . . . . . . . . . . . . . . 3
3.1 Splittings and shues of Reeb chords. . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Pointed matched circles for the genus 2 surface . . . . . . . . . . . . . . . . . 35
3.3 Orientation of a pointed matched circle. . . . . . . . . . . . . . . . . . . . . . 35
4.1 3-manifold represented by a bordered Heegaard diagram. . . . . . . . . . . . . 48
4.2 Bordered Heegaard diagram for the trefoil complement. . . . . . . . . . . . . 51
4.3 Heegaard diagrams for handlebodies. . . . . . . . . . . . . . . . . . . . . . . . 52
4.4 Another Heegaard diagram for a genus 2 handlebody . . . . . . . . . . . . . . 52
5.1 Examples of three kinds of codimension 1 degenerations. . . . . . . . . . . . . 61
5.2 A capping operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.3 Sources of curves at east . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.4 Visualizing shue curves, I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.5 Visualizing shue curves, II. . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.6 Schematic of a height 2 holomorphic comb. . . . . . . . . . . . . . . . . . . . 71
5.7 /
B
(x. y; o

; 1) may be a proper subset of /

B
(x. y; o

; 1), I. . . . . . . . . . 73
5.8 /
B
(x. y; o

; 1) may be a proper subset of /

B
(x. y; o

; 1), II. . . . . . . . . 73
5.9 1-parameter family of maps from an odd shue component. . . . . . . . . . 80
5.10 Annulus at east that does not exist. . . . . . . . . . . . . . . . . . . . . . . 82
5.11 Expected and unexpected behavior of /(x. y; o

;

1) /(x. y; o

). . . . . . 87
5.12 Degenerations discussed in Theorem 5.55. . . . . . . . . . . . . . . . . . . . . 88
5.13 Domains of model curves used to compute inv
t
. . . . . . . . . . . . . . . . . . 94
5.14 Jittered intersection number. . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.15 Parts of the proof of Lemma 5.66. . . . . . . . . . . . . . . . . . . . . . . . . 96
6.1 Local illustrations of possible contributions to
2
in

CFD. . . . . . . . . . . . 105
6.2 Labeling of triangle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.3 Curves and
H
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.4 A triangle is a strip with a boundary puncture. . . . . . . . . . . . . . . . . . 113
6.5 A degeneration at
12
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.6 Straightening -curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
v
vi LIST OF FIGURES
6.7 Illustration of a boundary bigon degenerating, in proof of invariance. . . . . . 124
6.8 Curves ,
H
and
t
in . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
6.9 Two compositions of triangles, as degenerations of a rectangle. . . . . . . . . . 129
7.1 Local illustrations of terms occurring the /

relation for

CFA. . . . . . . . . 136
7.2 Domain where :
3
is necessary for associativity. . . . . . . . . . . . . . . . . . 137
8.1 First step in making a Heegaard diagram with boundary nice. . . . . . . . . . 146
9.1 Schematic of an ideal-matched comb . . . . . . . . . . . . . . . . . . . . . . . 159
9.2 Illustration of the bered product in the proof of Proposition 9.40 . . . . . . . 168
9.3 Degenerating the hexagon. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
11.1 Bordered Heegaard diagrams for torus boundary. . . . . . . . . . . . . . . . . 191
11.2 Labeling of regions around (torus case). . . . . . . . . . . . . . . . . . . . 192
11.3 From bordered diagrams to doubly-pointed diagrams. . . . . . . . . . . . . . . 199
11.4 Diagram for longitude of solid torus. . . . . . . . . . . . . . . . . . . . . . . . 201
11.5 Reconstructing

CFD from CFK. . . . . . . . . . . . . . . . . . . . . . . . . . 205
11.6 Boundary degeneration as end of a one-dimensional moduli space. . . . . . . . 207
11.7 Sketch of some of the maps from Theorem 11.35. . . . . . . . . . . . . . . . . 211
11.8 Twisting, in proof of Theorem 11.35. . . . . . . . . . . . . . . . . . . . . . . . 212
11.9 Homology classes in the winding region. . . . . . . . . . . . . . . . . . . . . . 217
11.10 Homology classes in the winding region. . . . . . . . . . . . . . . . . . . . . . 217
11.11 Notation for the local multiplicities in the winding region. . . . . . . . . . . . 219
11.12 Pinching along C
2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
11.13 Illustration of the proof of Theorem 11.27 . . . . . . . . . . . . . . . . . . . . 229
11.14 Diagram for the (2. 1) cable. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
11.15 Cable of the trefoil. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
A.1 Bordered Heegaard diagrams for dieomorphisms. . . . . . . . . . . . . . . . . 237
A.2 Heegaard diagrams for Dehn twists of the torus. . . . . . . . . . . . . . . . . . 241
A.3 The DA bimodules for generators of the genus-one mapping class group . . . . 243
A.4 Deformed Heegaard diagram for identity map of torus . . . . . . . . . . . . . 244
A.5 The bimodule

CFAA(". 0). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
A.6 The bimodule

CFDD(". 0). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
CHAPTER 1
Introduction
1.1. Background
Since the pioneering work of Simon Donaldson, techniques from gauge theory have taken
a central role in the study of smooth four-manifold topology [DK90]. His numerical in-
variants, associated to closed, smooth four-manifolds with /
+
2
1, have shed much light
on our understanding of dierential topology in dimension four. Moreover, these invariants,
and the subsequent closely-related Seiberg-Witten invariants [Wit94] and Heegaard Floer
invariants [OSz04d] all t into a formal framework reminiscent of the topological quantum
eld theories proposed by Witten [Wit88]. Crudely speaking, these theories have the fol-
lowing form. To a closed three-manifold 1 one associates a (suitably graded) abelian group,
the Floer homology of 1 , and, to a four-manifold \ with boundary identied with 1 , a
homology class in the Floer homology of the boundary. If a closed, smooth four-manifold
A decomposes along 1 into a union of two four-manifolds with boundary A
1
and A
2
, then
the numerical (Donaldson, Seiberg-Witten, or Heegaard Floer) invariant of the closed four-
manifold is obtained as a suitable pairing between the relative invariants coming from A
1
and A
2
in a corresponding version of Floer homology of 1 .
As the name suggests, the rst such construction was proposed by Andreas Floer (for a
restricted class of three-manifolds) as a tool for studying Donaldsons theory. A complete
construction of the corresponding three-dimensional Floer theory for Seiberg-Witten invari-
ants was given by Kronheimer and Mrowka [KM07]. Heegaard Floer homology was dened
by Zoltn Szab and the second author.
The aim of the present work is to perform a corresponding construction one dimen-
sion lower. Specically, we produce an invariant which, loosely speaking, associates to a
parametrized, closed, oriented surface 1 a dierential graded algebra /(1) and associates
to a three-manifold whose boundary is identied with 1 a dierential graded module over
/(1). When a closed, oriented three-manifold can be decomposed along 1 into two pieces
1
1
and 1
2
, a suitable variant of Floer homology is gotten as a pairing of the dierential
graded modules associated to 1
1
and 1
2
.
We give now a slightly more detailed version of this picture, starting with some more
remarks about Heegaard Floer homology, and then a more precise sketch of the invariants
constructed in the present book.
Recall that there are several variants Heegaard Floer homology stemming from the fact
that, in its most basic form, the Heegaard Floer homology of a three-manifold is the homology
of a chain complex dened over a polynomial ring in an indeterminant l. The full theory
can be promoted to construct invariants for closed, smooth four-manifolds [OSz06], similar
in character (and conjecturally equal) to Seiberg-Witten invariants. In this book, we focus
on the specialization of Heegaard Floer homology for three-manifolds in the case where
l = 0, giving the three-manifold invariant denoted

HF(1 ). The corresponding simplied
theory is not rich enough to construct interesting closed four-manifold invariants, but it
1
2 1. INTRODUCTION
does already contain interesting geometric information about the underlying three-manifold
including, for example, full information about the minimal genera of embedded surfaces
in 1 [OSz04a, Ni09].
With this background in place, we proceed as follows. Let 1 be a closed, oriented two-
manifold. Equip 1 with some additional data, in the form of a minimal handle decomposition
and a basepoint, data which we refer to as a pointed matched circle (see Denition 3.16) and
denote by :. This handle decomposition can be thought of as giving a parameterization of
1. We associate to : a dierential graded algebra /(:).
Next, x a three-manifold with boundary 1 together with an orientation-preserving
dieomorphism : 1 1 . To this data (1. ) we associate a left dierential module over
/(:), the type 1 module of 1 , denoted

CFD(1. : 1 1 ) or, less precisely,

CFD(1 ).
To (1. ) we also associate the type module of 1 , denoted

CFA(1. : 1 1 ) or

CFA(1 ).
The type module of 1 invariant is a more general type of object: it is a right /

module
over /(:). Both invariants are dened by counting holomorphic curves. The denitions
of both of these modules depend on a number of auxiliary choices, including compatible
Heegaard diagrams and associated choices of almost complex structures. However, as is
typical of Floer homology theories, the underlying (/

) homotopy equivalence types of the

modules are independent of these choices, giving rise to topological invariants of our three-
manifold with boundary. When speaking informally about these invariants, we refer to them
collectively as the bordered Heegaard Floer invariants of 1 .
The relationship with closed invariants is encapsulated as follows. Suppose that 1
1
and 1
2
are two three-manifolds with boundary, where 1
1
is identied with 1 and 1
2
is identied
with 1. Then we can form a closed three-manifold 1 by identifying 1
1
and 1
2
along their
boundaries. The pairing theorem for bordered Heegaard Floer homology states that the
Heegaard Floer complex

HF of 1 is homotopy equivalence to the /

tensor product of

CFA(1
1
) with

CFD(1
2
).
Recall that

HF can be calculated algorithmically for any closed three-manifold, thanks
to the important work of Sarkar and Wang [SW10]. Nevertheless, the present work (and
forthcoming extensions) makes it possible to compute it in innite families: by cutting a
3-manifold into simpler pieces along surfaces and computing an invariant for each piece, we
can reduce the computation to a computation for each piece and an algebraic computation.
Indeed, we expect this to lead to a more ecient algorithm for computation in general.
Moreover, bordered Heegaard Floer homology provides a conceptual framework for or-
ganizing the structure of

HF. We give a few of these properties in this book, and return to
further applications elsewhere (see Section 1.6).
1.2. The bordered Floer homology package
The formal algebraic setting for the constructions herein lie outside the working toolkit
for the typical low-dimensional topologist: /

modules over dierential graded algebras,

and their /

tensor products (though, we should point out, these objects are now familiar
in symplectic topology, cf. [Fuk93, Kon95, Sei02]). We will recall the basics before giving
precise statements of our results. But before doing this, we sketch the outlines of the ge-
ometry which underpins the constructions, giving informal statements of the basic results of
this package.
1.2. THE BORDERED FLOER HOMOLOGY PACKAGE 3
Figure 1.1. An example of cutting a Heegaard diagram. The Heegaard
diagram shown may be degenerated by cutting along (or pinching) the curve
marked 2. Each half of the result will be a bordered Heegaard diagram.
Central to this geometric picture is the cylindrical reformulation of Heegaard Floer ho-
mology [Lip06a]. Recall that Heegaard Floer homology in its original incarnation [OSz04d]
is an invariant associated to a Heegaard diagram for a three-dimensional manifold 1 . Start
from a Heegaard diagram for 1 , specied by an oriented surface , equipped with two p-
tuples of attaching circles
1
. . . . .
g
and
1
. . . . .
g
, as well as an additional basepoint
. in the complement of these circles. One then considers a version of Lagrangian Floer
homology in the p-fold symmetric product of . In the cylindrical reformulation, disks in
the p-fold symmetric product are reinterpreted as holomorphic curves in the four-manifold
[0. 1] 1, with coordinates (r. :. t), satisfying certain constraints (dictated by the at-
taching circles) at : 0. 1 and asymptotic constraints as t approaches .
Starting from this cylindrical reformulation, suppose that our Heegaard diagram for 1
is equipped with a closed, separating curve 2 in . We suppose that 2 meets the following
combinatorial requirements: it is disjoint from the -curvesso cutting along 2 and the
-curves results in a surface with two componentsand although we do not require that 2
be disjoint from the -curves, we still require that the result of cutting along 2 and the
-curves results in a surface with two components. The constructions in this book emerge
when one considers limits of holomorphic curves in [0. 1] 1 as the complex structure
on is pinched along 2.
In the limit as the circle 2 is pinched to a node, the holomorphic curves limit to holo-
morphic curves in ((
1
H
2
) [0. 1] 1), where here
1
and
2
are the two components of
2 equipped with a cylindrical ends. The holomorphic curves from these two halves have
constrained limiting behavior as they enter the ends of
1
and
2
: there are nitely many
values of t where our holomorphic curves are asymptotic to arcs in the ideal boundaries
i
.
The data of the asymptotics (the values of t and the arcs in
i
) must match in the limit.
Thus, we may reconstruct holomorphic curve counts for [0. 1] 1 from holomorphic
curve counts in
i
[0. 1] 1.
The data on the Heegaard surface specied by 2 has a more intrisic, three-dimensional
interpretation. The -circles meet 2 generically in a collection of points, which come in
pairs (those points which belong to the same
i
). This is the structure of a matched circle
mentioned above. It species a closed surface 1. This can be seen in terms of Morse
functions, as follows. Think about the Heegaard diagram as the intersections of the ascending
and descending disks of a self-indexing Morse function on 1 with the middle level in the
usual way. We then construct 1 from 2 by owing 2 backwards and forwards under the
Morse ow. Concretely, 1 is obtained from a disk with boundary 2 by attaching one-handles
along the matched pairs of points, and then lling in the remaining disk with a two-handle.
(Our assumptions guarantee there is only one two-handle to be added in the end.) The two
4 1. INTRODUCTION
halves of a Heegaard diagram specify three-manifolds 1
1
and 1
2
meeting along their common
boundary 1.
In general terms, the dierential graded algebra /(:) is constituted from the Reeb
chords of a corresponding matched circle :. (The reader should not be thrown by the use
of the term Reeb chord here: the ambient contact manifold here is the circle, and hence
the Reeb chords are simply arcs in a circle, with boundaries on the feet of the 1-handles.
That said, the terminology is not entirely pointless; see Chapter 5.) The product on the
algebra and its dierential are dened combinatorially, but rigged to agree with (relatively
simple) holomorphic curve counts in the cylinder which interpolates between Reeb chords.
The precise algebra is dened in Chapter 3.
The type 1 module for the component 1
2
is dened as a chain complex over /(:),
generated by p
2
-tuples of intersection points of various of the
i
and
j
contained in
2
, with
a dierential given as a weighted count of rigid holomorphic curves, where the coecient
in /(:) of a rigid holomorphic curve measures the asymptotics of its Reeb chords at its
boundary. The precise denition is given in Chapter 6. In that chapter, we also prove a
more precise version (Theorem 6.16) of the following:
Theorem 1.1. Let 1
2
be a three-manifold equipped with an orientation-preserving dieomor-
phism : 1 1
2
. The homotopy equivalence class of the associated dierential module

CFD(1
2
. : 1 1
2
) is a topological invariant of the three-manifold 1
2
with boundary
parameterized by 1.
For the type module for the component 1
1
, on the other hand, the algebraic operations
are dened by counting holomorphic curves that are rigid when subject to certain height
constraints on their Reeb chords. Specically, we require that certain clumps of these Reeb
chords occur at the same height. The proper algebraic set-up for these operations is that
of an /

module over /(:), and the precise denition of the type module is given in
Chapter 7, where we also prove a more precise version (Theorem 7.17) of the following:
Theorem 1.2. Let 1
1
be a three-manifold equipped with an orientation-preserving dieo-
morphism : 1 1
1
. The /

homotopy equivalence class of the /

module

CFA(1
1
. :
1 1
1
) is a topological invariant of the three-manifold 1
1
with boundary parameterized
by 1.
Of course, dierential graded modules over a dierential graded algebra are special cases
of /

modules. Moreover, there is a pairing, the /

tensor product, between two /

modules `
1
and `
2
, giving rise to a chain complex `
1

`
2
, whose chain homotopy type
depends only on the chain homotopy types of the two factors; this generalizes the derived
tensor product of ordinary modules.
The Heegaard Floer homology of 1 can be reconstituted from the bordered Heegaard
Floer homology of its two components according to the following pairing theorem:
Theorem 1.3. Let 1
1
and 1
2
be two three-manifolds with parameterized boundary 1
1
=
1 = 1
2
, where 1 is specied by the pointed matched circle :. Fix corresponding bordered
Heegaard diagrams for 1
1
and 1
2
. Let 1 be the closed three-manifold obtained by gluing 1
1
and 1
2
along 1. Then

CF(1 ) is homotopy equivalent to the /

tensor product of

CFA(1
1
)
and

CFD(1
2
). In particular,

HF(1 )

= H

CFA(1
1
)

,(:)

CFD(1
2
)
_
.
1.4. THE CASE OF THREE-MANIFOLDS WITH TORUS BOUNDARY 5
We give two proofs of the above theorem. One proof (in Chapter 8) makes use of the
powerful technique of Sarkar and Wang [SW10]: we construct convenient Heegaard dia-
grams for 1
1
and 1
2
, where the holomorphic curve counts can be calculated combinatorially.
For such diagrams, the /

structure of the type module simplies immensely (higher

multiplications all vanish), /

tensor products coincide with traditional tensor products,

and the proof of the pairing theorem becomes quite simple. The other proof (in Chapter 9)
involves a rescaling argument to identify

CF(1 ) with another model for the /

tensor prod-
uct. This second proof gives geometric insight into the connection between the analysis and
the algebra. It also leads to pairing theorems for triangle and polygon maps [LOTa, LOTb].
1.3. On gradings
One further surprising aspect of the theory deserves further comment is the structure of
gradings.
Gradings in the Heegaard Floer theory of closed 3-manifolds have an unconventional form:
for each spin
c
structure there is a relative grading taking values in a cyclic abelian group.
Alternatively, this can be viewed as an absolute grading valued in some Z-set. Indeed, for
the case of Floer homology of Seiberg-Witten monopoles [KM07], this Z-set has an elegant
formulation as the set of isotopy classes of non-vanishing vector elds on the 3-manifold.
Gradings in bordered Floer homology have a correspondingly even less conventional form.
The algebra of a surface is graded by the Heisenberg group associated to the surfaces in-
tersection form. (This group also has a geometric interpretation; see Remark 3.48.) The
gradings for bordered Heegaard Floer modules take values in sets with actions of this Heisen-
berg group. This is explained in Chapter 10.
1.4. The case of three-manifolds with torus boundary
As explained in Chapter 11, for three-manifolds with boundary the torus 1, the (ap-
propriate summand of the) algebra /(1) is particularly simple: it is nite-dimensional (an
eight-dimensional subalgebra of 4 4 upper-triangular matrices) and has vanishing dieren-
tial.
The bordered Heegaard Floer homology of a solid torus bounded by 1 can be easily
calculated; see also [Lip06b, Section 5.3]. A fundamental result in Heegaard Floer ho-
mology, the surgery exact triangle, gives a long exact sequence relating the Heegaard Floer
homology groups of three three-manifolds which are obtained as dierent llings of the same
three-manifold with torus boundary. (The result was rst proved in [OSz04c], inspired by
a corresponding result of Floer for the case of his instanton homology, see [Flo95], compare
also [KMOSz07] for a Seiberg-Witten analogue.) This result can now be seen as a conse-
quence of the pairing theorem, together with a short exact sequence relating three type 1
modules associated to three dierent solid tori with boundary 1.
In a related vein, recall [OSz04b, Ras03] that there is a construction of Heegaard Floer
homology theory for knots 1 in a three-manifold 1 . Information about this knot Floer
homology turns out to determine the bordered Floer homology of the knot complement.
This is stated precisely and worked out in Theorem 11.27 in Section 11.7 for large framings
and in Theorem A.11 in Appendix A for general framings, using some results from [LOT10a].
This result provides a number of explicit, non-trivial examples of bordered Floer homology.
Indeed, using this connection between knot Floer homology and bordered Heegaard Floer
homology, and combining it with an adaptation of Theorem 1.3 (stated in Theorem 11.21
6 1. INTRODUCTION
below), we are able to calculate knot Floer homology groups of satellite knots in terms of
the ltered chain homotopy type of the knot ltration of the pattern knot, together with
some information (a type module) associated to the pattern. We illustrate this in some
concrete examples, where the type module can be calculated explicitly. These ideas are
further pursued in [Pet09, Lev09, Lev10].
1.5. Previous work
Some of the material in this book, especially from Chapters 4 and 5, rst appeared in
the rst authors Ph.D. thesis [Lip06b]. As in that work, we construct here an invariant
of bordered manifolds that lives in a suitable category of chain complexes. By modifying
the algebra and modules introduced in that work appropriately we are able to reconstruct
the invariant of a closed 3-manifold. Related constructions in the case of manifolds with
torus boundary have been worked out by Eftekhary [Eft08]. For the case of satellite knots,
there is extensive work by Hedden, including [Hed05a], which we used as a check of some of
our work. In a dierent direction, an invariant for manifolds-with-boundary equipped with
sutures on the boundary has been given by Andrs Juhsz [Juh06].
An introduction to some of the structures used in this book, in the form of a toy model
illustrating key features, is given in a separate paper [LOT09].
1.6. Further developments
The theory presented here is in some ways rather limited as a theory of 3-manifolds with
boundary: we deal only with a single, connected boundary component, for the somewhat
limited

HF theory. We return to these points in other papers: in other papers.
For a 3-manifold ` with two boundary components, together with parametriza-
tions of these boundary components and a framed arc connecting them we associate
several types of bimodules [LOT10a]. These bimodules are well-dened up to
homotopy equivalence, and satisfy appropriate pairing theorems. Some parts of this
theory are sketched in Appendix A.
As a special case, these bimodules allow us to change the parametrization of the
boundary. (For instance, if 1 is a torus this corresponds to changing the framing
on a knot.) This gives a representation of the mapping class group of 1 on the
derived category of modules over /(:). This mapping class group action is faith-
ful [LOT10d]. (Bimodules for generators of the mapping class group of the torus
are given explicitly in Appendix A.)
The type and 1 versions of the bordered invariant of a 3-manifold with boundary
are also related by certain dualizing bimodules [LOT10a]. (Again, this is sketched
in a little more detail in Appendix A.) They are also related by another kind of
duality, using the Hom-functor [LOT10e]. In particular,

CFA(1 ) can be calculated
explicitly from

CFD(1 ), and vice-versa.
Decomposing a three-manifold into simple pieces, and calculating the (bi)modules
associated to the pieces, one can calculate the Heegaard Floer invariant

HF for any
3-manifold, as well as the bordered invariants of any bordered 3-manifold [LOT10c].
Similar techniques, together with a pairing theorem for polygon maps, allow one to
compute the spectral sequence from Khovanov homology to

HF of the branched
double cover and, more generally, the link surgery spectral sequence [LOT10b,
1.7. ORGANIZATION 7
LOTa]. These methods also allow one to compute the maps on

HF associated to
4-dimensional cobordisms [LOTb].
Andrs Juhsz has constructed an invariant of sutured manifolds, called sutured
Floer homology [Juh06], which has been remarkably eective at studying geomet-
ric questions [Juh08]. Rumen Zarev has developed a bordered-sutured Floer theory
generalizing both bordered Floer homology and sutured Floer homology. This the-
ory allows one to recover the sutured invariants by tensoring the bordered Floer
invariants with a module recording the sutures [Zar09]. In turn, it also shows that
much of bordered Floer theory can be captured in terms of sutured Floer homol-
ogy [Zar10].
1.7. Organization
In Chapter 2 we briey recall the language of /

modules. Most of the material in that

chapter is standard, with the possible exception of a particular model, denoted , for the
/

tensor product. This model is applicable when one of the two factors has a particular
simple algebraic structure, which we call a type 1 structure; the modules

CFD have this
form.
In Chapter 3 we construct the dierential graded algebra associated to a closed, oriented
surface. Chapter 4 collects a number of basic properties of Heegaard diagrams for three-
manifolds with (parameterized) boundary, including when two such diagrams represent the
same 3-manifold. The generators of the bordered Floer modules are also introduced there,
and we see how to associate spin
c
-structures to these and discuss domains connecting them
(when they represent the same spin
c
-structure). In Chapter 5, we collect the technical
tools for moduli spaces of holomorphic curves which are counted in the algebraic structure
underlying bordered Floer homology.
With this background in place, we proceed in Chapter 6 to the denition of the type 1
module, and establish its invariance properties sketched in Theorem 1.1. In Chapter 7 we
treat the case of type modules, establishing a precise version of Theorem 1.2. We turn
next to the pairing theorem. Chapter 8 gives one proof of Theorem 1.3, using so-called nice
diagrams (in the sense of [SW10]). As a bi-product, this gives an algorithm to calculate
the bordered Floer homology of a three-manifold. We give a second, more analytic proof of
Theorem 1.3 in Chapter 9.
While gradings on the algebra are discussed in Chapter 3, discussion of gradings on the
invariants of bordered 3-manifolds and a graded version of the pairing theorem are deferred
to Chapter 10,
We conclude the book proper with a chapter on the torus boundary case, Chapter 11.
After reviewing the algebra associated to the torus, we use a computation of the bordered
Floer homology for solid tori to give a quick proof of the surgery exact triangle for

HF. In
this chapter, we also relate this invariant with knot Floer homology. We start by showing
how to recapture much of knot Floer homology from the bordered Floer invariants of a knot
complement. We then go the other direction, showing that the bordered Floer homology of
knot complements in o
3
(with suitable framings) can be computed from knot Floer homology;
this is Theorem 11.27.
In Appendix A, we give a quick introduction to [LOT10a], which gives a generalization
of the present work to the case of three-manifolds with two boundary components. Using
an important special case, one can describe the dependence of the bordered invariants on
8 1. INTRODUCTION
the parameterization of the boundary. We give explicit answers for this dependence in the
case of torus boundary. Using these computations, we are able to extend Theorem 11.27 to
arbitrary framings.
Acknowledgements
We thank Dror Bar-Natan, Eaman Eftekhary, Yakov Eliashberg, Matthew Hedden,
Mikhail Khovanov, Aaron Lauda, Dusa McDu, Tim Perutz and Zoltn Szab for helpful
conversations. We are also grateful to Atanas Atanasov, Tova Brown, Christopher Douglas,
Matthew Hedden, Jennifer Hom, Yank Lekili, Adam Levine, Ciprian Manolescu, Yi Ni, Ina
Petkova and Vera Vertsi for useful remarks on versions of this manuscript. We also thank
the valiant referee for a careful reading and many helpful suggestions.
CHAPTER 2
/

structures
In this work, we extensively use of the notion of an /

module. Although /

notions,
rst introduced by Stashe in the study of H-spaces [Sta63], have become commonplace
now in symplectic geometry (see for example [Fuk93, Kon95, Sei02]), they might not be
so familiar to low-dimensional topologists. In Sections 2.1 and 2.2, we review the notions of
primary importance to us now, in particular sketching the /

tensor product. Keller has

another pleasant exposition of this material in [Kel01].
The type 1 module

CFD(1) dened in Chapter 6 is an ordinary dierential graded
module, not an /

module, so the reader may wish to skip this chapter and concentrate on
the type 1 structure at rst. Similarly, the type module

CFA(1) of a nice diagram 1 (in
the technical sense, see Denition 8.1, which in turn is motivated by [SW10]) is an ordinary
dierential graded module. However, the notion of grading by a non-commutative group
(Section 2.5) is used in all cases.
In Section 2.3, we introduce a further algebraic structure which naturally gives rise to
a module, which we call a type 1 structure. In Section 2.4, we also give an explicit and
smaller construction of the /

tensor product when one of the two factors is induced from

a type 1 structure. Although type 1 structures have appeared in various guises elsewhere
(see Example 2.19 and Remark 2.25), this tensor product is apparently new. As the name
suggests, the type 1 module of a bordered three-manifold comes from a type 1 structure.
Finally, in Section 2.5 we introduce gradings of dierential graded algebras or /

algebras
with values in a non-commutative group.
2.1. /

algebras and modules

Although we will be working over dierential graded (dg) algebras (which are less general
than /

algebras) we recall the denition of /

algebras here as it makes a useful warm-up

for dening for /

modules, which we will need in the sequel.

To avoid complicating matters at rst, we will work in the category of Z-graded complexes
over a xed ground ring k, which we assume to have characteristic two. For our purposes,
this ground ring k will typically be a direct sum of copies of F
2
= Z,2Z. In particular, we
consider modules ` over k which are graded by the integers, so that
` =

dZ
`
d
.
If ` is a graded module and : Z, we dene `[:] to be the graded module dened by
`[:]
d
= `
dn
.
Note that since we are working over characteristic two and dont need to worry about
signs, the following discussion also carries over readily to the ungraded setting. The modules
in our present application are graded, but not by Z. We return to this point in Section 2.5.
9
10 2. A STRUCTURES
Denition 2.1. Fix a ground ring k with characteristic two. An /

algebra / over k is a
graded k-module , equipped with k-linear multiplication maps
j
i
:
i
[2 i]
dened for all i 1, satisfying the compatibility conditions
(2.2)

i+j=n+1
nj+1

=1
j
i
(c
1
c
1
j
j
(c

c
+j1
) c
+j
c
n
) = 0
for each : 1. Here,
i
denotes the k-module
i
..

k

k
. Note that throughout this
chapter, all tensor products are over k unless otherwise specied. We use / for the /

algebra and for its underlying k-module. An /

algebra / is strictly unital if there is an

element 1 / with the property that j
2
(c. 1) = j
2
(1. c) = c and j
i
(c
1
. . . . . c
i
) = 0 if i ,= 2
and c
j
= 1 for some ,.
In particular, an /

algebra is a chain complex over k, with dierential j

1
. In the
case where all j
i
= 0 for i 2, an /

algebra is just a dierential graded algebra over

k, with dierential j
1
and (associative) multiplication j
2
. We have assumed that k has
characteristic two; in the more general case, the compatibility equation must be taken with
signs (see, e.g., [Kel01]).
We can think of the algebraic operations graphically as follows. The module
i
is
denoted by drawing i parallel, downward-oriented strands. The multiplication operation
j
i
is represented by an oriented, planar tree with one vertex, i incoming strands and one
outgoing strand. With this convention, the compatibility relation with : inputs can be
visualized as follows. Fix a tree 1 with one vertex, : incoming edges, and one outgoing one.
Consider next all the planar trees o (with two vertices) with the property that if we contract
one edge in o, we obtain 1. Each such tree o represents a composition of multiplication
maps. The compatibility condition states that the sum of all the maps gotten by these
resolutions o vanishes. For example,
j
1
j
3
+
j
1
j
3
+
j
1
j
3
+
j
1
j
3
+
j
2
j
2
+
j
2
j
2
= 0.
Note that the last two terms are the usual associativity relation, and the remaining terms
say that associativity holds only up to a homotopy.
Another way to think of the compatibility conditions (Equation (2.2)) uses the tensor
algebra
T

([1]) :=

n=0

n
[:].
The maps j
i
can be combined into a single map
j: T

([1]) [2]
2.1. A ALGEBRAS AND MODULES 11
with the convention that j
0
= 0. We can also construct a natural degree 1 endomorphism
1: T

([1]) T

([1]) by
(2.3) 1(c
1
c
n
) =
n

j=1
nj+1

=1
c
1
j
j
(c

c
+j1
) c
n
.
Then the compatibility condition is that
j 1 = 0
or equivalently
1 1 = 0.
These conditions can also be expressed graphically as
1
j
or
1
1
.
Here, the doubled arrows indicate elements of T

([1]), while the single arrows indicate

elements of .
At some points later we will want to allow only nitely many nonzero j
i
. We give this
condition a name:
Denition 2.4. An /

algebra (. j
i

i=1
) is operationally bounded if j
i
= 0 for i su-
ciently large.
Denition 2.5. A (right) /

module / over / is a graded k-module `, equipped with

operations
:
i
: `
(i1)
`[2 i].
dened for all i 1, satisfying the compatibility conditions
0 =

i+j=n+1
:
i
(:
j
(x c
1
c
j1
) c
n1
)
+

i+j=n+1
nj

=1
:
i
(x c
1
c
1
j
j
(c

c
+j1
) c
n1
).
(2.6)
An /

module / over a strictly unital /

algebra is said to be strictly unital if for any

x `, :
2
(x1) = x and :
i
(xc
1
c
i1
) = 0 if i 2 and some c
j
= 1. The module
/ is said to be bounded if :
i
= 0 for all suciently large i.
Graphically, the module `
i1
is represented by i parallel strands, where the leftmost
strand is colored with ` (while the others are colored by ). The multiplication map :
i
is represented by the one-vertex tree with i incoming strands, the leftmost of which is `-
colored, and whose output is also `-colored. The compatibility condition can be thought of
the same as before, except now we have distinguished the leftmost strands with `. Thus,
the compatibility conditions can be thought of as stating the vanishing of maps induced by
sums of planar trees, with a xed number of inputs (and one output).
12 2. A STRUCTURES
As before, we can also write this condition in terms of the tensor algebra. The maps :
i
(and the j
i
) can be combined to form a degree 1 map :: ` T

([1]) ` T

([1]),
using the following analogue of Equation (2.3):
:(x
1
c
2
c
n
) =
n

=1
:

(x c
2
c

) c
n
+
n

j=2
nj+1

=1
c
1
j
j
(c

c
+j1
) c
n
. (2.7)
The /

relation is equivalent to the condition that : : = 0.

The /

relation in tensor form can also be expressed graphically. To do this conveniently,

let : T

be the canonical comultiplication

(c
1
c
n
) =
n

m=0
(c
1
c
m
) (c
m+1
c
n
).
If we use a dashed line to denote an element of / then the compatibility condition for an
/

module can be drawn as

:
:
+
1
:
= 0.
Example 2.8. Let / be a right /

module over /. Suppose moreover that, for all i 2,

:
i
= 0 and j
i
= 0. Then not only is / a dierential graded algebra, but also / is a module
over / (in the traditional sense), equipped with a dierential satisfying the Leibniz rule with
respect to the algebra action on the module, i.e., a dierential graded module.
Denition 2.9. Let /and /
t
be strictly unital right /

modules over a strictly unital /

algebra. A strictly unital homomorphism ) of /

modules, or simply an /

homomorphism,
is a collection of maps
)
i
: `
(i1)
`
t
[1 i]
indexed by i 1, satisfying the compatibility conditions
0 =

i+j=n+1
:
t
i
()
j
(x c
1
c
j1
) c
n1
)
+

i+j=n+1
)
i
(:
j
(x c
1
c
j1
) c
n1
)
+

i+j=n+1
nj

=1
)
i
(x c
1
c
1
j
j
(c

c
+j1
) c
n1
)
and the unital condition
)
i
(x c
1
c
i1
) = 0
2.1. A ALGEBRAS AND MODULES 13
if i 1 and some c
j
= 1.
We call a strictly unital homomorphism )
i
of /

modules bounded if )
i
= 0 for i
suciently large.
The compatibility condition can be formulated in terms of the tensor algebra as follows.
Promote the maps )
i

i=1
to a degree 0 map
) : ` T

([1]) `
t
T

([1])
by the formula:
(2.10) )(x
1
c
1
c
n
) =
n

=0
)
+1
(x c
1
c

) c
n
.
The /

relation now is equivalent to the condition that the map ) induces a chain map
from the chain complex ` T

([1]) with its induced dierential : (Equation (2.7)) to

the chain complex `
t
T

([1]) with its induced dierential :

t
.
The compatibility condition for /

homomorphisms can be drawn as

:
)
+

)
:
t
+
1
)
= 0.
Here, we use color to distinguish between / and /
t
.
For example, for any /

module /, the identity homomorphism " is the map with

"
1
(x) := x
"
i
(x
i1
) := 0 (i 0).
If ) is an /

homomorphism from / to /
t
, and p is an /

homomorphism from /
t
to /
tt
, we can form their composite p ), dened by
(p ))
n
(x (c
1
c
n1
)) :=

i+j=n+1
p
j
()
i
(x c
1
c
i1
) c
n1
).
Denition 2.11. Given any collection of maps
/
i
: `
i1
`
t
[i]
14 2. A STRUCTURES
with /
i
(x c
1
c
i1
) = 0 if i 1 and some c
j
= 1, we can construct a strictly unital
/

homomorphism ) by the expression

)
n
(x c
1
c
n1
) =

i+j=n+1
/
i
(:
j
(x c
1
c
j1
) c
n1
)
+

i+j=n+1
:
t
i
(/
j
(x c
1
c
j1
) c
n1
)
+

i+j=n+1
nj

=1
/
i
(x c
1
c
1
j
j
(c

c
+j1
) c
n1
).
If an /

homomorphism ) can be obtained from a map / in this way, we call ) null

homotopic. Two /

homomorphisms ). p : / /
t
are homotopic if their dierence is null
homotopic.
Two /

modules / and /
t
are (/

) homotopy equivalent if there are /

homomor-
phisms ) from / to /
t
and p from /
t
to / such that ) p and p ) are homotopic to
the identity.
Suppose that ). p : / /
t
are homotopic via a homotopy /. Then, we can promote
the components of /, /
i

i=1
to a map
/: ` T

([1]) `
t
T

([1])
following Equation (2.10). The condition that / gives the homotopy between ) and p is
equivalent to the more familiar condition
/ : + :
t
/ = ) + p.
Graphically, this equation is

:
/
+

/
:
t
+
1
/
= ) +
p
.
There are analogous denitions for left /

modules, maps, and homotopies.

2.2. /

tensor products
Consider the category of strictly unital /

modules over a dg algebra , whose morphism

sets are the strictly unital /

homomorphisms. In this category, the usual nave tensor

product `
A
` of a right (/

) module ` and a left module ` is not available: the

relation (: c) : :(c :) used in the denition of the tensor product is not transitive
in the /

setting. In fact, even in the original setting of dierential graded modules, this
tensor product is often not the right one: if ` and `
t
are quasi-isomorphic, then `
A
`
and `
t

A
` are not necessarily quasi-isomorphic.
2.2. A TENSOR PRODUCTS 15
To produce a functor which works for /

modules, we pass instead to the /

tensor
product, a generalization of the derived tensor product.
Denition 2.12. Let / be an /

algebra over k, / be a right /

module over / and A

be a left /

module over /. Then their /

tensor product is the chain complex

/

,
A := ` T

([1]) `
equipped with the boundary operator
(x c
1
c
n
y) :=
n+1

i=1
:
i
(x c
1
c
i1
) c
n
y
+
n

i=1
ni+1

=1
x c
1
j
i
(c

c
+i1
) c
n
y
+
n+1

i=1
x c
1
:
i
(c
ni+2
c
n
y).
Lemma 2.13. The operator on /

A dened above satises

2
= 0.
Proof. This is a straightforward consequence of the relations dening /

algebras and
modules.
An important case is when A = /. The resulting complex is called the bar resolution /
of /. As a k-module it is
`

= ` T

([1])

= (` T
+
([1]))[1]
where T
+
([1]) :=

i=1
([1])
i
is T

([1]) without the rst summand of k. / has the

further structure of a right /

module over /, with multiplications :

1
= as dened above
and, for i 2,
:
i
((x c
1
c
n
) /
1
/
i1
) :=
n

=1
x c
1
c
n
j
i+1
(c
n+1
c
n
/
1
/
i1
).
Note the range of summation: the multiplication j
i+1
is applied to all of the /
j
s and at
least one c
j
.
The following lemma is again straightforward and standard.
Lemma 2.14. / is an /

module over /.
Notice that if / is an honest dierential graded algebra then / is an honest dierential
graded module, i.e., all of the higher :
i
vanish. Furthermore, if / is also an honest
dierential graded module, this is the usual bar resolution. If in addition A is an honest
dierential graded module, then we could also dene /

A as /
A
A, the nave tensor
product with the bar resolution, but this is not possible in general.
16 2. A STRUCTURES
Remark 2.15. One way to understand the products on / and Lemma 2.14 is via /

bimodules. If / and B are two /

algebras, an /B bimodule A is a k-module ` with

maps
:
i,j
:
i
` 1
j
`
satisfying a natural version of the compatibility conditions. In particular, / is an //
bimodule in a natural way, and / is just /

,
/. A more general version of Lemma 2.14
is that if / is a / module and A is an /B bimodule, then /
,
A is a right B module.
Proposition 2.16. If / and / are strictly unital, then / is homotopy equivalent to /.
Proof. We dene an /

module map from / to / by

i
((x c
1
c
n
) /
1
/
i1
) := :
i+n
(x c
1
c
n
/
1
/
i1
).
Similarly, we dene an /

module map from / to / by

i
(x c
1
c
i1
) := x c
1
c
i1
1.
Also dene maps for a homotopy / from / to / by
/
i
((x c
1
c
n
) /
1
/
i1
) := x c
1
c
n
/
1
/
i1
1.
It is straightforward to verify that is the identity map, while is homotopic, via /,
to the identity map.
Proposition 2.17. An /

map ) : / /
t
induces a chain map )

": /

A
/
t

A. If ) is null homotopic, so is )

"
A
. In particular, if /
1
and /
2
are homotopy
equivalent right /

modules and A
1
and A
2
are homotopy equivalent left /

modules, then
/
1

A
1
is homotopy equivalent to /
2

A
2
.
Proof. For ) a collection of maps )
i
: `
(i1)
`[c i] for a xed c 0. 1,
dene )

"
A
by
()

"
A
)(x c
1
c
n
y) :=
n+1

i=1
)
i
(x c
1
c
i1
) c
i
c
n
y.
If ) is an /

map, so is )

"
A
, and if / is a null homotopy of ), then /

"
A
is a null
homotopy of )

"
A
. The same construction works, mutatis mutandis, for chain maps and
homotopies on the right.
2.3. Type 1 structures
There is a smaller model for the derived tensor product in the case where one of the
two factors has a special type: where it is the /

module associated to a type 1 structure

(Denitions 2.18). In this section, we introduce these type 1 structures and develop their
basic properties.
We rst specialize to the case where / is a dierential graded algebra, which is all that
we need for our present purposes.
Denition 2.18. Fix a dg algebra /. Let ` be a graded k-module, equipped with a map

1
N
: ` ( `)[1].
satisfying the compatibility condition that
(j
2
"
N
) ("
A

1
)
1
+ (j
1
"
N
)
1
: ` `
2.3. TYPE D STRUCTURES 17
vanishes. (When the module ` is xed, we drop the subscript ` from the notation for

1
.) We call the pair (`.
1
N
) a type 1 structure over with base ring k. Let (`
1
.
1
N
1
) and
(`
2
.
1
N
1
) be two type 1 structures. A k-module map
1
: `
1
`
2
is a 1-structure
homomorphism if
(j
2
"
N
2
) ("
A

1
)
1
N
1
+ (j
2
"
N
2
) ("
A

1
N
2
)
1
+ (j
1
"
N
2
)
1
= 0.
Given type 1 homomorphisms
1
: `
1
`
2
and
1
: `
2
`
3
, their composite
(
1

1
) is dened by (j
2
"
N
3
) ("
A

1
)
1
. A homotopy / between two 1-structure
homomorphisms
1
and
1
from (`
1
.
1
N
1
) to (`
2
.
1
N
2
) is a k-module map /: `
1
(
`
2
)[1] with
(j
2
"
N
2
) ("
A
/)
1
N
1
+ (j
2
"
N
2
) ("
A

1
N
2
) / + (j
1
"
N
2
) / =
1

1
.
Example 2.19. Assume is a dg algebra over k = F
2
. Let ` be a dg module which is free
as an -module. For a xed basis of ` over , let A denote the k-span of that basis. The
restriction of the boundary operator to A gives a map

1
: A /A = `
which satises the relations of a type 1 structure. Maps between such modules are 1-
structure homomorphisms.
Lemma 2.20. If (`.
1
) is a type 1 structure, then ` can be given the structure of a
left module, with
:
1
(c y) := [(j
2
"
N
) ("
A

1
) + j
1
"
N
](c y)
:
2
(c
1
(c y)) := j
2
(c
1
c) y.
Moreover, if
1
: `
1
`
2
is a 1-structure homomorphism, then there is an induced
map of dierential graded modules from `
1
to `
2
, dened by
(c y) (:
2
"
N
2
) ("
A

1
)
Similarly, a homotopy between two type 1-structure homomorphisms induces a chain homo-
topy between the associated chain maps.
Proof. The proof is straightforward.
Given a type 1 structure (`.
1
) over a dg algebra , we denote the -module from
Lemma 2.20 by A.
Note that the structure equation for a type 1 structure (`.
1
) is equivalent to the
condition that :
1
on A is, indeed, a dierential.
As a partial converse to Lemma 2.20, we have the following lemma. Recall that if
`
1
and `
2
are two dg modules over a dg algebra, then a dg homomorphism is a chain
map : `
1
`
2
which commutes with the algebra action. Equivalently, it is an /

homomorphism
i
whose components
i
vanish for all i 1.
Lemma 2.21. Let `
1
and `
2
be two type 1 structures over a dg algebra , and A
1
and A
2
be their associated dg modules. The correspondence from Lemma 2.20 gives an isomorphism
between the space of type 1 homomorphisms from `
1
to `
2
with the space of dg homomor-
phisms from A
1
to A
2
. Moreover, two type 1 homomorphisms are homotopic if and only if
the corresponding dg homomorphisms are homotopic by an -equivariant homotopy.
18 2. A STRUCTURES
Proof. Given : A
1
A
2
, we can realize as the homomorphism associated to the
map
1
: `
1
/ `
2
determined by
1
(x) = (1 x). Similarly, if /: A
1
A
2
is an
-equivariant homotopy (i.e., /(c r) = c /(r) for all c and r A
1
, then we can dene
/
1
(x) = /(1 x) for any x `
1
. It is straightforward to verify that these maps satisfy the
required properties.
Denition 2.22. Let be an /

algebra and ` a graded left k-module. A map

1
:
` ( `)[1] can be iterated to construct maps

k
: ` (
k
`)[/]
inductively by

0
= "
N

i
=
_
"
A
(i1)
1
_

i1
.
By construction, the these maps satises the basic equation
("
A
j
i
)
j
=
i+j
for all i. , 0, or graphically

.
We say that
1
is bounded if for all x `, there is an integer : so that for all i :,

i
(x) = 0.
We call a type 1 structure bounded if its structure map
1
is.
For a bounded map
1
, we can think of the
k
as tting together to form a map
: ` T

([1]) `.
(x) :=

k=0

k
(x).
If
1
is not bounded then, instead, the
k
t together to form a map to the completed tensor
algebra:
: ` T

([1]) ` :=
_

i=0
([1])
i
_
`.
(x) :=

k=0

k
(x).
We can also give a tensor algebra version of Denition 2.18. This also allows us to
generalize it to the /

setting.
2.3. TYPE D STRUCTURES 19
Denition 2.23. Fix an /

algebra / and a pair (`.

1
) as above. Assume that either
(`.
1
) is bounded or / is operationally bounded. Then we say (`.
1
) denes a type 1
structure if
(j "
N
) = 0.
or equivalently
(1 "
N
) = 0.
or graphically

1
= 0.
For two type 1 structures (`
1
.
1
N
1
) and (`
2
.
1
N
2
), a map
1
: `
1
`
2
can be used
to construct maps

k
: `
1
(
k
`
2
)[/ 1]

k
(x) :=

i+j=k1
("
A
(i+1)
j
N
2
) ("
A
i
1
)
i
N
1
.
The map
1
is bounded if
k
= 0 for / suciently large. (This is automatic if `
1
and `
2
are bounded.) We can dene
(x) :=

k=1

k
(x).
which maps to T

([1]) ` if
1
is bounded and T

([1]) ` in general. Assuming either

1
is bounded or / is operationally bounded, we say is a 1-structure homomorphism if
(j "
N
2
) = 0.
Similarly, is homotopic to if
(j "
N
2
) / =
for / constructed from some map /
1
: ` (`
2
)[1] in the analogous way (and with /
1
bounded if / is not operationally bounded).
The main complication with type 1 structures over an /

algebra is that the category of

type 1 structures over an /

algebra is an /

category, with composition given graphically

20 2. A STRUCTURES
as follows:

k
()
1
. . . . . )
k
) =

M
0
()
1
)
1

M
1
()
2
)
1
.
.
.
()
k
)
1

M
k
.
.
.
j
Note that the condition
1
()) = 0 is exactly the condition of being a type 1 homomorphism.
Specializing Denition 2.23 to the case of a dg algebra, we obtain Denition 2.18.
If (`
1
.
1
N
1
) and (`
2
.
1
N
2
) are both bounded then any homomorphism
1
: `
1
`
2
between them is automatically bounded.
There is a generalization of Lemma 2.20.
Lemma 2.24. Let / be an operationally bounded /

algebra, and (`.

1
) a type 1 structure
over it. Then ` can be given the structure of a left /

module A, with
:
i
:=

k=0
(j
i+k
"
N
) ("
A
i
k
).
Moreover, if
1
: `
1
`
2
is a 1-structure homomorphism, then there is an induced
/

homomorphism from A
1
to A
2
, whose i
th
component is given by

k=1
(j
i+k+1
"
N
1
)
k
.
Similarly, a homotopy between two 1-structure homomorphisms induces a homotopy between
the associated /

maps.
As in the case for dierential graded algebras, we will denote the /

module associated
to a type 1 structure (`.
1
) constructed in Lemma 2.20 by A.
2.4. ANOTHER MODEL FOR THE A TENSOR PRODUCT 21
Remark 2.25. Type 1 structures have appeared elsewhere in various guises. For example,
let ( be a small dg category. Dene an algebra /(() by
/(() =

a,b,Ob(()
Mor(c. /).
with multiplication given by composition if dened and zero otherwise. Then twisted com-
plexes (see [BK90]) over the split closure of ( are the same as type 1 structures over /(().
(Note that this diers slightly from the denition in [Sei08a] in that we have no condition
about being lower-triangular.)
Alternatively, type 1 structures can be viewed as dierential comodules over the bar
resolution, thought of as a coalgebra. See [LH03, Kel].
2.4. Another model for the /

tensor product
Having introduced type 1 structures, we now introduce a pairing between /

modules
and type 1 structures. When dened, this pairing is homotopy equivalent to the derived
tensor product with the /

module associated to the type 1 structure, but is smaller; for

instance, it is nite-dimensional if the modules are. As we will see in Chapter 9, this new
tensor product also matches better with the geometry and helps give a proof of the pairing
theorem.
Throughout this section, we restrict attention to the case where /

= (. j
i

i=1
) is
operationally bounded. Note that this includes the case of dg algebras.
Denition 2.26. Let / be a right /

module over / and (`.

1
) a type 1 structure.
Suppose moreover that either / is a bounded /

module or (`.
1
) is a bounded type
1 structure. Then, we can form the k-module / ` = `
k
`, equipped with the
endomorphism

(x y) :=

k=0
(:
k+1
"
N
)(x
k
(y)).
We call (` `.

) the box tensor product of / and (`.

1
).
Graphically, the dierential on the box tensor product is given by:

:
.
Example 2.27. If / is a dg algebra and / is a dierential module, then / ` is the
ordinary tensor product /
,
A.
Example 2.28. Let / be a nite-dimensional dg algebra over a ring k = F
2
c
1
F
2
c
m
,
a direct sum of copies of F
2
. Choose a F
2
-basis c
1
. . . . . c
n
for / which is homogeneous
with respect to the action by k, in the sense that for each c
i
, there are elements c

i
and c
r
i
22 2. A STRUCTURES
with c
i
= c

i
c
i
c
r
i
. Given a type 1 structure (`.
1
) over /, we have maps 1
i
: c

i
` c
r
i
`
for i = 1. . . . : uniquely determined by the formula

1
=

i
c
i
1
i
.
If ` is a right /

module over then we can write the dierential

more explicitly as
(2.29)

(x y) =

:
k+1
(x. c
i
1
. . . . . c
i
k
) (1
i
k
1
i
1
)(y).
where the sum is taken over all sequences i
1
. . . . . i
k
of elements in 1. . . . . : (including the
empty sequence with / = 0).
Lemma 2.30. Let / be an /

module and ` a type 1 structure. Assume that either `

or ` is bounded. Then the k-module ` ` equipped with the endomorphism

is a chain
complex.
Proof. The proof is easiest to understand in the graphical notation:

:
=

:
:
=

1
:
= 0.
The rst equality uses the denition of (or that is a comodule map), the second the
structure equation for an /

module, and the third the structure equation for a type 1

structure.
Example 2.31. Fix an /

algebra / and a left type 1 structure ` over /. We can view

/ also as a right /

module over /. In this case, the chain complex / ` constructed

above coincides with the chain complex underlying the /

module associated to (`.

1
)
from Lemma 2.20 (in the case where / is a dg algebra) or 2.24 (in the general case).
Lemma 2.32. Let ) : /
1
/
2
be a homomorphism of /

modules and
1
: `
1
`
2
a homomorphism of type 1 structures. Then:
(1) If either /
1
, /
2
and ) are bounded or (`
1
.
1
N
1
) is bounded then ) induces a chain
map ) "
N
1
: /
1
`
1
/
2
`
1
.
(2) If either (`
1
.
1
N
1
) and (`
2
.
1
N
2
) are bounded or /
1
is bounded then
1
induces a
chain map "
/
1

1
: /
1
`
1
/
1
`
2
.
(3) Suppose )
t
: /
1
/
2
is another homomorphism of /

modules which is homo-

topic to ). Suppose further that either (`
1
.
1
N
1
) is bounded or /, ), )
t
and the
homotopy between them are all bounded. Then ) "
N
1
is homotopic to )
t
"
N
1
.
(4) Suppose

1
: `
1
`
2
is another homomorphism of type 1 structures which
is homotopic to
1
. Suppose further that either /
1
is bounded or
1
,

1
and the
homotopy between them are all bounded. Then "
/
1

1
and "
/
1

1
are homotopic.
2.4. ANOTHER MODEL FOR THE A TENSOR PRODUCT 23
(5) Under boundedness conditions so that all of the products are dened, given a third
map p : /
2
/
3
, then (p )) "
N
1
is homotopic to (p "
N
1
) () "
N
1
).
(6) Under boundedness conditions so that all of the products are dened, given a third
map
1
: (`
2
.
1
N
2
) (`
3
.
1
N
3
), then "
/
1
(
1

1
) is homotopic to ("
/
1

1
)
("
/
1

1
).
(7) Under boundedness conditions so that all of the products are dened, () "
N
1
)
("
/
1

1
) is homotopic to ("
/
1

1
) () "
N
1
).
Proof. Given ), we dene the induced map ) "
N
by
() "
N
)(x y) :=

k=0
()
k+1
"
N
) (x
k
(y)).
where )
k
are the components of ). Similarly, given a homotopy / with components /
k
from
to
t
, we dene a homotopy / "
N
from ) "
N
to )
t
"
N
by
(/ "
N
)(x y) =

k=0
(/
k+1
"
N
) (x
k
(y)).
Graphically, the maps ) "
N
and / "
N
are given by

)
and

/
.
Given
1
, we dene "
/

1
by
("
/

1
)(x y) =

k=1
(:
k+1
"
N
) (x
k
(y)).
with
k
constructed from the map
1
: ` `
t
as in Denition 2.23. A homotopy /
from
1
to

1
is promoted to a homotopy from "
/

1
to "
/

1
by a similar formula.
Verication of the properties of these maps is a straightforward modication of the ar-
gument that

= 0.
Example 2.33. Continuing from Example 2.31, if
1
: `
1
`
2
is a homomorphism of
type 1 structures, then "
,

1
: / `
1
/ `
2
is the chain map underlying the /

homomorphism from A
1
to A
2
from Lemmas 2.20 or 2.24.
Proposition 2.34. Let / be an /

module over an operationally bounded /

algebra /
and (`.
1
) be a type 1 structure. Suppose that either / is a bounded /

module or (`.
1
)
is bounded. In the former case, suppose moreover that (`.
1
) is homotopy equivalent to a
bounded type 1 structure. Then the product /` is homotopy equivalent to the /

tensor
product /

A.
Proof. The boundedness assumption implies that / ` is dened. Moreover, since
respects homotopy equivalences (by Lemma 2.32), we may assume that ` is bounded.
By Proposition 2.16, / is /

homotopy equivalent to /, and hence by Lemma 2.32,

24 2. A STRUCTURES
/ ` / `. But this is the same complex with the same boundary operator as
/

A. For instance, as k-modules we have

/A

= (` T
+
([1])[1] `

= ` T

([1]) ( `)

= /

A.
(Recall that the underlying space of A is `. Also, the /

homotopy equivalence
between / and / is not bounded, so we need the boundedness assumption on `.)
The following consequence of Lemma 2.32 will be used in Section 11.2:
Denition 2.35. Let (`
1
.
1
), (`
2
.
2
), and (`
3
.
3
) be three type 1 structures over some
xed dg algebra , and let
1
: `
1
`
2
and
1
: `
2
`
3
be two homomorphisms
of type 1 structures. We say these form a short exact sequence if the induced modules A
1
,
A
2
, A
3
equipped with homomorphisms associated to between them "
,

1
and "
,

1
t
into a short exact sequence of dierential graded -modules. We abbreviate this, writing
0 `
1

1
`
2

1
`
3
0.
Proposition 2.36. If 0 `
1

1
`
2

1
`
3
0 is a short exact sequence of type 1
structures over a dg algebra and / is a bounded /

module, then there is an exact sequence

in homology
H

(/`
1
) H

(/`
2
) H

(/`
3
) H

(/`
1
)[1].
Proof. The short exact sequence can be viewed as a 3-step ltered type 1 structure
`. Since `
3
is a projective -module, this short exact sequence is split. The splitting
can be viewed as a null-homotopy of the 3-step ltered complex.
Next, form /` to get a three-step ltered complex, whose associated graded complex
is

3
i=1
/`
i
. (It is perhaps worth noting that the dierential of an element in /`
1

/` can have a component in /`
3
. In fact, the /`
3
component of the dierential
from / `
1
is precisely the null-homotopy of ("
/

1
) ("
/

1
).) By Lemma 2.32,
this complex /` is homotopy equivalent to the trivial complex; in particular, /` is
acyclic. It follows that there is a long exact sequence on the homology groups of the terms
in its associated graded complex. (This latter point is a small generalization of the fact that
a short exact sequence of chain complexes induces a long exact sequence in homology: it is
a degenerate case of the Leray spectral sequence.)
Remark 2.37. Proposition 2.36 is an explicit form of the statement that / is a trian-
gulated functor. Since / ` /

A in a natural way, this follows from the fact that

the /

tensor product is a triangulated functor; over a dg algebra, this, in turn, follows

from the fact that the derived tensor product is a triangulated functor.
2.5. Gradings by non-commutative groups
In earlier sections, we considered the familiar case of Z-graded modules. Here we record
basic denitions of /

algebras (including dierential algebras) that are graded over a

possibly non-commutative group G. The denitions are straightforward generalizations of
standard denitions. As before, we assume that our ground ring k has characteristic 2 so
that we need not worry about signs.
We start with ordinary dierential graded algebras.
2.5. GRADINGS BY NON-COMMUTATIVE GROUPS 25
Denition 2.38. Let (G. ) be a pair of a group G (written multiplicatively) with a dis-
tinguished element in the center of G. A dierential algebra graded by (G. ) is, rst of
all, a dierential algebra with a grading gr of by G (as a set), i.e., a decomposition
=

gG

g
. We say an element c in
g
is homogeneous of degree p and write gr(c) = p.
For homogeneous elements c and /, we further require the grading to be
compatible with the product, i.e., gr(c /) = gr(c) gr(/), and
compatible with the dierential, i.e., gr(c) =
1
gr(c).
For an ordinary Z-grading, we would take (G. ) = (Z. 1). (We choose the dierential to
lower the grading to end up with homology rather than cohomology.) We require to be
central because otherwise the identity
(c/) = c / + c /
would not be homogeneous in general. The map /
k
is a homomorphism from Z to G,
but even if is of innite order in G, there may not be a map from G to Z taking to 1, so
we do not in general get an ordinary Z-grading.
Remark 2.39. When working over a eld whose characteristic is not 2, we need a Z,2
grading (i.e., a homomorphism from G to Z,2 taking to 1) in order to, for instance,
specify the signs in the Leibniz relation. We will not consider this further here.
There is a natural extension of this denition to /

algebras. In general, if \ and \

are G-graded k-modules, we can turn \ \ into a G-graded k-module by setting
(2.40) gr( u) := gr() gr(u)
for homogeneous elements and u of \ and \, respectively. If \ is a G-graded module
and : Z, dene \ [:] to be the graded module with
(2.41) \ [:]
g
:= \

n
g
.
Since is central, we have (\ [:]) \

= (\ \)[:]

= \ (\[:]). Now the previous
denitions of the multiplications maps :
i
and their compatibility conditions carry over as
written.
We now turn to dierential modules.
Denition 2.42. Let be a dierential algebra graded by G, as in Denition 2.38. Let o
be a set with a right G action. A right dierential -module graded by o, or, simply, a right
o-graded module, is a right dierential -module ` with a grading gr of ` by o (as a set)
so that, for homogeneous elements c and r `,
gr(rc) = gr(r) gr(c) and
gr(r) = gr(r)
1
.
If the action of G on o is transitive, o is determined by the stabilizer G
s
of any point
: o, and we can identify o with G
s
G, the space of right cosets of G
s
.
More generally, if \ is graded by o and \ is graded by G, then Equation (2.40) gives
a natural grading of \ \ by o and Equation (2.41) denes a shift functor; then the
previous denitions for /

modules, homomorphisms, and homotopies carry through. (For

homomorphisms and homotopies, the two modules should be graded by the same set.) For
example, for /

modules ` over a graded /

algebra /, if r is a homogeneous element of `

26 2. A STRUCTURES
and c
1
. . . . . c

are homogeneous elements of /, then = :

+1
(r. c
1
. . . . . c

) is homogeneous
of degree
(2.43) gr() =
1
gr(r) gr(c
1
) gr(c

).
Left modules are dened similarly, but with gradings by a set 1 with a left action by G.
Suppose the pair (G. ) is such that has innite order. Then there is a partial order
on G given by p < / if / = p
n
for some : 0. So, it sometimes makes sense to say that
gr(c) < gr(/) for homogeneous elements c. / in a group-graded algebra. Similarly, for a G-set
o, if the subgroup acts freely on o then there is a partial order on o given by : < t if
t =
n
: for some : 0. Thus, for a set-graded module `, it sometimes makes sense to
write gr(r) < gr() for homogeneous elements r. `.
Finally we turn to tensor products.
Denition 2.44. Let be a dierential algebra graded by (G. ), let ` be a right -module
graded by a right G-set o, and let ` be a left -module graded by a left G-set 1. Dene
o
G
1 by
o
G
1 := (o 1), (:. pt) (:p. t) [ : o. t 1. p G.
Dene a (o
G
1) grading on `
k
` by
(2.45) gr(::) := [gr(:) gr(:)].
By inspection, this descends to the nave tensor product
`
A
` := `
k
`, :c: :c : [ : `. : `. c .
Furthermore, there is a natural action of (or any element in the center of G) on o
G
1
via
[: t] := [: t] = [: t].
(This denition fails to descend to the quotient for non-central elements in place of .) Again
by inspection, the boundary operator
(:
A
:) := (:)
A
: + :
A
(:)
acts by
1
on the gradings. Thus `
A
` is a chain complex with a grading by o
G
1,
which is a set with a Z-action.
Using Equation (2.45) and the same shift functor as before, we can extend the denitions
of both versions of the tensor product, chain maps and homotopies to this setting as well.
We now see a genuinely new feature of dg algebras where the grading is non-commutative:
even if has innite order in G, o, and 1, it may not have innite order on o
G
1. We
illustrate this with a simple, and pertinent, example.
Example 2.46. Let H be a variant of the Heisenberg group:
H = . . [ =
2
. is central .
(The usual Heisenberg group over the integers has instead of
2
in the rst relation; with
the exponent of 2, the group supports a Z,2 grading.) Let 1 be the subgroup of H generated
by . Then has innite order on H and on H,1 and 1H, the spaces of left and right
cosets of 1. However, the reduced product is
(1H)
H
(H,1) 1H,1
= 1
k

l
1 [ /. | Z.
2.5. GRADINGS BY NON-COMMUTATIVE GROUPS 27
These double cosets are not all distinct. In particular,
1
k

l
1 = 1
k

l
1
= 1
k

l+2k
1
= 1
k

l+2k
1.
Equalities like this generate all equalities between the double cosets. Thus the order of on
1
k

l
1 is 2/, which is not free when / ,= 0.
This example is exactly what we will see when we do 0-surgery on a knot in o
3
. (See
Section 10.6.)
CHAPTER 3
The algebra associated to a pointed matched circle
The aim of the present chapter is to dene the algebra /(:) associated to the boundary
of a three-manifold, parametrized by the pointed matched circle :. In Section 3.1, we start
by dening the strands algebra /(:), which is a simplied analogue of this algebra. The
desired algebra /(:), introduced in Section 3.2, is described as a subalgebra of /(:). In
Section 3.3, we explain the grading on /(:).
Although the algebras are dened combinatorially, the motivation for their denitions
comes from the behavior of holomorphic curves. The reader might nd it helpful to consult
Examples 6.136.15 and 7.137.16, as well as [LOT09], in conjunction with this chapter.
Our algebras, by construction, have characteristic 2. With a little more work, the algebras
could be lifted to give a construction over Z; for the algebra /(:) this is done in [Kho10].
Indeed, with considerably more work, one should be able to dene bordered Floer homology
over Z.
3.1. The strands algebra /(:. /)
We will describe an algebra /(:. /) in three dierent ways, which will be useful for
dierent purposes.
3.1.1. Algebraic denition. For non-negative integers : and / with : /, let /(:. /)
be the F
2
vector space generated by partial permutations (o. 1. ), where o and 1 are /-
element subsets of the set 1. . . . . : (hereafter denoted [:]) and is a bijection from o to 1,
with i (i) for all i o. For brevity, we will adopt a 2-line notation for these generators,
so that, for instance,

1 2 4
5 3 4
_
denotes the algebra element (1. 2. 4. 3. 4. 5. 1 5. 2
3. 4 4) inside /(3. 5).
For c = (o. 1. ) a generator of /(:. /), let inv(c) or inv() be the number of inversions
of : the number of pairs i. , o with i < , and (,) < (i). We make /(:. /) into
an algebra using composition. For c. / /(:. /), with c = (o. 1. ), / = (1. l. ), and
inv( ) = inv() + inv(), dene c / to be (o. l. ). In all other cases, (i.e., if the
range of is not the domain of , or if inv( ) ,= inv() + inv()), the product of two
generators is zero.
There is one idempotent 1(o) in /(:. /) for each /-element subset o of [:]; it is (o. o. "
S
).
It is easy to see that the 1(o) are all the minimal idempotents in /(:. /). (Minimal idem-
potents are the idempotents that can not be further decomposed as sums of orthogonal
idempotents). Let J(:. /) /(:. /) be the subalgebra generated by the idempotents.
We also dene a (co)dierential on /(:. /). For c = (o. 1. ) in /(:. /), let Inv() be
the set counted in inv(): the set (i. ,) [ i < ,. (,) < (i) . For Inv(), let

be
29
30 3. THE ALGEBRA ASSOCIATED TO A POINTED MATCHED CIRCLE
dened by

(/) =
_

_
(,) / = i
(i) / = ,
(/) otherwise.
Note that inv(

) < inv(), since the inversion (i. ,) was removed from Inv() and none are
added. It is also possible that inv(

) < inv() 1, since more inversions might be removed.

Dene
c :=

Inv()
inv()=inv()1
(o. 1.

).
Lemma 3.1. The set /(:. /), equipped with a product induced by composition of partial
permutations and with the above-dened endomorphism , is a dierential algebra.
We defer the proof until Section 3.1.2, in which we develop a convenient geometrical
language for /(:. /).
Denition 3.2. The dierential algebra /(:. /) is called the strands algebra with / strands
and : places. The direct sum /(:) =

k
/(:. /) is called the strands algebra with : places.
Remark 3.3. The function inv provides a Z-grading on /(:).
Remark 3.4. Setting the product in /(:. /) to be zero if the number of inversions in the
composition decreases is reminiscent of the nilCoxeter algebra [FS94, Kho01].
3.1.2. Geometric denition. There is an alternate description of the algebra /(:),
in terms of strands diagrams. We represent a triple (o. 1. ) by a diagram with : dots on
both the left and right (numbered from the bottom up), with a set of strands representing
connecting the subsets o on the left and 1 on the right. The condition that i (i)
means that strands stay horizontal or move up when read from left to right. For instance,
the diagram
1 1
2 2
3 3
4 4
5 5
represents the algebra element

1 2 4
5 3 4
_
.
It is easy to see that inv(c) is the minimal number of crossings in any strand diagram.
Accordingly, we may view basis elements of /(:) as strand diagrams as above, with no pair of
strands crossing more than once, considered up to homotopy. The summand /(:. /) /(:)
corresponds to strand diagrams with exactly / strands.
We can dene the product on /(:. /) in these pictures using horizontal juxtaposition.
The product c / of two algebra elements is 0 if the right side of c does not match the left
side of /; otherwise, we place the two diagrams next to each other and join them together.
If in this juxtaposition two strands cross each other twice, we set the product to be 0:
(3.5) = 0.
3.1. THE STRANDS ALGEBRA A(N, K) 31
This corresponds to the condition that, with c = (1. l. ) and / = (o. 1. ), we set c / = 0
when inv( ) ,= inv() + inv(). If there are no double crossings then the juxtaposition
is the product. For instance, the algebra element above can be factored as
1 1
2 2
3 3
4 4
5 5

1 1
2 2
3 3
4 4
5 5
=
1 1
2 2
3 3
4 4
5 5
.
The boundary operator c also has a natural graphical description. For any crossing in
a strand diagram, we can consider smoothing it:
.
The boundary of a strand diagram is the sum over all ways of smoothing one crossing of
the diagram, where we remove those diagrams from the sum for which there are two strands
which cross each other twice (compare Equation (3.5)). For instance, we have
(3.6)
_
_
_
_
1 1
2 2
3 3
4 4
5 5
_
_
_
_
=
1 1
2 2
3 3
4 4
5 5
+

7
0
1 1
2 2
3 3
4 4
5 5
+
1 1
2 2
3 3
4 4
5 5
.
It is straightforward to check that this agrees with the earlier description.
Proof of Lemma 3.1. Consider the larger algebra /(:. /) generated over F
2
by strands
diagrams but where we do not set double crossings to 0. (We also do not allow isotop-
ing away double crossings.) It is obvious that /(:. /) is a dierential algebra, i.e., that
the product is associative and the dierential satises the Leibniz rule and
2
= 0. Let
/
d
(:. /) /(:. /) be the sub-vector space of /(:. /) generated by all strands diagrams
with at least one double crossing. We claim that /
d
(:. /) is a dierential ideal in /(:. /);
since /(:. /)

= /(:. /),/
d
(:. /), the result follows. Clearly, /(:. /) is an ideal; and if
: /
d
(:. /) is a strand diagram with at least two double crossings then (:) /
d
(:. /). If
: is a strand diagram with a single double crossing, there are two terms in (:) which do
not lie in /
d
(:. /), but these terms cancel.
Remark 3.7. In combinatorial terms, the terms in the boundary of an algebra element c
correspond to the edges in the Hasse diagram of the Bruhat order on the symmetric group
that start from the permutation corresponding to c. More precisely, the Bruhat order on
the symmetric group may be dened by setting u u
t
if, when we write u = :
1
. . . :
r
as a reduced product of adjacent transpositions, we can delete some of the :
i
to obtain a
product representing u
t
. This turns out not to depend on the reduced expression for u
that we picked. (See, e.g., [BB05, Section 2.2].) If we represent a reduced expression for u
by drawing strands crossing each other at each adjacent transposition, then deleting an :
i
corresponds to smoothing a crossing. The Hasse diagram of a partially ordered set is the
graph connecting two elements u. u
t
if one covers the other, i.e., if u u
t
and there is no
other element u
tt
with u u
tt
u
t
. In the case of Bruhat order on the symmetric group,
this is equivalent to u u
t
and inv(u) = inv(u
t
) + 1 (see [BB05, Theorem 2.2.6], inter
alia), just as in /(:. /).
32 3. THE ALGEBRA ASSOCIATED TO A POINTED MATCHED CIRCLE
3.1.3. Reeb chords. There is one more description of /(:. /), relating it to the Reeb
chords we will see from the moduli spaces. Fix an oriented circle 2 with a set of : distinct
points a = c
1
. . . . . c
n
2, together with a basepoint . 2 a. We view 2 as a contact 1-
manifold, and a as a Legendrian submanifold of 2. A Reeb chord in (2.. a) is an immersed
(and consequently embedded) arc in 2 . with endpoints in a, with orientation induced by
the orientation on 2. The initial endpoint of is denoted

and the nal endpoint

+
.
A set =
1
. . . . .
j
of , Reeb chords is said to be consistent if

:=

1
. . . . .

j

and
+
:=
+
1
. . . . .
+
j
are both sets with , elements (i.e., no two
i
share initial or nal
endpoints). We adopt a two-line notation for consistent sets of Reeb chords so that, for
instance,
_
1 2
5 3
_
is the set of chords [1. 5]. [2. 3].
Denition 3.8. Let be a consistent set of Reeb chords. The strands algebra element
associated to , denoted c
0
() /(:), is dened by
c
0
() :=

S[S(

+
)=
(o

. o
+
.
S
)
where
S
[
S
= " and
S
(

i
) =
+
i
. Geometrically, this corresponds to taking the sum over
all ways of consistently adding horizontal strands to the set of strands connecting each

i
to
+
i
. For convenience, dene c
0
() to be 0 if is not consistent.
The strands algebra /(:) is generated as an algebra by the elements c
0
() and idem-
potents 1(o), with o a. In fact, the basis for /(:) over F
2
that we constructed in
Section 3.1.1 are the non-zero elements of the form 1(o)c
0
(), where o a is a /-element
subset, and is a consistent set of Reeb chords.
We give a description of the product on /(:. /) in terms of this basis, after introducing
two denitions.
Denition 3.9. A pair of Reeb chords (. ) is said to be interleaved if

<

<
+
<
+
.
They are said to be nested if

<

<
+
<
+
. An unordered pair of Reeb chords is said
to be interleaved (respectively nested) if either possible ordering is interleaved (respectively
nested).
Denition 3.10. If and are Reeb chords such that
+
=

, we say that and abut

and dene their join to be the Reeb chord [

.
+
] obtained by concatenating the two
chords.
Note that abutting is not a symmetric relation.
Denition 3.11. For two consistent sets of Reeb chords , , dene to be obtained
from the union by replacing every abutting pair
i
and
j
by their join
i

j
.
If
i
, let

++
i
:=
_

+
j

i
,
j
abut for some
j

+
i
otherwise.
Similarly dene

j
for
j
.
For and consistent sets of Reeb chords, we say that and are composable if
is consistent and has no double crossings; precisely, if is consistent and there is no pair

i
and
j
so that

i
<

j
,

j
<
+
i
, and
++
i
<
+
j
. Finally, we iterate this notion
to sequences of consistent sets of Reeb chords = (
1
. . . . .
n
): we say such a sequence is
composable if for all i = 1. . . . . : 1, the sets

i
j=1

j
and
i+1
are composable.
3.1. THE STRANDS ALGEBRA A(N, K) 33
Figure 3.1. Splittings and shues of Reeb chords. (a) A splitting. (b)
A weak splitting which is not a splitting. In both cases, the position at which
the splitting occurs is marked with a dotted line. (c) A shue. (d) A weak
shue which is not a shue. In all cases, the chords are shown on the top
line, and the associated strands diagrams on the bottom line.
Lemma 3.12. Given consistent sets of Reeb chords and and a subset o [4/] with
1(o)c
0
() ,= 0, we have
1(o)c
0
()c
0
() =
_
1(o)c
0
( ) and composable
0 otherwise.
Proof. This follows directly from the denitions.
We now describe the dierential on /(:) in terms of Reeb chords. Observe that there
are two ways a term in the boundary operator can appear, depending on whether we smooth
a crossing between two moving strands or between a moving strand and a horizontal strand.
Denition 3.13. For a Reeb chord
1
, a splitting of
1
is a pair of abutting chords (
2
.
3
)
so that
1
=
2

3
. For a set of Reeb chords, a weak splitting of is a set obtained
from by replacing a chord
1
with two chords
2
.
3
, where (
2
.
3
) is a splitting of

1
. A splitting of is a weak splitting of , replacing
1
with
2
.
3
, which is consistent
and does not introduce double crossings, i.e., there is no chord
4
nested in
1
and with

4
<
+
2
=

3
<
+
4
. See Figure 3.1.
For a set of Reeb chords , a weak shue of is a set obtained by replacing a nested pair
(
1
.
2
) of chords in with the corresponding interleaved pair ([

1
.
+
2
]. [

2
.
+
1
]). A shue
of is a weak shue with nested pair (
1
.
2
) that does not introduce double crossings, i.e.,
there is no chord
3
with
2
nested in
3
and
3
nested in
1
. Again, see Figure 3.1.
Lemma 3.14. We have
c
0
() =

a splitting
of
c
0
(
t
) +

a shue
of
c
0
(
t
).
Proof. This follows directly from the denitions.
See also Lemma 5.71 for alternate characterizations of composability, splittings, and
shues in terms of a grading on the algebra.
34 3. THE ALGEBRA ASSOCIATED TO A POINTED MATCHED CIRCLE
Remark 3.15. In the type module, these two types of terms in the boundary, splittings
and shues, appear respectively as join curve ends and odd shue curve ends, as dened in
Denition 5.52; see the proof of Proposition 7.12.
3.2. Matched circles and their algebras
We now turn to the algebra associated to a parametrized surface. We represent our
boundary surfaces by matched circles.
Denition 3.16. A matched circle is a triple (2. a. `) of an oriented circle 2, 4/ points
a = c
1
. . . . . c
4k
in 2 (as in Section 3.1.3), and a matching: a 2-to-1 function `: a [2/].
We require that performing surgery along the 2/ pairs of points in 2 (viewed as 0-spheres)
yields a single circle, rather than several disjoint circles.
A pointed matched circle : is a matched circle together with a basepoint . 2 a.
Denition 3.17. To each matched circle : we associate an oriented surface 1(:), or just
1, of genus /, by taking a disk with boundary 2 (respecting the orientation), attaching
oriented 2-dimensional 1-handles along the pairs specied by `, and lling the resulting
boundary circle with another disk. Alternately, 1 may be obtained from a polygon with 4/
sides, with the points a in the middle of the sides, by gluing pairs of sides specied by `
in an orientation-preserving way. The last disk glued in is a neighborhood of the unique
resulting vertex. We call 1(:) the surface associated to :.
If : denotes the orientation reverse of : (i.e., the same data except with the orientation
on 2 reversed) then 1(:) is the orientation reverse 1(:) of 1(:), canonically.
In the other direction from Denition 3.17, we have the following:
Denition 3.18. Given a closed, orientable surface 1 of genus /, let ) : 1 1 be a Morse
function with a unique index 0 critical point and a unique index 2 critical point, and such
that )(j
i
) = 1 for all index 1 critical points j
1
. . . . . j
2k
. Suppose that )
1
(3,2) is non-empty.
Fix also a Riemannian metric p on 1. Let 2 = )
1
(3,2), and let a 2 denote the ascending
spheres of the index 1 critical points. Dene `(c
j
) = i if c
j
is in the ascending sphere of
j
i
. Choose also a point . 2 a. Then : = (2. a. `. .) is a pointed matched circle, and
1(:)

= 1. We say that (). p) is compatible with the pointed matched circle :.
The points a in a matched circle (2. a. `) inherit a cyclic ordering from the orientation
of 2; in a pointed matched circle, this cyclic ordering is lifted to an honest ordering, which
we denote by <.
Example 3.19. Let 2 be a circle with four points, which we number 0. 1. 2. 3 in their cyclic
ordering, and let `: 1. . . . . 4 1. 2 be characterized by `(i) i (mod 2). Then 2 is
a matched circle which represents the torus. Indeed, it is the unique matched circle which
represents the torus.
Example 3.20. Let 2 be a circle with eight marked points, which we number 0. . . . . 7 in
their cyclic ordering. Let `: 1. . . . . 8 1. . . . . 4 be characterized by `(i) i (mod 4).
Then 2 is a matched circle which represents a surface of genus 2, called the antipodal pointed
matched circle. A dierent matched circle representing the same surface can be obtained
by forming the connected sum (in the obvious way) of two pointed matched circles for the
torus; this is called the split pointed matched circle. See Figure 3.2.
3.2. MATCHED CIRCLES AND THEIR ALGEBRAS 35
Figure 3.2. Two pointed matched circles for the genus 2 surface.
Left: the split pointed matched circle. Right: the antipodal pointed matched
circle.
Figure 3.3. Orientation of a pointed matched circle. Left: the orienta-
tion of a Heegaard surface, and the induced orientation of its boundary. Right:
the induced ordering of a: in this gure, j
1
<j
2
<j
3
<j
4
.
Denition 3.21. Fix a pointed matched circle :. For each subset s of [2/] we have one
idempotent 1(s) in the strands algebra /(4/), given by
(3.22) 1(s) :=

S[2k]
S a section of M over s
1(o).
where by a section of ` over s we mean a subset of o that maps bijectively to s via `.
That is, each element of s denes a pair of points in a; 1(s) is the sum over the 2
[s[
minimal
idempotents in /(4/. [s[) that pick one point out of each of these pairs. The ring of idempo-
tents associated to the pointed matched circle is the algebra J(:) generated by the elements
1(s) associated to all the subsets s [2/]. Dene
I :=

s[2k]
1(s)
to be the unit in J(:).
We will now dene /(:), our main algebra of interest.
Denition 3.23. For a given pointed matched circle :, the algebra associated to :, denoted
/(:), is the subalgebra of

2k
i=0
/(4/. i) generated (as an algebra) by J(:) and by Ic
0
()I
for every set of Reeb chords . The algebra element associated to , denoted c(), is the
projection I c
0
() I of c
0
() to /(:).
36 3. THE ALGEBRA ASSOCIATED TO A POINTED MATCHED CIRCLE
Set
/(:. i) = /(:) /(4/. / + i)
J(:. i) = J(:) /(4/. / + i).
and call /(:. i) the part of /(:) with weight i.
It is easy to verify that /(:) is closed under multiplication and .
In Chapter 2, we considered algebras (and modules) dened over a ground ring k. From
now on, we will view /(:) as dened over the ground ring k = J(:).
The algebra /(:) decomposes (as a direct sum of dierential graded algebras) according
to the weights of the algebra elements,
/(:) =
k

i=k
/(:. i).
There is a convenient basis for /(:) over F
2
, consisting of the non-zero elements of the
form 1(s)c(), where s [2/] and is a consistent set of Reeb chords. In two-line notation,
we express these basic generators 1(s)c() as
_
x
1
x
k
z
1
z
l
y
1
y
k

:= 1(`(r
1
). . . . . `(r
k
). `(.
1
). . . . . `(.
l
)c
__
x
1
x
k
y
1
y
k
__
.
Graphically, we denote the matching ` by drawing dotted arcs on the left and right of a
strand diagram. We draw the strands in an element /(:) as we did for elements of /(:) in
Section 3.1.2, except that we draw horizontal strands (which, if we expand out the sum for
1(s), appear in half the terms) with dashed lines. For instance, if : is the split genus-two
pointed matched circle, one basic generator is
(3.24) = +
which, in two-line notation, can be denoted
_
1 2
6

or
_
1 4
6

. In particular, note that in a

non-zero basic element of /(:), no initial (respectively terminal) ends of strands can be
matched with each other.
Example 3.25. Let : be the unique pointed matched circle representing the torus. The
algebra /(:. 0) is described in detail in Section 11.1. The algebra /(:. 1) is isomorphic to
F
2
, with generator (. . ) where is the unique bijection from to . The algebra /(:. 1)
has F
2
-basis
_
1 2

.
_
1 2
3

.
_
1 2
2 3

.
_
2 1
4

.
_
2 3
3 4

.
_
1 2
4 3

.
_
1 2
3 4

.
The non-trivial dierentials are given by

_
1 2
3

=
_
1 2
2 3

.
_
2 1
4

=
_
2 3
3 4

.
_
1 2
4 3

=
_
1 2
3 4

.
In particular, the homology H

(/(:. 1)) is F
2
, generated by the idempotent. Therefore, the
algebra /(:. 1) is formal (quasi-isomorphic to its homology).
It is not a coincidence that H

(/(:. 1))

= H

(/(:. 1)). In general, for a pointed

matched circle : for a surface of genus /, /(:. /) is 1-dimensional while /(:. /) is for-
mal and has 1-dimensional homology. (The rst fact is trivial; the second takes a little
3.2. MATCHED CIRCLES AND THEIR ALGEBRAS 37
work.) More generally, there is a duality relating /(:. i) and /(:
t
. i) for appropriate
pairs of pointed matched circles : and :
t
; see [LOT10e]. The fact that the algebra is
one-dimensional in the extremal weight relates to the fact that

HFK is monic for bered
knots; see [LOT10a, Theorem 7] for further discussion.
Finally, we give interpretations of /(:) directly in terms of Reeb chords. For any pointed
matched circle : and set of Reeb chords , if c
0
() ,= 0, then c() ,= 0 if and only if
[`(

)[ = [`(
+
)[ = [[, i.e., if and only if no two
i
have matched initial or nal endpoints.
Lemma 3.26. Let and be a pair of sets of Reeb chords so that c
0
() c
0
() ,= 0. Then
c()c() = 0 if and only if there is a matched pair j. in : satisfying one of the following
conditions:
(1) has a chord starting at j and another one starting at ;
(2) has a chord terminating at j and another one terminating at ; or
(3) has a chord which terminates in j while has a chord which starts in .
Further, if none of Conditions (1)(3) are satised then for any s such that 1(s) c() ,= 0
and 1(s) c( ) ,= 0 it follows that 1(s) c() c() ,= 0.
Proof. Any of the three conditions implies the vanishing of c() c(): in view of
Lemma 3.12, the rst is the case where Ic
0
()c
0
() = 0, the second is where c
0
()c
0
()I =
0, and the third is where c
0
() I c
0
() = 0.
Conversely, suppose none of Conditions (1)(3) hold. Let o := ( )

and 1 :=
( )
+
. The failure of Condition (1) (respectively (2)) implies that the restriction of
` to o (respectively 1) is injective, so 1(o) (respectively 1(1)) is a summand in I. Let
l := (o

)
+
; observe that 1(o)c
0
() = 1(o)c
0
()1(l). It follows from Lemma 3.12
(and the denitions) that l = (1
+
)

. By the failure of Condition (3), the restriction

of ` to l is injective, so l is a summand of I. Since 1(o) c
0
() 1(l) c
0
() 1(1) ,= 0,
it follows that c() c() ,= 0.
For the last part of the statement, since 1(s) c() ,= 0 there is a subset o [4/] such
that `(o) = s, `[
S
is injective and ()

o. Moreover, by the failure of Condition (3),

l := (o

)
+
contains

, and since 1(s) c() ,= 0, `[

U
is injective. Similarly,
1 := (l

)
+
=
_
o ( )

_
( )
+
has `[
T
injective since 1(s) c( ) ,= 0.
So, setting t := `(1) and u := `(l), we have
1(s) c
0
() 1(u) c
0
() 1(t) ,= 0.
as desired.
Remark 3.27. One might wonder to what extent /(:) is a topological invariant of the
surface specied by :. In fact, if : and :
t
are two dierent pointed matched circles repre-
senting dieomorphic surfaces, then neither /(:) and /(:
t
) nor their homologies H

(/(:)
and H

(/(:
t
)) are necessarily isomorphic; indeed, they need not have the same rank. (See
[LOT10a, Theorem 9].) However, it turns out that the module categories of /(:) and
/(:
t
) are derived equivalent [LOT10a, Theorem 1].
Remark 3.28. The summands /(:. i) /(:) for i ,= 0 act trivially on the modules

CFD(1 ) and

CFA(1 ) dened in Chapters 6 and 7. So, in the present work, only the
subalgebra /(:. 0) /(:) will be of interest. The other summands of /(:) do play an
important role in the case of three-manifolds with disconnected boundary; see [LOT10a],
and also Appendix A.
38 3. THE ALGEBRA ASSOCIATED TO A POINTED MATCHED CIRCLE
Remark 3.29. One can think of the index i in /(:. i) as corresponding to spin
c
-structures
on the surface 1(:), compare [LOT10a]. For a 3-manifold 1 with connected boundary 1 ,
only the middle spin
c
-structure on 1 extends over 1 .
Remark 3.30. We get pointed matched circles by splitting a pointed Heegaard diagram in
two along a circle 2, satisfying a suitable combinatorial hypothesis; see Lemma 4.31. The
points a are the intersections of 2 with the -circles and the matching ` records which
intersection points are on the same -circle. We declare one of the two components of the
complement of 2 in the Heegaard surface to be the left (west, or type ) side, and let 2
inherit the orientation as the boundary of this side. See Figure 3.3.
3.3. Gradings
Although the algebra /(:. /) has a Z-grading, for instance by the crossing number (i.e.,
number of inversions), it turns out that the subalgebra /(:) has no Z-grading.
Example 3.31. For : the pointed matched circle representing a torus, consider the two
algebra elements in /(:. 1)
r =
_
1 2
3

= and =
_
2 1
4

= .
On one hand, we have
r =
but on the other hand,
r =
(r) = .
Hence ((r) ) = r. which is inconsistent with any Z-grading in which r and are
homogeneous.
Instead, we grade /(:) by a non-commutative group. A grading of a dierential alge-
bra by a non-commutative group G is a straightforward, but not very standard, notion.
Briey, it is a grading of as an algebra by G together with a distinguished central element
G so that for any homogeneous elements c. / , we have gr(c /) = gr(c) gr(/) and
gr(c) =
1
gr(c). See Section 2.5 for the details of the algebraic setting.
In fact, we give two gradings of /(:) by non-commutative groups. In Section 3.3.1
we construct a canonical grading of /(:) by a group G
t
(:). In Section 3.3.2 we rene
this grading to a grading by a subgroup G(:) G
t
(:). The subgroup G(:) has a more
obvious topological interpretation and behaves better with respect to the operation of gluing
Heegaard diagrams (see Section 10.6), but the denition of the rened grading on /(:)
depends on some additional choices.
3.3. GRADINGS 39
3.3.1. Unrened grading on /(:). We start by giving a grading gr
t
of /(:. /) with
values in a non-commutative group G
t
(:). We then show that this grading has the property
that gr
t
is unchanged by adding horizontal strands, and so descends to a grading on /(:).
Let 2
t
= 2 .. The group G
t
(:) can be abstractly dened as a central extension by Z
of the relative homology group H
1
(2
t
. a) (which is isomorphic to Z
n1
). In particular, there
is an exact sequence of groups
Z

G
t
(:)
[]
H
1
(2
t
. a)
where in the rst map 1 maps to .
To dene the central extension concretely we make some more denitions. For j a and
H
1
(2
t
. a), dene the multiplicity :(. j) of j in to be the average multiplicity with
which covers the regions on either side of j. Extend : to a map H
1
(2
t
. a) H
0
(a)
1
2
Z
bilinearly.
For
1
.
2
H
1
(2
t
. a), dene
(3.32) 1(
1
.
2
) := :(
2
.
1
)
where is the connecting homomorphism : H
1
(2
t
. a) H
0
(a) from the long exact sequence
for a pair. The map 1 can be thought of as the linking number of
1
and
2
. Also, for
H
1
(2
t
. a), dene c() (
1
2
Z),Z by
c() :=
1
4
#(parity changes in ) (mod 1).
Here, a parity change in is a point j a so that the multiplicity of to the left of j has
dierent parity than the multiplicity of to the right of j, or, equivalently, so that :(. j)
is a half-integer. (Note that there are always an even number of such points.)
Denition 3.33. Dene G
t
(:) to be the group generated by pairs (,. ), where ,
1
2
Z,
H
1
(2
t
. a) and , c() mod 1, with multiplication given by
(3.34) (,
1
.
1
) (,
2
.
2
) := (,
1
+ ,
2
+ 1(
1
.
2
).
1
+
2
).
(Thus the map 1 is a 2-cocycle on H
1
(2
t
. a) which denes the central extension G
t
(:).) In
an element p = (,. ) of G
t
(:), we refer to , as the Maslov component of p and as the
spin
c
component of p. The distinguished central element of G
t
(:) is = (1. 0).
To see that G
t
(:) is, in fact, a group we make 1 more explicit and prove some elementary
properties.
Lemma 3.35. If =

n1
i=1
n
i
[j
i
. j
i+1
] and =

n1
i=1

i
[j
i
. j
i+1
], then
1(. ) =
n2

i=1
n
i

i+1
n
i+1

i
2
.
In particular, 1(. ) = 1(. ).
Note that n
i

i+1
n
i+1

i
can be interpreted as the determinant of a 2 2 submatrix of
the matrix
_
u
1
u
2
u
n1
v
1
v
2
v
n1
_
.
40 3. THE ALGEBRA ASSOCIATED TO A POINTED MATCHED CIRCLE
Proof. Recall that [j
i
. j
i+1
] = j
i+1
j
i
, from which it follows that =

n
i=1
(n
i1

n
i
)j
i
, where we set n
0
= n
n
= 0. Then
1(. ) =
n

i=1
(n
i1
n
i
)(
i1
+
i
)
2
=
n

i=1
n
i1

i
n
i

i1
2
+
n

i=1
_
n
i1

i1
2

n
i

i
2
_
=
n

i=1
n
i1

i
n
i

i1
2
since the second sum in the second line telescopes. The rst and last terms in the nal sum
vanish, proving the result.
Lemma 3.36. If . H
1
(2
t
. a), then
c( + ) c() + c() + 1(. ) (mod 1).
Proof. Consider the terms in 1(. ) = :(. ) corresponding to the dierent points
j a. A particular j contributes a half-integer to 1(. ) when j appears an odd number
of times in and :(. j) is a half-integer, i.e., j appears an odd number of times in .
This case (when j is odd in both and ) is exactly when the contribution of j to
c( + ) c() c() is 1,2 (mod 1).
Proposition 3.37. G
t
(:) is a group.
Proof. Lemma 3.36 guarantees that if (,
1
.
1
) and (,
2
.
2
) are in G
t
(:), then the product
is as well:
,
1
+ ,
2
+ 1(
1
.
2
) c(
1
) + c(
2
) + 1(
1
.
2
) (mod 1)
c(
1
+
2
) (mod 1).
The inverse of (,. ) is easily seen to be (,. ). We must also check that the product is
associative:
((,
1
.
1
) (,
2
.
2
)) (,
3
.
3
)
= (,
1
+ ,
2
+ ,
3
+ 1(
1
.
2
) + 1(
1
+
2
.
3
).
1
+
2
+
3
)
= (,
1
+ ,
2
+ ,
3
+ 1(
1
.
2
) + 1(
1
.
3
) + 1(
2
.
3
).
1
+
2
+
3
)
where we use bilinearity of 1. Associating in the opposite order gives the same result.
Denition 3.38. The grading gr
t
(c) of an element c /(:. /) with starting idempotent o
and ending idempotent 1 is dened as follows. The spin
c
(H
1
(2
t
. a)) component of the
grading of an element c = (o. 1. ), denoted [c], is dened to be the element of the relative
homology group H
1
(2
t
. a) given by summing up the intervals corresponding to the strands:
[c] :=

sS
[:. (:)].
The Maslov component of the grading is given by
(c) := inv(c) :([c]. o).
3.3. GRADINGS 41
So, the grading of an element c is given by
gr
t
(c) := ((c). [c]).
Note that [c] = 1 o H
0
(a), so by Lemma 3.35, :([c]. o) = :([c]. 1).
Proposition 3.39. The function gr
t
denes a grading (in the sense of Denition 2.38) on
/(:. /), with = (1. 0).
Proof. Let c be an algebra element with starting idempotent o and ending idempo-
tent 1. We rst check that gr
t
(c) G
t
(:). Modulo 1, we have
(c) :([c]. o)

1
2
#(o 1)
1
2
#(1 o)

1
4
#(o 1) c([c]).
where o 1 is the symmetric dierence of o and 1:
o 1 := (o 1) (1 o).
For the behavior of gr
t
under the action of , observe that (c) = (c) 1: neither o
nor [c] is changed by , so :([c]. o) is unchanged, and in the denition of only the inv(c)
term changes. Thus with = (1. 0), we have gr
t
(c) =
1
gr
t
(c), as desired.
We also check that gr
t
is compatible with multiplication: if c is given by (o. 1.
1
) and /
is given by (1. l.
2
) with c / ,= 0, then
[c /] = [c] + [/]
(c /) = inv(c /) :([c] + [/]. o)
= inv(c) + inv(/) :([c]. o) :([/]. 1) + :([/]. 1 o)
= (c) + (/) + :([/]. [c])
and so gr
t
(c /) = gr
t
(c) gr
t
(/).
Proposition 3.40. For a consistent set of Reeb chords, c
0
() is homogeneous with respect
to gr
t
. In particular, the grading gr
t
descends to a grading on the subalgebra /(:).
Proof. Observe that adding a horizontal strand at j a to an algebra element c =
(o. 1. ) does not change [c] and changes inv(c) and :([c]. o) by the same amount, :([c]. j).
Thus gr
t
(c) is unchanged by adding a horizontal strand (moving from /(:. /) to /(:. /+1)).
In particular the terms in the denition of c
0
() all have the same grading, proving the
rst statement. Since /(:) is generated by elements of the form 1(s)c
0
(), and 1(s) is
homogeneous with grading (0. 0), the second statement follows.
As in Section 2.5, there is a partial order on G
t
(:) by (,. ) < (,
t
.
t
) if =
t
and , < ,
t
.
3.3.2. Rened grading on /(:. i). We can also construct a rened grading gr on
/(:. i) with values in a smaller group G(:).
The group G(:) can be abstractly dened as a central extension of H
1
(1)

= Z
2k
. Again,
there is a sequence of groups
(3.41) Z

G(:)
[]
H
1
(1)
42 3. THE ALGEBRA ASSOCIATED TO A POINTED MATCHED CIRCLE
where 1 Z maps to G(:). This central extension is dened as an abstract group as
follows. Let denote the generator of the center of G(:) and for p G(:), let [p] be its
image in H
1
(1), as indicated above. Then, for p. / G(:), we have
(3.42) p/ = /p
2([g][h])
.
Equation (3.42) determines G(:) uniquely up to isomorphism.
In order to describe gr, rst suppose that we have an element of H
1
(2
t
. a) for which
`

() = 0; that is, for each pair j. of points in a identied by `, the sum of the
multiplicities of at j and at is 0. Then we can construct an element of H
1
(1(:))
from by embedding 2 in 1(:) to make a chain in 1(:), and then adding multiples of
the cores of the 1-handles we attached in the construction of 1(:) to form a cycle . Indeed,
we can identify H
1
(1) as ker(`

) inside H
1
(2
t
. a).
With this setup, we dene G(:) to be the subgroup (,. ) G
t
(4/) [ `

(()) = 0 of
G
t
(4/).
Lemma 3.43. G(:) is a group that ts in the exact sequence (3.41) and satises equa-
tion (3.42).
Proof. The fact that [] : G
t
(4/) H
1
(2
t
. a) is a homomorphism guarantees that G(:)
is a group. The image of G(:) under [] is, by denition, ker(`

) H
1
(1) and the kernel
of [] is (,. 0) [ , Z, so we have the desired exact sequence. Finally, for . ker(`

),
the corresponding elements . in H
1
(1) satisfy
= :(. ).
which implies Equation (3.42).
To dene the grading gr on /(:. i), pick an arbitrary base idempotent 1(s
0
) /(:. i)
and, for each other idempotent 1(s), pick a group element (s) = (1(s)) G
t
(4/) so that
`

([(s)]) = s s
0
. Then dene
(3.44) gr
_
1(s)c()1(t)
_
:= (s) gr
t
(c())(t)
1
.
We will call the choices s
0
, (s) the choice of grading renement data.
Lemma 3.45. The function gr is a grading with values in G(:) G
t
(4/).
Proof. To check the values are in G(:), we compute:
`

[gr(1(s)c()1(t))] = `

[(s)] + `

[c()] `

[(t)]
= (s s
0
) + (t s) (t s
0
)
= 0.
It is straightforward to check that gr is a grading.
Remark 3.46. Dierent choices of grading renement data give gradings on /(:. i) that are
conjugate in an idempotent-dependent way, in the following sense. For any two choices (s
0
. )
and (s
t
0
.
t
) giving gradings gr
1
and gr
2
respectively, there is a function : J(:. i) G(:)
so that, if c /(:) is homogeneous and satises c = 1(s)c1(t),
gr
2
(c) = (s) gr
1
(c) (t)
1
.
(The map is given by (s) =
t
(s)(s)
1
.) As a result, the categories of /(:. i)-modules
that are graded with respect to these dierent gradings are equivalent; see [LOT10a, Propo-
sition 3.10] for more details.
3.3. GRADINGS 43
Remark 3.47. An alternate way to construct the rened grading, which we employ in
Chapter 11 for the case where 1 is a torus, is to use a homomorphism from G
t
(4/) to G(:)
that xes G(:) as a subset of G
t
(4/) (after extending scalars from Z to ). For / 1,
however, there is no homomorphism G
t
(4/) G(:) xing G(:).
Remark 3.48. G(:) has a topological interpretation as the group of homotopy classes of
non-vanishing vector elds on 1 [0. 1] with xed behavior on the boundary. Precisely, x a
vector eld \
0
on 1 1 that is invariant by translation in 1. Then the homotopy classes of
vector elds on 1 [0. 1] that agree with \
0
on neighborhoods of 1 0 and 1 1 form
a group by concatenation; this group is isomorphic to G(:). This is unsurprising in light of
the fact that Seiberg-Witten Floer homology on 1 is graded by non-vanishing vector elds
on 1 [KM07, Section 28].
CHAPTER 4
Bordered Heegaard diagrams
Just as closed 3-manifolds can be represented by Heegaard diagrams, bordered 3-man-
ifolds can be described by a suitable type of Heegaard diagrams, which we call bordered
Heegaard diagrams. Just as Heegaard Floer homology of closed 3-manifolds is associated to
closed Heegaard diagrams, bordered Floer invariants of 3-manifolds are associated to bor-
dered Heegaard diagrams. This chapter is devoted to dening bordered Heegaard diagrams
and discussing their combinatorial / topological properties.
In Section 4.1, we dene bordered 3-manifolds and bordered Heegaard diagrams (Deni-
tions 4.2 and 4.3, respectively), and discuss the relevant existence and uniqueness properties
for bordered diagrams. In Section 4.2, we give some examples of bordered Heegaard dia-
grams, to give the reader a concrete picture. In Section 4.3, we describe the generators of the
bordered Heegaard Floer complex, and discuss the spin
c
structures associated to generators,
and the homology classes connecting them. As in the closed case, one cannot use an arbitrary
bordered Heegaard diagram to dene bordered Heegaard Floer homology: rather, one needs
to use bordered diagrams satisfying certain combinatorial conditions ensuring that the sums
encountered in the dierentials are nite. In Section 4.4, we describe these criteria. Finally,
in Section 4.5, we discuss how to glue bordered Heegaard diagrams and their generators to
obtain Heegaard diagrams and generators for closed manifolds.
4.1. Bordered Heegaard diagrams: denition, existence, and uniqueness
We recall rst the diagrams relevant for Heegaard Floer homology of closed three-
manifolds.
Denition 4.1. A pointed Heegaard diagram 1 is a quadruple (. . . .) consisting of
a compact, oriented surface with no boundary, of some genus p;
two p-tuples =
1
. . . . .
g
and =
1
. . . . .
g
of circles in , with the circles
in each tuple disjoint; and
a point . in ( ),
such that the -circles and -circles intersect transversally and and are each
connected.
We will also sometimes refer to pointed Heegaard diagrams as in Denition 4.1 as closed
Heegaard diagrams, to distinguish them from the bordered Heegaard diagrams introduced
soon.
When discussing pointed Heegaard diagrams (and their bordered variants), we slightly
abuse notation, letting denote both the set or tuple of -curves and the union of the -
curves, and similarly for . Note that the -circles and -circles are each maximal collections
of embedded circles with connected complement. An equivalent requirement to connectivity
of the complement is that the
i
be homologically linearly independent in H
1
(), spanning
a Lagrangian subspace, and likewise for the
i
.
45
46 4. BORDERED HEEGAARD DIAGRAMS
Thus fortied, we turn to the bordered case.
Denition 4.2. A bordered 3-manifold is a triple (1. :. ) where 1 is a compact, oriented
3-manifold with connected boundary 1, : is a pointed matched circle, and : 1(:) 1
is an orientation-preserving homeomorphism.
Two bordered 3-manifolds (1. :. ) and (1
t
. :
t
.
t
) are called equivalent if there is an
orientation-preserving homeomorphism : 1 1
t
so that
t
= .
Note that if : 1(:) 1 and
t
: 1(:) 1 are isotopic homeomorphisms, then,
by viewing the isotopy to a homeomorphism on a collar of 1 , we see that the bordered
3-manifolds (1. :. ) and (1. :.
t
) are equivalent.
We will often abuse notation and refer to a bordered 3-manifold 1 , suppressing : and
. Also, by a bordered 3-manifold we will implicitly mean an equivalence class of bordered
3-manifolds.
Denition 4.3. A bordered Heegaard diagram is a quadruple 1 = (. . . .) consisting of
a compact, oriented surface with one boundary component, of some genus p;
a p-tuple of pairwise-disjoint circles =
1
. . . . .
g
in the interior of ;
a (p + /)-tuple of pairwise-disjoint curves in , split into p / circles
c
=
(
c
1
. . . . .
c
gk
) in the interior of and 2/ arcs
a
= (
a
1
. . . . .
a
2k
) in with boundary
on (and transverse to ); and
a point . in () ( ),
such that the intersections are transverse and and are connected.
The conditions on the numbers of -circles and -circles and arcs again come naturally
from requiring that the tuples be maximal collections of curves of the appropriate type with
connected complement. Again, an equivalent requirement to connectivity of the complement
is that the
i
be homologically linearly independent in H
1
() and that the
i
be homologi-
cally linearly independent in H
1
(. ).
The boundary of a bordered Heegaard diagram naturally has the structure of a pointed
matched circle:
Lemma 4.4. Let (. . . .) be a bordered Heegaard diagram. Let 2 = and a = .
Dene a matching ` on a by `(
a
i
) = i. Then (2. a. `. .) is a pointed matched circle,
in the sense of Denition 3.16.
Proof. We need to verify that the result 2
t
of performing surgery on the points in a
according to the matching ` gives a (connected) circle. Let
t
be the result of performing
surgery on along each -circle, i.e., deleting a tubular neighborhood of each -circle and
gluing in two disks to the result. The circle 2
t
is homeomorphic to the boundary 1 of a
regular neighborhood of

2k
i=1

a
i
. But 1, in turn, is homeomorphic to (
t
nbd()),
and
t
nbd() is a disk.
Denition 4.5. If 1 is a bordered Heegaard diagram, we call the pointed matched circle
appearing on its boundary (in the sense of Lemma 4.4) the boundary of 1 and denote it 1.
Construction 4.6. Let 1 be a bordered Heegaard diagram. We will show how 1 species
a bordered 3-manifold, with boundary parameterized by 1(1).
The construction is illustrated in Figure 4.1. Write 1 = (. . . .), and write : = 1.
Let [c. 0] 2 denote a closed collar neighborhood of , so that 0 2 is identied with
. Choose also a closed tubular neighborhood 2 [0. 1] of 2 in 1(:).
4.1. BORDERED HEEGAARD DIAGRAMS: DEFINITION, EXISTENCE, AND UNIQUENESS 47
Let 1
0
denote the three-manifold obtained by gluing [0. 1] to [c. 0] 1(:), by
identifying ([c. 0] 2) [0. 1] [0. 1] with [c. 0] (2 [0. 1]) [c. 0] 1(:).
Next, attach a 3-dimensional 2-handle to each
i
1 [0. 1] 1
0
and to each

c
i
0 [0. 1] 1
0
. Call the result 1
1
. The manifold 1
1
has three boundary
components:
an o
2
which meets 1,
a surface
t
of genus 2/ which meets 0, and
a surface identied with 1(:), given by 0 1(:) [c. 0] 1(:).
Glue a 3-ball to the o
2
boundary component of 1
1
. Join each
a
i
0
t
to the co-core of
the corresponding handle in c1(:) to form a closed curve, and attach a 3-dimensional
2-handle along each of these circles. Call the result of these attachings 1
2
. The manifold
1
2
has one o
2
boundary component which meets 0 1
0
and a boundary component
identied with 1(:). Glue a 3-ball to the o
2
boundary component of 1
2
and call the result
1 . Let be the obvious identication of 1(:) with 1 . Then (1. :. ) is the desired
bordered 3-manifold.
Denition 4.7. Given a bordered Heegaard diagram we call the bordered 3-manifold given
by Construction 4.6 the bordered 3-manifold represented by 1 and denote it 1 (1).
Construction 4.6 introduces a few extra pairs of canceling handles. This is done to make
it t well with the Morse theory below. A quicker (and equivalent) construction is the
following: 1 (1) is obtained by thickening the Heegaard surface [0. 1] , attaching three-
dimensional two-handle to each
c
i
0 , and a three-dimensional two-handle to each

i
1 . The boundary of the result is naturally identied with 1(:).
Conversely, any 3-manifold with connected boundary is specied by some bordered Hee-
gaard diagram. One way to see this is in terms of Morse theory, following [Lip06b, Section
2.2]:
Denition 4.8. Fix a bordered 3-manifold (1. :. ). We say that a pair consisting of a
Riemannian metric p on 1 and a self-indexing Morse function ) on 1 are compatible with
(1. :. ) if
(1) the boundary of 1 is geodesic,
(2) the gradient vector eld )[
Y
is tangent to 1 ,
(3) ) has a unique index 0 and a unique index 3 critical point, both of which lie on 1 ,
and are the unique index 0 and 2 critical points of )[
Y
, respectively,
(4) the index 1 critical points of )[
Y
are also index 1 critical points of ), and
(5) the pair () .

p) on 1(:) are compatible with the pointed matched circle : in

the sense of Denition 3.18.
Lemma 4.9. Any bordered 3-manifold (1. :. ) is represented by some bordered Heegaard
diagram 1.
Proof. The rst step is to nd a Morse function ) and metric p compatible with (1. :. )
as in Denition 4.8. Choose a Morse function )
F
and metric p
F
on 1(:) compatible with :
in the sense of Denition 3.18 which is self-indexing except that )
F
takes is 3 on the index 2
critical point. Extend )
F

1
and

p
F
to ) and p on a collar neighborhood of 1 satisfying
Conditions (1), (2) and (4), and so that the index 0 and 2 critical points of )
F

1
are index
0 and 3 critical points of ). (Note that Condition (5) is automatically satised.) Extend
48 4. BORDERED HEEGAARD DIAGRAMS
Figure 4.1. The 3-manifold represented by a bordered Heegaard di-
agram. The bordered Heegaard diagram 1 is shown at the top; the four steps
in Construction 4.6 are illustrated below.
) and p arbitrarily to the rest of 1 . Now consider the graph formed by ows between the
index 0 and 1 critical points, which is connected. Since the descending ows from boundary
index 1 critical points remain on the boundary, we can and do cancel every interior index 0
critical point with an interior index 1 critical point. Similarly cancel interior index 3 critical
4.1. BORDERED HEEGAARD DIAGRAMS: DEFINITION, EXISTENCE, AND UNIQUENESS 49
points with index 2 critical points, so that Condition (3) is satised. Modify the function in
the interior so the result is self-indexing.
Now, given the boundary compatible pair (). p), construct a bordered Heegaard diagram
(. . . .) as follows. Let the Heegaard surface be )
1
(3,2), the curves be the inter-
section of the ascending disks of the index 1 critical points of ) (including those on 1 )
with , and the curves be the intersection of the descending disks of the index 2 critical
points of ) with .
The Morse theory description also implies that any two bordered Heegaard diagrams for
the same bordered 3-manifold can be connected by (bordered) Heegaard moves, as specied
in the following:
Proposition 4.10. Any pair of bordered Heegaard diagrams for equivalent bordered 3-man-
ifolds can be made dieomorphic by a sequence of
isotopies of the -curves and -circles, not crossing ,
handleslides of -curves over -circles and -circles over -circles, and
stabilizations and destabilizations in the interior of .
Before giving the proof of Proposition 4.10, for which we follow closely the proof of
[OSz04d, Proposition 2.2], we include two lemmas which will be useful for excluding stabi-
lizing by index zero (or three) critical points (compare [OSz04d, Lemmas 2.3 and 2.4]).
Lemma 4.11. Let 1 be a compact surface of genus / with a single boundary component.
Let =
1
. . . . .
h
be an /-tuple of embedded, pairwise disjoint, simple closed curves so
that 1 is a punctured disk. Suppose moreover that is a simple closed curve in 1 which
is disjoint from the . Then either is null-homologous or there is some
i
with the property
that is isotopic to a curve obtained by handlesliding
i
across some collection of the
j
with
, ,= i.
Proof. If we surger out
1
. . . . .
h
, we replace 1 by the disk 1 with 2/ marked points
j
1
.
1
. . . . . j
h
.
h
(where the pair of points j
i
.
i
corresponds to the zero-sphere which
replaced the circle
i
). Now, can be viewed as a Jordan curve in the disk 1. If separates
some j
i
from its corresponding
i
, then it is easy to see that is isotopic to the curve
gotten by handlesliding
i
over some collection of the
j
with , ,= i. Otherwise, was
null-homologous.
Lemma 4.12. Let 1 be a compact surface of genus / with a single boundary component and

1
. . . . .
d
a collection of embedded, pairwise disjoint, simple closed curves. Then any two
/-tuples of linearly-independent
i
s are related by a sequence of isotopies and handleslides.
Proof. This is proved by induction on /. The case where / = 0 is vacuously true.
Next, suppose we have two subsets with an element in common, which we label
1
. Surger
out
1
to obtain a surface 1
t
with one lower genus. Isotopies in 1
t
which cross the two
surgery points correspond to handleslides in 1 across
1
. Thus, in this case, the proof
follows from the induction hypothesis. Finally, consider the case where we have two disjoint
subsets
1
. . . . .
h
and
h+1
. . . . .
2h
. The curve
h+1
cannot be null-homologous, so,
after renumbering, we can obtain
h+1
by handlesliding
1
over some of the
2
. . . . .
h
,
according to Lemma 4.11. Thus, we have reduced to the case where the two subsets are not
disjoint.
50 4. BORDERED HEEGAARD DIAGRAMS
Proof of Proposition 4.10. The proof involves adapting standard handle calculus
as in [OSz04d, Section 2.1].
Any two bordered Heegaard diagrams 1
0
and 1
1
for equivalent bordered manifolds
(1
i
. :
i
.
i
) come from Morse functions and metrics ()
0
. p
0
) and ()
1
. p
1
) on 1
i
compatible
with the bordering (:. ) of 1
i
, in the sense of Denition 4.8. Using the equivalence between
the two bordered manifolds, we can view )
1
and p
1
as being dened on 1 = 1
1
. Moreover, we
may choose )
0
, p
0
, )
1
and p
1
so that for some neighborhood `(1 ) of 1 , )
1
[
N(Y )
= )
2
[
N(Y )
and p
1
[
N(Y )
= p
2
[
N(Y )
. We may connect )
0
and )
1
by a generic path )
t
of smooth functions
which are Morse functions for generic t and so that the family is constant on `(1 ).
Although the )
t
will not in general be self-indexing, there are still no ow lines from
higher-index critical points to lower-index. To )
t
at generic t we can still associate a gener-
alized Heegaard diagram 1
t
, possibly with extra - and -circles.
While )
t
remains a Morse function, the Heegaard diagram changes by isotopy of the -
and -curves. When )
t
fails to be a Morse function, there is either a birth-death singularity
involving a pair of critical points of adjacent index, or there is a ow line between two critical
points of the same index. The possibilities, and the way the generalized Heegaard diagram
changes, are:
Index 0 and 1 birthdeath: A disjoint -circle is added or deleted, without changing the
span of the -circles.
Index 1 and 2 birthdeath: The Heegaard diagram is stabilized or destabilized.
Index 2 and 3 birthdeath: A disjoint -circle is added or deleted, without changing the
span of the -circles.
Flow line between index 1 critical points: A handleslide of an -circle or -arc over
an -circle. Since our conditions imply that the descending disk of the index 1
critical points corresponding to -arcs lies entirely on the boundary, we cannot have
a handleslide over an -arc.
Flow line between index 2 critical points: A handleslide of a -circle over a -circle.
Note also that the critical points on the boundary cannot be involved in any cancellations,
since ) was xed on all of `(1 ).
Now, given any two bordered Heegaard diagrams 1
0
and 1
1
for 1 , in view of the above
remarks, we can stabilize 1
0
and 1
1
to obtain a pair of bordered Heegaard diagrams 1
t
0
and 1
t
1
with of the same genus and with the following property. Each of 1
t
0
and 1
t
1
can be
extended by adding new - and -circles if necessary to form a pair of generalized bordered
Heegaard diagrams 1
tt
0
and 1
tt
1
respectively, which in turn can be connected by isotopies
and handleslides (subject to the constraints from Proposition 4.10, i.e., -arcs are allowed to
slide over -circles, but not vice versa). The proposition is proved then once we show that
for each generalized Heegaard diagram 1
tt
of genus p, if 1
1
and 1
2
are any two ordinary
bordered Heegaard diagrams obtained from 1
tt
by picking out p / of the -circles and p
of the -circles, each homologically linearly independent, then 1
1
and 1
2
can be connected
by handleslides and isotopies.
To this end, consider any two (p /)-tuples of the -circles of 1
tt
chosen so that they,
along with the 2/ -arcs, span a p + /-dimensional subspace of H
1
(. ). The fact that
these two (p /)-tuples can be connected by isotopies and handleslides follows directly from
Lemma 4.12, applied to the surface 1 =
a
1
. . . . .
a
2k
. It follows similarly that any
two p-tuples of the circles which are linearly independent in H
1
() can be connected by
handleslides and isotopies. This completes the proof.
4.3. GENERATORS, HOMOLOGY CLASSES AND spin
C
STRUCTURES 51
Figure 4.2. Bordered Heegaard diagram for the trefoil complement.
This is based on the observation that +5 surgery on the trefoil is the lens space
1(5. 1), with the knot forming the core of the solid torus a Berge knot. The
diagram is on a genus 2 surface, represented as a torus with a handle attached
(at ).
Let (. . . .) be the result of attaching a cylindrical end to (. . . .), in the sense of
symplectic eld theory [EGH00]. Topologically = , and
a
i
=
a
i
(
a
i
). In due
course, we will choose a conformal structure on making it a punctured Riemann surface
and so that each
a
i
is radial near the puncture. Abusing terminology, we will often also
refer to (. . . .) as a bordered Heegaard diagram.
4.2. Examples of bordered Heegaard diagrams
Here are a few examples of bordered Heegaard diagrams.
Fix an oriented surface , equipped with a p-tuple of pairwise disjoint, homologically
independent curves, , and a (p 1)-tuple of pairwise disjoint, homologically independent
curves
c
=
c
1
. . . . .
g1
1
. Then (.
c
. ) is a Heegaard diagram for a three-manifold
with torus boundary, and indeed any such three-manifold 1 can be described by a Heegaard
diagram of this type. To turn such a diagram into a bordered Heegaard diagram, we proceed
as follows. Fix an additional pair of circles
1
and
2
in so that:

1
and
2
intersect, transversally, in a single point j and
both of the homology classes [
1
] and [
2
] are homologically independent from
[
c
1
]. . . . . [
c
g1
].
Let 1 be a disk around j which is disjoint from all the above curves, except for
1
and
2
,
each of which it meets in a single arc. Then, the complement of 1 species a bordered
Heegaard diagram for 1 , for some parametrization of 1 . A bordered Heegaard diagram
for the trefoil complement is illustrated in Figure 4.2. (See also Chapter 11 for a further
discussion of bordered three-manifolds with torus boundary.)
As another example, consider boundary connect sums. Fix bordered Heegaard diagrams
1
i
= (
i
.
i
.
i
. .
i
) for 1
i
with i = 1. 2. Take the boundary connect sum of
1
and
2
along
.
1
and .
2
, i.e., attach a rectangle to
1

2
by gluing two opposite sides to
1
and
2
along
intervals containing .
1
and .
2
. Introduce a new basepoint . on one of the two remaining
boundary components of the rectangle. (This involves a choice.) This procedure gives a
bordered Heegaard diagram for the boundary connect sum of 1
1
with 1
2
. See Figure 4.3,
where we form the boundary connect sum of two bordered Heegaard diagrams for genus-one
handlebodies to obtain a bordered Heegaard diagram for a genus-two handlebody.
Any bordered Heegaard diagram in which there are no circles necessarily represents
a genus p handlebody. In Figure 4.4, we have illustrated a genus two handlebody with
respect to a parameterization of the genus two surface which is dierent from the one used
in Figure 4.3.
4.3. Generators, homology classes and spin
c
structures
Fix a bordered Heegaard diagram 1, specifying some three-manifold 1 .
Denition 4.13. By a generator of 1 we mean a p-element subset x = r
1
. . . . . r
g
so that
52 4. BORDERED HEEGAARD DIAGRAMS

.
1 1
.
Figure 4.3. Heegaard diagrams for handlebodies. Top: a bordered
genus one handlebody. Bottom: the boundary connect sum of two copies of
the handlebody on top.

1 1
.
Figure 4.4. Another bordered Heegaard diagram for a genus two
handlebody, using a dierent parameterization of the boundary.
exactly one r
i
lies on each -circle,
exactly one r
i
lies on each -circle and
at most one r
i
lies on each -arc.
Let S(1) or S(. . ) denote the set of generators. Given a generator x, let o(x) denote
the set of -arcs which are occupied by x, i.e., o(x) := i [ x
a
i
,= [2/].
Soon (Chapter 5 and later) we will be interested in holomorphic curves in 1
s
1
t
,
where 1
s
= [0. 1] is the unit interval with parameter : and 1
t
is 1 with parameter t. These
curves have boundary on 11
t
and 01
t
. They will be asymptotic to p-tuples
of chords x1
s
and y 1
s
at , where x and y are generators. Each such curve carries a
relative homology class, and we let
2
(x. y) denote the set of these relative homology classes.
More precisely, we make the following denition.
Denition 4.14. Fix generators x and y. Let 1
s
= [0. 1] and 1
t
= [. ] be intervals.
We work in the relative homology group
H
2
_
1
s
1
t
.
_
(o

) 1
t
_
(G
x
) (G
y
)
_
.
4.3. GENERATORS, HOMOLOGY CLASSES AND spin
C
STRUCTURES 53
where
o

= 1
o

= 0
o

= ( .) 1
s
G
x
= x 1
s
G
y
= y 1
s
.
Let
2
(x. y), the homology classes connecting x to y, denote those elements of this group
which map to the relative fundamental class of x 1
s
y 1
s
under the composition of the
boundary homomorphism and collapsing the remainder of the boundary.
Denition 4.15. Given a homology class 1
2
(x. y), projecting to gives a well-dened
element of H
2
(. ), i.e., a linear combination of the components of ( ),
which we sometimes call regions. We call the linear combination of regions corresponding to
1 the domain of 1.
It is not hard to see that a homology class 1 is uniquely determined by its domain.
Denition 4.16. Given any point j ( ), the local multiplicity of 1 at j is the
local degree of the two-chain at the point j.
By hypothesis, the coecient in the domain of 1 of the region 1
z
adjacent to .
is 0.
Remark 4.17. We use the notation
2
for this object to agree with the original conven-
tions [OSz04d]. In the cylindrical setting of Heegaard Floer homology this is a homology
class rather than a homotopy class. Also, note that, unlike the closed case, we require the
multiplicity of a domain to be 0 at .; this is because we are only concerned with

HF.
Notice that concatenation (in the t factor) gives a product :
2
(x. y)
2
(y. w)

2
(x. w); this operation corresponds to addition of domains. If
2
(x. y) is non-empty then
concatenation makes
2
(x. y) into a free, transitive
2
(x. x)-set. Elements of
2
(x. x), or
their domains, are called periodic domains.
Lemma 4.18. There is a natural group isomorphism
2
(x. x)

= H
2
(1. 1 ).
Proof. The proof is a simple adaptation from the closed case ([Lip06a, Lemma 2.1] in
this cylindrical setting); the present case is also proved in [Lip06b, Lemma 2.6.1], which we
repeat here. Let .
t
be a point in the interior of near .. Let
t
= (,) .
t
. There is
an isomorphism

2
(x. x)

= H
2
(
t
[0. 1]. (1) ( 0))
gotten by adding (or subtracting) a copy of x [0. 1] 1. From the long exact sequence for
the pair (
t
[0. 1]. (1) ( 0)), we have
0 H
2
(
t
[0. 1]. (1) ( 0)) H
1
((1) ( 0)) H
1
(
t
).
(Note that H
2
(
t
[0. 1]) = 0.) Consequently,

2
(x. x)

= ker
_
H
1
(,) H
1
() H
1
(,)
_
.
But it follows by analyzing Construction 4.6 that this kernel is exactly H
2
(1. 1 ).
54 4. BORDERED HEEGAARD DIAGRAMS
We sometimes denote the image of 1
2
(x. x) in H
2
(1. 1 ) by [1].
We split up the boundary of the domain of 1 into three pieces,

(1), contained in ,

(1), contained in , and

(1), contained in . We orient them so that the oriented

boundary of the domain of 1 is

(1) +

(1) +

(1); in particular, this implies that

(

1) = (

1) + (

1) = y x. We will think of

1 as an element of H
1
(. a).
Denition 4.19. A homology class 1 connecting x to y is called provincial if

1 = 0.
Let

2
(x. y) denote the set of provincial homology classes connecting x to y.
In other words, a homology class 1
2
(x. y) belongs to

2
(x. y) if and only if the
domain of 1 does not include any of the regions adjacent to .
The concatenation maps obviously restrict to maps :

2
(x. y)

2
(y. w)

2
(x. w),
again making

2
(x. y) into a free, transitive

2
(x. x)-set (when non-empty). Elements of

2
(x. x), or their domains, are called provincial periodic domains.
Lemma 4.20. There is a natural group isomorphism

2
(x. x)

= H
2
(1 ).
Proof. Again, the proof is a simple adaptation of the closed case, and is given in
[Lip06b, Lemma 2.6.4]. There is an isomorphism

2
(x. y)

= H
2
( [0. 1]. (1) ( 0)).
Using the long exact sequence for the pair ([0. 1]. (1) ( 0)), it follows that
H
2
( [0. 1]. (1) ( 0))

= ker(H
1
() H
1
() H
1
()).
The Mayer-Vietoris sequence for the decomposition of 1 along identies this kernel with
H
2
(1 ).
Abusing notation, we may also sometimes denote the image of 1

2
(x. x) in H
2
(1 )
by [1].
Next we ask when
2
(x. y) and

2
(x. y) are non-empty. As in the closed case [OSz04d,
Section 2.6], the answer relates to spin
c
structures on 1 . Also as in the closed case, our
treatment is close to Turaevs [Tur97].
Choose a self-indexing Morse function ) and metric p on 1 inducing the bordered Hee-
gaard diagram (. . ). Let x be a generator. Using ), . and x we will construct
a non-vanishing vector eld
z
(x); this reduces the structure group of 11 from oC(3)
to oC(2) = l(1), and composing with the inclusion l(1) l(2) = spin
c
(3) given by
(c
i
)
_
e
i
0
0 1
_
denes a spin
c
structure on 1 .
Each r
i
x lies on a ow line
i
of ) from an index 1 critical point of ) to an index
2 critical point of ). The point . lies on a ow line from the index 0 critical point of )
to the index 3 critical point of ). Let 1 denote a regular neighborhood of
1

g
.
Since each ow line
i
or runs between critical points of opposite parity, it is possible to
extend )[
Y \B
to a vector eld
0
z
(x) on 1 so that
0
z
(x) is non-vanishing on 1. The vector
eld
0
z
(x) still has exactly / zeroes, one for each
a
i
not containing an r
i
. All of these
zeroes lie on 1 , however, so one can modify
0
z
(x) in a neighborhood of these zeroes to
produce a non-vanishing vector eld
z
(x). This modication can be done in a standard way,
depending only on o(x), and not on the generator x itself. The vector eld
z
(x) induces
the spin
c
structure s
z
(x). It is routine to verify that s
z
(x) does not depend on the choices
made in its construction.
Notice that the restriction of s
z
(x) to a collar neighborhood of 1 depends only on
the set of occupied -arcs o(x), and not on x itself. That is, for o a /-element subset of
4.4. ADMISSIBILITY CRITERIA 55

a
1
. . . . .
a
2k
, there is an induced spin
c
structure s
z
(o) on a collar neighborhood of 1 .
Given x, let s
rel
z
(x) denote the relative spin
c
structure on (1. nbd(1 )) induced by (.. x),
relative to s
z
(o).
1
Lemma 4.21. Given generators x and y,
2
(x. y) ,= if and only if s
z
(x) = s
z
(y). Further,

2
(x. y) ,= if and only if o(x) = o(y) and s
rel
z
(x) = s
rel
z
(y).
Proof. Again, this is a straightforward adaptation of the closed case, and is also proved
in [Lip06b, Lemmas 2.6.2 and 2.6.5].
On the one hand, the obstruction c(x. y) to nding a homology class connecting x to y is
gotten as follows: join x to y by a union of paths

( .) and by a union of paths

. Then x and y can be connected by a homology class if and only

can be made
null-homologous in by adding or subtracting some -curves and -circles, i.e., if and only
if the image c(x. y) of

in H
1
([0. 1]. 01(.)[0. 1])

= H
1
(1. 1 )
vanishes.
On the other hand, the dierence class s
z
(y) s
z
(x) is gotten as follows. As in the
denition of s
z
(x), after xing a Morse function and metric inducing the Heegaard diagram,
the generator x corresponds to p ows
x
1
. . . . .
xg
in 1 , and the generator y corresponds to
p ows
y
1
. . . . .
yg
in 1 . The dierence class is given by
s
z
(y) s
z
(x) = [
y
1
+ +
yg

x
1

xg
] H
1
(1. 1 )

= H
2
(1 ).
The chain
y
1
+ +
yg

x
1

xg
is homologous in (1. 1 ) to the chain

dening c(x. y).

The case for s
rel
z
is exactly analogous, except that now

, the group H
1
([0. 1].
01(.) [0. 1])

= H
1
(1. 1 ) is replaced by H
1
([0. 1]. 01),
and the group H
1
(1. 1 ) is replaced by H
1
(1 )

= H
2
(1. 1 ).
For s a spin
c
structure on 1 , dene S(1. s) S(1) to be those generators x so that
s
z
(x) = s.
4.4. Admissibility criteria
As in [OSz04d], we need certain additional hypothesis to ensure the denitions of the
dierentials involve only nite sums. The idea is as follows. Given a homology class 1

2
(x. y), in Chapter 5 we will construct a moduli space of holomorphic curves in the homology
class 1. In Chapters 6 and 7, we will use counts these moduli spaces, when they are 0-
dimensional, to dene the invariants of bordered 3-manifolds. For each 1, such a count will
be nite. But we will sum over all 1 (perhaps with given

), so we must ensure that only

nitely many such 1 can contribute. We will use the fact (Lemma 5.4) that if the moduli
space of curves in class 1 is non-empty then 1 must be positive, in the following sense:
Denition 4.22. A homology class 1
2
(x. y) is called positive if the coecient in the
domain of 1 of every component of ( ) is non-negative.
1
A spin
c
-structure on Y is a lift of the classifying map [TY ] : Y BSO(3) to a map s: Y Bspin
c
(3).
Given a spin
c
-structure s
V
on a subset V Y , a relative spin
c
-structure on Y relative to s
V
is a lift
s: Y Bspin
c
(3) with s[
V
= s
V
. The relative spin
c
-structures on (Y, V ) form an ane copy of H
2
(Y, V ).
While we normally consider spin
c
-structures up to homotopy (and relative spin
c
-structures up to homo-
topy relative to V ) to make sense of relative spin
c
-structures on (Y, V ) relative to s
V
, it is important that
the lift s
V
over V be given on the nose, not just up to homotopy.
56 4. BORDERED HEEGAARD DIAGRAMS
We will have two dierent analogues of the weak admissibility criterion of [OSz04d,
Denition 4.10], one of which we call provincial admissibility and the other of which we call
simply admissibility. We will see in Lemmas 4.28 and 4.29 that these conditions ensure that
the set of positive homology classes with given

(respectively all positive homology classes)

is nite.
In Chapters 6 and 7, provincial admissibility will be sucient to dene the invariants
of bordered 3-manifolds. However, in order for the tensor product involved in gluing two
bordered 3-manifolds to make sense, one of the two manifolds has to be admissible and not
just provincially admissible.
Denition 4.23. A bordered Heegaard diagram is called provincially admissible if every
provincial periodic domain has both positive and negative coecients.
Denition 4.24. A bordered Heegaard diagram is called admissible if every periodic domain
has both positive and negative coecients.
Proposition 4.25. Every bordered Heegaard diagram is isotopic to an admissible bordered
Heegaard diagram. Further, any two admissible bordered Heegaard diagrams can be connected
though a sequence of Heegaard moves in which every intermediate Heegaard diagram is ad-
missible. The same statements hold if admissible is replaced by provincially admissible.
Proof. This follows by winding transverse to the -circles, as in the case of closed
manifolds [OSz04d, Section 5].
As in the closed case, there are reformulations of the two admissibility criteria:
Lemma 4.26. A bordered Heegaard diagram (. . ) is admissible if and only if there is an
area function on such that (1) = 0 for any periodic domain 1. A bordered Heegaard
diagram (. . ) is provincially admissible if and only if there is an area function on
such that (1) = 0 for any provincial periodic domain 1.
Proof. The only if direction is immediate. For the if direction, the proof follows
exactly as in the closed case, see [OSz04d, Lemma 4.12]. For the readers benet, we recast
the result here in terms of Farkass lemma from the theory of convex sets [Far02]:
Lemma 4.27. Let \ be a vector space and j
i

N
i=1
\ a nite set of vectors. Then either
there is a nonzero linear functional / \

such that /(j

i
) 0 for all i or
there are constants c
i
0 such that

N
i=1
c
i
j
i
= 0.
Now, to prove the rst statement, enumerate the regions 1
1
. . . . . 1
N
of . Take \

to be
the vector space generated by the periodic domains in , and j
i
\ = \

to be evaluation
at the i
th
region 1
i
. By assumption, there is no nonzero 1 \

such that j
i
(\ ) 0 for all
i. Consequently, there are constants c
i
0 such that

N
i1
c
i
j
i
= 0. But then setting the
area of 1
i
to be c
i
is the desired area function.
The second statement follows by just the same reasoning, using only the provincial peri-
odic domains.
Proposition 4.28. Suppose that 1 is provincially admissible, as in Denition 4.23. Fix
generators x. y S(1) and an element / H
1
(2
t
. a). Then there are nitely many 1

2
(x. y) such that

1 = / and all coecients of 1 are non-negative.

4.5. CLOSED DIAGRAMS 57
Proof. Let be an area form as guaranteed by Lemma 4.26. If 1. 1
t

2
(x. y) with

1 =

1
t
= / then 1 1
t
is a provincial periodic domain. Consequently the areas
(1) and (1
t
) are equal. But there are only nitely many positive domains of any given
area.
Proposition 4.29. Suppose that 1 is admissible, as in Denition 4.24. There are nitely
many generators x. y S(1) and 1
2
(x. y) such that all coecients of 1 are non-
negative.
Proof. There are only nitely many generators, so we may rst x x and y. Let be
an area form as guaranteed by Lemma 4.26. If 1. 1
t

2
(x. y) then 1 1
t
is a periodic
domain. Consequently the areas (1) and (1
t
) are equal. But there are only nitely many
positive domains of any given area.
4.5. Closed diagrams
In this section we discuss the behavior of (bordered) Heegaard diagrams, generators and
homology classes under cutting and gluing.
We will use minus signs to denote orientation reversal, so, for instance, 1 is 1 with its
orientation reversed and 1 is 1 = (. . . .) with the orientation of reversed. Note
that 1 (1)

= 1 (1) and 1(:)

= 1(:), canonically.
Gluing bordered Heegaard diagrams corresponds to gluing bordered 3-manifolds:
Lemma 4.30. Let 1
1
= (
1
.
1
.
1
. .
1
) and 1
2
= (
1
.
1
.
1
. .
2
) be bordered Heegaard
diagrams for bordered 3-manifolds (1
1
. :.
1
) and (1
2
. :.
2
). Obtain a closed Heegaard
diagram 1
1

1
2
by gluing 1
1
and 1
2
along their boundaries, according to the markings
from their pointed matched circles.
1
1

1
2
= (
1

2
.
1

2
.
1

2
. .
1
= .
2
).
Then 1
1

1
2
represents 1
1

1
2
= 1
1

1
1
1
2
.
Proof. This is immediate from the Morse theory description.
Cutting bordered Heegaard diagrams corresponds to cutting bordered 3-manifolds:
Lemma 4.31. Let 1 = (. . . .) be a (closed) Heegaard diagram and 2 a separating curve
such that
2 is disjoint from the -circles (so that ( 2) has exactly two components),
( 2) has exactly two components, and
2 passes through the basepoint.
Write =
L

Z

R
. Then 1
L
= (
L
.
L
.
L
. .) and 1
R
= (
R
.
R
.
R
. .)
are bordered Heegaard diagrams. Further, there is a separating surface 1 1 (1) so that
1 (1
L
) and 1 (1
R
) are the two components of 1 (1) 1.
Proof. It is immediate from the hypotheses that 1
L
and 1
R
are bordered Heegaard
diagrams. The fact that there is a surface 1 1 (1) as in the statement of the lemma
follows from Lemma 4.30. Alternatively, 1 is the union of the ow lines through 2 with
respect to a Morse function and metric inducing 1.
Next we discuss how generators and homology classes behave under these cutting oper-
ations. Let 1
1
and 1
2
be bordered Heegaard diagrams with 1
1
= 1
2
. Fix generators
58 4. BORDERED HEEGAARD DIAGRAMS
x
1
S(1
1
) and x
2
S(1
2
). If the corresponding sets of occupied -arcs are disjoint, i.e.,
o(x
1
) o(x
2
) = , then the union x
1
x
2
can be viewed as a generator in S(1) for the
Heegaard Floer complex of the closed manifold 1 . We will call pairs of generators (x
1
. x
2
)
such that o(x
1
o(x
2
) = compatible pairs.
Lemma 4.32. Let (1
1
. .) = (
1
.
1
.
1
. .) and (1
2
. .) = (
2
.
2
.
2
. .) be two bordered
Heegaard diagrams for the bordered three-manifolds (1
1
. :.
1
) and (1
2
. :.
2
), with com-
patibly marked boundaries. Given x
1
. y
1
S(1
1
) and x
2
. y
2
S(1
2
), we have a natural
identication of
2
(x
1
x
2
. y
1
y
2
) with the subset of
2
(x
1
. y
1
)
2
(x
2
. y
2
) consisting of
pairs (1
1
. 1
2
) with

(1
1
)+

(1
2
) = 0. Moreover, given j
2
() and 1
i

2
(x
i
. y
i
)
for i = 1. 2 which agree along the boundary, the local multiplicity of 1
2
at j coincides with
the local multiplicity of the glued homology class in
2
(x
1
x
2
. y
1
y
2
) (under the above
identication) at the point j, now thought of as a point in =
1

2
.
Proof. The proof is straightforward.
For 1
1
and 1
2
which agree along the boundary, let 1
1
1
2
be the homology class in

2
(x
1
x
2
. y
1
y
2
) guaranteed by Lemma 4.32.
Finally, we give a criterion for when a closed Heegaard diagram is admissible.
Lemma 4.33. Let 1
1
and 1
2
be two bordered Heegaard diagrams with 1
1
= 1
2
. If 1
1
is admissible and 1
2
is provincially admissible, then 1 = 1
1

1
2
is admissible.
Proof. Let 1
1
1
2
be a positive periodic domain for 1. Since 1
1
is admissible, 1
1
= 0
and so

1
2
= 0. Then since 1
2
is provincially admissible, 1
2
= 0.
CHAPTER 5
Moduli spaces
Let 1 be a bordered 3-manifold. In order to dene invariants of 1 we will choose
a bordered Heegaard diagram 1 = (. . . .) for 1 and count J-holomorphic curves in
[0. 1] 1 with boundary conditions coming from and . In this chapter we give
the technical results concerning the moduli spaces of holomorphic curves that we use to
construct the invariants

CFD(1) and

CFA(1). We begin in Section 5.1 with an outline of
the main results that we will need and the ideas in the proof; the precise mathematics begins
in Section 5.2 below.
5.1. Overview of the moduli spaces
In this chapter, we will discuss J-holomorphic curves
1
n: (o. o) ( [0. 1] 1. (1 1) ( 0 1))
where o is a Riemann surface with boundary and punctures. Here we view , the interior
of , as a Riemann surface with one puncture, denoted j. As in Section 4.3, the coordinate
on the [0. 1] factor is denoted : and the coordinate on the 1 factor is denoted t.
We view the manifold [0. 1] 1 as having three dierent innities: [0. 1] +,
[0. 1] and j [0. 1] 1, which we refer to as +, and c respectively.
(Here, c stands for east.) The asymptotics of the holomorphic curves n we consider are:
At , n is asymptotic to a p-tuple of chords of the form r
i
[0. 1] , where
x = r
i

g
i=1
is a generator in the sense of Denition 4.13.
At c, n is asymptotic to a nite collection of Reeb chords
i
(1. t
i
). Here

i
is a Reeb chord in the ideal contact boundary (also called 2) with endpoints
on a = 2 (see Section 3.1.3). We refer to t
i
as the height of the chord
i
.
The set of east punctures of n is partially ordered by the heights of the corresponding Reeb
chords. This induces an ordered partition

1 = (1
1
. 1
2
. . . . . 1

) of the east punctures, where

each 1
i
consists of those punctures occurring at one particular height. (For

CFD(1) we only
need to consider discrete partitions where each 1
i
has only one puncture.)
With this background, we consider the moduli space

/
B
(x. y; o

;

1)
(Denition 5.10), which consists of curves from the decorated source o

, asymptotic to x at
and y and +, in the homology class 1 (as in Section 4.3) and respecting the ordered
partition

1 of the Reeb chords. Here the source surface o is decorated with labels describing
its asymptotics (Denition 5.2). We also consider the reduced moduli space, its quotient
/
B
(x. y; o

;

1) :=

/
B
(x. y; o

;

1),1
1
With respect to an appropriate J; see Denition 5.1 below.
59
60 5. MODULI SPACES
by translation in the t coordinate.
These moduli spaces are smooth manifolds (Proposition 5.6) whose dimensions are easy
to compute (Equation (5.9)). In fact, for an embedded curve the dimension is independent
of the source o and depends only on the asymptotics and homology class of the curve
(Section 5.7); this gives rise to the gradings on the algebra and on the modules.
As is typical in symplectic geometry, our invariants are dened by looking only at rigid
moduli spaces, those /
B
(x. y; o

;

1) which are 0-dimensional. To show that
2
= 0 and for
proofs of invariance, we also need to consider 1-dimensional moduli spaces, and, in particular,
we need appropriate compactications. These are provided by the spaces /
B
(x. y; o

;

1),
which include holomorphic combs. The idea is that, as in symplectic eld theory, some parts
of a holomorphic curve may go to innity relative to other parts. In our case, this can result
in a degeneration at or at c. When a curve degenerates at , the result is two or
more curves of the kind we have already discussed. A degeneration at c creates a new kind
of object: a curve in 2 1[0. 1] 1. Section 5.3 is devoted to discussing these curves at
east innity. This compactication is dened in Denition 5.21, and a schematic illustration
of a holomorphic comb can be found in Figure 5.6.
The main result we need for our purposes is a classication of the codimension 1 degen-
erations (Section 5.6). Proposition 5.37 and Lemma 5.50 say that for the curves we consider,
these degenerations come in four kinds:
Degenerating into a two-story building at . These are exactly the kinds of
degenerations that occur in Heegaard Floer homology for closed three-manifolds.
A boundary branch point of

n can go o to c, in the process splitting a Reeb

chord in two. Below, this is referred to as degenerating a join curve at c. (From
the point of view of the curve at c, two Reeb chords come together and merge as
we travel further away from the main surface.)
A collapse in the ordering of

1, i.e., the heights of two parts of

1 coming together.
In the process, some split curves can degenerate at c. These again correspond to
boundary branch points of

n approaching c, but in a combinatorially dierent

way from join curves. (In the curve at c, one (or more) Reeb chord splits apart
as we travel further away from the main surface.)
Degenerating a shue curve at east . This degeneration involves either an
interior branch point or two boundary branch points approaching east , and is
explained in more detail at the end of Section 5.3; see in particular Figure 5.5 for
a domain in which a shue curve degenerates. This type of degeneration is not
relevant for

CFD, only for

CFA. (In the curve at c, two Reeb chords recombine
with each other to create two dierent Reeb chords as we travel further away from
the main surface.)
Figure 5.1 shows examples in which the rst three kinds of degenerations occur.
Our next goal is to obtain an identity relating the number of degenerations of the
four types above. The usual way to do this is to show that if the expected dimension
ind(x. y; o

;

1) is 2, then the space /
B
(x. y; o

;

1) is a compact one-manifold with bound-
ary given by all holomorphic combs with the correct combinatorics to occur as one of these
degenerations. The identity would then follow from the fact that a compact one-manifold
has an even number of endpoints.
5.2. HOLOMORPHIC CURVES IN [0, 1] 1 61
Figure 5.1. Examples of three kinds of codimension 1 degenerations.
The large black dot represents a boundary branch point of

n. Left: degen-
erating into a two-story building. Center: degenerating a join curve. Right:
degenerating a split curve. The diagrams show the projection of the curve in
[0. 1] 1 to .
Unfortunately, available gluing technology seems ill equipped for this approach, the di-
culty being with split and shue curve degenerations. (See Remark 5.35 and the beginning
of Section 5.6.4 for more discussion of this point.) Instead, we use a somewhat indirect
argument to show that there are an even number of holomorphic combs of the types that
may appear as degenerations in an index 2 moduli space (Theorem 5.55). This is all that
we will need later to dene the invariants.
This chapter is organized as follows. Section 5.2 denes the moduli spaces of curves in
[0. 1] 1 of interest to us, and Section 5.3 the moduli spaces for curves at east innity. In
Section 5.4, we put these moduli spaces together to dene the compactied moduli spaces of
so-called holomorphic combs and prove that they form compactications. Next, in Section 5.5
we prove the gluing results for combs needed for the rest of the paper. In Section 5.6 we
restrict what kind of degenerations can occur in codimension-one; in particular, we show
that only those degenerations for which we proved gluing results are possible, leading to
Theorem 5.55. Finally, in Section 5.7 we study the index of the -operator at an embedded
curve, and use this to give further restrictions on the codimension-one boundaries of moduli
spaces of embedded curves.
Fortied by this overview, we turn to the mathematics itself.
5.2. Holomorphic curves in [0. 1] 1
In this section, we will dene various moduli spaces of holomorphic curves in [0. 1]1
with names like /
B
(x. y; o

), /
B
(x. y; o

; 1) and /
B
(x. y; o

;

1). In Chapters 6 and 7 we
will use the moduli spaces /
B
(x. y; o

;

1) to dene our invariants of bordered 3-manifolds.
These are, consequently, the spaces in which we will ultimately be interested. However, it
62 5. MODULI SPACES
is easier to formulate several technical results by describing the spaces /
B
(x. y; o

;

1) as
subspaces of /
B
(x. y; o

) and /
B
(x. y; o

; 1), which we introduce rst.

As before, let 1 be a bordered Heegaard diagram of genus p representing a manifold 1
with boundary of genus /. Choose a symplectic form

on , with respect to which the

boundary of is a cylindrical end. Let ,

be a complex structure on compatible with

.
With respect to ,

, the boundary is a puncture, which we denote j. We will assume that the

-arcs
a
are cylindrical near j in the following sense. Fix a punctured neighborhood l
p
of
j, and a symplectic identication : l
p

=
o
1
(0. ) 1

o
1
. Then we assume that both
,

[
Up
and (
a
i
l
p
) are invariant with respect to the 1-translation action in o
1
(0. ).
Let
e
be the result of lling in the puncture of , so ,

induces a complex structure on

e
.
We consider moduli spaces of holomorphic curves in [0. 1] 1. Let

: [0. 1] 1 .

|
: [0. 1] 1 [0. 1] 1.
:: [0. 1] 1 [0. 1], and
t : [0. 1] 1 1
denote the projections. We equip [0. 1] 1 with the symplectic form
|
:= d: dt and the
complex structure ,
|
with ,
|
_

s
_
=

t
, and equip [0. 1] 1 with the split symplectic
form

) +

|
(
|
). Note also that there is an 1-action on [0. 1] 1 by translation
on the t coordinate.
Denition 5.1. We say an almost complex structure J on [0. 1] 1 is admissible if the
following conditions are satised:
(J-1) The projection map
|
is J-holomorphic.
(J-2) For

s
and

t
the vector elds tangent to the bers of

induced by : and t
respectively, J

s
=

t
(and consequently the bers of

are J-holomorphic).
(J-3) The 1-action is J-holomorphic.
(J-4) The complex structure is split, i.e., J = ,

,
|
, near j [0. 1] 1.
For example, the split complex structure ,

,
|
is admissible in the above sense. Fix
now an admissible J; later, we will assume J is generic in certain senses.
Denition 5.2. By a decorated source o

we mean:
(1) a topological type of smooth (not nodal) Riemann surface o with boundary and
punctures on the boundary,
(2) a labeling of each puncture of o by one of +, , or c and
(3) a labeling of each c puncture of o by a Reeb chord in (2. a), as dened in Sec-
tion 3.1.3.
Given a decorated source o

, let o
e
denote the result of lling in the c punctures of o.
We will consider maps
n: (o. o) ( [0. 1] 1. (1 1) ( 0 1))
such that
(M-1) The map n is (,. J)-holomorphic with respect to some almost complex structure ,
on o.
5.2. HOLOMORPHIC CURVES IN [0, 1] 1 63
(M-2) The map n: o [0. 1] 1 is proper.
(M-3) The map n extends to a proper map n
e
: o
e

e
[0. 1] 1.
(M-4) The map n
e
has nite energy in the sense of Bourgeois, Eliashberg, Hofer, Wysocki
and Zehnder [BEH
+
03].
(M-5)
|
n is a p-fold branched cover. In particular,
|
n is non-constant on every
component of o.
(M-6) At each -puncture of o, lim
zq
(t n)(.) = .
(M-7) At each +-puncture of o, lim
zq
(t n)(.) = +.
(M-8) At each c puncture of o, lim
zq
(

n)(.) is the Reeb chord labeling .

(M-9)

n does not cover the region of adjacent to ..

(M-10) For each t 1 and each i = 1. . . . . p, n
1
(
i
0 t) consists of exactly one
point. Similarly, for each t 1 and each i = 1. . . . . p /, n
1
(
c
i
1 t)
consists of exactly one point.
We call Condition (M-10) weak boundary monotonicity. Sometimes later we will impose
the following additional condition, which we call strong boundary monotonicity:
(M-11) For each t 1 and each i = 1. . . . . 2/, n
1
(
a
i
1 t) consists of at most one
point.
However, our moduli spaces are easiest to dene if we do not initially impose this condition.
It follows from the Conditions (M-1)(M-10) that at , n is asymptotic to a p-tuple
of chords of the form r
i
[0. 1], where r
i
. By weak boundary monotonicity, the set
x = r
i
has the property that exactly one point of x lies on each -circle, and one point
on each -circle. However, more than one point of x may lie on the same -arc. We call
such a p-tuple a generalized generator. If the strong boundary monotonicity condition is also
satised then at most one r
i
lies on each -arc, and hence x is a generator in the sense of
Section 4.3. Exactly analogous statements hold at +.
Any curve n satisfying Conditions (M-1)(M-10) belongs to some homology class [n]

2
(x. y). (Here,
2
(x. y) is the obvious generalization of the notion from Section 4.3 to the
case of generalized generators.) We collect these curves into moduli spaces:
Denition 5.3. Given generalized generators x and y and a homology class 1
2
(x. y),
let

/
B
(x. y; o

)
denote the moduli space of curves satisfying Conditions (M-1)(M-10), with decorated source
o

, asymptotic to x at and y at + and in the homology class 1.

The domains of holomorphic curves (in the sense of Denition 4.15) are always positive.
Lemma 5.4. If the moduli space

/
B
(x. y; o

) is non-empty then the homology class 1 is

positive (in the sense of Denition 4.22).
Proof. Let n be an element of

/
B
(x. y; o

). The multiplicity of 1 at a region 1 is

given by the intersection number n (j [0. 1] 1). By Condition (J-2) of Denition 5.1,
the ber j [0. 1] 1 is J-holomorphic, so n (j [0. 1] 1) is the intersection number
between two J-holomorphic curves, which is positive by [McD94] or [MW95].
64 5. MODULI SPACES
Given n

/
B
(x. y; o

), for each puncture of o labeled by c, let c

q
(n) = t n
e
().
This gives an evaluation map ev
q
:

/
B
(x. y; o

) 1. Set
c =

qE(S

)
ev
q
:

/
B
(x. y; o

) 1
E(S

)
.
where 1(o

) (or just 1, when unambiguous) is the set of east punctures of o

.
We next use the evaluation maps to dene certain subspaces of the

/
B
(x. y; o

). Fix
a partition 1 = 1
i
of 1. By the partial diagonal
P
in 1
E
we mean the subspace of 1
E
dened by the set of equations r
p
= r
q
[ 1
i
1. j. 1
i
. That is,
P
is obtained by
requiring all coordinates in each part 1
i
of 1 to be equal. Now, we use the cycle
P
to cut
down the moduli spaces:
Denition 5.5. Given generalized generators x and y, 1
2
(x. y), a decorated source o

,
and a partition 1 of the c punctures of o

, let

/
B
(x. y; o

; 1) := ev
1
(
P
)

/
B
(x. y; o

).
Notice that

/
B
(x. y; o

) is the special case of

/
B
(x. y; o

; 1) when 1 is the discrete

partition, with one element per part.
We turn now to the meaning of genericity for J.
Proposition 5.6. There is a dense set of admissible J with the property that the moduli
spaces

/
B
(x. y; o

) are transversally cut out by the -equations. Indeed, for any countable
set `
i
of submanifolds of 1
E
, there is a dense set of admissible J which satisfy the further
property that ev:

/
B
(x. y; o

) 1
E
is transverse to all of the `
i
.
The proof is standard. See [MS04, Chapter 3] for a nice explanation in a slightly dierent
context, or [Lip06a, Proposition 3.8] for the proof of the rst half of the statement in the
present setting.
In particular, by Proposition 5.6, we can choose J satisfying the following denition:
Denition 5.7. An admissible almost complex structure is called suciently generic if the
moduli spaces

/
B
(x. y; o

; 1) are transversely cut out for all choices of x, y, 1, o

and 1.
From now on, we will always work with suciently generic almost complex structures,
unless we explicitly say otherwise.
The expected dimension of

/
B
(x. y; o

; 1) is not hard to calculate:

Proposition 5.8. The expected dimension ind(1. o

. 1) of

/
B
(x. y; o

; 1) is
(5.9) ind(1. o

. 1) = p (o) + 2c(1) +[1[.

Here [1[ denotes the number of parts in 1 and c(1) is the Euler measure of the domain
of 1.
Here the Euler measure of a region in () is its Euler characteristic minus 1,4 the
number of corners (intersections of -curves and -curves or -arcs and ), and is dened
to be additive under union. The Euler measure can be viewed as the integral of the Gaussian
curvature of the region with respect to any Riemannian metric for which the boundary, the
-curves and the -circles are all geodesics, meeting at right angles at the corners.
5.2. HOLOMORPHIC CURVES IN [0, 1] 1 65
Figure 5.2. A capping operation. Part of the surface

and the arcs

are shown; on the left side is and , and on the right side is [0. c) and

0
.
Proof of Proposition 5.8. This proof is given in [Lip06b, Proposition 4.5.1]; we
repeat it here for the readers convenience. In brief, the Proposition follows from the cor-
responding index formula for closed Heegaard Floer homology, [Lip06a, Formula (6)], by
a standard doubling argument at the east punctures, cf. [Bou06, Section 5]. (The for-
mula [Lip06a, Formula (6)] itself follows from the Riemann-Roch theorem by another dou-
bling argument.)
Let

denote the result of gluing a collar [0. c) to . Inside ( .) [0. c),

choose arcs connecting every pair of (ends of) -arcs, as shown in Figure 5.2. Call the new
arcs

0
, and let

0
. (This caps o the Reeb chords at east innity.) We call
the result the capped diagram. By attaching cylindrical ends to them, we can think of the

0
as arcs

0
lying in 1.
For each Reeb chord
i
there is a holomorphic disk 1

i
in ( 1 [0. 1] 1.

0

1 1) asymptotic to
i
. Given a map n /
B
(x. y; o

) where the c punctures of

o

are decorated by Reeb chords

i
1
. . . . .
in
one can preglue n to

n
j=1
1

i
j
to obtain an
approximately holomorphic map n

: o

[0. 1] 1.
Each height constraint decreases both [1[ and the expected dimension by one. So, it
suces to prove the formula in the case that 1 is the discrete partition. Recall from [Lip06a,
Formula (6), Section 4.1] that in the closed case the index at a map n: o [0. 1] 1 is
given by p (o) +2c(1). The closed formula applies to the linearization of the

-operator
at n

, giving ind(n

) = p (o

) + 2c(1

) = p (o) + 2c(1

), where 1

is the domain
in

corresponding to the domain 1 in . The index of 1

at any of the disks 1

is 1. It
follows that the index at n is
ind(n) = p (o) + 2(c(1) +[1[,2) [1[ +[1[.
where the [1[,2 comes from the eect of capping on the Euler measure, the [1[ from the
fact that ind(1

) = 1, and the +[1[ comes from the matching conditions at the punctures
(because of the Morse-Bott nature of the gluing). This proves the proposition.
Suppose n

/
B
(x. y; o

; 1). Then the t-coordinate of n induces a partial ordering on

the partition 1. This combinatorial data allows us to divide

/
B
(x. y; o

; 1) into dierent
strata.
66 5. MODULI SPACES
Denition 5.10. Given x, y, and o

as above and an ordered partition

1 of the c punctures
of o

, with 1 the corresponding unordered partition, let

/
B
(x. y; o

;

1) denote the (open)
subset of

/
B
(x. y; o

; 1) consisting of those holomorphic curves for which the ordering of 1

induced by t agrees with the ordering given in

1.
That is, for n

/
B
(x. y; o

; 1) and

1 = (1
1
. . . . . 1
m
), we have n

/
B
(x. y; o

;

1) if
and only if for all i < i
t
, 1
i
and
t
1
i
implies t n() < t n(
t
).
Given a partition 1 of 1(o

), there is an associated set [1] of multi-sets of Reeb chords,

given by replacing each puncture of o

in 1 by the associated Reeb chord. Similarly, given

an ordered partition

1 of 1(o

) there is an associated sequence [

1] of multi-sets of Reeb
chords.
The 1-action on [0. 1] 1 by translation on 1 induces an 1-action on each

/
B
(x. y;
o

; 1) and

/
B
(x. y; o

;

1). We denote this action by
R
for 1 1. The action is free
except in the trivial case that o

consists of p disks with two boundary punctures each, and

1 = 0. We say that a curve is stable if it is not this trivial case. We are interested primarily
in the quotient by this action:
Denition 5.11. Given x, y, 1, o

, and 1 or

1 as above, let
/
B
(x. y; o

; 1) :=

/
B
(x. y; o

; 1),1
/
B
(x. y; o

;

1) :=

/
B
(x. y; o

;

1),1.
denote the reduced moduli spaces.
The expected dimension of /
B
(x. y; o

; 1) (except in the trivial case 1 = 0) is given

by ind(1. o

. 1) 1.
Although the evaluation maps c
q
(for 1) do not descend to the quotient by trans-
lation, the dierence between any two of them, ev
p
ev
q
, does descend. We denote this
dierence by
c
p,q
: /
B
(x. y; o

;

1) 1. (5.12)
We can also combine all the evaluation maps into a single map
c : /
B
(x. y; o

;

1) 1
E
,1
where 1 acts diagonally by translation on 1
E
.
5.3. Holomorphic curves in 2 1 [0. 1] 1
In order to discuss the compactications of the moduli spaces /
B
(x. y; o

; 1) and
/
B
(x. y; o

;

1), we will need to consider certain holomorphic curves at east , i.e., in
21[0. 1] 1. We endow 21[0. 1] 1 with the obvious split symplectic form. Inside
this space, we have 4/ Lagrangian planes a 11 1. The space 2 1[0. 1] 1 has
four dierent ends: the ends in the rst 1-factor, which we call east and west innity
respectively, and the ends in the second 1-factor, which we call . Note that there
is an (1 1)-action on 2 1 [0. 1] 1, by translation in the two 1-factors. There are
projection maps

,
|
, :, and t, just as before.
Fix a split complex structure J = ,

,
|
on 2 1 [0. 1] 1.
Denition 5.13. By a bi-decorated source 1

we mean:
5.3. HOLOMORPHIC CURVES IN Z 1 [0, 1] 1 67
(1) A topological type of smooth (not nodal) Riemann surface 1 with boundary and
punctures on the boundary.
(2) A labeling of each puncture of 1 by either c or u (east or west), and
(3) A labeling of each puncture of 1 by a Reeb chord in (2. a).
Given a bi-decorated source 1

, we will consider maps

: (1. 1) ((2 .) 1 [0. 1] 1. a 1 1 1)
satisfying the following conditions:
(1) The map is (,. J)-holomorphic with respect to some almost complex structure ,
on o.
(2) The map is proper.
(3) The map : maps 1 to 1.
(4) At each west puncture of o labeled by a Reeb chord , lim
zq

n(.) is
2 .
(5) At each east puncture of o labeled by a Reeb chord , lim
zq

n(.) is
2 +.
Note that each component of such a holomorphic curve necessarily maps to a single point
under
|
by the maximum principal, since its boundary maps entirely to : = 1.
Again, we collect these holomorphic curves into moduli spaces:
Denition 5.14. For 1

a bi-decorated source let

A(1

) denote the moduli space of

holomorphic maps from 1 satisfying properties (1)(5) above.
A holomorphic map is called stable if there are no innitesimal automorphisms of . (If

is non-constant on each component of 1, one can verify that 1 is stable if and only if
it has a component which is not a twice-punctured disk.) We shall be interested in

A(1

)
only in the cases that the corresponding maps are stable.
Suppose that is a puncture of 1

. Then there is an evaluation map

c
q
:

A(1

) 1 given by ev
q
() = lim
zq
t (). (Here, writing lim is somewhat
silly, since t is constant on each connected component of 1

.) There are, consequently,

evaluation maps
c
w
:=

qW(S

)
ev
q
:

A(1

) 1
W(T

)
and
c
e
:=

qE(S

)
ev
q
:

A(1

) 1
E(T

)
where \(1

) or \ is the set of west punctures of 1

and 1(1

) or 1 is the set of east

punctures.
We dene sub-moduli spaces of

A(1

) by cutting-down by partial diagonals.

Denition 5.15. Given a bi-decorated source 1

and partitions 1
w
and 1
e
of the west and
east punctures of 1

, respectively, let

A(1

; 1
w
. 1
e
) = (ev
w
ev
e
)
1
(
Pw

Pe
).
If 1
w
is the discrete partition, we denote

A(1

; 1
w
. 1
e
) by

A(1

; 1
e
).
68 5. MODULI SPACES
The moduli spaces

A(1

) can be understood concretely. Indeed, they are determined by

the t-coordinates of the components of 1, and a holomorphic map from 1

to , which gives

(which in turn is determined by local branching data over ). However, in general the
moduli spaces

A(1

) are not transversally cut out. One situation in which they areand
indeed this is the case which is relevant for our applicationsis the following:
Proposition 5.16. Suppose 1

is a decorated source such that all components of 1 are

topological disks. Then

A(1

) is transversally cut out by the equation for any split complex

structure on 2 1 [0. 1] 1.
Proof. This is proved by explicit computation of the cokernel of the

-operator, which is
identied with a certain sheaf cohomology group. See McDu and Salamon [MS04, Section
3.3] for a nice exposition of these ideas in the absolute case. (The relative case can be
deduced from the absolute case via a doubling argument; see, for instance, Hofer, Lizan, and
Sikorav [HLS97, Section 4].) More details can also be found in [Lip06b, Lemma 4.1.2].
The (11)-translation action on 21[0. 1] 1 induces an (11)-action on

A(1

)
for any 1

; if

A(1

) is stable then (almost by denition) this (11)-action is free. As for

the earlier moduli spaces, we are primarily interested in the quotient space:
Denition 5.17. Given a stable decorated source 1

, let A(1

) =

A(1

),(1 1).
Certain holomorphic curves in 2 1 [0. 1] 1 will play a special role below. By
a trivial component we mean a component of 1

which is a topological disk with exactly

two boundary punctures, one east and one west, and both labeled by the same Reeb chord.
Holomorphic maps from such components are uninteresting. In particular, they are preserved
by the 1-translation action on 2 1.
A more interesting kind of component is a join component. This is a component of 1

which is a topological disk with two west punctures and one east puncture. In counter-
clockwise order, suppose the punctures are labeled by (c.
e
), (u.
1
) and (u.
2
). (We will
sometimes refer to the (u.
1
) puncture as the top puncture of the join component, and
the (u.
2
) puncture as the bottom puncture.) Then there is a holomorphic map from this
component if and only if
e
=
2

1
. If such a holomorphic map exists, it is unique up to
translation. A curve consisting entirely of one join component and some number of trivial
components is called a join curve.
Symmetrically, a split component is a component of 1

which is a topological disk with

two punctures labeled c and one labeled u. In counterclockwise order, suppose the punctures
are labeled by (u.
w
), (c.
1
) and (c.
2
). (We will sometimes refer to the (c.
2
) puncture as
the top puncture of the split component, and the (c.
1
) puncture as the bottom puncture.)
Then there is a holomorphic map from such this component if and only if
w
=
1

2
. Again,
if such a holomorphic map exists, it is unique up to translation. A stable curve consisting
entirely of some number of split components and some number of trivial components is called
a split curve. (Note the denition is not symmetric with that of join curves.)
The sources of these components are illustrated in Figure 5.3. Split and join components
can arise from the degenerations in Figure 5.1.
Finally, there is one more complicated kind of curve which we must consider, a shue
curve. A shue component is a topological disk with two c punctures and two u punctures,
with cyclic ordering (c. u. c. u) around the boundary. (A shue component is pictured in
Figure 5.4, and an example where one degenerates o is shown in Figure 5.5.) The map
5.4. COMPACTIFICATIONS VIA HOLOMORPHIC COMBS 69
Figure 5.3. Sources of curves at east . Left: a trivial component.
Center: a join component. Right: a split component.
Figure 5.4. Visualizing shue curves. Left: a shue-component. Right:
a schematic of the image of an odd shue-component projected to 1.
from a shue component to 2 1 has either one interior branch point or two boundary
branch points. Suppose that in counterclockwise order, the punctures of a shue component
are labeled (c.
1
). (u.
2
). (c.
3
). (u.
4
). Then the moduli space is non-empty if and only if

+
1
=
+
2
,

2
=

3
,
+
3
=
+
4
and

4
=

1
. If a join curve admits a representative, exactly
one of the following cases occurs:

1
.
3
is nested and
2
.
4
is interleaved, or

2
.
4
is nested and
1
.
3
is interleaved.
(Recall the denition of interleaved and nested pairs of Reeb chords from Denition 3.9.) In
fact, we will see in Proposition 5.36 that the two cases are not symmetric for our purposes:
when the rst occurs in a degeneration, it occurs an odd number of times. When the second
occurs, it occurs an even number of times. We will call the rst case an odd shue component
and the second case an even shue component, to remember the distinction.
A holomorphic curve in 2 1[0. 1] 1 consisting of one shue component and some
number of trivial components is called a shue curve; a shue curve can be odd or even,
depending on the type of its shue component.
5.4. Compactications via holomorphic combs
Just as the compactications of moduli spaces in Morse theory involve not just ow
lines but broken ow lines, in order to discuss the compactications of the moduli spaces
/
B
(x. y; o

; 1) we need to introduce a more general object than holomorphic curves. The

idea is that, because of the non-compactness of the target space [0. 1] 1, holomorphic
curves can degenerate into holomorphic buildings with several stories, as described by
Eliashberg, Givental, and Hofer [EGH00]. In our setting, however, there are two dierent
kinds of innities: and c. This leads us to the notion of holomorphic combs.
70 5. MODULI SPACES
Figure 5.5. A domain in a Heegaard diagram where a shue compo-
nent degenerates o. The domain is 1 = 1
1
+1
2
+1
3
+21
4
+1
5

2
(x. y),
as indicated by the shading. There is a 1-parameter family of holomorphic
curves in /
B
(x. y; o

; 1) where 1 is the partition

1
.
3
. One end of this
moduli space consists of a two-story holomorphic building connecting x to w
via 1
2
and w to y via 1
1
+1
3
+21
4
+1
5
. At the other end, either an interior
branch point of

n (in region 1
4
) or two boundary branch points of

n
(between 1
3
and 1
4
or between 1
4
and 1
5
) approach , resulting in the
degeneration of a shue curve.
By a simple holomorphic comb we mean a pair of holomorphic maps (n. ) where n maps
to [0. 1] 1 and maps to 2 1 [0. 1] 1, and such that the asymptotics of n at
east match with the asymptotics of at west . More precisely:
Denition 5.18. A simple holomorphic comb is a pair (n. ) with n /
B
(x. y; o

) and
A(1

) for some sources o

and 1

, together with a one-to-one correspondence between

1(o

) and \(1

), preserving the labeling by Reeb chords, and such that ev(n) = ev

w
()
inside 1
E(S

)
,1

= 1
W(T

)
,1.
More generally, we also want to allow the components at east to degenerate further:
Denition 5.19. A holomorphic story is a sequence (n.
1
. . . . .
k
) (for some / 0) where,
for some 1, o

, and 1

i
,
n /
B
(x. y; o

),

i
A(1

i
),
(n.
1
) is a simple holomorphic comb (if / 1),
there is a correspondence between 1(1

i
) and \(1

i+1
) for i = 1. . . . . / 1 which
preserves the labelings by Reeb chords, and
ev
e
(
i
) = ev
w
(
i+1
) in 1
E(T

i
)
,1

= 1
W(T

i+1
)
,1.
We refer to (n.
1
. . . . .
k
) as the horizontal levels of the holomorphic story.
We also need to allow degeneration at :
Denition 5.20. A holomorphic comb of height ` is a sequence (n
j
.
j,1
. . . . .
j,k
j
), , =
1. . . . . `, of holomorphic stories with n
j
a stable curve in /
B
j
(x
j
. x
j+1
; o

j
) for some sequence
of generalized generators x
1
. . . . . x
N+1
. We admit the case ` = 0, the trivial holomorphic
comb from x
1
to x
1
, which corresponds to a trivial (unstable) holomorphic curve. We refer to
the index , as the vertical level, so the comb (n
j
.
j,1
. . . . .
j,k
j
), , = 1. . . . . ` has ` vertical
levels (or stories).
5.4. COMPACTIFICATIONS VIA HOLOMORPHIC COMBS 71
Figure 5.6. A schematic of a height 2 holomorphic comb.
See Figure 5.6 for a schematic illustration of a height 2 holomorphic comb.
We will call a holomorphic comb toothless if it has no components at c. The spine of
a holomorphic comb is the sub-comb of components mapped to [0. 1] 1.
Finally, to compactify our moduli spaces, we will also need to drop Condition (1) of
Denition 5.2, and allow our holomorphic curves to have nodal sources. If we want to
explicitly allow nodes, we will refer to such combs as nodal holomorphic combs; if we want
to explicitly rule out nodal combs, we will refer to smooth combs. When the distinction is
unimportant, we will simply talk about holomorphic combs.
Note that a (nodal) holomorphic comb l = (n
j
.
j,1
. . . . .
j,k
j
)
N
j=1
naturally represents
a homology class 1 in
2
(x
1
. x
N+1
), where x
1
and x
N+1
are the rst and last generators
involved. (Specically, 1 = 1
1
1
N
, where 1
j
is the domain of n
j
. The
i,j
are
irrelevant to this homology class.) Further, l is asymptotic to a well-dened set of Reeb
chords at (far) east : these are the asymptotics at east innity of the
j,k
j
.
Next, we turn to what it means for a sequence of holomorphic curves to converge to a
holomorphic comb or, more generally, for a sequence of holomorphic combs to converge to
another holomorphic comb. This is a simple adaptation of denitions for holomorphic curves
converging to holomorphic buildings in [EGH00]; the reader is referred to [Lip06b, Section
4.2] for some more discussion of this point.
Given a simple holomorphic comb (n. ) with o

and 1

the decorated sources of n

and respectively, there is a natural way to (pre)glue o

and 1

to form a decorated
source o

: the surface o 1 is obtained by identifying small neighborhoods of the east

punctures of o with neighborhoods of the corresponding west punctures of 1. (Note that
72 5. MODULI SPACES
this is a purely topological operation, not involving any dierential equations. We use the
term preglue to distinguish from gluing of holomorphic curves.) The c punctures of 1

and punctures of o

carry over to decorate o

. The pregluing of sources extends

in an obvious way to general smooth holomorphic combs: for any smooth holomorphic comb
there is a corresponding (preglued) smooth surface. The construction also extends to nodal
holomorphic combs: there is a canonical way to deform away the nodes in the source to
produce a smooth Riemann surface.
From these constructions we can produce a compactication of /
B
(x. y; o

).
Denition 5.21. We make the following denitions.
/
B
(x. y; o

) is the space of all (possibly nodal) holomorphic combs whose preglued

surfaces are o

, in the homology class 1, with asymptotics x at , y at +.

/
B
(x. y; o

) is the closure of /
B
(x. y; o

) in /
B
(x. y; o

).
For j. 1(o

), c
p,q
: /
B
(x. y. o

) [. ] is the extension of the map ev

p,q
(see Equation (5.12)).
/
B
(x. y; o

; 1) is the space of all holomorphic combs respecting the partition 1.

Formally, we set
/
B
(x. y; o

; 1) :=

P
i
P
p,qP
i
ev
1
p,q
(0).
/
B
(x. y; o

; 1) is the closure of /
B
(x. y; o

; 1) in /
B
(x. y; o

; 1).
/
B
(x. y; o

;

1) is the closure of /
B
(x. y; o

;

1) in /
B
(x. y; o

; 1).
The space /
B
(x. y; o

; 1) is often a proper subset of /

B
(x. y; o

; 1), as the next

example illustrates. (Because of diculties with transversality at east , /
B
(x. y; o

) may
also be a proper subset of /
B
(x. y; o

).)
Example 5.22. Consider the portion of a Heegaard diagram shown in the left of Figure 5.7,
and let o

be the decorated source shown on the right of Figure 5.7. (We number the
east punctures of o

for convenience in referring to them.) Let 1 denote the partition

1. 3. 2. 4. The space

/
B
(x. y; o

) is then homeomorphic to 1(0. ) 1(0. ),

via the map
(ev
1
. ev
1
ev
2
. ev
3
. ev
3
ev
4
).
The space

/
B
(x. y; o

; 1) is the subspace r
1
= r
3
, r
2
= r
4
of 1 (0. ) 1 (0. ).
Now, consider the compactications. In the compactication, when (ev
1
ev
2
) 0 the
curve degenerates a split component at east , and similarly when (ev
3
ev
4
) 0. If
(ev
1
ev
2
) 0 and (ev
3
ev
4
) 0, there are two split components at east . There is
a one-parameter family of these curves at east , given by the relative 1-coordinate of the
branch point under

. Consequently the part of the space /

B
(x. y; o

) with (ev
1
ev
2
)
and (ev
3
ev
4
) small has the form shown in Figure 5.8, and /
B
(x. y; o

; 1) is the sub-
space pictured. By contrast, /
B
(x. y; o

; 1) contains the entire stratum in which two split

components have degenerated.
Next, we turn to the technical results justifying the denitions of the moduli spaces
/
B
(x. y; o

; 1).
5.4. COMPACTIFICATIONS VIA HOLOMORPHIC COMBS 73
Figure 5.7. An example illustrating that /
B
(x. y; o

; 1) may be a
proper subset of /
B
(x. y; o

; 1). Left: a portion of a Heegaard dia-

gram. The domain of interest is shaded in gray. Right: the decorated source
o

.
Figure 5.8. An illustration of various moduli spaces for the Hee-
gaard diagram from Figure 5.7. The colored walls form the boundary
of /
B
(x. y; o

), the open part of which one visualizes as the empty space in

front of the walls. The subspace where ev
1
= ev
3
is the gray horizontal plane.
The moduli space /
B
(x. y; o

; 1) is a thick, squiggly line segment; the space

/
B
(x. y; o

; 1) is a T formed by that line segment and an arc in

/
B
(x. y; o

).
74 5. MODULI SPACES
Proposition 5.23. The spaces /
B
(x. y; o

) are compact. That is, suppose that l

n
is
a sequence of holomorphic combs in a xed homology class, with a xed preglued topological
source. Then l
n
has a subsequence which converges to a (possibly nodal) holomorphic comb
l, in the same homology class as the l
n
. Similarly, /
B
(x. y; o

; 1) and /
B
(x. y; o

;

1)
are compact.
Proof. Compactness for /
B
(x. y; o

) is proved in [Lip06b, Proposition 4.2.1]; for the

readers convenience, we give a simplied (and slightly corrected) version of the proof here.
The diculty is that standard compactness results from symplectic eld theory do not apply
directly: since both and 1 are non-compact, the end of [0. 1] 1 does not satisfy
the hypotheses of a stable Hamiltonian structure. Still, our argument is essentially soft: we
deduce compactness in our setting from results of Bourgeouis, Eliashberg, Hofer, Wysocki,
and Zehnder [BEH
+
03], roughly by considering one innity at a time, using Condition (J-4)
of admissible almost complex structures.
To ease notation, we will restrict our attention to the case of a sequence n
n
: o
n

[0. 1] 1

n=1
of holomorphic curves in /
B
(x. y; o

), rather than a sequence of arbitrary

holomorphic combs. It is routine to adapt the argument to the general case.
We start by extracting the vertical level structure on the limit. Let
e
denote the result
of lling the puncture j of , so
e
is a compact Riemann surface. By Condition (J-4) of
Denition 5.1, the almost complex structure J extends to an almost complex structure J
e
on
e
. Similarly, let o
n
e
be the result of lling in the c punctures of o
n
; then, n
n
extends
to a J
e
-holomorphic map n
n
e
: o
n
e

e
[0. 1] 1. It is convenient to think of the points
o
n
e
o
n
as marked points in o
n
e
.
The ends
e
[0. 1] of
e
[0. 1]1 are (Levi-at) stable Hamiltonian structures.
So, [BEH
+
03, Theorem 10.2] applies to give a convergent subsequence of the sequence n
n
e
.
(Actually, [BEH
+
03, Theorem 10.2] is stated in the absolute case, i.e., for curves without
boundary. See Section 10.3 of that paper for discussion of its generalization to the relative
case.) In the process, we add some new marked points to the surfaces o
n
e
, so that every
component of the limit source is stable. By replacing the original sequence n
n
with this
subsequence, we may assume n
n
e
converges. The limit of n
n
e
is a holomorphic building
n
a,i
: o
a,i

e
[0. 1] 1, where n
a,i
denotes the i
th
level of the limit curve. Using the
fact that
|
n
a,i
has degree p, it is easy to see that each n
a,i
is asymptotic to a generalized
generator at .
The limit curve n
a,i
may be constant on some components of o
a,i
. These are called
ghost components; compare Denition 5.40. In particular, there may be ghost components
mapped to points (j. :. t) (
e
) [0. 1] 1; this happens, for instance, in the second
and third cases of Figure 5.1. Note, however, that if o
0
is some component of o
a,i
so that

n
a,i
(o
0
) = j then
|
n
a,i
is constant on o
0
, since
|
n
a,i
(o
0
) 1 1.
Next, we extract the horizontal level structure on the limit. Fix a neighborhood l
p
of
the puncture j in so that the almost complex structure J is split, J = ,

,
|
, over the
closure l
p
of l
p
. Also, choose l
p
small enough that l
p
= . Then

n
n
restricted to
(

n
n
)
1
(l
p
) forms a sequence of holomorphic curves with Lagrangian boundary conditions
l
p
( l
p
). Consequently, by [BEH
+
03, Theorem 10.2], we can nd a convergent
subsequence. (Again, more marked points are added to stabilize the sources.) Again, call
the subsequence n
n
. Let

n
b,i
: o
b,i
l
p
.

b,i,j
: 1
b,i,j
1
5.5. GLUING RESULTS FOR HOLOMORPHIC COMBS 75
denote the limit. (We have not yet dened n
b,i
or
b,i,j
, just o
b,i
,

n
b,i
and

b,i,j
.) We
explain the indexing. Forgetting the new marked points and collapsing unstable components
gives a map c:
_
i
o
b,i
_

_
i,j
1
b,i,j
_


i
o
a,i
. This denes the index i on o
b,i
and 1
b,i,j
.
The index , comes from the story structure coming from the end of .
In fact, collapsing gives more: composing n
a,i
with c we get a map from
_
i
o
b,i
_

_
i,j
1
b,i,j
_
to
e
[0. 1] 1. Composing further with
|
:
e
[0. 1] 1 [0. 1] 1 gives
maps

|
n
b,i
: o
b,i
[0. 1] 1.
|

b,i,j
: 1
b,i,j
[0. 1] 1.
(The map
|

b,i,j
is constant on each connected component of 1
b,i,j
. Each component of
1
b,i,j
either corresponds to a ghost component in o
a,i
or is a curve which is unstable when
we delete the new marked points; this second case will correspond to trivial components at
east .)
Let n
b,i
= (

n
b,i
) (
|
n
b,i
) and
b,i,j
= (

n
b,i,j
) (
|
n
b,i,j
). Then the sequence
n
n
converges to the holomorphic comb n
b,i
.
b,i,j
, as desired.
This proves compactness of /
B
(x. y; o

). Finally, the other spaces, /

B
(x. y; o

),
/
B
(x. y; o

; 1) and /
B
(x. y; o

;

1), are closed subspaces of /
B
(x. y; o

) by denition,
and hence compact.
5.5. Gluing results for holomorphic combs
We next turn to gluing results, which show that the space /
B
(x. y; o

; 1) is well-behaved
near some of the simplest holomorphic combs.
Proposition 5.24. Let (n
1
. n
2
) be a two-story holomorphic building with n
1
/
B
1
(x. y;
o

1
; 1
1
) and n
2
/
B
2
(y. w; o

2
; 1
2
). Then for suciently small open neighborhoods l
1
of
n
1
and l
2
of n
2
, there is an open neighborhood of (n
1
. n
2
) in /
B
1
B
2
(x. w; o

1
o

2
; 1
1
1
2
)
which is homeomorphic to l
1
l
2
[0. 1).
Proposition 5.25. Let (n. ) be a simple holomorphic comb with n /
B
(x. y; o

) and
A(1

; 1
e
). Let : = [1(o

)[ = [\(1

)[. Assume that the moduli spaces /

B
(x. y; o

)
and A(1

; 1
e
) are transversally cut out at n and respectively, and that ev: /
B
(x. y;
o

) 1
m
,1 and ev
w
: A(1

; 1
e
) 1
m
,1 are transverse at (n. ). Then, for suciently
small open neighborhoods l
u
of n and l
v
of , there is an open neighborhood of (n. ) in
/
B
(x. y; o

; 1
e
) which is homeomorphic to (l
u

ev
l
v
) [0. 1).
Proof of Propositions 5.24 and 5.25. The results follow by standard arguments.
Indeed, since each puncture of a holomorphic curve in our setting is mapped either to
or to east or west , the fact that we have two kinds of innities is irrelevant to a local
statement of this kind. The only complication to the usual arguments, then, is that the
asymptotics at east and west innity are degenerate (i.e., Morse-Bott). Thus, the proof can
be obtained by adapting [Bou06, Section 5.3] to the relative case or [Lip06a, Proposition
A.1] to the Morse-Bott case.
We will also use the following generalization of Proposition 5.25 to non-simple one-story
holomorphic combs in the proof of Proposition 5.33.
Proposition 5.26. Consider a height 1 holomorphic comb (n.
1
. . . . .
k
) with n /
B
(x. y;
o

),
i
A(1

i
) for i = 1. . . . . / 1 and
k
A(1

k
; 1
e
). Let :
i
= [\(1

i
)[. Assume that
76 5. MODULI SPACES
all the moduli spaces are transversally cut out, and that the map
/
B
(x. y; o

) A(1

1
) A(1

k
; 1
e
) 1
m
1
1
m
1
1
m
2
1
m
2
1
m
k
1
m
k
(n.
1
. . . . .
k
) (ev(n). ev
w
(
1
). ev
e
(
1
). ev
w
(
2
). . . . . ev
e
(
k1
). ev
w
(
k
))
is transverse to the diagonal (r
1
. r
1
. . . . . r
k
. r
k
) [ r
i
1
k
i
. Then, for suciently small
open neighborhoods l
u
of n and l
v
i
of
i
, there is an open neighborhood of (n.
1
. . . . .
k
) in
/
B
(x. y; o

1
. . . 1

k
; 1
e
) which is homeomorphic to (l
u

ev
l
v
1

ev

ev
l
v
k
)[0. 1)
k
.
Again, the proof is essentially standard.
As mentioned at the end of Section 5.1, we cannot prove that the compactied moduli
spaces are manifolds with corners in general. We will get around this diculty by proving
that certain evaluation maps are degree one, in an appropriate sense, at the corners; this
will be enough for the results in Section 5.6. Before stating the propositions, we make some
denitions.
Denition 5.27. Let (n. ) be a simple holomorphic comb with A(1

). A smeared
neighborhood of (n. ) in /
B
(x. y; o

; 1) is an open neighborhood of
(n.
t
) [
t
A(1

). (n.
t
) /
B
(x. y; o

; 1) .
There is an exactly analogous notion of a smeared neighborhood of (n. ) in /
B
(x. y; o

;

1).
Remark 5.28. The notion of a smeared neighborhood of (n. ) depends only on the combi-
natorics of (the decorated source of) , not the map itself.
Denition 5.29. Given a continuous map ) : A 1 of topological spaces and a point
j 1 , we say ) is proper near j if there is an open neighborhood l j such that )[
f
1
(U)
:
)
1
(l) l is proper.
We will use the following weak notion of stratied spaces.
Denition 5.30. An :-dimensional stratied space is a topological space A written as a
union of strata A
i

n
i=0
where A
i
is a smooth i-dimensional manifold and the closure of A
k
is contained in

ik
A
i
. We typically suppress the strata from the notation. Let A and 1
be stratied spaces. A stratied map ) : A 1 is a continuous map with the property that
the preimage of any stratum in 1
i
1 is a union of connected components of strata A
j
of
A, and the restriction of ) to each stratum of A
j
, thought of as a map into 1
i
, is a smooth
map.
Denition 5.31. Let A be a stratied space so that the top stratum is a smooth :-manifold,
and let ) : A 1
m
+
be a stratied map so that )
1
((0. )
m
) is the top stratum of A. Let
j 1
m
+
, and assume ) is proper near j. We say that ) is odd degree near j if there is an
open neighborhood l of j such that for any regular value j
t
l (0. )
m
, )
1
(j
t
) consists
of an odd number of points.
Lemma 5.32. Let A be a stratied space such that the union of the top two strata of A
forms an :-manifold with boundary. Let ) : A 1
m
+
be a stratied map which is proper
near 0. Assume that )
1
((0. )
m
) is the top stratum of A. Then
(1) If j. j
t
(0. )
m
near 0 are regular values of ) then [)
1
(j)[ [)
1
(j
t
)[ (mod 2).
(2) If the restriction of ) to the preimage of some facet 1
m1
+
of 1
m
+
is odd degree near
0 1
m1
+
then ) is odd degree near 0 1
m
+
.
5.5. GLUING RESULTS FOR HOLOMORPHIC COMBS 77
Proof. Let l be a neighborhood of 0 over which ) is proper. Let \ be )
1
_
l 1
m
+
_
.
Then )[
V
: \ (l (0. )
m
) is a proper map. So part (1) follows from standard degree
theory.
For part (2), let j 1
m1
+
be a regular value of )[
V
, and let 1 1
m1
+
be a ball
neighborhood of j small enough that )[
f
1
(B)
is a covering map. Write )
1
(1) = 1
1
H H
1
2+1
; each 1
i
is an (:1)-ball. For each i choose an (:1)-ball 1
t
i
whose interior lies in
the top stratum of A and with 1
t
i
= 1
i
. Note that by our assumptions on ), )(Int(1
t
i
))
is entirely contained in (0. )
m
. So )(1
i
1
t
i
) is a (singular) (:1)-sphere in 1
m
+
, and for
j
t
(0. )
m
suciently close to j, the sphere )(1
i
1
t
i
) has winding number 1 around
j
t
. It follows that ) has odd degree at regular values near j. Choosing j close enough to 0
gives the result.
Proposition 5.33. Suppose that (n. ) is a simple holomorphic comb in /
B
(x. y; o

; 1).
Assume that is a split curve and that there are two parts 1
1
and 1
2
of 1 such that, for
each split component of 1

, its bottom puncture belongs to 1

1
and its top puncture belongs
to 1
2
. Assume also that ind(1. o

. 1) = 2 so dim/
B
(x. y; o

; 1) = 1.
Let
1
1
2
and
2
1
1
be the top and bottom punctures, respectively, on one of the
split components of 1

. Then, for generic J, there is a smeared neighborhood l of (n. ) in

/
B
(x. y; o

; 1) so that the evaluation map ev

q
1
,q
2
: l 1
+
is proper near 0 and odd degree
near 0.
Proof. To see the ev is proper near 0 on l, note that ev
q
1
,q
2
is clearly proper on the
closure l of l. By examining the topology of o

, we see that l l is mapped to strictly

positive values by ev
q
1
,q
2
.
Let the split components of 1

be 1

i
for i = 1. . . . . `. For notational convenience,
we will assume that 1

has no trivial strips, so that the only parts of 1 are 1

1
and 1
2
.
Let 1
t
be the partition 1 with 1
2
split into the discrete partition; that is, 1
t
has ` + 1
parts, one of which is 1
1
and all others consisting of a single puncture. We will show that
there is a smeared neighborhood l
t
of (n. ) in /
B
(x. y; o

; 1
t
) so that the evaluation map
ev: l
t
1
N
+
is odd degree near 0 in the sense of Denition 5.31. (We see that ev is proper
near 0 as before.) This suces, since for generic J, the map ev from /
B
(x. y; o

; 1
t
) is
transverse to the diagonal in (0. )
N
by Proposition 5.6.
By Proposition 5.26, we can make a smeared neighborhood l of (n. ) inside /
B
(x. y;
o

; 1
t
) into a stratied space of a particularly nice kind; in particular, the union of the top
two strata is an `-manifold with boundary. The codimension 1 strata of l (of dimension
` 1) consist of simple holomorphic combs (n
t
.
t
), where
t
is a split curve with no more
than ` split components. Higher codimension strata of l are more complicated one-story
holomorphic combs.
We call such a neighborhood in an `-dimensional compactied moduli space /
B
(x. y;
o

; 1
t
) a split neighborhood: namely, a split neighborhood is an open set containing the
entire stratum of simple combs (n. ), where is a split curve with ` split components, with
all bottom punctures of belonging to the same part of 1
t
and all top punctures belonging
to dierent parts.
We will show by induction on ` that ev on a split neighborhood has odd degree near 0.
For ` = 1, by Proposition 5.25 we have a map from [0. 1) to 1
+
mapping 0 to 0 which is
proper near 0; such a map automatically has odd degree near 0. For ` 1, note that the
preimage of the interior of a facet of 1
N
+
consists of holomorphic curves (n
t
.
t
), where
t
has
78 5. MODULI SPACES
a single split component; thus, n
t
lives in a split neighborhood l
t
of a smaller moduli space,
of dimension ` 1. By induction, ev on l
t
has odd degree near 0, and so by Lemma 5.32,
ev on l also has odd degree near 0.
In Chapter 9, we will also need the following generalization of Proposition 5.33. To state
it, we need a little more terminology. A generalized split curve is a union of holomorphic
disks in 2 1[0. 1] 1 each of which has exactly one u puncture. Each component of a
generalized split curve is a generalized split component.
Proposition 5.34. Let (n. ) be a simple holomorphic comb in /
B
(x. y; o

;

1), where is
a generalized split curve. Write

1 = (1
1
. . . . . 1
k
), label the components of as 1
1
. . . . . 1
m
and label the east punctures of 1
i
by
i,1
. . . . .
i,n
i
. Assume that there are integers 0 =
/
1
. /
2
. . . . . /
k
. /
k+1
= : such that 1
j
=

j
+1,1
. . . . .

j+1
,n

j+1
. (That is, 1
j
consists of the
punctures on components 1

j
+1
. . . . . 1

j+1
.)
Suppose that ind(1. o

. 1) = 1.
Then there is a smeared neighborhood l of (n. ) in /
B
(x. y; o

) so that the evaluation

map
ev: l 1
n
1
1
+
11
n
2
1
+
1 1
n

2
1
+
1
n

2
+1
1
+
1 1
n

3
1
+
1
nm1
+
dened by
ev(n) = (
1
n
1
1
+
..
ev
q
1,2
(n) ev
q
1,1
(n). . . . . ev
q
1,n
1
(n) ev
q
1,n
1
1
(n).
1
..
ev
q
2,1
(n) ev
q
1
,n
1
(n).
1
n
2
1
+
..
ev
q
2,2
(n) ev
q
2,1
(n). . . . . ev
q
2,n
2
(n) ev
q
2,n
2
1
(n).
1
..
ev
q
3,1
(n) ev
q
2
,n
2
(n). . . . )
is proper near 0 and odd degree near 0.
Proof. The proof is similar to the proof of Proposition 5.33. The moduli space of
generalized split curves has ends where one degenerates o a split curve at (far) east or
west innity; the lowest-dimensional stratum consists of sequences of (ordinary) split curves.
Using this, one can inductively reduce to the case of gluing on split curves. Details are left
to the reader.
Remark 5.35. It would be convenient to extend Proposition 5.6 to the compactied moduli
spaces, and in particular to assert that for height 1 holomorphic combs, say, the extension
of ev to /
B
(x. y; o

) is transverse to the partial diagonals

P
. In particular, this would
eliminate the need for the cumbersome Propositions 5.33, 5.34 and 5.36. Such a statement
would require a smooth structure on the /
B
(x. y; o

), which we have not constructed.

Moreover, the validity of the statement would depend strongly on the smooth structure: in
the language of Hofer [Hof06], it depends on the gluing prole. (In fact, for the exponential
gluing prole advocated there, the extension of ev would generally not be transverse to
P
.)
We conclude this section with a result on shue curves. It states that, when they occur,
odd shue curves appear in the boundary of /
B
(x. y; o

; 1) (with ind(1. o

. 1) = 2) an
odd number of times, and even shue curves appear an even number of times.
5.5. GLUING RESULTS FOR HOLOMORPHIC COMBS 79
Proposition 5.36. Fix a generic J. Let (n. ) /
B
(x. y; o

; 1) be a simple holomorphic
comb with a shue curve. Assume that the two c punctures
i
and
j
of the shue
component belong to the same part 1
i
of 1. Assume also that ind(1. o

. 1) = 2.
(1) If is an odd shue curve then there is a smeared neighborhood l of (n. ) in
/
B
(x. y; o

; 1) so that l /
B
(x. y; o

; 1) is homeomorphic to

2m1
i=1
[0. 1) (for
some : N).
(2) If is an even shue curve then there is a smeared neighborhood l of (n. ) in
/
B
(x. y; o

; 1) so that l /
B
(x. y; o

; 1) is homeomorphic to

2m2
i=1
[0. 1) (for
some : N).
Proof. First some notation: let c
1
. . . . . c
4
be the endpoints of the Reeb chords on the
shue component, ordered so that c
1
< < c
4
. Let
ij
(with i < , and i. , 1. . . . . 4)
denote the Reeb chord from c
i
to c
j
. Let 1

denote the source of

The two cases are similar; we rst prove case (1). In this case, the c punctures of
the shue component of 1

are labeled by
14
and
23
. The u punctures are labeled by

13
and
24
. The moduli space A(1

) has a natural compactication A(1

) which is
homeomorphic to [0. 1]. (The parameter is obtained by moving the branch point(s); see
Figure 5.9 for a schematic illustration.) By our transversality (Propositions 5.6 and 5.16)
and gluing (Proposition 5.25) results, any curve in A(1

) can be glued to n to obtain a

curve in /
B
(x. y; o

).
To see when there exist such curves in /
B
(x. y; o

; 1) we consider the ends of the moduli

space A(1

). The two ends correspond to holomorphic combs (n.

1
.
2
) and (n.
3
.
4
). The
curves
1
. . . . .
4
are as follows:

1
is a split curve with one split component, labeled by (u.
13
), (c.
12
) and (c.
23
),
together with a trivial component labeled by (c.
24
) and (u.
24
).

2
is a join curve, with a join component labeled by (u.
24
), (u.
12
) and (c.
14
),
together with a trivial component labeled by (c.
23
) and (u.
23
).

3
is a split curve with one split component, labeled by (u.
24
), (c.
23
) and (c.
34
),
together with a trivial component labeled by (c.
13
) and (u.
13
).

2
is a join curve, with a join component labeled by (u.
34
), (u.
13
) and (c.
14
),
together with a trivial component labeled by (c.
23
) and (u.
23
).
See Figure 5.9.
Suppose one glues
1
to n, with some gluing parameter c
1
. Then in the resulting curve
the Reeb chord
23
will be above
12
. If one then glues
2
to the result, with some gluing
parameter c
2
c
1
, in the resulting curve the Reeb chord
23
will be above
14
.
By contrast, suppose one glues
3
to n, with some gluing parameter c
1
. Then in the
resulting curve the Reeb chord
34
will be above
23
. If one then glues
2
to the result, with
some gluing parameter c
2
c
1
, in the resulting curve the Reeb chord
14
will be above
23
.
So, since the two Reeb chords switched order as we moved from one end of A(1

) to
the other, for given c there is algebraically one curve
t
A so that gluing n with
t
with
gluing parameter c gives a curve with
14
and
23
at the same heighti.e., an element of
/
B
(x. y; o

; 1). Part (1) of the proposition follows.

Part (2) follows by a similar analysis. In this case, the c punctures of the shue compo-
nent are labeled by
13
and
24
, and the u punctures are labeled by
14
and
23
. A schematic
illustration of A(1

) may be obtained by turning Figure 5.9 upside down. An analysis as

80 5. MODULI SPACES
Figure 5.9. The one-parameter family of maps from an odd shue
component. The dark dots denote interior branch points, and the dark semi-
dots denote boundary branch points. The two ends of the moduli space are
shown on the left and the right. At the bottom, the two ends are sketched
in the style of Figure 5.4. The corresponding pictures for an even shue
component are obtained by reection.
above shows that at both ends of A, the Reeb chord
24
is above
13
. This implies that,
algebraically, there are zero curves in /
B
(x. y; o

; 1) for a given gluing parameter.

5.6. Degenerations of holomorphic curves
To construct our invariants we study degenerations of 1-dimensional moduli spaces. The
compactness theorem, Proposition 5.23, permits two kinds of degenerations to occur in
/
B
(x. y; o

; 1):
The source o can degenerate to a point in the boundary of the moduli space of Rie-
mann surfaces. Such degenerations correspond to the conformal structure pinching
along some circles and/or arcs, resulting in a surface with some nodes
i
. There are
three possible behaviors of the limit near each
i
:
The 1-coordinate of the map may converge to at
i
, so
i
corresponds to
a level-splitting in the resulting multi-story holomorphic comb. This is a case
of splitting at .
The holomorphic map may converge to the puncture j in on one side of
i
, so
the puncture
i
corresponds to a splitting at east ; or the analogous behavior
5.6. DEGENERATIONS OF HOLOMORPHIC CURVES 81
for combs with several east levels may occur. We call this splitting at east
.
The holomorphic map may extend continuously over
i
, sending
i
to some
point in [0. 1] 1 (or perhaps 2 1 [0. 1] 1), resulting in a nodal
holomorphic comb. We call this becoming nodal.
The source o may not degenerate, but the map could become singular. In this case,
the derivative blows up at some point or points. (In the proof of Proposition 5.23,
we added marked points in this case, so that it became a special case of the previous
point.) There are three sub-cases.
The derivative may blow up at a puncture of o. If the puncture is a puncture
then this is the other case of splitting at , where the curve that splits o
has a non-stable source. If it were an c puncture then this would be a case of
splitting or east , but there are no non-trivial curves at c with unstable
sources, so this does not occur.
The derivative may blow up in the interior of o, which results in bubbling a
holomorphic sphere.
The derivative may blow up on the boundary of o, which results in bubbling a
holomorphic disk.
The goal of this section is to restrict which of these degenerations can occur in codimen-
sion 1. Ultimately, we will rule out bubbling disks and spheres and becoming nodal, and will
restrict strongly which curves can split o at and c. In Section 5.7.3, we show that
a few more degenerations can not occur for embedded curves. The remaining degenerations
will be incorporated in the algebra of our invariants, in Chapters 6 and 7.
5.6.1. First restrictions on codimension-one degenerations. Let
/
B
(x. y; o

) := /
B
(x. y; o

; 1) /
B
(x. y; o

; 1).
Some restrictions follow easily from the index formula and other simple considerations:
Proposition 5.37. Suppose that /
B
(x. y; o

; 1) is 1-dimensional. Then for generic J

every holomorphic comb in /
B
(x. y; o

; 1) has one of the following forms:

(1) a two-story holomorphic building (n
1
. n
2
);
(2) a simple holomorphic comb (n. ) where is a join curve;
(3) a simple holomorphic comb (n. ) where is a shue curve;
(4) a one-story holomorphic comb (n.
1
. . . . .
k
) such that each
i
is a split curve, and,
furthermore, the result 1

1
. . . 1

k
of pregluing the sources of
1
. . . . .
k
is also a
split curve; or
(5) a nodal holomorphic curve, obtained by degenerating some arcs with boundary on
o.
(Recall the denitions of join curves, split curves and shue curves from the end of
Section 5.3; in particular, a join curve or a shue curve has only one non-trivial component,
while a split curve may have arbitrarily many.)
Proof of Proposition 5.37. We begin with some general remarks. Since
2
() = 0,

2
(. ) = 0 and
2
(. ) = 0, no disks or spheres can bubble o. As usual, Deligne-
Mumford type degenerations, in which a closed circle or circles degenerate, are codimension
82 5. MODULI SPACES
Figure 5.10. A putative annulus at east . Any near enough curve
has a boundary component with no punctures, which violates the maximum
modulus principle.
2, and hence do not occur in these moduli spaces. By contrast, it is possible for an arc with
boundary on o to get pinched in codimension 1 (and become a node); this is case (5).
Next, suppose we have a simple holomorphic comb (n. ) in /
B
(x. y; o

; 1). Let o
t
and 1
t
denote the sources of n and respectively, and let : be [\(1
t
)[.
The curve induces a partition 1
t
of \(1
t
), by specifying that j and belong to the
same part if they have the same t-coordinate in . Then, n /
B
(x. y; o
t
; 1
t
) (using the
identication of 1(o
t
) and \(1
t
)).
Notice that [1
t
[ [1[. Also notice that (o
t
) = (o)(1
t
)+:. Since every component
of 1
t
must have a west puncture, (1
t
) :, with equality if and only 1
t
is a disjoint union
of topological disks with one west puncture per component.
Since, by hypothesis, the expected dimension of /
B
(x. y; o

; 1) is 1, by Formula (5.9)
we have
ind(1. o

. 1) = p (o) + 2c(1) +[1[ = 2.

Now,
ind(1. o
t
. 1
t
) = p (o
t
) + 2c(1) +[1
t
[
= p (o) + (1
t
) : + 2c(1) +[1
t
[
= 2 [1[ + (1
t
) : +[1
t
[.
Since [1
t
[ [1[ and (1
t
) :, we therefore have ind(1. o
t
. 1
t
) 2.
Suppose that ind(1. o
t
. 1
t
) = 2. Then (1
t
) = : and [1[ = [1
t
[. Consequently, 1
t
consists entirely of topological disks, each with a single puncture labeled u. It follows from
Propositions 5.16 and 5.25 that can be glued to n, and so ind(1. o

. 1) ind(1. o
t
. 1
t
),
contradicting the hypothesis.
Next, suppose that ind(1. o
t
. 1
t
) = 1. There are now two cases:
[1[ = [1
t
[ and (1
t
) = : 1. In this case, either 1
t
consists of : 1 topological
disks and one topological annulus, or just :1 topological disks. It is easy to see
that the former does not occur: the image in 2 1 of the annular component of
1
t
is a strip with a slit in the middle. (See Figure 5.10.) Any curve close enough to
this comb will still have this slit and so would be forced, by the maximum modulus
theorem, to have a component on which
|
n is constant.
5.6. DEGENERATIONS OF HOLOMORPHIC CURVES 83
By contrast, the case that 1
t
consists of :1 topological disks can occur. In this
case, all but one of these disks has one u puncture, and one has two u punctures.
We claim that none of the disks has two consecutive c punctures (not separated by
a u puncture). Indeed, any two consecutive c punctures will have dierent heights
in any curve with source o
t
1
t
. But this contradicts the fact that [1[ = [1
t
[.
It follows that :2 of the disks in 1
t
are trivial components and the remaining
disk is either a join component or a shue component.
[1
t
[ = [1[ 1 and (1
t
) = :. In this case, 1
t
consists of : topological disks,
each with a single west puncture. On rst glance, it appears each disk may have an
arbitrary number of east punctures. However, by Propositions 5.6, 5.16 and 5.25,
we can glue n to in /(x. y; o

) (though perhaps not in /(x. y; o

; 1)). It is then
easy to see that the east punctures on a single component of 1
t
will have dierent
t-coordinates in all of the glued curves. It follows that if any component of 1
t
has
more than two east punctures then [1[ [1
t
[ 1, a contradiction. Therefore 1
t
is a split curve, as in Case (4) of the statement.
If ind(1. o
t
. 1
t
) 0 then, by Proposition 5.6, the moduli space /
B
(x. y; o
t
; 1
t
) is
empty.
Next, consider a general one-story holomorphic comb (n.
1
. . . . .

). The arguments
above then imply that the bi-decorated source 1

1
1

obtained by pregluing the

sources of
1
. . . . .

is a join curve or a split curve.

Finally, consider a general holomorphic comb l in /(x. y; o

; 1). It follows from

Formula (5.9) that each story of l drops the expected dimension by 1, so there are at most
two stories. Further, if there are two stories, each story individually has index 1, so by the
arguments above cannot have any components at east .
The rst four kinds of degenerations will be involved in the denitions of the invariants,
in Chapters 6 and 7. So we now show that nodal degenerations can not occur in certain
parts of the moduli spaces. We give names to three dierent kinds of nodal degenerations:
Denition 5.38. A boundary degeneration is a nodal holomorphic comb with the property
that some irreducible component of the source has no + or punctures and is mapped via
a non-constant map to .
Denition 5.39. A boundary double point is a holomorphic comb whose source is a nodal
curve and has a node j on the boundary with the property that the projection to [0. 1] 1
is not constant near either preimage point j
1
or j
2
of j in the normalization of the nodal
curve.
Denition 5.40. A ghost component of a holomorphic comb n is a component o
0
of the
source o of n such that n[
S
0
is constant. If n has a ghost component then we say n is haunted.
Lemma 5.41. Every nodal degeneration results in a boundary degeneration, a curve with
boundary double point, or a haunted curve.
Proof. If a nodal curve has no ghost components and no boundary degenerations then

|
n is non-trivial on every component, and hence n has a boundary double point.
We rule out boundary degenerations immediately:
Lemma 5.42. Boundary degenerations do not exist.
84 5. MODULI SPACES
Proof. Consider an irreducible component o of the source with no + or punctures;
and consider its corresponding map to [0. 1] 1. Projecting further to 1, we obtain a
harmonic function on o which is bounded on the boundary of o. Thus, this map is constant.
It follows that the map n maps into a single ber . Since we assumed that the map
was non-trivial, its image gives a non-trivial relative homology class 1 H
2
(. ; Z) with
:
z
(1) = 0. Such a two-chain violates homological linear independence of the -curves which
our bordered diagrams are required to satisfy, Denition 4.3.
Remark 5.43. The non-existence of boundary degenerations is an artifact of our denition of
bordered diagramsin particular, the homological linear independence of the - (respectively
-) curves. In more general settings, one may wish to consider diagrams with an excess of
- or -curves. In those settings, boundary degenerations do play a role (and the denitions
of the invariants need to be enhanced to take them into account).
Remark 5.44. In the proof of Proposition 5.23 we observed that holomorphic curves at c
can be viewed as ghost components of a curve in
e
[0. 1] 1. Since Denition 5.40 is
about curves in [0. 1] 1, it does not include such ghosts.
5.6.2. More on strong boundary monotonicity. A key tool in eliminating more
nodal degenerations will be strong boundary monotonicity, to which we now turn. The
denition of strong boundary monotonicity in Section 5.2 is a condition on curves; but we
will show in Lemma 5.47 that the condition in fact depends only on the asymptotics of the
curve. We will also see, in Lemma 5.49, that the strong boundary monotonicity condition is
closed. We start with a lemma about the boundaries of our holomorphic curves.
Lemma 5.45. Let n /(x. y; o

; 1). Let c be an arc in o mapped by n to

i
1 1
for some i. Then t n[
a
: c 1 is either strictly monotone or constant.
Proof. The map
|
n is holomorphic and has image contained in [0. 1] 1, and its
restriction to the arc c is contained in 11. Thus, if t n[
a
is non-constant then it cannot
have any critical points on the arc.
Denition 5.46. Let s be a /-element multi-set of elements of [2/] (i.e., a formal linear
combination of elements of [2/] with coecients in N0, so that the sum of the coecients
is /), and = (
1
. . . . .
n
) be a sequence of non-empty multi-sets of Reeb chords. Dene
o(s. ) := (s `(
+
)) `(

)
where
+
:=

+
i
and the union and dierence are interpreted in terms of multi-sets. Let

[i,j]
be the subsequence (
i
. . . . .
j
) of . (If i , the subsequence is empty.) We say that
the pair (s. ) is strongly boundary monotone if the following three conditions are satised:
(SB-1) The multi-set s is actually a subset of [2/] (i.e., s has no repeated entries),
(SB-2) `(

i+1
) o(s.
[1,i]
) and
(SB-3) `(
+
i+1
) is disjoint from o(s.
[1,i]
) `(

i+1
).
This is equivalent to the following conditions:
(SB-1
t
) For each i = 0. . . . . :, the multi-set o(s.
[1,i]
) is a /-element subset of [2/] with no
repeated elements.
(SB-2
t
) For each i, `(

i
) and `(
+
i
) (as multi-sets) have no elements with multiplicity
bigger than 1.
5.6. DEGENERATIONS OF HOLOMORPHIC CURVES 85
We also extend the denition to pairs (x. ), where x is a generalized generator, which we
call strongly boundary monotone if (o(x). ) is strongly boundary monotone.
Lemma 5.47. If (x. ) is strongly boundary monotone and n /
B
(x. y; o

;

1) with [

1] =
, then n is strongly boundary monotone. Conversely, if /
B
(x. y; o

;

1) contains a strongly
boundary monotone holomorphic curve n then (x. ) is strongly boundary monotone.
Proof. The fact that (x. ) being strongly boundary monotone implies n is strongly
boundary monotone is immediate from Lemma 5.45.
To see that n: o [0. 1] 1 strongly boundary monotone implies (x. ) strongly
boundary monotone, we also use the fact that there are no boundary degenerations, as
follows. Condition (SB-1) is immediate from the fact that x is a generator. To prove
Condition (SB-2), suppose that at some stage `(

i+1
) , o(s.
[1,i]
), so there is some chord

j

i+1
with

j
, o(s.
[1,i]
). Then there must also be a chord

i+1
with `(
+

) =
`(

j
), and an arc on o so that n is asymptotic to

at one end of ,
j
at the other
end of , and t n[
A
is constant. But by the open mapping principle applied to
|
n,
this implies that the component of o containing is a boundary degeneration, violating
Lemma 5.42. A similar argument implies Condition (SB-3).
We will sometimes say that a moduli space /
B
(x. y; o

;

1) is strongly boundary mono-
tone if (x. [

1]) is strongly boundary monotone; this is justied by Lemma 5.47.

We next show that strong boundary monotonicity is a closed condition on the compact-
ied moduli space. To make sense of this, we need to dene strong boundary monotonicity
for holomorphic combs. The following, somewhat weak denition will be sucient for our
purposes:
Denition 5.48. A holomorphic comb l is strongly boundary monotone if, after deleting
any ghost components, every story of the spine of l is strongly boundary monotone.
Lemma 5.49. The strong boundary monotonicity condition is closed. That is, if l
j
is a
convergent sequence of strongly boundary monotone holomorphic combs then the limit l of
l
j
is strongly boundary monotone.
Proof. This follows from the monotonicity of t n on arcs (Lemma 5.45) and the non-
existence of boundary degenerations (Lemma 5.42).
5.6.3. No nodal curves. In this section, we rule out boundary double points and
haunted curves for moduli spaces satisfying the strong boundary monotonicity condition.
We start with boundary double points.
Lemma 5.50. Suppose that (x. ) is strongly boundary monotone. Then the moduli spaces
/
B
(x. y; o

;

1) with [

1] = do not contain any curves with boundary double points.

Proof. Suppose that n
j
/
B
(x. y; o

;

1) converges to a holomorphic comb n with
source o containing a boundary double point. We have a node j in the source o whose
pre-images j
1
and j
2
in the normalization

o of o both get mapped non-trivially to [0. 1] 1.
Since j is on the boundary of o, it follows that j is mapped to some - or -curve. Now,
since the projection t n is a non-trivial harmonic function near both j
1
and j
2
, by the open
mapping principle, we can nd disjoint neighborhoods `
i
of j
i

o (for i = 1. 2), with the

property that both t n(`
1
) and t n(`
2
) are the same open interval 1 around t n(j). It
follows that, for large enough ,, n
j
is not strongly boundary monotone.
86 5. MODULI SPACES
Next we show that ghosts are vanishingly rare.
Lemma 5.51. Suppose that (x. [

1]) is strongly boundary monotone and ind(1. o

.

1) 2.
Then /
B
(x. y; o

.

1) does not contain any haunted curves.
Proof. This follows from Lemma 5.50 and a standard argument, a version of which we
give. Suppose that arcs and circles
1
. . . . .
n
in o collapse to form some ghost components.
Let o
1
. . . . . o
m
be the ghost components of the limit and o
0
the union of the other components
of the limit, o

o
decorated source corresponding to o
0
and n: o
0
o
m
[0. 1] 1
the limiting holomorphic curve. The arcs
i
come in two types: arcs which separate a
ghost-component from a non-ghost component, which we will call important arcs; and arcs
separating two ghost components, a ghost component from itself, two non-ghost components,
or a non-ghost component from itself.
Let G be the ghost graph with one vertex for each ghost component and one edge for
each node connecting two ghost components. Let G
1
. . . . . G
p
be the connected components
of G and 1
i
, i = 1. . . . . j, the union of the ghost components corresponding to G
i
. We
claim that each 1
i
has exactly one node corresponding to an important arc. Certainly 1
i
has at least one node corresponding to an important arc, and if some 1
i
has more than
one node corresponding to an important arc then n[
S
0
would have a double point, violating
Lemma 5.50. This argument also implies that smoothing the nodes in 1
i
gives a surface with
a single boundary component.
Each o
i
must be stable; in particular, no o
i
is a disk or sphere with 2 or fewer nodes. Since
exactly one node of 1
i
corresponds to an important arc, it follows that for each i = 1. . . . . j,
either G
i
is not a tree or some component of 1
i
is not a disk. Since the smoothing of 1
i
is a
surface with connected boundary, it follows that (o
0
) (o) + 2. Thus,
ind(1. o

0
.

1) = p (o
0
) + 2c(1) +[

1[
p (o) + 2c(1) +[

1[ 2
= ind(1. o

.

1) 2
0.
Hence, n[
S
0
belongs to a negative-dimensional moduli space, and so does not exist.
5.6.4. The codimension-one boundary of the moduli space of curves with a
given source. We have now ruled out bubbling and nodal degenerations, so the only ends
of the moduli spaces that remain are two-story buildings, collisions of levels and degen-
erating join, split and shue curves at c. At this point, we would like to assert that if
strong boundary monotonicity is satised then /
B
(x. y; o

;

1) is a compact 1-manifold with
boundary, and /
B
(x. y; o

;

1) consists of exactly those pieces. The diculty with this
statement is the case of shue curves and split curves with many split components: in these
cases, the transversality needed by Proposition 5.25 is absent. One would like to assert, for
instance, that simple combs (n
t
.
t
) in /
B
(x. y; o

;

1) /
B
(x. y; o

;

1) are isolated. The
maps n
t
are, indeed, isolated, but this is not obvious for the components
t
at east . (This
is related to the fact that the evaluation map from /
B
(x. y; o

;

1) collapses strata coming
from non-trivial moduli at east .)
5.6. DEGENERATIONS OF HOLOMORPHIC CURVES 87
Figure 5.11. A picture of the expected behavior of /(x. y; o

;

1)
/(x. y; o

) (left) and a behavior of /(x. y; o

;

1) /(x. y; o

) which
we have not ruled out (right). The reader might compare with Figure 5.8.
However, all we will need later is that the sum of the number of elements of the mod-
uli spaces occurring as the boundaries of /
B
(x. y; o

;

1) (when the latter space is one-
dimensional) is zero modulo two. This weaker statement is Theorem 5.55, below.
Before stating this theorem, we dene formally various moduli spaces that can appear
at the end of one-dimensional spaces, as illustrated in Figure 5.12, and place one more
restriction on collisions of levels.
Denition 5.52. Fix a one dimensional moduli space /
B
(x. y; o

;

1) (or / for short)
satisfying strong boundary monotonicity.
A two-story end of / is an element of /
B
1
(x. w; o

1
;

1
1
) /
B
2
(w. y; o

2
;

1
2
), where
w S(1), 1
1

2
(x. w) and 1
2

2
(w. y) are such that 1 = 1
1
1
2
, and o

= o

1
o

2
is a way of splitting o

in two which divides the ordered partition

1 as

1
1
<

1
2
.
A join curve end of / at level i is an element of /
B
(x. y; o
t
;

1
t
), where o
t
and

1
t
are obtained in the following way. Pick an east puncture of o

in the i
th
part of

1 and
a decomposition
q
=
a

b
. Then the decorated source o
t
is any source with a pair of
punctures c and / labeled by
a
and
b
, and such that o

is obtained from o
t
by pregluing
a join component to o
t
at the punctures c and /. The partition

1
t
is obtained from

1 by
replacing with c. / in the i
th
part.
An odd (respectively even) shue curve end of / at level i is an element of /
B
(x. y;
o
t
;

1
t
), where o
t
and

1
t
are obtained in the following way. Pick east punctures
1
and
2
of o

contained in the i
th
part of

1, and so that the corresponding Reeb chords are nested
(respectively interleaved). Then o
t
is any source together with punctures
t
1
and
t
2
such
that o

is obtained from o
t
by pregluing an odd (respectively even) shue component
1
t
to o
t
at the punctures
t
1
and
t
2
with
1
and
2
labeling the punctures of 1
t
. The
partition

1
t
is obtained from

1 by replacing
1
.
2
with
t
1
.
t
2
in the i
th
part.
A collision of levels i and i + 1 in / is an element of /
B
(x. y; o
t
; (1
1
. . . . . 1
i

1
i+1
. . . . . 1
n
)) with i satisfying 1 i < : and o
t
obtained from o

by contracting arcs
on o

connecting punctures labeled by abutting pairs of Reeb chords from 1

i
and 1
i+1
(and replacing those pairs of Reeb chords
i
.
j
by their joins
i

j
), and 1
i
1
i+1
denotes
the Reeb chords in o
t
coming from 1
i
and 1
i+1
. (The notions of abutting and are given
in Denition 3.10.)
88 5. MODULI SPACES
Figure 5.12. Schematics and examples of the four kinds of degener-
ations of Theorem 5.55. Far left: degenerating into a two-story building.
Center left: degenerating a join curve. Center right: degenerating a split curve.
Far right: degenerating a shue curve.
We extend this denition inductively to a collision of levels i. . . . . i+,: a collision of levels
i. . . . . i +, is a collision of levels i and i +1 in the result of a collision of levels i +1. . . . . i +,.
We will see (Lemma 5.54) that homological linear independence of the -curves often
prevents collisions of levels. First, a denition:
Denition 5.53. We say that two sets of Reeb chords
i
and
i+1
are weakly composable
if, for all
j

i
and
k

i+1
, if `(
+
j
) = `(

k
) then
+
j
=

k
. More generally, a
sequence = (
1
. . . . .
n
) is said to be weakly composable if for all i = 1. . . . . :1, the sets

1

i
and
i+1
are weakly composable.
Note that if two sets of Reeb chords are composable in the sense of Denition 3.11 then
they are weakly composable, but the converse is false.
Lemma 5.54. Suppose that /
B
(x. y; o

;

1) is strongly boundary monotone and /
B
(x. y;
o

;

1) contains a point corresponding to a collision of levels i and i +1 in

1. Then [1
i
] and
[1
i+1
] are weakly composable.
More generally, suppose that /
B
(x. y; o

;

1) contains a point corresponding to a col-
lision of levels i. i + 1. . . . . , in

1. Then ([1
i
]. [1
i+1
]. . . . . [1
j
]) is weakly composable.
Proof. Let n
j
be a sequence of curves in /
B
(x. y; o

;

1) whose spine converges to a
collision of levels. Fix chords [1
i
] and [1
i+1
] with the property that `(
+
) = `(

).
Our goal is to show that
+
=

.
Consider the arc on o along which n
j
leaves . The image

n
j
() is contained
in some curve
i
. Similarly, for the arc 1 on o

along which n
j
enters ,

n
j
(1) is
contained in the same curve
i
. By hypothesis, the sequences t(n
j
()) and t(n
j
()) converge
to the same point.
We must have = 1 by strong boundary monotonicity on n
j
. Therefore, by Lemma 5.45,
the restriction of tn to is constant, and hence n must be constant on the entire component
of o containing . Lemmas 5.42 and 5.51 imply that this component must be mapped to
east . It follows that the terminal point of coincides with the initial point of , as desired.
The more general case is similar, and is left to the reader.
5.6. DEGENERATIONS OF HOLOMORPHIC CURVES 89
We are now ready to state the theorem we will use to prove all the algebraic properties
of our modules.
Theorem 5.55. Suppose that (x. ) is strongly boundary monotone. Fix y, 1
2
(x. y),
o

, and

1 so that [

1] = and ind(1. o

. 1) = 2. Let / = /
B
(x. y; o

;

1). Then the
total number of
(ME-1) two-story ends of /,
(ME-2) join curve ends of /,
(ME-3) odd shue curve ends of / and
(ME-4) collision of levels i and i + 1 in /
is even. Moreover, in Case (ME-4) the parts 1
i
and 1
i+1
are weakly composable.
Proof. Let l
<
denote the open subset of /
B
(x. y; o

;

1) where two parts of

1 have
height dierence < c. Let l
shuf
denote the union of a smeared neighborhood of each shue
curve end in /
B
(x. y; o

;

1). It follows from the results above that
/
B
cropped
= /
B
(x. y; o

;

1) (l
<
l
shuf
)
is a compact 1-manifold with boundary (smoothness following from Proposition 5.6, com-
pactness from Proposition 5.23), the boundary of which consists of the following pieces:
Two-story holomorphic buildings (n
1
. n
2
). For c small enough, the number of these
correspond exactly to the number of two-story ends of /. (This is ensured by
Proposition 5.24. The fact that splittings must occur at generators, not just gener-
alized generators, follows from strong boundary monotonicity.)
Simple holomorphic combs (n. ) with a join curve. For c small enough the num-
ber of these boundary components is the number of join curve ends of /. (This
is an application of Proposition 5.25, with transversality hypotheses ensured by
Propositions 5.6 and 5.16)
An even number of points for each even shue curve end, and an odd number of
points for each odd shue curve end. (This is an application of Proposition 5.36.)
The subspace of /
B
(x. y;

1) where two parts dier in height by c. By Proposi-
tions 5.33 (if at least one split component degenerates) and 5.6 (if no split compo-
nents degenerate), for c small enough the number of boundary points of this form
agrees, modulo 2, with the number of collisions of two levels from o

. The fact that

the levels must be weakly composable follows from Lemma 5.54.
(In the rst case, we take c and l
shuf
small enough that no two-story buildings occur in
/
B
/
B
cropped
, and in the second case we take them small enough that no join curve
degenerations occur in /
B
/
B
cropped
. In the third case, c should be small enough that l
<
is disjoint from l
shuf
. In the fourth case, c should be small enough for Proposition 5.33 to
apply.)
We have accounted for all the boundary components, in view of Proposition 5.37 com-
bined with Lemma 5.50.
Remark 5.56. We will see in Lemma 5.70 that for moduli spaces of embedded curves, the
only collisions which occur are ones in which the colliding parts are composable (as dened
in Section 3.1.3) and not merely weakly composable, as well as similar restrictions on the
join curve and shue curve ends.
90 5. MODULI SPACES
5.7. More on expected dimensions
Recall from Proposition 5.8 that the expected dimension of /
B
(x. y; o

;

1) depends on
the topology of the source curve. Our aim here is to study an index formula which depends
on the source curve only through its homology class 1 and its asymptotics [

1]. We establish
two key properties of this index:
In cases where we have an embedded curve representing the homology class, this
formula agrees with our earlier index formula (Proposition 5.62 in Section 5.7.1).
The index is additive under juxtapositions (Proposition 5.68 in Section 5.7.2).
A fundamental consequence of the index formula is the following:
For collisions of levels occurring in the boundaries of two-dimensional, strongly
boundary monotone moduli spaces, the colliding levels are composable, rather than
just weakly composable (Lemma 5.70 in Section 5.7.3).
This last property is the justication for setting double crossings of strands to zero in the
algebra associated to a surface. There are also restrictions on join curve and odd shue
curve ends, also proved in Section 5.7.3.
5.7.1. The index at an embedded curve. Before stating the index formula, we
recall and extend some denitions from Section 3.3.1. For
1
.
2
H
1
(2
t
. a), recall that
1(
1
.
2
) = :(
2
.
1
) is the linking of
1
and
2
. Concretely, if the
i
are represented
by single Reeb chords
i
, we have
1([
1
]. [
2
]) =
_

_
1,2 if
+
1
=

2
1,2 if
+
2
=

1
0 if
1

2
= or
1

2
or
2

1
or
1
=
2
1 if

1
<

2
<
+
1
<
+
2
1 if

2
<

1
<
+
2
<
+
1
.
(The notation < is dened in Section 3.2.) Recall that for c /(:. /) with starting idempo-
tent o, the quantity (c) = inv(c) :([c]. o) is the Maslov component of the grading of c.
We can again express this more concretely for in terms of Reeb chords.
Lemma 5.57. For a set of Reeb chords so that c() ,= 0,
(c()) =

1
,
2

1([
1
]. [
2
])

[[
2
.
Proof. Recall from the proof of Proposition 3.40 that (c) is unchanged under adding
horizontal strands; thus we may pretend that the only strands in c() are those coming
from the Reeb chords in . Both terms in (c()), namely inv(c()) and :([].

), can
be written as sums over pairs of Reeb chords or single chords. A single chord contributes
1,2 to :([].

), while a straightforward case analysis shows that the contribution to the

sum from a pair of chords is 1 if they are interleaved, 1,2 if the endpoints abut, and 0
otherwise, as in the statement.
5.7. MORE ON EXPECTED DIMENSIONS 91
We will shorten notation and write 1(
1
.
2
) (respectively 1(
1
.
2
)) for 1([
1
]. [
2
]) (re-
spectively 1([
1
]. [
2
])) and () for (c()). Also, for a strongly boundary monotone se-
quence of sets of Reeb chords = (
1
. . . . .

), dene () by
(5.58) () :=

i
(
i
) +

i<j
1(
i
.
j
).
Lemma 5.59. The product of a sequence of elements in G
t
(:) is given by
(/
1
.
1
) (/
n
.
n
) =
_

i
/
i
+

i<j
1(
i
.
j
).

i
_
.
Proof. As in the proof of Proposition 3.37, expand the right-associated product on the
left hand side of the equation and use bilinearity of 1.
Lemma 5.60. For a strongly boundary monotone sequence of Reeb chords, () is the
Maslov component of gr
t
(c(
1
)) gr
t
(c(

)).
Proof. Clear from the denitions and Lemma 5.59.
For r and 1 a domain, dene :
x
(1) to be the average of the local multiplicities
of 1 in the four regions surrounding r, and for x S(1), dene :
x
(1) :=

xx
:
x
(1).
Denition 5.61. We say that a pair (1. ) with 1
2
(x. y) and a sequence of non-
empty sets of Reeb chords is compatible if the homology classes on the boundary agree,
[] =

1, and (x. ) is strongly boundary monotone. For such a pair, dene the embedded
Euler characteristic, embedded index, and embedded moduli space by

emb
(1. ) := p + c(1) :
x
(1) :
y
(1) ()
ind(1. ) := c(1) + :
x
(1) + :
y
(1) +[[ + ()
/
B
(x. y; ) :=
_
(S

)=
emb
(B,)
[

P]=
/
B
(x. y; o

;

1).
Dene /
B
(x. y; ) similarly.
Justication for this terminology is given by the following propositions.
Proposition 5.62. Suppose that /
B
(x. y; o

;

1) admits a holomorphic representative n.
Then
(5.63) (o) =
emb
(1. [

1])
if and only if n is embedded. When there is such an embedded, holomorphic n, the expected
dimension of /
B
(x. y; o

;

1) is given by
(5.64) ind(1. o

.

1) = ind(1. [

1]).
If /
B
(x. y; o

;

1) is strongly boundary monotone and admits a non-embedded holomorphic
representative then ind(1. o

.

1) ind(1. [

1]) 2.
Proof. We imitate the proof of [Lip06a, Proposition 4.2], to which the reader is referred
for a more leisurely account. We prove that if n is embedded, the Euler characteristic is as
stated; the other direction and Formula (5.64) are immediate consequences; we will return
to the inequality for non-embedded curves at the end of the proof.
92 5. MODULI SPACES
Fix an embedded holomorphic curve n /
B
(x. y; o

;

1). Let br(n) denote the rami-
cation number of

n. (Here, boundary branch points contribute 1,2.) By the Riemann-

Hurwitz formula,
(o) = c(o) + p,2 +

i
[1
i
[,2 = c(1) br(n) + p,2 +

i
[1
i
[,2.
So, to compute (o) we only need to compute br(n).
Let
R
(n) denote a copy of n translated by 1 units in the 1-direction. We will prove the
result by comparing n
R
(n) for 1 small and 1 large.
Assume now that the partition

1 is discrete, i.e., each part 1
i
consists of a single puncture
labeled by
i
; we will return to the general case at the end of the proof. Since n is embedded,
for small c the curves n and

(n) intersect only near where n is tangent to ,t, i.e., at

branch points of n. Since n and

(n) are J-holomorphic, their algebraic intersection number

is br(n). (Here and later, intersections along the boundary count for 1,2.)
On the other hand, for 1 large, from the vantage point of n,
R
(n) looks like p trivial strips
x[0. 1] 1; and from the vantage point of
R
(n), n looks like p trivial strips y [0. 1] 1.
Thus, the number of intersections of n and
R
(n) is n
R
(n) = :
x
(1) + :
y
(1) p,2; the
p,2 comes from the 2p corners. We will show that
(5.65) n

(n) n
R
(n) =

i<j
1(
i
.
j
).
and so
(o) = c(1) br(n) + p,2 +[

1[,2
= c(1) n

(n) + p,2 +[

1[,2
= c(1) n
R
(n) + p,2

i<j
1(
i
.
j
) +[

1[,2
= p + c(1) :
x
(1) :
y
(1) ([

1])
as desired.
It follows from a simple Schwartz reection argument that the intersection number n
r
(n)
can only change when an c puncture of
r
(n), mapped to some Reeb chord
1
, passes an
c puncture of n mapped to some Reeb chord
2
. Call this change 1
t
(
1
.
2
); that is,
1
t
(
1
.
2
) is the number of double points which disappear when
1
goes from below
2
to
above
2
. We claim that 1
t
(
1
.
2
) is universal, depending only on
1
and
2
, and not on n
itself. Equation (5.65) will then follow from a few model computations.
To see that 1
t
(
1
.
2
) is universal we employ a doubling argument, which also computes
1
t
(
1
.
2
). That is, we construct a new Heegaard diagram with boundary (
t
.
t
.
t
. .), which
we glue to along the boundary. We also construct holomorphic curves n
2
and n
1
which can
be glued to n and
r
(n) respectively, yielding curves nn
2
and
r
(n)n
1
, respectively. Since the
number of intersections between nn
2
and
r
(n)n
1
is topological, remaining constant under
deformations of nn
2
and
r
(n)n
1
, the number 1
t
(
1
.
2
) is equal to the number of double
points which appear when n
1
is slid up relative to n
2
.
In fact, it suces to construct
t
, n
1
, n
2
and so on locally near
1

2
. The constructions
of
t
, n
1
and n
2
can then be made to depend only on
1
and
2
, not on the curve n itself.
This implies that 1
t
(
1
.
2
) depends only on
1
and
2
.
5.7. MORE ON EXPECTED DIMENSIONS 93
Our model curves will have the domains shown in Figure 5.13. We distinguish 6 dierent
cases of how
1
and
2
can overlap.
(1)
+
1
=

2
. In this case, 1
t
(
1
.
2
) = 1,2. We choose the curves n
1
and n
2
to consist
of one strip each, as indicated in Figure 5.13 (upper left). When sliding n
1
up past
n
2
, one boundary double point appears.
(2)

1
=
+
2
. In this case, 1
t
(
1
.
2
) = 1,2. This case is the same as case (1), with
the roles of n
1
and n
2
exchanged.
(3)
1
is nested inside
2
or vice-versa. In this case, 1
t
= 0. Indeed, taking n
1
and n
2
to consist of one strip each, as indicated in Figure 5.13 (upper right), there are no
intersection points between n
1
and n
2
no matter how they are slid. (A degenerate
sub-case is when
1
=
2
.)
(4)
1
and
2
are disjoint. In this case, 1
t
(
1
.
2
) = 0. Taking n
1
and n
2
to each consist
of a single strip, as in Figure 5.13 (lower left) again gives no intersection points, no
matter how they are slid.
(5) (
1
.
2
) is interleaved, with

1
<

2
<
+
1
<
+
2
. In this case, 1
t
(
1
.
2
) = 1. Take n
1
and n
2
to consist of a single strip each, as in Figure 5.13 (lower right). Then if n
2
is
slid much higher than n
1
, there are no intersection points, while if n
2
is much lower
than n
1
, there is a single interior intersection point.
(6) (
2
.
1
) is interleaved, with

2
<

1
<
+
2
<
+
1
. In this case, 1
t
= 1. This is the
same as case the previous case, with the roles of n
1
and n
2
reversed.
This completes the proof in the case that

1 is a discrete partition.
In the general case, it is not true that br(n) is equal to the intersection number of n with

(n) for c small. That is, there may be intersection points of n and

(n) which run o to

east as c 0. This phenomenon is a simple variant of the change in number of double
points studied above, and it follows from the model computations used above that
n

(n) br(n) =

a,
b
[P
i
]
[1(
a
.
b
)[.
Otherwise, the argument proceeds as before. The result follows from Lemma 5.57.
Finally, if /
B
(x. y; o

;

1) admits a non-embedded holomorphic representative then
n

(n) br(n) = 2 sing(n) +

a,
b
[P
i
]
[1(
a
.
b
)[.
where sing(n) denotes the order of singularity of n (i.e., number of double-points in an
equivalent singularity; compare [MW95]). The rest of the argument goes through, giving
(o) = p + c(1) :
x
(1) :
y
(1) ([

1]) + 2 sing(n)
ind(n) = c(1) + :
x
(1) + :
y
(1) +[[ + () 2 sing(n)
= ind(1. ) 2 sing(n).
Strong boundary monotonicity rules out boundary double points in n (which contribute 1,2
to sing(n)), so sing(n) is a positive integer. Thus, ind(n) ind(1. ) 2, as desired.
5.7.2. The embedded index formula is additive. Since Proposition 5.62 assumed
the existence of a holomorphic curve, it is not obvious that ind(1. ) is additive in cases
when there is no holomorphic representative. To prove this, we will adapt Sarkars proof
94 5. MODULI SPACES
Figure 5.13. The domains of model curves used to compute inv
t
.
Top left: case (1). Case (2) is obtained by switching
1
and
2
. Top right:
case (3). Bottom left: case (4). Bottom right: case (5). Case (6) is ob-
tained by switching
1
and
2
. In the top two pictures, a schematic for
n
1
(o
1
). n
2
(o
2
) 1 1 is also shown.
of the same fact in the closed case [Sar10, Theorem 3.2] to our situation. We start with a
denition and some lemmas.
Suppose c is an oriented sub-arc of , with endpoints in , and / is an oriented
sub-arc of , with endpoints in . We dene the jittered intersection number c / as
follows. Let c
SW
, c
SE
, c
NW
, and c
NE
denote the four translates of c shown in Figure 5.14.
The endpoints of these translates are disjoint from , and the translates do not intersect
. The translates inherit orientations from the orientations of c and /. Consequently,
the intersection numbers c
SW
/. . . . . c
NE
/ are well dened. Dene
c / =
1
4
(c
SW
/ + c
SE
/ + c
NW
/ + c
NE
/) .
Note that instead of jittering c we could have jittered / with the same result. We extend the
denition of c / bilinearly to cellular 1-chains c contained in and / contained in . We
dene / c analogously. We also dene c c
t
for c and c
t
both contained in (or ) in exactly
the same way. (This denition is analogous to the denitions of :
x
(1) for Proposition 5.62
and :(. 1) in Section 3.3.1.)
The following lemma is due to Sarkar [Sar10, Section 3]. For the readers convenience
we repeat the proof here.
Lemma 5.66. The jittered intersection number c / has the following properties:
(1) c / = / c.
(2) If c and c
t
are both contained in then c c
t
= 0.
For the remaining items, let 1
2
(x. y), c =

(1), / =

(1), 1
t

2
(y. w), c
t
=

(1
t
)
and /
t
=

(1
t
).
(3) :
x
(1) :
y
(1) = c /.
(4) :
x
(1
t
) :
y
(1
t
) = c
t
/ and :
y
(1) :
w
(1) = c /
t
.
5.7. MORE ON EXPECTED DIMENSIONS 95
Figure 5.14. Jittered intersection number. Left: the four translates
c
SW
, c
SE
, c
NW
, and c
NE
of an arc c. Right: conventions for intersection
numbers.
(5) If 1 and 1
t
are provincial then c /
t
+ / c
t
= 0.
Proof. Properties (1) and (2) are obvious.
To understand Property (3), assume for simplicity that c and / are embedded arcs,
and consider the path /
SE
. The multiplicity of 1 at the starting point of /
SE
is :
y
SE
(1).
Each time /
SE
crosses c, the multiplicity of 1 increases by 1, for intersections contribut-
ing positively to c /, or decreases by 1, for intersections contributing negatively to c /.
Consequently, at the ending point of /
SE
, the multiplicity :
x
SE
is :
y
SE
+ c /
SE
and so
:
x
SE
(1) :
y
SE
(1) = c /
SE
. Averaging over the four possible directions gives the result.
Property (4) is proved similarly to Property (3), using /
t
instead of /. See Figure 5.15 for an
example of these two properties.
Finally, Property (5) follows from the fact that c + / and c
t
+ /
t
are null-homologous
1-chains, so (c + /) (c
t
+ /
t
) = 0. But by bilinearity and Property (2), this reduces to
c /
t
+ / c
t
= 0. (An alternate way to see Property (5), and see that the signs are write, is
to use x y = (

1) to compute :
x
(1
t
) :
y
(1
t
) = 1
t
((

1)) = (1
t
)

1, using an
extension of the jittered intersection number to 2-chains and 0-chains.)
We can extend Property (5) of Lemma 5.66 to non-provincial domains, as well.
Lemma 5.67. With notation as in Lemma 5.66, for arbitrary 1 and 1
t
in
2
(x. y) and

2
(y. w), respectively, we have c /
t
+ / c
t
= 1(

1.

1
t
).
Proof. We now have that c +/ +

1 and c
t
+/
t
+

1
t
are null-homologous 1-chains,
so their cap product in homology is 0. We can relate this to an extension of the jittered
intersection product by, for instance, treating as an extra -circle. Computing this
intersection number and dropping terms that always vanish gives:
(c + / +

1) (c
t
+ /
t
+

1
t
) = c /
t
+ / c
t
+ c

1
t
+

1 c
t
= 0.
Almost by denition, c

1
t
=
1
2
:(

1
t
. (

1)) and

1 c
t
=
1
2
:(

1. (

1
t
)). Both
of these terms are equal to
1
2
1(

1.

1
t
), as desired.
Proposition 5.68. The embedded index is additive in the following sense. Let 1
2
(x. y)
and 1
t

2
(y. w), and let and
t
be sequences of non-empty sets of Reeb chords with
[] =

1 and [

t
] =

1
t
. Then ind(1. ) + ind(1
t
.
t
) = ind(1 1
t
. (.
t
))
(Here, (.
t
) denotes the sequence obtained by simply concatenating and
t
.)
96 5. MODULI SPACES
x
Figure 5.15. Parts (3) and (4) of the proof of Lemma 5.66. Left:
part (3). The arrows show the orientations of c and / from 1. Notice that
c /
SE
= 1 and :
y
SE
(1) = :
x
SE
(1) (c /
SE
). Right: part (4). The
horizontal arrow indicates the orientation induced by 1
t
, the vertical arrow
the orientation induced by 1. Notice that c
t
/
NE
= 1 = :
x
NE
(1
t
)
:
y
NE
(1
t
).
Proof. It is clear that terms c(1) and [[ of Denition 5.61 are additive. Thus, we
must show that
:
x
(1) +:
y
(1) +:
y
(1
t
) +:
w
(1
t
) +() +(
t
) = :
x
(1+1
t
) +:
w
(1+1
t
) +((.
t
)).
On the other hand, we have
[:
x
(1
t
) :
y
(1
t
)] [:
y
(1) :
w
(1)] = c
t
/ c /
t
= c
t
/ + / c
t
1(

1.

1
t
)
= () + (
t
) ((.
t
)).
where the rst equality uses property (4) of Lemma 5.66, the second uses Lemma 5.67, and
the third uses property (1) and the denition of . This is equivalent to what we were trying
to show.
5.7.3. The codimension-one boundary of the moduli spaces of embedded
curves. We nish this chapter with two lemmas relating to which holomorphic curves occur
in the boundaries of moduli spaces of embedded curves. We start by showing that when
embedded holomorphic curves converge to holomorphic buildings, the stories are embedded.
Lemma 5.69. For a generic almost complex structure J and any pair (1. ) with ind(1. ) =
2, the two-story buildings which occur in the boundary of the embedded moduli space /
B
(x. y;
) (as in Theorem 5.55, say) are embedded. Equivalently, if (n
1
. n
2
) is such a two-story
building, with o

i
the source of n
i
, 1
i
the homology class of 1
i
and

1
i
the ordered partition
associated to n
i
, then:
(o

i
) =
emb
(1
i
.

1
i
)
ind(1
i
. 1
i
) = 1.
Proof. This follows from the inequality ind(n) ind(1. ) when n is non-embedded in
Proposition 5.62 and the fact that moduli spaces of negative expected dimension (ind 0)
are empty (Proposition 5.6).
5.7. MORE ON EXPECTED DIMENSIONS 97
The next lemma states that when two levels collide in a sequence of embedded curves, the
corresponding parts of a partition must be composable (as in Section 3.1.3), and not merely
weakly composable (Denition 5.53). Similar statements hold for splittings and shues.
Lemma 5.70. Let (x. ) satisfy the strong boundary monotonicity condition, and 1

2
(x. y) be compatible with . Suppose that ind(1. ) = 2. Suppose there is a holomor-
phic comb (n. ) in /
B
(x. y; ), and let
t
be the sequence of sets of Reeb chords appearing
as asymptotics of n. Then
(1) if two levels
i
and
i+1
collide in n,
i
and
i+1
are composable;
(2) if n is a join curve end at level i,
t
i
is a splitting of
i
; and
(3) if n is an odd shue curve end at level i,
t
i
is a shue of
i
.
As a preliminary, we rst state a characterization of composability, splittings, and shues
in terms of gradings. Recall that there is a partial order on G
t
(:) by comparing the Maslov
components: (/
1
.
1
) < (/
2
.
2
) if
1
=
2
and /
1
< /
2
.
Lemma 5.71. Let and be sets of Reeb chords such that c(), c() and c( ) are
non-zero. Then:
(1) The pair (. ) is composable if and only if gr
t
(c( )) = gr
t
(c()) gr
t
(c()).
(2) A weak splitting
t
of is a splitting if and only if gr
t
(c(
t
)) =
1
gr
t
(c()).
(3) A weak shue
t
of is a shue if and only if gr
t
(c(
t
)) =
1
gr
t
(c()).
In all cases, if equality does not hold then the left hand side has lower grading.
Proof. Clear.
Proof of Lemma 5.70. By Lemma 5.69, n is embedded.
For the rst statement, by Theorem 5.55 we know that
i
and
i+1
are weakly com-
posable. By Lemma 5.71, if
i
and
i+1
are weakly composable but not composable, then
gr
t
(c(
i

i+1
)) < gr
t
(c(
i
)) gr
t
(c(
i+1
)), so, by Lemma 5.60, (
t
) < (). Therefore
ind(1.
t
) < ind(1. ) +([
t
[ [[) = 1. Thus the embedded moduli space /
B
(x. y;
t
) has
negative expected dimension and is therefore empty.
Similarly, a join curve end at level i gives a weak splitting of
i
by denition. By
Lemma 5.71, for a weak splitting that is not a splitting we have gr
t
(c(
t
i
)) < gr
t
(c(
i
)) 1.
This implies that ind(1.
t
) < 1, so again /
B
(x. y;
t
) has negative expected dimension.
The case of odd shue curve ends is similar.
Remark 5.72. A similar analysis shows that even shue curve ends on the boundary of a
1-dimensional embedded moduli space all have negative expected dimension. We do not use
that, as we already showed in case (2) of Proposition 5.36 that such curves appear an even
number of times on the boundary of any moduli space.
CHAPTER 6
Type 1 modules
We now turn to dening the type 1 module

CFD of a bordered Heegaard diagram,
using the moduli spaces developed in Chapter 5. The denition, in Section 6.1, is fairly
straightforward. More work goes into proving that
2
= 0 (Section 6.2) and that the module
is independent of the choices made in its denition (Section 6.3). We also introduce a version
of

CFD with twisted coecients, in Section 6.4. We defer the construction of the grading
on

CFD to Chapter 10.
6.1. Denition of the type 1 module
Let 1 be a bordered Heegaard diagram (in the sense of Denition 4.3) which is provin-
cially admissible (in the sense of Denition 4.23) for a bordered three-manifold 1 , and let :
be the reverse of the pointed matched circle (in the sense of Denition 3.16) appearing on
the boundary of 1 = (. . . .). That is, : is (2. . `. .), where 2 = has the
opposite of the boundary orientation. Let /(:) be the associated algebra, as in Chapter 3.
We will abbreviate /(:) by / when it causes no confusion.
Fix a spin
c
structure s over 1 . The goal of this section is to dene a left dg module

CFD(1. s) over /.
Let S(1) be the set of generators of 1, as dened in Denition 4.13. That is, S(1)
consists of subsets of with
p elements in all,
exactly one element on each circle,
exactly one element on each circle, and
at most one element on each arc.
As in Section 4.3, let S(1. s) S(1) denote the subset of generators representing the
(absolute) spin
c
structure s. Let A(1. s) be the F
2
-vector space spanned by S(1. s) and
A(1) the F
2
-vector spanned by S(1), so that A(1) =

sspin
c
(Y )
A(1. s). Recall from
Section 4.3 that, for x S(1), o(x) [2/] denotes the set of -arcs occupied by x. Dene
1
D
(x) to be 1([2/] o(x)); that is, 1
D
(x) is the idempotent corresponding to the complement
of o(x). Now, dene an action on A(1. s) of the sub-algebra of idempotents J / via
(6.1) 1(s) x :=
_
x 1(s) = 1
D
(x)
0 otherwise.
(Here, s is a /-element subset of [2/], as in Formula (3.22).)
As an /-module,

CFD(1. s) is dened by

CFD(1. s) := /
7
A(1. s).
99
100 6. TYPE D MODULES
So, the module structure on

CFD(1. s) is quite simple; it is given by
(6.2) c (/ x) := (c/) x.
In particular, the summands /(:. i) of /(:) for i ,= 0 act trivially on

CFD(1. s).
By contrast, the dierential on

CFD(1. s) involves counting moduli spaces of holomor-
phic curves /
B
(x. y; ), where is a sequence (
1
. . . . .
n
) of Reeb chords. (Here we are
using both for a sequence of Reeb chords and for the corresponding sequence of one-element
sets of Reeb chords.)
Recall from Section 3.2 that c(
i
) denotes the element of / associated to
i
. For a
sequence of Reeb chords, let c( ) be the product c(
1
) c(
n
). (The reader is cautioned
that this typically does not coincide with the algebra element c() associated to a set of Reeb
chords =
1
. . . . .
n
.) Let be the sequence (
1
. . . . .
n
) of chords with reversed
orientation. (If is a Reeb chord in 1, is a Reeb chord in : = 1.)
Denition 6.3. Fix an almost-complex structure on [0. 1] 1 which is admissible
in the sense of Denition 5.1 and suciently generic in the sense of Denition 5.7. For
x. y S(1. s) and 1
2
(x. y), dene
c
B
x,y
:=

[ ind(B, )=1
#
_
/
B
(x. y; )
_
c( ).
Here there is also an implicit condition that (1. ) is compatible (i.e., that (x. ) is strongly
boundary monotone and

1 = [ ]); we will often omit such conditions from the notation.

Compactness of the moduli spaces /
B
(x. y; ) implies that the sum dening c
B
x,y
is nite.
Dene the dierential on

CFD(1. s) by
(I x) :=

yS(7,s)

B
2
(x,y)
c
B
x,y
y.
We will verify in Lemma 6.5 that if 1 is provincially admissible then the sums dening
are nite. Extend by linearity and the Leibniz rule to all of

CFD(1. s).
Let

CFD(1) :=

sspin
c
(Y )

CFD(1. s).
We will often abbreviate elements c x

CFD(1. s) writing, simply cx. In particular,
we blur the distinction between generators x S(1. s) and their associated elements Ix

CFD(1. s)
Remark 6.4. Consider the J-module A(1. s), equipped with the map
1
: A(1. s) /
7
A(1. s) given by

1
(x) := (I x).
This denes a type 1 structure over / with base ring J on A(1. s), in the sense of De-
nition 2.18. (The compatibility relation is equivalent to Proposition 6.7 below.)

CFD(1),
then, is the dierential module associated to this type 1 structure, as in Lemma 2.20. Fur-
ther, Theorem 6.16, below, and Lemma 2.21 imply that this type 1 structure is invariant
up to homotopy as a type 1 structure, not just as a module.
Lemma 6.5. If 1 is provincially admissible, then the sum in the denition of x on

CFD(1. s) is nite for every x S(1. s). If 1 is admissible, the map

1
(for the un-
derlying type 1 structure, as in Remark 6.4) is bounded in the sense of Denition 2.22.
6.2.
2
= 0 101
Proof. By Lemma 5.4, if /
B
(x. y; ) is non-empty then 1 has a positive domain. From
Proposition 4.28 we then see that if 1 is provincially admissible, then for a given x, y and
the union over 1 of 0-dimensional moduli spaces /
B
(x. y; ) is nite. There are only nitely
many possible generators y and algebra elements c /; furthermore, for any given c there
are only nitely many ways to write it as a product of Reeb chords, and so only nitely
many possible . The rst statement then follows.
If 1 is admissible, then by Proposition 4.29, there are only nitely many 1
2
(x. y)
with non-negative coecients, and hence only nitely many possible terms in any
k
(x).
More precisely, terms in
k
(x) count points in a product space

k
i=1
/
B
i
(x
i
. x
i+1
;
i
), where
we sum over all / + 1-tuples x
i

k+1
i=1
in S with x
1
= x and x
k+1
= y, /-tuples 1
i

2
(x
i
. x
i+1
)
k
i=1
and compatible sequences of Reeb chords
i

k
i=1
. Thus, for each non-zero

k
(x) we can nd such sequences x
i

k+1
i=1
and 1
i

2
(x
i
. x
i+1
)
k
i=1
with the property that
each 1
i
is positive. Adding the 1
i
, we obtain a corresponding 1
2
(x. x
k
) which is also
positive. Moreover, if we dene [1[ to be the sum of the coecients of 1 then [1[ /. Since
there are only nitely many options for 1, this gives an upper bound on /, as desired.
Remark 6.6. In the denition of the boundary operator, the requirement that y S(1. s)
is redundant: if c
B
x,y
,= 0, then 1
2
(x. y) is non-empty and hence, by Lemma 4.21,
s
z
(x) = s
z
(y). Thus, we could have dened

CFD(1) without reference to spin
c
structures,
and then noted that the complex splits as a direct sum over spin
c
(1 ).
6.2.
2
= 0
This section is devoted to proving that

CFD(1. s) is a dierential module, i.e.:
Proposition 6.7. The boundary operator on

CFD(1. s) satises
2
= 0.
The proof of Proposition 6.7 involves a number of pieces, some of which we give as
lemmas. The impatient reader is encouraged to skip to the examples after the proof, which
illustrate the most important cases which arise.
In our next two lemmas, we look at which ordered lists of Reeb chords actually con-
tribute to the boundary operator. The second lemma follows from the rst, which is more
general than our immediate needs, but will be useful in Chapter 9. The rst lemma is phrased
in terms of sequences of sets of Reeb chords (rather than merely sequences of Reeb chords).
In particular, if = (
1
. . . . .
n
) is a sequence of sets of Reeb chords, the notation means
the corresponding sequence of sets of Reeb chords where each Reeb chord has its orientation
reversed (but with the ordering unchanged). Similarly, for a sequence = (
1
. . . . .
n
) of
Reeb chords, denotes (
1
. . . . .
n
).
Lemma 6.8. Let s [2/] be a subset and = (
1
. . . . .
n
) be a sequence of non-empty
sets of Reeb chords. Let o(s.
[1,i]
) be as in Denition 5.46. Suppose that for i = 0. . . . . :,
`(

i+1
) o(s.
[1,i]
) and `((
i+1
)

) o([2/] s.
[1,i]
). (That is, (s. ) and ([2/]
s. ) satisfy Conditions (SB-1) and (SB-2) of the strong boundary monotonicity condition,
Denition 5.46.) Then:
(1) Both (s. ) and ([2/] s. ) are strongly boundary monotone.
(2) For i = 1. . . . . :, the sets o(s.
[1,i]
) and o([2/] s.
[1,i]
) are disjoint.
(3) For i = 1. . . . . :, `(

i
) and `(
+
i
) are disjoint.
102 6. TYPE D MODULES
Proof. We prove Conclusions (2) and (3) simultaneously by induction on i. The case
where i = 0 is the statement that s and [2/] s are disjoint. Assume the result for i. By
hypothesis, `((
i+1
)

) o([2/] s.
[1,i]
), which, by the inductive hypothesis, is disjoint
from o(s.
[1,i]
), which in turn contains `(

i+1
). Since reversing orientation switches the sets
of endpoints

i+1
and
+
i+1
, `(

i+1
) = `(
+
i+1
). Thus, we have veried Conclusion (3).
Thus, since `(
+
i+1
) = `(

i+1
) and `(
+
i+1
) are disjoint, it follows that o(s.
[1,i+1]
) and
o([2/] s.
[1,i+1]
) are disjoint, as well, verifying Conclusion (2).
Finally, we turn to Conclusion (1). To see that (s. ) is boundary monotone, it suces to
prove that `(
+
i+1
) is disjoint from o(s.
[1,i]
)`(

i+1
). But `(
+
i+1
) = `(

i+1
) o([2/]
s.
[1,i]
), which by Conclusion (2) is disjoint from o(s.
[1,i]
). By symmetry, ([2/] s. ) is
boundary monotone, as well.
Lemma 6.8 has the following consequence which is relevant for our present purposes:
Lemma 6.9. Suppose that (x. ) is strongly boundary monotone and 1
D
(x)c( ) ,= 0. Then
for each i, `(

i
) ,= `(
+
i
).
Proof. The condition that 1
D
(x)c( ) ,= 0 implies that (1
D
(x). ) = ([2/] o(x). )
is strongly boundary monotone. Since by hypothesis, (o(x). ) is strongly boundary mono-
tone as well, Lemma 6.8 applies, and gives the desired conclusion.
The next two lemmas identify how some of the terms in Theorem 5.55 relate to the
algebraic framework. Specically, a term in
2
x can come either from c
1
y appearing in x
and c
2
z appearing in y, contributing c
1
c
2
z to
2
x, or from cy appearing in x and c
t
appearing in c, contributing c
t
y to
2
x. Each type of term has its own interpretation in
terms of moduli spaces.
Lemma 6.10. For x. w. y S(1. s), 1
1

2
(x. w) and 1
2

2
(w. y), we have
c
B
1
x,w
c
B
2
w,y
=

1
,
2
[ ind(B
1
B
2
,(
1
,
2
))=2
#
_
/
B
1
(x. w;
1
) /
B
2
(w. y;
2
)
_
c((
1
.
2
)).
(Here, (
1
.
2
) denotes the sequence obtained by concatenating
1
and
2
.)
Proof. Clear.
Lemma 6.11. For x. y S(1. s) and 1
2
(x. y), we have
c
B
x,y
=

[ ind(B, )=2

i
,
i+1
abut

#/
B
(x. y; (
1
. . . . .
i

i+1
. . . . .
n
))c( ).
Proof. By denition,
c
B
x,y
=

[ ind(B, )=1
#
_
/
B
(x. y; )
_
c(
1
) c(
n
).
and so
c
B
x,y
=

[ ind(B, )=1

i
#
_
/
B
(x. y; )
_
c(
1
) ((c(
i
))) c(
n
)
=

[ ind(B, )=1

j
,
k
[
i
=
j

#
_
/
B
(x. y; )
_
c(
1
) c(
j
)c(
k
) c(
n
).
6.2.
2
= 0 103
To see the last step, note that if
j
and
k
abut (in that order), then
j
and
k
do not
abut; thus, for
i
=
j

k
,
(c(
i
)) =

j
,
k
[
i
=
j

c(
j
.
k
) =

j
,
k
[
i
=
j

c(
j
)c(
k
).
Now, from the denition of ind(1. ) (Denition 5.61),
ind(1. (
1
. . . . .
j
.
k
. . . . .
n
)) = ind(1. ) + 1 = 2.
(By contrast, ind(1. (
1
. . . . .
k
.
j
. . . . .
n
)) = ind(1. ).) The result follows by reindexing
the sum to run over
t
= (
1
. . . . .
j
.
k
. . . . .
n
).
Proof of Proposition 6.7. Roughly, the proof proceeds by considering the bound-
aries of the index 1 moduli spaces. More precisely, we appeal to Theorem 5.55.
Observe that

2
(cx) =
_
(c)x + c
_

w
c
x,w
w
__
= (
2
c)x + 2(c)(x) + c
__

w
(c
x,w
)w
_
+
_

y
c
x,w
c
w,y
y
__
.
where c
x,y
=

B
2
(x,y)
c
B
x,y
. So it suces to show that for all x and y,
(6.12) (c
x,y
) +

w
c
x,w
c
w,y
= 0.
(This is, of course, equivalent to the compatibility condition for the underlying type 1
structure, Denition 2.18.)
Fix generators x and y. Fix also 1
2
(x. y), and a non-zero algebra element c. Fix
an algebra element c and a sequence of Reeb chords so that ind(1. ) = 2. (The index
condition depends only on 1 and c, not the particular sequence ; see Lemma 5.60 and the
denition of ind(1. ).) By Theorem 5.55, applied to (x. y. 1. ) and the union, over all
sources o

with (o) =
emb
(1. ), of /
B
(x. y; o

;

1), the sum of the following terms is
equal to zero:
(1) The number of two-story ends, i.e., the number of elements of
/
B
1
(x. w;
1
) /
B
2
(w. y;
2
)
where 1 = 1
1
1
2
and = (
1
.
2
).
(2) The number of join curve ends, i.e., the number of elements of
/
B
(x. y; (
1
. . . . .
i1
.
j
.
k
.
i+1
. . . . .
n
))
where
i
=
j

k
.
(3) The number of split curve ends (necessarily with one split component), i.e., the
number of elements of
/
B
(x. y; (
1
. . . . .
i1
.
i

i+1
.
i+2
. . . . .
n
))
where
+
i
=

i+1
.
(4) The number of other collisions of levels, i.e., the number of elements of
/
B
(x. y; (
1
. . . . .
i1
.
i
.
i+1
.
i+2
. . . . .
n
))
where
+
i
,=

i+1
.
104 6. TYPE D MODULES
(Both the third and fourth contributions come from Case (ME-4) of Theorem 5.55. Shue
curve ends cannot happen when the partition is discrete.)
Next, we sum over all choices of vector with c = c( ). By Lemma 6.10, the sum of
terms of type (1) corresponds to the coecient in c of

w,B
1
,B
2
[ B
1
B
2
=B
c
B
1
x,w
c
B
2
w,y
, and by
Lemma 6.11 the sum of the terms of type (3) corresponds to the coecient of c in c
B
x,y
.
We will show that the remaining two types of terms cancel against each other.
For the terms of type (4), there are several cases that cannot contribute:
`(

i
) = `(

i+1
) or `(
+
i
) = `(
+
i+1
). This never occurs by Lemma 6.9 and
strong boundary monotonicity.
`(
+
i
) = `(

i+1
) and
+
i
,=

i+1
. This is ruled out by Lemma 5.54.
`(

i
) = `(
+
i+1
) and

i
,=
+
i+1
. In this case, c(
i
)c(
i+1
) = 0.
(
i
.
i+1
) are interleaved (in that order). This degeneration does not occur in codi-
mension one for embedded curves. More precisely, then the sets
i
and
i+1

are not composable by Denition 3.11, so by Lemma 5.70 this degeneration cannot
occur.
(
i+1
.
i
) are interleaved (in that order). Then c(
i
)c(
i+1
) = 0 as there are
double crossing strands.
The remaining possibilities are
(4a)

i
=
+
i+1
. These are exactly the moduli spaces which occur in Case (2), for the
factorization c(
1
) c((
i

i+1
)) c(
n
) of c. So, these two degenerations
cancel.
(4b) `(

i
). `(
+
i
) `(

i+1
). `(
+
i+1
) = , with
i
.
i+1
nested (in either order)
or disjoint. In this case, the same degeneration also occurs for the factorization
c(
1
) c(
i+1
)c(
i
) c(
n
) of c (also in Case (4b)). So, these two degen-
erations cancel.
It follows that the coecient of c in c
B
x,y
+

w,B
1
,B
2
[ B
1
B
2
=B
c
B
1
x,w
c
B
2
w,y
vanishes. Sum-
ming over 1 and noting that the choice of algebra element c was arbitrary, Equation (6.12)
follows.
The most interesting points in the proof of Proposition 6.7 are seen in the following
examples, illustrated in Figure 6.1; see also [LOT09]. We put the boundary of on the left
to indicate visually that the orientation of is reversed in the algebra.
Example 6.13. On the left of Figure 6.1 is a piece of a diagram with four generators. The
complex

CFD of this piece satises the relations
c. c =
_
1
2

c. d +
_
1
3

c. c
c. c = /. d
c. d =
_
2
3

/. d.
The fact that
2
= 0 follows from the relations in the algebra, more specically
_
1
2

_
2
3

=
_
1
3

.
This illustrates Case (1) in the proof of Proposition 6.7 (at the two two-story ends of the
moduli space), as well as the cancellation of Case (2) and Case (4a) (when the boundary
branch point goes out to ). The reader may also want to compare with Figure 5.1.
6.2.
2
= 0 105
Figure 6.1. Local illustrations of possible contributions to
2
in the
type 1 module. Left: a family of curves with one end a two-story building
and the other end a join curve end. Center: domain of a family of curves
with a two-story building at one end and a collision of two levels not resulting
in any split curves at the other. Right: a family of curves with a two-story
building at one end and a collision of two levels, degenerating a split curve, at
the other. (The thick lines indicate cuts; the boundary branch points at the
ends of the cuts are shown as black dots.)
Example 6.14. In the center of Figure 6.1, we have
c. c =
_
1 3
2

/. c +
_
3 1
4

c. d
/. c =
_
3 2
4

/. d
c. d =
_
1 4
2

/. d.
The fact that
2
= 0 follows from the fact that c
__
3
4
__
and c
__
1
2
__
commute in the algebra
or, more precisely, because
_
1 3
2

_
3 2
4

=
_
3 1
4

_
1 4
2

=
_
1 3
2 4

.
This is the self-cancellation of Case (4b) of the proof of Proposition 6.7.
Example 6.15. On the right of Figure 6.1, we have
c =
_
2 1
3

/ +
_
1 2
3

c
/ =
_
1 3
2

c.
Here,
2
= 0 because of a dierential in the algebra:

_
1 2
3

=
_
1 2
2 3

=
_
2 1
3

_
1 3
2

.
This illustrates Case (3) of the proof of Proposition 6.7, or equivalently Lemma 6.11.
The above examples are local: they are, of course, not the modules associated to ac-
tual bordered Heegaard diagrams. We refer the reader wishing to see global examples to
Section 11.2, where we compute the invariants of genus one handlebodies with certain bor-
derings.
106 6. TYPE D MODULES
6.3. Invariance
Theorem 6.16. Up to homotopy equivalence, the dierential module

CFD(1. s) is indepen-
dent of the choice of admissible, suciently generic almost complex structure and depends
on the provincially admissible Heegaard diagram 1 with : = 1 only through its induced
(equivalence class of ) bordered three-manifold (1. : 1(:) 1 ). That is, if 1 and 1
t
are
provincially admissible bordered Heegaard diagrams for (1. : 1(:) 1 ); J and J
t
are
admissible, suciently generic almost complex structures on [0. 1] 1 and
t
[0. 1] 1;
and s is a spin
c
structure on 1, then

CFD(1. s) (computed with respect to J) and

CFD(1
t
. s)
(computed with respect to J
t
) are homotopy equivalent dierential /(:)-modules.
The proof of Theorem 6.16 is modeled on the invariance proof for the closed case
[OSz04d] (as modied for the cylindrical case in [Lip06a], say). In particular, the proof
proceeds by showing that each of the three kinds of Heegaard moves from Proposition 4.10
induces a homotopy equivalence. (We also use Proposition 4.25 to ensure that all of the
intermediate diagrams are provincially admissible.)
There are two essentially new issues. The rst, and easier, is that we must keep track
of the algebra elements which occur as coecients in the maps. We will illustrate how this
is done by dening the chain map induced by deformation of the almost complex structure.
The second, more serious, issue is that the traditional proof of handleslide invariance uses
triangle maps, the general denitions of which are subtle for bordered Heegaard diagrams.
In the special case of handleslides the diculties which come up in triangle maps in general
can be circumvented, however. We illustrate this in a proof of invariance under handleslides
which carry one -arc over an -circle (the hardest case). These two issues having been
explained, the reader should have no diculty adapting the rest of the invariance proof to
our setting. For instance, the proof of isotopy invariance is similar to the proof of invariance
under change of complex structure, but notationally slightly more complicated.
6.3.1. The chain map for change of complex structure. Recall that the moduli
spaces /
B
(x. y; o

;

1) from Chapter 5 depended on a choice of a generic, admissible almost
complex structure J on [0. 1] 1, and consequently the dierential module

CFD(1. s)
also depends on this choice; to make this choice explicit, we will write

CFD(1. s; J) in this
section. As is usual in Floer homology theories, one proves independence of the almost
complex structure by constructing continuation maps, and proving that they are chain
homotopy equivalences; see, for example, [Flo89, Theorem 4]. More specically, let J
0
and
J
1
be two generic, admissible almost complex structures on [0. 1] 1, and let
0
and

1
denote the boundary operators on

CFD(1. s; J
0
) and

CFD(1. s; J
1
), respectively. Call a
(smooth) path J
r
[ : [0. 1] of almost complex structures from J
0
to J
1
(constant near
0 and 1) admissible if each J
r
is admissible. To a generic (in a sense to be made precise
presently) admissible path J
r
between J
0
and J
1
we will associate a continuation map
1
Jr
:

CFD(1. s; J
0
)

CFD(1. s; J
1
).
To the path J
r
we can associate a single almost complex structure J on [0. 1] 1 by
(6.17) J[
(x,s,t)
=
_

_
J
1
[
(x,s,t)
if t 1
J
t
[
(x,s,t)
if 0 t 1
J
0
[
(x,s,t)
if t 0.
6.3. INVARIANCE 107
(Here J[
(x,s,t)
means the map J induces on the tangent space at r , : [0. 1], and t 1.)
For 1
2
(x. y), let /
B
(x. y; o

; 1; J) denote the moduli spaces of holomorphic curves

dened in Denition 5.10 with respect to the almost complex structure J. (Note that here
there is no 1-action, since J is not 1-invariant.) For generic J, these moduli spaces are all
transversely cut out, as in Proposition 5.6. This is the genericity requirement we impose on
J
r
. For a sequence of non-empty sets of Reeb chords so that (x. ) is strongly boundary
monotone and 1 is compatible with , dene the embedded moduli space by
(6.18) /
B
(x. y; ; J) :=
_
(S

)=
emb
(B,)
/
B
(x. y; o

;

1; J).
Since we do not divide by 1-translation, we will primarily be interested in index 0 moduli
spaces, not the index 1 ones.
Then dene the map 1
Jr
on x S(1. s) by
(6.19) 1
Jr
(x) :=

yS(7,s)

B
2
(x,y)

[ ind(B, )=0
#
_
/
B
(x. y; ; J)
_
c( )y.
and extend 1
Jr
to all of

CFD(1. s; J
0
) by 1
Jr
(cx) = c1
Jr
(x). Compactness and provincial
admissibility imply that this sum is nite, as in Lemma 6.5.
It is immediate from Lemma 4.21 that 1
Jr
(x) respects the decomposition into spin
c
structures; see also Remark 6.6.
The next task is to show that 1
Jr
is a chain map. We will use the following analogue
of Theorem 5.55 for almost complex structures which are not translation invariant. The
statement is more general than we need for present purposes; we will also use it to prove
invariance of the type module in Section 7.3.
Proposition 6.20. Suppose that (x. ) satises the strong boundary monotonicity condition.
Fix y S(1), 1
2
(x. y), o

, and

1 such that [

1] = and ind(1. o

. 1) = 1. Consider
the parameterized moduli space /
B
(x. y; o

;

1; J). Then the total number of the following
ends of /
B
(x. y; o

;

1; J) is even:
(1) two-story ends, with either
(a) a J
0
-holomorphic curve followed by a J-holomorphic curve or
(b) a J-holomorphic curve followed by a J
1
-holomorphic curve;
(2) join curve ends;
(3) odd shue curve ends; and
(4) collision of levels 1
i
and 1
i+1
from

1, where [1
i
] and [1
i+1
] are weakly composable.
(The denition of two-story ends, join curve ends, odd or even shue curve ends, and col-
lisions of levels are given in Denition 5.52. As indicated, two-story ends come in two types:
elements of /
B
1
(x. w; o

1
;

1
1
; J
0
)/
B
2
(w. y; o

2
; J) and elements of /
B
1
(x. w; o

1
;

1
1
; J)
/
B
2
(w. y; o

2
; J
1
).)
Proof. The proof is essentially the same as the proof of Theorem 5.55.
Proposition 6.21. The map 1
Jr
is a chain map, i.e.,
1
1
Jr
+ 1
Jr

0
= 0.
Proof. The proof is essentially the same as the proof of Proposition 6.7. First, observe
that

1
(1
Jr
(cx)) + 1
Jr
(
0
(cx)) = (c)1
Jr
(x) + c
1
(1
Jr
(x)) + c1
Jr
(
0
(x)) + (c)1
Jr
(x).
108 6. TYPE D MODULES
So, it suces to prove that
1
(1
Jr
(x)) + 1
Jr
(
0
(x)) = 0 for any generator x S(1. s).
Now, as in the proof of Proposition 6.7, for given x, y, 1 and = (
1
. . . . .
n
) with
ind(1. ) = 1, Proposition 6.20 implies that the sum of the following terms is zero:
(1) The number of two-story buildings of the form
/
B
1
(x. w;
1
; J
0
) /
B
2
(w. y;
2
; J)
where 1 = 1
1
1
2
and = (
1
.
2
).
(2) The number of two-story buildings of the form
/
B
1
(x. w;
1
; J) /
B
2
(w. y;
2
; J
1
)
where 1 = 1
1
1
2
and = (
1
.
2
).
(3) The number of join curve ends, i.e., the number of elements of
/
B
(x. y; (
1
. . . . .
i1
.
j
.
k
.
i+1
. . . . .
n
); J)
where
i
=
j

k
.
(4) The number of split curve ends, i.e., the number of elements of
/
B
(x. y; (
1
. . . . .
i1
.
i

i+1
.
i+2
. . . . .
n
); J)
where
+
i
=

i+1
.
(5) The number of other collisions of levels, i.e., the number of elements of
/
B
(x. y; (
1
. . . . .
i1
.
i
.
i+1
.
i+2
. . . .
n
); J)
where
+
i
,=

i+1
.
The rst term corresponds to 1
Jt
(
0
(x)). The second corresponds to the terms in
1
(1
Jt
(x))
coming from applying
1
to the resulting generators w S(1. s). The fourth term corre-
sponds to the terms in
1
(1
Jt
(x)) coming from the dierential on the algebra /. The third
and fth terms cancel in pairs, as in the proof of Proposition 6.7.
One can similarly use a path of almost complex structures from J
1
to J
0
to dene a
chain map

CFD(1. s; J
1
)

CFD(1. s; J
0
). The proof that these maps are mutually inverse
chain homotopy equivalences is obtained by adapting the standard arguments in analogous
manners to the above, and we leave this to the interested reader. (Some more details are
given in the type case, in Proposition 7.19.)
6.3.2. Handlesliding an -arc over an -circle. Let (. . . .) denote a bordered
Heegaard diagram, which we assume for convenience to be admissible; this can be ar-
ranged by performing isotopies of the -curves. Write =
a
1
. . . . .
a
2k
.
c
1
. . . . .
c
gk
.
Provisionally, let
a,H
1
denote the result of performing a handleslide of
a
1
over
c
1
, and

H
=
a,H
1
.
a
2
. . . . .
c
gk
. Assume that the Heegaard diagram (.
H
. . .) is also admis-
sible. We want to show that

CFD(. . . .) and

CFD(.
H
. . .) are chain homotopy
equivalent.
6.3. INVARIANCE 109
Figure 6.2. The three-punctured disk (triangle) (left), and the
labeling inducing 1
,
H
,
(right). In counter-clockwise cyclic order, the
edges are labeled .
H
. .
6.3.2.1. Moduli spaces of triangles: basic denitions. As is traditional, we will produce a
chain map 1
,
H
,
:

CFD(.
H
. . .)

CFD(. . . .) by counting holomorphic triangles
in (. .
H
. . .). That is, let denote a disk with three boundary punctures. Label the
edges of by c
1
, c
2
and c
3
, counterclockwise, and let
ij
denote the puncture between c
i
and c
j
; see Figure 6.2. To dene 1
,
H
,
we will count holomorphic maps
n: (1. 1)
_
. (c
1
) (
H
c
2
) ( c
3
)
_
which do not cover ..
The
H
we use are actually a slight modication of the above, as shown in Figure 6.3.
Obtain
c,H
i
from
c
i
by performing a small Hamiltonian perturbation, so that
c,H
i
intersects

c
i
transversely in two points, is disjoint from all other -curves, and intersects
j
transver-
sally close to
j

c
i
. For i = 2. . . . . 2/,
a,H
i
is an isotopic translate of
a
i
, intersecting
a
i
transversely in a single point, and such that there are two short Reeb chords in running
from
a,H
i
to
a
i
(rather than the other way). The isotopy is chosen small enough that
a,H
i
is disjoint from all other -curves and intersects
j
transversally close to
j

a
i
. Obtain

a,H
1
by handlesliding
a
1
over
c
1
and then performing an isotopy so that
a,H
1
intersects
a
1
transversely in a single point, and such that there are two short Reeb chords in running
from
a,H
1
to
a
1
. We also arrange that
a,H
1
is disjoint from all other -curves and intersects
the
j
transversally at points close to
j
(
a
1

c
1
). Also dene a
H
to be the set
H
.
There is a natural bijection between a and a
H
.
Most of the notions from Chapter 5 generalize in straightforward manners to the setting
of triangles. Following Denition 5.1, we x an almost complex structure J on such
that:
(1) The projection map

: is (J. ,

)-holomorphic.
(2) The bers of the projection map

: are J-holomorphic.
(3) The almost complex structure is split near j . (Recall that j is the puncture
of ).
(4) For each puncture
i,j
of , choose a neighborhood l
i,j
which is biholomorphic to
[0. 1] [0. ), and such a biholomorphism. This identication induces an action of
1
+
on l
i,j
. We require that these actions be J-holomorphic.
110 6. TYPE D MODULES
Figure 6.3. The curves and
H
in . The -curves are solid while the

H
-curves are dashed. The intersection points which belong to generators
o
are marked. The annulus 1
a
1,1
+1
c
1,1
is shaded. The branch cut corresponding
to a possible holomorphic curve is also indicated by a thick arc.
(5) Identify with ([0. 1] 1) (1. 0) so that
12
is identied with (1. 0) (see Fig-
ure 6.4), and let
R
: ([0. 1] 1) (1. 0). (1. 1) ([0. 1] 1) (1. 1). (1. 0)
denote translation in the 1-factor. We require that the map
R
is J-holomorphic
for each 1.
We call such a J admissible. Notice that the restrictions of an admissible almost complex
structure J on to l
i,j
induce three admissible structures J
i,j
on [0. 1] 1.
Condition 5 can be thought of as a global version of Condition 4, and in particular implies
Condition 4 at the punctures
3,1
and
2,3
; together, Conditions 4 and 5 imply that J
1,2
is split.
Condition 5 is only needed for the computation of the embedded index in Proposition 6.25.
Similarly, we can dene decorated sources for maps to : a decorated source 1

is
an oriented surfaces with boundary and punctures, together with a labeling of each puncture
by an element of
1,2
.
2,3
.
3,1
. c and a further labeling of each c puncture either by a Reeb
chord with endpoints in a or by a Reeb chord with endpoints in a
H
.
Given a decorated source 1

we consider maps
n: (1. 1)
_
. (c
1
) (
H
c
2
) ( c
3
)
_
satisfying the obvious analogues of Conditions (M-1)(M-11) from Section 5.2. Such a holo-
morphic curve is asymptotic to a generator x for (. . . .) at
3,1
and a generator y for
(.
H
. . .) at
2,3
. The asymptotics at
1,2
, however, are more subtle. At each puncture
of 1

mapped to
1,2
, n is either asymptotic to a point in
H
(in the interior of ),
or is asymptotic to a Reeb chord in running from a
H
to a. In dening the map 1
,
H
,
we will use curves with only the former kind of asymptotics, but in proving that 1
,
H
,
is
6.3. INVARIANCE 111
a chain map we will need to consider curves with the latter asymptotics. Such asymptotics
present certain technical issues, as we will discuss below.
For each /-element subset o of 1. . . . . 2/ there is a distinguished p-element subset of

H
, denoted
o
. The set
o
contains the unique intersection point between
a
i
and

a,H
i
for i o, as well as the higher-graded intersection point between
c,H
i
and
c
i
for
each i = 1. . . . . p /. (The higher-graded intersection point r between
c,H
i
and
c
i
is the
one such that there are two holomorphic maps ([0. 1] 1. 1 1. 0 1) (.
H
. )
mapping to r.) We are thinking of
o
as a generator for counts of holomorphic curves
in (.
H
. . .), although we will not dene an object

CFD(.
H
. . .).
Like curves in [0. 1]1, curves in carry (relative) homology classes, in a manner
analogous to Denition 4.14. Let
2
(x. y.
o
) denote the set of homology classes connecting
x, y and
o
(not covering ., as before). For a homology class 1 in
2
(x. y.
o
) we can
consider its domain, the multiplicities with which

(1) covers each component of (

H
). The map from homology classes to domains is injective. Let /
B
(x. y.
o
; 1

)
denote the moduli space of holomorphic curves asymptotic to x at
3,1
, y at
2,3
and
o
at

1,2
, with decorated source 1

and in the homology class 1.

Along c
1
(respectively c
2
) a curve n /
B
(x. y.
o
; 1

) is asymptotic to a collection
of Reeb chords in (. a) (respectively (. a
H
)). The orientation on c
1
(respectively c
2
)
induces a partial order of these collections of Reeb chords. Thus, given ordered partitions

1
and

1
H
of Reeb chords in (. a) and (. a
H
) respectively, we may consider the subspace
/
B
(x. y.
o
; 1

;

1.

1
H
) /
B
(x. y.
o
; 1

)
consisting of those holomorphic curves so that the induced ordered partitions of the Reeb
chords along c
1
and c
2
are given by

1 and

1
H
respectively.
Lemma 6.22. If the moduli space /
B
(x. y.
o
; 1

;

1.

1
H
) is non-empty then, in the no-
tation of Denition 5.46, o = o(x. [

1]).
Proof. This is immediate from the denitions.
The arguments in Proposition 5.23essentially, considering the maps

n and

n
separatelyshow that the moduli spaces /
B
(x. y.
o
; 1

;

1.

1
H
) admit natural compact-
ications /
B
(x. y.
o
; 1

;

1.

1
H
). Other than degenerations at
12
, which we will discuss
presently, the limit curves are natural analogues of the holomorphic combs introduced in
Section 5.4. As for bigons in Chapter 5, we let /
B
(x. y.
o
; 1

;

1.

1
H
) denote the closure
of /
B
(x. y.
o
; 1

;

1.

1
H
) in /
B
(x. y.
o
; 1

;

1.

1
H
).
6.3.2.2. Homology classes and expected dimensions of triangles. The expected dimension
of /
B
(x. y.
o
; 1

;

1.

1
H
) is
(6.23) ind(1. 1

. 1. 1
H
) =
p
2
(1) + 2c(1) +[1[ +[1
H
[.
(The proof is similar to the proof of Proposition 5.8. See [Lip06a, p. 1018] for the closed
case.) Like Formula (5.9), Formula (6.23) depends on the topology of 1

, and not just

on the homology class and asymptotics. However, as was the case for embedded curves in
[0. 1] 1, the Euler characteristic of an embedded curve in is determined by the
homology class and asymptotics. To state the formula in a convenient form, rst note that
for any generator x S(. . ), there is a nearby generator x
t
S(.
H
. ) (but not vice
112 6. TYPE D MODULES
versa), as well as a canonical homology class 1
x
in
2
(x. x
t
.
o(x)
), the union of a number of
small triangles (contained in the isotopy region) and possibly an annulus with boundary on

a
1
,
a,H
1
and
c
1
.
Lemma 6.24. Any homology class 1
2
(x. y.
o
) can be written uniquely as
(1
x

12
1

H
,
)
23
1

H
,
for some 1

H
,

2
(
o(x)
.
o
) and 1

H
,

2
(x
t
. y).
Here
12
is the natural operation of attaching a homology class in (.
H
. . .) to a
homology class in (. .
H
. . .) at the corner
12
of , and likewise
23
is the operation
of attaching a homology class in (.
H
. . .) at the corner
23
of .
Proof. The space
2
(x. y.
o
) has a free and transitive action by the space of triply-
periodic domains, the space of 2-chains in which have local degree zero at . and whose
boundary is a formal linear combination of elements of
H
. The space of triply-
periodic domains in (. .
H
. . .) (and therefore
2
(x. y.
o
)) is isomorphic to the kernel
of the map
H
1
() H
1
(,) H
1
(
H
,
H
) H
1
(,).
Since H
1
(,) and H
1
(
H
,
H
) are the same subspace of H
1
(,), this kernel is iso-
morphic to
ker
_
H
1
() H
1
(,) H
1
(,)
_
ker
_
H
1
(,) H
1
(
H
,
H
) H
1
(,)
_

=
2
(x. x)
2
(
o
.
o
).
Thus, the action by
2
(x. x)
2
(
o
.
o
) on the set of triply-periodic domains is free and
transitive. The result follows.
For 1
2
(x. y.
o
), let /(1)
2
(x
t
. y) be the domain 1

H
,

2
(x
t
. y) whose
existence is guaranteed by Lemma 6.24.
Next, we prove an analogue of Sarkars index theorem for triangles [Sar10]. Given a
domain 1
2
(x. y.
o
), let

(1) (respectively

H
(1)) denote the part of 1 contained
in the -curves (respectively
H
-curves), oriented as 1. Let

H
(1)

(1) denote the

jittered intersection number of

H
(1) and

(1), as in Section 5.7.2. As for bigons in

Section 5.7, :
x
(1) denotes the point measure of 1 at x, and c(1) denotes the Euler measure
of 1. Also analogous to the bigon case, dene
(.
H
) := () + (
H
) + 1(.
H
)
=

i
[
i
[
2

i
[
H
i
[
2

a,
b

i
[1(
a
.
b
)[

a,
b

H
i
[1(
a
.
b
)[
+

i<j

i
,
j

j
1(
i
.
j
) +

i<j

H
i
,
j

H
j
1(
i
.
j
) +

i
,
j

H
j
1(
i
.
j
).
Proposition 6.25. If /
B
(x. y.
o
; 1

;

1.

1
H
) contains an embedded holomorphic curve n
then the expected dimension of /
B
near n is given by
(6.26) ind(n) = c(1) + :
x
(1) + :
y
(1) +

H
(1)

(1) p,2 + ([

1]. [

1
H
]).
6.3. INVARIANCE 113
Figure 6.4. A triangle is biholomorphic to a strip with a boundary
puncture. The edges are labeled by the corresponding arcs.
Proof. The proof is similar to the proof of Proposition 5.62. Identifying the triangle
with [0. 1] 1 (1. 0), so that
12
is identied with (1. 0), we can view n as a map to
[0. 1] 1. (See Figure 6.4.) Let o denote the source of n. Then
(o) = c(1) br(

n) +
3p
4
+

i
[1
i
[,2 +

i
[1
H
i
[,2.
where br denotes the order of branching and
3g
4
comes from the corners. Let
r
: [0. 1]
1 [0. 1] 1 denote translation in the 1 direction. Then for c suciently small,
br(

n) = n

(n)

a,
b
[P
i
]
[1(
a
.
b
)[

a,
b
[P
H
i
]
[1(
a
.
b
)[.
Translating farther, we see that for 1 suciently large,
(6.27) n

(n) = n
R
(n) +

H
(1)

(1) +
p
4
+

i<j

i
[P
i
],
j
[P
j
]
1(
i
.
j
)
+

i<j

i
[P
H
i
],
j
[P
H
j
]
1(
i
.
j
) +

i
[P
i
],
j
[P
H
j
]
1(
i
.
j
).
The 1(. ) terms in Equation (6.27) arise exactly as in the proof of Proposition 5.62; note
that this uses Condition (5), so that
r
(n) is holomorphic for any : 1. The term

H
(1)

(1) + p,4 comes from the number of intersection points that appear (or disappear) at
the boundary, as follows. For nitely many : (0. ), the
H
-boundary of n intersects
the -boundary of
r
(n). Generically, these intersections will be transverse, so it suces to
consider that case. We also assume that the complex structure on is chosen so that
and
H
meet at right angles. Let

n = n(o) and

H
n = n(o)
H
denote the subset
of n(o). Consider a point (

H
n) (

r
(n)), and let
1
= n
1
(),
2
= (
r
(n))
1
().
We can choose an identication of a neighborhood of with a neighborhood of (0. 1) in
C r + i [ r 1 and identications of neighborhoods of
1
and
2
with neighborhoods
of 0 in H so that, to rst order, on these neighborhoods the maps n and
r
(n) are given by
114 6. TYPE D MODULES
either
n(r + i) = (r + i. 1 + ir)

r
(n)(r
t
+ i
t
) = (
t
ir
t
. 1
t
+ ir
t
)
or
n(r + i) = (r + i. 1 + ir)

r
(n)(r
t
+ i
t
) = (
t
+ ir
t
. 1
t
+ ir
t
).
depending on whether corresponds to a positive or negative intersection in (

H
n)
(

r
(n)). (Again, we are using the fact that, by Condition (5),
r
(n) is holomorphic.)
Here, is a positive real number. Consider the rst case. The points in the domains of n
and
r+
(n) that map to the same point are, respectively,
r + i =

1 +
2
( + i)
r
t
+ i
t
=

1 +
2
( + i).
These points lie in the domains of n and
r+
(n) if 0, and not if < 0. So, for this local
model, one intersection point appears as we translate n up. Similarly, in the second case,
the points in the domain mapping to the intersection are
r + i =

1 +
2
( i)
r
t
+ i
t
=

1 +
2
( i).
These points lie in the domains of n and
r+
(n) if < 0, and not if 0. So, for this
local model, one intersection point disappears as we translate n up. This explains the term

H
(1)

(1), except for the part of

H
(1)

(1) coming from intersections at the

corners at
o
. For contributions coming from a corner of n intersecting an edge of
r
(n) or
vice-versa, the analysis is the same as before except that the intersection point that appears
or disappears is on the boundary. The expression

H
(1)

(1) also picks up 1,4 from

each of the p points in
o
, coming from the intersections of the corners of the domain with
themselves. These terms do not correspond to points appearing or disappearing; this is the
reason for the p,4 term in Equation (6.27).
As in the bigon case, n
R
(n) = :
x
(1) + :
y
(1) p,2. So,
(o) = c(1) br(

n) +
3p
4
+

i
[1
i
[
2
+

i
[1
H
i
[
2
= c(1) + :
x
(1) + :
y
(1) +

H
(1)

(1) p,2 + ([

1]. [

1
H
]).
(The last line uses Formula (6.23)). This proves the result.
For notational convenience, let
ind(1. .
H
) = c(1) + :
x
(1) + :
y
(1) +

H
(1)

(1) p,2 + (.
H
)

emb
(1. .
H
) = c(1) :
x
(1) :
y
(1) + p

H
(1)

(1) (.
H
).
6.3. INVARIANCE 115
Lemma 6.28. The function ind(1. .
H
) is additive with respect to gluing domains at
23
or
31
, in the following sense. If 1
2
(x. y.
o
), 1
,

2
(x
t
. x) and 1

H
,

2
(y. y
t
),
and (.
H
),
,
and

H
,
are sequences of non-empty sets of Reeb chords compatible with
1, 1
,
and 1

H
,
, respectively, then
ind(1
,

31
1
23
1

H
,
. (
,
. ). (
H
.

H
,
))
= ind(1
,
.
,
) + ind(1. .
H
) + ind(1

H
,
.

H
,
).
Proof. The proof is essentially the same as in the closed case [Sar10, Theorem 3.3]
(which also inspired the proof of Proposition 5.68).
Next we prove an analogue of Lemma 6.28 for gluing at
12
:
Lemma 6.29. Let 1
2
(x. y.
o
) and let (.
H
) be a pair of sequences of non-empty sets
of Reeb chords compatible with 1. Write = (
1
. . . . .
j
) and
H
= (
H
j+1
. . . . .
H
l
). Let
H
j
be the set of
H
-Reeb chords corresponding to
j
. Let 1
,
H
2
(
o
.
o
) (for some o
t
) be
a domain so that 1
12
1
,
H is compatible with ((
1
. . . . .
j1
). (
H
j
.
H
j+1
. . . . .
H
l
)). Then
(6.30) ind(1
12
1
,
H. (
1
. . . . .
j1
). (
H
j
.
H
j+1
. . . . .
H
l
)) = ind(1. .
H
).
Proof. The dierence between the two sides of Equation (6.30) is:
(6.31)
c(1
,
H) + :
x
(1
,
H) + :
y
(1
,
H)
+

H
(1)

(1
,
H) +

H
(1
,
H)

(1) +

H
(1
,
H)

(1
,
H)
+
j1

i=1
_
1(
i
.
H
j
) 1(
i
.
j
)
_
+
l

i=j+1
_
1(
H
j
.
H
i
) 1(
j
.
H
i
)
_
+ (
H
j
) (
j
).
By inspection, (
H
j
) = (
j
). Since

1
,
H =
H
j

j
, by multi-linearity of 1,
j1

i=1
_
1(
i
.
H
j
) 1(
i
.
j
)
_
= 1(
1
+ +
j1
.

1
,
H)
l

i=j+1
_
1(
H
j
.
H
i
) 1(
j
.
H
i
)
_
= 1(

1
,
H.
H
j+1
+ +
H
l
).
We next use the techniques of part (4) of Lemma 5.66 and Lemma 5.67. We have
o
x =
(

1 +
1
+ +
j
), so :
o
(1
,
H) :
x
(1
,
H) = 1
,
H (

1 +
1
+ +
j
), which
yields

H
(1
,
H)

(1) + 1(
1
+ +
j
.

1
,
H) + :
x
(1
,
H) = :
o
(1
,
H).
Similarly,

H
(1)

(1
,
H) + 1(

1
,
H.
H
j+1
+ +
H
l
) + :
y
(1
,
H) = :
o
(1
,
H).
So, Formula (6.31) reduces to
c(1
,
H) + 2:
o
(1
,
H) +

H
(1
,
H)

(1
,
H) + 1(
j
.

1
,
H).
116 6. TYPE D MODULES
Since [
H
j
] [
j
] =

1
,
H, we have
1(
j
.

1
,
H) = 1(
j
.
H
j
).
Similarly to part (3) of Lemma 5.66,
:
o
(1
,
H) +

H
(1
,
H)

(1
,
H) + 1(
j
.
H
j
) = :

(1
,
H).
By inspection of the possible domains for 1
,
H,
:
o
(1
,
H) + :

(1
,
H) + c(1
,
H) = 0.
This proves the result.
Corollary 6.32. If /
B
(x. y.
o
; 1

;

1.

1
H
) contains an embedded curve n then the expected
dimension of /
B
near n is given by
ind(n) = ind(/(1). ).
where is the sequence of Reeb chords gotten by replacing each
H
-chord in ([

1]. [

1
H
]) with
the corresponding -chord.
Proof. This is immediate from Proposition 6.25 and Lemmas 6.24, 6.28 and 6.29 and
the observation that, in the notation of Lemma 6.24, the small triangles 1
x
have index 0.
6.3.2.3. Denition of the homomorphism associated to a handleslide. For x S(. . ),
y S(.
H
. ), 1
2
(x
t
. y) and a sequence of non-empty sets of Reeb chords so that
(x. ) is strongly boundary monotone and (1. ) is compatible, dene
/
B
(x. y.
o
;
1
.
2
) :=
_
[

P]=
1
,[

P
H
]=
2
b(B

)=B
(T

)=
emb
(B

,
1
,
2
)
/
B

(x. y.
o
; 1

;

1
1
.

1
2
)
/
B
(x. y. ; ) :=
_
=(
1
,
2
)
/
B
(x. y.
o(x,
1
)
;
1
.
2
).
The triangle map from

CFD(. . . .) to

CFD(.
H
. . .) is dened by
1
,
H
,
(x) :=

B
2
(x

,y)

[ ind(B, )=0
#
_
/
B
(x. y. ; )
_
c( )y (6.33)
1
,
H
,
(cx) := c1
,
H
,
(x).
As in the denition of the boundary operator, if both (. . . .) and (.
H
. ) are provin-
cially admissible this sum is nite, as the 0-dimensional moduli spaces are compact and only
nitely many of them are non-empty.
The map 1
,
H
,
is dened on all of

CFD(. . . .). However, Lemma 4.21 and the
fact that s
z
(x) = s
z
(x
t
) imply that if we restrict to the summand

CFD(. . . .; s) of

CFD(. . . .) representing a xed spin

c
structure s then the image of 1
,
H
,
is contained
in the corresponding summand

CFD(.
H
. . .; s) of

CFD(.
H
. . .).
Our next goal is to prove that 1
,
H
,
is a chain map. The new technical point relates
to degenerations at
12
, to which we turn next.
6.3. INVARIANCE 117
6.3.2.4. Holomorphic curves at
12
. We start by completely analyzing what curves can
degenerate at
12
; but rst some more notation.
There are 2(p + /) components of (
H
) not adjacent to ..
For each -circle
c
i
(i = 1. . . . . p/) there is a pair of bigons 1
c
i,1
and 1
c
i,2
contained
entirely in the interior of , with boundaries in
c
i

c,H
i
. (The numbering 1 and 2
is arbitrary.)
For each -arc
a
i
with i 2 there is a pair of bigons 1
a
i,1
and 1
a
i,2
with boundary in

a
i

a,H
i
. (Again, the numbering 1 and 2 is arbitrary.) These bigons are adjacent
to ; one of the two punctures of each 1
a
i,j
corresponds to c in .
There is an annulus 1
a
1,1
with boundary in
a
1

a,H
1

c
1

c,H
1
and a bigon 1
a
1,2
with boundary in
a
1

a,H
1
. Both 1
a
i,1
and 1
a
i,2
are adjacent to .
We call the regions 1
a
i,1
and 1
a
i,2
(respectively 1
c
i,1
and 1
c
i,2
) twin regions. See Figure 6.3.
Lemma 6.34. Let
n: (o. o) ( [0. 1] 1. (0 1) (
H
1 1))
be a holomorphic curve asymptotic to a generator
o
at , with multiplicity 0 at .. Then
o is a disjoint union of bigons and possibly an annulus. Moreover:
(c-1) If o
i
and o
j
are distinct components of o then

(n(o
i
))

(n(o
j
)) = .
(c-2) If o
i
is a bigon then either n[
S
i
is a trivial strip (i.e., (

n)[
S
i
is constant and
(
|
n)[
S
i
is a dieomorphism) or (

n)[
S
i
is a dieomorphism onto its image,
which is one of the distinguished bigons 1
c
i,j
or 1
a
i,j
.
(c-3) The image of the annular component of o (if it exists) is one of 1
a
1,1
, 1
a
1,1
+ 1
c
1,1
,
1
a
1,1
+ 1
c
1,2
, 1
a
1,1
+ 1
c
1,1
+ 1
c
1,2
, 1
a
1,1
+ 21
c
1,1
or 1
a
1,1
+ 21
c
1,2
.
(c-4) If a region 1
a
i,1
(respectively 1
a
i,2
, 1
c
i,1
, 1
c
i,2
) is in the image of

n then its
twin region 1
a
i,2
(respectively 1
a
i,1
, 1
c
i,2
, 1
c
i,1
) is not in the image of

n, with
the possible exception that both 1
c
i,1
and 1
c
i,2
may be in the image of

n on an
annular component.
Proof. Suppose rst that

n does not cover the annulus 1

a
1,1
. Since n is asymptotic
to the generator
o
, each bigon 1
a
i,j
or 1
c
i,j
appears with multiplicity 0 or 1, and at most
one of each pair of twin bigons appears in the image of

n. The result follows.

Next, suppose that

n covers the annulus 1

a
1,1
. Then, as before, the domain of n is a
disjoint union of bigons away from a neighborhood of the annulus. The asymptotics at
imply that near the annulus, the domain has one of the forms 1
a
1,1
, 1
a
1,1
+1
c
1,1
, 1
a
1,1
+1
c
1,2
,
1
a
1,1
+ 1
c
1,1
+ 1
c
1,2
, 1
a
1,1
+ 21
c
1,1
or 1
a
1,1
+ 21
c
1,2
. The result follows.
6.3.2.5. Gluing results at
12
. Next we prove two gluing results for these curves at
12
.
(See Figure 6.7 for a typical example of this gluing.) Again, we need a little more terminology.
The holomorphic curves in Lemma 6.34 are asymptotic to a generator
o
at . At +,
they are asymptotic to p-tuples of the form
w
+
pd
= (u
1
. . . . . u

[0. 1] +) (
1
. . . . .
g
[0. 1] +)
where each u
i
is a point in
H
and each
i
is a Reeb chord in from some
a,H
j
to
some
a
k
. We call tuples w
+
pd
of this form positive partially-diagonal generators. (In fact, for
the curves in Lemma 6.34, , = /, but we do not require this condition for partially diagonal
generators.) We will abuse notation and also refer to the set w
+
pd
= u
1
. . . . . u

.
1
. . . . .
g

118 6. TYPE D MODULES

as a positive partially-diagonal generator. Negative partially-diagonal generators are dened
similarly, but with Reeb chords running from some
a
j
to some
a,H
k
, and are denoted by
w

pd
.
We collect the holomorphic curves at
12
into moduli spaces. Specically, for each gener-
ator
o
, positive partially-diagonal generator w
+
pd
and domain 1

H
,
2
(
o
. w
+
pd
), there is
a corresponding moduli space of (stable) holomorphic curves /
B
(
o
. w
+
pd
). (As usual, we
mod out by the action of 1 by translation.) Lemma 6.34 shows that the topological type
of the source of curves in /
B
(
o
. w
+
pd
) is determined by 1 and w
+
pd
, but when we want to
name the source explicitly we will also write this moduli space as /
B
(
o
. w
+
pd
; o

).
It follows from Lemma 6.34 that the moduli spaces /
B
(
o
. w
+
pd
) are typically high-
dimensional:
Corollary 6.35. Fix a generic split almost complex structure on [0. 1] 1. Let w
+
pd
=
u
1
. . . . . u

.
1
. . . . .
g
be a positive partially-diagonal generator. Then the moduli space
/
B
(
o
. w
+
pd
) for generic J is either empty or has dimension at least maxp / 1. 0.
Further:
(1) If p / = 0 and /
B
(
o
. w
+
pd
) is non-empty and has dimension 0 then 1 is the
bigon 1
c
i,j
for some c (and w
+
pd
=
o
).
(2) If /
B
(
o
. w
+
pd
) is non-empty and has dimension p / 1 then every coecient
in 1 is 0 or 1 and each connected component of the support of 1 is one of 1
a
i,j
for
some i 1; 1
a
1,2
; 1
a
1,1
1
c
1,1
; or 1
a
1,1
1
c
1,2
.
Conversely, each bigon has a unique holomorphic representative, and the sum of the number
of holomorphic representatives of 1
a
1,1
+ 1
c
1,1
and 1
a
1,1
+ 1
c
1,2
is odd.
Proof. This is a straightforward analysis of the moduli spaces corresponding to the
domains in Lemma 6.34, as follows.
Property (c-1) ensures that the moduli space

/
B
(
o
. w
+
pd
) (i.e., before we divide out by
the 1 action) splits as a product of moduli spaces associated to the component curves. By the
classication from Lemma 6.34, these factors correspond to moduli spaces representing the
domains in the following table, listed with their number of chords and expected dimension
(before dividing out by 1):
Expected dimension
Homology class Number of chords of moduli space
0 0 0
1
c
i,j
0 1
1
a
i,j
, i 1 1 1
1
a
1,2
1 1
1
a
1,1
1 0
1
a
1,1
+ 1
c
1,1
1 1
1
a
1,1
+ 1
c
1,2
1 1
1
a
1,1
+ 21
c
1,1
1 2
1
a
1,1
+ 1
c
1,1
+ 1
c
1,2
1 2
The only homology class where the number of chords exceeds the dimension is 1
a
1,1
; but in
this case, the dimension is 0, so (for generic J) the corresponding moduli space is empty, and
6.3. INVARIANCE 119
hence cannot appear as a factor of a non-empty moduli space. The desired inequality now
follows from the fact that dim/
B
(
o
. w
+
pd
) is the sum of the dimensions of the constituent
moduli spaces minus one (for the 1 action), while p / is the total number of chords.
The condition that p/ = 0 is equivalent to the condition that each constituent piece has
no chords in it, in which case the homology class must be a sum of domains 1
c
i,j
. Moreover,
each component contributes 1 to the dimension (before dividing out by 1). This establishes
Point (1).
The condition that p / 1 coincides with the dimension ensures that each constituent
piece must have exactly as many chords as its dimension: thus, each constituent piece must
be one of 1
a
i,j
for some i 1, 1
a
1,2
, 1
a
1,1
+1
c
1,1
or 1
a
1,1
+1
c
1,2
. This establishes Point (2).
Given a negative partially-diagonal generator w

pd
and generators x S(. ) and y
S(
H
. ) we can consider the moduli space
/
B
(x. y. w

pd
; 1

;

1.

1
H
)
of holomorphic triangles asymptotic to the generators x, y and w

pd
at
31
,
23
and
12
respec-
tively and with ordered partitions of Reeb chords

1 and

1
H
along c
1
and c
2
, respectively.
We can also consider the moduli space
/
B
(x. y. w

pd
; .
H
)
of embedded holomorphic triangles asymptotic to the sequences of non-empty sets of Reeb
chords and
H
along c
1
and c
2
, respectively.
We call a pair of partially-diagonal generators w
+
pd
= u
+
1
. . . . . u
+

.
1
. . . .
g
and w

pd
=
u

1
. . . . . u

m
.
t
1
. . . . .
t
gm
of opposite polarity compatible if u
+
1
. . . . . u
+

= u

1
. . . . . u

(so, in particular, : = /) and there is a bijection j:

1
. . . . .
g

t
1
. . . . .
t
gm
so that

i
and j(
i
) are adjacent in the sense that either
+
i
= j(
i
)

or

i
= j(
i
)
+
. We will call
the map j a matching between w
+
pd
and w

pd
.
Given a compatible pair of partially-diagonal generators w
+
pd
and w

pd
and a match-
ing j between them there are two associated sets of Reeb chords

(w
+
pd
. w

pd
. j) and

H(w
+
pd
. w

pd
. j), given by

H(w
+
pd
. w

pd
. j) :=
_

+
i
=(
i
)

i
j(
i
)

(w
+
pd
. w

pd
. j) :=
_

i
=(
i
)
+
j(
i
)
i
.
When the matching is implicit or determined by the rest of the data, as will usually
be the case below, we will drop it from the notation, writing simply

H(w
+
pd
. w

pd
) and

(w
+
pd
. w

pd
).
If w
+
pd
and w

pd
are compatible partially-diagonal generators then the product of mod-
uli spaces /
B
(
o
. w
+
pd
) /
B
(x. y. w

pd
; .
H
) sits inside the compactied moduli space
/
B
(x. y.
o
;
t
.
H,t
). where
t
is obtained from by adjoining the set

(w
+
pd
. w

pd
. j) at
the end and
H,t
is obtained from by adjoining the set

H(w
+
pd
. w

pd
. j) at the beginning.
(In principle, this construction depends on a choice of matching j, but boundary monotonic-
ity of /
B
(x. y. w

pd
; .
H
) implies that this matching is determined by the rest of the data.)
120 6. TYPE D MODULES
Figure 6.5. Example of a degeneration at
12
. At the left is a 1-
parameter family of holomorphic curves connecting r
1
,
1
and
1
; the pa-
rameter is the length of the cut. One end of this moduli space corresponds
to a degeneration at
12
into a curve connecting
1
and u
+
1
and a triangle
connecting r
1
,
1
and u

1
.
In other words, there is an obvious notion of a sequence of curves in /
B
(x. y.
o
;
t
.
H,t
)
converging to a pair /
B
(
o
. w
+
pd
) /
B
(x. y. w

pd
; .
H
). See Figure 6.5 for an example.
In order to understand the structure of /
B
(x. y.
o
;
t
.
H,t
) near the boundary stratum
/
B
(
o
. w
+
pd
) /
B
(x. y. w

pd
; .
H
) we will study some auxiliary moduli spaces. Consider
the product /
B
(
o
. w
+
pd
; o

) /
B
(x. y. w

pd
; 1

;

1.

1
H
). There is an obvious way to
preglue o

and 1

at the punctures corresponding to u

pd
, giving a source o

. The
preglued surface o

has punctures :
1
. . . . . :
a
labeled by the Reeb chords in

(w
+
pd
. w

pd
)
and :
H
1
. . . . . :
H
b
labeled by the Reeb chords in

H(w
+
pd
. w

pd
), where c + / = p / is the
number of chords in u

pd
. For any permutations
a
o
a
and
b
o
b
we can consider the
moduli space
/
B
(x. y.
o
; o

; (

1. (:
a(1)
. . . . . :
a(a)
)). ((:
H

b
(1)
. . . . . :
H

b
(b)
)

1
H
)).
(Recall that for sequences : and t, (:. t) denotes the concatenation of : and t.) There is
an obvious notion for a sequence of curves in this moduli space to converge to a curve in
/
B
(
o
. w
+
pd
; o

) /
B
(x. y. w

pd
; 1

;

1.

1
H
).
With this terminology in place, we have the following gluing results:
Proposition 6.36. A neighborhood of /
B
(
o
. w
+
pd
; o

) /
B
(x. y. w

pd
; 1

;

1.

1
H
) in
_
aSa,
b
S
b
/
B
(x. y.
o
; o

; (

1. (:
a(1)
. . . . . :
a(a)
)). ((:
H

b
(1)
. . . . . :
H

b
(b)
)

1
H
))
is homeomorphic to /
B
(
o
. w
+
pd
; o

) /
B
(x. y. w

pd
; 1

;

1.

1
H
) [0. c).
Proof. At rst glance, it seems that this requires a new kind of gluing argument. Specif-
ically, suppose that n

H
,
/
B
(
o
. w
+
pd
; o

) and n
,
H
,
/
B
(x. y. w

pd
; 1

;

1.

1
H
).
6.3. INVARIANCE 121
Figure 6.6. Straightening-out -arcs. The conformal map giving this
straightening is locally the map . .
t
for some appropriate t 0.
Write w
+
pd
= u
+
1
. . . . . u
+

.
1
. . . . .
g
and w

pd
= u

1
. . . . . u

.
t
1
. . . . .
t
g
. At a punc-
ture
i
of n

H
,
corresponding to
i
, lim
qq
i
t(n

H
,
()) = and lim
qq
i

(n

H
,
()) =
j, the puncture in (so the curve goes to innity in two directions at once). Simi-
larly, at a puncture
t
i
of n
,
H
,
corresponding to
t
i
, lim
qq

(n
,
H
,
()) =
12
and
lim
qq

(n
,
H
,
()) = j. Such diagonal asymptotics seem not to have been treated in
the literature.
Because of the simple form of the curves in /
B
(
o
. w
+
pd
), we can avoid the new analysis
by observing that our situation is biholomorphic to a familiar one, as follows.
Consider rst the case that / = p 1, so that each of w

pd
contains a single Reeb chord.
Recall that
e
denotes the result of lling in the puncture on . The various
a
i
and
a,H
i
meet at the puncture j in
e
, in general with a corner. To x notation, suppose that the
chord
1
runs from
a,H
i
to
a
j
and that
t
1
runs from
a
j
to
a,H
k
; the case that
t
1
runs
from
a
k
to
a,H
i
is similar. Let 1
z
be the component of ( ) adjacent to . and let

t
e
=
e
1
z
. We can choose a complex structure ,
t

(and smooth structure) on

t
e
so that
,
t

agrees with the xed complex structure ,

from Chapter 5 on
t
e
c and so that the
chord
1

t
1
corresponds to an 180

angle with respect to ,

t

. (See Figure 6.6. Note that

while the total angle at the corner of
t
e
may be more than 360

, the holomorphic curves

are contained in a subset with total angle 180

.) Let J
t
be an almost complex structure on

t
e
which agrees with J away from j and with ,
t

near j.
The holomorphic curve n
,
H
,
has image in
t
e
and is holomorphic with respect to
J
t
. Similarly, the holomorphic curve n

H
,
has image in
t
e
[0. 1] 1 and is holomorphic
with respect to the complex structure on
t
e
[0. 1] 1 induced by J
t
. Moreover, the gluing
problem for n
,
H
,
and n

H
,
in
t
e
[0. 1] 1 has the same form as the usual gluing of
holomorphic curves in Heegaard Floer theory at a generator. So, in this case the result
follows from the usual gluing theorem, [Lip06a, Proposition A.2], say.
The general case, where w

pd
contains more than one Reeb chord, follows from the above
argument with one more observation. The almost complex structure J
t
that we used above
depended on the endpoints of the chords
1
and
t
1
being glued; it may not be possible
to choose a single almost complex structure which straightens out all of the relevant arcs.
However, we may choose the complex structure J
t
on to depend on the point in
the source of n
,
H
,
. (In other words, for curves with source o, we are now considering
holomorphic sections of a topologically trivial bundle o o, with a complex
122 6. TYPE D MODULES
structure on o given by J
t
s
,
S
, where ,
S
is the complex structure on o and J
s
is an almost complex structure on ; compare [Sei08b, Section 3].) In particular, we
can choose J
t
= J
t
(:. d. j) (: , d , j o the source of n
,
H
,
) so that for j near the
puncture j
t
i
, the angle corresponding to
i

t
i
or
t
i

i
(as appropriate) is 180

. The result
then follows as before.
It follows from Proposition 6.36 that if /
B
(x. y.
o
; ) is a 1-dimensional moduli space
of holomorphic curves, where is the discrete partition, then, in degenerations at
12
, the
resulting curve for (
H
. ) has index 1, and hence at most one non-trivial component. In
particular, by Lemma 6.34, at most one -
H
-Reeb chord is involved.
Proposition 6.37. Suppose that 1
2
(x. y) and sequences of non-empty sets of Reeb
chords = (
1
. . . . .
m
),
H
= (
H
1
. . . . .
H
n
) are such that ind(1. (.
H
)) = 1. For a
suciently generic, admissible almost complex structure J, the boundary of the moduli space
/
B
(x. y.
o
; .
H
) (where o = o(x. )) consists of curves of the following types:
(MT-1) Two-story buildings corresponding to breaks at
31
, i.e., curves in
/
B
1
(x. u; (
1
. . . . .
i
)) /
B
2
(u. y.
o
; (
i+1
. . . . .
m
).
H
)
where 1
1

,
2
(x. u) and 1
2

2
(u. y) are such that 1
1
1
2
= 1, and 1 i :.
(MT-2) Two-story buildings corresponding to breaks at
23
, i.e., curves in
/
B
1
(x. u.
o
; . (
H
1
. . . . .
H
i
)) /
B
2
(u. y; (
H
i+1
. . . . .
H
n
))
where 1
1

2
(x. u) and 1
2

H
,
2
(u. y) are such that 1
1
1
2
= 1, and 1 i :.
(MT-3) Join curve ends.
(MT-4) Split curve ends.
(MT-5) Other collisions of levels among or
H
.
(MT-6) Shue curve ends.
(MT-7) Degenerations at
12
where the resulting partially diagonal generator involves only
intersection points (not Reeb chords), i.e., curves in
/
B
1
(
o
. w
+
pd
) /
B
2
(x. y. w

pd
; .
H
)
where w
+
pd
= w

pd
= u
1
. . . . . u
g

H
. Moreover, the homology class 1
1
(and
generator w
+
pd
) is such that /
B
1
(
o
. w
+
pd
) is 0-dimensional.
(MT-8) Degenerations where
m
approaches
12
, i.e., curves in
/
B
1
(
o
. w
+
pd
) /
B
2
(x. y. w

pd
; (
1
. . . . .
m1
).
H
)
where w

pd
are compatible partially-diagonal generators,

(w
+
pd
. w

pd
) =
m
and

H(w
+
pd
. w

pd
) = . Moreover, the homology class 1
1
(and generator w
+
pd
) is such
that moduli space /
B
1
(
o
. w
+
pd
) is ([
m
[1)-dimensional (compare Corollary 6.35).
(MT-9) Degenerations where
H
1
approaches
12
, i.e., curves in
/
B
1
(
o
. w
+
pd
) /
B
2
(x. y. w

pd
; . (
H
2
. . . . .
H
n
))
where w

pd
are compatible partially-diagonal generators,

(w
+
pd
. w

pd
) = and

H(w
+
pd
. w

pd
) =
H
1
. Moreover, the homology class 1
1
(and generator u
+
pd
) is such
that moduli space /
B
1
(
o
. w
+
pd
) is ([
H
1
[1)-dimensional (compare Corollary 6.35).
6.3. INVARIANCE 123
Conversely, each curve of type (MT-1), (MT-2), (MT-3), (MT-4), (MT-5), (MT-7) and
the odd case of (MT-6) corresponds to an odd number of ends of /
B
(x. y.
o
; .
H
). (By
contrast, we will see in the proof of Proposition 6.39 that not all curves of types (MT-8) and
(MT-9) correspond to ends of /
B
(x. y.
o
; .
H
).)
Proof. That any degeneration has one of forms (MT-1)(MT-6) or corresponds to a
degeneration at
12
follows from an argument analogous to the proof of Theorem 5.55. That
the degenerations at
12
have the specied forms follows from Proposition 6.36. For example,
for case (MT-8), Proposition 6.36 implies that without the height constraints imposed by

m
, the glued moduli space has dimension
dim(/
B
1
(
o
. w
+
pd
)) + dim(/
B
2
(x. y. w

pd
; (
1
. . . . .
m1
).
H
)) + 1.
So, the dimension of /
B
(x. y.
o
; .
H
) is
dim(/
B
1
(
o
. w
+
pd
)) + dim(/
B
2
(x. y. w

pd
; (
1
. . . . .
m1
).
H
)) + 1 [
m
[ + 1.
By Corollary 6.35, this is at least
[
m
[ 1 + 0 + 1 [
m
[ + 1 = 1.
with equality only if dim(/
B
1
(
o
. w
+
pd
)) = [
m
[ 1.
The converse follows from gluing arguments similar to those used to prove Theorem 5.55
(and Proposition 6.36).
As a warm-up to our next proposition, we have the following:
Lemma 6.38. For any 1, and
H
so that ind(1. (.
H
)) = 1, the moduli space /
B
(x. y.

o
; .
H
) has an even number of ends of type (MT-7).
Proof. By Corollary 6.35, each end of /
B
(x. y.
o
; .
H
) of type (MT-7) corresponds
to a pair of curves (n

H
,
. n
,
H
,
) where n

H
,
is a bigon with domain some 1
c
i,j
. There is
a corresponding end (n
t

H
,
. n
,
H
,
) where n
t

H
,
is a bigon with domain 1
c
i,j1
. The result
follows.
The following proposition is similar in spirit to Lemma 6.38, but for the bigons with
domains 1
a
i,j
and the annulus with domain 1
a
1,1
+ 1
c
1,1
or 1
a
1,1
+ 1
c
1,2
.
Proposition 6.39. Fix 1
2
(x. y) and a sequence of non-empty sets of Reeb chords
(
1
. . . . .
m
) such that ind(1. (
1
. . . . .
n
)) = 1. Let o
i
= o(x. (
1
. . . . .
i
)). For a suciently
generic almost complex structure J, the number of ends of the moduli space
/
B
i,i+1
:= /
B
(x. y.
o
i
; (
1
. . . . .
i
). (
H
i+1
. . . . .
H
n
))
where
i
approaches
12
has the same parity as the number of ends of
/
B
i1,i
:= /
B
(x. y.
o
i1
; (
1
. . . . .
i1
). (
H
i
. . . . .
H
n
))
where
H
i
approaches
12
.
(Since we have fairly weak control over the behavior of /
B
near the boundary, the
parity of the number of ends might deserve some explanation. For each c 0, consider
n /
B
i,i+1
[ d(

n(1
i
).
12
) c, where d denotes the distance in , viewed as the unit
disk in C with three boundary punctures, and 1
i
denotes the punctures of the source of n
mapped to
i
. For generic c, this is a manifold with boundary. The number of boundary
124 6. TYPE D MODULES
Figure 6.7. An illustration of a boundary bigon degenerating, and
a dierent one being glued on the other side. The domain (shaded)
changes slightly in the process.
components of this manifold modulo 2 is independent of c if c is generic and suciently
small. This is the (modulo 2) number of ends of /
B
i,i+1
where
i
approaches
12
.)
Proof. We start with the case that
i
consists of a single Reeb chord ; the argument
in this case is illustrated schematically in Figure 6.7. By Proposition 6.37, ends of /
B
i,i+1
at
which
i
approaches
12
correspond to a pair of curves (n

H
,
. n
,
H
,
). Proposition 6.37 and
Corollary 6.35 imply that the curve n

H
,
is either a bigon with domain 1
a
i,j
(for some i. ,)
or an annulus with domain 1
a
1,1
+ 1
c
1,1
or 1
a
1,1
+ 1
c
1,2
. In either case, there is algebraically
one such curve n

H
,
in the moduli space /
B
(
o
. w
+
pd
); this is obvious in the case of bigons,
and follows from [OSz04d, Lemma 9.4] in the case of annuli. Which domain 1
a
i,j
occurs in
the domain is determined by the chord : the region 1
a
i,j
has the point
+
in its boundary.
Proposition 6.36 implies that near (n

H
,
. n
,
H
,
), the compactied moduli space /
B
i,i+1
is
homeomorphic to [0. c).
There is a corresponding curve n
t

H
,
, again a bigon 1
i,j
or annulus 1
a
1,1
+1
c
1,1
or 1
a
1,1
+
1
c
1,2
, with

in the boundary of the domain of n

t

H
,
. The pair (n

H
,
. n
,
H
,
) is a point in
/
B
ii,i
. Again, Proposition 6.36 implies that near (n

H
,
. n
,
H
,
) the moduli space /
B
i1,i
is homeomorphic to [0. c). The result, in the case that
i
consists of a single chord, follows.
As a warm-up to the general case, we consider the case that
i
contains exactly two
Reeb chords,
1
and
2
. Our argument is analogous to the proof of Proposition 5.33. By
Proposition 6.37, when
i
approaches the puncture, the curve degenerates a curve n

H
,
in
(.
H
. . .) with two non-trivial components. By Corollary 6.35, each of these components
maps either to the bigon 1
a
i,j
or to the annulus 1
a
1,1
+ 1
c
1,1
or 1
a
1,1
+ 1
c
1,2
. There is also a
leftover component n
,
H
,
in . Let /(n

H
,
) denote the moduli space of curves in
([0. 1] 1. 0 1
H
1 1) of which n

H
,
is a member. The moduli space
/(n

H
,
) is homeomorphic to (an odd number of copies of) 1; dierent points are gotten
by sliding the two non-trivial components of n

H
,
relative to each other.
By Proposition 6.36, a neighborhood | of /(n

H
,
) n
,
H
,
in
/
B
(x. y.
o
i
; (
1
. . . . .
i1
.
1
.
2
). (
H
i+1
. . . . .
H
n
))
/
B
(x. y.
o
i
; (
1
. . . . .
i1
.
2
.
1
). (
H
i+1
. . . . .
H
n
))
6.3. INVARIANCE 125
is homeomorphic to 1[0. c). Choose this homeomorphism so that (0. c) is the distance
in between
12
and the closer of

(n(1
1
)) and

(n(1
2
)), where 1
1
and 1
2
are the punc-
tures corresponding to
1
and
2
, respectively. Given (0. c), for t suciently small the
point (t. ) lies in /
B
(x. y.
o
i
; (
1
. . . . .
i1
.
1
.
2
). (
H
i+1
. . . . .
H
n
)), say, while for t suf-
ciently large the point (t. ) lies in /
B
(x. y.
o
i
; (
1
. . . . .
i1
.
2
.
1
). (
H
i+1
. . . . .
H
n
)).
It follows that /
B
i,i+1
has an odd number of ends in /(n

H
,
) n
,
H
,
.
For each of the components of n

H
,
there is a matching bigon or annulus, as in the single
Reeb chord case; denote the union of the matching components by n
t

H
,
. (That is, n
t

H
,
has two non-trivial components whose domains contain the points

1
and

2
, respectively.)
A similar argument to the previous paragraph shows that /
B
i1,i
has an odd number of ends
in /(n
t

H
,
) n
,
H
,
. The result follows.
For the general case, write
i
=
1
. . . . .
m
. Consider a pair of curves (n

H
,
. n
,
H
,
)
/
B
i,i+1
. as above. Let o

denote the source of n

H
,
and 1

the source of n
,
H
,
. Let

1.

1
H
be the ordered partitions of punctures on n
,
H
,
along c
1
and c
2
induced by (
1
. . . . .
i1
)
and (
H
i+1
. . . . .
H
n
). Let :
1
. . . . . :
m
denote the punctures of o

mapped to
1
. . . . .
m
.
Consider the moduli space
_
Sm
/
B
(x. y.
o
; o

; (

1. (:
(1)
. . . . . :
(m)
)).

1
H
)
There is an evaluation map
ev:
_
Sm
/
B
(x. y.
o
; o

; (

1. (:
(1)
. . . . . :
(m)
)).

1
H
) (c
1
)
m
given by
ev(n) = (

n)(:
1
. . . . . :
m
).
The moduli space /
B
i,i+1
is the preimage under ev of the total diagonal (r. r. . . . . r) [ r
c
1
. We can also evaluate at only some of the punctures; let
ev
[1,j]
(n) = (

n)(:
1
. . . . . :
j
)
Let
: /
B
(
o
. w
+
pd
; o

) /
B
(x. y. w

pd
; 1

;

1.

1
H
) [0. c)

_
Sm
/
B
(x. y.
o
; o

; (

1. (:
(1)
. . . . . :
(m)
)).

1
H
)
denote the gluing map from Proposition 6.36. Recall from Corollary 6.35 that there is a
bijection between the nontrivial components of n

H
,
and the punctures :
i
. Gluing at the
punctures :
1
. . . . . :
j
(i.e., translating the components corresponding to :
j+1
. . . . . :
m
to
and then gluing at
12
) gives a gluing map

[1,j]
: \
[1,j]
/
B
(x. y. w

pd
; 1

;

1.

1
H
) [0. c)

_
Sm
/
B
(x. y.
o
; o

; (

1. (:
(1)
. . . . . :
(m)
)).

1
H
)
for the corresponding stratum \
[1,j]
of /
B
(
o
. w
+
pd
; o

) (where the curve at

12
has at
least two stories, and the punctures :
1
. . . . . :
j
are on a higher story than the punctures
:
j+1
. . . . . :
m
). Write \
[1,j]
= \
[1,j]
\
[j+1,m]
where \
[1,j]
is the moduli space corresponding
126 6. TYPE D MODULES
to the punctures :
1
. . . . . :
j
and \
[j+1,m]
is the moduli space corresponding to the punctures
:
j+1
. . . . . :
m
.
We will prove the following statement by induction on ::
Given any (n

H
,
. n
,
H
,
) as above there is an open neighborhood l in (c
1
)
j
con-
taining (
12
.
12
. . . . .
12
) and an c 0 so that
ev
[1,j]

[1,j]
: \
[1,j]
n
,
H
,
(0. c) (c
1
)
j
maps with odd degree onto l.
Given this statement for the case , = :, taking the preimage of the diagonal implies the
proposition.
The case , = 1 is clear from (the proof of) Proposition 6.36.
The inductive step follows from Lemma 5.32 with A = \
[1,j]
and ) = ev
[1,j]

[1,j]
. The
fact that the union of the top two strata of \
[1,j]
is a manifold with boundary is clear from
the structure of \
[1,j]
(analyzed completely in Corollary 6.35) or, alternatively, follows from
gluing theory. That ) is proper near 0 follows from the gluing construction. The inductive
hypothesis gives the fact that the restriction of ) to the (codimension 1) facet where :
1
is
on the bottom story and :
2
. . . . . :
m
are on the top story (adjacent to
12
), with image in
(c
1
)
m1
, has odd degree near 0.
6.3.2.6. The map associated to a handleslide is a chain map.
Proposition 6.40. 1
,
H
,
is a chain map, i.e., 1 + 1 = 0.
Proof. This follows from Proposition 6.37, Lemma 6.38 and Proposition 6.39, similarly
to the proof of Proposition 6.7. Ends of type (MT-1) correspond to 1
,
H
,
(x). Ends
of type (MT-2) correspond to part of 1
,
H
,
(x). Ends of type (MT-4) correspond to
the rest of 1
,
H
,
(x). Ends of type (MT-5) cancel with each other and with ends of
type (MT-3), as in the proof of Proposition 6.7. Ends of type (MT-7) cancel in pairs by
Lemma 6.38. Ends of type (MT-8) cancel with ends of type (MT-9) by Proposition 6.39.
The result follows.
To show that 1
,
H
,
is an isomorphism, we wish to construct a chain map the other
direction and a chain homotopy between their composition and the identity map. A count
of triangles in (.
H
. . ) does not work as desired, as the small bigons 1
a
i,j
(and annuli
1
a
1,1
+ 1
c
1,1
and 1
a
1,1
+ 1
c
1,2
) are not holomorphic: their orientations are consistent with
(.
H
. ), not (
H
. . ). Instead we dene a third set of circles, which we denote
t
.
Obtain
c,t
i
from
c
i
by performing a small Hamiltonian perturbation, so that
c,t
i
intersects
each of
c
i
and
c,H
i
transversely in two points and is disjoint from all other curves in
H
.
Let
a,t
i
be an isotopic translate of
a
i
, intersecting each of
a
i
and
a,H
i
in a single point, and
such that there are two short Reeb chords in running from
a,t
i
to
a,H
i
(and hence two
slightly longer Reeb chords running from
a,t
i
to
a
i
). The isotopy is chosen small enough
that
a,t
i
is disjoint from all other curves in
H
. See Figure 6.8. Now there is a chain map
1

H
,

,
dened using (.
H
.
t
. . .), in exactly the same way that 1
,
H
,
was dened
using (. .
H
. . .).
6.3.2.7. Completion of the proof of handleslide invariance. To prove handleslide invari-
ance, we also need to consider the map 1
,

,
dened in the same way as 1
,
H
,
but with

t
in place of
H
.
Proposition 6.41. The map 1
,

,
is an isomorphism of dierential /-modules.
6.3. INVARIANCE 127
Figure 6.8. The curves ,
H
and
t
in . The curves in are solid,
those in
H
are dashed, and those in
t
are dotted.
Proof. The proof that 1
,

,
is a chain map is parallel to, but marginally easier than,
the proof of Proposition 6.40, the only dierence being that the discussion of annuli is
irrelevant here.
Next we argue that 1
,

,
is in fact an isomorphism. The proof, which makes use of the
energy ltration, is a straightforward analogue of the proof for closed 3-manifolds [OSz04d,
Proposition 9.8]. We will use the following standard lemma.
Lemma 6.42. Let 1 : 1 be a map of ltered groups which is decomposed as a sum
1 = 1
0
+ / where 1
0
is a ltration-preserving isomorphism and / has strictly lower order
than 1
0
. Suppose that the ltration on 1 is bounded below. Then 1 is an isomorphism of
groups.
Choose an area form Area on such that if 1 is a periodic domain in (. . ) then
Area(1) = 0. (The existence of Area is guaranteed by Lemma 4.26.) Arrange also that
periodic domains in (.
t
. ) have area 0, and that the boundary bigons all have the same
area. It follows that for 1 a periodic domain in (.
t
. ), Area(1) = 0 as well.
For each spin
c
structure s on 1 , we dene a map T from S(. . . s) to 1. Choose one
generator x
0
S(. . . s) and declare T(x
0
) = 0. Then, for any other x S(. . . s)
pick
x
0
,x

2
(x
0
. x) and dene T(x) = Area(
x
0
,x
). Since periodic domains have area
0, T(x) is independent of the choice of
x
0
,x
. More generally, for c / such that cx ,= 0,
dene T(cx) = T(x). Since holomorphic curves have positive domains, the map T induces
a ltration on

CFD(. . ).
There is an obvious identication between S(. . . s) and S(.
t
. . s). As before,
let x
t
denote the generator corresponding to x, and let 1
x

2
(x. x
t
.
o(x)
) be the canonical
small triangle. Dene a map T
t
0
: S(.
t
. . s) 1 in the same way as T, still using the area
form Area but using x
t
0
in place of x
0
. Dene T
t
: S(.
t
. . s) 1 by T
t
= T
t
0
Area(1
x
0
).
By extending T
t
by T
t
(cy) = T
t
(y), we get a ltration T
t
on

CFD(.
t
. ).
128 6. TYPE D MODULES
To see that 1
,

,
respects the ltrations T and T
t
, suppose there is a term cy in
1
,

,
(x), and thus a positive domain 1
2
(x. y
t
.
o
), and pick a domain
x
0
,x

2
(x
0
. x).
Then, for an appropriate o
t
= o(x
0
) and domain
o,o

2
(
o
.
o
), the domain
x
0
,x
+1
1
x
0

o,o
connects x
t
0
and y
t
. Our assumptions guarantee that Area(
o,o
) = 0, so
T
t
(y
t
) = Area(
x
0
,x
+ 1 1
x
0

o,o
) Area(1
x
0
) = Area(
x
0
,x
) Area(1) < T(x)
as desired.
Now, if we choose Area so that for all x the triangle 1
x
is the unique triangle of minimal
area connecting x, y
t
and
o
for any y
t
and o (this is easily accomplished), then the top
order part of 1
,

,
with respect to T and T
t
is simply the map x x
t
. In particular, the
top order part of 1
,

,
is a group isomorphism.
Consequently, it follows from Lemma 6.42 that the /-module homomorphism 1
,

,
is an F
2
vector space isomorphism

CFD(. . . .)

CFD(.
t
. . .), and hence also a
/-module isomorphism.
Remark 6.43. We could weaken our assumptions and prove handleslide invariance for Hee-
gaard diagrams 1 that are provincially admissible (rather than admissible) by being slightly
more clever in the choice of ltration T, as follows. Pick a map ) : H
1
(. a) 1 so that
for any periodic domain 1 for (. . ), )(

1) + Area(1) = 0. (This is possible for any

provincially admissible diagram.) Then dene T(x) = Area(
x
0
,x
) + )(

x
0
,x
), and make
similar other adjustments. Alternatively, we can do an isotopy to make 1 admissible, do
the handleslide, and then do an inverse isotopy.
Proposition 6.44. The map 1
,

,
is chain homotopic to 1

H
,

,
1
,
H
,
.
Proof. The proof is in two steps. First, the composition 1

H
,

,
1
,
H
,
is a count
of pairs of triangles, in (. .
H
. . .) and (.
H
.
t
. . .). This map is chain homotopic
to counting pairs of triangles in (. .
t
. . .) and (. .
H
.
t
. .), via a chain homotopy
counting (index 1) holomorphic curves in a path of almost complex structures on the
product of and a rectangle. (See Figure 6.9 for a schematic illustration.) The reader
unfamiliar with such arguments is directed to any of [OSz04d, Theorem 8.16], [Lip06a,
Proposition 10.29] or [Lip06b, Lemma 7.2.5]. The only new types of degeneration are
curves in either (.
H
. . .) or (.
t
.
H
. .) touching , either small bigons or annuli.
As in Proposition 6.40, these degenerations cancel in pairs.
Second, a direct computation shows that the only rigid triangles in (. .
H
.
t
. .)
asymptotic to
o,
H
,
and
o,

,
H at two of the corners are asymptotic to
o,

,
at the
third cornerand that there is algebraically a single such triangle. Again, the reader is
directed to [OSz04d, Lemma 9.7] for detailsthe proof there applies without change in our
case.
Proposition 6.45. The /-modules

CFD(. . . .) and

CFD(.
H
. . .) are homotopy
equivalent as dg modules.
Proof. If we arrange that
t
and are close enough then the obvious identication of
generators of

CFD(. . . .) and

CFD(.
t
. . .) induces an isomorphism of /-modules
(because all of the moduli spaces counted in the denitions of are the same). Hence, by
Propositions 6.41 and 6.44, the composition of 1

H
,

,
1
,
H
,
and the evident identication
of generators is chain homotopic to an automorphism of

CFD(. . . .).
6.3. INVARIANCE 129
Figure 6.9. Two compositions of triangles, as degenerations of a rec-
tangle. The bold arrows indicate a path in the compactied moduli space of
rectangles, starting at a conguration giving 1

H
,

,
1
,
H
,
and ending at
a conguration giving 1
,

,
.
Choose a fourth set of curves
H,t
isotopic to
H
and dene another map 1

,
H,
,
:

CFD(.
t
. . .)

CFD(.
H,t
. . .). The same argument shows that the composition
of 1

,
H,
,
1

H
,

,
and the evident identication of generators is chain homotopic to an
automorphism of

CFD(.
H
. . .). The result follows.
6.3.3. Completion of the invariance proof.
Proof of Theorem 6.16. In Section 6.3.1, we saw that

CFD(1. s; J) is independent
of the choice of generic almost-complex structure J in its denition. Suppose now that 1
and 1
t
are two Heegaard diagrams which represent equivalent bordered three-manifolds (and
in particular 1 = : = 1
t
). Then, according to Proposition 4.10, we can pass from
1 to 1
t
by a sequence of Heegaard moves. Invariance under isotopies has not been spelled
out, but this can be done by writing down the continuation maps for isotopies modeled
on the continuation maps for variation of almost-complex structures as in Equation (6.19).
Invariance under handleslides of an arc over a circle was veried in Proposition 6.45 in the
case that the diagrams are admissible; in the general case, rst do an isotopy to make the
diagrams admissible, then a handleslide, and then undo the isotopy. The case of handleslides
of a circle over a circle is similar but easier, and is left to the reader. Invariance under
stabilization is straightforward: by stabilizing near the basepoint, one can see that the chain
complexes before and after the stabilization are isomorphic (just as in the closed case for

HF, [OSz04d, Theorem 10.1]).

130 6. TYPE D MODULES
In view of Theorem 6.16, we can now refer to

CFD without reference to a Heegaard dia-
gram, dening

CFD(1. . s) (where : 1(:) 1 ) to be

CFD(1. s), where 1 is any bor-
dered Heegaard diagram representing the bordered three-manifold (1. ). We will sometimes
drop the map from the notation, writing

CFD(1. s) to denote

CFD(1. : 1(:) 1 . s).
To obtain a topological invariant of the bordered three-manifold without reference to a
spin
c
structure, we simply sum over all spin
c
structures, dening

CFD(1 ) =

sspin
c
(Y )

CFD(1. s).
Proof of Theorem 1.1. With the above denition, this follows immediately from
Theorem 6.16.
6.4. Twisted coecients
As with the original Heegaard Floer homology [OSz04c, Section 8], there are versions of

CFD(1. s) with twisted coecients. We describe the (maximal) twist by H

2
(1. 1 ). (Recall
from Lemma 4.18 that, for any x S(1. s),
2
(x. x)

= H
2
(1. 1 ).) Besides its intrinsic
interest, we explain this version because, as we shall see in Chapter 10, the grading on the
twisted theory is easier to understand than the grading on the untwisted theory.
Fix a spin
c
structure s over 1 , and pick a base generator x
0
= x
0
(s) S(1. s) repre-
senting s. (For spin
c
structures s without any generators, the twisted chain complex is 0.)
Let S(1. x
0
) be the set of elements of the form c
B
0
x, where x S(1. s), 1
0

2
(x
0
. x),
and c is a formal variable. Let A(1. x
0
) be the F
2
vector space spanned by S(1. x
0
). There
is an action of J(:) on A(1. x
0
), where
1(s) c
B
0
x :=
_
c
B
0
x 1(s) = 1
D
(x)
0 otherwise,
as in Equation (6.1).
As a module over / = /(:) (where : = 1), dene the twisted chain complex

CFD(1. x
0
) to be /
7
A(1. x
0
).
There is a natural action of H
2
(1. 1 ) on

CFD(1. x
0
), by composition with the corre-
sponding periodic domain: for H
2
(1. 1 ) corresponding to 1

2
(x
0
. x
0
), dene
c
B
0
x := c
BB
0
x.
As in the untwisted case, the action of / on

CFD(1. s) is given by left multiplication:
c (/ c
B
0
x) := (c/) c
B
0
x.
For the boundary operator, the contributions of the holomorphic curves connecting x to y
are separated according to their homology classes:
(c
B
0
x) :=

yS(7,s)

B
2
(x,y)
c
B
x,y
c
B
0
B
y.
It is convenient to think of the action by H
2
(1. 1 ) as a right action of the group-ring
F
2
[H
2
(1. 1 )] on

CFD(1. x
0
. s). Writing the group-ring as sums

H
2
(Y,Y )
:

, where
6.4. TWISTED COEFFICIENTS 131
:

Z,2Z is zero for all but nitely H

2
(1. 1 ), we let
c
B
0
x
_

_
=

c
BB
0
x.
Proposition 6.46.

CFD(1. x
0
) is a dierential /F
2
[H
2
(1. 1 )] bimodule.
Proof. Here, we are thinking of F
2
[H
2
(1. 1 )] as a dg algebra with trivial dierential,
so that the statement means that

CFD(1. x. s) is a chain complex, and
for c /, :

CFD(1. x
0
. s), and c F
2
[H
2
(1. 1 )],
(c :) c = c (: c)
(c : c) = (c): c + c (:) c.
The proof of Proposition 6.7 readily adapts to show that

CFD(1. x
0
. s) is a chain com-
plex. The action by F
2
[H
2
(1. 1 )] commutes both with the dierential and with the action
by /. The result follows.
Theorem 6.47. Up to homotopy equivalence, the dierential module

CFD(1. x
0
) depends on
the generator x
0
only through its induced spin
c
structure s
z
(x) and on the Heegaard diagram
only through its (equivalence class of ) underlying bordered three-manifold (1. : 1(:)
1 ) (with : = 1). That is, if (1. x
0
) and (1
t
. x
t
0
) are provincially admissible bordered
Heegaard diagram for 1 with 1 = : = 1
t
, and x
0
S(1. s) and x
t
0
S(1
t
. s)
are base generators representing the same spin
c
structure over 1 , then

CFD(1. x
0
) and

CFD(1
t
. x
t
0
) are homotopy equivalent /(:)F
2
[H
2
(1. 1 )] bimodules.
Proof. First, we address the dependence of

CFD(1. x
0
) on the base generator x
0
.
Suppose that 1 is an admissible diagram, and that x
0
and x
1
are two dierent generators
which represent the same spin
c
structure over 1 . Then, by Lemma 4.21, we can nd some
/
2
(x
0
. x
1
). We use / to induce a map A(1. x
1
) A(1. x
0
) sending the generator c
B
x
to c
bB
x. To see that this respects the action by H
2
(1. 1 ), note that for H
2
(1. 1 ),
/ 1

= 1

/ (the map 1

from H
2
(1 ; Z) to
2
(x
0
. x
0
) gives a domain which does not
depend on the basepoint x
0
). The map A(1. x
1
) A(1. x
0
) extends to a bimodule map

b
:

CFD(1. x
1
)

CFD(1. x
0
). If x
0
= x
1
and /
2
(x
0
. x
1
) is the trivial homology class
then
b
is the identity. Moreover, given /
0

2
(x
0
. x
1
) and /
1

2
(x
1
. x
2
),
b
0
b
1
=
b
2

b
1
.
It follows that for any /
2
(x
0
. x
1
),
b
is an isomorphism.
To adapt the proof of Theorem 6.16 to the twisted setting, we must give twisted analogues
of the continuation and triangle maps. For example, for a suitable path of almost complex
structures J
r
, we dene
1
Jr
:

CFD(1. s; J
0
)

CFD(1. J
1
)
by adapting the map dened in Equation (6.19):
1
Jr
(c
B
0
x) =

yS(7,s)

B
2
(x,y)

[ ind(B, )=0
#
_
/
B
(x. y; ; J)
_
c( )c
B
0
B
y.
We modify the triangle map from Equation (6.33) similarly, writing
1
,
H
,
(c
B
0
x) :=

B
2
(x

,y)

[ ind(B, )=0
#
_
/
B
(x. y. ; )
_
c( )c
B

0
B
y.
132 6. TYPE D MODULES
where 1
t
0

2
(x
t
0
. x) is determined (via Lemma 6.24) by
1
x

13
1
0
= 1
x
0

12
1
,
H
23
1
t
0
.
With the maps in place, the proof Theorem 6.16 adapts easily to the twisted setting.
CHAPTER 7
Type modules
We now turn to dening the type module

CFA associated to a (provincially admissible)
bordered Heegaard diagram 1. This uses more of the moduli spaces from Chapter 5 than
the type 1 module does; in particular, we must consider moduli spaces of curves where the
(ordered) partition of Reeb chords has more than one chord per part.
7.1. Denition of the type module
Fix a bordered Heegaard diagram 1 = (. . . .), satisfying the provincial admissibility
criterion of Denition 4.23, and a spin
c
structure s over 1. The module

CFA(1. s) is a right
(/

) /(:)-module, where : = 1. Unlike

CFD, which is essentially free (projective)
over /(:), much of the data of the Heegaard diagram is encoded in the multiplication on

CFA(1. s), as well as in certain higher products.

The module

CFA(1. s) is generated over F
2
by the set of generators S(1. s). Given
x S(1. s), dene 1
A
(x) := 1(o(x)). (Recall that o(x) [2/] is the set of -arcs occupied
by x.) We dene a right action of J(:) on

CFA(1) by
(7.1) x 1(s) :=
_
x 1
A
(x) = 1(s)
0 otherwise.
Again, the summands /(:. i) of /(:) act trivially on

CFA(1. s) for i ,= 0.
Recall from Section 3.1.3 that to a set of Reeb chords we associate an algebra element
c().
Lemma 7.2. For x S(1. s) and a sequence of non-empty sets of Reeb chords, (x. ) is
strongly boundary monotone (Denition 5.46) if and only if the tensor product
1
A
(x)
7
c(
1
)
7

7
c(
n
)
is not zero.
Proof. In the notation of Denition 5.46, if 1
A
(x)
7
c(
1
)
7

7
c(
i
) ,= 0 then the
right idempotent of 1
A
(x)
7
c(
1
)
7

7
c(
i
) is 1(o(o(x).
[1,i]
)), i.e.,
(1
A
(x)
7
c(
1
)
7

7
c(
i
))1(o(o(x).
[1,i]
)) = 1
A
(x)
7
c(
1
)
7

7
c(
i
).
and then 1(o(o(x).
[1,i]
))c(
i+1
) ,= 0 if and only if `(

i+1
) o(o(x).
[1,i]
) and `(
+
i+1
)
is disjoint from o(o(x).
[1,i]
) `(

i+1
), which is exactly the condition of strong boundary
monotonicity.
As explained in Denition 2.5, an /

module structure on

CFA(1. s) is a family of maps
:
n+1
:

CFA(1. s)
7
/
7

7
/

CFA(1. s)
133
134 7. TYPE A MODULES
satisfying Equation 2.6. In particular, in view of Lemma 7.2, since the tensor products are
over J(:), it suces to dene
:
n+1
(x. c(
1
). . . . . c(
n
))
when (x.
1
. . . . .
n
) is strongly boundary monotone. (In the following sections we will use
commas to separate the tensor factors in the argument to :
n+1
and similar maps.)
Denition 7.3. Fix an admissible (in the sense of Denition 5.1), suciently generic (in
the sense of Denition 5.7) almost-complex structure on [0. 1] 1. For (x. ) strongly
boundary monotone (and
i
,= for each i), dene
:
n+1
(x. c(
1
). . . . . c(
n
)) :=

yS(7,s)

B
2
(x,y)
ind(B,)=1
#
_
/
B
(x. y;
1
. . . . .
n
)
_
y (7.4)
:
2
(x. I) := x (7.5)
:
n+1
(x. . . . . I. . . . ) := 0. : 1. (7.6)
where I is the unit in /(:). (Equations (7.5) and (7.6) state that

CFA(1) is strictly unital.)
We will sometimes use the alternate notation :
1
(x) = (r) and :
2
(x. c) = x c.
Let

CFA(1) =

sspin
c
(Y )

CFA(1. s).
Lemma 7.7. If 1 is provincially admissible, then for any xed generators x and y and
a consistent sequence of non-empty sets of Reeb chords, there are at most nitely many
1
2
(x. y) such that /
B
(x. y; ) is non-empty; in particular, the sums appearing in the
denition of :
k
are nite. Moreover, if 1 is admissible, then there are only nitely many
non-zero maps :
k
(i.e., in the language of Section 2.1, (

CFA(1). :
k
) is bounded).
Proof. The rst statement is immediate from Proposition 4.28 and the fact that if a
domain admits a holomorphic representative then all of its coecients must be non-negative
(Lemma 5.4). For the second, observe that if we let [1[ be the sum of all local multiplicities,
then :
n+1
counts domains with [1[ :. Thus, Proposition 4.29 gives the desired result.
Remark 7.8. Notice that the operation :
1
:

CFA

CFA is the dierential obtained by
counting only provincial holomorphic curves, i.e., curves which do not approach .
Remark 7.9. The value of :
n+1
when one of the arguments is I can be made more natural
by allowing the corresponding
i
to be empty (since c() = 1). The corresponding moduli
spaces should be interpreted as having a choice of height t
i
, and no other conditions at t
i
.
Except in the case of trivial strips, these moduli spaces cannot be 0-dimensional, and hence
they do not contribute to :
n+1
.
Remark 7.10. In Chapter 8 we will show that for special kinds of diagrams (the obvious
analogues of nice diagrams from [SW10]) the :
i
vanish for i 2, so

CFA is an ordinary
(not /

) dierential module. However, there are reasons for considering more general
diagrams. First, it is not clear how to prove invariance if one restricts only to nice diagrams.
Further, in spite of the added analytic and algebraic complications, it is often easier to
compute with a diagram which is not nice, as general diagrams often have a much smaller
set of generators S(1).
7.2. COMPATIBILITY WITH ALGEBRA 135
7.2. Compatibility with algebra
In this section we will prove that

CFA(1. s) is an /

module over /(:). Recall from

Denition 2.5 that the basic relation for an /

module (over an /

algebra over F
2
) is
0 =

i+j=n+1
:
i
(:
j
(x. c
1
. . . . . c
j1
). . . . . c
n1
)
+

i+j=n+1
nj

=1
:
i
(x. c
1
. . . . . c
1
. j
j
(c

. . . . . c
+j1
). . . . . c
n1
).
For the algebra /(:), the j
i
vanish for i 2. Therefore, we must check that
0 =

i+j=n+1
:
i
(:
j
(x. c
1
. . . . . c
j1
). . . . . c
n1
)
+
n1

=1
:
n
(x. c
1
. . . . . c

. . . . . c
n1
)
+
n2

=1
:
n1
(x. c
1
. . . . . c

c
+1
. . . . . c
n1
).
(7.11)
Proposition 7.12. The data (

CFA(1. s). :
i

i=1
) forms an /

module over /(:).

We give the proof shortly. The most interesting points in the proof of Proposition 7.12
can be seen in the following examples. (The reader may wish to compare Examples 7.137.15
with Examples 6.136.15.)
Example 7.13. On the left of Figure 7.1 is a piece of a diagram with four generators. The
non-trivial multiplications in the corresponding module

CFA are
/. d = c. c
c. d
_
1 3
2

= c. c
c. c
_
1 2
3

= c. c
/. d
_
2 1
3

= c. d
/. d
_
1 2
2 3

= c. c
There is an /

compatibility equation
(/. d)
_
1 2
3

+/. d
_
1 2
3

= 0.
(Recall that
_
1 2
3

=
_
1 2
2 3

.) Geometrically, this corresponds to considering the two-

dimensional moduli space of curves from /. d to c. c with a Reeb chord from 1 to 3.
This moduli space has two ends, one of which is a two-story building, and the other of which
is a join curve end.
136 7. TYPE A MODULES
Figure 7.1. Local illustrations of terms occurring the /

relation
for

CFA. The moduli spaces are the same as the moduli spaces in Figure 6.1
in Section 6.1, but the algebraic interpretations of the phenomena are dierent
for

CFA than for

CFD.
Example 7.14. In the center of Figure 7.1 is a piece of a diagram with four generators.
The non-trivial products are
/. d
_
1 3
2

= c. d
/. c
_
1 4
2

= c. c
c. d
_
3 2
4

= c. c
/. d
_
3 1
4

= /. c
/. d
_
1 3
2 4

= c. c.
Here, associativity follows from the fact that
_
/. d
_
1 3
2
_

_
3 2
4

= /. d
_
1 3
2 4

=
_
/. d
_
3 1
4
_

_
1 4
2

.
The second equality above can be seen geometrically as follows. Consider the two-dimensional
moduli space connecting /. d to c. c where the Reeb chord from 3 to 4 is encountered
before the one from 1 to 2. Then this moduli space has two ends, one of which is a two story
building, while at the other end the heights of two Reeb chords are the same.
Example 7.15. On the right in Figure 7.1 is a piece of a diagram with three generators.
The non-trivial products are
c
_
1
2

= /
c
_
1
3

= c
/
_
2
3

= c.
Associativity follows from the fact that
_
c
_
1
2
_

_
2
3

= c
_
1
3

.
Geometrically, at one end of the moduli space is a two story building. At the other end, two
levels collide and a split curve degenerates.
7.2. COMPATIBILITY WITH ALGEBRA 137
Figure 7.2. A domain where :
3
is necessary for associativity. The do-
main is shaded; the dark arcs denote cuts, whose ends are boundary branched
points of

n. A typical curve in the one-parameter family is indicated in

the center. On the left is a two-story end of the moduli space. On the right is
the other end, corresponding to degenerating a split curve.
Example 7.16. Figure 7.2 shows an example in which the higher product :
3
is needed for
/

associativity. The diagram has four generators, c. c, c. d, /. c and /. d. The

dierential is given by c. c = c. d and /. c = /. d. For an appropriate choice of
complex structure J, there is a non-trivial product (:
2
) given by /. c
_
1
3

= c. c; this
corresponds to the domain and branch cuts indicated in the right of the gure. One can
factor
_
1
3

=
_
1
2

_
2
3

. Obviously /. c
_
1
2

= 0, so ordinary associativity fails. There is,

however, for the same choice of complex structure J as above, a higher product
:
3
_
/. d.
_
1
2

.
_
2
3
_
= c. c.
with domain given by the lighter-colored region in the middle of Figure 7.2. Thus, the /

associativity condition is satised:

:
3
_
(/. c).
_
1
2

.
_
2
3
_
+
_
/. c
_
1
2
_

_
2
3

+/. c
_
1
3

= 0.
Geometrically, this corresponds to the one-parameter moduli space of curves asymptotic to
two Reeb chords (at dierent heights), shown at the left of Figure 7.2. One end (pictured in
the center) is a two-story building where one of the stories contains both Reeb chords. At the
other end (pictured at the right), the two Reeb chords collide and a split curve degenerates.
Proof of Proposition 7.12. Morally, the proof proceeds by considering the ends of
the index 2 moduli spaces. As in the type 1 module case, however, we will instead appeal
to Theorem 5.55.
Fix generators x and y, a domain 1
2
(x. y) and a sequence of non-empty sets of
Reeb chords = (
1
. . . . .
n
) so that (x. ) is strongly boundary monotone, 1 is compatible
with , and ind(1. ) = 2. Set
i
=
i,j
. Theorem 5.55 applied to the data (x. y. 1. )
(for all sources o

with (o

) =
emb
(1. )) shows that the sum of the following numbers
is zero:
(1) The number of two-story ends, i.e., the number of elements of
/
B
1
(x. w;
1
) /
B
2
(w. y;
2
)
where 1 = 1
1
1
2
and = (
1
.
2
).
(2) The number of join curve ends, i.e., the number of elements of
/
B
(x. y; (
1
. . . . .
a,b
i
. . . . .
n
))
138 7. TYPE A MODULES
where
i,j
=
a

b
and
a,b
i
is obtained from
i
by replacing
i,j
with
a
.
b
.
(3) The number of odd shue curve ends, i.e., the number of elements of
/
B
(x. y;
t
)
where
t
is obtained from by performing a weak shue on one
i
.
(4) The number of collisions of levels, i.e., the number of elements of
/
B
(x. y;
1
. . . . .
i

i+1
. . . . .
n
)
where
i
and
i+1
are weakly composable.
(We are also using Lemma 5.69 to ensure that the limiting curves are embedded.)
The two-story ends correspond to the rst term in Formula (7.11).
Next we discuss the second term in Formula (7.11). Lemma 3.14 implies that, for a
consistent set of Reeb chords,
c() =

a splitting
of
c(
t
) +

a shue
of
c(
t
).
On the other hand, sums (2) and (3) above correspond, respectively, to the sum over all
weak splittings and weak shues one of the sets of Reeb chords in . By Lemma 5.70, only
honest (not just weak) splittings or shues contribute, and therefore the second term in
Formula (7.11) corresponds to sums (2) and (3) above.
Finally, sum (4) corresponds to the third term in Formula (7.11), as follows. Since, by
Lemma 3.12, c(
i
)c(
i+1
) is c(
i

i+1
) if
i
and
i+1
are composable, and is zero otherwise,
the coecient of y in :
n1
_
x. c(
1
). . . . . c(
i
)c(
i+1
). . . . . c(
n
)
_
is equal to sum (4) if
i
and
i+1
are composable, and is 0 otherwise. But by Lemma 5.70, sum (4) is also 0 if
i
and
i+1
are not composable, and otherwise gives the third term of Formula (7.11).
7.3. Invariance
Theorem 7.17. Up to /

homotopy equivalence, the /

module

CFA(1. s) is independent
of the almost complex structure used to dene it, and depends on the provincially admissible
Heegaard diagram 1 with : = 1 only through its induced equivalence class of bordered
three-manifold (1. : 1(:) 1 ). That is, if 1 and 1
t
are provincially admissible bordered
Heegaard diagram for (1. : 1(:) 1 ) and s is a spin
c
structure on 1 then

CFA(1. s)
and

CFA(1
t
. s) are /

homotopy equivalent /

/(:)-modules.
As for invariance of

CFD, the proof of Theorem 7.17 diers from the standard arguments
in essentially two ways: the additional algebraic complications in dealing with /

maps and
homotopies, and the additional analytic complications in the case of handleslide invariance
because of the presence of east . Both of these issues are slightly more involved for

CFA
than for

CFD. We will take a similar approach as for

CFD: rst we will explain succinctly
the proof of invariance under change in almost complex structure, and then we will discuss
the (mild) new complications in the triangle maps used to prove handleslide invariance.
7.3.1. Change of complex structure. In this section, we address invariance of

CFA
under change in almost complex structure. In the course of this analysis, we will relate
the algebra of /

maps and their /

homotopies to holomorphic curve counts; because

7.3. INVARIANCE 139
this is somewhat more involved than the algebra of dierential graded algebras required in
Section 6.3.1, we will explain more steps in the proof.
Fix a Heegaard diagram 1 = (. . . .) and suciently generic, admissible almost
complex structures J
0
and J
1
. The complex structures J
0
and J
1
lead to potentially dier-
ent moduli spaces, and hence to two potentially dierent /

modules

CFA(1. s; J
0
) and

CFA(1. s; J
1
). We will show that these two /

modules are /

homotopy equivalent.
As in Section 6.3.1, call a (smooth) path J
r
[ : [0. 1] of almost complex structures
from J
0
to J
1
admissible if each J
r
is admissible. To the path J
r
we can associate a single
almost complex structure J on [0. 1] 1 by Formula (6.17), and dene embedded moduli
spaces /(x. y; ; J) by Formula (6.18).
Recall from Denition 2.9 that an /

morphism between /

modules /
1
and /
2
is a
compatible family of maps
)
n
: `
1

n1
..
/ / `
2
.
for : N. So dene an /

map )
Jr
= )
Jr
n

n=1
from

CFA(1. s; J
0
) to

CFA(1. s; J
1
) by
setting
)
Jr
n
(x. c(
1
). . . . . c(
n1
)) :=

yS(7,s)

B
2
(x,y)
ind(B,)=0
#
_
/
B
(x. y;
1
. . . . .
n1
; J)
_
y
)
Jr
n
(x. . . . . I. . . . ) := 0
and extending )
Jr
to all of

CFA(1. s; J
0
) /
(n1)
multilinearly over J(:). The above map
is dened over all of

CFA(1; J
0
); but, in view of Lemma 4.21, it clearly maps

CFA(1. s; J
0
)
to

CFA(1. s; J
0
).
Our rst task is to show that )
Jr
is an /

morphism. We again use Proposition 6.20.

Lemma 7.18. The map )
Jr
= )
Jr
n

nN
:

CFA(1. s; J
0
)

CFA(1. s; J
1
) is a strictly unital
morphism of /

modules, as dened in Denition 2.9.

Proof. The map )
Jr
is strictly unital by denition. It remains to check the /

relation
for morphisms, which we recall from Denition 2.9 (specialized for j
i
= 0 for i 2):
0 =

i+j=n+1
:
J
1
i
()
Jr
j
(x. c
1
. . . . . c
j1
). . . . . c
n1
)
+

i+j=n+1
)
Jr
i
(:
J
0
j
(x. c
1
. . . . . c
j1
). . . . . c
n1
)
+
n1

=1
)
Jr
n
(x. c
1
. . . . . c
1
. c

. c
+1
. . . . . c
n1
)
+
n2

=1
)
Jr
n1
(x. c
1
. . . . . c
1
. c

c
+1
. c
+2
. . . . . c
n1
).
This follows in a similar manner to Proposition 7.12. The rst term corresponds to the
two-story ends with a J-holomorphic curve followed by a J
1
-holomorphic curve. The second
term corresponds to two-story ends with a J
0
-holomorphic curve followed by a J-holomorphic
140 7. TYPE A MODULES
curve. The third term corresponds to join curve ends and odd shue curve ends. The fourth
term corresponds to collisions of two levels.
Similarly, x a generic path J
t
r
of admissible almost complex structures with J
t
0
= J
1
and
J
t
1
= J
0
. Let J
t
denote the associated complex structure on [0. 1] 1. Then J
t
r
denes
an /

morphism )
J

r
:

CFA(1. s; J
1
)

CFA(1. s; J
0
). We will show:
Proposition 7.19. The maps )
Jr
and )
J

r
are homotopy inverses to each other. In partic-
ular,

CFA(1. s; J
0
) is /

homotopy equivalent to

CFA(1. s; J
1
).
The proof is again essentially standard, but as the algebra may be unfamiliar we will give
a few details.
For 1 0, let J
R
J
t
denote the almost complex structure on [0. 1] 1 given by
J
R
J
t
[
(x,s,t)
:=
_

_
J
0
[
(x,s,t)
if 1 + 1 t
J
t
tR
[
(x,s,tR)
if 1 t 1 + 1
J
1
[
(x,s,t)
if 1 t 1
J
t
[
(x,s,t)
if 0 t 1
J
0
[
(x,s,t)
if t 0.
Associated to J
R
J
t
is an /

morphism )
J
R
J

:

CFA(1. s; J
0
)

CFA(1. s; J
0
).
Lemma 7.20. For 1 suciently large the map )
J
R
J

is the /

composite )
J

r
)
Jr
.
Proof. Recall from Section 2.1 that )
J

r
)
Jr
has
()
J

r
)
Jr
)
n
(x. c
1
. . . . . c
n1
) =

i+j=n+1
)
J

r
j
_
)
Jr
i
(x. c
1
. . . . . c
i1
). . . . . c
n1
_
.
The coecient of y in
)
J

r
j
_
)
Jr
i
(x. c(
1
). . . . . c(
i1
)). . . . . c(
n1
))
_
counts holomorphic curves in /
B
1
(x. w;
[1,i1]
; J) /
B
2
(x. w;
[i,n1]
; J
t
) where w
S(1. s), 1
1

2
(x. w), 1
2

2
(w. x), ind(1
1
.
[1,i1]
) = 0 and ind(1
2
.
[i,n1]
) = 0. It
follows from obvious non-1-invariant analogues of Propositions 5.23 and 5.24 (compactness
and gluing) that for 1 large enough such curves correspond to curves in
_
B
2
(x,y)
/
B
(x. y;
1
. . . . .
n1
; J
R
J
t
).
The result follows.
We next dene an /

homotopy / between )
J
R
J

and the identity map. Fix a path

of almost complex structures J
w

w[0,1]
on [0. 1] 1 with J
0
= J
R
J
t
and J
1
= J
0
,
and such that each J
w
satises conditions (J-1), (J-2) and (J-4) of Denition 5.1i.e., J
w
is
admissible except for not being 1-invariant. For a generic path J
w

w[0,1]
and any 1. o

. 1
with ind(1. o

. 1) = 1, there are nitely many u (0. 1) for which /

B
(x. y; o

; 1; J
w
)
7.3. INVARIANCE 141
is non-empty. So set
/
B
(x. y; ; J
w
) :=
_
w[0,1]
/
B
(x. y; ; J
w
)
/
n
(x. c(
1
). . . . . c(
n1
)) :=

yS(7,s)

B
2
(x,y)
ind(B,)=1
#
_
/
B
(x. y; ; J
w
)
_
y.
In this context, the analogue of Theorem 5.55 is:
Proposition 7.21. Suppose that (x. ) satises the strong boundary monotonicity condition.
Fix y, 1
2
(x. y), o

, and

1 such that both [

1] = and ind(1. o

.

1) = 0. Then, for a
generic path J
w

w[0,1]
, in the moduli space

w[0,1]
/
B
(x. y; o

;

1; J
w
), the total number
of all
(1) elements of /
B
(x. y; o

;

1; J
0
);
(2) elements of /
B
(x. y; o

;

1; J
1
);
(3) two-story ends, with either
(a) a J
0
-holomorphic curve followed by a J
w
-holomorphic curve or
(b) a J
w
-holomorphic curve followed by a J
0
-holomorphic curve;
(4) join curve ends;
(5) odd shue curve ends; and
(6) collisions of two levels 1
i
and 1
i+1
from

1, where [1
i
] and [1
i+1
] are composable,
is even.
Lemma 7.22. The map / is an /

homotopy between )
J
R
J

and the identity map on

CFA(1. J
0
).
Proof. We must check that
(7.23) )
J
R
J

1
(x) x = /
1
(:
J
0
1
(x)) :
J
0
1
(/
1
(x))
and that for : 1,
)
J
R
J

n
(x. c
1
. . . . . c
n1
) =

i+j=n+1
/
i
(:
J
0
j
(x. c
1
. . . . . c
j1
). . . . . c
n1
)
+

i+j=n+1
:
J
0
i
(/
j
(x. c
1
. . . . . c
j1
). . . . . c
n1
)
+
n1

=1
/
n
(x. c
1
. . . . . c

. . . . . c
n1
)
+
n2

=1
/
n1
(x. c
1
. . . . . c

c
+1
. . . . . c
n1
).
(7.24)
As expected, this follows from Proposition 7.21. The )
J
R
J

terms in Formulas (7.23)

and (7.24) correspond to item (1) of Proposition 7.21. The x term of Formula (7.23)
corresponds to item (2), since J
1
= J
0
and for an 1-invariant complex structure the only
index 0 holomorphic curves are trivial strips. As usual, two-story ends correspond to the
remaining two terms in Formula (7.23) as well as the rst two sums in Formula (7.24). The
142 7. TYPE A MODULES
join curve and shue curve ends account for the third sum in Formula (7.24). Collisions of
two levels account for the nal sum in Formula (7.24).
Proof of Proposition 7.19. This is immediate from Lemmas 7.20 and 7.22, together
with the obvious analogues for the composition )
Jr
)
J

r
.
7.3.2. Handlesliding an -arc over an -circle. The purpose of the present subsec-
tion is to prove handleslide invariance:
Proposition 7.25. Let (. . . .) be a bordered Heegaard diagram and (.
H
.
H
. .) the
result of performing a handleslide of one of the -curves over an -circle or one -circle
over another -circle. Then there is a strictly unital /

homotopy equivalence between

CFA(. . . .) and

CFA(.
H
. . .).
The most subtle case of handleslide invariance is handlesliding an -arc over an -circle;
we focus on this case. We will be brief, as most of the argument is the same as the type 1
case, in Section 6.3.2.
Fix a Heegaard triple diagram (. .
H
. ) as in Section 6.3.2. There are associated
moduli spaces of holomorphic curves in
/
B
(x. y. ;
1
. . . . .
n
) :=
n
_
i=0
/
B
(x. y.
o
i
; (
1
. . . . .
i
). (
H
i+1
. . . . .
H
n
))y.
Dene the /

morphism )
,
H
,
= )
,
H
,
n

n=1
associated to the handleslide by
(7.26) )
,
H
,
n
(x. c(
1
). . . . . c(
n1
)) :=

B
2
(x

,y)
ind(B,)=0
#
_
/
B
(x. y. ;
1
. . . . .
n1
_
)y
and extending to all of

CFA(1. s. J
0
) by multi-linearity over J(:).
Proof of Proposition 7.25. This follows along the same lines as the proof of Propo-
sition 6.45. That )
,
H
,
is a strictly unital /

morphism follows from Propositions 6.37

and 6.39, similarly to how Proposition 7.12 follows from Theorem 5.55. To prove that )
,
H
,
is invertible we consider its composition with another triangle map, )

H
,

,
, where
t
is a
small Hamiltonian perturbation of , as in Section 6.3.2.6. Associativity for holomorphic
triangle maps, the type analogue of Proposition 6.44, shows that )

H
,

,
)
,
H
,
is
homotopic to )
,

,
. It follows from Lemma 6.42 that )
,

,
is an isomorphism (noting
that an /

homomorphism )
i
is an isomorphism if and only if )
1
is an isomorphism).
Thus, )
,
H
,
is injective on homology (or, better, monic in the homotopy category). That
)
,
H
,
is surjective on homology (or, better, epic in the homotopy category) follows from a
similar argument using a fourth set of circles
H,t
, as in the proof of Proposition 6.44.
7.3.3. Completion of the invariance proof.
Proof of Theorem 7.17. Independence from the complex structure was veried in
Section 7.3.1. Handleslide invariance was outlined in Section 7.3.2. Stabilization invariance
is obvious, by stabilizing in the region containing the basepoint. Isotopy invariance is left to
the reader. In light of Proposition 4.10 this veries topological invariance.
7.4. TWISTED COEFFICIENTS 143
In view of Theorem 7.17, if (1. : 1(:) 1 ) is a bordered three-manifold and s is
a spin
c
structure over 1 , dene

CFA(1. : 1(:) 1 ) to be

CFA(1. s), where 1 is any
bordered diagram representing the bordered three-manifold. Let

CFA(1 ) :=

sspin
c
(Y )

CFA(1. s).
Proof of Theorem 1.2. This follows at once from Theorem 7.17.
7.4. Twisted coecients
We can also dene a twisted theory

CFA(1. x
0
), for x
0
a generator representing the
s spin
c
(1 ), in an analogous way to the denition of

CFD(1. x
0
) in Section 6.4.
Consider the generating set S(1. x
0
), consisting of elements of the form c
B
0
x, where
x S(1. s) and 1
0

2
(x
0
. x).

CFA(1. x
0
) is the F
2
vector space spanned by S(1. s). It
is given an action of J(:) generalizing Equation (7.1), via
1(s) c
B
0
x :=
_
x 1(s) = 1
A
(x)
0 otherwise
There is an action of H
2
(1. 1 ) on

CFA(1. s), by composition with the corresponding
periodic domain: For H
2
(1. 1 ) corresponding to 1
2
(x
0
. x
0
), dene
c
B
0
x := c
BB
0
x.
We can view this as a left action by the group ring F
2
[H
2
(1. 1 )].
The /

action by /(:) is specied by:

:
n+1
(c
B
0
x. c(
1
). . . . . c(
n
)) :=

yS(7)

B
2
(x,y)
ind(B,)=1
#
_
/
B
(x. y;
1
. . . . .
n
)
_
c
B
0
B
y
:
2
(c
B
0
x. I) := x
:
n+1
(c
B
0
x. . . . . I. . . . ) := 0. : 1.
where here I is the unit in /(:). Extend this to an /

bimodule action of F
2
[H
2
(1. 1 )]
and /(:) by setting
:
1,1,0
(. c
B
0
x) := c
B
0
x
:
m,1,n
(
1
. . . . .
m
. c
B
0
x. c
1
. . . . . c
n
) := 0 if :: 0 or : 1.
Proposition 7.27. The data (

CFA(1. s). :
i

i=1
) forms an /

bimodule over the rings

F
2
[H
2
(1. 1 )] and /(:).
Proof. This is a simple adaptation of the proof of Proposition 7.12.
Theorem 7.28. Up to /

homotopy equivalence, the /

bimodule

CFA(1. s) over F
2
[H
2
(1.
1 )] and /(:) is independent of the almost complex structure used to dene it, and depends
on the Heegaard diagram 1 only through its (equivalence class of ) underlying bordered three-
manifold (1. : 1(:) 1 ) and on the generator x
0
only through its induced spin
c
structure
s
z
(x
0
).
144 7. TYPE A MODULES
Proof. The proof of Theorem 7.17 can be adapted easily to the twisted context, fol-
lowing the model of Theorem 6.47. Note once again that all the maps commute with the
F
2
[H
2
(1. )]-action (rather than merely commuting up to homotopy).
CHAPTER 8
Pairing theorem via nice diagrams
We now give a proof of the pairing theorem (Theorem 1.3), adapting methods from Sarkar
and Wang [SW10].
Denition 8.1. A bordered Heegaard diagram (. . . .) is called nice if every region in
except for the region 1
z
adjacent to . is a (topological) disk with at most 4 corners. That
is, each region in the interior of is either a bigon or a quadrilateral and any region at the
boundary of except for 1
z
has two -arcs, one -arc and one arc of in its boundary.
Proposition 8.2. Any bordered Heegaard diagram (. . . .) can be turned into a nice
bordered Heegaard diagram (. .
t
. .) via
isotopies of the
i
not crossing and
handleslides among the
i
.
Proof. We start by doing a nger move as in Figure 8.1. That is, let c
1
. . . . . c
4k
denote
the ends of the -arcs intersecting , enumerated (say) counter-clockwise around , with
c
1
and c
4k
forming part of 1
z
. Let / be the -arc on the boundary of 1
z
which intersects
c
1
closest to . Do a nger move pushing / consecutively through c
1
, . . . , c
4k
. In the
resulting diagram, the regions 1
1
. . . . . 1
4k1
adjacent to other than 1
z
are all rectangles.
Now apply the Sarkar-Wang algorithm [SW10], which we recall briey, to the resulting
diagram. Assign to each region 1 its distance, which is the minimum number of -arcs which
a path in connecting 1 and 1
z
must cross. A region is called bad if it is not a bigon
or rectangle. The algorithm is inductive. At each step, choose a bad region 1 of maximal
distance and with minimal badness (i.e., maximal Euler measure) among such regions. Let
/
R
be a -arc separating 1 from a region of smaller distance. Do a nger move, pushing /
R
through one of the -edges of 1, and continue the nger move until reaching either a lower
distance region, a bigon, another bad region, or the region 1 again. In the last case, the
algorithm tells us to either choose a dierent edge through which to push /
R
or perform an
appropriate handleslide; otherwise, one repeats the process. The crucial point for us is that
the nger move will never reach any of the regions 1
1
. . . . . 1
2n1
. The proof in the closed
case thus implies that the process eventually yields a nice bordered Heegaard diagram.
Nice diagrams are automatically admissible:
Lemma 8.3. If 1 = (. . . .) is nice then 1 is admissible.
Proof. The proof is exactly the same as in the closed case ([LMW08, Corollary 3.2]).
One shows that the region containing . occurs both to the left and to the right of each
-circle: if all the regions to the left (say) of some
i
are bigons or rectangles then one sees
that the -circles are homologically linearly dependent.
Sarkar and Wang apply the index formula to show that for nice Heegaard diagrams (in the
closed case), the dierential counts only bigons and rectangles. This fact has the following
analogue for nice bordered Heegaard diagrams.
145
146 8. PAIRING THEOREM VIA NICE DIAGRAMS
Figure 8.1. First step in making a Heegaard diagram with boundary
nice: perform a nger-move to separate o the boundary.
Proposition 8.4. Let 1 be a nice bordered Heegaard diagram. Suppose that ind(1. ) = 1.
Then any holomorphic curve n /
B
(x. y; ) has one of the following three forms:
(1) The source o of n consists of p bigons. Each bigon is either a trivial strip or has a
unique puncture mapped to east . Further, has only one part.
(2) The source o consists of p bigons. All but one of the bigons are trivial strips, and
the other is mapped by

n to the interior of .
(3) The source o consists of p 1 bigons and one quadrilateral. All of the bigons are
trivial strips, and the quadrilateral is mapped by

n to the interior of .
Conversely, suppose the homology class 1 and asymptotic data admit a topological
map satisfying the condition of Case (1), (2), or (3), and ind(1. ) = 1. Then there is a
holomorphic map in /
B
(x. y; ), unique up to translation.
Proof. By Formula (5.9), for a rigid holomorphic curve we have
1 = p (o) + 2c(1(n)) +[1[.
Since o has at most p connected components, (o) p. Since the Heegaard diagram is nice,
c(1(n)) 0. Consequently, one of the following three cases holds:
p = (o), c(1(n)) = 0 and [1[ = 1. This corresponds to Case 1, above.
p = (o), c(1(n)) = 1,2 and [1[ = 0. This corresponds to Case 2, above.
p 1 = (o), c(1(n)) = 0 and [1[ = 0. This corresponds to Case 3, above.
It is clear that in each case the holomorphic representative is unique. The converse,
that such domains admit holomorphic representatives, follows from the Riemann mapping
theorem.
Remark 8.5. In the closed case, Sarkar and Wang show that for a rigid holomorphic curve
n in a nice diagram, the image of

n is an empty, embedded bigon or rectangle [SW10,

Theorem 3.3]. The analogue in the bordered case is that the domain of a rigid curve is an
empty embedded bigon, rectangle or half strip (rather than, say, merely immersed). This
follows immediately from the Proposition 8.4 and the result in the closed case, by doubling
the Heegaard diagram along the boundary and considering the doubled domain.
We do not need this stronger result. Proposition 8.4 already gives an algorithm for
counting holomorphic curves in a nice diagram: one checks which domains have index 1, and
are represented by orientation-preserving maps of a bigon, rectangle or half-strip. (This is
8. PAIRING THEOREM VIA NICE DIAGRAMS 147
analogous to the approach taken in [LMW08] for holomorphic triangles.) Proposition 8.4
also suces to prove the pairing theorem, our main goal.
Corollary 8.6. Let 1 be a nice Heegaard diagram with boundary, and

CFA =

CFA(1) the
associated type module. Then the higher /

operations :
i
(i 2) vanish on

CFA.
Proof. The higher multiplications :
i
(i 2) count rigid holomorphic curves in moduli
spaces /
B
(x. y; ) with [[ 1. But by Proposition 8.4, in a nice diagram there are no
such curves.
As a consequence of these remarks, we have a quick proof of Theorem 1.3.
Proof of Theorem 1.3. Let 1
1
= (
1
.
1
.
1
. .) and 1
2
= (
2
.
2
.
2
. .) be nice
Heegaard diagrams with boundary for 1
1
and 1
2
respectively. Their union 1 = (
1

2
.
1

2
.
1

2
. .) is a Heegaard diagram for 1 = 1
1

F
1
2
. We want to show that

CF(1)

=

CFA(1
1
)

CFD(1
2
).
It is easy to see that 1 is a nice diagram, since 1
1
and 1
2
are. Sarkar and Wang [SW10]
showed that the dierential for

CF(1) counts index 1 rectangles and bigons. (This is also a

special case of Proposition 8.4.) We compare this complex with

CFA(1
1
)

CFD(1
2
) from
Section 2.4, which by Proposition 2.34 serves as a model for the derived tensor product

.
(Note, in particular, that since 1
2
is admissible,

CFD(1
2
) is bounded.)
There is an obvious identication of generators of

CFA(1
1
)

CFD(1
2
) and

CF(1). Let
x
1
x
2


CFA(1)

CFD(1
2
). (Recall that the tensor products are over the ground ring
k = J(:), the subring of idempotents in /(:).)
By Proposition 8.4, the dierential (x
2
) can be written as

y
2
c
x
2
,y
2
y
2
, where each
term in c
x
2
,y
2
is either 1coming from a provincial rectangle or bigonor an algebra element
in which a single strand moves. Applying Corollary 8.6, the dierential in

CFA(1
1
)

CFD(1
2
) is given by

(x
1
x
2
) = (:
1
(x
1
)) x
2
+

y
2
:
2
(x
1
. c
x
2
,y
2
) y
2
.
The rst term in the sum counts holomorphic bigons and rectangles contained entirely in
(
1
.
1
.
1
. .). The second sum counts both rectangles and bigons contained entirely in
(
2
.
2
.
2
. .) (corresponding to terms where c
x
2
,y
2
is 1) and rectangles in (
1

2
.
1

2
.
1

2
. .) crossing
1
(corresponding to terms with a single moving strand in c
x
2
,y
2
).
But, since 1 is a nice diagram, this is exactly the dierential on

CF(1).
Remark 8.7. The idea to use the Sarkar-Wang algorithm to prove a gluing result is partly
inspired by work of Juhsz [Juh08].
Remark 8.8. Since for 1
1
a nice diagram

CFA(1
1
) is an ordinary /-module, we can
also consider the ordinary tensor product

CFA(1
1
)

CFD(1
2
). It is easy to see that this
agrees, on the one hand, with

CFA(1
1
)

CFD(1
2
). On the other hand, it is straightfor-
ward to check that

CFD(1
2
) is a projective /-module, and thus

CFA(1
1
)

CFD(1
2
)

=

CFA(1
1
)

CFD(1
2
). This gives an alternative to the use of Proposition 2.34.
With a little extra work, one can establish a graded version of the pairing theorem along
these lines; see Section 10.6.
CHAPTER 9
Pairing theorem via time dilation
The aim of the present section is to give an alternate proof of Theorem 1.3, which de-
scribes

HF of a three-manifold which is separated by a surface in terms of the bordered
invariants of the two pieces. Whereas the proof from Chapter 8 used special Heegaard dia-
grams, the current proof starts from any (suitably admissible) diagrams for the two pieces.
Our motivation for including this second proof is twofold. First, the proof illuminates the
connection between the analysis of holomorphic curves and the algebra underlying the def-
initions of the bordered Floer modules and their /

tensor products. Second, the proof

applies to other contexts, such as counting higher polygon maps; see [LOTa].
In slightly more detail, let 1 be a parameterized surface, and let 1
1
and 1
2
be three-
manifolds with compatibly parameterized boundaries, 1
1
= 1(:) = 1
2
= 1(:). Let
1 = 1
1

F
1
2
. Any two (suitably admissible) diagrams 1
1
for 1
1
and 1
2
for 1
2
can be
glued together to form an admissible Heegaard diagram 1 for 1 . Standard gluing theory
describes the moduli spaces counted in the dierential for

CF(1) in terms of bered products

of moduli spaces from 1
1
and 1
2
. Starting from this description, the dierential for

CF(1)
is then deformed to the dierential on the derived tensor product by a procedure we call
time dilation.
More precisely, we construct a family of chain complexes

CF(1; 1
1
. 1
2
) from bered
products of the moduli spaces for 1
1
and 1
2
, indexed by a real parameter 1 0 which,
informally, speeds up the time parameter t on the type 1 side (i.e., 1
2
), as compared
to time on the type side (1
1
). Standard gluing theory identies

CF(1; 1
1
. 1
2
) with

CF(1
1

1
2
). A continuation argument shows that the chain homotopy type of the complex

CF(1; 1
1
. 1
2
) is independent of 1. A further compactness and gluing argument shows that
for all suciently large 1, the behavior of the complex stabilizes. In the large 1 limit, the
complex is then identied with the product, in the sense of Denition 2.26, of the chain
complexes

CFA(1
1
) and

CFD(1
2
). Since this in turn is identied with the derived tensor
product (Proposition 2.34), Theorem 1.3 follows.
This chapter is organized as follows. In Section 9.1 we explain how, via a neck-stretching
argument, holomorphic curves for 1
1

1
2
correspond to certain pairs of holomorphic
curves for 1
1
and 1
2
(for suitable almost complex structures). In Section 9.2 we introduce
the chain complexes

CF(1; 1
1
. 1
2
) and show that they are all chain homotopy equivalent.
In Section 9.3 we study the behavior of holomorphic curves as the dilation parameter 1 goes
to . This is the most technically involved part of the chapter. In Section 9.4 we combine
the results from the previous sections to prove Theorem 1.3. Finally, in Section 9.6 we give a
local example of the steps in the argument; the reader may want to read this simple example
before delving into the details of the proof.
149
150 9. PAIRING THEOREM VIA TIME DILATION
9.1. Moduli of matched pairs
Let 1
1
= (
1
.
1
.
1
. .) and 1
2
= (
2
.
2
.
2
. .) be provincially admissible Heegaard
diagrams for 1
1
and 1
2
of genus p
1
and p
2
, respectively, inducing compatible markings of 1;
i.e., 1
1
= 1
2
. Let =
1

2
, =
1

2
, and =
1

2
. Assume that one of 1
1
or 1
2
is admissible. Then, by Lemma 4.33, (. . . .) is an admissible Heegaard diagram
(denoted 1) for 1 . Recall from Section 4.5 that generators S(1) of

CF(1) correspond to
certain elements (x
1
. x
2
) of S(1
1
) S(1
2
), called compatible pairs of generators; these are
the pairs (x
1
. x
2
) so that 1
A
(x
1
) = 1
D
(x
2
). Moreover, by Lemma 4.32,
2
(x. y) is naturally
a subset of
2
(x
1
. y
1
)
2
(x
2
. y
2
).
Denition 9.1. By a compatible pair of decorated sources we mean decorated sources o

1
and o

2
together with a bijection between 1(o

1
) and 1(o

2
), such that for each puncture

i
of o

1
, the Reeb chord labeling (
i
) is the orientation reverse of the Reeb chord labeling

i
. A compatible pair of sources o

1
and o

2
can be preglued to form a surface o

:= o

1
o

2
,
by identifying a neighborhood of
i
with a neighborhood of (
i
) for each i.
Denition 9.2. Fix two compatible pairs of generators x
i
. y
i
S(1
i
) (i = 1. 2), and
homology classes 1
1

2
(x
1
. y
1
) and 1
2

2
(x
2
. y
2
) inducing a homology class 1

2
(x. y). Fix a compatible pair of sources o

1
and o

2
connecting x
1
to y
1
and x
2
to y
2
respectively. Then dene the moduli space of matched pairs to be the bered product

//
B
(x
1
. y
1
. o

1
; x
2
. y
2
. o

2
) :=

/
B
1
(x
1
. y
1
; o

1
)
ev
1
=ev
2

/
B
2
(x
2
. y
2
; o

2
).
That is, the moduli space of matched pairs consists of pairs (n
1
. n
2
) with n
i


/
B
i
(x
i
. y
i
; o

i
)
such that ev(n
1
) = ev(n
2
), under the correspondence induced by .
Dene the index of a matched pair by
ind(1
1
. o

1
; 1
2
. o

2
) := ind(1
1
. o

1
. 1) + ind(1
2
. o

2
. 1) :
= p
1
+ p
2
+ 2c(1
1
) + 2c(1
2
) (o
1
) (o
2
) + :.
(9.3)
where 1 is the discrete partition on 1(o

1
) (or equivalently 1(o

2
)) and : = [1[ = [1(o

1
)[.
This is the expected dimension of the moduli space of matched pairs coming from its de-
scription as a bered product of two moduli spaces over 1
m
.
Lemma 9.4. For generic, admissible almost complex structures on
i
[0. 1] 1, the moduli
spaces of matched pairs

//
B
1
B
2
(x
1
. y
1
. o

1
; x
2
. y
2
. o

2
) are transversely cut out by the

-
equation and the evaluation map. Thus, the moduli space is a manifold whose dimension is
given by ind(1
1
. o

1
; 1
2
. o

2
).
Proof. For the transversality statement, by Proposition 5.6, for a generic almost com-
plex structure J
1
on
1
, the moduli spaces of holomorphic curves

/
B
1
(x
1
. y
1
; o

1
) are
transversally cut out for all (countably many) choices of x
1
, y
1
, 1
1
and o

1
. Again by
Proposition 5.6, for a generic almost complex structure J
2
on
2
, the moduli spaces of
holomorphic curves

/
B
2
(x
2
. y
2
; o

2
), for all choices of 1
2
, x
2
, y
2
and o

2
, are transversally
cut out and the evaluation maps ev:

/
B
2
(x
2
. y
2
; o

2
) 1
m
are transverse to the images
ev(

/
B
1
(x
1
. y
1
; o

1
)) for all choices of 1
1
, x
1
, y
1
and o

1
.
The statement about the expected dimension is immediate from the denitions and the
index formula, Formula (5.9).
9.1. MODULI OF MATCHED PAIRS 151
Lemma 9.5. Fix a compatible pair of generators x
1
and x
2
with x
i
S(1
i
). Let (n
1
. n
2
)
be any matched pair, i.e., n
i


/
B
i
(x
i
. y
i
; o

i
) such that ev(n
1
) = ev(n
2
), under the corre-
spondence induced by . Then, both n
1
and n
2
are strongly boundary monotone.
Proof. The condition that x
1
and x
2
can be combined to give a generator of the glued di-
agram is equivalent to the condition that o(x
1
) Ho(x
2
) = [2/]. Now, let = (
1
. . . . .
n
)
denote the sequence of (singleton sets of) Reeb chords in order on the n
1
side. Of course,
we encounter the same sequence of chords, only with orientations reversed, in the same
order on the n
2
side. Continuity of n
1
and n
2
ensures that `(

i+1
) o(x
1
.
[1,i]
) and
`(

i+1
) o(x
2
.
[1,i]
). So, Lemma 6.8 implies the result.
Translation induces an 1-action on

//
B
(x
1
. y
1
. o

1
; x
2
. y
2
. o

2
). This action is free
except where both sides of the matching are trivial strips. When the action is free we say the
curves in the moduli space are stable, and denote the quotient

//
B
(x
1
. y
1
. o

1
; x
2
. y
2
. o

2
),1
by //
B
(x
1
. y
1
. o

1
; x
2
. y
2
. o

2
).
Given a compatible pair o

1
and o

2
with o

= o

1
o

2
, consider the moduli space of holo-
morphic curves /
B
(x. y; o

) for the diagram 1. The expected dimension of /

B
(x. y; o

)
is given by
ind(1. o

) := p (o) + 2c(1).
(This is [Lip06a, Formula (6)], which is a special case of Equation (5.9).) If [1(o

1
)[ = :,
we have (o
1
o
2
) = (o
1
) +(o
2
) :. It follows from Formula (9.3) that the index of the
moduli space of curves from x to y represented by the pre-glued source o

coincides with
the index of the moduli space of matched pairs, i.e.,
ind(1
1
. o

1
; 1
2
. o

2
) = ind(1
1
1
2
. o

1
o

2
).
A stronger identication occurs at the level of moduli spaces.
Proposition 9.6. There are generic admissible complex structures on [0. 1] 1 and
on
i
[0. 1] 1 for i = 1. 2 with the property that for all x. y S(1), homology classes
1
2
(x. y), and sources o

with ind(1. o

) = 1, the number of elements in /

B
(x. y. o

)
is equal (modulo 2) to the number of elements in
_
S

=S

1
S

2
//
B
(x
1
. y
1
. o

1
; x
2
. y
2
. o

2
).
where x = x
1
x
2
and y = y
1
y
2
.
Proof. The result follows from standard compactness and gluing techniques, in a com-
pletely analogous way to Propositions 5.23 (but using [BEH
+
03, Theorem 10.3] in place
of [BEH
+
03, Theorem 10.2]) and 5.25 (using a Morse-Bott analogue of [Lip06a, Proposi-
tion A.2], say).
As always, the curves of primary importance to us are the embedded ones. For a Heegaard
diagram 1 for a closed manifold with genus p, x. y S(1), and 1
2
(x. y), dene

emb
(1) := p + c(1) :
x
(1) :
y
(1)
ind(1) := c(1) + :
x
(1) + :
y
(1).
(This is a special case of Denition 5.61.)
152 9. PAIRING THEOREM VIA TIME DILATION
Denition 9.7. The embedded matched moduli space

//
B
(x
1
. y
1
; x
2
. y
2
) is the union of

//
B
(x
1
. y
1
. o

1
; x
2
. y
2
. o

2
) over all pairs o

1
, o

2
with (o

1
o

2
) =
emb
(1). (Note there are
only nitely many terms in this union.) Also dene //
B
(x
1
. y
1
; x
2
. y
2
) to be the quotient
of

//
B
(x
1
. y
1
; x
2
. y
2
) by the 1-action (assuming 1 ,= 0, so the 1-action is free).
It follows from Lemma 9.4 that for generic J
1
and J
2
, //
B
(x
1
. y
1
; x
2
. y
2
) is a manifold
of dimension ind(1) 1.
Lemma 9.8. A matched pair of curves (n
1
. n
2
) is in the corresponding embedded matched
moduli space

//
B
(x
1
. y
1
; x
2
. y
2
) if and only if both n
i
are embedded and every pair of Reeb
chords appearing at the same height in the n
i
are either nested or disjoint.
Proof. Let (n
1
. n
2
) be any matched pair of curves. Let o

i
be the source of n
i
, and
let be sequence of sets of Reeb chords on n
1
. Let : = [1(o

1
)[ and 1
1
1
2
= 1. By
Proposition 5.62, (o
1
)
emb
(1
1
. ) and (o
2
)
emb
(1
2
. ).
We claim that
() + () :.
as follows. The 1(
i
.
j
) terms in the denition of (Formula (5.58)) contribute oppositely
to () and (), as the orientation on the circle is reversed, so the only contributions to
() +() are from the (
i
) terms. From Lemma 5.57 we see that each chord contributes
1, and each pair of chords
1
,
2
at the same level has a further negative contribution of
2[1(
1
.
2
)[, from which the claim follows.
Thus
(o

1
o

2
) = (o
1
) + (o
2
) :

emb
(1
1
. ) +
emb
(1
2
. ) :
= p + c(1) :
x
(1) :
y
(1) () () :

emb
(1).
The pair (n
1
. n
2
) is in the embedded moduli space if and only if we have equality at each step,
which is equivalent to both n
1
and n
2
being embedded and all [1(
1
.
2
)[ terms vanishing.
These last terms vanish if and only if for all
1
and
2
on the same level,
1
and
2
are nested
or disjoint, as desired.
Denition 9.9. Let

CF(1
1
. 1
2
) be the F
2
-vector space with basis those x
1
x
2
S(1
1
)
S(1
2
) so that x
1
x
2
is in S(1
1

1
2
). Dene a boundary operator
1
on

CF(1
1
. 1
2
) by

1
(x
1
x
2
) :=

y
1
,y
2

B[ind(B)=1
#
_
//
B
(x
1
. y
1
; x
2
. y
2
)
_
(y
1
y
2
).
Since, by Lemma 4.33, 1
1

1
2
is admissible, the argument from Lemma 6.5 shows that
this sum is nite.
Theorem 9.10.

CF(1
1
. 1
2
) is a chain complex isomorphic to the complex

CF(1
1

1
2
),
for suitable choices of almost-complex structures.
Proof. According to [Lip06a, Theorem 2],

CF(1
1

1
2
) can be calculated by the chain
complex whose dierential counts points in /
B
(x. y. o

), where we take the union over all

choices of 1
2
(x. y) and compatible o

so that ind(o

. 1) = ind(1) = 1. According to
Proposition 9.6, these counts agree with
1
.
9.2. DILATING TIME 153
Remark 9.11. Theorem 9.10 shows, in particular, that
2
1
= 0. This also follows as a special
case of Proposition 9.19, below.
9.2. Dilating time
We now generalize the notion of matched pairs by inserting a real parameter 1.
Denition 9.12. Fix a real number 1 0. Fix two compatible pairs of generators x
i
. y
i

S(1
i
) (i = 1. 2). Fix compatible sources o

1
and o

2
connecting x
1
to y
1
and x
2
to y
2
respectively. The moduli space of 1-matched pairs

//
B
(1; x
1
. y
1
. o

1
; x
2
. y
2
. o

2
) is dened
to be the bered product

/
B
1
(x
1
. y
1
; o

1
)
Tev
1
=ev
2

/
B
2
(x
2
. y
2
; o

2
).
That is, the moduli space of 1-matched pairs consists of pairs (n
1
. n
2
) with n
i


/
B
i
(x
i
. y
i
;
o

i
) and 1 ev(n
1
) = ev(n
2
). Let

//
B
(1; x
1
. y
1
; x
2
. y
2
) :=
_
S

1
,S

2
[(S

1
S

2
)=
emb
(B)

//
B
(1; x
1
. y
1
. o

1
; x
2
. y
2
. o

2
)
denote the embedded moduli space of 1-matched pairs.
The index of a 1-matched pair does not depend on the 1-parameter, and so is still given
by ind(1
1
. o

1
; 1
2
. o

2
) as dened in Equation (9.3).
Denition 9.13. The moduli space of 1-matched pairs has an 1-action given by
t
(n
1
. n
2
) :=
(
Tt
(n
1
).
t
(n
2
)). Let
//
B
(1; x
1
. y
1
. o

1
; x
2
. y
2
. o

2
) :=

//
B
(1; x
1
. y
1
. o

1
; x
2
. y
2
. o

2
),1
//
B
(1; x
1
. y
1
; x
2
. y
2
) :=

//
B
(1; x
1
. y
1
; x
2
. y
2
),1.
Lemma 9.14. For generic admissible almost complex structures J
i
on
i
[0. 1] 1, for
all choices of x
i
. y
i
S(1
i
) and 1
2
(x
1
x
2
. y
1
y
2
) so that 1 ,= 0,
for generic values of 1, //
B
(1; x
1
. y
1
; x
2
. y
2
) is a manifold of dimension ind(1)
1, and

T>0
//
B
(1; x
1
. y
1
; x
2
. y
2
) is a manifold of dimension ind(1).
Proof. As in Lemma 9.4, the rst statement follows from Proposition 5.6 and Equa-
tion (5.9). The proof of the second statement proceeds in a standard way; see, for instance,
[MS04, Section 3.4] for a nice explanation of this type of argument.
Denition 9.15. For 1 (0. ), let

CF(1; 1
1
. 1
2
) be the vector space with the same
basis as

CF(1
1
. 1
2
) from Denition 9.9 and with boundary operator

T
:

CF(1; 1
1
. 1
2
)

CF(1; 1
1
. 1
2
)
dened by

T
(x
1
x
2
) :=

y
1
,y
2

B[ind(B)=1
#
_
//
B
(1; x
1
. y
1
; x
2
. y
2
)
_
(y
1
y
2
).
154 9. PAIRING THEOREM VIA TIME DILATION
The next goal is to show that (
T
)
2
= 0. As usual, this requires considering the ends of
//
B
(1; x
1
. y
1
; x
2
. y
2
) when ind(1) = 2. To this end, we consider a compactication of
//
B
(1; x
1
. y
1
. o

1
; x
2
. y
2
. o

2
) by holomorphic combs, following Section 5.4.
Denition 9.16. Given x
i
. y
i
S(1
i
) for i = 1. 2, a 1-matched story from x = (x
1
. x
2
) to
y = (y
1
. y
2
) is a sequence (n
1
.
1
. . . . .
k
. n
2
) where
n
1
/
B
1
(x
1
. y
1
; o

1
), n
2
/
B
2
(x
2
. y
2
; o

2
), and
j
A(1

j
) for some sources o

i
and 1

j
;
the sources are equipped with bijective correspondences between 1(o

1
) and \(1

1
),
1(1

i
) and \(1

i+1
), and 1(1

k
) and 1(o

2
);
the correspondences preserve the labelings by Reeb chords, with orientation reversal
between 1(1

k
) and 1(o

2
); and
the following compatibility conditions hold:
ev(n
1
) = ev
w
(
1
).
ev
e
(
i
) = ev
w
(
i+1
) and
1 ev
e
(
k
) = ev(n
2
).
We call n
1
the west-most level of the 1-matched story and n
2
the east-most level of the
1-matched story. A 1-matched story is stable if either n
1
or n
2
is stable. (In particular, if
[1(o

1
)[ 0, the comb is automatically stable.)
A 1-matched comb of height ` is a sequence of stable 1-matched stories running from
x
j
= (x
j
1
. x
j
2
) to x
j+1
= (x
j+1
1
. x
j+1
2
) for some sequences of generators x
j
1
in 1
1
and x
j
2
in 1
2
(for , = 1. . . . . `).
As in Section 5.4, there is a natural compactication //
B
(x
1
. y
1
. o

1
; x
2
. y
2
. o

2
) of
//
B
(x
1
. y
1
. o

1
; x
2
. y
2
. o

2
) inside the space of stable 1-matched holomorphic combs. The
following is the analogue of Proposition 5.37, and the proof is similar.
Proposition 9.17. Suppose that ind(1
1
. o

1
; 1
2
. o

2
) = 2. Then for generic J, every 1-
matched comb in //
B
(1; x
1
. y
1
. o

1
; x
2
. y
2
. o

2
) has one of the following forms:
(1) a two-story 1-matched comb;
(2) a 1-matched comb (n
1
. . n
2
) where the only non-trivial component of is a join
component; or
(3) a 1-matched comb (n
1
. . n
2
) where the only non-trivial component of is a split
component.
Proof. The proof of Lemma 9.5 applies to show that the components of a 1-matched
pair are strongly monotone, as well. Thus, boundary double points cannot appear on the
boundaries of the 1-matched moduli spaces, owing to the strong boundary monotonicity of
the moduli spaces, Lemma 5.50. Boundary degenerations are ruled out by Lemma 5.42, and
ghost curves have codimension 2 as in Lemma 5.51.
Now suppose that we have a one-story 1-matched comb (n
1
.
1
. . . . .

. n
2
) in the bound-
ary of the moduli space, with source (o

1
t
. 1

1
. . . . . 1

. o

2
t
). Let 1

= 1

1
. . . 1

. By the
index hypothesis,
2 = p (o

1
) (o

2
) + 2c(1) + :
0
.
where :
0
= [1(o

1
)[ = [1(o

2
)[. We also have o

1
o

= o

1
t
1

2
t
and so
(o

1
) + (o

2
) :
0
= (o

1
t
) + (1

) + (o

2
t
) :
1
:
2
9.2. DILATING TIME 155
where :
1
= [1(o

1
t
)[ and :
2
= [1(o

2
t
)[. Let / be the number of components of 1

. Let
1
i
be the partition of 1(o

i
t
) according to the components of 1

, so that [1
i
[ = /. Then an
upper bound for the dimension of the space of limit curves for generic J is
ind(1
1
. o

1
t
. 1
1
) + ind(1
2
. o

2
t
. 1
2
) / 1
= p (o

1
t
) (o

2
t
) + 2c(1) + / 1
= (o

1
) (o

1
t
) + (o

2
) (o

2
t
) :
0
+ / + 1
= ((1

) /) + (/ :
1
) + (/ :
2
) + 1.
All three terms in parentheses are non-positive, so if the space of limit curves is non-empty,
at most one can be negative. If all are equal to 0, 1

consists of trivial strips and there is

no degeneration. If (1

) < /, one component of 1

is an annulus, which is ruled out as in

Proposition 5.37. If :
1
= / +1 or :
2
= / +1, then 1

is a join or split curve, respectively.

The arguments above also show that the index of a story in the degenerate curve is at
least 1, and that if the index is 1 there are no components at east innity. Thus the only
other possibility for a degeneration is a two-story matched curve.
In our setting, the matched combs with components at east innity cancel in pairs, as
boundary branch points move between
1
and
2
. Roughly speaking, the next proposition
states that rigid 1-matched combs (n
1
. . n
2
) appear with multiplicity one in the boundaries
of their corresponding one-dimensional moduli spaces.
Proposition 9.18. Let (n
1
. . n
2
) be a 1-matched comb of index two with source (o

1
, 1

,
o

2
), where 1

has one non-trivial component, which is either a join or split component.

Then there are arbitrarily small open neighborhoods l
1
of (n
1
. . n
2
) in
/
B
1
(x
1
. y
1
. o

1
1

) /
B
2
(x
2
. y
2
. o

2
)
with the property that l
1
meets //
B
(1; x
1
. y
1
. o

1
1

; x
2
. y
2
. o

2
) in an odd number of
points. Similarly, there are arbitrarily small open neighborhoods l
2
of (n
1
. . n
2
) in
/
B
1
(x
1
. y
1
. o

1
) /
B
2
(x
2
. y
2
. 1

2
)
with the property that l
2
meets //
B
(1; x
1
. y
1
. o

1
; x
2
. y
2
. 1

2
) in an odd number of
points.
Proof. Suppose that is a split curve. We label the east punctures of o

1
1

by c
i

i=0
so that c
0
and c
1
are the two east punctures which are in the same component in the split
curve. Let u
i

i=0
be the corresponding punctures of o

2
. There are open neighborhoods
`
1
/
B
1
(x
1
. y
1
; o

1
1

) of (n
1
. ) and `
2
/
B
2
(x
2
. y
2
; o

2
) of n
2
which admit evaluation
maps
ev
1
: `
1
1

ev
2
: `
2
1

.
where the i
th
coordinate of ev
2
is ev
w
i
,w
0
(as in Equation (5.12)), and the i
th
coordinate of
ev
1
is 1 ev
e
i
,e
0
. Let
)
j
: `
j
1
be the rst component of ev
j
for , = 1. 2, i.e., )
2
= ev
w
1
ev
w
0
and )
1
= ev
e
1
ev
e
0
.
By the assumptions on the index, dim`
1
+dim`
2
= / +1, so if we knew that the maps
ev
1
and ev
2
were smooth (and transverse) it would follow that the images intersect in a
156 9. PAIRING THEOREM VIA TIME DILATION
1-manifold, which would prove the result. The diculty is that we do not know that ev
1
is
smooth on the boundary. Instead, we will use linking number considerations.
Let l be a small neighborhood of n
1
inside /
B
1
(x
1
. y
1
; o

1
). Gluing (Proposition 5.25)
gives a map
: l [0. c] `
1
/
B
1
(x
1
. y
1
; o

1
1

)
which is a homeomorphism onto a neighborhood of (n
1
. ). Shrinking l slightly, we may
assume that is dened on l [0. c]. Note that (l 0) consists of pairs (n
t
1
.
t
) where

t
is a split curve whose height is determined by n
t
1
. Choose l so that
(1) l is a ball contained in /
B
1
(x
1
. y
1
; o

1
) /
B
1
(x
1
. y
1
; o

1
) and
(2) l is suciently small that ev
1
((l 0)) ev
2
(`
2
) consists of the single point
ev
1
((n
1
. )) = ev
1
((n
1
0)).
To see that the second requirement can be met, note that ev
1
((l 0)) = 0 ev
1
(l);
and by Proposition 5.6 we may assume that ev
2
is transverse to this submanifold (compare
Lemma 9.4).
The rest of the argument is similar to the proof of Proposition 5.33. The intersection
ev
1
((l 0)) ev
2
(`
2
) is empty and hence, for suciently small c:
(3) The intersection ev
1
((l [0. c])) ev
2
(`
2
) is empty.
Since (l c) is contained in /
B
1
(x
1
. y
1
; o

1
1

) /
B
1
(x
1
. y
1
; o

1
1

), )
1
is positive
on (l c). Therefore:
(4) For suciently small 0, ev
2
()
1
2
((. ])) is disjoint from ev
1
((l c)).
Consider now the ball 1
1
= l [0. c] and choose a closed ball 1
2
`
2
which is a neighbor-
hood of n
2
and contained in )
1
2
((. ]). Combining (2), (3) and (4), ev
1
(1
1
) intersects
ev
2
(1
2
) only in the single point ev
1
(n
1
. ), which is in the image ev
2
(1

2
) of the interior 1

2
of 1
2
. Therefore linking number considerations show that ev
1
(1
1
) ev
2
(1
2
) consists of an
odd number of points. We can therefore take l
1
in the statement of the proposition to be
(l [0. c)) 1

2
: then, the intersection of l
1
with //
B
(1; x
1
. y
1
. o

1
1

; x
2
. y
2
. o

2
)
consists of the point (n
1
. . n
2
), from ((l [0. c))) 1

2
, which is not in //
B
(1;
x
1
. y
1
. o

1
1

; x
2
. y
2
. o

2
), and an odd number of other points, from (l [0. c)) (1
2
).
A similar analysis holds for the other moduli space or if has a join component.
Proposition 9.19. The map
T
is a dierential.
Proof. Fix x
i
. y
i
S(1
i
) and a homology class 1
2
(x
1
x
2
. y
1
y
2
) with ind(1) =
2. Consider the ends of the moduli space /
B
(1; x
1
. x
2
; y
1
. y
2
) of embedded 1-matched
holomorphic curves representing 1. According to Proposition 9.17, these ends correspond
either to simple matched combs, which cancel in pairs according to Proposition 9.18, or to
two-story holomorphic buildings, with each story an index one 1-matched curve. Indeed,
each possible two-story building appears as an end an odd number of times, according to an
extension of Proposition 5.24 along the lines of Proposition 9.18. Hence, the number of such
two-story buildings must be even. But the number of such terms (over all homology classes
1 with index two) is the y
1
y
2
coecient of (
T
)
2
(x
1
x
2
).
The chain complex

CF(1; 1
1
. 1
2
) is

CF(1
1

1
2
) by Theorem 9.10. We will next show
that the homotopy type of

CF(1; 1
1
. 1
2
) is independent of the choice of 1.
9.2. DILATING TIME 157
Denition 9.20. Let 1
1
and 1
2
be positive real numbers, and x a smooth function :
1 1 with positive derivative so that
(t) =
_
1
1
t t 1
1
2
t t 1.
There are induced maps
m
: 1
m
1
m
dened by
m
(t
1
. . . . . t
m
) = ((t
1
). . . . . (t
m
)). Fix
two compatible pairs of generators x
i
. y
i
S(1
i
) (i = 1. 2). Fix compatible sources o

1
and
o

2
connecting x
1
to y
1
and x
2
to y
2
respectively. The moduli space of -matched pairs,
denoted
//
B
(; x
1
. y
1
. o

1
; x
2
. y
2
. o

2
)
is dened as the bered product
/
B
1
(x
1
. y
1
; o

1
)

m
ev
1
=ev
2
/
B
1
(x
2
. y
2
; o

2
).
where : = [1(o

1
)[. That is, the moduli space consists of pairs (n
1
. n
2
) with n
i
/
B
i
(x
i
. y
i
;
o

i
) such that
m
ev(n
1
) = ev(n
2
). Let
//
B
(; x
1
. y
1
; x
2
. y
2
) :=
_
S

1
,S

2
[(S

1
S

2
)=
emb
(B)
//
B
(; x
1
. y
1
. o

1
; x
2
. y
2
. o

2
)
denote the space of embedded -matched curves.
The space of -matched pairs has a natural compactication by -matched combs, de-
ned analogously to Denition 9.16, except using the map induced by rather than simply
rescaling on the left side.
The moduli spaces of -matched pairs do not usually have a natural action by 1.
For as above, dene
1
T
1
,T
2
:

CF(1
1
; 1
1
. 1
2
)

CF(1
2
; 1
1
. 1
2
)
by
1
T
1
,T
2
(x
1
x
2
) =

y
1
,y
2

B[ind(B)=0
#
_
//
B
(; x
1
. y
1
; x
2
. y
2
)
_
(y
1
y
2
).
We sketch now familiar arguments (generalizing the above proof of Proposition 9.19)
which show that 1
T
1
,T
2
is a chain map.
Lemma 9.21. Suppose that /
B
(; x
1
. y
1
. o

1
; x
2
. y
2
. o

2
) is 1-dimensional. Then for a
generic almost complex structure J, every -matched comb in /
B
(; x
1
. y
1
. o

1
; x
2
. y
2
. o

2
)
has one of the following forms:
(1) a two-story matched curve (n
t
1
. n
t
2
) (n
tt
1
. n
tt
2
), where (n
t
1
. n
t
2
) is a -matched curve
and (n
tt
1
. n
tt
2
) is a 1
2
-matched curve;
(2) a two-story matched curve (n
t
1
. n
t
2
) (n
tt
1
. n
tt
2
), where (n
t
1
. n
t
2
) is a 1
1
-matched curve
and (n
tt
1
. n
tt
2
) is a -matched curve;
(3) a -matched comb (n
1
. . n
2
) where the only non-trivial component of is a join
component; or
(4) a -matched comb (n
1
. . n
2
) where the only non-trivial component of is a split
component.
Proof. This follows along the same lines of Proposition 9.17 above. (Observe, once
again, that the proof of Lemma 9.5 applies to show that -matched combs also consist of
boundary monotone factors.)
158 9. PAIRING THEOREM VIA TIME DILATION
Proposition 9.22. 1
T
1
,T
2
induces a chain homotopy equivalence from

CF(1
1
; 1
1
. 1
2
) to

CF(1
2
; 1
1
. 1
2
). In particular, for any 1 0,

CF(1; 1
1
. 1
2
) is homotopy equivalent to

CF(1).
Proof. For xed x
i
. y
i
S(1
i
) and homology class 1 with ind(1) = 1, consider
the moduli space /
B
(; x
1
. y
1
; x
2
. y
2
). Lemma 9.21 accounts for the ends of these moduli
spaces. A straightforward adaptation of Proposition 9.18 shows that the ends enumerated in
Lemma 9.21 which do not a priori cancel in pairs are the two-story -matched curves. Gluing
as in the proof of Proposition 9.19 shows that those ends are in one-to-one correspondence
with the types of two-story buildings enumerated in Lemma 9.21. But these counts, which
must add up to zero, are precisely the coecients of y
1
y
2
in
T
2
1
T
1
,T
2
(x
1
x
2
) +1
T
1
,T
2

T
1
(x
1
x
2
).
Showing that 1
T
1
,T
2
is a chain homotopy equivalence also follows along familiar lines in
Floer homology. The homotopy inverse is provided by 1
T
2
,T
1
dened with a suitable function

21
, and the chain homotopy to the identity is provided by counting curves matched by a
suitable family of functions, depending on a parameter c, interpolating between multiplica-
tion by 1
1
and a function which agrees with (t) for t suciently small, and
21
(t +c) for
t suciently large.
The second statement is immediate from this and Theorem 9.10.
9.3. Dilating to innity
Now that we have established suitable independence of

CF(1; 1
1
. 1
2
) from 1, we turn to
the large 1 behavior of this complex. The work, of course, is in understanding the behavior
of 1-matched curves as 1 . In Section 9.3.1 we prove a compactness result, showing
that 1-matched curves converge to so-called ideal-matched curves as 1 . While general
ideal-matched curves can be fairly complicated, we show in Section 9.3.2 that for rigid moduli
spaces they are quite simple; this special class of ideal-matched curves are called simple ideal-
matched curves. It turns out that simple ideal-matched curves contain too much information
at east to correspond one-to-one with 1-matched curves. In Section 9.3.3 we trim this
extra information and prove the desired gluing result.
9.3.1. Ideal limits of 1-matched curves.
Denition 9.23. Let 1 := [. +] be the standard compactication of 1. For l a
holomorphic comb of height `, dene its time-parameter space 1(l) to be the union of `
copies of 1 modulo identifying + in the i
th
copy with in the (i + 1)
st
copy:
1(l) :=
_
N
_
i=1
1
i
__
_
(+)
i
()
i+1
_
.
When l is the trivial comb (` = 0), let 1(l) be a single point.
Denition 9.24. An ideal-matched holomorphic comb consists of
a pair (l
1
. l
2
) of strongly boundary monotone, stable holomorphic combs, for 1
1
and 1
2
, respectively;
an order-preserving map from the set of stories of l
2
to 1(l
1
); and
a correspondence from 1(l
2
) to 1(l
1
),
so that, for every east-most puncture j of l
2
,
9.3. DILATING TO INFINITY 159
Figure 9.1. Schematic illustration of an ideal-matched comb. On the
left is a two-story comb l
1
. In the center is the time-parameter space 1(l
1
).
On the right is a 4-story comb l
2
. The arrows indicate the map in the
denition of an ideal-matched comb.
if j is labeled by the Reeb chord , then (j) is labeled by and
if j is a puncture on the story C of l
2
, the t-coordinate of (j) is (C).
See Figure 9.1 for a schematic illustration of an ideal-matched comb. As usual, we will
often suppress from the notation. Let the total west source o

1
and total east source o

2
of
(l
1
. l
2
) be the result of pregluing all the sources of l
1
and l
2
, respectively. We view such
an ideal-matched comb as a point in
/
B
1
(x
1
. y
1
; o

1
;

1
1
) /
B
2
(x
2
. y
2
; o

2
; 1
2
)
where x
i
, y
i
, and 1
i
are as usual,

1
1
is the induced partition on 1(o

1
), and 1
2
is the discrete
partition on 1(o

2
).
Two ideal-matched combs are called equivalent if they dier only by translating some of
the components of l
2
and/or some of the components of l
1
together with the map .
Let
//
B
(; x
1
. y
1
. o

1
; x
2
. y
2
; o

2
)
denote the moduli space of equivalence classes of ideal-matched combs in the homology
class 1 connecting x
1
. x
2
to y
1
. y
2
and with total west and east sources o

1
and o

2
. Let
//
B
(; x
1
. y
1
; x
2
. y
2
) :=
_
S

1
,S

2
[(S

1
S

2
)=
emb
(B)
//
B
(; x
1
. y
1
. o

1
; x
2
. y
2
; o

2
).
In the denition, either l
1
or l
2
may be of height 0, in which case the other is an ordinary
provincial comb.
160 9. PAIRING THEOREM VIA TIME DILATION
Denition 9.25. Let l
j

j=1
be a sequence of 1
j
-matched combs, where 1
j
. Let
(l
1
. l
2
) be an ideal-matched holomorphic comb. From each 1
j
-matched comb l
j
extract a
pair of combs n
j
1
and n
j
2
by rst taking n
j
2
to consist of the east-most levels of each story and
n
j
1
to be the rest of each story and then throwing out any stories consisting of trivial strips.
We say that the sequence l
j
converges to (l
1
. l
2
) if the following conditions are satised:
The sequences n
j
1
and n
j
2
converge to l
1
and l
2
, respectively.
For suciently large , the identication between the east-most punctures of n
j
1
and
the east punctures of n
j
2
corresponds to the identication between the east-most
punctures of l
1
and the east-most punctures of l
2
.
For each story C of l
2
, (C) 1(l
1
) is determined as follows. Place marked points
j
j
on the source of n
2
j
so that the sequence j
j
converges to a point on the source
of C. Choose marked points
j
in the corresponding story of the source of n
1
j
so
that
1
j
t(n
1
j
(
j
)) = t(n
2
j
(j
j
)).
Then (C) is the limit of t(n
1
j
(
j
)) (which may be ).
It is clear that if the third condition of Denition 9.25 is satised for one choice of
sequences j
j
and
j
for the story C then it holds for any choice of sequences j
j
and
j
. If
C has an c puncture then (C) is given by the 1-coordinate of the corresponding c
puncture of l
1
, so is determined by the rest of the data in this case. By contrast, if C
is provincial then (C) is not determined by the rest of the data. Also, (C) = is only
possible for provincial stories C of l
2
.
Proposition 9.26. Fix decorated sources o

1
and o

2
and homology classes 1
1
and 1
2
. Let
(l
j
1
. l
j
2
)

j=1
be a sequence of 1
j
-matched holomorphic combs with total source (o

1
. o

2
) in
the homology class 1
1
1
2
. Suppose that 1
j
. Then (l
j
1
. l
j
2
)

j=1
has a subsequence
which converges to an ideal-matched holomorphic comb, with total source (o

1
. o

2
) and in the
homology class 1
1
1
2
.
Proof. The key point is that, by Proposition 5.23, the sequence (l
j
1
. l
j
2
)

j=1
limits to
some pair of holomorphic combs. Strong boundary monotonicity follows from Lemma 5.49.
Moreover, since the evaluation maps are continuous on the compactied moduli space, the
limit is an -matched (i.e., ideal-matched) holomorphic comb. More details follow.
We prove the claim in the case that each l
j
i
is a holomorphic curve; the general case is
similar, but the notation more complicated. So, let (n
j
1
. n
j
2
) be a sequence of 1
j
-matched
holomorphic curves connecting x
1
x
2
to y
1
y
2
with west source o

1
and east source o

2
. Re-
placing the sequence with a subsequence if necessary, we may assume that there is an ordering

i,1
. . . . .
i,n
of the punctures of o

i
so that if c < / then for any ,, we have ev
q
1,b
,q
1,a
(n
j
1
) 0
(and so also ev
q
2,b
,q
2,a
(n
j
2
) 0).
Consider the space
/
B
1
(x
1
. y
1
; o

1
;

1
1
) /
B
2
(x
2
. y
2
; o

2
;

1
2
) [1. ].
where

1
i
is the sequence of sets of punctures (
i,1
. . . . .
i,n
). By Proposition 5.23, this
space is compact. On the open part, there are evaluation maps ev
i
: /
B
i
(x
i
. y
i
; o

i
;

1
i
)
1
E(S

i
)
,1. We wish to extend these maps to /
B
i
(x
i
. y
i
; o

i
;

1
i
). To this end, identify
9.3. DILATING TO INFINITY 161
1
E(S

i
)
,1 with 1
n1
by using the ordering on the east punctures
i,1
. . . . .
i,n
and considering
dierences of successive coordinates. Thus the map ev
i
is given by
ev
i
(n
i
) = (ev
q
i,2
(n
i
) ev
q
i,1
(n
i
). ev
q
i,3
(n
i
) ev
q
i,2
(n
i
). . . . . ev
q
i,n
(n
i
) ev
q
i,n1
(n
i
))
= (ev
q
i,2
,q
i,1
(n
i
). ev
q
i,3
,q
i,2
(n
i
). . . . . ev
q
i,n
,q
i,n1
(n
i
)).
The image of ev
i
lies inside 1
n1
+
= [0. )
n1
, and the map ev
i
has a continuous extension
to a map ev
i
: /
B
i
(x
i
. y
i
; o

i
;

1
i
) [0. ]
n1
.
Now, consider the map
) : /
B
1
(x
1
. y
1
; o

1
;

1
1
) /
B
2
(x
2
. y
2
; o

2
;

1
2
) [1. ] [0. ]
n1
dened by
)(n
1
. n
2
. 1) = 1 ev
1
(n
1
) ev
2
(n
2
).
The preimage )
1
(0) is the union over all 1 [1. ) of the moduli space of 1-matched
holomorphic combs (with total sources o

1
and o

2
, in the homology class 1
1
1
2
, with the
partial order on the Reeb chords consistent with the order
i,1
. . . . .
i,n
), together with the
ber over 1 = , which we will discuss presently.
The space )
1
(0) is a closed subspace of a compact space, and hence compact. Con-
sequently, the sequence of points (n
j
1
. n
j
2
. 1
j
)

j=1
in )
1
(0) has a convergent subsequence.
The limit (l
1
. l
2
) of this subsequence lies in )
1
(0) 1 = . The limit (l
1
. l
2
) can be
endowed with the structure of an ideal-matched holomorphic comb as follows. The corre-
spondence is inherited from the on each (l
j
1
. l
j
2
). For each story of l
2
with at least
one c puncture, the map is induced from the rest of the structure. For each story n
i
of
l
2
with no c puncture, place a marked point j
i
on the source of n
i
; a sequence of marked
points j
i,j
on the source of n
j
2
converging to j
i
; and a sequence of points
i,j
on n
j
1
with
1
j
t(n
j
1
(
i,j
)) = t(n
j
2
(j
i,j
)). Since the time-parameter space 1(l
1
) is compact, we may take a
further subsequence of the (n
j
1
. n
j
2
) so that for each i, the sequence t(n
j
1
(
i,j
)) converges.
Then takes the i
th
story of l
2
to lim
j
t(
i,j
). It is immediate from the denitions that
our subsequence of (n
j
1
. n
j
2
) converges to (l
1
. l
2
. . ).
Remark 9.27. We have not claimed that the topology corresponding to Denition 9.25 is
Hausdor.
9.3.2. Simple ideal-matched curves, the limits of rigid 1-matched curves. We
will see in this section that there are strong restrictions on which ideal-matched combs come
from rigid 1-matched curves.
Denition 9.28. A simple ideal-matched curve is an ideal-matched comb (l
1
. l
2
) in which
l
1
has at most one story;
l
2
has no components at east innity;
every component of l
1
at east innity is a disk with a single west puncture; and
exactly one of the following two conditions holds:
(1) 1(o

1
) = 1(o

2
) = , one of l
1
or l
2
is the trivial comb of height 0 and the
other of l
1
or l
2
is a height 1 holomorphic comb with index 1, or
(2) each story of l
2
has at least one c puncture and has index 1 (with the dis-
crete partition), the map is injective, and l
1
has index 1 (with the induced
partition).
162 9. PAIRING THEOREM VIA TIME DILATION
In the above denition, the east innity components of l
1
are like split components,
except that there may be any number of east punctures; in the terminology of Section 5.5
they are generalized split components. Note that curves with generalized split components
that are not split components are not transversally cut with respect to the induced partition.
Lemma 9.29. Simple ideal-matched curves have ind(1
1
. o

1
; 1
2
. o

2
) = 1.
Proof. In Case (1) in the denition of simple ideal-matched curves the statement is
obvious. To prove Case (2), let

1 = (1
1
. . . . . 1
n
) denote the induced ordered partition of the
c punctures of o

1
(by their 1-coordinates), and let Q
i
denote the discrete partition of the c
punctures of o

i
. By denition,
ind(1
1
. o

1
.

1) = 1 ind(1
2
. o

2
. Q
2
) = :.
Note that [Q
1
[ = [Q
2
[ =

[1
i
[. By Formula (9.3),
ind(1
1
. o

1
; 1
2
. o

2
) = ind(1
1
. o

1
. Q
1
) + ind(1
2
. o

2
. Q
2
) [Q
1
[
= ind(1
1
. o

1
.

1) + ind(1
2
. o

2
. Q
2
) :
= 1 + : : = 1.
Lemma 9.30. With respect to generic almost complex structures J
1
and J
2
, if ind(1
1
. o

1
;
1
2
. o

2
) < 1 then //
B
(; x
1
. y
1
. o

1
; x
2
. y
2
; o

2
) is empty, and if ind(1
1
. o

1
; 1
2
. o

2
) = 1
then //
B
(; x
1
. y
1
. o

1
; x
2
. y
2
; o

2
) consists of simple ideal-matched curves (l
1
. l
2
).
Proof. This is similar to the proofs of Propositions 5.37 and 9.17. The case with no
east punctures is immediate. Otherwise, suppose rst that l
1
has a single story, and for
notational simplicity assume that l
2
also has a single story, so that all east punctures of
l
1
appear at a single height. Let o

1
t
and o

2
t
be the sources of the main components of
l
1
and l
2
, respectively, and similarly let 1

1
and 1

2
be the result of gluing together the
horizontal levels of the sources of the east components of l
1
and l
2
. Let /
2
be the number
of components of 1

2
, :
0
= [1(o

1
)[ = [1(o

2
)[, and :
i
= [1(o

i
t
)[ for i = 1. 2. Let 1
t
i
be the induced partition on 1(o

i
t
). For generic J, an upper bound for the dimension of
//
B
(; x
1
. y
1
. o

1
; x
2
. y
2
; o

2
) is
ind(1
1
. o

1
t
. 1
t
1
) + ind(1
2
. o

2
t
. 1
t
2
) 2
= p + 2c(1) (o

1
t
) (o

2
t
) + /
2
1
= p + 2c(1)
_
(o

1
) (1

1
) + :
1
_

_
(o

2
) (1

2
) + :
2
_
+ /
2
1
=
_
(1

1
) :
1
_
+
_
(1

2
) :
0
_
+
_
/
2
:
2
_
+
_
ind(1
1
. o

1
; 1
2
. o

2
) 1
_
.
When ind(1
1
. o

1
; 1
2
. o

2
) < 1, the sum is negative and so there are no ideal-matched combs.
When ind(1
1
. o

1
; 1
2
. o

2
) = 1, the three other terms in parentheses must be 0, which implies
that (l
1
. l
2
) is a simple ideal-matched curve as stated.
The case when l
2
has multiple stories is similar. If l
1
has more than one story, each
story must contribute at least one to ind(1
1
. o

1
; 1
2
. o

2
) by the above analysis, contradicting
the index assumption.
9.3.3. Trimmed curves and gluing. We showed in Section 9.3.1 that 1-matched
curves converge to ideal-matched curves as 1 . It is not true, however, that every ideal-
matched curve arises this way. For instance, the split curves that normally occur at c
for a simple ideal-matched curve have moduli, so the moduli spaces of simple ideal-matched
9.3. DILATING TO INFINITY 163
curves of expected dimension 0 are not, in fact, 0-dimensional. The philosophically correct
solution is probably to remember some microscopic information at the curves in c (e.g.,
maps to the upper half-plane rather than [0. 1] 1). For our purposes, however, it is enough
to simply trim o the components at c:
Denition 9.31. A trimmed simple ideal-matched curve is a pair of holomorphic combs
(u
1
. u
2
) connecting two compatible pairs of generators x
i
. y
i
S(1
i
) (i = 1. 2) such that
either:
(T-A) One of u
1
or u
2
is trivial and the other is a index 1 holomorphic curve with no c
punctures or
(T-B) (u
1
. u
2
) has the following properties:
(T-B1) The comb u
1
is a holomorphic curve for 1
1
(one-story building with no compo-
nents at east innity) which is asymptotic to the non-trivial sequence of non-empty
sets of Reeb chords = (
1
. . . . .
n
).
(T-B2) The curve u
1
has index 1 with respect to .
(T-B3) The comb u
2
is an :-story holomorphic building for 1
2
with no components at
east innity.
(T-B4) Each story of u
2
has index one.
(T-B5) Each of u
1
and u
2
is strongly boundary monotone.
(T-B6) For each i = 1. . . . . :, the east punctures of the i
th
story of u
2
are labeled, in order,
by a non-empty sequence of Reeb chords (
i
1
. . . . .
i

i
), which have the property
that the sequence of singleton sets of chords
i
= (
i
1
. . . . .
i

i
) is composable.
We will let be the concatenation of the sequences
1
. . . . .
n
.
(T-B7) The composition of the sequence of singleton sets of Reeb chords
i
on the i
th
story of u
2
(with reversed orientation) coincides with the i
th
set of Reeb chords

i
, in the partition for u
1
; i.e.,

i
=

j=1

i
j
.
Lemma 9.32. When Conditions (T-B1) through (T-B5) hold, then Conditions (T-B6)
and (T-B7) hold if and only if the following holds:
(T-B6
t
) For all i = 1. . . . . :, the east punctures of the i
th
story of u
2
are labeled, in order,
by a non-empty sequence of Reeb chords (
i
1
.
i

i
) which has the property
that
(9.33) 1(o(x
1
.
[1,i1]
))c(
i
) = 1(o(x
1
.
[1,i1]
))

j=1
c(
i
j
) ,= 0.
Here, o(x
1
.
[1,i1]
) denotes the set of -arcs occupied by u
1
between
i1
and
i
;
see Denition 5.46.
Proof. Suppose that (u
1
. u
2
) is a trimmed simple ideal-matched comb. Let y
i
denote
the initial point of the i
th
story of u
2
(so that y
1
= x
2
).
Claim 1. The sets o(x
1
.
[1,i1]
) and o(y
i
) are disjoint. In the case where i = 1, this is
the fact that x
1
and y
1
are a compatible pair of generators. The case where i 1 follows
from Conclusion (2) of Lemma 6.8, applied to , using Condition (T-B7).
164 9. PAIRING THEOREM VIA TIME DILATION
Let o
i
= o(x
1
.
[1,i1]
).
Claim 2. For each i = 1. . . . . :,
(9.34) 1(o
i
)c
0
(
i
) = 1(o
i
)

j=1
c
0
(
i
j
) ,= 0.
This follows from Claim 1, the composability of the
i
(Condition (T-B6)) and Lemma 3.12.
We wish to conclude that Equation (9.34) holds with c in place of c
0
.
For the next three claims, pick some j. a with `(j) = `() and j ,= . Let

i
j
1
. . . . .
i
jm
be the subsequence of chords on the i
th
level which start or end in j. .
Claim 3. There is no
i
j
such that (
i
j
)

= j and (
i
j
)
+
= . We argue this as follows.
Let
i
j
k
be the rst
i
j
such that (
i
j
k
)

= j and (
i
j
k
)
+
= . By strong boundary monotonicity
of u
2
(Condition (T-B5)), for / = 1. . . . . / 1, the chords
i
j

alternate between starting at

one of j. and terminating at one of j. . Similarly, since
i
j
k
terminates at , strong
boundary monotonicity of u
2
implies that
i
j
k1
starts at one of j. . Now, there are two
cases, according to the parity of /. If / is even, then there are at least two strands starting
in j. in

j
k
j=1

i
j
, and hence at least two strands starting at j and in
i
, violating
boundary monotonicity of u
1
(Condition (T-B5)). If / is odd, there is at least one strand
starting in j. in

j
k
j=1

i
j
, and hence j. appears in o
i
. But in this case,
i
j
1
terminates
at one of j. , and hence j. appears in o(y
i
). This violates Claim 1.
Claim 4. For / = 1. . . . . :,
i
j
k
alternates between starting at one of j. and ending
at one of j. . This follows immediately from boundary monotonicity of u
2
, together with
Claim 3.
Claim 5. There is no / such that (
i
j
k
)
+
and (
i
j
k+1
)

are both in j. but not equal

to each other. Suppose that there is such a /. From Claim 4, it follows that if / is odd
then
i
has a chord starting at one of j. , and
i
j
1
is a chord terminating at one of j. ;
hence j. is in the initial idempotent for both u
1
and u
2
, contradicting Conclusion (2)
from Proposition 6.8. If / is even, there are at least two strands leaving one of j or in
i
,
contradicting boundary monotonicity of u
1
.
Claim 4 ensures that for / = 1. . . . . :,
i
j
k
and
i
j
k+1
do not have matched initial
or terminal points. Together with Claim 5 (using a straightforward inductive applica-
tion of Lemma 3.26), this ensures that not only is 1(o
i
)

i
j=1
c
0
(
i
j
) ,= 0, but in fact
1(o
i
)

i
j=1
c(
i
j
) ,= 0.
Conversely, suppose that Conditions (T-B1)(T-B5) hold, along with Condition (T-B6
t
).
Then, the non-vanishing of 1(o
i
)

i
j=1
c(
i
j
) implies the non-vanishing of 1(o
i
)

i
j=1
c
0
(
i
j
),
which, by Lemma 3.12, ensures both the composability of
i
(i.e., Condition (T-B6)) and
the property that
1(o
i
)

j=1
c
0
(
i
j
) = 1(o
i
)c
0
_

i

j=1

i
j

_
.
Equation (9.33) now gives Conclusion (T-B7).
Let //
B
tsic
(x
1
. y
1
. o

1
; x
2
. y
2
; o

2
) denote the moduli space of trimmed simple ideal-
matched curves connecting x
1
. x
2
to y
1
. y
2
with homology class 1 and east/west sources
9.3. DILATING TO INFINITY 165
o

1
and o

2
, modulo 1-translation of u
1
and each story of u
2
. Let
//
B
tsic
(x
1
. y
1
; x
2
. y
2
) :=
_
(S

1
)=
emb
(B
1
,)
(S

2
)=
emb
(B
2
, )
//
B
tsic
(x
1
. y
1
. o

1
; x
2
. y
2
; o

2
).
Lemma 9.35. The spine of a simple ideal-matched curve is a trimmed simple ideal-matched
curve. That is, if (l
1
. l
2
) is a simple ideal-matched curve, where l
1
= (u
1
.
1
) (with
1
a
generalized split curve) and l
2
is toothless comb u
2
, then (u
1
. u
2
) is an embedded trimmed
simple ideal-matched curve.
Proof. Case (T-A) is trivial. In case (T-B), properties (T-B1), (T-B2), (T-B3), (T-B4),
and (T-B5) are all immediate from the fact that (l
1
. l
2
) is a simple ideal-matched curve.
Properties (T-B6) and (T-B7) are veried as follows.
First, weak composability follows from Property (T-B5) and Lemma 5.47.
Next, let denote the sequence of Reeb chords coming from u
2
, and be the sequence
of sets of Reeb chords on u
1
. Then is the partition of Reeb chords at west innity for the
generalized split curve
1
, while is the collection of Reeb chords at east innity of
1
, with
a discrete partition and an ordering compatible with . It follows that is obtained from
by contracting various consecutive sequences of Reeb chords.
The stronger condition of composability follows from the following index considerations
(akin to Lemma 5.70). By Lemma 5.71, we have an inequality:
[[

i=1
gr
t
(c(
i
))
[ [

i=1
gr
t
(c(
i
)).
with strict inequality if and only if some sequence in which is contracted to get an element

i
is not composable. Equivalently, by Lemma 5.60, if is the function from Equation (5.58)
then () ( ) with strict inequality if and only if there is a non-composable sequence
which is contracted.
It follows from Denition 5.61 that:
ind(1
1
. ) = ind(1
1
. ) +[[ [ [ + (() ( ))
= ind(1
1
. ) + ind(1
2
. ) [ [ + (() ( )) + ([[ ind(1
2
. ))
= ind(1
1
1
2
) + (() ( )) + ([[ ind(1
2
. )).
By Lemma 9.29, ind(1
1
1
2
) = 1, and we just veried that (() ( )) 0. Since [[
is equal to the number of stories of l
2
and each story has index 1, ([[ ind(1
2
. )) = 0.
Since there is a holomorphic curve, ind(1
1
. ) 1 by Proposition 5.62, so () = ( ) and
[[ = ind(1
2
. ). It follows that u
1
and u
2
are embedded and the Reeb chords in which
are collapsed in are composable.
It turns out that there is a combinatorially unique way to un-trim curves. To this end,
it helps to have the following:
Denition 9.36. Let = (
1
. . . . .
n
) be a sequence of Reeb chords. A subsequence
(
j
1
. . . . .
jm
) is called abutting if
+
j
k
=

j
k+1
for each / = 1. . . . . : 1. An amalgama-
tion of is a partition of into abutting sub-sequences. Given an amalgamation of , there
is an associated set of Reeb chords whose elements are formed by joining together the
chords in the abutting sub-sequences specied by the amalgamation. We call such a set
an amalgam of .
166 9. PAIRING THEOREM VIA TIME DILATION
Lemma 9.37. If = (
1
. . . . .
n
) is a sequence of Reeb chords with c
0
(
1
) c
0
(
n
) =
c
0
() ,= 0, then there is a unique amalgamation of with resulting amalgam .
Proof. For existence of the amalgamation, let ,
1
= 1, let ,
2
be the rst index after
,
1
so that

j
2
=
+
j
1
, and so on. This gives an abutting subsequence (
j
1
. . . . .
jm
). Since
=

j
, one of the chords in is
j
1

jm
. We take (
j
1
. . . . .
jm
) to be one part
of the desired amalgamation. The other parts of the amalgamation are gotten by iteratively
repeating this process with replaced by (
j
1
. . . . .
jm
). We call the resulting amalgamation
the maximal amalgamation of ; by construction the corresponding amalgam is .
For uniqueness, we argue as follows. By the non-vanishing of c
0
(
1
) c
0
(
n
), given
j :, the subsequence of Reeb chords in which start or end at j alternates between
starting at j and ending at j. In the maximal amalgamation, a chord ending at j always
is joined to the next chord starting at j. For an amalgamation which does not always join
these pairs, the number of chords starting or ending at j is greater than for the maximal
amalgamation. It follows that the only amalgamation whose amalgam is is the maximal
amalgamation: any other amalgamation results in an amalgam with more chords in it.
Lemma 9.38. Let (u
1
. u
2
) be a trimmed simple ideal-matched curve. Then there is a gener-
alized split curve : 1

1[0. 1] 1 such that (u

1
. . u
2
) is a simple ideal-matched
curve. Moreover, the combinatorial type of 1

is determined by (u
1
. u
2
).
Proof. Again Case (T-A) is trivial, so we consider Case (T-B). Observe that each
generalized split curve with c punctures labeled by gives an amalgamation of , depending
only on the combinatorial type of the source 1

: the parts of the amalgamation correspond

to the generalized split components making up . Furthermore, the u punctures of are
labeled by the corresponding amalgam of . In addition, in any simple ideal-matched curve,
must match the i
th
level of u
1
with the i
th
story of u
2
. So it suces to show that for
a trimmed simple ideal-matched curve with u
1
asymptotic to and the i
th
level of u
2
asymptotic to
i
, there is a unique amalgamation of
i
whose amalgam is
i
. But this
follows from Property (T-B6
t
) and Lemma 9.37.
Lemma 9.39. For a generic choice of almost complex structures J
1
and J
2
and a homology
class 1 = 1
1
1
2
, the moduli space //
B
tsic
(x
1
. y
1
. o

1
; x
2
. y
2
; o

2
) consists of a nite number
of points.
Proof. Consider a curve (u
1
. u
2
) //
B
tsic
(x
1
. y
1
. o

1
; x
2
. y
2
; o

2
). There are only
nitely many sequences of sets of Reeb chords compatible with 1
1
. Thus, by Condi-
tion (T-B2), u
1
lies in one of nitely many index 1 moduli spaces which, by Propositions 5.6
and 5.23 and Theorem 5.55 (transversality, compactness, and the analysis of the boundary
strata), is a compact 0-manifold. Similarly, there are a nite number of ways to decompose
1
2
as a sum 1
2
= 1
2,1
+ +1
2,n
with each 1
2,i
a positive domain, and a nite number of
sequences of sets of Reeb chords compatible with each 1
2,j
. Thus, each story of u
2
lies in
one of nitely many index 1 moduli spaces, each of which is a compact 0-manifold.
Proposition 9.40. For x
i
. y
i
S(1
i
); 1
1
, 1
2
so that ind(1
1
1
2
) = 1; generic J
1
and J
2
;
and generic 1 suciently large,
#//
B
1
B
2
(1; x
1
. y
1
; x
2
. y
2
) = #//
B
1
B
2
tsic
(x
1
. y
1
; x
2
. y
2
).
Proof. Let 1 = 1
1
1
2
. By Lemma 9.39, the moduli space //
B
tsic
(x
1
. y
1
; x
2
. y
2
)
consists of nitely many points; we label these points (u
i
1
. u
i
2
). Let (u
i
1
.
i
. u
i
2
) be an un-
trimming of (u
i
1
. u
i
2
), as given by Lemma 9.38.
9.3. DILATING TO INFINITY 167
Consider the union of moduli spaces
//
B
( 1
0
; x
1
. y
1
. o

1
; x
2
. y
2
. o

2
) :=
_
TT
0
(1. //
B
(1. x
1
. y
1
. o

1
; x
2
. y
2
. o

2
)).
Our goal is to nd disjoint neighborhoods J
i
of each (u
i
1
.
i
. u
i
2
) in //
B
1
B
2
( 1
0
;
x
1
. y
1
; x
2
. y
2
) so that, for 1 suciently large, //
B
(1; x
1
. y
1
; x
2
. y
2
)

i
J
i
and, for
each i, //
B
(1; x
1
. y
1
; x
2
. y
2
) intersects J
i
in an odd number of points. The neighborhoods
J
i
will have the form
J
i
= (|
i
1
|
i
2
) //
B
1
B
2
( 1
0
; x
1
. y
1
; x
2
. y
2
)
where |
i
1
/
B
1
(x
1
. y
1
; o

i,1
1

i
) is a smeared neighborhood (Denition 5.27) of (u
i
1
.
i
)
and |
i
2
/
B
2
(x
2
. y
2
; o

i,2
) is a neighborhood of u
i
2
.
The proof involves two steps, of course: compactness and gluing. The compactness
step is easier, and we start there. Fix any collection of neighborhoods J
i
as above. By
Proposition 9.26, any sequence of holomorphic curves in //
B
(1
j
; x
1
. y
1
; x
2
. y
2
) with 1
j

converges to an ideal-matched holomorphic comb, which by Lemma 9.30 must be a simple
ideal-matched holomorphic curve. By Lemma 9.35 and the uniqueness part of Lemma 9.38,
this limit lies in some J
i
. It follows that //
B
(1; x
1
. y
1
; x
2
. y
2
)

i
J
i
for 1 suciently
large.
We turn next to the gluing step. Fix a trimmed simple ideal-matched curve (u
1
. u
2
)
//
B
tsic
(x
1
. y
1
; x
2
. y
2
), and let (u
1
. . u
2
) be the un-trimming of (u
1
. u
2
). Let o

i
denote the
source of u
i
and 1

the source of . Let

1 = (j
1
. . . . . j
n
) be the sequence of c punctures of u
2
,
ordered by the story in which they occur and, within each story, by their 1-coordinates. Via
the correspondence (from Denition 9.24), we can also view

1 as a list of the c punctures
of . We can also group the j
i
according to the stories of u
2
or, equivalently, according to
their heights in ; write this grouping as

Q = (
Q
1
..
j
1
. . . . . j

1
.
Q
2
..
j

1
+1
. . . . . j

2
. . . . .
Qm
..
j

m1
+1
. . . . . j
m
).
Thus, : is the total number of c punctures of u
2
, : is the number of stories of u
2
and
/
i
/
i1
is the number of c punctures on the i
th
story of :, so : = /
m
.
Observe that, by the proof of Lemma 9.38, the ordering of the elements of Q
i
induced
by 1 is compatible with the induced ordering around the boundary of each generalized
split-component of .
View (u
1
. ) as an element of /
B
1
(x
1
. y
1
; o

1
1

;

1) and u
2
as an element of /
B
2
(x
2
. y
2
;
o

2
;

1). There are evaluation maps
ev
1
: /
B
1
(x
1
. y
1
; o

1
1

;

1) 1
n1
+
ev
2
: /
B
2
(x
2
. y
2
; o

2
;

1) 1
n1
+
given by
ev
i
(n
i
) = (ev
p
2
(n
i
) ev
p
1
(n
i
). ev
p
3
(n
i
) ev
p
2
(n
i
). . . . . ev
pn
(n
i
) ev
p
n1
(n
i
)).
Dene
1
: 1
n1
1
nm
and
2
: 1
n1
1
m1
by

1
(t
1
. . . . . t
n
) = (t
1
. . . . . t

1
1
. t

1
+1
. . . . . t

2
1
. t

2
+1
. . . . . . . . t
m1
. t
m
)

2
(t
1
. . . . . t
n
) = (t

1
. t

2
. . . . . t

m1
).
We can nd a smeared neighborhood |
1
of (u
1
. ) and a neighborhood |
2
of u
2
with the
following properties:
168 9. PAIRING THEOREM VIA TIME DILATION
Figure 9.2. Illustration of the bered product in the proof of Proposition 9.40.
(i-1) The map
2
ev
1
sends |
1
to a neighborhood of some vector
0
in the interior of
1
m1
+
(disjoint from 1
m1
+
). This uses continuity of the relative evaluation maps on
the compactied moduli space. The vector
0
records the height dierences of the
punctures on u
1
.
(i-2) The map
1
ev
2
sends |
2
to a neighborhood of some vector r
0
in the interior of 1
nm
+
.
Again, this uses continuity of the evaluation maps: the vector r
0
records the height
dierences of the punctures within each of the stories of u
2
.
(i-3) For all suciently small ,
1
ev
1
maps |
1
properly and with odd degree to r
1
nm
+
[ |r| < . This uses Proposition 5.34, and our observation about the ordering
of the points in each Q
i
(to see that the image of the punctures on each generalized
split component lands in 1
nm
+
1
nm
).
(i-4) For all suciently large 1,
2
ev
2
maps |
2
properly and with odd degree onto r
1
m1
+
[ |r| 1 (so that u
2
is mapped to ). This uses Proposition 5.24 (or
rather, an :-story analogue thereof).
(i-5) The closure of |
1
intersects (u
i
1
.
i
) in the one point (u
1
. ).
(i-6) The closure of |
2
intersects u
i
2
in the one point u
2
.
It follows from these observations and an intersection theory argument that for any generic,
large choice of 1, the bered product
(9.41) |
1
|
2
//
B
1
B
2
(1; x
1
. y
1
; x
2
. y
2
) = |
1 T ev
1
ev
2 |
2
consists of an odd number of points. We spell this out presently. First, we rephrase Condi-
tions (i-1) and (i-2) more concretely:
(i-1
t
) There is a
0
1
m1
+
and c 0 so that
2
ev
1
sends |
1
into 1
2
:= 1

(
0
) 1
m1
+
.
(i-2
t
) There is an r
0
1
nm
+
and 0 so that
1
ev
2
sends |
2
into 1
1
:= 1

(r
0
) 1
nm
+
.
See Figure 9.2.
9.4. COMPLETION OF THE PROOF OF THE PAIRING THEOREM 169
Think of the bered product |
1 T ev
1
ev
2 |
2
as a subset of 1
nm
+
1
m1
+
. By the above two
conditions, it follows that the bered product is contained in the subset 1
1
11
2
. Choose
1 large enough so that 1
1
1
1
is contained inside the ball of radius about the origin in
1
nm
+
, and so that 11
2
lies outside a ball of radius 1 around the origin in 1
m1
+
, where
and 1 are constants from Conditions (i-3) and (i-4), respectively. Then let 1
1
denote the
preimage under 1 ev
1
of 1
1
11
2
, and let 1
2
denote the preimage under ev
2
of 1
1
11
2
.
Both are compact, because
1
ev
1
is proper on the preimage of 1
1
1
1
and
2
ev
2
is proper
on the preimage of 1
2
.
Now, Condition (i-1
t
) shows that 1 ev
1
: 1
1
1
1
11
2
can be viewed as a relative
cycle, representing some homology class in H
nm
(1
1
11
2
. (1
1
) 11
2
; F
2
)

= F
2
. Indeed,
Condition (i-3) ensures that [1
1
] is the non-trivial class: the degree of
1
1 ev
1
: 1
1
1
1
can be interpreted as the intersection pairing of [1
1
] with an element of H
m1
(1
1
11
2
. 1
1

(11
2
); F
2
)

= F
2
.
Symmetrically, by Conditions (i-2
t
) and (i-4), ev
2
: 1
2
1
1
11
2
represents a generator
of H
m1
(1
1
11
2
. 1
1
(11
2
); F
2
)

= F
2
. The count of points in the bered product
1
1 T ev
1
ev
2 1
2
= |
1 T ev
1
ev
2 |
2
can now be interpreted as the intersection pairing of [1
1
]
with [1
2
] under the non-trivial intersection pairing
H
nm
(1
1
11
2
. (1
1
) 11
2
) H
m1
(1
1
11
2
. 1
1
(11
2
)) H
0
(1
1
11
2
)

= F
2
.
as desired.
9.4. Completion of the proof of the pairing theorem
We are now in a position to give our second proof of the pairing theorem, Theorem 1.3.
Proof of Theorem 1.3. Fix bordered Heegaard diagrams 1
1
and 1
2
for (1
1
.
1
:
1(:) 1
1
) and (1
2
.
2
: 1(:) 1
2
). By Proposition 4.25, we may assume that
both 1
1
and 1
2
are provincially admissible, and at least one of 1
1
or 1
2
is admissible. (We
could assume both diagrams are admissible, but we will not need this.) By Lemma 4.33,
this implies that 1 = 1
1

1
2
is a Heegaard diagram for 1 which is weakly admissible for
all spin
c
-structures. By Lemmas 6.5 and 7.7, our assumption also implies that at least one
of

CFA(1
1
) or

CFD(1
2
) is bounded.
By Proposition 2.34, the -product

CFA(1
1
)

CFD(1
2
) (Denition 2.26) is a model for

CFA(1
1
)

CFD(1
2
). By Proposition 9.22 (which uses Theorem 9.10),

CF(1 ) is homotopy
equivalent to

CF(1; 1
1
. 1
2
) for any 1 0. So, it remains to identify

CFA(1
1
)

CFD(1
2
)
with

CF(1; 1
1
. 1
2
) for 1 suciently large.
On the level of F
2
-vector spaces, the identication is straightforward. As a vector space,

CFA(1
1
)

CFD(1
2
) is given by A(1
1
)
7(:)
A(1
2
), which has a basis (x
1
. x
2
) S(1
1
)
S(1
2
) [ o(x
1
) o(x
2
) = . This is exactly the set of generators for

CF(1; 1
1
. 1
2
) (and for

CF(1)).
Next we identify the dierentials. It follows from the admissibility criteria that there
are only nitely many homology classes 1 which can contribute to the dierential on

CF(1; 1
1
. 1
2
) (compare Lemmas 6.5 and 7.7). So, for large enough 1, Proposition 9.40
implies that the dierential on

CF(1; 1
1
. 1
2
) counts trimmed simple ideal-matched curves.
170 9. PAIRING THEOREM VIA TIME DILATION
Let c
i
be the set of basic generators for the algebra /(:). Dene operators 1
i
:
A(1
2
) A(1
2
) by

2
(x
2
) =

i
c
i
1
i
(x
2
)
as in Example 2.28. Then, for large 1, the dierential on

CF(1; 1
1
. 1
2
) is given by
(9.42)
T
(x
1
x
2
) =

:
n+1
(x
1
. c
i
1
. . . . c
in
)(1
in
1
i
1
)(x
2
).
where the sum is taken over all nite sequences c
i
1
. . . . . c
in
of basic generators for /(:)
(including the empty sequence, where : = 0). To see this, consider a trimmed simple
ideal-matched curve (u
1
. u
2
) connecting x
1
x
2
to y
1
y
2
. The holomorphic building
u
2
corresponds to a sequence of dierentials 1
in
1
i
1
on

CFD(1
2
), where : is the
number of stories of u
2
. (This uses one direction of Lemma 9.32 to guarantee that the
algebra elements coming from u
2
are nonzero and agree with the algebra elements coming
from u
1
.) The curve u
1
corresponds to a higher product :
n+1
(x
1
. c
i
1
. . . . . c
in
) on

CFA(1
1
)
of x
1
with the corresponding algebra elements c
i
1
. . . . . c
in
. Every non-zero term on the
right of Equation (9.42) arises this way. (This uses the other direction of Lemma 9.32, to
guarantee that the corresponding curves u
1
and u
2
satisfy conditions (T-B6) and (T-B7) of
Denition 9.31.)
Equation (9.42) is exactly the dierential on

CFA(1
1
)

CFD(1
2
) (compare Equa-
tion (2.29)). This concludes the proof.
9.5. A twisted pairing theorem
Next, we give a version of the pairing theorem for the twisted-coecient variants of

CFD
and

CFA. Loosely speaking, it says that the derived tensor product of twisted bordered
invariants computes the twisted Heegaard Floer complex of the glued-up three-manifold.
Before stating the precise version, we introduce a little algebra.
Lemma 9.43. The left action on

CFA(1
1
. s
1
) by F
2
[H
2
(1
1
. 1
1
)] and the right action of

CFD(1
2
. s
2
) by F
2
[H
2
(1
2
. 1
2
)] induce a left F
2
[H
2
(1. 1)]-action on the tensor product

CFD(1
1
. s
1
)

CFA(1
2
. s
2
).
Proof. The two actions give

CFD(1
1
. s
1
)

CFA(1
2
. s
2
) the structure of a bimodule
over F
2
[H
2
(1
1
. 1)] and F
2
[H
2
(1
2
. 1)]; since F
2
[H
2
(1
2
. 1)] is commutative, the right action
by F
2
[1
2
. 1] can be viewed as a left action, giving the product complex the structure of a
left F
2
[H
2
(1
1
. 1)] F
2
[H
2
(1
2
. 1)]

= F
2
[H
2
(1. 1)]-module.
The quotient map H
2
(1 ) H
2
(1. 1) induces a ring map i : F
2
[H
2
(1 )] F
2
[H
2
(1. 1)].
Note that i is injective. As in [OSz04c, Section 8], the twisted Heegaard Floer complex

CF(1 ) is a chain complex of free F

2
[H
2
(1 )]-modules, and hence induces a chain complex of
free F
2
[H
2
(1. 1)]-modules i

CF(1 )) =

CF(1 )
F
2
[H
2
(Y )]
F
2
[H
2
(1. 1)].
Theorem 9.44. There is a homotopy equivalence of F
2
[H
2
(1. 1)]-modules.

CFA(1
1
. s
1
)

CFD(1
2
. s
2
) i

_

sspin
c
(Y )
s[
Y
i
=s
i
,i=1,2

CF(1. s)
_
.
9.6. AN EXAMPLE 171
D3
D2
D1
3
2
x3
y3
x2
y2
x1
y1
Figure 9.3. Degenerating the hexagon. A hexagon in the Heegaard
diagram (giving a ow from x = r
1
. r
2
. r
3
to y =
1
.
2
.
3
) is divided into
three pieces 1
1
, 1
2
, and 1
3
, grouped as 1
1
and 1
2
1
3
.
Proof. Lemma 9.43 implies that as modules the two sides agree. It remains to show
that the dierentials agree (up to homotopy). For this we can use either proof of the pairing
theorem, and only need to check that the group-ring coecients on the two sides agree.
Let x
i
, i = 1. 2, be the chosen base generator for 1
i
in the spin
c
-structure s
i
. Fix gen-
erators y = y
1
y
2
and w = w
1
w
2
and domains 1
i

2
(x
i
. y
i
). Given domains
C
i

2
(y
i
. w
i
), the group-ring coecient of w
1
w
2
in the dierential of c
B
1
x
1
x
2
c
B
2
coming from (C
1
. C
2
) is c
B
1
C
1
x
1
x
2
c
B
2
C
2
. The group-ring coecient of w in (x) coming
from C
1
C
2
is c
(B
1
B
2
)(C
1
C
2
)
. With respect to the identication from Lemma 9.43, these
coecients agree.
Remark 9.45. We can recover

CF(1 ) from i

CF(1 )) as follows. Since H

1
(1) is a free
abelian group, so is the cokernel of the projection map i : H
2
(1 ) H
2
(1. 1). So, we can
choose a splitting /: H
2
(1. 1) H
2
(1 ), with / i = "
H
2
(F)
. Then

CF(1 )

= /

(i

CF(1 ))).
9.6. An example
We conclude this chapter with a local example. Consider a hexagon connecting the
generator x = r
1
. r
2
. r
3
to y =
1
.
2
.
3
in a genus three Heegaard diagram 1 for some
three-manifold 1 . Here, the hexagon is a domain 1 in whose boundary, as we traverse
it, contains arcs in three distinct -curves and arcs in three distinct -curves, so that if the
arc 1
i
(with its boundary orientation) goes from r
i
to
i
. This domain corresponds
to an element 1
2
(x. y) with ind(1) = 1. In fact, it is not dicult to see that the
domain 1 always contributes y to the dierential of x, i.e., #/(x. y; 1) = 1; see, for
instance, [Ras03, Lemma 9.11]. We wish to see this contribution from the point of view of
time dilation in a bordered setting where the dividing curve 2, which separates 1 into two
bordered diagrams, crosses the hexagon in two arcs. Such a curve divides the hexagon into
three regions which we denote 1
i

3
i=1
with the convention that 1
i
connects r
i
to
i
. We
label so that 1
1
meets both arcs in 2 1, whereas 1
2
and 1
3
each meet only one apiece.
We label the Reeb chords corresponding to these arcs
2
and
3
, numbered so that
i
meets
1
i
for i = 2. 3. See Figure 9.3 for an illustration.
We start, as a warm-up, by considering the problem from the point of view of matched
pairs (i.e., using the complex

CF(1
1
. 1
2
) from Theorem 9.10). We verify that
#
_
//
B
(r
1
.
1
; r
2
. r
3
.
2
.
3
)
_
= 1.
Now, /(r
1
.
1
; 1
1
) consists of a single curve n. Then ev

3
,
2
(n) = ev

3
(n) ev

2
(n)
is some xed real number t
0
, the height dierence between the Reeb chords
3
and
2
.
172 9. PAIRING THEOREM VIA TIME DILATION
The number t
0
is necessarily positive, but the precise value of t
0
depends on the conformal
parameters in the picture.
The moduli space on the other side, /(r
2
. r
3
.
2
.
3
; 1
2
+ 1
3
) consists of a pair of
disjoint disks, which can be moved relative to each other subject to the constraint that
2
is below
3
. (The constraint comes from strong boundary monotonicity.) In particular, the
evaluation map ev

3
,
2
sends this moduli space homeomorphically to 1
0
.
It follows that there is a unique point in the bered product, as claimed.
We compare this against two dierent computations which arise by time dilation. The
two dierent computations arise by thinking of 1
1
as the part on the side or on the 1
side.
We start by placing 1
1
on the 1 side. In this case, the type module for 1
2
1
3
has
four generators r
2
r
3
= r
2
. r
3
, r
2

3
= r
2
.
3
,
2
r
3
=
2
. r
3
, and
2

3
=
2
.
3
and
the following operations:
:
2
(r
2
r
3
.
2
) =
2
r
3
:
2
(r
2

3
.
2
) =
2

3
:
2
(r
2
r
3
.
3
) = r
2

3
:
2
(
2
r
3
.
3
) =
2

3
:
2
(r
2
r
3
.
2

3
) =
2

3
.
The rst four of these are gotten by looking the union of a bigon with a trivial strip.
The last is a union of the two non-trivial bigons 1
2
+ 1
3
, and comes from the element
of /(r
2
. r
3
.
2
.
3
. 1
2
+ 1
3
) with the property that the ev

2
,
3
= 0. We have already
seen that there is a unique such element.
The type 1 module of 1
1
has two generators r
1
and
1
, and the bigon 1
1
is interpreted
as giving a dierential
r
1
=
2

3

1
.
The product of these two complexes has two generators, x = r
2
r
3
r
1
and y =
2

1
,
and a dierential single from x to y, consistent our earlier computation. Algebraically, this
dierential pairs the operation :
2
(r
2
r
3
.
2

3
) =
2

3
with the (unbroken) dierential on
the type 1 side. In the graphical notation of Chapter 2, this operation is given by
r
2
r
3
r
1

1
:
2

3

1
2
3
.
Analytically, the moduli spaces //
B
(1; r
2
r
3
.
2

3
; r
1
.
1
) consist of the curve n in 1
1
and the pair of disks in 1
2
with punctures at height dierence t
0
,1. Taking 1 , the
separation between these heights goes to 0; this corresponds to the stated :
2
.
Next, place 1
1
on the side. Now, the type module has two generators r
1
and
1
,
and a single non-trivial operation :
3
(r
1
.
2
.
3
) =
1
. The type 1 module for 1
2
1
3
is
9.6. AN EXAMPLE 173
generated by the four generators r
2
r
3
, r
2

3
,
2
r
3
, and
2

3
, with dierentials
r
2
r
3
=
2

2
r
3
+
3
r
2

3
r
2

3
=
2

2

2
r
3
=
3

2

3
.
On the product of these two complexes, we once again get two generators and a single
dierential. The generators are now labeled r
1
r
2
r
3
and
1

2

3
. Algebraically, the
dierential in this case is gotten by pairing the :
3
on the type side with the dierential
from r
2
r
3
to
2

2
r
3
followed by the dierential from
2
r
3
to
3

2

3
. In the graphical
notation of Chapter 2, this operation is given by
r
1
r
2
r
3

1
:
3

1

2

2
3
.
Analytically, the moduli spaces //
B
(1; x
1
. y
1
; r
2
r
3
.
2

3
) consist of the curve n in 1
1
and the pair of disks in 1
2
with punctures at height dierence 1 t
0
. Taking 1 , the
separation between these heights goes to ; this corresponds to the pair of
1
s.
CHAPTER 10
Gradings
We now turn to gradings on the structures we have already constructed in Chapters
69: Type 1 invariants

CFD(1); Type invariants

CFA(1); and their tensor products

CFA(1
1
)

CFD(1
2
). This extends the gradings on the algebra from Section 3.3, and we
will use the material on the dimension of the moduli spaces

/
B
(x. y; ) from Section 5.7.
As described in Denition 2.42, the gradings gr
t
on

CFD(1. s) and

CFA(1. s) take
values in sets o
t
D
(1. s) and o
t
A
(1. s) (depending on the spin
c
structure s) with actions of the
grading group G
t
(4/); a left action in the case of o
t
D
(1. s) (Section 10.4) and a right action
in the case of o
t
A
(1. s) (Section 10.3). We will also dene renements gr of these grading
taking values in sets with actions of the smaller group G(:) (Section 10.5). As discussed
in Section 10.6, the tensor product takes values in the product over G(:) of these two
grading sets. Before constructing these gradings, we collect some properties of the grading
groups and relationships with the expected dimensions of moduli spaces from earlier chapters
(Section 10.1) and give a useful grading on domains (Section 10.2).
10.1. Algebra review
For reference, we recall briey the relevant material from Sections 3.3 and 5.7.
In Section 3.3 we introduced two grading groups:
G
t
(4/), which is a central extension of H
1
(2
t
. a) by Z, and
G(:), which is a subgroup of G
t
(4/) isomorphic to a central extension of H
1
(1(:))
by Z.
We review G
t
(4/) here, and delay reviewing G(:) until Section 10.5 where we construct the
rened gradings.
Elements of G
t
(4/) are written (,. ), where ,
1
2
Z and H
1
(2
t
. a). We impose the
additional restriction that , c() (mod 1), where c() is one quarter the number of parity
changes in . The multiplication on G
t
(4/) is given by
(,
1
.
1
) (,
2
.
2
) := (,
1
+ ,
2
+ 1(
1
.
2
).
1
+
2
)
where 1(
1
.
2
) is the linking of the boundaries of
1
and
2
, given by
1(
1
.
2
) := :(
2
.
1
)
or, concretely,
1(
1
.
2
) =
_

_
1,2 if
+
1
=

2
1,2 if
+
2
=

1
0 if
1

2
= or
1

2
or
2

1
or
1
=
2
1 if

1
<

2
<
+
1
<
+
2
1 if

2
<

1
<
+
2
<
+
1
.
175
176 10. GRADINGS
We call , the Maslov component of (,. ) and the spin
c
component of (,. ). The distin-
guished central element of G
t
(4/) is the element = (1. 0); thus the dierential lowers the
rst component of the grading by 1.
More generally, we have (Lemma 5.59)
(,
1
.
1
) (,
n
.
n
) =
_

i
,
i
+

i<j
1(
i
.
j
).

i
_
.
The grading gr
t
(c) of an element c /(:) with starting idempotent o is given by
(c) := inv(c) :([c]. o)
gr
t
:= ((c). [c])
(Denition 3.38), where inv(c) denotes the number of inversions of c. For convenience, if
is a sequence of Reeb chords, let gr
t
( ) be the product

n
i=1
gr
t
(c(
i
)). (If c( ) ,= 0, then
gr
t
( ) = gr
t
(c( )).) Similarly set gr
t
() to be

n
i=1
gr
t
(c(
i
)).
The function can also be written, for a set of Reeb chords with c() ,= 0, as
(c()) =

1
,
2

[1(
1
.
2
)[
[[
2
(Lemma 5.57). We also dene
() := (c())
() :=

i
(
i
) +

i<j
1(
i
.
j
)
so that () is the Maslov component of gr
t
() (Lemma 5.60).
These quantities are related to the dimensions of the embedded moduli space as follows.
For 1
2
(x. y) and a compatible sequence of sets of Reeb chords,

/
B
(x. y; ) has
expected dimension
ind(1. ) = c(1) + :
x
(1) + :
y
(1) +[[ + ()
(Denition 5.61 and Proposition 5.62).
10.2. Domains
Before dening a grading on modules, we dene a grading on domains.
Denition 10.1. For x. y S(1) and 1
2
(x. y), dene p
t
(1)
1
2
Z H
1
(2
t
. a) by
(10.2) p
t
(1) := (c(1) :
x
(1) :
y
(1).

(1)).
Lemma 10.3. For any 1
2
(x. y), the element p
t
(1) is in G
t
(4/).
A special case of Lemma 10.3 is that for provincial domains 1, the expected dimension
c(1) + :
x
(1) + :
y
(1), which is a priori in
1
4
Z, is actually in Z.
Proof. We must show that c(1) :
x
(1) :
y
(1) c(

(1)) (mod 1). To do so,

we follow [OSz04d, Lemma 2.17] to construct a surface related to 1. First, consider

1 :=
/[] +1 where / is large enough that all local multiplicities of

1 are non-negative. Here,

1
is an element of
2
(x. y), the set of homology classes of curves connecting x to y which are
allowed to cross the basepoint .. (This is denoted
2
(x. y) in [OSz04d], but in this book we
10.2. DOMAINS 177
have reserved
2
for classes which have local multiplicity zero at ..) Since c([]) +:
x
([]) +
:
y
([]) = 1, we have c(1) + :
x
(1) + :
y
(1) c(

1) + :
x
(

1) + :
y
(

1) modulo 1.
Label the components of () by 1
i
for i = 1. . . . . `, and write

1 =

i
:(1
i
)1
i
with i 0. Then we will construct our surface by identifying copies of 1
i
:
1 :=
_
N

i=1
m(R
i
)

j=1
1
(j)
i
__

where each 1
(j)
i
is a dieomorphic copy of 1
i
. To describe the identication , rst note
that the -curves are divided by the -circles into shorter pieces, which we call -segments,
and similarly the -circles are divided by the -curves into -segments. For each - or -
segment c, let 1
1
(c) and 1
2
(c) be the two regions adjoining c, with :(1
1
(c)) :(1
2
(c)).
For r c, let r
(j)
1
and r
(j)
2
be the lifts of r to 1
(j)
1
(c) and 1
(j)
2
(c), respectively. Let
c
:=
:(1
2
(c)) :(1
1
(c)). Then the equivalence relation is generated by
r
(j)
1
r
(j+a)
2
for r in an -segment c, , = 1. . . . . :(1
1
(c))
r
(j)
1
r
(j)
2
for r in an -segment /, , = 1. . . . . :(1
1
(/)).
Let be the evident projection from 1 to , so

1 =

[1]. Topologically, 1 is a manifold

with boundary; requiring the map to be a local isometry makes 1 into a cone manifold.
We next analyze the local behavior of 1 around a point r
i

j
. Let the four regions
around r be 1
1
(r). . . . . 1
4
(r) in cyclic order, arranged so that :(1
1
(r)) :(1
i
(r)) for
i = 2. 3. 4. The fact that

1
2
(x. y) means that
:(1
1
(r)) + :(1
3
(r)) = :(1
2
(r)) + :(1
4
(r)) + C(r).
where C(r) 0. 1. 1. Specically, C(r) = 0 when r is in neither x nor y or when
r is in both x and y. For r let 1(r) be a small disk neighborhood of r and let
`(r) =
1
(1(r)). The geometry of `(r) depends on C(r) as follows:
If C(r) = 0 then `(r) is a union of :(1
1
(r)) disks and some half-disks, with
the boundary of the half-disks lying above either
i
or
j
; in this case, `(r) is a
manifold with boundary but no corners.
If C(r) = 1 then one component of `(r) has a cone point at r with total angle
2:(1
1
(r)) + ,2. The other components of `(r) are half-disks as in the C = 0
case.
If C(r) = 1, then one component of `(r) has a cone point at r with total angle
2(:(1
1
(r)) + 1) ,2. Again, the other components of `(r) are half-disks.
Let :
+
:= #r [ C(r) = 1 and :

:= #r [ C(r) = 1, and let /(r) be

the number of half-disks in `(r).
Recall that

1 denotes the part of

1 contained in . For each c

i
a, let
i
:=
:(

1. c
i
+ c) :(

1. c
i
c) be the dierence in multiplicities with which

1 covers
the regions on the two sides of c
i
. The surface 1 has corners at intersection points with
C(r) = 1 and at points above , and the number of corners of 1 over is
1
4

i
[
i
[.
Thus,
c(

1) = c(1)
1
4
_
:
+
+ :

+
4k

i=1
[
i
[
_
(mod 1).
178 10. GRADINGS
On the other hand,
:
x
(

1) + :
y
(

1) = :
x+y
(

1)

xx+y
_
1
4
C(r) +
1
2
/(r)
_

1
4
(:
+
:

) +
1
2

x[(x+y)
a
]
/(r).
For the second congruence, observe that on each -circle
j
there is one r
j
x and one

j
y, and /(r
j
) = /(
j
). A similar statement holds for each -circle, so it suces to sum
/(r) only when r is in an -arc.
It follows that
c(

1) + :
x
(

1) + :
y
(

1)
1
4
4k

i=1
[
i
[ +
1
2

x[(x+y)
a
]
/(r).
It remains to show that the right-hand side is equal to c(

1). Consider the contribution to

the right-hand side from a particular -arc
a

and its endpoints c

i
and c
j
.
If / is not in o(x) or o(y), the set of -arcs occupied by x and y, respectively, then
there is no contribution to the /(r) term and
i
=
j
, so the net contribution is
1
2
if
i
is odd and 0 otherwise, agreeing with the contribution to c(

1).
If / is in both o(x) and o(y), then we have points r

x
a

and

y
a

. We
have /(r

) = /(

), so the same argument as in the previous case applies.

If / is in exactly one of o(x) and o(y), we have
j
=
i
1. Assume without loss of
generality that [
j
[ = [
i
[ + 1. Let r

(x + y)
a

be the unique corner on this

arc. Then /(r

) = [
i
[, so
1
4
([
i
[ +[
j
[) +
1
2
/(r

)
1
4
(2[
i
[ + 1) +
1
2
[
i
[
1
4
.
Since exactly one of
i
and
j
is odd, the contribution of c
i
and c
j
to c(

1) is also
1
4
, as desired.
Lemma 10.4. For 1
1

2
(x. y) and 1
2

2
(y. w), we have
p
t
(1
1
)p
t
(1
2
) = p
t
(1
1
1
2
).
Proof. This is obvious for the spin
c
component. For the Maslov component, the proof
is similar to Proposition 5.68. If we set c
i
=

(1
i
) and /
i
=

(1
i
), the Maslov component
of p
t
(1
1
1
2
) is
(:
x
:
w
c)(1
1
1
2
)
= (:
x
:
y
c)(1
1
) + (:
y
:
w
c)(1
2
) + (:
y
:
w
)(1
1
) + (:
y
:
x
)(1
2
)
= (:
x
:
y
c)(1
1
) + (:
y
:
w
c)(1
2
) + c
1
/
2
c
2
/
1
= (:
x
:
y
c)(1
1
) + (:
y
:
w
c)(1
2
) + 1
7
(

(1
1
).

(1
2
)).
where the second equality uses Lemma 5.66 and the third equality uses Lemma 5.67. The
last formula is the Maslov component of p
t
(1
1
)p
t
(1
2
), as desired.
Denition 10.5. For a generator x S(1), let 1
t
(x) be the subset of G
t
(4/) given by
p
t
(1) [ 1
2
(x. x).
Corollary 10.6. 1
t
(x) is a subgroup of G
t
(4/).
10.3. TYPE A STRUCTURES 179
This grading on domains is closely related to ind(1. ).
Lemma 10.7. For a compatible pair (1. ) where = (
1
.
2
. . . . .

), we have
p
t
(1) =
ind(B,)
gr
t
().
Proof. Applying Lemma 5.60, Denition 5.61 and Denition 10.1, in turn, gives
gr
t
() = (().

1)
= (ind(1. ) / c(1) :
x
(1) :
y
(1).

1)
=
ind(B,)
p
t
(1).
10.3. Type structures
In this section we describe the (unrened) grading on

CFA(1). Per Denition 2.42, we
are looking for a right G
t
(4/)-set o
t
A
(1) and a map gr
t
: S(1) o
t
A
(1) so that
gr
t
(x) =
1
gr
t
(x)
gr
t
(xc) = gr
t
(x) gr(c)
and similarly for higher products. Since the dierent spin
c
structures do not interact in any
algebra action, we will construct one grading set o
t
A
(1. s) for each spin
c
structure s on 1 ;
then o
t
A
(1) is dened as

sspin
c
(Y )
o
t
A
(1. s).
We rst give the grading on the module with twisted coecients.
Denition 10.8. To dene the grading on the module

CFA(1. s), dene o
t
A
(1. s) to be
G
t
(4/) (independent of 1 and s), and set gr
t
(c
B
0
x) := p
t
(1
0
).
Proposition 10.9. The map gr
t
denes a grading on

CFA(1. s). Furthermore, for 1

2
(x. y), gr
t
satises
(10.10) gr
t
(c
B
0
x) p
t
(1) = gr
t
(c
B
0
B
y).
and any other grading gr with this property is related to gr
t
by left translation, in the sense
that there is an element p
0
G
t
(4/) so that for all x,
gr(x) = p
0
gr
t
(x).
Proof. Equation (10.10) is immediate from Lemma 10.4:
gr
t
(c
B
0
B
y) = p
t
(1
0
1) = p
t
(1
0
)p
t
(1) = gr
t
(c
B
0
x) p
t
(1).
To check that gr
t
is a grading, from Denition 2.5 (in the form in Equation (2.43)), we
want to show that for = (
1
. . . . .

) any sequence of sets of Reeb chords, :

+1
(c
B
0
x.
c(
1
). . . . . c(

)) is homogeneous of degree
1
gr
t
(c
B
0
x) gr
t
(). Let 1
2
(x. y) and let
= (
1
. . . . .

) be a sequence of sets of Reeb chords compatible with 1. If c

B
0
B
y appears
as a term in :
+1
(c
B
0
x. c(
1
). . . . . c(

)), then ind(1. ) = 1, so by Equation (10.10) and

Lemma 10.7,
gr
t
(c
B
0
B
y) = gr
t
(c
B
0
x) p
t
(1) =
1
gr
t
(c
B
0
x) gr
t
().
For any other grading gr satisfying Equation (10.10), let p
0
= gr(c
I
0
x
0
), where 1
0

2
(x
0
. x
0
) is the trivial domain. Then for any generator x and any 1
0

2
(x
0
. x),
gr(c
B
0
x) = gr(c
I
0
x) p
t
(1
0
) = p
0
gr
t
(c
B
0
x).
180 10. GRADINGS
For the untwisted theory, the presence of periodic domains means that we can not just
dene the relative grading of x and y to be p
t
(1) G
t
(4/) for some 1
2
(x. y), as the ana-
logue of Equation (10.10) would fail for any periodic domain 1 for which p
t
(1) is non-trivial.
Instead, we look for a right G
t
(4/)-set o
t
A
(1. s) satisfying an analogue of Equation (10.10).
The G
t
(4/) action on o
t
A
(1. s) will be transitive but not in general free; this generalizes the
fact that for closed 3-manifolds the gradings in a non-torsion spin
c
structure take values in
a quotient of Z.
Denition 10.11. Suppose that S(1. s) ,= . Pick any x
0
S(1. s) and dene o
t
A
(1. s)
to be the coset space
(10.12) o
t
A
:= 1
t
(x
0
)G
t
(4/).
Dene gr
t
: S(1. s) o
t
A
(1. s) on x by picking any 1
0

2
(x
0
. x) and setting
(10.13) gr
t
(x) := 1
t
(x
0
) p
t
(1
0
).
The value of gr
t
(x) is independent of 1
0
: If 1
1

2
(x
0
. x) is any other element, then
p
t
(1
1
(1
0
)
1
) 1
t
(x
0
). Thus, gr
t
(x) = 1
t
(x
0
) p
t
(1
1
).
For completeness, suppose next that S(1. s) = . Choose a diagram 1
t
isotopic to 1
and so that S(1
t
. s) ,= (by winding transverse to the -circles; compare [OSz04d, Lemma
5.2]). Dene o
t
A
(1. s) = o
t
A
(1
t
. s), and let gr
t
: S(1. s) o
t
A
(1. s) be the unique map from
the empty set to o
t
A
(1
t
. s).
Lemma 10.14. The function gr
t
: S(1. s) o
t
A
(1. s) above satises, for x. y S(1. s)
and 1
2
(x. y),
(10.15) gr
t
(x) p
t
(1) = gr
t
(y).
Furthermore, if S(1. s) ,= then o
t
A
(1. s) is universal, in the sense that for any other
G
t
(4/)-set

o and grading function gr on S(1. s) satisfying (10.15), there is a unique G
t
(4/)-
set map ) : o
t
A
(1. s)

o so that for any generator x, gr(x) = )(gr
t
(x)).
Proof. To check Equation (10.15), pick 1
0

2
(x
0
. x). Then 1
0
1
2
(x
0
. y), so we
have
gr
t
(y) = 1
t
(x
0
) p
t
(1
0
) p
t
(1) = gr
t
(x) p
t
(1).
(Note that we used the freedom to pick an arbitrary element of
2
(x
0
. y) to dene gr
t
(y).)
For universality, let
0
be gr(x
0
). Then Equation (10.15) implies that 1
t
(x
0
) stabilizes

0
. We can therefore set, for / G
t
(4/),
)(1
t
(x
0
) /) :=
0
/.
It is immediate that gr(x) = )(gr
t
(x)).
Observe that any right G
t
(4/)-set map from G
t
(4/) to itself is left translation as in
Proposition 10.9.
Corollary 10.16. If we use a dierent base generator x
t
0
S(1. s) to dene o
t
A
(1. s), the
grading of an element x with respect to x
t
0
is p
t
(1
0
) gr
t
(x) for any 1
0

2
(x
t
0
. x
0
).
Proof. This follows from noticing that 1
t
(x
t
0
) = p
t
(1
0
) 1
t
(x
0
) p
t
(1
0
)
1
.
Left translation by p
t
(1
0
) is an isomorphism of right G
t
(4/)-sets from 1
t
(x
0
)G
t
(4/) to
1
t
(x
t
0
)G
t
(4/).
10.3. TYPE A STRUCTURES 181
Proposition 10.17. For the grading gr
t
of Denition 10.11,

CFA(1. s) is a graded right
/

module in the sense of Denition 2.42.

Proof. This follows from Lemma 10.14 as in the proof of Proposition 10.9.
We now have the following graded renement of Theorems 7.17 and 7.28 (and indeed
Theorem 1.2):
Theorem 10.18. Up to graded /

homotopy equivalence, the G

t
(4/)-graded /

module

CFA(1. s) over /(1) is independent of the choices made in its denition. That is, for
any choices of provincially admissible Heegaard diagrams 1
1
and 1
2
, suciently generic
admissible almost complex structures J
i
for 1
i
and base generators x
i
0
S(1
i
. s), there is
an isomorphism of G
t
(4/)-sets
: o
t
A
(1
1
. x
1
0
) o
t
A
(1
2
. x
2
0
)
and an /

homotopy equivalence of /

modules
) = )
1+i
:

CFA(1
1
. J
1
) /(1)
i

CFA(1
2
. J
2
)
such that, for x S(1
1
) and homogeneous elements c
1
. . . . . c

/(1),
gr
t
()
1+
(x. c
1
. . . . . c

)) = (gr
t
(x)) gr
t
(c
1
) . . . gr
t
(c

.
The graded /

bimodule

CFA(1. s) over F
2
[H
2
(1. 1 )] and /(1) is similarly invari-
ant.
Proof. We prove invariance of the untwisted theory; the proof for the twisted theory is
similar.
Independence of the grading from the choice of base generator (assuming 1
1
= 1
2
and
J
1
= J
2
) is immediate from Corollary 10.16: the map p is the identity map (i.e., p
1
= " and
p
i
= 0 for i 1) and the map is given by (:) = p
t
(1
0
):, where 1
0

2
(x
2
0
. r
1
0
). (Left
translation is an isomorphism of right G
t
(4/)-sets.)
The rest of the proof follows the proof of the ungraded version, Theorem 7.17, noting
that the identications induced by changes of complex structure, isotopies, handleslides, and
stabilizations all respect gradings.
Stabilization invariance is immediate: stabilizing near the basepoint, the /

modules
before and after stabilization are identied, along with all the grading data. For changes of
complex structure, the fact that the continuation maps preserve gradings is a straightforward
adaptation of proof that

CFA is graded (Proposition 10.17). The key point is that now, if y
appears in )
Jr
1+
(x. c(
1
). . . . . c(

)), then there is a domain 1

2
(x. y) with ind(1. ) = 0
(rather than 1, as it would be for :
1+
in place of )
Jr
1+
), and hence
gr
t
(y) = gr
t
(x) gr
t
(c
1
) gr
t
(c

.
This is precisely what is needed for )
Jr
= )
Jr
n

n=1
to be a graded /

homomorphism.
The proof of isotopy invariance is the same as the proof of invariance under change of
complex structure except that the base generator may disappear during the isotopy. If there
are no generators in S(1
1
. s) or S(1
2
. s) then the result is obvious. Otherwise, we can
choose the isotopy so that there is an intermediate Heegaard diagram 1
3/2
so that there is
some generator x
1
0
in both S(1
1
. s) and S(1
3/2
. s), and a generator x
2
0
in both S(1
3/2
. s)
and S(1
2
. s). The same argument as for independence of complex structures gives a graded
182 10. GRADINGS
homotopy equivalence between

CFA(1
1
. s) and

CFA(1
3/2
. s), using x
1
0
to dene the grading,
and a graded homotopy equivalence between

CFA(1
3/2
. s) and

CFA(1
2
. s), using x
2
0
to dene
the grading. Together with the independence of the grading from the base generator, which
we already veried, this gives the invariance result for isotopies.
For handleslides, we must take into account the fact that the grading sets for the two
diagrams are dierent. Specically, suppose that 1
2
is gotten from 1
1
by a handleslide;
as in Section 7.3.2, suppose this is a handleslide of an -arc over an -circle, and write
1
2
= (.
H
. . .), 1
1
= (. . . .). Since we have already veried independence from
the basepoint, we may choose x
2
0
to be the generator connected to x
1
0
by a union of small
triangles 1
x
(see Section 6.3.2). Then, the grading set for 1
1
is 1
t
(x
1
0
)G
t
(4/), whereas the
grading set for 1
2
is 1
t
(x
2
0
)G
t
(4/). There is a map : 1
t
(x
0
) 1
t
(x
t
0
) dened by sending
1
2
(x
1
0
. x
1
0
) to the element (1)
2
(x
2
0
. x
2
0
) uniquely determined by
1
x

13
1 = 1
x

12
1
,
H
23
(1).
using Lemma 6.24. (Here, 1
,
H
2
(
o
.
o
) for appropriate o and o
t
.) By inspection,
p
t
((1)) = p
t
(1). So, 1
t
(x
1
0
) = 1
t
(x
2
0
), and o
t
A
(1
1
. x
1
0
) = o
t
A
(1
2
. x
2
0
).
Next we check that the handleslide map )
,
H
,
respects this identication of grading
sets. By Corollary 6.32 (and the denition of the map )
,
H
,
, Equation (7.26)),
)
,
H
,
1+
(x. c(
1
). . . . . c(

))
is given by a count of triangles in homology classes / so that ind(/(1). (
1
. . . . .

)) =
0. So, as for the map associated to changes of almost complex structure, if y occurs in
)
,
H
,
1+
(x. c(
1
). . . . . c(

)) then
gr
t
(y) = gr
t
(x) gr
t
(c
1
) gr
t
(c

.
This completes the proof.
10.4. Type 1 structures
We turn now to dening a grading on

CFD(1). One point that merits attention is that
the relevant grading groups G
t
(4/) and G(:) depend on the orientation of :. In particular,
in the denition of the product on G
t
(4/),
(,
1
.
1
) (,
2
.
2
) := (,
1
+ ,
2
+ 1(
1
.
2
).
1
+
2
)
(from Equation (3.34)), the correction term 1(. ) depends on whether we are in 1 or 1.
To avoid confusion, let : : 2 2 be the identity map, which is orientation-reversing, and
let 1: G
t
(4/) G
t
(4/) or G(:) G(:) be the induced map:
(10.19) 1(,. ) = (,. :

()).
It follows from the form of the product that 1 is a group anti-homomorphism, so, for instance,
1(p
t
(1
1
1
2
)) = 1(p
t
(1
2
))1(p
t
(1
1
)).
We have an analogue of Lemma 10.7 for the moduli spaces relevant for

CFD.
Lemma 10.20. For a compatible pair (1. ), we have gr
t
( )1(p
t
(1)) = (ind(1. ). 0).
10.4. TYPE D STRUCTURES 183
Proof. This can be deduced from Lemma 10.7, but we nd it clearer to give a direct
proof. The spin
c
component of 1(p
t
(1)) is the negative of the spin
c
component of gr
t
( ),
since all the boundary intervals are reversed in the latter. Thus the spin
c
component of
the result is 0, and Maslov component of the product is obtained by adding the Maslov
components of the factors. The Maslov component of 1(p
t
(1)) is, by denition, c(1)
:
x
(1) :
y
(1). On the other hand, if = (
1
. . . . .
n
) then
gr
t
( ) = (
1
2
. [
1
]) (
1
2
. [
n
])
=
_

n
2
+

i<j
1(:

([
i
]). :

([
j
]).

i
[
i
]
_
=
_

n
2

i<j
1([
i
]. [
j
]).

i
[
i
]
_
=
_
: ( ).

i
[
i
]
_
.
where we use Lemma 5.59, bilinearity of 1, and the denition of (Equation (5.58)). Thus,
by Denition 5.61, the Maslov component of the product is ind(1. ).
Denition 10.21. For

CFD(1. s) as for

CFA(1. s), dene o
t
D
(1. s) to be G
t
(4/), and set
gr
t
(c
B
0
x) := 1(p
t
(1
0
)).
Proposition 10.22. The map gr
t
above denes a grading on

CFD(1. s). Furthermore, for
1
2
(x. y), gr
t
satises
(10.23) 1(p
t
(1)) gr
t
(c
B
0
x) = gr
t
(c
B
0
B
y).
and any other grading gr with this property is related to gr
t
by right translation, in the sense
that there is an element p
0
G
t
(4/) so that for all x,
gr(x) = gr
t
(x) p
0
.
Proof. If there is a term of the form c( )c
B
0
B
y occurring in (c
B
0
x) then ind(1. ) =
1, and so
gr
t
(c( )c
B
0
B
y) = gr
t
(c( ))1(p
t
(1))gr
t
(c
B
0
x)
=
1
gr
t
(c
B
0
x).
The universality statement is analogous to the case for

CFA, Proposition 10.9.
Again, for the untwisted theory, we need to nd an appropriate left G
t
(4/)-set satisfying
a version of Equation (10.23).
Denition 10.24. To dene the grading on

CFD(1. s), take o
t
D
(1. s) to have the same
underlying set as o
t
A
(1. s), with the left action of p on o
t
D
(1. s) being given by the right
action of 1(p) on o
t
A
(1. s). Concretely, if S(1. s) ,= ,
o
t
D
(1. s) := G
t
(4/),1(1
t
(x
0
)).
and dene gr
t
: S(1. s) o
t
D
(1. s) on x by picking any 1
0

2
(x
0
. x) and setting
(10.25) gr
t
(x) := 1(p
t
(1
0
)) 1(1
t
(x
0
)).
184 10. GRADINGS
Lemma 10.26. The function gr
t
: S(1. s) o
t
D
(1. s) above satises, for x. y S(1. s)
and 1
2
(x. y),
(10.27) 1(p
t
(1)) gr
t
(x) = gr
t
(y).
Furthermore, o
t
D
(1. s) is universal, in the sense that for any other G
t
(4/)-set

o and grading
function gr on S(1. s) satisfying (10.27), there is a unique G
t
(4/)-set map ) : o
t
D
(1. s)

o
so that for any generator x, gr(x) = )(gr
t
(x)).
Proof. This follows from Lemma 10.14 by reversal of all products.
Proposition 10.28. For the grading gr
t
given by Denition 10.24,

CFD(1. s) is a graded
left dierential module
Proof. This is similar to the proof of Proposition 10.22, and is left to the reader.
Theorem 10.29. Up to graded homotopy equivalence, the G
t
(4/)-set graded dierential
module

CFD(1. s) over /(1) is independent of the choices made in its denition. That
is, for any choices of provincially admissible Heegaard diagrams 1
1
and 1
2
, suciently
generic admissible almost complex structures J
i
for 1
i
and base generators x
i
0
S(1
i
. s),
there is an isomorphism of G
t
(4/)-sets
: o
t
D
(1
1
. x
1
0
) o
t
D
(1
2
. x
2
0
)
and a homotopy equivalence of dierential modules
) :

CFD(1
1
. J
1
)

CFD(1
2
. J
2
)
such that, for each x S(1
1
) and each term cy in )(x) (where y S(1
2
. s) and c
/(1)),
gr
t
(c) gr
t
(y) = (gr
t
(x)).
The dg bimodule

CFD(1. s) over F
2
[H
2
(1. 1 )] and /(1) is similarly invariant.
Proof. This is similar to the proof of Theorem 10.18, and is left to the reader.
10.5. Rened gradings
We turn now to a rened grading gr on these modules, building on the rened grading
of the algebra. We start with a review of the small grading group G(:). The group G(:)
can be viewed either abstractly as a central extension or concretely as a subgroup of G
t
(4/).
Abstractly, it is the central extension of H
1
(1(:)), with a sequence of groups
Z

G(:)
[]
H
1
(1)
satisfying the relation
p/ = /p
2([g][h])
(Equation (3.42)). Concretely, the group G(:) is the subgroup of G
t
(4/) consisting of those
elements (,. ) where `

() = 0, that is, for each pair j. of points in a identied by the

matching ` dening :, the sum of the multiplicities of at j and at is 0.
The rened grading on /(:. 0) is built from grading renement data: a choice of a base
idempotent 1(s
0
) /(:. 0) and, for every other minimal idempotent 1(s), an element
10.5. REFINED GRADINGS 185
(s) = (1(s)) G
t
(4/) satisfying `

([(s)]) = s s
0
. The grading on the algebra is then
dened by
gr
_
1(s)c()1(t)
_
:= (s) gr
t
(c())(t)
1
(Equation (3.44)).
We now turn to rened gradings on domains. Recall rst that every generator x S(1)
has an associated set o(x) of occupied -arcs (Denition 4.13) and associated idempotents
1
A
(x) = 1(o(x)) and 1
D
(x) = 1([2/] o(x)). Then, for x. y S(1) and 1
2
(x. y), dene
p(1) := (1
A
(x))p
t
(1)(1
A
(y))
1
(10.30)
1(x) := p(1) [ 1
2
(x. x). (10.31)
Lemma 10.32. The function p satises p(1) G(:) and p(1
1
1
2
) = p(1
1
) p(1
2
).
Proof. To verify that p(1) G(:), compute
`

([p(1)]) = `

([(1
A
(x))]) + `

([p
t
(1)]) `

(([(1
A
(y))]))
= (o(x) s
0
) + (o(y) o(x)) (o(y) s
0
)
= 0.
The fact that p(1
1
1
2
) = p(1
1
) p(1
2
) is immediate from the denitions and Lemma 10.4.

In particular, 1(x) is a subgroup of G(:).

Denition 10.33. To dene the rened grading on

CFA(1. s), dene o
A
(1. s) to be G(1)
(independent of 1 and s), and set gr(c
B
0
x) := p(1
0
).
To dene the rened grading on

CFA(1. s), suppose rst that S(1. s) ,= . Pick any
x
0
S(1. s). Dene o
A
(1. s) and gr(x) by
o
A
(1. s) := 1(x
0
)G(1)
gr(x) := 1(x
0
) p(1).
For completeness, if S(1. s) = choose a diagram 1
t
isotopic to 1 and so that S(1
t
. s) ,= ,
let o
A
(1. s) = o
A
(1
t
. s) and let gr: S(1. s) o
A
(1. s) be the unique map from the empty
set to o
A
(1
t
. s).
Proposition 10.34. The function gr: S(1. s) o
A
(1. s) above denes a grading on

CFA(1. s). Likewise, gr: S(1. s) o

A
(1. s) denes a grading on

CFA(1. s).
Proof. The proof follows the proofs of Propositions 10.9 and 10.17, noting that if
:
n+1
(x. c
1
. . . . . c
n
) ,= 0, the idempotent of x agrees with the left idempotent of c
1
and
the right idempotent of c
i
agrees with the left idempotent of c
i+1
, so the corresponding s
cancel.
To dene the rened gradings on

CFD and

CFD, we rst need to x grading renement
data for /(1. 0). We choose it as follows. Fix a choice of (s
0
)
7
and
7
determining a
grading renement for /(1. 0) and dene
(s
0
)
7
:= [2/] (s
0
)
7

7
(s) := 1(([2/] s))
1
.
186 10. GRADINGS
With this choice,
`

([
7
(s)]) = `

([
7
([2/] s)]) = (([2/] s) (s
0
)
7
) = s (s
0
)
7
.
so (s
0
)
7
and
7
(s) dene a grading renement on /(1. 0); we call this the reverse
of the grading renement (s
0
)
7
and
7
. Then, for 1
2
(x. y),
(10.35)
1(p(1)) = 1
_

7
(1
A
(x))p(1)
7
(1
A
(y))
1
_
=
7
([2/] 1
A
(y))1(p(1))
7
([2/] 1
A
(x))
1
=
7
(1
D
(y))1(p(1))
7
(1
D
(x))
1
.
Denition 10.36. To dene the rened grading on

CFD(1. s), dene o
D
(1. s) to be
G(1) (independent of 1 and s), and set gr(c
B
0
x) := 1(p(1
0
)).
To dene the rened grading on

CFD(1. s), suppose rst that S(1. s) ,= . Pick any
x
0
S(1. s) and dene o
D
(1. s) and gr(x) by
o
D
(1. s) := G(1),1(1(x
0
))
gr(x) := 1(p(1)) 1(1(x
0
)).
For completeness, if S(1. s) = choose a diagram 1
t
isotopic to 1 and so that
S(1
t
. s) ,= , let o
D
(1. s) = o
D
(1
t
. s) and let gr: S(1. s) o
D
(1. s) be the unique
map from the empty set to o
D
(1
t
. s).
Proposition 10.37. The function gr: S(1. s) o
D
(1. s) above denes a grading on

CFD(1. s). Likewise, gr: S(1. s) o

D
(1. s) denes a grading on

CFD(1. s).
Proof. The proof follows the proofs of Propositions 10.22 and 10.28, noting that if
cx ,= 0, the right idempotent of c agrees with the idempotent of x, so the intermediate
cancels.
We have the following analogue of the graded invariance theorems (Theorems 10.29
and 10.18) with respect to the rened gradings.
Theorem 10.38. Up to graded /

homotopy equivalence, the G(1)-set graded /

module

CFA(1. s) over /(1) is independent of the choices of provincially admissible Heegaard

diagram, suciently generic admissible almost complex structure and base generator used to
dene it; and up to graded homotopy equivalence, the G(1)-set graded dierential module

CFD(1. s) over /(1) is independent of the choices of provincially admissible Heegaard

diagram, suciently generic admissible almost complex structure and base generator used
to dene it. Similar statements hold for F
2
[H
2
(1. 1 )]/(1) and F
2
[H
2
(1. 1 )]/(1)
bimodules

CFA(1. s) and

CFD(1. s).
Proof. This is immediate from the denitions and Theorems 10.29 and 10.18.
Remark 10.39. Theorem 10.38 does not assert that the rened gradings are independent
of the grading renement data. For some discussion of this point, see Remark 3.46.
10.6. TENSOR PRODUCT 187
10.6. Tensor product
We now turn our attention to a graded version of the pairing theorem. Let (1
1
.
1
:
1(:) 1
1
) and (1
2
.
2
: 1(:) 1
2
) be bordered three-manifolds which agree along
their boundary. As discussed in Section 4.5, given bordered Heegaard diagrams 1
i
for the
(1
i
.
i
), gluing 1
1
and 1
2
along their boundary gives a Heegaard diagram 1 = 1
1

1
2
for 1 = 1
1

1
2
.
For each spin
c
structure s
1
over 1
1
,

CFA(1
1
. s
1
) is graded by the set 1
1
(x
1
)G(:),
where x
1
S(1
1
) is a chosen base generator representing s
1
, and 1
1
(x
1
) denotes the image
of
2
(x
1
. x
1
) in G(:). Similarly, for each spin
c
structure s
2
over 1
2
,

CFD(1
2
. s
2
) is graded by
the set G(:),1(1
2
(x
2
)), where x
2
S(1
2
) is a chosen base generator representing s
2
, and
1
2
(x
2
) denotes the image of
2
(x
2
. x
2
) in G(:). (Here the grading renement on G(:)
is the reverse of the grading renement on G(:), so that the grading renement used in the
denition of the grading of

CFA(1
1
. s
1
) is the same as the grading renement used in the
denition of

CFD(1
2
. s
2
).) Correspondingly, the tensor product

CFA(1
1
. s
1
)

CFD(1
2
. s
2
)
inherits a grading gr

with values in the double-coset space

1
1
(x
1
)G(:),1(1
2
(x
2
)).
We choose the base generators x
1
and x
2
to occupy complementary -arcs, i.e., so that
x
1
x
2
S(1). Let x := x
1
x
2
.
The above double-coset space inherits an action by Z (acting on the Maslov component).
The quotient by this action is naturally identied with the quotient of H
1
(1) by the image of
H
2
(1
1
. 1) H
2
(1
2
. 1) under the boundary homomorphism, or, equivalently, with the image
of H
1
(1) in H
1
(1 ). More succinctly, we have
1
1
(x
1
)G,1(1
2
(x
2
))

[]

_
1
1
(x
1
)G,1(1
2
(x
2
))
_
,Z

=

Im
_
H
1
(1) H
1
(1 )
_
.
where [] denotes the map from the double-coset space to H
1
(1 ). A more invariant
version is
1
1
(x
1
)G,1(1
2
(x
2
))

_
1
1
(x
1
)G,1(1
2
(x
2
))
_
,Z

=

_
s spin
c
(1 )

s[
Y
i

= s
i
. i = 1. 2
_
.
where the map is dened by () := s(x) + []. We are implicitly using Poincar duality
to think of spin
c
(1 ) as an ane space over H
1
(1 ) (rather than the usual H
2
(1 )).
Let div(c
1
(s
t
)) be the divisibility of the rst Chern class of s
t
, i.e.,
H
1
(1 ; Z) c
1
(s
t
) = div(c
1
(s
t
)) H
3
(1 ; Z).
Lemma 10.40. For s
t
spin
c
(1 ),
1
(s
t
) is isomorphic as a Z-set to Z,div(c
1
(s
t
)).
Proof. Suppose that G(:) is a group element with the property that and
t

represent the same double coset. This means that there are 1
i

2
(x
i
. x
i
) with the property
that
=
t
p(1
1
) 1(p(1
2
)).
188 10. GRADINGS
Let
0
= (1
A
(x
1
)) = (1
D
(x
2
)). Then, by (10.30) and (10.35),
=
t

0
p
t
(1
1
)
1
0

0
1(p
t
(1
2
))
1
0
or, setting
t
=
1
0

0
,

t
=
t
p
t
(1
1
)
t
1(p
t
(1
2
)). (10.41)
Note that [] and [
t
] are the same element of H
1
(1).
This implies that

1
1
+

1
2
= 0, so 1
1
and 1
2
glue to a domain 1 = 1
1
1
2

2
(x. x)
(Lemma 4.32). Using the formula c(1) + 2:
x
(1) = c
1
(s(x)). [1] ([OSz04c, Proposi-
tion 7.4]), Equation (10.41) implies
1 =
t

e(B)2nx(B)
(0.

1
1
)
t
(0.

1
1
)
t1
=
tc
1
(s(x)),[B])2[][

B
1
]
=
tc
1
(s(x)),[B])2[][B]
=
tc
1
(()),[B])
.
where [][

1
1
] denotes the algebraic intersection number of [] and [

1
1
] in H
1
(1), while
[] [1] denotes the algebraic intersection number of [] H
1
(1 ) and [1] H
2
(1 ). The
second equality uses Equation (3.42) and the last uses the identity c
1
(()) = c
1
(s(x))+2[].
From this (and the fact that any periodic domain 1 on 1 can be decomposed into periodic
domains 1
1
and 1
2
(Lemma 4.32), the result follows.
Theorem 10.42. Fix s
i
spin
c
(1
i
). There is a homotopy equivalence
:

CFA(1
1
. s
1
)

CFD(1
2
. s
2
)

sspin
c
(Y )
s[
Y
i
=s
i
,i=1,2

CF(1. s)
which respects the identication between grading sets in the following sense:
(1) If :
1
:
2
is a homogeneous element with grading gr

(:
1
:
2
) 1
1
(x
1
)G,1
2
(x
2
),
then (:
1
:
2
) lies in the (gr

(:
1
:
2
)) spin
c
(1 ) summand of

CF(1 ).
(2) If :
1
:
2
and :
1
:
2
are homogeneous elements whose gradings are related by
gr

(:
1
:
2
) =
t
gr

(:
1
:
2
).
so that (gr

(:
1
:
2
)) = (gr

(:
1
:
2
)) = s
t
spin
c
(1 ) then (:
1
:
2
)
and (:
1
:
2
) are elements whose relative Maslov grading is given by t modulo
div(c
1
(s
t
)).
Proof. The map is the homotopy equivalence from Theorem 1.3. We must prove that
respects the grading sets.
To dene the identication of grading sets we chose complementary generators x
1
and x
2
with s(x
i
) = s
i
. (This is always possible after modifying 1
1
and 1
2
by an isotopy, which
changes

CFA(1
1
. s
1
) and

CFD(1
2
. s
2
) by a graded homotopy equivalence.) Assume for sim-
plicity that 1
A
(x
1
) = 1
D
(x
2
) is s
0
, the base idempotent used to dene the grading renement.
(Again, we may need to modify 1
1
and 1
2
by an isotopy to achieve this.) In this case, some
formulas simplify: 1(x
i
) = 1
t
(x
i
), and, for y
i
S(1
i
. s
i
),
gr(y
1
) = gr
t
(y
1
) (1
A
(y
1
))
1
and gr(y
2
) = (1
D
(y
2
)) gr
t
(y
2
).
10.6. TENSOR PRODUCT 189
Let y
1
S(1
1
) and y
2
S(1
2
) be generators with s(y
i
) = s
i
and such that y = y
1
y
2
is a generator for 1. Let 1
i

2
(x
i
. y
i
). Then
gr(y
1
) = 1
1
(x
1
) p(1
1
) o
A
(1
1
) = 1
1
(x
1
)G(:)
gr(y
2
) = 1(p(1
2
)) 1(1
2
(x
2
)) o
D
(1
2
) = G(:),1(1
2
(x
2
)).
Consequently,
gr(y
1
y
2
) = gr
t
(y
1
) (1
A
(y
1
))
1
(1
D
(y
2
)) gr
t
(y
2
)
= gr
t
(y
1
) gr
t
(y
2
).
where we have used 1
A
(y
1
) = 1
D
(y
2
). We wish to use this to calculate the dierence element
between the two spin
c
structures associated to x = x
1
x
2
and y = y
1
y
2
. The spin
c
component of gr
t
(y
1
) gr
t
(y
2
) is

1
1
+

1
2
. This represents a homology class in H
1
(1).
Moreover, the image of in H
1
(1 ) agrees with the dierence element c(x
1
x
2
. y
1
y
2
)
dened in [OSz04d, Section 2.6] (see also Lemma 4.21), whose Poincar dual is s
z
(x
1

x
2
) s
z
(y
1
y
2
). It follows that (gr
t
(y
1
) gr
t
(y
2
)) represents the same spin
c
structure as
y = y
1
y
2
. This veries Part 1.
For Part 2, let y
t
1
S(1
1
) and y
t
2
S(1
2
) be another pair of generators with s(y
i
) = s
i
and such that y
t
= y
t
1
y
t
2
is a generator for 1 in the same spin
c
structure as y. Fix some
1
t

2
(y. y
t
), where 1
t
= 1
t
1
1
t
2
. Thus 1
i
1
t
i

2
(x
i
. y
t
i
). Now, the grading of y
t
is
given by
gr(y
t
) = 1
1
(x
1
) p
t
(1
1
1
t
1
)1(p
t
(1
2
1
t
2
)) 1
2
(x
2
)
= 1
1
(x
1
) p
t
(1
1
)p
t
(1
t
1
)1(p
t
(1
t
2
))1(p
t
(1
2
)) 1
2
(x
2
)
= 1
1
(x
1
) p
t
(1
1
)
e(B

)ny(B

)n
y
(B

)
1(p
t
(1
2
)) 1
2
(x
2
)
=
e(B

)ny(B

)n
y
(B

)
gr(y).
In the second equality, we use Lemma 10.4 and the fact that 1 is an anti-homomorphism.
This proves Part 2.
Remark 10.43. To get the correct grading set on the tensor product, we need to use the
rened gradings: the double-coset space 1
t
1
(x
1
)G
t
(4/),1(1
t
2
(A
2
)) is larger than the double-
coset space 1
1
(x
1
)G(:),1(1
2
(A
2
)) in Theorem 10.42.
Remark 10.44. An alternative to the rened grading would be to consider a grading
groupoid, with objects corresponding to the dierent idempotents 1(s) and Mor(s
1
. s
2
) being
those elements p G
t
(4/) with `

([p]) = s
1
s
2
. Then o
A
and o
D
are equipped with
right and left actions of this grading groupoid, and the grading set for the tensor product is
the (suitably dened) product over the grading groupoid.
CHAPTER 11
Bordered manifolds with torus boundary
In this chapter, we specialize to the case of bordered manifolds with torus boundary. In
particular, we show that the bordered Floer invariant of a manifold with torus boundary is
closely related to its knot Floer homology. As a warm-up, after introducing some notation in
Section 11.1, we use the pairing theorem to give a quick proof of the surgery exact triangle,
in Section 11.2. We then turn to the relationship between bordered Floer homology for
manifolds with torus boundary and knot Floer homology. In Section 11.3 we recall the
conventions on knot Floer homology. The easier task of extracting (associated graded) knot
Floer homology from bordered Floer homology is addressed in Section 11.4. This follows
quickly from a suitable adaptation of the pairing theorem and a simple model calculation
for the solid torus.
Following this, Section 11.5 explains how to extract the type 1 module for a knot comple-
ment in o
3
from the (ltered) knot Floer homology. After further developing the holomorphic
curve machinery in Section 11.6, this result is proved in Sections 11.7 and 11.8. Finally, in
Section 11.9, we use bordered Floer homology to study satellite knots. (The knot Floer
homology groups of various families of satellite knots have also been studied using a more
direct analysis. See [Eft05], [Hed07], and [Hed05b].)
11.1. Torus algebra
A genus p bordered Heegaard diagram 1 for a 3-manifold 1 with torus boundary has two
-arcs, p 1 -circles, and p -circles. The -arcs are necessarily arranged as in Figure 11.1.
As a result, the parametrization of 1 is specied by a pair of a meridian and a longitude,
intersecting once, and the matching on 1 is always the same. Let /() = /(. 0) denote
the i = 0 part of the corresponding algebra. Since we will mostly be interested in

CFD(1),
we label the regions between the arcs in the opposite order to the induced orientation on
the boundary, as in Figure 11.2. (From the type side, the regions would be labeled in
0
.
1 2
3

a
2

a
2

a
1

a
1
0
.
1 2
3

a
2

a
2

a
1

a
1
Figure 11.1. Bordered Heegaard diagrams for torus boundary. Left:
the necessary arrangement of the -arcs. Opposite sides are identied as indi-
cated. There may be additional handles attached with corresponding -circles.
Right: a -circle to create an admissible diagram for a solid torus.
191
192 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
0
.
1 2
3

a
1

a
1

a
2

a
2
or
0
1
2
3

a
1

a
2

a
1

a
2
Figure 11.2. Labeling of the regions around . With respect to the
boundary orientation, they are labeled 0. 3. 2. 1 with . in the region labeled 0.
the opposite order around the puncture.) The point . is in the region labeled 0. Label the
-arcs so that regions 0 and 1 are separated from regions 2 and 3 by
a
2
.
The algebra /() has two minimal idempotents
0
and
1
, corresponding respectively
to
a
1
and
a
2
being occupied on the type side, and 6 other basic generators, denoted
graphically as follows:

1
:
2
:
3
:

12
:
23
:
123
:
The dierential is zero, and the non-zero products are

2
=
12

2

3
=
23

1

23
=
123

12

3
=
123
.
(All other products of two non-idempotent basic generators vanish identically.) There are
also compatibility conditions with the idempotents:

1
=
0

1

2
=
1

0

3
=
0

12
=
0

12

0

23
=
1

23

1

123
=
0

123

1
.
The unrened grading takes values in the group G
t
(), which we denote G
t
in this section.
It is generated by quadruples (,; c. /. c) with ,
1
2
Z, c. /. c Z and , is an integer if all of
c. /. c are even or if / is even and c and c are odd, and is a half-integer otherwise. The group
law is
(,
1
; c
1
. /
1
. c
1
) (,
2
; c
2
. /
2
. c
2
) =
_
,
1
+ ,
2
+
1
2

c
1
/
1
c
2
/
2

+
1
2

/
1
c
1
/
2
c
2

;
c
1
+ c
2
. /
1
+ /
2
. c
1
+ c
2
_
.
11.1. TORUS ALGEBRA 193
The G
t
-grading on the algebra is given by
gr
t
(
1
) =
_

1
2
; 1. 0. 0
_
gr
t
(
2
) =
_

1
2
; 0. 1. 0
_
gr
t
(
3
) =
_

1
2
; 0. 0. 1
_
.
The distinguished central element is = (1; 0. 0. 0).
There is also a rened grading, as in Section 3.3.2, with values in the group G(). We
will instead give a grading with values in a somewhat larger group G given explicitly by
triples (,; j. ) where ,. j.
1
2
Z and j + Z. The group law is
(,
1
; j
1
.
1
) (,
2
; j
2
.
2
) =
_
,
1
+ ,
2
+

j
1

1
j
2

2

; j
1
+ j
2
.
1
+
2
_
and the distinguished central element is = (1; 0. 0). (The group G() from Section 3.3.2
would have j. Z and 2, j+ (mod 2).) Let G
t

G
t
be the group dened in the same
way as G
t
but where ,. c. /. c (so G
t

is a -central extension of H
1
(2. a; )). There is a
homomorphisms G G
t

dened by (,; j. ) (,; j. j+. ). To dene the rened grading,

choose
0
as the base idempotent and set (
1
) :=
_
0;
1
2
. 0.
1
2
_
G
t

. From Equation (3.44)

we then nd
gr(
1
) =
_

1
2
;
1
2
.
1
2
_
gr(
2
) =
_

1
2
;
1
2
.
1
2
_
gr(
3
) =
_

1
2
;
1
2
.
1
2
_
.
(11.1)
Note that for any algebra element c, the gradings gr
t
(c) and gr(c) are related by the homo-
morphism from G
t
to G dened by
(11.2) (,; c. /. c)
_
,;
c + / c
2
.
c + / + c
2
_
.
Since the torus algebra is so simple, we can understand the structure of type 1 modules
over it more explicitly. (This will be helpful in Section 11.7, where we understand how to
construct the type 1 module of a knot complement from the knot Floer complex.)
Denition 11.3. Let \ = \
0
\
1
be a Z,2Z-graded vector space. A collection of (torus)
coecient maps is a collection of maps
1
I
: \
[i
0
1]
\
[in]
indexed by increasing sequences of consecutive integers 1 = i
0
. . i
n
1. 2. 3, where
[i] denotes the reduction of i (mod 2), satisfying the compatibility equation stated below.
(The empty sequence also has a corresponding map 1

or just 1, which maps \

i
to \
i
for
i = 0. 1.) For any increasing sequences of consecutive integers 1 1. 2. 3 we require that
(11.4)

I=JK[J<K
1
K
1
J
= 0.
where J < 1 means that each element of J is smaller than every element of 1.
194 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
Explicitly, Equation (11.4) in the case where 1 = says that 1 is a dierential; similarly,
1 = 1 says that 1
1
is a chain map with respect to the dierential 1; 1 = 1. 2 says that
1
12
is a null-homotopy of the composite 1
2
1
1
; 1 = 123 gives the following:
1 1
123
+ 1
3
1
12
+ 1
23
1
1
+ 1
123
1 = 0.
Lemma 11.5. A type 1 structure in the sense of Denition 2.18 over /() with base ring
J() corresponds to a pair of Z,2Z-graded vector spaces equipped with coecient maps in
the sense of Denition 11.3.
Proof. Let \ be a vector space underlying a type 1 structure over /(). The projection
\ \
0
is induced by multiplication by the idempotent
0
, while the projection \ \
1
is
induced by multiplication by the idempotent
1
. The coecient maps can be extracted from
the map
1
by the formula

1
= I 1 +

i
1
i
+

i,j[j=i+1

ij
1
ij
+
123
1
123
.
Equation (11.4) gives the components of the compatibility condition for
1
.
In particular, thinking of

CFD(1 ) as a type 1 structure, as in Remark 6.4, we can talk
about its coecient maps.
Remark 11.6. The coecient maps depend on the type 1 structure itself, not just on its
corresponding dierential module.
11.2. Surgery exact triangle
Recall that Heegaard Floer homology admits a surgery exact triangle [OSz04c]. Specif-
ically, for a pair (`. 1) of a 3-manifold ` and a framed knot 1 in `, there is an exact
triangle
(11.7)

HF(`
1
)

HF(`
0
)

HF(`

)
where `
1
, `
0
, and `

are 1, 0, and innity surgery on 1, respectively. As a simple

application of bordered Floer theory, we prove a version of this result. A similar computation
was carried out in [Lip06b, Section 5.3]. We do not, however, verify that the maps in our
present triangle agree with those in the original version; see [LOTa] or [LOTb].
We prove this exact triangle by constructing three provincially admissible bordered Hee-
gaard diagrams 1
1
, 1
0
, and 1

for the three solid tori lled at the corresponding slopes,

and exhibiting a short exact sequence relating the chain complexes

CFD(1

). For any
knot complement with (admissible) bordered Heegaard diagram 1
t
, we can then take the
derived tensor product of

CFA(1
t
) with this short exact sequence. The exact triangle in
Equation (11.7) is the induced exact triangle on homology (see Proposition 2.36).
11.3. PRELIMINARIES ON KNOT FLOER HOMOLOGY 195
The three diagrams we use are
(11.8) 1

:
0
.
1 2
3
:
:
1
1
:
0
.
1 2
3
c c
/
/
1
0
:
0
.
1 2
3
: :
A generator for

CFD(1

) consists of a single intersection point between the -circle in 1

and an -arc. These intersections are labeled above.

The boundary operators on the

CFD(1

) (and the support of the relevant domains) are

given by
: =
23
:
.
c =
3
/
.
+
1
/
.
/ = 0 : =
12
:
.
There is a short exact sequence
0

CFD(1

CFD(1
1
)

CFD(1
0
) 0
where the maps and are given by
(:) = /
.
+
2
c
.
(c) = :
.
(/) =
2
:
.
.
Indeed, one can guess the maps and by considering holomorphic triangles, as indicated
above. It is straightforward to verify that = 0, that is injective, that is surjective,
and that ker = im.
11.3. Preliminaries on knot Floer homology
We focus now on the relationship between Heegaard Floer invariants of three-manifolds
with torus boundary and knot Floer homology. Before delving into the details, we recall
some of the basics of knot Floer homology, with a special emphasis on the bigradings. For
simplicity of notation, we restrict attention to the case where the ambient three-manifold
is o
3
. More detailed accounts may be found in the original papers [OSz04b, Ras03].
The knot Floer invariant of a knot is the ltered chain homotopy type of a ltered chain
complex over F
2
[l]. Specically, a knot in o
3
can be specied by a suitable Heegaard
diagram 1
K
= (. . . u. .), where u and . are a pair of basepoints in the complement
of the - and -circles. The knot Floer complex is dened as a chain complex CFK

(1)
generated over F
2
[l] by S
K
, the usual generators of Heegaard Floer homology with respect
to the Heegaard diagram 1
K
, endowed with the dierential
(11.9)

(x) :=

yS
K

B
2
(x,y)
ind(B)=1
#
_
/
B
(x. y)
_
l
nw(B)
y.
196 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
In the above equation,
2
(x. y) refers to the set of homology classes of curves connecting x
to y which are allowed to cross both u and .; this is denoted
2
(x. y) in [OSz04b, Ras03],
but in this book we have reserved
2
for classes which have local multiplicity zero at ..
The complex is endowed with an Alexander ltration. The Alexander depth (ltration
degree) of the generators is characterized, up to an overall translation, by
(11.10)
(x) (y) = :
z
(1) :
w
(1)
(l x) = (x) 1
for any x. y S
K
and homology class 1
2
(x. y). The boundary operators do not
increase . Furthermore, the boundary operators decrease (by one) the Maslov grading `,
dened up to overall translation by
`(x) `(y) = ind(1) 2:
w
(1) (11.11)
`(l x) = `(x) 2. (11.12)
(Note that the knot enters the denition only through the Alexander ltration.) We will
explain how to remove the indeterminacy in the Alexander and Maslov gradings below, in
Equations (11.16) and in Equations (11.15) and (11.17), respectively.
In some contexts, it is natural to consider a dierent grading `, the .-normalized Maslov
grading, characterized up to an overall translation, by
`(x) `(y) = ind(1) 2:
z
(1) (11.13)
`(l x) = `(x). (11.14)
Any two of , `, and ` determine the remaining one, by the relation
(11.15) ` = ` 2.
Let gCFK

(1) denote the associated graded object (with respect to the Alexander
ltration). Explicitly, this is a bigraded complex with a dierential

(x) :=

yS
K

B
2
(x,y)
ind(B)=1,nz(B)=0
#
_
/
B
(x. y)
_
l
nw(B)
y.
The functions and ` induce Alexander and Maslov gradings on gCFK

(1); the Alexander

grading is preserved by the dierential, and the Maslov grading is decreased by one by the
dierential.
The homology groups of gCFK

(1) are the knot Floer homology groups of 1,

HFK

(1) =

r,d
HFK

d
(1. :).
The index : is the Alexander grading and d is the Maslov grading.
The above constructions can be specialized to l = 0, where we have a dierent version of
knot Floer homology. The l = 0 specialization of CFK

(1) is a ltered chain complex over

F
2
denoted

CFK(1). We denote the associated graded complex to

CFK(1) by

gCFK(1).
Explicitly,

gCFK(1) is a bigraded complex generated over F
2
by S
K
, and endowed with the
dierential

(x) :=

yS
K

B
2
(x,y)
ind(B)=1,nz(B)=nw(B)=0
#
_
/
B
(x. y)
_
y.
11.3. PRELIMINARIES ON KNOT FLOER HOMOLOGY 197
(Sometimes, when we wish to call attention to a xed doubly-pointed Heegaard diagram, we
denote

gCFK(1) by

gCFK(1. u. .).) The homology of

gCFK(1) is denoted

HFK(1) =

r,d

HFK
d
(1. :).
The indeterminacy (up to an overall translation) of the Alexander grading can be removed
by requiring that the function have the following symmetry property for all : Z:
(11.16) #x S
K
[ (x) = : #x S
K
[ (x) = : (mod 2).
The graded Euler characteristic of

HFK coincides with the (symmetrized) Alexander poly-
nomial
K
in the sense that

d,r
(1)
d
rank

HFK
d
(1. :) 1
r
.
=
K
(1).
where
.
= indicates that the two Laurent polynomials agree up to multiplication by 1
k
(for
some /). The symmetry property from Equation (11.16), which pins down the indeterminacy
in the Alexander grading, can be reformulated as the condition that

d,r
(1)
d
rank

HFK
d
(1. :) 1
r
=
K
(1).
The indeterminacy in the Maslov grading can be removed by rst removing the inde-
terminacy in the .-normalized Maslov grading, and demanding that Equation (11.15) con-
tinue to hold. In turn, to pin down the algebraic grading `, consider the chain complex
gCFK

(1),(l = 1). Note that neither the Alexander nor the Maslov gradings descend
to this complex; but the .-normalized Maslov grading does. The homology of the result-
ing complex has rank one. The algebraic grading is then normalized by requiring that this
homology is supported in grading equal to zero; i.e.,
(11.17) H

_
gCFK

(1)
l = 1
_
=
_
F
2
= 0
0 otherwise,
where denotes the algebraic grading `.
It is more traditional to dene the Maslov gradings in terms of holomorphic disks cross-
ing the basepoint .. This is normalized by considering

CF(o
3
), dened as holomorphic
disks which do not cross the basepoint u (i.e., set l = 0 in the chain complex dened in
Equation (11.9) and ignore its Alexander ltration), and require that the resulting homology
is supported in degree zero with respect to the Maslov grading. Our present formulation
denes gradings in terms of disks which never cross .. The equivalence of the two points of
view follows from the fact that gCFK

(1),(l = 1) is the chain complex

CF(o
3
), using the
basepoint . in place of u.
Knot Floer homology naturally gives rise to an integral valued concordance invariant for
knots [OSz03]. One construction of comes from considering the ltered chain homotopy
type of

CFK(o
3
). From this point of view, (1) is the minimal : for which the generator
of H

CF(o
3
)) can be represented as a sum of generators in Alexander grading less than or
equal to :. In an alternative formulation, has an interpretation from the associated graded
object HFK

(1), namely,
(11.18) (1) = max: [ d 0. l
d
HFK

(1. :) ,= 0.
198 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
(See, for example, [OSzT08, Lemma A.2].)
The above denition of the Alexander grading carries over with little change to the more
general case where 1 1 is a null-homologous knot in an arbitrary closed, oriented three-
manifold (rather than just o
3
). The Maslov grading is a little more subtle; but that is
because Maslov gradings on closed manifolds are more complicated. Our present applica-
tions, however, primarily concern the case where 1 = o
3
.
11.4. From

CFD to HFK

Let 1 be a null-homologous knot in a 3-manifold 1 . Recall from Section 4.2 that we can
construct a bordered Heegaard diagram for 1 nbd(1) as follows. Fix a Heegaard diagram
(
g
.
c
1
. . . . .
c
g1
.
1
. . . . .
g
) for 1 nbd(1), in the classical sense. Let j and
denote a meridian and longitude for 1, respectively. Arrange that and j intersect in a
single point j, and that they are disjoint from all
c
i
. Set
a
1
:= j and
a
2
:= j j.
Then
( j.
a
1
.
a
2
.
c
1
. . . . .
c
g1
.
1
. . . . .
g
)
is a bordered Heegaard diagram for 1 nbd(1).
Number the four regions in (. . ) adjacent to j, in counter-clockwise cyclic order (i.e.,
the opposite of the induced boundary orientation), so that the meridian
a
2
separates regions
0 and 3.. Let . and u be basepoints in regions 0 and 2, respectively. (See Figure 11.8.) Then
(. j.
c
1
. . . . .
c
g1
.
1
. . . . .
g
)
is a doubly-pointed Heegaard diagram for 1 in 1 . So, with notation as in Section 11.1,
for the Heegaard diagram (. . . .) it is immediate that (\
0
. 1) is the knot Floer chain
complex

CFK(1. 1). (The chain complex (\
1
. 1) is the longitude Floer complex studied by
Eftekhary [Eft05].)
It follows that for any bordered Heegaard diagram (. . . .) for 1 nbd(1), with
framing such that
a
2
is a meridian of 1, the complex (\
0
. 1) is homotopy equivalent to

CFK(1. 1). To see this, dene an /-module \

0
to be one-dimensional over F
2
with
0
acting as the identity and all other basic generators of / acting by 0. Then the tensor
product \
0

CFD(. . . .) is exactly (\
0
. 1). By Lemma 2.32, the homotopy type of
this tensor product depends only on the homotopy type of

CFD(. . . .). Thus, we have
the following.
Proposition 11.19. Let (1. .) be any bordered Heegaard diagram for 1 nbd(1), with
framing such that
a
2
is a meridian of 1. Then, with notation as in Section 11.1, (\
0
. 1) is
homotopy equivalent to

CFK(1. 1). In particular,

CFK(1. 1) is determined by

CFD(1. .).
The relative Maslov grading on

CFK(1. 1) within each spin
c
structure (on the knot
complement) is induced from the grading on

CFD(1. .) as follows. Two generators represent
the same spin
c
-structure on 1 1 if and only if they can be connected by a provincial domain.
Consequently, the gradings of two such generators dier by a power of , and this -power is
their Maslov grading dierence. One can also determine the Alexander grading, the relative
Maslov grading between generators in dierent Alexander gradings and, in the case that
1 = o
3
, the absolute Maslov grading; we return to these points later.
There is a simple geometric interpretation of \
0
, which we explain after a more general
discussion.
11.4. FROM

CFD TO HFK

199
Figure 11.3. Turning a bordered diagram for 1
t
= 1 nbd(1 )
into a doubly-pointed bordered diagram for (1. 1). This diagram can
be thought of as a bordered Heegaard diagram with two basepoints for the
standard 1-handle attachment cobordism ` from a surface of genus p to a
surface of genus p +1. Gluing the left boundary of this diagram, or an obvious
analogue, to a bordered Heegaard diagram for (1
t
. 1
t
= 1
2
#
g
), so that
the top portion of this diagram is glued to the 1
2
summand in 1
t
, gives a
doubly-pointed Heegaard diagram for (1
t

`. 1), where 1 is the core of

the 1-handle in `.
A doubly-pointed bordered Heegaard diagram is a bordered Heegaard diagram (. . )
for (1. 1 ), together with basepoints . and u in (). We further assume . . A
doubly-pointed bordered Heegaard diagram species a knot 1 in 1 : let

(respectively

)
denote a path in (respectively ) from . to u. Let

denote the result of pushing

the interior of

into the -handlebody and

the result of pushing the interior of

into
the -handlebody. Then 1 =

. We orient 1 as

.
Every knot in a bordered three-manifold can be realized by a doubly-pointed bordered
Heegaard diagram. Specically, let 1 be a bordered three-manifold with genus p boundary,
and let 1 1 be a knot. By connecting 1 to the boundary by an arc and deleting a
regular neighborhood of 1, we can write 1 as the union of a three-manifold 1
t
with genus
p +1 boundary and a standard cobordism ` (a two-handle attachment) from the surface of
genus p +1 to a surface with genus p. Gluing the (partial) diagram shown in Figure 11.3 to
a bordered Heegaard diagram for 1
t
gives a doubly-pointed bordered Heegaard diagram for
1 in 1 .
To a doubly-pointed bordered Heegaard diagram (1. .. u), we can associate type 1 and
type modules as before, now working over a ground ring of F
2
[l], with the understanding
that holomorphic disks crossing u with multiplicity : contribute a factor of l
n
. (Holomor-
phic disks crossing . are still forbidden.) We denote the resulting modules by CFA

(1. .. u)
and CFD

(1. .. u). Setting l = 1 recaptures

CFA(1. .) and

CFD(1. .), while setting
l = 0 gives a pair of modules denoted

CFA(1. .. u) and

CFD(1. .. u) where we count only
those holomorphic disks with multiplicity zero at u.
Remark 11.20. We do not claim that these new modules CFD

(1. .. u), etc., are invariants

of the pair (1. 1). Because the proofs of invariance in Sections 6.3 and 7.3 require one
200 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
basepoint, ., to remain on the boundary in the course of the handleslides and isotopies, the
module CFD

(1. .. u) is a priori an invariant of the graph in (1. 1 ) obtained by attaching

a point in 1 to 1 .
For a doubly-pointed bordered Heegaard diagram with boundary :, we can enhance the
grading set to include an Alexander grading. Specically, the enhanced grading group has
the form

G(:) = G(:) Z. (This is a direct product.) We call the new Z summand the
Alexander factor. Given x. y S(1) and 1
2
(x. y), dene
p(1) := (p(1). :
w
(1) :
z
(1)).
where p(1) is the grading from Equation (10.30). We can use this enhanced grading group to
dene an enhanced grading on CFD

(1. .. u). Following the discussion from Section 10.4,

a spin
c
structure s on 1 gives rise to a grading on the type 1 module with values in the
coset space

o
D
(1. s) := (

G,

1(x
0
)).
where x
0
is any intersection point representing s, and

1(x
0
) denotes the image of the space of
periodic domains in

G under the homomorphism p. Namely, we dene gr(x) := p(1)

1(x
0
),
where 1
2
(x. x
0
). Note that when we extend this grading to CFD

(1. .. u), we take

the convention that multiplication by l drops the Alexander grading by one and the Maslov
grading by two. That is, gr(lr) = (
2
. 1) gr(r).
Suppose next that (1
1
. .) is a bordered Heegaard diagram for (1
1
. 1
1
= 1) and that
(1
2
. .. u) is a doubly-pointed bordered Heegaard diagram for (1
2
. 1
2
= 1), equipped
with the knot 1. Then, the grading set of the tensor product

CFD(1
1
. .) CFA

(1
2
. .. u)
is naturally the double-coset space

1(x
1
)

G,

1(x
2
). (In the denition of

1(x
1
), we take the
convention that :
w
(1) = 0 for any domain 1 since u is thought of as located on the 1
2
side
of the Heegaard diagram.)
Projecting onto the Alexander factor in the double-coset space induces a map from the
grading set to a transitive Z-set. Indeed, the transitive Z-set is isomorphic to Z,:
w
(
2
(x. x))
for any generator x of 1
1

Z
1
2
. For any 1
2
(x. x) representing the homology class
H
2
(1
1
1
2
), the quantity :
w
(1) is the intersection of 1 with . In particular, if 1
is null-homologous, the Z-set is (non-canonically) isomorphic to Z. In this way, when 1 is
an integer homology sphere, the tensor product CFA

(1
1
. u. .)

CFD(HF
2
. .) inherits a
(relative) Z grading, which we call its Alexander grading.
We now have a pairing theorem for knot Floer homology.
Theorem 11.21. Let (1
1
. .. u) = (.
1
.
1
. .. u) be a doubly-pointed bordered Heegaard di-
agram for (1
1
. 1
1
= 1. 1) and (1
2
. .) = (
2
.
2
.
2
. .) be a bordered diagram for (1
2
. 1
2
=
1). Then, there are homotopy equivalences
gCFK

(1
1

F
1
2
. 1) CFA

(1
1
. u. .)

CFD(1
2
. .)

CFA(1
2
. .) CFD

(1
1
. u. .)

gCFK(1
1

F
1
2
. 1)

CFA(1
1
. u. .)

CFD(1
2
. .)

CFA(1
2
. .)

CFD(1
1
. u. .).
which respect gradings in an obvious sense. In particular, when 1 is null-homologous for
1
1

F
1
2
, the above homotopy equivalences respect the relative .-normalized Maslov and
11.4. FROM

CFD TO HFK

201
0
1 2
3
. u
a
2

a
2

a
1

a
1
Figure 11.4. The doubly-pointed Heegaard diagram 1
st
for the lon-
gitude in the solid torus. This is a genus 1 Heegaard diagram with bound-
ary, with -arcs
a
1
and
a
2
, and a -circle
1
intersecting
a
1
in a single point
and disjoint from
a
2
. Note that the boundary markings follow the conventions
of a type module.
Alexander gradings. (The Alexander and .-normalized Maslov gradings on the knot Floer
homology appearing on the left-hand-side are as recalled in Section 11.3; the relative Alexan-
der grading on the right-hand-side is dened immediately above, and the relative Maslov
grading is induced by projecting onto the Maslov component of the double-coset space).
Proof. The identication of complexes follows from an extension of either of the two
proofs of Theorem 1.3 to the doubly-pointed context, without any signicant change. The
statement about Alexander gradings now follows from Lemma 4.32.
Proposition 11.19 can be seen as a special case of Theorem 11.21, as follows. Consider the
doubly-pointed Heegaard diagram1
st
with boundary for the solid torus shown in Figure 11.4.
If we restrict to holomorphic curves which cross neither . nor u, the resulting type module

CFA(1
st
) has a single generator over F
2
, no dierential, and indeed is isomorphic to the
module \
0
described before the statement of Proposition 11.19. Thus, Proposition 11.19
follows from the l = 0 version of Theorem 11.21 (with the solid torus on the type side).
To reconstruct CFK

of a knot from

CFD, we must describe CFA

(1
st
). But this is
easily done.
Lemma 11.22. The type module CFA

(1
st
) for 1
st
, representing the solid torus equipped
with a longitudinal unknot, has a single generator x
0
and higher products given by
:
3+i
(x
0
.
3
.
i
..

23
. . . . .
23
.
2
) = l
i+1
x
0
for all i 0.
Proof. The only curves which can contribute are disks covering the region containing u
a total of / times. There is exactly one such disk for each /, and it contributes the term in
the statement with / = i + 1.
Thus, according to Theorem 11.21, HFK

(1. 1) can be calculated as a derived tensor

product of

CFD(1 1) with CFA

(1
st
), generalizing Proposition 11.19.
In a dierent direction, Theorem 11.21 gives a technique for studying the knot Floer
homology of satellite knots. We return to this point in Section 11.9 (where we give also some
examples), contenting ourselves for now with the following corollary.
Corollary 11.23. Suppose that 1
1
and 1
2
are knots in o
3
such that

CFD(o
3
1
1
)

CFD(o
3
1
2
) (with respect to the 0-framing on o
3
1
1
and o
3
1
2
, say). Let 1
C
1
(respectively
202 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
1
C
2
) denote the satellite of 1
1
(respectively 1
2
) with pattern C. Then HFK

(o
3
. 1
C
1
)

=
HFK

(o
3
. 1
C
2
).
Proof. According to Theorem 11.21, the knot Floer homology of the satellite knot is
obtained as the derived tensor product of the type 1 module of the companion knot (1
1
or
1
2
) with the type module of a doubly-pointed Heegaard diagram for the pattern knot in
the solid torus.
11.5. From CFK

to

CFD: Statement of results
We now turn to the opposite direction, computing

CFD(o
3
1) from CFK

(1), Theo-
rem 11.27 below. The statement will refer to some general algebraic properties of the chain
complex CFK

(1), which is a free, Z-ltered, Z-graded chain complex over F

2
[l]; and in
particular, it refers to certain convenient bases for this chain complex. After setting up the
needed algebra, we give the statement of the theorem.
Denition 11.24. Let \
i
\
i+1
be an (ascending) ltration of a vector space
\ , so \ =

i=
\
i
. Assume that

i=
\
i
= 0. Given \ , dene () := infi Z [
\
i
; we call () the ltration level of . Let gr(\ ) denote the associated graded vector
space, i.e.,
gr(\ ) :=

i=
gr
i
(\ ) gr
i
(\ ) := \
i
,\
i1
.
Composing the projection \
i
\
i
,\
i1
and the inclusion \
i
,\
i1
gr(\ ) gives a map

i
: \
i
gr(\ ). For \ dene [] :=
A(v)
().
A ltered basis of a ltered vector space \ is a basis
1
. . . . .
n
for \ such that
[
1
]. . . . . [
n
] is a basis for gr(\ ). Similarly, we say
1
. . . . .
m
is ltered linearly-in-
dependent if [
1
]. . . . . [
m
] is linearly independent.
There is a ltered analogue of the basis extension theorem: if
1
. . . . .
m
is ltered
linearly-independent then there is a ltered basis for \ containing
1
. . . . .
m
.
We now look at ltered F
2
[l] modules in general. Let C be a nitely-generated free
F
2
[l] module equipped with
an integer-valued grading, which we refer to as the Maslov grading `,
an integer-valued ascending ltration, the Alexander ltration , and
a dierential, denoted .
Assume these data satisfy the following compatibility conditions:
The dierential drops the Maslov grading by one: for homogeneous C,
`() = `() 1.
The dierential respects the Alexander ltration: () ().
Multiplication by l is compatible with the Maslov grading, in the sense that `(l
) = `() 2.
Multiplication by l drops the Alexander ltration by at least 1, in the sense that
(l ) () 1.
Given such a C, there are two naturally associated chain complexes. One complex,
which we call the vertical complex C
v
:= C,(l C), is a ltered complex which inherits
the Alexander ltration. Its homology is called the vertical homology, denoted H
v
(C). Next
dene C

:= C
F
2
[U]
F
2
[l. l
1
]. This has two ltrations, namely the Alexander ltration
and the ltration by powers of l
1
. (Precisely, the :
th
ltered part of the second ltration
11.5. FROM CFK

TO

CFD: STATEMENT OF RESULTS 203
contains generators l
t
x, where t : and x is a generator of C.) Then the horizontal
complex C
h
is dened to be the degree 0 part of the associated graded complex to C

with
respect to the Alexander ltration. (Note that C
v
is the degree 0 part of the associated
graded space to C

with respect to the ltration by l powers.) The complex C

h
still has
the ltration by powers of l
1
and inherits a dierential, whose homology is called the
horizontal homology, denoted H
h
(C).
We illustrate C

graphically by picking a ltered basis

i
over F
2
[l] of C and plotting
a generator over F
2
of C F
2
[l. l
1
] of the form l
x

i
with Alexander depth on the
plane at the position (r. ). Then the dierential of a generator at (r. ) can be graphically
represented by arrows connecting the point at (r. ) with the coordinates of other generators.
These arrows necessarily point (non-strictly) to the left and down. The complex C
v
is
obtained by restricting to one vertical column, and C
h
is obtained by restricting to one
horizontal row.
Denition 11.25. We call C reduced if strictly drops either the Alexander or l power
ltration; i.e., if for all x, we have x = y
1
+ l y
2
where (y
1
) < (x).
We call a ltered basis
i
over F
2
[l] for C vertically simplied if for each basis vector
i
,
either
i
l C or
i

i+1
(mod l C). In the latter case we say that there is a vertical
arrow from
i
to
i+1
; so, in a vertically simplied basis, there is at most one vertical arrow
starting or ending at each basis element. (The fact that
2
= 0 says that there cannot be
a vertical arrow both starting and ending at a single element.) The length of the vertical
arrow is the dierence (
i
) (
i+1
).
Similarly, we call a ltered basis
i
over F
2
[l] for C horizontally simplied if for each
basis vector
i
, either (
i
) < (
i
) or there is an : so that
i
= l
m

i+1
+ c where
(
i
) = (l
m

i+1
) and (c) < (
i
). In the latter case we say that there is a horizontal
arrow from
i
to
i+1
; so, in a horizontally simplied basis, there is at most one horizontal
arrow starting or ending at each basis element. The length of the horizontal arrow is the
integer :.
Remark 11.26. These notions could alternatively be formulated without reference to a
basis. Vertical (respectively horizontal) arrows correspond to non-vanishing dierentials in
the spectral sequence belonging to the ltered complex C
v
(respectively C
h
).
Every nitely generated chain complex C as above is homotopy equivalent to one which is
reduced; moreover, for a reduced complex, we can nd a vertically or horizontally simplied
basis. See Proposition 11.52.
In particular, these constructions apply to knot Floer homology CFK

(1). The com-

plexes C
v
(CFK

(1)) and C
h
(CFK

(1)) (ignoring the ltration) are both canonically iso-

morphic to

CF(o
3
) and so if 1 is a knot in o
3
, then H
v
(CFK

(1))

= H
h
(CFK

(1))

=

HF(o
3
)

= F
2
. As such, after passing to a reduced model for CFK

(1), a vertically (re-

spectively horizontally) simplied basis
i
(respectively
i
) has a preferred element which
(after reordering) we label
0
(respectively
0
), characterized by the property that it has no
in-coming or out-going vertical (respectively horizontal) arrows.
Theorem 11.27. Let CFK

(1) be a model for a chain complex for a knot 1 o

3
which
is reduced. Let 1 be the bordered three-manifold o
3
nbd 1. Given any suciently large
integer :, we consider 1 with framing :. The associated type 1 module

CFD(1 ) can be
extracted from CFK

(1) by the following procedure.

204 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
In the idempotent
0
,

CFD(1 ) is isomorphic as an F
2
-module to CFK

(1).
1
Let
i

be a vertically simplied basis for CFK

(1). To each vertical arrow of length / from

i
to

i+1
, we introduce a string of basis elements
i
1
. . . . .
i

for
1

CFD(1 ) and dierentials

i
D
1

i
1
D
23

D
23

i
k
D
23

i
k+1
D
23

D
23

i

D
123

i+1
where
i
and
i+1
are viewed as elements of

CFD(1 ). (The reader is cautioned to note the
directions of the above arrows.)
Similarly, endow CFK

(1) with a horizontally simplied basis

i
. To each horizontal
arrow from
i
to
i+1
of length /, we introduce a string of basis elements
i
1
. . . . .
i

connected
by a string of dierentials

i
D
3

i
1
D
23

D
23

i
k
D
23

i
k+1
D
23

D
23

i

D
2

i+1
.
(As before,
i
and
i+1
are viewed as elements of

CFD(1 ), via the identication of
0

CFD(1 )
with CFK

(1).)
There is a nal string, the unstable chain, of generators j
1
. . . . . j
m
connecting
0
and
0
,
connected by a string of dierentials,

0
D
1
j
1
D
23
j
2
D
23

D
23
j
m
D
3

0
.
The length : of this string is : + 2(1).
As stated, Theorem 11.27 only computes

CFD for suciently negative framing param-
eters. In [LOT10a], we give bimodules which allow us to deduce the type 1 module for
arbitrary framings; this is summarized in Appendix A. It turns out that most of the state-
ment of Theorem 11.27 goes through for arbitrary integral framings; the only part which
depends on the framing is the unstable chain, connecting
0
and
0
, which can be explicitly
described. In particular, for framing 2(1), this entire chain can be replaced by a single
map
(11.28)
0
D
12

0
.
See Theorem A.11 for the complete statement.
The procedure for translating knot Floer complexes to type 1 modules given by Theo-
rem 11.27 is illustrated in Figure 11.5, with one proviso: to keep the pictures small, we have
illustrated type 1 modules with framing parameter : = 2(1), for which one needs to refer
to Theorem A.11.
It is interesting to note that the above construction of

CFD uses only the vertical and
horizontal arrows, while CFK

in general can contain diagonal arrows. In concrete terms, the

vertical arrows correspond to counting holomorphic disks with :
w
() = 0, while horizontal
ones correspond to counts of holomorphic disks with :
z
() = 0. Diagonal arrows come from
disks crossing both . and u.
We will deduce Theorem 11.27 from a more complicated statement, Theorem 11.35,
which does not make reference to a choice of basis, and which comes more directly from
a Heegaard diagram. This basis-free version is stated and proved in Section 11.7, and we
deduce Theorem 11.27 from it in Section 11.8. Before turning giving these proofs, we discuss
some further counts of holomorphic curves, in Section 11.6.
1
That is, the submodule
0

CFD(Y ) is identied with CFK

(K) as an F
2
-module.
11.5. FROM CFK

TO

CFD: STATEMENT OF RESULTS 205
D
1
D
2
D
23
D
3
D
1
D
2
D
3
D
12
D
23
D
123
D
123
D
12
D
2
D
3
D
123
D
1
D
2
D
3
D
1
D
123
D
2
D
23
D
123
D
3
D
23
D
1
D
123
D
1
D
123
D
1
D
2
D
3
D
2
D
3
D
12
Figure 11.5. Reconstructing the type 1 module. A table with data
from knot Floer homology on the left, where vertical dierentials correspond to

z
and horizontal ones corresponding to
w
; the length of the arrow represents
the change in Alexander grading. Top: the left-handed (3. 4) torus knot.
Middle: the (2. 1) cable of the left-handed trefoil. Bottom: the gure eight
knot. The corresponding type 1 modules are on the right, with black dots for
generators in the idempotent
0
and white ones for generators in the idempotent

1
. The unstable chain is denoted by the dotted arrow. Here we take framing
parameter 2(1) as in Equation (11.28). Explicitly, these are framings 6,
4 and 0, respectively.
206 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
11.6. Generalized coecient maps and boundary degenerations
So far, the holomorphic curves we count have not covered the basepoint, and in particular
have not involved the Reeb chords that contain the basepoint. In this section, we explain a
certain further structure on the type 1 module coming from counting certain holomorphic
curves which contain
0
in their boundary as well. These additional maps will be used
in Section 11.7 to construct

CFD from CFK

. This further structure exists for arbitrary

(connected) boundary, but for notational simplicity we will restrict our attention to the torus
boundary case.
Fix a bordered Heegaard diagram (1. .) for (1
3
. 1
2
), and label the regions around as
in Section 11.1. Recall that for 1 = i
0
. . . . . i
n
an interval in 1. 2. 3 there is an associated
map 1
I
: \
[i
0
1]
\
[in]
. This is induced by the dierential on

CFD, which in turn is dened
by counting holomorphic curves in [0. 1] 1, cf. Chapter 6. By moving the basepoint .
from the region labeled 0 to one of the regions labeled 1, 2 or 3, we obtain maps 1
I
: \ \
for intervals 1 in 0. 1. 2. 3 with respect to the cyclic ordering, of length at most 3. (For
instance, putting . in region 1 the new maps we obtain are 1
0
, 1
30
and 1
230
.) We will call
these new maps generalized coecient maps.
We dene four more generalized coecient maps, 1
0123
, 1
1230
, 1
2301
and 1
3012
as follows.
By dropping Condition (M-9) on holomorphic maps from Section 5.2, we can consider moduli
spaces of holomorphic curves asymptotic to Reeb chords
0
,
01
and so on. Collecting these
into moduli spaces as in Section 6.1, for x a generator of

CFD(1. .) in idempotent
0
, we
set
1
0123
(x) :=

yS(7)
B
2
(x,y)
[ ind(B, )=1
#
_
/
B
(x. y; )
_
y
where the sum over is over (
0
.
1
.
2
.
3
). (
012
.
3
). (
0
.
123
). (These are the three se-
quences of Reeb chords for which

c(

) = c(
0123
) in a suitable generalization of /.)
The maps 1
1230
, 1
2301
and 1
3012
are dened similarly.
By moving the basepoint ., it follows from the discussion in Section 11.1 that these maps
satisfy the compatibility conditions

I=JK[ J<K
1
K
1
J
= 0
where < now denotes the cyclic ordering on 0. 1. 2. 3 and 1 is any proper cyclic interval in
0. 1. 2. 3. The compatibility equations involving all four Reeb chords are, perhaps, more
surprising, as there is now a right hand side.
Proposition 11.29. The maps 1
0123
and 1
2301
satisfy
1 1
0123
+ 1
3
1
012
+ 1
23
1
01
+ 1
123
1
0
+ 1
0123
1 = ": \
1
\
1
(11.30)
1 1
2301
+ 1
1
1
230
+ 1
01
1
23
+ 1
301
1
2
+ 1
2301
1 = ": \
1
\
1
. (11.31)
The maps 1
1230
and 1
3012
satisfy similar formulas.
(Here, " denotes the identity map.)
Remark 11.32. The grading shifts of the new coecient maps may be computed directly.
Alternatively, they are uniquely specied by the grading shifts for the original grading shift
maps, together with the above formulas, which are homogeneous.
11.6. GENERALIZED COEFFICIENT MAPS AND BOUNDARY DEGENERATIONS 207
0
.
1 2
3
0
.
1 2
3
Figure 11.6. Boundary degeneration as end of a one-dimensional
moduli space. Left: a genus 1 Heegaard diagram (. . ) for the solid
torus. Here,
1
is disjoint from
a
1
and intersects
a
2
in a single point. Right:
one point in a one-parameter family of holomorphic curves in [0. 1] 1,
with one end a two-story building and the other end a boundary degeneration.
The reason that the right hand sides of Equations (11.30) and (11.31) are the identity
map rather than zero is the existence of certain boundary degenerations. We can see these
boundary degenerations explicitly in simple cases. For example, let (. . . .) be a genus
1 Heegaard diagram for a solid torusfor instance, the diagram in Figure 11.6. Then
is a rectangle (disk with four boundary punctures). There is a holomorphic map
n: o [0. 1] 1, where o is the union of a trivial strip and a rectangle, and the
rectangle is mapped in the obvious way to , and by a constant map to [0. 1] 1.
Such curves look quite strange in the cylindrical setting. (In particular, it is not clear
when they are transversely cut out.) We therefore invoke, for the rst and only time in
this book, the tautological correspondence between holomorphic curves in [0. 1] 1 and
holomorphic disks in Sym
g
(). That is, recall that there is a bijective correspondence
Holomorphic maps
: [0. 1] 1 Sym
g
()
with (0 1)

i
and (1 1)

Holomorphic curves
n in [0. 1] 1
with boundary in 0 1
and in 1 1
so that
|
n is a p-fold branched cover.
(In the simplest case, one uses a complex structure Sym
g
(,

) on Sym
g
() and a split complex
structure ,

,
|
on [0. 1] 1. More generally, admissible almost complex structures J
on [0. 1] 1 correspond to working with paths of complex structures on Sym
g
().) See,
for instance, [Lip06a, Section 13] for a detailed explanation.
To extend this to the bordered setting, recall that
e
denotes the result of lling-in the
puncture j of . Viewing Sym
g
() as the complement of the divisor j Sym
g1
(
e
) in
Sym
g
(
e
), we see that Sym
g
() is a symplectic manifold with a cylindrical end, in a way
compatible with the complex structure Sym
g
(,

). The end of Sym

g
() is modeled on the
unit normal bundle l(Sym
g1
(
e
)) to j Sym
g1
(
e
), an o
1
-bundle over Sym
g1
(
e
);
compare [BEH
+
03, Example 2.2 and Remark 5.9]. There is a Reeb-like vector eld

1 on
l(Sym
g1
(
e
)) tangent to the o
1
-bers.
The Lagrangian subspace
a
i

c
1

c
g
(i = 1. 2) intersects the ideal boundary of
Sym
g
() in the subspace (
a
i
)
c
1

c
g
. So, for instance, the Reeb chord
1
corresponds
to a (p1)-dimensional family of

1-chords between
a
1
and
a
2
, parameterized by the points
in
c
1

c
g
; and similarly for the other Reeb chords.
The tautological correspondence extends to a correspondence between holomorphic maps
n: o [0. 1] 1 asymptotic to a sequence of Reeb chords (
1
. . . . .
n
) at c and
208 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
generators x and y at and holomorphic maps
: (([0. 1] 1) (1. t
1
). . . . . (1. t
n
)) Sym
g
()
asymptotic to a chord of the form
i
(c
i,1
. . . . . c
i,g1
) at the puncture (1. t
i
) and to x and
y at . Here, the values of the t
i
and the points c
i,j

j
are allowed to vary, with the
restriction that t
1
< t
2
< < t
n
. In particular, the 1-coordinates of the c punctures of n
correspond to the values of the t
i
.
With respect to this correspondence, the boundary degeneration mentioned above comes
from a disk in Sym
g
() with boundary entirely in the -tori; we will view this as a map from
the upper half-plane H with some boundary punctures. There is nothing strange about this
disk. In particular, for generic J on the symmetric product Sym
g
(), the moduli spaces of
these disks will be transversely cut out.
We collect the boundary degenerations in Sym
g
() into moduli spaces:
Denition 11.33. Given i = 1 or 2, x
a
i

c
1

c
g1
, a sequence = (
1
. . . . .
n
) of
Reeb chords, and an almost complex structure J on Sym
g
(), let /
[]
(x; ; J) denote the
moduli space of J-holomorphic maps
: H (t
1
. . . . . t
n
) Sym
g
().
where the t
i
are real numbers (points in H) which are allowed to vary with the given order,
such that the boundary of H is mapped into
a
1

a
2

c
2

c
g1
, the homology class
of is [], is asymptotic to x at and to a chord of the form
i
(c
i,1
. . . . . c
i,g1
) at
t
i
.
Proposition 11.34. There is a non-empty open subset of nearly-symmetric almost com-
plex structures J on Sym
g
(), as in [OSz04d, Section 3], such that the moduli spaces
/
[]
(x; ; J), where is a cyclic permutation of (
0
.
1
.
2
.
3
) and x S(1), are trans-
versely cut out and have an odd number of points. Further, for these almost complex struc-
tures, all /
[]
(x; ; J) for other are empty.
Proof. As already discussed, it is standard that /
[]
(x; ; J) is transversely cut out
for generic J. To prove the moduli space has an odd number of points, we start by studying
a model case. Consider the Heegaard diagram (
1
.
1
.
1
) shown in Figure 11.6. In this
diagram, there is a single generator x. With respect to any almost complex structure ,
on
1
, there is a one-parameter family of holomorphic disks in
1
= Sym
g
1
(
1
), in the
homology class [
1
], with asymptotics (
0
.
1
.
2
.
3
). One end of this moduli space is a
broken holomorphic curve and the other end is the boundary degeneration under discussion.
Since the domain of these curves has multiplicity 1 somewhere (in fact, everywhere), for a
generic , on , this moduli space of Sym
g
(,)-holomorphic curves is transversely cut out,
including at its boundary. So, the count of boundary degenerations in Sym
g
(
1
) =
1
is 1.
Now, for the general case, let (. . ) be any bordered Heegaard diagram with torus
boundary. Then (. ) can be decomposed as a connect sum (
1
.
1
) # (
2
.
2
) where
(
1
.
1
) is the -curves of the diagram discussed above and
2
consists of closed curves.
A simple adaptation of the arguments showing stabilization invariance of HF of closed 3-
manifolds [OSz04d, Section 10] shows that for almost complex structures with a long neck
between (
1
.
1
) and (
2
.
2
), there are an odd number of holomorphic curves in /
[]
(x;
(
0
.
1
.
2
.
3
); J). Similar arguments apply for other cyclic orderings of (
0
.
1
.
2
.
3
). This
proves the rst claim in the proposition.
11.7. FROM CFK

TO

CFD: BASIS-FREE VERSION 209
Curves with other asymptotics (e.g., (
0
.
123
)) cannot occur in (
1
.
1
.
1
) by inspection.
Hence, if there is a long neck between (
1
.
1
) and (
2
.
2
), they do not occur for (. ).
This proves the second claim in the proposition.
Proof of Proposition 11.29. We will prove Relation (11.30); the other relations fol-
low similarly. Consider the moduli space
_
ind(B,(
0
,
1
,
2
,
3
))=2
/
B
(x. y; (
0
.
1
.
2
.
3
)).
For x ,= y, the ends of this moduli space all correspond to terms on the left side of For-
mula (11.30). By contrast, for x = y it follows from Proposition 11.34, together with a stan-
dard gluing argument (compare [OSz08, Theorem 5.1]) and the tautological correspondence,
that an odd number of ends of the moduli space correspond to boundary degenerations. The
result is immediate.
11.7. From CFK

to

CFD: Basis-free version
In this section we state and prove a basis-free version of Theorem 11.21, Theorem 11.35
below. We deduce Theorem 11.21 from Theorem 11.35 in the next section. Before stating
the basis-free version, we introduce a little more notation for knot Floer homology.
Fix a doubly-pointed Heegaard diagram 1 for 1, with basepoints u and .. We write
C for the chain complex

gCFK(1. u. .) computing

HFK(1) and C(:) for its summand in
Alexander grading :. There are maps
i
: C(:) C(: +i) dened by counting holomorphic
disks with :
z
() = i and :
w
() = 0 if i 0, and :
z
() = 0 and :
w
() = i if i 0.
Set
w
:=

i0

i
and
z
:=

i0

i
. The maps
w
and
z
each give dierentials on C.
Furthermore, C has two ltrations,
C(:) :=

rs
C(:) and C(:) :=

rs
C(:).
preserved by
w
and
z
, respectively. We will principally use not the complex (C(:).
w
),
but the complex (C(:).
w
), dened to be the quotient of (C.
w
) by the subcomplex (C(
: + 1).
w
). We likewise need (C(:).
z
), with a similar denition.
The above constructions can be interpreted using the notions from Section 11.5. The
complex CFK

(1) is a nitely generated, free, Z-ltered, Z-graded chain complex over

F
2
[l] satisfying compatibility conditions spelled out in that section. The complex (C.
w
)
is the horizontal complex associated to CFK

(1), while (C.

z
) is the vertical complex.
In particular, (C(:).
w
) is identied with the :
th
summand of gCFK

(1. :) with respect

to the Alexander grading, while (C(:).
z
) is identied with the :
th
ltered complex of

CFK(1). Thus (C(:).

w
) and (C(:).
z
) for : suciently small both have homology
isomorphic to F
2
; likewise for (C(:).
z
) and (C(:).
w
) for : suciently large.
Theorem 11.35. Let 1 o
3
be a knot with meridian j and 0-framed longitude , and let
C,
w
, and
z
be the data associated as above. Given a positive integer :, let
\
0
:=

sZ
\
0
s
and \
1
:=

sZ+
n+1
2
\
1
s
210 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
where
\
0
s
:= C(:)
\
1
s
:=
_

_
C
_
: +
n1
2
_
:
n
4
F
2
[:[ <
n
4
C
_
:
n1
2
_
:
n
4
.
Let \ := \
0
\
1
. Then, for : large enough, the following maps 1
I
satisfy the compatibility
condition (11.4):
The dierential 1 restricted to \
0
s
= C(:) is the dierential
0
on the knot complex;
for :
n
4
, 1 restricted to \
1
s
= C
_
: +
n1
2
_
is the dierential
w
; for :
n
4
,
1 restricted to \
1
s
= C
_
:
n1
2
_
is the dierential
z
; for [:[ <
n
4
, 1 restricted
to \
1
s
= F
2
is identically zero.
The map
1
1
: \
0
s
= C(:) \
1
s+
n1
2
= C(:)
is the obvious inclusion of the subcomplex.
The map 1
2
is nonzero only on \
1
s
for :
n
4
, in which case
1
2
: \
1
s
= C
_
: +
n1
2
_
\
0
s+
n+1
2
= C
_
: +
n+1
2
_
is given by

w
: C
_
: +
n1
2
_
C
_
: +
n+1
2
_
where : C C
_
: +
n+1
2
_
is the projection.
The map
1
3
: \
0
s
= C(:) \
1
s
n1
2
= C(:)
is the obvious inclusion of the subcomplex.
The map 1
12
is identically zero.
For : <
n
4
, the map
1
23
: \
1
s
= C
_
: +
n1
2
_
\
1
s+1
= C
_
: +
n+1
2
_
is the obvious inclusion map; for the : with :
n
4
< : + 1,
1
23
: \
1
s
= (C.
w
) \
1
s+1
= F
2
is a chain map inducing an isomorphism in homology; for
n+2
4
< : <
n2
4
, 1
23
:
\
1
s
= F
2
\
1
s+1
= F
2
is the isomorphism; for the : with : <
n
4
: + 1,
1
23
: \
1
s
= F
2
\
1
s+1
= (C.
z
)
is a chain map inducing an isomorphism on homology; for :
n
4
,
1
23
: \
1
s
= C
_
:
n1
2
_
\
1
s+1
= C
_
:
n+1
2
_
is the obvious projection map.
The map
1
123
: \
0
s
\
1
s+
n+1
2
is the composite of one component of
w
,

1
w
: C(:) C(: + 1).
with the inclusion of C(: + 1) in (C( : + 1).
z
) = \
1
s+
n+1
2
.
11.7. FROM CFK

TO

CFD: BASIS-FREE VERSION 211
D23
D23
D23
D23
D23
D23
D23
D23
D23
D23 D23
D23
D23
D23
D23
D23
D23
D23
D23 D23 D23
V
1
s+
3
2
F F F C
s1
C
s
C
s+1
C
s2
C
s1
C
s
C
s
C
s+1
C
s+2
C
s+3
C
s+1
C
s+2
C
s+3
. . . . . .
C
s
V
1
s
n3
2
V
1
s
n1
2
V
1
s+
n1
2
V
0
s
V
0
s+1
D3
D3
C
s2
D1
C
s+1
V
1
s+
n+1
2
V
1
s+
1
2
V
1
s
1
2
D123
D2
D23
D1
Figure 11.7. A sketch of some of the maps from Theorem 11.35. The
horizontal coordinate represents the grading by relative spin
c
structures. The
shaded squares correspond to \
0
, while the unshaded ones represent \
1
.
Moreover, the associated type 1 module is homotopy equivalent to

CFD(o
3
nbd(1)), marked
by the curves j and : j+. The : grading on \
0
\
1
is identied with the spin
c
grading
on

CFD, induced from a projection of G to
1
2
Z which descends to o
D
(1).
Figure 11.7 sketches some of the maps from Theorem 11.35.
Remark 11.36. The complexes C( :) appear in Heddens cabling and Whitehead doubling
formulas [Hed05b, Hed07].
In the theorem, we have associated a type 1 module to the complex C, equipped with
boundary operators
w
and
z
and ltrations, which in turn can be thought of as associated
to CFK

(1), a Z-ltered, Z-graded complex over F

2
[l].
Proposition 11.37. Up to homotopy equivalence, for suciently large : the type 1 module
associated in Theorem 11.35 depends only on the Z-ltered, Z-graded chain homotopy type
of the F
2
[l]-chain complex CFK

(1).
Proof. A Z-ltered, Z-graded homotopy equivalence of F
2
[l]-chain complexes induces
homotopy equivalences on the associated complexes C( :) and C( :), which in turn in-
duces homotopy equivalences between the associated \
0
and \
1
s. It is immediate from the
denitions that these homotopy equivalences commute with the coecient maps in Theo-
rem 11.35.
We prove Theorem 11.35 by considering a bordered Heegaard diagram 1(:) for the knot
complement with suciently large framing :, in a way analogous to the surgery formula
for Heegaard Floer homology [OSz04b]. For an analogous choice of bordered Heegaard
diagram 1, the corresponding type 1 module,

CFD(1) = \
0
\
1
, has generators which
correspond to the generators of \ = \
0
\
1
as stated in the theorem, except for \
1
s
with
[:[ <
n
4
. Moreover, again except for \
1
s
with [:[ <
n
4
, the coecient maps are as stated in
the theorem. We then construct a homotopy equivalence which xes the cases with [:[ <
n
4
.
We give the proof after some preliminary lemmas.
212 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
r
2
r
1
r
0
r
1
r
2
u

a
1

a
2
r
3
2
3 0
1
r

g
.
Figure 11.8. Twisting. At the top, we have pictured a portion of a doubly-
pointed Heegaard diagram for a knot. Below it, we have a bordered diagram
for the knot complement, with suitably twisted framing. The gray shading
represents a domain from a r to br (for some a and b) which, in the top
diagram, crosses u with multiplicity one and . with multiplicity zero. There
is a corresponding (provincial) domain below connecting a r
i
to b r
i+1
,
provided that i 2.
To describe our bordered Heegaard diagram 1(:) we start with a doubly-pointed Hee-
gaard diagram (
0
.
0
.
0
. .. u) for the knot 1. We stabilize this diagram to obtain a new
diagram 1
K
:= (. . . .. u). Here, is gotten by attaching a two-dimensional one-handle
to
0
, with feet near the markings . and u; is gotten from
0
by introducing a new
circle
g
, contained in the new one-handle; and is gotten from
0
by introducing a new
circle
g
, which meets
g
in a single transverse intersection point, and is disjoint from the
other -circles. In particular, the circle
g
is a meridian for the knot.
Identify a neighborhood J of
g
in with [1. 1] o
1
so that
g
is identied with
0 o
1
and
g
J is identied with [1. 1] c
i/2
. Let denote a 0-framed longitude
of 1 in , disjoint from the
i
for i ,= p, and with J = [1. 1] 1. Let

a
1
= ( ( J)) (t. c
in(t+1)
) J [ t [1. 1].
That is,
a
1
is obtained from the 0-framed longitude by winding : times around the meridian

g
inside J. We refer to J as the winding region. Note that
a
1
intersects
g
in : points
in J, and intersects
g
in a single point j. Dene
a
1
:=
a
1
j and
a
2
:=
g
j. Then,
with notation as in Section 4.2,
1(:) := (.
a
1
.
a
2
.
1
. . . . .
g1
. )
is a bordered Heegaard diagram for o
3
nbd(1), with framing :. We will also write 1 for
1(:) when the framing is clear. We label the four quadrants around j by 0. . . . . 3, arranged
in a counterclockwise order (i.e., the order induced by nbd(j)), so that, if you forget
a
1
,
0 and 1 are in the region which contains . in 1
K
. As usual, we place the basepoint in
11.7. FROM CFK

TO

CFD: BASIS-FREE VERSION 213
region 0. For convenience of notation, we will henceforth assume that : is divisible by 4.
See Figure 11.8 for an illustration.
We now turn to the grading on

CFD(1). Following Section 10.4, the grading takes
values in the coset space G,1(x
0
), where 1(x
0
) is the span of p(1) over all periodic domains
1
2
(x
0
. x
0
). For a homology solid torus like a knot complement, the group of periodic
domains is isomorphic to Z. Let 1
0
be a generator of
2
(x
0
. x
0
). Since the knot has framing :,
we can choose the sign of 1
0
so that 1
0
has local multiplicities (0. 1. 1:. :) in the regions
(0. 1. 2. 3), respectively. Thus 1(x
0
) G is (; 1. :) for some Z (depending on
x
0
). For present purposes it suces to consider only the spin
c
component of the grading on

CFD(1), which we can recover with the homomorphism from G to

1
2
Z given by
(11.38) (,; j. ) :j .
(Note that p(1(x
0
)) is in the kernel of this map.)
More generally, let o: S(1)
1
2
Z be a function characterized up to overall translation
by the formula
(11.39) o(x) o(y) =
_
n+1
2
_
:
0
(1)
_
n1
2
_
:
1
(1)
_
n+1
2
_
:
2
(1) +
_
n1
2
_
:
3
(1).
where 1 is any homology class in
2
(x. y) (i.e., classes possibly covering the basepoint .).
Lemma 11.40. Let 1 be a homology solid torus, i.e., a three-manifold with torus boundary
and H
1
(1 ; Z)

= Z. Pick j. H
1
(1 ) so that generates the kernel of the map on H
1
induced by inclusion of the boundary while the image of j generates H
1
(1 ). Let

CFD(1 )

=
\
0
\
1
be the decomposition by idempotents of the type 1 invariant of 1 , with boundary
marked by j and : j + . Then a function o satisfying (11.39) gives a decomposition
\
i

=

s
1
2
Z
\
i
s
.
well-dened up to overall translation. The coecient maps respect this grading in the follow-
ing sense:
1
1
: \
0
s
\
1
s+
n1
2
1
2
: \
1
s
\
0
s+
n+1
2
1
3
: \
0
s
\
1
s
n1
2
1
12
: \
0
s
\
0
s+n
1
23
: \
1
s
\
1
s+1
1
123
: \
0
s
\
1
s+
n+1
2
.
Proof. We rst check that there is a function o satisfying Equation (11.39). To see this,
observe that there is a natural splitting
2
(x. y) =
2
(x. y) Z[]. Now, the right-hand-side
of Equation (11.39) remains unchanged when we add a copy of [] to the domain 1; so it
suces to show that the right-hand-side of Equation (11.39) depends only on the generators
x and y, when 1 is chosen to lie in
2
(x. y) (i.e., when 1 satises :
0
(1) = 0). For such
1, o(1) is obtained by taking p(1) and applying rst the map in Formula (11.2) followed
by the map in Formula (11.38). Since Formula (11.38) was chosen to annihilate periodic
domains, it follows that the result is independent of tho choice of 1.
The fact that the coecient maps change grading as stated is immediate from the de-
nition of o, Formula (11.39).
Let S
K
denote the set of generators for the knot Floer complex

CFK(1
K
), and S(:)
denote the generators for

CFD(1(:)). Order the intersection points of (
1

2
) with
g
in
the winding region in the order they are encountered along
g
, r
i

i=
n
2
i=
n
2
, so that r
0
is the
214 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
intersection point of
2
with
g
. (Here and below, we assume for notational convenience that
: is even.) For each generator x S
K
there are : + 1 corresponding generators in S(:),
gotten by adding a point in (
1

2
)
g
in the winding region. That is, we let x
i

i=
n
2
i=
n
2
denote the sequence of generators with the property that the (
1

2
)
g
-component of x
i
is r
i
and, outside the winding region, x
i
agrees with x.
Lemma 11.41. For a given :, we can choose a function o on S(:) as above so that it is
related to the Alexander grading : S
K
Z of elements in S
K
by the formula
(11.42) o(x
k
) = (x) / +
_
(: + 1) sgn(/)
2
_
.
where sgn(/) = 1, 0, or 1 if / < 0, / = 0, or / 0 respectively.
Proof. The region 3 is a domain connecting x
0
to x
1
, and for / < 0 there is a domain
connecting x
k
to x
k1
with multiplicities (0. 0. 1. 1) in regions (0. 1. 2. 3). (See Figure 11.8
and the middle of Figure 11.10.) Similarly, region 1 connects x
0
to x
1
and, for / 0, the
generators x
k
and x
k+1
are connected by a domain with multiplicities (1. 1. 0. 0). It follows
from Equation (11.39) that
o(x
k
) = o(x
0
) / +
_
(: + 1) sgn(/)
2
_
.
According to Equation (11.10), the Alexander grading on knot Floer homology is character-
ized (up to translation) by the property that if x. y S
K
and
2
(x. y) is any domain in
1
K
, then (x) (y) +:
z
() :
w
() = 0. We can convert into a domain
t
in 1(:) that
connects x
0
to y
0
with :
0
() = :
1
() = :
z
() and :
2
() = :
3
() = :
w
() (as is evident
from Figure 11.8). So,
o(x
0
) = (x) + c.
for some constant independent of the intersection point x S
K
. We can modify o by an
overall translation to take c = 0.
We now have a collection of generators in S(:) which have corresponding elements of
S
K
. The remaining generators (coming from intersections of
a
2
with the -curves outside
of the winding region) are called exterior generators. Note that the non-exterior generators
are the generators of the form x
i
for some x S
K
, i :,2. :,2 + 1. . . . . :,2.
Lemma 11.43. For any doubly-pointed Heegaard diagram 1
K
of a knot in o
3
, there is a
constant c (independent of the framing parameter :) with the following property. For the
function o: S(:) as dened in Lemma 11.41, the generators y S(:) with [o(y)[ c
satisfy the following properties:
(c-1) y is not an exterior generator, so y = x
i
for some x S
K
, i Z; and
(c-2) the sign of o(y) agrees with the sign of i.
Proof. First, we nd c
1
so that generators y with [o(y)[ c
1
satisfy Property (c-1).
Note that the set of exterior generators is independent of the framing parameter :. Start
with some initial framing parameter :, x some x S
K
and consider x
1
S(:). For each
exterior generator y, we can nd a domain 1
0

2
(y. x
1
) with :
0
(1
0
) = :
1
(1
0
) = 0 and
:
2
(1
0
) = :
3
(1
0
) = c(y). (Start with an arbitrary domain in
2
(y. x
1
) and add copies of []
and the periodic domain to make :
0
= :
1
= 0; since y and x
1
are in the same idempotent,
11.7. FROM CFK

TO

CFD: BASIS-FREE VERSION 215
we will then have :
2
= :
3
.) Now, in 1(:), the domain 1
0
can be used to construct a new
domain 1
n

2
(y. xmn
2
1
) with :
0
(1
n
) = :
1
(1
n
) = 0 and :
2
(1
n
) = :
3
(1
n
) = c(y). Since
there are only nitely many possible exterior generators, we see that [o(y) o(xmn
2
)[ [
y an exterior generator is bounded independently of :. It follows from Lemma 11.41 that
[o(y)[ for exterior generators is bounded by some constant c
1
independent of :, giving
Property c-1.
Next, we claim that there is a constant c
2
with the property that if o(x
i
) c
2
, then i 0.
Specically, choose c
2
max
xS
K
(x). Then Equation (11.42) shows that o(x
0
) c
2
.
Similarly, if x
i
is a non-exterior generator with i 0 then, since i
n
2
, Equation (11.42)
guarantees that
o(x
i
) = (x) + i
: + 1
2
(x)
1
2
c
2

1
2
< c
2
.
This guarantees Property (c-2) for o(y) 0. A symmetric argument ensures the property
also in the case where o(y) < 0.
Lemmas 11.4411.47 below state that for an appropriate class of almost complex struc-
tures, the coecient maps have the desired forms from Theorem 11.35. The boundary of
the winding region J consists of two circles C
1
and C
2
parallel to the meridional -circle;
we label these so that C
1
is on the 2. 3 side of J and C
2
is on the 0. 1 side of winding; see
Figure 11.9. Fix an admissible complex structure J
0
on [0. 1] 1. We now consider
a two-parameter family of complex structures J
T
1
,T
2
on [0. 1] 1 gotten by inserting
a necks of lengths 1
1
and 1
2
along C
1
and C
2
, respectively. We say that a result holds for
suciently pinched almost complex structures if there is a 1
1
so that for any 1
1
1
1
there
is a 1
2
so that for any 1
2
1
2
the result holds for all generic almost complex structures of
the form J
T
1
,T
2
.
Lemma 11.44. Let 1 o
3
be a knot with Heegaard diagram 1
K
, and let c be the
constant from Lemma 11.43. For any suciently large surgery coecient : the following
holds on 1(:). Given any suciently pinched almost complex structure, for each generator
x S
K
, with corresponding generators x
k
in S(:), we have:
1
23
(x
k
) = x
k1
if / < 0 and o(x
k
) c 1;
1
01
(x
k
) = x
k+1
if / 0 and o(x
k
) c + 1;
1
3
(x
0
) = x
1
; and
1
1
(x
0
) = x
1
.
That is, for each x S
K
, we have a string of generators connected by coecient maps as
follows:

D
23
x
3
D
23
x
2
D
23
x
1
D
3
x
0
D
1
x
1
D
01
x
2
D
01
x
3
D
01
. . .
Proof. The domains realizing the maps displayed in the statement can be found inside
the winding region. For example, the domain realizing 1
3
carrying x
0
to x
1
is the region
marked by 3 in Figure 11.8. Domains from x
k
to x
k1
are annuli
k
which have a boundary
component in
a
1
as in the middle of Figure 11.10.
216 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
The case of 1
23
. We analyze the map 1
23
, verifying it has the form stated, in the
following eight steps.
Step 1. The term x
k1
occurs in 1
23
(x
k
). By cutting
K
along
a
1
at r
k
, we obtain
a bigon realizing a non-zero coecient of x
k1
in 1
23
(x
k
). The rest of the proof (for the
operator 1
23
) consists in showing that this is the only term in 1
23
(x
k
).
Step 2. If y
k1
occurs as a term in 1
23
(x
k
) then y = x. Suppose that y
k1
occurs in 1
23
(x
k
). Considering the corners shows that any domain
2
(x
k
. y
k1
) is a
disjoint union of
k
with a dierent positive domain
t
. As such, the corresponding moduli
space naturally has an (1 1)-action (translating the two components independently). If
this moduli space is generic and one-dimensional, it follows that the other component
t
is
trivial, so y
k1
= x
k1
.
Step 3. No exterior generators occur in 1
23
(x
k
). If an exterior generator y appears
in 1
23
(x
k
), then o(y) = o(x
k
) + 1 c, in view of Lemma 11.40 and our assumption
on o(x
k
). By Lemma 11.43 (and specically Property (c-1)), it follows that y is not an
exterior generator.
Step 4. No terms of the form y
j
where , 0 occur in 1
23
(x
k
). This is same argument
as in Step 3, using Property (c-2)) this time.
In view of Steps 3 and 4, only terms of the form y
k+i
with / + i < 0 can appear in
1
23
(x
k
). Suppose now that such a term does, indeed, appear, and let 1
2
(x
k
. y
k+i
)
be a corresponding homology class with a pseudo-holomorphic representative. Our goal now
is to study properties of this domain.
Step 5. Pinching induces a decomposition of the domain 1.
The boundary of the winding region J consists of two circles C
1
and C
2
parallel to the
meridional -circle, where C
1
is the component on the side of the regions 2 and 3. Fix a
complex structure J and, as in the denition of suciently pinched, consider degenerating
J (and so ) along J. Suppose that there is a

/(x
k
. y
k+i
) with multiplicity +1
in regions 2 and 3 and 0 in regions 0 and 1. We will show by positivity considerations
that i 1; in fact, we will see that the homology class 1 can be constructed as a kind
of connected sum of a domain 1
0

2
(x. y) with :
w
(1
0
) = i + 1 and :
z
(1
0
) = 0 with a
homology class 1
1

2
(r
k
. r
k+i
).
More precisely, let 1
2
(x
k
. y
k+i
) be a homology class giving a term in 1
23
(x
k
).
The domain 1 has multiplicity zero at the regions 0 and 1, no corners outside of the winding
region on
g
or
a
1
, and in fact, both of its corners on
g
(and on
a
1
) occur in the negative
half of the winding region. (Specically the corners occur at r
k
and r
k+i
, where both /
and / +i are negative; here we are using Step 4.) It follows that 1 induces a two-chain 1
0
on the destabilized surface
0
, having local multiplicity zero at ., and some multiplicity ,
at the point u (which in turn corresponds to C
1
, shrunk to a point).
The domain at the winding region induces another two-chain, supported in a disk. Specif-
ically, let J
L
be the part of the winding region to the left of
a
2
, so J
L
contains the regions
2 and 3. As we degenerate our curve C
1
to a point, J
L
becomes a disk
t
1
whose outer
boundary is
a
2
, the meridian of the knot from Figure 11.8, with a preferred central point
1
,
the node obtained by degenerating C
1
. See Figure 11.10. The two-chain 1
1
induced from 1
must have local multiplicity , at
1
. There are two distinguished arcs in
t
1
connecting
1
to the boundary. One is a portion of
g
. The other is a portion of
a
1
which winds around,
meeting
g
in a sequence of intersection points r
k

0
k=n/2
. Recall that for each / there is a
homology class
k
connecting r
k
to r
k1
, with multiplicity +1 at the regions 2 and 3, and
11.7. FROM CFK

TO

CFD: BASIS-FREE VERSION 217
Figure 11.9. Pinching o the winding region. The Heegaard surface
is degenerated at the two brown curves to give a nodal surface with two com-
ponents: one of these is the de-stabilized knot diagram
0
(equipped with a
pair of arcs
g
and
g
connecting the u and . punctures) and the other as a
sphere, which contains the winding region.

a
1

a
2
3
2
x
0
x
1
x
2
x
3
3
2
x
0
x
1
x
2
x
3
3
2
x
0
x
1
x
2
x
3
Figure 11.10. Homology classes in the winding region. The winding
region to the left of the meridian, from Figure 11.8, is drawn inverted so that
the outer boundary of the winding region corresponds to the dark marked
point, and the meridional arc corresponds to the outer boundary of the picture.
In the center and right we illustrate two homology classes:
1

2
(x
1
. x
2
)
and its complement,
1

2
(x
2
. x
1
) from the proof of Lemma 11.44.
multiplicity zero at
1
. There is also a homology class
k
connecting r
k1
to r
k
, which can
be thought of as the complement of
k
. It is easy to see that a positive domain in this disk
region which connects intersections r
k
and r
k+i
of the - and -arcs and has multiplicity
one in regions 2 and 3 necessarily decomposes as a composite
k

k

k+1

k+i
. In
particular, for our initial 1
2
(x
k
. y
k+i
), we get 1
1

2
(r
k
. r
k+i
), which we decom-
pose as
k

k

k+1

k+i
. Consequently, the multiplicities of 1
1
near the boundary
of the winding region are all equal (and indeed equal to i +1). Thus, the homology class 1
0
on
0
has multiplicity , = i + 1 at u.
Step 6. Holomorphic curves have constrained behavior under the degeneration. We turn
next to the behavior of holomorphic curves as we stretch the neck along J. By an argument
218 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
analogous to the proof of Proposition 9.6, a sequence of curves n
r
holomorphic with respect
to neck lengths : converges to a pair (n
0
. n
1
) consisting of a holomorphic curve (or
comb) n
0
in
0
[0. 1] 1 and a holomorphic curve (or comb) n
1
in J [0. 1] 1, where
we view J as a sphere with three punctures, corresponding to the two circles where J is
joined to the rest of and the puncture j in J (i.e., c of ). (For the purposes of some
of the arguments here, it is useful to bear in mind that
0
is equipped with a little extra
structure now, in the form of two arcs, remnants of
g
and
g
, which connect the punctures
u and ..) For notational convenience, write
1
to denote J viewed as a three-punctured
sphere. Let
1
denote the puncture of
1
corresponding to the curve C
1
.
The maps n
i
are analogous to the matched discussed in Section 9.1 (Denition 9.2), but
can have Reeb orbits in addition to Reeb chords. More precisely, in the case that n
0
and n
1
are curves rather than combs,
n
0
is a holomorphic map o
0

0
[0. 1] 1 (with respect to an appropriate
almost complex structure J on
0
[0. 1] 1), where o
0
is a surface with boundary,
boundary punctures, and interior punctures. At each boundary puncture, n
0
is
asymptotic to one of:
a chord in x [0. 1] at or y [0. 1] at +,
a chord in C
1
starting and ending on
a
1
C
1
,
a chord in C
1
starting and ending on
g
C
1
, or
a chord in C
1
starting on
g
and ending on
a
1
or starting on
a
1
and ending on

g
.
We call punctures of these types punctures, boundary punctures, boundary
punctures and mixed boundary punctures, respectively.
At each interior puncture, n
0
is asymptotic to the circle C
1
at some point (r. )
[0. 1] 1 (i.e.,

n
1
and
|
n (r. )); or possibly to a multiple cover of
C
1
.
n
1
is a map o
1

1
[0. 1] 1 (with respect to an appropriate almost complex
structure J on
1
[0. 1] 1), where o
1
is a surface with boundary, boundary
punctures, and interior punctures. At each boundary puncture, n
1
is asymptotic to
one of:
a chord in r
i
[0. 1] at ,
a chord at c (at some point (1. t) [0. 1] 1).
a chord in C
1
starting and ending on
a
1
C
1
,
a chord in C
1
starting and ending on
g
C
1
, or
a chord in C
1
starting on
g
and ending on
a
1
or starting on
a
1
and ending on

g
.
We call punctures of these types punctures, c punctures, boundary punc-
tures, boundary punctures and mixed boundary punctures, respectively.
At each interior puncture, n
0
is asymptotic to the circle C
1
at some point (r. )
[0. 1] 1 (i.e.,

n
1
and
|
n (r. )); or possibly to a multiple cover of
C
1
.
There is a bijection
a
between the boundary punctures of o
0
and the bound-
ary punctures of o
1
so that for each boundary puncture
a
i
of o
0
, n
0
t(n
0
(
a
i
)) =
t(n
1
(
a
(
a
i
))); and similarly for the boundary punctures.
There is a bijection
c
between the interior punctures of o
0
and the interior punctures
of o
1
so that for each interior puncture
c
i
of o
0
,
|
(n
0
(
c
i
)) =
|
(n
1
(
c
(
c
i
))).
11.7. FROM CFK

TO

CFD: BASIS-FREE VERSION 219
2
3
a
1
a
2
a
3
a
4
a
0
Figure 11.11. Notation for the local multiplicities in the winding
region. The multiplicities are denoted c
0
. . . . . c
4
, in order of distance from
the region containing
3
.
More generally, n
0
and n
1
can be combsthe obvious analogues of combs from Section 5.4,
but allowing Reeb orbits. (Indeed, in the presence of mixed boundary punctures, n
1
is forced
to be a holomorphic comb with at least two vertical levels.)
Step 7. Holomorphic curves with Reeb chords do not occur in the limit. We show next
that the limiting curves (n
0
. n
1
) have no mixed boundary punctures, boundary punctures
or boundary punctures.
If n
1
has a mixed boundary puncture then n
1
necessarily has at least two stories. Let
n
1,1
be the rst story of n
1
that contains a mixed boundary puncture, so n
1,1
is asymptotic
to some r
i
[0. 1] at and to a chord
1
connecting
a
1
to
g
at +. Number the local
multiplicities of n
1,1
in the various regions c
i

n/2+1
i=0
, ordered by distance from the region
containing the Reeb chord
3
, as indicated in Figure 11.11. Note that c
0
and c
1
coincide with
the local multiplicities of n
1,1
around
3
and
2
, respectively. In particular, c
0
. c
1
0. 1.
We rule out the four possibilities for c
0
and c
1
in turn.
If c
0
= 1 and c
1
= 0 then the domain n
1,1
consists entirely of the region 3. This
contradicts the fact that n
1,1
is asymptotic to a mixed chord at +. If c
0
= 0 and c
1
= 1 then
c
j
= , for , = 1. . . . . i and c
j
= i for , i (since n
1,1
has a corner at r
i
and no other
corners in the interior of J). But the fact that n
1,1
is asymptotic to a mixed chord at +
implies that c
n/2
= c
n/21
1, a contradiction. If c
0
= c
1
= 0 then 0 = c
0
= c
1
= = c
i
.
Since r
i
is an initial corner, it follows that c
i+1
= 1, a contradiction to the fact that n
1,1
is holomorphic. Finally, if c
0
= c
1
= 1 then c
j
= 1 for , i, c
i+1
= 0 and c
i+2
= 1, which
again contradicts the fact that n
1,1
is holomorphic. (Here we use the assumption that : is
suciently large, to guarantee that / + 2 < :,2 from the assumption on o(x
i
).) It follows
that n has no mixed chords.
Next, suppose n
0
has an boundary puncture. Then the source o
0
of n
0
has some
boundary component which is mapped to
a
1
. Since x and y have no components in
a
1

0
,
t n
0
remains bounded on this component of o
0
. By the maximum principle, it follows that

|
n
0
is constant on this component of o
0
. In particular, the boundary of this component
of o
0
is contained entirely in . Thus,

0
n
0
gives a homological relation between
0
.
Since these curves and arcs are homologically linearly independent, and the two-chain has
multiplicity 0 at .
0
(as established in Step 5), it follows that all its local multiplicities
vanish. (See also Lemma 5.42 for a similar argument.) Thus, n
0
(and hence n
1
) has no
220 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
boundary punctures. A similar argument shows that n
0
and n
1
have no boundary
punctures.
Step 8. Dimension constraints on holomorphic curves with Reeb orbits rule out all
holomorphic curves which are not supported in the winding region. In sum, the holomorphic
curves n
0
and n
1
are asymptotic to some collection of Reeb orbits (copies of C
1
(r. ),
or multiple covers of these) at
1
. Generically, all of these Reeb orbits will be simple (not
multiply covered), so it suces to consider this case. As in Step 5, we decompose 1
1
as
1
1
=
k

k

k+i1
, where now i is the number of Reeb orbits of n
0
and n
1
. (In
the case i = 0 there are no
j
in the decomposition and no Reeb orbits.) The (expected)
dimension of the moduli space of curves containing n
1
is ind(n
1
) = 2i +1, and the (expected)
dimension of the moduli space of curves near n
0
is given by ind(1
0
). Finally, each Reeb orbit
contributes a two-dimensional matching condition between n
0
and n
1
, so
ind(1. (
2
.
3
)) = ind(n
0
) + ind(n
1
) 2i
= ind(1
0
) + 1.
Thus, since ind(1. (
2
.
3
)) = 1, ind(1
0
) = 0, so n
0
is a trivial holomorphic curve. It follows
that i = 0, and 1 is exactly the class
k
, showing that the only possible curve is the one
analyzed in Step 1.
This completes the proof for 1
23
.
The case of 1
01
. This case is symmetric to the case of 1
23
, via a rotation of the winding
region (which also reverses the sign of the indexing of the r
i
and the sign of the function o).
The case of 1
3
. Suppose that
2
(x
0
. y) is a positive domain with local multiplicity
one at the region marked with 3 and zero at the regions marked by 0, 1 and 2. Considering
local multiplicities around the intersection point r
1
, we see that r
1
is forced to be a corner
point, and indeed that the intersection of with the winding region J is forced to be
the bigon domain from r
0
to r
1
. It follows (as in Step 2 of the analysis of 1
23
) that for
generic almost-complex structures, the only positive, index one
2
(x
0
. y) which admits
a pseudo-holomorphic representative is this bigon, connecting x
0
to y = x
1
.
The case of 1
1
. This is symmetric to the case of 1
3
via a rotation of the winding
region.
Lemma 11.45. For any suciently large :, if y S(:) satises [o(y)[ <
n
4
then 1
230
(y) =
1
2
(y) = 1
0
(y) = 1
012
(y) = 0.
Proof. By the same argument as used to prove Lemma 11.40, the grading shifts of these
maps are given by
1
0
: \
1
s
\
0
s
n+1
2
1
2
: \
1
s
\
0
s+
n+1
2
1
230
: \
1
s
\
0
s
n1
2
1
012
: \
1
s
\
0
s+
n1
2
.
Thus, if [o(y)[ <
n
4
then, writing 1

for any of the above maps, we nd [o(1

(y))[
n2
4
.
By Equation (11.41), the o-grading of any generator x
0
of \
0
coincides with the Alexander
grading of the corresponding x S
K
. Thus, taking : 4 max
xx
S
[(x)[ + 2, 1

(y) lies in
the zero group.
Lemma 11.46. The following holds for any suciently pinched almost complex structure
on [0. 1] 1. For each positive domain 1
2
(x. y) in 1
K
with x. y S
K
, :
w
(1) =
/ 0, and :
z
(1) = 0, there is a corresponding sequence of domains 1
i

2
(x
ik
. y
i
) for
/
n
2
i 0 such that:
11.7. FROM CFK

TO

CFD: BASIS-FREE VERSION 221
1
i
is positive.
If i < 0 or / = 0 then 1
i
is provincial, and if i = 0 and / 0 then 1
i
crosses
region 2 with multiplicity one and has multiplicity 0 at regions 0, 1 and 3.
If i < 0 or / = 0 then ind(1
i
. ()) = ind(1), and if i = 0 and / 0 then
ind(1
i
. (
2
)) = ind(1).
Outside the winding region, 1
i
agrees with 1 (in the obvious sense).
In the other direction, suppose 1
t

2
(x
ik
. y
i
) satises :
0
(1
t
) = :
1
(1
t
) = :
3
(1
t
) = 0 and
:
2
(1
t
) 0. 1. Let (1
t
) be the empty sequence of sets of Reeb chords if :
2
(1
t
) = 0 and
the sequence (
2
) if :
2
(1
t
) = 1. Suppose that
ind(1
t
. (1
t
)) = 1
/(1
t
. (1
t
)) ,= .
Then 1
t
= 1
i
for some 1
2
(x. y), and #/(1
t
. (1
t
)) = #/(1).
Similarly, for each positive domain 1
2
(x. y) in 1
K
with x. y S
K
, :
z
(1) = / 0,
and :
w
(1) = 0, there is a corresponding sequence of domains 1
i

2
(x
i+k
. y
i
) for 0 i
n
2
/ such that:
1
i
is positive.
If i 0 or / = 0 then 1
i
is provincial, and if i = 0 and / 0 then 1
i
crosses
region 0 with multiplicity one and has multiplicity 0 at regions 1, 2 and 3.
If i 0 or / = 0 then ind(1
i
. ()) = ind(1), and if i = 0 and / 0 then
ind(1
i
. (
0
)) = ind(1).
Outside the winding region, 1
i
agrees with 1 (in the obvious sense).
In the other direction, suppose 1
t

2
(x
i+k
. y
i
) satises :
1
(1
t
) = :
2
(1
t
) = :
3
(1
t
) = 0 and
:
0
(1
t
) 0. 1. Let (1
t
) be the empty sequence of sets of Reeb chords if :
0
(1
t
) = 0 and
the sequence (
0
) if :
0
(1
t
) = 1. Suppose that
ind(1
t
. (1
t
)) = 1
/(1
t
. (1
t
)) ,= .
Then 1
t
= 1
i
for some 1
2
(x. y), and #/(1
t
. (1
t
)) = #/(1).
Proof. Given 1, the domain 1
i
is constructed using the procedure illustrated by the
shaded region in Figure 11.8; 1
i
is uniquely determined by its corners r
ik
and r
i
, its local
multiplicities at the regions 0. 1. 2. 3, and the condition that 1
i
agrees with 1 outside the
winding region. It is immediate from the index formulas that 1
i
has the same index as 1.
Suppose next that 1
t

2
(x
s
. y
s+t
) for :. t 0 is an index one homology class
with :
0
(1
t
) = :
1
(1
t
) = :
3
(1
t
) = 0 which has a holomorphic representative. We claim that
1
t
= 1
s
for some 1
2
(x. y) with :
w
(1) = /, :
z
(1) = 0, and #/(1
t
. (1
t
)) = #/(1).
By stretching along the boundary of the winding region, as in the proof of Lemma 11.44,
we see that 1
t
can be written as a connected sum of a homology class 1
0
with a standard
chain
s,t

2
(r
s
. r
s+t
). We have that ind(
s,t
. (1
t
)) = 2t, and ind(1
t
) = ind(1
0
) +
ind(
s,t
) 2t. Thus, ind(1
0
) = 1.
We claim the moduli space is obtained as a bered product

/(
0
)
Sym
t
(|)

/(
s,t
).
As in the proof of Lemma 11.44, this follows from compactness and gluing results, together
with the following argument to rule out curves asymptotic to Reeb chords at the boundary
of J (so the only asymptotics at J are Reeb orbits). Chords with both endpoints on
or both endpoints on are ruled out exactly as in the proof of Lemma 11.44: such chords
222 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
force components of the curve in
0
to be constant. Mixed Reeb chords are ruled out by
considering the local multiplicities at the regions in J, similarly to the proof of Lemma 11.44;
for the bottom-most story containing a mixed puncture the local multiplicities c
i
are all 1
for i suciently large, which is inconsistent with being asymptotic to a mixed Reeb chord
at +.
Having identied our moduli space with the ber product

/(1
0
)
Sym
t
(|)

/(
s,t
), we
next claim that the map

/(
s,t
) Sym
t
(|), gotten by projecting the preimage of the
connect sum point , has degree one (onto Sym
t
(|)), so that the moduli space has the same
count modulo two as

/(1
0
), as desired.
In the case where t = 1, this can be seen directly. Consider the moduli space
s,1
, which,
in the notation of the proof of Lemma 11.44, is
s1
; see also the right of Figure 11.10. Our
claim is that the map of

/(
s1
) to [0. 1] 1 has degree one or, equivalently, after we
divide out the space

/(
s1
) by translations, the preimage of is mapped with degree one
onto the interval [0. 1]. But the quotient of

/(
s1
) by translations can be parameterized
by depth of the cut at r
s
. In the limit as the cut goes out to the marked point along the
-circle, the preimage of is mapped (under projection to [0. 1]) to 1; if the cut goes out
along the -circle, then the preimage of is mapped to 0. Hence the evaluation map has
degree one when t = 1, as stated.
For t 1 we show that the evaluation map has degree one as follows. By a straightforward
analysis of domains in the winding region (see Step 5 in the proof of Lemma 11.44),
s,t
can
be written as a juxtaposition of owlines with Maslov index two. In fact, any decomposition
of
s,t
consisting of non-negative and non-constant homology classes must necessarily be a
decomposition into homology classes each of which has non-zero multiplicity at . It follows
that the map

/(
s,t
) Sym
t
(|) is proper. To calculate its degree, consider the bers of
points in Sym
t
(|) where one of the components has 1 coordinate much larger than all the
others. Such holomorphic disks are in the image of the gluing map of a moduli space with
multiplicity 1 with another moduli space with multiplicity t 1. By induction, it follows
that the evaluation map has degree one.
Lemma 11.47. For any suciently pinched almost complex structure on [0. 1] 1, for
:
n
4
the map 1
23
: \
1
s
\
1
s+1
is as given by Theorem 11.35.
Proof. We introduce some preliminary notation. A curve u representing 1
23
has local
multiplicity zero at the regions marked by 0 and 1. Thus, if we let denote the
complement in of the regions marked by 0 and 1, the source of u maps to [0. 1] 1.
We prove the lemma in nine steps.
Step 1. 1
23
(x
1
) = 0 for any x S
K
, since there are no domains which can represent
such a map. On the one hand, any domain with initial corner at r
1
and vanishing local
multiplicities at 0 and 1 must in fact vanish at all four regions near r
1
. On the other hand,
a domain which represents 1
23
must have a cut (along
a
1
) which goes out the boundary
(separating regions 2 and 3). Thus, it cannot have vanishing multiplicities around r
1
.
Step 2. Degenerate the surface to extract a boundary degeneration. Consider 1
23
(x
i
) for
some 1 < i. We are interested in terms y
j
in 1
23
(x
i
) where , 0. We use a degeneration
argument similar to the proof of Lemma 11.44. As in the denition of suciently pinched
almost complex structures let C
2
be the boundary component of the winding region on the
0. 1 side (i.e., the right side as shown in Figure 11.9). Pinch the surface along C
2
. The
resulting nodal surface has two irreducible components: a topological disk which we denote
11.7. FROM CFK

TO

CFD: BASIS-FREE VERSION 223
Figure 11.12. Pinching along C
2
. This is half the pinching from Fig-
ure 11.9 (done on ).
J
R
corresponding to part of the winding region; and another component
t
corresponding
to most of the diagram and containing the regions 2 and 3. See Figure 11.12. Let
2
denote
the node (or puncture) corresponding to C
2
after the degeneration. Holomorphic curves
giving terms in 1
23
(x
i
) degenerate to pairs of holomorphic curves (n. ) where n is a curve
in J
R
[0. 1] 1 and is a curve whose interior lies in
t
[0. 1] 1.
By hypothesis, the components of x
i
and y
j
on
a
1
both lie in J
R
, and has punctures
mapped to the Reeb chords
2
and
3
. Let o
0
be the component of (the source of) part of
the boundary of which is mapped to
a
1
. Then one component of o
0
is mapped entirely to

a
1
, and the t-coordinate of remains bounded on this component. Consequently, as in the
proof of Lemma 5.45,
|
is constant on o
0
, so [
S
0
is a boundary degeneration.
Write the source of as o = o
0
Ho
1
.
Step 3. The curve consists of the boundary degeneration and trivial strips. By Propo-
sition 5.62, 1
23
counts domains 1 with
(11.48) 0 = c(1) + :
x
(1) + :
y
(1).
Let 1
2
be the domain in J
R
determined by n and 1
1
the domain in
t
determined by [
S
1
.
Let :

2
(1
i
) denote the local multiplicity of 1
i
at
2
. Then,
:

2
(1
2
) = :

2
(1
1
) + 1
:
x
i
(1
2
) + :
x
(1
1
) = :
x
(1) (p 1)
:
x
j
(1
2
) + :
y
(1
1
) = :
y
(1) (p 1)
c(1
2
) + c(1
1
) = c(1) 2 + 2p + 2:

2
(1
2
).
(11.49)
(In these formulas, for the purposes of the Euler measure c, the point
2
is treated as a node,
not a puncture.) Combining Equations (11.49) and (11.48) gives
(11.50) c(1
1
) + :
x
(1
1
) + :
y
(1
1
) + c(1
2
) + :
x
i
(1
2
) + :
x
j
(1
2
) = 2:

2
(1
1
) + 2.
A direct analysis of the domains in the winding region shows that
c(1
2
) + :
x
i
(1
2
) + :
x
j
(1
2
) = 2:

2
(1
2
).
224 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
It follows that c(1
1
) + :
x
(1
1
) + :
y
(1
1
) = 0. This is the expected dimension of the param-
eterized moduli space of holomorphic curves in the domain 1
1
. It follows that [
S
1
consists
of trivial strips, so 1
1
= 0 and :

2
(1
2
) = 1. In particular, y = x.
Step 4. Analyzing n. The homology class of n is uniquely determined by the properties
that its local multiplicities at the regions 0 and 1 are zero, r
i
is a corner, and its local
multiplicity
2
is 1. It follows that n is connects r
i
to r
i1
. Combining this with the
previous step, it follows that y
j
= x
i1
. It also follows from the above considerations that
there is a unique holomorphic curve n in this homology class.
At this point, we have shown that either 1
23
(x
i
) = x
i1
or 1
23
(x
i
) = 0. It remains to
exclude the latter case; this involves a gluing argument.
Step 5. Relaxing the almost complex structure. We would like to assert that there
is a unique holomorphic curve in the moduli space of with the Reeb chord in C
2
at a
given t-coordinate, and that we can glue this curve to n to get a holomorphic curve in
. Unfortunately, if p 1 then the curve is not transversally cut out for any choice of
admissible almost complex structure in the sense of Denition 5.1.
Instead, we relax Condition (J-1) of the denition of admissible complex structures,
to [Lip06a, Condition (J5
t
)], which we restate here as:
(J
t
-1) There is a 2-plane distribution on [0. 1] such that the restriction of to is
non-degenerate, J preserves , and the restriction of J to is compatible with .
We further assume that is tangent to near ( ) [0. 1].
This is sucient for [BEH
+
03] to give compactness for the moduli spaces of holomorphic
curves. We call such almost-complex structures relaxed.
Given an almost complex structure J on [0. 1] 1, let /(; ; J) denote the moduli
space of J-holomorphic curves containing . For a generic almost complex structure satisfying
(J
t
-1) rather than (J-1), the moduli spaces /(; ; J) is transversally cut out (and, in
particular, does not contain any boundary degenerations).
Step 6. Independence of /(. ; J) from the -curves. Let
t
be another set of -
curves that agree with near x and let /(;
t
; J) denote the moduli space of holomorphic
curves with respect to
t
corresponding to /(; ; J). We claim that for J a generic almost
complex structure satisfying condition (J
t
-1) close enough to an almost complex structure J
0
satisfying condition (J-1), #/(;
t
; J) = #/(; ; J). Indeed, taking J J
0
, elements
of /(;
t
; J) converge to the unions of -boundary degenerations and trivial strips. Let l
x
be a small neighborhood of x so that l
x
=
t
l
x
. It follows that for J close enough
to J
0
and u /(;
t
; J), u(o) ( l
x
); and similarly for elements of /(; ; J).
Thus, for J close enough to J
0
, /(; ; J) = /(;
t
; J).
Step 7. A model computation when p = 1. In the special case where p = 1, the
moduli space /(; ; ,

,
|
) is transversally cut out, and consists of a single point. This
is analogous to Proposition 5.16.
Step 8. Stabilizing the model computation. Let
t
=
t
1
. . . . .
t
g1
be a (p 1)-tuple of
curves so that
t
i

i
is a single point and
t
i

j
= for i ,= , (and the
t
j
are disjoint from

a
1
and
a
2
). Choose a curve so that
t
has two connected components, one containing

t
and the other containing
a
1
. Stretching the neck along , /(;
t
; J) breaks up as a
bered product of two moduli spaces over a disk. One is the moduli space from the model
computation in Step 7, which has a single, transversally cut-out point. The other maps
11.7. FROM CFK

TO

CFD: BASIS-FREE VERSION 225
degree 1 onto the disk, by the same argument used to prove stabilization invariance in the
cylindrical setting [Lip06a, Proposition A.3].
Step 9. Conclusion of the proof. Steps 6 and 8 imply that /(; ; J) is transversally cut
out, and consists of an odd number of points, for some relaxed almost-complex structure J,
which we can take to be arbitrarily close to some xed, admissible J
0
(suciently pinched).
By a standard gluing argument, this moduli space can be glued to n to give an odd number
of holomorphic curves contributing to 1
23
(x
i
). Thus, the number of points in the moduli
space of J-holomorphic curves /(x
i
. x
i1
; (
2
.
3
); J) is odd. Let J
0
be a suciently pinched,
generic almost complex structure satisfying the more restrictive Condition (J-1). Since the
moduli space of J
0
-holomorphic curves /(x
i
. x
i1
; (
2
.
3
); J
0
) is transversally cut out, it
follows that #/(x
i
. x
i1
; (
2
.
3
); J
0
) is odd, as well. Thus, 1
23
(x
i
) = x
i1
, as desired.
Proof of Theorem 11.35. We choose : larger than 4c, where c is the constant from
Lemma 11.43, so that all exterior generators x have [o(x)[
n
4
, and large enough that
Lemmas 11.44 and 11.45 hold.
Let \ = \
0
\
1
be the type 1 module for

CFD(o
3
nbd(1)). We see from
Lemma 11.41 that \
0
s
= C(:), and its dierential 1 is clearly identied with the dier-
ential on knot Floer homology.
Consider next generators x representing elements of \
1
. In the case where [o(x)[
n
4
,
according to Lemmas 11.41 and 11.43, all generators with the same o-grading are of the form
x
t
for x S
K
, and t uniquely determined by (the Alexander grading of) x. This gives the
identication of \
1
s
with \
1
s
for [:[
n
4
as stated in the theorem. For instance, for :
n
4
,
\
1
s
contains x
k
for o(x) = : +
n+1
2
/ and can therefore be identied with C
_
: +
n1
2
_
,
as desired. Moreover, Lemma 11.46 then identies the dierential 1 on \
1
s
with the stated
dierential 1 on \
1
s
. We return to studying \
1
with [o(x)[
n
4
later in the proof. We say
what we can about the other coecient maps rst, though.
Any domain representing 1
1
must start at a generator of the form x
0
. By Lemma 11.44,
the map 1
1
is represented by the bigon from x
0
to x
1
. Similarly, any domain representing
1
3
is represented by the bigon from x
0
to x
1
. We conclude that 1
1
and 1
3
are the maps
from x
0
to x
1
and x
1
, respectively, as desired.
Next, any domain representing 1
2
must start at a generator of the form x
s
with : < 0
and end at a generator of the form y
0
. By Lemma 11.46, such a domain must be induced
by some domain
2
(x. y) with :
z
() = 0 and :
w
() = / 1.
An analogous argument shows that 1
123
has the stated form, except that we necessarily
have :
w
() = 1.
There are no domains which could represent a map of the form 1
12
, for such a domain
would need to have a cut going out to the intersection point r
0
, which has the wrong
idempotents.
The fact that 1
23
has the stated form on \
1
s
for :
n
4
was established in Lemma 11.44.
The fact that 1
23
has the given form for :
n
4
follows from Lemma 11.47.
We have now shown that all the coecient maps have the right form for \
1
s
with [:[
n
4
and for \
0
s
. We claim next that for all [:[
n
4
, H

(\
1
s
. 1)

= F
2
, and in fact
1
23
: \
1
s
\
1
s+1
is a chain map (with respect to 1) inducing an isomorphism on homology. We investigate the
terms in Equation (11.30) and Equation (11.31), bearing in mind Lemma 11.45. Specically,
for x with [o(x)[
n
4
, 1
3
1
012
(x) = 0, 1
1
1
230
(x) = 0, and 1
301
1
2
(x) = 0 for
226 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
degree reasons (comparing Lemma 11.40 with the observation that \
0
s
= 0 for all [:[
n
4
).
Similarly, since
1
0
: \
1
s
\
0
s
n+1
2
.
(by the same considerations as in Lemma 11.40), it follows that 1
123
1
0
(x) = 0 for [o(x)[
n
4
. Thus 1
01
and 1
23
are homotopy inverses in this range, and so the map induced on
homology by 1
23
: \
1
s
\
1
s+1
is an isomorphism. Then since \
1

n+1
4
is (C.
w
), a chain
complex with homology isomorphic to F
2
, we can conclude that
H

(\
1
s
. 1)

= F
2
for all [:[
n
4
.
It follows that for [:[
n
4
, we have a chain map : \
1
s
F
2
inducing an isomorphism
on homology. The map
: \ \
given by
(x) =
_
(x) if x \
1
s
with [:[ <
n
4
x otherwise
induces a homotopy equivalence between the type 1 module determined by \ and the one
determined by \ .
11.8. Proof of Theorem 11.27
We next deduce Theorem 11.27 from Theorem 11.35. But rst, we show that CFK

(1)
can be horizontally or vertically reduced, as in Denition 11.25 (and as needed for the
statement of Theorem 11.27). To this end, we have the following preliminary lemma:
Lemma 11.51. Let \ and \ be nite-dimensional ltered vector spaces and ) : \ \ be
a (possibly un-ltered) isomorphism between the underlying vector spaces. Then there is a
ltered basis for \ that maps under ) to a ltered basis for \.
Proof. The ltration \
j
\ of \ induces a second ltration on \ , by the subsets
)
1
\
j
, which, together with the original ltration on \ gives a (Z Z)-ltration on \ by
subcomplexes \
s,t
= \
s
)
1
\
t
. Correspondingly, the ltration level of an element \
is dened to be the pair of integers (:. t) where : = supi Z [ \
i
, t = sup, Z [
)() \
j
. Let [[]] denote the projection of to \
s,t
,(\
s1,t
+ \
s,t1
). We can form the
associated ZZ-graded vector space gr
ZZ
(\ ) =

s,t
\
s,t
. A (ZZ)-ltered basis is a basis
for \ whose projection under [[]] is a basis for gr
ZZ
(\ ).
Now, pick a (Z Z)-ltered basis for \ (which can be obtained by taking a basis for
gr
ZZ
\ and lifting it to a basis for \ ). Such a basis is a fortiori a ltered basis for either
the ltration by \ or the ltration induced by \; i.e., it can be thought of as a (Z-)ltered
basis for \ whose image under ) is a ltered basis of \, as needed.
Proposition 11.52. Let C be a Z-ltered, Z-graded, nitely-generated chain complex over
F
2
[l] which is free as an F
2
[l]-module. Then C is Z-ltered, Z-graded homotopy equivalent
to a chain complex C
t
which is reduced. Further, one can choose a basis for C
t
over F
2
[l]
which is vertically simplied or, if one prefers, a basis which is horizontally simplied instead.
11.8. PROOF OF THEOREM 11.27 227
Proof. We show that C can be reduced by induction on its rank. If C is not reduced, by
denition it admits a non-trivial dierential which does not change the Alexander ltration,
so there is some C with ,= 0 and () = (). Then and generate a subcomplex
of C whose quotient complex Q is ltered homotopy equivalent to C with rank two less than
the rank of C.
Next, we argue that C can be vertically simplied. Consider C
v
= C,(l C) with its dif-
ferential
v
. By Lemma 11.51 applied to the restriction of
v
to a map C
v
,Ker(
v
) Im(
v
),
we can nd a ltered basis
2
.
4
. . . . .
2m
for Im(
v
) and vectors
1
.
3
. . . . .
2m1
in C
v
so that
v
(
2i1
) =
2i
and
1
.
3
. . . . .
2m1
descends to a ltered basis of C
v
,Ker(
v
). The
set
1
.
2
. . . . .
2m
is linearly independent, since
v
is a dierential. Extend
2
.
4
. . . . .
2m

to a ltered basis
2
.
4
. . . . .
2m
.
2m+1
. . . . .
n
for Ker(
v
). Then
1
. . . . .
n
is a ltered
basis for C
v
, and can be thought of as a basis for C over F
2
[l], as well. As a basis for C,

1
. . . . .
n
is vertically simplied.
A similar argument applies for C
h
= gr
0
(C
F
2
[U]
F
2
[l. l
1
]) (where we are now taking
the associated graded object with respect to the Alexander grading) to give a ltered basis

1
. . . . .
n
for C
h
so that
h
(
2i1
) =
2i
for 1 i : and
h
(
i
) = 0 for i 2:. Pick
(arbitrary) lifts
i
of
i
to C

= C
F
2
[U]
F
2
[l. l
1
]; by assumption (
i
) = 0, and each
i
can be written uniquely as
i
= l
k
i

i
for some
i
C with non-zero projection to C,lC.
Then the
i
are the desired horizontally simplied basis.
Proof of Theorem 11.27. We will write

CFD for

CFD(o
3
nbd(1)) with a suitably
large framing :. Let C be a reduced complex homotopy equivalent to CFK

(1), and let

i
and
i
be bases of C which are vertically and horizontally simplied, respectively.
These bases are related by
(11.53)
i
=

j
i

j
for a suitable change of basis matrix . As in Theorem 11.35,
i
and
i
can also be
thought of as two dierent bases for \
0
s
over F
2
or, alternatively, two bases for
0

CFD over
/(). We will modify and extend these to two bases for all of

CFD which
on
0

CFD are still related by Formula (11.53),

agree on
1

CFD, and
split as a direct sum of a contractible complex and a complex with the coecient
maps from Theorem 11.27.
Here, to dene the coecient maps, we view

CFD as a type 1 structure as in Remark 6.4.
To give an initial basis for
1

CFD, recall that all

i
except for
0
come in pairs
j
and

j+1
, where
z

j
=
j+1
. Fix i so that there is a height / vertical arrow from
i
to
i+1
. Then
the generators for \
1
s
for : 0 coming from this pair have coecient maps like
(11.54)

i
D
1

i
1

D
23

i
k
D
23

i
k+1
D
23

D
23

D
23

i
D
23

D
23

i+1
D
1

i

D
23

That is, the only time we get a contribution from this pair to the 1

-homology of \
1
s
for
: 0 is for (
i+1
) +
_
n1
2
_
< : (
i
) +
_
n1
2
_
. We call these generators
i
k
as in the
228 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
statement of the theorem, where / = (
i
) : +
_
n1
2
_
+ 1. For : = (
i+1
) +
_
n1
2
_
, we
have a canceling pair of generators
i
(from
i
) and
i
(from
i+1
). We also have canceling
pairs for smaller :.
Similarly, for a horizontal arrow from
i
to
i+1
of width /, we get generators and coe-
cient maps like
(11.55)

i
D
3

i
1
D
23

D
23

i
k
D
23

i
k+1
D
23

D
23

D
23

D
2

i
D
23

i+1
D
3

i
D
23

(The
i
k
are elements of \
1
s
with : < 0.) In particular, there is an undesirable term in
1
3
(
i+1
). In addition, we usually have 1
123
(
i
) = 0; the exception is when the arrow has
width equal to one, when
(11.56) 1
123
(
i
) = 1
1
(
i+1
).
Consider next
0
, which has no vertical dierential either into or out of it. We see that
its corresponding element comes from a chain

D
23

3
D
23

2
D
23

1
D
1

0
.
where the chain of elements
i
extends back to \
1
s
where : is small in magnitude (and
hence \
1
s
is one-dimensional). Similarly, consider the element
0
which has no horizontal
dierential either into or out of it. (Remember that
0
and
0
can coincide.) Associated to

0
is a similar string

0
D
3
j
1
D
23
j
2
D
23
j
3
D
23

where the chain of elements j
i
extends forward to \
1
s
where : is suciently small in mag-
nitude that \
1
s
is one-dimensional). Thus, for some suciently large c and /, we must have
that
a
= j
b
; i.e., we can connect the two chains to a single one of the form

0
D
3
j
1
D
23
j
2
D
23
j
3
D
23

D
23
j
m
D
1

0
We have now constructed bases for all the \
1
s
:

i
k
and j
k
for : :,4,

k
= j
k
for :,4 < : < :,4, and

i
k
and
k
for : :,4.
However, we have extra generators for the modules and extra terms in the coecient maps.
To bring the module to the desired form, we must change basis so as to
split o the chain of canceling generators in (11.55) by arranging for 1
3
(
i+1
) and
1
23
(
i

) to vanish;
eliminate the generators with canceling dierentials in (11.54), in particular
j
and
j
; and
bring 1
123
to the desired form, by eliminating the non-trivial terms belonging to
width one arrows (Equation (11.56)).
We do this by successively changing basis (and, at some point, dropping out
j
).
We start by modifying the base to eliminate the unwanted terms in 1
3
(from
i+1
when
there is a horizontal arrow from
i
). To this end, we change basis by

t
i+1
:=
i+1
+
3

i
.
11.8. PROOF OF THEOREM 11.27 229
/

c d

.
.
.
.
.
.
D

.|
|
|
|
|
|
|
|
|
/
2
D
23

c
2
D

{
{
{
{
{
{
{
{
D
23

/
1
D
23

c
1
D
2
{
{
{
{
{
{
{
{
D
23

D
23

a
a
a
a
a
a
a
a
a
a
/
2
D
23

Y
Y
Y
Y
Y
Y
Y
Y
Y
/
1
/
0
D
1

D
3

c
0
D
1

D
3

D
123
c
1
c
2
D
23

D
23

D
23

c
2
D
23

c
1
c
0
D
3

D
1

d
0
D
3

D
1

D
123

d
1
d
2
D
23

D
23

c
1
D
23

d
1
D
23

D
2
c
c
c
c
c
c
c
c
c
c
2
D
23

d
2
D
23

g
g
g
g
g
g
g
g
.
.
.
.
.
.
D

f
f
f
f
f
f
f
f
f
/
0
+
3
c
2
D
1

c
1
D
2

c
0
D
3

D
1

/
1
c
1
+
23
/
1
c
0
+
3
d
2
+
1
/
2
D
123

d
1
D
2

d
0
+
1
c
2
+
1
c
2
D
3

D
123

Figure 11.13. An illustration of the proof of Theorem 11.27. The

top illustrates a summand in the knot complex; the middle illustrates a corre-
sponding piece coming from Theorem 11.35; the bottom is obtained by making
the substitutions coming from the proof of Theorem 11.27 and dropping acyclic
summands.
230 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
With this substitution, we have that 1
3
(
t
i+1
) = 0, 1
2
(
i

) =
t
i+1
, and 1
23
(
i

) = 0. Let
t
j

denote the corresponding change of basis applied to the

j
so that Formula (11.53) remains
true. Note that the other coecient maps are unchanged; in particular 1
1
(
t
j
) = 1
1
(
j
), and
there are no coecient maps other than 1
2
which enter
t
i+1
. Also, the chain of canceling
generators from this horizontal arrow now splits o as a direct sum, so we can drop them.
We pass now to a submodule to eliminate
i
and
i
. To do this, for each vertical arrow
from
i
to
i+1
of height / (and corresponding new elements
t
i
and
t
i+1
), we replace
t
i+1
by

tt
i+1
:=
t
i+1
+
1

i
.
Now we have 1
123
(
tt
i+1
) =
i

+ 1
123
(
i+1
). Again, there is a corresponding modication to
the other generators, replacing
t
i
by
tt
i
. Now the submodule without the chain of canceling
generators from (11.54) is easily seen to be homotopy equivalent to the original module.
Finally, we wish to bring 1
123
to the desired form, eliminating the terms from horizontal
arrows of length one. This is done as follows. Suppose that
i
is the initial point of some
vertical arrow. Replace
i
1
by

i
1
:=
i
1
+
23
1
123
(
i
).
With this new basis, then, 1
123
(
tt
i
) = 0 (as its contribution is absorbed into 1
1
). Moreover,
this change does not aect any of the other coecient maps.
It remains to make one last change of basis, corresponding to terminal points
i+1
of
vertical arrows, so that 1
123
(
i+1
) is the element taking the place of
i

. A priori we have
1
123
(
tt
i+1
) =
i

+ 1
123
(
i+1
). In the case where the length of the arrow from
i
to
i+1
is
greater than one, we make the change by letting

i

:=
i

+ 1
123
(
i+1
).
Since 1
123
(
i+1
) is in the kernel of 1
23
(from the description in Theorem 11.35), all other
coecient maps remain unchanged.
In the case where the length of the arrow from
i
to
i+1
is one, proceed as follows. First,
we nd an element
i
so that

i
=
23
1
123
(
i+1
).
In particular, 1

(
i
) = 0 and 1
23
(
i
) = 1
123
(
i+1
). This element is found as follows. Recall
that
i
and
i+1
correspond to elements x
i
and x
i+1
in C(:) and C(: 1), respectively.
Let y :=
1
(x
i
) C(: + 1), i.e., the image of x
i
under the map which counts holomorphic
disks which cross u exactly once. (Recall from Theorem 11.35 that 1
123
(
i+1
) is
1
(x
i+1
)
considered as an element of C(:).) We then let
i
=
z
(y), where we consider y as a chain
in C( : 1). Then
i
, thought of as a generator of the type 1 module, has the desired
properties. Making the substitution

ttt
i+1
:=
tt
i+1
+
1

i
.
we see that 1
1
(
ttt
i+1
) = 1
1
(
tt
i+1
) and 1
123
(
ttt
i+1
) has the desired form.
This module now has the form promised in Theorem 11.27.
As a nal point, we can relate the length : of the chain of j
i
to the framing parameter
and the knot invariant (1), as follows. We let x be some intersection point which appears
with non-zero multiplicity in
0
, and y be one which appears with non-zero multiplicity
in
0
. The number 2(1) is the Alexander dierence between x and y. Concretely, there is
some
2
(x. y) with :
w
() = (1) and :
z
() = (1). By adding copies of , we can
turn this into a domain connecting the corresponding intersection points, thought of now
11.9. CABLES REVISITED 231
0
1
2 3
u
.
c
/
1
/
2
Figure 11.14. Heegaard diagram for computing the type module
for the (2. 1) cable. This is a doubly-pointed Heegaard diagram for the (2. 1)
cable (of the unknot), thought of as a knot in the solid torus. The basepoint
. lies in the region marked with a 0.
as generators for

CFD(1), whose local multiplicity is zero near 0 and 1, and whose local
multiplicity at the regions 2 and 3 is 2(1). Combining this with our chain of domains going
through the j
i
gives a periodic domain with local multiplicities (0. 1. 1 :. :) at the four
regions around (0. 1. 2. 3), where : = : + 2(1).
In Figure 11.13 we have illustrated some of the changes of basis from Theorem 11.27 in
the case where the knot complex has a summand which is a square, as illustrated on the
top of the gure. (Such a summand appears, e.g., in the knot Floer complex for the gure
eight knot, on the bottom of Figure 11.5.) Initially, Theorem 11.35 gives a corresponding
summand as illustrated in the middle of Figure 11.13. Going through the cancellations
prescribed in the proof of Theorem 11.27, we end up with the complex on the bottom of
Figure 11.13.
11.9. Cables revisited
In this section, we illustrate Theorems 11.21 and 11.27 by showing how to compute
HFK

of a knot 1 obtained as a cable of the left-handed trefoil 1 in o

3
. (Specically, we
compute HFK

of the (2. 3)-cable of 1.) Although the knot Floer homology groups of this
particular knot were already known [Hed05a], the techniques here extend to give a formula
for HFK

of any particular satellite in terms of CFK

of the pattern.
Figure 11.14 is a Heegaard diagram for the (2. 1) cabling operation. Since this diagram
has genus equal to one, the holomorphic curves are straightforward to count. The type
module CFA

(C) associated to this diagram is given as follows. It has three generators c,

/
1
and /
2
, and the following relations:
(11.57)
:
1
(/
1
) = l /
2
:
2
(c.
1
) = /
2
:
3+i
(c.
3
.
i
..

23
. . . . .
23
.
2
) = l
2i+2
c. i 0
:
4+i
(c.
3
.
i
..

23
. . . . .
23
.
2
.
1
) = l
2i+1
/
1
. i 0.
Consider next the type 1 module for the left-handed trefoil knot 1 with framing 2, repre-
sented graphically in Figure 11.15. This module has generators r
1
. r
2
. r
3
in the
0
idempotent
232 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
and
1
.
2
in the
1
idempotent, and coecient maps:
1
3
(r
1
) =
1
1
12
(r
1
) = r
3
1
2
(
1
) = r
2
1
123
(r
2
) =
2
1
1
(r
3
) =
2
.
(Although this framing is not large enough for Theorem 11.27 to apply,

CFD(1) may still
be computed; see Theorem A.11, whose proof uses results from [LOT10a].)
Let 1 denote the (2. 3)-cable of the (0-framed) left-handed trefoil (which is also the
(2. 1)-cable of the 2-framed left-handed trefoil). By Theorem 11.21, gCFK

(1) is com-
puted by the tensor product CFA

(C)

CFD(1), which has generators c r
i
and /
k

j
.
Dierentials are easily computed; they are:
(c r
1
) = l
2
c r
2
(c r
2
) = 0 (c r
3
) = /
2

2
(/
1

1
) = l /
2

1
(/
2

1
) = 0 (/
1

2
) = l /
2

2
(/
2

2
) = 0.
The relative Alexander grading is determined by this data and the facts that l
4
c r
3
has
the right grading to appear in (c r
1
), and l
3
(c r
1
) has the right grading to appear
in (/
1

1
). (The dierentials and Alexander gradings are also displayed in Figure 11.15.)
The concordance invariant can be read from this data, bearing in mind the interpreta-
tion of in terms of the Alexander grading on HFK

, as in Equation (11.18). Specically,

we nd that HFK

has a free summand, generated by the element l cr

3
+/
1

2
, an element
with Alexander grading 3, which in turn is equal to . Thus, (c
2,3
1
2,3
) = 3.
To calculate the bigraded knot Floer homology complexes, we must rst grade the factors
in the tensor product. The (enhanced) grading on

CFD(1) is given by
(11.58)
gr(r
1
) = (1; 0. 2; 0),/ gr(r
2
) = (
1
2
; 0. 1; 0),/
gr(r
3
) = (0; 0. 0; 0),/ gr(
1
) = (
3
2
;
1
2
.
3
2
; 0),/
gr(
2
) = (
1
2
;
1
2
.
1
2
; 0),/.
where, in the notation (c; /
1
. /
2
; c), (c; /
1
. /
2
) denotes the component in the ordinary grading
group, as in Section 11.1, and c denotes the Alexander component. The indeterminacy /
will be calculated shortly.
The displayed gradings are uniquely determined from the gradings on the algebra, the
normalization gr(r
3
) = (0; 0. 0; 0), and the algebraic structure of

CFD(1). For example,
given the grading of
2
, we can calculate the grading of r
2
, since
123

2
appears in r
2
; so
that

1
gr(r
2
) = gr(
123
) gr(
2
).
Given the computations in Equations (11.58), the indeterminacy / is determined from the
fact that
12
r
3
appears in r
1
, so / divides
gr(
12
)
1

1
(1; 0. 2; 0) = (
3
2
; 1. 2; 0).
Thus, since (
3
2
; 1. 2; 0) is primitive,
/ = (
3
2
; 1. 2; 0).
(It is not a coincidence that / is primitive: any loop passing through the unstable chain once
will have a 1 as the rst spin
c
component of its grading.)
11.9. CABLES REVISITED 233

2
r
3
1
12
cr
2
l
2
l
4
1
1
1
123
r
2

1
r
1
1
3
1
2
cr
1
/
1

1
/
1

2
/
2

2
1
cr
3
l
l
3
l
/
2

1
Figure 11.15. Cable of the trefoil. On the left, the coecient maps
for the left-handed trefoil. On the right, the result of cabling. Arrows with
solid arrows represent dierentials; they are labeled by their corresponding
coecients (powers of l). Dotted arrows do not represent dierentials; they
represent domains, and the labels over them represent changes in Alexander
gradings.
As in the case of

CFD(1), the gradings on CFA

(C) can be calculated from the gradings

on the algebra and the algebraic form of this module (Equations (11.57)), combined with
the normalization that gr(c) = (0; 0. 0; 0) (say). This leads to the (enhanced) grading on the
type module:
gr(c) = p(0; 0. 0; 0) gr(/
1
) = p(
1
2
;
1
2
.
1
2
; 1) gr(/
2
) = p(
1
2
;
1
2
.
1
2
; 0).
where p = (
1
2
; 0. 1; 2).
The gradings of elements in the tensor product lie in the double-coset space
(
1
2
; 0. 1; 2)G,(
3
2
; 1. 2; 0).
Indeed, choosing appropriate coset representatives, we nd that
(11.59)
gr(cr
1
) = (2; 0. 0; 4) gr(cr
2
) = (1; 0. 0; 2) gr(cr
3
) = (0; 0. 0; 0)
gr(/
1

1
) = (6; 0. 0; 7) gr(/
2

1
) = (5; 0. 0; 6) gr(/
1

2
) = (0; 0. 0; 1)
gr(/
2

2
) = (1; 0. 0; 0)
The rst component here is the .-normalized Maslov grading ` of Equation (11.15), a priori
up to an overall translation. Similarly, the last component is the Alexander grading, again
up to an overall translation. We discuss next how to compute the absolute Maslov and
Alexander gradings; for simplicity, we explain this for

HFK, the homology of the l = 0
specialization of the complex.
The l = 1 specialization of the complex has homology generated by cr
3
, an element
whose grading has rst component equal to zero. So, it follows from Equation (11.17) that
the rst component of the grading is, in fact, the absolute .-normalized Maslov grading,
rather than a non-trivial translate of it.
234 11. BORDERED MANIFOLDS WITH TORUS BOUNDARY
Next, observe that the Poincar polynomial for the tensor product, after setting l = 0,
is given by

d,r

d
N
t
r
rank H
(A,N)=(r,d)
(CFA

(C)

CFD(1)) = t
7

6
N
+t
6

5
N
+t
4

2
N
+t
2

N
+t
0

0
N
.
Here, the formal variable t records the Alexander factor of the grading, while the formal vari-
able
N
records the .-normalized grading ` of Equation (11.15), which is the rst coordinate
of Equation (11.59), up to an additive constant. This polynomial is not symmetric under
exchanging t and t
1
, but t
4
times it is. So, by Equation (11.16), the absolute Alexander
gradings are 3 more than the last components in Equation (11.59). With respect to this
rescaling, the Poincar polynomial is

d,r

d
N
t
r
rank H
(A,N)=(r,d)
(CFA

(C)

CFD(1)) = t
3

6
N
+ t
2

5
N
+
2
N
+ t
2

N
+ t
3
.
Since ` = ` 2, we can recover the Poincar polynomial of 1 in terms of the usual
Alexander and Maslov gradings:

d,r

d
t
r
rank H
d
(CFA

(C)

CFD(1). :) = t
3
+ t
2
+
2
+ t
2

5
+ t
3

6
.
Here, the formal variable t records Alexander gradings as before, while records the usual
Maslov gradings. This answer is in agreement with [Hed05a, Theorem 1.0.6].
APPENDIX A
Bimodules and change of framing
The modules

CFA(1 ) and

CFD(1 ) depend (up to homotopy equivalence) not only on the
3-manifold 1 but also on the parametrization of 1 by a reference surface 1; this dependence
can already be seen for the modules associated to solid tori computed in Section 11.2. The
result of reparametrization (e.g., change of framing on a knot complement) is captured by
tensoring with certain bimodules. Furthermore, the modules

CFA(1 ) and

CFD(1 ) are
related to each other by certain dualizing bimodules. These topics are explored in detail
in [LOT10a]. In this chapter we present highlights from that paper, and then exhibit the
bimodules relevant to the case of torus boundary. When combined with Theorem 11.27, this
gives a complete computation of the bordered Floer invariants of knot complements in o
3
,
with any framing.
We start by briey stating the main results on bimodules in Section A.1. We then sketch
the denitions of the bimodules in Section A.2. We give explicitly the bimodules for a set of
generators of the mapping class group of the torus in Section A.3 and use these bimodules
in Section A.4 to nish the computation of

CFD of knot complements in terms of the knot
Floer homology.
A.1. Statement of results
Fix a pointed matched circle :. Recall from Denition 4.2 that a bordered 3-manifold
with boundary 1(:) is a triple (1. :. ) where 1 is a compact, oriented 3-manifold with
boundary, : is a pointed matched circle and : 1(:) 1 is an orientation-preserving
dieomorphism. We have associated to (1. ) homotopy equivalence classes of modules

CFD(1. ) (Chapter 6) and

CFA(1. ) (Chapter 7) over /(:) and /(:), respectively.
Here, is the map but viewed as a map 1(:) 1 ; compare Theorem 1.1.
Theorem A.1. Given a dieomorphism : 1(:) 1(:), there is an /

bimodule
,(:)

CFDA()
,(:)
, well-dened up to (/

) homotopy equivalence, such that for any bor-

dered three-manifold (1. : 1(:) 1 ),

CFA(1.
1
)

CFA(1. )

,(:)

CFDA()

CFD(1. )

CFDA()

,(:)

CFD(1. ).
Moreover, given another map : 1(:) 1(:),

CFDA( )

CFDA()

,(:)

CFDA().
As we will discuss in Section A.2,

CFDA() is dened in terms of a Heegaard diagram
with two boundary components, one of which is treated in type 1 fashion and the other of
which is treated in type fashion.
235
236 A. BIMODULES AND CHANGE OF FRAMING
Next we turn to the duality between

CFD and

CFA. By `
R,S
we mean a bimodule `
with commuting right actions by rings 1 and o; similarly, by
R,S
` we mean a bimodule `
with commuting left actions by 1 and o.
Theorem A.2. For any pointed matched circle : there are dierential graded bimodules

CFAA(")
,(:),,(:)
and
,(:),,(:)

CFDD(") such that for any bordered (1. ),

CFD(1. )

CFA(1. )

,(:)

CFDD(")

CFA(1. )

CFAA(")

,(:)

CFD(1. ).
As the notation suggests, these dualizing bimodules correspond to the identity map
": 1 1. In fact, for any mapping class : 1 1 there are associated bimodules

CFDD() and

CFAA(); see [LOT10a].
A.2. Sketch of the construction
In this section we outline the construction of the bimodules discussed above. Details
can be found in [LOT10a], although the constructions are close enough to the construction
of

CFD and

CFA that the reader might nd it amusing to ll them in on his or her own.
We start by discussing bordered Heegaard diagrams representing (mapping cylinders of)
dieomorphisms (Section A.2.1). We then give the denitions of the bimodules

CFDA()
(Section A.2.2) and

CFDD(") and

CFAA(") (Section A.2.3).
A.2.1. Bordered Heegaard diagrams for dieomorphisms. The bimodule for a
dieomorphism ,

CFDA(), is associated to a bordered Heegaard diagram for . Such
a diagram is a slight generalization of the notion from Chapter 4 to allow more than one
boundary component. (Such diagrams already made a brief appearance in Section 11.4.)
Specically, an arced bordered Heegaard diagram with two boundary components is a
quadruple 1 = (. . . z) where
is an oriented surface of genus p, with two boundary components
L
and
R
;
=
a,L
1
. . . . .
a,L
2k
.
a,R
1
. . . . .
a,R
2k
.
c
1
. . . . .
c
g2k
where the
a,L
i
are embedded arcs
with endpoints on
L
, the
a,R
i
are embedded arcs with endpoints on
R
, and the

c
i
are embedded circles in ; and all of the
a,L
i
,
a,R
i
and
c
i
are pairwise disjoint,
and is connected;
=
1
. . . . .
g
are circles in such that is connected; and
z is an arc in ( ) connecting
L
and
R
.
As for ordinary bordered Heegaard diagrams, we let = ; we will be sloppy about
the distinction between and .
The data 1 species a 3-manifold 1 with two boundary components in a manner exactly
analogous to Section 4.1. Moreover, setting :
i
to be the pointed matched circle specied by

i
(i 1. 1), the diagram 1 species a homeomorphism
i
from 1
i
= 1(:
i
) to
i
1 , as
well as a framed arc connecting
L
1 and
R
1 . We call 1 a bordered Heegaard diagram for
(1.
L
.
R
).
Assume that 1 is homeomorphic to [0. 1] 1
R
. There is a homeomorphism : [0. 1]
1
R

=
1 extending
R
: 1
R
1 . Moreover, this homeomorphism is unique up to homeo-
morphisms : [0. 1] 1
R
[0. 1] 1
R
such that [
1F
R
is the identity. It follows that
A.2. SKETCH OF THE CONSTRUCTION 237
Figure A.1. Bordered Heegaard diagrams for dieomorphisms. Left:
the identity bordered Heegaard diagram for the genus 1 surface. Center: the
same diagram, with a regular neighborhood of
a,L
1

a,L
2k

L
z in
dark gray and a closed curve as a dashed line. Right: a bordered Heegaard
diagram for the Dehn twist around the curve .
[
0F
R
is well-dened up to homotopy, and hence (because 1
R
is a surface) up to iso-
topy. (Note that [
0F
R
is orientation-reversing.) Consequently, ([
0F
R
)
1

L
is an
orientation-reversing homeomorphism 1
L
1
R
, well-dened up to isotopy. If we assume
that 1
L
= 1
R
then ([
0F
R
)
1

L
can be regarded as an element (1) of the mapping
class group MCG(1
R
). Paying attention to the arc z, we obtain a map
: bordered Heegaard diagrams for ([0. 1] 1. 1. 1) MCG
0
(1).
where MCG
0
denotes the strongly based mapping class group of any surface 1 = 1(:);
here, strongly based refers to the mapping class group of dieomorphisms xing a small
disk in 1.
Every element of MCG
0
(1) arises this way. Indeed, given a dieomorphism : 1 1,
we can explicitly construct a bordered Heegaard diagram realizing as follows. Given a
pointed matched circle :, let "
:
= (
k
. . . z) be the bordered Heegaard diagram with

L
"
:
=
R
"
:
= :, with no -circles and where
a,L
i
and
a,R
i
each intersect
i
in a single
point and are disjoint from
j
for i ,= ,. (See Figure A.1.) Then "
:
is a bordered Heegaard
diagram for the identity map of 1(:). Further, a regular neighborhood of

a,L
1

a,L
2k

L

in is canonically homeomorphic to 1 minus two disks. By removing the arc z from this
neighborhood, we obtain 1 minus one disc. Under this identication, let : be
the result of extending MCG
0
(1) by the identity. Then, (. (). . z) is a bordered
Heegaard diagram for . (Equivalently, we could have applied
1
to .)
The hypothesis that 1
L
= 1
R
in the above construction is not crucial. In the more
general case (where 1
L
and 1
R
are dierent), one obtains Heegaard diagrams corresponding
to elements of a mapping class groupoid. See [LOT10a] for further details.
A.2.2. The denition of

CFDA. Let 1 = (
g
. . . z) be a bordered Heegaard di-
agram for MCG
0
(1
R
). The bimodule

CFDA() =

CFDA(1) is dened by counting
238 A. BIMODULES AND CHANGE OF FRAMING
holomorphic curves in [0. 1]1. As usual, we choose an almost complex structure so that
the two boundary components of become cylindrical ends of , and hence of [0. 1] 1.
We call the end corresponding to
L
the left boundary and the end corresponding to
R

right boundary.
The bimodule

CFDA(1) is dened by treating the left boundary as a type 1 end (i.e.,
analogously to Chapter 6) and the right boundary a Type end (i.e., analogously to Chap-
ter 7).
More precisely, let S(1) denote the set of p-tuples of points x in so that exactly one
r
i
lies on each - and each -circle, and no two r
i
lie on the same -arc. Given x S(1),
let o
L
(x) denote the set of
L
-arcs occupied by x and o
R
(x) the set of
R
-arcs occupied by
x. Let A(1) be the F
2
vector space spanned by S(1). We dene left and right actions of
J(1
R
) on S(1) by
1(s) x 1(t) :=
_
x s = [2/] o
L
(x) and t = o
R
(x)
0 otherwise.
where s and t are subsets of [2/], as in Formula (3.22).
As a left module,

CFDA(1) is just /(1
L
)
7
A(1). Note that, for the rst time in this
book, we have a module on which /(:. i) may act non-trivially for i ,= 0.
Our next task is to dene the /

bimodule structure on

CFDA(1). (See, for instance,
[Kel01, Section 3] for basic notions relating to /

bimodules.) The left and right actions

strongly commute, i.e.,
:
i,1,j
(c
1
. . . . . c
i
. x. /
1
. . . . . /
j
) = 0
if i. , 0, it only remains to describe the right action. The bimodule

CFDA(1) is also
strictly unital, i.e.,
:
0,1,1
(x. I) = x
:
0,1,n
(x. . . . . I. . . . ) = 0. : 1.
The denition of the operations :
0,1,n
(x. c(
1
). . . . . c(
n
)) uses holomorphic curves. Let
/
B
(x. y;
D
;
A
) denote the moduli space of embedded holomorphic curves in the homol-
ogy class 1, asymptotic to the sequence of Reeb chords in
D
at west (with no height
constraints), and to the ordered partition
A
= (
A
1
. . . . .
A
n
) of Reeb chords at east . The
rest of the bimodule structure is determined by
:
0,1,n
(x. c(
1
). . . . . c(
n
)) :=

yS(7)

B
2
(x,y)

D
[ind(B;
D
;
A
)=1
_
#/
B
(x. y;
D
;
A
)
_
c(
D
)y.
(Compare Denitions 6.3 and 7.3. The case : = 0 is essentially the dierential on

CFD.)
We prove in [LOT10a] that this does, indeed, dene an /

bimodule. Invariancethe
rst part of Theorem A.1is proved similarly to invariance of

CFD or

CFA. Both of the
proofs of the Pairing Theorem (Chapters 8 and 9) extend easily to prove the second and
third parts of Theorem A.1. Again, this is explained in more detail in [LOT10a].
Remark A.3. Just as

CFD and

CFA have been graded by left or right G-sets, bimodules
are graded by sets with left and right actions by G. The bimodule

CFDA() is graded
by G(1
R
), thought of as a set with commuting left and right actions by G(1
R
)

= G(1
L
).
A.2. SKETCH OF THE CONSTRUCTION 239
The construction of the grading is a simple modication of the discussions in Sections 10.4
and 10.3.
A.2.3. The denition of

CFDD(") and

CFAA("). The bimodules

CFDD(") and

CFAA(") are both dened in terms of holomorphic curves with respect to the identity Hee-
gaard diagram "
:
= (. . . z). The bimodule

CFDD(") is dened by treating both bound-
ary components as Type 1 boundary components, and the bimodule

CFAA(") is dened by
treating both boundary components as Type boundary.
More precisely,

CFDD(") is dened as follows. Recall that : is the pointed matched
circle obtained by reversing the orientation on :. The bimodule

CFDD(") is generated
over /(:) /(:) by the 2
2k
=

i
_
2k
i
_
ways of choosing i points from
L,a
1

1
.
L,a
2

2
. . . . .
L,a
2k

2k
(and the 2/ i points on the complementary
R,a
i
); we can identify this
with the set S("
:
) from Section A.2.2 by looking at o
L
, the set of strands occupied on the
left. There are left actions of J(:) and J(:) on A("
:
) by, respectively,
1(s) x :=
_
x s = [2/] o
L
(x)
0 otherwise.
1(s) x :=
_
x s = [2/] o
R
(x)
0 otherwise.
(Note that o
R
(x) = [2/] o
L
(x).) These actions commute. Consequently, extending scalars
from the idempotents to the whole algebra gives commuting left actions of /(:) and /(:);
this is the bimodule structure on

CFDD(").
The dierential on

CFDD(") is given as follows. Let /
B
(x. y;
L
.
R
) denote the moduli
space of embedded holomorphic curves in the homology class 1, asymptotic to the sequence
of Reeb chords
L
on the left and
R
on the right. (We do not constrain the relative ordering
of Reeb chords on the left and right.) Then
(A.4) x :=

yS(7)

B
2
(x,y)

L
,
R
[ind(B;
L
;
R
)=1
#
_
/
B
(x. y;
L
;
R
)
_
c(
L
) c(
R
) y.
Next we turn to

CFAA("). As an F
2
-vector space, the bimodule

CFAA(") is just A("
:
).
There are commuting right actions of J(:) and J(:) on A("
:
) by
x 1(s) :=
_
x s = o
L
(x)
0 otherwise.
x 1(s) :=
_
x s = o
R
(x)
0 otherwise.
The bimodule

CFAA(") is strictly unital, i.e.,
:
1,1,0
(x; I; ) := :
1,0,1
(x; ; I) := x
:
1,m,n
(x; . . . . I. . . . ; . . . ) := :
1,m,n
(x; . . . ; . . . . I. . . . ) := 0. : + : 1.
240 A. BIMODULES AND CHANGE OF FRAMING
The other higher multiplications are dened using holomorphic curves. Let /
B
(x. y;
L
;
R
)
denote the moduli space of embedded holomorphic curves in the homology class 1, asymp-
totic to the ordered partition = (
1
. . . . .
m
) at (
L
) [0. 1] 1 and asymptotic to the
ordered partition = (
1
. . . . .
n
) at (
R
) [0. 1] 1. Then dene
:
1,m,n
(x; c(
1
). . . . . c(
m
); c(
1
). . . . . c(
n
)) :=

yS(7)

B
2
(x,y)
ind(B;;)=1
#
_
/
B
(x. y; ; )
_
y.
Theorem A.2 now follows from a version of the pairing theorem, which can be proved
using either the nice diagrams technique of Chapter 8 or the time dilation argument of
Chapter 9.
Remark A.5. The gradings on

CFAA(") and

CFDD(") are dened analogously to Sec-
tions 10.3 and 10.4 respectively. In the case of

CFAA(") (respectively

CFDD(")) the G-
grading set is G(:), viewed as having two right (respectively left) actions, one by G(:) and
one by G(:)

= G(:)
op
. In fact, both of these actions are the standard action of the group
on itself by translation.
Remark A.6. For each of

CFDA(1),

CFDD(") and

CFAA("), the summands /(:. i) inside
/(:) act non-trivially even for i ,= 0, and the bimodules decompose as direct sums

CFDA(1) =
k

i=k

CFDA(1. i)

CFDD(") =
k

i=k

CFDD(". i)

CFAA(") =
k

i=k

CFAA(". i).
Of course, only the summands corresponding to i = 0 contribute to tensor products with

CFD or

CFA as in Theorem A.1. The other summands do arise naturally if one studies
self-gluing of Heegaard diagrams, which in turn relates to the Floer homology of open books.
A.3. Computations for 3-manifolds with torus boundary
In this section we state the relevant bimodules for three-manifolds with torus boundary.
These are used in Section A.4 to deduce a version of Theorem 11.27 for arbitrary integral
framings of knots, Theorem A.11. (In fact, the data from this section can be used to calculate
the result for arbitrary rational framings; but that will be left to the interested reader.) The
computations of the bimodules are given in [LOT10a].
There is a unique pointed matched circle representing the surface of genus one. Let
/ = /(:. 0) denote the i = 0 summand of the corresponding algebra. Label the generators
of / as in Chapter 11.
A.3. COMPUTATIONS FOR 3-MANIFOLDS WITH TORUS BOUNDARY 241

1

1

3

3

1
: :
Heegaard diagram for
1

Heegaard diagram for

1

Heegaard diagram for

Heegaard diagram for

.
j
: j
:

j
:
j
. .

j
:

3

3

1
.
j
:
:
j

1
Figure A.2. Heegaard diagrams for mapping class group elements.
Genus 2 diagrams for

,
1

and
1

are shown. In each of the four

diagrams, there are three generators in the i = 0 summand.
A.3.1. Type DA invariants of Dehn twists. The mapping class group is generated
by Dehn twists

and

along a meridian and a longitude, respectively (i.e.,

takes an
:-framed knot complement to an : + 1-framed knot complement). Recall from Section 11.4
that
a
2
is a meridian. Thus,

corresponds to a Dehn twist around a curve j dual to the

curve
a
1
from Figure 11.1.
Most of this section is devoted to describing the i = 0 part

CFDA(. 0) of the type DA
bimodules for these Dehn twists and their inverses. First, we dispense with the i = 1
parts. The summand /(. 1) of /() is isomorphic to F
2
, and

CFDA(. 1)

= F
2
for any
mapping class . The summand /(. 1) of /() is 7-dimensional, but is quasi-isomorphic
to F
2
. Thus, /(. 1)-bimodules are determined (up to quasi-isomorphism) by corresponding
F
2
-bimodules. For any mapping class , the F
2
-bimodule corresponding to

CFDA(. 1) is,
again, F
2
.
Heegaard diagrams for

,
1

and
1

are illustrated in Figure A.2. Each of the

four bimodules

CFDA(

. 0),

CFDA(
1

. 0),

CFDA(

. 0) and

CFDA(
1

. 0) has the form

/
7
A where A is 3-dimensional over F
2
. The vector space A has a basis consisting of p,
q and either r or s. The generators are compatible with the idempotents as follows:
(A.7)
0
p
0
= p
1
q
1
= q
1
r
0
= r
0
s
1
= s.
As in Section A.2.2, the left module structure is induced from the isomorphism

CFDA

=
/
7
A. Higher products of the form :
i,1,j
(c
1
. . . . . c
i
. x. /
1
. . . . . /
j
) vanish if i, 0. So, it
remains to give the higher products of the form :
0,1,j
for , 0. We do this for each of the
four bimodules in turn; see also Figure A.3.
242 A. BIMODULES AND CHANGE OF FRAMING
The bimodule

CFDA(

. 0) is generated by p, q and r. The idempotents act as above;

the other non-trivial algebra actions are given as follows:
:
0,1,1
(p.
1
) =
1
q :
0,1,1
(p.
12
) =
123
r
:
0,1,1
(p.
123
) =
123
q :
0,1,2
(p.
3
.
2
) =
3
r
:
0,1,2
(p.
3
.
23
) =
3
q :
0,1,1
(q.
2
) =
23
r
:
0,1,1
(q.
23
) =
23
q :
0,1,0
(r) =
2
p
:
0,1,1
(r.
3
) = q.
The bimodule

CFDA(
1

. 0) is generated by p, q and r. The idempotents act as above;

the other non-trivial algebra actions are given as follows:
:
0,1,0
(p) =
3
r :
0,1,1
(p.
1
) =
1
q
:
0,1,1
(p.
12
) =
1
r :
0,1,1
(p.
123
) =
123
q
:
0,1,2
(p.
123
.
2
) =
12
p :
0,1,1
(q.
2
) = r
:
0,1,1
(q.
23
) =
23
q :
0,1,2
(q.
23
.
2
) =
2
p
:
0,1,1
(r.
3
) =
23
q :
0,1,2
(r.
3
.
2
) =
2
p.
The bimodule

CFDA(

. 0) is generated by p, q and s. The idempotents act as above;

the other non-trivial algebra actions are given as follows:
:
0,1,2
(q.
2
.
1
) =
2
s :
0,1,2
(q.
2
.
12
) =
2
p
:
0,1,2
(q.
2
.
123
) =
23
q :
0,1,1
(p.
1
) =
12
s
:
0,1,1
(p.
12
) =
12
p :
0,1,1
(p.
123
) =
123
q
:
0,1,1
(p.
3
) =
3
q :
0,1,0
(s) =
1
q
:
0,1,1
(s.
2
) = p :
0,1,1
(s.
23
) =
3
q.
The bimodule

CFDA(
1

. 0) is generated by p, q and s. The idempotents act as above;

the other non-trivial algebra actions are given as follows:
:
0,1,0
(q) =
2
s :
0,1,1
(p.
1
) = s
:
0,1,1
(p.
12
) =
12
p :
0,1,1
(p.
123
) =
123
q
:
0,1,1
(p.
3
) =
3
q :
0,1,2
(p.
12
.
1
) =
1
q
:
0,1,1
(s.
2
) =
12
p :
0,1,1
(s.
23
) =
123
q
:
0,1,2
(s.
2
.
1
) =
1
q.
Remark A.8. The computations leading to the results above are eased by noting that

CFDA(
1

. 0) is inverse to

CFDA(

. 0), i.e.,

CFDA(
1

. 0)

CFDA(

. 0)

CFDA(". 0);
the Heegaard diagram for

is obtained from that for

1

by a horizontal reection
(across the r-axis); and
the Heegaard diagram for
1

is obtained from that for

by a horizontal reection.
Thus there is essentially only one computation to do.
A.3. COMPUTATIONS FOR 3-MANIFOLDS WITH TORUS BOUNDARY 243
p q
r

1
+
123

123
+
3
(
3
,
23
)

1
2
3

1
2
+

3
,

2
)

2
3

23

23

1
1

3
p q
r

1
+
123

123

12
(
123
,
2
)

1
+

1
2

2
(
23
,
2
)
1

2

23

23

3
,

2
)

2
3

3
q p
s

2
(
2
,
12
)

2
,

1
)

23
(
2
,
123
)

1
2

1

12

12

3
+
123

123

1
+

2
3
1

2
q p
s

12

12
1

3
+
123

123
+
1
(
12
,
1
)

2
,

1
)
+

1
2
3

2
3

1
2

2
Figure A.3. Type DA bimodules for torus mapping class group action.
These are the bimodules associated to

,
1

,
1

respectively. The no-

tation is as follows. Consider the bimodule for

. The label
1

1
on the
horizontal arrow indicates that :
0,1,1
(p.
1
) contains a term of the form
1
q.
Similarly, the label
3
(
3
.
23
) on that arrow indicates that :
0,1,2
(p.
3
.
23
)
contains a term of the form
3
q.
Remark A.9. As the reader may well have expected, up to homotopy equivalence the type
DA bimodule

CFDA(". 0) associated to the identity map is isomorphic to / as an //
bimodule.
A.3.2. The type AA invariant of the identity. To compute the type AA and DD
identity bimodules for the torus, it is convenient to use the Heegaard diagram pictured
in Figure A.4. Note that the boundary segments on both sides have been labeled to be
consistent with the type conventions.
244 A. BIMODULES AND CHANGE OF FRAMING
d
c
/
c
c
2
.
c
1

2
)
1
)
2
Figure A.4. A Heegaard diagram for the genus 1 identity map. This
diagram has four more intersection points in than the standard Hee-
gaard diagram for the identity map, introduced by two nger moves. As a
result of these nger moves, the diagram is admissible, not merely provincially
admissible, and is useful for computing

CFAA(").
The resulting type AA bimodule has six generators, denoted /)
1
, /)
2
, c
1
c, c
2
c, cc and /d.
Its non-trivial /

-operations are given by:

:
1,1,0
(/)
1
;
1
; ) = cc :
1,0,0
(/)
1
; ; ) = /)
2
:
1,1,1
(/)
1
;
12
;
2
) = /d :
1,1,1
(/)
1
;
12
;
23
) = /)
2
:
1,1,1
(/)
1
;
123
;
2
) = c
2
c :
1,3,1
(/)
1
;
3
.
2
.
1
;
2
) = c
2
c
:
1,0,0
(c
1
c; ; ) = c
2
c :
1,0,1
(c
1
c; ;
1
) = cc
:
1,1,1
(c
1
c;
2
;
12
) = /d :
1,1,1
(c
1
c;
23
;
12
) = c
2
c
:
1,1,1
(c
1
c;
2
;
123
) = /)
2
:
1,1,1
(cc;
2
;
23
) = /)
2
:
1,1,1
(cc;
2
;
2
) = /d :
1,1,1
(cc;
23
;
2
) = c
2
c
:
1,0,1
(/d; ;
3
) = /)
2
:
1,1,0
(/d;
3
; ) = c
2
c.
Here, to avoid confusion, we denote algebra elements acting on one side (which we think of
as /(:. 0)) by
i
while on the other side (which we think of as /(:. 0)) they are denoted
by
i
(as indicated in Figure A.4).
Remark A.10. The type AA identity bimodule of the torus has a model which is 2-
dimensional over F
2
. However, this model has innitely many non-trivial higher products.
These can be described explicitly; see [LOT10c, Section 8.4].
A.3.3. The type DD invariant of the identity. The type DD bimodule can also
be calculated using the same Heegaard diagram. The result, which is given graphically in
A.4. FROM HFK TO

CFD FOR ARBITRARY INTEGRAL FRAMINGS 245
/)
1
c
1
c
cc
/d
/)
2
c
2
c

1
(

2
,

2
)

3
1
+
(

1
2
,

2
3
)
1
+
(

2
3
,

1
2
)
(

1
2
,

2
)
(

2
,

2
3
)
(

2
,

1
2
)
(

2
3
,

2
)
(

1
2
3
,

2
)
+
(

3
,

2
,

1
,

2
)
(

2
,

1
2
3
)
Figure A.5. The type AA bimodule

CFAA(". 0). The labels on the arrows
indicate the /

operations; for instance, the label (

23
.
2
) on the arrow from
cc to c
2
c means that :
1,1,1
(cc.
23
.
2
) contains a term c
2
c.
Figure A.6, is given as follows:
/)
1
= (
3
1) cc + /)
2
c
1
c = (1
3
) cc + c
2
c + (
2

123
) /)
2
cc = (
2

2
) /d /d = (1
1
) /)
2
+ (
1
1) c
2
c
/)
2
= 0 c
2
c = 0.
(We have relabeled the
i
and
i
to conform to our conventions for type 1 boundary.)
This simplies to a bimodule with two generators r and with
r = (
1

3
+
3

1
+
123

123
) = (
2

2
) r.
A.4. From HFK to

CFD for arbitrary integral framings
We turn now to the generalization of Theorem 11.27 for arbitrary (integral) framings.
Theorem A.11. Let 1 o
3
be a knot and let CFK

(1) be a reduced model for the knot

Floer complex of 1. Let 1 be the bordered three-manifold o
3
nbd 1, given any integral
framing :. The associated type 1 module

CFD(1 ) can be extracted from CFK

(1) using
the procedure described in Theorem 11.27, except for the unstable chain, whose precise form
depends on the framing parameter :. There are three cases for the form of the unstable
chain. When : < 2(1), the unstable chain has the form

0
D
1
j
1
D
23
j
2
D
23

D
23
j
m
D
3

0
.
where : = 2(1) :. When : = 2(1), the unstable chain has the form

0
D
12

0
.
246 A. BIMODULES AND CHANGE OF FRAMING
/)
1
c
1
c
cc
/d
/)
2
c
2
c

3

3

1
1
1 1

1
2
3
r

3
+
3

1
+
123

123

2
Figure A.6. The type DD bimodule

CFDD(". 0). The labels on the
arrows indicate dierentials; for instance, the arrow from cc to /d signies
a term (
2

2
) /d in cc. The form on the left comes from the Heegaard
diagram in Figure A.4, and is bounded in an analogous sense to Denition 2.22.
The form on the right is homotopy equivalent, but no longer bounded.
Finally, when : 2(1) the unstable chain has the form

0
D
123
j
1
D
23
j
2

D
23
j
m
D
2

0
.
where : = : 2(1).
(The reader is warned that : here has the opposite sign from Theorem 11.27.)
Proof. In view of Theorem A.1, we need only see what happens as we iteratively tensor
the 1 module for a knot complement with suciently small framing parameter, as calculated
in Theorem 11.27, with the bimodule

CFDA(

. 0).
We denote the generators in the idempotent
0
for

CFD(1 ) by r
i
, and the generators
in idempotent
1
by
j
. In particular, we write r
0
and r
1
for the elements
0
and
0
in the
statement of the above theorem.
In the tensor product, each generator r
i
gives rise to a pair of generators p r
i
and
r r
i
, while each generator
j
gives rise to a generator q
j
. Note that the generators
p r
i
are the generators in the
0
-idempotent for the tensor product.
If we think of the r
i
as the generators in a horizontally-simplied basis in the sense of
Denition 11.25, then if there are 2/ + 1 basis elements, / of them are initial points of
horizontal arrows. For each initial point of a horizontal arrow r
i
, we can cancel r r
i
with
some corresponding element of the form q
1
; i.e., if we have a chain
r
i
D
3

1
D
23

D
23

D
2
r
j
;
A.4. FROM HFK TO

CFD FOR ARBITRARY INTEGRAL FRAMINGS 247
this is transformed to
p r
i
D
3
q
2
D
23
q
3
D
23

D
23
q

D
23
r r
j
D
2

D
23

_
D
2
r r
i
D

q
1
p r
j
.
Setting
t
i
= q
i+1
for i = 1 . . . / 1 and
t

= r r
j
and dropping the canceling pair r r
i
and q
1
, our new chain has the same form as the original chain.
Vertical chains are taken to chains of the same form (without any need for cancellation).
Finally, we must observe the behavior of the unstable chain under tensor product. To
this end, recall the generator r
0
which is neither the initial nor the nal intersection point
of a horizontal arrow. There is then a chain of the form
r
0
D
3

1
D
23

2
D
23

3
D
23

D
23

m
D
1
r
1
with : 1. Under the tensor product, canceling a dierential from r r
0
to q
1
as
before, this chain is carried to a similar chain from with length one less, from p r
0
to
p r
1
.
In the case where : = 1, we have the chain
r
0
D
3

1
D
1
r
1
.
This chain is transformed to another conguration where we can once again cancel the pair
r r
0
and q
1
, to get the chain
p r
0
D
12
p r
1
.
Now if we start with a chain of the form
r
0
D
12
r
1
.
this in turn is transformed into
p r
0
D
2
r r
0
D
123
p r
1
D
2
r r
1
.
The arrow from r r
1
should be viewed as part of another chain.
The remaining elements to lie in an unstable chain of the form
r
0
D
2

1
D
23

2

D
23

m
D
123
r
1
with : = 1. Now it is easy to see inductively that tensoring with our bimodule takes such
an unstable chain with parameter : to one with parameter : + 1.
The precise relationship between the length (and type) of the unstable chain, the framing
parameter, and follows as in the proof of Theorem 11.27. (Indeed, it is a formal conse-
quence of the above proof, together with the relationship between these three quantities for
suciently large framing parameter as established in Theorem 11.27).
Index of Notation
General 249
/

structures 249
Algebra associated to a pointed matched circle 250
Gradings 251
Heegaard diagrams 252
Moduli spaces 253
Type D modules 254
Invariance of

CFD 255
Type A modules 256
The pairing theorem 256
Manifolds with torus boundary 257
Bimodules and change of framing 258
General
" Identity morphism. 13
Orientation reversal. 57
[n] The set 1, . . . , n. 29
The symmetric dierence of two sets; S T := (S T) (T S). 41
/

structures
/ An /

algebra. Also used for algebra associated to a surface. 10

A The k-module underlying the /

algebra /. 10
A
i
The i
th
tensor power of A. 10
D Dierential on bar (tensor) algebra of an /

algebra. 11

1
Structure map of type D structure. 16

k
Iterate of structure map
1
on a type D structure. 18
Sum of all
k
for a type D structure. Also used for connecting homomorphism on
H

. 18

k
Analogue of
k
for a homomorphism of type D structures. 19
Analogue of for a homomorphism of type D structures. 19
f
i
A morphism of /

modules. 13
k Ground ring, a sum of F
2
s. 9
/ An /

module. 11
M The k-module underlying an /

module /. 11
/ Bar resolution of /. 15
M Module underlying bar resolution of /. 15

i
The (higher) products on an /

algebra. 10
m
i
The (higher) products on an /

module. 11
m Dierential on bar resolution of an /

module / gotten by considering all m

i
at
once. 12
(N,
1
) A type D structure. 17
A dg (resp. /

) module associated to a type D structure N over a dg (resp. /

) algebra.
17, 20
T

(A) Tensor (bar) algebra on A. 10

T
+
(A) Augmentation ideal in tensor algebra T

(A). 15
249
250 INDEX OF NOTATION

The /

tensor product. 15
f

" The /

tensor product of f with the identity map. 16

Box tensor product, a smaller model for /

tensor product. 21

Dierential on . 21
"
N
Box tensor product of a homomorphism with the identity on N. 22
Algebra associated to a pointed matched circle
a n points in an oriented circle. 34
/(n, k) Strands algebra on n places and k strands. 29
/(n) Strands algebra on n places, /(n) =

k
/(n, k). 30
/(:) Algebra associated to pointed matched circle :. /(:) /(4k). 35
/(:, i) Summand of /(:) with weight i. /(:, i) /(4k, k +i). 36
a
0
() Element of /(n) associated to consistent set of Reeb chords . 32
a() Element of /(:) associated to set of Reeb chords. 35
a( ) Product a(
1
) a(
n
) where = (
1
, . . . ,
n
). 100
a( ) Product a(
1
) a(
n
) where = (
1
, . . . ,
n
). 100
F(:) Surface associated to pointed matched circle :. 34
I(S) Idempotent in /(n, k) associated to set S [n], [S[ = k. 29
I(s) Idempotent of /(:) associated to set s of matched pairs. 35
J(n, k) Ring of idempotents in /(n, k). 29
J(:) Ring of idempotents of /(:). 35
J(:, i) Summand of J(:) with weight i. 36
I Unit of /(:). 35, 134
inv(a) Number of inversions of element a of /(n, k). 29
Inv() Set of inversions of partial permutation . 29
inv() Number of inversions of partial permutation . 29
< Ordering on a induced by orientation on Z and basepoint z. 34
M Matching function of a matched circle. 34
_
1 2
5 3
_
Consistent set of Reeb chords [1, 5], [2, 3]. 32
_
1 4
6

Element of /(:) shown in Equation (3.24). 36

1 2 4
5 3 4
_
Element of /(n, k) given by partial permutation 1 5, 2 3, 4 4. 29
1 1
2 2
3 3
4 4
5 5
Graphical notation for the algebra element

1 2 4
5 3 4
_
. 30

Result of resolving crossing in corresponding to . 29

A Reeb chord in Z. 32

Initial endpoint of Reeb chord . 32

Initial endpoint of chord being glued to left of . 32

+
Terminal endpoint of Reeb chord . 32

++
Terminal endpoint of chord being glued to right of . 32
A set of Reeb chords. 32

Initial endpoints of set of Reeb chords . 32

+
Terminal endpoints of set of Reeb chords . 32
A sequence (
1
, . . . ,
n
) of sets of Reeb chords. 84
(,

) Concatenation of the sequences and

. 95

[i,j]
Subsequence (
i
, . . . ,
j
) of . 84
Sequence obtained by reversing the orientation of each Reeb chord in (but keeping
the same order). 101
A sequence (
1
, . . . ,
n
) of Reeb chords or the associated sequence (
1
, . . . ,
n
) of
singleton sets of Reeb chords. 100
The sequence (
1
, . . . ,
n
). 101
Join of abutting Reeb chords. 32
Join of composable sets of Reeb chords. 32
INDEX OF NOTATION 251
Z An oriented circle. 32
: Pointed matched circle. 34
z Basepoint for pointed matched circle. 34
Gradings
[a] Homology class in H
1
(Z

, a) of algebra element a. 40
Element of H
1
(F(:)) associated to appropriate element of H
1
(Z

, a). 42
A
g
Part of A in grading g. 25
div(c
1
(s)) Divisibility of c
1
(s). 187
Connecting homomorphism H
1
(Z

, a) H
0
(a). 39
() One quarter the number of parity changes in . 39, 175
G A grading group. 25
G

(n) Big grading group on n positions; Z G

(n) H
1
(Z

, a). 39, 175

G(:) Small grading group for :; Z G(n) H
1
(F(:)). 41, 184
g

(B) (e(B) n
x
(B) n
y
(B),

(B)). Also given by

ind(B,)

i=1
gr

(a(
i
)) if (B, ) is
compatible. 176, 179
(G, ) Group with distinguished central element. 25
gr Grading on group-graded dierential algebra or module.
gr G(:)-grading on /(:); or G(:)-set grading on

CFA or

CFD. 42, 185, 186
gr

Grading in G

(n) of element of /(:); or G

(4k)-set grading on

CFA or

CFD. 41,
176, 180, 183
gr

(4k)-set grading on

CFA or

CFD. 179, 183
gr

( )

n
i=1
gr

(a(
i
)). 176
gr

()

n
i=1
gr

(a(
i
)). 176
gr G(:)-set grading on

CFA or

CFD. 185, 186
gr(mn) Grading on tensor product of modules. 26
gr

Grading on

CFA

CFD. 187
g(B) (I
A
(x))g

(B)(I
A
(y))
1
. 185
G
s
Stabilizer of s. 25
G
s
G Right cosets of G
s
. 25
(a) Maslov component of grading of algebra element a. 40, 90, 176
() Shorthand for (a()). Given explicitly by

{1,2}
[L(
1
,
2
)[
||
2
. 91, 176
() The Maslov component of gr

(a(
1
)) gr

(a(

)), i.e.,

i
(
i
) +

i<j
L(
i
,
j
).
91, 176
(j, ) Element of G

(n); j
1
2
Z is the Maslov component; H
1
(Z, a) is the spin
c
compo-
nent. 39, 175
(j
1
,
1
) (j
2
,
2
) = (j
1
+j
2
+L(
1
,
2
),
1
+
2
).. 175
Distinguished central element of a grading group or, in particular, of G

(n) or G

(:).
25, 39
< Partial order on a grading group or set given by a < b if b =
n
a for some n N. 26,
41
L(
1
,
2
) m(
2
,
1
). 39, 90, 175
L(
1
,
2
) L([
1
], [
2
]). 91
L(
1
,
2
) L([
1
], [
2
]). 91
m(, p) Multiplicity of p in . 39
() Projection G/R(P
2
(x
2
))
_
s spin
c
(Y )

s[
Yi

= s
i
, i = 1, 2
_
gotten by modding-out
Z-action. 187
P(x) Subgroup of G(:) given by g(B) [ B
2
(x, x). 185
P

(x) Subgroup of G

(4k) given by g

(B) [ B
2
(x, x). 178
S
G
T Fiber product of G-sets. 26
(s) = (I(s)) Elements of G

(4k) used to rene G

(4k)-grading to G(:)-grading. 42, 184

r Identity map Z Z (orientation-reversing). 182
252 INDEX OF NOTATION
R Map G

(4k) G

(4k) or G(:) G(:) given by R(j, ) = (j, r

()). 182
S

A
G

(4k)-grading set for

CFA. 180
S
A
G(:)-grading set for

CFA. 185
S
D
G(:)-grading set for

CFD. 186
V [n] Shift functor applied to V (Z-graded or G-graded sense). 9, 25
Heegaard diagrams
Attaching circles and arcs in for a bordered Heegaard diagram corresponding to
index 1 critical points. 46
Curves in non-compact surface corresponding to . 51

a
i
The i
th
curve in
a
. 46

a
Attaching arcs among . 46

a
Curves in corresponding to
a
. 51, 62

c
Attaching circles among . 46

c
i
The i
th
curve in
c
. 46
a
SW
, a
SE
, a
NW
, a
NE
Small translates in ( ) of a . 94
a b Intersection number or jittered intersection number of a and b. 94
B A homology class in
2
(x, y). 53, 59

(B) Portion of the boundary of B contained in . 54

(B) Portion of the boundary of B contained in . 54

(B) Portion of the boundary of B contained in . 54

Attaching circles in a closed or bordered Heegaard diagram corresponding to index 2
critical points. 45, 46

i
The i
th
curve in . 46
[B] Element of H
2
(Y ) represented by B
2
(x, x). 54
Concatenation of homology classes corresponding to addition of domains. 53, 54
1 Pointed matched circle appearing at the boundary of a bordered Heegaard diagram 1
(Lemma 4.4). 46
D
z
Region (component of ( )) adjacent to z . 53, 145
e Formal variable used in twisted coecients. 130
f A Morse function. 34, 47
g A Riemannian metric. 34, 47
S(1) The set of generators for the Heegaard diagram 1. 52
S(1, s) The set of generators for the Heegaard diagram 1 in the spin
c
-structure s. 55, 130
S(, , ) Same as S(1) for 1 = (, , , z). 52
S(1, x
0
) The set of generators for the twisted invariants for the Heegaard diagram 1 based at
the generator x
0
. 130, 143

Gluing 3-manifolds, Heegaard diagrams, etc. along their boundaries. 57

Operation of gluing homology classes; also used for pregluing holomorphic curves. 58
1 A bordered Heegaard diagram or pointed Heegaard diagram. 45, 46
n
x
(B) Average multiplicity of B near x. 91
n
x
(B)

xix
n
xi
(B). 91
o(x) Set of occupied -arcs in the generator x. 52, 185
o(s, ) Multi-set of -arcs obtained from multi-set s by deleting initial points of , adding
terminal points of . 84
o(s,
[1,i]
) Same as o(s, ) but for initial segment (
1
, . . . ,
i
) of . 84

2
(x, y) Set of homology classes from x to y (with local multiplicity 0 at z). 53

2
(x, y) Set of homology classes from x to y with possibly non-zero local multiplicity at z;

2
(x, y)

= Z
2
(x, y). 176, 196

2
(x, y) Set of provincial homology classes from x to y. 54
Compact Heegaard surface with one boundary component. 46
Result of attaching a cylindrical end to . 51, 62

e
Closed Riemann surface obtained by lling puncture of . 62, 74
INDEX OF NOTATION 253
(x, y) Obstruction to connecting x and y by a homology class; equal to s
z
(y) s
z
(x). 55,
189
s
rel
z
(x) Relative spin
c
-structure on (Y, Y ) induced by the generator x and basepoint z. 55
s
z
(x) spin
c
-structure on Y induced by the generator x and basepoint z. 55
x A generator, a set of g points in , one on each circle and at most one on each arc.
52
x
i
An element of the generator x. 52
X(1) F
2
-vector space spanned by S(1). 99
X(1, s) F
2
-vector space spanned by S(1, s). 99
X(1, x
0
) F
2
-vector space spanned by S(1, x
0
). 130
(Y, :, ) A bordered 3-manifold. 46, 235
Y (1) The bordered 3-manifold specied by a bordered Heegaard diagram 1. 47
z Basepoint on for a bordered Heegaard diagram or basepoint in for a closed
Heegaard diagram. 45, 46
Moduli spaces

emb
(B, ) Euler characteristic of embedded curve in homology class B with asymptotics , given
by
emb
(B, ) = g +e(B) n
x
(B) n
y
(B) (). 91

P
Partial diagonal x
p
= x
q
[ P
i
P, p, q P
i
associated to partition P. 64
e(B) Euler measure of domain B. 64
ev Evaluation map /
B
(x, y; S

;

P) 1
E
/1. 64, 66
ev
q
Evaluation map at the puncture q, a map

/
B
(x, y; S

;

P) 1 or

A
B
(T

) 1.
64, 66, 67
ev
e
Evaluation map at the east punctures of T

, a map

A
B
(T

) 1
E(T

)
. 67
ev
p,q
Dierence ev
p
ev
q
. 66
ev
p,q
Extension of ev
p,q
to /
B
(x, y; S

). 72
ev
w
Evaluation map at the west punctures of T

, a map

A
B
(T

) 1
W(T

)
. 67
ind(B, ) Expected dimension of moduli space of embedded curves in homology class B with
asymptotics ; ind(B, ) = e(B) +n
x
(B) +n
y
(B) +[[ +(). 91, 176
ind(B, S

, P) Expected dimension of

/
B
(x, y; S

; P). 64
e East innity: end of [0, 1] 1 at the puncture of . 59
End as t of [0, 1] 1. 59
+ End as t of [0, 1] 1. 59
j
D
Standard complex structure on |. 62
j

Complex structure on (xed). 62

/
B
(x, y; ) Open moduli space of embedded curves in homology class B with asymptotics x, y
and ; counts of these moduli spaces are used to dene

CFD and

CFA. 91
/
B
(x, y; ) Compactied moduli space of embedded curves in homology class B with asymptotics
x, y and . 91

/
B
(x, y; S

) Moduli space of curves with decorated source S

asymptotic to x and y. 63
/
B
(x, y; S

) Space of holomorphic combs with preglued source S

connecting x to y in homology
class B. Nave compactication of /
B
(x, y; S

). 72
/
B
(x, y; S

) Closure of /
B
(x, y; S

) in /
B
(x, y; S

). 72
/
B
(x, y; S

) /
B
(x, y; S

; P) /
B
(x, y; S

; P). 81

/
B
(x, y; S

; P) Moduli space of curves with decorated source S

asymptotic to x at and y at
+, with partition of e punctures P. 64
/
B
(x, y; S

; P) Quotient

/
B
(x, y; S

; P)/1. 66
/
B
(x, y; S

; P) Space of holomorphic combs with preglued source S

connecting x to y in homol-
ogy class B, respecting partition P of east punctures. Nave compactication of
/
B
(x, y; S

; P). 72
/
B
(x, y; S

; P) Closure of /
B
(x, y; S

; P) in /
B
(x, y; S

; P). 72
254 INDEX OF NOTATION

/
B
(x, y; S

;

P) Moduli space of curves with decorated source S

asymptotic to x at and y at
+, with ordered partition

P of e punctures. 59, 66
/
B
(x, y; S

;

P) Quotient

/
B
(x, y; S

;

P)/1. 59, 66
/
B
(x, y; S

;

P) Compactication of the space /
B
(x, y; S

;

P), via holomorphic combs; the closure
of /
B
(x, y; S

;

P) in /
B
(x, y; S

;

P). 60, 72

A(T

) Moduli space of curves in Z 1 [0, 1] 1 with bi-decorated source T

. 67
A(T

) Quotient

A(T

)/(1 1). 68

A(T

; P
w
, P
e
) Subspace of

A(T

) with partitions P
w
/ P
e
of west / east punctures. 67

A(T

; P
e
) Special case of

A(T

; P
w
, P
e
) where P
w
is the discrete partition. 67
p Puncture of corresponding to . 62
Identication of neighborhood U
p
of p with S
1
(0, ). 62
U
p
Neighborhood of puncture p of . 62
P A partition of the e punctures of a decorated source S

. 64
[P] Multi-sets of Reeb chords associated to partition P of east punctures. 66
[P[ Number of parts of partition P. 64

P An ordered partition of e punctures. 59, 66

[

P] Sequence of multi-sets of Reeb chords associated to ordered partition

P of east
punctures. 66
P
i
A set of e punctures; a part in a partition or ordered partition. 59, 64
P
e
A partition of the e punctures of a bi-decorated source T

. 67
P
w
A partition of the w punctures of a bi-decorated source T

. 67

D
Projection [0, 1] 1 [0, 1] 1. 62

Projection [0, 1] 1 . 62
Pregluing of holomorphic curves; also used for gluing homology classes. 71, 150
1
E
/1 Quotient of 1
E
by simultaneous translation; target space for evaluation maps. 66

i
A Reeb chord. 59
s Coordinate on [0, 1]. 62
S

A decorated source. 59, 62

E(S

) Set of east punctures of S

. 64
E Same as E(S

). Used when S

is clear from context. 64

S Riemann surface underlying a decorated source S

. 59
T

A bi-decorated source. 66
E(T

) Set of east punctures of T

. 67
E Same as E(T

). Used when T

is clear from context. 67

T Riemann surface underlying a bi-decorated source T

. 67
W(T

) Set of west punctures of T

. 67
W Same as W(T

). Used when T

is clear from context. 67

D
Standard symplectic form on |. 62

Symplectic form on (xed). 62

t Coordinate on 1. 59, 62
t
i
Height (t-coordinate) of Reeb chord
i
. 59

R
Translation of [0, 1] 1 by R 1. 66
u A holomorphic curve in [0, 1] 1. 63
[u] Homology class of u. 63
U A holomorphic comb. 71
(u, v) A simple holomorphic comb. 70
v A holomorphic curve in Z 1 [0, 1] 1. 67
Type D modules
a
B
x,y
Coecient of y in x coming from homology class B. 100
a
x,y
Coecient of y in x; the sum over B of a
B
x,y
. 103

CFD(Y ) Type D invariant

CFD(1) associated to any bordered Heegaard diagram 1. 130
INDEX OF NOTATION 255

CFD(Y, s) Summand of

CFD(Y ) corresponding to spin
c
-structure s. 130

CFD(1) Type D invariant associated in Section 6.1 to bordered Heegaard diagram 1. 100

CFD(1, s) Summand of

CFD(1) corresponding to spin
c
-structure s. 99

CFD(1, x
0
) Totally twisted variant of

CFD(1), with base generator x
0
. 130
I
D
(x) Idempotent for

CFD for generator x; I
D
(x) = I([2k] o(x)). 99, 185
Invariance of

CFD

H
-curves gotten by handlesliding
a
1
over
c
1
and performing an appropriate isotopy.
109
a
H

H
. 109

a,H
i
i
th
arc in
H
; for i 2 a small perturbation of
a
i
intersecting
a
i
in one point; for
i = 1, a handleslide of
a
1
over
c
1
, appropriately perturbed. 109

c,H
i
i
th
circle in
H
; a small perturbation of
c
i
intersecting
c
i
in two points. 109

Appropriately chosen small perturbation of . 126

a,
i
i
th
arc in

, a small deformation of
a
i
. 126

c,
i
i
th
circle in

, a small deformation of
c
i
. 126
b(B) Domain in
2
(x

, y) (in (,
H
, )) corresponding to domain B
2
(x, y,
o
), as in
Lemma 6.24. 112

CFD(1, s; J) Module

CFD(1, s) computed with respect to the almost complex structure J. 106
A triangle (disk with three boundary punctures). 109
e
1
, e
2
, e
3
Edges of . 109
v
12
, v
23
, v
13
Vertices of ; v
ij
is adjacent to e
i
and e
j
. 109
F
Jr
Continuation map associated to path J
r
of almost complex structures. 106, 107
F
,
H
,
Continuation map

CFD(,
H
, , z)

CFD(, , , z) for a handleslide gotten by
counting triangles. 109, 116
ind(B, T

, P, P
H
) Expected dimension of

/
B
(x, y,
o
; T

;

P,

P
H
). 111
/
B
(x, y; ; J) Open moduli space of embedded curves with respect to t-dependent complex structure
J in homology class B with asymptotics x, y and . 107
/
B
(x, y; S

; P; J) Moduli space of curves for t-dependent almost complex structure J with decorated
source S

asymptotic to x at and y at +, with partition of e punctures P.

107
/
B
(x, y,
o
; T

) Moduli space of holomorphic triangles with decorated source T

asymptotic to x at
v
31
, y at v
23
and
o
at v
12
. 111
/
B
(x, y, ; ) Moduli space of embedded holomorphic triangles asymptotic to x at v
31
, y at v
23
, some

o
at v
12
and along e
1
and e
2
. 116
/
B
(x, y, w

pd
; ,
H
) Moduli space of embedded holomorphic triangles asymptotic to partially-diagonal
generator w

pd
at v
12
, along e
1
and
H
along e
2
. 119
/
B
(x, y,
o
;
1
,
2
) Moduli space of embedded holomorphic triangles asymptotic to x at v
31
, y at v
23
,

o
at v
12
,
1
along e
1
and
2
along e
2
. 116
/
B
(x, y,
o
; T

;

P,

P
H
) Subspace of

/
B
(x, y,
o
; T

) with ordered partition of east punctures

P along
e
1
and

P
H
along e
2
. 111
/
B
(x, y,
o
; T

;

P,

P
H
) Naive compactication of /
B
(x, y,
o
; T

;

P,

P
H
) via combs. 111
/
B
(x, y, w

pd
; T

;

P,

P
H
) Moduli space of triangles with source T

asymptotic to partially diagonal

generator w

pd
at v
12
. 119
/
B
(x, y,
o
; T

;

P,

P
H
) Closure of /
B
(x, y,
o
; T

;

P,

P
H
) in /
B
(x, y,
o
; T

;

P,

P
H
). 111
/
B
(
o
, w
+
pd
) Moduli space of holomorphic curves in (,
H
, ). 118
/
B
(
o
, w
+
pd
; S

) Same as /
B
(
o
, w
+
pd
); used when we want to name the source. 118
Matching between partially diagonal generators. 119

2
(x, y,
o
) Set of homology classes for (, ,
H
, ) connecting x, y, and
o
. 111

(w
+
pd
, w

pd
) Sequence of -Reeb chords associated to partially diagonal generators w

pd
and match-
ing . 119
256 INDEX OF NOTATION

H(w
+
pd
, w

pd
) Sequence of
H
-Reeb chords associated to partially diagonal generators w

pd
and match-
ing . 119

ij
Operation of attaching a homology class at corner v
ij
of . 112

o
Top-graded generator for (,
H
, ) with idempotent I
D
(o). 111
T
x
Canonical small triangle homology class in
2
(x, x

,
o(x)
). 112
T

Decorated source for maps to . 110

w

pd
Positive / negative partially-diagonal generators. 118
x

Nearest generator in S(
H
, ) to x S(, ). 111
Type A modules
(x) Alternate notation for m
1
(x). 134

CFA(1) Type A invariant associated in Section 7.1 to bordered Heegaard diagram 1. 134

CFA(1, s) Summand of

CFA(1) corresponding to spin
c
-structure s. 133, 134

CFA(1, x
0
) Totally twisted variant of

CFA(1), with base generator x
0
. 143
f
Jr
n
/

chain homotopy equivalence on

CFA associated to change in almost complex struc-
ture. 139
I
A
(x) Idempotent for

CFA for generator x; I
A
(x) = I(o(x)). 133, 185
m
n+1
/

products on

CFA. 134
x a Alternate notation for m
2
(x, a). 134
The pairing theorem
1
i
= (
i
,
i
,
i
, z) Provincially admissible Heegaard diagrams with 1
1
= 1
2
. 150
, ,
1

2
,
1

2
and
1

2
, respectively. 150

T
(x
1
x
2
) T-deformed dierential, on

CF(T; 1
1
, 1
2
). 153
j
CF(T; 1
1
, 1
2
) Analogue of

CF(1
1

1
2
) computed using T-matched holomorphic curves. 153
S

The pregluing S

1
S

2
of sources S

i
for curves in 1
i
. 150
F
T1,T2
Chain homotopy equivalence

CF(T
1
; 1
1
, 1
2
)

CF(T
2
; 1
1
, 1
2
) from counting -
matched curves. 157
ind(B
1
, S

1
; B
2
, S

2
) Expected dimension of

//
B
0
(x
1
, y
1
, S

1
; x
2
, y
2
, S

2
). 150

//
B
(x
1
, y
1
, S

1
; x
2
, y
2
, S

2
) Matched moduli space,

/
B1
(x
1
, y
1
; S

1
)
ev1=ev2

/
B2
(x
2
, y
2
; S

2
). 150
//
B
(x
1
, y
1
, S

1
; x
2
, y
2
, S

2
) Quotient

//
B
(x
1
, y
1
, S

1
; x
2
, y
2
, S

2
)/1. 151
//
B
(x
1
, y
1
, S

1
; x
2
, y
2
, S

2
) Compactication of //
B
(x
1
, y
1
, S

1
; x
2
, y
2
, S

2
). 154

//
B
(x
1
, y
1
; x
2
, y
2
) Embedded matched moduli space, union of matched moduli spaces with (S

1

S

2
) =
emb
(B). 152
//
B
(x
1
, y
1
; x
2
, y
2
) Quotient

//
B
(x
1
, y
1
; x
2
, y
2
)/1. 152

//
B
(T; x
1
, y
1
, S

1
; x
2
, y
2
, S

2
) T-matched moduli space, the bered product

/
B1
(x
1
, y
1
; S

1
)
T ev1=ev2

/
B2
(x
2
, y
2
; S

2
). 153
//
B
(T; x
1
, y
1
, S

1
; x
2
, y
2
, S

2
) Quotient

//
B
(T; x
1
, y
1
, S

1
; x
2
, y
2
, S

2
)/1. 153

//
B
(T; x
1
, y
1
; x
2
, y
2
) Embedded T-matched moduli space, union of T-matched moduli spaces with
(S

1
S

2
) =
emb
(B). 153
//
B
(T; x
1
, y
1
; x
2
, y
2
) Quotient

//
B
(T; x
1
, y
1
; x
2
, y
2
)/1. 153
//
B
( T
0
; x
1
, y
1
, S

1
; x
2
, y
2
, S

2
) Union of T-matched moduli spaces for T T
0
. 167
//
B
(; x
1
, y
1
, S

1
; x
2
, y
2
, S

2
) Moduli space of -matched pairs. 157
//
B
(; x
1
, y
1
; x
2
, y
2
) Union of -matched moduli spaces with (S

1
S

2
) =
emb
(B). 157
//
B
tsic
(x
1
, y
1
; x
2
, y
2
) Moduli space of embedded trimmed simple ideal-matched curves, given by the
union

{S

1
,S

2
|(S

1
S

2
)=
emb
(B)}
//
B
tsic
(x
1
, y
1
, S

1
; x
2
, y
2
; S

2
). 165
//
B
tsic
(x
1
, y
1
, S

1
; x
2
, y
2
; S

2
) Moduli space of trimmed simple ideal-matched curves (modulo transla-
tion). 164
Correspondence between the punctures of U
1
and U
2
in an ideal-matched holomorphic
comb. 158
INDEX OF NOTATION 257
T Dilation parameter, a positive real number. 153
T(U) Time parameter space of a holomorphic comb U. 158
(t) Dilation function interpolating between T
1
and T
2
. Also used for grading renement
data. 157
Manifolds with torus boundary
A Alexander depth (ltration). 196

G(:) G(:) Z; the extra Z is the Alexander factor. 200

S
D

G-set for CFD

(1, z, w) given by

G/

P(x
0
). 200

P(x
0
) g(
2
(x
0
, x
0
))

G. 200
gr Enhanced grading on CFD in presence of knot given by g(B)

P(x
0
) for B

2
(x, x
0
). 200
g Enhanced grading of domains in presence of knot, (g(B), n
w
(B) n
z
(B))

G. 200
/() Algebra /(, 0) associated to the unique pointed matched circle for a torus in the
central (i = 0) spin
c
structure. 191

a
1
,
a
2
-arcs in a Heegaard diagram with torus boundary, labeled as in Section 11.1. 192
C
h
Horizontal complex of a Z-ltered, Z-graded chain complex over F
2
[U]. 203
C
v
Vertical complex of a Z-ltered, Z-graded chain complex over F
2
[U]. 202
D
I
Coecient map for type D structure over /() corresponding to I 1, 2, 3; or
generalized coecient maps for I 0, 1, 2, 3 a cyclic interval. 193, 204, 206, 210
S
K
Set of generators for

CFK(1
K
). 213
S(n) Set of generators for

CFD(1(n)). 213
G

(4), the unrened grading group for the torus algebra. 192
G G(), the rened grading group for the torus algebra. 193
(m; a, b, c) Element of G

. 192
(m; a, b) Element of G. 193
(
0
,
0
,
0
, z, w) Doubly-pointed Heegaard diagram for K S
3
. 212
1
K
:= (, , , z, w) Doubly-pointed Heegaard diagram for K from stabilizing (
0
,
0
,
0
, z, w). 212

g
New -circle in 1
K
, contained in the new handle. 212

g
New -circle in 1
K
meeting
g
in a single point. 212
1(n) := (,
a
1
,
a
2
,
1
, . . . ,
g1
, ) Bordered Heegaard diagram for S
3
nbd(K) with framing n, coming
from 1
K
. 212
1
1
, 1
0
, 1

Standard Heegaard diagrams for 1-, 0- and -framed solid tori, respectively. 195
S Function S(1)
1
2
Z characterized by Equations (11.39) and (11.42). Also used for
source of holomorphic curves. 213, 214
B
0
Generator for periodic domains in 1(n), with local multiplicities (0, 1, 1 n, n) in
regions (0, 1, 2, 3). 213
H
h
Horizontal complex of a Z-ltered, Z-graded chain complex over F
2
[U]. 203
H
v
Vertical complex of a Z-ltered, Z-graded chain complex over F
2
[U]. 202

0
,
1
Indecomposable idempotents in /(), labeled so that
0
corresponds to
a
1
occupied
on type A side (unoccupied on type D side). 192
CFK

(K) Knot Floer complex, a ltered complex generated over F

2
[U] by S
K
. 195
gCFK

(K) Associated graded complex to CFK

(K). 196

CFK(K) U = 0 specialization of CFK

(K), a ltered complex generated over F

2
by S
K
. 196

gCFK(K) Associated graded complex to

CFK(1). 196
C Shorthand for

gCFK(K). 209
C(r) Summand

gCFK(1, r) of

gCFK(K) in Alexander grading r. 209
C(s)

rs
C(r). 209
C(s)

rs
C(r). 209

i
Map C(r) C(r +i) dened by counting holomorphic disks with n
z
() = i and
n
w
() = 0 if i 0, and n
z
() = 0 and n
w
() = i if i 0. 209

i0

i
. 209
258 INDEX OF NOTATION

i0

i
. 209
S
K
Generators for knot Floer complex. 195
(, , , w, z) A doubly pointed Heegaard diagram for a knot in a 3-manifold. 195
HFK

(K) Knot Floer homology, homology of gCFK

(K). 196

HFK(K) Homology of

gCFK(K). 197
CFA

(1, z, w) Type A variant of bordered knot Floer complex. 199

CFA(1, z, w) U = 0 specialization of CFA

(1, z, w). 199

CFD

(1, z, w) Type D variant of bordered knot Floer complex. 199

CFD(1, z, w) U = 0 specialization of CFD

(1, z, w). 199

M Maslov grading on knot Floer complex. 196
N Renormalized Maslov grading on knot Floer complex. N = M 2A. 196
n The framing parameter in Section 11.7, a large positive integer. 209

k
Distinguished homology class in winding region connecting x
k
to x
k1
with multiplicity
+1 at regions 2 and 3. 215, 216

k
Distinguished homology class in winding region connecting x
k1
to x
k
with multiplicity
0 at regions 2 and 3. 217

1
,
2
,
3
,
12
,
23
,
123
Non-idempotent basic generators for /(). 192
The tau invariant from knot Floer homology. 197
U Formal variable used in knot Floer homology. 195
gr(Ux) (
2
, 1) gr(x). 200
V
1
s
C( s +
n1
2
) if s n/4; F
2
if [s[ < n/4; and C( s
n1
2
) if s n/4. 210
V
1

sZ+
n+1
2
V
1
s
. 209
V
0
s
C(s). 210
V
0

sZ
V
0
s
. 209
V V
0
V
1
. 210
W
i
Summand of

CFD(1) (for 1 as in Section 11.7) corresponding to idempotent i. 211

s
1
2
Z
W
i
s
Decomposition of W
i
according to a grading function S. 213
J The winding region. 212
x
i

i=
n
2
i=
n
2
Intersections between
g
and
a
1

a
2
, contained in winding region. 213
x
i

i=
n
2
i=
n
2
Generators of

CFD(1(n)) corresponding to generator x S
K
. 214
Bimodules and change of framing
1 = (, , , z) Arced bordered Heegaard diagram. Also used for other kinds of Heegaard diagrams.
236

L
,
R
Left and right boundaries of , Y or 1. 236
A(Z)

CFDA()
A(Z)
Type DA bimodule associated to mapping class . 235
m
i,1,j
(a
1
, . . . , a
i
, x, b
1
, . . . , b
j
) /

bimodule structure on

CFDA. 238

CFAA(")
A(Z),A(Z)
Type AA bimodule associated to the identity map of F(:); tensoring with this turns
type D invariants into type A invariants. 236, 239
m
1,i,j
(x; a
1
, . . . , a
i
; , b
1
, . . . , b
j
) /

bimodule structure on

CFAA("
Z
). 240
A(Z),A(Z)

CFDD(") Type DD bimodule associated to the identity map of F(:); tensoring with this turns
type A invariants into type D invariants. 236, 239
Dierential on

CFDD("
Z
) given by Equation (A.4). 239
F
L
, F
R
Surfaces F(:
L
) and F(:
R
). 236
S(1) The set of generators for the arced bordered Heegaard diagram 1. 238
M
R,S
Bimodule with commuting right actions of R and S. 236
R,S
M Bimodule with commuting left actions of R and S. 236
o
L
(x), o
R
(x) Set of
L
arcs and of
R
arcs occupied by x, respectively. 238

L
,
R
Homeomorphisms
L
: F(:
L
)
L
Y and
R
: F(:
R
)
R
Y specied by a bordered
Heegaard diagram for Y . 236
MCG
0
Strongly based mapping class group. 237
INDEX OF NOTATION 259
:
L
, :
R
Left and right boundaries of 1, i.e., :
L
=
L
1 and :
R
=
R
1. 236
p, q, r, s Generators for type DA invariants of
1

,
1

, with idempotents specied by Equa-

tion (A.7). 241

Dehn twist of the standard torus around its longitude. 241

Dehn twist of the standard torus around its meridian. 241

Index of Denitions
/

algebra, see algebra, /

homomorphism, see homomorphism of /

modules
module, see module, /

homotopy equivalent, 14
tensor product, see tensor product, /

1-action
on [0, 1] 1, 62, 66
on

/
B
, 66
on matched pairs, 151, 153, 157, 165
1 1-action, 66, 68
(J1)(J4), 62
(M1)(M11), 62, 63
(SB-1

), (SB-2

), 84
(SB-1), (SB-2), (SB-3), 84
abutting
Reeb chords, 32
sequence of Reeb chords, 165
admissible
complex structure, see complex structure,
admissible
Heegaard diagram, 56
alternate characterization of, 56
provincially, 56
Alexander factor (of grading), 200
Alexander ltration, 196, 202
Alexander grading, see grading, Alexander
algebra
/

, 10
operationally bounded, 11
strictly unital, 10
associated to pointed matched circle, 35
grading on, see grading
strands, see strands algebra
almost complex structure, see complex structure
amalgamation of chords, 165
arced bordered Heegaard diagram, 236
area function, 56
arrow
horizontal, 203
length, 203
vertical, 203
arrows
diagonal, 204
bar resolution, 15
basic generator, 36
basis
ltered, 202
horizontally simplied, 203
vertically simplied, 203
becoming nodal, 81
bordered
3-manifold, 46
cutting, 57
equivalent, 46
gluing, 57
represented by Heegaard diagram, 47
Heegaard diagram, see Heegaard diagram,
bordered
bottom puncture, 68
boundary connect sum
of bordered 3-manifolds, 51
of bordered Heegaard diagrams, 51
boundary degeneration, see degeneration,
boundary, 207
boundary double point, see degeneration,
boundary double point
boundary monotonicity
strong, 63, 84
for comb, 85
for moduli space, 85
for pair (s, ), 84
for pair (x, ), 85
weak, 63
bounded
/

module, 11, 134

homomorphism of /

modules, 13
operationally b. /

algebra, 11
type D homomorphism, 19
type D structure, 18
box product, see tensor product, box
261
262 INDEX OF DEFINITIONS
Bruhat order, 31
bubble, sphere or disk, 81
building, see holomorphic building
chord, see Reeb chords
circle, matched, see matched circle
coecient maps, 193
generalized, 206
collision of levels, see end, collision of levels
comb, see holomorphic comb
compactied moduli space, see moduli space,
compactied
compatible pair
(B, ), 91
of decorated sources, 150
of generators, 58, 150
complex
Z-graded, Z-ltered over F
2
[U], 202
horizontal, 203
reduced, 203
vertical, 202
complex structure
admissible, 62
on , 110
path of, 106
on |, 62
on , 62
on [0, 1] 1, 62
on
e
, 62
on Z 1 [0, 1] 1, 66
suciently generic, 64
for path, 107
suciently pinched, 215
component
generalized split, 78, 162
ghost, 74, 83
join, 68
shue, 68
even, 69
odd, 69
split, 68
composable
sequence of sets of Reeb chords, 32
sets of Reeb chords, 32
weakly, 88
composite of /

homomorphisms, 13
concatenation, 53, 54
consistent set of Reeb chords, 32
convergence
to holomorphic comb, 71
to ideal-matched comb, 160
curve
generalized split, 78
haunted, 83
ideal-matched
simple, 161
trimmed simple, 163
join, 60, 68
shue, 60, 69
even, 69, 79
odd, 69, 79
split, 60, 68
cutting
bordered 3-manifolds, 57
bordered Heegaard diagrams, 57
cylindrical end, 51, 62
decorated source, see source, decorated
degeneration of holomorphic curves, 60, 80, see
also end
becoming nodal, 81
boundary, 83, 207
does not occur, 83
boundary double point, 83
does not occur, 85
splitting at , 80
splitting at east , 81
degree near p, odd, 76
Dehn twists, 241
destabilization, 49
dg module, 12
diagonal, partial, 64
diagram, Heegaard, see Heegaard diagram
dierential
on /(n, k), 30, 31
on

CFA, 134
on

CFD, 100
on product, 21
dierential graded module, 12
dimension, see expected dimension
disk bubble, 81
domain, 53
of triangle, 111
periodic, 53
positive, 56
provincial, 54
triply-periodic, 112
double point, boundary, see degeneration,
boundary double point
drop shadow, 80
duality theorem, 236
east innity
bimodule case, 238
holomorphic curves at, 66, 67
of Z 1 [0, 1] 1, 66
east puncture, see puncture, east
east-most level, 154
INDEX OF DEFINITIONS 263
embedded
Euler characteristic, 91
index, see index, embedded
moduli space, 91
end, see also degeneration
collision of levels, 87
cylindrical, see cylindrical end
join curve, 87
shue curve, 87
two-story, 87
energy, 63
energy ltration, 127
equivalent bordered 3-manifolds, 46
Euler measure, 64
even
shue component, see component, shue, even
shue curve, see curve, shue, even
exact triangle, surgery, 194
expected dimension, see also index
at embedded curve, 9093
of /
B
(x, y; S

; P), 66
of

/
B
(x, y; S

; P), 64
exterior generator, 214
Farkass lemma, 56
ltered
basis, 202
linearly-independent, 202
ltration, Alexander, 196, 202
formal dg algebra, 36
G-graded algebra, see grading
generalized
coecient maps, see coecient maps,
generalized
generator, 63
split component, see component, generalized
split
split curve, see curve, generalized split
generator, 52
compatible pair of, 58, 150
exterior, 214
generalized, 63
gluing g.s, 58
partially-diagonal, 117
negative, 118
positive, 117
generators
partially-diagonal
compatible, 119
matching of, 119
generic, see complex structure, suciently generic
ghost component, 74, 83
gluing
bordered 3-manifolds, 57
bordered Heegaard diagrams, 57
generators, 58
holomorphic curves, 72, 75
homology classes, 58
prole, 78
grading
spin
c
component, 39
Alexander
bordered case, 200
relationship to S, 214
dierential algebra graded by a group, 25
dierential module graded by G-set, 25
for pairing theorem, 188
group
big, 39
dependence on orientation, 182
rened, 184
small, 41
group-valued, on algebra, 25
homogeneous element of, 25
Maslov, 202
on CFK, 196
z-normalized, 196
Maslov component, 39, 176
multiplication by U on, 200
on /(:), 3941, 176
rened, 4143, 184185
on

CFA, 181
rened, 185
on

CFAA(") and

CFDD("), 240
on

CFD, 184
rened, 186
on

CFDA, 238
on

CFA, 179
rened, 185
on

CFD, 183
rened, 186
on tensor product of group-graded vector
spaces, 25
on tensor product of set-graded modules, 26
renement data, 42, 184
dependence on, 42
spin
c
component, 176
handlebody, bordered Heegaard diagrams for, 52
handleslide, 49
Hasse diagram, 31
haunted curve, 83
rare, 86
Heegaard diagram
bordered, 46
arced, 236
264 INDEX OF DEFINITIONS
cutting, 57
doubly pointed, 199
for handlebody, 52
for solid torus, 194
for trefoil complement, 51
gluing, 57
represents bordered 3-manifold, 47
capped, 65
closed, 45
doubly pointed for knot, 195
nice, 145
pointed, 45
Heegaard moves, 49
destabilization, 49
handleslide, 49
invariance under, see invariance
isotopy, 49
stabilization, 49
height, 59
Heisenberg group, 26
holomorphic building, 71
two-story, 60
holomorphic comb, 60, 70
-matched, 157
height, 70
zero, 70
homology class of, 71
horizontal level of, 70
ideal-matched, 158
nodal, 71
simple, 70
smooth, 71
spine of, 71
story of, 70
T-matched, 154
toothless, 71
trivial, 70
vertical level of, 70
holomorphic story, 70
horizontal level of, 70
T-matched, 154
holomorphic triangle, 109
homogeneous element of group-graded algebra,
see grading, group-valued
homology class
connecting generators, 53
gluing, 58
of comb, 71
of curve, 63
positive, 56
provincial, 54
homomorphism
of type D modules, 19
of /

modules, 12
composition, 13
homotopic, 14
identity, 13
null homotopic, 14
of type D modules, 17, 19
bounded, 19
homotopic, 19
homotopy of, 17
homotopic, see homomorphism, homotopic
homotopy equivalent, see module, homotopy
equivalent
horizontal
arrow, 203
length, 203
complex, 203
homology, 203
level of holomorphic story, 70
horizontally simplied basis, 203
ideal-matched holomorphic comb, 158
idempotents
minimal, 20, 29
of /(:), 35
of /(n, k), 29
index, see also expected dimension
embedded, 91
is additive, 95
of matched pair, 150
innity
, 59
east, 59, 66, 238
of [0, 1] 1, 59
of Z 1 [0, 1] 1, 66
west, 66, 238
interleaved Reeb chords, 32
intersection number, jittered, 94
invariance
of

CFA, 138143
graded, 181
of

CFD, 106130
graded, 184
inversions, 29
isotopy, see Heegaard moves, isotopy
jittered intersection number, 94
join
component, see component, join
curve, see curve, join
curve end, see end, join curve
of Reeb chords, 32
knot Floer
complex, 195
homology, 196
pairing theorem, 200
INDEX OF DEFINITIONS 265
left boundary, 238
length
of arrow, 203
length of arrow, 203
local multiplicity, 53
longitude Floer complex, 198
mapping class group, 237
-oid, 237
Maslov, see grading, Maslov
matched circle, 34
pointed, see pointed matched circle
matched pairs, see moduli space of matched pairs
matching, 34
minimal idempotent, see idempotents, minimal
module
/

, 11
bounded, 11
strictly unital, 11
dierential graded, 12
type D, 17
bounded, 18
coecient maps, see coecient maps
moduli space, 59, 63, 64, 66
compactied, 72
is compact, 72
embedded, 91
of -matched pairs, 157
of combs, see holomorphic comb
of curves at east , 67
of matched pairs, 150
embedded, 152
expected dimension, 150
of T-matched pairs, 153
reduced, 60, 66
rigid, 60
monotonicity, see boundary monotonicity
Morse function
compatible with bordered 3-manifold, 47
compatible with pointed matched circle, 34
multiplicity of point in homology class, 39
nice diagram, 145
nodal
becoming, 81
holomorphic comb, see holomorphic comb,
nodal
nose, 55
null homotopic, see homomorphism, null
homotopic
occupied -arcs, 52, 185
odd
degree near p, 76
shue component, see component, shue, odd
shue curve, see curve, shue, odd
operationally bounded /

algebra, 11
ordered partition, 59
ordering of Reeb chords at e, 59
orientation reversal, 34, 57, 99, 100, 102, 182, 239
pairing theorem
for change of framing, 235
for knot Floer, 200
graded, 188
relating

CFA and

CFD, 236
twisted coecients, 170
partial diagonal, 64
partially-diagonal generator, see generator,
partially-diagonal
partition
ordered, see ordered partition, 66
periodic domain, see domain, periodic
pointed Heegaard diagram, see Heegaard
diagram, pointed
pointed matched circle, 34
boundary of bordered Heegaard diagram, 46
orientation reverse, 34
surface associated to, 34
positive homology class, 56
pregluing sources, 72
giving closed source, 150
proper near p, 76
provincial, see homology class, provincial
puncture
bottom, 68
boundary of is a, 62
east, 59, 62, 67
top, 68
west, 67
reduced
complex, 203
moduli space, 60, 66
Reeb chord, 59
Reeb chords, 32
abutting, 32, 165
amalgamation, 165
composable
sequence of sets of, 32
sets of, 32
interleaved, 32
join of, 32
nested, 32
set of
consistent, 32
shue of, 33
splitting of, 33
weak shue of, 33
266 INDEX OF DEFINITIONS
weak splitting of, 33
splitting of, 33
set of, 33
renement data, see grading renement data
region, 53
twin, 117
winding, 212
right boundary, 238
S-graded module, 25
section of matching, 35
set-graded, see grading
shue
component, see component, shue
curve, see curve, shue
curve end, see end, shue curve
simple
holomorphic comb, see holomorphic comb,
simple
simple ideal-matched curve, see curve, simple
ideal-matched
smeared neighborhood, 76
smooth holomorphic comb, see holomorphic
comb, smooth
smoothing a crossing, 31
solid torus, Heegaard diagram for, 194
source
bi-decorated, 66
decorated, 59, 62
for map to , 110
sphere bubble, 81
spin
c
component of grading, 39, 176
spin
c
-structure, 54
associated to generator, 55
associated to vector eld, 54
relative
associated to generator, 55
spin
c
-structure
relative, 55
spine
of comb, 71
of ideal-matched curve, 165
split
component, see component, split
curve, see curve, split
splitting
at , 80
at east , 81
stabilization, 49
stable
holomorphic curve, 66
matched pair, 151
T-matched story, 154
stable holomorphic curve
at east , 67
story, see holomorphic story
strands algebra, 30
element associated to a
0
(), 32
strands diagram, 30
stratied
map, 76
space, 76
strictly unital
/

algebra, 10
/

module, 11

CFA is, 134

homomorphism of /

modules, 12
strong boundary monotonicity, see boundary
monotonicity, strong
strongly based mapping class group, 237
suciently generic
complex structure, 64
path of almost complex structures, 107
suciently pinched complex structure, 215
surface associated to pointed matched circle, 34
surgery exact triangle, 194
symplectic form
on |, 62
on , 62
on [0, 1] 1, 62
T-matched
comb, 154
story, 154
tautological correspondence, 207
tensor product
/

, 15
is triangulated functor, 24
box, 21
is triangulated functor, 24
of homomorphisms, 22
time-parameter space, 158
toothless comb, 71, 165
top puncture, 68
torus
algebra, 192
coecient maps, see coecient maps
rened grading group, 193
unrened grading group, 192
total east source, 159
total west source, 159
transversality, 64
for matched pair, 150
trefoil, 51
triangle, holomorphic, 109
triply-periodic domain, 112
trivial
component, 68
INDEX OF DEFINITIONS 267
holomorphic comb, see holomorphic comb,
trivial
twin regions, 117
twisted coecients
for

CFD, 130
pairing theorem with, 170
two-story end, see end, two-story
type A invariant, 133134
for solid torus, 201
type D
homomorphism, see homomorphism, type D
invariant, 99101
module, see module, type D
vertical
arrow, 203
length, 203
complex, 202
homology, 202
level of comb, 70
vertically simplied basis, 203
weakly composable sets of chords, 88
weight of algebra element, 36
west innity
bimodule case, 238
of Z 1 [0, 1] 1, 66
west puncture, see puncture, west
west-most level, 154
winding region, 212
z-normalized Maslov grading, 196
Bibliography
[BB05] Anders Bjrner and Francesco Brenti, Combinatorics of Coxeter groups, Graduate Texts in
Mathematics, vol. 231, Springer, 2005.
[BEH
+
03] Frdric Bourgeois, Yakov Eliashberg, Helmut Hofer, Kris Wysocki, and Eduard Zehnder, Com-
pactness results in symplectic eld theory, Geom. Topol. 7 (2003), 799888, arXiv:math.SG/
0308183.
[BK90] A. I. Bondal and M. M. Kapranov, Framed triangulated categories, Mat. Sb. 181 (1990), no. 5,
669683.
[Bou06] Frdric Bourgeois, A Morse-Bott approach to contact homology, Ph.D. thesis, Stanford Univer-
sity, 2006.
[DK90] Simon K. Donaldson and Peter B. Kronheimer, The geometry of four-manifolds, Oxford Math-
ematical Monographs, Oxford University Press, 1990.
[Eft05] Eaman Eftekhary, Longitude Floer homology and the Whitehead double, Algebr. Geom. Topol.
5 (2005), 13891418, arXiv:math.GT/0407211.
[Eft08] , Floer homology and splicing knot complements, 2008, arXiv:0802.2874.
[EGH00] Yakov Eliashberg, Alexander Givental, and Helmut Hofer, Introduction to symplectic eld theory,
Geom. Funct. Anal. (2000), no. Special Volume, Part II, 560673, arXiv:math.SG/0010059,
GAFA 2000 (Tel Aviv, 1999).
[Far02] Julius Farkas, ber die Theorie der einfachen Ungleichungen, Journal fr die Reine und Ange-
wandte Mathematik 124 (1902), 127.
[Flo89] Andreas Floer, Symplectic xed points and holomorphic spheres, Comm. Math. Phys. 120
(1989), no. 4, 575611.
[Flo95] , Instanton homology and Dehn surgery, The Floer memorial volume, Progr. Math., no.
133, Birkhuser, 1995, pp. 7797.
[FS94] Sergey Fomin and Richard P. Stanley, Schubert polynomials and the nil-Coxeter algebra, Adv.
Math. 103 (1994), no. 2, 196207.
[Fuk93] Kenji Fukaya, Morse homotopy, A

-category, and Floer homologies, Proceedings of GARC

Workshop on Geometry and Topology 93 (Seoul), Lecture Notes Ser., vol. 18, Seoul Nat. Univ.,
1993, pp. 1102.
[Hed05a] Matthew Hedden, On knot Floer homology and cabling, Ph.D. thesis, Columbia University, 2005.
[Hed05b] , On knot Floer homology and cabling, Algebr. Geom. Topol. 5 (2005), 11971222, arXiv:
math.GT/0406402.
[Hed07] , Knot Floer homology of Whitehead doubles, Geom. Topol. 11 (2007), 22772338, arXiv:
math.GT/0606094.
[HLS97] Helmut Hofer, Vronique Lizan, and Jean-Claude Sikorav, On genericity for holomorphic curves
in four-dimensional almost-complex manifolds, J. Geom. Anal. 7 (1997), no. 1, 149159.
[Hof06] H. Hofer, A general Fredholm theory and applications, Current developments in mathematics,
2004, Int. Press, Somerville, MA, 2006, pp. 171, arXiv:math/0509366v1.
[Juh06] Andrs Juhsz, Holomorphic discs and sutured manifolds, Algebr. Geom. Topol. 6 (2006), 1429
1457, arXiv:math/0601443.
[Juh08] Andrs Juhsz, Floer homology and surface decompositions, Geom. Topol. 12 (2008), no. 1,
299350, arXiv:math/0609779.
[Kel] Bernhard Keller, Koszul duality and coderived categories (after K. Lefvre), http://people.
math.jussieu.fr/~keller/publ/kdc.dvi.
[Kel01] , Introduction to A-innity algebras and modules, Homology Homotopy Appl. 3 (2001),
no. 1, 135, arXiv:math.RA/9910179.
269
270 BIBLIOGRAPHY
[Kho01] Mikhail Khovanov, Nilcoxeter algebras categorify the Weyl algebra, Comm. Algebra 29 (2001),
no. 11, 50335052, arXiv:math.RT/9906166.
[Kho10] , How to categorify one-half of quantum gl(1 [ 2), 2010, arXiv:1007.3517.
[KM07] Peter Kronheimer and Tomasz Mrowka, Monopoles and three-manifolds, New Mathematical
Monographs, vol. 10, Cambridge University Press, Cambridge, 2007.
[KMOSz07] Peter Kronheimer, Tomasz Mrowka, Peter S. Ozsvth, and Zoltn Szab, Monopoles and lens
space surgeries, Ann. of Math. (2) 165 (2007), no. 2, 457546, arXiv:math.GT/0310164.
[Kon95] Maxim Kontsevich, Homological algebra of mirror symmetry, Proceedings of the International
Congress of Mathematicians, Vol. 1, 2 (Zrich, 1994) (Basel), Birkhuser, 1995, pp. 120139.
[Lev09] Adam Simon Levine, Slicing mixed Bing-Whitehead doubles, 2009, arXiv:0912.5222.
[Lev10] , Knot doubling operators and bordered Heegaard Floer homology, 2010, arXiv:1008.3349.
[LH03] Kenji Lefvre-Hasegawa, Sur les A

-catgories, Ph.D. thesis, Universit Denis Diderot Paris

7, 2003, arXiv:math.CT/0310337.
[Lip06a] Robert Lipshitz, A cylindrical reformulation of Heegaard Floer homology, Geom. Topol. 10
(2006), 9551097, arXiv:math.SG/0502404.
[Lip06b] , A Heegaard-Floer invariant of bordered 3-manifolds, Ph.D. thesis, Stanford University,
Palo Alto, CA, 2006.
[LMW08] Robert Lipshitz, Ciprian Manolescu, and Jiajun Wang, Combinatorial cobordism maps in hat
Heegaard Floer theory, Duke Math. J. 145 (2008), no. 2, 207247, arXiv:math/0611927.
[LOTa] Robert Lipshitz, Peter S. Ozsvth, and Dylan P. Thurston, Bordered Floer homology and the
spectral sequence of a branched double cover II: the spectral sequences agree, in preparation.
[LOTb] , Computing cobordism maps with bordered Floer homology, in preparation.
[LOT09] , Slicing planar grid diagrams: a gentle introduction to bordered Heegaard Floer homol-
ogy, Proceedings of Gkova Geometry-Topology Conference 2008, Gkova Geometry/Topology
Conference (GGT), Gkova, 2009, pp. 91119, arXiv:0810.0695.
[LOT10a] , Bimodules in bordered Heegaard Floer homology, 2010, arXiv:1003.0598.
[LOT10b] , Bordered oer homology and the spectral sequence of a branched double cover I, 2010,
arXiv:1011.0499.
[LOT10c] , Computing

HF by factoring mapping classes, 2010, arXiv:1010.2550.
[LOT10d] , A faithful linear-categorical action of the mapping class group of a surface with bound-
ary, 2010, arXiv:1012.1032.
[LOT10e] , Heegaard Floer homology as morphism spaces, 2010, arXiv:1005.1248.
[McD94] Dusa McDu, Singularities and positivity of intersections of J-holomorphic curves, Holomorphic
curves in symplectic geometry, Progr. Math., vol. 117, Birkhuser, Basel, 1994, pp. 191215,
With an appendix by Gang Liu.
[MS04] Dusa McDu and Dietmar Salamon, J-holomorphic curves and symplectic topology, American
Mathematical Society Colloquium Publications, vol. 52, American Mathematical Society, Prov-
idence, RI, 2004.
[MW95] Mario J. Micallef and Brian White, The structure of branch points in minimal surfaces and in
pseudoholomorphic curves, Ann. of Math. (2) 141 (1995), no. 1, 3585.
[Ni09] Yi Ni, Non-separating spheres and twisted Heegaard Floer homology, 2009, arXiv:0902.4034.
[OSz03] Peter S. Ozsvth and Zoltn Szab, Knot Floer homology and the four-ball genus, Geom. Topol.
7 (2003), 615639, arXiv:math.GT/0301149.
[OSz04a] , Holomorphic disks and genus bounds, Geom. Topol. 8 (2004), 311334, arXiv:math.
GT/0311496.
[OSz04b] , Holomorphic disks and knot invariants, Adv. Math. 186 (2004), no. 1, 58116, arXiv:
math.GT/0209056.
[OSz04c] , Holomorphic disks and three-manifold invariants: properties and applications, Ann. of
Math. (2) 159 (2004), no. 3, 11591245, arXiv:math.SG/0105202.
[OSz04d] , Holomorphic disks and topological invariants for closed three-manifolds, Ann. of Math.
(2) 159 (2004), no. 3, 10271158, arXiv:math.SG/0101206.
[OSz06] , Holomorphic triangles and invariants for smooth four-manifolds, Adv. Math. 202
(2006), no. 2, 326400, arXiv:math.SG/0110169.
BIBLIOGRAPHY 271
[OSz08] , Holomorphic disks, link invariants and the multi-variable Alexander polynomial, Algebr.
Geom. Topol. 8 (2008), no. 2, 615692, arXiv:math/0512286.
[OSzT08] Peter S. Ozsvth, Zoltn Szab, and Dylan P. Thurston, Legendrian knots, transverse knots
and combinatorial Floer homology, Geom. Topol. 12 (2008), no. 2, 941980, arXiv:math.GT/
0611841.
[Pet09] Ina Petkova, Cables of thin knots and bordered Heegaard Floer homology, 2009, arXiv:0911.2679.
[Ras03] Jacob Rasmussen, Floer homology and knot complements, Ph.D. thesis, Harvard University,
Cambridge, MA, 2003, arXiv:math.GT/0306378.
[Sar10] Sucharit Sarkar, Maslov index formulas for Whitney n-gons, 2010, arXiv:math/0609673v3.
[Sei02] Paul Seidel, Fukaya categories and deformations, Proceedings of the International Congress
of Mathematicians, Vol. II (Beijing, 2002) (Beijing), Higher Ed. Press, 2002, pp. 351360,
arXiv:math.SG/0206155.
[Sei08a] , Fukaya categories and Picard-Lefschetz theory, Zurich Lectures in Advanced Mathe-
matics, European Mathematical Society (EMS), Zrich, 2008.
[Sei08b] , Lectures on four-dimensional Dehn twists, Symplectic 4-manifolds and algebraic sur-
faces, Lecture Notes in Math., vol. 1938, Springer, Berlin, 2008, pp. 231267, arXiv:math/
0309012.
[Sta63] James D. Stashe, Homotopy associativity of H-spaces. I, II, Trans. Amer. Math. Soc. 108
(1963), 275312.
[SW10] Sucharit Sarkar and Jiajun Wang, An algorithm for computing some Heegaard Floer homologies,
Ann. of Math. (2) 171 (2010), no. 2, 12131236, arXiv:math/0607777.
[Tur97] Vladimir Turaev, Torsion invariants of Spin
c
-structures on 3-manifolds, Math. Res. Lett. 4
(1997), no. 5, 679695.
[Wit88] Edward Witten, Topological quantum eld theory, Comm. Math. Phys. 117 (1988), no. 3, 353
386.
[Wit94] , Monopoles and four-manifolds, Math. Res. Lett. 1 (1994), 769796, arXiv:hep-th/
9411102.
[Zar09] Rumen Zarev, Bordered Floer homology for sutured manifolds, 2009, arXiv:0908.1106.
[Zar10] , Joining and gluing sutured Floer homology, 2010, arXiv:1010.3496.