You are on page 1of 58

a

r
X
i
v
:
1
0
1
0
.
3
4
9
6
v
1


[
m
a
t
h
.
G
T
]


1
8

O
c
t

2
0
1
0
JOINING AND GLUING SUTURED FLOER HOMOLOGY
RUMEN ZAREV
Abstract. We give a partial characterization of bordered Floer ho-
mology in terms of sutured Floer homology. The bordered algebra and
modules are direct sums of certain sutured Floer complexes. The algebra
multiplication and algebra action correspond to a new gluing map on
SFH. It is dened algebraically, and is a special case of a more general
join map.
In a follow-up paper we show that this gluing map can be identied
with the contact cobordism map of Honda-Kazez-Matic. The join map
is conjecturally equivalent to the cobordism maps on SFH dened by
Juhasz.
1. Introduction
Heegaard Floer homology is a family of invariants for 3 and 4manifold
invariants dened by counting pseudo-holomorphic curves, originally intro-
duced by Ozsvath and Szab o. The most simple form associates to an ori-
ented 3manifold Y a graded homology group

HF(Y ) [OS04b, OS04a].
While Heegaard Floer theory for closed 3manifolds has been very suc-
cessful, a lot of the applications involve manifolds with boundary. In [Juh06]
Juhasz introduced sutured Floer homology, or SFH, which generalizes

HF
to sutured manifolds. Introduced by Gabai in [Gab83], they are 3manifolds
with boundary, and some extra structure. In the context of Heegaard Floer
homology, the extra structure can be considered to be a multicurve , called
a dividing set, on the boundary of the 3manifold Y . Sutured Floer homol-
ogy associates to such a pair (Y, ) a homology group SFH(Y, ).
Among other applications, sutured Floer homology has been used to solve
problems in contact topology, via a contact invariant for contact manifolds
with boundary, and a map associated to contact cobordisms, dened by
Honda, Kazez, and Matic in [HKM07, HKM08]. This map has been used by
Juhasz in [Juh09] to dene a map on SFH associated to a cobordism (with
corners) between two sutured manifolds.
A shortcoming of sutured Floer homology is that there is little relationship
between the groups SFH(Y,
1
) and SFH(Y,
2
), where
1
and
2
are two
dividing sets on the same manifold Y . For example one can nd many exam-
ples where one of the groups vanishes, while the other does not. Moreover,
The author was partially supported by NSF grant number DMS-0804121.
1
2 RUMEN ZAREV
the groups SFH(Y
1
,
1
) and SFH(Y
2
,
2
) are not sucient to reconstruct

HF(Y ), where Y = Y
1
Y
2
is a closed manifold.
To overcome these shortcomings, Lipshitz, Ozsvath, and Thurston in-
troduced in [LOT08] a new Heegaard Floer invariant for 3manifolds with
boundary called bordered Floer homology. To a parametrized closed con-
nected surface F they associate a DG-algebra /(F). To a 3manifold Y
with boundary Y

= F they associate an /

module

CFA(Y ) over /(F)
(dened up to /

homotopy equivalence). This invariant overcomes both


of the above shortcomings of SFH. On the one hand, given two parametriza-
tions of the surface F, the modules

CFA(Y ) associated to these parametriza-
tions can be computed from each other. On the other hand, if Y
1
and Y
2
are
two manifolds with boundary dieomorphic to F, the group

HF(Y
1

F
Y
2
)
can be computed from

CFA(Y
1
) and

CFA(Y
2
).
The natural question arises: How are these two theories for 3manifolds
with boundary related to each other? Can SFH(Y, ) be computed from

CFA(Y ), and if yes, how? Can



CFA(Y ) be computed from the sutured
homology of Y , and if yes, how?
In [Zar09] we introduced bordered sutured Floer homology, to serve as a
bridge between the two worlds. We used it to answer the rst part of the
above questionto each dividing set on F we can associate a module

CFD() over /(F), such that SFH(Y, ) is simply the homology of the
derived tensor product

CFA(Y )


CFD().
In the current paper we answer the second half of this question. We show
that for a given parametrization of F, the homologies of the bordered algebra
/(F) and the module

CFA(Y ) associated to a 3manifold Y are direct
sums of nitely many sutured Floer homology groups. Moreover we identify
multiplication in H

(/(F)) and the action of H

(/(F)) on H

CFA(Y ))
with a certain gluing map on sutured Floer homology.
1.1. Results. The rst result of this paper is to dene the gluing map dis-
cussed above. Suppose (Y
1
,
1
) and (Y
2
,
2
) are two sutured manifolds. We
say that we can glue them if there are subsets F
1
and F
2
of their boundaries,
where F
1
can be identied with the mirror of F
2
, such that the multicurve

1
F
1
is identied with
2
F
2
, preserving the orientations on
i
. This
means that the regions R
+
and R

on the two boundaries are interchanged.


We will only talk of gluing in the case when F
i
have no closed components,
and all components of F
i
intersect the dividing sets
i
.
Denition 1.1. Suppose (Y
1
,
1
), (Y
2
,
2
), F
1
and F
2
are as above. The
gluing of (Y
1
,
1
) and (Y
2
,
2
) along F
i
is the sutured manifold (Y
1

F
i
Y
2
,
1+2
). The dividing set
1+2
is obtained from (
1
F
1
)
F
i
(
2
F
2
)
as follows. Along each component f of F
i
the orientations of
1
and
2
disagree. We apply the minimal possible positive fractional Dehn twist along
f that gives a consistent orientation.
JOINING AND GLUING SFH 3


}
1
}
2
}
1

F
i
}
2
F
1
F
2
Figure 1. Gluing two solid balls along F = D
2
D
2
, to obtain a solid torus.
The R
+
regions have been shaded.
An illustration of gluing is given in Figure 1. We dene a gluing map
on SFH corresponding to this topological construction.
Theorem 1. Let (Y
1
,
1
) and (Y
2
,
2
) be two balanced sutured manifolds,
that can be glued along F. Then there is a well dened map

F
: SFH(Y
1
,
1
) SFH(Y
2
,
2
) SFH((Y
1
,
1
)
F
(Y
2
,
2
)),
satisfying the following properties:
(1) Symmetry: The map
F
for gluing Y
1
to Y
2
is equal to that for gluing
Y
2
to Y
1
.
(2) Associativity: Suppose that we can glue Y
1
to Y
2
along F
1
, and Y
2
to Y
3
along F
2
, such that F
1
and F
2
are disjoint in Y
2
. Then the
order of gluing is irrelevant:

F
2

F
1
=
F
1

F
2
=
F
1
F
2
.
(3) Identity: Given a dividing set on F, there is a dividing set

on F [0, 1], and an element

SFH(F [0, 1],

), satisfying
the following. For any sutured manifold (Y,

) with F Y and

F = , there is a dieomorphism (Y,

)
F
(F [0, 1],

)

=
(Y,

). Moreover, the map


F
(,

) is the identity of SFH(Y,

).
One application of this result is the following characterization of bordered
Floer homology in terms of SFH and the gluing map. Fix a parametrized
closed surface F, with bordered algebra A = /(F). Let F

be F with a
disc removed, and let p, q F

be two points. We can nd 2


2g(F)
distin-
guished dividing sets on F, which we denote
I
for I 1, . . . , 2g, and
corresponding dividing sets

I
=
I
F

on F

. Let
IJ
be a dividing set
on F

[0, 1] which is

I
along F

0,

J
along F

1, and half of a
negative Dehn twist of p, q [0, 1] along F

[0, 1].
4 RUMEN ZAREV
Theorem 2. Suppose the surfaces F and F

, the algebra A, and the dividing


sets
I
,

I
, and
IJ
are as described above. Then there is an isomorphism
H

(A)

I,J{1,...,2g}
SFH(F

[0, 1],
IJ
),
and the multiplication map
2
on H

(A) can be identied with the gluing map

F
. It maps SFH(F

[0, 1],
IJ
) SFH(F

[0, 1],
JK
) to SFH(F

[0, 1],
IK
), and sends all other summands to 0.
The module

CFA can be similarly described.
Theorem 3. Suppose Y is a 3manifold with boundary Y

= F. There is
an isomorphism
H

CFA(Y )
A
)

I{1,...,2g}
SFH(Y,
I
),
and the action m
2
of H

(A) on H

CFA(Y )) can be identied with the gluing


map
F
. It maps SFH(Y,
I
) SFH(F

[0, 1],
IJ
) to SFH(Y,
J
), and
sends all other summands to 0.
The gluing construction and the gluing map readily generalize to a more
general join construction, and join map, which are 3dimensional analogs.
Suppose that (Y
1
,
1
) and (Y
2
,
2
) are two sutured manifolds, and F
1
and F
2
are subsets of their boundaries, satisfying the conditions for gluing. Suppose
further that the dieomorphism F
1
F
2
extends to W
1
W
2
, where W
i
is
a compact codimension0 submanifold of Y
i
, and W
i
Y
i
= F
i
. Instead
of gluing Y
1
and Y
2
along F
i
, we can join them along W
i
.
Denition 1.2. The join of (Y
1
,
1
) and (Y
2
,
2
) along W
i
is the sutured
manifold
((Y
1
W
1
)
W
i
\F
i
(Y
2
W
2
),
1+2
),
where the dividing set
1+2
is constructed exactly as in Denition 1.1. We
denote the join by (Y
1
,
1
)
W
i
(Y
2
,
2
).
An example of a join is shown in Figure 2. Notice that if W
i
is a collar
neighborhood of F
i
, then the notions of join and gluing coincide. That is,
the join operation is indeed a generalization of gluing. In fact, throughout
the paper we work almost exclusively with joins, while only regarding gluing
as a special case.
Theorem 4. There is a well-dened join map

W
: SFH(Y
1
,
1
) SFH(Y
2
,
2
) SFH((Y
1
,
1
)
W
(Y
2
,
2
)),
satisfying properties of symmetry, associativity, and identity, analogous to
those listed in Theorem 1.
JOINING AND GLUING SFH 5

}
1
}
2
}
1

W
i
}
2
J
1
J
2
Figure 2. Join of two solid tori along D
2
[0, 1], to obtain another solid
torus. The R
+
regions have been shaded.
The join map is constructed as follows. We cut out W
1
and W
2
from
Y
1
and Y
2
, respectively, and regard the complements as bordered sutured
manifolds. The join operation corresponds to replacing W
1
and W
2
by an
interpolating piece T J
F,+
. We dene a map between the bordered su-
tured invariants, from the product

BSA(W
1
)

BSA(W
2
) to the bimodule

BSAA(T J
F,+
). We show that for an appropriate choice of parametriza-
tions, the modules

BSA(W
1
) and

BSA(W
2
) are duals, while

BSAA(T J
F,+
)
is the dual of the bordered algebra for F. The map is then an /

version of
the natural pairing between a module and its dual. The proof of invariance
and the properties from Theorems 1 and 4 is purely algebraic. Most of the
arguments involve /

versions of standard facts in commutative algebra.


The proofs of Theorems 2 and 3 involve several steps. First, we nd a
manifold whose bordered sutured invariant is the bordered algebra, as a
bimodule over itself. Second, we nd manifolds whose bordered sutured
invariants are all possible simple modules over the algebra. Finally, we
compute the gluing map explicitly in several cases.
1.2. Further implications and speculations. In a follow-up paper [Zar]
we prove that the gluing map map
F
can be identied with the contact
6 RUMEN ZAREV
cobordism map from [HKM08]. This is somewhat surprising as that the
denition of that map uses contact structures and open books, while our
denition uses bordered sutured Floer homology and is purely algebraic.
The equivalence of the two maps also gives a purely contact-geometric in-
terpretation of the bordered algebra.
There is no analog of the join map in the setting of Honda, Kazez, and
Matic. However, there is a natural pair-of-pants cobordism
Z
W
: (Y
1
,
1
) (Y
2
,
2
) (Y
1
,
1
)
W
(Y
2
,
2
),
and we conjecture that the join map
W
is equivalent to the cobordism
map F
Z
W
that Juhasz associates to such a cobordism, by counting pseudo-
holomorphic triangles.
Though Theorems 2 and 3 give a pretty good description of bordered
Floer homology in terms of sutured Floer homology, it is not complete.
For instance, to be able to recover the pairing theorem for bordered Floer
homology, we need to work either with the full bordered DG-algebra /(F),
or with its homology H

(/(F)), considered as an /

algebra. That is,


H

(/(F)) inherits higher multiplication maps


i
, for i 2 from the DG-
structure on /(F). Theorem 2 only recovers
2
. Similarly, H

CFA(Y )) has
higher actions m
i
, for i 2 by H

(/(F)), while Theorem 3 only recovers


m
2
.
We believe that these higher structures can be recovered by a similar
construction. There are maps,
i
, for i 2, dened algebraically, similar
to , of the following form:

i
: SFC(Y
1
) SFC(Y
i
) SFC(Y
1
. . . Y
i
).
Here SFC denotes the chain complex dening the homology group SFH.
The rst term
2
induces the usual join on homology.
Conjecture 5. The following two statements hold:
(1) The collection of maps
i
, for i 2 can be used to recover the
higher multiplications
i
on H

(/(F)), and the higher actions m


i
of
H

(/(F)) on H

CFA(Y )).
(2) The map
i
can be computed by counting pseudo-holomorphic (i+1)
gons in a sutured Heegaard multidiagram.
Analogs of sutured Floer homology have been dened in settings other
than Heegaard Floer homologyfor instanton and monopole Floer homol-
ogy in [KM10], and for embedded contact homology in [CGHH10]. We
believe that analogs of the join and gluing maps can be used to extend
bordered Floer homology to those settings.
Organization. We start by introducing in more detail the topological con-
structions of the gluing join operations in Section 2. In Section 3 we recall
briey the denitions of the bordered sutured invariants

BSA and

BSD. We
JOINING AND GLUING SFH 7
also discuss how the original denitions involving only arcs can be ex-
tended to diagrams using both arcs, and to some mixed diagrams using
both. Section 3.4 contains computations of several

BSA invariants needed
later.
We dene the join map in Section 4, on the level of chain complexes. The
same section contains the proof that it descends to a unique map on homol-
ogy. In the following Section 5 we prove the properties from Theorems 1
and 4. Finally, Section 6 contains the statement and the proof of a slightly
more general version of Theorems 2 and 3.
Throughout the paper, we make use of a diagrammatic calculus to com-
pute /

morphisms, which greatly simplies the arguments. Appendix A


contains a brief description of this calculus, and the necessary algebraic as-
sumptions. Appendix B gives an overview of /

bimodules in terms of the


diagrammatic calculus, as they are used in the paper.
Acknowledgments. The author is grateful to his advisor Peter Ozsvath,
and to Robert Lipshitz, Dylan Thurston, and Shea Vela-Vick for many pro-
ductive discussions about this work. Shea Vela-Vick, Robert Lipshitz, and
Peter Ozsvath also gave much appreciated feedback on earlier versions of
this paper. A signicant portion of the work described here was carried out
at the Mathematical Sciences Research Institute, which kindly hosted the
author as a program associate in the program Homology theories of knots
and links, during the Spring of 2010.
2. Topological preliminaries
We recall the denition of a sutured manifold and some auxiliary notions,
and dene what we mean by gluing and surgery.
Remark. Throughout the paper all manifolds are oriented. We use M to
denote the manifold M with its orientation reversed.
2.1. Sutured manifolds and surfaces.
Denition 2.1. As dened in [Juh06], a balanced sutured manifold is a
pair } = (Y, ) consisting of the following:
An oriented 3manifold Y with boundary.
A collection of disjoint oriented simple closed curves in Y , called
sutures.
They are required to satisfy the following conditions:
Y can be disconnected but cannot have any closed components.
Y is divided by into two complementary regions R
+
() and R

()
such that R

(Y ) = . (R
+
and R

may be disconnected.)
Each component of Y contains a suture. Equivalently, R
+
and R

have no closed components.


(R
+
) = (R

).
In [Zar09] we introduced the notion of a sutured surface.
8 RUMEN ZAREV
Denition 2.2. A sutured surface is a pair T = (F, ) consisting of the
following:
A compact oriented surface F.
A nite collection F of points with sign, called sutures.
They are required to satisfy the following conditions:
F can be disconnected but cannot have any closed components.
F is divided by into two complementary regions S
+
() and S

()
such that S

(Y ) = . (S
+
and S

may be disconnected.)
Each component of F contains a suture. Equivalently, S
+
and S

have no closed components.


A sutured surface is precisely the 2dimensional equivalent of a balanced
sutured manifold. The requirement (S
+
) = (S

) follows automatically
from the other conditions.
From T = (F, ) we can construct two other sutured surfaces: T =
(F, ), and T = (F, ). In both of T and T, the orientation of the
underlying surface F is reversed. The dierence between the two is that in
T the roles of S
+
and S

are preserved, while in T they are reversed.


Denition 2.3. Suppose T = (F, ) is a sutured surface. A dividing set
for T is a nite collection of disjoint embedded oriented arcs and simple
closed curves in F, with the following properties:
= , as an oriented boundary.
divides F into (possibly disconnected) regions R
+
and R

with
R

= () S

.
We can extend the denition of a dividing set to pairs (F, ) which do not
quite satisfy the conditions for a sutured surface. We can allow some or all
of the components F to be closed. We call such a pair degenerate. In that
case we impose the extra condition that each closed component contains a
component of .
Note that the sutures of a sutured manifold (Y, ) can be regarded as
a dividing set for the (degenerate) sutured surface (Y, ).
Denition 2.4. A partially sutured manifold is a triple } = (Y, , T)
consisting of the following:
A 3manifold Y with boundary and 1dimensional corners.
A sutured surface T = (F, ), such that F Y , and such that the
1dimensional corner of Y is F.
A dividing set for (Y F, ) (which might be degenerate).
Note that a partially sutured manifold } = (Y, , T
1
T
2
) can be thought
of as a cobordism between T
1
and T
2
. On the other hand, the par-
tially sutured manifold } = (Y, , ) is just a sutured manifold, although
it may not be balanced. We can concatenate } = (Y, , T
1
T
2
) and
}

= (Y

, T
2
T
3
) along T
2
= (F
2
,
2
) and T
2
= (F
2
,
2
) to
JOINING AND GLUING SFH 9
+

+
+
(a) The sutured surface T.
+

+
+
(b) A dividing set of T.
T
[0, 1]
(T, )
(c) The cap for .
Figure 3. A sutured annulus T, with a cap associated to a dividing set.
obtain
}
F
2
}

= (Y
F
2
Y

, T
1
T
3
).
We use the term concatenate to distinguish from the operation of gluing of
two sutured manifolds described in Denition 2.10.
A partially sutured manifold whose sutured surface is parametrized by an
arc diagram is a bordered sutured manifold, as dened in [Zar09]. We will
return to this point in section 3, where we give the precise denitions.
An important special case is when Y is a thickening of F.
Denition 2.5. Suppose is a dividing set for the sutured surface T =
(F, ). Let W = F [0, 1], and W

= F [0, 1]/ , where (p, t) (p, t

)
whenever p F, and t, t

[0, 1]. We will refer to the partially sutured


manifolds
J

= (W, 1 [0, 1], (F 0, 0)),


J

= (W

, 1, (F 0, 0))
as the caps for T associated to .
Since J

is just a smoothing of J

along the corner F 1, we will


not distinguish between them. An illustration of a dividing set and a cap
is shown in Figure 3. In this and in all other gures we use the convention
that the dividing set is colored in green, to avoid confusion with Heegaard
diagrams later. We also shade the R
+
regions.
Notice that the sutured surface for J

is T. This means that if } =


(Y,

, T) is a partially sutured manifold, we can concatenate } and J to


obtain (Y,

). That is, the eect is that of lling in F Y by .


Denition 2.6. Suppose T = (F, ) is a sutured surface. An embedding
J } of the partially sutured J = (W,
W
, T) into the sutured } =
(Y,
Y
) is an embedding W Y with the following properties:
F W is properly embedded in Y as a separating surface.
10 RUMEN ZAREV
J
}
} J
T T
Figure 4. Examples of a partially sutured manifold J embedding into the
sutured manifold }, and the complement }J, which is also partially sutured.
W F = Y W.

W
=
Y
W.
The complement Y W also inherits a partially sutured structure. We
dene
} J = (Y W,
Y

W
, T).
The denition of embeddings easily extends to J } where both J =
(W,
W
, T) and } = (Y,
Y
, T

) are partially sutured. In this case we require


that W is disjoint from a collar neighborhood of F

. Then there is still a


complement
} J = (Y W,
Y

W
, T

T).
In both cases } is dieomorphic to the concatenation J
F
(} J).
Examples of a partial sutured manifold and of an embedding are given in
Figure 4.
2.2. Mirrors and doubles; joining and gluing. We want to dene a
gluing operation which takes two sutured manifolds (Y
1
,
1
) and (Y
2
,
2
),
and surfaces F Y
1
and F Y
2
, and produces a new sutured manifold
(Y
1

F
Y
2
,
3
). To do that we have to decide how to match up the dividing
sets on and around F and F. One solution is to require that we glue
F R
+
(
1
) to F R
+
(
2
), and F R

(
1
) to F R

(
2
). Then
(
1
F) (
2
F) is a valid dividing set, and candidate for
3
. The
problem with this approach is that even if we glue two balanced sutured
manifolds, the result is not guaranteed to be balanced.
Another approach, suggested by contact topology is the following. We
glue F R
+
to F R

, and vice versa. To compensate for the fact that


the dividing sets
1
F and
2
F do not match up anymore, we introduce
a slight twist along F. In contact topology this twist appears when we
smooth the corner between two convex surfaces meeting at an angle.
JOINING AND GLUING SFH 11
J = (W, , T) J = (W, , T)
T
T
+

+
+

+
Figure 5. A partially sutured manifold J and its mirror J.
It turns out that the same approach is the correct one, from the bordered
sutured point of view. To be able to dene a gluing map on SFH with nice
formal properties, the underlying topological operation should employ the
same kind of twist. However, its direction is opposite from the one in the
contact world. This is not unexpected, as orientation reversal is the norm
when dening any contact invariant in Heegaard Floer homology.
As we briey explained in Section 1, we will also dene a surgery pro-
cedure which we call joining, and which generalizes this gluing operation.
We will associate a map on sutured Floer homology to such a surgery in
Section 4.2.
First we dene some preliminary notions.
Denition 2.7. The mirror of a partially sutured manifold J = (W, , T),
where T = (F, ) is J = (W, , T). Alternatively, it is a partially
sutured manifold (W

, T

), with an orientation reversing dieomorphism


: W W

, such that:
F is sent to F

(orientation is reversed).
is sent to

(orientation is preserved).
R
+
() is sent to R

), and vice versa.


S
+
() is sent to S

), and vice versa.


Whenever we talk about a pair of mirrors, we will implicitly assume that
a specic dieomorphism between them has been chosen. An example is
shown in Figure 5.
There are two partially sutured manifolds, which will play an important
role.
Denition 2.8. A positive (respectively negative) twisting slice along the
sutured surface T = (F, ) is the partially sutured manifold T J
F,
=
(F [0, 1], , T T) where we identify T with F 0, and T with
F 1. The dividing set is obtained from [0, 1] by applying
1
n

th of a positive (respectively negative) Dehn twist along each component of


12 RUMEN ZAREV
T T
[0, 1]
(a) T J
F,+
T T
[0, 1]
(b) T J
F,
Figure 6. Positive and negative twisting slices. The dividing sets are [0, 1],
after a fractional Dehn twist has been applied. The R
+
regions have been
shaded.
F
1
2
, containing n points of . (The twists might be dierent for
dierent components.)
Examples of twisting slices are shown in Figure 6.
Denition 2.9. Let }
1
and }
2
be sutured manifolds, and J = (W, , T)
be partially sutured. Suppose there are embeddings J }
1
and J }
2
.
We will call the new sutured manifold
}
1

W
}
2
= (}
1
J)
F
T J
F,+

F
(}
2
J)
the join of }
1
and }
2
along J.
Intuitively, this means that we cut out J and J and concatenate the
complements together. There is a mismatch of R
+
with R

along the bound-


ary, so we introduce a positive twist to x it. An example of gluing was
shown in Figure 2.
Another important operation is gluing.
Denition 2.10. Suppose that }
1
= (Y
1
,
1
, T) and }
2
= (Y
2
,
2
, T) are
two partially sutured manifolds, and
0
is a dividing set for T = (F, ). We
dene the gluing of the sutured manifolds (Y
1
,
1


0
) and (Y
2
,
2


0
)
along (F,
0
) to be the concatenation
}
1

F
T J
F,+

F
}
2
,
and denote it by
(Y
1
,
1

0
)
(F,
0
)
(Y
2
,
2

0
).
An example of gluing was shown in Figure 1. It is easy to see that gluing
is a special case of the join. Recall that the concatenation (Y,

, T)
F
J

is the sutured manifold (Y,

). Thus we can identify gluing along (F,


0
)
with join along J

0
.
Another useful object is the double of a partially sutured manifold.
JOINING AND GLUING SFH 13
Denition 2.11. Given a partially sutured manifold J = (W, , T), where
T = (F, ), dene the double of J to be the be sutured manifold obtained
by concatenation as follows:
D(J) = J
F
T J
F,

F
J.
All the operations we have dened so far keep us in the realm of balanced
sutured manifolds.
Proposition 2.12. If we join or glue two balanced sutured manifolds to-
gether, the result is balanced. The double of any partially sutured manifold
J is balanced.
Proof. There are three key observations. The rst one is that (R
+
)(R

)
is additive under concatenation. The second is that when passing from J
to its mirror J, the values of (R
+
) and (R

) are interchanged. Finally,


for positive and negative twisting slices (R
+
) = (R

).
The operations of joining and gluing sutured manifolds have good formal
properties described in the following proposition.
Proposition 2.13. The join satises the following:
(1) Commutativity: }
1

W
}
2
is canonically dieomorphic to }
2

W
}
1
.
(2) Associativity: If there are embeddings J }
1
, (J J

) }
2
,
and J

J
3
then there are canonical dieomorphisms
(}
1

W
}
2
)
W
}
3

= }
1

W
(}
2

W
}
3
)

= (}
1
}
3
)
WW
}
2
.
(3) Identity: }
W
T(J)

= }.
Gluing satises analogous properties.
Proof. These facts follow immediately from the denitions.
3. Bordered sutured Floer homology
We recall the denitions of bordered sutured manifolds and their invari-
ants, as introduced in [Zar09].
3.1. Arc diagrams and bordered sutured manifolds. Parametriza-
tions by arc diagrams, as described below are a slight generalization of those
originally dened in [Zar09]. The latter corresponded to parametrizations
using only arcs. While this is sucient to dene invariants for all possible
situations, it is somewhat restrictive computationally. Indeed, to dene the
join map we need to exploit some symmetries that are not apparent unless
we also allow parametrizations using arcs.
Denition 3.1. An arc diagram of rank k is a triple : = (Z, a, M) con-
sisting of the following:
A nite collection Z of oriented arcs.
14 RUMEN ZAREV
A collection of points a = a
1
, . . . , a
2k
Z.
A 2to1 matching M: a 1, . . . , k of the points into pairs.
A type: or .
We require that the 1manifold obtained by performing surgery on all the
0spheres M
1
(i) in Z has no closed components.
We represent arc diagrams graphically by a graph G(:), which consists of
the arcs Z, oriented upwards, and an arc e
i
attached at the pair M
1
(i) Z,
for i = 1, . . . , k. Depending on whether the diagram is of or type, we
draw the arcs to the right or to the left, respectively.
Denition 3.2. The sutured surface T(:) = (F(:), (:)) associated to
the arc diagram : is constructed in the following way. The underlying
surface F is produced from the product Z[0, 1] by attaching 1handles along
the 0spheres M
1
(i) 0, for i = 1, . . . , k. The sutures are = Z
1/2, with the positive region S
+
being above, i.e. containing Z 1.
The sutured surface associated to a arc diagram is constructed in the
same fashion, except that the 1handles are attached on top, i.e. at
/
1
(i) 1. The positive region S
+
is still above.
Suppose F is a surface with boundary, G(:) is properly embedded in
F, and = G(:) F are the vertices of valence 1. If F deformation
retracts onto G(:), we can identify (F, ) with T(:). In fact, the embed-
ding uniquely determines such an identication, up to isotopies xing the
boundary. We say that : parametrizes (F, ).
As mentioned earlier, all arc diagrams considered in [Zar09] are of type.
Let : = (Z, a, M) be an arc diagram. We will denote by : the diagram
obtained by reversing the orientation of Z (and preserving the type). We
will denote by : the diagram obtained by switching the typefrom to
, or vice versaand preserving the triple (Z, a, M). There are now four
related diagrams: :, :, :, and :. The notation is intentionally similar
to the one for the variations on a sutured surface. Indeed, they are related
as follows:
T(:) = T(:), T(:) = T(:).
To illustrate that, Figure 7 has four variations of an arc diagram of rank
3. Figure 8 shows the corresponding parametrizations of sutured surfaces,
which are all tori with one boundary component and four sutures. Notice
the embedding of the graph in each case.
Denition 3.3. A bordered sutured manifold } = (Y, , :) is a partially
sutured manifold (Y, , T), whose sutured surface T has been parametrized
by the arc diagram :.
As with partially sutured manifolds, } = (Y, , :
1
:
2
) can be thought
of as a cobordism from T(:
1
) to T(:
2
).
JOINING AND GLUING SFH 15
e
1
e
2
e
3
(a) : of -type
e
1
e
2
e
3
(b) : of -type
e
1
e
2
e
3
(c) : of -type
e
1
e
2
e
3
(d) : of -type
Figure 7. Four variants of an arc diagram
S
+
S
+
S

+
+

(a) T(:)
S
+
S
+
S

+
+

(b) T(:)
S

S
+
+
+

(c) T(:)
S

S
+
+
+

(d) T(:)
Figure 8. Parametrizations of surfaces by the arc diagrams in Figure 7
3.2. The bordered algebra. We will briey recall the denition of the
algebra /(:) associated to an type arc diagram :. Fix a diagram : =
(Z, a, M) of rank k. First, we dene a larger strands algebra /

(Z, a), which


is independent of the matching M. Then we dene /(:) as a subalgebra of
/

(Z, a).
Denition 3.4. The strands algebra associated to (Z, a) is a Z/2algebra
/

(Z, a), which is generated (as a vector space) by diagrams in [0, 1] Z of


the following type. Each diagram consists of several embedded oriented arcs
or strands, starting in 0 a and ending in 1 a. All tangent vectors
on the strands should project non-negatively on Z, i.e. they are upward-
veering. Only transverse intersections are allowed.
The diagrams are subjects to two relationsany two diagrams related by
a Reidemeister III move represent the same element in /

(Z, a), and any


diagram in which two strands intersect more than once represents zero.
Multiplication is given by concatenation of diagrams in the [0, 1]direction,
provided the endpoints of the strands agree. Otherwise the product is zero.
The dierential of a diagram is the sum of all diagrams obtained from it by
taking the oriented resolution of a crossing.
16 RUMEN ZAREV
(a) a
1
(b) a
2
(c) a
3
(d) a
4
Figure 9. Four generators of /(:).
We refer to a strand connecting (0, a) to (1, a) for some a a as horizontal.
Notice that the idempotent elements of /

(Z, a) are precisely those which are


sums of diagrams with only horizontal strands. To recover the information
carried by the matching M we single out some of these idempotents.
Denition 3.5. The ground ring J(:) associated to : is a ground ring,
in the sense of Denition A.1, of rank 2
k
over Z/2, with canonical basis
(
I
)
I{1,...,k}
. It is identied with a subring of the strands algebra /

(Z, a),
by setting
I
=

J
D
J
. The sum is over all J a such that M[
J
: J I
is a bijection, and D
J
is the diagram with horizontal strands [0, 1] J.
For all I 1, . . . , k, the generator
I
is a sum of 2
#I
diagrams.
Denition 3.6. The bordered algebra /(:) associated to : is the subal-
gebra of J(:) /

(Z, a) J(:) consisting of all elements subject to the fol-


lowing condition. Suppose M(a) = M(b), and D and D

are two diagrams,


where D

is obtained from D by replacing the horizontal arc [0, 1]a by the


horizontal arc [0, 1] b. Then contains D as a summand i it contains
D

as a summand.
We use J(:) as the ground ring for /(:), in the sense of Denition B.3.
The condition in Denition 3.6 ensures that the canonical basis elements of
J(:) are indecomposable in /(:).
It is straightforward to check that Denition 3.6 is equivalent to the def-
inition of /(:) in [Zar09].
Examples of several algebra elements are given in Figure 9. The dotted
lines on the side are given to remind us of the matching in the arc diagram
:. All strands are oriented left to right, so we avoid drawing them with
arrows. The horizontal lines in Figure 9b are dotted, as a shorthand for the
sum of two diagrams, with a single horizontal line each. For the elements in
this example, we have a
1
a
2
= a
3
, and a
1
= a
4
.
The situation for arc diagrams of type is completely analogous, with
one important dierence.
JOINING AND GLUING SFH 17
(a) /(:) (b) /(:) (c) /(:) (d) /(:)
Figure 10. Four elements in the algebras :, :, :, and :, which corre-
spond to each other.
Denition 3.7. The bordered algebra /(:) associated to a arc diagram
:, is dened in the exact same way as in Denitions 3.6, except that moving
strands are downward veering, instead of upward.
The relationship between the dierent types of algebras is summarized in
the following proposition.
Proposition 3.8. Suppose : is an arc diagram of either or type. The
algebras associated to :, :, :, and : are related as follows:
/(:)

= /(:)

= /(:)
op
,
/(:)

= /(:).
Here A
op
denotes the opposite algebra of A. That is, an algebra with
the same additive structure and dierential, but the order of multiplication
reversed.
Proof. This is easily seen by reecting and rotating diagrams. To get from
/(:) to /(:) we have to rotate all diagrams by 180 degrees. This means
that multiplication switches order, so we get the opposite algebra.
To get from /(:) to /(:) we have to reect all diagrams along the
vertical axis. This again means that multiplication switches order.
An example of the correspondence is shown in Figure 10.
3.3. The bordered invariants. We will give a brief sketch of the deni-
tions of the bordered invariants from [Zar09], which apply for the case of
arc diagrams. Then we discuss the necessary modications when arcs
are involved.
For now assume : = (Z, a, M) is an arc diagram.
Denition 3.9. A bordered sutured Heegaard diagram 1 = (, , , :)
consists of the following:
A compact surface with no closed components.
18 RUMEN ZAREV
A collection of circles
c
and a collection of arcs
a
, which are
pairwise disjoint and properly embedded in . We set =
a

c
.
A collection of disjoint circles , properly embedded in .
An embedding G(:) , such that Z is sent into , preserving
orientation, while
a
is the image of the arcs e
i
in G(:).
We require that
0
(Z)
0
((
c

a
)) and
0
(Z)
0
()
be surjective.
To such a diagram we can associate a bordered sutured manifold (Y, , :)
as follows. We obtain Y from [0, 1] by gluing 2handles to 1
and
c
0. The dividing set is = ( Z) 1/2, and F(:) is a
neighborhood of Z [0, 1]
a
0.
As proved in [Zar09], for every bordered sutured manifold there is a unique
Heegaard diagram, up to isotopy and some moves.
The bordered invariants are certain homotopy-equivalence classes of /

modules (see Appendix B). For a given Heegaard diagram 1, we can form
the set of generators ((1) consisting of collections of intersection points of
.
The invariant

BSA(1)
A(Z)
is a right typeA /

module over /(:), with


Z/2basis ((1). The ground ring J(:) acts as follows. The only idempotent
in J(:) which acts nontrivially on x ((1) is
I(x)
where I(x) 1, . . . , k
records the arcs which contain a point of x.
The structure map m of

BSA(1) counts certain holomorphic curves in
Int [0, 1] R, with boundary on ( 1 R) ( 0 R). Each
such curve has two types of asymptoticsends at () [0, 1] , and
ends at 0 h where h R is nite. The possible ends at are
in 1-to-1 correspondence with elements of /(:).
The expression m(x, a
1
, . . . , a
n
), y

counts curves as above, which have


asymptotics x [0, 1] at , y [0, 1] at +, and a
1
, a
2
, . . . , a
n
at some
nite values h
1
< h
2
< . . . < h
n
.
We write

BSA(}) for the homotopy equivalence class of

BSA(1). (In-
variance was proven in [Zar09].)
The invariant
A(Z)

BSD(1) is a left typeD /

module over /(:) =


/(:)
op
, with Z/2basis ((1). (See Appendix B.2 for typeD modules, and
the meaning of upper and lower indices). The ground ring J(:) acts as
follows. The only idempotent in J(:) which acts nontrivially on x ((1)
is
I
c
(x)
where I
c
(x) 1, . . . , k records the arcs which do not contain a
point of x.
The structure map of

BSD(1) counts a subset of the same holomorphic
curves as for

BSA(1). Their interpretation is somewhat dierent, though.
Equivalently,
A(Z)
op

BSD(1) =

BSA(1)
A(Z)

A(Z),A(Z)
op
I, where I is a cer-
tain bimodule dened in [LOT10a].
Again, we write

BSD(}) for the homotopy equivalence class of

BSD(1).
(Invariance was proven in [Zar09].)
JOINING AND GLUING SFH 19
We can also construct invariants
A(Z)
op

BSA(}) and

BSD(})
A(Z)
purely
algebraically from the usual

BSA and

BSD. Indeed, as discussed in Appen-
dix B.6, any right Amodule is a leftA
op
module and vice versa.
If } is bordered by T(:
1
) T(:
2
), we can similarly dene several bimod-
ules invariants for }:
A(Z
1
)
op

BSAA(})
A(Z
2
)
A(Z
1
)
op

BSDA(})
A(Z
2
)
A(Z
1
)
op

BSAD(})
A(Z
2
) A(Z
1
)
op

BSDD(})
A(Z
2
)
For the invariants of diagrams little changes. Suppose : is a type
arc diagram. Heegaard diagrams will now involve arcs as the images
of e
i
G(:), instead of arcs. We still count holomorphic curves in
Int [0, 1] R. However, since there are curves hitting instead of ,
the asymptotic ends at 1h are replaced by ends at 0h,
which again correspond to elements of /(:). The rest of the denition is
essentially unchanged.
The last case is when } is bordered by T(:
1
) T(:
2
), where :
1
is a
diagram of type and :
2
is of type. We can extend the denition of

BSAA(}) as before. There are now four types of asymptotic ends:


The ones at which correspond to generators x, y ((1).
1 h (or ends) which correspond to /(:
1
).
0 h (or ends) which correspond to /(:
2
).
Each holomorphic curve will have some number k 0 of ends, and some
number l 0 of ends. Such a curve contributes to the structure map
m
k|1|l
which takes k elements of /(:
1
) and l elements of /(:
2
).
To summarize we have the following theorem.
Theorem 3.10. Let } be a bordered sutured manifold, bordered by T(:
1
)
T(:
2
), where :
1
and :
2
can be any combination of and types. Then
there are bimodules, well dened up to homotopy equivalence:
A(Z
1
)

BSAA(})
A(Z
2
)
A(Z
1
)

BSDA(})
A(Z
2
)
A(Z
1
)

BSAD(})
A(Z
2
) A(Z
1
)

BSDD(})
A(Z
2
)
If }
1
and }
2
are two such manifolds, bordered by T(:
1
) T(:
2
) and
T(:
2
) T(:
3
), respectively, then there are homotopy equivalences

BSAA(}
1
}
2
)

BSAA(}
1
)
A(Z
2
)

BSDA(}
2
),

BSDA(}
1
}
2
)

BSDD(}
1
)
A(Z
2
)

BSAA(}
2
),
etc. Any combination of bimodules for }
1
and }
2
can be used, where one is
typeA for /(:
2
), and the other is typeD for /(:
2
).
The latter statement is referred to as the pairing theorem. The proof of
Proposition 3.10 is a straightforward adaptation of the corresponding proofs
when dealing with only type diagrams. An analogous construction involv-
ing both and arcs in the purely bordered setting is given in [LOT10b].
20 RUMEN ZAREV
3.4. Mirrors and twisting slices. In this section we give two compu-
tations of bordered invariants. One of them relates the invariants for a
bordered sutured manifold J and its mirror J. The other gives the
invariants for a positive and negative twisting slice.
Recall that if J = (W, , T(:)), its mirror is J = (W, , T(:)) =
(W, , T(:)).
Proposition 3.11. Let J and J be as above. Let M
A(Z)
be a repre-
sentative for the homotopy equivalence class

BSA(J)
A(Z)
. Then its dual
A(Z)
M

is a representative for
A(Z)

BSA(J). Similarly, there are homo-
topy equivalences
_

BSD(J)
A(Z)
_

A(Z)

BSD(J),
_
A(Z)
op

BSA(J)
_


BSA(J)
A(Z)
op
,
_
A(Z)
op

BSD(J)
_


BSD(J)
A(Z)
op
.
A similar statement holds for bimodulesif J is bordered by T(:
1
)
T(:
2
), then the corresponding bimodule invariants of J and J are duals
of each other.
Proof. We prove one case. All others follow by analogy. Let 1 = (, , , :)
be a Heegaard diagram for J. Let 1

= (, , , :) be the diagram ob-


tained by switching all and curves. (Note that if : was an type
diagram, this turns it into the type diagram :, and vice versa.)
The bordered sutured manifold described by 1

is precisely J. Indeed,
it is obtained from the same manifold [0, 1] by attaching all 2handles on
the opposite side, and taking the sutured surface T also on the opposite side.
This is equivalent to reversing the orientation of W, while keeping the orien-
tations of and Z the same. (Compare to [HKM07], where the
EHinvariant for contact structures on (Y, ) is dened in SFH(Y, +).)
The generators ((1) and ((1

) of the two diagrams are the same. There


is also a 1to1 correspondence between the holomorphic curves u in the
denition of

BSA(1)
A(Z)
and the curves u

in the denition of

BSA(1

)
A(Z)
.
This is given by reecting both the [0, 1]factor and the Rfactor in the
domain Int [0, 1]R. The asymptotic ends are reversed. The ends
of u are sent to the ends of u

, and vice versa, while their heights h on the


Rscale are reversed. When turning ends to ends, the corresponding
elements of /(:) are reected (as in the correspondence /(:)

= /(:)
op
from Proposition 3.8).
This implies the following relation between the structure maps m of

BSA(1) and m

of

BSA(1

):

m(x, a
1
, . . . , a
n
), y

_
=

(y

, a
op
n
, . . . , a
op
1
), x

_
.
JOINING AND GLUING SFH 21
Turning

BSA(1

) into a left module over (/(:)


op
)
op
= /(:), we get the
relation

m(x, a
1
, . . . , a
n
), y

_
=

(a
1
, . . . , a
n
, y

), x

_
.
This is precisely the statement that

BSA(1)
A(Z)
and
A(Z)

BSA(1

) are
duals, with ((1) and ((1

) as dual bases.
A similar statement for purely bordered invariants is proven in [LOT10b].
Proposition 3.12. Let : be any arc diagram, and let A = /(:). The
twisting slices T J
F(Z),
are bordered by T(:) T(:). They have bi-
module invariants
A

BSAA(T J
F(Z),
)
A

A
A
A
,
A

BSAA(T J
F(Z),+
)
A

A
A

A
.
Proof. Since T J
F(Z),
are mirrors of each other, by Proposition 3.11, it is
enough to prove the rst equivalence. The key ingredient is a very convenient
nice diagram 1 for T J
F(Z),
. This diagram was discovered by the author,
and independently by Auroux in [Aur], where it appears in a rather dierent
setting.
Recall from [Zar09] that a nice diagram is a diagram, (, , , :) where
each region of ( ) is either a boundary region, a rectangle, or a
bigon. The denition trivially extends to the current more general setting.
Nice diagrams can still be used to combinatorially compute bordered sutured
invariants.
The diagram is obtained as follows. For concreteness assume that : is
of type. To construct the Heegaard surface , start with several squares
[0, 1] [0, 1], one for each component Z Z. There are three identications
of Z with sides of the squares:
sending Z to the left sides 0 [0, 1], oriented from 0 to 1.

sending Z to the right sides 1 [0, 1], oriented from 1 to 0.


sending Z to the top sides [0, 1] 1, oriented from 1 to 0.
For each matched pair a, b = M
1
(i) a Z, attach a 1handle at
(a, b). Add an arc
a
i
from (a) to (b), and a arc
a
i
from

(a) to

(b), both running through the handle corresponding to a, b. To see that


this gives the correct manifold, notice that there are no or circles, so
the manifold is topologically [0, 1]. The pattern of attachment of the
1handles shows that = F(:). It is easy to check that and the arcs are
in the correct positions.
This construction is demonstrated in Figure 11. The gure corresponds
to the arc diagram : from Figure 7c.
Calculations with the same diagram in [Aur] and [LOT10b] show that
the bimodule

BSAA(1) is indeed the algebra A as a bimodule over itself.
While the statements in those cases are not about bordered sutured Floer
homology, the argument is purely combinatorial and caries over completely.
We give a brief summary of this argument. Intersection points in
are of two types:
22 RUMEN ZAREV
:
:
Figure 11. Heegaard diagram for a negative twisted slice T J
F,
.

(a) Dierential.
=
(b) Left action.
=
(c) Right action.
Figure 12. Examples of domains counted in the diagram for T J
F,
. In each
case the domain goes from the black dots to the white dots. Below them we
show the corresponding operations on the algebra.
x
i

a
i

a
i
inside the 1handle corresponding to M
1
(i), for i
1, . . . , k. The point x
i
corresponds to the two horizontal strands
[0, 1] M
1
(i) in /(:).
y
ab

a
M(a)

a
M(b)
, inside the square regions of 1. The point y
ab
corresponds to a strand (0, a) (1, b) (or a b for short) in /(:).
The allowed combinations of intersection points correspond to the allowed
diagrams in /(:), so

BSA(1)

= /(:) as a Z/2vector space.


Since 1 is a nice diagram the dierential counts embedded rectangles in
1, with sides on and . The rectangle with corners (y
ad
, y
bc
, y
ac
, y
ad
)
corresponds to resolving the crossing between the strands a d and b c
(getting a c and b d).
The left action m
1|1|0
of A counts rectangles hitting the :part of the
boundary. The rectangle with corners ((a), y
ac
, y
bc
, (b)) corresponds to
concatenating the strands a b and b c (getting a c). The right
action is similar, with rectangles hitting the :part of the boundary.
Some examples of domains in 1 contributing to m
0|1|0
, m
1|1|0
, and m
0|1|1
are shown in Figure 12. They are for the diagram 1 from Figure 11.
JOINING AND GLUING SFH 23
= =

M

M
m
M
A MM

A
A

A M M A
A
A M M A
A
Figure 13. Denition of the join map .
4. The join map
In this section we will dene the join and gluing maps, and prove some
basic properties. Recall that the gluing operation is dened as a special case
of the join operation. The gluing map is similarly a special case of the join
map. Thus for the most part we will only talk about the general case, i.e.
the join map.
4.1. The algebraic map. We will rst dene an abstract algebraic map,
on the level of /

modules.
Let A be a dierential graded algebra, and
A
M be a left /

module over
it. As discussed in Appendix B.6, the dual M

A
is a right /

module over
A. Thus
A
(MM

)
A
is an /

bimodule. On the other hand, since A is a


bimodule over itself, so is its dual
A
A

A
. We dene a map M M

which is an /

analog of the natural pairing of a module and its dual.


Denition 4.1. The algebraic join map
M
:
A
(M M

)
A

A
A

or just when unambiguousis an /

bimodule morphism, dened as


follows. It is the unique morphism satisfying
(1)

i|1|j
(a
1
, . . . , a
i
, p, q

, , a

1
, . . . , a

j
), a

_
=

m
i+j+1|1
(a

1
, . . . , a

j
, a

, a
1
, . . . , a
i
, p), q

_
,
for any i, j 0, p M, q

, and a

A.
Eq. (1) is best represented diagrammatically, as in Figure 13. Note that

M
is a bounded morphism if and only if M is a bounded module.
As discussed in Appendix B.4, morphisms of /

modules form chain


complexes, where cycles are homomorphisms. Only homomorphisms de-
scend to maps on homology.
Proposition 4.2. For any
A
M, the join map
M
is a homomorphism.
Proof. It is a straightforward but tedious computation to see that
M
= 0
is equivalent to the structure equation for m
M
.
A more enlightening way to see this is to notice that by turning the
diagram in Figure 13 partly sideways, we get a diagram for the homotopy
equivalence h
M
: A

M M, shown in Figure 14. Taking the dierential

M
and turning the resulting diagrams sideways, we get precisely h
M
.
We know that h
M
is a homomorphism and, so h
M
= 0.
The equivalences are presented in Figure 15.
24 RUMEN ZAREV
= h
M
m
M
A A A M
M
A A A M
M
Figure 14. The homotopy equivalence h
M
: A

M M.
+ + + +

A

A

M

M

M

M
m
M
m
M
(a) The dierential
M
which needs to vanish to show that
M
is an /

bimodule homomorphism.
+ + + +

A
h
M
h
M
h
M
h
M
h
M
m
M
m
M
(b) The dierential h
M
of the homotopy equivalence h
M
.
Figure 15. Proof that is a homomorphism, by rotating diagrams.
We will prove two naturallity statements about that together imply
that descends to a well dened map on the derived category. The rst
shows that is natural with respect to isomorphisms in the derived cat-
egory of the DG-algebra A, i.e. homotopy equivalences of modules. The
second shows that is natural with respect to equivalences of derived cat-
egories. (Recall from [Zar09] that dierent algebras corresponding to the
same sutured surface are derived-equivalent.)
Proposition 4.3. Suppose
A
M and
A
N are two /

modules over A, such


that there are inverse homotopy equivalences : M N and : N M.
Then there is an /

homotopy equivalence of A, Abimodules


: M M

N N

,
and the following diagram commutes up to /

homotopy:
M M

M
))
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
N N


N
//
A

.
JOINING AND GLUING SFH 25
Proposition 4.4. Suppose A and B are dierential graded algebras, and
B
X
A
and
A
Y
B
are two typeDA bimodules, which are quasi-inverses. That
is, there are /

homotopy equivalences
A
(Y X)
A

A
I
A
,
B
(X Y )
B

B
I
B
.
Moreover, suppose H

(B

) and H

(XA

) have the same rank (over


Z/2).
Then there is a B, Bbimodule homotopy equivalence

X
: X A

.
Moreover, for any /

module
A
M, such that X M is well dened, the
following diagram commutes up to /

homotopy:
X M M

id
X

M
id
X

XM
**
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
X A


X
//
B

.
Notice the condition that X M be well dened. This can be satised
for example if M is a bounded module, or if X is reletively bounded in A
with respect to B. Before proving Propositions 4.3 and 4.4 in Section 4.3,
we will use them to dene the join .
4.2. The geometric map. Suppose that }
1
and }
2
are two sutured mani-
folds, and J = (W, , T) is a partially sutured manifold, with embeddings
J }
1
and J }
2
. Let : be any arc diagram parametrizing the sur-
face T. Recall that J = (W, , T). Also recall the twisting slice
T J
F,+
, from Denition 2.8. The join }
1

W
}
2
of }
1
and }
2
along J was
dened as
}
1

W
}
2
= (}
1
J)
F
T J
F,+

F
(}
2
J).
Let A = /(:) be the algebra associated to :. Let
A
M, U
A
, and
A
V
be representatives for the bordered sutured modules
A

BSA(J),

BSD(}
1

J)
A
, and
A

BSD(}
2
J), respectively such that U M and M

V
are well-dened. (Recall that the modules are only dened up to homotopy
equivalence, and that the product is only dened under some boundedness
conditions.) We proved in Proposition 3.11 that M

A
is a representative
for

BSA(J)
A
, and in Proposition 3.12 that
A
A

A
is a representative for

BSAA(T J
F,+
).
From the K unneth formula for SFH of a disjoint union, and from Theo-
rem 3.10, we have the following homotopy equivalences of chain complexes.
SFC(}
1
}
2
)

= SFC(}
1
) SFC(}
2
)

BSD(}
1
J)
A

BSA(J)
_

BSA(J)
A

BSD(}
2
J)
_
U
A

A
(M M

)
A

A
V.
26 RUMEN ZAREV
SFC(}
1

W
}
2
)


BSD(}
1
J)
A

BSAA(T J
F,+
)
A

BSD(}
2
J)
U
A

A
A

A

A
V.
Denition 4.5. Let }
1
, }
2
and J be as described above. Dene the geo-
metric join map

M
: SFC(}
1
) SFC(}
2
) SFC(}
1

W
}
2
)
by the formula
(2)
M
= id
U

M
id
V
: U M M

V U A

V.
Note that such an induced map is not generally well dened (it might
involve an innite sum). In this case, however, we have made some bound-
edness assumptions. Since U M and M

V are dened, either M must


be bounded, or both of U and V must be bounded. In the former case,
M
is also bounded. Either of these situations guarantees that the sum dening

M
is nite.
Theorem 4.6. The map
M
from Denition 4.5 is, up to homotopy, in-
dependent on the choice of parametrization :, and on the choices of repre-
sentatives M, U, and V .
Proof. First, we will give a more precise version of the statement. Let :

be
any other parametrization of T, with B = /(:

), and let
B
M

, U
B
and
B
V

, be representatives for the respective bordered sutured modules. Then


there are homotopy equivalences and making the following diagram
commute up to /

homotopy:
U M M

//

U A

//
U

.
The proof can be broken up into several steps. The rst step is indepen-
dence from the choice of U and V , given a xed choice for A and M. This
follows directly from the fact id and id are DG-functors.
The second step is to show independence from the choice of M, for xed
A, U, and V . This follows from Proposition 4.3. Indeed, suppose : M
M

is a homotopy equivalence with homotopy inverse : M

M. Then

: M

is also a homotopy equivalence inducing the homotopy


equivalence
id
U

id
V
: U M M

V U M

V.
By Proposition 4.3,
M

M
(

), which implies
id
U

M
id
V
(id
U

M
id
V
) (id
U

id
V
).
The nal step is to show independence from the choice of algebra A. We
will cut }
1
and }
2
into several pieces, so we can evaluate the two dierent
versions of from the same geometric picture.
JOINING AND GLUING SFH 27
T T

J
J

T Q
Figure 16. The various pieces produced by slicing J at two surfaces parallel
to T.
Let T

and T

be two parallel copies of T in J, which cut out


J

= (W

, T

) and J

= (W

, T

), where J

J.
Let T = J

and Q = J J

(see Figure 16). Both T and Q are


topologically F [0, 1].
Parametrize T and T

by :, and T

by :

, where /(:) = A, and


/(:

) = B. Let
B
X
A
and
A
Y
B
be representatives for
B

BSAD(T)
A
and
A

BSAD(Q)
B
, respectively. Note that Q
F
T is a product bordered sutured
manifold, and thus has trivial invariant
A

BSAD(Q T)
A

A
I
A
. By the
pairing theorem, this implies Y X
A
I
A
. Similarly, by stacking T and Q
in the opposite order we get X Y
B
I
B
.
There are embeddings J

, J

}
1
and J

, J

}
2
and two
distinct ways to cut and glue them together, getting }
1

W
}
2

= }
1

W
}
2
.
This is illustrated schematically in Figure 17.
Let
A
M be a representative for
A

BSA(J

). By the pairing theorem,


B
(X M) is a representative for
B

BSA(J

). Notice that T J
F

,+

= T
T J
F

,+
T and
B
B

B
and
B
(XA

)
B
are both representatives for
its

BSAA invariant. In particular, they have the same homology. Finally,
let U
B
and
B
V be representatives for

BSD(}
1
J

)
B
and
B

BSD(}
2
J

),
respectively.
The two join maps
M
and
XM
are described by the following equa-
tions.

M
= id
UX

M
id
X

V
:
(U X) M M

(X

V ) (U X) A

(X

V ),

XM
= id
U

XM
id
V
:
U (X M) (M

) V U B

V.
We can apply Proposition 4.4. The boundedness condition can be satised
by requiring that X and Y are bounded modules. There is a homotopy
28 RUMEN ZAREV
}
1
J

T J

T }
2
J

(a) Cutting }
1
and }
2
in two dierent places.
}
1
J

T
T J
F

,+ T }
2
J

(b) The join by J

.
}
1
J

T J
F

,+ }
2
J

(c) The join by J

.
Figure 17. Two ways of cutting and pasting to get the join of }
1
and }
2
.
equivalence
X
: X A

B, and a homotopy
XM

X

(id
X

M
id
X
). These induce a homotopy
(id
U

X
id
V
)
M
= id
U
(
X
(id
X

M
id
X
)) id
V
id
U

XM
id
V
=
XM
.
This nishes the last step. Combining all three gives complete invariance.
Thus we can refer to
W
from now on.
4.3. Proof of algebraic invariance. In this section we prove Proposi-
tions 4.3 and 4.4.
Proof of Proposition 4.3. The proof will be mostly diagrammatic. There are
two modules
A
M and
A
N, and two inverse homotopy equivalences, : M
N and : N M. The dualizing functor
A
Mod Mod
A
is a DG-functor.
Thus it is easy to see that

= ( id
N
) (id
M

)
is also a homotopy equivalence. Let H: M M be the homotopy between
id
M
and .
We have to show that the homomorphism
(3)
M
+
N
(

)
is null-homotopic (see Figure 18a). Again, it helps if we turn the diagram
sideways, where bar resolutions come into play. Let h
M
: A

M M and
h
N
: A

N N be the natural homotopy equivalences.
JOINING AND GLUING SFH 29
+
m
M
m
N


(a) Representation of Eq. (3).
+
h
M
h
N

(b) Representation of Eq. (4).


+

h
M
H
(c) Null-homotopy of (4).
+

m
M
H
(d) Null-homotopy of (3).
Figure 18. Diagrams from the proof of Proposition 4.3.
Turning the rst term in Eq. (3) sideways, we get h
M
. Turning the second
term sideways we get h
N
(id
A

). Thus we need to show that
(4) h
M
+ h
N
(id
A

)
is null-homotopic (see Figure 18b).
There is a canonical homotopy h

: A

M N between h
M
and
h
N
(id
A

), given by
h

(a
1
, . . . , a
i
, (a

, a

1
, . . . , a

j
, m)) = (a
1
, . . . , a
i
, a

, a

1
, . . . , a

j
, m).
Thus we can build the null-homotopy h

+ H h
M
(see Figure 18c).
Indeed,
( h

) = h
M
+ h
N
(id
A

),
(H h
M
) = id
M
h
M
+ h
M
.
Alternatively, we can express the null-homotopy of the expression (3)
directly as in Figure 18d.
Proof of Proposition 4.4. Recall the statement of Proposition 4.4. We are
given two dierential graded algebras A and B, and three modules
B
X
A
,
A
Y
B
, and
A
M. We assume that there are homotopy equivalences X Y
B
I
B
and Y X
A
I
A
, and that X A

and B

have homologies of
the same rank.
We have to construct a homotopy equivalence
X
: X A

,
and a homotopy
XM

X
(id
X

M
id
X
).
30 RUMEN ZAREV

X
B X A X B
B
(a) Denition of .
= h
X

X
B B B X
X A
B B B X
X A
(b) View as a map B

X X.
Figure 19. Two views of the homotopy equivalence from Eq. (5).
= =

X
m
M

X
m
M

X
m
X

M
B X M M X B
B
B XM MX B
B
B XM MX B
B
Figure 20. Equality of the direct and induced maps for X M.
We start by constructing the morphism . We can dene it by the fol-
lowing equation:
(5)

(
X
)
i|1|j
(b
1
, . . . , b
i
, (x, a

, x

), b

1
, . . . , b

j
), b

_
=

i+j+1|1|1
(b

1
, . . . , b

j
, b

, b
1
, . . . , b
i
, x), (x

, a)

_
.
Again, it is useful to turn it sideways. We can reinterpret
X
as a
morphism of typeAD modules B

X X. In fact, it is precisely the
canonical homotopy equivalence h
X
between the two. Diagrams for
X
and
h
X
are shown in Figure 19. Since the h
X
is a homomorphism, it follows
that
X
is one as well.
Next we show that
XM
is homotopic to
X
(id
X

M
id
X
). They
are in fact equal. This is best seen in Figure 20. We use the fact that
X
and
X
commute with merges and splits.
Finally, we need to show that
X
is a homotopy equivalence. We will do
that by constructing a right homotopy inverse for it. Combined with the
fact that the homologies of the two sides have equal rank, this is enough to
ascertain that it is indeed a homotopy equivalence.
Recall that X Y I. Thus there exist morphisms of typeAD B, B
bimodules f : I XY , and g: XY I, and a null-homotopy H: I I
JOINING AND GLUING SFH 31
of id
I
g f. Note that g

: I

is a map of typeDAmodules,
and (
B
I
B
)

=
B
I
B
.
Let
Y
: Y B

A be dened analogous to
X
. Construct the
homomorphism
= (id
X

Y
id
X
) (f id
B
id
Y
id
X
) (id
I
id
B
g

):
I B

I X A

.
We need to show that
X
is homotopic to id
B
, or more precisely to
the canonical isomorphism : I B

I B

. A graphical representation
of
X
is shown in Figure 21a. It simplies signicantly, due to the
fact that B is a DG-algebra, and
B
only has two nonzero terms. The
simplied version of
X
is shown in Figure 21b. As usual, it helps to
turn the diagram sideways. We can view it as a homomorphism B

I I
of typeAD B, Bbimodules. As can be seen from Figure 21c, we get the
composition
(6) g (h
X
id
Y
) (id
B

f) = g h
XY
(id
B

f): B

I I.
On the other hand, the homomorphism : I B

I B

, if written
sideways, becomes the homotopy equivalence h
I
: B

I I. See Figure 22
for the calculation. In the second step we use some new notation. The caps
on the thick strands denote a map Bar B K to the ground ring, which
is the identity on B
0
, and zero on B
i
for any i > 0. The dots on the I
strands denote the canonical isomorphism of I B

I and B

as modules
over the ground ring.
Finding a null-homotopy for +
X
is equivalent to nding a null-
homotopy B

I I of h
I
+g h
XY
(id
B

f). There is a null-homotopy

f
: B

I B

X Y of f h
I
+ h
XY
(id
B

f). Recall that H was
a null-homotopy of id
I
+g f. Thus we have
(H h
I
+g
f
) = (id
I
h
I
+g f h
I
)
+ (g f h
I
+g h
XY
(id
B

F)
= h
I
+g h
XY
(id
B

F),
giving us the required null-homotopy.
To nish the proof, notice that if
X
is homotopic to id
B
, then it
is a quasi-isomorphism, i.e. a homomorphism whose scalar component is a
quasi-isomorphism of chain complexes. Moreover, when working with Z/2
coecients, as we do, quasi-isomorphisms of /

modules and bimodules


coincide with homotopy equivalences.
In particular we have that (
X
)
0|1|0
= (
X
)
0|1|0

0|1|0
induces an
isomorphism on homology (in this case the identity map on homology). In
particular induces an injection, while
X
induces a surjection. Combined
with the initial assumption that B

and X A

have homologies of
equal rank, this implies that (
X
)
0|1|0
and
0|1|0
induce isomorphisms on
homology. That is,
X
and are quasi-isomorphisms, and so homotopy
32 RUMEN ZAREV

B
f
g
B B
I
B
B
B
I
B B
B
(a) Before simplication.

B
f
g
B B
I
B
B
B
I
B B
B
(b) After simplication.

B
f
g
B B B B
I
B
B
I
B
B
(c) Written sideways.
Figure 21. Three views of
X
: I B

I B

.
JOINING AND GLUING SFH 33
= =

I
h
I
B I B I B
B
B I B I B
B
B I B I B
B
B B B I
I B
Figure 22. The equivalence of the morphism and h
I
.
equivalences. This concludes the proof of Proposition 4.4, and with it, of
Theorem 4.6.
5. Properties of the join map
In this section we give some formulas for the join and gluing maps, and
prove their formal properties.
5.1. Explicit formulas. We have abstractly dened the join map
W
in
terms of

BSA(W)
but so far have not given any explicit formula for it.
Here we give the general formula, as well as some special cases which are
somewhat simpler.
If we want to compute
W
for the join }
1

W
}
2
, we need to pick a
parametrization by an arc diagram :, with associated algebra A, and repre-
sentatives U for

BSD(}
1
)
A
, V for
A

BSD(}
2
), and M for
A

BSA(J). Then
we know SFC(}
1
) = U M, SFC(}
2
) = M

V , and SFC(}
1

W
}
2
) =
U A

V . As given in Denition 4.5, the join map


W
is

W
= id
U

M
id
V
: U M M

V U A

V.
In graphic form this can be seen in Figure 23a.
This general form is not good for computations, especially if we try to
write it algebraically. However
W
has a much simpler form when M is a
DG-type module.
Denition 5.1. An /

module M
A
is of DG-type if it is a DG-module,
i.e., if its structure maps m
1|i
vanish for i 2. A bimodule
A
M
B
is of
DG-type if m
i|1|j
vanish, unless (i, j) is one of (0, 0), (1, 0) or (0, 1) (i.e. it
is a DG-module over AB).
A typeDA bimodule
A
M
B
is of DG-type if
1|1|j
vanish for all j 2. A
typeDD bimodule
A
M
B
is of DG-type if
1|1|1
(x) is always in AX 1+
1 X B (i.e. it is separated). All type Dmodules M
A
are DG-type.
The product of any combination of DG-type modules is also DG-type.
All modules

BSA,

BSD,

BSAA, etc., computed from a nice diagram are of
DG-type.
34 RUMEN ZAREV
Proposition 5.2. Let the manifolds }
1
, }
2
, and J, and the modules U,
V , and M be as in the above discussion. If M is DG-type, the formula for
the join map
W
simplies to:
(7)
W
(u mn

v) =

m
M
(a, m), n

_
u a

v,
where the sum is over a Z/2basis for A. A graphical representation is given
in Figure 23b.
Finally, an even simpler case is that of elementary modules. We will see
later that elementary modules play an important role for gluing, and for the
relationship between the bordered and sutured theories.
Denition 5.3. A typeA module
A
M (or similarly M
A
) is called elemen-
tary if the following conditions hold:
(1) M is generated by a single element m over Z/2.
(2) All structural operations on M vanish (except for multiplication by
an idempotent, which might be identity).
A typeD module
A
M (or M
A
), is called elementary if the following
conditions hold:
(1) M is generated by a single element m over Z/2.
(2) (m) = 0.
Notice that for an elementary module M = 0, m we can decompose m
as a sum m =
1
m+ +
k
m, where (
i
) is the canonical basis of the ground
ring. Thus we must have
i
m = m for some i, and
j
m = 0 for all i ,= j.
Therefore, elementary (left) modules over A are in a 1to1 correspondence
with the canonical basis for its ground ring.
We only use elementary typeA modules in this section but we will need
both types later.
Remark. For the algebras we discuss, the elementary typeA modules are
precisely the simple modules. The elementary typeD modules are the those
A
M for which A M
A
Mod is an elementary projective module.
Proposition 5.4. If
A
M = m, 0 is an elementary module corresponding
to the basis idempotent
M
, then the join map
W
reduces to
(8)
W
(u mm

v) = u
M

v.
Graphically, this is given in Figure 23c.
Moreover, in this case, SFC(}
1
) = U M

= U
M
U and SFC(}
2
) =
M V

=
M
V V as chain complexes.
Proposition 5.2 and Proposition 5.4 follow directly from the denitions of
DG-type and elementary modules.
JOINING AND GLUING SFH 35

U

V
m
M
U
U
V
V
M M
A
(a) The general case.
m
M
U
U
V
V
M M
A
(b) M of DG-type.

M
U
U
V
V
M M
A
(c) M elementary.
Figure 23. Full expression for join map in three cases.
5.2. Formal properties. In this section we will show that the join map
has the formal properties stated in Theorem 1. A more precise statement of
the properties is given below.
Theorem 5.5. The following properties hold:
(1) Let }
1
and }
2
be sutured and J be partially sutured, with embeddings
J }
1
and J }
2
. There are natural identications of the
disjoint unions }
1
}
2
and }
2
}
1
, and of of the joins }
1

W
}
2
and }
2

W
}
1
. Under this identication, there is a homotopy

W

W
.
(2) Let }
1
, }
2
, and }
3
be sutured, and J
1
and J
2
be partially sutured,
such that there are embeddings J
1
}
1
, (J
1
J
2
) }
2
, and
J
2
}
3
. The following diagram commutes up to homotopy:
SFC(}
1
}
2
}
3
)

W
1
//

W
2

W
1
W
2
++
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
SFC(}
1
}
2
}
3
)

W
2

SFC(}
1
}
2
}
3
)

W
1
//
SFC(}
1
}
2
}
3
)
(3) Let J be partially sutured. There is a canonical element [
W
] in
the sutured Floer homology SFH(T(J)) of the double of J. If
is any representative for [
W
], and there is an embedding J },
then
(9)
W
(, ) id
SFC(Y)
: SFC(}) SFC(}).
Proof. We will prove the three parts in order.
For part (1), take representatives U
A
for

BSD(}
1
J),
A
V for

BSD(}
2

J), and
A
M for

BSA(J). The main observation here is that we can turn
left modules into right modules and vice versa, by reecting all diagrams
along the vertical axis (see Appendix B.6). If we reect the entire diagram
for
M
, domain and target chain complexes are turned into isomorphic ones
and we get a new map that is equivalent.
The domain U
A

A
MM

A
V becomes V
A
op

A
opM

M
A
op
A
op
U,
and the target U
A

A
A

A
V becomes V
A
op

A
op(A

)
op
A
op
A
op
U.
36 RUMEN ZAREV
Notice that V
A
op
is

BSD(}
2
J),
A
opU is

BSD(}
1
J), and
A
opM

is

BSA(J). In addition (A

)
op
= (A
op
)

. Since the map


M
is completely
symmetric, when we reect it, we get
M
. Everything else is preserved, so
reecting
W
gives precisely
W
. This nishes part (1).
For part (2), the equivalence is best seen by working with convenient
representatives. Pick the following modules as representatives: U
A
for

BSD(}
1
J
1
),
A
X
B
for

BSDD(}
2
(J
1
J
2
)),
B
V for

BSD(}
1
),
A
M for

BSA(J
1
) and
B
N for

BSD(J
2
). We can always choose M, N, and X to
be of DG-type in the sense of Denition 5.1. Since X is of DG-type, taking
the product with it is associative. (This is only true up to homotopy in
general). Since M and N are DG-type, we can apply Proposition 5.2 to get
formulas for
W
1
and
W
2
. The two possible compositions are shown in
Figures 24a and 24b.
To compute
W
1
W
2
, notice that (U V )
A,B
op
represents

BSDD((}
1

}
3
)(J
1
J
3
)),
A,B
op
X represents

BSDD(}
2
(J
1
J
2
)), and
A,B
op(M
N

) is a DG-type module representing



BSAA(J
1
J
2
). To compute the
join map, we need to convert them to single modules. For typeDD mod-
ules, this is trivial (any A, Bbimodule is automatically an A Bmodule
and vice versa). For typeAA modules, this could be complicated in gen-
eral. Luckily, it is easy for DG-type modules. Indeed, if P
A,B
is DG-type,
the corresponding ABmodule P
AB
is also DG-type, with algebra action
m
1|1
(, a b) = m
1|1|0
(, a) m
1|0|1
(, b) = m
1|0|1
(, b) m
1|1|0
(, a).
In the denition of bimodule invariants in [Zar09], the procedure used to
get

BSAA from

BSA, and

BSDD from

BSD is exactly the reverse of this
construction.
Thus, we can see that (U V )
AB
op
represents

BSD((}
1
}
3
) (J
1

J
3
)),
AB
op
X represents

BSD(}
2
(J
1
J
2
)), and
AB
op(M N

)
represents

BSA(J
1
J
2
). It is also easy to check that
A
A

B
op(B
op
)

B
op

=
AB
op(AB
op
)

AB
op.
We can see a diagram for
W
1
W
2
in Figure 24c. By examining the
diagrams, we see that the three maps are the same, which nishes part (2).
Part (3) requires some more work, so we will split it in several steps. We
will dene
M
for a xed representative M of

BSD(J). We will prove that
[
M
] does no depend on the choice of M. Finally, we will use a computa-
tional lemma to show that Eq. (9) holds for
M
.
First we will introduce some notation. Given an /

module
A
M over
A = /(:), dene the double of M to be
(10) T(M) = M

(
A
I
A
A
A
I
A
) M.
Note that if M =

BSA(J), then T(M) =

BSA(J)

BSDD(T J
F,
)

BSA(J) SFC(T(J)). Next we dene the diagonal element


M
T(M)
JOINING AND GLUING SFH 37
m
M
m
N
U M MX N N V
U A X B V
(a)
W2

W1
m
M
m
N
U M MX N N V
U A X B V
(b)
W1

W2
m
M
m
N
UV MN

N X
UV AB
op
X
(c)
W1W2
Figure 24. Three ways to join }
1
, }
2
, and }
3
.
=

M
1
M
A
I
A
A
A
I
A
M M
A
I
A
A
A
I
A
M
Figure 25. The diagonal element
M
.
as follows. Pick a basis (m
1
, . . . , m
k
) of M over Z/2. Dene
(11)
M
=
k

i=1
m
i
( 1 ) m
i

.
It is easy to check that this denition does not depend on the choice of
basis. Indeed there is a really simple diagrammatic representation of M,
given in Figure 25. We think of it as a linear map from Z/2 to T(M). It
is also easy to check that
M
= 0. Indeed, writing out the denition of

M
, there are are only two nonzero terms which cancel.
The proof that [
M
] does not depend on the choices of A and M is very
similar to the proof of Theorem 4.6, so will omit it. (It involves showing
independence from M, as well as from A via a quasi-invertible bimodule
A
X
B
.)
Lemma 5.6. Let A be a dierential graded algebra, coming from an arc
diagram :. There is a homotopy equivalence
c
A
:
A
I
A
A

A
I
A

A
A
A

A
I
A
,
given by
(c
A
)
1|1|0
_
a

b
_
=
_
b if a is an idempotent,
0 otherwise.
38 RUMEN ZAREV
= c
A
1
A
I
A
A
A
I
A
A
A
I
A
A
A
I
A
A
A
A
I
A
A
A
I
A
Figure 26. The cancellation homotopy equivalence c
A
: I A

I A I.
Here we use to denote the unique element with compatible idempo-
tents in the two versions of I. (Both versions have generators in 1to1
correspondence with the basis idempotents.)
Remark. As we mentioned earlier, one has to be careful when working with
typeDD modules. While and

are usually associative by themselves,
and with each other, this might fail when a DDmodule is involved, in which
case we only have associativity up to homotopy equivalence. However, this
could be mitigated in two situations. If the DDmodule is DG-type (which
fails for
A
I
A
), or if the typeA modules on both sides are DG-type, then
true associativity still holds. This is true for A and A

, so the statement of
the lemma makes sense.
Proof of Lemma 5.6. Note that we can easily see that there is some homo-
topy equivalence (I A

I) A I, since the left-hand side is

BSDD(T J
F,+
)

BSAA(T J
F,
)

BSDA(T J
F,+
T J
F,
),
while the right side is

BSDA(T [0, 1]), and those bordered sutured mani-
folds are the same. The diculty is in nding the precise homotopy equiv-
alence, which we need for computations, in order to cancel A

and A.
First, we need to show that c
A
is a homomorphism. This is best done
graphically. The denition of c
A
is represented in Figure 26. The notation
we use there is that
A
I
A
is a jagged line, without a direction, since I is its
own dual.
A
I
A
is represented by a dashed line. As before the line can start
or end with a dot, signifying the canonical isomorphism given by .
We need to show that c
A
= 0. Note that by denition c
A
only has a
1[1[0term. On the other hand on I A

I A has only 1[1[0 and


1[1[1terms, while on I has only a 1[1[1term.
Thus only four terms from the denition of c
A
survive. These are shown
in Figure 27. Expanding the denition of on I A

I A in terms of
the operations of I, A, and A

, we get seven terms. We can see them in


Figure 28. The terms in Figures 28a28d correspond to Figure 27a, while
those in Figures 28e28g correspond to Figures 27b27d, respectively. Six
of the terms cancel in pairs, while the one in 28b equals 0.
JOINING AND GLUING SFH 39

1|1|0
c
A
(a)

1
c
A
(b)

2
c
A

I
(c)

1|1|1
c
A
(d)
Figure 27. Nontrivial terms of c
A
.

2
1
(a)

1
1
1
(b)

2
1
1
(c)

1
1
1
(d)

1
1
(e)

2
1
(f)

2
1
(g)
Figure 28. Elementary terms of h.
Showing that c
A
is a homotopy equivalence is somewhat roundabout.
First we will show that the induced map
id
A
c
A
: A (
A
I
A
A

A
I
A
A) A
A
I
A

= A
is a homotopy equivalence. It is easy to see that the map is
(id
A
c
A
)
0|1|0
(a b

c) =
_
a c if b is an idempotent,
0 otherwise.
In particular, it is surjective. Indeed, id
A
c
A
(a 1

1) = a for
all a A. Thus the induced map on homology is surjective. But the source
and domain are homotopy equivalent for topological reasons (both represent

BSAA(T J
F,
)). This implies that id
A
c
A
is a quasi-isomorphism, and a
homotopy equivalence. But (I A

I) A I and A(I A

I) I for
topological reasons, so A is an equivalence of derived categories. Thus,
40 RUMEN ZAREV
= = =
1
1
1
1
1
1
A A A

I

I

IAIM

IAIM

IM
m
M
m
M
A A
M
M
cA
I M I M I M I M
A I M A I M A I M A I M
Figure 29. Proof that
M
(,
M
) id.
c
A
itself must have been a homotopy equivalence, which nishes the proof
of the lemma.
We will now use Lemma 5.6, to show that for any } there is a homo-
topy
W
(,
M
) id
SFC(Y)
. Let c
A
be the homotopy equivalence from the
lemma. There is a sequence of homomorphisms as follows.
I M
id
IM

M

I M T(M)

= I M M

I AI M
id
I

M
id
IAIM

I A

I AI M
c
A
id
IM

I M
The composition of these maps is shown in Figure 29. As we can see from
the diagram, it is equal to id
I
id
M
. If U =

BSD(} J), then by applying
the functor id
U
to both homomorphisms, we see that
(id
U
c
A
id
IM
)
M
(id
SFC(Y)

M
) = id
SFC(Y)
,
which is equivalent to Eq. (9).
5.3. Self-join and self-gluing. So far we have been talking about the join
or gluing of two disjoint sutured manifolds. However, we can extend these
notions to a self-join or self-gluing of a single manifold. For example if there
is an embedding (JJ) }, then we can dene the self-join of } along
J to be the concatenation
}
W,
= (Y (J J))
FF
T J
F,+

= }
WW
T(J).
It is easy to see that if J and J embed into dierent components of },
this is the same as the regular join.
JOINING AND GLUING SFH 41
Similarly, we can extend the join map to a self-join map

W,
: SFC(}) SFC(}
WW
T(J)) SFC(}
W,
),
by setting

W,
=
WW
(,
W
).
Again, if J and J embed into disjoint components of },
W,
is, up
to homotopy, the same as the regular join map
W
. This follows quickly
from properties (2) and (3) in Theorem 5.5.
6. The bordered invariants in terms of SFH
In this section we give a (partial) reinterpretation of bordered and bor-
dered sutured invariants in terms of SFH and the gluing map . This is a
more detailed version of Theorem 2.
6.1. Elementary dividing sets. Recall Denition 2.3 of a dividing set.
Suppose we have a sutured surface T = (F, ) parametrized by an arc
diagram : = (Z, a, M) of rank k. We will dene a set of 2
k
distinguished
dividing sets.
Before we do that, recall the way an arc diagram parametrizes a sutured
surface, from Section 3.1. There is an embedding of the graph G(:) into
F, such that Z = (Recall Figure 8). We will rst dene the elementary
dividing sets in the cases that : is of type. In that case the image of Z
is a push-o of S
+
into the interior of F. Denote the regions between S
+
and Z by R
0
. It is a collection of discs, one for each component of S
+
. The
images of the arcs e
i
G(:) are in the complement F R
0
.
Denition 6.1. Let I 1, . . . , k. The elementary dividing set for T
associated to I is the dividing set
I
constructed as follows. Let R
0
be the
region dened above. Set
R
+
= R
0

_
iI
(e
i
) F.
Then
I
= (R
+
) S
+
.
If : is of type, repeat the same procedure, substituting R

for R
+
and
S

for S
+
. For example the region R
0
consists of discs bounded by S

Z.
Examples of both cases are given in Figure 30.
We refer to the collection of
I
for all 2
k
many subsets of 1, . . . , k
as elementary dividing sets for :. The reason they are important is the
following proposition.
Proposition 6.2. Let : be an arc diagram of rank k, and let I 1, . . . , k
be any subset. Let
I
be the idempotent for A = /(:) corresponding to
horizontal arcs at all i I, and let
I
c be the idempotent corresponding to
the complement of I. Let J
I
be the cap associated to the elementary dividing
set
I
.
Then the following hold:
42 RUMEN ZAREV
1
2
3
4
1
2
3
4
(a) type diagram.
1
2
3
4
1
2
3
4
(b) type diagram.
Figure 30. Elementary dividing sets for an arc diagram. In each case we
show the arc diagram, its embedding into the surface, and the dividing set

{2,3}
. The shaded regions are R
+
.
A A B B
Figure 31. Heegaard diagram 1 for the cap J
2,3
corresponding to the di-
viding set from Figure 30a.

A

BSD(J) is (represented by) the elementary typeD module for
I
.

A

BSA(J) is (represented by) the elementary typeA module for
I
c .
Proof. The key fact is that there is a particularly simple Heegaard diagram
1 for J
I
. For concreteness we will assume : is a type diagram, though
the case of a type diagram is completely analogous.
The diagram 1 = (, , , :) contains no circles, exactly one arc
a
i
for each matched pair M
1
(i), and k#I many circles. Each circle has
exactly one intersection point on it, with one of
a
i
, for i / I. This implies
that there is exactly one generator x ((1), that occupies the arcs for I
c
.
This implies that

BSD(J
I
) and

BSA(J
I
) are both x, 0 as Z/2modules.
The actions of the ground ring are
I
x = x for

BSD(J
I
) and
I
c x = x for

BSA(J
I
). This was one of the two requirements for an elementary module.
The connected components of ( ) are in 1to1 correspondence
with components of R
+
. In fact each such region is adjacent to exactly
one component of Z. Therefore, there are only boundary regions and
no holomorphic curves are counted for the denitions of

BSD(J
I
) and

BSA(J
I
). This was the other requirement for an elementary module, so
the proof is complete. The diagram 1 can be seen in Figure 31.
We will dene one more type of object. Let T be a sutured surface
parametrized by some arc diagram :. Let I and J be two subsets of
1, . . . , k. Consider the sutured manifold J
I
T J
F,
J
J
. Since
J
I
and J
J
are caps, topologically this is F [0, 1]. The dividing set can
JOINING AND GLUING SFH 43
be described as follows. Along F 0 it is
I
0, along F 1 it is

J
1, and along F [0, 1] it consists of arcs in the [0, 1] direction with
a partial negative twist.
Denition 6.3. Let
IJ
denote the dividing set on (F [0, 1]), such that
(F [0, 1],
IJ
) = J
I
T J
F,
J
J
.
6.2. Main results. The main results of this section are the following two
theorems. We will give the proofs in the next subsection.
Theorem 6.4. Let T be a sutured surface parametrized by an arc diagram
:. The homology of A = /(:) decomposes as the sum
(12) H

(A) =

I,J{1,...,k}

I
H

(A)
J
=

I,J{1,...,k}
H

(
I
A
J
),
where
(13)
I
H

(A)
J

= SFH(F [0, 1],
IJ
).
Multiplication
2
descends to homology as

H
=
(F,
J
)
: SFH(F [0, 1],
IJ
) SFH(F [0, 1],
JK
)
SFH(F [0, 1],
IK
) ,
(14)
and is 0 on all other summands.
Theorem 6.5. Let } = (Y, , T) be a bordered sutured manifold where T
parametrized by :. Then there is a decomposition
H

BSA(Y )
A
_
=

I{1,...,k}
H

BSA(})
_

I
=

I{1,...,k}
H

BSA(Y )
I
_
,
(15)
where
(16) H

BSA(Y )
_

I

= SFH(Y,
I
).
Moreover, the m
1|1
action of A on

BSA descends to the following action
on homology:
(17) m
H
=
(F,
I
)
: SFH(Y,
I
)SFH(F I,
IJ
) SFH(Y,
J
),
and m
H
= 0 on all other summands.
Similar statements hold for left Amodules
A

BSA(}), and for bimodules
A

BSAA(})
B
.
Theorem 6.4 and 6.5, give us an alternative way to think about bordered
sutured Floer homology, or pure bordered Floer homology. (Recall that as
shown in [Zar09], the bordered invariants

CFD and

CFA are special cases
of

BSD and

BSA.) More remarkably, as we show in [Zar], H

(A),
H
, and
m
H
can be expressed in purely contact-geometric terms.
44 RUMEN ZAREV
For practical purposes, A and

BSA can be replaced by the /

algebra
H

(A) and the /

module H

BSA) over it. For example, the pairing


theorem will still hold. This is due to the fact that (using Z/2coecients),
an /

algebra or module is always homotopy equivalent to its homology.


We would need, however, the higher multiplication maps of H

(A), and
the higher actions of H

(A) on H

BSA). The maps


H
and m
H
that we just
computed are only single terms of those higher operations. (Even though A
is a DG-algebra, H

(A) usually has nontrivial higher multiplication.)


6.3. Proofs. In this section we prove Theorems 6.4 and 6.5. Since there is
a lot of overlap of the two results and the arguments, we will actually give a
combined proof of a mix of statements from both theorems. The rest follow
as corollary.
Combined proof of Theorem 6.4 and Theorem 6.5. First, note that Eq. (12)
and Eq. (15) follow directly from the fact that the idempotents generate the
ground ring over Z/2.
We will start by proving a generalization of Eq. (13) and Eq. (16). The
statement is as follows. Let T and T

be two sutured surfaces parametrized


by the arc diagrams : and :

of rank k and k

, respectively. Let A = /(:)


and B = /(:

). Let } = (Y, , T T

) be a bordered sutured manifold,


and let M =
A

BSAA(})
B
.
Fix I 1, . . . , k and J 1, . . . , k

. Let J
I
and J

J
be the respective
caps associated to the dividing sets
I
on T and

J
on T

. Then the
following homotopy equivalence holds.
(18)
I


BSAA(})
J
SFC(Y,
I

J
).
The proof is easy. Notice that the sutured manifold (Y,
I

J
) is just
J
I
}J

J
. By the pairing theorem, SFC(Y,
I

J
)

BSD(J
I
)

BSAA(})

BSD(J

J
). But by Proposition 6.2,

BSD(J
I
) = x
I
, 0 is the
elementary module corresponding to
I
, while

BSD(J

J
) = y
J
, 0 is the
elementary idempotent corresponding to

J
. Thus we have

BSD(J
I
)

BSAA(})

BSD(J

J
) = x
I


BSAA(}) y
J

=
I


BSAA(})

J
.
Eq. (13) follows from Eq. (18) by substituting the empty sutured sur-
face = (, ) for T. Its algebra is /() = Z/2, so
Z/2

BSAA(})
B
and

BSA(})
B
can be identied.
Eq. (16) follows from Eq. (18) by substituting T(:) for both T and T

,
and T J
F,
for }. Indeed,

BSAA(T J
F,
) /(:), as a bimodule over
itself, by Proposition 3.12.
Next we prove Eq. (17). Let U
A
be a DG-type representative for

BSA(})
A
,
and let M
I
be the elementary representative for
A

BSA(J
I
). Since both are
JOINING AND GLUING SFH 45
DG-type, we can form the associative product
U
A
I
A
M
I


BSA(Y )

BSD(J
I
)
SFC(Y,
I
).
Similarly, pick M
J
to be the elementary representative for
A

BSA(J
J
). We
also know that
A
A
A
is a DG-type representative for
A

BSAA(T J
F,
)
A
.
We have the associative product
M
I

A
I
A
A
A
I
A
M
J


BSD(J
I
) A

BSD(J
J
)
SFC(F [0, 1],
IJ
).
Gluing the two sutured manifolds along (F,
I
) results in
} T J
F,+
T J
F,
J
J

= } J
J
= (Y,
J
),
so we get the correct manifold.
The gluing map can be written as the composition of

M
I
: (U I) M
I
M
I

(I A I M
J
)
(U I) A

(I AI M
J
),
id
U
c
A
id
IM
J
: U (I A

I A) I M
J
U I M
J
,
where c
A
is the homotopy equivalence from Lemma 5.6.
Luckily, since M
I
is elementary,
M
I
takes the simple form from Propo-
sition 5.4. In addition, since U and M
J
are DG-type, id hid is also very
simple. As can be seen in Figure 32, the composition is in fact
u x
I
c x
I
c

a x
J
c m
1|1
(u, a) x
J
c.
Since x
I
c corresponds to
I
, this translates to the map

(F,
I
)
: (U
I
) (
I
A
J
) U
J
,
(u
I
) (
I
a
J
). m(u, a)
J
Note that even though we picked a specic representative for

BSA(})
A
,
the group H

BSA(})) and the induced action m


H
of H

(A) do not depend


on this choice. Finally, Eq. (14) follows by treating A as a right module over
itself.
Appendix A. Calculus of diagrams
This appendix summarizes the principles of the diagrammatic calculus we
have used throughout the paper. First we describe the algebraic objects we
work with, and the necessary assumptions on them. Then we describe the
diagrams representing these objects.
46 RUMEN ZAREV
=
id
UI

MI
id
IAIMJ
id
U
c
A
id
IMJ
m
U

I
1
U I
M
I
M
I I A I
M
J U I
M
I
M
I I A I
M
J
U I M
J
U I M
J
Figure 32. The gluing map
MI
on SFC(Y,
I
) SFC(F [0, 1],
IJ
),
followed by the chain homotopy equivalence id c
A
id.
A.1. Ground rings. The two basic objects we work with are a special class
of rings, and bimodules over them. We call these rings ground rings.
Denition A.1. A ground ring K is a nite dimensional Z/2algebra with
a distinguished basis (e
1
, . . . , e
k
) such that multiplication is given by the for-
mula
e
i
e
j
=
_
e
i
if i = j,
0 otherwise.
Such a basis for K is called a canonical basis.
The canonical basis elements are uniquely determined by the property
that e
i
cannot be written as a sum u + v, where u and v are nonzero and
u v = 0. Each element of K is an idempotent, while 1
K
= e
1
+ + e
k
is
an identity element.
We consider only nite dimensional bimodules
K
M
L
over ground rings K
and L, and collections (
K
M
iL
)
iI
where I is a countable index set (usually
I = 0, 1, 2, . . ., or some Cartesian power of the same), and each M
i
is a
nite-dimensional K, Lbimodule. It is often useful to think of the collection
(M
i
) as the direct sum

iI
M
i
, but that sometimes leads to problems, so
we will not make this identication.
There are some basic properties of bimodules over ground rings as dened
above.
Proposition A.2. Suppose K, L, and R are ground rings with canonical
bases (e
1
, . . . , e
k
), (e

1
, . . . , e

l
), and (e

1
, . . . , e

r
), respectively.
A bimodule
K
M
L
is uniquely determined by the collection of Z/2
vector spaces
e
i
M e

j
, i 1, . . . , k, j 1, . . . , l,
which we will call the components of M.
JOINING AND GLUING SFH 47
A K, Lbilinear map f : M N is determined by the collection of
Z/2linear maps
f[
e
i
Me

j
: e
i
M e

j
e
i
N e

j
.
The tensor product (
K
M
L
)
L
(
L
N
R
) has components
e
i
(M
L
N) e

j
=
l

p=1
(e
i
M e

p
)
Z/2
(e

p
N e

j
).
The dual
L
M

K
of
K
M
L
has components
e
i
M

j

= (e

j
M e
i
)

,
and the double dual (M

is canonically isomorphic to M.
Proof. These follow immediately. The fact that M

= M is due to the fact
the fact that each component is a nite dimensional vector space.
Finally, when dealing with countable collections we introduce the follow-
ing conventions. For consistency we can think of a single module M as a
collection (M
i
) indexed by the set I = 1.
Denition A.3. Let K, L, and M be as in Proposition A.2.
An element of (M
i
)
iI
is a collection (m
i
)
iI
where m
i
M
i
.
A bilinear map f : (
K
M
iL
)
iI
(
K
N
j
L
)
jJ
is a collection
f
(i,j)
: M
i
N
j
(i, j) I J.
Equivalently, a map f is an element of the collection
Hom
K,L
((M
i
)
iI
, (N
j
)
jJ
) = (Hom(M
i
, N
j
))
(i,j)IJ
.
The tensor
K
(M
i
)
L

L
(N
j
)
R
is the collection
((M N)
(i,j)
)
(i,j)IJ
= (M
i
N
j
)
(i,j)IJ
.
The dual ((M
i
)
iI
)

is the collection (M
i

)
iI
.
Given bilinear maps f : (M
i
) (N
j
) and g: (N
j
) (P
p
), their
composition g f : (M
i
) (P
p
) is the collection
(g f)
(i,p)
=

jJ
g
(j,p)
f
(i,j)
.
Note that the composition of maps on collections may not always be
dened due to a potentially innite sum. On the other hand, the double
dual (M
i
)

is still canonically isomorphic to (M


i
).
48 RUMEN ZAREV
A.2. Diagrams for maps. We will use the following convention for our
diagram calculus. There is a TQFT-like structure, where to decorated planar
graphs we assign bimodule maps.
Proposition A.4. Suppose K
0
, K
1
, . . . , K
n
= K
0
are ground rings, n 0,
and
K
i1
M
iK
i
are bimodules, or collections of bimodules. Then the following
Z/2spaces are canonically isomorphic.
A
i
= M
i
M
i+1
M
n
M
1
M
i1
/ ,
B
i,j
= Hom
K
i
,K
j
(M

i
M

1
M

n
M

j+1
, M
i+1
M
j
),
C
i,j
= Hom
K
j
,K
i
(M

j
M

i+1
, M
j+1
M
n
M
1
M
i
),
for 0 i j n, where the relation in the denition of A
i
is k x x k,
for k K
i1
.
Proof. The proof is straightforward. If all M
i
are single modules, then we are
only dealing with nite-dimensional Z/2vector spaces. If some of them are
collections, then the index sets for A
i
, B
i,j
and C
i,j
are all the same, and any
individual component still consists of nite dimensional vector spaces.
This property is usually referred to as Frobenius duality. Our bimodules
behave similar to a pivotal tensor category. Of course we do not have a real
category, as even compositions are not always dened.
Denition A.5. A diagram is a planar oriented graph, embedded in a disc,
with some degree1 vertices on the boundary of the disc There are labels as
follows.
Each planar region (and thus each arc of the boundary) is labeled by
a ground ring K.
Each edge is labeled by a bimodule
K
M
L
, such that when traversing
the edge in its direction, the region on the left is labeled by K, while
the one on the right is labeled by L.
An internal vertex with all outgoing edges labeled by M
1
, . . . , M
n
, in
cyclic counterclockwise order, is labeled by an element of one of the
isomorphic spaces in Proposition A.4.
If any of the edges adjacent to a vertex are incoming, we replace the
corresponding modules by their duals.
When drawing diagrams we will omit the bounding disc, and the boundary
vertices. We will usually interpret diagrams consisting of a single internal
vertex having several incoming edges M
1
, . . . , M
m
on top, and several
outgoing edges N
1
, . . . , N
n
on the bottom, as a bilinear map in Hom(M
1

M
m
, N
1
N
n
). See Figure 33 for an example.
Under some extra assumptions, discussed in Section A.3, a diagram with
more vertices can also be evaluated, or interpreted as an element of some set,
corresponding to all outgoing edges. The most common example is having
two diagrams T
1
and T
2
representing linear maps
M
f
1
//
N
f
2
//
P.
JOINING AND GLUING SFH 49
F
F
F
M
1
M
2
M
3
M
4
M
5
M
2
M
3
M
4
M
5
M
1
M

1
M

5
M

4
M

3
M

2
Figure 33. Three equivalent diagrams with a single vertex. The label F is
interpreted as an element of A
1
= M
1
M
5
/ , B
1,4
= Hom(M

1

M

5
, M
2
M
3
M
4
), and C
1,4
Hom(M

4
M

3
M

2
, M
5
M
1
), respectively.
=
D
F G
H
M N P M N P
Q R
S
Figure 34. Evaluation of a complex diagram.
Stacking the two diagrams together, feeding the outgoing edges of T
1
into
the incoming edges of T
2
, we get a new diagram T, corresponding to the
map f
2
f
1
: M P. More generally, we can contract along all internal
edges, pairing the elements assigned to the two ends of an edge. As an
example we will compute the diagram D in Figure 34. Suppose the values
of the vertices F, G, and H are as follows:
F =

i
m
i
q
i
s
i
M QS,
G =

j
s

j
r

j
p

j
S

,
H =

k
q

k
n
k
r
k
Q

N R.
Then the value of D is given by
D =

i,j,k

q
i
, q

k
_
Q

s
i
, s

j
_
S

r
k
, r

j
_
R
m
i
n
k
p

j
M N P

.
Edges that go from boundary to boundary and closed loops can be in-
terpreted as having an identity vertex in the middle. As with individual
vertices, we can rotate a diagram to interpret it as an element of dierent
spaces, or dierent linear maps.
Note that the above construction might fail if any of the internal edges
corresponds to a collection, since there might be an innite sum involved.
The next section discusses how to deal with this problem.
50 RUMEN ZAREV
A.3. Boundedness. When using collections of modules we have to make
additional assumptions to avoid innite sums. We use the concept of bound-
edness of maps and diagrams.
Denition A.6. An element (m
i
)
iI
of the collection (M
i
)
iI
is called
bounded if only nitely many of its components m
i
are nonzero. Equiv-
alently, the bounded elements of (M
i
) can be identied with the elements of

i
M
i
.
For a collection (M
i,j
)
iI,jJ
there are several dierent concepts of bound-
edness. An element (m
i,j
) is totally bounded if it is bounded in the above
sense, considering I J as a single index-set. A weaker condition is that
(m
i,j
) is bounded in J relative to I. This means that for each i I, there are
only nitely many j J, such that m
i,j
is nonzero. Similarly, an element
can be bounded in I relative to J.
Note that f : (M
i
) (N
j
) is bounded in J relative to I exactly when f
represents a map from

i
M
i
to

j
N
j
. In computations relatively bounded
maps are more common than totally bounded ones. For instance the identity
map id: (M
i
) (M
i
) and the natural pairing , : (M
i
)

(M
i
) K are
not totally bounded, but are bounded in each index relative to the other.
To be able to collapse an edge labeled by a collection (M
i
)
iI
in a diagram,
at least one of the two adjacent vertices needs to be labeled by an element
relatively bounded in the Iindex. For a given diagram D we can ensure
that it has a well-dened evaluation by imposing enough boundedness con-
ditions on individual vertices. (There is usually no unique minimal set of
conditions.) Total or relative boundedness of D can also be achieved by a
stronger set of conditions. For example, if all vertices are totally bounded,
the entire diagram is also totally bounded.
Appendix B. /

algebras and modules


In this section we will present some of the background on /

algebras
and modules, and the way they are used in the bordered setting. A more
thorough treatment is given in [LOT10a].
As in Appendix A, we always work with Z/2coecients which avoids
dealing with signs. Everything is expressed in terms of the diagram calculus
of Appendix A. As described there, all modules are nite dimensional,
although we also deal with countable collections of such modules. There is
essentially only one example of collections that we use, which is presented
below.
B.1. The bar construction. Suppose K is a ground ring and
K
M
K
is a
bimodule over it.
Denition B.1. The bar of M is the collection
Bar M = (M
i
)
i=0,...,
,
of tensor powers of M.
JOINING AND GLUING SFH 51
Bar M Bar M Bar M
Bar M Bar M M Bar M Bar M Bar M Bar M
(a) Split maps.
Bar M Bar M Bar M
Bar M Bar M M Bar M Bar M M Bar M
(b) Merge maps.
Figure 35
There are two important maps on the bar of M.
Denition B.2. The split on Bar M is the map s: Bar M Bar M
Bar M with components
s
(i,j,k)
=
_
id: M
i
(M
j
) (M
k
) if i = j +k,
0 otherwise.
The merge map Bar M Bar M Bar M is similarly dened.
Merges and splits can be extended to more complicated situations where
any combination of copies of Bar M and M merge into Bar M, or split from
Bar M. All merges are associative, and all splits are coassociative.
Like the identity map, splits and merges are bounded in incoming indices,
relative to outgoing, and vice versa. To simplify diagrams, we draw merges
and splits as merges ans splits of arrows, respectively, without using a box
for the corresponding vertex (see Figure 35).
B.2. Algebras and modules. The notion of an /

algebra is a general-
ization of that of a dierential graded (or DG) algebra. While the algebras
that arise in the context of bordered Floer homology are only DG, we give
the general denition for completeness. We will omit grading shifts.
Denition B.3. An /

algebra A over the base ring K consists a K


bimodule
K
A
K
, together with a collection of linear maps
i
: A
i
A,
i 1, satisfying certain compatibility conditions. By adding the trivial map

0
= 0: K A, we can regard this as a map = (
i
): Bar A A. This
induces a map : Bar A Bar A, given by splitting Bar M into three copies
of itself, applying to the middle one, and merging again (see Figure 36a).
The compatibility condition is = 0, or equivalently = 0 (see
Figure 36b).
The algebra is unital if there is a map 1: K A (which we draw as a
circle labeled 1 with an outgoing arrow labeled A), such that
2
(1, a) =

2
(a, 1) = a, and
i
(. . . , 1, . . .) = 0 if i ,= 2.
The algebra A is bounded if is bounded, or equivalently if is relatively
bounded in both directions.
Notice that a DG-algebra with multiplication m and dierential d is just
an /

algebra with
1
= d,
2
= m, and
i
= 0 for i 3. Moreover,
DG-algebras are always bounded.
52 RUMEN ZAREV

=
(a) in terms of .
= 0

= 0


(b) Compatibility conditions.
Figure 36. Denition of /

algebras
Since DG-algebras are associative, there is one more operation that is
specic to them.
Denition B.4. The associative multiplication : Bar A A for a DG-
algebra A is the map with components

i
(a
1
a
i
) =
_
a
1
a
2
a
i
i > 0,
1 i = 0.
There are two types of modules: typeA, which is the usual notion of
an /

module, and typeD. There are four types of bimodules: typeAA,


typeDA, etc. These can be extend to tri-modules and so on. We describe
several of the bimodules. Other cases can be easily deduced.
Suppose A and B are unital /

algebras with ground rings K and L,


respectively. We use the following notation. A typeA module over A will
have A as a lower index. A typeD module over A will have A as an upper
index. Module structures over the ground rings K and L are denoted with
the usual lower index notation.
Denition B.5. A typeAA bimodule
A
M
B
consists of a bimodule
K
M
L
over the ground rings, together with a map m = (m
i|1|j
): Bar A M
Bar B M. The compatibility conditions for m are given in Figure 37.
The bimodule M is unital if m
1|1|0
(1
A
, m) = m
0|1|1
(m, 1
B
) = m, and
m
i|1|j
vanishes in all other cases where one of the inputs is 1
A
or 1
B
.
The bimodule can be bounded, bounded only in A, relatively bounded in A
with respect to B, etc. These are dened in terms of the index sets of Bar A
and Bar B.
Denition B.6. A typeDA bimodule
A
M
B
consists of a bimodule
K
M
L
over the ground rings, together with a map = (
1|1|j
): MBar B AM.
This induces another map = (
i|1|j
): MBar B Bar AM, by splitting
Bar B into i copies, and applying imany times (see Figure 38a). The
compatibility conditions for and are given in Figure 38b.
The bimodule M is unital if
1|1|1
(m, 1
B
) = 1
A
m, and
1|1|i
vanishes
for i > 1 if one of the inputs is 1
B
.
JOINING AND GLUING SFH 53
m
m

A
m

B
m
+ + = 0
Figure 37. Structure equation for a typeAA module.
.
.
.
=
.
.
.
.
.
.

(a) in terms of .
+ + = 0 = 0

B

B


A
(b) Structure equation for a typeDA module.
Figure 38
+ + = 0

A

B

A

B
Figure 39. Structure equation for a typeDD module.
Again, there are various boundedness conditions that can be imposed.
TypeDD modules only behave well if the algebras involved are DG, so
we only give the denition for that case.
Denition B.7. Suppose A and B are DG-algebras. A type DDmodule
A
M
B
consists of a bimodule
K
M
L
over the ground rings, together with a
map
1|1|1
: M A M B satisfying the condition in Figure 39.
We omit the denition of one-sided typeA and typeD modules, as they
can be regarded as special cases of bimodules. TypeA modules over A
can be interpreted as typeAA bimodules over A and B = Z/2. Similarly,
typeD modules are type DAmodules over Z/2.
B.3. Tensor products. There are two types of tensor products for /

modules. One is the more traditional derived tensor product



. It is gen-
erally hard to work with, as M

N is innite dimensional over Z/2 even
when M and N are nite dimensional. This is bad for computational rea-
sons, as well as when using diagramsit violates some of the assumptions
54 RUMEN ZAREV
= + +
m
M

N
m
M
m
N
B
(a)
A
M
B

B
N
C
= + +

N

M
m
N
B
1 1
(b)
A
M
B

B
N
C
Figure 40. Structure maps for two types of

products.
of Appendix A. Nevertheless, we do use it in a few places throughout the
paper.
Throughout the rest of this section assume that A, B, and C are DG-
algebras over the ground rings K, L, and P, respectively.
Denition B.8. Suppose
A
M
B
and
B
N
C
are two typeAA bimodules. The
derived tensor product (
A
M
B
)

B
(
B
N
C
) is a typeAA bimodule
A
(M

N)
B
dened as follows. Its underling bimodule over the ground rings is
K
(M

N)
P
= (
K
M
L
)
L
_

i=0
L
B
L
i
_

L
(
L
N
P
)
= M
L
Bar B
L
N.
Here were slightly abusing notation in identifying Bar B with a direct sum.
The structure map as an /

bimodule over A and C is m


M

N
, as shown
in Figure 40a.
Similarly, we can take the derived tensor product of a DA module and
an AA module, or a DA module and an AD module. The former is demon-
strated in Figure 40b.
The other type of tensor product is the square tensor product . It is
asymmetric, as it requires one side to be a typeD module, and the other
to be a typeA module. The main advantage of over

is that M N is
nite dimensional over Z/2 whenever M and N are. Its main disadvantage
is that M N is only dened subject to some boundedness conditions on
M and N.
Denition B.9. Suppose
A
M
B
is a typeAA bimodule and
B
N
C
is a type
DA bimodule, such that at least one of M and N is relatively bounded in
B. The square tensor product (
A
M
B
)
B
(
B
N
C
) is a typeAA bimodule
A
(M N)
C
dened as follows. Its underlying bimodule over the ground
JOINING AND GLUING SFH 55
= m
MN
m
M
N
(a)
A
M
B

B
N
C
=
MN

M
N
(b)
A
M
B

B
N
C
=
MN
m
M
N

C
(c)
A
M
B

B
N
C
=
MN

M
N

C
(d)
A
M
B

B
N
C
Figure 41. Structure maps for the four types of products.
rings is
K
(M N)
P
= (
K
M
L
)
L
(
L
N
P
),
and its structure map is m
MN
as shown in Figure 41a.
There are three other combinations depending on whether the modules
are of type D or A with respect A and C. All combinations are shown in
Figure 41.
B.4. Morphisms and homomorphisms. There are two dierent notions
of morphisms when working with /

modules and bimodules. The more


natural one is that of homomorphisms, which generalize chain maps. How-
ever, if we work only with homomorphisms, too much information is lost.
For this reason we also consider the more general morphisms. These gen-
eralize linear maps of chain complexes, which do not necessarily respect
dierentials.
Denition B.10. A morphism f : M N between two bimodules M and
N of the same type is a collection of maps of the same type as the struc-
ture maps for M and N. For example, f :
A
M
B

A
N
B
has components
f
i|1|j
: Bar A M Bar B N. The spaces of morphisms are denoted by
A
Mor
B
(M, N), etc.
Suppose A and B are DG-algebras. The bimodules of each type, e.g.
A
Mod
B
, form a DG-category, with morphism spaces
A
Mor
B
, etc. The
dierentials and composition maps for each type are shown in Figures 42
and 43, respectively.
Denition B.11. A homomorphism f : M N of bimodules is a mor-
phism f which is a cycle, i.e., f = 0. A null-homotopy of f is a morphism
H, such that H = f. The space of homomorphisms up to homotopy is
denoted by
A
Hom
B
, etc.
Notice that the homomorphism space
A
Hom
B
(M, N) is exactly the ho-
mology of
A
Mor
B
(M, N). This gives us a new category of bimodules.
56 RUMEN ZAREV
= + + +
f
m
M f

A

B
f
m
N f f
(a) TypeAA.
= + + +
f f
f
f
f

A

A

B
(b) TypeDA.
= + + +
f f
f
f f

A

B

A

B

A

B
(c) TypeDD.
Figure 42. Dierentials of the dierent types of morphisms.
=
gf
f
g
(a) TypeAA.
=
gf
g
f

A
(b) TypeDA.
=
gf
g
f

A

B
(c) TypeDA.
Figure 43. Compositions of the dierent types of morphisms.
Having homomorphisms and homotopies allows us to talk about homo-
topy equivalences of modules. For example, if
A
M
B
is a bimodule, then
A

M M M

B, via canonical homotopy equivalences. For example,
there is h
M
: A

M M, which we used in several places.
JOINING AND GLUING SFH 57
= f id
N
f

N
(a)
= id
M
f
m
M

N
f

P
(b)
=
f

id
N
f
(c)
Figure 44. Three types of induced maps on tensor products.
=
A M A
op
M A
op
M
B M B
op
M B
op
M
m
M
m
M m
op
M
Figure 45. Passing from
A
Mod
B
to
B
op Mod
A
op
by reection.
B.5. Induced morphisms. Suppose f : M N is a bimodule morphism.
This induces morphisms
f

id: M

P N

P f id: M P N P,
whenever the tensor products are dened. The main types of induced mor-
phisms are shown in Figure 44. The functors id and

id are DG-functors.
That is, they preserve homomorphisms, homotopies, and compositions.
B.6. Duals. There are two operations on modules, which can be neatly
expressed by diagrams. One is the operation of turning a bimodule
A
M
B
into a bimodule
B
opM
A
op. (Similarly, typeDA bimodules become typeAD
bimodules, etc.) Diagrammatically this is achieved by reecting diagrams
along the vertical axis. See Figure 45 for an example.
The other operation is dualizing modules and bimodules. If
A
M
B
has
an underlying bimodule
K
M
L
over the ground rings, then its dual
B
M

A
has an underlying bimodule
L
M

K
= (
K
M
L
)

. Diagrammatically this is
achieved by rotating diagrams by 180 degrees. Again, there are variations
for typeD modules. See Figure 46 for an example.
Since the structure equations are symmetric, it is immediate that both
of these operations send bimodules to bimodules, as long as we restrict to
modules nitely generated over Z/2.
58 RUMEN ZAREV
=
A M
A
M A M
B M
B
M B M
m
M
m
M
m
M

Figure 46. Passing from


A
Mod
B
to
B
Mod
A
by rotation.
This gives equivalences of the DG-categories
A
Mod
B

=
B
op Mod
A
op

=
_
B
Mod
A
_
op
,
etc. One can check that both constructions extend to tensors, induced
morphisms, etc.
References
[Aur] Denis Auroux, Fukaya categories of symmetric products and bordered
Heegaard-Floer homology, J Gokova Geometry Topology (to appear),
arXiv:1001.4323.
[CGHH10] Vincent Colin, Paolo Ghiggini, Ko Honda, and Michael Hutchings, Sutures
and contact homology I, 2010, arXiv:1004.2942.
[Gab83] David Gabai, Foliations and the topology of 3-manifolds, J. Dierential Geom-
etry 18 (1983), 445503.
[HKM07] Ko Honda, William Kazez, and Gordana Matic, The contact invariant in su-
tured Floer homology, 2007, arXiv:0705.2828.
[HKM08] , Contact structures, sutured Floer homology and TQFT, 2008,
arXiv:0807.2431.
[Juh06] Andras Juhasz, Holomorphic discs and sutured manifolds, Algebraic & Geo-
metric Topology 6 (2006), 14291457, arXiv:math.GT/0601443.
[Juh09] , Cobordisms of sutured manifolds, 2009, arXiv:0910.4382.
[KM10] Peter Kronheimer and Tomasz Mrowka, Knots, sutures, and excision, J Dif-
ferential Geometry 84:2 (2010), 301364, arXiv:0807.4891.
[LOT08] Robert Lipshitz, Peter Ozsvath, and Dylan Thurston, Bordered Heegaard Floer
homology: Invariance and pairing, 2008, arXiv:0810.0687.
[LOT10a] , Bimodules in bordered Heegaard Floer homology, 2010,
arXiv:1003.0598.
[LOT10b] , Heegaard Floer homology as morphism spaces, 2010,
arXiv:1005.1248.
[OS04a] Peter Ozsvath and Zolt an Szabo, Holomorphic disks and three-manifold in-
variants: Properties and applications, The Annals of Mathematics 159 (2004),
no. 3, 11591245, arXiv:math/0105202.
[OS04b] , Holomorphic disks and topological invariants for closed three-
manifolds, The Annals of Mathematics 159 (2004), no. 3, 10271158,
arXiv:math/0101206.
[Zar] Rumen Zarev, Equivalence of gluing maps for SFH, in preparation.
[Zar09] , Bordered Floer homology for sutured manifolds, 2009,
arXiv:0908.1106.
Department of Mathematics, Columbia University, New York, NY 10027
E-mail address: rzarev@math.columbia.edu