Sikimeti Ma'u - Gluing Pseudoholomorphic Quilted Disks

a
r
X
i
v
:
0
9
0
9
.
3
3
3
9
v
1

[
m
a
t
h
.
S
G
]

1
7

S
e
p

2
0
0
9
GLUING PSEUDOHOLOMORPHIC QUILTED DISKS.
SIKIMETI MAU
Abstract. We construct families of quilted surfaces parametrized by the Stashe multipli-
hedra, and dene moduli spaces of pseudoholomorphic quilted disks using the holomorphic
quilts theory of Wehrheim and Woodward. We prove a gluing theorem for isolated, regular
pseudoholomorphic quilted disks. This analytical result is a fundamental ingredient for the
construction of A functors associated to Lagrangian correspondences.
1. Introduction
1.1. The setting. The Fukaya category of a symplectic manifold (M, ) is an algebraic con-
struct built out of certain Lagrangian submanifolds and their associated Floer chain groups.
Roughly speaking the objects of Fuk(M) are Lagrangian submanifolds, while morphisms are
elements of the Floer chain groups, CF(L, L
). The A
structure of the Fukaya category comes

from a whole sequence of higher compositions,
n
: CF(L
n1
, L
n
) . . . CF(L
0
, L
1
) CF(L
0
, L
d
)
for n 1, which satisfy a sequence of quadratic relations called the A
associativity relations.
The maps
n
are dened geometrically by counts of pseudoholomorphic (n + 1)-gons, and are
natural extensions of the dierential and Donaldson product in Lagrangian Floer homology,
which count pseudoholomorphic bigons and triangles, respectively. The main dierence in the
analytical theory when n 3 is that the complex structure on the domain of the pseudoholo-
morphic maps is allowed to vary, parametrized by the space of disks with n+1 distinct points on
the boundary modulo complex isomorphism. This parameter space is identied with /
R
0,n+1
, a
component of the real locus of the Deligne-Mumford space /
0,n+1
. Its stable compactication
realizes the Stashe associahedron K
n
, and this is where the intrinsic A
structure comes
from.
In this paper we consider a similar set-up that uses holomorphic quilts, introduced by
Wehrheim and Woodward [10, 11]. For d 1, we construct families of quilts parametrized
by moduli spaces of pointed quilted disks, and study moduli spaces of pseudoholomorphic
quilted disks. A quilted disk with d + 1 marked points on the boundary is a conguration
(D, C, z
0
, . . . , z
d
) where D is the unit disk in C, (z
0
, . . . , z
d
) is a cyclically ordered tuple of
distinct points in D, and C D is an inner circle such that C D = z
0
. The moduli
space of (d+1)-pointed quilted disks, denoted
d,0
, consists of such tuples modulo complex au-
tomorphisms of D. These moduli spaces were studied in [4]
1
, and their Grothendieck-Knudsen
type compactications realize the multiplihedra, a family of polytopes that appeared alongside
1
In [4] the moduli spaces R
d,0
were called M
d,1
.
1
2 SIKIMETI MAU
the associahedra in the work of Stashe in the 60s [8]. The multiplihedra are intrinsically
connected to A
maps between A
spaces, just as the associahedra are intrinsically connected

to A
spaces.
z0
z1
z2
z3
Figure 1.1. The third multiplihedron, J
3
=
3,0
.
1.2. Functorial motivation. A Lagrangian correspondence from (M,
M
) to (N,
N
) is a La-
grangian submanifold L of the product symplectic manifold M
N := (M N, (
M
,
N
)).
Lagrangian correspondences are a notion of morphism between symplectic manifolds, so one
would hope for topological invariants associated to symplectic manifolds to exhibit some sort
of functoriality with respect to them. Lagrangian correspondences arent morphisms on the
nose, since they cant always be composed
2
. Following [10], a generalized Lagrangian cor-
respondence from M
A
to M
B
is a sequence of Lagrangian correspondences, that starts with
M
A
and ends in M
B
(passing through any number of intermediate symplectic manifolds in
between). A generalized Lagrangian correspondence is represented as a directed sequence
L = (M
A
,
A
)
L
01
(M
1
,
1
)
L
12
(M
2
,
2
)
L
23
. . . (M
r
,
r
)
L
r,r+1
(M
B
,
B
).
Generalized Lagrangian correspondences can be composed by concatenation. We denote con-
catenation by #, i.e. L#L
:= M
A
L
M
B
L
M
C
.
Quilted Lagrangian Floer theory allows one to dene Floer homology HF(L, L
) for pairs
L, L
of admissible generalized Lagrangian correspondences from M

A
to M
B
[10, 12]. Thus a
pair of symplectic manifolds (M
A
,
A
), (M
B
,
B
) has an associated enlarged Donaldson-Fukaya
category, Don
#
(M
A
, M
B
), whose objects are sequences of admissible Lagrangian correspon-
dences from M
A
to M
B
, and Hom(L
0
, L
1
) := HF(L
0
, L
1
). When M
A
= pt, i.e., the zero di-
mensional symplectic manifold, we say that a generalized Lagrangian correspondence from pt
2
There is a geometric notion of composition of Lagrangian correspondences, following Weinstein, but the
resulting object may well be singular.
GLUING PSEUDOHOLOMORPHIC QUILTED DISKS. 3
z0
z2
z3
z1
z0
z1
z2
z3
double
MA
MB
z1 z2 z3
Figure 1.2. Schematic representation of a generalized 3 + 1-gon. The case
drawn is for L
0
having 3 arrows (i.e., 2 intermediate symplectic manifolds in
the sequence), L
1
4 arrows, L
2
two arrows and L
3
one arrow. The big (outer)
dotted square corresponds to the marked point z
0
at .
to M
B
is a generalized Lagrangian submanifold of M
B
.
3
We write Don
#
(M
B
) := Don
#
(pt, M
B
)
for the corresponding enlarged Donaldson-Fukaya category.
It was shown in [10] that a generalized Lagrangian correspondence between M
A
and M
B
determines a functor
(L
AB
) : Don
#
(M
A
) Don
#
(M
B
),
and furthermore that the assignment L
AB
(L
AB
) extends to a functor
: Don
#
(M
A
, M
B
) Fun(Don
#
(M
A
), Don
#
(M
B
)),
where Fun(Don
#
(M
A
), Don
#
(M
B
)) is the category of functors from Don
#
(M
A
) to Don
#
(M
B
).
The eventual goal is to extend both these functors to the chain level; that is, with the corre-
sponding enlarged Fukaya categories.
The enlarged Fukaya category Fuk
#
(M
A
, M
B
) is a generalization of the Fukaya category. Its
objects are admissible generalized Lagrangian correspondences from M
A
to M
B
, and morphisms
between objects are elements of Floer chain groups. Higher compositions
n
: CF(L
n1
, L
n
) . . . CF(L
0
, L
1
) CF(L
0
, L
n
)
are dened by counting isolated pseudoholomorphic generalized (n + 1)-gons. The quilted do-
mains behind the
n
are obtained from the (n +1)-pointed disks of the usual Fukaya category
by doubling then attaching strips. Given an (n + 1)-pointed disk, we rst take an identical
copy of it, then attach quilted strips of a xed width joining corresponding boundary compo-
nents. The number of components in the quilted strips depends on the number of intermediate
manifolds in the generalized Lagrangian correspondences being used (Figure 1.2).
For a xed collection of generalized Lagrangian correspondences, the complex structures on
the components of the quilted (n+1)-gon depend only on the complex structure on the (n+1)-
pointed disk. Thus the quilted domains behind the maps
n
are still parametrized by the moduli
space /
R
0,n
. The gluing result of this paper (Theorem 1) applies with only cosmetic changes to
isolated regular pseudoholomorphic quilted polygons, which is an analytical ingredient behind
3
An actual Lagrangian submanifold is a sequence with exactly one arrow, L = {pt
L
(MB, B)}.
4 SIKIMETI MAU
proving that the compositions
n
in the enlarged Fukaya category satisfy the A
associativity
relations, In particular, Fuk
#
(M
A
, M
B
) has the structure of a non-unital A
category. In the
special case that M
A
= pt, we write Fuk
#
(M
B
) := Fuk
#
(pt, M
B
).
Given a Lagrangian correspondence L
AB
from M
A
to M
B
, there is an associated non-unital
A
functor (L
AB
) : Fuk
#
(M
A
) Fuk
#
(M
B
). We give a very rough sketch of the construc-
tion here, referring to [3] for details.
Recall the rather combinatorial denition of a non-unital A
functor [7, Chapter I, (1b)]:

Denition. A non-unital A
functor between non-unital A
categories / and B consists of

a map : Ob / Ob B on objects, together with a sequence of maps
d
, d 1,
d
: Hom
A
(X
0
, X
1
) . . . Hom
A
(X
d1
, X
d
) Hom
B
((X
0
), (X
d
))[1 d],
where the
d
t together with the
j
A
,
k
B
to satisfy the A
functor relations:
i,j
(1)
e
(a
d
, . . . , a
i+j+1
,
j
A
(a
j+i
, . . . , a
i+1
), a
i
, . . . , a
1
) = (1.1)
r,i
1
+...ir=d
r
B
(
ir
(a
d
, . . . , a
dir
), . . . ,
i
1
(a
i
1
, . . . , a
1
)),
where = [a
1
[ +. . . +[a
i
[ i.
Here, / = Fuk
#
(M
A
) and B = Fuk
#
(M
B
). Dene (L
AB
) as follows. On objects,
(L
AB
) : Ob Fuk
#
(M
A
) Ob Fuk
#
(M
B
), L L#L
AB
,
i.e., concatenate L
AB
to each generalized Lagrangian submanifold of M
A
. Now suppose that
d 1, and L
0
, L
1
, . . . , L
d
are generalized Lagrangian submanifolds of M
A
. Abbreviate L
0,AB
:=
L
0
#L
AB
, likewise L
d,AB
. Dene the multilinear maps
(L
AB
)
d
: CF(L
d1
, L
d
) . . . CF(L
0
, L
d
) CF(L
0
#L
AB
, L
d
#L
AB
)
(p
d
, . . . , p
1
)
qI(L
0,AB
,L
d,AB
)
N
q
q),
where the coecient N
q
is a signed count of a 0-dimensional moduli space of pseudoholomorphic
quilted disks, which are pairs (r, u) where r
d,0
parametrizes a quilted domain o
r
, and u is
a pseudoholomorphic quilt map from o
r
, with a seam condition given by L
AB
(Figure 1.3).
MB
L
0
L
1
L
2
L
4
L
3
q
p
1
p
2
p
3
p
4
L
AB
MA
Figure 1.3. A quilted disk with boundary conditions.
The codimension one boundary strata of the multiplihedra come in two distinct types. The
rst type of facet is a pointed quilted disk joined to a pointed disk; call this a facet of Type 1.
The second type is a pointed disk with several pointed quilted disks; call this a facet of Type
2. These facets are closely related to the A
functor relations, which have four distinct kinds

of terms, corresponding in this construction to: a) broken tuples whose domains correspond to
facets of Type 1, b) broken tuples whose domains correspond to facets of Type 2, and c) either
incoming or outgoing broken Floer trajectories (Figure 1.4 ).
Figure 1.4. Four cases of gluing, from left to right: facet of Type 1, facet of
Type 2, incoming Floer trajectory, outgoing Floer trajectory.
The purpose of this paper is to prove the following statement.
Theorem 1. Let (M
A
,
A
) and (M
B
,
B
) be compact symplectic manifolds, and L
AB
a La-
grangian correspondence from M
A
to M
B
. Suppose that L
0
, . . . , L
d
are generalized Lagrangian
submanifolds of M
A
, all of whose intermediate symplectic manifolds are compact. Write
L
0,AB
:= (L
0
, L
AB
), L
d,AB
:= (L
d
, L
AB
) for the generalized Lagrangian submanifolds of M
B
obtained by concatenating L
AB
to L
0
and L
d
. Let x
0
1(L
0,AB
, L
d,AB
) and for i = 1, . . . , d let
x
i
1(L
i1
, L
i
). Given either:
(a) a regular pair
(r
1
, u
1
) /
de+1,0
(x
0
, x
1
. . . , x
i1
, y, x
i+e+1
, . . . , x
d
)
0
(r
2
, u
2
) /
e
(y, x
i
, x
i+1
, . . . , x
i+e
)
0
where 2 e d, 1 i d e, and y 1(L
i1
, L
i+e
);
(b) or a regular (k + 1)-tuple
(r
0
, u
0
) /
k
(x
0
, y
1
, . . . , y
k
)
0
(r
1
, u
1
) /
d
1
,0
(y
1
, x
1
, . . . , x
d
1
)
0
(r
2
, u
2
) /
d
2
,0
(y
2
, x
d
1
+1
, . . . , x
d
1
+d
2
)
0
. . .
(r
k
, u
k
) /
d
k
,0
(y
k
, x
d
1
+...+d
(k1)
+1
, . . . , x
d
1
+...+d
k1
+d
k
)
0
where d
1
+ . . . + d
k
= d, d
i
1 for each i, and y
i
1(L
d
1
+...+d
(i1)
, L
d
1
+...+d
i
) (inter-
preting d
0
as 0);
(c) or a regular pair
(r, u) /
d,0
(x
0
, . . . , x
i1
, y, x
i+1
, . . . , x
d
)
0
v

/
1
(y, x
i
)
0
6 SIKIMETI MAU
where 1 i d, and y 1(L
i1
, L
i
);
(d) or a regular pair
v

/
1
(x
0
, y)
0
(r, u) /
d,0
(y, x
1
, . . . , x
d
)
0
where y 1(L
0,AB
, L
d,AB
),
there is an associated continuous gluing map
g : (R
0
, ) /
d,0
(x
0
, . . . , x
d
)
1
dened for some R
0
>> 0, such that g(R) Gromov converges to the given pair/tuple as R .
Moreover, the gluing map surjects onto suciently small Gromov neighborhoods of the given
broken pairs/tuples.
1.3. Organization. Section 2 covers preliminaries on quilts, holomorphic quilt maps, and the
multiplihedra. Section 3 constructs the families of quilted surfaces that we will use. For analytic
simplicity we require all our quilted surfaces to have striplike ends, and we emphasise that:
In this paper, all strips are the standard complex strip,
Z

= z C[0 Imz 1.
Thus the analysis along the striplike ends reduces to the usual theory of pseudoholomorphic
strips, since a quilted strip consisting of strips of equal width can be folded into an ordinary
strip with boundary conditions in a big product manifold (Figure 1.5).
A somewhat ad hoc aspect of the construction comes from the fact that if one simply takes the
canonical bundle of quilted disks over the multiplihedron, there is no natural quilted striplike
end over the marked point z
0
, so we will manually construct the bundles o
d,0

d,0
of
quilted surfaces parametrized by the multiplihedra, with quilted striplike ends built in.
Section 4 denes the moduli spaces of pseudoholomorphic pointed quilted disks. The moduli
spaces incorporate parameters coming from the nite dimensional spaces
d,0
parametrizing
the domains. The framework is based on [7] Part II, sections 7 through 9, which denes
pseudoholomorphic polygons for the Fukaya category. We x a universal choice of perturbation
data for all compactied parameter spaces
d,0
at once, with recursive compatibility properties
with respect to boundary strata, and study solutions of the inhomogeneous pseudoholomorphic
map equations in an appropriate Banach bundle setting. We dene local trivializations in which
the moduli space of holomorphic quilted disks is modeled on zeros of a Fredholm map T
S,r,u
,
and write out the linearized operator D
S,r,u
explicitly.
Section 5 covers the estimates needed to prove Theorem 1. The gluing map is an application
of the Implicit Function Theorem, and the strategy of proof is standard, though the details are
lengthy.
M
5
M
1
M
2
M
3
M
4
L
1
L
1,2
L
2,3
L
3,4
L
4,5
L
5
L
1
L
2,3
L
4,5
M
4
M
1
M
3
M
5
M
2
L
4,5
M
1
M
2
M
3
M
4
M
5
L
1
L
2,3
L
1,2
L
3,4
L
5
L
1,2
L
3,4
L
5
Figure 1.5. Unfolding an ordinary strip into a quilted strip.
Appendices A and B are included for completeness. Appendix A establishes Sobolev em-
bedding estimates that are relevant to the quadratic estimate (Section 5.5). In order to obtain
estimates that are independent of the gluing parameter, we need to establish Sobolev embed-
ding constants that are uniform for all domains in the parametrized family; that is, constants
that do not vary with the parameters coming from the multiplihedra or associahedra. The
main ingredient behind the uniform embedding constants is the fact that the non-compact
striplike ends of all domains in the families satisfy the same cone condition. Appendix B
contains explicit convexity estimates for the solutions of Floers inhomogeneous equation near
non-degenerate intersections.
Acknowledgements. Most of the work in this paper was part of my Ph.D. thesis, carried
out at Rutgers University under the supervision of Chris Woodward. This paper owes a great
deal to his ideas and advice. I also thank Katrin Wehrheim for enormously helpful discussions
on analysis and quilts.
8 SIKIMETI MAU
2. Preliminaries
2.1. Quilts. We review the basic set-up of quilts, following [11].
Denition 2.1.1. A quilted surface S with strip-like ends consists of the following data:
(a) A collection S = (S
)
A
, where each S
is a pointed Riemann surface with markings
1
, . . . ,
n
S
, and for each marking a striplike end, which is a holomorphic

embedding
: R
0
[0, 1] S such that
(s, 0),
(s, 1) S for all s R

0
, and
lim
s
(s, t) uniformly in t. Write c(S
) for the set of boundary components of

S
1
, . . . ,
n
. In general the boundary components in c(S
) are dieomorphic to
R or S
1
, depending on whether they are incident to any marked points or not.
(b) A collection ( of seams, whose elements are tuples
( = (, I), (
, I
),
(I,I
where ,
/, and I c(S
), I
c(S
), I ,= I
, and
(I,I
)
: I

I
is a real-
analytic identication of the boundary components. If I and I
are noncompact, each

of them intersects a striplike end at both ends in this case we require that
(I,I
)
be
compatible with the strip-like ends: that is, if
(s, 0) I then there is a marked point
with
(s, 1) I
and
(I,I
)
(
(s, 0)) =
(s, 1). Similarly if is a marked point on

S
with
(s, 1) I, then there is a marked point
such that
(s, 0) I
and
(I,I
)
(
(s, 1)) =
(s, 0).
(c) Boundary components that do not appear in any seam are called true boundary com-
ponents of S.
2.2. Quilt maps. Label each surface S
S with a symplectic manifold (M
), each seam
= (, I), (, I
) with a Lagrangian correspondence L
, and each true boundary

component (, I) with a Lagrangian submanifold L
(,I)
M
. Denote the labels collectively

by M = (M
)
A
, L = (L
)
C
(L
(,I)
)
A,IE(S)
.
Denition 2.2.1. A (smooth) quilt map u : S M with boundary conditions in L is a tuple
u
A
of smooth maps u
: S
satisfying:
(a) for each true boundary component (, I), u
I
L
(,I)
;
(b) for each seam = (, I), (, I
), u
(
1
I
(s, 0)) u
(
1
I
(s, 0)) L
.
Let C
(S, M; L) be the set of smooth quilt maps from S into target manifolds M with bound-
ary conditions in L.
For each /, let J
: S
(M
) be a choice of
-compatible almost complex

structure J
(z) varying smoothly with z S
. Along a striplike end
(s, t), (s, t) [0, )

[0, 1], we take J
(s, t)) =: J
(t), i.e., to be independent of s. We write j
for the complex

structure on S
.
Denition 2.2.2. A pseudoholomorphic quilt map u : S M with boundary in L is a smooth
quilt map with boundary in L such that for each , J
(z, u
(z)) du
(z) = du
(z).
2.3. Multiplihedra. We describe two realizations of the multiplihedra following [4].
Moduli of quilted disks. A pointed quilted disk is a tuple (D, C, z
0
, . . . , z
d
), where D is the
closed unit disk in C, (z
0
, . . . , z
d
) is an ordered conguration of points in D (compatible with
the orientation of D), and C D is a circle that is tangent to D at the distinguished point
z
0
. The group PSL(R) of complex automorphisms of D acts freely and properly on the space
of pointed quilted disks, and the moduli space of pointed quilted disks is the quotient by this
action, which we denote by
d,0
. The space
d,0
is compactied by stable nodal quilted disks.
The combinatorial type of a stable nodal quilted disk is a stable colored, rooted ribbon tree:
Denition 2.3.1. A colored, rooted ribbon tree T = (E
(T), E(T), V (T), V

col
(T)) is a tree with
vertices V (T), a collection of semi-innite edges E
(T) = e
0
, e
1
, . . . , e
n
, and a collection of
nite edges E(T), together with a distinguished subset V
col
(T) V (T) of colored vertices, such
that
(a) Each semi-innite edge in E
(T) is incident to a single vertex in V (T),

(b) Each nite edge in E(T) is incident to exactly two vertices in V (T),
(c) the ribbon structure at every vertex ensures that the planar structure of T has the
semi-innite edges e
0
, . . . , e
n
arranged in counter-clockwise order; the edge e
0
is called
the root, and e
1
, . . . , e
n
are called the leaves;
(d) each non-self-crossing path from a leaf e
i
to the root e
0
contains exactly one colored
vertex.
The colored ribbon tree is stable if every uncolored vertex has valency 3, and each colored
vertex has valency 2.
Denition 2.3.2. A nodal (d+1)-quilted disk S is a collection of quilted and unquilted marked
disks, identied at pairs of points on the boundary. The combinatorial type of S is a colored
rooted ribbon tree T, where the colored vertices represent quilted disks, and the remaining
vertices represent unquilted disks. A nodal quilted disk is stable if and only if
(a) Each quilted disk component contains at least 2 singular or marked points;
(b) Each unquilted disk component contains at least 3 singular or marked points.
Let
T
denote the stable, nodal quilted disks with combinatorial type T. As a set,
d,0
:=
_
T
T
,
where T ranges over all stable, d-leafed colored rooted ribbon trees.
Metric colored ribbon trees. There is a dual realization of the multiplihedra as metric trees,
which is analogous to the realization of the associahedra as a space of metric ribbon trees. A
metric colored ribbon tree is a colored ribbon tree and a map : E(T) [0, ] of edge lengths
which satisfy the condition that the colored vertices are all the same distance from the root.
If k is the number of colored vertices of T, this condition imposes k 1 relations on the edge
lengths, and denes a cone (
T
[0, )
|E(T)|
of dimension [E(T)[ k + 1.
10 SIKIMETI MAU
z0
z2
z3
z4
z5
z6
z7
z4
z0
z2
z3
z5
z6
z7
z1
z1
Figure 2.1. A pointed quilted disk in
7,0
, and a nodal pointed quilted disk
in the compactication
7,0
.
Denition 2.3.3. A tuple of lengths is admissible if it is in the cone (.
Example 2.3.4. For the tree in Figure 2.2, the edge lengths satisfy the relations
1
+
2
+
3
=
1
+
2
+
4
=
1
+
2
+
5
=
1
+
6
=
7
.
These are equivalent to the relations
3
=
4
=
5
,
3
+
2
=
6
, and
6
+
1
=
7
.
1
2
3 5
4
6
7
8
Figure 2.2. A colored ribbon tree with interior edges labeled by gluing lengths (
1
, . . . ,
8
).
For each stable colored, rooted ribbon tree T, set
C
T
= : E(T) [0, ]
(
T
, C
d,0
=
_
_
T
C
T
_
/
where the equivalence relation identies an edge of length 0 in T with the tree obtained from
T by contracting that edge. In [4] it is shown that C
d,0
=
d,0
.
3. Families of quilts
3.1. Pointed disks and associahedra. The construction of higher compositions in the Fukaya
category is based on families of pointed disks. We recall an explicit map in [2] identifying the
moduli space of metric ribbon trees with the moduli space of pointed disks.
Let d 2 and let (T, ) = (V (T), e
0
, e
1
, . . . , e
d
, E(T), : E(T) R
+
) be a metric ribbon
tree: V (T) are vertices of a tree T, e
0
, e
1
, . . . , e
d
are semi-innite (exterior) edges, and E(T)
are nite (interior) edges, and for each e E(T), (e) 0 is the length of e. Thicken
each semi-innite edge e
i
to a semi-innite strip, Z
i
= [0, ) [0, 1], and thicken each edge
e E(T) of length (e) to a nite strip Z
e
= [0, (e)] [0, 1]. Consider the punctured strips
Z
i
(0,
1
2
), i = 0, . . . , d and Z
e
(0,
1
2
), ((e),
1
2
), e E(T). Use the cyclic ribbon structure of
incident edges at each vertex v to dene identications of adjacent half-intervals at the ends of
adjacent strips (see Figure 3.1). The resulting Riemann surface with boundary has a hole for
each vertex in the tree, but each hole can be conformally lled to produce a Riemann surface
that is biholomorphic to a d+1-pointed disk, and the semi-innite strips Z
0
, . . . , Z
d
determine
holomorphic embeddings of striplike ends at the points.
v
e
a
e
b
e
c
Z
ea
Z
eb
Z
ec
Figure 3.1. Identifying strips at a trivalent vertex.
3.1.1. Attaching strips. A generalized d + 1-pointed disk (Figure 3.2) is a quilted surface ob-
tained by attaching strips to boundary components of a d + 1-pointed disk. In the above
construction, the boundary components of the Riemann surfaces are naturally identied with
the boundary of an innite strip R [0, 1]. The identication is unique up to translation, and
can be made translationally invariant for all surfaces in the family by requiring that a common
boundary point (for instance, a xed point on one of the striplike ends) always be identied
with 0. Thus one can x a unique attaching map of a standard strip to that boundary com-
ponent. Proceeding inductively one can attach further strips to the boundary of an attached
strip, without introducing any additional degrees of freedom.
3.2. Quilts parametrized by multiplihedra. Let d 1. We will construct a ber bundle
o
d,0

d,0
of quilted surfaces that are dieomorphic to pointed quilted disks, but are
equipped with strip-like ends consisting of quilted strips, such that all strips have width 1. We
will do it in two steps: rst, constructing the quilted surfaces, and second, xing the complex
structures.
12 SIKIMETI MAU
z0
z1
z2
z4
z3
Figure 3.2. A quilted surface parametrized by a marked disk in
4
.
3.2.1. Constructing the quilted surfaces. The construction is based on the fact that the mul-
tiplihedron is parametrized by metric trees with some additional metric data. The quilted
surfaces that we dene will be pointed Riemann surfaces, as constructed in Section 3.1, which
are parametrized by the underlying metric tree. The extra metric data will be used to x an
embedded 1-submanifold.
For d 1, we will dene the bundle over the boundary of the multiplihedron J
d
in terms of
lower strata, extend the bundle over a neighborhood of the boundary by gluing striplike ends,
and nally x a smooth interpolation of the bundle over the interior of the d-th multiplihedron.
Base step: For d = 1, let Z = R [0, 1] be the standard strip of unit width. Fix a smooth,
connected embedded 1-manifold L Z such that
L Z
1
= R
0

1
3
,
2
3
,
L Z
0
= ,
L Z = .
Denote this quilted surface by o
1,0
. As shorthand, for x, y R we let o
1,0
x
, o
1,0
y
denote the
pieces of o
1,0
corresponding to Z
x
, Z
y
respectively.
s s = 1
L
s = 0
Figure 3.3. The basic piece o
1,0
.
Inductive hypothesis: Assume that all bundles o
e,0

e,0
have been constructed for 1 e < d.
Assume furthermore that the bundles o
e,0

e,0
are compatible with the bundles o
e

e
constructed in the previous section, i.e., there is a commuting diagram
(3.1)
o
e,0

e,0
_
o
e

e
where the vertical maps are forgetful maps. That is,
e,0

e
is the map from stable metric
colored trees to stable metric trees that forgets the additional colored structure, and o
e,0
o
e
is the map from a quilted pointed Riemann surface to an unquilted pointed Riemann surface,
that forgets the embedded submanifold L.
Inductive step: We extend the bundle o
d

d
over an open neighborhood of
d
, as follows.
The codimension one facets of
d
are indexed by two types of colored tree (Figure 3.4). Facets
of Type 1 are denoted by F
(e,i)
where the label (e, i) is such that 0 i d, 2 e d i.
We denote the indexing tree by T
(e,i)
. Facets of Type 2 are denoted by F
s
1
+s
2
+...+sr
, where the
labels s
1
+. . . +s
r
satisfy r 2, 1 s
i
d 1, s
1
+ s
2
+ . . . s
r
= d. We denote the indexing
tree by T
s
1
+...+sr
.
Figure 3.4. Combinatorial types of facets of the multiplihedron: Type 1 (left)
and Type 2 (right).
Type 1: Consider a boundary facet F
(e,i)

d,0
; denote the single edge of T
(e,i)
by a. Every
point in
d,0
is represented by a metric colored tree, (T, ), and such a metric colored tree is
in F
(e,i)
if and only if
(a) there exists a morphism p : T T
(e,i)
of trees,
(b) for the unique edge p
1
(a) T, (p
1
(a)) = .
Cutting the edge p
1
(a) produces a pair (T
1
,
1
)
de+1,0
, and (T
2
,
2
)
e
, where
i
:= [
T
i
. Note that the edge a of T has become the leaf e
i+1
of T
1
and the root e
0
of T
2
. Let
o
(T
1
,
1
)
and o
(T
2
,
2
)
be the corresponding surfaces, constructed by inductive hypothesis.
Let := (p
1
(a)) [0, ] be a normal coordinate to the facet F
(e,i)
, i.e., the length of the
edge p
1
(a). For 1, truncate the strips Z
e
i
and Z
e
0
at length /2 and identify the two
strips along the truncations; this replaces the semi-innite strips Z
e
i
and Z
e
0
with a nite strip
of length .
Type 2: Now consider a boundary facet F
s
1
+...+sr

d,0
; denote the r edges of T
s
1
+...+sr
by
a
1
, . . . , a
r
. A metric colored tree (T, ) is in F
s
1
+...sr
if and only if
14 SIKIMETI MAU
(a) there exists a morphism p : T T
s
1
+...sr
of trees,
(b) for i = 1, . . . , r, (p
1
(a
i
)) = .
Cutting the edges p
1
(a
i
) in T produces a tuple (T
0
,
0
)
r
, (T
1
,
1
)
s
1
,0
, . . . , (T
r
,
r
)
sr,0
. Note that each edge a
i
has become the leaf e
i
of T
0
and the root e
0
of T
i
. By inductive
hypothesis the surfaces o
(T
0
,
0
)
, o
(T
1
,
1
)
, . . . , o
(Tr,r)
have already been constructed. We x a
normal coordinate to the facet F
s
1
+...+sr
, taking := (p
1
(a
1
)); the relations on the colored
tree T then determine the values of admissible lengths (p
1
(a
2
)), . . . , (p
1
(a
r
)). Truncate
both the semi-innite strip Z
e
i
of T
0
and the semi-innite strip Z
e
0
of T
i
at length (p
1
(a
i
))/2,
and identify them along the truncations.
These constructions are only well-dened for large , dening the bundle o
d,0

d,0
over
an open neighborhood of
d,0
. The bundle is compatible by construction with the forgetful
maps, i.e., the diagram (3.1) commutes. We extend the bundle over the remainder of the
interior of
d,0
by choosing a family of smooth isotopies. All of the surfaces in the family
have the same striplike ends, so we may assume that the isotopies are compactly supported.
Furthermore, we may also choose the isotopies to be compatible with the forgetful maps, that
is, we may choose them such that (3.1) commutes. This completes the inductive step.
3.2.2. Fixing the complex structures. The nal step is to x an area form on each quilted
surface in the bundle o
d,0

d,0
, and a compatible complex structure, using the fact that
the space of area forms is convex, and the space of compatible complex structures contractible.
Given coordinates (s, t) we say that the area form, and complex structure are standard if they
are respectively ds dt, and j(
s
) =
t
, j(
t
) =
s
. We x the area form and complex
structure as follows. First, on each striplike end, the area form and complex structure are
chosen to be the standard ones on each strip. In particular, on the quilted striplike end, the t
coordinate corresponds to a rescaling of the corresponding coordinate of the underlying strip
by a factor of 3. Second, choose a tubular neighborhood for each boundary component, and
a tubular neighborhood of the seam. For the tubular neighborhood of the seam, x local
coordinates (s, t) so that along the striplike end they coincide with the local coordinates xed
there, and then take the standard area form and complex structure for these coordinates.
For each boundary component, one takes a tubular neighborhood of that component in the
coordinates of the underlying surface, rescaling if necessary in the t direction in order to get
local coordinates which match the standard coordinates on the striplike ends, and then take
the standard area form and complex structure with respect to those coordinates. Lastly, using
convexity of area forms one can smootly extend the area form over the remainder of the
quilt, and then smoothly extend the compatible complex structure over the remainder of the
quilt. To achieve consistency one builds up choices of area form and complex structure over
all bundles o
d,0

d,0
inductively, using the gluing procedure to determine area form and
complex structure for all quilts in the bundle over a neighborhood of
d,0
, and then smoothly
interpolating both over the interior of
d,0
.
3.2.3. Attaching strips. Each boundary component is by construction equipped with a tubular
neighborhood on which the complex structure is the standard complex structure for a strip.
Strips of unit width can be real-analytically attached to the boundary components of the quilted
surfaces in the families o
d,0
; we call the generalized quilted surfaces obtained in this manner
generalized pointed quilted disks (Figure 3.5).
z0
z2
z4
z3
z1
Figure 3.5. A generalized quilted disk, parametrized by a quilted marked disk.
Denition 3.2.1. Borrowing more terminology from [7], we say that the thin part of a quilt
o
r
consists of the striplike ends, together with the images of truncated striplike ends that were
glued to form o
r
. The complement of the thin part will be called the thick part. We call this
the thick/thin decomposition of o
r
.
4. Pseudoholomorphic quilted disks
The framework that we use is based on that of [7], Part II, sections 7 through 9.
4.1. Floer and perturbation data. To each striplike end of S, with Lagrangian boundary
conditions given by L, we assign a Floer datum, which is a regular pair
(H, J) = ((H
k
)
k=1,...,m
, (J
k
)
k=1,...,m
)
of a Hamiltonian perturbation and an almost complex structure of split type.
Denition 4.1.1. Let L, L
be generalized Lagrangian submanifolds of M, and let (H, J) be

a Floer datum for the pair. A generalized intersection of a pair of generalized Lagrangians
L = pt
L
n
M
n
L
n,n+1
. . . M
1
L
1,0
M
0
= M,
L
= pt
L
m
M
m
L
m,m1
. . . M
1
L
1,0
M
0
= M
given a Hamiltonian perturbation H is a tuple of paths
x = (x
i
)
m
i=n
[ x
i
: [0, 1] M
i
, x
i
(t) = X
H
i
t
(x
i
(t)), (x
i1
(1), x
i
(0)) L
i1,i
.
We denote the set of generalized intersections by 1(L, L
).
Fix a collection of generalized Lagrangian submanifolds L
0
, . . . , L
d
of a symplectic manifold
M
A
. Given a Lagrangian correspondence L
AB
between M
A
and M
B
, write
L
0,AB
:= L
0
#L
AB
, L
d,AB
:= L
d
#L
AB
for the concatenations, which are generalized Lagrangian submanifolds of M
B
.
Assign each pair (L
i
, L
i+1
) a Floer datum (H
i
, J
i
), and assign (L
0,AB
, L
d,AB
) a Floer datum
(H
AB
, J
AB
). A perturbation datum is a pair (K, J), K in the perturbation datum is a smooth
16 SIKIMETI MAU
family of 1-forms on the bers S
r
which take values in the space of Hamiltonian functions on
M. Thus, a choice of K determines a 1-form Y on each ber S
r
, taking values in the space of
Hamiltonian vector elds on M. A perturbation datum is compatible with the striplike ends
and Floer data if it is equal on each striplike end to the Floer datum for the pair of Lagrangians
labeling the end. We x a universal choice of perturbation data (i.e., for all bundles o
d,0
and
o
d
), which is consistent with the recursive constructions of the bundles, i.e., the following
conditions hold: rst, for each d 1 (resp. d 2) there is a subset U
0,d
(resp.
d
),
where gluing parameters are suciently small, such that the perturbation data is given by the
Floer data on the thin parts of the surfaces o
r
, r U. Second, if (

K,

J) is the perturbation
datum on U o
d,0
(resp. o
d
) obtained by gluing perturbation data of the lower strata, then
the perturbation datum in the universal family for o
d,0
(resp. o
d
) extends smoothly over
d,0
(resp.
d
) and agrees with (

K,

J) over the boundary
d,0
(resp.
d
).
Fix a quilted surface S
r
0
, r
0

d,0
, which is labeled by the Lagrangians L
0
, . . . , L
d
and the
Lagrangian correspondence L
AB
. In a neighborhood U of r
0
all of the quilted surfaces are
dieormorphic, giving rise to a family of dieomorphisms parametrized by points in U,
: U o
r
0
o
U
, (r, ) : o
r
0
=
o
r
which identify strip-like ends, i.e., (r,
(r
0
, s, t)) =
(r, s, t). Each quilted surface S

r
is
equipped with complex structures j
Sr
which can be pulled back by
r
to give a family of
complex structures on S
r
0
parametrized by r, j(r) :=
r
(j
Sr
).
The inhomogeneous pseudo-holomorphic map equation for (r, u) B is
(4.1)
_
du(z) +J(r, u, z) du(z) j(r) = Y (r, u, z) +J(r, u, z) Y (r, u, z) j(r)
u(C) L
C
for all seams/boundary components C with label L
C
.
The compatibility of the perturbation data with the Floer data along the striplike ends means
that the above equation reduces to Floers equation along the striplike ends. In particular,
solutions with nite energy converge exponentially along each striplike end to a generalized
intersection of the pair of generalized Lagrangians labeling that striplike end.
A tuple (y
0
, x
1
, . . . , x
d
) 1(L
0,AB
, L
d,AB
) 1(L
0
, L
1
) . . . 1(L
d1
, L
d
) of generalized
intersections determines a ber bundle B
S
U
U
d,0
, whose ber over r is the space of
smooth quilt maps from o
r
to M with boundary in L and which converge along striplike ends
to the given tuple of intersection points. There is another ber bundle
c
S
U
B
S
U
,
whose ber c
(r,u)
=
0,1,r
(o
r
, u
TM) of (0, 1) forms on o

r
taking values in the pullback bundle
u
TM, and the (0, 1) part is with respect to J(r, u) and j(r).
Since equation (4.1) can be written as (du Y )
0,1
= 0, we will also abbreviate it as (
)(r, u) = 0, where (r, u) := (du)
0,1
and (r, u) := (Y (r, u))
0,1
, with the 0, 1 taken with respect
to J(r, u, z) and j(r).
Thus : B c denes a section whose intersection with the zero-section B
0
is the set
of solutions of (4.1), whose striplike ends converge to the prescribed intersections. The energy
of such a solution (r, u) is the quantity
E(r, u) =
1
2
_
Sr
[du Y [
2
dvol
Sr
=
1
2
A
_
Sr,
[du
[
2
dvol
Sr,
where the norms [ [ on TM
come from the induced metric g

J(r)
, and the area form dvol
Sr,
comes from the construction of the bundles in Section 3.
Denition 4.1.2. The moduli space of holomorphic quilted disks /
d,1
(y
0
, x
1
, . . . , x
d
) consists
of all (nite energy) solutions (r, u) to (4.1) which converge along the strip-like ends labeled
0
, . . . ,
d
to the respective intersections (y
0
, x
1
, . . . , x
d
). For a collection (x
0
, x
1
, . . . , x
d
)
1(L
0
, L
d
)1(L
0
, L
1
). . .1(L
d1
, L
d
), the moduli space of holomorphic disks /
d
(x
0
, x
1
, . . . , x
d
)
consists of all (nite energy) solutions of the analogous equations for the surfaces o
d

d
,
with generalized Lagrangian boundary conditions L
0
, . . . , L
d
. Similarly, given x, y 1(L, L
),
the moduli space of Floer trajectories is

/(x, y) := /(x, y)/R, where /(x, y) consists of all
(nite energy) solutions to Floers inhomogeneous equation converging to x and y.
4.2. Sobolev spaces. Consider a quilt consisting of surfaces S = (S
)
A
and set ( of seams
and boundary components, equipped with labelings of the surfaces by target symplectic man-
ifolds M = (M
)
A
and labelings of the seams and boundary components by Lagrangian
boundary conditions L = (L
C
)
CC
. Let u : S M be a smooth quilt map with boundary in
L, and let J = (J
(z), z S
)
A
be a smooth family of compatible almost complex structures
varying over the surfaces. Given an area form on each patch of S, dene Sobolev norms on
sections of the pull-backs of the tangent bundles by
W
1,p
(S, u
TM) :=
A
W
1,p
(S
, u
TM
),
L
p
(S,
0,1
J
u
TM) :=
A
L
p
(S
,
0,1
J
u
TM
)
where the norms on TM
come from the induced metrics g

J(z)
(, ) := (, J
(z)).
4.3. Metric connections and local trivializations. We recall the following lemma of Frauen-
felder on the existence of metrics with useful properties with respect to a given almost-complex
structure and a given Lagrangian submanifold.
Lemma 4.3.1 (Lemma 4.3.3 in [5]). Let (M, ) be a symplectic manifold equipped with an
almost complex structure J, and let L M be a Lagrangian submanifold. There exists a
Riemannian metric g = , ) on M such that
(i) J(p)v, J(p)w) = v, w) for p M and v, w T
p
M,
(ii) J(p)T
p
L is the orthogonal complement of T
p
L for every p L,
(iii) L is totally geodesic with respect to g.
Let be the Levi-Civita connection of g. The Hermitian connection

associated to an
almost complex structure J is
(4.2)

v
X :=
v
X
1
2
J(
v
J)X.
18 SIKIMETI MAU
Lemma 4.3.2. Under the assumptions of Lemma 4.3.1, L is totally geodesic with respect to
, the Hermitian connection (4.2).

Proof. Write P : TM
L
TL for orthogonal projection of TM onto TL dened pointwise on
L. It suces to check that for every p L, and every X, Y T
p
L,

X
Y T
p
L. By denition,
X
Y =
X
Y
1
2
J(
X
J)Y.
where is the Levi-Civita connection of g. By assumption, L is totally geodesic with respect
to , therefore
X
Y T
p
L for all p L and all X, Y T
p
L. So it is enough to show that
J(
X
J)Y T
p
L. Since the orthogonal complement of T
p
L is JT
p
L, it reduces to showing that
for all X, Y, Z T
p
L, J(
X
J)Y JZ.
g(J(
X
J)Y, JZ) = g((
X
J)Y, Z) = d(g(JY, Z))(X) g(J
X
Y, Z) g(JY,
X
Z) = 0.
Now let o be a family of quilted surfaces of type either o

d

d
or o
d,0

d,0
, as
constructed in Section 3. We x a basepoint r
0
, and a neighborhood | of r
0
. Write
S := o
r
0
for the quilted surface over r
0
. All the quilted surfaces over U are dieomorphic,
and moreover the dieomorphisms can be chosen to be the identity on the striplike ends, so
the bundle o
U
U can be viewed as a single quilted surface S with a family of complex
structures parametrized by r U, such that the complex structures j(r) are all standard on
the striplike ends.
Each patch S
of the quilt S is labeled by a target symplectic manifold (M
) and an
-compatible almost complex structure J
= J
(r
0
), which determine an induced metric
g
J(r
0
)
. We collectively write g for these metrics, and write exp for the exponential maps of
their associated Hermitian connections (4.2). Note that once a base point r
0
is xed, this xes
a metric on each M
for each z S
.
The Lagrangian boundary conditions are not necessarily totally geodesic with respect to
the exponential maps in exp. We therefore introduce a collection of auxiliary metrics g
Q
,
parametrized by the quilt S, to satisfy the prescribed boundary conditions at seams and bound-
ary components (c.f. [11, Remark 2.2]).
For each boundary component or seam C labeled by a Lagrangian L
C
M
, x a
metric g
C
on M
that satises properties (i), (ii) and (iii) of Lemma 4.3.1 with respect to
the almost complex structure (J
) J
and the Lagrangian L

C
. A -tubular neighborhood
of C in the quilt is a copy of R[, ], where the side where t < 0 is labeled by M
, the side
where t > 0 is labeled by M
, and the seam t = 0 is labeled by L

C
. We fold this to R [0, ]
labeled by the product manifold M
and the Lagrangian boundary condition L

C
. Over
the interval [0, ] x a smooth interpolation (of metrics on M
) between the metric g

C
and the induced metric g
J
g
J
. Note that both g

C
and g
are (J
) J
-invariant, so
we can choose the interpolating metrics to be (J
) J
-invariant too. On the complements

of the tubular neighborhoods, we take g
Q
to be the induced metric g.
We write g
Q
for the resulting collection of metrics parametrized by the quilt S. We emphasise
that on each tubular neighborhood of a seam, we view g as a metric on the associated product
manifold, parametrized by the folded tubular neighborhood. Let exp
Q
denote the collection of
exponential maps of the associated Hermitian connections (4.2). We will use exp
Q
to dene
quadratic corrections to exp.
Dene a function Q : S
0
(S, u
TM)
0
(S, u
TM) as follows. Let

0
(S, u
TM) be
a section of the pull-back bundle over the quilt S. In a tubular neighborhood of a true boundary
component of the patch S
, with local coordinates z = s+it Ri[0, ), dene Q((z, )(z) by

the condition that exp
u(z)
((z) + Q(z, )(z)) = exp
Q
u(z)
((z)). In a tubular neighborhood of
a seam between patches S
l
and S
r
, suppose we have local coordinates z = s +it Ri[, ],
where S
l
corresponds to t < 0 and S
r
corresponds to t > 0. For z = s + it S
l
we dene
Q(z, )(z) and Q(z, )(z) (where z = s it S
r
) by the condition that
(exp
u
l
(z)
(
l
(z) +Q(z, )(z)), exp
ur(z)
(
r
(z) +Q(z, )(z)) = exp
Q
(u
l
(z),ur(z))
(
l
(z),
r
(z))
as points on the product manifold M
l
M
r
. For a point z S in the complement of the
tubular neighborhoods, Q(z, ) = 0 for all
0
(S, u
TM).
Lemma 4.3.3. Q(z, ) dened above has the following properties.
(a) For all z S, Q(z, 0) = 0.
(b) For all z S, and
0
(S, u
TM),
Q(z, )[
=0
= dQ(z, 0)(0, ) = 0.
Proof. (a) follows fromu = exp
Q
u
(0) = exp
u
(Q(z, 0)). (b) follows from dierentiating exp
Q
u
() =
exp
u
( +Q(z, )) with respect to at = 0, which gives = +
Q(z, )[
=0
.
The following lemma will be relevant for explicit computations of linearized operators.
Lemma 4.3.4. Let (M, ) be a symplectic manifold, p M and T
p
M. Fix a connection
on TM, with exponential map exp. Suppose that Q : T
p
M T
p
M satises Q(0) = 0 and
DQ(0) = 0. For [0, ), and t [0, 1], let
(, t) = exp
p
(t +Q(t))
(, t) = exp
p
(t( +Q())).
(Note that these paths coincide at t = 0, 1.) Let
p
() denote parallel transport along the
curve t (, t), and let () denote parallel transport along the curve t (, t). Then,
for every T
p
M, we have that
(4.3)
d
d
=0
p
()
1
p
() = 0.
Thus, in particular, for any vector eld f along the curve exp( +Q()),
(4.4)
d
d
=0
p
()
1
f =
=0
.
20 SIKIMETI MAU
Proof. Let p
= (, 1) = (, 1) be the common end point of the two paths for xed . It

suces to show that the dierence
p
()
p
() T
p
M
is only quadratic in . For suciently small both curves must be contained in a local
coordinate chart and we can calculate the dierence directly. Let x
1
, . . . , x
n
be local coordinates
for a neighborhood of p in M. Write
((, t)) for the parallel transports of T along

(, t), and write
((, t)) for the parallel transports of along (, t). In terms of the local
coordinate components, for j = 1, . . . , n we have from the parallel transport equations and the
fundamental theorem of calculus that
(p
) =
j
(p)
1
_
0
j
ih
(
(t))
i
(t))
d
h
dt
dt

j
(p
) =
j
(p)
1
_
0
j
ih
(
(t))
i
(t))
d
h
dt
dt
where
j
ih
are the Christoel symbols for in these coordinates. Since
(p) = =
(p), we
have that in each coordinate, j = 1, . . . , n,
(p
)
j
(p
) =
1
_
0
(
j
ih
(
(t))
i
(t))
d
h
dt

j
ih
(
(t))
i
(t))
d
h
dt
) dt.
We now compute the derivative of this quantity with respect to at = 0.
d
d
=0
(
j
(p
)
j
(p
)) =
d
d
=0
1
_
0
(
j
ih
(
(t))
i
(t))
d
h
dt

j
ih
(
(t))
i
(t))
d
h
dt
dt
=
1
_
0
j
ih
(
(t))
i
(t))
d
h
dt

j
ih
(
(t))
i
(t))
d
h
dt
_
=0
dt.
We now show that the integrand is zero. When = 0, (0, t) = (0, t) = p and therefore
d
h
dt
[
=0
=
d
h
dt
[
=0
= 0. Hence, the integrand above reduces to
j
ih
(p)
i

_
d
h
dt

d
h
dt
_
=0
and now it is enough to show that

_
d
h
dt

d
h
dt
_
=0
= 0. Using the equality of mixed
partials for the smooth function
h
(, t)
h
(, t) we can write
t
(
h
(, t)
h
(, t))
_
[
=0
=

t
_

(
h
(, t)
h
(, t))[
=0
_
.
Now using the denitions of (, t) and (, t) we see that
(, t)[
=0
= t +DQ(0)(t) = t,
while
(, t) = t +tDQ(0)() = t, so the quantities above must be zero.

4.4. Local trivializations. We now dene local trivializations of the bundles B[

U
U and
c B[
U
. A small neighborhood of (r
0
, u
0
) in B
U
consists of pairs (r, u) where r U
parametrizes the complex structure j(r) on S, and u : S M is a map satisfying the
prescribed Lagrangian boundary conditions and limits along the striplike ends. For u suf-
ciently close to u
0
, there is a unique
0
(S, u
0
TM) such that exp
Q
u
0
() = u. Write
Q
u
() : T
u
M T
exp
Q
u
()
M for parallel transport along curves exp
Q
u
() = exp( + Q()),
[0, 1], with respect to the Hermitian connection

. We emphasize that these curves are
not geodesics for the connection

, since they are dened with the corrected exponential map
exp
Q
and not with exp.
Dene a projection map
[ ]
0,1
(r)
=
r
:
1
(S, u
TM)
0,1
r
(uS, u
TM),
1
2
( +J(r) j(r))
i.e., projection onto the (0, 1) part with respect to J(r) and j(r). Note that if
0,1
r
(S, u
TM),
then
r
() = . Also, if r
is close to r, then
r
determines an isomorphism
r
:
0,1
r
(S, u
TM)
0,1
r
(S, u
TM).
The bers of c over a small neighborhood of (r
0
, u
0
) can be identied using the isomorphisms
S,r
0
,u
0
(, )
1
:
0,1
exp
r
0
()
(S, exp
Q
u
0
()
TM)
0,1
r
0
(S, u
0
TM)

Q
u
0
()
1
1
2
( +J(r
0
, u, z) j(r
0
)) .
Fix a metric on the compact nite dimensional space once and for all, and an exponential
map exp
r
0
: T
r
0
U U. Use the exponential map on and the maps exp
Q
on M to identify
B[
U
= T
r
0
0
(S, u
0
TM).
The vertical part of the section : B c at (r
0
, u
0
) is given by the non-linear map
T
S,r
0
,u
0
: T
r
0
0
(S, u
0
TM)
0,1
r
0
(S, u
0
TM)
(, )
S,r
0
,u
0
(, )
1
( )(exp
r
0
, exp
Q
u
0
),
which extends to a non-linear function between Banach completions
T
S,r
0
,u
0
: T
r
0
W
1,p
(S, u
0
TM) L
p
(S,
0,1
J(r),j(r)
u
0
TM).
4.5. The linearized operator. We give explicit computations of the linearized operators.
For convenience we introduce the notation
T
S,r,u
(, ) :=
S,r,u
(, )
1
(exp
r
, exp
u
)
T
S,r,u
(, ) :=
S,r,u
(, )
1
(exp
r
, exp
u
),
22 SIKIMETI MAU
(T for perturbation term). The corresponding linearized operators are given by
dT
S,r,u
(0, 0)
. .
=:D
S,r,u
= d
T
S,r,u
(0, 0)
. .
=:
e
D
S,r,u
dT
S,r,u
(0, 0)
. .
=:P
S,r,u
.
Abbreviating r
:= exp
r
(), u
:= exp
Q
u
(), we can write
D
S,r,u
(, ) =
d
d
=0
Q
u
()
1
1
2
_
(r
, u
) +J(r, u
) (r
, u
) j(r)
_
=
1
2
_
d
d
=0
Q
u
()
1
(r
, u
) +J(r, u)
d
d
=0
Q
u
()
1
(r
, u
) j(r)
_
=
_
d
d
=0
Q
u
()
1
(r
, u
)
_
0,1
where [ ]
0,1
denotes projection onto the (0, 1) part with respect to J(r, u) and j(r). By direct
calculation we get
d
d
=0
Q
u
()
1
(r
, u
) =
d
d
=0
Q
u
()
1
1
2
(du
+J(r
, u
) du
j(r
))
=
1
2
_
du
J(r
, u) du j(r)
+J(r, u)

du
j(r) +J(r, u) du
j(r
)
_
=
_
du
_
0,1
+
1
2
J du j(r) +
1
2
J(r, u) du
j.
Thus the linearized operator

D
S,r,u
splits into the two parts
D
S,r,u
(, ) =
_
du
_
0,1
. .
=:
e
D
(r)
u
()
+
_
1
2
J du j(r) +
1
2
J(r, u) du
j
_
0,1
. .
=:
e
D
(u)
r
()
, (4.5)
and identical arguments as in [5, Proposition 3.1.1] give the explicit local formula
(4.6)

D
(r)
u
() = []
0,1
1
2
J(r, u)(
J)(r, u)
J(r)
(u),
in terms of the Levi-Civita connection of the metrics g
J(r)
. Similarly,
P
S,r,u
(, ) =
_
d
d
=0
Q
u
()
1
1
2
(Y (r
, u
) +J(r
, u
) Y (r
, u
) j(r
))
_
0,1
=
_
Y (r, u
)
_
0,1
. .
=:P
(r)
u
()
+ [
Y ]
0,1
+ [
J Y j +J Y
j]
0,1
. .
=:P
(u)
r
()
(4.7)
and P
(r)
u
is explicitly given locally by
(4.8) P
(r)
u
() =
_
Y
_
0,1
1
2
_
J
J Y
_
0,1
.
4.6. Gromov convergence.
Denition 4.6.1. Consider a sequence (r
n
, u
n
)
n=1
/
d,0
(x
0
, . . . , x
d
).
(a) For 2 e d 1, we say that the sequence Gromov converges to the broken pair
(r
1
, u
1
) /
de+1,0
(x
0
, x
1
, . . . , x
i
, y, x
i+e+1
, . . . , x
d
)
(r
2
, u
2
) /
e
(y, x
i+1
, . . . , x
i+e
)
if
r
n
r
1
#
0
r
2
in the topology of
d,0
near the boundary point r
1
#
0
r
2

d,0
,
E(u
n
) E(u
1
) +E(u
2
),
u
n
converges uniformly on compact subsets of o
r
1
to u
1
, and converges uniformly
on compact subsets of o
r
2
to u
2
.
(b) For 1 s
1
, . . . , s
k
d 1 such that s
1
+ . . . + s
k
= d, we say the sequence Gromov
converges to the broken tuple
(r
0
, u
0
) /
k
(x
0
, y
1
, . . . , y
k
)
(r
1
, u
1
) /
s
1
,0
(y
1
, x
1
, . . . , x
s
1
)
(r
2
, u
2
) /
s
2
,0
(y
2
, x
s
1
+1
, . . . , x
s
1
+s
2
)
. . .
(r
k
, u
k
) /
s
k
,0
(y
k
, x
ds
k
+1
, . . . , x
d
)
if
r
n
r
0
#
0
(r
1
, . . . , r
k
)
d,0
in the topology of
d,0
near the boundary,
E(u
n
) E(u
0
) +E(u
1
) +. . . +E(u
k
),
u
n
r
j
to u
j
, for j = 0, . . . , k.
(c) For i 1, . . . , d, we say that the sequence Gromov converges to the broken pair
(r, u) /
d,0
(x
0
, x
1
, . . . , x
i1
, y, x
i+1
, . . . , x
d
)
v

/
1
(y, x
i
)
if
r
n
r in
d,0
, where r is in the interior of
d,0
,
E(u
n
) E(u) +E(v),
u
n
r
to u, and there is a sequence
n

R (shift parameters) such that if (s, t) denote coordinates on the strip R [0, 1],
and
i
: R
0
[0, 1] o
r
is the i-th striplike end of o
r
, then the sequence of
shifted maps u
n
(
i
(s +
n
, t)) converges uniformly on compact subsets of R[0, 1]
to a xed parametrization of the Floer trajectory v.
(d) We say that the sequence Gromov converges to the broken pair
(r, u) /
d,0
(y, x
1
, . . . , x
d
)
v

/
1
(x
0
, y)
if
r
n
r in
d,0
, where r is in the interior of
d,0
,
E(u
n
) E(u) +E(v),
24 SIKIMETI MAU
u
n
r
to u, and there is a sequence
n

R (shift parameters) such that if (s, t) denote coordinates on the strip R [0, 1],
and
0
: R
0
[0, 1] o
r
is the 0-th striplike end of o
r
, then the sequence of
shifted maps u
n
(
0
(s+
n
, t)) converges uniformly on compact subsets of R[0, 1]
to a xed parametrization of the Floer trajectory v.
4.7. Gromov neighborhoods. We now dene what we call Gromov neighborhoods of a broken
quilt of Type 1, 2 or 3. For > 0, we will dene a subset U
B
d,0
. Under these denitions,
a sequence (r
, u
) /
d,0
(x
0
, . . . , x
d
) will Gromov converge to a broken quilt if, and only if,
given > 0 there is a
0
such that (r
, u
) U
for all
0
.
Type 1. Let (r
1
, u
1
) and (r
2
, u
2
) be a broken pair,
(r
1
, u
1
) /
de+1,0
(x
0
, x
1
. . . , x
i1
, y, x
i+e+1
, . . . , x
d
)
0
(r
2
, u
2
) /
e
(y, x
i
, x
i+1
, . . . , x
i+e
)
0
.
A small neighborhood of the point r
1
#
0
r
2

d,0
is of the form
U

= U
1
U
2
[0, )
where U
1

de+1,0
is a neighborhood of r
1
, and U
2

e
2
, and the
interval [0, ) represents the gluing parameter. Recall that a gluing parameter corresponds
to a gluing length R() = log().
Fix a metric on U
1

de+1,0
and a metric on U
2

e
, and dene a metric topology on
U

= U
1
U
2
[0, ) by
dist
U
(r
1
#
r
2
, r
1
#
2
) := supdist
U
1
(r
1
, r
1
), dist
U
2
(r
2
, r
2
), [
[.
By the construction of the surface bundles o , we can suppose that the neighborhood
U is suciently small that the corresponding neighborhoods U
1

de+1,0
and U
2

e
are
also small enough that all surfaces in the bundles over them are dieomorphic to each other
by dieomorphisms preserving the striplike ends. Write o
r
1
#
r
2
= o
r
1
o
r
2
/ , where o
r
i
represents the truncation of o
r
i
along the prescribed striplike end at s = R(), and is the
identication of the two truncated surfaces along the cuts.
Denition 4.7.1. Let > 0 be given. Dene a Gromov neighborhood U
B
d,0
of the pair
(r
1
, u
1
), (r
2
, u
2
) as follows: (r, u) U
if
r = r
1
#
r
2
U with dist
U
( r
1
#
r
2
, r
1
#
0
r
2
) < ,
[E(u
1
) +E(u
2
) E(u)[ < ,
dist
M
(u(z), u
1
(z)) < for all z o
r
1
,
dist
M
(u(z), u
2
(z)) < for all z o
r
2
.
The metrics on the target manifolds M are those induced by their symplectic forms and the
choice of compatible almost complex structures J = J(z).
Type 2. Let (r
0
, u
0
), . . . , (r
k
, u
k
) be a broken tuple of the form
(r
0
, u
0
) /
k
(x
0
, y
1
, . . . , y
k
)
(r
1
, u
1
) /
s
1
,0
(y
1
, x
1
, . . . , x
s
1
)
(r
2
, u
2
) /
s
2
,0
(y
2
, x
s
1
+1
, . . . , x
s
1
+s
2
)
. . .
(r
k
, u
k
) /
s
k
,0
(y
k
, x
ds
k
+1
, . . . , x
d
).
A small neighborhood of the point r
1
#
0
r
2

d,0
is of the form
U

= U
0
U
1
. . . U
k
[0, )
where U
0

k
0
, U
i

s
i
,0
i
for i = 1, . . . , k, and
the interval [0, ) represents the gluing parameter. Fixing a metric on U
0
, . . . , U
i
determines a
metric topology on U

= U
1
U
2
[0, ) by
dist
U
(r
0
#
(r
1
, . . . , r
k
), r
0
#
(r
1
, . . . , r
k
)) := supdist
U
0
(r
0
, r
0
), . . . , dist
U
k
(r
k
, r
k
), [
[.
Taking the neighborhood U to be suciently small we can assume that all surfaces parametrized
by U
0
, . . . , U
k
are dieomorphic via dieomorphisms that are constant on the striplike ends.
Write o
r
0
#
(r
1
, . . . , r
k
) = o
r
0
o
r
1
. . .o
r
k
/ , where each o
r
i
is the truncation of the surface
o
r
i
along the prescribed striplike end at s = R() = log(), and is the identications of
the surfaces along the truncated ends.
Denition 4.7.2. Let > 0 be given. Dene a Gromov neighborhood U
B
d,0
of the tuple
(r
0
, u
0
), . . . , (r
k
, u
k
) as follows: (r, u) U
if
r = r
0
#
r
1
, . . . , r
k
U with dist
U
(r, r
0
#
0
r
1
, . . . , r
k
) < ,
[E(u
0
) +E(u
1
) +. . . +E(u
k
) E(u)[ < ,
dist
M
(u(z), u
i
(z)) < for all z o
r
i
, i = 0, . . . , k.
Type 3. Let (r
0
, u
0
) /
d,0
(x
0
, . . . , x
i1
, y, x
i+1
, . . . , x
d
) be a pseudoholomorphic quilted disk,
and let v

/
1
(y, x
i
) be a quilted Floer trajectory. We x a parametrization of the Floer
trajectory v : R [0, 1] M. As in previous sections we write
i
: [0, ) [0, 1] o
r
for the
i-th striplike end, and Z
i
for the image of this striplike end in o
r
.
Denition 4.7.3. Let > 0 be given, and dene R() = log(). Dene a Gromov neighbor-
hood U
B
d,0
of the pair (r
0
, u
0
), v as follows: (r, u) U
if
dist
R
d,0(r, r
0
) < ,
[E(u
0
) +E(v) E(u)[ < ,
for z o
R()
r
0
, dist
M
(u(z), u
0
(z)) < ,
there exists some 2R() such that dist
M
(u(
i
(s + , t), v(s, t)) < for (s, t)
[R(), R()] [0, 1].
26 SIKIMETI MAU
5. Gluing
Denition 5.0.4. A pseudoholomorphic quilt (r, u) is regular if the linearized operator D
S,r,u
is surjective. Similarly we say that a generalized Floer trajectory v is regular if the associated
linearized operator D
v
is surjective.
The strategy of proof for gluing is standard:
1: Dene a pre-glued curve, (r
R
, u
R
), for gluing lengths R >> 0.
2: Compute that |( )(r
R
, u
R
)|
0,p
(R) where (R) 0 as R .
3: Show that D
S,r
R
,u
R
is surjective, and construct a right inverse Q
R
by rst constructing
an approximate right inverse, T
R
, with the same image.
4: Show there is a uniform bound |Q
R
| C for suciently large R.
5: Show that for each R, the function T
S,r
R
,u
R
satises a quadratic estimate
(5.1) |dT
S,r
R
,u
R
(, ) D
S,r
R
,u
R
| c([[ +||
W
1,p)
with the constant c independent of R.
These steps are the content of Sections 5.1 through 5.5 respectively. In Section 5.6 these
ingredients are used to dene, with the help of an implicit function theorem, a gluing map asso-
ciated to each regular tuple, and we show that the image of the gluing map is contained in the
one-dimensional component of the relevant moduli space of quilted disks, /
d,0
(x
0
, . . . , x
d
)
1
. In
Section 5.7 we show that the gluing map is surjective, in the sense that if a pseudoholomorphic
quilted disk (r, u) /
d,0
(x
0
, . . . , x
d
)
1
is in a suciently small Gromov neighborhood of the
broken tuple, then it is in the image of the gluing map for that tuple.
5.1. Pregluing. There are three types of pre-gluing to consider - two types arise from the two
types of facet of codimension one in the boundary of
d,0
(corresponding to whether the inner
circle has bubbled through or not), and the other type arises from a Floer trajectory breaking
o.
Denition 5.1.1. Given a gluing parameter > 0, dene R() := log() to be the gluing
length corresponding to . So R() as 0.
Type 1. Assume that we have a regular pair
(r
1
, u
1
) /
d
1
,0
(x
0
, . . . , x
i1
, y, x
i+1
, . . . , . . . x
d
1
)
0
, where i 1, . . . , d,
(r
2
, u
2
) /
d
2
(y, z
1
, . . . , z
d
2
)
0
.
The surface parametrized by r
1
is a quilted disk, and the surface parametrized by r
2
is a
marked, unquilted disk. A marked point labeled
on r
1
is identied with a marked point
+
on r
2
, identifying the pair r
1
#r
2
with a nodal quilted disk. Along the strip-like ends labelled
by
, u
1
and u
2
converge exponentially to y.
r1
r2
s = 0
s = R s = R
s = 0
+
u1
u2
R
R/2
y(t)
y(t)
R/2
R/2 1
R/2 1
exp
y
(11)
exp
y
(22)
Figure 5.1. Case 1 of pregluing (left), and gluing u
1
and u
2
along the neck(right).
The quilted surface r
R
:= r
1
#
R
r
2
: Truncate the surface o
r
1
along the striplike end labeled by
at s = R, and truncate the surface o

r
2
along the striplike end labeled by
+
at s = R,
then identify the two truncated surfaces along s = R. Explicitly, one identies
+(R, t)
(R, 1 t); see Figure 5.1.

Let z = (s, t) denote the variables on the striplike end of o
r
1
, where s [0, ) and t [0, 1].
For s >> 0, u
1
(s, t) is exponentially close to y(t) 1(L, L
), and we dene
1
(s, t) T
y(t)
M by
exp
Q
y(t)
(
1
(s, t)) = u
1
(s, t)
where exp
Q
is the tuple of t-parametrized quadratically corrected exponential maps for the
pair L, L
, such that L is totally geodesic when t = 0, and L
is totally geodesic when t = 1.

Similarly, we can dene
2
(s, t) T
y(t)
M, for very large s, by
exp
Q
y(t)
(
2
(s, t)) = u
2
(s, t).
Let : R [0, 1] be a smooth cut-o function such that (s) = 1, s 1, and (s) = 0, s
0. Dene a pair of intermediate approximate pseudoholomorphic quilted surfaces (r
i
, u
R
i
) for
i = 1, 2, by
u
R
i
(z) =
_
_
_
u
1
(z), z o
r
i

(s R/2)
exp
Q
x
((s R/2)
i
(s, t)), R/2 1 s R/2
y(t), s R/2.
The pre-glued map u
1
#
R
u
2
: o
r
1
#
R
r
2
M is (Figure 5.1)
u
1
#
R
u
2
(z) =
_
u
R
1
(z), z o
r
1

((R, ) [0, 1])

u
R
2
(z), z o
r
2

+((R, ) [0, 1]).

28 SIKIMETI MAU
Type 2. Consider a collection
(r
0
, u
0
) /
k
(x
0
, y
1
, . . . , y
k
)
0
(r
1
, u
1
) /
d
1
,0
(y
1
, z
1
, z
2
, . . . , z
d
1
)
0
(r
2
, u
2
) /
d
2
,0
(y
2
, z
d
1
+1
, z
d
1
+2
, . . . , z
d
1
+d
2
)
0
. . .
(r
k
, u
k
) /
d
k
,0
(y
k
, z
d
1
+d
2
+...+d
k1
+1
, z
d
1
+d
2
+...+d
k1
+2
, . . . , z
d
1
+d
2
+...d
k1
+d
k
)
0
of regular pseudoholomorphic quilts. Let
(0)
,
(1)
, . . . ,
(k)
label the striplike ends of o
r
0
where
the quilt map u
0
converges to x
0
, y
1
, . . . , y
k
respectively. For i = 1, . . . , k, each quilt o
r
i
has a
distinguished strip-like end
(0)
i
on which u
i
: o
r
i
M converges to y
i
.
The quilted surface r
R
:= r
0
#
R
r
1
, . . . , r
k
: Strictly speaking a gluing parameter > 0 and
R = log() gives rise to a tuple of admissible gluing lengths, (R+a
1
, . . . , R+a
k
), where the
a
1
, . . . , a
k
are determined by the coordinates of r
0
, r
1
, . . . , r
k
in their respective moduli spaces.
However, for suciently large R we can shift the position of s = 0 on the striplike ends so
as to take the gluing lengths along all the striplike ends to be R. Truncate o
r
0
along each of
the striplike ends labeled
(1)
, . . . ,
(d)
at s = R, and truncate the surfaces o
r
1
, . . . , o
r
k
along
their respective striplike ends labeled
(0)
1
, . . . ,
(0)
k
at s = R, then identify truncations labeled
by pairs
(i)
,
(0)
i
along s = R (Figure 5.2).
The approximate pseudoholomorphic map. Let z = (s, t) be coordinates along the striplike
ends. For s >> 0, dene
i
(s, t) T
y
i
(t)
M by
exp
Q
y
i
(t)
(
i
(s, t)) = u
0
(
(i) (s, t)).

Similarly, for i = 1, . . . , k and s >> 0 dene
(0)
i
(s, t) T
y
i
(t)
M by
exp
Q
y
i
(t)
(
(0)
i
(s, t)) = u
i
(
(0)
i
(s, t)).
We introduce the intermediate approximate pseudoholomorphic quilted surfaces (r
0
, u
R
0
), (r
1
, u
R
1
), . . . , (r
k
, u
R
k
)
dened as follows. The map u
R
0
: o
r
0
M is dened piecewise by
u
R
0
(z) =
_
_
u
0
(z), z o
r
0

k
i=1
(i) (s R/2 1)
exp
Q
y
i
(t)
((s R/2)
i
(s, t)), z =
(i) (s, t), s [R/2 1, R/2]

y
i
(t), z =
(i) (s, t), s R/2.

For i = 1, . . . , k the intermediate maps u
R
i
: o
r
i
M are dened by
u
R
i
(z) =
_
_
u
i
(z), z o
r
i

(0)
i
(s R/2 1)
exp
Q
y
i
(t)
((s R/2)
i
(s, t)), z =
(0)
i
(s, t), s [R/2 1, R/2]
y
i
(t), z =
(0)
i
(s, t), s R/2.
r0
(2)
(1)
(0)
3
(0)
2
(0)
1
(3)
(4)
(0)
r1
r2 r3
(0)
4
r4
Figure 5.2. Pregluing in Case 2.
Then the pre-glued quilt map is
u
0
#
R
u
1
, . . . , u
k
(z) =
_
_
u
R
0
(z), z o
r
0

k
i=1
(i) (s R)
u
R
i
(z), z o
r
i

(0)
i
(s R).
Type 3. (A Floer trajectory breaks o.) Consider a pair
(r
1
, u
1
) /
d,0
(x
0
, x
1
, . . . , x
d
)
0
, v

/(x
i
, y)
0
of a regular pseudoholomorphic quilted disk and a regular Floer trajectory. Assume without
loss of generality that lim
s
v(s, t) = x
i
(t), lim
s
v(s, t) = y(t), and let label the
striplike end of o
r
1
for which lim
s
u
1
(
(s, t)) = x
i
(t). The Floer trajectory v is dened
only up to an R translation, so x a parametrization.
The quilted surface. In this case, r
R
= r
1
.
The approximate pseudoholomorphic map. For s >> 0 we know that u
1
(
(s, t)) and v(s, t)

are exponentially close to x
i
(t). For such s, dene (s, t), (s, t) T
x
i
(t)
M by the conditions
30 SIKIMETI MAU
that
exp
Q
x
i
(t)
((s, t)) = u
1
(
(s, t)), exp

Q
x
i
(t)
((s, t)) = v(s, t).
Dene an approximate pseudoholomorphic quilt and an approximate trajectory by
u
R
1
(z) =
_
_
_
u
1
(z), z o
r
1

(s R/2 1)
exp
Q
x
i
(t)
((s R/2)(s, t)), z =
(s, t), s [R/2 1, R/2]

x
i
(t), z =
(s, t), s R/2.

v
R
(s, t) =
_
_
_
v(s 2R, t), s 3R/2 + 1
exp
Q
x
i
(t)
((s + 3R/2)(s 2R, t)), s [3R/2, 3R/2 + 1]
x
i
(t), s 3R/2.
Then the preglued quilt map is
u
R
(z) =
_
u
R
1
(z), z o
r
1

(s R)
v
R
(s, t), z
(s R).
5.2. Estimates for the preglued quilts.
Proposition 5.2.1. For suciently large R
0
0, there is a monotone decreasing function
: [R
0
, ) [0, ) such that
|( )(r
R
, u
R
)|
0,p
(R)
and (R) 0 as R .
Proof. We estimate it for each of the three types of pregluing separately.
Type 1. Let (
J,

K) denote the approximate perturbation datum on the preglued quilt o
r
R
that
is inherited from the perturbation data (J
1
, K
1
) and (J
2
, K
2
) for o
r
1
and o
r
2
respectively via
the pregluing procedure. Let (J(r
R
), K(r
R
)) be the perturbation datum on o
r
R
that comes from
the universal choice of perturbation data over the family of quilted surfaces parametrized by the
multiplihedron. In general (
J,

K) and (J(r
R
), K(r
R
)) are not the same, but the assumption
of consistency implies that for large values of R, the data agree on the thin part of o
r
R
,
while on the complement of the thin part, which consists of two compact components coming
from o
r
1
and o
r
2
, (J(r
R
), K(r
R
)) converges uniformly to (J
1
, K
1
) or (J
2
, K
2
) respectively as
R .
Given that
(r
R
, u
R
) (r
R
, u
R
) =
1
2
(du
R
+J(r
R
) du
R
j(r
R
))
1
2
(Y (r
R
) +J(r
R
) Y (r
R
) j(r
R
))
we will estimate this on dierent parts of the preglued surface o
r
R
.
Consider a striplike end Z o
r
R
. It corresponds to a striplike end on either o
r
1
or o
r
2
,
that was not truncated in the pregluing step; let us denote this striplike end by Z too, where
Z o
r
i
for i = 1 or 2. Since j(r
R
, z), J(r
R
, u(z), z) and Y (r
R
, z) are independent of R and are
the same as the corresponding data on o
r
1
,
( )(r
R
, u
R
)
Z
= ( )(r
1
, u
1
)
Z
= 0
since we assumed that ( )(r
1
, u
1
) = 0. This argument proves that ( )(r
R
, u
R
) is zero
on all of the striplike ends of the glued surface o
r
R
.
Next, for i = 1, 2 let S
i
denote the complement of the striplike ends on o
r
i
, and let S
i
also
denote its image in o
r
R
after pregluing. Note that S
i
is compact, and that u
R
S
i
= u
i
S
i
.
Hence,
( )(r
R
, u
R
)
S
i
= ( )(r
R
, u
i
)
S
i
=
1
2
[du
i
+J(r
R
, u
i
)du
i
j(r
R
)]
S
i
1
2
[Y (r
R
, u
i
) +J(r
R
, u
i
)Y (r
R
, u
i
)j(r
R
)]
S
i
.
Since we know that ( )(r
i
, u
i
) = 0, and that j(r
R
), J(r
R
, u
i
) and Y (r
R
, u
i
) converge uni-
formly to j(r
i
), J(r
i
, u
i
) and Y (r
i
, u
i
) on S
i
, it follows that for suciently large R there is a
monotone decreasing function
1
(R) 0 as R such that the uniform pointwise estimate
[( )(r
R
, u
R
)(z)[
1
(R) holds for all z S
i
.
Finally, consider the neck of o
r
R
along which the pregluing took place. Let Z
1
and Z
2
denote the striplike ends of o
r
1
and o
r
2
that were truncated along s = R; and by slight
abuse of notation let Z
1
and Z
2
also denote the images of the truncations after pregluing. By
symmetry it is enough to consider what happens on Z
1
. There are three regions to consider,
0 s R/2 1, R/2 1 s R/2, R/2 s R.
On the region 0 s R/2 1, ( )(r
R
, u
R
) = ( )(r
1
, u
1
) = 0.
On the region R/2 s R, we have that u
R
(s, t) = x
0
(t), and so
dx
0
(t) +J(t, x
0
(t)dx
0
(t)j X
H
t
J(t, x
0
(t))X
H
t
j = 0
because
s
x
0
(t) = 0 and
t
x
0
(t) = X
H
t
together imply that dx
0
(t) X
H
t
= 0.
On the region R/2 1 s R/2, since u
1
(s, t) converges exponentially to y(t) as s
, we know that [(s, t)[ becomes exponentially small in s. Therefore, [(s)
1
(s, t)[ is also
exponentially small in s. Now, dy(t) X
H
t
(y(t)) = 0, so
[dexp
Q
y(t)
((s)
1
(s, t)) X
H
t
(exp
y(t)
((s)
1
(s, t))[ [d exp
Q
y(t)
((s)
1
(s, t)) dy(t)[
+[X
H
t
(y(t) X
H
t
(exp
Q
y(t)
((s)
1
(s, t)))[
and since exp
Q
y(t)
((s)
1
(s, t)) is exponentially close to y(t), there is a monotone decreasing
function
2
(R) 0 as R such that
[dexp
Q
y(t)
((s)
1
(s, t)) X
H
t
(exp
Q
y(t)
((s)
1
(s, t))[
2
(R)
uniformly in t for all s R/2 1.
32 SIKIMETI MAU
Using all these estimates, we have
_
_
_
_
Sr
R
[( )(r
R
, u
R
)[
p
dvol
Sr
R
_
_
_
1/p
_
_
_
S
1
[( )(r
R
, u
R
)[
p
dvol
S
1
_
_
1/p
+
_
_
_
S
2
[( )(r
R
, u
R
)[
p
dvol
S
2
_
_
1/p
+
_
_
_
1
_
0
R/2
_
R/21
[( )(r
R
, u
R
)[
p
ds dt
_
_
_
1/p

1
(R) vol(S
1
)
1/p
+
1
(R) vol(S
2
)
1/p
+
2
(R).
Here vol
S
i
is the volume of the compact subset S
i
o
r
i
with respect to a xed volume form
on o
r
i
.
Type 2. The estimate for Type 2 of the pregluing construction is very similar to that of Type
1. Repeating the same arguments on the corresponding parts of o
r
R
leads to an estimate
|T
S,r
R
,u
R
(0, 0)|
0,p,R

1
(R)
k
i=0
vol(S
i
)
1/p
+k
2
(R).
Type 3. This type of pregluing is slightly dierent from the previous two types. Let Z denote
the striplike end of o
r
1
on which the pregluing takes place. The perturbation data is xed, so
on the complement of Z, ( )(r
1
, u
R
) = ( )(r
1
, u
1
) = 0. Thus ( )(r
1
, u
R
) is only
supported on Z.
Write z = (s, t). For s [0, R/2 1], ( )(r
1
, u
R
) = ( )(r
1
, u
1
) = 0. For s
[R/2, 3R/2], ( )(r
1
, u
R
) = (
t
y(t) X
H
t
(y(t))
0,1
= 0 since y(t) is the Hamiltonian ow of
H
t
. For s [3R/2 + 1, ), ( )(r
1
, u
R
) = v(s 2R, t) X
H
t
(v(s 2R, t) = 0, as v is a
Floer trajectory.
Therefore ( )(r
1
, u
R
) is only supported on the region where s [R/2 1, R/2] or
s [3R/2, 3R/2 + 1]. On the rst part, for suciently large R, [(s, t)[ is exponentially small
in R when s R/2 1, and so for the same reasons as in the previous two calculations we can
nd a monotone decreasing
2
(R) 0 as R such that
[dexp
Q
y(t)
((s R/2)(s, t)) X
H
t
(exp
Q
y(t)
((s R/2)(s, t))[
2
(R)
uniformly in t, whenever s [R/2 1, R/2]. Similarly, for suciently large R, [(s 2R, t)[ is
exponentially small and we can nd an
2
(R) as above such that
[dexp
Q
y(t)
((s + 3R/2)(s 2R, t)) X
Ht
(exp
Q
y(t)
((s + 3R/2)(s 2R, t)[
3
(R))
uniformly in t, whenever s [3R/2, 3R/2 + 1]. Putting the estimates together gives
|T
S,r
1
,u
R
(0, 0)|
0,p,R

_
_
_
1
_
0
R/2
_
R/21
[( )(u
R
)[
p
ds dt
_
_
_
1/p
+
_
_
_
1
_
0
3R/2+1
_
3R/2
[( )(u
R
)[
p
ds dt
_
_
_
1/p

2
(R) +
3
(R).
5.3. Constructing a right inverse. We construct an approximate right inverse for the lin-
earized operator of the preglued surface and curve, then show that it is suciently close to
being a right inverse that an actual right inverse can be obtained from it via a convergent
power series. We treat the three types of pregluing separately.
Type 1. Assume that (r
1
, u
1
) and (r
2
, u
2
) are regular and isolated, for the perturbation data
(J
1
, K
1
) and (J
2
, K
2
) respectively. We will show that D
R
:= D
S,r
1
#
R
r
2
,u
1
#
R
u
2
, dened with
perturbation datum (J(r
R
), K(r
R
)), is also a surjective Fredholm operator, with a right inverse
Q
R
that is uniformly bounded for suciently large R.
Recall the intermediate quilt maps (r
1
, u
R
1
), (r
2
, u
R
2
). For large R, the functions u
R
i
, i = 1, 2
are W
1,p
-small perturbations of u
i
. The perturbation data (J(r
R
), K(r
R
)) are just the Floer
data along the neck of o
r
R
. Let (J
i
(r
R
), K
i
(r
R
)) denote the perturbation datum on o
r
i
given
by the restriction of (J(r
R
), K(r
R
)) to the truncation of o
r
i
, extended trivially over the rest of
the striplike end of o
r
i
by the Floer data. The assumption of consistency implies that for large
R, the data (J
i
(r
R
), K
i
(r
R
)) is a compact perturbation of (J
i
, K
i
).
The property of being a surjective Fredholm operator is stable under W
1,p
-small and com-
pact perturbations, hence for suciently large R the pairs (r
i
, u
R
i
) are regular with respect
to (J
i
(r
R
), K
i
(r
R
)). Let Q
i,R
denote a right inverse for the linearized operator D
S,r
i
,u
R
i
. We
observe that Q
i,R
is actually a left and right inverse, since by assumption ker D
S,r
i
,u
R
i
has
dimension 0, so D
S,r
i
,u
R
i
is an isomorphism.
Returning to the linearized operator
D
R
: T
r
R
d
1
+d
2
,0
W
1,p
(o
r
R
, u
R
TM) L
p
(o
r
R
,
0,1
J
u
R
TM)
we construct an approximate inverse
T
R
: L
p
(o
r
R
,
0,1
J
u
R
TM) T
r
R
d
1
+d
2
,0
W
1,p
(o
r
R
, u
R
TM).
Let L
p
(o
r
R
,
0,1
J
u
R
TM). Set
1
(z) =
_
(z), z r
R
1
,
0, z r
1
r
R
1
,
2
(z) =
_
.
(z), z r
R
2
,
0, z r
2
r
R
2
The abrupt cut-o does not matter because the norms involving
1
and
2
are just L
p
norms,
so
1
L
p
(o
r
1
,
0,1
J
(u
R
1
)
TM),
2
L
p
(o
r
2
,
0,1
J
(u
R
2
)
TM).
34 SIKIMETI MAU
Using Q
1,R
and Q
2,R
we get
Q
1,R
1
=: (
1
,
1
) T
r
1
d
1
+1,0
W
1,p
(o
r
1
, (u
R
1
)
TM)
Q
2,R
2
=: (
2
,
2
) T
r
2
d
2
+1
W
1,p
(o
r
2
, (u
R
2
)
TM).
Our nal step is to glue these into a single element of T
r
R
d
1
+d
2
,0
W
1,p
(o
r
R
, u
R
TM). For
large R, local gluing charts near the boundary of
d
1
+d
2
,1
give an isomorphism
T
r
R
d
1
+d
2
,0

= T
r
1
d
1
+1,0
T
r
2
d
2
+1
T
R
R
where the last component represents the gluing parameter. Using this isomorphism we get a
well-dened element
1
#
R
2
:= (
1
,
2
, 0) T
r
R
d
1
+d
2
,0
.
Now x a smooth cut-o function : R
0
[0, 1] such that (s) = 1 for s 1 and (s) = 0
for s 1/2, and 0

2. Let
R
(s) := (s/R).
1
#
R

2
(z) :=
_
1
(z) if z o
r
1
Z
1
1
(s, t) +
R
(s)
2
(2R s, 1 t) if z Z
1
, s [0, R]
R
(s)
1
(2R s, 1 t) +
2
(s, t) if z Z
2
, s [0, R]
2
(z) if z S
r
2
Z
2
.
To simplify notation, dene
1,R
: o
r
R
R by
1,R
(z) :=
_
_
1 if z o
r
1
Z
1
,
1 if z Z
1
, s [0, R],
R
(s), if z Z
2
, s [0, R],
0 if z o
r
2
Z
2
and make a corresponding denition of
2,R
. With this notation, we can write
1
#
R
2
=
1,R
1
+
2,R
2
. We dene
T
R
:= (
1
#
R
2
,
1
#
R
2
).
Now we need to check that for all (c
Sr
R
)
u
, for suciently large R,
|D
S,r
R
,u
R
T
R
|
0,p,R

1
2
||
0,p,R
.
Now,
D
S,r
R
,u
R
T
R
= D
S,r
R
,u
R
(
1
#
R
2
,
1
#
R
2
)
= D
(u
R
)
r
R

1
#
2
+D
r
R
(u
R
)
1
#
R
2
.
By construction D
(u
R
)
r
R
and D
(u
R
1
)
r
1
agree on the support of
1
, which is o
r
1
Z
1
. Similarly D
(u
R
2
)
r
2
agrees with D
(u
R
)
r
R
on the support of
2
which is o
r
2
Z
2
. Hence,
D
(u
R
)
r
R

1
#
R
2
= D
(u
R
1
)
r
1

1
+D
(u
R
2
)
r
2

2
.
Since D
(r
R
)
u
R
and D
(r
1
)
u
R
1
agree on the support of
1,R
1
, we may write
D
(r
R
)
u
R
1,R
1
= D
(r
1
)
u
R
1
1,R
1
= (
s
1,R
)
1
+
1,R
D
(r
1
)
u
R
1
1
,
D
(r
R
)
u
R
2,R
2
= D
(r
2
)
u
R
2
(
2,R
)
2
= (
s
2,R
)
2
+
2,R
D
r
2
u
R
2
2
.
By construction, D
(u
i
)
r
i

i
+D
(r
i
)
u
R
i
i
= D
(r
i
)
u
R
i
Q
i,R
i
=
i
, where Q
i,R
is the inverse of D
S,r
i
,u
R
i
. So,
D
S,r
R
,u
R
T
R
= D
(u
R
)
r
R

1
#
R
2
+D
(r
R
)
u
R
1
#
R
2

= D
(u
R
1
)
r
1

1
+D
(u
R
2
)
r
2

2
+D
(r
R
)
u
R
(
1,R
1
+
2,R
2
)
= D
(u
R
1
)
r
1

1
+D
(u
R
2
)
r
2

2
+D
(r
1
)
u
R
1
1,R
1
+D
(r
2
)
u
R
2
2,R
= D
(u
R
1
)
r
1

1
+D
(u
R
2
)
r
2

2
+ (
s
1,R
)
1
+
1,R
D
(r
1
)
u
R
1
1
+ (
s
2,R
)
1
+
2,R
D
(r
2
)
u
R
2
2

=
1,R
(D
(u
1
)
r
1

1
+D
r
1
u
R
1
1
) +
2,R
(D
(u
2
)
r
2

2
+D
(r
2
)
u
R
2
2
)
+(
s
1,R
)
1
+ (
s
2,R
)
2

=
1,R
1
+
2,R
2
+ (
s
1,R
)
1
+ (
s
2,R
)
2

=
1
+
2
+ (
s
1,R
)
1
+ (
s
2,R
)
2
= (
s
1,R
)
1
+ (
s
2,R
)
2
as the support of
i
is precisely where
i,R
= 1, and
1
+
2
= . We can nd a c > 0 and an
R
0
0 such that operator norms |Q
i,R
| c for all R R
0
. Therefore
|
i
|
1,p,R
|Q
i,R
i
|
R
c|
i
|
0,p,R
so we can estimate
|D
S,u
R
,r
R
T
R
|
0,p,R
= |(
s
1,R
)
1
+ (
s
2,R
)
2
|
0,p,R
2/R(|
1
|
0,p,R
+|
2
|
0,p,R
)
2c
R
||
0,p,R
and 2c/R 1/2 for suciently large R.
0
, u
0
), (r
1
, u
1
), . . . , (r
k
, u
k
) are all regular and isolated for the pertur-
bation data prescribed on their strata. For a gluing length R, form the preglued surface and
map, abbreviating
(r
R
, u
R
) := (r
0
#
R
(r
1
, . . . , r
k
), u
0
#
R
(u
1
, . . . , u
k
)).
36 SIKIMETI MAU
For i = 0, . . . , k, denote by (J
i
(r
R
), K
i
(r
R
)) the perturbation datum on o
r
i
given by the re-
striction of (J(r
R
), K(r
R
)) to the truncation of o
r
i
, extended trivially over the rest of the
striplike end of o
r
i
by the given Floer data. The assumption of consistency implies that for
large R, the data (J
i
(r
R
), K
i
(r
R
)) is a compact perturbation of (J
i
, K
i
). Also, by construc-
tion the intermediate maps u
R
i
are W
1,p
small perturbations of u
i
. The properties of being
Fredholm and surjective are stable under W
1,p
-small perturbations, hence for suciently large
R, (r
0
, u
R
0
), . . . , (r
k
, u
R
k
) are regular with respect to (J
i
(r
R
), K
i
(r
R
)). The linearized operators
D
S,r
i
,u
R
i
are surjective with zero dimensional kernel hence are isomorphisms, so let Q
i,R
be the
inverse of D
S,r
i
,u
R
i
. For convenience, for each i we will denote by o
R
r
i
the truncation of o
r
i
that
appears in the pre-glued surface o
r
R
, denote by Z
1
, . . . , Z
k
the striplike ends of o
r
1
, . . . , o
r
k
that
are truncated in the pre-gluing process, and denote by Z
1
, . . . , Z
k
the corresponding striplike
ends of the surface o
r
0
.
We construct an approximate right inverse T
R
for the linearized operator D
R
as follows. Let
L
p
(o
r
R
,
0,1
J
u
R
TM). Set
0
(z) =
_
(z), z o
R
r
0
0, else
, . . . ,
k
(z) =
_
(z), z o
R
r
k
,
0, else.
with
i
L
p
(o
r
i
,
0,1
J
(u
R
i
)
TM) for i = 0, . . . , k. Dene

(
0
,
0
) := Q
0,R
(
0
) T
r
0
k+1
W
1,p
(o
r
0
, (u
R
0
)
TM)
. . .
(
k
,
k
) := Q
k,R
(
k
) T
r
k
d
k
+1,0
W
1,p
(o
r
k
, (u
R
k
)
TM).
The nal step is to glue these k + 1 things together to get a single element of T
r
R
d
1
+d
2
,0
W
1,p
(o
r
R
, u
R
TM). For large R, r
R
is near the boundary of
d
1
+...+d
k
,0
, where local charts
identify
T
r
R
d
1
+...+d
k
,1

= T
r
0
k+1
T
r
1
d
1
+1,0
. . . T
r
k
d
k
+1,0
T
R
R,
the last component coming from the gluing parameter. With this identication set
0
#
R
(
1
, . . . ,
k
) := (
0
,
1
, . . . ,
k
, 0).
Fix a smooth cuto function : R
0
[0, 1] such that (s) = 1 for s 1 and (s) = 0 for
s 1/2, and 0

2. Let
R
(s) := (s/R). Dene
0,R
: o
r
R
[0, 1] by the condition that
0,R
(z) =
_
_
_
1, z o
R
r
0
,
R
(s), z Z
i
, s [0, R], i = 1, . . . , k,
0, else,
and for each i 1, . . . , k, let
i,R
(z) =
_
_
_
1, z o
R
r
i
,
R
(s), z Z
i
o
r
0
, s [0, R],
0, else,
then set
0
#
R
(
1
, . . . ,
k
) :=
k
i=0
i,R
i
. We dene the approximate inverse by T
R
:=
(
0
#
R
(
1
, . . . ,
k
),
0
#
R
(
1
, . . . ,
k
). The construction leads to the estimate
D
S,r
R
,u
R
T
R
= D
(u
R
)
r
R

0
#(
1
, . . . ,
k
) +D
(r
R
)
u
R
0
#(
1
, . . . ,
k
)
=
k
i=0
D
(u
R
i
)
r
i

i
+
k
i=0
D
(r
i
)
u
R
i
i,R
i

=
k
i=0
D
(u
R
i
)
r
i

i
+
k
i=0
i,R
D
(r
i
)
u
R
i
i
+
k
i=0
(
s
i,R
)
i
=
k
i=0
i,R
(D
(u
R
i
)
r
i

i
+D
(r
i
)
u
R
i
i
) +
k
i=0
(
s
i,R
)
i
=
k
i=0
i,R
i
+
k
i=0
(
s
i,R
)
i
=
k
i=0
(
s
i,R
)
i
.
Since (
i
,
i
) = Q
i,R
i
there is a c > 0 such that
|
i
|
0,p,R
|Q
i,R
i
|
1,p,R
c|
i
|
0,p,R
,
for all i. Combine everything into a total estimate
|D
S,u,r
T |
0,p,R

k
i=0
|
/R
i
|
0,p,R
|
s
i,R
|
i=0
|
i
|
0,p,R

2
R
k
i=0
c|
i
|
0,p,R
2c
R
||
0,p,R
and for suciently large R, 2c/R 1/2.
1
, u
1
) is a regular isolated pseudoholomorphic quilted disk, and v is a
regular isolated Floer trajectory, for their respective perturbation data. For large R, the inter-
mediate maps (r
1
, u
R
1
) and v
R
are W
1,p
-small perturbations of (r
1
, u
1
) and v respectively, and
since being Fredholm and surjective are properties that are stable under small perturbations,
it follows that (r
1
, u
R
1
) and v
R
are also regular for suciently large R. Let Q
1,R
be the inverse
of D
S,r
1
,u
R
1
(which is invertible), and let Q
v
R be a right inverse for D
v
R respectively, whose
image is the L
2
orthogonal complement of the kernel of D
v
R. Let denote the innite strip
R[0, 1], and label by Z the striplike end of o
r
1
on which the pregluing takes place. Construct
an approximate right inverse
T
R
: L
p
(o
r
1
,
0,1
J
u
R
TM) T
r
1
d+1,0
W
1,p
(o
r
1
, u
R
TM)
38 SIKIMETI MAU
of D
S,r
1
,u
R
as follows. Let L
p
(o
r
1
,
0,1
J
u
R
TM). Set
1
(z) = (z), z o
R
r
1
,
0, else,
and
v
= 1
1
. Then
1
L
p
((o
r
1
,
0,1
J
(u
R
1
)
TM), and
v
L
p
(,
0,1
J
(v
R
)
TM),
and we set
Q
1,R
1
:= (
1
,
1
) T
r
1
d+1,0
W
1,p
(o
r
1
, (u
R
1
)
TM)
Q
v
R
v
:=
2
W
1,p
(, (v
R
)
TM).
We need to glue
1
and
2
together to get an element of W
1,p
(o
r
1
, (u
1
#
R
v)
TM). Fix a smooth

cut-o function : R
0
[0, 1] such that (s) = 0 for s 1/2 and (s) = 1 for s 1, and
0

2. Let
R
(s) := (s/R). Dene
1,R
: o
r
1
[0, 1] by
1,R
(z) = 1, z o
r
1
Z,
1
R
(s R/2), z = (s, t) Z,
and
2,R
(z) = 0, z o
r
1
Z,
R
(s), z = (s, t) Z.
Putting
1
#
R
2
:=
1,R
1
+
2,R
2
we dene T
R
:= (
1
,
1
#
R
2
). Now,
D
S,r
1
,u
R
T
R
= D
S,r
1
,u
R
(
1
,
1
#
R
2
)
= D
(u
R
1
)
r
1

1
+D
(r
1
)
u
R
1
1,R
1
+D
v
R
2,R
2

= D
(u
R
1
)
r
1

1
+ (
s
1,R
)
1
+
1,R
D
(r
1
)
u
R
1
1
+ (
s
2,R
)
2
+
2,R
D
v
R
2

=
1,R
(D
(u
R
1
)
r
1

1
+D
(r
1
)
u
R
1
1
) +
2,R
D
v
R
2
+ (
s
1,R
)
1
+ (
s
2,R
)
2
= (
s
1,R
)
1
+ (
s
2,R
)
2
,
and choosing a c > 0, R
0
0 such that the operator norms |Q
1,R
| c, |Q
v
R| c for all
R R
0
, we have an estimate
|D
S,r
1
,u
R
T
R
|
0,p,R

2
R
(|
1
|
0,p,R
+|
2
|
0,p,R
)
2
R
(c|
1
|
0,p,R
+c|
2
|
0,p,R
) =
2c
R
||
0,p,R
so 2c/R 1/2 for suciently large R. We obtain an actual right inverse Q
R
= T
R
(D
R
T
R
)
1
using a power series
(D
R
T
R
)
1
= (Id + (D
R
T
R
Id))
1
=
k=0
(1)
k
(D
R
T
R
Id)
k
,
which is convergent for large R because of the estimate |D
R
T
R
Id| 1/2.
5.4. Uniform bound on the right inverse. By construction, the image of Q
R
= T
R
(D
R
T
R
)
1
is the same as the image of T
R
. The operator norm of (D
R
T
R
)
1
can be uniformly estimated,
by the results of the previous section. Thus it remains to nd a uniform bound for |T
R
| in
order to get a uniform bound |Q
R
|.
Type 1. T
R
is a composition of operations. The initial cut-os dene a map
2
: L
p
(o
r
R
, u
R
TM) L
p
(o
r
1
, (u
R
1
)
TM) L
p
(o
r
2
, (u
R
2
)
TM)
where the norm on the product L
p
(o
r
1
, (u
R
1
)
TM)L
p
(o
r
2
, (u
R
2
)
TM) is the sum of the norms.

So by construction, the operator norm |
1
2
| = 1. The next step is the map Q
1,R
Q
2,R
whose domain is the product L
p
(o
r
1
, (u
R
1
)
TM) L
p
(o
r
2
, (u
R
2
)
TM) and range is the product

T
r
1
W
1,p
(o
r
1
, (u
R
1
)
TM) T
r
2
W
1,p
(o
r
2
, (u
R
2
)
TM). The operator norm is estimated

|Q
1,R
Q
2,R
| |Q
1,R
| + |Q
2,R
| 2c where c is a uniform bound on the operator norms
of Q
1,R
, Q
2,R
for large R. (Such a uniform bound c exists because for large R, Q
1,R
and Q
2,R
converge to Q
1
and Q
2
, the respective inverses of D
S,r
1
,u
1
and D
S,r
2
,u
2
.) The nal step in the
construction of T
R
uses an operator
1

2
: W
1,p
(o
r
1
, (u
R
2
)
TM) W
1,p
(o
r
2
, (u
R
2
)
TM) W
1,p
(o
r
R
, (u
R
)
TM)
dened using the cut-o functions
1,R
and
2,R
, which by construction satises
|(
1

2
)(
1
,
2
)|
1,p,R
|
1
|
1,p,R
+|
2
|
1,p,R
where the right hand side is the norm of (
1
,
2
) on the product W
1,p
(o
r
1
, (u
R
2
)
TM)
W
1,p
(o
r
2
, (u
R
2
)
TM). Hence |
1
2
| = 1. Putting everything together we get that |T
R
| 2c.
Type 2. By almost identical arguments as above we can conclude that |T
R
| (k+1)c, where
c is a uniform bound for large R on the operator norms |Q
i,R
|, for i = 0, 1, . . . , k.
Type 3. By almost identical arguments as above we can conclude that |T
R
| 2c, where c is
a uniform bound for large R on the operator norms |Q
1,R
| and |Q
v,R
|.
5.5. Quadratic estimate. The goal of this section is to establish a quadratic estimate,
|dT
S,r,u
(, ) D
S,r,u
| c(||
W
1,p +[[)
where the norm on the left is the operator norm. The W
1,p
norm on the right will depend
on a choice of volume form on S, and in the proof of the gluing theorem it will be a dierent
volume form for dierent values of the gluing parameter R. Note, however, that the denition
of the operator T
S,r,u
doesnt depend in any way on the volume form on S. The crucial thing
to ensure is that the constant c in the estimate can be chosen to be independent of the gluing
parameter R.
We assume that we are working in a local trivialization of the ber bundle o , in a
neighborhood U of r. We will write S := o
r
. Recall that in this neighborhood of r, all the
bers o
r
are dieomorphic to S, and so the varying almost complex structures can be pulled
back to S, as can the perturbation data.
40 SIKIMETI MAU
Denition 5.5.1. For a xed volume form on S, let
c
p
(dvol
S
) := sup
0=fC
(S)W
1,p
(S)
|f|
L
|f|
W
1,p
.
Lemma 5.5.2. For each d 1, there is a constant c
0
such that c
p
c
0
for all the quilted
surfaces o
r
in the bundle o
d
.
Proof. If one xes volume forms for each patch in the quilts in a given family o
d
or o
d,0
, then
such a constant exists for each patch by the embedding statements of Appendix A, which can
be uniformly bounded over a given family o
d,0
or o
d
.
Proposition 5.5.3 (c.f. Proposition 3.5.3 in [5]). Let p > 2, r and let S := o
r
be its
quilted surface with strip-like ends. Let S
thick(thin)
denote the thick (resp. thin) part of a thick-
thin decomposition. Then, for every constant c
0
> 0, there exists a constant c > 0 such that the
following holds for every volume form dvol
S
on S such that c
p
(dvol
S
) c
0
and vol(S
thick
) c
p
0
.
If u W
1,p
(S, M), W
1,p
(S, u
TM), r , and T
r
satisfy
(5.2) |du|
L
p c
0
, ||
L
c
0
, [[ c
0
,
then
(5.3) |dT
S,r,u
(, ) D
S,r,u
| c(||
W
1,p +[[).
Here | | denotes the operator norm on the space of bounded linear operators from T
r

W
1,p
(S, u
TM) to L
p
(S,
0,1
J
u
TM).
Proof. For each component manifold M of M, a point x M and a vector T
x
M determine
a linear map
E
x
() : T
x
M T
exp
x

M,

d
d
=0
exp
x
( +
)
and a bilinear map
x
() : T
x
M T
x
M T
exp
x

M, (
, )

x
( +
=0
Similarly, for each pair of manifolds (M
1
, M
2
) in M labeling the patches on either side of a
seam (we include true boundary components by considering them as a seam with the other side
labeled by pt), a point x := (x
1
, x
2
) M
1
M
w
=: M together with a vector := (
1
,
2
)
T
x
M determine a linear map
E
Q
x
() : T
x
M T
exp
Q
x
()
M,

d
d
=0
exp
Q
x
( +
)
and a bilinear map
Q
x
() : T
x
M T
x
M T
exp
x

M, (
, )

Q
x
( +
=0
We use the notation
r
() to denote the inverse of the projection [ ]
0,1
(r)
. That is,
r
()
1
() =
[]
0,1
(r)
=
r
=
1
2
( +J(r) j(r)). We therefore have, for each r , and T
r
, a
linear map
E
r
() : T
r
T
r
,
d
d
=0
exp
r
( + )
and a family of bilinear maps (parametrized by the quilted surface) given by
r,z
() : T
r
0,1
(r)
(T
z
S)
J
T
x
M
0,1
(r)
(T
z
S)
J
T
x
M
( , )
d
d
=0
r

r
( + )( ).
All the manifolds are compact, and by construction the almost complex structures J(r, z)
(which determine the local metrics and exponential maps) vary only over a compact subset of
the bundle o, and the quadratic corrections Q near each seam/boundary component only vary
over a compact interval. Therefore, there is a constant c
1
such that for all [[ c
0
and [[ c
0
,
[E
x
()
[ c
1
[
[, [
x
()(
, )[ c
1
[[[
[[[, [E
Q
x
()
[ c
1
[
[, [
Q
x
()(
, ) c
1
[[[[
[[[,
[E
r
() [ c
1
[ [, [
r,z
(; , )[ c
1
[[[ [[ [
for all norms determined by the metrics g
J(r,z)
. The inequalities for the bilinear forms are
possible because both are 0 when = 0 and = 0.
From here on in the calculations, we will simplify notation and write u instead of u, and
instead of ; but it is important to remember that they are dened on quilts. Starting from
the identity
S,r,u
( + , +
)T
S,r,u
( + , +
) =
J
(exp
r
( + ), exp
u
( + ))
we covariantly dierentiate both sides respect to at = 0 within our trivialization. For
typesetting reasons, abbreviate T := T
S,r,u
, u
:= exp
u
(), r
:= exp
r
().
(5.4)

=0
S,r,u
( + , +
)T( + , +
) = D
S,r,u
(E
r
() , E
u
()
).
Expanding the left hand side of (5.4) yields
[
r
(; ,
u
()T(, )]
0,1
r
+ [
u
(;
,
r
()T(, ))]
0,1
r
+
S,r,u
(, )dT(, )( ,

)].
Therefore we have
dT
S,r,u
(, )( ,

) D
S,r,u
( ,

)
S,r,u
(, )
1
_
u
(;

,
r
()T(, ))
_
0,1
r
. .
A
=
S,r,u
(, )
1
[
r
(; ,
u
()T(, ))]
0,1
r
. .
B
+
S,r,u
(, )
1
D
S,r,u
(E
r
() , E
x
()
) D
S,r,u
( ,

)
. .
C
.
We estimate A, B and C in turn.
Estimating A. By denition
S,r,u
(, )
1
=
r

u
()
1
. By construction, the projection
r
satises [
r
[ [[ when using the induced norm at r, and
u
() is an isometry. Under
42 SIKIMETI MAU
the assumption that [[ c
0
, we can nd a constant c
2
(which depends on c
0
) such that the
operator norms |
r
()| c
2
and |
r
| c
2
. So we get a pointwise estimate
[A[ [
u
(;

,
r
()T(, ))[ c
1
c
2
[[[
[[
r
()T(, )[.
Moreover
T(, ) =
S,r,u
(, )
1
[du
Y (r
, u
))]
0,1
r
,
= [T(, )[ c
2
[du
Y (r
, u
)[ c
2
[du
[ +c
2
[Y (r
, u
)[.
We can nd a constant c
3
, which depends on the xed choice of c
0
, and the xed choice of
almost-complex structures J(r, z) in the perturbation datum, and the xed choice of quadratic
corrections Q for the seams (but given xed choices, this constant can be made independent
of the parameters (r, z)), such that
[d exp
Q
u
[ c
3
([du[ +[[ +[[)
whenever |du|
L
p c
0
and ||
L
c
0
. There is a further constant, call it c
3
as well, that
gives a uniform bound [Y (r, z, x)[ c
3
for all (r, z) o, x M, and all metrics [ [ on TM
induced by J(r, z). Taking L
p
norms of these estimates, we get
[A[ c
1
c
2
2
c
3
[[[
[([du[ +[[ +[[ + 1)

= |A|
p
c
1
c
2
2
c
3
_
||
|du|
p
+||
||
p
+||
||
p
+||
|
p
_
c
1
c
2
2
c
3
_
c
3
0
||
1,p
|
|
1,p
+c
2
0
||
1,p
|
|
1,p
+c
2
0
|
|
1,p
||
1,p
+c
0
||
1,p
|
|
1,p
_
c
A
(||
1,p
+[[)(|
|
1,p
+[ [),
where c
A
= c
A
(c
0
, c
1
, c
2
, c
3
).
Estimating B. First we have
[B[ [
r
(; ,
u
()T(, ))[ [
r
(; ,
u
()
T(, ))[ +[
r
(; ,
u
()T(, ))[
where the notation T =

T+T (from section 4.5) just separates the piece of the inhomogeneous
pseudoholomorphic equation from the perturbation piece. For the rst term,
[
r
(; ,
u
()
T(, ))[ c
1
[[[ [du
[ c
1
c
3
[[[ [([du[ +[[ +[[)
= |
r
(; ,
u
()
T(, ))|
p
c
1
c
2
[[[ [(|du|
p
+||
p
+||
p
)
a
B
(||
1,p
+[[)(|
|
1,p
+[ [),
where a
B
= a
B
(c
0
, c
1
, c
3
). For the other part of B, observe that
r
(; ,
u
()T(, )) = 0 on
striplike ends and the images of glued striplike ends, i.e., the thin part of the quilt. To see that
it vanishes on the striplike ends, note that j and J are independent of r , so
r
() is the
identity. On the regions of S in the images of the glued striplike ends, i.e. the thin parts, J
is independent of r , but j = j(r) does change with r. However, it is independent of r in
the t direction, so the bilinear form vanishes on T(, )(
t
) here. The bilinear form does not
automatically vanish in the
s
direction, however, the form T itself vanishes precisely in the s
direction because the Hamiltonian perturbations are always independent of s on these parts,
so Y (r, z)(
s
) = 0 here. Thus,
r
(; ,
u
()T(, )) is only supported on the thick part of S.
With this in mind, we have that
|
r
(; ,
u
()T(, ))|
p
c
1
[[[ [|Y (r, z)|
L
p
(S
thick
)
c
1
c
3
[[[ [(vol(S
thick
))
1/p
c
0
c
1
c
3
[[[ [
b
B
(||
1,p
+[[)(|
|
1,p
+[ [),
where b
B
= b
B
(c
0
, c
1
, c
3
), and so taking c
B
= a
B
+b
B
we have that
|B|
p
c
B
(||
1,p
+[[)(|
|
1,p
+[ [).
Estimating C. Writing D
S,r,u
= D
(r)
u
+D
(u
)
r
as in section 4.5, we have
S,r,u
(, )
1
D
S,r,u
(E
r,u
(, )( ,

)) D
S,r,u
( ,

)
= [
S,r,u
(, )
1
D
(r)
u
E
u
()(
) D
(r)
u
(
)]
. .
C(a)
+[
S,r,u
(, )
1
D
(u
)
r
E
r
()( ) D
(u)
r
( )]
. .
C(b)
.
Estimating C(a). By denition
S,r,u
(, )
1
=
u
()
1

r
, and
u
() is an isometry, so
u
()C(a) =
r
D
r
u
(E
u
()(
))
u
()D
(r)
u
(
) =
_
D
r
u
(E
u
()(
))
u
()D
(r)
u
(
)
_
0,1
(r)
= [C(a)[ [D
(r)
u
(E
u
()(
))
u
()D
(r)
u
(
)[
[
D
(r)
u
(E
u
()(
))
u
()
D
(r)
u
(
)[
. .
C(a)(i)
+[P
(r)
u
(E
u
()(
))
u
()P
(r)
u
(
)[
. .
C(a)(ii)
.
Estimating C(a)(i). By formula (4.6) for

D
(r)
u
we have
[
D
(r)
u
(E
u
()(
))
u
()
D
(r)
u
(
)[ [[(E
u
()(
))]
0,1
r

u
()[
]
0,1
r
[
+
1
2
[J(r
)(
Eu()(
)
J)(r
, u
)
J(r)
u

u
()J(r)(
J)(r, u)
J(r)
u[. (5.5)
For [[ c
0
, we can nd a constant c
4
such that at every (r, z) o, the operator norm
|
r

r
| c
4
[[ uniformly for all (r, z) o and x M. Hence the rst term in (5.5) can be
estimated by
[[(E
u
()(
))]
0,1
r

u
()[
]
0,1
r
[ [(E
u
()(
))
u
()
[ +[(
r

r
)
u
()
[
[(E
u
()(
)) E
u
()
[ +[(E
u
()
u
())
[ +c
4
[[[
[
[(E
u
())
[ +[(E
u
()
u
())
[ +c
4
[[[
[.
Since E
x
(0)
x
(0) = 0, there is a uniform constant, again call it c
4
, such that |E
x
()
x
()| c
4
for all [[ c
0
. There is also a uniform constant, again call it c
4
, such that
44 SIKIMETI MAU
pointwise, the operator norm |E
u
()| c
4
[[([du[ +[[) whenever [(z, u(z))[ c
0
. Hence,
|[(E
u
()(
))]
0,1
r

u
()[
]
0,1
r
|
p
c
4
_
||
(|du|
p
+||
p
)|
+||
|
p
+[[|
|
p
_
c
4
_
c
3
0
||
1,p
|
|
1,p
+c
2
0
||
p
||
|
1,p
+c
0
||
1,p
|
|
p
+[[|
|
p
_
C(||
1,p
+[[)(|
|
1,p
+[ [).
where C = C(c
0
, c
4
) is a uniform constant. Up to the factor of
1
2
, the second term in (5.5) can
be expanded to
J(r
)(
Eu()(
)
J)(r
, u
)
J(r)
u

u
()J(r)(
J)(r, u)
J(r)
u
= (J(r
) J(r))(
Eu()(
)
J)(r
, u
)
J(r)
u
+ J(r)((
Eu()(
)
J)(r
, u
) (
Eu()(
)
J)(r, u
))
J(r)
u
+ J(r)(
Eu()(
)
J)(r, u
))(
J(r)
u

J(r)
u
)
+ J(r)(
Eu()(
)
J)(r, u
)[
J(r)
u

u
()
J(r)
u]
+ J(r)[(
Eu()(
)
J)(r, u
)
u
()(
J)(r, u)]
u
()
J(r)
u
There is constant c
5
so that |
J| c
5
[[ for all (r, z) o, x M, T
x
M. Furthermore
we can choose c
5
so that for all (r, z) o, T
r
satisfying [[ c
0
, x M and T
x
M
such that [[ c
0
, we have
|J(r
) J(r)| c
5
[[, |J(r
)| c
5
, [j(r
)[ c
5
, [(
J)(r
, x) (
J)(r, x)| c
5
[[[[,
[
J(r)
u

u
()
J(r)
u[ c
5
[[, [(
Eu()(
)
J)(r, u
)
u
()(
J)(r, u)[ c
5
[[[
[,
and for all T
z
S T
x
M, we have [[]
1,0
r
[]
1,0
r
[ c
5
[[[[, where [ ]
1,0
r
denotes projection
onto the complex linear part with respect to J(r) and j(r). Thus the above expression can be
estimated by
c
1
c
3
c
4
5
[[[
[([du[ +[[) +c
1
c
3
c
3
5
[
[[[([du[ +[[) +c
1
c
3
c
2
5
[
[[[([du[ +[[)
+c
1
c
2
5
[
[[[ +c
5
[[[
[[du[,
and it follows readily that there is a constant C = C(c
0
, c
1
, c
3
, c
5
) such that the L
p
norm of
this sum is estimated by C([[ +||
1,p
)([ [ +|
|
1,p
). Combining this with the estimate for the
rst term in (5.5) we get that there is a constant C = C(c
0
, c
1
, c
3
, c
4
, c
5
) such that
(5.6) |C(a)(i)|
p
C([[ +||
1,p
)([ [ +|
|
1,p
).
Estimating C(a)(ii). By formula (4.8) for P
r
u
, we have
C(a)(ii) = [
Eu()(
)
Y (r
, u
)]
0,1
r

u
()[
Y (r, u)]
0,1
r
1
2
[J(r
, u
) (
Eu()(
)
J)(r
, u
) Y (r
, u
)]
0,1
r
+
1
2
u
()[J(r, u) (
J)(r, u) Y (r, u)]

0,1
r
= (
r

r
)
Eu()(
)
Y (r
, u
) + [
Eu()(
)
Y
u
()
Y ]
0,1
r
1
2
_
r

r
_
(J (
Eu()(
)
J) Y )(r
, u
1
2
[
_
J
Eu()(
)
J Y
_
(r
, u
)
_
J
Eu()(
)
J Y
_
(r, u
)]
0,1
r
1
2
[
_
J
Eu()(
)
J Y
_
(r, u
)
u
()
_
J
J Y
_
(r, u)]
0,1
r
There is a constant c
6
such that for all (r, z) o, T
r
satisfying [[ c
0
, x M, T
x
M
such that [[ c
0
, and for all T
x
M, we have
[
Y [ c
6
[[,
[ (J
J Y ) (r
, x) (J
J Y ) (r, x)[ c
6
[[[[,
[
Eu()()
Y
u
()
Y [ c
6
[[[[,
[
_
Ex()()
J Y
_
(r, exp
Q
x
())
x
() (
J) Y ) (r, x)[ c
6
[[[[.
Thus we get a pointwise estimate
[C(a)(ii)[ c
1
c
4
c
6
[[[
[ +c
6
[[[
[ +
1
2
c
1
c
3
c
4
c
2
5
[[[
[ +c
0
c
6
[[[
[ +c
6
[[[
[
= |C(a)(ii)|
p
C([[ +||
1,p
)([ [ +|
|
1,p
)
where C = C(c
0
, c
1
, c
3
, c
4
, c
5
, c
6
). This completes the quadratic estimate for C(a).
Estimating C(b). As in Section 4.5, we write D
(u)
r
=

D
(u)
r
+P
(u)
r
,
C(b) =
u
()
1
D
(u
)
r
E
r
()( )

D
(u)
r

. .
C(b)(i)
u
()
1
r
P
(u
)
r
E
r
()( ) +P
(u)
r

. .
C(b)(ii)
.
Estimating C(b)(i). Using the explicit formula (4.5), we get
2
u
()C(b)(i) =
r

r
(
Er()( )
J du
j)(r
, u
)
r
u
() (

J du j)(r, u)
. .
+
r
r
(J du

Er()( )
j)(r
, u
)
r
u
()(J du

j)(r, u).
. .
We can nd a constant c
7
such that for all (r, z) o, all T
r
such that [[ c
0
, all T
r
,
and all x M and T
x
M with [[ c
0
, we have |
J| c
7
[[, |
j| c
7
[[, |
Y | c
7
[[,
|
Er()()
J(r
, x)
J(r, x)| c
7
[[[[, [
x
()(
J)(r, x) (
J)(r, exp
x
())
x
()[ c
7
[[[[,
46 SIKIMETI MAU
[du

u
()du[ c
7
[[, and |j(r
) j(r)| c
7
[[. Expanding we get
=
r
(
r

r
)(
Er()( )
J du
j)(r
, u
)
+
r
_
Er()( )
J(r
, u
)

J(r, u
)
_
du
j(r
)
+
r
(

J)(r, u
) du
(j(r
) j(r))
+
r
(

J)(r, u
) (du

u
()du) j(r)
+
r
(

J(r, u
)
u
()
u
()

J(r, u)) du j(r)
= [[ c
1
c
3
c
4
c
5
c
7
[[[ [([du
[) +c
3
c
5
c
7
[[[ [([du
[)
+c
3
c
2
7
[ [([du
[)[[ + 2c
7
[ [[[
= ||
p
C([[ +||
1,p
)([ [ +|
|
1,p
)
where C = C(c
0
, . . . , c
7
). Now expanding we get
=
r
(
r

r
)J(r
, u
)du
Er()( )
j(r
)
+
r
(J(r
, u
) J(r, u
))du

Er()( )
j)(r
)
+
r
J(r, u
)du
(
Er()( )
j(r
)

j)(r)
+
r
J(r, u
)(du

u
()du)

j(r)
= [[ c
1
c
3
c
4
c
5
c
7
[[([du
[)[ [ +c
1
c
2
7
[[([du
[)[ [ +c
3
c
7
([du
[)[[[ [ +c
2
7
[[[ [
= ||
p
C([[ +||
1,p
)([ [ +|
|
1,p
)
where C = C(c
0
, . . . , c
7
). Estimating C(b)(ii). By the explicit formula (4.7), we can write
u
()C(b)(ii) =
r
r
(
Er()( )
Y )(r
, u
)
u
()
r
(

Y )(r, u)
. .
a
+
r
r
(
Er()( )
JY j)(r
, u
)
u
()
r
(

JY j)(r, u)
. .
b
+
r
r
(JY
Er()( )
j)(r
, u
)
u
()
r
(JY
)(r, u)
. .
c
.
There is a constant c
8
such that for all (r, z) o, all T
r
such that [[ c
0
, all
T
r
, and all x M and T
x
M with [[ c
0
, we have [
Er()()
Y (r
, x)
(r, x)[
c
8
[[[[, [
Y (r, exp
x
)
x
()
Y (r, x)[ c
7
[[[[, [
Y (r, x)[ c
8
[[. Note also that Y is
independent of r on the thin parts of the surfaces in o (i.e., striplike ends and images of
striplike ends under gluing), so in particular,
Y is only supported on the thick part of each

quilt o
r
. Therefore, expanding a we have
a =
r
(
r

r
)
Er()( )
Y (r
, u
)
+
r
_
Er()( )
Y (r
, u
)

Y (r, u
)
_
+
r
(

Y (r, u
)
u
()

Y (r, u))
= |a|
p
c
1
c
4
c
8
[[[ (vol(S
thick
))
1/p
+c
8
[[[ [(vol(S
thick
))
1/p
+c
8
[ [||
p
C([[ +||
1,p
)([ [ +|
|
1,p
)
where C = C(c
0
, . . . , c
8
). The estimates for b and c are very similar, so we omit them. Now
combining all estimates proves the desired estimate (5.9).
5.6. The gluing map. To dene the gluing map, we use an innite dimensional implicit
function theorem, quoted here from [5, Appendix A].
Theorem 5.6.1 (Theorem A.3.4 in [5]). Let X and Y be Banach spaces, U X an open
subset of X, and f : U Y a continuously dierentiable map. Suppose that for x
0
U,
D := df(x
0
) : X Y is surjective, and has a bounded right inverse Q : Y X, and that
> 0, C > 0 are constants such that |Q| C, B
(x
0
) U, and
|x x
0
| = |df(x) D|
1
2C
.
Suppose that x
1
X satises |x
1
x
0
|

8
, and |f(x
1
)|

4C
. Then there exists a unique
x X such that
f(x) = 0, x x
1
imQ, and |x x
0
| .
Moreover, |x x
1
| 2C|f(x
1
)|.
We apply it to our situation as follows. For each R, take a local trivialization of a small
neighborhood of the preglued curve (r
R
, u
R
). Identify
X := T
r
R
W
1,p
(o
r
R
, u
R
TM), Y := L
p
(o
r
R
,
0,1
J
R
u
R
TM),
f := T
S,r
R
,u
R
, x
0
:= (0, 0).
For 1/(2Cc), the estimate (5.1) says that whenever ([[ +||
1,p
) then
|dT
S,r
R
,u
R
(, ) D
S,r
R
,u
R
| c([[ +||
W
1,p) c 1/(2C)
For suciently large R the constants C and c are independent of R, thus can also be chosen
to be independent of R. So now let x
1
:= (0, 0). Then
|T
S,r
R
,u
R
(0, 0)|
0,p
= |( )(r
R
, u
R
)|
0,p
c (R)

4C
for suciently large R. Thus by the Implicit Function Theorem there exists a unique (
R
,
R
)
T
r
R
W
1,p
(o
r
R
, u
R
TM) such that
T
S,r
R
,u
R
(
R
,
R
) = 0, (
R
,
R
) imQ
R
, [
R
[ +|
R
|
W
1,p .
Now T
S,r
R
,u
R
(
R
,
R
) = 0 if and only if (exp
r
R

R
, exp
u
R
R
) /
d,1
(x
0
, . . . , x
d
). Thus we can
dene a gluing map
g : [R
0
, ) /
d,0
(x
0
, . . . , x
d
), R (exp
r
R

R
, exp
u
R
R
)
The implicit function theorem also implies
[
R
[ +|
R
|
W
1,p 2C|T
S,r
R
,u
R
(0, 0)|
0,p
2C(R) 0
as R . In particular, the glued curves g(R) converge to the preglued curves (r
R
, u
R
),
hence as R they Gromov converge to the same limiting broken tuple.
For a gluing length R >> 0, write (r
R
, u
R
) for the preglued curve, and ( r
R
, u
R
) for the
corresponding glued curve.
In a local trivialization about the preglued curve (r
R
, u
R
), the implicit function theorem
implies that the moduli space of pseudoholomorphic quilts in a neighborhood of (r
R
, u
R
) is
modeled on a complement of im Q
R
. In particular, to show that the image of the gluing map
48 SIKIMETI MAU
is contained in the 1-dimensional component of the moduli space of pseudoholomorphic quilted
disks, its enough to check that im Q
R
has codimension 1. Since ker D
R
is a complement to
imQ
R
, an equivalent statement is that dim ker D
R
= 1. By construction, the right inverse
Q
R
has the same image as the approximate right inverse T
R
.
Proposition 5.6.2. The images of the gluing maps are contained in the one-dimensional
component of the moduli space, /
d,0
(x
0
, . . . , x
d
)
1
.
The proposition will follow from:
Lemma 5.6.3. For each type of gluing construction, with approximate right inverse T
R
,
codim im T
R
= dim ker D
R
= 1.
Proof. We prove Lemma 5.6.3 for each type of gluing construction in turn. Since the construc-
tions behave somewhat dierently, the line of proof will be as follows: for Types 1 and 2, we will
prove that codim im T
R
= 1 by nding an explicit description of im T
R
and a one-dimensional
complement. For Type 3, we will prove the existence of an isomorphism ker D
v

= ker D
S,r
1
,u
R
;
then, since dim ker D
v
= 1, it will follow that dim ker D
S,r
1
,u
R
= 1 too.
Type 1. Suppose that (, ) T
r
R
d,0
W
1,p
(o
r
R
, (u
R
)
TM). Let
1
+
1,2
+
2
= 1 be a
smooth partition of unity on o
r
R
such that the support of
1
is on the part of o
r
R
that comes
from the truncation of o
r
1
on the neck at s = R, the support of
2
is on the part of o
r
R
that
comes from the truncation of o
r
2
on the neck at s = R, and the support of
1,2
is on the subset
of the neck of o
r
R
corresponding to R/2 s R in o
r
1
as well as the corresponding piece
from o
r
2
. For W
1,p
(o
r
R
, (u
R
)
TM), it is clear that
1
W
1,p
(o
r
1
, (u
R
1
)
TM),
2
W
1,p
(o
r
2
, (u
R
2
)
TM),
1,2
W
1,p
(o
r
1
, (u
R
1
)
TM) W
1,p
(o
r
2
, (u
R
2
)
TM).
For T
r
R
d,0
, we use local charts near the boundary to write =
g
+
1
+
2
, where
1
T
r
1
,
2
T
r
2
and
g
R represents the component in the direction of the gluing
parameter. Thus
(, ) = (
g
, 0) + (
1
,
1
) + (0,
1,2
) + (
2
,
2
)
and the result follows if we can show that the nal three terms on the right hand side are all
in imT
R
. The cases of (
1
,
1
) and (
2
,
2
) have identical proofs, so we prove it for the rst
case.
Consider D
S,r
R
,u
R
(
1
,
1
) L
p
(o
r
R
, (u
R
)
TM). It is supported on the image of the trun-

cation o
r
1
s > R and so it can be identied with image of D
S,r
1
,u
R
1
(
1
,
1
). The operators
D
S,r,u
preserve basepoints on the surface o
r
, so the support of W
1,p
(o
r
, u
TM) on o
r
is
the same as the support of D
S,r,u
(, ) L
p
(o
r
, u
TM). Applying this fact we can conclude

that the support of D
S,r
R
,u
R
(
1
,
1
) is precisely the support of the cut-o function used in the
construction of T
R
, and therefore
T
R
D
S,r
R
,u
R
(
1
,
1
) = Q
1
D
S,r
1
,u
R
1
(
1
,
1
) = (
1
,
1
)
where the last equality follows from the standing assumptions that D
S,r
1
,u
R
1
is surjective and
has trivial kernel, so that it is an isomorphism; as such its right-inverse Q
1
is a left-inverse too.
Hence, (
1
,
1
) imT
R
.
For the other piece, consider D
S,r
R
,u
R
(0,
1,2
). The support of
1,2
is on the part of the
neck where all three of the operators D
S,r
R
,u
R
, D
S,r
1
,u
R
1
and D
S,r
2
,u
R
2
coincide, i.e.,
D
S,r
R
,u
R
(0,
1,2
) = D
S,r
1
,u
R
1
(0,
1,2
) = D
S,r
2
,u
R
2
(0,
1,2
).
By assumption both D
S,r
1
,u
R
1
and D
S,r
2
,u
R
2
are isomorphisms (being surjective with trivial
kernel). Writing = D
S,r
R
,u
R
(0,
1,2
), recall that in the construction of T
R
, one uses a cut-o
function to write =
1
+
2
, where the supports of
1
and
2
intersect only at the truncation
line s = R, and then the right inverse Q
1
is used on
1
and the right inverse Q
2
is used on
2
. We
know that Q
1
= (0, ) and Q
2
= (0, ). The operator T
R
is dened by T
R
= Q
1
1
+Q
2
2
.
So it is enough to show that Q
1
1
= Q
2
1
. So suppose for the sake of contradiction that
Q
1
1
Q
2
1
,= 0. If so, applying D
S,r
1
,u
R
1
, we could write
D
S,r
1
,u
R
1
(Q
1
1
Q
2
1
) = D
S,r
1
,u
R
1
Q
1
1
D
S,r
1
,u
R
1
Q
2
1
= D
S,r
1
,u
R
1
Q
1
1
D
S,r
2
,u
R
2
Q
2
1
=
1

1
= 0,
contradicting the assumption that ker D
S,r
1
,u
R
1
= 0. Hence, T
R
= (0,
1,2
). In summary, we
have:
(a) If (, ) T
r
R
d,0
W
1,p
(o
r
R
, (u
R
)
TM) is such that = (0,

1
,
2
) RT
r
1
de+1,0
T
r
2
e
= T
r
R
d,1
, then (, ) im T
R
,
(b) If (, 0) T
r
R
d,0
W
1,p
(o
r
R
, (u
R
)
TM) is such that = (

g
, 0, 0) RT
r
1
de+1,0
T
r
2
e
= T
r
R
d,0
, then (, 0) / im T
R
.
Since these elements span T
r
R
d,0
W
1,p
(o
r
R
, (u
R
)
TM), it follows that

(5.7) im T
R
= (, )
g
= 0.
Type 2. Suppose that (, ) T
r
R
d,0
W
1,p
(o
r
R
, (u
R
)
TM).
Let
0
+
1
+ . . . +
k
+
0,1
+ . . . +
0,k
= 1 be a smooth partition of unity on o
r
R
such
that for i = 0, . . . , k, the support of
i
is on the part of o
r
R
that comes from the truncation
of o
r
i
on the neck at s = R, and the support of
0,i
is on the subset of the neck of o
r
R
corresponding to R/2 s R in o
r
0
as well as the corresponding piece from o
r
i
. Then for
50 SIKIMETI MAU
W
1,p
(o
r
R
, (u
R
)
TM), it is clear that
0
W
1,p
(o
r
0
, (u
R
0
)
TM),
1
W
1,p
(o
r
1
, (u
R
1
)
TM),
. . .
k
W
1,p
(o
r
k
, (u
R
k
)
TM),
0,1
W
1,p
(o
r
0
, (u
R
0
)
TM) W
1,p
(o
r
1
, (u
R
1
)
TM),
. . .
0,k
W
1,p
(o
r
0
, (u
R
0
)
TM) W
1,p
(o
r
k
, (u
R
k
)
TM).
For T
r
R
d,0
, we use local charts near the boundary to write =
0
+
1
+ . . . +
k
+
g
,
where for i = 0, . . . , k,
i
T
r
i
, and
g
R represents the component of the tangent vector
in the direction of the gluing parameter. Then as in the previous case we can write
(, ) = (
g
, 0) + (
0
,
0
) + (
1
,
1
) +. . . + (
k
,
k
) + (0,
0,1
) +. . . (0,
0,k
)
and the result follows if all terms except the rst on the right hand side of the above expression
are in imT
R
. The same argument as used for Type 1 proves that all terms on the right except
for (
g
, 0) are in the image of T
R
, and the result follows. In summary:
(a) If (, ) T
r
R
d,0
W
1,p
(o
r
R
, (u
R
)
TM) is such that

g
= 0, then (, ) im T
R
,
(b) If (, 0) T
r
R
d,0
W
1,p
(o
r
R
, (u
R
)
TM) is such that =

g
, then (, 0) / im T
R
.
Since these elements span T
r
R
d,0
W
1,p
(o
r
R
, (u
R
)
TM), it follows that

(5.8) im T
R
= (, )
g
= 0.
Type 3. In this case we have glued (r
1
, u
1
) /
d,0
(x
0
, . . . , y, . . . , x
d
)
0
to a Floer trajectory
v

M(y, x
i
)
0
. Since the complement of im T
R
is ker D
R
, our goal is to show that the vector
space ker D
R
as the same dimension as ker D
v
, which by assumption is 1. From this it would
follow that the image of the gluing map lies in a one dimensional component of the moduli
space.
To show that the nite dimensional vector spaces ker D
v
and ker D
R
have the same di-
mension, it is enough to produce a pair of injective linear maps, : ker D
v
ker D
R
and
: ker D
R
ker D
v
. Write f
R
for the pregluing map.
Claim 1: := (1 Q
R
D
R
)df
R
: ker D
v
ker D
R
is injective.
By hypothesis ker D
v
is one-dimensional, so we know that an explicit basis is
s
v. It
suces therefore to show that for suciently large R, there is a constant c > 0 such that
(5.9) |
s
v|
1,p
c|(1 Q
R
D
R
)df
R
(
s
v)|
1,p
.
We prove the estimate by proving two separate inequalities
|df
R
(
s
v)|
1,p,R
c
1
|
s
v|
1,p
(5.10)
|D
R
df
R
(
s
v)|
0,p,R
(R)|
s
v|
1,p
(5.11)
where c
1
> 0 and (R) 0 as R . Together these imply (5.9), since the uniform bound
on the right inverse
|Q
R
|
1,p
C||
0,p
holds for R suciently large, and so
|(1 Q
R
D
R
)df
R
(
s
v)|
1,p
|df
R
(
s
v)|
1,p
|Q
R
D
R
df
R
(
s
v)|
1,p
c
1
|
s
v|
1,p
C|D
R
df
R
(
s
v)|
0,p
c
1
|
s
v|
1,p
C(R)|
s
v|
1,p
c
1
|
s
v|
1,p
(5.12)
for some c > 0 for suciently small (R).
To prove (5.10), we write v
(s, t) := v(s +, t). With this notation,
=0
=
s
v. Thus,
df
R
(
s
v)(s, t) =
d
d
=0
v
#
R
u
1
,
which by construction is supported only on the region s 3R/2 on the striplike end. The
pre-gluing map on this region is
f
R
(v
, u
1
) = v
#
R
u
1
(s, t) =
_
exp
Q
y(t)
((s +
3R
2
)
(s 2R, t)), s [
3R
2
,
3R
2
+ 1]
v
(s 2R, t), s
3R
2
+ 1.
and we need to take the derivative with respect to . On the region s 3R/2+1 we have that
df
R
(
s
v)(s, t) =
d
d
=0
v
(s 2R, t) = (
s
v)(s 2R, t).
Note that
|df
R
(
s
v)(s, t)|
1,p
|df
R
(
s
v)(s, t)|
1,p;[3R/2+1,)
= |(
s
v)(s 2R, t)|
1,p;[3R/2+1,)
= |
s
v|
1,p;[R/2+1,)
.
By exponential convergence of
s
v, there is a c
1
> 0 such that for all suciently large R,
|
s
v|
1,p;[R/2+1,)
c
1
|
s
v|
1,p
which proves (5.10).
To prove (5.11), observe rst that by construction, D
R
df
R
(
s
v) is supported only on the
interval s [3R/2, 3R/2 + 1]. It follows that the L
p
norm of D
R
df
R
(
s
v) is controlled by the
W
1,p
norm of df
R
(
s
v) on that interval. Let
R
(s) := (s+3R/2) and
R
(s, t) := (s2R, t).
52 SIKIMETI MAU
For s [3R/2, 3R/2 + 1],
df
R
(
s
v)(s, t) =
d
d
=0
exp
Q
y(t)
(
R
(s 2R, t))
=
d
d
=0
exp
Q
y(t)
(
R
(s)
R
(s +, t))
= dexp
Q
y(t)
(
R
R
)
d
d
=0
(
R
R
(s +, t))
=
R
dexp
Q
y(t)
(
R
R
)(
s
)(s 2R, t).
The identity
exp
Q
y(t)
(s 2R +, t) = v(s 2R +),
implies that
dexp
Q
y(t)
((s 2R, t))(
s
)(s 2R, t) =
s
v(s 2R).
The linear operator dexp
Q
y
() : T
y
M T
exp
Q
y

M is the identity for = 0, so is invertible
for small . For suciently large R, the exponential convergence of trajectories means that
(s2R, t) is uniformly small for s [3R/2, 3R/2+1]. Therefore on this interval, we can write
df
R
(
s
v)(s, t) =
R
(s)dexp
Q
y(t)
(
R
R
)[dexp
Q
y(t)
((s 2R, t))]
1
s
v(s 2R).
Thus there is a constant c
2
0 such that
|df
R
(
s
v)(s, t)|
1,p;[3R/2,3R/2+1]
c
2
|
s
v(s 2R, t)|
1,p;[3R/2,3R/2+1]
= c
2
|
s
v(s, t)|
1,p;[R/2,R/2+1]
(R)|
s
v(s, t)|
1,p
where the last inequality and the term (R) reects the fact that the ratio
|
s
v|
1,p;[R/2,R/2+1]
/|
s
v|
1,p
goes to 0 as R . This proves (5.11), hence also Claim 1.
Now let : R [0, 1] be a smooth cut-o function such that (s) = 0 for s 1/4 and
(s) = 1 for s 1/4, and 0

3. Dene a shifted and rescaled cut-o function
R
(s) :=
((sR)/R). Then
R
= 0 for s 3R/4 and
R
= 1 for s 5R/4, and 0

R
=

/R 3/R.
Claim 2: := (1 Q
u
R
1
D
u
R
1
)(1
R
) (1 Q
v
RD
v
R)
R
: ker D
R
ker D
u
R
1
ker D
v
R is
injective.
Let ker D
R
. Then = (1
R
) +
R
. On the support of (1
R
), the linearized
operators D
R
and D
u
R
1
coincide, so
D
u
R
1
(1
R
) = D
R
(1
R
) =

R
+ (1
R
)D
R
=

R
.
Hence |D
u
R
1
(1
R
)|
0,p
3/R||
0,p
. Similarly on the support of
R
, the linearized operators
D
R
and D
v
R coincide, so
D
v
R
R
= D
R
R
=

R
+
R
D
R
=

R
,
= |D
v
R
R
|
0,p
3/R||
0,p
.
Let c
1
and c
2
be uniform bounds for the right inverses Q
u
R
1
and Q
v
R respectively. Then we
have:
(1 Q
u
R
1
D
u
R
1
)(1
R
) (1 Q
v
RD
v
R)() = [(1
R
) +Q
u
R
1
(

R
),
R
Q
v
R
(

R
)]
Combine the identity (1
R
) +
R
= with the estimates
|Q
u
R
1
(

R
)|
1,p
3c
1
/R||
0,p
|Q
v
R(

R
)|
1,p
3c
2
/R||
0,p
to get that for R suciently large, there is a constant c 0 such that
|(1 Q
u
R
1
D
u
R
1
)(1
R
) (1 Q
v
RD
v
R)()|
1,p
c||
0,p
.
This proves Claim 2, completing the proof of Lemma 5.6.3.
5.7. Surjectivity of the gluing map. The nal step is to prove the surjectivity of the gluing
maps near the broken tuples. Our goal will be to show that for suciently small and su-
ciently large R, the gluing map associated to the given tuple surjects onto /
d,0
(x
0
, x
1
, . . . , x
d
)
1
, where the neighborhoods U
were dened in Section 4.7. We will prove surjectivity sepa-

rately for the dierent types of gluing constructions.
Type 1.
Proposition 5.7.1. Let
(r
1
, u
1
) /
de+1,0
(x
0
, . . . , x
i
, y, x
i+e+1
, . . . , x
d
)
0
and
(r
2
, u
2
) /
e
(y, x
i+1
, . . . , x
i+e
)
0
be regular, and let U
be a Gromov neighborhood of the pair. Given > 0, there is an > 0

such that the following holds. If (r, u) U
/
d,0
(x
0
, . . . , x
d
)
1
, then there is a pre-glued
curve (r
R
, u
R
) and a (, ) T
r
R
d,0

0
(o
r
R
, u
R
TM) such that exp
r
R
= r, exp
Q
u
R
= u,
[[ +||
1,p
< , and (, ) im Q
R
.
Proof. We will prove it by contradiction. Suppose there were a > 0, and sequences
0,
0, R
= log(
) , and (r
, u
) /
d,0
(x
0
, . . . , x
d
)
1
, and r
1,
r
1
, r
2,
r
2
such that r
= r
1,
#
r
2,
, with the following properties as : E(u
) E(u
1
) + E(u
2
),
and writing (r
R
, u
R
) for the preglued curve constructed with gluing length R
,
(5.13) inf[[ +||
1,p
(r
, u
) = (exp
r
R
, exp
Q
u
R
) .
Our goal is to contradict 5.13.
Writing =
g
+
1
+
2
where
g
is the component of in the direction of the gluing
parameter, it follows from the choice of gluing length R
that
g
= 0. For suciently small
,
54 SIKIMETI MAU
the condition exp
r
R
= r
determines uniquely, so let us call it
. The convergence of r
i,
to r
i
for i = 1, 2 implies that [
[ 0.
By assumption, u
u
1
uniformly on compact subsets of o
r
1
and u
u
2
uniformly
on compact subsets of o
r
2
; moreover since the maps are pseudoholomorphic, convergence on
these compact subsets is uniform in all derivatives. The preglued maps u
R
have the same
convergence properties, so for large there is a unique section

0
(o
r
R
, (u
R
)
TM)
such that exp
Q
u
R
= u
. So it is enough to show that |
|
1,p
< for suciently large ,
contradicting (5.13). Equivalently, we will show that the L
p
norms of
,
s
and
t
can
be made arbitrarily small by taking suciently large.
It follows from the uniform convergence in all derivatives on compact subsets that on such
subsets of o
r
1
o
r
2
, the L
p
norms of
,
s
and
t
all go to zero as . We can

choose these compact subsets to be such that their complement is on the striplike ends and
neck. Hence, without loss of generality, it suces to prove that the L
p
norms of
,
s
and
t
converge to zero along the striplike ends and neck of the preglued surfaces. The
exponential convergence of u
as well as u
R
along the striplike ends means that the L
p
norms
of
and its rst derivatives can be made arbitrarily small too; so the essential thing to prove
is that the L
p
norm of
along the neck of the preglued surface can be made arbitrarily small
with suciently large .
The neck consists of two nite strips of length R
identied along an end to form a single

strip,
[R
, R
] [0, 1]

= [0, R
] [0, 1] [0, R
] [0, 1]/ ,
where is the identication of (R
, 1 t) of the rst strip with (R
, t) of the second, for

t [0, 1].
Let
0
> 0 be given. Fix R > 0 large enough that lim
E(u
; [R, R
] [0, 1] [R, R
]
[0, 1]/ ) <
0
. Without loss of generality we can assume that
0
> 0 is small enough that
[
s
u
[ satises, by Proposition B.7, [

s
u
[ ce
2
s
for all and for all s [R, R
], for some
c, > 0. Since u
satises Floers inhomogeneous pseudoholomorphic equation on this strip,

we deduce
(5.14) [
t
u
(s, t) X
H
t
(u
(s, t))[ ce
2
s
.
Let
t
be the ow of the Hamiltonian vector eld X
]ulHt
, and consider the function u
:=
1t
(u
(s, t)). Then
t
u
= (
1t
)
(
t
u
X
H
t
(u
))
and (5.14) implies
dist( u
(s, 1) u
(s, 0))
_
1
0
[
t
u
[ dt ce
2
s
.
Since u
(s, 1) L
i
and u
(s, 0)
1
(L
i+e
), this means that both are very close to an inter-
section p
1
(L
i+e
) L
i
. The assumption of transverse intersection implies that there is a
constant a > 0 such that
dist( u
(s, 0), p) a dist( u
(s, 0), L
1
)
a dist( u
(s, 0), u
(s, 1))
a c e
2
s
.
Now for every other t,
dist( u
(s, t), p) dist( u
(s, t), u
(s, 0)) + dist( u
(s, 0), p)
ce
2
s
(t) +a c e
2
s
b e
2
s
.
In terms of the original function u
, and writing
1t
x(t) = p, this estimate translates into
dist(u
(s, t), x(t))
be
2
s
.
By construction, the preglued curves u
R
satisfy a similar inequality, and so
dist(u
(s, t), u
R
(s, t)) dist(u
(s, t), x(t)) + dist(x(t), u

R
(s, t)) C e
2
s
= [
(s, t)[ C
2
s
.
Taking the L
p
norm on a strip [R, R
] [0, 1] gives
_
R
R
_
1
0
[
[
p
ds dt C
_
R
R
e
p
2
s
ds =
C
p
2
(e
p
2
R
e
p
2
R
),
and this can be made arbitrarily small by choosing R large enough. By symmetry the same
estimate holds for the strip [R, R
] [0, 1] on the other side of the neck. Thus, the L

p
norm of
on the neck can be made arbitrarily small as .
Now we consider the L
p
norms of
s
and
t
. First note that since u
converges uniformly
in all its derivatives on compact subsets of o
r
1
and o
r
2
to the limits u
1
and u
2
, we see that
on such compact subsets we have uniform estimates for [
s
[ 0 and [
t
[ 0. On the
striplike ends of o
r
, the exponential convergence of u
and u
R
to the same limits mean that
the L
p
norms here can be made arbitrarily small. Therefore what we need to show is that the
L
p
norms of
s
and
t
on the neck, which varies in length with , can be made arbitrarily

small with large .
Write exp
Q
: TM M, and consider dexp
Q
: T(TM) TM. At a xed point (p, ) TM
we can take a tangent vector (, ), T
p
M, T
p
M, and write
dexp
Q
(p,)
(, ) = D
1
exp
Q
(p,)
() +D
2
exp
Q
(p,)
()
where D
1
exp
Q
p,
corresponds to varying the basepoint p while keeping all else xed, and D
2
exp
Q
p,
corresponds to xing the basepoint p and varying the tangent vector . In particular D
1
exp
Q
p,0
and D
2
exp
Q
(p,0)
are the identity, so for small they are invertible. Hence
s
u
=
s
exp
Q
u
R
= (D
1
exp
Q
)
(u
R
,
s
u
R
+ (D
2
exp
Q
)
(u
R
,
)
(
s
)
(D
2
exp
Q
)
1
(u
R
,
)
(
s
u
(D
1
exp
Q
)
(u
R
,
s
u
R
) =
s
(5.15)
(D
2
exp
Q
)
1
(u
R
,
)
(
t
u
(D
1
exp
Q
)
(u
R
,
t
u
R
) =
t
. (5.16)
56 SIKIMETI MAU
First we analyze (5.15). The operators (D
2
exp
Q
)
1
(u
R
,
)
and (D
1
exp
Q
)
(u
R
,
)
can be uni-
formly bounded for |
< , so by Proposition B.7 we get on each strip [R, R
] [0, 1] on
either side of the neck,
[
s
[ c
1
[
s
u
[ +c
2
[
s
u
R
[ Ae
2
s
for some constant A > 0, and therefore
_
R
R
_
1
0
[
s
[
p
ds dt
A
p
2
(e
p
2
R
e
p
2
R
)
which can be made arbitrarily small by choosing R large enough. The same estimate holds by
symmetry on the other side of the neck. From this we conclude that |
s
|
L
p can be made
arbitrarily small for large .
Now we analyze (5.16). We can write
t
u
= J
t
(u
)(
s
u
) +X
H
t
(u
t
u
R
= J
t
(u
R
)(
s
u
R
) +X
H
t
(u
R
) +E
(s, t)
where E
(s, t) is an error term that is supported only on the compact interval s [R
/2, R
/2+
1] of each of the two strips making up the neck, with [E
(s, t)[
0 as . Together
with (5.16) this yields a pointwise estimate
[
t
[ = [(D
2
exp
Q
)
1
(u
R
,
)
(
t
u
(D
1
exp
Q
)
(u
R
,
t
u
R
)[
c[
t
u
(D
1
exp
Q
)
(u
R
,
t
u
R
[
c
1
[
s
u
[ +c
2
[
s
u
R
[ +c
3
[X
H
t
(u
) (D
1
exp
Q
)
(u
R
,
)
X
H
t
(u
R
)[ +c
4
[E
(s, t)[
c
5
([
s
u
[ +[
s
u
R
[ + dist(u
, u
R
) +[E
(s, t)[).
From this, applying the estimates for [
s
u
[, [
s
u
R
[ and dist(u
, u
R
) and [E
(s, t)[ we get

_
R
R
_
1
0
[
t
[
p
ds dt c
6
_
R
R
_
1
0
_
[
s
u
[
p
+[
s
u
R
[
p
+dist(u
, u
R
)
p
+[E
(s, t)[
p
_
ds dt
c
7
_
R
R
e
2
s
ds +
_
R/2
R/21
ds
c
8
(e
2
R
e
2
R
) +c
7
,
and it is clear that this can be made arbitrarily small by taking R large enough.
This provides a contradiction to (5.13). Hence, given > 0, there is an > 0 such that
whenever (r, u) /
d,0
(x
0
, . . . , x
d
) U
, there is a gluing length R such that r = exp

r
R
, u =
exp
Q
u
R
, with [[ +||
1,p
, and with
g
= 0. By (5.7), this implies that (, ) im Q
R
.
Surjectivity for Type 2.
Proposition 5.7.2. Let (r
0
, u
0
), (r
1
, u
1
), . . . , (r
k
, u
k
) be regular, and let U
be a Gromov neigh-
borhood of the tuple. Given > 0, there is an > 0 such that the following holds. If
(r, u) U
/
d,0
(x
0
, . . . , x
d
)
1
, then there is a pre-glued curve (r
R
, u
R
) and a (, ) T
r
R
d,0
0
(o
r
R
, u
R
TM) such that exp
r
R
= r, exp
Q
u
R
= u, [[ +||
1,p
< , and (, ) im Q
R
.
Proof. Again observe that for suciently small > 0, if dist
R
(r, r
0
#
0
r
1
, . . . , r
k
) < , the
local charts near the boundary provide a unique way of writing r = r
0
#
r
r
1
, . . . , r
k
with
0
r
and with each r
i
in an -neighborhood of r
i
. So now suppose that there were a > 0
and sequences
0, (r
, u
) /
d,0
(x
0
, . . . , x
d
)
1
, and 0 <
, R
= log
such
that r
= r
0,
#
r
1,
, . . . , r
k,
, and dist
M
(u(z), u
i
(z))
on all compact subsets of o

r
i
, but
(5.17) inf[[ +||
1,p
= exp
R
, u
= exp
Q
u
R
.
Then the convergence would be uniform in all derivatives on those compact subsets, and we
could choose the compact subsets to be large enough that their complements comprise the
striplike ends and the k necks of the glued surfaces o
r
0
#
{r
1
,...,r
k
}
= o
r
R
. However on these
striplike ends and these necks, the energy of u
must approach 0, and the same exponential

decay arguments would imply that u
= exp
Q
u
R
for some

0
(o
r
R
, u
R
TM) with
|
|
1,p
0 as . Since [
[ 0 also, we would get a contradiction to (5.17). Hence

given > 0 we could nd an > 0 such that (r, u) /
d,0
(x
0
, . . . , x
d
)
1
U
could be written
as r = exp
r
R
, u = exp
Q
u
R
for some preglued curve (r
R
, u
R
), with [[ + ||
1,p
. Moreover,
this is such that
g
= 0, so it follows from (5.8) that (, ) im Q
R
.
Surjectivity for Type 3.
Proposition 5.7.3. Let (r
0
, u
0
) and v be regular, and U
a Gromov neighborhood of the pair.

Given > 0, there is an > 0 such that the following holds. If (r, u) U
/
d,0
(x
0
, . . . , x
d
)
1
,
then (r, u) is in the image of the gluing map.
Proof. In this case it suces to prove the following:
Claim 1: Given R
1
>> 0, and > 0, there is an > 0 such that the following holds. If
(r, u) U
/
d,0
(x
0
, . . . , x
d
)
1
, then there is an R R
1
, and a preglued curve (r
R
, u
R
) and a
(, ) T
r
R
d,0
0
(o
r
R
, u
R
TM) such that exp
r
R
= r, exp
Q
u
R
= u, and [[ +||
1,p
< .
To see how Claim 1 implies Proposition 5.7.3, the argument is as follows. From Section 5.6 we
know that the image of the gluing map is contained in the one dimensional component of the
moduli space of pseudoholomorphic quilted disks, /
d,0
(x
0
, . . . , x
d
)
1
. The image of [R
0
, ) is
a connected component of this one-dimensional manifold. The implicit function theorem also
tells us that in a local trivialization about a preglued curve (r
R
, u
R
), we get a local chart for
/
d,0
(x
0
, . . . , x
d
)
1
. This chart contains g(R), so the piece of the manifold /
d,0
(x
0
, . . . , x
d
)
1
covered by the chart intersects the image of the gluing map. So we want to show that if R is
58 SIKIMETI MAU
suciently big, and is suciently small, then the piece of the one-manifold determined by
the local chart about (r
0
, u
R
) is contained in the image of the gluing map.
The preglued curves u
R
are dened on the same domain o
r
0
, but by construction any two
of them will dier by a translation in the s direction suciently far along the striplike end
Z
i
. The magnitude of the distance between these translations depends on distances between
points in v, and the size of the dierence in gluing lengths. Since v is non-constant we can
chooose > 0 small enough that for any R
1
, there will eventually be an R
> R
1
such that the
preglued curves u
R
for R R
can not be written u

R
= exp
Q
u
R
1
with ||
1,p
. In particular,
for R R
the preglued curves (r

0
, u
R
) are not in a -neighborhood of the local trivialization
about (r
0
, u
R
1
), and similarly (r
0
, u
R
1
) is not in a -neighborhood of the local trivialization
about (r
0
, u
R
). So considering the gluing map g : [R
0
, ) /
d,0
(x
0
, . . . , x
d
)
1
, by making
> 0 smaller if necessary we can assume that the hypotheses of the implicit function theorem
are satised. Then we can x an R
1
> R
0
such that the respective -neighborhoods of (r
0
, u
R
1
)
and (r
0
, u
R
0
) are disjoint. If we suppose, as in Claim 1, that (r, u) /
d,0
(x
0
, . . . , x
d
)
1
is such
that r = exp
r
0
, u = exp
Q
u
R
for some R R
1
, and [[ +||
1,p
< , then (r, u) is in the local
chart around (r
0
, u
R
). But since R R
1
we see that g(R
0
) is not in this chart, and we can
choose an R
2
>> R large enough that g(R
2
) is also not in that chart, but by the connectedness
of the image of the gluing map this means that the whole chart is contained in the image of
the gluing map.
Proof of Claim 1 For the sake of contradiction suppose that the assertion were false. Then
there would be some > 0, and sequences
0, R
= log
2R
, and
(r
, u
) /
d,0
(x
0
, . . . , x
d
)
1
such that
dist
R
(r
, r
0
) <
,
[E(u
0
) +E(v) E(u
)[ <
,
dist
M
(u
(z), u
0
(z)) <
for all z o
R
r
0
, and
dist(u
(s +
, t), v
(
s, t)) <
for all s [R
, R
],
and yet for every ,
(5.18) inf[[ +||
1,p
= exp
r
0
(), u
= exp
Q
u
R
() .
For large , the condition exp
r
0
= r
uniquely determines =:
, and the convergence

implies that [
[ 0. So the quantity [[ becomes insignicant in (5.18). We will arrive at a

contradiction by showing that the norms ||
1,p
in (5.18) must also go to 0 for large . The
assumptions show that u
converges to u
0
uniformly on compact subsets of o
r
0
, and since both
are pseudoholomorphic the convergence is uniform in all derivatives. On the striplike end Z
i
,
u
(s +
n
, t) converges uniformly on compact subsets of R[0, 1] to v(s, t); and since they are
pseudoholomorphic curves the convergence is uniform in all derivatives. Moreover, the preglued
curves (r
0
, u
) converge in the same way. For each the energy of u
restricted to the subsets

[R(
),
n
R(
)] [0, 1] and s
n
+ R(
) of the striplike end Z

i
goes to zero. Thus, the
proof reduces to the same calculations as done for Type 1. That is, the uniform estimates
of convergence on those compact subsets of o
r
0
, combined with exponential decay estimates
based on the vanishing energy of the strips in the complement of those compact subsets, show
that for suciently large there is a unique section

0
(o
r
0
, u
) for which exp

Q
u
= u
,
and |
|
1,p
0, contradicting 5.18.
Appendix A. Sobolev embeddings.
We collect relevant W
1,p
embedding statements for domains in R
2
, to verify that for each
quilted surface S constructed in Section 2, there exists a constant c
p
such that
|f|
L
(S)
c
p
(S)|f|
W
1,p
(S)
and that for the families constructed in Chapter 4, there is a uniform bound c
p
(S) c
0
. The
following theorem is a consequence of general Sobolev estimates (see Theorem B.1.11, [5]).
Theorem A.1. Let S R
2
be a compact Lipschitz domain. Let u C
(S). Then there is a

constant c, depending only on p, such that
sup
S
[u(s, t)[ c|u|
W
1,p
std
.
Here the W
1,p
std
norm refers to the standard volume form ds dt on R
2
.
For a general volume form dvol
S
, we have the following consequence of Theorem A.1.
Corollary A.2. Let S R
2
be a compact Lipschitz domain, and dvol
S
= f(s, t)ds dt a
volume form on S. Let u C
(S). Then there is a constant c = c(p) (in fact it is the same

constant as in Theorem A.1) such that
sup
S
[u(s, t)[
c
(f
min
)
1/p
|u|
W
1,p
(S)
for all z S, where | |
W
1,p
(S)
denotes the W
1,p
norm dened by the volume form dvol
S
, and
f
min
= min
S
f(s, t).
60 SIKIMETI MAU
Proof. Since f(s, t)ds dt is a volume form, f(s, t) > 0 for all (s, t) S. Moreover, S is
compact so f achieves a minimum f
min
> 0. Hence, using the constant c of Theorem A.1,
sup
(s,t)S
[u(s, t)[ c|u|
W
1,p
= c
_
_
__
S
([u[
p
+[du[
p
)ds dt
_
_
1/p
= c
_
_
__
S
([u[
p
+[du[
p
)
f(s, t)
f(s, t)
ds dt
_
_
1/p
c
_
_
__
S
([u[
p
+[du[
p
)
f(s, t)
f
min
ds dt
_
_
1/p
=
c
f
min
_
_
__
S
([u[
p
+[du[
p
f(s, t))ds dt
_
_
1/p
=
c
f
min
|u|
W
1,p
(S)
.
Similar estimates hold for unbounded domains R

2
whose geometry satises a cone condition;
see [1], Chapter 5.
Denition A.0.4. A domain R
2
satises the cone condition if there is a nite cone
C = C(r
c
,
c
) such that each x is the vertex of a nite cone C
x
contained in and
congruent to C.
Theorem A.3. Let R
2
. Suppose that satises the cone condition for some nite cone
C = C(r
c
,
c
), and let p > 2. Then there is a constant c = c(r
c
,
c
, p) > 0 such that for every
f C
(S) W
1,p
(), and every x ,
[f(x)[ c|f|
W
1,p.
Theorem A.4. Let S = S
1
. . . S
l
be a surface dened by a union of open sets S
i
, such that
each S
i
is one of the following types:
(a) The closure of S
i
is dieomorphic to a compact Lipschitz domain

S
i
R
2
.
(b) The closure of S
i
is dieomorphic to a domain

S
i
R
2
that satises the cone condition,
for some cone C
i
= (r
i
,
i
), and the volume form on S restricted to S
i
is the pull-back
of the standard volume form on R
2
.
Then there is a constant c, depending on p, the cones in the cone condition, and the volume
form dvol
S
restricted to the S
i
s of type (a), such that
sup
S
[u[ c|u|
W
1,p
(S)
for all u C
(S) W
1,p
(S).
Proof. Let
i
be a partition of unity subordinate to the cover S
1
. . . S
l
of S. Then
u =
l
i=1
i
u, and each
i
u C
(S
i
) W
1,p
(S
i
). By Corollary A.2 and Theorem A.3, there is
a constant c
i
which depends on p and, in the case of the subsets of type (a), the volume form
dvol
S
restricted to those components, such that
sup
S
[
i
u[ = sup
S
i
[
i
u[ c
i
|
i
u|
W
1,p
(S
i
)
= c
i
|
i
u|
W
1,p
(S)
c
i
|u|
W
1,p
(S)
.
Hence, sup
S
[u[ sup
S
l
i=1
[
i
u[
l
i=1
sup
S
[
i
u[
_
l
i=1
c
i
_
|u|
W
1,p
(S)
= c|u|
W
1,p
(S)
.
Appendix B. Exponential decay

Results on exponential decay for Floer trajectories with small energy follow from two things:
a convexity estimate for the energy density of a trajectories in a suciently small neighborhood
of a generalized intersection point, and a mean-value inequality that converts L
2
energy den-
sity estimates to pointwise estimates. The results we collect here are based on the convexity
estimates in [6], and mean-value inequality in [9].
Without loss of generality, we only need to consider solutions to
s
u +J
t
t
u = 0, (B.1)
u(s, 0) L
0
, u(s, 1) L
1
,
where L
0
and L
1
are transversely intersecting Lagrangians. This is because a solution of the
inhomogeneous equation,
s
u +J
t
(
t
u X
Ht
(u)) = 0 (B.2)
u(s, 0) L
0
, u(s, 1) L
1
,
can be translated into a solution of type (B.1) by setting u(s, t) =
1t
(u(s, t)), where
t
is the
time t ow of the Hamiltonian vector eld X
Ht
, and

J
t
:= (
1
1t
)
J
t
, which satises
s
u +

J
t
t
u = 0,
u(s, 0)
1
(L
0
), u(s, 1)
0
(L
1
) = L
1
,
and by assumption the Hamiltonian perturbation is such that
1
(L
0
) intersects L
1
transversely.
62 SIKIMETI MAU
Consider a solution u : I [0, 1] M of (B.1) where I = [T, T], [T, ) or (, T] for
some T > 0. We assume that I is xed. The energy density of u on this strip is the function
(B.3) e(s, t) := (
s
u(s, t), J
t
s
u(s, t)) = [
s
u(s, t)[
2
Jt
= [
t
u(s, t)[
2
Jt
Let =
2
s
+
2
t
be the Laplace operator on I [0, 1].
Lemma B.1. There are constants A
1
, a, B
1
, b, C
1
, c 0, which depend on M, , J
t
and the
Lagrangians L
0
, L
1
, such that if u : I [0, 1] M satises (B.1), then its energy density, e,
satises
e A
1
e ae
2
(B.4)
e
t
t=0
B
1
e be
3/2
(B.5)
e
t
t=1
C
1
e +ce
3/2
. (B.6)
Proof. First we prove (B.4). We abbreviate =
s
u, =
t
u,; so +J
t
(u) = 0. Consider the
(t-dependent) Riemannian metric , )
Jt
on M, and let be its Levi-Civita connection. Then
s
, )
Jt
= 2
s
, )
Jt
t
, )
Jt
= 2
t
, )
Jt
+(,

J
t
) = 2
t
, )
Jt
+J
t
,

J
t
)
Jt
from which we get that
e(s, t) = 2
s
s
, )
Jt
+ 2
t
t
, )
Jt
+
t

J
t
, J
t
)
Jt
= 2
s
s
, )
Jt
+ 2
s
,
s
)
Jt
+2
t
t
, )
Jt
+ 2
t
,
t
)
Jt
+ 2(
t
,

J
t
)
+
t
(

J
t
), J
t
)
Jt
+

J
t
,
t
(J
t
))
Jt
+(

J
t
,

J
t
J
t
)
= 2
s
s
, )
Jt
+ 2
s
,
s
)
Jt
+2
t
t
, )
Jt
+ 2
t
,
t
)
Jt
+ 2J
t
t
,

J
t
)
Jt
+
t
(

J
t
), J
t
)
Jt
+

J
t
,
t
(J
t
))
Jt

J
t
,

J
t
)
Jt
= 2(
2
s
+
2
t
), )
Jt
+ 2
s
,
s
)
Jt
+ 2
t
,
t
)
Jt
+(
t

J
t
) +

J
t
t
, J
t
)
Jt
+

J
t
, (
t
J
t
) +J
t
t
)
Jt

J
t
,

J
t
)
Jt
Since
t
J
t
=
J +

J and
t

J
t
=

J
t
+

J
t
, there is a constant c
2
> 0 that depends on J
t
and , and exists because of the compactness of M, such that the operator norms
|
J
t
|, |

J
t
| c
2
|
t
J
t
| c
2
([[
Jt
+ 1) = c
2
([[
Jt
+ 1)
|
t

J
t
| c
2
([[
Jt
+ 1) = c
2
([[
Jt
+ 1)
where the equalities follow from the identity = J
t
, with J
t
an isometry for the metric
, )
Jt
. We temporarily drop the subscript J
t
from our notation for the metric for the next few
calculations, since there is no danger of confusion. So we can estimate
[(
t

J
t
), J
t
)[ c
2
([[ + 1)[[
2
[

J
t
t
, J
t
)[ c
2
[
t
[[[
[

J
t
, (
t
J
t
))[ c
2
2
([[ + 1)[[
2
[

J
t
, J
t
t
)[ c
2
[[[
t
[
[

J
t
,

J
t
)[ c
2
2
[[
2
= e(s, t) 2[(
2
s
+
2
t
)[[[ + 2[
s
[
2
+ 2[
t
[
2
(c
2
+c
2
2
)[[
3
(c
2
+ 2c
2
2
)[[
2
2c
2
[
t
[[[.
It remains to estimate [(
2
s
+
2
t
)[. Since we are using a Levi-Civita connection,
t
=
s
.
Write R(, ) for the curvature tensor associated to . Then
s
+
t
t
=
s
s
+
t
=
s
s
+ (
t
s

s
t
) +
s
=
s
(
s
+
t
) +R(, ).
Analyzing
s
+
t
:
s
+
t
=
s
(J
t
) +
t
(J
t
)
= (
s
J
t
) J
t
s
+ (
t
J
t
) +J
t
= (
s
J
t
) + (
t
J
t
)
=
s
(
s
+
t
) = (
2
s
J
t
) (
s
J
t
)
s
+(
s
t
J
t
) + (
t
J
t
)
s
.
Putting things together gives
2
s
+
2
t
= (
2
s
J
t
) (
s
J
t
)
s
+(
s
t
J
t
) + (
t
J
t
)
s
+R(, )().
Since
s
J
t
=
J
t
and
t
J
t
=
J
t
+

J
t
3
depending only on , J
t
, and
the fact that M is compact, such that we can estimate operator norms
|
s
J
t
| c
3
[[
|
s
(
s
J
t
)| c
3
([
s
[ +[[
2
)
|
s
(
t
J
t
)| c
3
([
s
[ +[[[[ +[[)
= c
3
([
t
[ +[[
2
+[[).
Also by the compactness of M there is a uniform constant c
4
> 0 such that [R(X, Y )Z[
c
4
[X[[Y [[Z[ for all X, Y, Z T
p
M, and all p M. Hence,
[
2
s
+
2
t
[ c
3
([
s
[ +[[
2
)[[ +c
3
[[[
s
[ +c
3
([
t
[ +[[
2
+[[)[[
+c
2
([[ + 1)[
s
[ +c
4
[[
2
[[
= c
3
[
s
[[[ +c
3
[[
3
+c
3
[[[
t
[ +c
3
[
t
[[[ +c
3
[[
3
+c
3
[[
2
+c
2
[[[
s
[ +c
2
[
s
[ +c
4
[[
3
= c
2
[
s
[ + (c
3
+c
2
)[
s
[[[ + 2c
3
[
t
[[[ + (2c
3
+c
4
)[[
3
+c
3
[[
2
64 SIKIMETI MAU
leading to the estimate
e 2c
2
[
s
[[[ 2(c
3
+c
2
)[
s
[[[
2
4c
3
[
t
[[[
2
2(2c
3
+c
4
)[[
4
2c
3
[[
2
+2[
s
[
2
+ 2[
t
[
2
(c
2
+c
2
2
)[[
3
(c
2
+ 2c
2
2
)[[
2
2c
2
[
t
[[[
2[
s
[
2
+ 2[
t
[
2
c[
s
[([[
2
+[[) c[
t
[([[
2
+[[) c([[
4
+[[
3
+[[
2
where c > 0 is chosen large enough to absorb the other constants. We apply the estimate
2xy x
2
+y
2
to the mixed terms,
[
s
[([[
2
+[[)

2
2
[
s
[
2
+
1
2
2
([[
2
+[[)
2
[
t
[([[
2
+[[)

2
2
[
t
[
2
+
1
2
2
([[
2
+[[)
2
.
Choosing > 0 suciently small that B
2
/2 < 2c
1
leads to an inequality
e(s, t)
A([[
4
+[[
3
+[[
2
)
where

A depends only on J
t
, M, . The term [[
3
can be absorbed into the other terms, in the
sense that we can nd constants

A
1
0, a 0 such that
A([[
2
+[[
3
+[[
4
)

A
1
[[
2
+a[[
4
A
1
[[
2
Jt
+a[[
4
Jt
= A
1
e +ae
2
.
Thus e A
1
e ae
2
, proving (B.4).
Now we prove (B.5), (B.6). In each case we x a reference Riemannian metric , ) on M
that renders L
0
(respectively L
1
) totally geodesic, and write for its associated Levi-Civita
connection. Since the manifold M is compact, there are uniform constants c
1
, c
2
> 0 such that
for all p M, X T
p
M, and for all t [0, 1],
c
1
[X[
Jt
[X[ c
2
[X[
Jt
.
Then in either case we have
e
t
=
t
(
s
u, J
t
(u)
s
u)
= 2(
t
s
u, J
t
(u)
s
u) +(
s
u, (
t
J
t
)
s
u) + (
tu
)(
s
u, J
t
(u)
s
u).
For the rst term,
s
u =
s
t
u
=
s
(J
t
(u)
s
u)
= (
su
J
t
)
s
u +J
t
(u)
s
s
u
= [(
t
s
u, J
t
(u)
s
u)[ [(
su
J
t
)
s
u, J
t
(u)
s
u)[ +[(J
t
(u)
s
s
u, J
t
(u)
s
u)[
c
3
[
s
u[
3
+[(
s
s
u,
s
u)[.
Evaluating at t = i where i is 0 or 1, the boundary conditions on u imply that
s
u(s, i)
T
u(s,i)
L
i
, and since L
i
is totally geodesic
s
s
u(s, i) T
u(s,i)
L
i
. Thus, at t = i, (
s
s
u,
s
u) =
0. Since
t
J
t
=
tu
J
t
+

J
t
4
0 such that |
t
J
t
| c
4
([
t
u[ +1). With
this, the second term can be estimated by
[(
s
u, (
t
J
t
)
s
u)[ c
5
[
s
u[
2
([
t
u[ + 1)
c
6
([
s
u[
3
+[
s
u[
2
)
c
7
([
s
u[
3
Jt
+[
s
u[
2
Jt
)
= c
7
(e
3/2
+e).
To estimate the last term, by compactness there is a uniform constant c
8
> 0 such that
[(
tu
)(
s
u, J
t
(u)
s
u)[ c
8
[
t
u[[
s
u[
2
c
2
c
8
[
t
u[
J
i
[
s
u[
2
J
i
= c
2
c
8
[
s
u[
3
J
i
= c
2
c
8
e
3/2
.
Combining all estimates gives
e
t
t=i
De +de
3/2
, proving (B.5), (B.6).
We use the mean-value inequalities of [9] for n = 2.

4
Theorem B.2 (Theorem 1.1, [9]). There exist constants C, > 0 such that the following
holds. Let B
r
(0) R
2
be the ball of radius 0 < r 1. Suppose that the non-negative function
e C
2
(B
r
(0), [0, )) satises for some A
0
, A
1
, a 0,
e A
0
A
1
e ae
2
and
_
Br(0)
e a
1
.
Then
e(0) CA
0
r
2
+C(A
1
+r
2
)
_
Br(0)
e.
Theorem B.3 (Theorem 1.3, [9]). There exists a constant C,and for all a, b 0 there exists
(a, b) > 0 such that the following holds: Consider the (partial) ball D
r
(y) H
2
for some r > 0
and y H
2
. Suppose that e C
2
(D
r
(y), [0, )) satises for some A
0
, A
1
, B
0
, B
1
0
_
e A
0
A
1
e ae
2
,
H
2
e B
0
+B
1
e +be
3/2
and
_
Dr(y)
e (a, b).
Then
e(y) CA
0
r
2
+CB
0
r +C(A
1
+B
2
1
+r
2
)
_
Dr(y)
e.
We apply these theorems to the estimates of Lemma B.1. In the following we write I
1/2
=
s I[[s 1/2, s + 1/2] I, i.e., all points in the interval I which are at least distance 1/2
from the endpoint(s) of I.
4
In [9] the Laplace operator is
2
s

2
t
, , so at rst sight some inequalities appear reversed from those in
[9].
66 SIKIMETI MAU
Corollary B.4. Suppose that u : I [0, 1] M is a solution of (B.1). Then there is a > 0
such that if E(u) < , then there is a constant C depending on M, J
t
, L
0
, L
1
such that for every
s I
1/2
,
(B.7) e(s, t) C
_
s+1/2
s1/2
_
1
0
e dt ds.
Proof. This will follow from the theorems and Lemma B.1, since the derivative in the outer
normal direction at t = 0 is precisely
e
t
t=0
, while at t = 1 it is precisely
e
t
t=1
.
So given the estimates of Lemma B.1, x small enough that for each s I
1/2
, the uniform
inequality
_
D
1/2
(s,t)
e E(u)
is enough to imply that the mean-value inequalities of the Theorems hold; the choice of such a
therefore depends on the constants a, b and c in the estimates of Lemma B.1. Then it follows
directly from the mean-value inequalities of the Theorems that there exists a uniform constant
C such that
e(s, t) C
_
D
1/2
(s,t)
e C
_
s+1/2
s1/2
_
1
0
e dt ds.
We now study solutions of (B.1) with energy E(u) < . If is small enough, then the
mean-value inequality provides a uniform pointwise bound on the sizes of
t
u and
s
u for
(s, t) I
1/2
[0, 1],
(B.8) [
t
u[
2
Jt
= [
s
u[
2
Jt
C
.
In particular, for a xed s I
1/2
, the path
s
: [0, 1] M dened by
s
(t) := u(s, t) satises
dist
M
(
s
(0),
s
(1)) C
1
_
1
0
|
s
|
Jt
dt
C
1
(
_
1
0
|
t
u|
2
Jt
dt)
1/2
C
1
(C
)
1/2
. (B.9)
By the transversality of the intersection L
0
L
1
, and the compactness of M, intersection points
are isolated in M so if is small enough, all paths
s
are close to the same intersection point
p L
0
L
1
. Thus, given any neighborhood of this p L
0
L
1
, one can choose > 0 small
enough that each path
s
, s I
1/2
, is contained in that neighborhood.
Dene f : I
1/2
R by
(B.10) f(s) =
1
2
_
1
0
[
t
u[
2
Jt
dt.
It follows from [6] that given an intersection point p L
0
L
1
, there is a neighborhood U of p
such that the the function f(s) satises a convexity estimate
(B.11)

f(s)
2
f(s)
for some > 0. Therefore, choose small enough that all paths
s
are contained in such a
neighborhood. Combining this convexity estimate with the mean value inequality again, we
can prove the following standard exponential decay results for strips.
Proposition B.5. Let u : [0, ) [0, 1] M be a solution of (B.1), such that E(u) < .
Then there exist constants > 0 and A > 0 such that
(B.12) [
s
u(s, t)[ A e
s
.
Proof. We can safely ignore a compact subset [0, T][0, 1] of the strip, which can be bounded by
some xed constant. So choosing suciently large T we can assume without loss of generality
assume that the strip is of the form [T, ) [0, 1], the energy of u restricted to this strip is
less than for which the convexity estimate (B.11) holds. Note that the fact that E(u) <
implies that
s
u 0 as s , so in particular f(s) 0 as s . Then for s T, we have
that
f(s)
2
f(s).
This convexity estimate on f(s) implies (explained, for instance, in [6]) an inequality f(s)
ce
s
for some c > 0, i.e.,
f(s) =
_
1
0
[
s
u(s, t)[
2
Jt
dt ce
s
.
Now by (B.7) we get
[
s
u[
2
Jt
C
_
s+1/2
s1/2
f(s) ds C c e
(s1/2)
= A e
s
.
Proposition B.6. There is a > 0 so that the following holds. For any solution v : [, ]
[0, 1] M of (B.1) with E(v) < , there is a > 0 such that
(B.13) E(v; [ +T, T] [0, 1]) e
T
E(v)
for all 1 T /2.
Proof. Take to small enough that E(v) < implies the a priori estimate (B.8) for all +1
s 1, as well as the convexity estimate (B.11). Let us write
E(T) := E(v; [ +T, T] [0, 1]).
Then in terms of f, we have
E(T) =
_
T
+T
f(s)ds
68 SIKIMETI MAU
is a monotone decreasing function of T. Taking derivatives with respect to T,
E
(T) = f( T) f( +T),
E
(T) =

f( T)

f( +T) =
_
T
+T
f(s)ds
2
_
T
+T
f(s)ds =
2
E(T).
Thus e
T
(E
(T) + E(T)) is monotone increasing. An inequality E
(T) + E(T) > 0 would

imply that
e
T
(E
(T) +E(T)) > > 0,

so that E
(T) +E(T) > e

T
would imply imply(e
T
E(T))
> e
2T
, i.e., E(T) grows expo-
nentially, which is impossible. Hence E
(T) +E(T) 0, so e
T
E(T) is monotone decreasing,
and e
T
E(T) E(0) = E(v) implies E(T) e
T
E(v).
Applying the mean-value inequality we get the following description of the behavior of long
pseudo-holomorphic strips with small energy.
Proposition B.7. There is a > 0 so that the following holds. For any solution v : [, ]
[0, 1] M of (B.1) with E(v) < , there is a > 0 and A > 0 which depend only on M, , J
t
,
L
0
and L
1
, such that
(B.14) [
s
v[
2
Jt
Ae
|s|
for all s [ + 1, 1].
Proof. This is an application of the mean-value inequality to the previous lemma. By (B.7) we
have, for each s [ + 1/2, 1/2],
[
s
u(s, t)[
2
Jt
C
_
|s|1/2
|s|+1/2
_
1
0
[
s
u[
2
Jt
ds dt = CE([s[ 1/2)
CE(v)e
(|s|1/2)
Ce
/2
e
|s|
=: Ae
|s|
.
References
[1] Robert A. Adams. Sobolev spaces. Academic Press [A subsidiary of Harcourt Brace Jovanovich, Publishers],
Pure and Applied Mathematics, Vol. 65, New York-London, 1975.
[2] Kenji Fukaya and Yong-Geun Oh. Zero-loop open strings in the cotangent bundle and Morse homotopy.
Asian J. Math., 1(1):96180, 1997.
[3] S. Mau, K. Wehrheim, and C. Woodward. A functors for Lagrangian correspondences. In preparation,
2009.
[4] S. Mau and C. Woodward. Geometric realizations of the multiplihedron and its complexication. Preprint
arXiv:0802.2120.
[5] Dusa McDu and Dietmar Salamon. J-holomorphic curves and symplectic topology, volume 52 of American
Mathematical Society Colloquium Publications. American Mathematical Society, Providence, RI, 2004.
[6] Joel W. Robbin and Dietmar A. Salamon. Asymptotic behaviour of holomorphic strips. Ann. Inst. H.
Poincare Anal. Non Lineaire, 18(5):573612, 2001.
[7] Paul Seidel. Fukaya categories and Picard-Lefschetz theory. Zurich Lectures in Advanced Mathematics.
European Mathematical Society (EMS), Z urich, 2008.
[8] James Dillon Stashe. Homotopy associativity of H-spaces. I, II. Trans. Amer. Math. Soc. 108 (1963),
275-292; ibid., 108:293312, 1963.
[9] Katrin Wehrheim. Energy quantization and mean value inequalities for nonlinear boundary value problems.
J. Eur. Math. Soc. (JEMS), 7(3):305318, arXiv:math/0405484, 2005.
[10] Katrin Wehrheim and Chris Woodward. Functoriality for Lagrangian correspondences in Floer theory.
Preprint arXiv:0708.2851.
[11] Katrin Wehrheim and Chris Woodward. Pseudoholomorphic quilts. Preprint arXiv:0905.1369.
[12] Katrin Wehrheim and Chris Woodward. Quilted Floer cohomology. Preprint arXiv:0905.1370.

Sikimeti Ma'u - Gluing Pseudoholomorphic Quilted Disks

Cargado por

Información del documento

Descripción original:

Título original

Derechos de autor

Formatos disponibles

Compartir este documento

Compartir o incrustar documentos

Opciones para compartir

¿Le pareció útil este documento?

¿Este contenido es inapropiado?

Copyright:

Formatos disponibles

Sikimeti Ma'u - Gluing Pseudoholomorphic Quilted Disks

Cargado por

Copyright:

Formatos disponibles

a

structure of the Fukaya category comes

spaces, just as the associahedra are intrinsically connected

of admissible generalized Lagrangian correspondences from M

functor [7, Chapter I, (1b)]:

functor between non-unital A

categories / and B consists of

functor relations, which have four distinct kinds

is a pointed Riemann surface with markings

, and for each marking a striplike end, which is a holomorphic

(s, 1) S for all s R

) for the set of boundary components of

are noncompact, each

(s, 0) I then there is a marked point

(s, 1). Similarly if is a marked point on

(s, 1) I, then there is a marked point

S with a symplectic manifold (M

) with a Lagrangian correspondence L

, and each true boundary

. Denote the labels collectively

-compatible almost complex

(z) varying smoothly with z S

. Along a striplike end

(s, t), (s, t) [0, )

(t), i.e., to be independent of s. We write j

for the complex

(T), E(T), V (T), V

(T) is incident to a single vertex in V (T),

be generalized Lagrangian submanifolds of M, and let (H, J) be

(r, s, t). Each quilted surface S

TM) of (0, 1) forms on o

come from the induced metric g

come from the induced metrics g

, the Hermitian connection (4.2).

Now let o be a family of quilted surfaces of type either o

of the quilt S is labeled by a target symplectic manifold (M

-compatible almost complex structure J

and the Lagrangian L

, and the seam t = 0 is labeled by L

and the Lagrangian boundary condition L

) between the metric g

. Note that both g

-invariant too. On the complements

TM) as follows. Let

, with local coordinates z = s+it Ri[0, ), dene Q((z, )(z) by

= (, 1) = (, 1) be the common end point of the two paths for xed . It

((, t)) for the parallel transports of T along

(, t) = t +tDQ(0)() = t, so the quantities above must be zero.

4.4. Local trivializations. We now dene local trivializations of the bundles B[

at s = R, and truncate the surface o

(R, 1 t); see Figure 5.1.

, such that L is totally geodesic when t = 0, and L

is totally geodesic when t = 1.

((R, ) [0, 1])

+((R, ) [0, 1]).

(i) (s, t)).

(i) (s, t), s [R/2 1, R/2]

(i) (s, t), s R/2.

(s, t)) and v(s, t)

(s, t)), exp

(s, t), s [R/2 1, R/2]

(s, t), s R/2.