Vous êtes sur la page 1sur 48

a

r
X
i
v
:
1
4
0
8
.
6
7
9
4
v
1


[
m
a
t
h
.
S
G
]


2
8

A
u
g

2
0
1
4
THE FAMILY FLOER FUNCTOR IS FAITHFUL
MOHAMMED ABOUZAID
Abstract. Family Floer theory yields a functor from the Fukaya category of a symplec-
tic manifold admitting a Lagrangian torus bration to a (twisted) category of perfect
complexes on the mirror rigid analytic space. This functor is shown to be faithful by a
degeneration argument involving moduli spaces of annuli.
1. Introduction
Applications of Fukaya categories to symplectic topology require an algebraic model for
these categories: this involves of nding a collection of Lagrangians which generate the cat-
egory in the sense that the Fukaya category fully faithfully embeds in the category of perfect
modules over the corresponding A

algebra. For closed symplectic manifolds, the known


strategies for understanding such categories of modules rely on realising them, in an instance
of homological mirror symmetry, as modules over the endomorphism algebra of (complexes
of) coherent sheaves on an algebraic variety, or a non-commutative deformation thereof.
Such descriptions are possible in a limited class of examples, which include Calabi-Yau hy-
persurfaces in projective space [14, 16] and toric varieties [11, 5]. It reasonable to expect
that these methods will lead to descriptions of Fukaya categories of complete intersections
in toric varieties [4].
The goal of the family Floer program is to both give a more compelling proof of these
equivalences, and to extend the class of examples for which they can be proved. Keeping
with tradition, we shall call the symplectic side the A-side, and the algebro-geometric side
the B-side. The current strategies rely on matching computations on the two sides, without
having a good reason for the agreement. Moreover, these computations work for a very
special class of symplectic structures; in the typical case of the K3 surface, homological
mirror symmetry is only understood for the restriction of the Fubini-Study form to the
quartic hypersurface, whereas the rank of the second cohomology group is 22.
There are essentially only two previous results on family Floer cohomology. In [9], Fukaya
gave a very general result constructing the local charts of the B-side, which were shown by
Tu to admit compatible identications over the overlaps [18]. In [8, Section 6] Fukaya also
outlined a strategy for assigning to Lagrangians (complexes of) coherent sheaves on the
mirror, under some convergence assumptions which should yield a complex analytic mirror.
This paper extends these results by (1) constructing a map of morphism spaces from the
A-side to the B-side, (2) constructing a map of morphism spaces from the B-side to the
A-side, (3) showing that the composition of these two maps is the identity on the A-side,
leading to the main result. The formal results are stated in Section 8. The Appendix also
includes the construction of an A

functor, which should be thought of as the family Floer


mirror functor.
Date: August 29, 2014.
The author was supported by NSF grant DMS-1308179.
1
2 M. ABOUZAID
x
y
F
q
F
q
L
L
L
L
L
F
q L
F
q
Figure 1.
Remark 1.1. In order to focus on the new ideas, we restrict the setting that we consider
by assuming that (1) the ambient symplectic manifold admits a Lagrangian torus bration
all of whose bres are smooth and bound no holomorphic discs, and (2) one can choose an
almost complex structure for each Lagrangian so that it bounds no holomorphic discs. The
requirement that the Lagrangians bound no holomorphic disc is really only technical, and
meant to avoid discussing foundations of multivalued perturbations in Lagrangian Floer
theory [10] (and multiplying the length of the paper by a potentially large factor). The
reader may consult the introduction to [2] for a discussion of the more serious diculties
one would encounter in the presence of singular bres.
Since the construction of the homotopy from the composition to the identity uses a
moduli space of annuli, faithfulness can be seen as the analogue of the generation criterion
[1]. Heuristically, the strategy for the proof is the following: let X be a symplectic manifold
equipped with a Lagrangian torus bration over a base Q (we denote the bre over q Q
by F
q
), and L a Lagrangian in X. Consider moduli spaces of holomorphic discs with 3
marked points, on which we impose Lagrangian boundary conditions given by L and a bre.
We shall consider two avours for this moduli space (see the leftmost diagram in Figure
1): in the rst case, one marked point is distinguished as an input mapping to L, and the
remaining two are outputs mapping to intersection points x and y of L with a bre, while in
the second case, the intersections of L with a bre correspond to inputs, while the marked
point on L is an output.
In the classical versions of Floer theory, one would consider the subcategory of the Fukaya
category of X whose objects are bres, and the Yoneda module over this subcategory as-
sociated to L. Upon adding many more marked points, the rst of these moduli spaces
denes the map from the Floer cohomology L to the endomorphism algebra of this Yoneda
module, and the second moduli space denes a map which one could hope to show is a right
inverse by gluing the two triangles to an annulus, and degenerating this annulus to two discs
meeting at an interior point; one of the discs has Lagrangian boundary conditions on an
arbitrary bre and the other has Lagrangian boundary conditions on L and carries the two
boundary marked points. Since the moduli space of discs with boundary on an arbitrary
bre gives us a copy of the ambient space X, the rst type of disc imposes no constraint, so
we are simply considering the moduli space of discs with boundary on L (and two marked
point). This moduli space represents the identity on Floer cohomology.
Trying to implement this strategy in this setting runs into a convergence problem: since
the bres are disjoint, they are Floer theoretically orthogonal, so the Yoneda module dened
by L is a direct product of the corresponding modules for all bres. The map back to
Floer theory is not well-dened because it is the sum of innitely many terms. The correct
THE FAMILY FLOER FUNCTOR IS FAITHFUL 3
framework for this argument is in fact family Floer cohomology, and the main diculty that
arises is due to the need to make compatible families of perturbation in dening the Floer
cohomology of L with every bre; in the classical case, one can choose such perturbations
independently for all pairs of objects.
We end this introduction with a brief outline of the paper: Sections 2 and 3 introduce
the abstract moduli spaces and the corresponding spaces of maps which will serve to con-
struct the mirror to a Lagrangian and the maps back and forth from the Floer complex to
the morphism spaces on the B-side. Familiarity with the ideas explained in [2] may help
understand what are otherwise likely to appear as unmotivated constructions.
Section 4 is the heart of the paper, ensuring the compatibility of the Floer theoretic
constructions which yield the maps from the A-side to the B-side and vice versa; in the family
Floer setting, this is particularly delicate because the Lagrangian boundary conditions are
varying over the bres, and one cannot choose the data arbitrarily for dierent bres. The
key idea it to choose a very ne triangulation of the base of the bration, make controlled
choices at the vertices of this triangulation, and associate to higher dimensional cells families
of equations which interpolate between these.
Sections 5 and 6 show that the moduli spaces constructed in the previous section can
be glued to obtain a moduli space of annuli parametrised by the base. The main delicate
point is that it is not possible to perform this construction in such a way that the annuli
over every point in the base are obtained by gluing the degenerate annuli corresponding to
that point. This is responsible for the notion of an annulus gluing function introduced in
Denition 6.1.
Sections 7 and 8 are simply a matter of bookkeeping, as all the hard work appears in
the previous four sections. The results of [2], to which the reader is referred, are repeatedly
used. Finally, in the Appendix, the A

functor is constructed. This entails giving map


at the linear level between morphisms on the A-side and the B-side, whose construction is
quite a bit simpler than the construction in the main part of the paper. The additional
complexity of the papers main construction comes from the need to see such a map arise
in the boundary of a moduli space of annuli.
Acknowledgments. Discussions with Denis Auroux, Kenji Fukaya, Paul Seidel, and Ivan
Smith, exploring potential applications of family Floer cohomology, were helpful in justifying
the development of these techniques.
The author was supported by NSF grant DMS-1308179.
2. Families of Riemann surfaces
2.1. Adamss universal curve.
2.1.1. Adamss family of paths. In [6], Adams constructed a family of paths in the d-simplex
from the initial to the terminal vertex which are parametrised by the d 1-cube; see Figure
2. To give an explicit formula for this family, let A
d
= [0, 1)
d1
and A
d
= [0, 1]
d1
, introduce
the polytope
(2.1) T
d
=
_
(r
1
, . . . , r
d1
, s) A
d
R[0 s

r
i
+ 1
_
4 M. ABOUZAID
0 1
2
Figure 2.
and denote by s
j
the dierence s

j1
i=1
r
j
. One model for the Adams map is:
a: T
d

d
R
d+1
(2.2)
a(r
1
, . . . , r
d1
)(s) =
_

_
(1 s
1
, s
1
, . . . , 0) 0 s
1
r
1
((1 s
2
)(1 r
1
), (1 s
2
)r
1
, s
2
, . . . , 0) 0 s
2
r
2

((1 s
d
)

d1
i=1
(1 r
i
), (1 s
d
)

d1
i=2
(1 r
i
)r
1
,

d1
i=3
(1 r
i
)r
2
, . . . , (1 s
d
)r
d1
, s
d
) 0 s
d
1.
(2.3)
Let T
d
denote the complement in T
d
of the inverse image under a of the vertices of
d
;
i.e. remove the locus where s = 0 or some coordinate s
j
equals 1. Denote by T
r
the ber
over a point r A
d
of the projection map T
d
A
d
. This is a union of intervals, numbering
one more than the coordinates in r which equal 1.
Some compatibility between the spaces T
d
for dierent values of d will be required. To
this end, let K be a nite ordered set, and denote by
K
the associated [K[ 1-dimensional
simplex which naturally embeds in R
K
.
Denition 2.1. The compactied Adams moduli space of paths A
K
is the cube on K
min K, max K.
The associated partially ordered set consists of pairs of subsets I and J of K such that
(2.4) min K, max K I J
The partial ordering is such that the pair I J precedes I

whenever I I

J.
Consider also the open subset A
K
A
K
corresponding to the inclusion [0, 1) [0, 1].
There is an alternate description of the boundary strata which is often more useful:
introduce the notation
(2.5) K

i
= j K[j i and K

i
= j K[i i.
Lemma 2.2. The boundary stratum corresponding to I J admits a natural product de-
composition
(2.6) A
IJ

= A
J

i
d
A
J

i
d
J

i
d1
A
J

i
1
J

i
0
A
J

i
0
,
where I = min K, i
0
, i
1
, . . . , i
d
, max K.
THE FAMILY FLOER FUNCTOR IS FAITHFUL 5
Given i K max K, min K, the above result for facets yields a natural identication
A
K\{i}

= A
{minK,max K}K\{i}
(2.7)
A
K

i
A
K

= A
{minK,i,max K}K
(2.8)
corresponding to the locus where the i
th
coordinate vanishes or equals 1. The union of the
images of these inclusions over all i K max K, min K covers the boundary of A
K
.
Let T
K
denote the copy of the space T
d
over A
K
. The bre over each point in A
K
is
an interval, while those over its complement are disjoint unions of intervals. The inclusion
maps in Equations (2.7) and (2.8) induce natural inclusions
T
K\{i}
T
K
(2.9)
T
K

i
A
K

A
K

i
T
K

i
T
K
. (2.10)
Remark 2.3. The case [K[ = 1 is a degenerate case of the above discussion. Whenever it
appears, set A
K
to be a point, and T
K
to be empty.
2.1.2. From paths to strips. Consider the projection map
(2.11) f
IJ
: A
K

= [0, 1]
K\{minK,maxK}
[0, 1]
J\I

= A
IJ
.
Expressing the boundary stratum A
IJ
as a product of Adams moduli spaces yields a
projection A
IJ
A
I

J
whenever I I

J, so that the following diagram


commutes:
(2.12) A
K

A
IJ

A
I

J
.
Fix an order-preserving identication of the interval (0, 1) with the positive real numbers
and consider the projection
(2.13) g
I
: A
K

= [0, +]
K\{minK,max K}
[0, ]
I\{minK,max K}
for each subset I of K; the image of a point will yield a parameter used the glue the bres
of T
K
over A
IJ
to obtain bres over points in a neighbourhood of this stratum in A
K
.
Whenever I

I, these t in a commuting diagram


(2.14) A
K

(0, ]
I

\{min K,max K}

(0, ]
I\{minK,max K}
.
Given a union T of d+1 copies of the real line, and a sequence of d numbers

S = (S
1
, . . . , S
d
)
in [0, ], introduce the equivalence relation which identies points in the subintervals
[S
i
, +) and (, S
i
] of the copies of R labelled by i and the element succeeding i.
The quotient of T by this equivalence relation will be denoted T#

S. It is often convenient
to think of T#S as the union of intervals
(2.15)
_
(, S
1
]

[S
1
, S
2
]

[S
d
, +)
_
/
in which successive boundary points are identied.
6 M. ABOUZAID
0202 012012
0
2
0
2
0
1
2
Figure 3. The moduli space |
012
over A
012
.
Denition 2.4. If R < S
i
for all i, the R-thick part of T#

S is the image of
(2.16) [R, R]

[R, R]

[R, R].
Inductively choose identication of the bre of T
K
over a point r with unions of real
lines
(2.17) R

R T
r
and neighbourhoods of the boundary strata
(2.18) A
IJ
A
K
,
together with a gluing map T
fIJr
#g
I
(r) T
r
whenever r lies in A
IJ
. The existence
of such a gluing map yields a map
(2.19) f

IJ
T
K
T
K
[A
IJ
.
Since these spaces of choices are contractible, the commutativity of Diagrams (2.12) and
(2.14) implies that such neighbourhoods and identications may be chosen so that there
are commuting identications of bres whenever I I

J in the following sense:


the product decomposition of the boundary strata of A
K
yields a map g
I

\I
from A
IJ
to
(0, ]
I

\I
which ts in a commutative diagram of dieomorphisms
(2.20) T
f
I

J
r
#g
I
(r)

_
T
f
I

J
fIJr
#g
I

\I
(f
IJ
r)
_
#g
I
(r)

T
r
.
Denition 2.5. The universal curve over A
K
is the map
(2.21) |
K
T
K
[0, 1] A
K
.
The bres |
r
are equipped via the choice in Equation (2.17) with identications
(2.22) |
r

= R [0, 1]

R [0, 1] . . . R [0, 1],


hence with a complex structure which in the right hand coordinates (s, t) is j
s
=
t
on
each component. Denote by
(2.23) |
{0}
K
|
K
|
{1}
K
THE FAMILY FLOER FUNCTOR IS FAITHFUL 7
the two copies of T
K
embedded in |
K
as the product of the real line with the boundary of
the interval. Denote by B

B = R [0, 1] the half-strips:


(2.24) B
+
= [0, +) [0, 1] and B

= (, 0] [0, 1],
and x maps with disjoint images

min(K)
: A
K
B
+
|
K
(2.25)

max(K)
: A
K
B

|
K
(2.26)
whose restrictions to r A
K
give positive (resp. negative) strip like ends on the bre
|
r
, which agree up to translation with the natural ones coming from the last (respectively
rst) factor in Equation (2.22), and which are compatible with the gluing maps near every
boundary stratum.
Let F : |
J
Z be a map from |
J
to a topological space Z. Such a map is constant
along the positive (respectively negative) end if there is a map f : [0, 1] Z such that
the restriction of F to each bre |
r
agrees with f near s = + on the last component
(respectively near on the rst component).
The choices of gluing maps for the moduli space T
K
yield a map
(2.27) G
IJ
: f

IJ
|
K
|
K
[A
IJ
for all pairs I J labelling boundary strata of A
K
.
Denition 2.6. The map F is obtained by gluing if its restriction to neighbourhoods of all
boundary strata yields a commutative diagram
(2.28) f

IJ
|
K

|
K
[A
IJ

|
K
[A
IJ

Z.
Since the gluing map G
IJ
is surjective, F is determined, on a neighbourhood of a
boundary stratum, by its restriction to the stratum and the gluing map. In addition,
continuity implies that its restriction to the boundary is constant along each glued end.
The notion of an R-thick part of the image of G
IJ
is inherited from Denition 2.4: this
consists of [I[ 1 components, each of which is labelled by a pair of successive elements of
I, and is naturally identied with [R, R] [0, 1]. Let F
IJ
: A
IJ
Z be a function
obtained by gluing, and assume Z is a Frechet manifold. We shall say that a section of
F

IJ
TZ is consistent if there is a constant R so that the support is contained in the interior
of the R-thick part and the restriction to the i
th
component of the R-thick part vanishes
to innite order along the strata for which the corresponding pair of gluing parameters are
innite. A consistent perturbation of F
IJ
is the image under exponentiation of a consistent
section.
Denition 2.7. A map F : |
K
Z is obtained by perturbed gluing if its restriction to
A
IJ
for all pairs I J agrees with a consistent perturbation of a function obtained by
gluing.
8 M. ABOUZAID
2.2. Adams paths in a prism. Equip K , + with the total ordering obtained by
extending the ordering on K via (+, i) < (, j) for all i, j K. Write every subset of
K , + as a union K

K
+
+. The constructions of the previous section,
applied to this ordered set, yields spaces and maps
(2.29) |
K,K+

T
K,K+
a

K,K+
A
K,K+
,
and an open subset A
K,K+
A
K,K+
over which the bres of |
K,K+
are connected.
There are inclusions of the boundaries of each bre |
{i}
K,K+
|
K,K+
for i = 0, 1.
The additional data of the decomposition yields a natural map
(2.30)
K,K+
[0, 1].
The bre of T
K,K+
over A
K,K+
is therefore equipped with a distinguished component
characterised by the property that its projection to [0, 1] is not constant. To state the
compatibility of this distinguished component with the boundary decomposition of A
K,K+
,
note that such a stratum is labelled by pairs I

and J

such that
min K
+
I
+
J
+
K
+
(2.31)
max K

. (2.32)
Whenever I

are singletons (i.e. respectively consist only of min K


+
and max K

), the
corresponding boundary stratum is naturally identied with the moduli space A
J,J+
.
Lemma 2.8. The codimension 1 boundary strata of A
K,K+
are:

iK\max K
A
K\i,K+
(2.33)

iK+\minK+
A
K,K+\i
(2.34)

iK\max K
A
K

,i
A
K

,i
,K+
(2.35)

iK+\minK+
A
K,K

+,i
A
K

+,i
(2.36)
The restriction of |
K,K+
to these strata is naturally isomorphic to the union of pullbacks
of the universal curves on each factor.

Using the above description of the boundary strata, inductively choose identications of
bres with unions of real lines, compatibly with the choices made in Section 2.1 for |
J
for
a subset J of K; note that these moduli spaces appear in the bres over Equations (2.35)
and (2.36), and the compatibility condition is that the restriction of the identication to
such a boundary stratum agrees with the one previously made. Choose families of positive
(respectively negative) strip like ends

: A
K,K+
B

|
K,K+
(2.37)
THE FAMILY FLOER FUNCTOR IS FAITHFUL 9
whose restriction to the boundary strata are compatible with the inductive choices, and
those made in Section 2.1.
Finally, in the choices of identication made in Equation (2.17) as part of the gluing
procedure, extend the coordinates of the component distinguished by the map in Equation
(2.30), and consider the resulting an open embedding
(2.38) : A
K,K+
B |
K,K+
which is a dieomorphism over A
K,K+
. On each bre of |
K,K+
, let z
1
be the image of
(0, 1) under .
Assume now that max K
+
min K

. Let z
in
denote the boundary marked point z
1
. For
each r
K,K+
x the following positive strip like ends near z
in

in
: B
+
B |
r
(2.39)
(s, t)

1 e
st

1
, (2.40)
where the complex coordinates on B are given by its embedding in C. By construction,
these strip-like ends are compatible with gluing. Since |
{1}
K,K+
is naturally ordered via its
identication with a union of real line, the points preceding or succeeding z
in
dene subsets
of the boundary:
(2.41) |
zin<
K,K+
|
{1}
K,K+
|
zin>
K,K+
.
Assuming that max K

min K
+
repeat the same procedure to obtain a marked point
z
ou
on each bre of |
K,K+
, and pick negative strip-like ends

ou
: B

B |
r
(2.42)
whenever r
K,K+
which are compatible with gluing. The points preceding or succeeding
z
ou
yield subsets |
zou<
K,K+
|
{1}
K,K+
|
zou>
K,K+
.
Given I K, introduce the notation
(2.43) K

I
K

maxI
and K

I
K

minI
.
where the sets K

i
and K

i
are as in Equation (2.5). Given a nested pair I K, dene the
following subsets of K +, :
K
in
I
(K

I
, K

I
+) (2.44)
K
ou
I
(K

I
, K

I
+). (2.45)
Lemma 2.9. The minimal element of K
in
I
is (min K, +), and the maximal element is
(max K, ). The minimal element of K
ou
I
is (max I, +) and the maximal element is (min I, ).

2.3. Strips with one input marked point. Let



K be a nested sequence of totally ordered
sets whose maximal element is K and minimal element is K
0
.
Denition 2.10. The Adams moduli space with one input A

K;in
is the cube
(2.46) A

K;in
[0, 1]

K\K
.
The cells of A

K;in
are given by pairs of subsets K

I

J

K, and write A

J;in
for the corresponding stratum. Dene the open subset A

K;in
A

K;in
corresponding to the
inclusion [0, 1) [0, 1]. It is the union of the strata for which K =

I.
10 M. ABOUZAID
max K max J min J min K
Figure 4. A bre of the universal curve over A

J
;in
A

J
.
To dene the universal curve over this space, let min

K K denote the set of minimal
elements of subsets of K lying in

K, and max

K the set of maximal elements. Assign to a
sequence

J

K a subset of K
in
K0
(2.47)
in

J (max

J , min

J +).
The map of posets

I

J
in

I
in

J induces a natural map


(2.48)
in
: A

K;in
A
K
in
K
0
.
Denition 2.11. The universal curve over A

K;in
is the projection map
(2.49) |

K;in

in
(|
K
in
K
0
) A

K,in
.
Similarly dene T

K;in
. By construction, these spaces are equipped with natural maps
(2.50) |

K;in

T

K;in
a

K;in
A

K;in
,
where, for the sake of consistency, the notation

K;in

K
in
K
0
is used.
By construction, the following diagram commutes:
(2.51) A

J;in

K;in

A
in

Iin

J

A
K
in
K
0
To describe the restriction of |

K;in
to the boundary strata, it is convenient to introduce
more notation: considering

K as a totally ordered subset (with respect to inclusion), the
natural maps from

K to min K and max K respectively reverse and preserve ordering. This
denes maps min and max from A

K
to A
K
, and pullback along these maps yields moduli
spaces over A

K
(2.52) min

|
min

K

max

|
max

K
.
From the choices made in Section 2.1, the bres are identied with unions of strips. Fix
the induced family of positive strip-like ends
+
with values in max

|
max

K
, and negative
strip-like ends

with values in min

|
min

K
.
Given an element J

K, write

K

J
and

K

J
as before for the elements preceding and
succeeding J. The boundary stratum of A

K;in
labelled by K

I

J is
(2.53) A

J;in

= A

I
0
A

I
0
;in
,
THE FAMILY FLOER FUNCTOR IS FAITHFUL 11
where the rst factor is the cube on

J

I0

I, and maps by min and max to


A
min

Imin

J

I
0
A
K

I
0
(2.54)
A
max

Imax

J

I
0
A
K

I
0
, (2.55)
where the fact that K

J is used. The second factor in Equation (2.53) maps by
in
to
(2.56) A
max

J

I
0
,min

J

I
0
A
I0

J
0
,I0

J
0
A
I0
in
J
0
.
The product of the right hand sides in Equations (2.54)-(2.56) is a stratum of A
K
in
K
0
, and
there is a commutative diagram
(2.57) A

J;in

= A

I
0
A

I
0
;in

K;in

A
K

I
0
A
I0
in
J
0
A
K

I
0

A
K
in
K
0
The pullback of the universal curves on the three factors of the space at the bottom left is
then naturally isomorphic to the restriction of |

K;in
.
Equations (2.33)-(2.36) yield a more explicit description of the codimension 1 strata:
Lemma 2.12. The boundary of A

K;in
is covered by the following codimension 1 strata:
_
I

K\{K}
A

K\{I};in
(2.58)
_
I

K\{K}
A

I
A

I
;in
(2.59)
The restriction of |

K;in
to the rst stratum is naturally isomorphic to |

K\{I};in
, and the
restriction to the second is given by
_
max

|
max

K

min

|
min

K

I
_
A

I
;in

I
|

I
;in
. (2.60)
Proof. Equation (2.58) corresponds the case when the coordinate labelled by I vanishes,
and the other type of boundary stratum to the case this coordinate equals 1. In Equation
(2.60), the projection from K
in
K0
to K can be used in order to identify max

K

I
, as a
subset of K
in
K0
with max

K

I
, and similarly for min

K

I
.
Pulling back Equations (2.37)-(2.38), denes strip-like ends

: A

K,in
B

K,in
(2.61)
as well as an open embedding
(2.62) : A

K,in
B |

K,in
whose restriction to each point in A

K,in
gives a distinguished component in the correspond-
ing bre.
Dene subsets |
{i}

K;in
|

K;in
for i = 0, 1, and a decomposition of |
{1}

K;in
into |
zin<

K;in

|
{1}

K;in
|
zin>

K;in
corresponding to the points preceding and succeeding z
in
.
12 M. ABOUZAID
0 0
0 0 0 2 0
0 2 1 0 0 2 1 0
Figure 5. The moduli space A

K;ou
, with vertices labelled by bres of
|

K;ou
(

K = 0 02 012).
2.4. Strips with one output marked point. The construction of moduli spaces with
outputs is entirely analogous to that of the moduli space with inputs, replacing the partially
ordered set K
in
K0
by K
ou
K0
in the construction, and a few other minor changes.
Denition 2.13. The Adams moduli space with one output, A

K;ou
is the cube
(2.63) A

K;ou
[0, 1]

K\K0
.
The cells of A

K;ou
are given by pairs of subsets K
0


I

J

K, and the union of those
strata for which

I = K
0
correspond to the open subset A

K;ou
.
There is a natural map
(2.64)
ou
: A

K;ou
A
K
ou
K
0
which assigns to a sequence

J

K the pair (min

J , max

J +).
Denition 2.14. The universal curve over A

K;ou
is the projection map
(2.65) |

K;ou

ou
(|
K
ou
K
0
) A

K,ou
.
Similarly dene T

K;ou
. By construction, these spaces are equipped with natural maps
(2.66) |

K;ou

T

K;ou
a

K;ou

K
ou
K
0
A

K;ou
.
By pullback, Equation (2.37) determines families of strip like ends

and
+
, and a family
of distinguished components which are trivialised by a map we still denote .
THE FAMILY FLOER FUNCTOR IS FAITHFUL 13
As in Equation (2.57), the boundary strata of A

K;ou
and their images under
ou
t in a
commutative diagram:
(2.67) A

J;ou

= A

I
;ou
A

K;ou

A
I

K
0
A
J
ou
I
A
I

K
0

A
K
ou
K
0
.
For codimension 1 strata, the boundary of A

K;ou
admits a natural decomposition as a
union
_
I

K\K0
A

K\I;ou
(2.68)
_
I

K\K0
A

I
;ou
A

I
, (2.69)
and the bres of |

K;ou
are given by |

K\I;ou
and
A

I
;ou

_
max

|
max

K

min

|
min

K

I
_

I
;ou
A

I
. (2.70)
Pullback denes strip like ends

, an embedding
ou
of a strip into a distinguished
component of each bre, and subsets |
{i}

K;ou
|

K;ou
for i = 0, 1, corresponding to the
boundaries of each bre over A

K;ou
. These bres are equipped with a marked point z
ou
, and
the ordering on |
{1}

K;ou
yields a decomposition |
zou<
K;in
|
{1}
K;ou
|
zout>
K;ou
which corresponds
to points preceding or succeeding the marked point z
ou
.
3. Lagrangian Floer theory
3.1. Geometric assumptions. Let (X, ) be a closed symplectic manifold, and : X Q
a Lagrangian bration. Let H and D be the space of Hamiltonian dieomorphisms and all
dieomorphisms of X. Let J denote the space of -tame almost complex structures on X.
Assume that
2
(Q) = 0 which excludes the presence of holomorphic spheres in X. For any
J J, there is a natural isomorphism of vector bundles TX

=

(TQ)
R
C. In particular,
there is a natural quadratic complex volume form on TQ.
Assume that all Lagrangians are tautologically unobstructed in the sense that
(3.1) there exists J
L
J so that L bounds no J
L
-holomorphic discs.
and are graded with respect to , i.e. the map L RP
1
induced by is null-homotopic,
and that a lift of this null-homotopy to the universal cover of RP
1
is xed.
Whenever L and L

are both graded Lagrangians, and x L L

is a transverse inter-
section point, there is a well-dened Maslov index deg(x) Z as explained in [15, Section
(12b)].
3.2. The Floer equation. Given a point q Q and a Lagrangian L, pick a Hamiltonian
dieomorphism so that L which is transverse to F
q
. A family J = J
t
J
t[0,1]
so that
(3.2) J
1
=

(J
L
)
14 M. ABOUZAID
yields a holomorphic curve equation on B with Lagrangian boundary conditions
u: B X
s
u = J
t

t
u (3.3)
u(s, 0) F
q
u(s, 1) L. (3.4)
Given a pair of points (x, y) L F
q
, denote by /
q
(x, y) the Gromov-Floer compact-
ication of the quotient by translation in the R factor of B of the space of solutions to
Equations (3.3) and (3.4) which in addition satisfy the asymptotic conditions:
(3.5) lim
s
u(s, t) = x lim
s+
u(s, t) = y.
For generic J, /
q
(x, y) is a manifold with boundary such that
(3.6) dim
R
/
q
(x, y) = deg(x) deg(y) 1,
and the codimension 1 strata of the boundary are given by the union
(3.7)

LFq
/
q
(x, z) /
q
(z, y).
If L and L

are graded Lagrangians which are transverse, and which both satisfy Condition
(3.1), pick a family J
t
of almost complex structures such that J
0
= J
L
and J
1
= J
L
, and
which is constant in a neighbourhood of the point t = 1/2. To each pair x, y LL

, there
corresponds a moduli space /(x, y) with boundary conditions given by L along R 0
and L

along R 1. Equation (3.6) holds in this case as does the analogue of Equation
(3.7). Finally, dene /(x, y) to be the compactied moduli space of solutions to the Floer
equation with one interior marked point lying on the interval R1/2. This moduli space
can be thought of as the continuation moduli space for a constant Hamiltonian family (see
the next Section), and is the product of a closed interval with /(x, y), unless x = y, in
which it case it is a point, corresponding to the constant solution.
3.3. Continuation maps. Let K be a totally ordered set.
Denition 3.1. A consistent family of continuation data parametrised by |
K
is a map
(3.8)
K
= (
K
, J
K
,
K
): |
K
HJ D
such that (i)
K
is constant along each end of a bre of |
K
, (ii) the maps
K
and
K
, are
obtained by gluing, and J
K
by perturbed gluing, and (iii) for each z |
{1}
K
(3.9) J
K
(z) = (
K
(z)
K
(z))

J
L
.
Remark 3.2. Only the restrictions of
K
and
K
to |
{1}
K
shall be used, so the reader may
assume for simplicity that these maps (but not J) factor via the projection |
K
T
K
.
From such consistent families, one can dene, for each r A
K
, a holomorphic curve
equation with moving Lagrangian boundary conditions:
u: |
r
X
s
u(z) = J
K
(z)
t
u(z) (3.10)
u(z) F
q
if z |
{0}
r
u(z)
K
(z)L if z |
{1}
r
. (3.11)
Given a pair of points x L
minK
F
q
and y L
maxK
F
q
, denote by
(3.12) /
K
(x, y) A
K
THE FAMILY FLOER FUNCTOR IS FAITHFUL 15
the moduli space of solutions to Equations (3.10)-(3.11), which is addition satisfy the as-
ymptotic conditions:
(3.13) lim
s
u

(s, t) = x lim
s+
u
+
(s, t) = y.
For a generic family J
K
, the Gromov-Floer compactication /
q,K
(x, y) is a manifold
with boundary such that
dim
R
/
q,K
(x, y) = deg(y) deg(x) + dim(A
K
) (3.14)
= deg(y) deg(x) +[K[ 2. (3.15)
The generalisation of Equation (3.7) is that the codimension 1 strata of the boundary come
in two families:

iK

zLiFq
/
q,K

i
(x, z) /
q,K

i
(z, y) (3.16)

iK\{minK,max K}
/
q,K\{i}
(x, y). (3.17)
Breaking of Floer strips is incorporated in the rst case, corresponding to i = max K or
i = min K. So the rst strata project to the interior of A
K
or to the boundary stratum
where the cube coordinate corresponding to i vanishes, and the second to the case where
this coordinate equals 1.
Remark 3.3. Since the choice of data in Denition 3.1 takes place for a xed K, there is a
slight abuse of notation in Equations (3.16) and (3.17), since the moduli spaces that appear
are dened with respect to the Floer data restricted from |
K
. This issue will be addressed
in Section 4, where such data will be chosen inductively so that there is no ambiguity in the
description of the boundary strata.
3.4. Continuation in a prism. Let L
+
and L

be graded Lagrangians which are trans-


verse, and which both satisfy Condition (3.1) for almost complex structures J
L
+ and J
L
.
Let K

, K
+
be subsets of K , + as in Section 2.2.
Denition 3.4. A consistent family of continuation data in a prism parametrised by |
K
is a map
(3.18)
K,K+
= (
K,K+
, J
K,K+
,
K,K+
): |
K,K+
HJ D
such that (i) the restriction to every end is constant, (ii) J
K,K+
is obtained by perturbed
gluing and the other maps by gluing in a neighbourhood of each boundary stratum, and
(3.19) J
K,K+
(z) =
_
_

K,K+
(z)
K,K+
(z)
_

J
L
+ if z |
z1<
K,K+
_

K,K+
(z)
K,K+
(z)
_

J
L
if z |
z1>
K,K+
.
We obtain, for each r A
K,K+
, a holomorphic curve equation
s
u(z) = J(z)
t
u(z)
with moving Lagrangian boundary conditions
u: |
r
X u(z) F
q
if z |
{0}
r
(3.20)
u(z) (z)L

if z |
z1>
r
u(z) (z)L
+
if z |
z1<
r
. (3.21)
16 M. ABOUZAID
max K max J min J
L

L
min K
Figure 6. A schematic picture of the holomorphic curve problem for curves
over A

J
;in
A

J
.
3.5. Continuation with one input. Let K

and K
+
be subsets of such that max K
+

min K

, and let L and L

be graded Lagrangians satisfying Condition (3.1) for almost


complex structure J
L
and J
L
. In the setting of the previous section, set L
+
= L and
L

= L

.
Given points x

maxK
F
q
, x L
minK
F
q
and x
in
L L

, dene
(3.22) /
q,K,K+
(x

; x
in
, x) A
K,K+
to be the moduli space of solutions to Equations (3.20)-(3.21), which in addition satisfy the
conditions:
(3.23) lim
s
u

(s, t) = x

u(z
in
) = x
in
lim
s+
u
+
(s, t) = x.
For a generic family of parametrised Floer data, the Gromov-Floer compactication
/
q,K,K+
(x

; x
in
, x) is a manifold with boundary such that
dim
R
/
q,K,K+
(x

; x
in
, x) = deg(x

) deg(x
in
) deg(x) +[K

[ +[K
+
[ 2. (3.24)
If

K is a nested sequence of subsets of K, with minimal element K
0
, let
(3.25) /
q,

K;in
(x

; x
in
, x) A

K;in
be the pullback of /
q,K
in
K
0
(x

; x
in
, x) by the map from A

K;in
to A
K
in
K
0
(see Equation (2.44)).
For generic Floer data, this is a manifold with boundary such that
dim
R
/
q,

K;in
(x

; x
in
, x) = deg(x

) deg(x
in
) deg(x) +[

K[ 1. (3.26)
The boundary decomposition of A

K;in
leads to the following decomposition of the bound-
ary of /
q,

K;in
(x

; x
in
, x)

max K
Fq
/
q
(x

, y

) /
q,

K;in
(y

; x
in
, x) (3.27)

yLminKFq
/
q,

K;in
(x

; x
in
, y) /
q
(y, x) (3.28)

yinLL

/
q,

K;in
(x

; y
in
, x) /(y
in
, x
in
) (3.29)

K\K
/
q,

K\I;in
(x

; x
in
, x) (3.30)

K\K

yLminIFq

max I
Fq
max

/
q,max

K

I
(x

, y

)
A

I
min

/
q,min

K

I
(y, x) (3.31)
/
q,

I
;in
(y

; x
in
, y).
THE FAMILY FLOER FUNCTOR IS FAITHFUL 17
Lemma 3.5. Assume that

K consists of [K[ elements, and that dim
R
/
q,

K;in
(x

; x
in
, x) =
1. The only contributions to the stratum in Equation (3.31) are given by I such that (i)
max I = max K and min is injective on

K

I
, in which case the boundary contribution is:
(3.32)

yLminIFq
/
q,

I
;in
(x

; x
in
, y) /
q,min

K

I
(y, x).
or (ii) min I = min K and max is injective on

K

I
:
(3.33)

max I
Fq
/
q,max

K

I
(x

, y

) /
q,

I
;in
(y

; x
in
, x),
or (iii) I = K i, for i ,= min K, max K, in which case the corresponding stratum is
(3.34) /
q,

I
;in
(x

; x
in
, x).
Proof. Under the assumption on

K, successive elements of this sequence dier by exactly
one element of K, yielding a bijection between

K

I
and K/I. In this case, the map
(3.35) max min: A

I
A
min K
A
max K
is submersive in the interior unless all element of

K

I
have the same maximum or mini-
mum. This implies that the bre product in Equation (3.31) necessarily has non-vanishing
dimension unless either max I = max K or min K = min I, and the map on the other factor
is injective.
To arrive at Equations (3.33) and (3.32), we note that, whenever max

K

I
or min

K

I
consist of a single element, the corresponding rigid moduli spaces in Equation (3.31) consist
only of constant solutions, hence these factors are omitted. In the case where both max I =
max K and min K = min I, A
minK
A
max K
is a point, so injectivity implies that A

I
consists of two elements, hence that I is the penultimate element of

K, leading to the
stratum in Equation (3.34).
3.6. Continuation with one output. Let K

and K
+
be subsets of K such that min K
+

max K

. Set L

= L and L
+
= L

.
Given points x L
max K
F
q
, x

minK+
F
q
, and x
ou
L L

, denote by
(3.36) /
q,K,K+
(x, x
ou
; x

) A
K,K+
the compactied moduli space of solutions to the Cauchy Riemann equation determined by
such data, with asymptotic conditions
(3.37) lim
s
u

(s, t) = x u(z
ou
) = x
ou
lim
s+
u
+
(s, t) = x

.
If

K is a sequence of nested subsets with minimal element K
0
and maximal element K,
let /
q,

K;ou
(x, x
ou
; x

) be the pullback of /
q,K
ou
K
0
(x, x
ou
; x

) under the projection map from


A

K;ou
to A
K
ou
K
0
. This is a manifold with boundary such that
dim
R
/
q,

K;ou
(x, x
ou
; x

) = deg(x) + deg(x
ou
) deg(x

) +[

K[ n 1. (3.38)
The boundary decomposition of |

K;ou
leads to the following decomposition of the boundary
18 M. ABOUZAID
min K
0 min I max I
L L

max K
0
Figure 7. A schematic picture of the holomorphic curve problem for curves
over A

I
;ou
A

I
.
of /
q,

K;ou
(x, x
ou
; x

max K
0
Fq
/
q,

K;ou
(x, x
ou
; y

) /
q
(y

, x

) (3.39)

yLminK
0
Fq
/
q
(x, y) /
q,

K;ou
(y, x
ou
; x

) (3.40)

youLL

/(x
ou
, y
ou
) /
q,

K;ou
(x, y
ou
; x

) (3.41)

K\K0
/
q,

K\I;ou
(x, x
ou
; x

) (3.42)

K\K0

yLminIFq

max I
Fq
min

/
q,min

K

I
(y

, x

)
A

I
max

/
q,max

K

I
(x, y) (3.43)
/
q,

I
;ou
(y, x
ou
; y

).
We also have the analogue of Lemma 3.5:
Lemma 3.6. Assume that

K consists of [K[ elements, and that dim
R
/
q,

K;ou
(x, x
ou
; x

) =
0. The only contribution to the stratum in Equation (3.43) is given by I such that (i)
max I = max K
0
, and min is injective on

K

I
:
(3.44)

yLminI Fq
/
q,min

K

I
(x, y) /
q,

I
;ou
(y, x
ou
; x

),
or (ii) min I = min K
0
, and max is injective on

K

I
(3.45)

max I
Fq
/
q,

I
;ou
(x, x
ou
; y

) /
q,max

K

I
(y

, x

).
or (iii) I = K
0
i, for i ,= min K, max K, in which case the corresponding stratum is
(3.46) /
q,

I
;ou
(x, x
ou
; x

).

4. Floer theory and convergence


4.1. Uniformly small choices of perturbations. Let L and L

be Lagrangians which
are tautologically unobstructed with respect to a pair of tame almost complex structures
J
L
and J
L
. Fix an embedded path J
t

1
t=0
from J
L
to J
L
. Pick a sequence of contractible
neighbourhoods of this path
(4.1) J
0
J
1
J
2
n+2
so that the closure of J
i
lies in the interior of J
i+1
.
THE FAMILY FLOER FUNCTOR IS FAITHFUL 19
Let H
0
be a contractible neighbourhood of the identity in the space of Hamiltonian
dieomorphisms such that
(4.2)

(J
t
) =

J
t

1

J
0
for all H
0
. Let L and L

denote the families of Lagrangians obtained by applying


elements of H
0
to L and L

.
Fix a neighbourhood D of the identity in Di(X) such that D(J
i
) J
i+1
, where the
action is by pushforward. Given a subset W Q
2
, dene
(4.3) D
W
(L) W L D
to be the subset consisting of triples (p, q, L, ) such that
(F
q
) = F
p
(L) = L, (4.4)
and similarly for L

. Dropping the rst condition yields a family D(L) over L, and dropping
the second, a family D
W
over W. For a xed element of L, write
(4.5) D(L) = D[(L) = L.
For all q Q, pick
q
H
0
so that
q
L is transverse to F
q
. Since Q is compact, there is
a nite cover U of Q and maps
U
,

U
H
0
for all U U, so that
U
(L) and

U
(L

) are
transverse to F
q
if q U. Moreover, if the elements of U are suciently small, there is a
sequence of neighbourhoods
(4.6) Id = D
0
D
1
D
2
n+1
D
which are invariant under inversion so that
(4.7) D
i
D
j
D
k
and we have acyclic brations D
k
(L) L, D
k
(L

) L

, and
(4.8) D
k
U
2
(
U
L)

D
k
U
2

D
k
U
2
(
U
L

)
s
s
s
s
s
s
s
s
s
s
U
2
.
Pick a partially ordered set labelling the vertices of a simplicial triangulation of Q
which is suciently ne that, for each element i , there is an integral ane polygon P
i
containing the open star of the vertex q
i
Q, and contained in U U whenever q
i
U,
such that
(4.9) P
j
P
i
if i j.
For each i, pick
i
,

i
H
0
arbitrarily among those Hamiltonian dieomorphisms
U
and

U
where P
i
U. It is immediate that D
i
P
2
i
, D
i
P
2
i
(
i
L) and D
i
P
2
i
(
i
L

) are acyclic brations


over P
2
i
.
Fix a basepoint q
i
P
i
for each i , and write q
I
for q
maxI
. Choose generic families
J
i
, J

i
: [0, 1] J
1
so that
(4.10) J
i,0
= (
i
)

J
L
and J

i,0
= (

i
)

J
L

and all moduli spaces of Floer trajectories /


qi
(x, y) and /
qi
(x

, y

) dened with respect


to these families of almost complex structures are regular if x, y
i
(L) F
i
and x

, y

i
(L

) F
i
. We write
(4.11)
i
= (
i
, J
i
, Id): [0, 1] H
0
J D,
20 M. ABOUZAID
where the rst and last maps are constant, and similarly for

i
. Moreover, pick sections
(4.12)
i
: P
i
D
1
{qi}Pi
(
i
L).
Remark 4.1. In order not to increase the awkwardness of the notation, we shall adopt the
following conventions: let Z be a topological space. Given maps
(4.13) (, ): Z DHJ D,
with = (, J, ), use pushforward and composition (pointwise in Z) to dene
(4.14)

() (,

J, ): Z HJ D.
Also, write
1
for the pointwise inverse of .
4.2. Locally constant families of continuation maps. Pick a map
(4.15)
I
: P
I
|
I
D
2
|I|1
for each totally ordered subset I of , which is the identity on q
I
|
I
and continuation
data
(4.16)
I
= (
I
, J
I
,
I
): |
I
H
0
J
2
|I|1
D
2
|I|1
.
Remark 4.2. The remainder of this section is likely to be less incomprehensible if the reader
keeps in mind the holomorphic curve problem on the moduli space |
I
is dened with
Lagrangian boundary conditions F
I
over the boundary labelled 0. The map
I
is introduced
to transport this holomorphic curve problem to nearby bres.
In addition to the conditions imposed on
I
in Denition 3.1, assume that the restrictions
of (
I
,
I
) agree with
(
minI
(q
I
)

min I
,
minI

1
minI
(q
I
)) along the end
+
(4.17)
(
max I
,
max I
) along the end

(4.18)
(
I\i
,
I\i
) on |
I\i
(4.19)
(
I

i
,
I

i
) on |
I

i
A
I

i
(4.20)
_

i
(q
I
)

i
,
I

i

1
I

i
(q
I
)
_
on A
I

i
|
I

i
. (4.21)
Moreover, on the boundary of each bre,

I
(q, z) D
2
|I|1
(
I,z
(L)) if z |
{1}
I
(4.22)

I
(q, z) D
2
|I|1
qI ,q
if z |
{0}
I
. (4.23)
Lemma 4.3. There are choices of maps (
I
,
I
) which are obtained by perturbed gluing in
a neighbourhood of every boundary stratum so that Equations (4.17)-(4.23) are satised.
Proof. Proceed by induction on the number of elements of I. In the base case, [I[ = 2 so
A
I
is a point, |
I
= B, and |
{1}
I

= R . Pick a map
I
which agrees with
min I
and
max I
at the two ends. Then pick
I
agreeing with Id near the negative end and with
minI
(q
I
)
near the positive end, and so that
I
(
I
(z)L) =
I
(z)L for all z |
{1}
I
. Note that both
values at the endpoints lie in D
1
, so such a path may be chosen in D
1
by the assumption
that the forgetful map from D
k
(L) to L is an acylic bration. Also choose a family of
almost complex structures J
I
whose pullback under the negative end is J
max I
and under
the positive end is (
minI
(q
I
))

J
minI
, and which agrees with (
I

I
)

J along |
{1}
I
. Since
THE FAMILY FLOER FUNCTOR IS FAITHFUL 21
all these almost complex structures lie in J
1
, the image of J
I
may be required to also lie in
J
1
since this space was assumed to be contractible.
Pick a map
I
on P
I
|
I
subject to the condition that along the ends, Equations (4.17)
and (4.18) hold. These constraints imply that the image of
I
along the ends lies in D
2
;
extend it to a map from P
I
|
I
to D
2
, with the requirement that Equations (4.22) and
(4.23) hold along the two boundaries of the strip. This completes the base case.
Given an ordered subset K of , assume by induction that continuation data and families
of dieomorphisms for all subsets I of K which satisfy Equations (4.18)-(4.23) have been
chosen. Dene (
K
,
K
) along the boundary strata of |
K
using Equation (4.19)-(4.21).
The inductive hypothesis for (
I
,
I
) implies that this construction, which is a priori dened
only on each separate top-dimensional cell of the boundary of |
K
, restricts to the same map
on each codimension 2 cells, hence denes a map from the boundary of |
K
. Moreover, the
conditions imposed on the ends in Equations (4.18)-(4.23) show that the restriction of these
data to an end of a component of a bre in |
K
which is labelled by i K is given by
(
i
(q
K
))


i
).
One can therefore use gluing to extend these data from the boundary of |
K
to its interior
as follows: rst pick the extension
K
by gluing with image in H
0
. Then pick the extension

K
by gluing which preserves the Lagrangian boundary conditions, and whose restrictions
to the positive and negative ends are respectively given by
min K
(q
K
) and the identity. By
induction, the images of
I
and
I
both lie in D
2
|I|1
D
2
|K|2
, so such an extension may
be chosen to have image in D
2
|K|1
. The next step is to extend J
K
by perturbed gluing
subject to the condition that it agree with the pushforward of J under
K

K
on |
{1}
K
,
with the pushforward of J
minK
under
minK
(q
K
) along the positive end, and with J
max(K)
along the positive end. The same argument as above shows that the image of J
K
may be
chosen to lie in J
2
|K|1
.
Finally, by the inductive hypothesis, the image of
I
lies in D
2
|I|1
D
2
|K|2
. By
Equation (4.19)-(4.21) the map dened on the boundary strata of |
K
has image in D
2
|K|1
.
Since this space is contractible, choose an extension to a map
K
dened on |
K
. Construct
this map by gluing in a neighbourhood of the boundary strata, requiring in addition that
Equations (4.17) and (4.18) hold along the ends and Equations (4.22) and (4.23) along
the boundary; these properties can be achieved given the assumptions on the acyclicity of
D
2
|K|1
(L) and D
2
|K|1
P
2
K
. This completes the construction of the data (
I
,
I
) by induction
on [I[.
Denote the pushforward
K

K
by:

P
K
: P
K
|
K
HJ
2
|K|
D
2
|K|
(4.24)
The conditions imposed on
K
in the previous section imply that, for each q P
K
,

P
K
(q) is a compatible family of continuation data in the sense of Denition 3.1. For each
k K, the dieomorphism
k
(q) maps F
k
to F
q
, and preserves
k
L, hence maps a point
x
k
F
k

k
L to
(4.25) x
k
(q)
k
(q)x
k
F
q

k
L.
Lemma 4.4. Given y F
minK

min K
L and x F
max K

max K
L, composition with

K
(q) yields a homeomorphism
(4.26) /
qK,K
(x(q
K
), y(q
K
))

= /
q,K
(x(q), y(q)).

22 M. ABOUZAID
When there is no ambiguity, we shall write /
K
(x, y) for this moduli space. Fix generic
choices of continuation data
K
, for which all such moduli spaces are regular.
4.3. Locally constant continuation maps in a prism: input. Start by assuming that
data (
I
,
I
) and (

I
,

I
) for the Lagrangians L and L

have been chosen for all totally


ordered subsets I of . As in Section 3.5, assume that I

and I
+
are subsets of whose
union is I such that max I
+
min I

.
Remark 4.5. To make sense of this section, the reader should consult Figure 6. The holo-
morphic curve problem in the presence of an input along the boundary has Lagrangian
boundary conditions F
max I
.
Pick a continuation datum with inputs
I,I+
, and a family of dieomorphisms
I,I+
which agree with the identity on q
I
|
I,I+
and are obtained by gluing

I,I+
: |
I,I+
H
0
D
2
|I

|+|I
+
|1
J
2
|I

|+|I
+
|1
(4.27)

I,I+
: P
I
|
I,I+
D
2
|I

|+|I
+
|1
. (4.28)
The restrictions of these maps to subsets of |
I,I+
are required to agree with
(
minI
(q
I
)

min I
,
minI

1
minI
(q
I
)) along the end
+
(4.29)
(

max I
,

max I
) along the end

(4.30)
(
I\{i},I+
,
I\{i},I+
) on |
I\{i},I+
(4.31)
(
I,I+\{i}
,
I,I+\{i}
) on |
I,I+\{i}
(4.32)
_

P
I

+,i
(q
I
),
I

+,i

+,i
(q
I
)
_
1
_
on A
I,I

+,i
|
I

+,i
(4.33)
(
I,I

+,i
,
(I,I

+,i
)
) on |
I,I

+,i
A
I

+,i
(4.34)
(

,i
,

,i
) on |
I

,i
A
I

,i
,I+
(4.35)
_
_

(I

,i
,I+)
(q
I
)
_

,i
,I+
,
(I

,i
,I+)

(I

,i
,I+)
(q
I
)
_
1
_
on A
I

,i
|
I

,i
,I+
(4.36)
Moreover, on the boundary of each bre, the following conditions hold:

I,I+
(q, z) D
2
|I

|+|I
+
|1
(
I,I+,z
(L)) if z |
zin<
I,I+
(4.37)

I,I+
(q, z) D
2
|I

|+|I
+
|1
(
I,I+,z
(L

)) if z |
zin>
I,I+
(4.38)

I,I+
(q, z) D
2
|I

|+|I
+
|1
qmax I

,q
if z |
{0}
I,I+
. (4.39)
The existence of such data again follows from an inductive argument on the number of
elements of I

and I
+
, as in the proof of Lemma 4.3.
The pushforward of
I,I+
by
I,I+
denes a map

P
I,I+
: P
max I
|
I,I+
H
0
J
2
n+2
D
2
n+2
. (4.40)
As in Section 4.2, given x
min I+
LF
min I+
, x
in
LL

, and x


max I
L

F
max I
,
evaluation of
P
I,I+
at q denes a moduli space /
q,I,I+
(x

(q); x
in
, x(q)), where x

(q) and
x(q) are the images of x

and x under

max I
(q) and
min I+
(q).
By construction, composition with
I,I+
(q) yields a homeomorphism
(4.41) /
qmax I

,I,I+
(x

(q
max I
); x
in
, x(q
max I
))

= /
q,I,I+
(x

(q); x
in
, x(q)).
THE FAMILY FLOER FUNCTOR IS FAITHFUL 23
We choose the continuation data generically so that all such moduli spaces are regular.
4.4. Locally constant continuation maps in a prism: output. As in Section 4.4,
assume that I

and I
+
are totally ordered subsets of such that min I
+
max I

.
Remark 4.6. Figure 7, and the fact that the holomorphic curve problem on the moduli
space |
I,I+
in the presence of an output along the boundary is dened with Lagrangian
boundary conditions F
min I+
may be helpful in understanding the construction.
Pick a continuation datum with outputs
I,I+
, and a family of dieomorphisms
I,I+
which are obtained by gluing and agree with the identity on q
minI+
|
I,I+
:

I,I+
: |
I,I+
H
0
D
2
|I

|+|I
+
|1
J
2
|I

|+|I
+
|1
(4.42)

I,I+
: P
min I+
|
I,I+
D
2
|I

|+|I
+
|1
(4.43)
The restrictions of these maps to subsets of |
I,I+
are required to be
_

minI+
,

min I+
_
along the end
+
(4.44)
(
_

maxI
(q
min I+
)
_


maxI
,
max I

1
max I
(q
minI+
)) along the end

(4.45)
(
I\{i},I+
,
I\{i},I+
) on |
I\{i},I+
(4.46)
(
I,I+\{i}
,
I,I+\{i}
) on |
I,I+\{i}
(4.47)
_

P
I

+,i
(q
min I+
),

+,i

+,i
(q
min I+
)
_
1
_
on A
I,I

+,i
|
I

+,i
(4.48)
((
(I

,I

+,i
)
(qmin I
+
))
I

,I

+,i
,
(I

,I

+,i
)

1
(I

,I

+,i
)
(qmin I
+
)) on |
I,I

+,i
A
I

+,i
(4.49)
(
P
I

,i
(q
min I+
),
I

,i

1
I

,i
(q
min I+
)) on |
I

,i
A
I

,i
,I+
(4.50)
_

,i
,I+
,
(I

,i
,I+)
_
on A
I

,i
|
I

,i
,I+
(4.51)
On the boundary of each bre:

I,I+
(q, z) D
2
|I

|+|I
+
|1
(
I,I+,z
(L

)) if z |
zin<
I,I+
(4.52)

I,I+
(q, z) D
2
|I

|+|I
+
|1
(
I,I+,z
(L)) if z |
zin>
I,I+
(4.53)

I,I+
(q, z) D
2
|I

|+|I
+
|1
qmin I
+
,q
if z |
{0}
I,I+
. (4.54)
It is crucial at this stage that P
I
was required to include the closure of the open star of the
cell corresponding to I, as expressions of the form
I
(q
i
), with i an arbitrary element of I,
must be dened.
The pushforward of
I,I+
by
I,I+
denes a map

P
I,I+
: P
minI+
|
I,I+
H
0
J
2
n+2
D
2
n+2
. (4.55)
Given x
max I
L F
max I
, x
ou
L L

, and x


min I+
L

F
min I+
, the data

P
I,I+
(q) yield a moduli space /
q,I,I+
(x(q), x
ou
; x

(q)), for each point q P


max I+
, where
x(q) and x

(q) are the images of x and x

under
max I
(q) and

minI+
(q). Composition
with
P
I,I+
(q) yields a homeomorphism
(4.56) /
qmin I
+
,I,I+
(x, x
ou
; x

)

= /
q,I,I+
(x, x
ou
; x

(q)).
24 M. ABOUZAID
4.5. Moduli spaces associated to cells of . Combining the discussions of Section 2.3
and 3.5, yields a moduli space
(4.57) /
q,

K
(x

(q); x
in
, x(q)),
whenever q P
max

K
, by taking the bre product of /
q,K
in
K
0
(x

(q); x
in
, x(q)) with A

K;in
over A
K
in
K
0
. Given a totally ordered subset K of , dene
(4.58) /
q,K;in
(x

(q); x
in
, x(q)) =
_
|

K|=|K|
K=max

K
/
q,

K
(x

(q); x
in
, x(q)),
where the moduli spaces on the right are glued along the boundary strata obtained by
omitting an element of

K. This moduli space is again independent of q by applying the
homeomorphism in Equation (4.41), so it is denoted /
K;in
(x

; x
in
, x).
This moduli space is parametrised by a manifold of dimension [K[ 1, obtained by taking
the union of the spaces A

K;in
, for

K a maximal length sequence with largest element K,
glued along the boundary strata in Equation (2.58). Assuming that all its strata are dened
using regular continuation data, the dimension of the interior is
(4.59) deg(x

) deg(x
in
) deg(x) +[K[ 1.
Lemma 4.7. If the dimension is 1, then /
K;in
(x

; x
in
, x) is a 1-dimensional manifold with
boundary

yinLL

/
K;in
(x

; y
in
, x) /(y
in
, x
in
) (4.60)

iK

yiLFi
/
K

i
;in
(y

; x
in
, y) /
K

i
(y, x) (4.61)

iK

iL

Fi
/
K

i
(x

, y

) /
K

i
;in
(y

; x
in
, y) (4.62)

iK\{minK,max K}
/
K\{i};in
(x

; x
in
, x) (4.63)
Proof. The description of the boundary of the moduli space follows from Lemma 3.5. The
key point is that /
K;in
(x

; x
in
, x) is obtained by gluing the moduli spaces /

K;in
(x

; x
in
, x)
along the strata given in Equation (3.30), so these correspond to the boundary. The strata
in Equations (3.32)-(3.34) correspond to Equations (4.61)-(4.63).
There are parallel results for moduli spaces with outputs:
(4.64) /

K;ou
(x(q), x
ou
; x

(q)),
is a bre product over A
K
ou
K
0
. Given I B, form the union
(4.65) /
q,I;ou
(x(q), x
ou
; x

(q)) =
_
I=min

K
/

K
(x(q), x
ou
; x

(q)).
As this moduli space is independent of q, denote it /
I;ou
(x, x
ou
; x

).
For generic data, the interior is a manifold of dimension
(4.66) deg(x) + deg(x
ou
) deg(x

) +n + 1 [I[.
THE FAMILY FLOER FUNCTOR IS FAITHFUL 25
Figure 8. The pairs barycentric subdivision of a simplex.
Lemma 4.8. If this dimension is 1, then /
I;ou
(x, x
ou
; x

) is a 1-dimensional manifold with


boundary

youLL

/(x
ou
, y
ou
) /
I;ou
(x, y
ou
; x

) (4.67)

max J=min I

yLminJFmin J
/
J
(x, y) /
JI;ou
(y, x
ou
; x

) (4.68)

min J=max I

Lmax JFmax J
/
IJ;ou
(x, x
ou
; y

) /
J
(y

, x

) (4.69)

I{j}B
min I<j<max I,j / I
/
I{j};ou
(x, x
ou
, x

). (4.70)
5. Moduli space of degenerate annuli
5.1. Dual and pair subdivisions. As in Section 4.1, let denote a partially ordered set
labelling a simplicial triangulation of Q. In particular, every maximal totally ordered subset
of consists of n + 1 elements. Given a totally ordered subset J of , denote by
J
the
corresponding simplex included in Q. The barycenters of the top dimensional simplices are
the vertices of the dual subdivision, corresponding to the partially order set

which is
obtained from by reversing the order; write
J
for the cell dual to
J
. This cell can
be realised as the polyhedron associated to the partially ordered set K[J K equipped
with the ordering which reverses inclusions (i.e. J is the unique maximal element of this
ordering, and corresponds to the top dimensional face). In particular,
(5.1)
I
=
_
IJ
|J|=|I|+1

J
.
The pairs subdivision P of is the partially ordered set whose elements are nested
pairs I K of totally ordered subsets of . The cells associated to totally ordered subsets
of P correspond to pairs of intersecting cells associated to totally ordered subsets of
and

. Moreover, these cells are cubes: this can be seen geometrically by considering the
dual subdivision of a simplex, or algebraically by noting that the maximal elements of P
are pairs i K, where K has length n + 1: the corresponding cell of Q is associated to
the set of elements preceding this pair, which consists of all pairs I J contained in K
and containing i. This is exactly the totally ordered sets associated to the cube on the set
26 M. ABOUZAID
K i as such a pair corresponds to the face in which all coordinates labelled by I i
vanish, while those corresponding to K J equal 1.
5.2. The pairs barycentric subdivision. Let B denote the barycentric subdivision of
whose elements are totally ordered subsets, and whose partial ordering is given by inclusion.
Denote a totally ordered subset of B by

I, and the corresponding simplex by

I
.
Let PB denote the pairs subdivision of the barycentric subdivision, which will hence-
forth be called the pairs barycentric subdivison. Cells correspond to pairs

I

J where

J is
a totally ordered subset of . Write

J
for the corresponding cube embedded in Q.
Let I
0
and J
0
(respectively I and J) denote the minimal (respectively maximal) elements
of

I and

J. There are natural maps
(5.2) A

I
;ou

J

A

I
0
;in
where the ordered sets

J

I
and

J

I0
respectively consist of elements of

J which are contained
in I
0
and which contain I. At the level of partially ordered sets the rst map takes a pair
of subsets

I
1


J
1
to their intersections with

J

I
, and the second to their intersection with

J

I0
:
(5.3)

I
1,
I


J
1,
I

I
1


J
1

I
1,
I0


J
1,
I0
Lemma 5.1. If

I = I the product map is a bijection:
(5.4)

J
A

I
;ou
A

I
0
;in
.
Proof. By construction,

J
is the cube on

J I, whereas A

I
0
;in
and A

I
;ou
are respec-
tively the cubes on

J

I0
I and

J

I
I. By assumption, I
0
= I, and every element of

J may
be uniquely written as the union of an element of

J

I0
and

J

I
. Equation (5.4) realises the
the induced product decomposition
(5.5) [0, 1]

J\I

= [0, 1]

I
0
\I
[0, 1]

I
\I
.

Composing Equation (5.2) with the maps dened in Equation (2.48) yields maps
(5.6) A
J
ou
I

J
ou

in

A
I0
in
J
0
5.3. Degenerate annuli over a cell of the pairs barycentric subdivision. Given a
pair

I

J, dene
min

I
(

J) = minJ
1
[I
0
J
1
I, and J
1


J (5.7)
max

I
(

J) = maxJ
1
[I
0
J
1
I, and J
1


J. (5.8)
For each cell of the pairs barycentric subdivision, this denes maps
(5.9) A
I

I
0

J
min

max

A
I

I
0
which, at the level of partially ordered sets, are given by
(5.10) min

I
(

I
1
) min

I
(

J
1
)

I
1


J
1

max

I
(

I
1
) max

I
(

J
1
) .
THE FAMILY FLOER FUNCTOR IS FAITHFUL 27
min I
1
0
max I
1
0
max I
0
min I
0
max I
min I
max I
1
min I
1
Figure 9. A bre of (

I
1

J
1
) for

I
1
= I
1
0
I
0
I I
1
.
Dene the universal curve (

J
over
(5.11) A

J
A
J
ou
I
A
I

I
0
A
I

I
0
A
I0
in
J
0
to be the union of the pullback of the universal curves of each factor.
Denition 5.2. The universal family of degenerate annuli (

J
) over

J
is the
pullback of (

J
by the product of
ou
, max, min, and
in
.
The reader should consult Figure 9 which shows how to interpret the bre of (

J
)
as a degenerate annulus by arranging its constituent strips end-to-end.
Given a pair of strata

I
1

J
1

J
, there is a commutative diagram for the restrictions
of
in
and
ou
:
(5.12) A
I
1,in
0,J
1
0
A
I

0,I
1
0
A
I
1,in
0,J
1
0
A
I

0,I
1
0




A
I
in
0,J
0

I
1

J
1

in
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
ou
.

J
in

ou

A
J
1,ou
I
1
A
I
1,
I
A
J
1,ou
I
1
A
I
1,
I



A
J
ou
I
.
On the other hand, the maps min and max t into the diagram
(5.13) A
I
1,
I
1
0
A
I

0,I
1
0
A
I

I
0
A
I
1,
I
?
_

A
I

I
0

I
1

J
1

A
I
1,
I
1
0
A
I
1,
I
A
I

I
0
A
I

0,I
1
0
?
_

A
I

I
0
.
Letting
(5.14)

I
1

J
1
A

J
A
J
1,ou
I
1
A
I
1,
I
A
I

I
0
A
I

0,I
1
0
A
I

0,I
1
0
A
I

I
0
A
I
1,
I
A
I
1,in
0,J
1
0
,
28 M. ABOUZAID
there is a combined diagram
(5.15)

I
1

J
1

I
1

J
1

I
1

J
1
A


?
_

J
.
The maps on the bottom row are inclusions of boundary strata, and the universal curve over
the left and right products of Adams path spaces restrict to the natural universal curve in
the product in the middle. In particular:
Lemma 5.3. There is a natural identication
(5.16) (

I
1

J
1
)

= (

J
)[

I
1

J
1
.
The above result implies the existence of a family of degenerate annuli over Q
(5.17) (

(Q) Q
whose restriction to each top-dimensional cell of the pairs barycentric subdivision is naturally
isomorphic to

ou
|
J
ou
I

in
|
I0
in
J
0
. The boundaries of the bres will be denoted
(
{0},
Q
=

ou
|
{0}
J
ou
I

in
|
{0}
I0
in
J
0
and (
{1},
Q
=

ou
|
{1}
J
ou
I

in
|
{1}
I0
in
J
0
(5.18)
and have natural marked points z
in
and z
ou
, which induce a decomposition
(5.19) (
{1},
Q
= (
zin<zou,
Q
(
zou<zin,
Q
coming from the decompositions of |
{1}
J
ou
I
and |
{1}
I0
in
J
0
.
5.4. Floer data on degenerate annuli. On each top-dimensional cell of the pairs barycen-
tric subdivision, consider the map

{I}

J
dened as the composition
(5.20) (

J
) =

in
|
I
in
J
0

ou
|
J
ou
I

H
0
J
2
n+2
D
2
n+2

{I}

J
|
I
in
J
0

{I}

J
|
J
ou
I

P
max I
|
I
in
J
0
P
max J
|
J
ou
I
(
P
I
in
J
0
,
P
J
ou
I
)

Over a codimension 1 stratum, the restrictions of these maps agree under the natural iden-
tications; for the rst map, this follows from Equations (4.31)-(4.36), while for the second
map this follows from the analogous conditions imposed in Equations (4.46)-(4.51). Taking
the union over all top-dimensional cells, we obtain a map
(5.21)

Q
= (

Q
, J

Q
,

Q
): (

(Q) H
0
J
2
n+2
D
2
n+2
.
Given intersection points x
in
, x
ou
L L

, dene
(5.22) /

J
(x
ou
; x
in
)
to be the moduli space of nite energy maps from a bre of (

J
) to X such that

s
u(z) = J

Q
(z)
t
u(z) u(z)

Q
(z)(L) if z (
zin<zou,
(

J
) (5.23)
u(z) F
q
if z (
{0},
(q) u(z)

Q
(z)(L

) if z (
zou<zin,
(

J
), (5.24)
THE FAMILY FLOER FUNCTOR IS FAITHFUL 29
and which converge to x
in
and x
ou
at the ends. The virtual dimension of this space is
(5.25) dim

J
n + deg x
ou
deg x
in
.
Since the Floer data on each component of a bre of (

J
) were chosen generically in
Sections 4.2, 4.3, and 4.4, this moduli space is empty whenever the virtual dimension is
negative. Assuming that deg x
ou
= deg x
in
, all moduli spaces are therefore empty, except
those for which dim

J
= n, i.e.

I = I, and

J has length n + 1. Dene
(5.26) /

I
(x
ou
; x
in
)
_
I

J
/

J
(x
ou
; x
in
),
and denote the component consisting of curves whose boundary is null-homologous by
(5.27) /

[0]
(x
ou
; x
in
) /

I
(x
ou
; x
in
),
and by /

[0],I
(x
ou
; x
in
) its Gromov-Floer compactication. For each class H
1
(F
qI
, Z),
let
/
qI,I,;ou
(x, x
ou
, x

) /
qI ,I;ou
(x, x
ou
, x

) (5.28)
/
qI ,I,;in
(x

; x
in
, x) /
qI ,I;in
(x

; x
in
, x) (5.29)
denote the components consisting of curves u such that [u] = (in both cases, use the
path from x to x

to dene this homology class).


Lemma 5.4. If deg x
ou
= deg x
in
, /

[0],I
(x
ou
; x
in
) is a 0-dimensional manifold, which
agrees with its Gromov-Floer compactication, and admits a decomposition
(5.30)

xFIL
x

FIL

H1(Fq
I
,Z)
/
qI,I,;ou
(x, x
ou
; x

) /
qI ,I,;in
(x

; x
in
, x).

Write /

[0]
(x
ou
; x
in
) for the union of the moduli spaces /

[0],I
(x
ou
; x
in
) over all cells I of
.
5.5. Gluing description of Floer data on degenerate annuli. As in Lemma 5.3, con-
sider an inclusion of cells of the pairs barycentric subdvisision

I
1

J
1

J
. Pulling back
the gluing map in Equation (2.13) for each of the factors of (

J
, yields a map
(5.31) g

I
1
:

J
[0, ]
min

I
1
\min

I
[0, ]
max

I
1
\max

I
.
This maps provides a gluing description of the restriction of (

J
) to a neighbourhood
of

I
1

J
1
. Indeed, composing the map from

J
to A

J
with the maps in Equations
(2.11) for each factor of the target denes a map

J
A

I
1

J
1 . Denote the pullback of
the universal curve under this map by
(5.32) (

I
1

J
1
(

J
)

J
.
The gluing map in Equation (2.19) yields a map of families of surfaces over

I
1

J
1
(5.33) (

I
1

J
1
(

I
1

J
1
) (

I
1

J
1
).
Dene the continuation data
(5.34)

I
1

J
1
= (

I
1

J
1
, J

I
1

J
1
,

I
1

J
1
): (

I
1

J
1
(

J
) H
0
J
2
n+2
D
2
n+2
to be the composition of

Q
with the gluing map.
30 M. ABOUZAID
If V

I
1

J
1
is a neighbourhood of

I
1

J
1
whose intersection with every cell

J
lies in

I
1

J
1
, we dene
(5.35) (

I
1

J
1
(V

I
1

J
1
) V

I
1

J
1
to be the union of the restrictions
(5.36) (

I
1

J
1
(

I
1

J
1
)[

I
1

J
1
V

I
1

J
1
,
glued along the natural identications across strata of the pairs barycentric subdivision.
The compatibility of the Floer data across strata yields maps
(5.37)

I
1

J
1
: (

I
1

J
1
(V

I
1

J
1
) H
0
J
2
n+2
D
2
n+2
.
6. Cardys relation
6.1. Moduli of smooth annuli over Q. Let ( [0, ) be the family of curves over the
positive real line whose bre at S is the annulus
(6.1) R/4SZ [0, 1]

= z C[e
/2S
[z[ 1,
where the left hand side is equipped with coordinates (s, t) and complex structure j
s
=
t
,
and the identication is via the coordinates e
2
t+is
4S
. For S = 0, the bre is the unit disc.
Let (
S
denote the bre over S. Given any subset U Q, let
(6.2) ((U) U [0, )
denote the pullback of this family of annuli via projection to the rst factor, with (
S
(U)
the bre over S. Fix the holomorphic embeddings

in
,
ou
: [S, S] [0, 1] R/4SZ [0, 1] (6.3)
which are the identity on the [0, 1] factor, and whose images are respectively the natural
inclusions of [S, S] [0, 1] and [S, 3S] [0, 1]. Note in particular that the image of (0, 1)
under the rst map is (0, 1), and under the second map is (2S, 1); in the model of the annulus
embedded in the unit disc, these marked points are 1. Denote the corresponding sections
of ((Q) by
z
in
, z
ou
: Q ((Q). (6.4)
We write (
{0}
(Q) for the family of circles over ((Q) corresponding to R/4SZ0, (
zin<zou
(Q)
for the family of intervals (0, 2S)1, and (
zou<zin
(Q) for the family of intervals (2S, 0)
1. Let (
{1}
(Q) denote the union of these intervals.
Fixing an identication of the complement of 1 in the unit disc with the strip B,
mapping 1 to , denes positive and negative strip-like ends on neighbourhoods of
z
in
, z
ou
. Choose a family of strip-like ends

: Q(0, ] B

((Q) (6.5)
covering neighbourhoods of z
in
and z
ou
, which agree with these strip-like ends for S close
to 0, and agree with the ends obtain by gluing whenever S is suciently large.
THE FAMILY FLOER FUNCTOR IS FAITHFUL 31
Figure 10. Constructing the cover V

J
by induction.
6.2. Contractible choices of gluing. Let

I
1

J
1

J
be cells of the pairs barycen-
tric subdivision,

I
1

J
1
the neighbourhood xed in Section 5.5, and

I
1

J
1
its
interior.
Fix an open cover V

J
of Q whose closures V

I

J
satisfy (see Figure 10)
V

I
1

J
1

_

I
1

J
1

I
1

J
1
(6.6)
V

I
1

J
1
V

I

J
= unless the corresponding cells are nested. (6.7)
Denition 6.1. An annulus gluing function on V

J
is a map
(6.8) g
C
: V

J
(0, +] (0, ]
min

I
(0, ]
max

I
which is smooth on V

J
(0, +) , and such that the sum of the coordinates in each of
the factors of the right agrees with the projection to (0, +].
Since addition of coordinates denes a smooth bre bundle with contractible bres
(0, )
d
(0, ), which extends to an acylic bration (0, ]
d
(0, ], the space of
choices of g
C
extending any given choice on a subset of the domain is contractible.
Restricting the domain of Equation (6.8) to in the second factor, Equation (5.31)
yields a map on the intersection of V

I
1

J
1
with each stratum
(6.9) V

I
1

J
1

I
1

J
1
g

I
1

[0, ]
min

I
1
\min

I
[0, ]
max

I
1
\max

I
[0, ]
min

I
1
\min

min

I
[0, ]
max

I
1
\max

max

I
.
The construction of gluing parameters is compatible with inclusions of strata by Equation
(2.14), so we obtain a continuous map
(6.10) g
C

: V

J
(0, ]
min

I
(0, ]
max

I
.
If

I
1

J
1

J
, the product of Equation (5.31) with the choice of a map g
C
on V

J
yields a map
(6.11) V

J
V

I
1

J
1
(0, ]
min

I
1
(0, ]
max

I
1
.
Since the space of choices is contractible, such a map may be extended to a map on V

I
1

J
1
.
Proceeding by descending induction on the dimension of

J
shows:
32 M. ABOUZAID
Lemma 6.2. There is a choice of annulus gluing maps g
C
restricting to g
C

on the bound-
ary, and which extends the map in Equation (6.11) for every nested pair.
6.3. Gluing maps from degenerate to smooth annuli. Let V

J
be a neighbourhood
of

J
as in the previous section. Recall that the components of the bres of (

J
(V

J
)
over a point in V

J
are identied with strips, and have ends labelled by (min I
0
, max I
0
),
(min I, max I), or successive elements of either min

I or max

I.
Since the closure of V

J
does not intersect any cell which does not include

J
, the
Floer data are constant outside the union of the nite strips
[g
C
maxI0
(v, S), g
C
minI0
(v, S)] [0, 1] (6.12)
[g
C
minI
(v, S), g
C
maxI
(v, S)] [0, 1] (6.13)
[g
C
j
(v, S), g
C
i
(v, S)] [0, 1] (6.14)
whenever S is large enough, where i < j are successive elements of min

I or max

I in the
last equation.
Restricting the embeddings
in
and
ou
, denes maps
[g
C
maxI0
(v, S), g
C
minI0
(v, S)] [0, 1] R/4SZ [0, 1] (6.15)
[g
C
minI
(v, S), g
C
max I
(v, S)] [0, 1] R/4SZ [0, 1], (6.16)
where the image of the second map is [2S g
C
minI
(v, S), 2S +g
C
max I
(v, S)] [0, 1].
To dene the embeddings corresponding to the other components of (

J
(V

J
), let
(6.17) R
j
(v, S) = 2

kmin

I
jk
g
C
k
(v, S)
for j min

I and dene the embedding in R/4SZ [0, 1]


(6.18) [g
C
j
(v, S), g
C
i
(v, S)] [0, 1] [R
j
(v, S) g
C
j
(v, S), R
j
(v, S) + g
C
j
(v, S)] [0, 1]
for successive elements min I
0
,= i < j of min

I. Since R
min I
(v, S) = 2S by assumption,
these embeddings intersect only on the boundary, and their images cover the nite strip
(6.19) [g
C
minI0
(v, S), 2S g
C
minI
(v, S)] [0, 1] R/4SZ [0, 1].
Similarly, if j max

I, dene
(6.20) R
j
(v, S) = 2

kmax

I
jk
g
C
k
(v, S),
and consider the embedding
(6.21) [g
C
j
(v, S), g
C
i
(v, S)] [0, 1] [R
i
(v, S) g
C
j
(v, S), R
i
(v, S) +g
C
i
(v, S)] [0, 1]
The union of these strips is
(6.22) [2S +g
C
maxI
(v, S), g
C
maxI0
(v, S)] [0, 1] R/4SZ [0, 1].
Note that the annulus is covered by the images of Equations (6.15), (6.16), (6.19), and (6.22).
Collapsing the innite ends of each strip containing the domains of Equations (6.15), (6.16),
(6.18), and (6.21) to the corresponding boundary interval, yields a surjective map
(6.23) G
C

J
: (

J
(V

J
) (
S
(V

J
).
THE FAMILY FLOER FUNCTOR IS FAITHFUL 33
6.4. Floer data on annuli. Floer data on annuli are given by a map

Q
= (
Q
, J
Q
,
Q
): ((Q) H
0
J D (6.24)
such that (i) the pullback of
Q
under
in
and
ou
agrees with (Id, J
t
, Id), (ii) the restriction
of J
Q
to (
zin<zou
(Q) agrees with the pushforward of J under
Q
, and (iii) the restriction of
J
Q
to (
zou<zin
(Q) agrees with the pushforward of J

under
Q
.
By gluing,

J
induces Floer data

J
on a neighbourhood of S = in ((V

J
) by
requiring the commutativity of the diagram
(6.25) (

J
(V

J
)

(
S
(V

J
)

H
0
J D.
whenever S is suciently large. The maps (
Q
,
Q
) are said to be obtained by gluing if their
restriction to (
S
(V

J
) agree with (

J
,

J
) whenever S is suciently large.
As in the case of strips, a more general class of almost complex structures is needed.
Dene the R-thick subset of ((V

J
) to be the union of the images of [R, R] [0, 1] under
the maps in Equations (6.15), (6.16), (6.18), and (6.21). A section of the pullback of TJ by
J

J
is consistent if it is supported in the interior of the R-thick part, and it vanishes to
innite order at S = .
Denition 6.3. The data
Q
are obtained by perturbed gluing if
Q
and
Q
are obtained
by gluing, and the restriction of J
Q
to a neighbourhood of S = in ((V

J
) agrees with a
consistent perturbation of J

J
.
6.5. Moduli spaces of annuli. The identication of the complement of z
in
and z
ou
in (
0
with a strip yields data
(6.26) (
0
Q
, J
0
Q
,
0
Q
) (Id, J
t
, Id): (
0
(Q) H
0
J D.
There is a natural embedding (
S
(Q) (
0
(Q). Say that Floer data (
Q
, J
Q
,
Q
) are
obtained by gluing near S = 0 if they agree with the restriction of the above data for S
suciently small. This agrees with the usual notion of gluing under the assumption, made
in Section 3.2, that J
t
is constant in a neighbourhood of t = 1/2. Such data are obtained by
perturbed gluing if there is a compact subset of the interior of the punctured disc wherein
the almost complex structure agrees with the restriction of the almost complex structure
J
0
Q
up to a perturbation which vanishes to innite order at S = 0.
Choose Floer data (
Q
, J
Q
,
Q
) which are obtained by perturbed gluing near S = 0
and S = , and whose pullbacks under the strip-like ends

agree with the restrictions


of (
0
Q
, J
0
Q
,
0
Q
). Given intersection points x
in
, x
ou
L L

, dene /
S
(x
ou
; x
in
) to be the
union over q Q of the space of nite-energy maps from bres of (
S
(Q) to X with boundary
conditions

s
u(z) = J
Q
(z)
t
u(z) u(z) F
q
if z (
{0}
(q) (6.27)
u(z)
Q
(z)(L) if z (
zin<zou
(Q) u(z)
Q
(z)(L

) if z (
zou<zin
(Q), (6.28)
and which converge to x
in
and x
ou
at the ends. Let
(6.29) /
(0,)
(x
ou
; x
in
)

S(0,)
/
S
(x
ou
; x
in
),
which is topologised as a parametrised moduli space over Q[0, ).
34 M. ABOUZAID
Lemma 6.4. For generic Floer data
Q
, /
(0,)
(x
ou
; x
in
) is a smooth manifold and
(6.30) dimA
(0,)
(x
ou
; x
in
) = 1 + deg x
ou
deg x
in
.
Proof. The formula for the virtual dimension is a special case of [15, Section 12c]. The
assertion that generic Floer data yield smooth moduli spaces follows from [7, Theorem
5.1].
Denote the space of annuli such that the image of the boundary component mapping to a
bre is null-homologous by /
(0,)
[0]
(x
ou
; x
in
). Let /
[0,]
[0]
(x
ou
; x
in
) denote its Gromov-Floer
compactication:
Lemma 6.5. Whenever deg x
ou
= deg x
in
, /
[0,]
[0]
(x
ou
; x
in
) is a compact 1-dimensional
manifold if the data are chosen generically, and its boundary is stratied as follows:
/(x
ou
; x
in
) (6.31)

xLL

/
(0,)
[0]
(x
ou
; x) /(x; x
in
) (6.32)

xLL

/(x
ou
; x) /
(0,)
[0]
(x; x
in
) (6.33)
/

[0]
(x
ou
; x
in
) (6.34)
Proof. Having excluded interior sphere bubbling by a topological assumption, and disc bub-
bling by the careful choice of almost complex structure along the boundary, the virtual di-
mension of all strata of the Gromov-Floer compactication is negative if deg x
ou
< deg x
in
, so
the regularity of the choices of Floer data implies that they are empty. If deg x
ou
= deg x
in
,
the only strata which do not have negative virtual dimension are those given in the state-
ment. The rst and last stratum respectively correspond to S = 0 and S = , and the
middle two correspond to breaking along the ends.
The proof that /
[0,]
[0]
(x
ou
; x
in
) is a manifold with boundary follows from a standard glu-
ing argument. For S = 0, this takes the form of interior gluing of the element of /(x
ou
; x
in
)
with the family of constants (ghost) discs on bres of . For S = , there is a natural
projection map
(6.35) (

[0]
(x
ou
; x
in
) Q.
Choose an open subset V

J
containing the image of a map u in the left hand side. The
gluing description of Floer data on ((V

J
) yields a gluing chart near this boundary point.

7. Twisted sheaves of perfect complexes


7.1. The rigid-analytic T-dual. Let be a partially ordered set, labelling a suciently
ne simplicial triangulation of Q as in Section 4, and let P
i
Q be suciently small integral
ane polygons such that P
j
P
i
whenever i < j. Given I B, let
(7.1) P
I
=

iI
P
i
= P
max I
.
We x basepoints q
i
P
i
, which induce basepoints q
I
P
I
. The ane structure on Q
denes an embedding of P
i
in T
qi
Q, which we identify with H
1
(F
i
, R), where F
I
is the bre
THE FAMILY FLOER FUNCTOR IS FAITHFUL 35
at q
I
. Denote the image of every point q P
i
in H
1
(F
i
, R) by q q
i
. Whenever i < j,
parallel transport yields an ane map, taking the origin to q
j
q
i
:
(7.2) H
1
(F
j
, R)

= H
1
(F
i
, R).
Let k be a eld, and the associated universal Novikov eld
(7.3) =

i=0
a
i
T

i
[a
i
k,
i
R, lim
i

i
= +.
This is a non-archimedean eld, whose non-zero elements, denoted

, are equipped with a


valuation that assigns to a non-zero series the exponent of its leading order term.
Let Y
I
denote the inverse image of P
I
under the valuation map from H
1
(F
I
,

) to
H
1
(F
I
, R). Since

splits as R U

, the ane isomorphism in Equation (7.2) induces an


isomorphism mapping Y
j
to Y
i
.
(7.4) H
1
(F
j
,

)

= H
1
(F
i
,

).
Denition 7.1. The rigid-analytic T-dual of X is the quotient
(7.5) Y =

i
Y
i
/
where the equivalence relation is induced by Equation (7.4).
Each Y
i
is an anoid domain, equipped with the ring of regular functions consisting of
Laurent series on H
1
(F
i
, Z) which converge in Y
i
:
(7.6) O
i
=
_

AH1(Fi,Z)
f
A
z
A
qi
, f
A
[v P
i
, lim
|A|+
val(f
A
) +v, A = +
_
.
Since Equation (7.2) maps P
j
to P
i
, the inclusion Y
j
Y
i
induces a (ring) map O
i
O
j
.
The space Y is therefore a rigid analytic space in the sense of Tate. Given I = i
0
< i
1
<
. . . < i
d
B, dene the O
i0
module
(7.7) O
I
= O
i
d

Oi
d

Oi
1
O
i1
.
7.2. The twisting cocycle. Let X
I
denote the inverse image of P
I
in X. Fix Lagrangian
sections
i
: P
i
X
i
Moreover, choose functions f
ij
: P
ij
R such that brewise addition
by df
ij
agrees with the transition map between the restrictions of
i
and
j
to P
ij
. We
obtain a function
(7.8) f
ij
+f
jk
f
ik
: P
ijk
R
whose dierential at q
k
lies in H
1
(F
k
, Z) and dene
(7.9)
ijk
= T
fij(q
k
)+f
jk
(q
k
)f
ik
(q
k
)
z
d(fij+f
jk
f
ik
)
q
k
O

ijk
.
If v
ijk
is a

Cech representative of a class v H
2
(Q; Z
2
), dene
(7.10)
v
ijk
= (1)
v
ijk

ijk
.
Denote the corresponding cohomology class by
v
H
2
(Y, O

). Given a pair (I, J) of


ordered subsets of such that min J = max I, it is convenient to introduce the notation
(7.11)
v
J,I
=
v
minI,max I,maxJ
.
36 M. ABOUZAID
7.3. Twisted sheaves.
Denition 7.2. An (
v
)
1
-twisted pre-sheaf of perfect O
Y
-modules consists of (i) a nite
rank graded free O
i
-module F(i) for each i and (ii) a degree 2 [I[ map
(7.12) F
I
: O
I

Oi
0
F(i
0
) F(i
d
)
for each I B. These data are required to satisfy the equation:
(7.13)

i=min I,maxI
(1)
|I

i
|+|a|
F
I\i
a =

iI

v
I

i
,I

i
F
I

i
F
I

i
a.
Remark 7.3. The sign conventions used are those of [15]. At the cohomological level, the
justication for the terminology is given in [2, Section 2.4], though the reader should be
aware that the cocycle is on the wrong side of [2, Equation (2.33)]): from an algebraic point
of view, the natural denition consists of maps G
I
satisfying
(7.14)

i=minI,max I
(1)
|I

i
|+|a|
(
v
iii+
)
1
G
I\i
a =

iI
G
I

i
G
I

i
a.
where i

are respectively the elements of I immediately preceding and succeeding i. These


two notions are equivalent by setting G
I
= F
I

iI

v
min Iii+
.
If F
ij
is a quasi-isomorphism, we call such an object an (
v
)
1
-twisted sheaf of perfect O
Y
-
modules. Henceforth, we shall call such objects sheaves, specifying
v
only when necessary
for clarity.
Sheaves form a dierential graded category, with morphisms given by
(7.15) Hom(F, F

) =

IB
Hom
Omin I
(F(min I), F(max I))[1 [I[],
where F(max I) is an O
minI
-module via restriction. Decomposing every element T of this
direct sum as T =

I
T
I
, the dierential is:
(7.16)
1
T(a) =

iI

v
I

i
,I

i
F
I

i
T
I

i
(a) + (1)
|I

i
|+(1|T|)

v
I

i
,I

i
T
I

i
F
I

i
(a)
+

i=min I,maxI
(1)
|I

i
|1+|a|+|T|
T
I\i
(a)
For a summand T
I
, we may rewrite the above as
(7.17)
1
T
I
(a) =

JB
min J=max I

v
J,I
F
J
T
I
(a)+

JB
minI=max J
(1)
|I

i
|+(1|TI|)

v
I,J
T
I
F
J
(a) +

I{j}B
minI<j<max I
(1)
|I

j
|1+|a|+|TI|
T
I
(a)
The composition of morphisms in this category is given, on an element a of the component
of Hom(F, F

) corresponding to a totally ordered subset I, by the formula


(7.18)
2
(S, T)(a) =

iI
(1)
(|S|1)|I

i
|

v
I

i
,I

i
S
I

i
T
I

i
(a).
THE FAMILY FLOER FUNCTOR IS FAITHFUL 37
8. Floer cochains and morphism of sheaves
8.1. Relative Pin
+
structures and orientation lines. Let L and L

be graded La-
grangians as in Section 3.1. In order to work over a eld of arbitrary characteristic, assume
that the class w H
2
(Q, Z
2
) is the second-Stiefel Whitney class of a vector bundle E. We
require that
(8.1) the restrictions of

w to L and L

respectively agree with w


2
(L) and w
2
(L

).
Remark 8.1. The most important cases of interest in are the trivial case and w = w
2
(Q).
As in [10], one can drop the condition that w be a Stiefel-Whitney class: the restriction of w
to the 3-skeleton may be represented as the second Stiefel-Whitney class of a vector bundle.
Since an orientation is a discrete datum, the uniqueness up to homotopy of a retraction from
a 2-dimensional complex to the 3-skeleton suces to establish consistency of orientations,
and no higher coherence is required, even in the parametrised setting.
Assumption (8.1) implies that

E[L TL and

E[L

TL

admit Pin
+
structures
which we x; this is the choice of a relative Pin
+
structure. Given an intersection point
x L L

, pick a path
x
of linear Lagrangian subspaces of T
x
X starting at T
x
L and
ending at T
x
L

, in the homotopy class prescribed by the graded lift of these Lagrangian


subspaces (see [15, Section (11j)]). We obtain a vector bundle over the interval with bre

x
(t)

E.
The choice of relative Pin
+
structures on L and L

yields Pin
+
structures on the restric-
tions of this bundle to 0 and 1. Let
x
denote the free abelian group generated by the
two choices of extensions of this Pin
+
structure on the boundary to the interior, with the
relation that their sum vanishes. Let o
x
denote the determinant line of the Cauchy-Riemann
operator on complex linear maps from C to TX with Lagrangian boundary conditions
x
(t)
(extended by L and L

outside the interval), and [o


x
[ its orientation line. We dene
(8.2)
x
= [o
x
[
x
.
For each vertex i of the triangulation of Q, x a Pin
+
structure on T

qi
Q E
qi
. By
the isomorphism of the tangent space of bres with the cotangent space of the base, this
induces such a structure on
(8.3) TF
i

E[F
i
.
For each pair i < j in , x a Pin
+
structure on the restriction of T

Q E to the corre-
sponding edge
ij
of the triangulation of , which agrees with the Pin
+
structure chosen on
the ends. Given a triple i < j < k in , there is a canonical homotopy between
ik
and the
concatenation of
ij
and
jk
. Dene a

Cech cochain with coecients in Z
2
:
(8.4) v
ijk
=
_
0 if the induced Pin
+
structures agree
1 otherwise.
Lemma 8.2. Equation (8.4) is a cocycle representing w
2
(Q) w
2
(E).
Proof. The corresponding formula for orientable vector bundles is well-known, see e.g. [13].
The general case follows from the existence of a canonical isomorphism between the set of
Pin
+
structures on a bundle E and Spin structures on (det E)
3
E, see [12].
Let x
i
L F
i
be an intersection point. The family of Cauchy-Riemann operators
associated to x
i
(q)
i
(q)(x
i
) and the choice of Pin
+
structure on the restriction of T

QE
38 M. ABOUZAID
to an edge
ij
, dene a local system over
ij
whose bre at q is
xi(q)
, hence a map
(8.5)
xi

xi(qj )
.
Given a simplex
ijk
, the map
xi

xi(q
k
)
dened by
ik
agrees with the composition
(8.6)
xi

xi(qj )

xi(q
k
)
if and only if the Pin
+
structure on the restriction of TQE to the boundary of
ij
extends
to the interior. From Lemma 8.2, we conclude:
Corollary 8.3. Equation (8.6) and (8.5) dier by (1)
v
ijk
.
8.2. Other auxiliary choices. Fix the local choice of Lagrangian section
i
made in Sec-
tion 7.2. The intersection of this section with F
i
equips this bre with a basepoint. For
each intersection point x
i
L F
i
, choose a function
(8.7) g
x
: P
i
R
such that the section corresponding to x is obtained by berwise addition of dg
x
to
i
. This
choice determines a path on F
i
from x to the basepoint. Given a map u from a strip to
X, with one boundary component mapping to F
i
converging at both ends to intersection
points with
i
L, dene
(8.8) [u] H
1
(F
i
, Z)
to be the homology class of the loop in F
i
obtained by concatenating the restriction of u
to the boundary component mapping to F
i
with the paths from these intersection points to
the basepoint.
Assume that u satises a holomorphic curve equation with moving Lagrangian boundary
conditions along R 1 given by a Hamiltonian family
s
, parametrised by s R. Let
(8.9) H: R M R
be the Hamiltonian generating this family, normalised so that integral of H
s
with respect
to a xed volume form vanishes. Dene the energy of u to be
(8.10) c(u) =
_
B
u

()
_
R
H
s
(u(s, 1))ds.
The main result about energies of strips that shall be used is [2, Lemma 3.2]: if the boundary
arc along F
q
converges to intersection points x and y, and x

and y

are the corresponding


intersection points of L with a nearby bre F
q
, then
(8.11) c(u

) c(u) = q

q, [u] +g
x
(q) g
y
(q) +g
y
(q

) g
y
(q

),
whenever u

be a curve with boundary conditions L and F


q
, which is homotopic to the
concatenation of u with the strip with boundary conditions L, F
q
, and F
q
determined by
the lifts of the functions g
x
and g
y
.
A generalisation of this result to annuli shall be required: x the normalised Hamiltonian
H: ((Q) R generating a moving path of Lagrangians along the boundary of ((Q). For
a bre in A
S
(x
ou
; x
in
), we obtain an expression for the energy
(8.12) c(u) =
_
C
S
u

()
_
s[0,4S]
H
(1,s)
(u(1, s)).
Extending this map to A

(x
ou
; x
in
) by the sum of the energies of each constituent strip
yields a real-valued function c on A
[0,]
(x
ou
; x
in
), which is easily seen to be locally constant
if the boundary condition F
q
is xed. If we change boundary conditions, Equation (8.11)
THE FAMILY FLOER FUNCTOR IS FAITHFUL 39
implies that the energy of (nearby) curves in A

(x
ou
; x
in
) with boundary on F
q
and F
q

dier by q

q, , where is the homology class of the boundary in H


1
(F
q
, Z). This
implies that the energy is locally constant on A

[0]
(x
ou
; x
in
). By expressing the dierence of
energy of nearby annuli as the integral of over an annulus in X connecting the boundary
conditions, this result extends to all annuli:
Lemma 8.4. c is locally constant on A
[0,]
[0]
(x
ou
; x
in
).
8.3. Local construction. Given i , we recall in this section the construction of an
O
i
-module F(L) associated to the Lagrangian L: given a pair of intersection points x
i
, y
i

F
i

i
L and an element u /
qi
(y
i
, x
i
), denote the orientation line of the linearised
Cauchy-Riemann operator at u by
(8.13)
u
= [ det D
u
[ = [ coker D
u
[

[ ker D
u
[,
where the absolute value symbol stands for the line of orientation, and [V [

is the dual line.


Index theory determines a canonical isomorphism (see [15, Remark 11.6]):
(8.14)
u

xi

=
yi
.
Assuming that deg(y
i
) = deg(x
i
) + 1, the moduli space /(y
i
, x
i
) consists only of rigid
curves, and ker D
u
is 1-dimensional, which implies that it is generated by translation in the
s-direction. Fixing the orientation of this kernel corresponding to the positive direction,
yields a map
(8.15)
u
:
xi

yi
.
The family Floer module and the dierential are given by
F(L, i)

xiiLFi
O
i

xi
(8.16)
F
i
: F(L, i) F(L, i)[1] (8.17)
F
i
[
xi
=

yi

uMq
i
(yi,xi)
T
E(u)
z
[u]
i

u
. (8.18)
That this map is well-dened (i.e. the expression

uM(yi,xi)
T
E(u)
z
[u]
gives a function
in O
i
), is a consequence of Gromov compactness, as rst noticed by Fukaya [9], see [2,
Proposition 3.3].
8.4. From local to global. Let K be a totally ordered subset of . Given x
min K

min K
L F
minK
and x
max K

max K
L F
max K
, recall that
(8.19) /
K
(x
max K
; x
min K
) /
qK,K
(x
max K
; x
minK
(q
K
)).
There is a natural isomorphism
(8.20) [T
u
/
K
(x
max K
; x
minK
)[
xminK(qmax K)
[A
K
[
xmax K
.
Assuming that deg(x
max K
) = deg(x
min K
) + 2 [K[, rigidity yields an isomorphism
(8.21)
u
:
xminK

xmax K
40 M. ABOUZAID
by xing (i) the isomorphism in Equation (8.5) and (ii) the orientation of A
K
arising from
its description as a product of intervals, and the ordering on K. Dene
F
K
: F(L, min K) F(L, max K)[2 [K[] (8.22)
F
K
[
xminK
=

xmax K
T
fminK,max K
z
dfijdgmax K+dgminK
maxK
(8.23)

uMK(xmax K;xminK)
T
E(u)
z
[u]
maxK

u
.
This map is convergent by [2, Proposition 3.11].
There is a natural bijection between the boundary of the 1-dimensional moduli spaces
/
K
(x
max K
; x
min K
), given in Equations (3.16)-(3.17), and the terms of Equation (7.13):
(8.24)
/
K

i
/
K

i
F
K

i
F
K

i
/
K\{i}
F
K\{i}
To conclude that Equation (7.13) holds, it suces to show that the coecient of each term
is correct, i.e prove the cancellation of the terms in Equation (7.13) which have as coecient
a xed monomial.
Fix an energy E and a homology class H
1
(F
qmax K
, Z), and let /
E,
qmax K,K
(x
max K
; x
min K
)
be the corresponding component of the moduli space.
Lemma 8.5. The contribution to Equation (7.13) of every boundary stratum of /
E,
qmax K,K
(x
max K
; x
minK
)
agrees with
(8.25) T
fminK,max K(qmax K)
z
dfminK,max Kdgmax K+dgminK
maxK
T
E(u)
z
[u]
maxK

u
.
Proof. The only strata for which this does not follow immediately from the denition are
those corresponding to /
K

i
/
K

i
, for E corresponds to the energy of a holomorphic
curve with boundary conditions F
max K
, while the map F
K

i
is dened using a moduli space
with boundary conditions F
i
. The comparison between energies is given in [2, Lemma 3.2],
and the proof of [2, Lemma 4.2] applies verbatim.
Combining this with the sign computation from Corollary 8.3, implies:
Lemma 8.6. The maps F
K
dene an (
v
)
1
-twisted sheaf of perfect O
Y
-modules.
8.5. From Floer to

Cech. For Lagrangian branes (L, L

), consider the Floer complex


(8.26) CF

(L, L

) =

xLL


x
.
Every rigid curve u /(y, x) denes a map
u
:
x

y
, yielding the dierential
(8.27)
1
[
x
=

uM(y,x)
T
E(u)

u
.
Given x
min K
L F
min K
, x

max K
L

F
max K
, and x
in
L L

, let
(8.28) /
K;in
(x

max K
; x
in
, x
minK
) =
_
K=max

K
/
q,

K
(x

max K
; x
in
, x
min K
(q
K
))
as in Section 4.5. Each element of the right hand side is a parametrised moduli space
(8.29) /
q,

K
(x

max K
; x
in
, x
min K
(q
K
)) A

K;in
THE FAMILY FLOER FUNCTOR IS FAITHFUL 41
If [

K[ = [K[, subsequent elements of



K dier by one element of K. We x the orientation
of A

K;in
coming from the ordering of these elements. Together with the isomorphisms in
Equation (8.5), this results in a natural isomorphism
(8.30)
u
:
xin

xminK

xmax K
for every rigid element u /
K;in
(x

max K
; x
in
, x
minK
). Dene
C
K
: CF

(L, L

) F(L, min K) F(L

, max K)[1 [K[] (8.31)


C
K
[
xin

xmin K
=

max K

uMK;in(x

max K
;xin,xminK)
T
E(u)
z
[u]
maxK

u
, (8.32)
and denote the direct sum of these maps by
(8.33) C: CF

(L, L

) Hom(F(L), F(L

)).
The boundary strata of /
K;in
(x

max K
; x
in
, x
min K
) listed in Equations (4.60)-(4.63) are
matched with the terms appearing in the equation for a chain map with respect to the
Floer dierential and the dierential in Equation (7.16) as follows:
(8.34)
/
K;in
/ C
K
(
1
id)
/
K

i
;in
/
K

i
C
K

i
(id F
K

i
)
/
K

i
/
K

i
;in
F
K

i
C
K

i
/
K\{i};in
C
K\{i}
.
As in Lemma 8.6, this implies that C is a chain map.
8.6. From

Cech to Floer. Let I be a totally ordered subset of . Given x

min I
L


F
min I
, x
max I
L F
max I
, and x
ou
L L

, let
(8.35) /
I;ou
(x
min I
, x
ou
; x

max I
) =
_
I=min

K
/
q,

K
(x
min I
(q
K
), x
ou
; x

max I
)
as in Section 4.5. Each stratum of the right hand side is a parametrised moduli space over
A

K;ou
. If [

K[ = n [I[ +2, subsequent elements of



K dier by one element of the maximal
element of

K. Fix the orientation of A

K;ou
coming from the ordering of these elements.
Together with the isomorphisms in Equation (8.5), this gives a natural isomorphism
(8.36)
u
:
x

max I

xminI

xou
for every rigid element u /
I;ou
(x
min I
, x
ou
; x

max I
). Given Hom(
xminI
,
x

max I
), the
trace of the composition with
u
yields a map
(8.37) tr(
u
): Z
xou
.
Using the fact that Laurent polynomials are dense, one denes a map
P
u
: O
I
Hom(
xminI
,
x

max I
)[1 [I[]
xou
(8.38)
z

I

_
T
E(u)
tr(
u
) if [u] =
0 otherwise.
(8.39)
Using the decomposition of Hom
OminI
(F
minI
(L), F
max I
(L

)) as a direct sum
(8.40)

xminIminI LFminI
x

max I
max I LFmax I
O
I
Hom(
xmin I
,
x

max I
)[1 [I[],
42 M. ABOUZAID
dene the components of a map P to the Floer complex:
P
I
: Hom
Omin I
(F
min I
(L), F
max I
(L

)) CF

(L, L

) (8.41)
P
I
[O
I
Hom(
xmin I
,
x

max I
)[1 [I[] =

xou

uMI;ou(xminI ,xou;x

max I
)
P
u
. (8.42)
To establish that P is a chain map, it is convenient to use the expression for the dierential
on morphisms of sheaves given in Equation (7.16), which yields the equation
(8.43)
1
P(z

I
) =

JB
minJ=max I

v
J,I
P
IJ
(F
J
(L

) (z

I
)) +

JB
min I=maxJ
(1)
|I

i
|(1||)

v
I,J
P
JI
((z

I
) F
J
(L))
+

I{j}B
minI<j<max I
(1)
|I

j
|1deg(xminI)+||
P(z

I
)
assuming as in Equation (8.39) that Hom(
xminI
,
x

max I
)[1 [I[].
The correspondence between the boundary strata of /
I;ou
(x
min I
, x
ou
; x

max I
) given in
Equation (4.67)-(4.70) and the terms in the chain map equation is as follows:
(8.44)
//
I;ou

1
P
I
()
/
IJ;ou
/
J
P
IJ
(F
J
(L

) )
/
J
/
JI;ou
P
JI
( F
J
(L))
/
I{j};ou
P
I{j}
.
Lemma 8.7. P is a chain map.
Sketch of proof: To check the coecients, restrict to the subset of /
I;ou
(x
min I
, x
ou
; x

max I
)
consisting of curves with [u] = . The coecient in of each term in P is given by
the energy of the corresponding curve with one boundary condition on the bre F
I
. We
conclude that the rst and last lines in Equation (8.44) also have coecients given by the
energy of corresponding broken curve. The remaining two cases require the argument used
in Lemma 8.5 and Equation (8.11). We consider only the case corresponding to the term
P
JI
( F
J
(L

)), leaving the other to the reader.


Let u be a curve contributing to F
J
(L

). The coecient c(u) appearing in Equation


(8.23) is the area of a curve with boundary on F
max J
. The dierence with the area of the
corresponding curve with boundary on F
max I
is given by
(8.45) q
J
q
I
, [u] +g
xmax J
(q
J
) g
xminJ
(q
J
) +g
xminJ
(q
I
) g
xmaxJ
(q
I
).
The rst term appears when changing coecients from z
I
to z
J
, i.e.
(8.46) z

I
= T
,qJqI
z

J
.
The remaining coecients arise as the sums of the exponents of the coecients of
v
I,J
and
the coecients in the denition of F
J
(L

), see Equation (8.23).


8.7. Homotopy from the composition to an isomorphism. Comparing Equations
(8.32) and (8.39), implies that the composition
(8.47) P C: CF

(L, L

) CF

(L, L

)
THE FAMILY FLOER FUNCTOR IS FAITHFUL 43
is given on
xin
by counts of elements of (

[0]
(x
ou
, x
in
), and that the corresponding Novikov
coecient is given by the energy.
By Lemma 6.5, the moduli space A
[0,]
[0]
(x
ou
, x
in
) yields a cobordism between (

[0]
(x
ou
, x
in
)
and /(x
ou
; x
in
). If deg(x
ou
) = deg(x
in
), the only elements of /(x
ou
; x
in
) are constant maps,
and the corresponding map is the identity.
Proposition 8.8. The composition P C is homotopic to (1)
n(n1)
2
Id.
Proof. It remains to orient A
[0,]
[0]
(x
ou
, x
in
) relative to
xin
and
xou
. Given u A
[0,]
[0]
(x
ou
, x
in
),
with boundary condition on q
u
, there is a natural isomorphisms
(8.48) [T
u
A
[0,]
[0]
(x
ou
, x
in
)[
xin

= [(0, )[ [T
q
Q[ [TF
q
[ [TX[
1

xou
induced by degenerating the domain of the linearisation of the Cauchy-Riemann operator
at u to the union of two discs meeting at a point, and noting that one of those discs has
Lagrangians boundary condition TF
q
, hence has determinant line naturally isomorphic to
TF
q
. Using the isomorphism TX

= TQTF
q
, we conclude that every rigid element induces
a map
xin

=
xou
.
By construction, this map agrees with that dened by rigid elements of /(x
ou
; x
in
).
However, the orientations at (

[0]
(x
ou
, x
in
) dier from the product orientation by (1)
n(n1)
2
,
as rst noticed by Fukaya, Oh, Ohta, and Ono [11, Proposition 3.9.1] (see also [3, Lemma
5.5.23]). This accounts for the sign in the statement.
Appendix A. The family Floer functor
A.1. The A

structure. Let L X be a closed Lagrangian immersion in X in generic


position, i.e. so that L meets itself in pairs of transverse double points. Assume that (i)
L is tautologically unobstructed; i.e there exists a tame almost complex structure for which
there are no J
L
holomorphic maps D
2
X such that the complement of one point on
the boundary lift to L, (ii) L is a graded Lagrangian, i.e. there is a xed lift to R of the
S
1
-valued phase on L and (iii) the pullback of w H
2
(X, Z
2
) agrees with w
2
(L).
Remark A.1. The union of a nite collection of transverse immersed Lagrangians in generic
positions which satisfy these conditions for the same J
L
again satises these properties. The
constructions of this section can be carried out for dierent almost complex structures, but
the notation becomes more cumbersome.
Choose a Hamiltonian H: X R whose time-t ow
t
lies in the contractible set H
0
chosen in Section 4.1, and such that
1
L is transverse to L. Pick a family J : [0, 1] J
0
of almost complex structures on X parametrised by t [0, 1] which agree with
t

J
L
at
t = 0, 1. Generically, all moduli spaces of strips /(x, y) are regular for x, y
1
L L.
The orientation lines from Equation (8.2), with dierential from Equation (8.27) dene the
Floer complex
(A.1) CF

(L) =

xL
1
L

x
.
Let !
d+1
denote the moduli space of discs with d + 1 marked points on the boundary,
one of which is distinguished as outgoing. Let |
d+1
denote the universal punctured curve
over this moduli space. For all d, x a consistent family of negative strip-like ends
0
at the
44 M. ABOUZAID
max K min K
A
4
K\{i}
max K i
A
1
K

i
A
3
K

i
min K min K max K
A
3
K
!
3
Figure 11. Representative boundary strata of A
4
K
.
outgoing point, and of positive strip-like ends
i

d
i=1
at all other points as in [15, Section
(9g)]. Pick a consistent family of Floer data
(A.2) (
d+1
, J
d+1
): |
d+1
H
0
J
0
such that (i) J
d+1
(z) =
d+1

(z)J
L
whenever z lies on the boundary of a bre, (ii) the
pullback of (
d+1
, J
d+1
) under the strip-like ends agrees with (, J).
Remark A.2. It is only for notational convenience that
d+1
is dened as a map on |
d+1
,
since only its values on the boundary of each bre will ever be used. The fact that H
0
is
contractible implies that there is no obstruction to extending a function to H
0
.
Given a sequence x
j

d
j=0
of intersection points between L and
1
L, let /
d+1
(x
0
; x
d
, . . . , x
1
)
denote the moduli space of maps from a bre of |
d+1
to X such that
du(z) j = J
d+1
(z) dt u(z)
d+1
(z)L if z , (A.3)
and u converges at the k
th
strip-like end to x
k
. These moduli spaces are regular for generic
choices of almost complex structures J
d+1
, and a choice of orientation of !
d+1
yields a map
(A.4)
u
:
x
d

x1

x0
whenever u is rigid. Dene the A

structure on CF

(L) to be given by the operations


(A.5)
d
[
x
d

x1
=

x0L
1
L
deg(x0)=2d+

d
j=1
deg(xj)
T
E(u)

u
.
A.2. Adams moduli spaces with d marked points. Let K be a totally ordered set,
and |
K
the universal curve over A
K
from Denition 2.5. Given an integer d, let A
d
K
denote
the compactied moduli space of bres of |
K
equipped with d boundary punctures along
|
{1}
K
. Let |
d
K
denote the punctured universal curve over A
d
K
. The boundary strata of the
universal curve, shown in Figure 11, are as follows
|
d
K\{i}
A
d
K\{i}
i K (A.6)
|
d2
K

i
A
d1
K

i
A
d2
K

i
|
d1
K

i
A
d2
K

i
A
d1
K

i
i K, 0 d
i
, d
1
+d
2
= d (A.7)
|
d2
K
!
d1+1
A
d2
K
|
d1+1
A
d2
K
!
d1+1
1 j d
2
, 2 d
1
, d
1
+d
2
= d + 1. (A.8)
THE FAMILY FLOER FUNCTOR IS FAITHFUL 45
Fix consistent families of positive strip-like ends
j

d
j=1
at all punctures, and for all bres
of |
d
K
; the consistency requirement is an inductive choice on d and [K[, with the base case
being the choices of strip like ends on bres of |
d+1
over !
d+1
.
Denition A.3. A compatible family of continuation data parametrised by |
d
K
is a map
(A.9)
d
K
= (
d
K
, J
d
K
,
d
K
): |
d
K
HJ D
such that (i) the pullback under every end of the strip labelled by k K is given by a triple of
maps (
k
, J
k
, Id), (ii) the pullback under the end
j
is given by (, J, Id) (iii) the maps
d
K
,
and
d
K
, are obtained by gluing, and J
d
K
by perturbed gluing, and (iv) for each z |
d,{1}
K
(A.10) J
d
K
(z) =
_

d
K
(z)
d
K
(z)
_

J
L
.
Assuming that
k
L is transverse to F
q
for all k K, we obtain a holomorphic curve
equation
s
u(z) = J
d
K
(z)
t
u(z) on the space of maps from bres of |
d
K
to X, with boundary
conditions
u(z) F
q
if z |
{0}
K
u(z)
d
K
(z)L if z |
d,{1}
K
. (A.11)
A.3. Choices of continuation data with multiple inputs. Fix the choices made in
Section 4.1, i.e. nested sequences J
i

2
n+2
i=1
of almost complex structures and D
i

2
n+2
i=1
of
dieomorphisms, the simplicial triangulation of Q, the associated cover P
i
by polyhedra
with basepoints q
i
, and maps
i
H
0
mapping L to a Lagrangian transverse to F
i
. These
choices should be made so that D
i
(L) is an acyclic bration over L, and both D
i
P
2
i
and
D
i
P
2
i
(
i
L) are acyclic brations over P
2
i
. A section
i
of D
i
qi,Pi
(
i
L) is xed.
Assume that the maps (
I
,
I
) from Section 4.2 are chosen, and dene
(A.12)
d+1
= (
d+1
, J
d+1
, Id): |
d+1
H
0
J
0
D.
For each ordered subset of I, pick continuation data and families of dieomorphisms

d
I
= (
d
I
, J
d
I
,
d
I
): |
d
I
H
0
J
2
|I|1
D
2
|I|1
(A.13)

d
I
: P
I
|
d
I
D
2
|I|1
(A.14)
where
d
I
is obtained by gluing near every boundary stratum, and is the identity on q
I
|
d
I
.
In addition, we require that the restrictions of (
d
I
,
d
I
) agree with
(
min I
(q
I
)

minI
,
minI

1
minI
(q
I
)) along the end
+
(A.15)
(
max I
,
max I
) along the end

(A.16)
(
j
, Id) along the end
j
(A.17)
(
d
I\i
,
d
I\i
) on |
d
I\i
(A.18)
(
d2
I

i
,
d2
I

i
) on |
d2
I

i
A
d1
I

i
(A.19)
_

d1
I

i
(q
I
)

d1
I

i
,
d1
I

i
(
d1
I

i
(q
I
))
1
_
on A
d2
I

i
|
d1
I

i
(A.20)
(
d2
I
,
d2
I
) on |
d2
I
!
d1
(A.21)
(
d1
, Id) on A
d2
I
|
d1
. (A.22)
46 M. ABOUZAID
Moreover, on the boundary of each bre:

d
I
(q, z) D
2
|I|1
(
I,z
(L)) if z |
d,{1}
I
(A.23)

d
I
(q, z) D
2
|I|1
qI,q
if z |
d,{0}
I
. (A.24)
Such data can be constructed by a double induction: assuming that they have been chosen
for all pairs (d
1
, I) whenever d
1
< d, construct the data for d by induction on the number
of elements of I as in Lemma 4.3.
The pushforward of
d
K
by
d
K
yields compatible families parametrised by P
K
:

d,P
K
: P
K
|
d
K
H
0
J
2
|K|
D
2
|K|
. (A.25)
Given x F
qmin K

min K
L and y F
qK

max K
L, let x(q) and y(q) be as in Equation
(4.25). If x
j
L
1
L for 1 j d, the compactied moduli space of solutions to
the holomorphic curve equation determined by
d,P
K
(q), with boundary conditions as in
Equation (A.11), and asymptotic conditions x(q) at the positive end of the strip, y(q) at
the negative end, and x
j
along the j
th
strip like end will be denoted:
(A.26) /
d
K
(y; x
d
, . . . , x
1
, x).
Composition with
d,P
K
(q) yields a dieomorphism
(A.27) /
d
K
(y; x
d
, . . . , x
1
, x)

= /
d
qK,K
(y(q
K
); x
d
, . . . , x
1
, x(q
K
)).
Using the description of the boundary of A
d
K
given in Equations (A.6)-(A.8), the boundary
of /
d
K
(y; x
d
, . . . , x
1
, x) decomposes as follows:
/
d
K\{i}
(y; x
d
, . . . , x
1
, x) (A.28)

FiiL
/
K

i
(y, y

) /
d
K

i
(y

; x
d
, . . . , x
1
, x) (A.29)

FiiL
/
d
K

i
(y; x
d
, . . . , x
1
, x

) /
K

i
(x

, x) (A.30)

1d1,d2
d1+d2=d

FiiL
/
d2
K

i
(y; x
d
, . . . , x
d1+1
, x

) /
d1
K

i
(x

; x
d1
, . . . , x
1
, x) (A.31)
/
d2
K
(y; x
d
, . . . , x
d1+j+1
, x
0
, x
j
, . . . , x
1
, x) !
d1+1
(x
0
; x
d1+j
, . . . , x
j+1
). (A.32)
In the above, the strata corresponding to Equation (A.7) for which d
1
or d
2
vanish are listed
separately, while the breaking of strips at the ends is incorporated into Equations (A.29),
(A.30), or (A.32) depending on whether the breaking takes place at

,
+
, or one of the
ends
j
, for 1 j d.
A.4. The A

functor. Assume now that the Floer data are chosen generically so that the
moduli spaces in the previous section are manifolds of the expected dimension. Fixing an
orientation of the moduli space A
d
K
yields a map
(A.33)
u
:
x
d

x1

x

y
THE FAMILY FLOER FUNCTOR IS FAITHFUL 47
associated to every rigid element u /
d
K
(y; x
d
, . . . , x
1
, x). Dene a map C
d
K
CF

(L)
d
F(L, min K) F(L, max K)[2 [K[ d] (A.34)
C
d
K
[
x
d

x1

x
=

ymax KLFq
K

uM
d
K
(y;x
d
,...,x1,x)
T
E(u)
z
[u]
max K

u
. (A.35)
The direct sum of these maps over all K will be denoted
(A.36) C
d
: CF

(L)
d
Hom(F(L), F(L)).
Recall that an A

homomorphism from an A

algebra to a dierential graded algebra


consists of such maps which satisfy the equation:
(A.37)
1
_
C
d
(a
d
, . . . , a
1
)
_
+

d1+d2=d

2
_
C
d2
(a
d
, . . . , a
d1+1
), C
d1
(a
d1
, . . . , a
1
)
_
=

d1,d2,j
d1+d2=d+1
(1)

j
i=1
|aj|j
C
d2
(a
d
, . . . , a
d1+j+1
,
d1
(a
d1+j
, . . . , a
j+1
), a
j
, . . . , a
1
).
The terms on the right hand side correspond to the boundary strata in Equation (A.32),
those in the second term of the left to Equation (A.31), and the rst term to Equations
(A.28) and (A.29). To see the last part, use the denition of the dierential in Equation
(7.16):
Proposition A.4. The maps C
d
are the components of an A

homomorphism from CF

(L)
to the endomorphism algebra of C(L) as an (
v
)
1
-twisted sheaf.
References
[1] Mohammed Abouzaid, A geometric criterion for generating the Fukaya category, Publ. Math. Inst.
Hautes

Etudes Sci. 112 (2010), 191240.
[2] , Family Floer cohomology and mirror symmetry, Proceedings of the International Congress of
Mathematicians., 2014.
[3] , Symplectic cohomology and Viterbos theorem, available at arXiv:1312.3354.
[4] Mohammed Abouzaid, Denis Auroux, and Ludmil Katzarkov, Lagrangian brations on blowups of toric
varieties and mirror symmetry for hypersurfaces, available at arXiv:1205.0053.
[5] Mohammed Abouzaid, Kenji Fukaya, Yong-Geun Oh, Hiroshi Ohta, and Kaoru Ono, Homological
mirror symmetry for compact toric manifolds, in preparation.
[6] J. F. Adams, On the cobar construction, Proc. Nat. Acad. Sci. U.S.A. 42 (1956), 409412.
[7] Andreas Floer, Helmut Hofer, and Dietmar Salamon, Transversality in elliptic Morse theory for the
symplectic action, Duke Math. J. 80 (1995), no. 1, 251292.
[8] Kenji Fukaya, Floer homology for families - report of a project in progress, available at
http://www.math.kyoto-u.ac.jp/
~
fukaya/familyy.pdf.
[9] , Cyclic symmetry and adic convergence in Lagrangian Floer theory, Kyoto J. Math. 50 (2010),
no. 3, 521590.
[10] Kenji Fukaya, Yong-Geun Oh, Hiroshi Ohta, and Kaoru Ono, Lagrangian intersection Floer theory:
anomaly and obstruction. Part I, AMS/IP Studies in Advanced Mathematics, vol. 46, American Math-
ematical Society, Providence, RI, 2009.
[11] , Lagrangian Floer theory and mirror symmetry on compact toric manifolds, available at
arXiv:1009.1648.
[12] R. C. Kirby and L. R. Taylor, Pin structures on low-dimensional manifolds, Geometry of low-
dimensional manifolds, 2 (Durham, 1989), London Math. Soc. Lecture Note Ser., vol. 151, Cambridge
Univ. Press, Cambridge, 1990, pp. 177242.
[13] D. A. McLaughlin, Local formulae for Stiefel-Whitney classes, Manuscripta Math. 89 (1996), no. 1,
113.
[14] Paul Seidel, Homological mirror symmetry for the quartic surface, available at arXiv:0310414.
48 M. ABOUZAID
[15] , Fukaya categories and Picard-Lefschetz theory, Zurich Lectures in Advanced Mathematics,
European Mathematical Society (EMS), Z urich, 2008.
[16] Nick Sheridan, Homological Mirror Symmetry for Calabi-Yau hypersurfaces in projective space, avail-
able at arXiv:1111.0632.
[17] W. P. Thurston, Some simple examples of symplectic manifolds, Proc. Amer. Math. Soc. 55 (1976),
no. 2, 467468.
[18] Junwu Tu, On the reconstruction problem in mirror symmetry, Adv. Math. 256 (2014), 449478, DOI
10.1016/j.aim.2014.02.005. MR3177298