a

r
X
i
v
:
m
a
t
h
/
0
2
1
0
2
5
7
v
2


[
m
a
t
h
.
S
G
]


4

D
e
c

2
0
0
4
MODULI OF J-HOLOMORPHIC CURVES WITH LAGRANGIAN
BOUNDARY CONDITIONS AND OPEN GROMOV-WITTEN
INVARIANTS FOR AN S
1
-EQUIVARIANT PAIR
CHIU-CHU MELISSA LIU
Abstract. Let (X, ) be a symplectic manifold, J be an -tame almost com-
plex structure, and L be a Lagrangian submanifold. The stable compacti-
cation of the moduli space of parametrized J-holomorphic curves in X with
boundary in L (with prescribed topological data) is compact and Hausdor in
Gromovs C

-topology. We construct a Kuranishi structure with corners in

the sense of Fukaya and Ono. This Kuranishi structure is orientable if L is
spin. In the special case where the expected dimension of the moduli space is
zero, and there is an S
1
action on the pair (X, L) which preserves J and acts
freely on L, we dene the Euler number for this S
1
equivariant pair and the
prescribed topological data. We conjecture that this rational number is the
one computed by localization techniques using the given S
1
action.
Contents
1. Introduction 2
1.1. Background 2
1.2. Main results 3
2. Surfaces with Analytic or Dianalytic Structures 5
2.1. Analyticity and dianalyticity 5
2.2. Various categories of surfaces 6
2.3. Topological types of compact symmetric Riemann surfaces 8
3. Deformation theory of Bordered Riemann Surfaces 8
3.1. Deformation theory of smooth bordered Riemann surfaces 8
3.2. Nodal bordered Riemann surfaces 12
3.3. Deformation theory for prestable bordered Riemann surfaces 17
4. Moduli of Bordered Riemann Surfaces 23
4.1. Various moduli spaces and their relationships 23
4.2. Decomposition into pairs of pants 25
4.3. Fenchel-Nielsen coordinates 26
4.4. Compactness and Hausdorness 29
4.5. Orientation 31
5. Moduli Space of Stable Maps 32
5.1. Prestable and stable maps 32
5.2. C

topology 33
5.3. Compactness and Hausdorness 35
6. Construction of Kuranishi Structure 36
6.1. Kuranishi structure with corners 36
Date: February 1, 2008.
1
2
6.2. Stable W
k,p
maps 38
6.3. Deformation of the domain 45
6.4. Local Charts 51
6.5. Transition functions 70
6.6. Orientation 71
7. Virtual fundamental chain 73
7.1. Construction of virtual fundamental chain 73
7.2. S
1
action 78
7.3. Invariants for an S
1
-equivariant pair 80
7.4. Multiple covers of the disc 82
References 83
1. Introduction
1.1. Background. String theorists have been making predictions on enumerative
invariants using dualities. One of the most famous examples is the astonishing
predictions of the number of rational curves in a quintic threefold in [CdGP]. To
understand these predictions, mathematicians rst developed Gromov-Witten the-
ory to give the numerical invariants a rigorous mathematical denition so that
these predictions could be formulated as mathematical statements, and then tried
to prove these statements. The predictions in [CdGP] are proven in [Gi, LLY].
Recently, string theorists have produced enumerative predictions about holomor-
phic curves with Lagrangian boundary conditions by studying dualities involving
open strings [OV, LMV, AV, AKV, MV]. Moreover, assuming the existence of a
virtual fundamental cycle and the validity of localization formulas, mathematicians
have carried out computations which coincide with these predictions [KL, GZ]. In
certain cases, these numbers can be reproduced by considering relative morphisms
[LS]. It is desirable to give a rigorous mathematical denition of these enumera-
tive invariants, so that we may formulate physicists predictions as mathematical
theorems, and then try to prove these theorems.
Gromov-Witten invariants count J-holomorphic maps from a Riemann surface
to a xed symplectic manifold (X, ) together with an -tame almost complex
structure J. These numbers can be viewed as intersection numbers on the moduli
space of such maps. We want the moduli space to be compact without boundary
and oriented so that there exists a fundamental cycle which allows us to do inter-
section theory. The moduli space of J-holomorphic maps can be compactied by
adding stable maps, whose domain is a Riemann surface which might have nodal
singularities. The stable compactication is compact and Hausdor in the C

topology dened by Gromov [Gr]. The moduli space is essentially almost complex,
so it has a natural orientation. In general, the moduli space is not of the expected
dimension and has bad singularities, but there exists a virtual fundamental cycle
which plays the role of fundamental cycles [LT1, BF, FO, LT2, Sie1]. These are
now well-established in Gromov-Witten theory.
The open Gromov-Witten invariants that I want to establish shall count J-
holomorphic maps from a bordered Riemann surface to a symplectic manifold X
as above such that the image of the boundary lies in a Lagrangian submanifold
L of X. To compactify the moduli space of such maps, Sheldon Katz and I [KL]
3
introduced stable maps in this context. The stable compactication is compact and
Hausdor in the C

topology, as in the ordinary Gromov-Witten theory. However,

orientation is a nontrivial issue in the open Gromov-Witten theory. Moreover, the
boundary is of real codimension one, so the compactied moduli space does not
close up as in the ordinary Gromov-Witten theory, and the best we can expect
is a fundamental chain.
A fundamental chain is not satisfactory for intersection theory. For example,
the Euler characteristic of a compact oriented manifold without boundary can be
dened as the number of zeros of a generic vector eld, counted with signs deter-
mined by the orientation. This number is independent of the choice of the vector
eld, and thus well-dened. For a compact oriented manifold with boundary, one
can still count the number of zeros of a generic vector eld with signs determined
by the orientation, but the number will depend on the choice of the vector eld.
Therefore, we need to specify extra boundary conditions to get a well-dened num-
ber.
1.2. Main results. Let (X, ) be a symplectic manifold of dimension 2N, and L
be a Lagrangian submanifold. To compactify the moduli space of parametrized
J-holomorphic curves in X with boundary in L, Gromov introduced cusp curves
with boundary [Gr], which I will call prestable maps. A prestable map to (X, L)
is a continuous map f : (, ) (X, L) such that f : (

) (X, L) is
J-holomorphic, where is a prestable (i.e., smooth or nodal) bordered Riemann
surface, and :

is the normalization map [KL, Denition 3.6.2].
A smooth bordered Riemann surface is of type (g, h) if it is topologically a
sphere with g handles and h holes. The boundary of consists of h disjoint cir-
cles B
1
, . . . , B
h
. We say has (n, m) marked points if there are n distinct marked
points in its interior and m
i
distinct marked points on B
i
, where m = (m
1
, . . . , m
h
),
m
i
0. By allowing nodal singularities, we have the notion of a prestable Riemann
surface of type (g, h) with (n, m) marked points and an ordering B
1
, . . . , B
h
of the
boundary components. An isomorphism between two such prestable bordered Rie-
mann surfaces is an isomorphism of prestable bordered Riemann surfaces which
preserves the marked points and ordering of boundary components. An isomor-
phism between two prestable maps f : X and f

X is an isomorphism
:

in the above sense such that f = f

. A prestable map is stable if

its automorphism group is nite. This is the analogue of Kontsevichs stable maps
[Ko] in the ordinary Gromov-Witten theory.
For H
2
(X, L; Z), = (
1
, . . . ,
h
) H
1
(L; Z)
h
, and Z, dene

M
(g,h),(n, m)
(X, L [ , , )
to be the moduli space of isomorphism classes of stable maps f : (, )
(X, L), where is a prestable bordered Riemann surface of type (g, h) with (n, m)
marked points and an ordering B
1
, . . . , B
h
of the boundary components, f

[] = ,
f

[B
i
] =
i
, i = 1, . . . , h, and (f

T
X
, (f[

T
L
) = . Here (f

T
X
, ((f[

T
L
)
is the Maslov index dened in [KL, Denition 3.3.7].
Theorem 1.1.

M
(g,h),(n, m)
(X, L [ , , ) is compact and Hausdor in the C

topology.
Here the C

topology is the one dened by Gromovs weak convergence [Gr].

The stability condition is necessary for Hausdorness. The compactness follows
4
from [Gr, Ye], which will be explained in Section 5.3. I do not claim any originality
for Theorem 1.1.
The boundary of the moduli space corresponds to degeneration of the domain or
blowup of the map. An interior node corresponds to a (real) codimension 2 stratum,
while a boundary node corresponds to a codimension 1 stratum. Blowup of the map
at an interior point leads to the well-known phenomenon of bubbling o of spheres
which is codimension 2, while blowup at a boundary point leads to bubbling o of
discs which is codimension 1. The intersection of two or more codimension 1 strata
forms a corner. It is shown in Section 6 that
Theorem 1.2.

M
(g,h),(n, m)
(X, L [ , , ) has a Kuranishi structure with corners
of (real) virtual dimension +(N 3)(2 2g h) +2n+m
1
+ +m
h
, where 2N
is the (real) dimension of X. The Kuranishi structure is orientable if L is spin or
if h = 1 and L is relatively spin (i.e., L is orientable and w
2
(T
L
) = [
L
for some
H
2
(X, Z
2
)).
The case for the disc with only boundary marked points (g = n = 0, h = 1) is
proven in [FO
3
]. I will describe briey what a Kuranishi structure with corners is
and refer to Section 6.1 for the complete denition. A chart of a Kuranishi structure
with corners is a 5-uple (V, E, , , s), where V is a smooth manifold (possibly with
corners), is a nite group acting on V , E is a -equivariant vector bundle over V ,
s : V E is a -equivariant section, and maps s
1
(0)/ homeomorphically to
an open set of the moduli. The dimension of V , rank of E, and the nite group
might vary with charts, but d = dimV rankE is xed and is the virtual dimension
of the Kuranishi structure with corners. det(TV ) (det E)
1
can be glued to an
orbibundle, the orientation bundle, and the Kuranishi structure with corners is
orientable if its orientation bundle is a trivial real line bundle.
If s intersects the zero section transversally, s
1
(0) is a manifold (possibly with
corners) of dimension d. In general, s
1
(0) might have dimension larger than d
and bad singularities due to the nontransversality of s. The virtual fundamental
chain can be constructed by perturbing s to a transversal section. Locally it is a
singular chain with rational coecients in V/ which is a rational combination of
the images of d-dimensional submanifolds of V .
A virtual fundamental chain is not satisfactory for intersection theory. For ex-
ample, when X is a Calabi-Yau threefold and L is a special Lagrangian submani-
fold,

M
(g,h),(0,

0)
(X, L [ , , ) is empty for ,= 0, and the expected dimension of

M
(g,h),(0,

0)
(X, L [ , , 0) is zero for any g, h, , . The virtual fundamental chain
is a zero chain with rational coecients, and we would like to dene the invariant

(g,h)
(X, L [ , , ) Q to be the degree of this zero chain. However, this number
depends on the perturbation, so we need to impose extra boundary conditions to
obtain a well-dened number. Now assume that
There is an S
1
action : S
1
X X which preserves J and L.
The restriction of to L is free.
The virtual dimension of

M
(g,h),(0,

0)
(X, L [ , , ) is zero.
Under the above assumptions I can, using the S
1
action , impose boundary con-
ditions to get a well-dened rational number

(g,h)
(X, L, [ , , )
5
which is an invariant of the equivariant pair (X, L, ), but not an invariant of
the pair (X, L). I conjecture that these rational numbers are the ones computed
by localization techniques using the S
1
action [KL, GZ]. The computations in
[KL, LS, GZ] coincide with physicists predictions.
Acknowledgments: First and foremost, I would like to thank my thesis advisor
Shing-Tung Yau for leading me to the eld of mirror symmetry and providing the
best environment to learn its newest developments. Secondly, I thank Cliord
Taubes and Gang Liu for answering my questions on symplectic geometry and
carefully reading the draft. I thank Sheldon Katz for being so generous to cooperate
with me. The cooperation [KL] is a very instructive experience for me and led me
to this project. I thank Cumrun Vafa for suggesting to me this fruitful problem and
patiently explaining his works to me. I thank Jason Starr for being an incredibly
generous and patient mentor of algebraic geometry. I thank Xiaowei Wang for
always being a source of mathematical knowledge and moral support. I thank
Arthur Greenspoon for his numerous valuable suggestions on the rst draft, and
Chien-Hao Liu for his meticulous proofreading. I thank Spiro Karigiannis for his
great help on my English. In addition, it is a pleasure to thank William Abiko,
Selman Akbulut, Raoul Bott, Kevin Costello, Yong Fu, Kenji Fukaya, Tom Graber,
Irwin Kra, Kefeng Liu, Curtis McMullen, Maryam Mirzakhani, Yong-Geun Oh,
Kaoru Ono, Scott Wolpert, and Eric Zaslow for helpful conversations. Finally,
special thanks go to Ezra Getzler for corrections and renements of the part on
moduli spaces of bordered Riemann surfaces.
2. Surfaces with Analytic or Dianalytic Structures
In this section, we review some denitions and facts of surfaces with analytic or
dianalytic structures, following [AG, Chapter 1] closely. This section is an expansion
of Section 3.1 and 3.2 of [KL].
The marked bordered Riemann surfaces dened in Section 2.2.5 are directly re-
lated to open Gromov-Witten theory.
2.1. Analyticity and dianalyticity.
Denition 2.1. Let A be a nonempty open subset of C, f : A C a map. f is
analytic on A if
f
z
= 0, and antianalytic on A if
f
z
= 0. f is said to be dianalytic
if its restriction to each component of A is either analytic of antianalytic.
Denition 2.2. Let A and B be nonempty subsets of C
+
= z C [ Imz 0. A
continuous function f : A B is analytic (resp. antianalytic on A if it extends
to an analytic (resp. antianalytic) function f
C
: U C, where U is an open
neighborhood of A in C. f is said to be dianalytic on A if its restriction to each
component of A is either analytic or antianalytic.
Theorem 2.3 (Schwarz reection principle). Let A and B be nonempty subsets
of C
+
= z C [ Imz 0. A continuous function f : A B is dianalytic
(resp. analytic) if it is dianalytic (resp. analytic) on the interior of A and satises
f(A R) R.
Denition 2.4. A surface is a Hausdor, connected, topological space together
with a family / = (U
i
,
i
) [ i I such that U
i
[ i I is an open covering of
and each map
i
: U
i
A
i
is a homeomorphism onto an open subset A
i
of C
+
.
6
/ is called a topological atlas on , and each pair (U
i
,
i
) is called a chart of /.
The boundary of is the set
= x [ i I s.t. x U
i
,
i
(x) R

ij

i

1
j
:
j
(U
i
U
j
)
i
(U
i
U
j
) are surjective homeomorphisms, called
the transition functions of /. / is called a dianalytic (resp. analytic) atlas if all
its transition functions are dianalytic (resp. analytic).
2.2. Various categories of surfaces.
2.2.1. Riemann surfaces.
Denition 2.5. A Riemann surface is a surface equipped with the analytic structure
induced by an analytic atlas on it.
A Riemann surface is canonically oriented by its analytic structure.
2.2.2. Symmetric Riemann surfaces.
Denition 2.6. A symmetric Riemann surface is a Riemann surface together
with an antiholomorphic involution : . is called the symmetry of .
Denition 2.7. A morphism between symmetric Riemann surfaces (, ) and
(

) is an analytic map f :

such that f =

f.
Denition 2.8. A symmetric Riemann surface with (n, m) marked points is a sym-
metric Riemann surface (, ) together with 2n+m distinct points p
1
, . . . , p
2n+m
in
such that (p
i
) = p
n+i
for i = 1, . . . , n and (p
i
) = p
i
for i = 2n+1, . . . , 2n+m.
2.2.3. Klein surfaces.
Denition 2.9. A Klein surface is a surface equipped with the dianalytic structure
induced by a dianalytic atlas on it.
A Riemann surface can be viewed as a Klein surface. A Klein surface can be
equipped with an analytic structure compatible with the dianalytic structure if and
only if it is orientable. In particular, an orientable Klein surface without boundary
admits a compatible structure of a Riemann surface.
Denition 2.10. A morphism between Klein surfaces and

is a continu-
ous map f : (, ) (

) such that for any x there exist analytic

charts (U, ) and (V, ) about x and f(x) respectively, and an analytic function
F : (U) C such that the following diagram commutes:
U
f
V

(U)
F
C

C
+
where (x +iy) = x +i[y[ is the folding map.
Given a Klein surface , there are three ways to construct an unramied double
cover of . We refer to [AG, 1.6] for the precise denition of an unramied double
cover and detailed constructions. The complex double
C
is an orientable Klein
surface without boundary. The orienting double
O
is an orientable Klein surface.
It is disconnected if and only if is orientable, and it has nonempty boundary if
7
and only of does. The Schottkey double
S
is a Klein surface without boundary.
It is disconnected if and only if the boundary of is empty, and it is nonorientable
if and only if is.
If is orientable, then
C
=
S
, and
O
is disconnected (the trivial double
cover). If = , then
C
=
O
, and
S
is disconnected. In particular, if
comes from a Riemann surface, then all three covers are the trivial disconnected
double cover.
Example 2.11. Let be a M obius strip. Then
C
is a torus,
S
is a Klein bottle,
and
O
is an annulus.
2.2.4. Bordered Riemann surfaces.
Denition 2.12. A bordered Riemann surface is a compact surface with nonempty
boundary equipped with the analytic structure induced by an analytic atlas on it.
Remark 2.13. A bordered Riemann surface is canonically oriented by the analytic
(complex) structure. In the rest of this paper, the boundary circles B
i
of a bordered
Riemann surface with boundary = B
1
. . . B
h
will always be given the
orientation induced by the complex structure, which is a choice of tangent vector to
B
i
such that the basis (the tangent vector of B
i
, inner normal) for the real tangent
space is consistent with the orientation of induced by the complex structure.
Denition 2.14. A morphism between bordered Riemann surfaces and

is
a continuous map f : (, ) (

) such that for any x there exist

analytic charts (U, ) and (V, ) about x and f(x) respectively, and an analytic
function F : (U) C such that the following diagram commutes:
U
f
V

(U)
F
C
A bordered Riemann surface is topologically a sphere with g 0 handles and
with h > 0 discs removed. Such a bordered Riemann surface is said to be of type
(g, h).
A bordered Riemann surface can be viewed as a Klein surface. Its complex
double and Schottkey double coincide since it is orientable.
2.2.5. Marked bordered Riemann surfaces. The following renement of an earlier
denition is suggested to the author by Ezra Getzler.
Denition 2.15. Let h be a positive integer, g, n be nonnegative integers, and m =
(m
1
, . . . , m
h
) be an h-uple of nonnegative integers. A marked bordered Riemann
surface of type (g, h) with (n, m) marked points is an (h + 3)-uple
(, B; p; q
1
, . . . , q
h
)
whose components are described as follows.
is a bordered Riemann surface of type (g, h).
B = (B
1
, . . . , B
h
), where B
1
, . . . , B
h
are connected components of , ori-
ented as in Remark 2.13.
p = (p
1
, . . . , p
n
) is an n-uple of distinct points in

.
q
i
= (q
i
1
, . . . , q
i
m
i
) is an m
i
-uple of distinct points on the circle B
i
.
8
Let

0 = (0, . . . , 0). Note that a marked bordered Riemann surface of type (g, h)
with (n,

0) marked points is a bordered Riemann surface togther with an ordering

of the h boundary components.
Denition 2.16. A morphism between marked bordered Riemann surfaces of type
(g, h) with (n, m) marked points
(, B; p; q
1
, . . . , q
h
) (

, B

; p

; (q

)
1
, . . . , (q

)
h
)
is an isomorphism of bordered Riemann surface f :

such that f(B

i
) = (B

)
i
for i = 1, . . . , h, f(p
j
) = p

j
for j = 1, . . . , n, and f(q
i
k
) = (q

)
i
k
for k = 1, . . . , m
i
.
Remark 2.17. The category of marked bordered Riemann surfaces of type (g, h)
with (n, m) marked points is a groupoid since every morphism in Denition 2.16 is
an isomorphism.
2.3. Topological types of compact symmetric Riemann surfaces. A com-
pact symmetric Riemann surface is topologically a compact orientable surface with-
out boundary together with an orientation reversing involution , which is clas-
sied by the following three invariants:
(1) The genus g of .
(2) The number h = h() of connected components of

, the xed locus of .

(3) The index of orientability, k = k() 2the number of connected compo-
nents of

.
These invariants satisfy:
(1) 0 h g + 1.
(2) For k = 0, h > 0 and h g + 1 (mod 2).
(3) For k = 1, 0 h g.
The above classication was realized already by Felix Klein (see e.g. [Kl], [We],
[Se]). This classication is probably better understood in terms of the quotient
Q() = /), where ) = id, is the group generated by . The quotient
Q() is orientable if k = 0 and nonorientable if k = 1, hence the name index
of orientability. Furthermore, h is the number of connected components of the
boundary of Q(). If Q() is orientable, then it is topologically a sphere with
g 0 handles and with h > 0 discs removed, and the invariants of (, ) are
( g, h, k) = (2g + h 1, h, 0). If Q() is nonorientable, then it is topologically a
sphere with g > 0 crosscaps and with h 0 discs removed, and the invariants of
are ( g, h, k) = (g +h 1, h, 1).
From the above classication we see that symmetric Riemann surfaces of a given
genus g fall into [
3 g+4
2
] topological types.
3. Deformation theory of Bordered Riemann Surfaces
In this section, we study deformation theory of bordered Riemann surfaces. We
refer to [KL, Section 3] for some preliminaries such as doubling constructions and
the Riemann-Roch theorem for bordered Riemann surfaces.
3.1. Deformation theory of smooth bordered Riemann surfaces. Let be
a bordered Riemann surface, (
C
, ) be its complex double (see e.g. [KL, Section
3.3.1] for the denition). Analytically, (
C
, ) is a compact symmetric Riemann
surface. Algebraically, it is a smooth complex algebraic curve X which is the
complexication of some smooth real algebraic curve X
0
, i.e., X = X
0

R
C (see
9
[Ha, Chapter II, Exercise 4.7]). Alternatively, (X, S) is a complex algebraic curve
with a real structure (see [Sil, I.1]), where S is a semi-linear automorphism in the
sense of [Ha, Chapter II, Exercise 4.7] which induces the antiholomorphic involution
on
C
.
3.1.1. Algebraic approach. First order deformations of the complex algebraic curve
X is canonically identied with the complex vector space Ext
1
OX
(
X
, O
X
), where

1
X
is the sheaf of Kahler dierentials on X. The obstruction lies in Ext
2
OX
(
X
, O
X
) =
0. Similarly, the rst order deformation of the real algebraic curve X
0
is identied
with the real vector space Ext
1
OX
0
(
X0
, O
X0
), and the obstruction vanishes. We
have
Ext
1
OX
(
X
, O
X
)

= Ext
1
OX
0
(
X0
, O
X0
)
R
C
since X = X
0

R
C. The semi-linear automorphismS induces a complex conjugation
S : Ext
1
OX
(
X
, O
X
) Ext
1
OX
(
X
, O
X
).
The xed locus Ext
1
OX
(
X
, O
X
)
S
gives the rst order deformation of (X, S) as
a complex algebraic curve with a real structure, and is naturally isomorphic to
Ext
1
OX
0
(
X0
, O
X0
).
More explicitly, X can be covered by complex ane curves which is a complete
intersection of hypersurfaces dened by polynomials with real coecients. Defor-
mation of X are given by varying the coeents (in C). Deformation of (X, S) is
given by varying the coeents in R. The above polynomials with real coecients
also dene the real algebraic curve X
0
, and varying the coeents in R gives the de-
formation of X
0
. The complex conjugation of coecients corresponds to the above
complex conjugation S on Ext
1
OX
(
X
, O
X
).
X is a smooth algebraic variety, so Ext
1
OX
(
X
, O
X
) is isomorphic to the sheaf
cohomology group H
1
(X,
X
), where
X
is the tangent sheaf of X, and
Ext
2
OX
(
X
, O
X
)

= H
2
(X,
X
) = 0.
Similarly, we have
Ext
1
OX
0
(
X0
, O
X0
)

= H
1
(X,
X0
), Ext
2
OX
0
(
X0
, O
X0
)

= H
2
(X,
X0
) = 0.
We now return to the original bordered Riemann surface .
Denition 3.1. Let O

be the sheaf of local holomorphic functions on with real

boundary values.
Let

be a sheaf of O

-modules, together with an R derivation d : O

,
which satisfy the following universal property: for any sheaf of O

-modules T, and
for any R derivation d

: O

T, there exists a unique O

-module homomorphism
f :

T such that d

= f d. We call

the sheaf of Kahler dierentials on .

Let

= Hom
O
(

, O

) =

, the dual of

in the category of sheaves of

O

-modules.
Note that

and

are locally free sheaves of O

-modules of rank 1. Analyt-

ically,

is the sheaf of local holomorphic 1-forms on whose restriction to

are real 1-forms, and

is the sheaf of holomorphic vector elds with boundary

values in T

. There are natural isomorphisms

Ext
i
O
(
1

, O

= Ext
i
OX
0
(
1
X0
, O
X0
), H
i
(,

= H
i
(X
0
,
X0
)
10
for i 0. Since the rst order deformations of the bordered Riemann surface ,
of the symmetric Riemann surface (
C
, ), and of the real algebraic curve X
0
are
identied, the rst order deformation of is canonically identied with
Ext
1
O
(

, O

= H
1
(,

),
and the obstruction lies in
Ext
2
O
(

, O

= H
2
(,

) = 0.
3.1.2. Analytic approach. We rst give the denitions of a dierentiable family of
compact symmetric Riemann surfaces and a dierentiable family of bordered Rie-
mann surfaces, which are modications of [Ko, Denition 4.1].
Denition 3.2. Suppose given a compact symmetric Riemann surface (M
t
,
t
) for
each point t of a domain B of R
m
. (M
t
,
t
)[t B is called a dierentiable family
of symmetric Riemann surfaces if there are a dierentiable manifold /, a surjective
C

map : /B and a C

map : // such that

(1) The rank of the Jacobian matrix of is equal to m at every point of /.
(2) For each t B,
1
(t) is a compact connected subset of /.
(3)
1
(t) = M
t
.
(4) There are a locally nite open covering |
i
[ i I of / and C

functions
z
i
: |
i
C such that

_
|
i

1
(t), z
i
[
Ui
1
(t)
_
[ i I, |
i

1
(t) ,=
is an analytic atlas for M
t
.
(5) = , and [

1
(t)
=
t
: M
t
M
t
is an antiholomorphic involution.
Denition 3.3. Suppose given a bordered Riemann surface M
t
for each point t
of a domain B of R
m
. M
t
[ t B is called a dierentiable family of bordered
Riemann surfaces if there are a dierentiable manifold with boundary / and a
surjective C

map : /B such that

(1) The rank of the Jacobian matrix of is equal to m at every point of /.
(2) For each t B,
1
(t) is a compact connected subset of /.
(3)
1
(t) = M
t
.
(4) There are a locally nite open covering |
i
[ i I of / and C

functions
z
i
: |
i
C
+
such that

_
|
i

1
(t), z
i
[
Ui
1
(t)
_
[ i I, |
i

1
(t) ,=
is an analytic atlas for M
t
.
The complex double
C
of a bordered Riemann surface is a complex manifold
of dimension 1. Innitesimal deformation of
C
can be identied with
H
1
(
C
, T

C
) = H
1
(X,
X
),
and the obstruction lies in
H
2
(
C
, T

C
) = H
2
(X,
X
) = 0.
(See [Ko].) The dierential d of is an antiholomorphic involution on the holo-
morphic line bundle T

C

C
which covers :
C

C
. d induces a complex
conjugation
: H
1
(
C
, T

C
) H
1
(
C
, T

C
)
11
which gets identied with the action of S on Ext
1
OX
(
X
, O
X
) under the isomor-
phism
H
1
(
C
, T

C
)

= Ext
1
OX
(
X
, O
X
).
The pair (T

C
, d) is the holomorphic complex double of the Riemann-Hilbert
bundle (T

, T

) (, ), where a Riemann-Hilbert bundle and its holomorphic

complex double are dened in [KL, Section 3.3.4]. There is an isomorphism (see
[KL, Section 3.4])
H
1
(
C
, T

C
)


= H
1
(, , T

, T

).
From the above discussion, we know that H
1
(
C
, T

C
)

gives rst order deformation
of the symmetric Riemann surfaces (
C
, ). Given a dierential family of symmetric
Riemann surface (

t
,
t
) such that (

0
,
0
) = (
C
, ), K
t
=

t
/
t
) is a family of
Klein surfaces. Each K
t
is homeomorphic to , which is orientable, so it admits two
analytic structures compatible with its dianalytic structure, and one is the complex
conjugate of the other. We get two dierentiable families M
t
, M

t
=

M
t
of bordered
Riemann surfaces, one is a deformation of , and the other is a deformation of

.
So H
1
(, , T

, T

) should give innitesimal deformations of .

Recall that an innitesimal deformation of
C
determines a

Cech 1 cocycle in
H
1
(

/,
C
) H
1
(
C
, T

C
), where

/ is an analytic atlas of
C
, and
C
is the sheaf
of local holomorphic vector elds on
C
[Ko]. (The inclusion is an isomorphism
if

/ is acyclic). Following argument similar to that in [Ko], we now show that an
innitesimal deformation of determines a

Cech 1 cocycle in H
1
(/,

), where /
is an analytic atlas of , and

is the sheaf of local holomorphic vector elds on

T

with boundary values in T

.
Let M
t
[ t B be a dierentiable family of bordered Riemann surfaces,
M
0
= . We use the notation in Denition 3.3. Then
/ = (U
i
,
i
)
_
|
i

1
(0), z
i
[
Ui
1
(t)
_
[ i I, |
i

1
(t) ,=
is an analytic atlas of . Without loss of generality, we may assume that /is acyclic.
We dene t-dependent transition functions f
ij
by z
i
= f
ij
(z
j
, t) = f
ik
(z
k
, t). Then
f
ij
(z
j
, t) R if z
j
R by 4. of Denition 3.3.
z
i
= f
ik
(z
k
, t) = f
ij
(f
jk
(z
k
, t), t)
f
ik
t
=
f
ij
z
j
f
jk
t
+
f
ij
t
Multiplying by

zi
and noting that
fij
zj
=
zi
zj
, we have
f
ik
t

z
i
=
f
jk
t

z
j
+
f
ij
t

z
i
Let (x
i
, y
i
) be real coordinates dened by z
i
= x
i
+ iy
i
, then the boundary is
dened by y
i
= 0, and the tangent line to the boundary is spanned by

xi
.
Under the isomorphism T

T
0,1

given by v (v iJv)/2, we have

x


z
and

y
i

z
. So
ik

f
ik
t

t=0

zi
denes a

Cech 1 cochain in C
1
(/,

) which
satises the cocycle condition
ik
=
ij
+
jk
.
By exactly the same argument in [Ko] we see that another system of coordinates
will give rise to a

Cech 1 cocycle

= +, where is a

Cech 0 cochain. Therefore,
12
1
0
0
0
1 1
0
1
0 0
2
Figure 1. strata of

M
2
the innitesimal deformation of is given by
H
1
(/,

) = H
1
(, , T

, T

).
3.2. Nodal bordered Riemann surfaces. To compactify the moduli of bordered
Riemann surface, we will allow nodal singularities. The complex double of a bor-
dered Riemann surface is a complex algebraic curve with real structure. The stable
compactication of moduli of such curves parametrizes stable complex algebraic
curves with real structure [SS, Se], or equivalently, stable compact symmetric Rie-
mann surfaces. Naively, the quotient of a stable compact symmetric Riemann sur-
face by its antiholomorphic involution will give rise to a stable bordered Riemann
surface. We will make this idea precise in this section.
Let be a (smooth) bordered Riemann surface of type (g, h). Note that if
: is an automorphism (Denition 2.14), then its complex double ([KL,
Section 3.3.2])
C
:
C

C
is an automorphism of (
C
, ) (Denition 2.7). This
gives an inclusion Aut() Aut(
C
, ). It is easy to see that the following are
equivalent:
is stable, i.e., Aut() is nite.

C
is stable.
The genus g = 2g +h 1 of
C
is greater than one.
The Euler characteristic () = 2 2g h of is negative.
We start with g = 2. Let

M
2
be the moduli of stable complex algebraic curves
of genus 2. The strata of

M
2
are shown in Figure 1.
The dual graph and the underlying topological surface of a prestable curve are
shown. In the dual graph of the curve C, each vertex corresponds to an irreducible
13
1
1
3
3
3
3 6
3
6 3
6
1
3 3
Figure 2. strata of

M
(0,3)
. Each stable bordered Riemann sur-
face above represents a topological type, and the number above
it is the number of strata associated to it. Strata associated
to the same topological type are related by relabelling the three
boundary circles. There are one 3-dimensional stratum, nine 2-
dimensional strata, twenty-one 1-dimensional strata, and fourteen
0-dimensional strata. We will see in Example 4.7 that

M
(0,3)
can
be identied with the associahedron K
5
dened by J. Stashe [St].
component of C, labeled by the genus of the normalization, while each edge cor-
responds to a node of C, whose two end points correspond to the two irreducible
components which intersect at this node.
If g = 2, (g, h) can be (0, 3) (Figure 2) or (1, 1) (Figure 3).
Denition 3.4. Let (x, y) be coordinates on C
2
, and A(x, y) = ( x, y) be the complex
conjugation. A node on a bordered Riemann surface is a singularity isomorphic to
one of the following:
(1) (0, 0) xy = 0 (interior node)
(2) (0, 0) x
2
+y
2
= 0/A (boundary node of type E)
(3) (0, 0) x
2
y
2
= 0/A (boundary node of type H)
A nodal bordered Riemann surface is a singular bordered Riemann surface whose
singularities are nodes.
A type E boundary node on a bordered Riemann surface corresponds to a bound-
ary component shrinking to a point, while a type H boundary node corresponds
14
Figure 3. strata of

M
(1,1)
to a boundary component intersecting itself or another boundary component. The
boundary of a nodal Riemann surface is a union of points and circles, where one
circle might intersect other circles in nitely many points.
The notion of morphisms and complex doubles can be easily extended to nodal
bordered Riemann surfaces. The complex double of a nodal bordered Riemann
surface is a nodal compact symmetric Riemann surface.
Example 3.5. Consider C

= X
2
Y
2
+ Z
2
= 0 P
2
, where [X, Y, Z] are
homogeneous coordinates on P
2
, and R. C

is invariant under the standard

complex conjugation A([X, Y, Z]) = [

X,

Y ,

Z] on P
2
, so

= A[
C
is an antiholo-
morphic involution on C

. For ,= 0, (C

) is a symmetric Riemann surface of

type (0, 1, 0), and C

) is the disc, which is a bordered Riemann surface. C

0
has
two irreducible components X + Y = 0 and X Y = 0 which are projective
lines, and the intersection point [0, 0, 1] of the two lines is a node on C
0
. Both lines
are invariant under the antiholomorphic involution
0
. C
0
/
0
) is a nodal bordered
Riemann surface: it is the union of two discs whose intersection is a boundary node
of type H.
Example 3.6. Consider C

= X
2
+ Y
2
+ Z
2
= 0 P
2
, where R. C

is
invariant under the complex conjugation A on P
2
, so

= A[
C
is an antiholo-
morphic involution on C

. Set

= C

). For > 0, (C

) is a symmetric
Riemann surface of type (0, 0, 1), and

is the real projective plane; for < 0,

(C

) is a symmetric Riemann surface of type (0, 1, 0), and

is the disc. C
0
has two irreducible components X +

1Y = 0 and X

1Y = 0 which
are projective lines, and their intersection point [0, 0, 1] is a node on C
0
. The an-
tiholomorphic involution
0
interchanges the two irreducible components of C
0
and
leaves the node invariant, so
0

= P
1
, which is a smooth Riemann surface without
15
normalization degeneration
(g, h) = (1, 3)
(g, h) = (1, 2)
Figure 4. boundary node of type E
degeneration normalization
(g, h) = (1, 3)
(g, h) = (1, 2)
Figure 5. boundary node of type H1
boundary. However, we would like to view it as a disc whose boundary shrinks to a
point which is a boundary node of type E.
Denition 3.7. Let be a nodal bordered Riemann surface. The antiholomorphic
involution on its complex double
C
can be lifted to :

C
, where

C
is
the normalization of
C
(viewed as a complex algebraic curve). The normalization
of =
C
/) is dened to be

=

C
/ ).
From the above denition, the complex double of the normalization is the nor-
malization of the complex double, i.e.,

C
=

C
.
Let be a smooth bordered Riemann surfaces of type (g, h). The following are
possible degenerations of whose only singularity is a boundary node.
E. One boundary component shrinks to a point. The normalization is a smooth
bordered Riemann surface of type (g, h 1) (Figure 4).
H1. Two boundary components intersect at one point. The normalization is a
smooth bordered Riemann surface of type (g, h1), and the two preimages
of the node are on the same boundary component (Figure 5).
H2. One boundary component intersects itself, and the normalization of the
surface is connected. The normalization is a smooth bordered Riemann
surface of type (g 1, h + 1), and the two preimages of the node are on
dierent boundary components (Figure 6).
H3. One boundary component intersects itself, and the normalization of the
surface is disconnected. The normalization is a disjoint union of two smooth
16
degeneration normalization
(g, h) = (1, 3) (g, h) = (0, 4)
Figure 6. boundary node of type H2
degeneration normalization
(g, h) = (1, 3) (g
1
, h
1
) = (1, 2) (g
2
, h
2
) = (0, 2)
Figure 7. boundary node of type H3
bordered Riemann surfaces of types (g
1
, h
1
) and (g
2
, h
2
) such that g =
g
1
+ g
2
and h = h
1
+ h
2
1, and each connected component contains one
of the two preimages of the node (Figure 7).
Denition 3.8. A prestable bordered Riemann surface is either a smooth bordered
Riemann surface or a nodal bordered Riemann surface.
Let be a prestable bordered Riemann surface,

be its normalization. Let

C
1
, . . . ,

C

1
, . . . ,

be the connected components of

, where

C
i
is a smooth
Riemann surface of genus g
i
, and

i
is a smooth bordered Riemann surface of
type (g
i
, h
i
). Let be the number of connecting interior nodes (Figure 8), and

E
,
H1
,
H2
,
H3
be the numbers of boundary nodes described in E, H1, H2, H3,
respectively.
The topological type (g, h) of is given by
g = g
1
+ + g

+g
1
+ +g

+ +
H2
h = h
1
+ +h

+
E
+
H1

H2

H3
It is now straightforward to extend the notion of marked bordered Riemann
surfaces to prestable bordered Riemann surfaces.
Denition 3.9. Let h be a positive integer, g, n be nonnegative integers, and m =
(m
1
, . . . , m
h
) be an h-uple of nonnegative integers. A prestable marked bordered
Riemann surface of type (g, h) with (n, m) marked points is an (h + 3)-uple
(, B; p; q
1
, . . . , q
h
)
whose components are described as follows.
17
connecting interior node disconnecting interior node
(g, h) = (2, 1)
Figure 8. connecting and disconnecting nodes
is a prestable bordered Riemann surface of type (g, h).
B = (B
1
, . . . , B
h
), where =

h
i=1
B
i
, and each B
i
is an immersed circle.
The circles B
1
, . . . , B
h
may intersect with each other at boundary nodes,
and become h disjoint embedded circles under smoothing of all boundary
nodes.
p = (p
1
, . . . , p
n
) is an n-uple of distinct smooth points in

.
q
i
= (q
i
1
, . . . , q
i
m
i
) is an m
i
-uple of distinct smooth points on B
i
.
Denition 3.10. A morphism between prestable marked bordered Riemann sur-
faces of type (g, h) with (n, m) marked points
(, B; p; q
1
, . . . , q
h
) (

, B

; p

; (q

)
1
, . . . , (q

)
h
)
is an isomorphism of prestable bordered Riemann surfaces :

such that
(B
i
) = (B

)
i
for i = 1, . . . , h, (p
j
) = p

j
for j = 1, . . . , n, and (q
i
k
) = (q

)
i
k
for
k = 1, . . . , m
i
. A prestable marked bordered Riemann surfaces of type (g, h) with
(n, m) marked points is stable if its automorphism group is nite.
Remark 3.11. The category of prestable marked bordered Riemann surfaces of
type (g, h) with (n, m) marked points is a groupoid since every morphism in De-
nition 3.10 is an isomorphism.
Example 3.12. Consider the case (g, h) = (0, 2), n = 0, m = (1, 0) (annuli with
one boundary marked point). The moduli space

M
(0,2)(0,(1,0))
is an interval [0, 1].
There are three strata: t (0, 1), t = 0, t = 1 (Figure 9)
Example 3.13. Consider the case (g, h) = (0, 2), n = 0, m = (2, 0) (annuli with
two boundary marked points on the same boundary circle). The moduli space

M
(0,2)(0,(2,0))
is a pentagon. There are eleven strata (Figure 10).
Example 3.14. Consider the case (g, h) = (0, 2), n = 0, m = (1, 1) (annuli with
one marked point on each boundary circle). The moduli space

M
(0,2)(0,(1,1))
is a
disc z C [ [z[ 1. There are four strata: 0 < [z[ < 1, [z[ = 1 but z ,= 1, z = 1,
z = 0 (Figure 11)
3.3. Deformation theory for prestable bordered Riemann surfaces. The
algebraic approach of deformation theory for smooth bordered Riemann surfaces
in Section 3.1 can be easily extended to nodal bordered Riemann surfaces. We will
also consider marked points.
18
1
1 1
t = 0 t = 1
t (0, 1)
Figure 9. strata of

M
(0,2)(0,(1,0))
1
2
1
2
1
2
1
2
1
2
1
2
1
2
1 2
1
2
1
2
1
2
Figure 10. strata of

M
(0,2)(0,(2,0))
Let (, B; p; q
1
, . . . , q
h
) be a marked prestable bordered Riemann surface of type
(g, h) with (n, m) marked points. We want to study its innitesimal deformation.
The ordering of the boundary circles is irrelevant to the innitesimal deformation,
so in this section we will ignore it and write (; p; q), where
q = (q
1
, . . . , q
m
) = (q
1
1
, . . . , q
1
m
1, q
2
1
, . . . , q
2
m
2 , . . . , q
h
1
, . . . , q
h
m
h),
19
0 < [z[ < 1
[z[ = 1, z ,= 1
z = 0 z = 1
Figure 11. strata of

M
(0,2)(0,(1,1))
and m = m
1
+ +m
h
.
The complex double (
C
, ) of is a nodal complex algebraic curve X which
is the complexication of some real algebraic curve X
0
, i.e., X = X
0

R
C. Al-
ternatively, (X, S) is a complex algebraic curve with a real structure, where S is a
semi-linear automorphism which induces the antiholomorphic involution on
C
.
Algebraically, the complex double of (; p; q) is a nodal complex algebraic curve
with (2n +m) marked points (X, x), where
x = (x
1
, . . . , x
2n+m
) = (p
1
, . . . , p
n
, p
1
, . . . , p
n
, q
1
, . . . , q
m
).
Here we identify with the image under the inclusion i :
C
, and denote (p)
by p.
Let
D
x
= x
1
+ +x
2n+m
be the divisor in X associated to x. The set of rst order deformation of the
pointed complex algebraic curve (X, x) is canonically identied with the complex
vector space
Ext
1
OX
(
X
(D
x
), O
X
),
and the obstruction lies in
Ext
2
OX
(
X
(D
x
), O
X
).
We claim that Ext
2
OX
(
X
(D
x
), O
X
) = 0. Three terms in the local to global
spectrum sequence contribute to Ext
2
OX
(
X
(D
x
), O
X
):
H
0
(X, cxt
2
OX
(
X
(D
x
), O
X
)),
H
1
(X, cxt
1
OX
(
X
(D
x
), O
X
)),
H
2
(X, cxt
0
OX
(
X
(D
x
), O
X
)).
20
The curve X has only nodal singularities, so
cxt
2
OX
(
X
(D
x
), O
X
) = 0,
thus
H
0
(X, cxt
2
OX
(
X
(D
x
), O
X
)) = 0.
The sheaf cxt
1
OX
(
X
(D
x
), O
X
) is supported on nodes, so
H
1
(X, cxt
1
OX
(
X
(D
x
), O
X
)) = 0.
Finally,
H
2
(X, cxt
0
OX
(
X
(D
x
), O
X
)) = 0
since X is one dimensional. So we get the desired vanishing.
The semi-linear automorphism S induces a complex conjugation
S : Ext
1
OX
(
X
(D
x
), O
X
) Ext
1
OX
(
X
(D
x
), O
X
).
The xed locus Ext
1
OX
(
X
(D
x
), O
X
)
S
gives the rst order deformation of (X, x, S)
as a pointed complex algebraic curve with a real structure.
We will study the group Ext
1
OX
(
X
(D
x
), O
X
) and the action of S on it more
closely. We rst introduce some notation.
Let C
1
, ..., C

be the irreducible components of which are (possibly nodal)

Riemann surfaces, and let
1
, ...,

be the remaining irreducible components of

, which are (possibly nodal) bordered Riemann surfaces. Then the irreducible
components of X are
C
1
, . . . , C

,

C
1
, . . . ,

C

, (
1
)
C
, . . . , (

)
C
.
Let

C
i
denote the normalization of C
i
, i = 1, . . . , , and let

i
denote the
normalization of
i
, i

= 1, . . . ,

. Then

C
1
, . . . ,

C

1
, . . . ,

are the connected components of

, the normalization of , and

C
1
, . . . ,

C

C
1
, . . . ,

, (

1
)
C
, . . . , (

)
C
.
are the connected components of

X, the normalization of X.
Let r
1
, . . . , r
l0

be interior nodes of , and s

1
, . . . , s
l1
be boundary
nodes of . Then X has 2l
0
+l
1
nodes r
1
, . . . , r
l0
, r
1
, . . . , r
l0
, s
1
, . . . , s
l1
.
Let p
j


be the preimage of p
j
under the normalization map :

X X, j =
1, . . . , n. Dene

p
j
, q
j
similarly. Let r

, r
l0+
be the preimages of r

, = 1, . . . , l
0
,
and dene s

, s
l1+
similarly.
Consider

X with marked points
x = ( p
1
, . . . , p
n
,

p
1
, . . . ,

p
n
, q
1
, . . . , q
m
, r
1
, . . . , r
2l0
,

r
1
, . . . ,

r
2l0
, s
1
, . . . , s
2l1
),
which can be written as a disjoint union of pointed curves
(

C, y
i
), (

C, y
i
), ((

i
)
C
, z
i

) = ((

i
)
C
, p
i

, p
i

, q
i

),
where
y
i
= (y
i
1
, . . . , y
i
ni
), y
i
= (y
i
1
, . . . , y
i
ni
),
p
i

= (p
i

1
, . . . , p
i

n
i

), p
i

= ( p
i

1
, . . . , p
i

n
i

), q
i

= (q
i

1
, . . . , q
i

m
i

),
21
i = 1, . . . , , i

= 1, . . . ,

, p
i

1
, . . . , p
i

n
i

i
, q
i

1
, . . . , q
i

m
i

i
, and

i=1
n
i
+

=1
n
i
= n + 2l
0
,

i=1
m
i
= m+ 2l
1
.
From the local to global spectrum sequence, we have an exact sequence
0 H
1
(X, cxt
0
OX
(
X
(D
x
), O
X
)) Ext
1
OX
(
X
(D
x
), O
X
)
H
0
(X, cxt
1
OX
(
X
(D
x
), O
X
)) 0,
where
cxt
0
OX
(
X
(D
x
), O
X
) = Hom
OX
(
X
(D
x
), O
X
) =
1
X
(D
x
)

.
We have the following elementary fact:
Lemma 3.15.

1
X
(D
x
)

X
(D
x
),
where :

X X is the normalization map, and D
x
is the divisor corresponding
to the marked points x in

X.
Proof. The equality obviously holds for smooth points. It suces to show that

1
Y
=

Y
for Y = SpecC[x, y]/(xy), which follows from a local calculation. 2
The map :

X X is an ane morphism, so by [Ha, Chapter III, Exercise 4.1]
we have
H
1
(X, cxt
0
OX
(
X
(D
x
), O
X
)) = H
1
(X,

X
(D
x
)) =

i=1
(W
i

W
i
)

=1
W

where the vector spaces

W
i
= H
1
(

C
i
, T

Ci
(D
y
i )),

W
i
= H
1
(

C
i
, T

Ci
(D
y
i )),
W

i
= H
1
((

i
)
C
, T
(

i
)
C
(D
z
i
)
correspond to deformations of pointed curves (

C
i
, y
i
), (

C
i
, y
i
), ((

i
)
C
, z
i

), respec-
tively.
Another local calculation shows that
H
0
(X, cxt
1
OX
(
X
(D
x
), O
X
))

=
l0

=1
(V

)
l1

=1
V

,
where
V

= T
r

X T
r
l
0
+

X,

V

= T
r

X T
r
l
0
+

X, V

= T
s

X T
s
l
0
+

X
correspond to smoothing of the nodes r

, r

, s

in X, respectively.
Now the action of S on Ext
1
OX
(
X
(D
x
), O
X
) is clear: it acts on W
i


W
i
and
V

by (a, b) (

b, a), and it acts on W

i
and V

by a a. The real vector

space W

S
corresponds to deformation of the pointed symmetric Riemann surface
((

i
)
C
, z
i

, ). We also have W

=

W
i
, where

W
i
= H
1
(

i
,

i
, T

(p
i

1
p
i

n
i

), T

(q
i

1
. . . q
i

m
i

))
corresponds to deformation of the pointed bordered Riemann surface (

i
, p
i

, q
i

).
22
crosscap
t>0 t=0 t<0
( g, h, k) = (4, 3, 0)
( g, h, k) = (4, 2, 1)
Figure 12. type E
t>0 t<0 t=0
=
Mobius strip
=
( g, h, k) = (4, 3, 0)
( g, h, k) = (4, 2, 1)
Figure 13. type H1
The action of S on V

can be understood by studying local models

x
2
+y
2
= 0/A (type E) ,
x
2
y
2
= 0/A (type H) ,
where the antiholomorphic involution is A(x, y) = ( x, y). The deformation of x
2

y
2
= 0 is given by x
2
y
2
= , where C is small. A acts on deformation
by , so the deformation of (x
2
y
2
= 0, A) is given by (x
2
y
2
= , A),
where R is small. x
2
y
2
= /A, R small gives a deformation of the
boundary nodes, and there is a topological transition from > 0 to < 0 (Figure
12, 13, 14, 15). C corresponds to V

, R corresponds to V

S
=

V

, and
0 corresponds to

V
+

.
Above discussion can be summarized as follows.
(1) The innitesimal deformation of the pointed complex algebraic curve (X, x)
is given by
Ext
1
OX
(
X
(D
x
), O
X
)

i=1
(W
i

W
i
)

=1
W

i

l0

=1
(V

)
l1

=1
V

.
(2) The innitesimal deformation of the pointed complex algebraic curve with
a real structure (X, x, S) is given by
Ext
1
OX
(
X
(D
x
), O
X
)
S

=

i=1
(W
i

W
i
)
S

=1
W

l0

=1
(V

)
S

l1

=1
V

S
.
23
t>0 t=0 t<0
=
=
( g, h, k) = (4, 3, 0) ( g, h, k) = (4, 3, 1)
Figure 14. type H2
=
t>0 t=0 t<0
orientation is reversed
( g, h, k) = (4, 3, 0)
( g, h, k) = (4, 3, 0)
Figure 15. type H3
(3) The innitesimal deformation of the pointed prestable bordered Riemann
surface (; p; q) is given by

i=1
W
i

=1

W
i

l0

=1
V

l1

=1

V
+

.
4. Moduli of Bordered Riemann Surfaces
For stable complex algebraic curves, the moduli of complex structures can be
identied with the moduli of hyperbolic structures. Under this identication, an-
alytic methods are applicable to the study of moduli of stable curves. In [Ab],
Abiko constructed a topology on

M
g,n
, the moduli of stable complex algebraic
curves of genus g with n marked points, and showed that

M
g,n
is compact and
Hausdor in this topology. Similarly, Seppala [Se] constructed a topology for the
moduli space of stable real algebraic curves of a given genus g > 1, and showed
that the moduli space is compact and Hausdor in the topology. In this section, we
describe how the above works can be modied to study stable bordered Riemann
surfaces.
4.1. Various moduli spaces and their relationships. Let
(, B; p; q
1
, . . . , q
h
)
24
be a marked stable bordered Riemann surface of type (g, h) with (n, m) marked
points. The moduli of the ordering of the circles B
1
, . . . , B
h
is given by the permu-
tation group of h elements. Let (
C
, ) be the complex double of , and let
x = (x
1
, . . . , x
2n+m
) = (p
1
, . . . , p
n
, p
1
, . . . , p
n
, q
1
, . . . , q
m
)
(
C
, , p
1
, x) is a stable symmetric Riemann surface of genus g = 2g + h 1 with
(n, m) marked points. Removing x
1
, . . . , x
n
from
C
, where n = 2n+m, we obtain
(S, ), a stable symmetric Riemann surface of genus g with n punctures. Let S

be the complement of nodes in S. There is a one to one correspondence between

connected components of S

and irreducible components of S. Each connected

components of S

is a smooth punctured Riemann surface. The stability condition

is equivalent to the statement that each connected component of S

has negative
Euler characteristic. Therefore, there is a unique complete hyperbolic metric in the
conformal class of Riemannian metrics on S

determined by the complex structure.

Let

M
g, n
be the moduli of stable compact Riemann surfaces of genus g with n
marked points, or equivalently, the moduli of stable complex algebraic curves of
genus g with n marked points. Let

P
g, n
be the moduli of stable oriented hyperbolic
surfaces of genus g with n punctures. From the above discussion we know that
there is a surjective map :

M
g, n


P
g, n
. is generically n! to one since the
marked points are ordered, while the punctures are not. The ber over a point in

P
g, n
represented by the surface S consists of less than n! points if and only if there
is an automorphism of S permuting its punctures.
Let

M
R
g,(n,m)
be the moduli of stable symmetric compact Riemann surface of
genus g with (n, m) points, and let M
R
( g,h,k),(n,m)
be the moduli of smooth symmetric
compact Riemann surface of type ( g, h, k) with (n, m) marked points. Note that
M
R
( g,0,k),(n,m)
is empty if m > 0. M
R
( g,h,k),(n,m)
are disjoint subsets of

M
R
g,(n,m)
, and
their closures

M
R
( g,h,k),(n,m)
(in the topology dened later) cover

M
R
g,(n,m)
.
There is an involution A :

M
g, n


M
g, n
, given by
[(, x
1
, . . . , x
n
)] [(

, (x
n+1
), . . . , (x
2n
), (x
1
), . . . , (x
n
), (x
2n+1
), . . . , (x
2n+m
))]
where :

is the canonical anti-holomorphic map from to its complex
conjugate

. Let

M
A
g, n
denote the xed locus of A. Then there is a surjective map

M
R
g,(n,m)


M
A
g, n
, given by forgetting the symmetry . This map is generically
injective. It fails to be injective exactly when the automorphism group of (, x) is
larger than that of (, , x).
Let

P
R
g, n
be the moduli of stable symmetric oriented hyperbolic surfaces of genus
g with n punctures, and let P
R
( g,h,k), n
the moduli of smooth symmetric oriented
hyperbolic surfaces of type ( g, h, k) with n punctures. P
R
( g,h,k), n
are disjoint subsets
of

P
R
g, n
, and their closures

P
R
( g,h,k), n
(in the topology to be dened later) cover

P
R
g, n
.
There is an involution A

:

P
g, n


P
g, n
, given by [S] [

S]. There is a surjective

map

P
R
g, n


P
A

g, n
, given by forgetting the symmetry . This map is generically
injective. It fails to be injective exactly when the automorphism group of S is
larger than that of (S, ).
We have the following commutative diagrams:
25

M
g, n
A

M
g, n

P
g, n
A

P
g, n

M
R
g,(n,m)


M
A
g, n

_
A

P
R
g,(n,m)


P
A

g, n
where the generic ber of
R
consists of 2
n
n!m! points. The factor 2
n
corresponds
to the permutation of the two points in each of the n conjugate pairs. The factor n!
corresponds to the permutation of the n conjugate pairs. The factor m corresponds
to the permutation of the m marked points xed by the symmetry. Similarly,
the generic ber of
A
consists of 2
n
n!m! points. For a generic point in

P
A

g, n
, its
preimage under consists of n! points, but only 2
n
n!m! lie in the xed locus of A.
Neither
R
nor
A
is surjective because the number of punctures xed by the
symmetry can be any integer between 0 and n, not only m.
We are interested in the moduli space

M
(g,h),(n, m)
of stable bordered Riemann
surfaces of type (g, h) with (n, m) marked points. There is a nite to one map

M
(g,h),(n, m)


M
R
g,(n,m)
given by complex double, so there is a nite to one map

M
(g,h),(n, m)


P
R
( g,h,0), n


P
R
g, n
. We will rst study

P
g, n
and

P
R
g, n
, following
[Ab, Se].
4.2. Decomposition into pairs of pants. A pair of pants P is a sphere from
which three disjoint closed discs (or points) have been removed. It is the interior
a stable bordered Riemann surface of type (0, 3). There is a unique hyperbolic
structure compatible with the complex structure of P such that the boundary
curves are geodesics. Conversely, given a hyperbolic structure on P such that
the boundary curves are geodesics, the conformal structure is determined up to
conformal or anti-conformal equivalence by the lengths l
1
, l
2
, and l
3
of the three
boundary curves ([Ab, Chapter II (3.1), Theorem]).
4.2.1. Riemann surfaces with punctures. A Riemann surface S of genus g with n
punctures can be decomposed into pairs of pants. More precisely, there are 3 g3+ n
disjoint curves
1
, . . . ,
3 g3+ n
on , each of which is either a closed geodesic (in
the hyperbolic metric) or a node, such that the complement of
3 g3+ n
i=1

i
is a
disjoint union of 2 g 2 + n pairs of pants P
1
, . . . , P
2 g2+ n
. A boundary component
of the closure of a pair of pants in this decomposition is either a decomposing curve
or a puncture. We call
T = P
1
, P
2
, . . . , P
2 g2+ n

a geodesic decomposition of S into pairs of pants.

Suppose that there exists an anti-holomorphic involution : S S. Then (S, )
is a stable symmetric Riemann surface of genus g with n punctures, and is an
isometry of the hyperbolic metric. Let
T = P
1
, P
2
, . . . , P
2 g2+ n

be a geodesic decomposition of S into pairs of pants. Then

(T) = (P
1
), (P
2
), . . . , (P
2 g2+ n
)
is another geodesic decomposition of S into pairs of pants. T is said to be -
invariant if (T) = T. The argument in [Se, Section 4], combined with [Ab,
Chapter II (3.3), Lemma 3], shows that
26
Theorem 4.1. Let (S, ) be a stable symmetric Riemann surface of genus g with
n punctures. There exists a -invariant geodesic decomposition of into pairs of
pants such that the decomposing curves are simple closed geodesics of length less
than C( g, n), where C( g, n) is a constant depending only on g, n.
4.2.2. Riemann surfaces with boundary and punctures. Let (, B; p) be a stable
bordered Riemann surface of type (g, h) with (n,

0) marked points, and suppose

that has no boundary nodes. Let S be the complement of marked points in ,
then S is a surface of type (g, h, n) in the sense of [Ab], namely, S is obtained
by removing h open discs and n points from a compact (possibly nodal) Riemann
surface of genus g, where the discs and points are all disjoint. S is stable in the
sense that its automorphism group is nite. Let S

be the complement of nodes in

S. Each connected component of S

is a smooth Riemann surface with boundary

or punctures. The stability condition on S is equivalent to the statement that each
connected component of S

has a negative Euler characteristic, so there exists a

unique hyperbolic metric on S

in the conformal class determined by the complex

structure such that the boundary circles are geodesics.
Let S be a stable surface of type (g, h, n). The S can be decomposed into pairs
of pants. More precisely, there are 3g +h3 +n disjoint curves
1
, . . . ,
3g+h3+n
on S, each of which is either a closed geodesic (in the hyperbolic metric) or a node,
such that the complement of
3g+h3+n
i=1

i
is a disjoint union of 2g + h 2 + n
pairs of pants P
1
, . . . , P
2g+h2+n
. A boundary component of the closure of a pairs
of pants in this decomposition is a decomposing curve, a boundary component or
a puncture. We call
T = P
1
, P
2
, . . . , P
2g+h2+n

a geodesic decomposition of S into pairs of pants. We have the following result

([Ab, Chapter II (3.3), Lemma 3]):
Theorem 4.2. Let S be a stable surface of type (g, h, n). There is a geodesic decom-
position of S into pairs of pants such that the decomposition curves are simple closed
geodesic with length less than C(g, h, n, L
1
, . . . , L
h
), where C(g, h, n, L
1
, . . . , L
h
)
is a constant depending only on g, h, n, and the lengths of the h border curves
L
1
, . . . , L
h
.
We will see later that the moduli of stable surfaces of type (g, h, n) is of (real)
dimension 6g + 3h 6 + 2n, and L
1
, . . . , L
h
are among the 6g + 3h 6 + 2n real
parameters. They are not good coordinates for compactness since L
1
, . . . , L
h
can
be arbitrarily large. Actually, the length of some border curve tends to innity as
S acquires a type H boundary node. To deal with boundary nodes and boundary
marked points, we go to the complex double, where the local coordinates of the
moduli can be chosen to be bounded.
4.3. Fenchel-Nielsen coordinates.
Denition 4.3. Let S be a stable symmetric Riemann surface of genus g with n
punctures. A geodesic pants decomposition T is oriented if
(1) The pairs of pants in T is ordered.
(2) The boundary components of each pair of pants in T is ordered.
(3) Any decomposing curve which is not a node is oriented.
27

2,3

3,1

1,2
Figure 16. a pair of pants
Remark 4.4. The orientability of a geodesic decomposition of a surface of type
(g, h, n) can be dened similarly, with the additional assumption that the boundary
components are ordered.
Let P be a pair of pants with hyperbolic structure with ordered boundary com-
ponents
1
,
2
,
3
. The base points
i
on
i
, i = 1, 2, 3, are shown in Figure 16.
In Figure 16,
1,2
is the geodesic which realizes the distance between
1
,
2
, etc.
If T is oriented, then each boundary curve of a pair of pants in T has a base point.
Each decomposing curve has two distinguished points since it appears twice as a
boundary of some pair of pants in T. The orderings in 1. and 2. of Denition 4.3
determine an ordering of the decomposing curves
j
, j = 1, . . . , 3 g 3 + n and
an ordering on the two distinguished points
1
j
,
2
j
on each decomposing curve. We
dene l
j
(T) to be the length of
j
, and
j
(T) be the distance one travels from
1
j
to

2
j
along
j
, in the direction determined by 3. in Denition 4.3. Set
j
(T) = 2
j
lj
if l
j
,= 0. We have l
j
0, 0
j
< l
j
, and 0
j
< 2.
Denition 4.5. Let

S, S be two stable Riemann surfaces of genus g with n punc-
tures. Let

S

, S

be 1-dimensional complex manifolds obtained from

S, S by re-
moving the nodes. A strong deformation :

S S is a continuous map such
that
(1) If r is a node on

S, then ( r) is a node on S.
(2) If r is a node on S, then
1
(r) is a node or an embedded circle on a
connected component of S

.
(3) f[

1
(S

)
:
1
(S

) S

is a dieomorphism.
Note that 1. implies that
1
(S

)

S

, so 3. makes sense. There is a strong

deformation :

S S if and only if

S can be obtained by deforming S as a
quasiprojective variety over C.
Remark 4.6. A strong deformation : (

S, ) (S, ) between stable symmetric

Riemann surfaces can be dened similarly, with the additional assumption that
= . A strong deformation between two surfaces of type (g, h, n) can also
be dened similarly.
We now describe Fenchel-Nielsen coordinates for various category of surfaces.
(1) Let S be a stable Riemann surface of genus g with n punctures. Let T
be an oriented geodesic decomposition of S into pairs of pants, and let
28

1
,
2
, . . . ,
3 g3+ n
be the decomposing curves of T. Suppose that there is
a strong deformation :

S S. Let
j
be the closed geodesic homotopic to

1
(
j
). There exists another strong deformation

such that

(
j
) =
j
,
so T is pulled back under

to an oriented geodesic decomposition T

S
.
[S] (l
1
(T
S
),
1
(T
S
), . . . , l
3 g3+ n
(T
S
),
3 g3+ n
(T
S
))
denes local coordinates (l
1
,
1
, . . . , l
3 g3+ n
,
3 g3+ n
) on

P
g, n
. Therefore,
both

M
g, n
and

P
g, n
are (6 g 6 + 2 n) dimensional.
(2) Let (S, ) be a stable symmetric surface of genus g with n punctures,
and let T be an oriented geodesic decomposition of S into pairs of pants
which is invariant under . By considering -invariant decompositions
into pants in a neighborhood of (S, ) in T
R
g, n
we obtain local coordinates
(l
1
,
1
, . . . , l
3 g3+ n
,
3 g3+ n
). However, these parameters are not indepen-
dent. If (
i
) =
j
, i ,= j, then l
i
= l
j
, and
i
= c
j
for some constant c.
If (
i
) =
i
, then
i
= 0. So there are 3 g 3+ n independent parameters,
and the dimension of T
R
g, n
is 3 g 3 + n.
(3) Let S be a stable surface of type (g, h, n), and let R
1
, . . . , R
h
be its border
curves. Let T be an oriented geodesic decomposition of S into pairs of
pants, and let
1
, . . . ,
3g+h3+n
be the decomposing curves. We have
local coordinates (l
1
,
1
, . . . , l
3g+h3+n
,
3g+h3+n
, L
1
, . . . , L
h
), where l
j
is
the length of
j
,
j
is the angle of gluing along
j
, and L
i
is the length
of the border curve R
i
. Therefore, the moduli of stable surfaces of type
(g, h, n) is 6g +3h6 +2n = 3 g 3 + n, where g = 2g +h1, and n = 2n.
This is consistent with the previous paragraph since the complex double of
S is a stable symmetric Riemann surface with genus g = 2g + h 1 and
n = 2n punctures.
(4) Let (, B; p; q
1
, . . . , q
h
) be a stable marked bordered Riemann surface of
type (g, h) with (n, m) marked points. (
C
, , x) be its complex double, and
(S, ) be the symmetric Riemann surface obtained from by removing the
marked points, as in Section 4.1. Choose a -invariant geodesic decomposi-
tion of S into pants, we have local coordinates (l
1
,
1
, . . . , l
3 g3+ n
,
3 g3+ n
),
as described in 1., where
g = 2g +h 1, n = 2n +m
1
+ +m
h
,
3 g 3 + n = 6g + 3h 3 + 2n +m
1
+ +m
h
.
We have seen that half of the 2(6g+3h6+2n+m
1
+ +m
h
) parameters
are independent, so the dimension of

M
(g,h),(n, m)
is
6g + 3h 6 + 2n +m
1
+ +m
h
.
In the following example, we describe the Frenchel-Nelson coordintates of the
moduli space

M
0,3
of a pair of pants explicitly.
Example 4.7. The hexigon in Figure 17 is obtained by cutting the pair of pants
in Figure 16 along the geodesics
1,2
,
2,3
,
3,1
.
1
is the geodesic which realizes the
distance between
1
and
2,3
, etc.
Let l
1
, l
2
, l
3
, l
4
, l
5
, l
6
, l
7
, l
8
, l
9
be twice the lengths of
1
,
2
,
3
,
2,3
,
3,1
,
1,2
,
1
,
2
,
3
,
respectively. The degeneration l
i
= 0 corresponds to a real codimension one stratum
V
i
of

M
0,3
. Let V
ij
= V
i
V
j
and V
ijk
= V
i
V
j
V
k
.

M
0,3
can be identied with
29

2

3

1,2

2,3

3,1

3
V
12
V
13
V
23
V
45
V
56
V
46
V
14
V
25
V
36
V
27
V
37
V
57
V
67
V
18
V
38
V
48
V
68
V
19
V
29
V
49
V
59
Figure 17.
the associahedron K
5
dened by J. Stashe [St]. The conguration of the strata in

M
0,3

= K
5
is shown in Figure 17. There is one 3-dimensional stratum. There are
nine 2-dimensional strata:
V
1
, V
2
, V
3
, V
4
, V
5
, V
6
, V
7
, V
8
, V
9
.
There are twenty-one 1-dimensional strata:
V
12
, V
13
, V
23
, V
45
, V
56
, V
46
, V
14
, V
25
, V
36
, V
27
, V
37
, V
57
, V
67
, V
18
, V
38
, V
48
, V
68
, V
19
, V
29
, V
49
, V
59
.
There are fourteen 0-dimensional strata:
V
123
, V
456
, V
237
, V
257
, V
367
, V
567
V
138
, V
148
, V
368
, V
468
V
129
, V
149
, V
259
, V
459
.
There is a one-to-one correspondence between the 0-dimensional strata and Frenchel-
Nelson coordinate charts of

M
0,3
: the Frenchel-Nelson coordinates near V
ijk
are
l
i
, l
j
, l
k
.
4.4. Compactness and Hausdorness. We rst dene a topology on

M
(g,h),(n, m)
,
following [Ab, Chapter II (3.4)], and [Se, Section 5]. We will call it Fenchel-Nielsen
topology.
Denition 4.8. A strong deformation between two stable marked bordered Rie-
mann surfaces (

,

B; p; q
1
, . . . , q
h
) and (, B; p; q
1
, . . . , q
h
) of type (g, h) with (n, m)
marked points is a continuous map :

such that
(1) (

B
i
) = B
i
, ( q
i
k
) = q
i
k
, ( p
j
) = p
j
.
(2) If r is an interior node on

, then ( r) is an interior node on .
(3) If s is a boundary node of type E (H) on

, then ( s) is a boundary node
of type E (H) on .
(4) If r is an interior node on , then
1
(r) is an interior node or a circle.
(5) If s is a boundary node of type E, then
1
(r) is a boundary node of type
E or a border circle.
30
(6) If s is a boundary node of type H, then
1
(s) is a boundary node or type
H or an arc with ends in .
(7) f[

1
(S

)
:
1
(S

) S

is a dieomorphism.
Let
: (

,

B; p; q
1
, . . . , q
h
) (, B; p; q
1
, . . . , q
h
)
be a strong deformation. Let (

S, ), (S, ) be the symmetric Riemann surfaces

obtained by removing marked points from

C
,
C
, respectively. Dene
C
:

C
by

C
(z) =
_
(z) if z
(z) if z

Let denote the restriction of

C
to

S. Then : (

S, ) (S, ) is a strong
deformation.
Given , > 0 and
= [(, B; p; q
1
, . . . , q
h
)]

M
(g,h),(n, m)
,
we will dene a neighborhood U(, , ) of in

M
(g,h),(n, m)
. Let (S, ) be the
symmetric Riemann surface obtained by removing marked points from
C
. Let
= [(

,

B; p; q
1
, . . . , q
h
)]

M
(g,h),(n, m)
,
and (

S, ) be the associated symmetric Riemann surface. Then

M
(g,h),(n, m)
if
(1) There exists a -invariant oriented geodesic decomposition of S into pairs
of pants.
(2) There exists a strong deformation
: (

,

B; p; q
1
, . . . , q
h
) (, B; p; q
1
, . . . , q
h
)
i n the sense of Denition 4.8. So we have a strong deformation :

S S
as above.
(3) Let l
j
,
j
and

l
j
,

j
be the Fenchel-Nielsen coordinates for T and

(T),
respectively. Set d = 6g + 3h 6 + 2n + m. We have [l
j

l
j
[ < for
j = 1, . . . , d, and [
j

j
[ < if l
j
> 0.
U(, , ) [ , > 0,

M
(g,h),(n, m)

form a basis of the Fenchel-Nielsen topology.

U(, , ) can be described more precisely. Set z
j
= l
j
e
ij
, then up to permutation
and complex conjugation of some z
k
we have
(z
1
, z
2
, . . . , z
2d11
, z
2d1
, z
2d1+1
, . . . , z
d
) = ( z
2
, z
1
, . . . , z
2d
, z
2d1
, z
2d1+1
, . . . , z
d
),
so the xed locus of consists of points of the form
( z
2
, z
2
, . . . , z
2d11
, z
2d1
, x
1
, . . . , x
d2
),
where 2d
1
+ d
2
= d, z
2
, z
4
, . . . , z
2d1
C, and x
1
, . . . , x
d2
R. The coordinates
take values in the xed locus of , and x
i
are nonnegative on

M
(g,h),(n,m)
because
negative values correspond to nonorientable surfaces, as we have seen in Section 3.3.
We conclude that U(, , ) is homeomorphic to

U/, where

U is an open subset
of C
d1
[0, )
d2
, and is the automorphism group of . The transition functions
between charts are real analytic ([Wolp, Appendix]), so Fenchel-Nielsen coordinates
give

M
(g,h),(n, m)
the structure of an orbifold with corners. The topology determined
31
by the structure of an orbifold with corners coincides with Fenchel-Nielsen topology.
Therefore, we may equip

M
(g,h),(n, m)
with a metric which induces Fenchel-Nielson
topology. In particular, the topology is Hausdor, and compactness is equivalent
to sequential compactness. A straightforward generalization of the argument in
[Se, Section 6] shows that

M
(g,h),(n, m)
is sequentially compact in Fenchel-Nielsen
topology. Therefore,
Theorem 4.9.

M
(g,h),(n, m)
is Hausdor and compact in Fenchel-Nielson topology.
4.5. Orientation.

M
(g,h),(n, m)
is an orbifold with corners, so we may ask if it is
orientable as an orbifold. By Stashes results in [St], we have
Theorem 4.10.

M
(0,1),(0,(m))
has (m1)! isomorphic connected components, which
correspond to the cyclic ordering of the m boundary marked points. Each connected
component of

M
(0,1),(0,(m))
is homeomorphic to R
m3
.
Lemma 4.11. Suppose that (g, h, n) ,= (0, 1, 0), and m
i
> 0. If

M
(g,h),(n, m)
is
orientable, then

M
(g,h),(n,(m
1
,...,m
i
+1,...,m
h
))
is orientable.
Proof. Assume that

M
(g,h),(n,(m
1
,...,m
i
,...,m
h
)))
is orientable. Consider the map
(1) F :

M
(g,h),(n,(m
1
,...,m
i
+1,...,m
h
))


M
(g,h),(n, m)
,
given by forgetting the last boundary marked point on the i-th boundary circle.
Under our assumption, the ber of F over [(, B; p; q
1
, . . . , q
h
)] is a union of m
i
intervals and inherits the orientation of B
i
. So

M
(g,h),(n,(m
1
,...,m
i
+1,...,m
h
)))
is ori-
entable. 2
Lemma 4.12. Supposet that (g, h, n) ,= (0, 1, 0), and m
i
= 0. If

M
(g,h),(n,(m
1
,...,m
i
+1,...,m
h
)))
is orientable, then

M
(g,h),(n, m)
is orientable.
Proof. Assume that

M
(g,h),(n,(m
1
,...,m
i
+1,...,m
h
)))
is orientable. Let T be the tangent
bundle of

M
(g,h),(n, m)
, which is an orbibundle over

M
(g,h),(n, m)
. To show that

M
(g,h),(n, m)
is orientable, it suces to show that the restriction of T to every loop
in

M
(g,h),(n, m)
is orientable. Let N
(g,h),(n, m)
be the interior of

M
(g,h),(n, m)
. More
precisely, N
(g,h),(n, m)
corresponds to surfaces with no boundary nodes. Since every
loop in

M
(g,h),(n, m)
is homotopic to a loop in N
(g,h),(n, m)
, it suces to show that
N
(g,h),(n, m)
is orientable.
Suppose that = [(, B; p; q
1
, . . . , q
h
)] N
(g,h),(n, m)
. Then B
i
is an embedded
circle in , oriented as in Remark 2.13, and the ber of the map F in (1) over
can be identied with B
i
. So N
(g,h),(n, m)
is orientable. 2
It is shown in [IS2] that
Theorem 4.13. Suppose that (g, h, n) ,= (0, 1, 0). Then

M
(g,h),(n,(1,...,1))
is a com-
plex orbifold.
Theorem 4.10, Lemma 4.11, Lemma 4.12, and Theorem 4.13 imply that
Theorem 4.14.

M
(g,h),(n, m)
is orientable.
Let

Q
(g,h),n
be the moduli space of stable bordered Riemann surfaces of type
(g, h) with n interior points. There is an h! to one map

M
(g,h),(n,

0)


Q
(g,h),n
, given
by forgetting the ordering of boundary components. Then

Q
(g,h),n
is nonorientable.
32

3

4
Figure 18.

Q
(g,h),n
is nonorientable
For example, consider

M
(1,2),(0,0)
represented by the surface as show in
Figure 18. The local coordinates are (l
1
,
1
, l
2
,
2
, l
3
, l
4
), where l
j
is the length of

j
for j = 1, . . . , 4, and
1
,
2
are gluing angles for
1
,
2
, respectively. There is
an automorphism of order 2 of which rotates the above Figure 14 by 180

.
(
1
) =
2
, (
3
) =
4
, and (l
1
,
1
, l
2
,
2
, l
3
, l
4
) = (l
2
,
2
, l
1
,
1
, l
4
, l
3
), which
is orientation reversing.
In general, an automorphism induces permutation of d
1
decomposing curves and
permutation of d
2
bordered curves. The former corresponds to permutation of
pairs (l
j
,
j
), j = 1, . . . , d
1
, which is orientation preserving. The later corresponds
to permutation of (l
d1+1
, . . . , l
d1+d2
) which is orientation preserving if and only if it
is an even permutation. When we consider

M
(g,h),(n,

0)
, automorphisms permuting
border curves are not allowed.
5. Moduli Space of Stable Maps
5.1. Prestable and stable maps. Let (X, ) be a compact symplectic manifold,
and let L be a Lagrangian submanifold. Let J be an -tame almost complex
structure.
Denition 5.1. A prestable map is a continuous map u : (, ) (X, L) such
that J d u = d u j, where is a prestable bordered Riemann surface, u = u ,
:

is the normalization map (Denition 3.7).
Denition 5.2. A prestable map of type (g, h) with (n, m) marked points consists
of a prestable marked bordered Riemann surface of type (g, h) with (n, m) marked
points (, B; p; q
1
, . . . , q
h
) and a prestable map u : (, ) (X, L).
Denition 5.3. A morphism between prestable maps of type (g, h) with (n, m)
marked points
(, B; p; q
1
, . . . , q
h
; u) (

, B

; p

; (q

)
1
, . . . , (q

)
h
; u

)
is an isomorphism
: (, B; p; q
1
, . . . , q
h
) (

, B

; p

; (q

)
1
, . . . , (q

)
h
)
between prestable marked bordered Riemann surfaces of type (g, h) with (n, m) points
such that u = u

.
Denition 5.4. A prestable map of type (g, h) with (n, m) marked points is stable
if its automorphism group is nite.
33
5.2. C

topology. Let H
2
(X, L; Z), = (
1
, . . . ,
h
) H
1
(L; Z)
h
be such
that
1
+ +
h
= , where : H
2
(X, L; Z) H
1
(L; Z) is the connecting
map in the long exact sequence for relative homology groups. Let h be a positive
integer, g, n be nonnegative integers, m = (m
1
, . . . , m
h
) be an h-uple of nonnegative
integers, and be an integer. Given the above data, dene

M
(g,h),(n, m)
(X, L [ , , )
to be the moduli space of isomorphism classes of stable maps of type (g, h) with
(n, m) marked points
(, B; p; q
1
, . . . , q
h
; u)
such that u

[] = , u

[B
i
] =
i
for i = 1, . . . , h, and (u

TX, u

TL) = . Here
(u

TX, u

TL) is the Maslov index dened in [KL, Denition 3.3.7, Denition

3.7.2]. From now on, we assume that L is oriented, so (u

TX, u

TL) is even, and

we may restrict ourselves to even . We will also assume that none of the
i
is
zero, so the domain cannot have boundary nodes of type E.
Let

M
(g,h),(n, m)
(X, L) be the moduli space of isomorphism classes of stable maps
of type (g, h) with (n, m) marked points. Then

M
(g,h),(n, m)
(X, L [ , , ) are
disjoint subsets of

M
(g,h),(n, m)
(X, L) for dierent (, , ).
Let be a prestable bordered Riemann surface, and let :

be the
normalization. Let u : (, ) (X, L) be a continuous map such that u = u :
(

) (X, L) is C

w.r.t. g
0
on X and some Hermitian metric h on

, l 1.
Dene
a(u) =
_

u
x

u
y

_
u
x
,
u
y
_
2
g(x, y)dx dy,
where (x, y) are local isothermal coordinates on

, and g(x, y)dxdy is the volume
form for the metric h. If u is an embedding, a(u) is the area of u() w.r.t. g
0
. If u
is a prestable map, then a(u) =
1
2
| du |
L
2= (u

[]) [], where

| du |
2
L
2=
_

u
x

2
+

u
y

2
_
g(x, y)dx dy.
(u

[]) [] only depends on the relative homology class u

[] H
2
(X, L; Z), so
we have a function a :

M
(g,h),(n,m)
[0, ), which takes the constant value []
on

M
(g,h),(n, m)
(X, L [ , , ).
With the above denition,

M
(g,h),(n, m)
(X, L [ , , ) is a set. We will equip it
with the structure of a topological space, and show that it is sequentially compact
and Hausdor in this topology. This topology was introduced by Gromov [Gr].
We want to say two stable maps are close if the complex structures on the domain
are close, and the maps are close. To measure the closeness, we use metrics on the
domain and on the target. For the target X, J is an -tame complex structure,
so g
0
(X, Y ) =
1
2
((X, JY ) + (Y, JX)) is a Riemannian metric on X such that J
is an isometry. For the domain, by a Hermitian metric h on a prestable bordered
Riemann surface we mean a Hermitian metric on

h on

, the normalization of
.
We now introduce some notation. Let :

be the normalization map.
Given a node r and a small positive number , let B

(r) = (B

(r
1
) B

(r
2
)),
where
1
(r) = r
1
, r
2
, and B

(r

) is the geodesic ball of radius for = 1, 2. Let

34
be suciently small so that B

(r) are disjoint for r

sing
, where
sing
denotes
the set of nodes on . Set N

() =

rsing
B

(r), K

() = N

().
Denition 5.5 (C

topology). Let = (, B; p; q
1
, . . . , q
h
; u) be a prestable map
of type (g, h) with (n, m) marked points. For a Hermitian metric h on and

1
, . . . ,
4
> 0, a neighborhood U(, h,
1
, . . . ,
4
) of u in

M
(g,h),(n, m)
(X, L) in the
C

topology is dened as follows. A prestable map

= (

, B

; p

; (q

)
1
, . . . , (q

)
h
; u

)
belongs to U(, h,
1
, . . . ,
4
) if
(1) There is a strong deformation
: (

, B

; p

; (q

)
1
, . . . , (q

)
h
) (, B; p; q
1
, . . . , q
h
)
such that
1
is dened on K
1
().
(2) | j (
1
)

|
C

(K
1
())
<
2
, where j, j

are complex structures on ,

, respectively.
(3) | u u

1
|
C

(K
1
())
<
3
.
(4) [a(u) a(u

)[ <
4
.
(1) says that

can be obtained by deforming , or equivalently,

is in the
same or a higher stratum in

M
(g,h),(n, m)
, the moduli space of prestable marked
bordered Riemann surfaces of type (g, h) with (n, m) marked points. (2) says that
(, B; p; q
1
, . . . , q
h
), (

, B

; p

; (q

)
1
, . . . , (q

)
h
)
are close in the C

topology (Denition 5.6). (3) says that the maps u, u

are C

close away from the nodes. (4) implies that a :

M
(g,h),(n, m)
(X, L) [0, ) is a
continuous function.
(, B; p; q
1
, . . . , q
h
; u), (

, B

; p

; (q

)
1
, . . . , (q

)
h
; u

)
represent the same point in

M
(g,h),(n, m)
(X, L) if and only if there is a Hermitian
metric h on such that
(1) There is a homeomorphism
: (

, B

; p

; (q

)
1
, . . . , (q

)
h
) (, B; p; q
1
, . . . , q
h
)
which induces a dieomorphism

.
(2) | j (
1
)

|
C

()
= 0, where j, j

are complex structures on ,

,
respectively.
(3) | u u

1
|
C

()
= 0.
(4) [a(u) a(u

)[ = 0.
So the C

topology is actually a topology on the moduli space

M
(g,h),(n, m)
(X, L).

M
(g,h),(n, m)
(X, L [ , , ) is a closed subspace of

M
(g,h),(n, m)
(X, L) and is equipped
with the subspace topology.
When X is a point, we get the C

topology of

M
(g,h),(n, m)
.
Denition 5.6. Let = (, B; p; q
1
, . . . , q
h
) be a stable (prestable) bordered Rie-
mann surface of type (g, h) with (n, m) marked points. For a Hermitian metric
h on and
1
,
2
> 0, a neighborhood U(, h,
1
,
2
) of u in

M
(g,h),(n, m)
in the
C

topology is dened as follows. A stable (prestable) bordered Riemann surface

= (

, B

; p

; (q

)
1
, . . . , (q

)
h
) belongs to U(, h,
1
,
2
) if
35
(1) There is a strong deformation :

such that
1
is dened on
K
1
().
(2) | j (
1
)

|
C

(K
1
())
<
2
, where j, j

are complex structures on ,

, respectively.
5.3. Compactness and Hausdorness. The following is the main theorem of
this section.
Theorem 5.7.

M
(g,h),(n, m)
(X, L [ , , ) is Hausdor and sequentially compact
in the C

topology.
The Hausdorness can be proven as in the case of curves without boundary, see
e.g. [Sie1, Proposition 3.8]. The compactness is a consequence of the following
theorem.
Theorem 5.8 (Gromovs Compactness Theorem). Let
l
be a sequence in

M
(g,h),(n m)
(X, L) such that a(
l
) < C for all l ^. Then there is a subsequence
of
l
convergent in the C

topology.
Gromovs compactness theorem [Gr, 1.5] for J-holomorphic curves without bound-
ary was carried out in details in [PW, Ye]. The case with boundary was proved in
[Ye] (see also [IS1, IS2]). In [Ye], the moduli space is compactied by the moduli
space of cusp curves, or prestable maps in this paper. We will describe how the
proof in [Ye] gives Theorem 5.8.
The C

topology can be equivalently dened as follows.

Denition 5.9 (C

Topology). A sequence

l
= (
l
, B
l
; p
l
; q
1
l
, . . . , q
h
l
; u
l
)
converges to = (, B; p; q
1
, . . . , q
h
; u) in the C

topology if for each

1
, . . . ,
4
>
0, there is an integer N such that for l N,
(1) There is a strong deformation
l
:
l
such that
1
l
is dened on
K
1
().
(2) | j (
1
l
)

j
l
|
C

(K
1
())
<
2
.
(3) | u u
l

1
l
|
C

(K
1
())
<
3
.
(4) [a(u) a(u
l
)[ <
4
.
Recall that

M
(g,h),(n, m)
denotes the moduli space of prestable bordered Rie-
mann surfaces of type (g, h) with (n, m) marked points. There is a map F :

M
(g,h),(n, m)
(X, L [ , , )

M
(g,h),(n, m)
, given by forgetting the map.

M
(g,h),(n,m)
has innitely many strata since one can keep on going to lower and lower strata by
adding non-stable components spheres and discs.
We claim that the image of F is covered by only nitely many strata, or equiv-
alently:
Lemma 5.10. The domains in

M
(g,h),(n, m)
(X, L [ , , ) have only nitely many
topological types.
Proof. There is a map

M
(g,h),(n, m)


M
(g,h),(n, m)
, given by contracting non-stable
components. Since a stable bordered Riemann surfaces of type (g, h) with (n, m)
points can have only nitely many possible topological types, it suces to get an
upper bound for the number of non-stable irreducible components. The restriction
of a stable map to a non-stable irreducible component is nonconstant, so there is
36
a lower bound > 0 for the area of the restriction of the map to each non-stable
component by [Ye, Lemma 4.3, Lemma 4.5]. Therefore, the number of non-stable
irreducible components cannot exceed ( [])/. 2
Let
l
be a sequence in

M
(g,h),(n, m)
(X, L) such that a() < C for all l N. By
Lemma 5.10, there is a subsequence of
l
such that the domains are of the same
topological type. By normalization we obtain several sequences of stable maps with
smooth domains of the same topological type and with uniform area bound. Note
that each node gives rise to two marked points on the normalization. It suces to
show that each sequence has a subsequence convergent in the C

topology. So we
may assume that the domain is a smooth marked bordered Riemann surface or a
smooth curve with marked points. In this case, it is proven in [Ye] that there is a
subsequence which converges to a prestable map in the C

topology. However, it
is straightforward to check that the limit produced in [Ye] is actually a stable map.
6. Construction of Kuranishi Structure
6.1. Kuranishi structure with corners. We rst quote the following denition
from [FO
3
, A2.1.1-A2.1.4], which is a slight modication of [FO, Denition 5.1].
Denition 6.1 (Kuranishi neighborhood). Let M be a Hausdor topological space.
A Kuranishi neighborhood (with corners) of p M is a 5-uple (V
p
, E
p
,
p
,
p
, s
p
)
such that
(1) V
p
is a smooth manifold (with corners), and E
p
is a smooth vector bundle
on it.
(2)
p
is a nite group which acts smoothly on E
p
V
p
.
(3) s
p
is a
p
-equivariant continuous section of E
p
.
(4)
p
: s
1
p
(0) M is a continuous map which induces a homeomorphism
from s
1
p
(0)/
p
to a neighborhood of p in M.
We call E
p
the obstruction bundle and s
p
the Kuranishi map.
The following equivalence relation is weaker than the one in [FO, Denition 5.2],
so the resulting equivalence class is larger.
Denition 6.2. Let M be a Hausdor topological space. Two Kuranishi neigh-
borhoods (with corners) (V
1,p
, E
1,p
,
1,p
,
1,p
, s
1,p
) and (V
2,p
, E
2,p
,
2,p
,
2,p
, s
2,p
) of
p M are equivalent if
(1) dimV
1,p
rankE
1,p
= dimV
2,p
rankE
2,p
d.
(2) There is another Kuranishi neighborhood (with corners) (V
p
, E
p
,
p
,
p
, s
p
)
of p such that dimV
p
rankE
p
= d.
(3) There are homomorphisms h
i
:
i,p

p
for i = 1, 2.
(4) For i = 1, 2, there is a
i,p
-invariant open neighborhood V
i
of
1
i,p
(p), an
h
i
-equivariant embedding
i
: V
i
V
p
, and an h
i
-equivariant embedding of
vector bundles

i
: E
i,p
[
Vi
E
p
which covers
i
.
(5)

i
s
i,p
= s
p

i
for i = 1, 2.
(6)
i,p
=
p

i
for i = 1, 2.
In this case, we write (V
1,p
, E
1,p
,
1,p
,
1,p
, s
1,p
) (V
2,p
, E
2,p
,
2,p
,
2,p
, s
2,p
)
The following denition is a combination of [FO
3
, A2.1.5-A2.1.11] and [FO,
Denition 5.3].
37
Denition 6.3 (Kuranishi structure). Let M be a Hausdor topological space.
A Kuranishi structure (with corners) on M assigns a Kuranishi neighborhood (or
a Kuranishi neighborhood with corners) (V
p
, E
p
,
p
,
p
, s
p
) to each p M and a
4-uple (V
pq
,

pq
,
pq
, h
pq
) to each pair (p, q) where p M, q
p
(s
1
p
(0)) such that
(1) V
pq
is an open subset of V
q
containing
1
q
(q).
(2) h
pq
is a homomorphism
q

p
.
(3)
pq
: V
pq
V
p
is an h
pq
-equivariant embedding.
(4)

pq
: E
q
[
Vpq
E
p
is an h
pq
-equivariant embedding of vector bundles which
covers
pq
.
(5)

pq
s
q
= s
p

pq
.
(6)
q
=
p

pq
.
(7) If r
q
(s
1
q
(0) V
pq
), then

pq

qr
=

pr
in a neighborhood of
1
r
(r).
(8) dimV
p
rank E
p
is independent of p and is called the virtual dimension of
the Kuranishi structure (with corners).
(V
pq
,

pq
,
pq
, h
pq
) is called a transition function from (V
q
, E
q
,
q
,
q
, s
q
) to (V
p
, E
p
,
p
,
p
, s
p
).
Remark 6.4. Let M be a Hausdor space with a Kuranishi structure with corners
/ =
_
(V
p
, E
p
,
p
,
p
, s
p
) : p M, (V
pq
,

pq
,
pq
, h
pq
) : q
p
(s
1
p
(0))
_
of virtual dimension d. Let M =
pM

p
(s
1
p
(0) V
p
), where V
p
is the union
of corners in V
p
. Then
/ =
_
(V
p
, E
p
[
V
,
p
,
p
, s
p
) : p M, (V
pq
,

pq
,
pq
, h
pq
) : q
p
(s
1
p
(0) V
p
)
_
is a Kuranishi structure with corners of virtual dimension d 1 on M.
Denition 6.5. Let M be a Hausdor topological space. Two Kuranishi structures
/
1
=
_
(V
1,p
, E
1,p
,
1,p
,
1,p
, s
1,p
) : p M, (V
1,pq
,

1,pq
,
1,pq
, h
1,pq
) : q
1,p
(s
1
1,p
(0))
_
and
/
2
=
_
(V
2,p
, E
2,p
,
2,p
,
2,p
, s
2,p
) : p M, (V
2,pq
,

2,pq
,
2,pq
, h
2,pq
) : q
2,p
(s
1
2,p
(0))
_
on M are equivalent if there is another Kuranishi structure
/ =
_
(V
p
, E
p
,
p
,
p
, s
p
) : p M, (V
pq
,

pq
,
pq
, h
pq
) : q
p
(s
1
p
(0))
_
on M such that for all p M, (V
1,p
, E
1,p
,
1,p
,
1,p
, s
1,p
), (V
2,p
, E
2,p
,
2,p
,
2,p
, s
2,p
),
and (V
p
, E
p
,
p
,
p
, s
p
) satisfy the relation described in Denition 6.2. In this case,
we write /
1
/
2
.
Let (V
p
, E
p
,
p
,
p
, s
p
) be a Kuranishi neighborhood (with corners) of p. If s
p
intersects the zero section of E
p
transversally, then

M
p
= s
1
p
(0) is a smooth
submanifold (with corners) of V
p
of dimension dimV
p
rankE
p
, and there is an
exact sequence of smooth vector bundles
0 T

M
p
TV
p
[

Mp
dsp
E
p
[

Mp
over

M
p
. So T

M
p
is equivalent to the two term complex [TV
p
[

Mp
dsp
E
p
[

Mp
] as an
element of the Grothendieck group KO(

M
p
).
38
Both TV
p
and E
p
are
p
-equivariant vector bundles over V
p
, so TV
p
/
p
, E
p
are
orbibundles over the orbifold (with corners) U
p
= V
p
/
p
. We call TU
p
TV
p
/
p
the tangent bundle of the orbifold (with corners) U
p
, and TM
p
T

M
p
/
p
is the
tangent bundle of the orbifold (with corners) M
p
=

M
p
/
p
.
In general,

M
p
might be singular, so T

M
p
does not exist. Nevertheless,

M
p
is a
topological space, so KO(

M
p
) makes sense. We dene
T
vir

M
p
= [TV
p
[

Mp
dsp
E
p
[

Mp
] KO(

M
p
)
to be the virtual tangent bundle of

M
p
, and
T
vir
M
p
= [TU
p
[
Mp
dsp
(E
p
/
p
)[
Mp
]
to be the virtual tangent bundle of M
p
. We have T
vir

X
p
= T

M
p
and T
vir
M
p
=
TM
p
when s
p
intersects the zero section transversally. The transition functions
(V
pq
,

pq
,
pq
, h
pq
) in Denition 6.3 enable us to glue T
vir
M
p
to obtain the virtual
tangent bundle T
vir
M of the Kuranishi structure on M.
det TV
p
(det E
p
)
1
glue to a real line orbibundle det(T
vir
M), the orientation
bundle of the Kuranishi structure (with corners). It is a real line bundle if the action
of each
p
on det TV
p
(det E
p
)
1
is orientation preserving. We say a Kuranishi
structure is orientable if its orientation bundle is a trivial real line bundle. If / and
/

are equivalent Kuranishi structures (with corners), then / is orientable if and

only if /

is.
In the ordinary Gromov-Witten theory, there is an algebraic approach to dene
Gromov-Witten invariants when the target is a smooth projective variety [BF].
In the algebraic approach, the moduli of stable maps is a Deligne-Mumford stack,
which is locally etale covered by ane schemes. In Denition 6.3, s
1
p
(0) is the
analogue of an ane scheme an ane scheme is the zero locus of polynomials,
while s
1
p
(0) is the zero locus of smooth functions.
The moduli space of stable maps admits a perfect obstruction theory, which is an
element in the derived category locally isomorphic to a two term complex of vector
bundles [E
1
E
0
]. Given a perfect obstruction theory, the virtual dimension is
dened to be rankE
0
rank E
1
, and a virtual fundamental class of the virtual
dimension can be constructed. The two term complex [TV
p
dsp
E
p
] is the analogue
of [E

0
E

1
].
A Kuranishi structure can be viewed as the analytic counterpart of a Deligne-
Mumford stack together with a perfect obstruction theory. We will show that
Theorem 6.6.

M
(g,h),(n, m)
(X, L [ , , ) has a Kuranishi structure of virtual
dimension
+ (N 3)(2 2g h) + 2n +m
1
+ + m
h
,
where 2N is the dimension of X. The Kuranishi structure is orientable if L is spin
or if h = 1 and L is relatively spin (i.e., L is orientable and w
2
(TL) = [
L
for
some H
2
(X, Z
2
)).
6.2. Stable W
k,p
maps. Let (X, ) be a compact symplectic manifold together
with an -tame almost complex structure J, and let L be a compact Lagrangian sub-
manifold of X, as before. To construct a Kuranishi structure on

M
(g,h),(n, m)
(X, L [
, , ), we need to enlarge the category of stable maps. We rst specify metrics on
the target (Section 6.2.1) and on the domain (Section 6.2.2) so that we can dene
39
norms on relevant Banach spaces. The denition of stable W
k,p
maps is given in
Section 6.2.3. The virtual dimension in Theorem 6.6 is computed in Section 6.2.4.
6.2.1. Metric on the target. Let g
0
be the Riemannian metric on X dened by
g
0
(v, w) =
1
2
((v, Jw) + (w, Jv)). We will modify g
0
to obtain a Riemannian
metric g
1
such that L is totally geodesic w.r.t. g
1
.
Lemma 6.7. Given a Riemannian vector bundle (V, h) over a compact Riemannian
manifold (M, g), there is a Riemannian metric g on the total space of V such that
(1) For any x M, the restriction of g to the ber V
x
over x is h(x).
(2) The zero section i
0
: (M, g) (V, g) is an isometric embedding.
(3) i
0
(M) is totally geodesic in (V, g).
Proof. Let : V M be the canonical projection. Choose a connection on V
which is compatible with h. This gives a decomposition TV =

V H, where
H

=

TM. Let x M, w V
x
, so that (x, w) V . Given T
(x,w)
V , there
is a unique decomposition =
v
+
h
, where
v

V , and
h
H. Dene a
quadratic form Q on T
(x,w)
V by Q(, ) = h(x)(
v
,
v
) +g(x)(

(
h
),

(
h
)). Then
Q determines an inner product g(x, w) on T
(x,w)
V . g is a Riemannian metric on V
which clearly satises 1. and 2.
For (3), let x
0
, x
1
i
0
(M) be close enough such that there is a unique length
minimizing geodesic : [0, 1] (V, g) such that (0) = x
0
, (1) = x
1
. It suces
to show that this geodesic lies in i
0
(M). We have
l() =
_
1
0
g(

)dt
=
_
1
0
(h( )(

v
,

v
) +g( )(( )

, ( )

))dt

_
1
0
g( )(( )

, ( )

))dt
=
_
1
0
g((i
0
)

, (i
0
)

)dt
= l(i
0
),
The equality holds since is length minimizing. So

v
0, and = i
0
:
[0, 1] L. 2
The Riemannian metric g
0
on X gives an orthogonal decomposition
TX[
L
= TL N
L/X
,
where N
L/X
is the normal bundle of L in X. Let exp
0
denote the exponential map
TX X determined by g
0
. For R > 0, let B
R
(TX) denote the ball bundle of radius
R in TX. There exists R > 0 such that exp
0
maps B
R
(N
L/X
) dieomorphically
to its image in X. For r R, let N
r
(L) denote the image of B
r
(N
L/X
) under
exp
0
. We have a dieomorphism G : N
R
(L) B
R
(N
L/X
) which is the inverse
of exp
0
[
BR(N
L/X
)
. Construct a Riemannian metric g on N
L/X
as in Lemma 6.7.
Let be a smooth cut-o function dened on X such that = 1 on NR
3
(L) and
= 0 on X N2R
3
(L). Dene g
1
= G

g + (1 )g
0
on N
R
(L). Then g
1
= g
0
on
N
R
(L) N2R
3
(L), so g
1
extends to a Riemannian metric on X such that g
1
= g
0
on X N
R
(L) and on TX[
L
.
40
X is compact, so there exists some constant C
0
> 0 such that C
1
0
g
0
g
1

C
0
g
0
. The C

topology (Denition 5.5) dened by g

1
is equivalent to that dened
by g
0
. From now on, all the parallel transports, exponential maps, and norms are
dened by g
1
instead of g
0
.
6.2.2. Metric on the domain. Let = [(, B; p; q
1
, . . . , q
h
)]

M
(g,h),(n, m)
, and let

M
(g,h),(n, m)
(X, L [ , , )

denote the ber of

F :

M
(g,h),(n, m)
(X, L [ , , )

M
(g,h),(n, m)
over . Choose a Hermitian metric

h on
C
which is compact, at near nodes,
and invariant under the antiholomorphic involution :
C

C
. Let h be the
restriction of

h to . Then the border curves of are geodesics in the Riemannian
metric determined by h. We further require that
(1) If s is an interior node, then there is an isometric holomorphic embedding
B

(r) C
2
, where C
2
is equipped with the standard metric, such that the
image is (x, y) C
2
[ xy = 0, [x[ < , [y[ < .
(2) If s is a boundary node of type H, then there is an isometric holomorphic
embedding B

(r) C
2
/A, where A(x, y) = ( x, y), such that the image is
(x, y) C
2
[ xy = 0, [x[ < , [y[ < /A.
(3) h is invariant under Aut .
We call such a Hermitian metric an admissible metric.
6.2.3. W
k,p
maps and C
l
maps.
Denition 6.8. Let be a prestable bordered Riemann surface. A continuous map
u : (, ) (X, L) is a W
k,p
map on if u = u : (

) (X, L) is of class
W
k,p
in the sense of [MS, Appendix B], where :

is the normalization map.
In the above denition, we assume that kp > 2, so the embedding W
k,p
C
0
is
compact.
Denition 6.9. A prestable W
k,p
map of type (g, h) with (n, m) marked points
consists of a prestable marked bordered Riemann surface of type (g, h) with (n, m)
marked points (, B; p; q
1
, . . . , q
h
) and a prestable W
k,p
map u : (, ) (X, L).
Denition 6.10. A morphism between prestable W
k,p
maps of type (g, h) with
(n, m) marked points
(, B; p; q
1
, . . . , q
h
; u) (

, B

; p

; (q

)
1
, . . . , (q

)
h
; u

)
is an isomorphism
: (, B; p; q
1
, . . . , q
h
) (

, B

; p

; (q

)
1
, . . . , (q

)
h
)
between prestable bordered Riemann surfaces of type (g, h) with (n, m) points such
that u = u

.
Denition 6.11. A prestable W
k,p
map of type (g, h) with (n, m) marked points
is stable if its automorphism group is nite.
C
l
maps and stable C
l
maps are dened similarly.
Let W
k,p
(g,h),(n, m)
(X, L [ , , ), C
l
(g,h),(n, m)
(X, L [ , , ) be the moduli space
of isomorphism classes of stable W
k,p
, C
l
maps of type (g, h) with (n, m) marked
41
points satisfying the topological conditions as in the denition of

M
(g,h),(n, m)
(X, L [
, , ), respectively. There are maps
F
k,p
: W
k,p
(g,h),(n, m)
(X, L [ , , )

M
(g,h),(n, m)
and
C
l
: C
l
(g,h),(n, m)
(X, L [ , , )

M
(g,h),(n, m)
given by forgetting the map. Recall that forgetting the map also gives
F :

M
(g,h),(n, m)
(X, L [ , , )

M
(g,h),(m,n)
.
Given

M
(g,h),(m,n)
, let

M
(g,h),(n, m)
(X, L [ , , )

, W
k,p
(g,h),(n, m)
(X, L [ , , )

,
C
l
(g,h),(n, m)
(X, L [ , , )

denote the ber of F, F

k,p
, F
l
over , respectively. From
now on, we will write
M

=

M
(g,h),(n, m)
(X, L [ , , )

W
k,p

= W
k,p
(g,h),(n, m)
(X, L [ , , )

C
l

= C
l
(g,h),(n, m)
(X, L [ , , )

for convenience.
Let exp denote the exponential map of the Riemannian metric g
1
on X, and let
h be an admissible metric on . For a stable W
k,p
map u on and > 0, dene
U
k,p
(u, ) = exp
u
(w) [ w W
k,p
(, , u

TX, (u[

TL), | w |
W
k,p < ,
where W
k,p
(, , u

TX, (u[

TL) will be dened in Section 6.2.4, and the

norms are dened by g
1
, h. Similarly, for a stable C
l
map u on and > 0,
dene
U
l
(u, ) = exp
u
(w) [ w C
l
(, , u

TX, (u[

TL), | w |
C
l < ,
where C
l
(, , u

TX, (u[

TL) will be dened in Section 6.2.4. Note that

exp
u
(w)[

L because L is totally geodesic w.r.t. g

1
. Then
U
k,p
(u, ) [ u is a stable W
k,p
map on , > 0
generate the W
k,p
topology on W
k,p

, while
U
l
(u, ) [ u is a stable C
l
map on , > 0
generate the C
l
topology on C
l

.
Dene
W
k,p
(u, ) = w W
k,p
(, , u

TX, (u[

TL) [ | w |
W
k,p<
and
C
l
(u, ) = w C
l
(, , u

TX, (u[

TL) [ | w |
C
l < .
For suciently small > 0, w exp
u
(w) gives
U
k,p
(u, )

= W
k,p
(u, )/Aut(, u)
and
U
l
(u, )

= C
l
(u, )/Aut(, u).
So W
k,p

and C
l

are Banach orbifolds. They are Banach manifolds if Aut is trivial.

M

is contained in both W
k,p

and C
l

since stable maps are C

([Ye, Theorem
2.1]).
42
6.2.4. Virtual dimension. Let = [(, B; p; q
1
, . . . , q
h
; u)] M

. Let C
1
, . . . , C

be irreducible components of which are (possibly nodal) Riemann surfaces, and

let
1
, . . . ,

be the remaining irreducible components of , which are (possibly

nodal) bordered Riemann surfaces. Let

C
i
denote the normalization of C
i
for
i = 1, . . . , , let

i
denote the normalization of
i
for i

= 1, . . . ,

, and let
:

be the normalization map.
Let r
1
, . . . , r
l0

be interior nodes of , and s

1
, . . . , s
l1
be boundary
nodes of . Let p
j


be the preimage of p
j
under for j = 1, . . . , n, q
j
be the
preimage of q
j
under for j

= 1, . . . , m, r

, r
l0+
be the preimages of r

under
for = 1, . . . , l
0
, and s

, s
l1+
be the preimages of s

under for

= 1, . . . , l
1
.
Set u = u . Then u : (

) (X, L) is J-holomorphic. For l > 1, let

C
l
(

, u

TX, ( u[

TL)
denote the vector space of C
l
sections of u

TX with boundary values in ( u[

TL.
Then
C
l
(

, u

TX, ( u[

TL)
=

i=1
C
l
(

C
i
, ( u[

Ci
)

TX)

=1
C
l
(

i
,

i
, ( u[

TX, ( u[

TL)
Let C
l
(, , u

TX, (u[

TL) be the kernel of

C
l
(

, u

TX, ( u[

TL)
l0

=1
T
r
X
l1

=1
T
s

L
s
_
s( r

) s( r
l0+
)
l0
=1
, s( s

) s( s
l1+
)
l1

=1
_
,
and set
C
l1
(,
0,1
u

TX)
=

i=1
C
l1
(

C
i
,
0,1

C
i
( u[

Ci
)

TX)

=1
C
l1
(

i
,
0,1

i
( u[

TX)
The linearization of

J,
at the map u gives rise to operators
C
l
(

C
i
, ( u[

Ci
)

TX) C
l1
(

C
i
,
0,1

C
i
( u[

Ci
)

TX), i = 1, . . . , ,
C
l
(

i
,

i
, ( u[

TX, ( u[

TL) C
l1
(

i
,
0,1

i
( u[

TX) i

= 1, . . . ,

,
and thus the operator
D
u
: C
l
(, , u

TX, (u[

TL) C
l1
(,
0,1
u

TX)
Similarly, we have
D
u
: W
k,p
(, , u

TX, (u[

TL) W
k1,p
(,
0,1
u

TX).
The following proposition follows from straightforward computations.
Proposition 6.12. Let be a connection on TX. Then
D
u
() =
1
2
( du +J du j +

J du j +T(, du) +JT(, du j))

where T(v, w) =
v
w
w
v [v, w] is the torsion of .
43
Lemma 6.13. Let be a smooth bordered Riemann surface of type (g, h), and
u : (, ) (X, L) be a J-holomorphic map. Then
D
u
: C

(, , u

TX, (u[

TL) C

(,
0,1
u

TX)
is a Fredholm operator of index +N(22g h), where = (u

TX, (u[

TL),
and 2N = dim
R
X.
Proof. Let be the Levi-Civita connection. is torsion-free, so by Proposi-
tion 6.12,
D
u
() =
1
2
( du +J du j +

J du j) .
We have D
u
= D

+R, where
D

() =
1
2
( du +J du j)
R() = =
1
2

J du j
D

denes a holomorphic structure on u

TX such that D

=

. By [KL,
Theorem 3.4.2], D

is a Fredholm operator of index + N(2 2g h). R is a

compact operator, so D
u
= D

+R is Fredholm of index +N(2 2g h). 2

Proposition 6.14. Let u : (, ) (X, L) be a prestable map. Then
D
u
: C

(, , u

TX, (u[

TL) C

(,
0,1
u

TX)
is a Fredholm operator of index +N(1 g), where = (u

TX, (u[

TL), and
g is the arithmetic genus of
C
.
Proof. We use the above notation. Set
C
0
= C

(, , u

TX, (u[

TL)
C
1
= C

(,
0,1
u

TX)

C
0
i
= C

(

C
i
, ( u[

Ci
)

TX)

C
1
i
= C

(

C
i
,
0,1

C
i
( u[

Ci
)

TX)
C
0
i
= C

i
,

i
, ( u[

TX, ( u[

TL)
C
1
i
= C

i
,
0,1

i
( u[

TX)
where i = 1, . . . , , i

= 1, . . . ,

. The linearization of

J,
gives rise to Fredholm
operators

D
i
:

C
0
i


C
1
i
for i = 1, . . . , and D
i
: C
0
i
C
1
i
for i

= 1 . . . ,

. We
have the following commutative diagram:
0 C
0

i=1

C
0
i

=1
C
0
i


l0
=1
T
r
X

l1

=1
T
s

L 0
Du

_ D

_
0 C
1

i=1

C
1
i

=1
C
1
i
0 0
where D =

i=1

D
i

=1
D
i
, and the rows are exact. So D
u
is Fredholm.
Given a Fredholm operator D, let Ind(D) denote the virtual real vector space
Ker(D) Coker(D),
whose dimension
dimInd(D) = dimKer(D) dimCoker(D)
44
is the Fredholm index of D. With the above notation, we have
dimInd(D
u
) =

i=1
dimInd(

D
i
) +

=1
dimInd(D
i
) 2Nl
0
Nl
1
.
Suppose that

C
i
is of genus g
i
, and

i
is of type (g
i
, h
i
). We have
dimInd(

D
i
) = 2 deg(( u[

Ci
)

TX) + 2N(1 g
i
),
dimInd(D
i
) = (( u[

TX, ( u[

i
)

TL) +N(2 2g
i
h
i
),
where the second equality follows from Proposition 6.13.
C
has 2l
0
+l
1
nodes and
2 +

irreducible components
C
1
, . . . , C

,

C
1
, . . . ,

C

, (
1
)
C
, . . . , (

)
C
,
where the genus of (

)
C
is g
i
= 2g
i
+h
i
1, so the arithmetic genus of
C
is
g = 2

i=1
g
i
+

=1
g
i
+ 2l
0
+l
1
2

+ 1
= 2

i=1
( g
i
1) +

=1
(2g
i
+ h
i
2) + 2l
0
+l
1
+ 1
by [HM, (3.1)]. Finally,
= (u

TX, (u[

TL) = 2

i=1
deg(( u[

Ci
)

TX+

=1
(( u[

TX, ( u[

TL),
so we conclude that
dimInd(D
u
) = (u

TX, (u[

TL) +N(1 g). 2

Remark 6.15. Corollary 6.14 remains true for
D
u
: W
k,p
(, , u

TX, (u[

TL) W
k1,p
(,
0,1
u

TX)
D
u
: C
l
(, , u

TX, (u[

TL) C
l1
(,
0,1
u

TX)
Set c

u
= C

(,
0,1
u

TX) for u C

. The c

u
t together to form
a Banach orbibundle c

. There is a section s
J
: C

, dened by
u

J,
u , and M

is the zero locus of s

J
. If has no nontrivial automorphism,
then C

is a Banach manifold, and c

is a Banach bundle. In this case,

if M

is nonempty and D
u
is surjective for all u M

, then M

is a smooth
manifold of dimension +N(2 2g h) by the implicit function theorem. We call
+N(2 2g h) the virtual dimension of M

. In general, M

is singular, and the

actual dimension of M

can be larger than the virtual dimension.

The dimension of

M
(g,h),(n, m)
is 6g +3h6+2n+m
1
+ +m
h
, so the expected
(or virtual) dimension of

M
(g,h),(n, m)
(X, L [ , , ) is
+ (N 3)(2 2g h) + 2n +m
1
+ + m
h
,
which will also be the virtual dimension of the Kuranishi structure on

M
(g,h),(n, m)
(X, L [ , , ).
45
Similarly, we have
c
l

C
l

, s
J
: C
l

c
l

,
c
k,p

W
k,p

, s
J
: W
k,p

c
k,p

,
and M

is the zero locus of s

J
in the above spaces.
6.3. Deformation of the domain. Let
= [(, B; p; q
1
, . . . , q
h
; u)]
be a point in

M
(g,h),(n, m)
(X, L [ , , ). We have seen in Section 3.3 that the
innitesimal deformation of the domain
= [(, B; p; q
1
, . . . , q
h
)]
is given by
H
,domain
=

i=1
W
i

=1

W
i

l0

=1
V

l1

=1

V
+

,
where
W
i
= H
1
(

C
i
, T

Ci
(D
y
i )),

W
i
= H
1
(

i
,

i
, T

(p
i

1
p
i

n
i

), T

(q
i

1
. . . q
i

m
i

)),
V

= T
r

T
r
l
0
+

= C,

= T
s

T
s
l
1
+

= R,

V
+

= [0, )

V

are dened as in Section 3.3.

V
+

is only a semigroup, while the others are vector

spaces. Set
H
,deform
=

i=1
W
i

=1

W
i

H
,interior
=
l0

=1
V

H
,boundary
=

V
+
1


V
+
l1
H
,smooth
= H
,interior
H
,boundary
Then H
,deform
corresponds to tangent directions of the stratum that belongs to,
while H
,smooth
corresponds to normal directions to this stratum. H
,interior
corre-
sponds to smoothing of interior nodes, and H
,boundary
corresponds to smoothing
of boundary nodes.
Let (Aut )
0
denote the identity component of Aut . (Aut )
0
is a normal
subgroup of Aut , and the quotient Aut

= Aut /(Aut )
0
is a nite group.
Aut acts on H
,deform
, and (Aut )
0
acts trivially, so Aut

acts on H
,deform
.
We choose an admissible metric h on in the sense of Section 6.2.2. Let
1
be a
small positive number, and dene N
1
(), K
1
() as in the paragraph right before
Denition 5.5. Then N
1
(), K
1
() are invariant under Aut . We may choose a
subspace

H of the space of smooth Beltrami dierentials such that the elements in

H vanish on N
1
(), and the natural map

H H
,deform
is an isomorphism. We
may further assume that

H is invariant under the action of Aut

, so that Aut

46
acts on

H and the isomorphism

H H
,deform
is Aut

-equivariant. From now on,

we will identify

H with H
,deform
.
6.3.1. Deformation within the stratum. Let j() be the complex structure on
determined by

H, and let
(,0,0)
be the prestable bordered Riemann surface
corresponding to (, j()). In particular, j(0) is the original complex structure j
on , and
(0,0,0)
= . Let

(,0,0)
= (
(,0,0)
, B; p; q
1
, . . . , q
h
).
Let
(,0,0)
:
(,0,0)
be the identity map. Then
(1)
(,0,0)
:
(,0,0)
is a strong deformation in the sense of Denition 4.8.

(,0,0)
:
(,0,0)
is a homeomorphism.
(2) j =
_

1
(,0,0)
_

j() on N
1
().
(3) | j
_

1
(,0,0)
_

j() |
C

(K
1
())
< C[[, where [[ is the Weil-Petersson
norm of the Beltrami dierential .
Note that any two norms on H
,deform
are equivalent since H
,deform
is nite
dimensional. Let B
2
H
,deform
be the ball of radius
2
> 0 centered at the
origin. From the above discussion, we see that there is a family of prestable bordered
Riemann surfaces of type (g, h) with (n, m) marked points
(,0,0)
[ B
2
. More
precisely, we have
( : ( B
2
; s; t
1
, . . . , t
h
)),
where s = (s
1
, . . . , s
n
), t
i
= (t
i
1
, . . . , t
i
m
i
), and a contraction : ( . Dieomor-
phically, ( = B
2
, is the projection to the rst factor, is the projection of
the second factor, s
j
, t
i
k
: B
2
( are constant sections corresponding to marked
points p
j
, q
i
k
, respectively. Holomorphically,
1
() =
(,0,0)
.
Let
0
be the volume form on determined by h. Then
0
is a Kahler form. Let
h
(,0,0)
be the Hermitian metric on
(,0,0)
determined by

(,0,0)

0
and j(). Then

(,0,0)
is an isometry near the boundary and nodes, so h
(,0,0)
is an admissible
metric on
(,0,0)
.
There is a map i : B
2


M
(g,h),(n, m)
, given by
(,0,0)
. Given
2
> 0, there
exists
2
> 0 such that i(B
2
) U(,
1
,
2
), where U(,
1
,
2
) is the neighborhood
of in

M
(g,h),(n, m)
in the C

topology dened in Denition 5.6. i(B

2
) is a
neighborhood of in the stratum of

M
(g,h),(n, m)
which belongs to.
6.3.2. Smoothing of interior nodes. Let s be an interior node of , and let i :
B
1
(s) C
2
be a holomorphic isometry such that i(B
1
(s)) = (x, y) C
2
[ xy =
0, [x[ <
1
, [y[ <
1
. Let s
1
, s
2


be the preimages of s under :

.
Up to permutation of s
1
, s
2
, there exist unique e
1
T
s1

, e
2
T
s2

such that
(i )

(e
1
) = (1, 0) C
2
, and (i )

(e
2
) = (0, 1) C
2
.
Given v T
s1

T
s2

, we have v = te
1
e
2
for some t C. Suppose that
t = r
2
e
i
, where 0 < r <
1
3
. Let
t
be the bordered Riemann surface obtained
from by replacing
B
1,0
= (x, y) C
2
[ xy = 0, [x[ <
1
, [y[ <
1

with
B
1,t
= (x, y) C
2
[ xy = t, [x[ <
1
, [y[ <
1

47
More precisely, B
1,t
is obtained by identifying
x = re
i
x C [ [x[ = r x C [ r [x[ <
1

with
t
x
= re
i()
y C [ [x[ = r y C [ r [y[ <
1
.
There is a strong deformation

t
: B
1,t
=
__
x,
t
x
_

r
2

1
< [x[ <
1
_
B
1,0
given by

t
_
x,
t
x
_
=
_
(

x
r

2
)x, 0
_
if r [x[ <
1
_
0, (

t
xr

2
)
t
x
_
if
r
2

1
< [x[ r
where : R [0, 1] is a smooth function such that
(1) 0

(s) 1.
(2) (s) = 0 for s 1.
(3) (s) > 0 for s > 1.
(4) (s) = 1 for s 4.
Lemma 6.16. Let f, g be smooth functions on C such that f(0) = g(0). Let F be
the continuous function on z C [
r
2
1
< [z[ <
1
dened by
F(z) =
_
f((

z
r

2
)z) if r [z[ <
1
g
_
(

t
zr

2
)
t
z
_
if
r
2
1
< [z[ r
Then F is smooth.
Proof. The lemma follows from
(1) Both h
1
(z) = f((

z
r

2
)z), h
2
(z) = g
_
(

t
zr

2
)
t
z
_
are smooth for
r
2
1
<
[z[ <
1
.
(2) The derivatives of h
1
of any order vanish when [z[ = r, and the same is
true for h
2
. 2
Let A(r, R) denote the annulus (u, v) R
2
[ r
2
u
2
+v
2
R
2
, and let

A(r, R)
denote its interior. Let D
R
denote the closed disc (u, v) R
2
[ u
2
+v
2
R
2
.
Lemma 6.17. Let f, g, F be dened as in Lemma 6.16. Dene h :

A(
r
2
1
,
1
) C
by h(u, v) = F(u +iv). Then
max
A(
r
2
,2r)
[h[ = maxmax
D2r
[f[, max
D2r
[g[,
max
A(
r
2
,2r)
[h[ 9

2 maxmax
D2r
[f

[, 4 max
D2r
[g

[,
where [h[
2
= [h
u
[
2
+[h
v
[
2
.
48
Proof. We have
h(u, v) =
_
f((
u
2
+v
2
r
2
)(u +iv)) if r
2
u
2
+v
2
<
2
1
,
g
_
(
r
2
u
2
+v
2
)
t
u+iv
_
if
r
4

2
1
< u
2
+v
2
r
2
,
so
sup
A(
r
2
,2r)
[h[ = maxmax
D2r
[f[, max
D2r
[g[.
We also have
h
u
(u, v) =
_
f

((
u
2
+v
2
r
2
)(u +iv))(

(
u
2
+v
2
r
2
)
2u(u+iv)
r
2
+(
u
2
+v
2
r
2
)) if r
2
< u
2
+v
2
<
2
1
,
g

((
r
2
u
2
+v
2
)
t
u+iv
)(

(
r
2
u
2
+v
2
)
2r
2
u
(u
2
+v
2
)
2
t
u+iv
(
r
2
u
2
+v
2
)
t
(u+iv)
2
) if
r
4

2
1
< u
2
+ v
2
< r
2
,
and h
u
(u, v) = 0 for u
2
+v
2
= r
2
, so
sup
A(
r
2
,2r)
[h
u
[ max9 max
D2r
[f

[, 36 max
D2r
[g

[.
Similarly,
sup
A(
r
2
,2r)
[h
v
[ max9 max
D2r
[f

[, 36 max
D2r
[g

[. 2
In particular, let (f(x), g(y)) = (x, 0), (0, y). We see that
t
is smooth as a map
to C
2
.
t
is a dieomorphism when [x[ , = r, and
1
t
(0, 0) =
_
x,
t
x
) [ [x[ = r
_
.
Choose a Hermitian metric h
t
on B
1,t
such that it is induced by inclusion in C
2
on B
2r,t
and
t
is an isometry outside B
3r,t
.
We now have a family of prestable bordered Riemann surfaces of type (g, h) with
(n, m) marked points

t
= (
t
, B; p; q
1
, . . . , q
h
)
together with a family of admissible metrics h
t
on
t
such that
(1) There are strong deformations
t
:
t
such that on K
3r
(), where
r =
_
[t[,
1
t
is dened and is an isometry.
(2) j =
_

1
t
_

j
t
on K
3r
(), where j
t
is the complex structure on
t
.
Let D

2
1
/9
= te
1
e
2
[ [t[ <
2
1
/9 T
s1

T
s2

. The map D

2
1
/9


M
(g,h),(n, m)
given by te
1
e
2

t
denes a parametrized curve in

M
(g,h),(n, m)
whose tangent
line at
0
= is T
s1

T
s2

H
,interior
.
Let = (v
1
, . . . , v
l0
) H
,interior
, where v

= T
r

T
r
l
0
+

. Applying
the above construction to each interior node on
(,0,0)
, we obtain

(,,0)
= (
(,,0)
, B; p; q
1
, . . . , q
h
).
Given 0 < d
1
, . . . , d
l0
<
2
1
/9, let D(d
1
, . . . , d
l0
) denote the polydisc D
d1
. . . D
d
l
0
in H
,interior
, where D
d
is the disc of radius d

centered at the origin in V

. We
have a family
( : ( B
2
D(d
1
, . . . , d
l0
); s; t
1
, . . . , t
h
)
of prestable bordered Riemann surfaces of type (g, h) with (n, m) marked points
together with a family of admissible metrics h
(,,0)
on
(,,0)
. There is a con-
traction : ( whose restriction to
1
(, ) =
(,,0)
is a strong deformation

(,,0)
:
(,,0)
such that
(1)
1
(,,0)
is dened on K
3

||
(), and
1
(,,0)

(,0,0)
is an isometry on
K
3

||
(
(,0,0)
).
49
(2) | j
_

1
(,,0)
_

j(, ) |
C

(K
1
())
=| j
_

1
(,0,0)
_

j() |
C

(K
1
())
<
C[[, where j(, ) is the complex structure on
(,,0)
.
6.3.3. Smoothing of boundary nodes. Let s be a boundary node of , and let i :
B
1
(s) C
2
be a holomorphic isometry such that
i(B
1
(s)) = (x, y) C
2
[ xy = 0, [x[ <
1
, [y[ <
1
, Imx 0, Imy 0,
Let s
1
, s
2


be the preimages of s under :

. Up to permutation of
s
1
, s
2
, there exist unique e
1
T
s1

, e
2
T
s2

such that (i )

(e
1
) = (1, 0) C
2
,
and (i )

(e
2
) = (0, 1) C
2
.
Given v T
s1

T
s2

, we have v = te
1
e
2
for some t R. We construct
t
as in Section 6.3.2. We have seen in Section 3.3 that there is topological transition
when t changes sign. We may assume that
t


M
(g,h),(n, m)
for t 0.
Let

= (v

1
, . . . , v

l0
) H
,boundary
, where v

V
+

. Apply the above con-

struction to each boundary node on
(,,0)
, we obtain
(,,

)
and
(,,

)
. Given
0 < d

1
, . . . , d

l1
<
2
1
/9, let D

(d

1
, . . . , d

l0
) = [0, d

1
) . . . [0, d

l1
)

V
1
. . .

V
l1
.
We have a universal family
( : ( B
2
D(d
1
, . . . , d
l0
) D

(d

1
, . . . , d

l1
); s; t
1
, . . . , t
h
)
of prestable bordered Riemann surfaces of type (g, h) with (n, m) marked points
together with a family of admissible metrics h
(,,

)
on
(,,

)
. There is a con-
traction : ( whose restriction to
1
(, ,

) is a strong deformation

(,,

)
:
(,,

)
such that
(1)
1
(,,

)
is dened on K
3

||+|

|
(), and
1
(,,

)

(,0,0)
is an isometry
on K
3

||+|

|
(
(,0,0)
).
(2) | j
_

1
(,,

)
_

|
C

(K
1
())
< C[[.
If we embed
C
in a complex projective space P
N
, then
(,,

)
is smooth
as a map to P
N
, where :

(,,

)

(,,

)
is the normalization map.
Given
1
,
2
> 0, choose
2
as before. Suppose that both max
_
[d

[ [ =
1, . . . , l
0
and max
_
[d

[ [

= 1, . . . , l
1
are less than
1
3
. Then the image of the
map
i : B
2
D(d
1
, . . . , d
l0
) D

(d

1
, . . . , d

l1
)

M
(g,h),(n, m)
given by (, ,

)
(,,

)
lies in the neighborhood U(,
1
,
2
) of in the C

topology.
6.3.4. Action of the automorphism group. We write D
d
for D(d
1
, . . . , d
l0
) and D

for D

(d

1
, . . . , d

l1
). In this section, we study the action of Aut on B
2
D
d
D

and the universal family over it.

We rst consider deformation within the stratum. Let
2
: (
2
B
2
be
the universal family, so that
1
2
() =
(,0,0)
. Aut acts on B
2
by j( ) =
(
1
)

j(). Therefore, it acts on the the universal family. Given Aut , we

50

1

2
Figure 19. =
(0,0)

(t,0)

(0,t)
Figure 20.
(t,0)
and
(0,t)
have the following commutative diagram:
(
2

(
2

2
B
2

B
2
:
(,0,0)

(,0,0)
is an isomorphism. In particular, if (Aut )
0
, then
: B
2
B
2
is the identity map, and we have the following commutative diagram:

(,0,0)

(,0,0)

(,0,0)

(,0,0)

We now consider smoothing of nodes. Let
2,d,d
: (
2,d,d
B
2
D
d
D

d
be
the universal family, so that
1
2,d,d
(, ,

) =
(,,

)
. Aut acts on D
d
D

d
by
(te
1
e
2
) = t

e
1

e
2
. Given Aut , we have the following commutative
diagram
(
2,d,d

(
2,d,d

2
,d,d

2
,d,d

B
2
D
d
D

B
2
D
d
D

:
(,,

)

(,,

)
is an isomorphism. For example, the prestable bordered
Riemann surface in Figure 15 has two interior nodes. The smoothing of the two
interior nodes is parametrized by = (
1
,
2
). Let
(1,2)
be the corresponding
bordered Riemann surfaces obtained by smoothing the two interior nodes on .
has an automorphism of order two which rotates Figure 19 by 180

. It
acts on by (
1
,
2
) = (
2
,
1
) and gives an isomorphism
(1,2)

(2,1)
by
rotating 180

. The case
2
= 0 is shown in Figure 20.
51
6.4. Local Charts. Let
= [(, B; p; q
1
, . . . , q
h
; u)]
be a point in

M
(g,h),(n, m)
(X, L [ , , ), as at the beginning of Section 6.3. Consider
D
u
: W
1,p
(, , u

TX, (u[

TL) L
p
(,
0,1
u

TX)
for a large integer p. L
p
(,
0,1
u

TX) is a complex vector space, but W

1,p
(, , u

TX, (u[

TL)
is only a real vector space. Aut acts on both W
1,p
(, , u

TX, (u[

TL) and
L
p
(,
0,1
u

TX), and D
u
is Aut -equivariant.
Lemma 6.18. Let ImD
u
denote the image of D
u
. We can choose a subspace E

of L
p
(,
0,1
u

TX) such that

(1) ImD
u
+E

= L
p
(,
0,1
u

TX).
(2) E

is nite dimensional.
(3) Elements in E

are smooth sections supported in K

1
().
(4) E

is a complex subspace of L
p
(,
0,1
u

TX).
(5) E

is Aut -invariant.
Proof. We rst claim that given any L
p
(,
0,1
u

TX), there exists g

W
1,p
(, , u

TX, (u[

TL) such that

= D
u
g has support in K
1
().
Actually, there exists g

dened on N
21
() such that D
u
g

= on N
21
(). Let
be a smooth function on which is 1 on N
1
() and 0 on K
21
(). Let g = g

on
N
21
(), and 0 on K
21
(). Then

= D
u
g is supported in K
1
().
By Corollary 6.14, we can nd a nite dimensional subspace E

of
L
p
(,
0,1
u

TX) such that ImD

u
E

= L
p
(,
0,1
u

TX). We may
assume that E

consists of smooth sections since any section in

L
p
(,
0,1
u

TX) can be approximated by sections in C

(,
0,1
u

TX).
The above claim shows that we may assume that all sections in E

have compact
support in K
1
(). Let E

be the smallest Aut -invariant complex subspace which

contains E

. Then E

satises (1)(5). 2
Let F

= L
p
(,
0,1
u

TX) E

, where E

is the orthogonal complement

of E

in L
2
(,
0,1
u

TX). Then F

is a closed subspace of
L
p
(,
0,1
u

TX), thus a Banach subspace of L

p
(,
0,1
u

TX). F

=
L
p
(,
0,1
u

TX)/E

. Let
: L
p
(,
0,1
u

TX) F

be the L
2
-orthogonal projection. (1) implies that D
u
is surjective. Multiplication
by i preserves the L
2
inner product, so (4) implies that F

is a complex vector space.

The action of Aut also preserves the L
2
inner product, so (5) implies that Aut
acts on F

, and D
u
: W
1,p
(, , u

TX, (u[

TL) F

is Aut -equivariant.
Set H
,map
= Ker( D
u
). We have
dimH
,map
= +N(1 g) + dimE

,
where = (u

TX, (u[

TL), 2N = dim
R
X, and g is the arithmetic genus of

C
, as before.
The innitesimal deformation of the domain is given by
H
,aut
=

i=1
U
i

=1

U
i
,
52
where
U
i
= H
0
(

C
i
, T

Ci
(x
i
1
x
i
ni
))

U
i
= H
0
(

i
,

i
, T

(p
i

1
p
i

n
i

), T

(q
i

1
. . . q
i

m
i

))
U
i
= 0 if and only (

C
i
, x
i
1
, , x
i
ni
) is stable, and

U
i
= 0 if and only if
(

i
, (p
i

1
, , p
i

n
i

), (q
i

1
, . . . , q
i

m
i

)) is stable.
Let = (, B; p; q
1
, . . . , q
h
) as before. Then H
,aut
is the tangent space to Aut
at the identity map. u is nonconstant on nonstable components, so Aut
u
1
induces an inclusion of vector spaces H
,aut
H
,map
. Let H

,map
be the
L
2
-orthogonal complement of H
,aut
in H
,map
. Set
H

= H
,domain
H
,map
,
H

= H
,domain
H

,map
.
With the above denitions, we are ready to state the main theorem of this
section:
Theorem 6.19. Let

M
(g,h),(n, m)
(X, L [ , , ) be equipped with the C

topology.
There are a neighborhood V

of 0 in H

such that Aut acts on V

, an Aut -
equivariant map s

: V

such that s

(0) = 0, and a continuous map

:
s
1

(0)

M
(g,h),(n, m)
(X, L [ , , ) such that s
1

(0)/Aut

M
(g,h),(n, m)
(X, L [
, , ) gives a homeomorphism onto a neighborhood of in

M
(g,h),(n, m)
(X, L [
, , ).
(V

, E

, Aut ,

, s

) is a Kuranishi neighborhood of .
6.4.1. Pregluing: construction of approximate solutions. In this section, we will
modify u near nodes to obtain approximate J-holomorphic maps
u
,
: (
(0,,

)
,
(0,,

)
) (X, L),
where the notation
(,,

)
was introduced in Section 6.3.1, 6.3.2, 6.3.3 and will be
used repeatedly in the rest of Section 6.4.
We rst consider a neighborhood of an interior node. We will follow the con-
struction in [MS, Appendix A] closely. Recall that
B
1,t
= (x, y) C
2
[ xy = t, [x[ <
1
, [y[ <
1
.
When t ,= 0, we have
B
1,t
=
__
x,
t
x
_

x C,
r
2

1
< [x[ <
1
_
,
where [t[ = r
2
. We assume that r <

2
1
4
.
Let B
1
= x C [ [x[ <
1
. Two nonconstant J-holomorphic maps f, g :
B
1
X such that f(0) = g(0) = p determine a stable map u : B
1,0
X dened
by u(x, 0) = f(x), u(0, y) = g(y). Suppose that the images of f, g are contained
in the geodesic ball B
r1
(x) with respect to g
1
, where r
1
is the injectivity radius of
(M, g
1
). We dene u
t
: B
1,t
X by
u
t
_
z,
t
z
_
= exp
p
_

1
_
z

r
_
(exp
p
)
1
(f(z)) +
1
_
r

r
z
_
(exp
p
)
1
(g
_
t
z
_
)
_
,
53
where
1
: C [0, 1] is a smooth cuto function such that

1
(z) =
_
1, if [z[ 2
0, if [z[ 1
[
1
[ 2
Then
u
t
_
z,
t
z
_
=
_
_
_
u
_
0,
t
z
_
= g(
t
z
) if
r
2
1
< [z[ <
r

r
2
,
p if r

r [z[

r,
u(z, 0) = f(z) if 2

r < [z[ <

1
.
Dene f
t
, g
t
: B
1
X by
f
t
(x) = exp
p
_

1
_
x

r
_
(exp
p
)
1
(f(x))
_
g
t
(y) = exp
p
_

1
_
y

r
_
(exp
p
)
1
(g(y))
_
Then
f
t
(x) =
_
f(x) if 2

r < [x[ <

1
p if [x[ <

r
g
t
(y) =
_
g(y) if 2

r < [x[ <

1
p if [x[ <

r
f
t
, g
t
determine a map v
t
: B
1,0
X dened by v
t
(x, 0) = f
t
(x), v
t
(0, y) = g
t
(y).
Dene F
t
, F : B
1
X by f
t
(x) = exp
p
(F
t
(x)), f(x) = exp
p
(F(x)). Then F
t
(0) =
F(0) = 0, and
F
t
(x) =
1
_
x

r
_
F(x)
Lemma 6.20.
| F
t
F |
W
1,p
(B
1
)
C max

B
2

r
[F[r
1
p
,
where C is a universal constant, and

B
2

r
= x C [ [x[ 2

r.
Proof. F
t
(x) F(x) = 0 for [x[ 2

r, and for [x[ < 2

r,
[F
t
(x) F(x)[ = [(
1
_
x

r
_
1)F(x)[ [F(x)[ (max

B
2

r
[F[)[x[
[(F
t
(x) F(x))[ =

1
_
x

r
_
F(x) +
1
_
x

r
_
F(x)

r
[F(x)[ +[F(x)[ 5 max

B
2

r
[F[
The desired estimate is obtained by integrating over B
2

r
, and noting that r < 1.
2
We embed (X, g
1
) isometrically in R
l
for some large l. Maps to X can be viewed
as maps to R
l
, so we may subtract one map from another, and dene their L
1,p
norms.
54
Corollary 6.21.
| v
t
u |
W
1,p
(B
1
)
C(max

B
2

r
[u[)r
1
p
,
where C depends on the C

norms of exp [
Br
1
(N
L/X
)
and its inverse.
Lemma 6.22.
|

J
u
t
|
L
p Cr
1
p
where C depends on the C
1
norm of u, the C
1
norm of J, the C

norms of
exp[
Br
1
(N
L/X
)
and its inverse.
Proof. For r [z[ <
1
, we have
u
t
_
z,
t
z
_
= f
t
(z),
so

J
u
t
(z) = 0 for [z[ >

r. For r [z[

r, we have

J
u
t
_
z,
t
z
_
=

J
f
t
(z) =

J
(f
t
f)(z).
Let z = x +iy ,then
2

J
(f
t
f)
_

x
_
=
f
t
x
+J(f
t
)
f
t
y

f
x
J(f)
f
y
=

x
(f
t
f) + (J(f
t
) J(f))
f
y
+J(f)

y
(f
t
f) + (J(f
t
) J(f))

y
(f
t
f)

J
(f
t
f)
_

x
_

C
1
(1 + sup [J[ + sup[J[[f
t
f[)[(f
t
f)[ +C
2
sup [J[[f
t
f[[f[
C
3
(1 + sup [J[ + sup[J[ sup [f[[z[)[(f
t
f)[ +C
4
sup[J[ sup [f[[z[ sup[f[
C
5
[(f
t
f)[ +C
6
[z[
Similarly,

J
(f
t
f)
_

y
_

C
5
[(f
t
f)[ +C
6
[z[
The case
r
2
1
< [z[ < r can be estimated similarly. Therefore,
|

J
u
t
|
L
p C
7
(| (v
t
u) |
L
p +

rr
1
p
) Cr
1
p
where C depends on the C
1
norms of u and of J, C

norms of exp
1
[
Br
1
(N
L/X
)
and
its inverse. 2
We next consider boundary nodes. For t [0, ), dene
B
+
1,t
= (x, y) C
2
[ xy = t, [x[ <
1
, [y[ <
1
, Imx 0, Imy 0.
Then for t = r
2
> 0,
B
+
1,t
=
__
x,
r
2
x
_

x C,
r
2

1
< [x[ <
1
, Imx 0
_
Let B
+
1
= x C [ [x[ <
1
, Imx 0. Let I
1
= B
+
1
R, and I
1,t
= B
+
1,t

R R. Two nonconstant J-holomorphic maps f, g : (B
+
1
, I
1
) (X, L) such that
55
f(0) = g(0) = p determine a stable map u : (B
+
1,0
, I
1,0
) (X, L) dened by
u(x, 0) = f(x), u(0, y) = g(y). One can construct u
t
: (B
+
1,t
, I
1,t
) (X, L) and
v
t
: (B
+
1,0
, I
1,0
) (X, L) as before.
Applying the above construction to each node, we obtain
u
,
: (
(0,,

)
,
(0,,

)
) (X, L)
v
,
: (, ) (X, L)
Lemma 6.22 implies
Lemma 6.23.
|

J
u
,
|
L
p C([[ +[

[)
1
2p
where C depends on the C
1
norms of u and J, C

norms of exp [
Br
1
(N
L/X
) and
its inverse.
The linearization of

J,
at the stable W
1,p
map v
,
is
D
v
,

: W
1,p
(, , v

,
TX, (v
,
[

TL) L
p
(,
0,1
v

,
TX)
Lemma 6.24.
lim
(,

)0
| D
v
,

|=| D
u
|
Proof. We have a bundle isomorphism
P
0
: (u

TX, (u[

TL) (v

,
TX, (v
,
[

TL)
given by parallel transport along the unique length minimizing geodesic from u(z)
to v
,
(z). This also gives
P
1
:
1
u

TX
1
v

,
TX
and
P

1
= P
1
i :
0,1
u

TX
0,1
v

,
TX,
where i :
0,1
u

TX
1
u

TX is the inclusion, and :

1
v

,
TX

0,1
v

,
TX is the projection. P
0
, P

1
induces

P
1
,

,0
: W
1,p
(, , v

,
TX, (v
,
[

TL) W
1,p
(, , u

TX, (u[

TL)

P
,

,1
: L
p
(,
0,1
u

TX) L
p
(,
0,1
v

,
TX)
Dene
D

,
=

P
,

,1
D
u

P
1
,

,0
: W
1,p
(, , v

,
TX, (v
,
[

TL)
L
p
(,
0,1
v

,
TX),
then
lim
(,

)(0,0)
| D

,
|=| D
u
| .
From Lemma 6.12, we see that
| (D
v
,

,
)w | C
2
| u v
,
|
C
0| w |
L
p + | w |
C
0| du dv
,
|
L
p
C
3
| u v
,
|
W
1,p| w |
W
1,p
| D
v
,

,
| C
3
| u v
,
|
W
1,p
which tends to 0 as (,

) (0, 0). 2
56
6.4.2. Gluing: construction of exact solutions. The goal of this section is to con-
struct a local parametrization of solutions to

J
v = 0 near the approximate
J-holomorphic map u
,
constructed in Section 6.4.1. The main result in this
Section is Proposition 6.32.
Let B
2
D(d
1
, . . . , d
l0
) D

(d

1
, . . . , d

l1
) be the neighborhood of the origin
in H
,domain
as in Section 6.3.3. We write D
d
for D(d
1
, . . . , d
l0
), and D

d
for
D

(d

1
, . . . , d

l1
), as in Section 6.3.4. We have seen that there is a family of prestable
bordered Riemann surfaces
( : ( B
2
D
d
D

d
; s; t
1
, . . . , t
h
)
of type (g, h) with (n, m) marked points, together with a family of admissible met-
rics, such that
1
(0) = . There is a map ( whose restriction to each ber
of is a smooth strong deformation
(,,

)
:
(,,

)
.
Let B be the image of the map i : B
2
D
d
D

d


M
(g,h),(n, m)
given by
(, ,

)
(,,

)
. Then B is a neighborhood of in the C

topology. Us-
ing the family of admissible metrics, we dene W
B
=

B
W
1,p

. Let M
B
=

M
(g,h),(n, m)
(X, L [ , , ) W
B
. Then M
B
.
We have a Cartesian diagram
W

B
W
B

B
2
D
d
D

d
=

B
i
B
Let S :

B W

B
be given by (, ,

) u
,
.
We rst extend E

L
p
(,
0,1
u

TX) to a trivial bundle over S(0

D
d
D

d
). Recall that elements in E

are supported on K
1
(). Since v
,
= u
on K
1
(), E

can be viewed as a subspace E

,
of L
p
(,
0,1
v

,
TX). Let
F
,
be the L
2
-orthogonal complement of E
,
in L
p
(,
0,1
v

,
TX), and let

,
: L
p
(,
0,1
v

,
TX) F
,

be the L
2
-orthogonal projection. Then for suciently small ,

,
D
v
,

: W
1,p
(, , v

,
TX, (v
,
[

TL) F
,

is a surjection. Let H
(,

)
= Ker(
,
D
v
,

), and let H

(,

)
be its L
2
-orthogonal
complement in W
1,p
(, , v

,
TX, (v
,
[

TL). Then we have an isomorphism

,
D
v
,

: H

,
F
,

whose inverse is

Q
,
: F
,
H

We have
Q
,
= i

Q
,
: F
,
W
1,p
(, , v

,
TX, (v
,
[

TL)
where
i : H

(,

)
W
1,p
(, , v

,
TX, (v
,
[

TL)
is the inclusion. Q
,
is a right inverse of
,
D
v
,

.
By Lemma 6.24, we may choose d, d

suciently small such that

Q
,
exists and
| Q
,
| M for all (,

) D
d
D

d
, where M is a constant.
57
We now extend E S(0 D
d
D

d
) to a neighborhood U

B
of S(

B) in W

B
.
Let

= (
(,,

)
, f), where f : (
(,,

)
,
(,,

)
) (X, L) is a stable W
1,p
map
such that
sup

(,,

)
d
1
(u
,
(z), f(z))
is less than the injectivity radius of g
1
, where d
1
is the geodesic distance of g
1
.
We have a bundle isomorphism
P
0
: (u

,
TX, (u
,
[

(,,

)
)

TL) (f

TX, (f[

(,,

)
)

TL)
given by the parallel transport along the unique length minimizing geodesic from
u
,
(z) to f(z), which gives
P
1
:
1

(,,

)
u

,
TX
1

(,,

)
f

TX
P

1
= P
1
i :
0,1

(,,

)
u

,
TX
0,1

(,,

)
f

TX
where i :
0,1

(,,

)
u

,
TX
1

(,,

)
u

,
TX is the inclusion, and

(,,

)
f

TX
0,1

(,,

)
f

TX is the projection. We have

P : L
p
(
(,,

)
,
0,1

(,,

)
u

,
TX) L
p
(
(,,

)
,
0,1

(,,

)
f

TX).
Let E

=

PE

= E

. Then we have a trivial bundle E U

B
together with a
trivialization : E

= U

B
E

.
Let F

be the L
2
-orthogonal complement of E

in L
p
(
(,,

)
,
0,1

(,,

)

f

TX), and let

: L
p
(
(,,

)
,
0,1

(,,

)
f

TX) F

be the L
2
-orthogonal projection.
We will use Q
,
, the right inverse of
,
D
v
,

, to construct an approximate
right inverse Q

(,,

)
of
(,,

)
D
(,,

),u
,

, where
(,,

)
=
(
(,,

)
,u
,
)
,
and D
(,,

),u
,

is the linearization of

J,
(,,

)
at u
,
. We will use the cuto
function constructed in [MS, A.1]. The construction in this section works for W
1,p
but not for general W
k,p
.
Lemma 6.25. For any r (0, 1), there is a smooth cuto function
r
: C [0, 1]
such that

r
(z) =
_
1 if [z[ r

r
0 if [z[ r
_
|z|r

r

4
[ log r[
.
Proof. We will follow the proof of [MS, Lemma A.1.1]. We rst dene a cuto
function of class W
1,2
by

r
(z) =
_

_
1 for [z[ r

r
2
_
log |z|
log r
1
_
for r

r [z[ r
0 for [z[ r
Then we have
[
r
(z)[ =
2
[z[[ log r[
58
for r

r [z[ r, so
_
r

r|z|r
[
r
(z)[
2
=
4
[ log r[
.
To obtain a smooth function
r
, take the convolution with
N
(z) = N
2
(Nz)
where N is large and : C R is a smooth function with support in the unit ball
and mean value 1. 2
Let p > 2 be xed as before.
Lemma 6.26. For every r (0, 1), there exists a smooth cuto function
r
: C
[0, 1] as in Lemma 6.25 such that
| (
r
)w |
L
p C | w |
W
1,p [ log r[
1
p
1
for any w W
1,p
(C) with w(0) = 0.
Proof. It follows from the proof of [MS, Lemma A.1.2]. 2
We now look at the local model of an interior node. Let u : B
1,0
X be a
stable map, and construct smooth maps u
t
: B
1,t
X, v
t
: B
1,0
X as before.
We now dene linear maps
e
t
: L
p
(B
1,t
,
0,1
B
1,t
u

t
TX) L
p
(B
1,0
,
0,1
B
1,0
v

t
TX)
g
t
: W
1,p
(B
1,0
, v

t
TX) W
1,p
(B
1,t
, u

t
TX)
Given s L
p
(B
1,t
,
0,1
B
1,t
u

t
TX), we dene
e
t
(s) L
p
(B
1,0
,
0,1
B
1,0
v

t
TX)
by
e
t
(s)(x, 0) =
_
s(x,
t
x
) if r [x[
1
0 if [x[ < r
e
t
(s)(0, y) =
_
s(
t
y
, y) if r [y[
1
0 if [y[ < r
The above denition is valid since
u
t
_
z,
t
z
_
=
_
v
t
(0,
t
z
) if
r
2
1
[z[ r
v
t
(z, 0) if r [x[
1
Given w W
1,p
(B
1,0
, v

t
TX), we dene
g
t
(w) W
1,p
(B
1,t
, u

t
TX)
by
g
t
(w)
_
z,
t
z
_
=
_

_
w(z, 0) if

r [z[
1
w(z, 0) + (1
r
(
t
z
))(w(0,
t
z
) w(0, 0)) if r [z[

r
w(0,
t
z
) + (1
r
(z))(w(z, 0) w(0, 0)) if r

r [z[ r
w(0,
t
z
) if
r
2
1
[z[ r

r
Lemma 6.27. If s L
p
(B
1,t
,
0,1
B
1,t
u

t
TX), w W
1,p
(B
1,0
, v

t
TX) satisfy
D
vt
w = e
t
(s), then
| D
ut
g
t
(w) s |
L
p C | w |
W
1,p ([ log r[
1
p
1
+ | u |
W
1,p r
1
p
)
59
Proof. We have
D
ut
g
t
(w)
_
z,
t
z
_
=
_
D
vt
w(z, 0) = e
t
(s)(z, 0) = s(z,
t
z
) for

r [z[
1
,
D
vt
w(0,
t
z
) = e
t
(s)(0,
t
z
) = s(z,
t
z
) for
r
2
1
[z[ r

r.
We now consider r [z[

r. We have
D
ut
g
t
(w)
_
z,
t
z
_
= D
vt
w(z, 0) +D(1
r
_
t
z
_
)(w(0,
t
z
) w(0, 0))
+(1
r
_
t
z
_
)(D
vt
w(0,
t
z
) D
vt
w(0, 0))
where D is a derivation, so
[D(1
r
_
t
z
_
)[ C
1

t
z
2

r
_
t
z
_

.
We also have
D
vt
w(z, 0) = e
t
(s)(z, 0) = s(z,
t
z
),
D
vt
w(0,
t
z
) = e
t
(s)(0,
t
z
) = 0,
and
[D
vt
w(0, 0)[ C
2
| v
t
|
W
1,p [w(0, 0)[
C
2
| v
t
|
W
1,p| w |
C
0
C
3
| v
t
|
W
1,p| w |
W
1,p
Let h
0
be the metric on A(r,

r) given by A(r,

r) C, and h
1
be the metric on
A(r,

r) given by the embedding z A(r,

r) (z,
t
z
) C
2
. Then
h
1
=
_
1 +
_
r
[z[
_
4
_
h
0
,
so h
0
h
1
2h
0
. To estimate the L
1,p
norm dened by any metric which is an
interpolation of h
0
and h
1
, it suces to calculate in h
0
.
_
r|z|

r
[D
ut
g
t
(w)(z,
t
z
) s(z,
t
z
)[
p
i
2
dz d z
C
p
1
_

r|z|r

r
_
t
z
_

w(0,
t
z
) w(0, 0)

p
_
r
[z[
_
2p
i
2
dz d z
+C
4
| v
t
|
p
W
1,p
| w |
p
W
1,p
r
60
where
_
r|z|

r
_
t
z
_

p
[w(0,
t
z
) w(0, 0)[
p
_
r
[z[
_
2p
i
2
dz d z
=
_
r

r|y|r
[
r
(y)[
p
[w(0, y) w(0, 0)[
p
_
[y[
r
_
2p4
i
2
dy d y
=
_
r

r|y|r
[
r
(y)[
p
[w(0, y) w(0, 0)[
p
i
2
dy d y
C
5
| w |
p
W
1,p
[ log r[
1p
We nally consider the case r

r [z[ r. Let y =
t
z
, then r [y[

r, and
g
t
(w)
_
z,
t
z
_
= g
t
(w)
_
t
y
, y
_
= w(0, y) + (1
r
_
t
y
_
)(w(
t
y
, 0) w(0, 0)),
which is the same as the case r [z[

r. So we conclude that
| D
ut
g
t
(w) s |
L
p C | w |
W
1,p ([ log r[
1
p
1
+ | u |
W
1,p r
1
p
)
since v
t
converges to u in W
1,p
norm. 2
We next look at the local model of a boundary node. Let u : (B
+
1,0
, I
1,0
)
(X, L) be a stable map, and construct smooth maps u
t
: (B
+
1,t
, I
1,t
) (X, L),
v
t
: (B
+
1,0
, I
1,0
) (X, L) as before. We dene linear maps
e
t
: L
p
(B
+
1,t
,
0,1
B
1,t
u

t
TX) L
p
(B
+
1,0
,
0,1
B
1,0
v

t
TX)
g
t
: W
1,p
(B
+
1,0
, I
1,0
, v

t
TX, (v
t
[
I
1
,0
)

TL) W
1,p
(B
+
1,t
, I
1,t
, u

t
TX, (u
t
[
I
1
,t
)

TL)
in exactly the same way as for B
1,t
, B
1,0
. Then we have
Lemma 6.28. If s L
p
(B
+
1,t
,
0,1
B
1,t
u

t
TX), w W
1,p
(B
+
1,0
, I
1,0
, v

t
TX, (v
t
[
I
1
,0
)

TL)
satisfy D
vt
w = e
t
(s), then
| D
ut
g
t
(w) s |
L
p C | w |
W
1,p ([ log r[
1
p
1
+ | u |
W
1,p r
1
p
).
We now apply above construction to each node to obtain linear maps
e
,
: L
p
(
(0,,

)
,
0,1

(0,,

)
u

,
TX) L
p
(,
0,1
v

,
TX)
g
,
: W
1,p
(, , v

,
TX, (v
,
[

TL)
W
1,p
(
(0,,

)
,
(0,,

)
, u

,
TX, (u
,
[

(0,,

)
)

TL)
Let
Q

(0,,

)
= g
,
Q
,

,

_
e
,
[
F
,

_
: F
,

W
1,p
(
(0,,

)
,
(0,,

)
, u

,
TX, (u
,
[

(0,,

)
)

TL).
The operator norm of Q

(0,,

)
has a uniform bound independent of ,

since that
of Q
,
has a uniform bound independent of ,

. We now show that Q

(0,,

)
is
an approximate right inverse of
(0,,

)
D
(0,,

),u
,

.
61
Proposition 6.29.
|
_

(0,,

)
D
(0,,

),u
,

(0,,

)
_
s s |
L
p C([ log([[ +[

[)[
1
p
1
) | s |
L
p
where C depends on | u |
W
1,p.
Proof. Let (,

) = (
(0,,

)
, u
,
). Given s F
(,

)
, let
s
1
=
,
e
,
(s) F
,
,
t
1
= e
,
(s) s
1
E
,
,
w = Q
,
(s
1
) W
1,p
(, , v

,
TX, (v
,
[

TL),
then

,
D
v
,

w = s
1
= e
,
(s) t
1
.
so D
v
,

w = e
,
(s) + t
2
for some t
2
E
,
. There is a unique t E
(,

)
such
that e
,
(t) = t
2
. We have
D
v
,

w = e
,
(s +t),
so by Lemma 6.27, 6.28,
| D
(0,,

),u
,

g
,
(w) (s +t) |
L
p
_

1
,

(N
1
())
_
C
1
| w |
W
1,p
_
[ log([[ + [

[)[
1
p
1
+ ([[ +[

[)
1
2p
_
where C
1
depends on | u |
W
1,p .
We have

(0,,

)
D
(0,,

),u
,

,
s s =
(0,,

)
_
D
(0,,

),u
,

g
,
(w) (s +t)
_
| w |
W
1,p = | Q
,

,
e
,
(s) |
W
1,p
C
2
|
,
e
,
(s) |
L
p
C
3
| s |
L
p
So
|
(0,,

)
D
(0,,

),u
,

(0,,

)
s s |
L
p
_

1
,

(N
1
())
_
C
4
_
[ log([[ +[

[)[
1
p
1
+ ([[ +[

[)
1
2p
_
| s |
L
p
C
5
[ log([[ +[

[)[
1
p
1
| s |
L
p . 2
Let
p
(,,

)
: L
p
(
(,,

)
,
0,1

(,,

)
u

,
TX) L
p
(
(,,

)
,
0,1

(0,,

)
u

,
TX)
be the map determined by the bundle isomorphismPi :
0,1

(,,

)

0,1

(0,,

)
,
where i :
0,1

(,,

)

1

(,,

)

=
1

(0,,

)
is the inclusion, and P :
1

(0,,

0,1

(0,,

)
is the projection. Let
Q

(,,

)
= g
,
Q
,

,
e
,
(p
(,,

)
[
F
(,,

)
) : F
(,,

)
W
1,p
(
(,,

)
,
(,,

)
, u

,
TX, (u
,
[

(,,

)
)

TL)
where F
(,,

)
= F
(
(,,

)
,u
,
)
. We have
62
Proposition 6.30.
|
_

(,,

)
D
(,,

),u
,

(,,

)
_
ss |
L
p C([[ +[ log([[ +[

[)[
1
p
1
) | s |
L
p
Proof. We identify
W
1,p
(
(,,

)
,
(,,

)
, u

,
TX, (u
,
[

(,,

)
)

TL)
with
W
1,p
(
(0,,

)
,
(0,,

)
, u

,
TX, (u
,
[

(0,,

)
)

TL),
and embed both
L
p
(
(,,

)
,
0,1

(,,

)
u

,
TX)
and
L
p
(
(0,,

)
,
0,1

(0,,

)
u

,
TX)
into
L
p
(
(0,,

)
,
1

(0,,

)
u

,
TX).
We extend the domain of Q

(0,,

)
to L
p
(
(0,,

)
,
0,1

(0,,

)
u

,
TX), and that
of Q

(,,

)
to L
p
(
(0,,

)
,
0,1

(0,,

)
u

,
TX). In other words, we have
Q

(0,,

)
= g
,
Q
,

,
e
,
, Q

(,,

)
= Q

(0,,

)
p
(,,

)
.
With the above understanding, we have
_

(,,

)
D
(,,

),u
,

(,,

)
_
s s
= (
(,,

)
D
(,,

),u
,

(0,,

)
D
(0,,

),u
,

) Q

(,,

)
s
+
_

(0,,

)
D
(0,,

),u
,

(0,,

)
_
p
(,,)
s p
(,,)
s
+p
(,,)
s s,
where
|
(,,

)
D
(,,

),u
,

(0,,

)
D
(0,,

),u
,

| C
1
[[,
| p
(,,)
Id | C
2
[[, and | Q

(,,

)
| C
3
[[ for all (, ,

) B
2
D
d
D

d
.
We also have
|
_

(0,,

)
D
(0,,

),u
,

(0,,

)
_
p
(,,)
s p
(,,)
s |
C
4
_
[ log([[ +[

[)[
1
p
1
_
| p
(,,)
s |
L
p
C
4
_
[ log([[ +[

[)[
1
p
1
_
| s |
L
p
where C
4
is the constant C in Proposition 6.30. Therefore,
|
_

(,,

)
D
(,,

),u
,

(,,

)
_
s s |
L
p
(C
1
C
3
+C
4
+C
2
)([[ +[ log([[ +[

[)[
1
p
1
) | s |
L
p . 2
Corollary 6.31. There exist
2
, d
1
, . . . , d
l0
, d

1
, . . . , d

l1
> 0 such that for every
(, ,

) B
2
D
d
D

d
there is a right inverse Q
(,,

)
of
(,,

)
D
(,,

),u
,

such that the operator norm | Q

(,,

)
| C for some constant C.
63
Proof. By Proposition 6.29, there exist
2
, d
1
, . . . , d
l0
, d

1
, . . . , d

l1
> 0 suciently
small such that
|
_

(,,

)
D
(,,

),u
,

(,,

)
_
s s |
L
p
1
2
| s |
L
p .
for any (, ,

) B
2
D
d
D

Let A
,,
=
(,,

)
D
(,,

),u
,

(,,

)
I, where I is the identity map.
Then | A
,,
|
1
2
, so I + A
,,
is invertible and | (I + A
,,
)
1
| 2. Let
Q
(,,

)
= Q

(,,

)
(I +A
,,
)
1
, then Q
(,,

)
has the desired properties. 2
Let
2
, d
1
, . . . , d
l0
, d

1
, . . . , d

l1
be chosen as in Corollary 6.31. Then
(,,

)

D
(,,

),u
,

is surjective for (, ,

) B
2
D
d
D

d
. We will construct a linear
isomorphism
i
(,,

)
: Ker( D
u
) Ker(
(,,

)
D
(,,

),u
,

).
Given w W
1,p
(, , u

TX, (u[

TL), we will cut it o near nodes to obtain

g

(0,,

)
(w) W
1,p
(
(0,,

)
,
(0,,

)
, u

,
TX, (u
,
[

TL).
We rst look at the local model of an interior node. Let u : B
1,0
X be a
stable map. We have constructed smooth maps u
t
: B
1,t
X for small t C such
that
u
t
_
z,
t
z
_
=
_
_
_
u
_
0,
t
z
_
) if
r
2
1
< [z[ <
r

r
2
,
p if r

r [z[

r,
u(z, 0) if 2

r < [z[ <

1
.
where p = u(0, 0).
Let r =
_
[t[ as before, and set s = (4r)
1
3
. For w W
1,p
(B
1,0
, u

TX, (u[
B
1
,0
)

TL),
dene w
t
W
1,p
(B
1,t
, u

TX, (u
t
[
B
1
,t
)

TL) by
w
t
_
z,
t
z
_
=
_
(1
s
(
t
z
))w(0,
t
z
) +
s
(
t
z
)P(z)w(0, 0) if
r
2
1
< [z[ r
(1
s
(z))w(z, 0) +
s
(z)P(z)w(0, 0) if r [z[ <
1
where
s
is the cuto function in Lemma 6.25, and P(z) is the parallel transport
along the unique length minimizing geodesic from p to u
t
(z,
t
z
). We have
w
t
_
z,
t
z
_
=
_

_
w(0,
t
z
) if
r
2
1
< [z[ 2

2
3
r
5
3
P(z)w(0, 0) if
r

r
2
[z[ 2

r
w(z, 0) if (4r)
1
3
[z[ <
1
We apply above construction to each interior node and similar construction to
each boundary node to obtain a linear map
g

(0,,

)
: W
1,p
(, , u

TX, (u[

TL)
W
1,p
(
(0,,

)
,
(0,,

)
, u

,
TX, (u
,
[

(0,,

)
)

TL),
which can also be viewed as a map
g

(,,

)
: W
1,p
(, , u

TX, (u[

TL)
W
1,p
(
(,,

)
,
(,,

)
, u

,
TX, (u
,
[

(,,

)
)

TL).
The restriction of g

(,,

)
to Ker( D
u
) is injective by the unique continuity
theorem ([Ar]). Lemma 6.26 implies the following estimate for w Ker( D
u
):
|
(,,

)
D
u
,

(,,

)
w |
W
1,p C([[ + log([[ +[

[)
1
p
1
) | w |
W
1,p .
64
Let
i
(,,

)
= (Id Q
(,,

)

(,,

)
D
(,,

),u
,

) g

(,,

)
Then i
(,,

)
: Ker( D
u
) Ker(
(,,

)
D
(,,

),u
,

) is injective for (, ,

)
suciently small. It is actually a linear isomorphism since
dimKer(
(,,

)
D
(,,

),u
,

) = IndD
u
+ dimE

= dimKer( D
u
).
We are now ready to nd exact solutions near the approximate solution u
,
.
Proposition 6.32. There exist
2
, d
1
, . . . , d
l0
, d

1
, . . . , d

l1
,
1
,
2
> 0 suciently
small such that for all (, ,

) B
2
D
d
D
d
, if
w Ker(
(,,

)
D
(,,

),u
,

), | w |
W
1,p
1
,
then there exists a unique
h(, ,

, w) L
p
(
(,,

)
,
0,1

(,,

)
u

,
TX)
such that

(,,

J,
(,,

)
exp
u
,

(w +Q
(,,

)
h(, ,

, w)) = 0
and
| h(, ,

, w) |
L
p
2
.
Proof. We assume that (, ,

) B
2
D
d
D

d
, and | w |
W
1,p
1
, where
2
,
d = (d
1
, . . . , d
l0
), d

= (d

1
, . . . , d

l1
), and
1
will be determined later.
We will use Newtons method to nd h(, ,

, w) as in [MS, Theorem 3.3.4].

For convenience, write v for u
,
, Q for Q
(,,

)
,

for

J,
(,,

)
. Set h
0
= 0, and
h
n+1
= h
n
P
n


exp
v
(w +Qh
n
)
where P
n
is the parallel transport along the geodesic t [0, 1] exp
v
((1 t)(w +
Qh
n
)). Let D
n
denote the linearization of

J,
(,,

)
at v
n
= exp
v
(w + Qh
n
), and
write
n
for
(
(,,

)
,vn)
. We have
P
n+1


exp
v
(w +Qh
n+1
)
= P
n+1


exp
v
(w +Qh
n
Q P
n


exp
v
(w +Qh
n
))
= P
n


exp
v
(w +Qh
n
)
P
n

n
D
n
(d exp
v
)(w +Qh
n
)(Q P
n


exp
v
(w +Qh
n
))
+R(

exp
v
(w +Qh
n
)),
where
| R(

exp
v
((w +Qh
n
) |
L
p C
1
|

exp
v
((w +Qh
n
)) |
2
L
p,
| P
n

n
D
n
(d exp
v
)(w +Qh
n
) D
(,,

),v
| C
2
([w[ +[Qh
n
[).
Therefore,
|

exp
v
(w +Qh
n+1
) |
L
p
C
3
_
| w |
C
0 + | Qh
n
|
C
0 + |

exp
v
(w +Qh
n
) |
L
p
_
|

exp
v
(w +Qh
n
) |
L
p .
We also have
h
n
=
n1

k=0
P
k


exp
v
(w +Qh
k
),
65
so
|

exp
v
(w +Qh
n+1
) |
L
p
C
3
_
| w |
W
1,p +
n

k=0
|

exp
v
(w +Qh
k
) |
L
p
_
|

exp
v
(w +Qh
n
) |
L
p .
Let a
n
=|

exp
v
(w +Qh
n
) |
L
p 0 and b =| w |
W
1,p 0, so that
a
n+1
C
3
_
b +
n

k=0
a
k
_
a
n
We will show that a
0
, b
1
6C3
a
n+1

1
2
a
n
. We prove by induction. For n = 0,
we have
a
1
C
3
(b +a
0
)a
0
C
3
_
1
6C
3
+
1
6C
3
_
a
0
=
1
3
a
0

1
2
a
0
.
Now suppose that a
n+1
a
n
for n = 0, 1, . . . , m. Then
m+1

k=0
a
k
2a
0
,
so
a
m+2
C
3
(b + 2a
0
)a
m+1
C
3
_
1
6C
3
+
1
3C
3
_
a
m+1
=
1
2
a
m+1
.
Let D = D
(,,

),u
,

. Then


exp
v
w =

v + Dw +R(w),
where | R(w) |
L
p C
4
| w |
C
0| w |
W
1,p, and Dw = 0. The proof of Lemma 6.23
can be modied to show that
|

v |
L
p C
5
_
[[ + ([[ +[

[)
1
2p
_
.
So
a
0
= |

exp
v
w |
L
p
C
5
_
[[ + ([[ +[

[)
1
2p
_
+C
6
| w |
2
W
1,p
C
5
_

2
+ ([d[ +[d

[)
1
2p
_
+C
6

2
1
,
which is less than
1
6C3
for suciently small
2
, d, d

,
1
. b =| w |
W
1,p
1
, which is
also less than
1
6C3
for small
1
. So if
2
, d, d

,
1
are suciently small, h
n
converges
uniformly in L
p
norm, and the limit h(, ,

, w) satises


exp
v
(w +Qh(, ,

, w)) = 0,
| h(, ,

, w) |
L
p 2a
0
2C
5
(
2
+ ([d[ +[d

[)
1
2p
) + 2C
6

2
1
.
This proves the existence of h(, ,

, w).
We now turn to the uniqueness of h(, ,

, w). Write s
1
for h(, ,

, w), and
suppose that s
2
satises


exp
v
(w +Qs
2
) = 0
and
| s
2
|
L
p 2C
5
(
2
+ ([d[ +[d

[)
1
2p
) + 2C
6

2
1
.
66
Let P
1
, P
2
denote the parallel transports along the geodesics t [0, 1] exp
v
((1
t)(w +Qs
1
)), exp
v
((1 t)(w +Qs
2
)), respectively. Let D

denote the linearization

of

J,
(,,

)
at v

= exp
v
(w +Qs
1
), and let

denote
(
(,,

)
,v

)
. We have
0 = P
2


exp
v
(w +Qs
2
)
= P
2


exp
v
(w +Qs
1
+Q(s
2
s
1
))
= P
1


exp
v
(w +Qs
1
) +P
1

(d exp
v
)(w +Qs
1
) Q(s
2
s
1
)
+R(s
2
s
1
)
= P
1

(d exp
v
)(w +Qs
1
) Q(s
2
s
1
) +R(s
2
s
1
),
s
1
s
2
= (P
1

(d exp
v
)(w +Qs
1
) D) Q(s
2
s
1
) +R(s
2
s
1
),
where
| P
1

(d exp
v
) D | C
2
([w[ +[Qs
1
[),
and
| R(s
2
s
1
) |
L
p| s
2
s
1
|
2
L
p .
So
| s
1
s
2
|
L
p C
7
(| w |
W
1,p + | s
1
|
L
p + | s
2
|
L
p) | s
1
s
2
|
L
p
C
7
_

1
+ 4C
5
_

2
+ ([d[ +[d

[)
1
2p
_
+ 4C
6

2
1
_
| s
1
s
2
|
L
p,
where
C
7
_

1
+ 4C
5
_

2
+ ([d[ +[d

[)
1
2p
_
+ 4C
6

2
1
_

1
2
for suciently small
2
, d, d

,
1
. We conclude that s
1
= s
2
.
The proposition is true if
2
, d
1
, . . . , d
l0
, d

1
, . . . , d

l1
,
1
,
2
> 0 are chosen such
that
C
5
_

2
+ ([d[ +[d

[)
1
2p
_
+C
6

2
1

1
6C
3
,
1

1
6C
3
,
C
7
_

1
+ 4C
5
_

2
+ ([d[ +[d

[)
1
2p
_
+ 4C
6

2
1
_

1
2
,
and
2
= 2C
5
(
2
+ ([d[ +[d

[)
1
2p
) + 2C
6

2
1
. 2
6.4.3. Kuranishi neighborhood. Let
2
, d
1
, . . . , d
l0
, d

1
, . . . , d

l1
> 0 be chosen as in
Proposition 6.32. Let V
,map
H
,map
= Ker( D
u
) be a neighborhood of the
origin such that
| i
(,,

)
w |
W
1,p <
1
for all w V
,map
, (, ,

) B
2
D
d
D

d
.
Write

B for B
2
D
d
D

d
, as at the beginning of Section 6.4.2. Dene a map
:

B V
,map
W
B
(, ,

, w) [(
(,,

)
, u
(,,

,w)
)]
where W
B
is dened as at the beginning of Section 6.4.2, and
u
(,,

,w)
= exp
u
,

_
i
(,,

)
w +Q
(,,

)
h(, ,

, i
(,,

)
w)
_
.
Then

(
(,,

)
,u
(,,

,w)
)

J,
(,,

)
u
(,,

,w)
= 0,
so (, ,

, w)

J,
(,,

)
u
(,,

,w)
denes a map s :

B V
,map
E

such
that (s
1
(0))

M
(g,h),(n, m)
(X, L [ , , ). Actually, (s
1
(0)) contains a
neighborhood of in

M
(g,h),(n, m)
(X, L [ , , ). To see this, note that any
67

M
(g,h),(n, m)
(X, L [ , , ) which is suciently close to in C

topology
can be written in the form

= (
(,,

)
, exp
u
,

(w))
where w W
1,p
(
(,,

)
,
(,,

)
, u

,
TX, (u
,
[

(,,

)
)

TL) is small. There

exist unique w
0
Ker(
(,,

)
D
(,,

),u
,

) and h L
p
(
(,,

)
,
0,1

(,,

)

u

,
TX) such that w = w
0
+Q
(,,

)
h. We may further assume that w is so small
such that w
0
= i
(,,

)
w
1
for some w
1
V
,map
, and | h |
L
p
2
. Since

J,
(,,

)
exp
u
,

(w
0
+Q
(,,

)
h) = 0,
we have h = h(, ,

, w
0
) by the uniqueness part of Proposition 6.32. So

=
(, ,

, w
1
), and s(, ,

, w
1
) = 0.
Now assume that is stable, so that Aut is nite, and V

,map
= V
,map
. Then
there is an isomorphism :
(1,1,

1
)

(2,2,

2
)
if and only if (
2
,
2
,

2
) =
(

1
,

1
,

1
) for some

Aut , and the action of

on the universal
family restricts to on
(1,1,

1
)
. We may choose

B, V
,map
small enough such that
if u
(,,

,w1)
= u
(,,

,w2)
for (, ,

)

B, w
1
, w
2
V
,map
, Aut ,
then Aut .
Given Aut and w V
,map
, let
(,,

)
w be the unique vector in V
,map
such that u
(,,

,w)

1
= u
(,,

,
(,,

)
w)
. Then Aut acts on

B V
,map
by (, ,

, w) = ( , ,

,
(,,

)
w). From the above discussion,
(
1
,
1
,

1
, w
1
) = (
2
,
2
,

2
, w
2
) if and only if (
2
,
2
,

2
, w
2
) = (
1
,
1
,

1
, w
1
)
for some Aut . Let V

be an Aut -invariant neighborhood of the origin

in

B V
,map
H
,domain
H
,map
= H

. Let s

: V

be the restric-
tion of s. Then s

is Aut -equivariant. The restriction of gives a continu-

ous map

: s
1

(0)

M
(g,h),(n, m)
(X, L [ , , ) such that s
1

(0)/Aut

M
(g,h),(n, m)
(X, L [ , , ) is injective. It is actually a homeomorphism onto a
neighborhood of in

M
(g,h),(n, m)
(X, L [ , , ) since C

topology and W
1,p
topology are equivalent on

M
(g,h),(n, m)
(X, L [ , , ). This completes the proof
of Theorem 6.19 when the domain of is stable.
Remark 6.33. The section s in the above construction is only continuous, not
smooth. It is not hard to show that it is smooth within the stratum, but its de-
pendence on ,

is not even C
1
(see Lemma 6.23). This is because the smooth
structure of V

is canonical within the stratum but dependent on our particular

gluing construction in the direction (,

) transversal to the stratum.

If = (, B; p; q
1
, . . . , q
h
) is not stable, we add minimal number of marked
points to obtain a stable marked bordered Riemann surface

= (, B, p; q
1
, . . . , q
h
)
where p = (p
1
, . . . , p
n+ n
), q
i
= (q
i
1
, . . . , q
i
m
i
+ m
i
), p
n+1
, . . . , p
n+ n
are additional
interior marked points, and q
i
m
i
+1
, . . . , q
i
m
i
+ m
i
are additional marked points on B
i
.
Note that when counting the above minimal number, an interior marked point
counts twice, while a boundary marked point counts once. We have
2 n + m
1
+ m
h
= dim
R
H
,aut
.
Let = (

, u) M
(g,h),(n+ n, m+

m)
(X, L [ , , ). Then H
,domain
= H
,domain
and H
,aut
= 0. If the additional marked points are chosen in K
3

|d|+|d

|
(), then
68
the construction of
(,,

)
also yields deformation

(,,

)
of

for (, ,

)
B
2
D
d
D

d
=

B H
,domain
= H
,domain
. Both Aut

and Aut are subgroups
of Aut , and Aut = Aut

Aut . We may choose E

= E

, so that H
,map
=
H
,map
, H

= H

.
There is a map
F : M
(g,h),(n+ n, m+

m)
(X, L [ , , )

M
(g,h),(n, m)
(X, L [ , , )
dened by forgetting p
n+1
, . . . , p
n+ n
, q
m+1
, . . . , q
m+ m
and then contracting non-
stable components which are mapped to points. We have F( ) = . We will
construct a multi-valued map A from a neighborhood U

of in

M
(g,h),(n, m)
(X, L [
, , ) to M
(g,h),(n+ n, m+

m)
(X, L [ , , ) such that F A is the identity map, and
A() = .
The additional marked points are on non-stable components, where u is not a
constant. Let 2N be the dimension of X, as before. For j = n+1, . . . , n+ n, we may
assume that there is a geodesic ball in X centered at u(p
j
) such that its intersection
with the image of u is an embedded holomorphic disc D
j
, and u
1
(D
j
) D
j
is a
trivial cover. u
1
(D
j
) might consist of more than one connected components if u
is not injective. Let B
j
X be an embedded (2N 2)-dimensional ball which is
the image of N
pj
under exp
u(pj)
, where N
pj
is a small ball centered at the origin
in the orthogonal complement of u

(T
pj
) in T
u(pj)
X. Then D
j
and B
j
intersect
orthogonally at u(p
j
). We choose N
pj
suciently small such that B
j
intersects the
image of u at a single point u(p
j
).
Similarly, for j
i
= m
i
+1, . . . , m
i
+ m
i
, we may assume that there is a geodesic ball
in X centered at u(q
j
i ) such that its intersection with the image of u is an embedded
holomorphic half disc D
+
j
i
, and u
1
(D
+
j
i
) D
+
j
i
is a trivial cover. Let B

j
i
L be
an embedded (N 1)-dimensional ball which is the image of N

q
j
i
under exp
u(q
j
i )
,
where N

q
j
i
is a small ball centered at the origin in the orthogonal complement of
u

(T
q
j
i
) in T
u(q
j
i )
L. Then I
j
i = D
+
j
i
L

= [0, 1] and B

j
i
intersect orthogonally
in L at u(q
j
i ). We choose N

q
j
i
suciently small such that B

j
i
intersects the image
of u at a single point u(q
j
i ).
In a small neighborhood of in

M
(g,h),(n, m)
(X, L [ , , ), intersecting with B
j
for j = n+1, . . . , n+ n and B

j
i
for j
i
= m
i
+1, . . . , m
i
+ m
i
determines additional
marked points and gives the desired multi-valued map A : U

M
(g,h),(n+ n, m+

m)
(X, L [
, , ) for some neighborhood of in

M
(g,h),(n, m)
(X, L [ , , ). A is single-valued
if Aut = Aut . Let

=
_

= (

, B

; p

; (q

)
1
, . . . , (q

)
h
; u

) F
1
(U

) [
u

(p
j
) B
j
for j = n + 1, . . . , n + n, u

(q
j
i ) B

j
i for j
i
= m
i
+ 1, . . . , m
i
+ m
i
_
,
and let

U

be the connected component containing . Then

U

= A(U

), and the
ber of

U

is nite.
Let

W be the (real) codimension (2 n+ m
1
+ + m
h
) subspace of W
1,p
(, , u

TX, (u[

TL)
dened by

W =
_
w W
1,p
(, , u

TX, (u[

TL)

w(p
j
) T
u(pj)
B
j
for j = n + 1, . . . , n + n
w(q
j
i ) T
u(q
j
i )
B

j
i
for j
i
= m
i
+ 1, . . . , m
i
+ m
i
_
69
Then H
,aut

W = 0, so
W
1,p
(, , u

TX, (u[

TL) = H
,aut

W.
Let

W
(,,

)
=
_
w W
1,p
(
(,,

)
,
(,,

)
, u

,
TX, (u
,
[

(,,

)
)

TL)

w(p
j
) T
u(pj)
B
j
for j = n + 1, . . . , n + n
w(q
j
i ) T
u(q
j
i )
B

j
for j
i
= m
i
+ 1, . . . , m
i
+ m
i
_
.
Then
W
1,p
(
(,,

)
,
(,,

)
, u

,
TX, (u
,
[

(,,

)
)

TL) = i
(,,

)
H
,aut

W
(,,

)
for (, ,

) suciently small. Let

p
(,,

)
: W
1,p
(
(,,

)
,
(,,

)
, u

,
TX, (u
,
[

(,,

)
)

TL)

W
(,,

)
be the projection. We have
(,,

)
D
(,,

),u
,

p
(,,

)
=
(,,

)
D
(,,

),u
,

since i
(,,

)
H
,aut
Ker(
(,,

)
D
(,,

),u
,

). By replacing Q
(,,

)
with
p
(,,

)
Q
(,,

)
, we may assume that ImQ
(,,

)


W
(,,

)
.
Let

H
,map
= H
,map

W

= H

,map
. We may modify the isomorphism i
(,,

)
:
H
,map
Ker(
(,,

)
D
(,,

),u
,

) such that
i
(,,

)

H
,map
= Ker(
(,,

)
D
(,,

),u
,

)

W
(,,

)
.
For (, ,

, w)

BV
,map
=

BV
,map
, dene

(, ,

, w) = (

(,,

)
, u
(,,

,w)
).
For w

W, we have i
(,,

)
w +Q
(,,

)
h(, ,

, i
(,,

)
w)

W
(,,

)
, so
u
(,,

,w)
= exp
u
,

(i
(,,

)
w +Q
(,,

)
h(, ,

, i
(,,

)
)

W
(,,

)
satises u
(,,

,w)
(p
j
) B
j
for j = n + 1, . . . , n + n, and u
(,,

,w)
(q
j
) B

j
for
j = n+1, . . . , n+ n. So

(, ,

, w)

U

if w

W, s(, ,

, w) = 0, and , ,

, w
are suciently small.
Conversely, if (

, u

) is a stable map near in

U

, then

=
(,,

)
for some
(, ,

)

B, and u

= exp
u
,

(w) for some w

W
(,,

)
. There exist unique w
0

Ker(
,,
D
(,,

),u
,

)

W
(,,

)
and h L
p
(
(,,

)
,
0,1

(,0,0)
u

,
TX)
such that w = w
0
+Q
(,,

)
h. We have h = h(, ,

, w
0
) since

J,
(,,

)
exp
u
,

(w
0
+Q
(,,

)
h) = 0.
Let w
1
= i
1
(,,

)
w
0


H
,map
. Then u

= u(, ,

, w
1
), and s(, ,

, w
1
) = 0. Let

V
,map
= V
,map


H
,map
. Then Aut acts on

B

V
,map
. Let

V

be an Aut
invariant neighborhood of the origin in

B

V
,map
. It corresponds to a neighborhood
V

of the origin in

B H

,map
under the isomorphism

H
,map

= H

,map
. We have
the following commutative diagram
(s
1
(0)

V

)/Aut

_F
(s
1
(0) V

)/Aut

where

and

are injective.

is a homeomorphism onto its image.

70
6.5. Transition functions. For each / =

M
(g,h),(n, m)
(X, L [ , , ) and
each choice of E

, we have constructed a Kuranishi neighborhood (V

, E

, Aut ,

, s

).
Remark 6.34. A dierent choice of E

yields a dierent, but equivalent Kuranishi

neighborhood in the sense of Denition 6.2. Actually, if E
1,
, E
2,
L
p
(,
0,1

TX) are two dierent choices, set E

= E
1,
+E
2,
. Then (V

, E

, Aut ,

, s

)
can serve as the (V
p
, E
p
,
p
,
p
, s
p
) in Denition 6.2.
In this section, we will modify the Kuranishi neighborhoods we constructed so
that we can construct transition functions between them. We will follow [FO,
Section 15] closely. We refer the reader to [FO, Section 15] for more details.
/ is compact in C

topology, so there exist

1
, . . . ,
l
/ such that
U

i
=
i
(s
1
i
(0)) [ i = 1, . . . , l
is an open cover of /, and there is U
i
U

i
such that U
i
[ i = 1, . . . , l is still an
open cover of /. / is compact and Hausdor, so the closure K
i
of U
i
is compact
for i = 1, . . . , l.
Let E
i
V
i
= V
i
be the obstruction bundle constructed from E
i
. We may
choose
i
and E
i
such that if
= (, B; p; q
1
, . . . , q
h
; u) K
i1
. . . K
i
k
and / K
i
if i ,= i
k
, then the subspace

E

of L
p
(,
0,1
u

TX) spanned by
(E
i1
)

, . . . , (E
i
k
)

is actually a direct sum

E

= (E
i1
)

(E
i
k
)

. We use

E

to construct a Kuranishi neighborhood (

,

E

, Aut ,

, s

) as in Section 6.4.2,
6.4.3. We shrink

V

such that

( s
1

(0)) U
i1
U
i
k
K
c
j1
K
c
j
lk
, where
i
1
, . . . , i
k
, j
1
, . . . , j
lk
= 1, . . . , l.
Suppose that

(, ,

, w), where (, ,

, w)

V

, s

(, ,

, w) = 0.
Then Aut

can be identied with the stabilizer (Aut )

(,.

,w)
of the action Aut
on

V

. This gives a monomorphism h

: Aut

Aut . Without loss of gener-

ality, we may assume that

K
1
K
k
, and

/ K
i
for i = k + 1, . . . , l. We
have K
1
K
k
. We may assume that K
1
K
k
, and / K
i
for i = k

+ 1, . . . , l, where k

k. It follows from the construction that there is

an Aut

-invariant neighborhood V

of 0 =
1

) in

V

such that we have the

following commutative diagram:

[
V

_
V

where

is induced by the inclusion E

1
E
k
E
1
E
k
. Both

and

are h

-equivariant embedding of codimension rankE

k+1
+ +rankE
k
.
The Kuranishi neighborhoods (

,

E

, Aut ,

, s

) satisfy the properties listed

in Denition 6.1, and the transition functions (V

, h

) satisfy the prop-

erties listed in Denition 6.3. From now on, we will write (V

, E

, Aut ,

, s

) for
the (

V

,

E

, Aut ,

, s

) we just constructed.
Remark 6.35. The

in the above construction are smooth when restricted

to a stratum. It is possible to rene the above modication of Kuranishi neighbor-
hoods such that

are smooth (see [FO, Section 15]). Such an renement is

71
articial since our smooth structure in directions transversal to a stratum is not nat-
ural, as discussed in Remark 6.33. For simplicity of exposition, we will assume the
smoothness of

in Section 7, though such an assumption is not absolutely

necessary for our purpose.
By Remark 6.34, the equivalence class of the Kuranishi structure constructed
above is independent of various choices in our construction.
6.6. Orientation. Recall that the orientation bundle of the Kuranishi structure is
the real line orbibundle obtained by gluing det(TV

) det(E

)
1
. In our case,
(det(TV

)det(E

)
1
)
(,,

,w)

= det(IndD
(,,

),u
,

)det
_
Ind(T

(,,

)
, T

(,,

)
)
_
1
where Ind(D) denotes the virtual real vector space Ker(D)Coker(D), as in Section
6.2.4.
Theorem 6.36. The Kuranishi structure constructed above is orientable if L is
spin or if h = 1 and L is relatively spin (i.e., L is orientable and w
2
(TL) = [
L
for some H
2
(X, Z
2
)).
To orient the index of the linearization D

J
of

J
at a stable map, we need the
following generalization of [FO
3
, Proposition 21.3].
Lemma 6.37. Let (E, E
R
) (, ) be a Riemann-Hilbert bundle over a prestable
bordered Riemann surface without boundary nodes , and let E
R
be a totally real
subbundle of E[

. Then an ordering of the connected components of and a

trivialization of E
R
determine an orientation of Ind(E, E
R
), where Ind(E, E
R
) is
dened as in [KL, Denition 3.4.1].
Proof. Let B
1
, ..., B
h
be the ordered connected components of . An isomorphism
E
R

= R
n
is a collection of isomorphisms E
R
[
b
i

= B
i
R
n
. Let > 0 be such
that A
i
= B(B
i
, ), the collar neighborhood of B
i
in of radius w.r.t. some
admissible metric on , are disjoint. By tensoring with C, we have trivializations
E[
B
i

= B
i
C
n
. By deforming the Hermitian connection, we may assume that the
connection is at on A =

h
i=1
B(B
i
,
2
3
) and there are parallel sections s
1
, ..., s
h
on E[
A
such that for x , s
i
(x) corresponds to (x, e
i
) under the isomorphism
E
R

= R
n
, where e
1
, . . . e
n
are the standard basis of R
n
.
The boundary of A
i
= B(B
i
,
1
2
) is the disjoint union of two circles, B
i
and
(B

)
i
. We shrink (B

)
i
to obtain a family of prestable bordered Riemann surface

t
, t [0, 1], such that
1
= ,
t
are homeomorphic to , and
0
is obtained
from by shrinking each (B

)
i
to a point.
0
= C D
1
. . . D
h
is a prestable
bordered Riemann surfaces, where C is a complex algebraic curve of genus g, and
D
i
is a disc which intersects C at an interior node on
0
, for i = 1, . . . , h.
We extend (E, E
R
) to a family of Riemann-Hilbert bundles (E(t), E
R
(t))
(
t
,
t
) such that s
1
, . . . , s
h
extend to a neighborhood of
t[0,1]

t
in
t[0,1]

t
and give a holomorphic trivialization of E(t) in a collar neighborhood of
t
and
a trivialization of E
R
(t). In particular, they are dened on D
i

0
to give an
identication (E(0), E
R
(0))[
Di
= (C
n
, R
n
).
We use the notation in [KL, Section 3.4]. Ind(E(t), E
R
(t)) is a family of virtual
real vector spaces over [0, 1]. We have
Ind(E, E
R
) = Ind(E(1), E
R
(1))

= Ind(E(0), E
R
(0)),
72
so it suces to orient Ind(E(0), E
R
(0)). We have a long exact sequence
0 H
0
(
0
,
0
, E(0), E(0)
R
) H
0
(C, F)
h

i=1
H
0
(D
i
, D
i
, C
n
, R
n
)
e
C
n
H
1
(
0
,
0
, E(0), E(0)
R
) H
1
(C, F)
h

i=1
H
1
(D
i
, D
i
, C
n
, R
n
) 0
where F = E(0)[
C
is a holomorphic vector bundle of degree
1
2
(E, E
R
), and e is
given by
H
0
(C, F)
h

i=1
H
0
(D
i
, D
i
, C
n
, R
n
) C
n
(
0
,
1
, ...,
h
) (
0
(p
1
)
1
(0), ...,
0
(p
h
)
h
(0))
p
i
C and 0 D
i
are identied to form an interior node of
0
. E
R
R
n
gives the
orientation on H
0
(D
i
, D
i
, C
n
, R
n
) R
n
via the evaluation map. H
1
(D
i
, D
i
, C
n
, R
n
) =
0. H
0
(C, F), H
1
(C, F) and C
n
are complex vector spaces, thus canonically oriented.
Therefore,
Ind(E(0), E
R
(0)) = H
0
(
0
,
0
, E(0), E(0)
R
) H
1
(
0
,
0
, E(0), E(0)
R
)
is oriented. This orientation depends on the trivialization of E
R
and the ordering of
connected components of , since the trivialization of E
R
determines the orienta-
tion on each H
0
(D
i
, D
i
, C
n
, R
n
)

= R
n
and the ordering of connected components
of determines the ordering of these h copies of R
n
. 2
Proof of Theorem 6.36. It suces to show that the orientation bundle is trivial
when restricted to each loop : S
1

M
(g,h),(n, m)
(X, L [ , , ), (t) =
t
. From
the construction of Kuranishi structure we see that we may deform to a family
of smooth maps
t
V
t
such that the domain of
t
are smooth bordered Riemann
surfaces. We rst assume that these domains are stable. The tangent bundle of

M
(g,h),(n,m)
is orientable by Theorem 4.14, so it suces to show that the index
bundle IndD

is orientable along the loop (t) =

t
. Note that Aut preserves
the orientation of IndD

since it does not permute boundary components of the

domain.
Let : (, ) S
1
(X, L) be given by (z, t) = u
t
(z), where u
t
is the
map for
t
. Since the connected components of are ordered, by Lemma 6.37 it
suces to show that ([
S
1 )

TL is stably trivial.
We rst assume that L is relatively spin, i.e, L is orientable and w
2
(TL) =
[
L
for some H
2
(X, Z
2
). Choose a cellular decomposition on X such that
X
(2)
L = L
(2)
. There exists a real orientable vector bundle of rank 2 over X
(3)
such that w
2
(V ) = [
X
(3) H
2
(X
(3)
, Z
2
). Then
w
2
((TL V )[
L
(2) ) = 0 H
2
(L
(2)
, Z
2
),
so (TL V )[
L
(2) is spin, thus stably trivializable on L
(2)
.
We write
(TL V )[
L
(2) R
k
= R
N+k+2
.
Let

be homotopic to such that

( S
1
) L
(2)
. It suces to show that
(

[
S
1 )

TL is stably trivial. We have

(

[
S
1 )

TL (

[
S
1 )

V R
k
= R
N+k+2
73
(

[
S
1 )

h
i=1
R
i
S
1
= S
1
i
S
1
We may view V as a complex line bundle. We need to show that (

[
S
1 )

V
is trivial, or equivalently, n
i
= (deg

[
RiS
1)

V = 0 for i = 1, . . . , h. We have
h

i=1
n
i
= deg(

[
S
1 )

V
= (i

[ S
1
]) c
1
(

V )
= (i

[ S
1
]) c
1
(

V )
= 0.
So (

[
S
1)

V is trivial if h = 1. For h > 1, we assume that L is spin, so that we

may take V = R
2
, the trivial bundle.
We nally consider nonstable cases:
(1) (g, h) = (0, 1), n = 1, m = (0).
(2) (g, h) = (0, 1), n = 0, m = (1).
(3) (g, h) = (0, 1), n = 0, m = (2).
(4) (g, h) = (0, 2), n = 0, m = (0, 0).
Cases (2) and (3) are treated in [FO
3
]. For (1) the domain is isomorphic to the
unit disc with one interior marked point at the origin, and the automorphism group
U(1) = e
i
[ R of the domain can be oriented by

. For Case 4, the domain

is isomorphic to an annulus z C [ 1 [z[ r for some r (1, ), which is
oriented by

r
, and automorphism group U(1) of the domain is oriented as above.
2
7. Virtual fundamental chain
7.1. Construction of virtual fundamental chain. We consider the general set-
ting in Section 6.1.
Denition 7.1. Let M be a Hausdor space with a Kuranishi structure (with
corners)
/ =
_
(V
p
, E
p
,
p
,
p
, s
p
) : p M, (V
pq
,

pq
,
pq
, h
pq
) : q
p
(s
1
p
(0))
_
.
A Hausdor topological space W is an ambient space of / if
(1) M is a subspace of W.
(2)
p
: V
p
W,
p
(x) M if and only if s
p
(x) = 0.
(3)
q
=
p

pq
.
(4) There is a subset W W such that
p
(x) W if and only if x V
p
,
where V
p
is the union of corners of V
p
. We take W = if / is a
Kuranishi structure. We dene M = M W.
(5) V
p
/
p
W is injective.
Remark 7.2. If /
1
and /
2
are equivalent Kuranishi structures (with corners)
in the sense of Denition 6.5, and they have the same ambient space W, we will
implicitly assume that W is also an ambient space for the / in Denition 6.5, and

i,p
=
p

i
: V
i,p
W in Denition 6.2.
74
Example 7.3. J = W
1,p
(g,h),(n, m)
(X, L [ , , ) is an ambient space of the Ku-
ranishi structure with corners on / =

M
(g,h),(n, m)
(X, L [ , , ) constructed in
Section 6. J J is the subset corresponding to maps whose domain has at least
one boundary node.
Now assume that M is a compact, Hausdor topological space with an oriented
Kuranishi structure with corners /, and that / has an ambient space W. Let d be
the virtual dimension of /. The virtual fundamental chain we will construct is a
singular d-chain M
K,
o
d
(W, Q) such that M
K,
o
d1
(W, Q) o
d1
(W, Q),
so it represents a relative singular d-cycle

M
K,
o
d
(W, W, Q).
M is compact, so there exist nitely many p
1
, . . . , p
l
M such that
U

j
=
pj
(s
1
pj
(0)) [ j = 1, . . . , l is an open cover of M.
There exists a
pj
-invariant neighborhood V
j
of
1
pj
(p
j
) in V
pj
such that
V
j
V
pj
and U
j
=
pj
(s
1
pj
(0) V
j
) [ j = 1, . . . , l is still an open cover
of M.
Let
j
: V
pj
[0, 1] be a smooth function with compact support such that

j
1 on the closure of V
j
. For any = (
1
, . . . ,
l
), where
i
: V
pj
E
pj
is
a small continuous section, not necessarily
pj
-equivariant, we construct a section

p
:

V
p
E
p
[

Vp
for each p X, where

V
p
is an
p
-invariant neighborhood of

1
p
(p).
Given p M, if p U
j
, then there is a
p
-invariant neighborhood V
pj p
of
1
p
(p)
such that we have the following commutative diagram:
E
p
[
Vp
j
p

p
j
p
E
pj

_
V
pjp
p
j
p
V
pj
For any section s : V
pj
E
pj
, let

pj p
s : V
pjp
E
p
[
Vp
j
p
be the unique sec-
tion satisfying

pj p

pj p
s = s
pjp
. Let

V
p
=
pU

j
V
pj p
. We dene
p
=

pU

pj p
(
j

j
) :

V
p
E
p
[

Vp
.
There exist p
1
, . . . , p

l
M such that

U
j
=
pj
(s
1
pj
(0)

V
pj
) [ j = 1, . . . ,

l is
an open cover of M. We may choose = (
1
, . . . ,
l
) such that s
pj
+
pj
:

V
pj

E
pj
[

V
p
j
is smooth and intersects the zero section transversally for j = 1, . . . ,

l. So

j
= (s
pj
+
pj
)
1
(0) is a d-dimensional submanifold of

V
pj
, where d is the virtual
dimension of the Kuranishi structure. The orientation of the Kuranishi structure
induces an orientation on

M

j
. If

V
pj
has corners, we may further require that

M

j
intersect all the corners of

V
pj
transversally, i.e.,

M

j
is a neat submanifold of

V
pj
in
the sense of [Hi, Chapter 1, Section 4]. We call such =
p
:

V
p
E
p
[

Vp
[ p M
a generic perturbation. Note that the dierence between two generic perturbations
is smooth.
From our choice of
pj
, we have
(2)
pj
(

M

j
)
p
j

V
p
j

) =
pj
(

V
pj
)
p
j

(

M

j
).
for any j, j

1, . . . ,

l.
75
Choose a triangulation of

M

j
so that it becomes a simplicial complex, and all
its corners are subcomplexes. By (2), we may assume that there is a compact
subcomplex K
j
of

M

j
such that

l
j=1

pj
(K
j
) =

l
j=1

pj
(

M

j
).
For any j, j

1, . . . ,

l,
1
pj
_

pj
(K
j
)
p
j

(K
j
)
_
is a subcomplex of
the (d 1)-dimensional simplicial complex K
j
.
Let
d
j,
[ = 1, . . . , N
j
be the set of d-simplices in the triangulation of K
j
.
Each
d
j,
has an orientation induced by that of

M

j
. The inclusion i
j,
:
d
j,

K
j


V
pj
can be viewed as a singular d-chain in

V
pj
. We dene
K

j
=
Nj

=1
i
j,
o
d
(

V
pj
; Z) o
d
(

V
pj
; Q)
M
K,
=

j=1
1
[
pj
[

pj

j
o
d
(W; Q)
where [
pj
[ is the cardinality of
pj
. It follows from our construction that M
K,

o
d1
(W) o
d1
(W), so it represents a singular relative d-cycle

M
K,
o
d1
(W, W; Q),
which represents a class [M
K,
]
rel
H
d
(W, W; Q).
Note that up to subdivision, M
K,
o
d
(W; Q) depends on =
p
:

V
p

E
p
[

Vp
[ p M but not on the choice of p
1
, . . . , p

l
M.
Proposition 7.4. The class [M
K,
]
rel
H
d
(W, W; Q) is independent of the choice
of =
p
:

V
p
E
p
[

Vp
[ p M. So we may write [M
K
]
rel
for this class.
Proof. We rst observe that a Kuranishi structure with corners / on M with ambi-
ent space W gives rise to a Kuranishi structure with corners /[0, 1] on M[0, 1]
with ambient space W [0, 1]. To see this, consider (p, t) M [0, 1]. Let
(V
p
, E
p
,
p
,
p
, s
p
) be the Kuranishi neighborhood of p assigned by the Kuranishi
structure on M. Let V
(p,t)
= V
p
[0, 1], and let
p
: V
(p,t)
V
p
be the pro-
jection to the rst factor. Let E
(p,t)
=

p
E
p
V
(p,t)
, and let s
(p,t)
=

p
s
p
:
V
(p,t)
E
(p,t)
. We dene
(p,t)
=
p
id : V
(p,t)
= V
p
[0, 1] W [0, 1],
where id is the identity map on [0, 1]. Finally, let
p
act on [0, 1] trivially. Then
(V
(p,t)
, E
(p,t)
,
p
,
(p,t)
, s
(p,t)
) is a Kuranishi neighborhood of (p, t). The transition
functions can be constructed from those of the Kuranishi structure on M in an
obvious way.
Let =
p
:

V
p
E
p
[

Vp
[ p M,

p
:

V
p
E
p
[

Vp
[ p M be two
choices of small generic perturbation of / in the above construction. There exists
a generic perturbation
=
(p,t)
: V
(p,t)
E
(p,t)
[ (p, t) M [0, 1]
of / [0, 1] such that
i

p,0

(p,
1
2
)
=
p
: V
p
E
p
, i

p,1

(p,
1
2
)
=

p
: V
p
E
p
,
where i
p,t
: V
p
V
(p,
1
2
)
= V
p
[0, 1] is the inclusion x (x, t).
76
From the paragraph right before Proposition 7.4, we may use p
1
, . . . , p

l
in the
construction of both M

and M

. Let
Y
j
= (s
( pj ,
1
2
)
+
( pj,
1
2
)
)
1
(0)

V
pj
[0, 1].
Then
Y
j
(

V
pj
[0, 1]) = i
pj,1
(

M

j
) i
pj,0
(

M

j
)
_
Y
j
(

V
pj
[0, 1])
_
,
where
(

V
pj
[0, 1]) = i
pj,1
(

V
pj
) i
pj,0
(

V
pj
) (

V
pj
[0, 1])
is the union of corners of

V
pj
[0, 1]. This gives rise to a singular (d + 1)-chain B
j
in

V
pj
[0, 1] such that
B
j
= (i
pj ,1
)

j
(i
pj,0
)

j
+C
j
+D
j
,
where (i
pj,1
)

j
comes from i
pj,1
(

M

j
), (i
pj,0
)

j
comes from i
pj,0
(

M

j
), D
j
comes from Y
j
(

V
pj
[0, 1]), and C
j
will get canceled:

j=1
(
pj
id)

C
j
= 0 o
d
(W [0, 1]; Q).
Let
j
:

V
pj
[0, 1]

V
pj
be the projection to the rst factor. We have
(
j

B
j
) = K

j
K

j
+ (
j
)

C
j
+ (
j
)

D
j
.
Let
B =

j=1
1
[
pj
[

pj

B
j
o
d+1
(W; Q),
and
D =

j=1
1
[
pj
[

pj

D
j
o
d
(W; Q) o
d
(W; Q).
Then
B = M
K,
M
K,
+D o
d
(W; Q)
since

j=1
1
[
pj
[

pj

C
j
=

_
_

j=1
1
[
pj
[
(
pj
id)

C
j
_
_
= 0,
where : W [0, 1] W is the projection to the rst factor. We have

B =

M
K,


M
K,
o
d
(W, W; Q), so
[M
K,
]
rel
= [M
K,
]
rel
H
d
(W, W; Q). 2
Proposition 7.5. Let /
1
and /
2
be two equivalent Kuranishi structures with cor-
ners on a compact Hausdor topological space M. Let W be an ambient space of
both /
1
and /
2
. Then [M
K1
]
rel
= [M
K2
]
rel
.
77
Proof. Let
/
1
=
_
(V
1,p
, E
1,p
,
1,p
,
1,p
, s
1,p
) : p M, (V
1,pq
,

1,pq
,
1,pq
, h
1,pq
) : q
1,p
(s
1
1,p
(0))
_
/
2
=
_
(V
2,p
, E
2,p
,
2,p
,
2,p
, s
2,p
) : p M, (V
2,pq
,

2,pq
,
2,pq
, h
2,pq
) : q
2,p
(s
1
2,p
(0))
_
/ =
_
(V
p
, E
p
,
p
,
p
, s
p
) : p M, (V
pq
,

pq
,
pq
, h
pq
) : q
p
(s
1
p
(0))
_
be as in Denition 6.5. Let
i
=
i,p
: V
i,p
E
i,p
[ p M be a generic
perturbation which can be used to dene a virtual fundamental chain M
Ki,i

o
d
(W; Q), for i = 1, 2.
i
can be extended to a generic perturbation
i
=
i,p
:
V
p
E
p
[ p M such that

i

i,p
=
i,p

i
. We have M
Ki,i
= M
K,i

o
d
(W; Q). So
[M
K1
]
rel
= [M
K,1
]
rel
= [M
K,2
]
rel
= [M
K2
]
rel
H
d
(W, W; Q). 2
Remark 7.6. Let M be a compact Hausdor topological space with an oriented
Kuranishi structure /, and let W be an ambient space of /. Then the above con-
struction yields M
K,
o
d
(W, Q) such that M
K,
= 0 o
d1
(W, Q), so it repre-
sents a class [M
K,
] H
d
(W; Q). The proof of Proposition 7.4 shows that this class
is independent of the choice of , so we may write [M
K
] for this class. The proof
of Proposition 7.5 shows that if /

is another Kuranishi structure on M such that

/

/ and W is also an ambient space of /

, then [M
K
] = [M
K
] H
d
(W; Q).
Let i : M W be the inclusion. Then [M
K
] = i

[M] H
d
(W; Q), where i

[M] is
dened in [FO, Section 6].
Example 7.7. Let X be a Calabi-Yau 3-fold, and let H
2
(X; Z). Let M
g,0
(X, )
denote the moduli space of stable maps f from a genus g prestable curve C to
X such that f

[C] = . Then M
g,0
(X, ) has an oriented Kuranishi structure
[FO]. The virtual dimension of is 0 for any g 0 and H
2
(X; Z). This
Kuranishi structure has an ambient space W
1,p
g,0
(X, ), the moduli space of stable
W
1,p
maps from a genus g prestable curve C to X such that f

[C] = . Then
[M
g,0
(X, )
K
] H
0
(W
1,p
g,0
(X, ); Q), and deg[M
g,0
(X, )
K
] Q is some Gromov-
Witten invariant [FO, Section 17].
Example 7.8. Let X be a Calabi-Yau 3-fold, L be a special Lagrangian subman-
ifold.

M
(g,h),(n, m)
(X, L [ , , ) is empty for ,= 0. The tangent bundle of L is
trivial, so L is spin. We have constructed an orientable Kuranishi structure with
corners / on /=

M
(g,h),(0,

0)
(X, L [ , , 0). The equivalence class of / is indepen-
dent of choices in our construction. The virtual dimension of / is 0 for all g, h, , .
The Kuranishi structure has an ambient space J = W
1,p
(g,h),(0,

0)
(X, L [ , , 0).
[/
K
]
rel
H
0
(J, J; Q) = 0, so we cannot get a nontrivial number from [/
K
]
rel
.
Most enumerative predictions about holomorphic curves with Lagrangian bound-
ary conditions concern the special case in Example 7.8. Our goal is to give a rigor-
ous mathematical denition of these highly nontrivial enumerative numbers. From
Example 7.8 we know that the relative cycle [M
K
]
rel
H
d
(W, W; Q) is not the
right object for our purpose. We want to remember the virtual fundamental chain
M
K,
o
d
(W, Q) which contains more information. For example, when the virtual
dimension d = 0, the 0-chain M
K,
represents a class [M
K,
] H
0
(W; Q) since any
0-chain is a cycle. Let ,

be two perturbations. From the proof of Proposition 7.5,

B = M
K,
M
K,
+D,
78
where B o
1
(W, Q), D o
0
(W, Q). So [M
K,
] depends on . We want to
impose extra constraint on such that if and

both satisfy the constraint

then the above D is zero. In particular, if and

satisfy the same boundary

conditions in the sense that
pj
=

pj
on

V
pj
for all j = 1, . . . ,

l, then D = 0, so
[M
K,
] = [M
K,
] H
0
(W; Q), and deg[M
K,
] = deg[M
K,
] Q.
7.2. S
1
action.
Denition 7.9. Let M be a Hausdor topological space, and let : S
1
M M
be a continuous S
1
action. A Kuranishi structure with corners
/ =
_
(V
p
, E
p
,
p
,
p
, s
p
) : p M, (V
pq
,

pq
,
pq
, h
pq
) : q
p
(s
1
p
(0))
_
.
on M is -equivariant on the boundary if the Kuranishi neighborhood (V
p
, E
p
,
p
,
p
, s
p
)
of any p M is -equivariant in the sense that
(1) There is a free continuous S
1
action on V
p
which commutes with the action
of
p
and leaves V
p
invariant.
(2) E
p
V
p
is an S
1
-equivariant vector bundle.
(3) s
p
: V
p
E
p
is an S
1
-equivariant section.
(4)
p
: s
1
p
(0) M is S
1
-equivariant, where S
1
acts on M by .
Remark 7.10. Let / be as above, and let / be the Kuranishi structure with
corners on M dened in Remark 6.4. Then the quotients of the Kuranishi neigh-
borhoods and transition functions of / by the free S
1
action dene a Kuranishi
structure //S
1
of virtual dimension d 2 on M/S
1
.
Remark 7.11. Let M be a Hausdor topological space with a continuous S
1
-action
: S
1
M M. Then there exist an -equivariant Kuranishi structure with
corners only if the action leaves M invariant and has no xed point on M.
Denition 7.12. Let M be a Hausdor topological space with a continuous S
1
-
action : S
1
M M. Let
/
1
=
_
(V
1,p
, E
1,p
,
1,p
,
1,p
, s
1,p
) : p M, (V
1,pq
,

1,pq
,
1,pq
, h
1,pq
) : q
1,p
(s
1
1,p
(0))
_
and
/
2
=
_
(V
2,p
, E
2,p
,
2,p
,
2,p
, s
2,p
) : p M, (V
2,pq
,

2,pq
,
2,pq
, h
2,pq
) : q
2,p
(s
1
2,p
(0))
_
be two Kuranishi structures with corners on M which are -equivariant on the
boundary. /
1
and /
2
are are -equivalent if
There is another Kuranishi structure
/ =
_
(V
p
, E
p
,
p
,
p
, s
p
) : p M, (V
pq
,

pq
,
pq
, h
pq
) : q
p
(s
1
p
(0))
_
on M which is -equivariant on the boundary such that for all p M,
(V
1,p
, E
1,p
,
1,p
,
1,p
, s
1,p
), (V
2,p
, E
2,p
,
2,p
,
2,p
, s
2,p
), and (V
p
, E
p
,
p
,
p
, s
p
)
satisfy the relation described in Denition 6.2.
The
i
and

i
in Denition 6.2 are S
1
-equivariant.
In this case, we write /
1

/
2
.
Note that /
1

/
2
/
1
/
2
.
79
Denition 7.13. Let M be a Hausdor topological space with a continuous S
1
action : S
1
M M. Let
/ =
_
(V
p
, E
p
,
p
,
p
, s
p
) : p M, (V
pq
,

pq
,
pq
, h
pq
) : q
p
(s
1
p
(0))
_
be a Kuranishi structure with corners on M which is -equivariant on the boundary.
An ambient space W of / is -equivariant if extends to an action S
1
W W,
and
p
: V
p
W is S
1
-equivariant.
We now assume that M is a compact, Hausdor topological space with a con-
tinuous S
1
action : S
1
M M. Suppose that / is an oriented Kuranishi
structure with corners of virtual dimension zero on M. We further assume that /
is -equivariant on the boundary, and W is an -equivariant ambient space of /.
In this case, a generic perturbation =
p
: V
p
E
p
[ p M of / is a
perturbation such that s
p
+
p
intersects the zero section transversally at isolated
points and is nowhere zero on V
p
. Let p denote the equivalent class of p M in
M/S
1
. The virtual dimension of
//S
1
=
_
((

V
p
,

E
p
,
p
,

p
, s
p
: p M/S
1
, (

V
p q
,

p q
,

p q
,

h
p q
) : q

p
( s
1
p
(0))
_
is 2, so a generic perturbation =

:

V
p


E
p
[ p M/S
1
of //S
1
is a
perturbation such that s
p
+
p
is nowhere zero for all p M/S
1
. Let Q
p
: V
p

V
p
= V
p
/S
1
be the natural projection. Then Q

= (Q

)
p
= Q

p
: V
p

E[
Vp
[ p M is a generic perturbation of /. We call such a perturbation an
-equivariant perturbation. A generic perturbation of / can always be extended
to a generic perturbation of /, so there exist a generic perturbation of / whose
restriction to / is -equivariant.
Proposition 7.14. Let M be a compact Hausdor topological space with a con-
tinuous S
1
action : S
1
M M. Let / be a Kuranishi structure with corners
of virtual dimension 0 on M which is -equivariant on the boundary. Let W be a
-equivariant ambient space of /. Let =
p
: V
p
E
p
[ p M be a generic
perturbation such that [
K
=
p
[
Vp
: V
p
E
p
[
Vp
[ p M is -equivariant.
Then [M
K,
] H
0
(W; Q) does not depend on the perturbation , so we may write
[M

K
] for this class. If /

/, and W is also a -equivariant ambient space of /

,
then [M

K
] = [M

K
] H
0
(W; Q).
Proof. Let
=
p
: V
p
E
p
[ p M,

p
: V
p
E
p
[ p M
be two generic perturbations of / such that [
K
= Q

[
K
= Q

, where
=
p
:

V
p


E
p
[ p M/S
1
,

p
:

V
p


E
p
[ p M/S
1

are perturbations for //S

1
such that s
p
+
p
, s
p
+

p
are nowhere zero for all
p M/S
1
. There exists a generic perturbation
=
( p,t)
: V
( p,t)
E
( p,t)
[ ( p, t) M/S
1
[0, 1]
such that
i

p,0

( p,
1
2
)
=
p
:

V
p


E
p
, i

p,1

( p,
1
2
)
=

p
:

V
p


E
p
,
where i
p,t
: V
p
V
( p,
1
2
)
= V
p
[0, 1] is the inclusion x (x, t). The virtual
dimension of //S
1
[0, 1] is 1, so being generic simply means s
( p,t)
+
( p,t)
80
is nonzero for all ( p, t) M/S
1
[0, 1]. The pull back Q

of is a generic
perturbation of / [0, 1] such that s
(p,t)
+ (Q

)
(p,t)
is nowhere zero for any
(p, t) M [0, 1]. We may extend Q

to a generic perturbation of / [0, 1]

such that
i

p,0

(p,
1
2
)
=
p
: V
p
E
p
, i

p,1

(p,
1
2
)
=

p
: V
p
E
p
as in the proof of Proposition 7.4. We proceed as the proof of Proposition 7.4, and
use the notation there. Y
j
does not intersect

V
pj
[0, 1], so D = 0, and the 1-chain
B o
1
(W; Q) satises
B = M
K,
M
K,
o
0
(W; Q).
So we have [M
K,
] = [M
K,
] H
0
(W; Q).
The last statement can be proved as Proposition 7.5. 2
7.3. Invariants for an S
1
-equivariant pair. Let (X, ) be a compact symplec-
tic manifold together with an -tame almost complex structure J, and L be a
Lagrangian submanifold. We assume that L is spin or that h = 1 and L is relative
spin so that / =

M
(g,h),(n, m)
(X, L [ , , ) has an orientable Kuranishi struc-
ture. We x an orientation by choosing a stable trivialization of TL or TL V
on the 2-skeleton L
(2)
of L, where V is chosen as in the proof of Theorem 6.36.
J = W
1,p
(g,h),(n, m)
(X, L [ , , ), is an ambient space of the Kuranishi structure on
/.
Denition 7.15. An admissible S
1
action on (X, L) is an S
1
action : S
1
X
X such that
preserves J and L.
The restriction of to L is free.
We now assume that there is an admissible S
1
action on (X, L). Given t = e
i

S
1
, let f
t
: X X be the J-holomorphic dieomorphism given by x t x. We
have an S
1
action : S
1
J J given by (t, [(, u)]) t [(, u)] = [(, f
t
u)].
The action preserves / J since f
t
is J-holomorphic for all t S
1
.
If (, B; p; q
1
, . . . , q
h
; u) represents a xed point of the S
1
action, then =
C D
1
D
h
, where C is a genus g prestable curve, and D
k
is a disc which
intersects C at an interior node for k = 1, . . . , h. u(D
k
) is an orbit of the S
1
action on L. Let J
S
1
and /
S
1
denote the xed loci of the S
1
action on J and
/, respectively. Then /
S
1
= J
S
1
/, J
S
1
J = , and /
S
1
/= .
Theorem 7.16. Let (X, L) be as above. Then there exists an oriented Kuran-
ishi structure on / =

M
(g,h),(n, m)
(X, L [ , , ) which is -equivariant on the
boundary. J = W
1,p
(g,h),(n, m)
(X, L [ , , ) is a -equivariant ambient space of /.
Proof. Let
/ =
_
(V

, E

, Aut ,

, s

) : /, (V

, h

),

(s
1
(0))
_
be a Kuranishi structure constructed as in Section 6. For /, we will modify
the Kuranishi neighborhood of such that 1.-4. in Denition 7.9 hold.
Given = (, u) /, let H
,domain
, H
,map
= Ker( D
u
) be dened as in
Section 6. Recall that Aut u
1
induces an inclusion H
,aut
H
,map
.
Similarly, t S
1
f
t
u induces an inclusion H
,circle
= T
1
S
1
H
,map
, where
T
1
S
1
= R is the tangent line of S
1
= e
i
[ R at 1 S
1
. Note that H
,circle

81
H
,aut
if and only if /
S
1
, which is excluded since we consider /. We
may choose H

,map
such that H
,circle
H

,map
and H
,map
= H
,aut
H

,map
.
Choose a subspace H

,map
in H

,map
such that H

,map
= H
,circle
H

,map
.
Let V

,map
be a small neighborhood of 0 in H

,map
, and > 0 be small. Then
for suciently small (, ,

) B
2
D
d
D

d
and w

,map
, there are unique
w

,map
and (, ) such that
u
(,,

,w

)
= f
e
i u
(,,

,w

)
u
(,,

,w

,e
i
)
,
where the notation is the same as in Section 6. Both Aut and S
1
act on B
2
D
d

d
V

,map
S
1
such that u
(,,

,w

,t)
= u
(,,

,w

,t)

1
and t

(, ,

, w

, t) =
(, ,

, w

, t

t) for Aut , t

S
1
, (, ,

, w

, t) B
2
D
d
D

d
V

,map
S
1
.
Note that the action of Aut commutes with that of S
1
. So we may choose a
neighborhood V

of 0 in B
2
D
d
D

d
V

,map
such that V

S
1
is invariant
under the action of Aut . There is an S
1
equivariant map

: V

S
1
J
(, ,

, w

, t) [(
(,,

)
, u
(,,

,w

,t)
)]
(V

S
1
)/Aut J is injective if and only if S
1

= 1, where S
1

is the stabilizer
at of the S
1
action. In general, S
1

is a nite group because M is not a xed

point of the S
1
action.
The automorphism of M/S
1
is
Aut = Aut [ u = f
t
u for some t S
1
,
and
S
1

= t S
1
[ u = f
t
u for some Aut .
Aut is a normal subgroup of Aut , and we have an exact sequence
1 Aut Aut S
1

1.
The action of Aut on V

S
1
extends to Aut , and (V

S
1
)/Aut J is
injective.
Let q

: V

S
1
V

be the projection to the rst factor, and let E V

be the
restriction of the obstruction bundle over V

. Let

V

= V

S
1
,

E

= q

E

V

,
s

= q

(s

[
V

). Then (

,

E

, Aut ,

, s

) is a Kuranishi neighborhood of /
in / which satises 1.-4. in Denition 7.9.
Finally, we proceed as in Section 6.5 to construct new transition functions. 2
The -equivalence class of the Kuranishi structure with corners which is -
equivariant on the boundary constructed in the above proof does not depend on
various choices.
Now consider the case m = n = 0, and the virtual dimension
d = + (N 3)(2 2g h) = 0.
We dene the Euler characteristic of /=

M
(g,h),(0,

0)
(X, L[, , ) to be

(g,h)
(X, L, [, , ) = deg[/

K
] Q,
where / is a Kuranishi structure with corners which is -equivariant on the bound-
ary, constructed in the proof of Theorem 7.16. This number is well-dened by
Proposition 7.14.
82

(g,h)
(X, L, [, , ) is an invariant for the equivariant pair (X, L, ), but not
an invariant for the pair (X, L). In other words, it is possible that

(g,h)
(X, L,
1
[, , ) ,=
(g,h)
(X, L,
2
[, , )
for two dierent admissible S
1
actions
1
,
2
on (X, L).
7.4. Multiple covers of the disc. We consider the special case studied in [OV,
KL, LS]. Let (z, u, v) and ( z, u, v) be the two charts of O
P
1 (1)O
P
1(1), related
by ( z, u, v) = (
1
z
, zu, zv). Let X be the total space of O
P
1 (1) O
P
1(1). There
is an antiholomorphic involution
A : X X
(z, u, v) (
1
z
, z v, z u)
in terms of the rst chart. The xed locus L = X
A
is a special Lagrangian submani-
fold of the noncompact Calabi-Yau 3-fold. For any integer a, let
a
: S
1
X X be
the S
1
action on X dened by (e
i
, (z, u, v)) (e
i
z, e
i(a1)
u, e
ia
v) on the rst
chart. Then
a
is an admissible S
1
action on (X, L). Let D
2
= (z, 0, 0) [ [z[ 1
be a disc in the rst chart, oriented by the complex structure. Then D
2
L. Let
= [D
2
] H
2
(X, L; Z)

= H
2
(P
1
, S
1
; Z) and = [D
2
] H
1
(L; Z)

= H
1
(S
1
; Z).
By Schwartz reection principle (see e.g [KL, Section 3.3.2]), any nonconstant
holomorphic map f : (, ) (X, L) can be extended to a nonconstant holomor-
phic map f
C
:
C
X, whose image must lie in P
1
. So we have
/=

M
(g,h),(0,

0)
(X, L [ d, (n
1
, . . . , n
h
), 0) =

M
(g,h),(0,

0)
(P
1
, S
1
[ d, (n
1
, . . . , n
h
), 2d)
as topological spaces. Therefore,

M
(g,h),(0,

0)
(X, L [ d, (n
1
, . . . , n
h
), 0) is com-
pact in C

topology. Note that the virtual dimension of the Kuranishi structure

corners is 0 for

M
(g,h),(n,

0)
(X, L [ d, (n
1
, . . . , n
h
), 0) and 2(d + 2g + h 2)
for

M
(g,h),(n,

0)
(P
1
, S
1
[ d, (n
1
, . . . , n
h
), 0). In particular, these two Kuranishi
structures on / are not equivalent. The following numbers are dened:
C(g, h[d; n
1
, . . . , n
h
[a) =
(g,h)
(X, L,
a
[ d, (n
1
, . . . , n
h
), 0) Q,
where n
1
, . . . , n
h
are positive integers, and d = n
1
+ +n
h
.
Let : (

M
g,h
be the universal family,

be the relative dualizing sheaf, and

s
i
:

M
g,h
( be the section corresponding to the i-th marked point. Let E =

be the Hodge bundle on

M
(g,h)
, and
i
= c
1
(s

).
Conjecture 7.17. Let a be a positive integer. Then
(1)
dh
C(0; h[d; n
1
, . . . , n
h
[a) = C(0; h[d; n
1
, . . . , n
h
[1 a)
= (a(1 a))
h1
h

i=1
_
n
i
a 1
n
i
1
_
d
h3
.
For g > 0,
(1)
dh
C(g; h[d; n
1
, . . . , n
h
[a) = C(g; h[d; n
1
, . . . , n
h
[1 a)
= (a(1 a))
h1
h

i=1
_
n
i
a 1
n
i
1
_

_
(

M
g,h)
U(1)
c
g
(E

())c
g
(E

((a 1)))c
g
(E

(a))
2h3

h
i=1
( n
i

i
)
.
83
The above formulae for C(g; h[d; n
1
, . . . , n
h
[a) are calculated in [KL] by localiza-
tion techniques using the S
1
action
a
. Actually, the denition of
(g,h)
(X, L, [
, , ) is inspired by R. Botts interpretation of the computations in [KL]. The
localization formula, and in particular the proof of Conjecture 7.17, is left to future
work.
Finally, the assumption of the existence of an admissible S
1
action is too restric-
tive. The S
1
action disappears when we perturb the almost complex structure J or
the Lagrangian submanifold L, so the invariant is not even dened for other almost
complex structures, and it is not clear in which sense
(g,h)
(X, L, [ , , ) is an
invariant. It is desirable to nd a natural way to impose boundary conditions for
the general case.
References
[Ab] W. Abiko, The real analytic theory of Teichm uller space, Lecture Notes in Mathematics,
Vol. 820. Springer-Verlag, Berlin-New York, 1980.
[Ar] N. Aronszajn, A unique continuation theorem for solutions of elliptic partial dierential
equations or inequalities of second order, J. Math. Pures Appl. (9) 36 (1957), 235249.
[AG] N.L. Alling and N. Greenleaf, Foundations of The theory of Klein Surfaces, Lecture
Notes in Mathematics, Vol. 219, Springer-Verlag, Berlin-New York, 1971.
[AKV] M. Aganagic, A. Klemm, and C. Vafa, Disc Instantons, Mirror Symmetry, and the
Duality Web, Z. Naturforsch. A 57 (2002), no. 1-2, 128
[AV] M. Aganagic and C. Vafa, Mirror Symmetry, D-Branes and Counting Holomorphic
Discs, preprint, 2000, hep-th/0012041.
[BEGG] E. Bujalance, J.J. Etayo, J.M. Gamboa, and G. Gromadzki, Automorphism Groups of
Compact Bordered Klein Surfaces, Lecture Notes in Mathematics, Vol. 1439, Springer-
Verlag, Berlin-New York, 1990.
[BF] K. Behrend and B. Fantechi, The intrinsic normal cone, Invent. Math. 128 (1997), no.
1, 4588.
[CdGP] P. Candelas, X. de la Ossa, P. Green and L. Parkes, A pair of Calabi-Yau manifolds as
an exactly soluble superconformal eld theory, Nuclear Physics B359 (1991), 2174 and
in Essays on Mirror Manifolds (S.-T. Yau, ed.), International Press, Hong Kong 1992,
3195.
[FO] K. Fukaya and K. Ono, Arnold conjecture and Gromov-Witten invariant, Topology 38
(1999), no. 5, 9331048.
[FO
3
] K. Fukaya, Y.-G. Oh, H. Ohta, and K. Ono, Lagrangian intersection Floer theory
anomaly and obstruction, preprint 2000.
[Gi] A. Givental, Equivariant Gromov-Witten Invariants, Internat. Math. Res. Notices
(1996), 613663.
[GV] R. Gopakumar and C. Vafa, M-Theory and Topological StringsII , preprint 1998,
hep-th/9812127.
[GP] T. Graber and R. Pandharipande, Localization of virtual classes, Invent. Math. 135
(1999), 487-518
[GZ] T. Graber, E. Zaslow Open-String Gromov-Witten Invariants: Calculations and a Mir-
ror Theorem, hep-th/0109075.
[Gr] M. Gromov, Pseudoholomorphic curves in symplectic manifolds, Invent. Math. 82
(1985), no. 2, 307347.
[Ha] R. Hartshorne, Algebraic geometry, Graduate Texts in Mathematics, 52, Springer-Verlag,
New York-Heidelberg, 1977.
[Hi] M.W. Hirsch Dierential topology, Graduate Texts in Mathematics, 33, Springer-Verlag,
New York-Heidelberg, 1976.
[HM] J. Harris and I. Morrison, Moduli of curves, Graduate Texts in Mathematics, 187,
Springer-Verlag, New York, 1998.
[IS1] S. Ivashkovich and V. Shevchishin, Gromov compactneww theorem for J-complex curves
with boundary Internat. Math. Res. Notices 2000, no. 22, 11671206.
84
[IS2] S. Ivashkovich and V. Shevchishin, Holomorphic structure on the space of Riemann
surfaces with marked boundary, Tr. Mat. Inst. Steklova 235 (2001), Anal. i Geom. Vopr.
Kompleks. Analiza, 98109; translation in Proc. Steklov Inst. Math. 2001, no. 4 (235),
91102.
[IS3] S. Ivashkovich, V. Shevchishin, Reection principle and J-complex curves with boundary
on totally real immersions, Commun. Contemp. Math. 4 (2002), no. 1, 65106.
[Kl] F. Klein,

Uber Realitatsverhaltnisse bei der einem beliebigen Geschlechte zugehorigen
Normalkurve der , Math. Ann., Vol. 42, 1892.
[Ko] K. Kodaira, Complex manifolds and deformation of complex structures, Grundlehren-
der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences],
283. Springer-Verlag, New York, 1986.
[KL] S. Katz and C.C. Liu, Enumerative Geometry of Stable Maps with Lagrangian Boundary
Conditions and Multiple Covers of the Disc, Adv. Theor. Math. Phys. 5 (2001), no. 1,
149.
[LLY] B. Lian, K. Liu and S.-T. Yau, Mirror principle I, Asian J. Math. 1 (1997), 729763.
[LS] J. Li and Y.S. Song, Open string instantons and relative stable morphisms, Adv. Theor.
Math. Phys. 5 (2001), no. 1, 6791.
[LT1] J. Li and G. Tian, Virtual moduli cycles and Gromov-Witten invariants of algebraic
varieties, J. Amer. Math. Soc. 11 (1998), no. 1, 119174.
[LT2] J. Li and G. Tian, Virtual moduli cycles and Gromov-Witten invariants of general
symplectic manifolds, Topics in symplectic 4-manifolds (Irvine, CA, 1996), 4783, First
Int. Press Lect. Ser., I, Internat. Press, Cambridge, MA, 1998.
[LT] G. Liu and G. Tian, Floer homology and Arnold conjecture, J. Dierential Geom. 49
(1998), no. 1, 174.
[LMV] J.M.F. Labastida, M. Mari no and C. Vafa Knots, links and branes at large N, Jour.
High En. Phys. 0011 (2000) 007.
[MS] D. McDu and D. Salamon, J-holomorphic curves and quantum cohomology University
Lecture Series, 6. American Mathematical Society, Providence, RI, 1994.
[MV] M. Mari no and C. Vafa Framed knots at large N, hep-th/0108064.
[OV] H. Ooguri and C. Vafa, Knot invariants and topological strings, Nuclear Phys. B577
(2000), 69-108.
[PW] T. Parker and J.G. Wolfson, Pseudo-holomorphic maps and bubble trees, J. Geom. Anal.
3 (1993), no. 1, 6398.
[Sa] I. Satake, The Gauss-Bonnet theorem for V -manifolds, J. Math. Soc. Japan 9 (1957),
464-492.
[Se] M. Sepp al a, Moduli spaces of stable real algebraic curves, Ann. Sci.

Ecole norm.Sup. (4)
24 (1991).
[Sie1] B. Siebert, Gromov-Witten invariants of general symplectic manifolds, dg-ga/9608005
[Sie2] B. Siebert, Symplectic Gromov-Witten invariants, New trends in algebraic geometry
(Warwick, 1996), 375424, London Math. Soc. Lecture Note Ser., 264, Cambridge Univ.
Press, Cambridge, 1999.
[Sil] R. Silhol, Real algebraic surfaces, Lecture Notes in Mathematics, Vol. 1392. Springer-
Verlag, Berlin, 1989
[St] J.D. Stashe, Homotopy associativity of H-spaces I, II, Trans. Amer. Math. Soc. 108
(1963), 275292, 293312.
[SS] M. Sepp al a and R. Silhol, Moduli spaces for real algebraic curves and real abelian vari-
eties, Math. Z. 201 (1989), 151-165.
[SU] J. Sacks and K. Uhlenbeck, The existence of minimal immersions of 2-spheres, Ann. of
Math. (2) 113 (1981), no. 1, 124.
[Th] W.P. Thurston, The geometry and topology of three-manifolds, Princeton mimeographed
lecture notes, 1979.
[We] G. Weichhold,

Uber symmetrische Riemannsche Flachen und die Periodizitatsmodulen
der zugerhorigen Abelschen Normalintegrale erstes Gattung , Leipziger dissertation,
1883.
[Wolf] J.G. Wolfson, Gromovs compactness of pseudo-holomorphic curves and symplectic ge-
ometry,J. Dierential Geom. 28 (1988), no. 3, 383405.
[Wolp] S. Wolpert, On the Weil-Petersson geometry of the moduli space of curves, Amer. J.
Math. 107 (1985), no. 4, 969997.
85
[Ye] R. Ye, Gromovs compactness theorem for pseudo holomorphic curves, Trans. Amer.
Math. Soc. 342 (1994), no. 2, 671694.
E-mail address: ccliu@math.harvard.edu
Harvard University, Department of Mathematics, One Oxford Street, Cambridge,
MA 02138, USA