Advanced Studies in Pure Mathematics 20, 1992

Aspects of Low Dimensional Manifolds

pp. 1-92

Floer Homology for Oriented -Manifolds

$3$

Kenji Fukaya

Dedicated to Professor Akio Hattori on his sixtieth birthday

Contents

\S 1 Introduction
\S 2 Perturbation
\S 3 Local structure of moduli space
\S 4 Sum formula for index bundles
\S 5 Dimension of moduli space
\S 6 Orientation of moduli space
\S 7 Partial compactification of moduli space
\S 8 Taubes construction
\S 9 Decay estimate
\S 10 Local action on the end of moduli space
\S 11 Extension of the line bundle to the boundary
\S 12 Boundary operators
\S 13 Independence of the metrics and the perturbations

\S 1. Introduction
In [F], A. Floer introduced a new invariant for homology 3-spheres.
In this paper we generalize his invariant to arbitrary closed and oriented
3-manifolds. In the case when the first homology group of the manifold
is torsion free and nonzero, we also define invariants for $s<$
$I_{k}^{s}(M)$

, which, in the case $s=0$ , is a generalization of Floers one. The

$3$

construction of this invariant is closely related also to the Donaldsons

polynomial for closed 4-manifolds [D4]. The construction is based on
the study of the moduli space of selfdual connections over $M$ and $\times R$

its compactification.

Received December 12, 1989.

A part of this paper is written during the authors stay in University of
Maryland, College Park Campus.
 K. Fukaya

In this section, we describe briefly the construction of our invariant.

Throughout this paper, we let $M$ be an oriented 3-manifold, a Rie- $\sigma$

mannian metric on it. It induces the Hodge -operator, : $*$ $*_{\sigma}$

$\Lambda^{k}(M)\rightarrow$

$\Lambda^{3-k}(M)$ . We consider the trivial $SU(2)$ bundle over $M$ . Let

$A(M)=\{d+a|a\in\Gamma(M, \Lambda^{1}\otimes su(2))\}$

be the set of all smooth connections of it. (In later sections, we work
with Sobolev spaces but in this section we omit those details.) Put

$\hat{\mathcal{G}}(M)=$
{ $g$ : $M\rightarrow SU(2)|C^{\infty}$ -maps},
$\mathcal{G}(M)=\{g \in\hat{\mathcal{G}}(M)|degg =0\}$ ,
$B(M)=A(M)/\hat{\mathcal{G}}(M)$ ,
$\overline{B}(M)=A(M)/\mathcal{G}(M)$ ,

where $\mathcal{G}(M)$
acts on $A(M)$ by

$g^{*}(d+a)=d+g^{-1}dg+g^{-1}ag$ .

Following Taubes [T4] and Floer [F], we define a functional $c\epsilon$

: $\overline{B}(M)\rightarrow$

$R$ by

(1.1) $c\epsilon(a)=\int_{M}$ Tr $(\frac{1}{2}a\wedge da+\frac{1}{3}a\wedge a\wedge a)$

(Here and hereafter, we shall write in place of $d+a.$ ) It is well known $a$

that the right hand side is -invariant. The gradient flow of this
$\mathcal{G}(M)$

functional is described by

(1.2) $\frac{\partial a_{t}}{\partial t}=*_{\sigma}F^{a_{t}}$

.

The idea of Floer and Taubes is to use this gradient flow in order to define
the -dimensional homology group of $B(M)$ . It is not in general
$\infty/2$

true that is a Morse-Smale flow, then in [T4], [F], they used a

$gradc\epsilon$

perturbation of it. In their case, where $M$ is a homology sphere, the

singular locus $SB(M)$ and the set of critical points of the flow $gradc\epsilon$

intersect at one point, the trivial connection. (Recall that the singular
locus of $B(M)$ is the set of reducible connections, and a critical point of
the flow is a flat connection.) In our case the intersection is
$gradc\epsilon$

(1.3) $Hom(\pi_{1}(M), U(1))/Z_{2}$ .

 Floer Homology for Oriented 3-Manifolds 3

which is $b_{1}(M)$ -dimensional. In \S 2, using the sum of the traces of the

holonomy along the generators of $H_{1}(M;Z)$ , we shall find a functional
: $B(M)\rightarrow R$ , such that the equation
$f$

(1.4) $*_{\sigma}F^{a}-grad_{a}f=0$

has only a finite number of solutions, each of which is nondegenerate

(see \S 2 for definition.) A connected component of elements of the set of
elements $SB(M)$ , the reducible connections, satisfying (1.4) is identified
to an element of

(1.5.1) $Hom(TorH_{1}(M;Z), U(1))/Z_{2}$ .

And each connected component is identified to

(1.5.2) $Hom(\frac{H_{1}(M\ovalbox{\tt\small REJECT} Z)}{TorH_{1}(M\cdot Z)}, Z_{2})$

or its quotient by $Z_{2}$

. Put

(1.6.1) $Fl$ $=$

{ $a\in\overline{B}(M)|$
a $satis5^{r}(1.4)$ },
(1.6.2) $Fl_{0}=$ { $a\in Fl|$ $a$ is irreducible}.
For $a$ , $b\in Fl_{0}$ , we set

$a_{t}$ satisfies (1.7),
$\Lambda 4(a, b)=\{a_{t}$ $ a_{t}.\cdot(-\infty, \infty)\rightarrow\overline{B}(M)t\rightarrow\infty t\rightarrow-\infty$
$\}$
.
$\lim a_{t}=b$ , $\lim a_{t}=a$

(The precise definition is in \S 3.) Here

(1.7) $\partial a_{t}=*_{\sigma}F^{a}-grad_{a_{t}}f$

$\overline{\partial t}$
.

In a way similar to [F], we can find a map $\mu$ : $Fl_{0}\rightarrow Z$

such that

$dim$ A4 $(a, b)=\mu(a)-\mu(b)$ ,

for , $b\in Fl_{0}$ $(5.)$ We can also prove that

$a$ $\mathcal{M}(a, b)$
is orientable (6).
Then, following Witten [W1] and Floer [F], we put

(1.8)
$C_{k}^{0}=\mu(a)=ka\in Fl_{O}\oplus Z[a]$
4 K. Fukaya

We define a boundary operator : as follows. (Again our $\partial$

$C_{k}^{0}\rightarrow C_{k-1}^{0}$

construction is the same as Floers.) The action of $R$ on $M\times R$ induces

a free action of $R$ on . We put, for $a\in Fl_{0}$ , $\mu(a)=k$ , $\Lambda 4(a, b)$

$\partial([a])=\sum_{\mu(b)=k-1}\langle\partial a, b\rangle[b]$
,

where is the dierence of the number of connected components

$\langle\partial a, b\rangle$

of A4 $(a, b)$ for which the direction of its orientation and the $R$ action
coincide and the number of connected components for which the orien-
tation is the opposite direction to the $R$-action. In a way similar to [F],
we can prove . Then we define $\partial\partial=0$

$I_{k}^{0}(M)=.\frac{Ker\partial.C_{k}^{0}\rightarrow C_{k-1}^{0}}{Im\partial\cdot C_{k+1}^{0}\rightarrow C_{k}^{0}}.$

,

which, we shall prove, is an invariant of M. (In fact, we need to fix a

basis of $H_{1}(M;Z).)$
As is pointed out by Donaldson, Atiyah [A] and Witten [W2], Floer
homology is closely related to the Donaldson polynomial [D4]. In fact,
in the case when $M$ is a homology sphere and is a boundary of a 4-
manifold satisfying some additional assumptions, it is possible to define
a relative Donaldson polynomial, which has a value in . But in $I_{k}^{0}(M)$

the case when the first Betti number of $M$ is positive, it seems that the
above boundary operator is not enough for such a purpose. Then we
construct other boundary operators. To motivate our construction we
recall the definition of relative Donaldson polynomial very briefly. (Our
description is not precise since it is anounced that the precise description
will appear in [DFK].) Let $X$ be a 4 manifold such that its boundary
$\partial X=M$ is a homology sphere. Let , , , $a\in Fl_{0}$ . By $[\Sigma_{1}]$ $\cdots$ $[\Sigma_{\ell}]\in H_{2}(X)$

$\Lambda 4_{k}(X;a)$ , we denote the set of all gauge classes of self dual connections

with $c^{2}(\nabla)=k$ ,
$\nabla$
. Define a line bundle on it by
$\nabla|_{\partial X}=a$ $\mathcal{L}_{\Sigma_{i}}$

$\mathcal{L}_{\Sigma_{i}}(\nabla)=top\wedge(Ker6_{\nabla 1\Sigma}i)^{*}\otimes top\wedge Coker6_{\nabla 1\Sigma}i$

,

where is a Dirac operator on

$6_{\nabla 1\Sigma}i$
$\Sigma_{i}$
twisted by the restriction of $\nabla$
to
. We put
$\Sigma_{i}$

$Q_{\ell}([\Sigma_{1}], \cdots, [\Sigma_{\ell}])(a)=\int_{\mathcal{M}_{k}(X;a)}c^{1}(\mathcal{L}_{\Sigma_{1}})\cup\cdots\cup c^{1}(\Sigma_{\ell})$ .

Here we choose ,
$k$
so that $dim\lambda\Lambda_{k}(X, a)$ $=$
$\ell$

. $ 2\ell$
We regard
$Q_{\ell}([\Sigma_{1}], \cdots, [\Sigma_{\ell}])$
as a cochain, an element of $Hom(C_{m}, 0)$ with $m=$
 Floer Homology for Oriented 3-Manifolds

. Under an appropriate assumtion this cochain is a cocycle and its

$\mu(a)$

cohomology class is an invariant of $X$ .

In case $\partial X_{1}=\partial X_{2}=M$ , $X=X_{I}$ , , $II_{M}$ $X_{2}$ $\Sigma_{1}\cdots\Sigma_{\ell_{1}}\subset X_{2}$

, one can prove, under appropriate assumption, that

$\Sigma_{1}\ldots\Sigma_{\ell_{2}}\subset X_{2}$

$Q\ell_{1}+\ell_{2}(\Sigma_{1}, \cdots, \Sigma_{\ell_{1}},\Sigma_{1}, \cdots, \Sigma_{\ell_{2}})$

(1.9)
$=\langle Q_{\ell_{1}}(\Sigma_{1}, \cdots, \Sigma_{\ell_{1}}), Q_{\ell_{2}}(\Sigma_{1}, \cdots, \Sigma_{\ell_{2}})\rangle$
,

where is a coupling between Floer cohomologies of $M$ and $M^{-}$ ,

$\langle , \rangle$

( $M$ with opposite orientation). Note that in case $H_{1}M=0$ , we have

$H_{2}X=H_{2}X_{1}\oplus H_{2}X_{2}$ .
Now we remove the assumption $H_{1}M=0$ . Assume, for example
$H_{1}X_{1}=H_{1}X_{2}=0$ . Then we have Mayer-Vietoris exact sequence:

$H_{2}X_{1}\oplus H_{2}X_{2}\rightarrow H_{2}X\rightarrow H_{1}M\rightarrow 0$ .

Fix a section : $H_{1}M\rightarrow H_{2}X$ . This is equivalent to choose, for each

$s$

$[\gamma]\in H_{1}M$ , surfaces with such that $s([\gamma])=$

$\Sigma_{(i)}(\gamma)\subset X_{i}$ $\partial\Sigma_{i}(\gamma)=\gamma$

. To generalize (1.9) one needs to calculate

$[\Sigma_{(1)}(\gamma)\cup\Sigma_{(2)}(\gamma)]=[\Sigma(\gamma)]$

$Q_{\ell_{1}+\ell_{2}+\ell_{3}}(\Sigma_{1}, \cdots, \Sigma_{\ell_{1}}, \Sigma(\gamma_{1}), \cdots, \Sigma(\gamma\ell_{3}), \Sigma_{1}, \cdots, \Sigma_{\ell_{2}})$

,

in terms of invariants of $X_{1},X_{2}$ . So it is natural to consider cochains

such as
$Q_{\ell+\ell}(\Sigma_{(1)}, \cdots, \Sigma_{\ell}, \Sigma_{(1)}(\gamma_{1}), \cdots, \Sigma_{1}(\gamma_{\ell}))(a)$

$=\int_{\Lambda 4_{k}(X_{1},a)}\mathcal{L}_{\Sigma_{1}}\cup\cdots\cup \mathcal{L}\Sigma_{\ell}\cup \mathcal{L}_{\Sigma_{(1)(\gamma_{1})}}\cup\cdots\cup \mathcal{L}_{\Sigma_{(1)}(\gamma_{l})}$ .

But one finds that this cochain is not a cocycle in general. Hence in our
situation, the relative Donaldson polynomial should not take a value on
usual Floer cohomology but a generalization of it. Our purpose is to
find such a generalization.

We assume that $H_{1}(M;Z)$ is torsion free. Choose a set of closed

loops representing a basis of $H_{1}(M;Z)$ . Put
$\{\gamma_{1}, \cdots, \gamma_{d}\}$ $\Sigma_{i}=\gamma_{i}\times R\subset$

R. Let
$ M\times$
, , $b\in Fl_{0}$ . It induces a connection of a trivial
$a_{t}\in \mathcal{M}(a, b)$ $a$

$SU(2)$ bundle over . Let be the Dirac operator on twisted by

$\Sigma_{i}$ $6_{a_{t}}$ $\Sigma_{i}$

the connection. We may assume that for each $a\in Fl_{0}$ . It $a(\gamma_{i})\neq 1$

implies that is Fredholm. Put $6_{a_{t}}$

$Det6_{a_{t}}=top\wedge(Ker6_{a_{t}})^{*}\otimes top\wedge Coker6_{a_{t}}$

.
6 K. Fukaya

By taking and moving on , we obtain a complex

$(Det6_{a_{t}})^{\otimes 2}$ $a_{t}$ $\mathcal{M}(a, b)$

line bundle on A4 $(a, b)$ , which is denoted by . (The reason why we $\mathcal{L}_{i}^{(2)}$

have to take the square will be explained in \S 7.) Now, let , $b\in Fl_{0}$ with $a$

$\mu(a)-\mu(b)=2\ell+1$ . Put . Then we can define $\overline{\Lambda 4}(a, b)=\mathcal{M}(a, b)/R$

the Chern number

$\int_{\overline{\lambda 4}(a,b)}c^{1}(\mathcal{L}_{i_{1}}^{(2)})\cup\cdots\cup c^{1}(\mathcal{L}_{i_{\ell}}^{(2)})\in Z$ .

This number is denoted by . (Since has a boundary, $\langle\partial_{i_{1},i\ell}\cdots,a, b\rangle$ $\overline{\mathcal{M}}(a, b)$

the above number is, in fact, not well defined. This problem is discussed
in 12.) We define , : by $\partial_{i_{1},i_{\ell}}\cdots$ $C_{k}^{0}\rightarrow C_{k-2\ell-1}^{0}$

$\partial_{i_{1}}$

, , $i_{\ell}([a])=\sum_{b}\langle\partial_{i_{1},ip}\cdots,a, b\rangle[b]$ .

Now we can state the main result of this paper. Let $\alpha\in\{1, \cdots, d\}^{\ell}/S_{\ell}$
.
(Here stands for the symmetric group.) We put
$S_{\ell}$ $\partial_{\alpha}=\partial_{\alpha_{1}}$
, $\cdot,\alpha\ell$
.

Theorem 1.10. If $\#\alpha<3$ , and if $H_{1}(M;Z)$ is torsion free, then

$\sum_{\alpha^{1}\cup\alpha^{2}=\alpha}\partial_{\alpha^{1}}\partial_{\alpha^{2}}=0$
.

Remark 1.11. In case when $\alpha=(1,1)$ the formula is:

$\partial\partial_{1,1}+2\partial_{1}\partial_{1}+\partial_{1,1}\partial=0$
.

Remark 1.12. For $Q\alpha>2$ the formula is $noi$ correct. We discuss

the reason in \S 12. There we also discuss why the formula may not be
correct for $s>0$ if $H_{1}(M;Z)$ has a torsion.

Now let $S^{p}H_{1}(M;Z)$ be the symmetric power. We put

$C_{k}^{s}=\oplus_{s}\ell\leq S^{\ell}H_{1}(M;Z)\otimes C_{k+2\ell}^{0}$

.

Define $\partial_{k}^{s}$
: $C_{k}^{s}\rightarrow C_{k-1}^{s}$
by

$\partial_{k}^{s}(\gamma_{\alpha}\otimes[a])=\sum_{\alpha^{1}\cup\alpha^{2}=\alpha}\gamma_{\alpha^{1}}\otimes\partial_{\alpha^{2}}[a]$
,

where $\gamma_{\alpha}=\gamma_{\alpha_{1}}\otimes\cdots\otimes\gamma_{\alpha_{\ell}}$
. Theorem 1.10 immediately implies
 Floer Homology for Oriented 3-Manifolds 7

Corollary 1.13. Suppose that $H_{1}(M;Z)$ is torsion free. For $s<3$

we have
$\partial_{k-1}^{s}\partial_{k}^{s}=0$
.

We put
$I_{k}^{s}(M)=\frac{Ker\partial_{k}^{s}}{Im\partial_{k-1}^{s}}$
.

Theorem 1.14. Suppose that $H_{1}(M;Z)$ is torsion free. $I_{k}^{s}(M)$

does not depend on the choices of the metrics, s, etc, and is an in- $\gamma_{i}$

variant of , equipped with a basis of

$M$ $H_{1}(M;Z)$ .

By construction we have an exact sequence of complexes

$0\rightarrow C_{k}^{s}\rightarrow C_{k}^{s+1}\rightarrow S^{s+1}(H_{1}(M;Z))\otimes C_{k+2s+2}^{0}\rightarrow 0$

It follows that:
Theorem 1.15. Suppose that $H_{1}(M;Z)$ is torsion free. There
exists a long exact sequence
$\rightarrow I_{k}^{s}(M)\rightarrow I_{k}^{s+1}(M)\rightarrow S^{s+1}(H_{1}(M, Z))\otimes I_{k+2s+2}^{0}(M)\rightarrow$

for $s=0$ or 1. The exact sequence is also an invariant of $M$ .

The proof of these theorems is based on the detailed analysis of the
end of the moduli space . The results on it is in \S 7. In fact, we
$\Lambda 4(a, b)$

shall prove more general results than we need to construct our invariants.
In the course, we develop various techniques, which might be useful in
other situations.
Using our invariant , we can partially generalize the definition
$I_{k}^{s}(M)$

of relative Donaldson polynomial to the case when the boundary is not

necessary a homology sphere. Those applications will appear elsewhere.
The organization of this paper is as follows.
In 2,3, we perturb the equation.
In \S 4, we review the sum formula for the index of the elliptic oper-
ators. We also discuss the sum formula of the family of indices.
This result is used in \S 5 to define the degree . In \S 5 we study also $\mu$

neighborhoods of various reducible connections.

In \S 6 we define the orientation of the moduli space. The fact that
every oriented 3-manifolds bounds an oriented 4-manifold, is essentially
used in the proof.
7-11 are devoted to the study of the end of moduli space . $\mathcal{M}(a, b)$

The results of these sections are stated in \S 7.

 K. Fukaya

In \S 8, we prove that the patching procedure of selfdual connections

as in [T1] is possible in our situation, where various reducible connections
must be dealt with.
In \S 9, we shall prove that the selfdual connections constructed in \S 8,
contains all the connections in the end of the moduli space, except the
concentrated ones. For this purpose, we establish a decay estimate such
as in [FU].
Combining the results of \S \S 8,9 we obtain a chart for a neighborhood
of each point at infinity. In order to patch those charts, we introduce,
in \S 10, the local action of the groups. This notion is a generalization of
one introduced in [CG] to study the end of Riemannian manifolds. We
use it to study the end of the moduli space.
The line bundle is constructed and is extended to the boundary $\mathcal{L}_{i}^{(2)}$

in \S 11. For this purpose we use the sum theorem for index bundles in
\S 4 and the existence of the lift of the local action to the bundle.
Using the results of 7-11, we define the boundary operator in \S 12
and prove Theorem 1.10. As is remarked before, the Chern number
of the bundle is not well defined. We shall prove in \S 12 that the
$\mathcal{L}_{i}^{(2)}$

boundary operator is well defined modulo isomorphism. In \S 12, we also

discuss the case when $s=3$ and describe why Theorem 1.10 does not
hold in that case.
Finally we shall prove Theorems 1.14 and 1.15 in \S 13.
As the reader can find easily, this paper heavily depends on the
brilliant ideas due to Donaldson, Floer, Taubes e.t.c. in their papers.
Before this work is completed the author is informed (without the precise
statement) that A. Floer generalized his invariant to homology . $S^{1}\times S^{2}$

\S 2. Perturbation
Let be the Sobolev space of the sections, namely the set of sec-
$L_{\ell}^{p}$

tions norms of whose -th derivatives are finite. Put

$L^{p_{-}}$ $\ell$

$A_{\ell}^{p}(M)=\{d+a|a\in L_{\ell}^{p}(M, \wedge^{1}\otimes su(2))\}$

$\mathcal{G}\ell(M)=the$ set of maps : $M\rightarrow SU(2)$ of $L_{\ell}^{2}$

-class.

is denoted by
$A_{\ell}^{2}$
. We choose suciently large $A_{\ell}$ $\ell$

and fix it throughout

this paper. acts on . (See [FU].) Put$\mathcal{G}_{\ell+1}$ $A_{\ell}$

$B_{\ell}(M)=A_{\ell}(M)/\mathcal{G}_{\ell+1}(M)$ .

Let $a\in A_{\ell}(M)$ . Then the set

(2.1) $\{u\in L_{\ell}^{2}(M, \wedge^{1}\otimes su(2))|d_{a}^{*}u=0\}$

 Floer Homology for Oriented 3-Manifolds 9

is the orthonormal complement of in TaAi(M). In the case $T_{a}\mathcal{G}_{\ell+1}a$

when is irreducible, the set (2.1) can be identified to $T_{[a]}B_{\ell}(M)$ . (See

$a$

[FU].) We let the set (2.1) be denoted by $T_{[a]}B_{\ell}(M)$ also in the case
when is reducible. In that case,
$a$ is a singular point of $B_{\ell}(M)$ . $[a]$

The purpose of this section is to perturb the functional and the $c\epsilon$

equation (1.2), so that (1.4) has only a finite number of solutions each
of which is nondegenerate. We put

$H_{1}(M;Z)=\frac{H_{1}(M,Z)}{Torsion}.$
.

First we deal with singular points on

$Hom(H_{1}(M;Z), SU(2))/conjugate$ $\subset B_{\ell}(M)$ .

Choose a set of loops representing a basis of $H_{1}(M;Z)$ . $\{\ell_{1}^{0}, \cdots, \ell_{d}^{0}\}$

Extend . to an embedding : $\ell_{?}^{0}$

. Choose a nonnegative $\ell_{i}^{0}$

$S^{1}\times D^{2}\rightarrow M$

function on with compact support such that

$u$
$D^{2}$

$\int_{D^{2}}u(x)dx=1$ .

For a loop : $S^{1}\rightarrow M$ and $a\in A(M)$ , let $h_{\ell}(a)\in

$\ell$
SU(2)$ be the
holonomy along . Define a functional on $B_{\ell}(M)$ by
$\ell$
$f_{0}$

(2.2) $f_{0}(a)=\epsilon\sum_{i=1}^{d}\int Tr(h_{\ell_{i}^{O}(\cdot,x)}(a))u(x)dx$ ,

where is a small positive number. Then by [F] $1b.1$ , $grad_{a}f_{o}\in$

$\epsilon$

TaBi(M) is well defined. Similarly we can define the hessian, $Hess_{a}f_{0}$ :

$T_{[a]}B_{\ell}(M)\rightarrow T_{[a]}B_{\ell}(M)$ .
Here we examine the set, $FR$ , of the flat reducible connections
in $B_{\ell}(M)$ . The set of the conjugacy classes of the elements of
$Hom(TorH_{1}(M, Z)$ , $U(1))$ has a one to one correspondence to $\pi_{0}(FR)$ .
For $\varphi\in Hom(Tor H_{1}(M, Z)$ , $U(1))$ , let be the corresponding com- $FR_{\varphi}$

ponent. is dieomorphic to
$FR_{\varphi}$
if , and is dieomor- $T^{d}$
$Im(\varphi)\not\subset\{\pm 1\}$

phic to if $T^{d}/Z_{2}$
$Im(\varphi)\subset\{\pm 1\}$ . Let $1\in Hom(Tor H_{1}(M, Z)$ , Z) be the
trivial representation.

Lemma 2.3. There exists a neighborhood $U$ of $FR_{1}$ such that, for
suciently small , the set of elements of $U$ satisfying $\epsilon$

(2.4) $*_{\sigma}F^{a}-grad_{a}f_{0}=0$
10 K. Fukaya

is identifified to $Hom(H_{1}(M, Z)$ , $Z_{2})\simeq\{\pm 1\}^{d}$

.

Proof. By identifying $FR_{1}=\{(e^{i\theta_{1}}, \cdots, e^{i\theta_{d}})\}/Z_{2}$ , we have

(2.5) $f_{0}(e^{i\theta_{1}}, \cdots, e^{\theta_{d}})=2\epsilon\sum\cos\theta_{i}$

.

The lemma follows immediately.

Lemma 2.6. Let $a\in Hom(H_{1}(M, Z)$ , $Z_{2})$ . Then $c\epsilon-f_{0}$

is non-
degenerate at . In other words $a$

$*_{\sigma}d_{a}-Hess_{a}f_{0}$ : $T_{[a]}B_{\ell}(M)\rightarrow T_{[a]}B_{\ell-1}(M)$

is invertible.

Remark 2.7. $Hess_{a}$ $c\epsilon=*_{\sigma}d_{a}$

. See $[F],[T4]$ .

Proof. We have

$Ker*_{\sigma}d_{a}\simeq H^{1}(M;R)\otimes su(2)\simeq su(2)^{d}$ .

On this space $Hess_{a}f_{0}$ is given by $-\epsilon\sum x_{i}^{2}$

. Hence the lemma follows
from the invertibility of the matrix

$\left(\begin{array}{ll}A+\epsilon E & \epsilon B\\\epsilon C & \epsilon D\end{array}\right)$

for small $\epsilon$

and invertible $A$

and $D$ .

We take in (2.2) such that Lemma 2.6 holds and fix it.
$\epsilon$

Next we use a method similar to [D3] and [F]. Let $p_{0}\in M$ and
$v_{0}\in T_{p0}M$ . Choose an embedding : , such that $I(0)=p_{0}$ , $I$ $D^{2}\rightarrow M$

and that is transversal to . Let

$I_{*}(T_{0}D^{2})$ be the set of $v_{0}$ $\Gamma_{1}(p_{0}, I, v)$

smooth embeddings such that $\ell(1,0)=p_{0}$ , , $\ell(0, x)=I(x)$ . $\frac{D\ell}{dt}(1, 0)=v_{0}$

We put
$\Gamma_{m}=(p_{0},v_{O},I)\cup(\Gamma_{1}(p_{0}, v_{0}, I))^{m}$
.

Let $L_{m}=SU(2)^{m}/SU(2)$ , where $SU(2)$ acts by conjugation. Define a

map
: $A_{\ell}(M)\times\Gamma_{m}\rightarrow Map(D^{2}, SU(2)^{m})$
$\overline{\Phi}$

by
$\overline{\Phi}(a, (\ell_{1}, \cdots, \ell_{m}))(x)=(h_{\ell_{1}(\cdot,x)}(a), \cdots, h_{\ell_{m}(\cdot,x)}(a))$
.
 Floer Homology for Oriented 3-Manifolds 11

$\overline{\Phi}$

induces a map
$\Phi$
: $B_{\ell}(M)\times\Gamma_{m}\rightarrow Map(D^{2}, L_{m})$ .

Following [F], we choose $(\beta_{i})_{i\in}z_{+}(\beta_{i}>0)$ . and put

$C^{\beta}(L_{m}, R)=\{\psi\in C^{\infty}(L_{m}, R)|||\psi||_{\beta}<\infty\}$ ,

where
$||\psi||_{\beta}=\sum_{i=1}^{\infty}\beta_{i}\max_{x\in L_{m}}|D^{i}\psi(x)|$ .

Fix a function $u:D^{2}\rightarrow[0, \infty)$ as before and define

$\Phi$
: $B_{\ell}(M)\times\Gamma_{m}\times C^{\beta}(L_{m}, R)\rightarrow R$

by

$\Phi([a], (\ell_{1}, \cdots, \ell_{m}), \psi)=\int_{D^{2}}\psi(\Phi([a], (\ell_{1}, \cdots, \ell_{m})(x))u(x)dx$ .

For $v\in\Gamma_{m}\times C^{\beta}(L_{m}, R)$

, we put $f_{v}([a])=\Phi([a], v)$ . For $\lambda=(\ell_{1}, \cdots, \ell_{m})$

$\in L_{m}$ and $\lambda=(\ell_{1}, \cdots, \ell_{m})\in L_{m}$ , we say $\lambda$ $\prec\lambda$
if $\{\ell_{1}, \cdots, \ell_{m}\}$

$\subset\{\ell_{1}, \cdots, \ell_{m}\}$

Lemma 2.8. There exists and $\delta>0$ such that for each $\lambda_{0}\in\Gamma_{m_{O}}$

, the set of $\psi\in C^{\beta}((L_{m}), R)$ satisfying the following conditions

$\lambda_{0}\prec\lambda$

is of fifirst category in . $\{\psi|||\psi||_{\beta}<\delta\}$

(2.8.1) The set $Fl(\psi)$

of the solution of
$*_{\sigma}F^{a}=grad_{a}(f_{0}+f_{(\lambda,\psi)})$ .

is fifinite.
(2.8.2) For each $a\in Fl(\psi)$ the map
$*_{\sigma}d-Hess_{[a]}(f_{0}+f_{(\lambda,\psi)})$ : $T_{a}B_{\ell}(M)\rightarrow T_{a}B_{\ell-1}(M)$

is invertible.

As is well known, (2.8.2) implies (2.8.1). Hence the prob-

Proof.
lem is local on $B_{\ell}(M)$ . The argument in a neighborhood of irreducible
connections is the same as [F] $2c.1$ . Then we study the neighborhood of
the set of reducible connections. Precisely, we first take a perturbation
so that (2.8.2) holds in a neighborhood of the set of the reducible con-
nections, next we perturb again so that (2.8.1) and (2.8.2) holds, in the
set of irreducible connections, as well.
12 K. Fukaya

Let $\varphi\in Hom(TorH_{1}(M, Z)$ , $SU(2))$ . In the case when , $Im\varphi\subset\{\pm 1\}$

the proof of Lemma 2.6 works in a neighborhood of . Then we $FR_{\varphi}$

assume that . By the proof of Lemma 2.3,

$Im(\varphi)\not\subset\{\pm 1\}$ is a Morse $f_{0}$

function on and has exactly

$Fl_{\varphi}$
singular points on it. The same $2^{d}$

holds for if is small. Hence it suces to work at a

$f_{0}+f_{\lambda,\psi}$ $||\psi||_{\beta}$

neighborhood of each singular point . Choose a neighborhood $U$ of $a_{0}$ $a_{0}$

with is of bounded norm. $L_{\ell}^{2}$

Sublemma 2.9. The set of such $that*_{\sigma}d_{a}-Hess_{a}(f_{0}+f_{\lambda,\psi})$ $\psi$

is invertible for each a $\in U\cap Fl(\psi)$ , is open.

Proof First we remark that the set

$Fl(\psi)=\{[a]\in B_{\ell}(M)|*_{\sigma}F^{a}=grad_{a}(f_{0}+f_{\lambda+\psi})\}$

is independent of because the equation is elliptic modulo gauge trans-

$\ell$

formation. Hence we can find a bounded subset in $L$ $L_{\ell+2}^{2}(M, \wedge^{1}\otimes su(2))$

such that if

(2.10.1) $||\psi-\psi||_{\beta}<\delta$

(2.10.2) $[a]\in Fl(\psi)$

(2.10.3) $[a]\in U$

then $[a]=[a_{0}+u]$ for some $u\in L$ . Now, if the sublemma is false, then,
there exists and such that
$\psi,\psi_{i}$
$a_{i}$

(2.11.1) $\lim_{i\rightarrow\infty}||\psi_{i}-\psi||_{\beta}=0$ ,
(2.11.2) $[a_{i}]\in Fl(\psi_{i})$ ,
(2.11.3) $[a_{i}]\in U$ ,
(2.11.4) $*_{\sigma}d_{a_{i}}-Hess_{a_{i}}(f_{0}+f_{\lambda,\psi_{i}})$ is not invertible,
(2.11.5) $*_{\sigma}d_{a}-Hess_{a}(f_{0}+f_{\lambda,\psi})$ is invertible for each $a\in Fl(\psi)\cap U$ .

We can choose $u_{i}\in L$ such that $[a_{0}+u_{i}]=[a_{i}]$ . By Rellichs

Theorem, we can find a subsequence such that converges to in $u_{i}$ $u_{\infty}$

. Hence by (2.11.1),(2.11.2) and (2.11.3), we have $[a_{0}+u_{\infty}]=$

$L_{\ell+1}^{2}$

$[a_{\infty}]\in U\cap Fl(\psi)$ . Therefore is invertible. $*_{\sigma}d_{a_{\infty}}-Hess_{a_{\infty}}(f_{0}+f_{\lambda,\psi})$

On the other hand, we remark that the map

$A_{\ell+1}(M)\times L_{\ell}^{2}(M, \wedge^{1}\otimes su(2))\rightarrow L_{\ell-1}^{2}(M, \wedge^{1}\otimes su2))$

: $(a, u)\mapsto*_{\sigma}d_{a}u-Hess_{a}(f_{0}+f_{\lambda},\psi)u$

is continuous. (See [FU]). It follows that is $*_{\sigma}d_{a_{i}}-Hess_{a_{i}}(f_{0}+f_{\lambda,\psi_{i}})$

invertible for suciently large . This contradicts (2.11.4). The proof of $i$

Sublemma 2.9 is now complete.

 Floer Homology for Oriented 3-Manifolds 13

Hence it suces to show that the set of $\psi$

for which

$*_{\sigma}d_{a_{0}}-Hess_{a_{O}}(f_{0}+f_{(\lambda,\psi)})$

is surjective, is dense. We can choose a loop $\ell_{0}$

so that $\varphi(\ell_{0})\not\in\{\pm 1\}$

and assume . Put$\{\ell_{0}\}\prec\lambda=(\ell_{1}, \cdots, \ell_{m})$

$\overline{\Phi}(a_{0}, \lambda)(0)=(g_{1}, \cdots, g_{m})$

.

We have

(2.12) $\{g\in SU(2)|g^{-1}(g_{1}, \cdots, g_{m})g=(g_{1}. \cdots, g_{m})\}\simeq U(1)$

Hence $[g_{1}, \cdots, g_{m}]$

is contained in $U(1)^{m}/Z_{2}\subset SU(2)^{m}/SU(2)$ and is a
regular point of $U(1)^{m}/Z_{2}$ . Put

$B_{\ell}^{red}(M)=$
{ $[a]\in B_{\ell}^{red}(M)|$ $a$ is reducible.}
It follows from (2.12) that is a regular point of $[a_{0}]$ $B_{\ell}^{red}(M)$ . Therefore,
by a $U(1)$ analogue of [F] $2c.1$ , we may assume that

(2.13) $*_{\sigma}d_{a_{O}}-Hess_{a_{O}}(f_{0}+f_{\lambda},\psi)$ : $T_{[a_{0}]}(B_{\ell}^{red}(M))\rightarrow T_{[ao]}(B_{\ell}^{red}(M))$

is invertible. Put

$K_{\psi}=\{u\in T_{[ao]}B_{\ell}(M)|*_{\sigma}d_{a_{0}}u-Hess_{a_{O}}(f_{0}+f_{\lambda,\psi})u=0\}$

By the invertibility of (2.13) we have

(2.14) $K_{\psi}\cap T_{[a_{O}]}B_{\ell}^{red}(M)=\{0\}$ .

The group

(2.15) $U(1)=\{g\in \mathcal{G}_{\ell}(M)|g^{*}a_{0}=a_{0}\}$

acts on . By (2.14) and the finite dimensionality of

$K_{\psi}$ $K_{\psi}$
, we can
identify $K\psi\simeq C^{k}$ . Therefore by taking suciently large $\lambda$

and $m$ we
may assume that

$P:K_{\psi}\rightarrow T_{(g_{1},\cdot\cdot,g_{m})}SU(2)^{m}$

is injective, where $P$ is the dierential at of the map : $[a]\mapsto$ $[a_{0}]$

: $A_{\ell}(M)$ $SU(2)^{m}$ .
$\overline{\Psi}(a, \lambda)(0)$
By (2.8), $U(1)$ acts on $\rightarrow$

, ,
$T_{(g_{1}}$
, which we can identify to $C^{m}\oplus R^{m}$ . The map $P$
$g_{\mathfrak{m}})SU(2)^{m}$
14 K. Fukaya

is $U(1)$ invariant. Hence we may assume that $P(K_{\psi})\subset C^{m}$ . We define

a function in a neighborhood of
$\psi$ $(g_{1}, \cdots, g_{m})$ by

(2.16) $\psi(\exp_{(g_{1},,g_{m})}\cdots(z_{1}, \cdots, z_{m}, t_{1}, \cdots, t_{m}))=-\sum|z_{i}|^{2}$ ,

and extend it to a $SU(2)$ invariant function on $SU(2)^{m}$ . We obtain a

function on , for which we use the same symbol. Now it is easy to
$L_{m}$

see that
$*_{\sigma}d_{a_{O}}-Hess_{a_{0}}(f_{0}+f_{\lambda,\psi+\epsilon\psi})$

is invertible for each suciently small . The proof of Lemma 2.7 is now $\epsilon$

completed.

Note that a linear function is used in [F] for the perturbation in a

neighborhood of an irreducible connection. Here we use quadratic func-
tion to perturb the equation in a neighborhood of a reducible connection.

Remark 2.17. We choose the perturbation so that the zero eigen-

values $of*_{\sigma}d-Hess_{a}(f_{0}+f_{(\lambda,\mu)})$ is perturbed to positive one, if is $a$

a reducible connection and if the corresponding eigenspace is identified

to with respect to the $U(1)$ action. The set of such connections is a
$C^{k}$

subset of first category in an open set. This choice is used in the proof
of Theorem 5.6. (See Remark 5.7.)

Now we put $f=f_{0}+f_{\lambda,\psi}$ $fr$ generic $\psi$

, and define $Fl$ and $Fl_{0}$
by
(1.6.1) and (1.6.2).

\S 3. Local structure of moduli space

Let : $M\times R\rightarrow M$ be the projection,

$p$ be the pull back $p^{*}(\wedge^{i}M)$

of the vector bundles on R. Let be a number suciently close

$ M\times$ $\delta$

to 0. Choose a -map : $C^{\infty}$

, such that $||t||=|t|$ outside a
$||||$ $R\rightarrow[0, \infty)$

compact subset, put . For a smooth section of

$e_{\delta}(t)=e^{\delta||t||}$ $u$ $ p^{*}(\wedge^{i}M)\otimes$

$su(2)$ with compact support, we put

$(||u||_{\ell,\delta}^{p})^{p}=\sum_{k\leq\ell}\int_{M\times R}e_{\delta}(t)|\nabla^{k}u|^{p}dxdt$ .

Let $L_{\ell,\delta}^{p}(M\times R, su(2)\otimes p^{*}(\wedge^{i}M))$ be the completion with respect to

this norm. We put

$\mathcal{L}_{\ell,\delta}^{i}=L_{\ell,\delta}^{2}(M\times R, su(2)\otimes p^{*}(\wedge^{i}M))$ .

 Floer Homology for Oriented 3-Manifolds 15

Define $L_{\ell,\delta}^{p}(M\times R, su(2)\otimes\wedge^{i}(M\times R))$ in a similar way. Let $ L_{p\delta}^{p},(M\times$

,
$R$ be the subspace of
$su(2)\otimes\bigwedge_{\pm}^{2}(M\times R))$ $L_{\ell,\delta}^{p}(M\times R, su(2)\otimes\wedge^{2}(M\times R))$

consisting of the elements satisfying , respectively. Here and $u$ $\overline{*}_{\sigma}u=\pm u$

hereafter denotes the Hodge *operator on $M\times R$ with respect to

$\overline{*}_{\sigma}$

the product metric . The Hodge operator on $M$ induces : $\sigma\oplus dt^{2}$
$*_{\sigma}$

. We define isomorphisms
$p^{*}(\wedge^{k}M)\rightarrow p^{*}(\wedge^{3-k}M)$

$I_{\pm}^{2}$
: $ L_{\ell,\delta}^{p}(M\times R, su(2)\otimes p^{*}(\wedge^{1}M))\rightarrow$

$L_{\ell,\delta}^{p}(M\times R, su(2)\otimes\bigwedge_{\pm}^{2}(M\times R))$

$I^{1}$
: $ L_{\ell,\delta}^{p}(M\times R, su(2)\otimes p^{*}(\wedge^{0}M\oplus\wedge^{1}M))\rightarrow$

$L_{\ell,\delta}^{p}(M\times R, su(2)\otimes\wedge^{1}(M\times R))$

$I^{0}$
: $L_{\ell,\delta}^{p}(M\times R, su(2))\rightarrow L_{\ell,\delta}^{p}(M\times R, su(2))$

by
$I_{\pm}^{2}(\alpha)=\alpha\pm(*_{\sigma}\alpha)\wedge dt$

$ I^{1}(\varphi, \alpha)=\varphi dt+\alpha$

$I^{0}=identify$ .

We put
$\Omega_{\ell,\delta}^{0}=L_{\ell,\delta}^{2}(M\times R, su(2))$

$\Omega_{\ell,\delta}^{1}=L_{\ell,\delta}^{p}(M\times R, su(2)\otimes\wedge^{1}(M\times R))$

$\Omega_{\ell,\delta}^{2}=L_{\ell,\delta}^{2}(M\times R, su(2)\otimes\bigwedge_{-}^{2}(M\times R))$

and identify , , , by . $\mathcal{L}_{\ell,\delta}^{0}\simeq\Omega_{\ell,\delta}^{0}$ $\mathcal{L}_{\ell,\delta}^{0}\oplus \mathcal{L}_{\ell,\delta}^{1}\simeq\Omega_{\ell,\delta}^{1}$ $\mathcal{L}_{p\delta}^{1},\simeq\Omega_{\ell,\delta}^{2}$

$I^{i}$

For , $b\in Fl$ , choose a connection $d+A^{a,b}$ of the trivial $SU(2)$

$a$

bundle on $M\times R$ such that $A^{a,b}=b$ if $t>1$ and that $A^{a,b}=a$ if

$t<-1$ . We put

$A_{\ell,\delta}(a, b)=\{d+A^{a,b}+\alpha|\alpha\in\Omega_{\ell,\delta}^{1}\}$ .

Clearly this space is independent of the choice of . Hereafter we $A^{a,b}$

write $A$ in place of $d+A$ . Let be the set of all locally $\mathcal{G}_{\ell,\delta}^{0}(M\times R)$

map : $M\times R\rightarrow SU(2)$ such that there exists

$L_{\ell}^{2}$
$g$ satisfying $\psi\in \mathcal{L}_{\ell,\delta}$

$\exp\psi=g$ outside a compact subset.

Lemma 3.1. $\mathcal{G}_{\ell+1,\delta}^{0}(M\times R)$ acts on $A_{\ell,\delta}(a, b)$ by

$g^{*}A=g^{-1}dg+g^{-1}Ag$ .
16 K. Fukaya

The action is free if $\delta$

is positive or , $a$ $b\in Fl_{0}$ .

We omit the proof. (See $[FU],[T3],[F].$ )

For $a\in A_{\ell}(M)$ , $A\in A_{\ell,\delta}(a, b)$ , we put

$G_{a}=\{g\in \mathcal{G}_{\ell+1}(M)|g^{*}a=a\}$

$G_{A}=$ { $g:M\times R\rightarrow G|g$ is a locally $L_{\ell+1}^{2}$

map satisfying $g^{*}A=A.$ }
Remark 3.2. $G_{A}\subset G_{a}\cap G_{b}$ .

Put

$B_{\ell,\delta}^{reg}(a, b)=\{[A]|A\in A_{\ell,\delta}(a, b), G_{A}\neq\{\pm 1\}\}$

$T_{[A]}B_{\ell,\delta}(a, b)=\{\alpha\in\Omega_{\ell,\delta}^{1}|e_{\delta}d_{A}^{*}e_{\delta}^{-1}\alpha=0\}$
.

$G_{A}$ acts on $B_{\ell,\delta}(a, b)$ and $T_{[A]}B_{\ell}(a, b)$ .

Lemma 3.3. The map : $\alpha\mapsto[A+\alpha]$ ,

$T_{[A]}B_{\ell,\delta}(a, b)\rightarrow B_{\ell,\delta}(a, b)$

induces a -invariant dieomorphism from a neighborhood of 0 onto

$G_{A}$ $a$

neighborhood of $A$ , if , $b\in Fl_{0}$ , or if $\delta>0$ . $a$

The proof is in [FU], [T3], [F].

Lemma 3.4. acts on $G_{a}\times G_{b}$ $B_{\ell,\delta}(a, b)$ . The action is compatible
with the diagonal inclusion : $G_{A}\rightarrow G_{a}\times G_{b}$ .

Proof. For each and choose a map :

$g_{1}\in G_{a}$ $g_{2}\in G_{b}$ $g$
$ M\times R\rightarrow$

$SU(2)$ such that $g_{t}=g_{1}$ if $<-1$ and that $g_{t}=g_{2}$ if $t>1$ . For$t$

$[A]\in B_{\ell,\delta}(a, b)$ the element is contained in , and $g^{*}A$ $A_{\ell,\delta}(a, b)$ $[g^{*}A]$

depends only on $[A]$ and , . Clearly this induces a desired action. $g_{1}$ $g_{2}$

Hereafter we put

$g_{1}[A]g_{2}^{-1}=(g_{1}, g_{2})[A]$

for $A\in B_{\ell,\delta}(a, b)$ , , $g_{2}\in G_{b}$ . Then

$g_{1}\in G_{a}$ $G_{a}$
and $G_{b}$
act from left and
right on $B_{\ell,\delta}(a, b)$ , respectively.

Remark 3.5. The action is trivial if $\delta<0$ .

Now we consider a dierential equation

(3.6) $F^{A}-\overline{*}_{\sigma}F^{A}-grad_{a_{t}}f\wedge dt+*_{\sigma}grad_{a_{t}}f=0$ ,

 Floer Homology for Oriented 3-Manifolds 17

for . Here we put

$A\in A_{\ell,\delta}(a, b)$ $A=I^{1}(a_{t}, \varphi)$ . Letbe the set $\overline{\mathcal{M}}_{\ell,\delta}(a, b)$

of all solutions of (3.6) in $A_{\ell,\delta}(a, b)$ . Since $grad_{g_{t}^{*}a_{t}}f=g_{t}^{-1}(grad_{a_{t}}f)g_{t}$ ,

it follows that
$F^{g^{*}A}-\overline{*}_{\sigma}F^{g^{*}A}-grad_{g_{t}^{*}a_{t}}f\wedge dt+*_{\sigma}grad_{g_{t}^{*}a_{t}}f=$

$g^{-1}(F^{A}-\overline{*}_{\sigma}F^{A}-grad_{a_{t}}f\wedge dt+*_{\sigma}grad_{a_{t}}f)g$ .

Therefore $\overline{M}_{\ell,\delta}(a, b)$

is $\mathcal{G}_{\ell+1,\delta}^{0}$
invariant. We put

$\mathcal{M}_{\ell,\delta}(a, b)=\overline{M}_{\ell,\delta}(a, b)/\mathcal{G}_{\ell+1,\delta}^{0}$

.

By a standard elliptic regularity estimate, is independent of $\mathcal{M}_{\ell,\delta}(a, b)$

. Then we omit and write

$\ell$

. $\ell$
$\lambda\Lambda_{\delta}(a, b)$

Here we remark that the set is identified to the set $G_{a}\backslash \mathcal{M}_{\delta}(a, b)/G_{b}$

in \S 1. In fact, the elements of the set

$\mathcal{M}(a, b)$ have a one to one $\Lambda 4(a, b)$

correspondence to the set of s satisfying (1.7) and , $a_{t}$

$\lim_{t\rightarrow-\infty}a_{t}=a$

. Put
$\lim_{t\rightarrow\infty}[a_{t}]=[b]$ . There exists such that $\lim_{t\rightarrow\infty}a_{t}=b$ $g_{\infty}$

. Choose
$g_{\infty}^{*}b=b$ such that , . It is
$g_{t}$
$\lim_{t\rightarrow-\infty}g_{t}=1$ $\lim_{t\rightarrow\infty}g_{t}=g_{\infty}$

easy to see that . This element depends only on

$g^{*}(d+a_{t})\in \mathcal{M}_{\delta}(a, b)$ $[a_{t}]$

and is independent of . Conversely, if , we can find such $a_{t}$

$A\in\overline{\lambda\Lambda}_{\delta}(a, b)$
$g$

that has no factor. Let

$g^{*}A$ $(g^{*}A)(\cdot, t)=a_{t}$ . Then $dt$
. $[a_{t}]\in \mathcal{M}(a, b)$

Remark 3.7. It is not in general true that the set of loops joining
and
$[a]$ in $B_{\ell}(M)$ has one to one correspondence to
$[b]$ . This is $B_{\ell,\delta}(a, b)$

valid if the loop is contained in $B_{\ell}(M)-SB_{\ell}(M)$

For $A\in A_{\ell}(a, b)$ , we define $D_{A}$ : $\Omega_{\ell}^{1}\rightarrow\Omega_{\ell-1}^{2}$

by

$D_{A}\alpha=(d_{A}-\overline{*}_{\sigma}d_{A})\alpha-Hess_{a_{t}}f(u_{t})$ ,

where $\alpha=I_{1}(u_{t}, \varphi)$ , $d+A=d+a_{t}+\psi dt$ . If we identify $\Omega_{\ell,\delta}^{1}\simeq \mathcal{L}_{\ell,\delta}^{1}\oplus \mathcal{L}_{\ell,\delta}^{0}$

,
$\Omega_{\ell-1,\delta}^{2}\simeq \mathcal{L}_{\ell-1,\delta}^{1}$
, we have

(3.8) $ D_{A}(u, \varphi)=-\frac{\partial u}{\partial t}+(*_{\sigma}d_{a_{t}}-Hess_{a_{t}}f-\psi_{t}\wedge)u+d_{a_{t}}\varphi$

.

Recall that $\mathcal{M}(a, b)$ is a $C^{\infty}$

-manifold in a neighborhood of $[A]$ if $D_{A}$ is
surjective.
Lemma 3.9. There exists and such that, for each , $\lambda_{0}$
$m_{0}$
$\lambda_{0}\prec\lambda$

the set of $\psi\in C^{\beta}(L_{m}, R)$ satisfying the following is of fifirst category in
an open set. Let , $b\in Fl$ , . $a$ $f=f_{\lambda,\psi}$

(3.9.1) $\Lambda 4_{\delta}(a, b)$

is $a$
fifinite dimensional smooth manifold.
18 K. Fukaya

(3.9.2) For each $[A]\in \mathcal{M}_{\delta}(a, b)$ , $D_{A}$ is surjective.

We write
Proof. , while proving Lemma 3.9. In the $\mathcal{M}_{\delta}^{\psi}(a, b)$ $D_{A}^{\psi}$

set of irreducible connections, the proof of [F] $2c.2$ works. Hence we

study in the neighborhood of reducible connections. Put
$\mathcal{M}_{\delta}^{\psi}(a, b)$

$B_{\ell,\delta}^{red}(a, b)=\{[A]\in B_{\ell,\delta}(a, b)|G_{A}=U(1)\}$

$\mathcal{M}_{\delta}^{red,\psi}(a, b)=B_{\ell,\delta}^{red}(a, b)\cap \mathcal{M}_{\delta}^{\psi}(a, b)$

Then by a $U(1)$ analogue of the argument by Floer [F] $2c.2$ , we may

assume that is a -manifold, and, for each $[A]$
$\mathcal{M}_{\delta}^{red,\psi}(a, b)$ $C^{\infty}$ $\in$

, the map
$\mathcal{M}_{\ell,\delta}^{red,\psi}(a, b)$

$D_{A}^{red}$
: $ L_{\ell,\delta}^{2}(M\times R,u(1)\otimes\wedge^{1}(M\times R))\rightarrow$

$L_{\ell-1,\delta}^{2}(M\times R, u(1)\otimes\bigwedge_{-}^{2}(M\times R))$

is surjective. Let $[A]\in \mathcal{M}_{\delta}^{red,\psi}$

. Choose a neighborhood $U$ of $[A]$ in
, which is bounded in
$B_{\delta,\ell}^{\psi}(a, b)$ $L_{\ell}^{2}$
norm.

Sublemma 3.10. The set of all $\psi$

such that $D_{A}^{\psi}$
is surjective for
all A $\in U\cap \mathcal{M}_{\delta}^{\psi}(a, $
b), is open.

The proof is similar to one for Sublemma 2.9 and is omited.

Sublemma 3.11. For each $\epsilon>0$

and , there exists
$\psi$
and $\psi$ $a$

neighborhood $U$ of $A$ , such that $||\psi||_{\beta}<\epsilon$

and that , is surjective
$D_{A}^{\psi+\psi}$

for each $[A]\in U\cap\lambda\Lambda_{\delta}^{\psi+\psi}(a, b)$

.

By an argument similar to the proof of Sublemma 2.9, it

Proof.
suces to find such that , and that is surjective. We
$\psi$ $||\psi||_{\beta}<\epsilon$
$D_{A}^{\psi+\psi}$

put

$Cok=Ker(D_{A}^{\psi})^{*}\subset \mathcal{L}_{\ell,\delta}^{1}$
,
$Ker=\{u\in \mathcal{L}_{\ell,\delta}^{1}|D_{A}u=0, d_{a_{t}}^{*}u_{t}=0\}$

The group $U(1)\simeq G_{A}$ acts on $Ker$ and $Cok$ . By the surjectivity of
, we have $Cok\simeq C^{k}$ as $U(1)$ module. By the index calculation in
$D_{A}^{\psi,red}$

\S 5, we can find a $U(1)$ invariant subspace $K$ of $Ker$ which is isomorphic

to as $U(1)$ module. (See Remark 5.7.) Choose an isomorphism
$C^{k}$

$Q$ : $Cok\rightarrow K$
. For each , let , be the projection of $t$ $K_{t}$ $Cok_{t}\subset T_{[a_{t}]}B_{\ell}(M)$
 Floer Homology for Oriented 3-Manifolds 19

$K$ and By the unique continuation theorem ([Ar]), the projections

$Cok$ .

, $Cok\rightarrow Cok_{t}$ are isomorphisms. Let

$K\rightarrow K_{t}$ : be the $Q_{t}$ $Cok_{t}\rightarrow K_{t}$

projection of $Q$ . We can choose suciently large $m$ and such that $\lambda$

the curve is injective, and :

$t\mapsto\overline{\Psi}(a_{t}, \lambda)(0)=a_{t}$ $P_{t}$ $ T_{[a_{t}]}(\mathcal{B}_{\ell}(M))\rightarrow$

$T_{a_{t}}SU(2)^{m}$ is injective on $K_{t}+Cokt$ for each . Since the action of $t$

$U(1)$ has no trivial component on Cokt, it follows that $P_{t}(K_{t}+Cok_{t})$ is

transversal to the tangent vector of the curve . Hence we can find a $a_{t}$

function $\psi_{0}\in C^{\beta}(L_{m}, R)$ such that

$(Hess_{a_{t}}\psi o)(P_{t}V, PtW)=\langle Q_{t}V, W\rangle$ ,

for each $V\in Cok_{t}$ and $W\in K_{t}$ . It is easy to see that $\psi=\psi+\delta\psi_{0}$
has
the required property.
Lemma 3.9 follows easily from Sublemmas 3.10 and 3.11.

\S 4. Sum formula for index bundles

It seems that many parts of this section are well known to experts.
But we include it here because of the lack of appropriate reference and
because we need a part of the proof in \S 11. However we omit the detail
of the proof since the results are essentially known. First we shall work
in the following situation.
Situation 4.1. Let $X^{n+1}$ be an oriented complete Riemannian man-
ifold, $E$ , $F$ be vector bundles on it, $K$ a compact subset. Suppose that
$X-K$ is isometric to the direct product $M\times(0, \infty)$ . Let $V$ be a vector
bundle on $M$ and : , and : be isomorphisms
$\Psi_{E}$ $E\rightarrow p^{*}V$ $\Psi_{F}$ $F\rightarrow p^{*}V$

of vector bundles. (Here : (0, is the projection.) Let $p$

$ M\times$ $\infty)\rightarrow M$

:
$D^{0}$
and $D$ :
$\Gamma(V)\rightarrow\Gamma(V)$ be elliptic operators of first $\Gamma(E)\rightarrow\Gamma(F)$

order. Suppose that is selfadjoint. Assume that $M$ is decomposed

$D^{0}$

to such that
$M_{+}$ $II$ $M_{-}$

$D=\Psi_{F}^{-1}(\pm\frac{\partial}{\partial t}+D^{0})\Psi_{E}$

respectively on $M_{\pm}\times(0, \infty)$ . Let $\{\lambda_{i}|i\in Z\}$ be the set of all eigenvalues
of . Put
$D^{0}$ $\lambda_{0}=\min_{i\in Z}\lambda_{i}^{2}$
.

Theorem 4.2. Suppose . then $D$ is Fredholm. Moreover, $\lambda_{0}>0$

for , there exists fifinite dimensional subspace

$\lambda<\lambda_{0}$
of $L^{2}(E)$ , such
$a$ $L_{\lambda}$

that
(4.2.1) If $u\in L_{\lambda}^{\perp}$
then $|Du|>\sqrt{\lambda}|u|$ . $\circ Here$ $L_{\lambda}^{\perp}is$ $a$
orthonormal
complement of $L_{\lambda}$
20 K. Fukaya

(4.2.2) $L_{\lambda}$
is generated by the vectors $v$ satisfying $D^{*}Dv=\lambdav$ with
$\lambda\leq\lambda$
.

We omit the proof. See $[LM],[T3]$ . Theorem 4.2 implies that

Index $=dimKer$ D-dim $KerD^{*}$

is well defined.

Situation 4.3. Let , , , , , , be as in Situation 4.1.

$X_{i}$ $M_{i}$ $E_{i}$ $F_{i}$ $V_{i}$ $D_{i}$ $D_{i}^{0}$

We assume that there are unions of connected components, say $M_{1,+}^{0}$

and , of $M_{1,+}$ and $M_{2,-}$ respectively, and an orientation reversing

$M_{2-}^{0}$

dieomorphism from to , by which we can identity $M_{1,+}^{0}$ , $M_{2,-}^{0}$ $V_{1}$ $D_{1}^{0}$

and , . We patch $X_{1}-M_{1,+}^{0}\times(T, \infty)$ and $X_{2}-M_{2,-}^{0}\times(T, \infty)$ by

$V_{2}$ $D_{1}^{0}$

the dieomorphism to obtain $X(T)$ . (Figure

$M_{1,+}^{0}\times\{T\}\rightarrow M_{2,-}^{0}\times\{T\}$

1)

Figure 1.

Let $E(T)$ (resp. $F(T)$ ) be a vector bundle on $X(T)$ obtained by

patching and (resp. $E_{1}$
) by (resp. $E_{2}$ $F_{i}$ $\Psi_{E_{2}}^{-1}\Psi_{E_{1}}$ $\Psi_{F_{2}}^{-1}\Psi_{F_{1}}$
). Define an
operator $D:\Gamma(E(T))\rightarrow\Gamma(F(T))$ by

$D_{1}$
on $X_{1}$

$D=\{$
$D_{2}$
on $X_{2}$
 Floer Homology for Oriented 3-Manifolds 21

Theorem 4.4. If $\lambda_{0}>0$ then we have

Index $=Index$ $D_{1}+IndexD_{2}$ .

Proof. Let is not an eigenvalue

$0<\lambda<\lambda_{0}$ . We may assume that $\lambda$

of or$D^{*}D$ . Let be the vector space generated by the

$D_{i}^{*}D_{i}$ $L_{\lambda}\subset L^{2}(E)$

vectors such that $D^{*}Dv=\lambdav$

$v$ with . Define , , $\lambda<\lambda$ $L_{\lambda}^{*}\subset L^{2}(F)$ $L_{\lambda}^{i}$

in the same way. Note that an embedding $X_{1}-M_{1,+}^{0}\times[T,$

$L_{\lambda}^{i*}$ $\infty\grave{)}\rightarrow X$

can be extended to an embedding $X_{1}-M_{1,+}^{0}\times[2T, \infty)$ . Let $ M_{1,+}^{0}\times$

$[0,2T]\rightarrow X$ be its restriction. Put $d(t)=\min(|t|, |2T-t|)$ .

Lemma 4.5. If $u\in L_{\lambda}$

then
$|\nabla^{k}\varphi|(I(x, t))<C_{k}e^{-\sqrt{\lambda_{O}-\lambda}d(t)}||u||_{L^{2}}$
.

We may assume $D^{*}Du=\lambdau$ ,

Proof. $\lambda<\lambda$
. Let $\varphi_{1}$
, $\cdots$
be the
eigenvectors of . We put $D_{0}^{*}D_{0}$

$u(I(x, t))=\sum_{i=1}^{\infty}u_{i}(t)\varphi_{i}(x)$ .

Since
$D^{*}D=-\frac{\partial^{2}}{\partial t^{2}}+(D^{0})^{2}$
,
we have
$-\frac{d^{2}u_{i}}{dt^{2}}+\lambda_{i}^{2}u_{i}=\lambdau_{i}$
.

It follows that
$|u_{i}(t)|\leq Ce^{-\sqrt{\lambda_{0}-\lambda}d(t)}\max\{|u_{i}(0)|, |u_{i}(T)|\}$
,
from which the lemma follows by the standard estimates for elliptic
operators.
Let $\chi$ : $[-1, 1]\rightarrow[0,1]$ be a nondecreasing $C^{\infty}$
function such that
0 if $t<-1$
$\chi(t)=\{$
1 if $t>1$ .

We define : as follows. (Here

$P_{i}$ $L_{\lambda}\rightarrow\Gamma_{c}(X_{i}, E_{i})$ $\Gamma_{c}$
stands for the set
of smooth sections with compact support.)
$(P_{1}u)(x, t)=(1-\chi(\frac{t-T}{T}))u(x, t)$ if $(x, t)\in M_{1,+}^{0}\times[0,2T]$

$\{$ $(P_{1}u)(x, t)=0$ if $(x, t)\in M_{1,+}^{0}\times[2T, \infty)$

$(P_{1}u)(z)=u(z)$ if $z\not\in M_{1,+}^{0}\times[0, \infty)$

22 K. Fukaya

$(P_{2}u)(x, t)=\chi(\frac{t-T}{T})u(x, t)$ if $(x, t)\in M_{2,-}^{0}\times[0,2T]$

$\{$ $(P_{1}u)(x, t)=0$ if $(x, t)\in M_{2,-}^{0}\times[2T, \infty)$

$(P_{1}u)(z)=u(z)$ if $z\not\in M_{2,-}^{0}\times[0, \infty)$

Let be the orthonormal projection of

$P_{i}(u)$ to . Put $P_{i}(u)$ $L_{\lambda}^{i}$ $P_{\lambda}=$

:
$(P_{1}, P_{2})$ . Then using Lemma 4.5 we can prove that is
$L_{\lambda}\rightarrow L_{\lambda}^{1}\oplus L_{\lambda}^{2}$ $P_{\lambda}$

an isomorphism for large . Similarly we can construct an isomorphism

$T$

:
$P_{\lambda}^{*}$
. On the other hand, $D$ defines an isomorphism:
$L_{\lambda}^{*}\rightarrow L_{\lambda}^{1*}\oplus L_{\lambda}^{2*}$

$ L_{\lambda}\cap$
. Therefore
$(Ker D)^{\perp}\rightarrow L_{\lambda}^{*}\cap(KerD^{*})^{\perp}$

Index $D=dim$ $L_{\lambda}-dimL_{\lambda}^{*}$ .

Similarly, we have

$IndexD_{i}=dimL_{\lambda}^{i}-dimL_{\lambda}^{i*}$ .

The theorem follows immediately. (Recall that Index $D^{T}$

does not de-
pend on $T.$ )

Remark 4.6. By the same method, we can prove that, if $D_{0}$

is
invertible, then the neighborhood of the set $Ce^{-\sqrt{\lambda_{0}-\lambda}T/C_{-}}$

{eigenvalues of $D^{T*}D^{T}$
smaller than $\lambda_{0}$

}
contains the set

{eigenvalues of $D_{1}D_{1}^{*}$
smaller than $\lambda_{0}$

}
{eigenvalues of
$\cup$ $D_{2}D_{2}^{*}$
smaller than $\lambda_{0}$

}.
Also the $Ce^{-\sqrt{\lambda_{0}-\lambda}T/C_{-}}$
neighborhood of the later set contains the former
set.

Moreover we can prove the following:

Corollary 4.7. In Situation 4.1, let , be unions of com- $M_{+}^{0}$ $M_{-}^{0}$

ponents of $M_{+}$ , $M_{-}$ , respectively. Suppose that , together with $D_{0},V$ $M_{+}^{0}$

on it, is dieomorphic to . Construct $X(T)$ , $E(T)$ , $F(T)$ , , $e.t.c$ . $M_{-}^{0}$ $D^{T}$

as before. {Figure 2) Then we have

Index $D^{T}=Index$ V.

In \S 6 and \S 11, we need also a family version of Theorem 4.4.

 Floer Homology for Oriented 3-Manifolds 23

$M_{+}^{0}=M_{-}^{0}$

$X(T)$

Figure 2.

Situation 4.8. Let $Y$ be a manifold, : , : $Z\rightarrow Y$ $p_{i}$ $W_{i}\rightarrow Y$ $q$

be fibre bundles. Let be vector bundles and $\overline{E}_{i},\overline{F}_{\dot{x}}:\rightarrow W_{i},\overline{V}\rightarrow Z$

:
$\overline{D}_{i}$

: be families of elliptic operators.

$\Gamma(\overline{E}_{i})\rightarrow\Gamma(\overline{F}_{i}),\overline{D}^{0}$ $\Gamma(\overline{V})\rightarrow\Gamma(\overline{V})$

Suppose that $p_{i}^{-1}(y)=X_{i}(y)$ , $q^{-1}(y)=M(y),\overline{E}_{i}|_{X_{i}(y)}=E_{i}(y)$ , $F_{i}(y)$ ,

$V(y)$ are as in Situation 4.3, for each
, $D_{i}(y)$ , $D^{0}(y)$ $y\in Y$ . As before,
we can construct, , $W(T)\rightarrow Y,\overline{E}(T),\overline{F}(T)\rightarrow W(T)$
$D(T)$ : $\Gamma(\overline{E}(T))\rightarrow$

. As in [AS], the index bundles

$\Gamma(\overline{F}(T))$

Index $D_{i}$
, $IndexD^{T}\in K(Y)$ ,

are well defined if $D^{0}(y)$ is invertible.

Theorem 4.9. Suppose $D^{0}(y)$ is invertible for each $y$ , then we
have
Index $D_{1}+IndexD_{2}=IndexD^{T}$ ,
in $K(Y)$ .

Theorem 4.9 follows from the proof of Theorem 4.4, since $P_{\lambda}$
and
$L_{\lambda}$
, e.t.c. there depend smoothly on operators.
24 K. Fukaya

Remark 4.10. The results of this section hold in the case when,
for example, in Situation 4.1 the operator $D$ is not exactly equal to
, but the dierence is estimated by $Ce^{-|t|/C}$ . (See
$\Psi_{F}^{-1}(\pm\frac{\partial}{\partial t}+D^{0})\Psi_{E}$

[T3].)

\S 5. Dimension of moduli space

We put . Recall that the action of

$\overline{\mathcal{M}}_{\delta}(a, b)=G_{a}\backslash \mathcal{M}_{\delta}(a, b)/G_{b}$

is trivial if
$G_{a}\times G_{b}$ $\delta<0$ . We can prove that is independent $\overline{M}_{\delta}(a, b)$

of . Hence we write
$\delta$

. $\overline{\mathcal{M}}(a, b)$

Theorem 5.1. There exists a map $\mu$ : $Fl$ $\rightarrow Z$

such that $\mu(1)=0$
and that

(5.1) $dim\overline{\Lambda 4}(a, b)=\mu(a)-\mu(b)-dimG_{a}$ ,

except the component containing no irreducible connection.

Proof. First we assume that , $b\in Fl_{0}$ . In this case $a$ $dim\overline{\mathcal{M}}(a, b)=$

$dim\mathcal{M}_{\delta}(a, b)$ . We can use the perturbed Atiyah-Hitchin-Singer complex

(5.2) $\Omega_{\ell+1,0}^{0}\rightarrow\Omega_{\ell,0}^{1}d_{A}\rightarrow\Omega_{\ell-1,0}^{2}D_{A}$
.

(definitions of operators and spaces are in \S 3), to calculate the dimension

as
$dim\mathcal{M}_{\delta}(a, b)=dim\frac{KerD_{A}}{Imd_{A}}$ .

Since $a\in Fl_{0}$ , it follows that $d_{A}$

is injective. By Lemma 3.9, is $D_{A}$

surjective. Hence $dim\mathcal{M}_{\delta}(a, b)$ is equal to the index of the complex

(5.2). We put
$(D_{A}, d_{A}^{*})$ : $\Omega_{\ell,0}^{1}\rightarrow\Omega_{\ell,0}^{2}\oplus\Omega_{\ell-1,0}^{0}$
.

Then we have:
$dim\mathcal{M}_{\delta}(a, b)=Index(D_{A}, d_{A}^{*})$ .

We $identi5^{r}\Omega_{\ell}^{1}$
and $\Omega_{\ell}^{2}\oplus\Omega_{\ell}^{0}$
to $\mathcal{L}_{\ell,\delta}^{1}\oplus \mathcal{L}_{p\delta}^{0}$

, as in \S 3. For $a\in A_{\ell}(M)$ ,

define

$D_{a}$
: $L_{\ell}^{2}(M, (\wedge^{1}\oplus\wedge^{2})\otimes su(2))\rightarrow L_{\ell}^{2}(M, (\wedge^{1}\oplus\wedge^{2})\otimes su(2))$

by
$D_{a}(u, \varphi)=(*_{\sigma}d_{a}u-Hess_{a}u+d_{a}\varphi, d_{a}^{*}u)$ .
 Floer Homology for Oriented 3-Manifolds 25

Then when $t$

the operator
$\rightarrow\infty$ $(D_{A}, d_{A}^{*})$ is asymptotic to and $-\frac{\partial}{\partial t}+D_{b}$

when $t$
$\rightarrow-\infty$
it is asymptotic to . Since , $b\in Fl_{0}$ it follows
$-\frac{\partial}{\partial t}+D_{a}$ $a$

that
$d_{a}$
: $L^{2}(M, su(2))\rightarrow L^{2}(M, \wedge^{1}\otimes su(2))$

is injective. Hence by (2.8.2), and are invertible. Therefore $D_{a}$ $D_{b}$

by Theorem 4.3, is Fredholm for each $A\in B_{\ell,\delta}(a, b)$ . Since

$(D_{A}, d_{A}^{*})$

is connected, it follows that its index is independent of $A$ .

$B_{\ell,\delta}(a, b)$

Therefore, we can use Theorem 4.4 to show

Index(VB, $ d_{C}^{*}$
) $=Index(VB, d_{A}^{*})+Index(VB, d_{B}^{*})$ ,

for ,
$A\in \mathcal{M}_{\delta}(a, b)$ , $B\in \mathcal{M}_{\delta}(b, c)$ $C\in \mathcal{M}_{\delta}(a, c)$ , $a$ , , $c\in Fl_{0}$ . In the case
$b$

when $b$
is reduced, way we can prove

Index(VB, $ e_{\delta}d_{C}^{*}e_{\delta}^{-1}$
) $=Index(D_{A}, e_{\delta}d_{A}^{*}e_{\delta}^{-1})+Index(D_{B}, e_{\delta}d_{B}^{*}e_{\delta}^{-1})$

$-dimG_{b}$ ,

in a similar way, for $\delta>0$ . Therefore the theorem follows by putting

$\mu(a)=Index(D_{A}, e_{\delta}d_{A}^{*}e_{\delta}^{-1})-3$ ,

for an element $[A]\in B_{\ell,\delta}(1, a)$ .

Next we study the neighborhood of a reducible connection

. There are two cases:
$A\in \mathcal{M}_{\delta}(a, b)$

Case I. $dimG_{a}=dimG_{b}=3$ , $G_{A}=U(1)$ .

Case . $II$
$dimG_{a}=dimG_{b}=1$ , $G_{A}=U(1)$ .

In case , there exists : $TorH_{1}(M, Z)\rightarrow\{\pm 1\}\subset U(1)$ such that

$I$
$\varphi$

,
$a$ . (See \S 2.) Then we can renumber the loops ,
$b\in RF_{\varphi}$ , , which $\ell_{1}^{0}$
$\cdots$ $\ell_{d}^{0}$

we choose at the beginning of \S 2, such that

$a(\ell_{i}^{0})=1\Leftrightarrow i\leq p$

$b(\ell_{i}^{0})=1\Leftrightarrow i\leq p+k$ .

(At this point, it is not yet clear that $k>0.$ )

Replacing the element by a gauge equivalent one, we may assume $b$

that there exists $a_{t}\in A_{\ell}(M)$ such that $d+A=d+a_{t}$ . (Namely $A$
has no component.) The group $U(1)=G_{A}$ acts on the complex
$dt$

. It follows that its index is a $U(1)$ module.

$(D_{A}, e_{\delta}d_{A}^{*}e_{\delta}^{-1})$
26 K. Fukaya

Lemma 5.3.

Index(PA, $ e_{\delta}d_{A}^{*}e_{\delta}^{-1}$
) $\simeq\{$
$C^{k+1}\oplus R^{k+1}$
if $\delta>0$

$C^{k-1}\oplus R^{k-1}$
if $\delta<0$ .

Proof. We replace the complex $(D_{A}, d_{A}^{*})$ by $(D_{A,1}+\epsilon, d_{A}^{*}+\epsilon)$

, where

$ D_{A,1}(u, \varphi)=-\frac{\partial u}{\partial t}+*_{\sigma}d_{a_{t}}u+d_{a_{t}}\varphi$

.

Put
$Index(D_{A,1}+\epsilon, d_{A}^{*}+\epsilon)=C^{k_{1}}\oplus R^{k_{2}}$ .

The trivial $su(2)$ bundle together with (nontrivial) connection $d+a_{t}$ on

$M\times R$ splits into a real line bundle and a complex line bundle , $\mathcal{L}^{R}$ $\mathcal{L}^{C}$

since $d+a_{t}$ is reducible. Note that the image of holonomy representation

of and is contained in
$a$ , the center of $SU(2)$ . Therefore the line
$b$ $\{\pm 1\}$

bundles together with their connections, have canonical trivializations

on their ends. Hence we can apply Corollary 4.7 to obtain bundles
$\overline{\mathcal{L}}^{R}$

and on such that

$\overline{\mathcal{L}}^{C}$

$M\times S^{1}$

$k_{1}=dimc$ Index $((P_{-}d_{A}, d_{A}^{*})\otimes\overline{\mathcal{L}}^{C})$

$k_{2}=dim_{R}$ Index $((P_{-}d_{A}, d_{A}^{*})\otimes\overline{\mathcal{L}}^{R})$

.

Here
$\overline{\mathcal{L}}^{C}\rightarrow d_{A}\wedge^{1}(M\times S^{1})\otimes\overline{\mathcal{L}}^{C}P_{-}d_{A}\rightarrow\bigwedge_{-}^{2}(M\times S^{1})\otimes\overline{\mathcal{L}}^{C}$
,

and similarly for Therefore, as in Atiyah-Hichin-Singer [AHS], we

$\overline{\mathcal{L}}^{R}$

have

$k_{1}=\int_{M\times S^{1}}(2+\frac{p^{1}(M\times S^{1})}{3})(1+c^{1}(\overline{\mathcal{L}}^{C})+\frac{c^{1}(\overline{\mathcal{L}}^{C})\wedge c^{1}(\overline{\mathcal{L}}^{C})}{2})$

$=0$ ,

since

$c^{1}(\overline{\mathcal{L}}^{C})=\sum_{i=p+1}^{p+k}[\ell_{i}^{0}]\cup[S^{1}]$
.

Similarly $k_{2}=0$ .
 Floer Homology for Oriented 3-Manifolds 27

Next we compare the index of to one of $(D_{A,1}+\epsilon, d_{A}^{*}+\epsilon)$

. For this purpose, we use the notion of spectral flow

$(D_{A}, e_{\delta}d_{A}^{*}e_{\delta}^{-1})$

due to Atiyah-Patodi-Singer [APS]. Put

$D_{a_{t},1}(u, \varphi)=(*_{\sigma}d_{a_{t}}u+d_{a_{t}}\varphi, d_{a_{t}}^{*}\varphi)$ .

The spectral flow of the operator gives the index of $(D_{A}+$ $ D_{a_{t},1}+\epsilon$

,
$\epsilon$

. The operator
$d_{A}^{*}+\epsilon)$
has zero as eigenvalue. The eigenspace is $D_{a,1}$

identified to $(C\oplus R)^{d+1}\simeq(H_{0}(M;R)\oplus H_{1}(M;R))\otimes su(2)$ . Replacing

by
$D_{A}$
is equivalent to push these eigenvalues a bit to positive
$ D_{A}+\epsilon$

direction. Next we examine the eect of the perturbation. We put

$D_{a_{t},2}(u, \varphi)=(*_{\delta}d_{a_{t}}u-Hess_{a_{t}} f(u)+d_{a_{t}}\varphi, d_{a_{t}}^{*}\varphi)$ .

We take the basis of $H_{1}(M;R)\otimes su(2)$ such that

$(z_{1}, \cdots, z_{d}, t_{1}, \cdots, t_{d})$

and
$z_{i}$ correspond to . Then, by (2.5) and our choice of and ,
$t_{\dot{\iota}}$ $\ell_{i}^{0}$
$a$
$b$

replacement of by is equivalent to push the zero eigenvalues

$D_{a_{t},1}$ $D_{a_{t},2}$

corresponding , , and , , a bit to positive direction and

$z_{1}$ $\cdots$
$z_{p}$
$t_{1}$ $\cdots$ $t_{p}$

the others to negative direction while , and to push the zero $t$
$\rightarrow-\infty$

eigenvalue corresponding to , , and , , a bit to positive $z_{1}$

$\cdots$
$z_{p+k}$
$t_{1}$ $\cdots$ $t_{p+k}$

direction and the others to negative direction while . It follows $ t\rightarrow\infty$

from $k_{1}=k_{2}=0$ that the index of the spectral flow is . $D_{a_{t},2}$

$C^{k}\oplus R^{k}$

Finally we examine the eect replacing by . If $D_{a_{t}}$ $(D_{a_{t}}, e_{\delta}d_{A}^{*}e_{\delta}^{-1})$

$\delta>0$ , this is equivalent to push the zero eigenvalues in $H_{0}(M;R)\otimes su(2)$

to positive direction while and push them to negative direction $ t\rightarrow\infty$

while . If $\delta<0$ this is equivalent to the perturbation to the

$ t\rightarrow-\infty$

opposite direction. Lemma 5.3 follows.

Lemma 5.3 implies $k>0$ . Using Lemma 5.3, we have a description

of the moduli space in a neighborhood of reducible connections. First
let $k=1$ , $\delta>0$ . The group $SU(2)\times SU(2)\times R$ acts on . Here $\sqrt{}\backslash 4_{\delta}(a, b)$

$SU(2)\times SU(2)\simeq G_{a}\times G_{b}$ acts on by Lemma 3.4, and the $\mathcal{M}_{\delta}(a, b)$

action of $R$ is induced by its action on R. Since $G_{A}=U(1)$ there $ M\times$

exists an embedding

$\frac{SU(2)\times SU(2)}{U(1)}\times R\rightarrow \mathcal{M}_{\delta}(a, b)$ .

By Lemma 5.3, this map is a dieomorphism onto a connected compo-

nent containing $[A]$ . It follows that all the connections on this component
is reducible. In the case $k\geq 2$ we can use a similar argument. Summing
up we obtain
28 K. Fukaya

Theorem 5.4. Suppose $dimG_{a}=dimG_{b}=3$ , $dimG_{A}=1$ , $[A]\in$

, $\delta>0$ . Then $\mu(a)=3k+\mu(b)$ for some $k\leq d$ and that there
$\mathcal{M}_{\delta}(a, b)$

exists a dieomorphism from

$\frac{SU(2)\times C^{k-1}\times SU(2)}{U(1)}\times R^{k}$

onto a neighborhood of the orbit of $[A]$ . The dieomorphism

$G_{a}\times G_{b}\times R$

is compatible with $G_{a}\times G_{b}\times R\simeq SU(2)\times SU(2)\times R$ action.

Remark 5.5. In case $k=1$ the formula (5.1) does not hold for this
component. This is similar to the fact that the virtual dimension of the
trivial connection on is -3. In case $k>1$ the neighborhood of $[A]$
$S^{3}$

in is dieomorphic to the product of $CCP^{k-1}\times R^{k}$ . Here $C$

$\overline{\mathcal{M}}(a, b)$

means the cone. (Compare [D1].)

By a similar but simpler argument we can examine the case when
$G_{0}=U(1)$ and obtain:

Theorem 5.6. Let $G_{a}=G_{b}=G_{A}=U(1)$ , and $A\in\Lambda 4_{\delta}(a, b)$

. Then $\mu(a)=\mu(b)+k$ for some

$\delta>0$ $k\leq d$ . All the connections
contained in the connected component of containing $[A]$ are $\lambda\Lambda_{\delta}(a, b)$

reducible.

Remark 5.7. We used the above index calculation in the proof of

Sublemma 3.10. The fact we used there is that the $C$ -part of the index
is always of nonnegative dimension.
If we use dierent perturbation from one we gave in 2,3, (for ex-
ample if we change the sign in Formula (2.16) from point to point) then
the above fact is no longer true. As the consequence, Lemma 3.9 does
not necessary hold in that case, and we have an obstruction in second
homology of Atiyah-Hitchin-Singer complex.
Finally we remark:

Lemma 5.8. Let $[a]$ , $[b]\in Fl$ , $b=g^{*}a$ , where $g$ : $M\rightarrow SU(2)$ and
$degg=k$ . Then,
$\mu(b)=8k+\mu(a)$ .

For the proof see [F].

\S 6. Orientation of moduli space

Lemma 6.1. $\mathcal{M}_{\delta}(a, $
b) is orientable.
 Floer Homology for Oriented 3-Manifolds 29

Proof.
Let be the determinant $D\mathcal{E}\mathcal{T}(a, b)=D\mathcal{E}\mathcal{T}(D_{A}, e_{\delta}d_{A}^{*}e_{\delta}^{-1})$

bundle of the Atiyah-Hitchin-Singer complex (5.2). We can extend

to a real line bundle on
$D\mathcal{E}\mathcal{T}(a, b)$
. On , the bun- $B_{\ell,\delta}(a, b)$ $\mathcal{M}_{\delta}(a, b)$

dle is isomorphic to the bundle of

$D\mathcal{E}\mathcal{T}(a, b)$
-forms. Hence $dim\mathcal{M}_{\delta}(a, b)$

it suces to show :
Lemma 6.2. The bundle $D\mathcal{E}\mathcal{T}(a, b)$
on $B_{\ell,\delta}(a, b)$ is trivial.

Proof. Since is not simply connected, the argument in

$\mathcal{M}_{\ell,\delta}(a, b)$

[D1], [F], can not be applied directly to our situation. Instead we shall
proceed as follows. Since 3-dimensional oriented cobordism group is
trivial, we can find oriented manifolds such that , $\overline{X}_{\pm}$ $\partial\overline{X}_{+}=M$ $\partial\overline{X}_{-}=$

$M^{-}$
, where $M^{-}$ is the manifold $M$ with opposite orientation. Let $W$
be a closed oriented 4-manifold obtained by patching and along $X_{+}$ $X_{-}$

$M$ . Take trivial $SU(2)$ bundles on them. Let $A_{\ell}(W)$ be the set of all $L_{\ell}^{2}$

connection on $W$ , and be the group of transformations. We put $\mathcal{G}_{\ell}(W)$

$B_{\ell}(W)=A_{\ell}(W)/\mathcal{G}_{\ell+1}(W)$ . Put a metric on , such $X\pm=\overline{X}_{\pm}-\partial\overline{X}_{\pm}$

that $X\pm-K\pm is$ isometric to $M\times(0, \infty)$ for some compact subset $ K\pm\cdot$

Let be a function on $X\pm such$ that

$e_{\delta}$
outside . For $e_{\delta}(x, t)=e^{-\delta||t||}$ $K_{\pm}$

$a\in Fl$ choose a connection $d+A^{a}$ on such that $A^{a}=a$ outside . $X_{\pm}$ $K_{\pm}$

Put

$L_{\ell,\delta}^{2}(X_{\pm}, \wedge^{1}\otimes su(2))=\left\{\begin{array}{lll} & .alocallyuisL_{\ell}^{2}section & \\ & of\wedge^{1}\otimes su(2) & \\u & \sum_{k=0}^{\ell}\int_{x_{\pm}} & e_{\delta}|\nabla^{k}u|<\infty\end{array}\right\}$

$A_{\ell,\delta}(X_{\pm}, a)=\{d+A^{a}+u|u\in L_{\ell,\delta}^{2}(X_{\pm}, \wedge^{1}\otimes su(2))\}$ .

Define $\mathcal{G}_{\ell,\delta}^{0}$
as in \S 2. Put

$B_{\ell,\delta}(X_{\pm}, a)=A_{\ell,\delta}(X_{\pm}, a)/\mathcal{G}_{\ell+1,\delta}^{0}(X_{\pm})$

.

Let be the determinant bundle of Atiyah-Hitchin-Singer com-

$D\mathcal{E}\mathcal{T}_{\pm}(a)$

plex on . First we shall prove that

$B_{\ell,\delta}(X_{\pm}, a)$ is trivial. For $D\mathcal{E}\mathcal{T}_{\pm}(a)$

simplicity, we assume that $a\in Fl_{0}$ . It suces to show that is $D\mathcal{E}\mathcal{T}_{\pm}(a)$

trivial on each compact subset of . We define a map Pat : $ L\pm$ $\mathcal{B}_{\ell,\delta}(X\pm, a)$

as follows. Define a Riemannian manifold $X(T)$ by

$L_{+}\times L_{-}\rightarrow B_{\ell,\delta}(W)$

patching and along $M$ as in Situation 4.3. Then $M\times[0,2T]$ is

$X_{+}$ $X_{-}$

embedded in $X(T)$ . Choose a function : $[-1, 1]\rightarrow[0,1]$ by $C^{\infty}$

$\chi$

0 if $t<-1$
$\chi(t)=\{$
1 if $t>1$ .
30 K. Fukaya

For $[d+A]\in L_{+}$ , $[d+B]\in L_{-}$ define Pat([A], $[B]$ ) by

Pat([A], $[B]$ ) $(z)=A(z)$ if $z\in X_{+}-M\times(0, \infty)$

$\{$
Pat([A], $[B]$ ) $(x, t)=(1-\chi(\frac{t-T}{T}))A(x, t)+\chi(\frac{t-T}{T})B(x, t)$

Pat([A], $[B]$ ) $(z)=B(z)$ if $z\in X_{-}-M\times(0, \infty)$

Let be the determinant bundle of the Atiyah-

$D\mathcal{E}\mathcal{T}_{X(T)}\rightarrow B_{\ell}(X(T))$

Hitchin-Singer complex on $X(T)$ . By Theorem 4.9, we have

Pat* $(D\mathcal{E}\mathcal{T}_{X}(T))\simeq D\mathcal{E}\mathcal{T}_{+}(a)\otimes D\mathcal{E}I_{-}(a)$

.

For suciently large $T$ . By [D3], $D\mathcal{E}\mathcal{T}_{X(T)}$ is trivial. It follows that

is trivial.
$D\mathcal{E}I_{\pm}(a)$

Next, Let , $L\subset B_{\ell,\delta}(a, b)$ $L\subset B_{\ell,\delta}(X^{+}, a)$ be compact subsets. In a
similar way, we define a map Pat : . By Theorem
$L\times L\rightarrow B_{\ell,\delta}(X^{+}, b)$

4.9, we have

Pat* $(D\mathcal{E}\mathcal{T}_{+}(b))\simeq D\mathcal{E}\mathcal{T}(a, b)\otimes D\mathcal{E}\mathcal{T}_{+}(a)$

.

Therefore the trivializations of and induces a triv- $D\mathcal{E}\mathcal{T}_{+}(a)$ $D\mathcal{E}\mathcal{T}_{+}(b)$

ialization of , if , $b\in Fl_{0}$ . The case when $and/or$ are

$D\mathcal{E}\mathcal{T}(a, b)$ $a$ $a$ $b$

reducible can be proved in a similar way, by using a perturbation of

the complex around the boundaries. The proof of Lemma 6.2 is now
complete.

\S 7. Partial compactification of moduli space

Let , be the quotients of

$\mathcal{M}_{\delta}(a, b)$
and by
$\overline{\mathcal{M}}^{J}(a, b)$
$\mathcal{M}_{\delta}(a, b)$ $\overline{\mathcal{M}}(a, b)$

the $R$-action. The proof of the theorems in \S 1 is based on the following

Theorems 7.1 and 7.3 on the structure of the ends of . Hereafter $\overline{\sqrt{}\backslash \Lambda}(a, b)$

we fix suciently small positive number and write e.t.c. in $\delta$

$\Lambda 4(a, b)$

place of . $\Lambda 4_{\delta}(a, b)$

Theorem 7.1. For , $b\in Fl$ , let$a$

$C\overline{\mathcal{M}}^{J}(a, b)$
be the disjoint union
of
$\overline{\mathcal{M}}^{J}(a, c_{0})\times\prod_{i=0}^{k-1}\overline{\mathcal{M}}^{J}(c_{i}, c_{i+1})\times\overline{\mathcal{M}}(c_{k}, b)$
,

for $c_{0}$ , $\cdots$

, $c_{k}\in Fl$ , with $\mu(a)>\mu(c_{0})>\cdots>\mu(c_{k})>\mu(b)$ . Put
$m=dim\overline{\mathcal{M}}(a, b)$
.
 Floer Homology for Oriented 3-Manifolds 31

Then we can defifine a smooth structure on $C\overline{\mathcal{M}}(a, b)$

such that the
following holds.
(7.1.1) If

$x\in\overline{\mathcal{M}}(a, c_{0})\times\prod_{i=0}^{k-1}\overline{\mathcal{M}}(c_{i}, c_{i+1})\times\overline{\mathcal{M}}^{J}(c_{k}, b)$

,

with $G_{c_{x}}=\{\pm 1\}$ . Then a neighborhood of in is dieomor- $x$

$C\overline{\mathcal{M}}(a, b)$

phic to $[0, \infty)^{k+1}\times R^{m-k-1}$ .

(7.1.2) If , with $G_{c}=U(1)$ , $G_{A}=$
$x=([A], [B])\in\overline{\mathcal{M}}(a, c)\times\overline{\mathcal{M}}(c, b)$

$G_{B}=\{\pm 1\}$ . Then a neighborhood

of is dieomorphic to $R^{m}$ . $x$

(7.1.3) If , with $G_{c}=SU(2)$ ,

$x=([A], [B])\in\overline{\mathcal{M}}(a, c)\times\overline{\lambda\Lambda}(c, b)$

$G_{A}=G_{B}=\{\pm 1\}$ . Then a neighborhood of is dieomorphic to $x$

$\frac{C^{2}}{Z_{2}}\times R^{m-4}$
.

(7.1.4) If $x=(A, B, C)\in\overline{\mathcal{M}}(a, c_{1})\times\overline{\mathcal{M}}(c_{1}, c_{2})\times\overline{\mathcal{M}}(c_{2}, b)$

, with
$G_{c_{1}}=G_{c_{2}}=SU(2)$ , $G_{B}=U(1)$ , $G_{A}=G_{C}=\{\pm 1\}$ , $3k=\mu(c_{1})-\mu(c_{2})$ .
Then a neighborhood of $x$ is dieomorphic to

$((\frac{SO(3)\times C^{k-1}\times SO(3)}{U(1)}\times(0, \infty]^{2})/\sim)\times R^{m-2k-5}$ ,

where $\sim is$

defifined by
$([g_{1}, z, g_{2}], (\infty, t))\sim[g_{1}g, z, g_{2}]$ , $(\infty, t))$

$([g_{1}, z, g_{2}], (t, \infty))\sim[g_{1}, z, gg_{2}]$ , $(t, \infty))$ .

(7.1.5) If , with
$x=([A], [B], [C])\in\overline{\vee\wedge\Lambda}(a, c_{1})\times\overline{\mathcal{M}}^{J}(c_{1}, c_{2})\times\overline{\mathcal{M}}^{J}(c_{2}, b)$

$G_{c_{1}}=G_{c_{2}}=G_{B}=U(1)$ , $G_{A}=G_{C}=\{\pm 1\}$ . Then a neighborhood of $x$

is dieomorphic to $R^{m}$
.
(7.1.6) Let $\Lambda\in R_{+}$ . Then the set

$\overline{\mathcal{M}}^{J}(a, b;\Lambda)=\{[A]\in\overline{\mathcal{M}}(a, b)|\sup|F^{A}|<\Lambda\}$

is relatively compact in $C\overline{\mathcal{M}}(a, b)$

.
(7.1.7) The orientations of $\overline{\mathcal{M}}^{J}(c_{i}, c_{i+1})$
are compatible in $C\overline{\mathcal{M}}(a, b)$
.

Remark 7.2. (7.1.1). . . (7.1.5) above do not cover all the possible
cases. The general case is the combination of them and the reader can
easily supply it.
32 K. Fukaya

Next we construct the bundles in \S 1. Choose a set of loops

representing a basis of $H_{1}(M;Z)$ . Put
$\{\gamma_{1}, \cdots, \gamma_{d}\}$
R. $\Sigma_{i}=\gamma_{i}\times R\subset M\times$

The surface has a canonical spin structure. For

$\Sigma_{i}$
, we let $A\in A_{\ell,\delta}(a, b)$

$6_{A}^{i}$
: $\Gamma_{c}(\Sigma_{i}, su(2)\otimes C)\rightarrow\Gamma_{c}(\Sigma_{i}, su(2)\otimes C)$

be the Dirac operator twisted by the connection $A$ . For each , $b\in Fl$ , $a$

is a Fredholm operator. (We add since

$ 6_{A}^{i}+\epsilon$
is not Fredholm when $\epsilon$ $6_{A}^{i}$

or is reducible.) Then we obtain a complex line bundle

$a$ $b$

$\mathcal{L}_{i}(a, b)\rightarrow B_{\ell,\delta}(a, b)$

by
$\mathcal{L}_{i}(a, b)|_{[A]}=top\wedge(Ker(6_{A}^{i}+\epsilon))^{*}\otimes top\wedge Coker(6_{A}^{i}+\epsilon)$
.

(Note the action of is free on ( , )). The action of $\mathcal{G}_{\ell,\delta}$ $A_{\ell,\delta}$ $a$
$b$ $G_{a}\times G_{b}$

on is lifted to this line bundle. The group

$B_{\ell,\delta}(a, b)$ acts trivially $\{\pm 1\}$

on . The lift of the action of

$B_{\ell,\delta}(a, b)$ to is not necessary $\{\pm 1\}$ $\mathcal{L}_{i}(a, b)$

trivial. (Compare [D2], where the similar action is trivial because the
numerical index of the Dirac operator on a closed surface is zero.) Then
we consider the tensor product . It induces a complex $\mathcal{L}_{i}(a, b)\otimes \mathcal{L}_{i}(a, b)$

line bundle on , the set of irreducible connections in

$\overline{\mathcal{L}}_{i}^{(2)}(a, b)$ $\overline{\Lambda 4}_{*}^{J}(a, b)$

. (If we want to define the first Chern class

$\overline{\mathcal{M}}(a, b)$
itself, $c^{1}(\mathcal{L}_{i}(a, b))$

we have to invert 2.)

Theorem 7.3. Collection of line bundles

$\mathcal{L}_{i}^{(2)}(a, c_{0})\otimes\cdots\otimes \mathcal{L}_{i}^{(2)}(c_{k}, b)\rightarrow\overline{\mathcal{M}}_{*}^{J}(a, c_{0})\times\prod_{i=0}^{k-1}\overline{\mathcal{M}}_{*}(c_{0}, c_{i+1})\times\overline{\mathcal{M}}_{*}(c_{k}, b)$

,

can be patched together to give a complex line bundle on $C\overline{\mathcal{M}}_{*}^{J}(a, b)$

.

Here and hereafter stands for the set of irreducible connections. $\mathcal{M}_{*}$

We can not extend the line bundle to the neighborhood of the connec-
tions described in Theorems 5.4 and 5.6. This is the reason why Theorem
1.10 does not hold for $s>2$ when $H_{1}(M;Z)$ is torsion free and $s>0$
when $H^{1}(M;Z)$ has a
detail in 12.)
. (We shall explain this point a bit more
torsi.o $n$

The proofs of Theorems 7.1 and 7.3 occupy 7-11. We include the
analysis of the structure of moduli space and the line bundle on it in
the neighborhood of the connection described in Theorems 5.4 and 5.6,
though the author does not know how to use it to deduce a topological
 Floer Homology for Oriented 3-Manifolds 33

information. In order to explain the outline of the proofs of Theorems

7.1 and 7.3, we introduce the following notion. (Compare Donaldson
[D2].)

Definition 7.4. Let , , be com-

$K_{0}\subset\overline{\Lambda 4}(a, c_{0})$ $\cdots$
$K_{k}\subset\overline{\mathcal{M}}^{J}(c_{k}, b)$

pact subsets and , $T$ , $C>0$ . We say that

$\epsilon$

is a standard $[A]\in\overline{\mathcal{M}}^{J}(a, b)$

model of type $(K_{0}, \cdots, K_{k}, T, \epsilon, C)$ , if there exist $[A_{i}]\in K_{i}$ , $S_{i+1}>$
$T+S_{i}$ , and $[A]=[A]$ , with the following property.

Let : $M\times[-T, T]\rightarrow M\times R$ be the embedding defined by $I_{i}(x, t)=$

$I_{i}$

$(x, t+S_{i})$ . Then we have

(7.4.1) $||I_{i}^{*}(A)-A_{i}||_{C^{p}}(x, t)<\epsilon$ ,

(7.4.2) $|A-c_{i}|_{C^{\ell}}(x, t)<$

$C\exp\{-\min\{|S_{i}+T/2-t|, |S_{i+1}-T/2-t|\}/C\}$ ,

if $t$
$\in[S_{i}+T/2, S_{i+1}-T/2]$ .

$F^{A}|$

Figure 3.

The proof of Theorem 7.1 is based on the following two Theorems

7.5 and 7.6.
Theorem 7.5. There exists $C$ such that, for each $T$ , , , we $\Lambda$
$\epsilon>0$

can fifind a compact subset of for each , $b\in Fl$ , with the
$K_{a,b}$ $\overline{\mathcal{M}}(a, b)$
$a$
34 K. Fukaya

following property. If , $\sup|F^{A}|<\Lambda$ , and if $[A]\not\in K_{a,b}$ ,

$[A]\in\overline{\mathcal{M}}(a, b)$

then there exist , $c_{0}$

$\cdots$
, $c_{k}\in Fl$ such that $[A]$ is a standard model of type
$(K_{a,c_{0}}, \cdots, K_{c_{k},b}, T, \epsilon, C)$ .

Theorem 7.6. For each compact set , , $K_{0}\subset\overline{\mathcal{M}}^{J}(a, c_{0})$ $\cdots$ $K_{k}$

$\subset$
and $C$ , there exist
$\overline{\mathcal{M}}^{J}(c_{k}, b)$
$=$ and $T$ $=$ $\epsilon$
$\epsilon(K_{0}, \cdots, K_{k}, C)$

$T(K_{0}, \cdots, K_{k}, C)$ , such that the set

of elements of which is $\Lambda 4(a, b)$

a standard form of type $(K_{0}, \cdots, K_{k}, \epsilon, T, C)$ is parametrized by

$\overline{K}_{0}\times_{G_{c_{0}}}\overline{K}_{1}\times c_{c_{1}}\cdots\times c_{c_{k}}\overline{K}_{k}\times(T, \infty)^{k+1}$

.

Here $\overline{K}_{i}\subset\sqrt{}\backslash \Lambda(c_{i-1}, c_{i})$

is the lift of $K_{i}$
.

Here is the quotient of by the action $g([A], [B])=$

$\overline{K}_{0}\times_{G_{O}}\overline{K}_{1}$ $\overline{K}_{0}\times\overline{K}_{1}$

$([A]g^{-1}, g[B])$ of . The proof of Theorem 7.6 is in \S 8. For the proof of

$G_{0}$

Theorem 7.1, we need a bit more complicated version of Theorem 7.5.

Theorem 7.5. For each $\Lambda>0$ we can fifind $K_{a,b}\subset\overline{\mathcal{M}}(a, b)$

and
$C_{k}$
such that the conclusion of Theorem 7.5 holds for

$\epsilon_{k}=\epsilon(K_{a,c_{O}}, \cdots, K_{c_{k},b}, C_{k})$ ,

$T_{k}=T(K_{a,c_{0}}, \cdots, K_{c_{k},b}, C_{k})$

where $\epsilon(\cdots)$
, $T(\cdots)$ , and $C(\cdots)$ are as in Theorem 7.6.

The proof of Theorem 7.5 is in \S 9. Now we are ready to explain

the outline of the proof of Theorem 7.1. Let , $b\in Fl_{0}$ . Choose $a$ $K_{c,c}$

for $\mu(a)\geq\mu(c)\geq\mu(c)\geq\mu(b)$ , as in Theorem 7.5. For $c=(c_{0}, \cdots, c_{k})$ ,

Let and $T(c)$ be the number in Theorem 7.6. Define an equivalence
$\epsilon(c)$

relation $\sim on$

$\overline{K}_{a,c_{0}}\times\cdots\times\overline{K}_{c_{k},b}\times(T(c), \infty]^{k+1}$

by

$(x_{0}, \cdots, x_{k+1}, t_{0}, \cdots, t_{k+1})\sim(x0, \cdots, x_{i}g, g^{-1}x_{i+1}, \cdots, t_{k+1})$

$\{$
for each $t_{0}$
, $\cdots$
, $t_{k+1}$

$(x_{0}, \cdots, x_{k+1}, t_{0}, \cdots, t_{k+1})\sim(x_{0}, \cdots, x_{i}g, x_{i+1}, \cdots, t_{k+1})$

if $ t_{i}=\infty$ .
 Floer Homology for Oriented 3-Manifolds 35

Put

,
$\overline{X}(c)=\underline{\overline{K}_{a,c_{0}}\times\cdots\times\overline{K}_{c_{k},b}\times(T(c),\infty]^{k+1}}\sim$

$X(c)=G_{a}\backslash \overline{X}(c)/G_{b}$
,

$\overline{X}^{o}(c)=\mathring{\underline{\overline{K}_{a,c}\times\cdots\times\overline{K}_{c_{k},b}\times(T(c),\infty)^{k+1}}}\sim$
,

$\mathring{X}(c)=G_{a}\backslash \overline{X}^{o}(c)/G_{b}$
.

By Theorem 7.6, we have a dieomorphism

$\Phi_{c}$
: $\mathring{X}(c)\rightarrow\overline{\mathcal{M}}(a, b)$
.

to its image. If $c\subset c$

, we have, by Theorem 7.6,

$\Phi_{c,c}$
: $X(c)\rightarrow G_{a}\backslash \mathcal{M}(a, c_{0})\times_{G_{c_{\acute{O}}}}\cdots\times c_{c_{k}^{l}}\mathcal{M}(c_{k}, b)/G_{b}\times[T, \infty]^{k+1}$ .

We put
$U(c, c)=\{z\in X(c)|\Phi_{c,c}(z)\in\mathring{X}(c)\}$
.
If is true, then we are able to use these maps to define the
$\Phi_{c}\Phi_{c,c}=\Phi_{c}$

smooth structure on $C\Lambda 4(a, b)$ . But the above equality does not exactly
hold but holds modulo some small dierence. Hence we have to perturb
them. The argument needed for it is in \S 10, where we define the notion
of local action and construct it on the end of . To extend line $\lambda\Lambda(a, b)$

bundle we use an argument similar to the proof of the theorems in \S 4

and a lift of the local action to the line bundle.

\S 8. Taubes construction
We prove Theorem 7.6 in this section. Theorem 7.6 corresponds
Donaldson [D2] \S 4. There Donaldson used the alternating method.
His method might work in our situation, where we have to deal with var-
ious types of reducible connections. But, since the organization needed
for alternating method is a bit complicated, we use here more direct
argument. (Maybe this is one Donaldson suggested in [D2] $p302.$ )
For simplicity of notation, we shall prove a special (but the most
dicult) case. Let , , , $b\in Fl$ such that $G_{a}=G_{b}=\{\pm 1\}$ , $G_{c_{1}}=$
$a$ $c_{1}$ $c_{2}$

$G_{c_{2}}=SU(2)$ , $\mu(c_{1})=\mu(c_{2})+3$ , and be a component $\overline{K}\subset\Lambda 4(c_{1}, c_{2})$

consisting of reducible connections. (We have, by Theorem 5.4,

$\overline{K}\simeq\frac{SU(2)\times SU(2)}{U(1)}$ . )
36 K. Fukaya

Let be compact subsets and

$K_{1}\subset\overline{\mathcal{M}}(a, c_{1})$
, $K_{2}\subset\overline{\mathcal{M}}(c_{2}, b)$ $\overline{K}_{1}\subset$

be their lifts. We shall construct a dieomor-

$\lambda 4(a, c_{1}),\overline{K}_{2}\subset \mathcal{M}(c_{2}, b)$

phism : $\Phi_{K,K_{1},K_{2}}$ , whose

$\overline{K}_{1}\times c_{c_{1}}\overline{K}\times_{G_{c_{2}}}\overline{K}_{2}\times[T, \infty)^{2}\times R\rightarrow \mathcal{M}(a, b)$

image contains all standard model of type $(K_{1}, K, K_{2}, T, \epsilon, C)$ .
Choose a finite open covering

$U_{1}^{1}\cup\cdots\cup U_{N}^{1}\supseteq K_{1}$

$U_{1}^{2}\cup\cdots\cup U_{N}^{2}\supseteq K_{2}$

,

and sections : . Let : $\overline{s}_{j}^{i}$ $U_{j}^{i}\rightarrow\overline{K}_{i}$ $s_{j}^{1}$ $U_{j}^{1}\rightarrow A_{\ell,\delta}(a, c_{1})$ , $s_{j}^{2}$
: $ U_{j}^{2}\rightarrow$

be their lifts. Choose also an open covering

$A_{\ell,\delta}(c_{2}, b)$

$V_{1}\cup\cdots\cup V_{N}=SU(2)$ ,

such that $V_{k}$

is contractible. We have maps

$J_{k}^{1}$
: $V_{k}\times R\rightarrow SU(2)$

$J_{k}^{2}$
: $V_{k}\times R\rightarrow SU(2)$

such that

$J_{k}^{1}(g, t)=1$ if $t$

$<-1$
$\{$

$J_{k}^{1}(g, t)=g$ if $t>0$

$J_{k}^{2}(g, t)=1$ if $t>1$
$\{$

$J_{k}^{2}(g, t)=g$ if $t<0$ .

Let be a representative of
$d+a_{t}^{0}\in A_{\ell,\delta}(c_{1}, c_{2})$ point. $G_{c_{1}}\backslash \overline{K}/G_{c_{2}}=one$

Choose a nonincreasing smooth function : $R\rightarrow[0,1]$ such that $\chi$

1 if $t$
$<0$
$\chi(t)=\{$
0 if $t>1$ .

Now, we define a map

$\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}$
: $U_{j_{1}}^{1}\times V_{k_{1}}\times V_{k_{2}}\times U_{j_{2}}^{2}\times[T, \infty)^{2}\times R\rightarrow A_{\ell,\delta}(a, b)$
,
 Floer Homology fo7. Oriented 3-Manifolds 37

as follows. Let $A_{i}=s_{j_{i}}^{i}([A_{i}])$ , .

$S_{?}$ $\in[T, \infty)$ , $S\in R$ , $g_{i}\in V_{k_{i}}$ . Then

$\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}([A_{1}], g_{1}, g_{2}, [A_{2}], S_{1}, S_{2}, S)$

$=(J_{k_{1}}(g_{1}, \cdot)^{*}A_{1})(x, t-S)$ for $t$

$<S+S_{1}/3$

$=\chi(\frac{t-S-S_{1}/3}{S_{1}/3})g_{1}^{*}A_{1}(x, t-S)$

$+(1-\chi(\frac{t-S\prime-S_{1}/3}{S_{1}/3}))a_{t-S-S_{1}}^{0}$

for $t$
$\in[S+S_{1}/3, S+2S_{1}/3]$

$=a_{t-S-S_{1}}^{0}$ for $t\in[S+2S_{1}/3, S+S_{1}+S_{2}/3]$

$=\chi(\frac{t-S-S_{1}-S_{2}/3}{S_{2}/3})a_{t-S_{1}-S}^{0}$

$+(1-\chi(\frac{t-S-S_{1}-S_{2}/3}{S_{2}/3}))g_{2}^{*}A_{2}(x, t-S-S_{1}-S_{2})$

for $t\in[S+S_{1}+S_{2}/3, S+S_{1}+2S_{2}/3]$

$=(J_{k_{2}}^{2}(g_{2}, \cdot)^{*}A_{2})(s, t-S-S_{1}-S_{2})$ for $t$

$>S+S_{1}+2S_{2}/3$ .

Here $J_{k}^{x}(g, \cdot)$

is regarded as a map $M\times R\rightarrow SU(2)$ and a gauge trans-
formation.

Figure 4

We remark that, by the compactness of $K_{1}$

, we have a constant $ C\sim$

such that
$|(d+A_{1})-(d+a)|<Ce^{t/C}$ ,
(8.1) $\{$

$|(d+A_{1})-(d+c_{1})|<Ce^{-t/C}$ ,
38 K. Fukaya

for $A_{1}\in K_{1}$ . (Compare the decay estimate in next section.) A similar
estimate holds for and $K$ . Using (8.1) we can prove the following:
$K_{2}$

Lemma 8.2. If
$[A_{1}]\in U_{j_{1}}^{1}\cap U_{j_{1}}^{1}$ ,
$[A_{2}]\in U_{j_{2}}^{1}\cap U_{j_{2}}^{2}$ ,
$g_{1}\in V_{k_{1}}\cap V_{k_{2}}$ ,
$g_{2}\in V_{k_{2}}\cap V_{k_{2}}$ ,

then there exists a gauge transformation $\hat{g}$

, such that
$\hat{g}^{*}\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}([A_{1}], g_{1}, g_{2}, [A_{2}], S_{1}, S_{2}, S)(t, x)=$

$\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}([A_{1}], g_{1}, g_{2}, [A_{2}], S_{1}, S_{2}, S)(t, x)$

,

if $t\not\in[S+S_{1}/3, S+2S_{1}/3]\cup[S+S_{1}+S_{2}/3, S+S_{1}+2S_{2}/3]$ , and

$|\hat{g}^{*}\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}-\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}|<e(S_{1}, S_{2})$
.

Here and hereafter, we put

$e(S_{1}, S_{2})=C$ $\exp(-\min\{S_{1}, S_{2}\}/C)$ .

Choose an embedding $U(1)\subset SU(2)$ such that is invariant by $a_{t}^{0}$

the image. By Lemma 8.2 and the construction, we can apply the par-
tition of unity associated to the coverings and to prove the $\{U_{j}^{1}\}$ $\{U_{j}^{2}\}$

following:
Lemma 8.3. There exists
$\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}$
: $U_{j_{1}}^{1}\times V_{k_{1}}\times V_{k_{2}}\times U_{j_{2}}^{2}\times[T, \infty)^{2}\times R\rightarrow A_{\ell,\delta}(a, b)$
,

such that

(8.3.2) $|\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}^{JJ}-\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}|<e(S_{1}, S_{2})$

,

(8.3.2) the maps $\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}$

can be patched together to give a map

$\Phi_{K_{1},K,K_{2}}$ : $\overline{K}_{1}\times_{SU(2)}\frac{SU(2)\times SU(2)}{U(1)}\times_{SU(2)}\overline{K}_{2}\times[T, \infty)^{2}\times R$

$\rightarrow B_{\ell,\delta}(a, b)$

.

By (8.1) we have:
 Floer Homology for Oriented 3-Manifolds 39

Lemma 8.4. Let $[A]\in Im\Phi_{K_{1},K,K_{2}}$ then

$|F^{A}+\overline{*}_{\sigma}F^{A}-grad_{a_{t}}f\wedge dt-*_{\sigma}grad_{a_{t}}f|_{L_{\ell}^{2}}<e(S_{1}, S_{2})$ .

We put
$|u|_{\ell,S_{1},S_{2},S}=|u|_{L_{\ell}^{2}(M\times R)}+|u|_{L_{\ell}^{1}(M\times S,S+S_{1}+S_{2})}$ .

Then we have also

Lemma 8.4. Let $[A]\in Im\Phi_{K_{1},K,K_{2}}$ then

$|F^{A}+\overline{*}_{\sigma}F^{A}-grad_{a_{t}}f\wedge dt-*_{\sigma}grad_{a_{t}}f|\ell,s_{1},s_{2},s<e(S_{1}, S_{2})$ .

We shall apply Taubes method as in [FU], to deform to $\Phi_{K_{1},K,K_{2}}$

a map to A4 $(a, b)$ . For this purpose, the following estimate is essential.
Lemma 8.5. There exists $\lambda>0$ independent of $S_{i}$
such that if
$A\in Im\Phi_{K_{1},K,K_{2}}$ , we have
$u\in\Omega_{\ell}^{2}$

$|D_{A}D_{A}^{*}u|_{L_{\ell-2}^{2}}>\lambda|u|_{L_{\ell}^{2}}$ .

This lemma is an immediate consequence of Lemma 3.9 and Remark

4.6. Furthermore since $a\rightarrow grad_{a}f$ is a map with respect to the $C^{2}$ $L_{\ell}^{2}$

norm for large , it follows that $\ell$

$grad_{a_{t}+u_{t}}f=grad_{a_{t}}f+(Hess_{a_{t}}f)(u_{t})+E(a, u)$

with
$|E(a, u)|_{L_{\ell}^{2}}\leq C|u|_{\ell}^{2}$

$|E(a, u)|_{l,S_{1},S_{2},S}\leq C|u|_{\ell,S_{1},S_{2},S}^{2}$ .

Hence we can apply the argument of [FU] pp.132-139, and obtain

Lemma 8.6. There exists $T_{0}$

, and $\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}$
: $ U_{j_{1}}^{1}\times V_{k_{1}}\times V_{k_{2}}\times$

$U_{j_{2}}^{2}\times[T_{0}, \infty)\times R\rightarrow\overline{M}(a, b)$

such that

(8.6.1) $\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}$
can be patched together to give a map

$\Phi_{K_{1},K,K_{2}}$ : $\overline{K}_{1}\times_{SU(2)}\frac{SU(2)\times SU(2)}{U(1)}\times_{SU(2)}\overline{K}_{2}\times[T, \infty)^{2}\times R$

$\rightarrow\sqrt{}\backslash \Lambda(a, b)$

.
40 K. Fukaya

(8.6.2) $|\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}^{JJ}-\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}|_{C^{1},\ell,S_{1},S_{2},S}<e(S_{1}, S_{2})$ .

The definition of the norm in (8.6.2) is as follows. $ U_{j_{1}}^{1}\times V_{k_{1}}\times V_{k_{2}}\times$

has a natural Riemannian metric. We define a norm

$U_{j_{2}}^{2}\times[T_{0}, \infty)\times R$

on by using
$A_{\ell,\delta}(a, b)$ -norm. Then the norm in (8.6.2) is
$(\ell, S_{1}, S_{2}, S)$

the -norm with respect to this metric and norm.

$C^{1}$

Note that the linear equation solved in [FU] pp.132-139 is gauge

invariant. (8.6.1) follows from this fact.
We shall prove that the map is an immersion, surjective $\Phi_{K_{1},K,K_{2}}$

to the set of standard model, and that injective.

Let , , , and
$g_{1}$ be an orthonormal
$g_{2}\in V_{k_{1}}$ $V_{k_{2}}$ $\Pi\subset T_{(g_{1},g_{2})}(V_{k_{1}}, V_{k_{2}})$

complement of $(U(1). (g_{1}, g_{2}))$ . $T_{(g_{1},g_{2})}$

Lemma 8.7. There exists $C$ independent of $S_{1}$

, $S_{2}$
such that, for
each $ v\in\Pi$ we have:
$|\Phi_{j_{1},j_{2},k_{1},k_{2}*}(v)|_{\ell,s_{1}s_{2},s}\geq C|v|$ ,

for suciently large $S_{i}$

. Here we choose $[A_{i}]\in U_{j_{i}}^{i}$
, $S_{i},S$ and regard
$\Pi\subset T_{([A_{1}],g_{1},g_{2},[A_{2}],S_{1},S_{2},S)}(U_{j_{1}}^{1}\times V_{k_{1}}\times V_{k_{2}}\times U_{j_{2}}^{2}\times[T, \infty)^{2}\times R)$ .

Remark 8.8. The lemma does not hold if we replace the $||\ell,s_{1},s_{2}$
, s-
norm by -norm, since and are reducible.
$L_{\ell}^{2}$
$c_{1}$ $c_{2}$

Proof. For simplicity, we put $g_{1}=g_{2}=1$ . Set

$A=\overline{\Phi}_{j_{i},j_{2},k_{1},k_{2}}([A_{1}], 1,1, [A_{2}], S_{1}, S_{2}, S)$

$v=(\overline{v}_{1}, \overline{v}_{2})\in su(2)\oplus su(2)$ .

Define $v_{i}$ : $R\rightarrow su(2)$ , by

$v_{i}(t)=\frac{d}{ds}J_{k_{i}}^{i}(1+s\overline{v}_{i}, t)|_{s=0}$

Then by definition
$(d^{A_{1}}v_{1})(x, t-S)$

for $t<S$
(8.9) $\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}*}(v_{1}, v_{2})=\{$ $(d^{A_{2}}v_{2})(x, t -S_{1}-S_{2}-S)$

for $t>S+S_{1}+S_{2}$
0 otherwise.
Let the dierential form in the above formula be denoted by $w$ . Lemma
8.7 is a consequence of the following:
 Floer Homology for Oriented 3-Manifolds 41

Lemma 8.10. There exists $C$ such that

$|w-d^{A}u|_{\ell,S_{1},S_{2},S}>C(|v_{1}|+|v_{2}|)$

for each $u\in\Omega_{\ell+1}^{0}$

and suciently large $S_{i}$
.

(In the statement we omit $\delta$

, since $a$ and $b$

are irreducible.)

Proof. We prove by construction. Then we assume that we have

$\overline{v}_{i}^{n}\in su(2)$
with , and , such that
$|\overline{v}_{i}^{n}|=1$ $ S_{i}^{n}\rightarrow\infty$ $[A_{i}^{n}],u^{n}$

$\lim_{n\rightarrow\infty}|w^{n}-d^{A_{i}^{n}}u^{n}|_{\ell,s_{1}^{n},s_{2}^{n},s}=0$ .

Since and move on compact sets, we may assume that they are
$[A_{i}^{n}]$ $\overline{v}_{i}^{n}$

independent of . Hence we have $n$

$ S_{i}^{n}\rightarrow\infty$

$|w^{n}-d^{A^{n}}u^{n}|_{\ell,s_{1}^{n},s_{2}^{n},s}\rightarrow 0$ .

Here $w^{n}$
is as in (8.9) with $S_{i}=S_{i}^{n}$ , and

$A^{n}=\overline{\Phi}_{j_{1},j_{2},k_{1},k_{2}}([A_{1}], 1,1, [A_{2}], S_{1}^{n}, S_{2}^{n}, S)$

.

(Since everything is invariant by the $R$ action, we may assume that $S$

is independent of ) By construction, there exists independent of $n.$ $\alpha$ $n$

such that

(8.11)
$|d+A^{n}-d|_{C^{\ell}}<Ce^{-\beta_{1}(t)/C}$ if $ t\in S+\alpha$ , $[S+S_{1}^{n}-\alpha]$

$|d+A^{n}-d|_{C^{\ell}}<Ce^{-\beta_{2}(t)/C}$ if $t\in[S+S_{1}^{n}+\alpha, S+S_{1}^{n}+S_{2}^{n}-\alpha]$ ,

where

$\beta_{1}(t)=d(t, \partial[S+\alpha, S+S_{1}^{n}-\alpha])$

$\beta_{2}(t)=d(t, \partial[S+S_{1}^{n}+\alpha, S+S_{1}^{n}+S_{2}^{n}-\alpha])$ .

Hence, by (8.9), we have, for each $\alpha>\alpha$

, that

$|du^{n}|_{L_{\ell}^{1}(S+\alpha.S+S_{1}^{n}+S_{2}^{n}-\alpha)}<\epsilon_{n}+Ce^{-\alpha/C}$ ,
42 K. Fukaya

where $\epsilon_{n}\rightarrow 0$
. Therefore there exists $s_{1}^{n}$
, $s_{2}^{n}\in su(2)$ such that

$|u^{n}-s_{1}^{n}|_{C^{p\prime}}(x, t)<C\epsilon_{n}+Ce^{-\beta_{1}(t)/C}$

if $t\in[S+\alpha, S+S_{1}^{n}-\alpha]$

$|u^{n}-s_{2}^{n}|_{C^{\ell}}(x, t)<C\epsilon_{n}+Ce^{-\beta_{2}(t)/C}$

if $t\in[S+S_{1}^{n}+\alpha, S+S_{1}^{n}+S_{2}^{n}-\alpha]$ .

(This is the step we can not work with norm.) $L^{2}$

Then patching with and , we have , $u$ $s_{1}^{n}$ $s_{2}^{n}$ $u_{1}^{n}$ $u_{2}^{n}$
, $ u_{3}^{n}\in L_{\ell+1}^{2}(M\times$

$R$ , $su(2))$ such that

(8.12.1) $|d^{A_{1}}(v_{1}-u_{1}^{n})|_{C^{\ell}}<C\epsilon_{n}$

(8.12.2) $|d^{A_{2}}(v_{2}-u_{2}^{n})|_{C^{\ell}}<C\epsilon_{n}$

(8.12.3) $|d^{a_{t}^{0}}u_{3}|_{C^{p\prime}}<C\epsilon_{n}$

(8.12.1) $|u_{1}^{n}(t, x)-s_{1}^{n}|_{C^{\ell}}<Ce^{-t/C}$

(8.12.1) $|u_{2}^{n}(t, x)-s_{2}^{n}|_{C^{p\prime}}<Ce^{t/C}$

(8.12.1) $|u_{3}^{n}(t, x)-s_{2}^{n}|_{C^{\ell!}}<Ce^{-t/C}$

(8.12.7) $|u_{3}^{n}(t, x)-s_{1}^{n}|_{C^{\ell!}}<Ce^{t/C}$

( $u_{1}^{n},u_{2}^{n}$
, and $u_{3}^{n}$
are constructed from the restrictions of
to $S+$ $u^{n}$ $(-\infty,$

,
$S_{1}^{n}/3]$ ,
$[S+S_{1}^{n}+2S_{2}^{n}/3, \infty)$ , respectively.) $[S+2S_{1}^{n}/3, S+S_{1}^{n}+S_{2}^{n}/3]$

We may assume that $\lim s_{1}^{n}=s_{1}$ and $\lim s_{2}^{n}=s_{2}$ . Therefore, by

(8.12.3), (8.12.6),(8.12.7) and the fact $G_{a_{t}^{0}}=U(1)$ imply that $ s_{1}=s_{2}\in$
$u(1)\subset su(2)$ . ( $u(1)$ is a Lie algebra of $G_{a_{y}^{0}}=U(1).$ ) Hence, using the

fact that is perpendicular to $u(1)\subset su(2)\oplus su(2)$ , we can find

$(\overline{v}_{1}, \overline{v}_{2})$

such that
$t_{0}$

(8.13) $|v_{1}-u_{1}^{n}|(x, t_{0})>C$

or
$|v_{2}-u_{2}^{n}|(x, -t_{0})>C$ ,

for some $C$ independent of . Suppose, for example (8.13) holds. By $n$

scaling, we can find such that $(u^{n})$

$\infty>C_{2}>|(u^{n})|(x, t_{0})>C_{1}>0$

$|d^{A_{1}}(u^{n})|_{C^{\ell}}<\epsilon_{n}\rightarrow 0$
.
 Floer Homology for Oriented 3-Manifolds 43

Therefore, by taking a subsequence, converges to such that $(u^{n})$ $u$

$d^{A_{1}}u=0$ , with respect to the compact uniform topology. This contra-

dicts the irreducibility of . The proof of Lemma 8.10 is now complete. $A_{1}$

An estimate similar to Lemma 8.7 for $TK_{i}$ direction and $[T, \infty)^{2}\times R$

direction is easier. Then, combined with (8.6.2), they imply:

Lemma 8.14. If $V$ is a tangent vector of
$\overline{K}_{1}\times_{SU(2)}\frac{SU(2)\times SU(2)}{U(1)}\times_{SU(2)}\overline{K}_{2}\times[T, \infty)^{2}\times R$ ,

at $([A_{1}], g_{1}, g_{2}, [A_{2}], S_{1}, S_{2}, S)$ , then we have

$|\Phi_{K_{1},K,K_{2},*}(V)|_{\ell,S_{1},S_{2},S}>C|V|$ .

Lemma 8.14 implies that $\Phi_{K_{1},K,K_{2}}$ is of maximal rank.

Remark 8.15. By H\"olders inequality, we have

$||\ell,s_{1},s_{2},s<C(S_{1}+S_{2})||_{L^{2}}$ .

Hence, Lemma 8.14 implies

$|\Phi_{K_{1},K,K_{2}*}(v)|_{L^{2}}>\frac{C|v|}{S_{1}+S_{2}}$ .

It seems that this reflects the fact that the sectional curvature $K$ of
at $\Phi(A_{1}, g_{1}, g_{2}, A_{2}, S_{1}, S_{2}, S)$ is estimated as $|K|<C(S_{1}+S_{2})^{2}$ .
$\mathcal{M}(a, b)$

Lemma 8.16. For each $C$ , there exist $ T,S,\epsilon$

, such that if $[A]$ is $a$

standard model of type $(K_{1}, K, K_{2}, T, \epsilon, C)$ , then

$[A]\in\Phi_{K_{1},K,K_{2}}(\overline{K}_{1}\times c_{c_{1}}\overline{K}\times c_{c_{2}}\overline{K}_{2}\times[S, \infty)^{2}\times R)$

.

Proof. The definition of the standard model implies that there exist
$[A_{1}],[A_{2}],g_{1},g_{2},S_{1},S_{2},S$ such that

$|\overline{\Phi}_{i_{1},i_{2},k_{1},k_{2}}([A_{1}], g_{1}, g_{2}, [A_{2}], S_{1}S_{2}, S)-A|_{L_{\ell}^{2}}<e(S_{1}, S_{2})$ .

Here is a representative of $A$ , and

$A$ , . Let : $A_{j}\in U_{i_{j}}$ $g_{j}\in V_{i_{j}}$
$\ell$
$[0, 1]\rightarrow$

be the straight line connecting them. The length of is smaller

$A_{\ell,\delta}(a, b)$
$\ell$

than $e(S_{1}, S_{2})$ . By [FU] pp. 132-139, we can deform this path to a path $\ell$

in connecting
$\overline{\mathcal{M}}(a, b)$
and $A$ . The
$\overline{\Phi}_{i_{1},i_{2},k_{1},k_{2}}([A_{1}], g_{1}, g_{2}, [A_{2}], S_{1}, S_{2}, S)$
44 K. Fukaya

length of is also estimated by

$\ell$
$e(S_{1}, S_{2})$ . By using Lemma 8.14, we can
lift this path to $\overline{\ell}:[0,1]\rightarrow\overline{K}_{1}\times_{G_{c_{1}}}\overline{K}\times_{G_{c_{2}}}\overline{K}_{2}\times[T, \infty)^{2}\times R$
such that
$\overline{\ell}(0)=([A_{1}], g_{1}, g_{2}, [A_{2}], S_{1}, S_{2}, S)$
. Therefore

$\Phi_{K_{1},K,K_{2}}(\overline{\ell}(1))=[A]$
,

as required.
Finally we shall prove that $\Phi_{K_{1},K,K_{2}}$ is injective.
Lemma 8. 17. If
$\Phi_{K_{1},K,K_{2}}([A_{1}], g_{1}, g_{2}, [A_{2}],S_{1}, S_{2}, S)=$

$\Phi_{K_{1},K,K_{2}}([A_{1}], g_{1}, g_{2}, [A_{2}], S_{1}, S_{2}, S)$

then
$|A_{\dot{x}}-A_{i}|\ell,s_{1},s_{2},s<e(S_{1}, S_{2})$

$|S_{i}-S_{i}|<e(S_{1}, S_{2})$

$|S-S|<e(S_{1}, S_{2})$ ,

and there exists $h\in SU(2)$ such that

$|hg_{i}-g_{i}|<e(S_{1}, S_{2})$ .

The proof is similar to the proof of Lemma 8.7. Suppose

Proof.
, ,
$A_{j}\in U_{i_{j}}$ , . The proof of the statement on
$A_{j}\in U_{i_{j}}$ $g_{j}\in V_{k_{j}}$ $g\in V_{k_{j}}$
$S_{i}$

and is easy, then we assume that $S_{i}=S_{i}$ , $S=S$ , for simplicity. By

$S$

assumption, there exists a gauge transformation : $M\times R\rightarrow SU(2)$ $\hat{g}$

such that
$\hat{g}^{*}\overline{\Phi}_{i_{1},i_{2},k_{1},k_{2}}([A_{1}], g_{1}, g_{2},[A_{2}], S_{1}, S_{2}, S)=$

$\overline{\Phi}_{i_{1},i_{2},k_{1},k_{2}}([A_{1}], g_{1}, g_{2}, [A_{2}], S_{1}, S_{2}, S)$

.

Then
$|\hat{g}^{*}\overline{\Phi}_{i_{1},i_{2},k_{1},k_{2}}([A_{1}], g_{1}, g_{2}, [A_{2}], S_{1}, S_{2}, S)-$

$\overline{\Phi}_{i_{1},i_{2},k_{1},k_{2}},([A_{1}], g_{1}, g_{2}, [A_{2}], S_{1}, S_{2}, S)|\ell,s_{1},s_{2},s<e(S_{1}, S_{2})$ .

Therefore, we have
$Ce^{-\beta_{1}(t)/C}$
if $t$
$\in[S+\alpha, S+S_{1}-\alpha]$
$|d\hat{g}|_{C^{\ell}}<\{$

$Ce^{-\beta_{2}(t)/C}$
if $t$
$\in[S+S_{1}+\alpha, S+S_{1}+S_{2}-\alpha]$ .
 Floer Homology for Oriented 3-Manifolds 45

Here $\beta_{i}$
is as in (8.11). Hence we have $g_{i}^{0}\in SU(2)$ such that

$|\hat{g}-g_{1}^{0}|<Ce^{-\beta_{1}(t)/C}$
if $t$
$\in[S+\alpha, S+S_{1}-\alpha]$

$|\hat{g}-g_{2}^{0}|<Ce^{-\beta_{2}(t)/C}$
if $t\in[S+S_{1}+\alpha, S+S_{1}+S_{2}-\alpha]$ .

Hence as in the proof of Lemma 8.10, we obtain $\hat{g}_{i}$

: $M\times R\rightarrow SU(2)$ ,

$i=1,2,3$ , such that

(8.18.1) $|(\hat{g}_{1}J_{k_{1}}^{1}(g_{1}, \cdot))^{*}A_{1}-J_{k_{1}}^{1},(g_{1}, \cdot)^{*}A_{1}|_{L_{p}^{2}}<e(S_{1}, S_{2})$

(8.18.2) $|(\hat{g}_{2}J_{k_{2}}^{2}(g_{2}, \cdot))^{*}A_{2}-J_{k_{2}}^{2}(g_{2}, \cdot)^{*}A_{2}|_{L_{\ell}^{2}}<e(S_{1}, S_{2})$

(8.18.3) $|\hat{g}_{3}^{*}a_{t}^{0}-a_{t}^{0}|_{L_{\ell}^{2}}<e(S_{1}, S_{2})$

and
(8.18.4) $|\hat{g}_{1}(x, t)-g_{1}^{0}|_{C^{\ell}}<Ce^{-t/C}$

(8.18.5) $|\hat{g}_{2}(x, t)-g_{2}^{0}|_{C^{\ell}}<Ce^{t/C}$

(8.18.6) $|\hat{g}_{3}(x, t)-g_{2}^{0}|_{C^{p}}<Ce^{-t/C}$

(8.18.7) $|\hat{g}_{3}(x, t)-g_{1}^{0}|_{C^{\ell}}<Ce^{t/C}$

(8.18.3),(8.18.6),(8.18.7) and $G_{a_{t}^{0}}=U(1)$ implies that we have $h\in U(1)$

such that
$|g_{i}^{0}-h|<e(S_{1}, S_{2})$ .
Hence (8.18.1),(8.18.2),(8.18.4),(8.18.5) and the irreducibility of $A_{i}$
, $A_{i}$

imply
$|g_{i}-hg_{i}|<e(S_{1}, S_{2})$

$|A_{i}-A_{\dot{x}}|_{L_{p}^{2}}<e(S_{1}, S_{2})$ .

The proof of Lemma 8.17 is now complete.

Lemma 8.19. For suciently large $T$ , the map $\Phi_{K_{1},K,K_{2}}$ is in-
jective.

Proof.
Let , , , , , , , $A_{i}$
be as in the proof of Lemma
$A_{i}$
$g_{i}$
$g_{i}$ $S_{i}$ $S_{i}$ $S$ $S$

8.17. Replacing by , we may assume that $|g_{i}-g_{i}|<e(S_{1}, S_{2})$ .

$g_{i}$
$hg_{i}$

Hence we can find a path : $\ell$

$[0, 1]\rightarrow\overline{K}_{1}\times_{G_{c_{1}}}\overline{K}\times_{G_{c_{2}}}\overline{K}_{2}\times[T, \infty)^{2}\times$

$R$ connecting $([A_{1}], g_{1}, g_{2}, [A_{2}], S_{1}, S_{2}, S)$ and . $(A_{1}, g_{1}\prime, g_{2}\prime, A_{2}, S_{1}, S_{2}, S)$

The length of is smaller than $e(S_{1}, S_{2})$ . We may assume that

$\ell$

and $A_{j}$

are in the same

$A_{j}$ , and that and are in the same
$U_{j_{i}}^{i}$
. Therefore $g_{j}$
$g_{j}$ $V_{k_{j}}$

the map
:
$\overline{\ell}=\overline{\Phi}_{U_{j_{1}}^{1},U_{j_{2}}^{2},V_{j_{1}},V_{j_{2}}}\circ\ell$ $[0, ^{1}]\rightarrow\overline{\mathcal{M}}(a, b)$
46 K. Fukaya

is well defined. Note and the length of with respect to $\overline{\ell}(0)=\overline{\ell}(1)$

$\overline{\ell}$

the -norm is smaller than $e(S_{1}, S_{2})$ . Hence we can find $H$ :
$||p,S_{1},S_{2}$

such that
$D^{2}\rightarrow A_{\ell,\delta}(a, b)$ . By [FU] pp.132-139, we can $H|_{\partial D^{2}}=\overline{\ell}$

deform $H$ to $H$ : such that $H=H$ on . Since

$D^{2}\rightarrow\overline{\mathcal{M}}_{\ell,\delta}(a, b)$ $\partial D^{2}$

the diameter of $H(D^{2})$ is smaller than $e(S_{1}, S_{2})$ , we can lift $H$ to
, by Lemma 8.14. We conclude $\ell(0)=$
$\overline{K}_{1}\times_{G_{c_{1}}}\overline{K}\times_{G_{c_{2}}}\overline{K}_{2}\times[T, \infty)^{2}\times R$

. The proof of Lemma 8.19 is complete.

$\ell(1)$

Thus, we have proved that the set of the standard model of type
in
$(K_{1}, K, K_{0}, T, \epsilon, C)$ is parametrized by $\Lambda 4(a, b)$

$\overline{K}_{1}\times_{SU(2)}\frac{SU(2)\times SU(2)}{U(1)}\times_{SU(2)}\overline{K}_{2}$ .

We divide it by $G_{a}\times G_{b}=\{\pm 1\}\times\{\pm 1\}$ and obtain

$\overline{K}_{1}X_{SU(2)}\frac{SO(3)\times SO(3)}{U(1)}\times_{SU(2)}\overline{K}_{2}$ .

This proves Theorem 7.6, in our case. The proof of the general case is
the same, but the notations will be more complicated.

Remark 8.20. It seems that the proofs of Lemmas 8.17 and

8.19 reflect the fact that the injectivity radius of at $\overline{\mathcal{M}}(a, b)$

$\Phi_{K_{1},K,K_{2}}([A_{1}], g_{1}, g_{2}, [A_{2}], S_{1}, S_{2}, S)$ is larger than . $C(\frac{1}{|S_{1}|+|S_{2}|})$

\S 9. Decay estimate

In this section we shall prove Theorem 7.5. This theorem corre-

sponds to [FU] \S 9. There Weitzenbeck formula was used for the proof.
We can not use it here because, in our case, $M$ is not and because $S^{3}$

we perturbed the equation.

Lemma 9.1. There exist $\epsilon,\lambda$

and independent of $T$ such that
$C$

if $d+a_{t}$ is a $su(2)$ connection on $M\times[-T, T]$ without component, $dt$

$c\in Fl$ and

if
(9.2.1) $|a_{t}-c|_{L_{l}^{2}}<\epsilon$

(9.2.1) $\partial a_{t}=*_{\sigma}F^{a_{t}}-grad_{a_{t}}f$

$\overline{dt}$

(9.2.3) $d_{c}^{*}a_{0}=0$ ,
 Floer Homology for Oriented 3-Manifolds 47

then we have

(9.3) $|a_{t}-c|_{L_{\ell}^{2}}\leq Ce^{-\lambda\beta_{T}(t)}$ .

Here $\beta_{T}(t)=\inf\{T-t, T+t\}$ .

Proof. We put $u(t)=a_{t}-c$ . We have

$*_{\sigma}F^{c+u(t)}-grad_{c}f$

$=*_{\sigma}d_{c}u(t)-Hess_{c}f(u(t))+E(u(t))$ ,

with

(9.4) $|E(u(t))|_{L_{p}^{2}}\leq C|u(t)|_{L_{\ell}^{2}}^{2}$ ,

for suciently large . Decompose $\ell$

$u(t)=\alpha(t)+\beta(t)$ with

$d_{c}^{*}\alpha(t)=0$
$\{$

$\beta(t)\in Imd_{c}$

Then we have

(9.5.1) $|\alpha(t)|_{L_{p}^{2}}<C\epsilon$ , $|\beta|_{L_{\ell}^{2}}<C\epsilon$ ,

(9.5.2) $\frac{\partial\alpha(t)}{\partial t}=*_{\sigma}d_{c}\alpha(t)-Hess_{c}f(\alpha(t))+E_{1}(\alpha(t), \beta(t))$

(9.5.3) $\frac{\partial\beta(t)}{\partial t}=E_{2}(\alpha(t), \beta(t))$

,

with

(9.6) $|E_{i}(\alpha(t), \beta(t))|_{L_{p}^{2}}<C(|\alpha(t)|_{L_{p}^{2}}+|\beta(t)|_{L_{\ell}^{2}})^{2}$

We decompose
$\alpha(t)=\alpha_{+}(t)+\alpha_{-}(t)$ ,

where ,
$\alpha_{+}$ belong to the spaces spanned by positive and negative
$\alpha_{-}$

eigenspaces $of*_{\sigma}d_{c}-Hess_{c}$ , respectively. (Note that by Lemma 2.8,

$f$

zero is not an eigenvalue $of*_{\sigma}d_{c}-Hess_{c}f.$ ) We put $g\pm(t)=|\alpha\pm(t)|_{L^{2}}$ ,

$h(t)=|\beta(t)|_{L^{2}}$ . By (9.2.2) and (9.4), we have

$|E_{1}(\alpha(t), \beta(t))|_{L^{\infty}}<C(g+(t)+g-(t)+h(t))^{2}$ .
48 K. Fukaya

Therefore, we have

(9.7.1) $\frac{dg_{+}}{dt}\geq\lambda g_{+}-C_{0}(g_{-}+h)^{2}$

,

(9.7.2) $\frac{dg-}{dt}\leq-\lambda g-+C_{0}(g_{+}+h)^{2}$

,

(9.7.3) $|\frac{dh}{dt}|\leq C_{0}(g_{+}+g-+h)^{2}$ .

Hence, by elliptic regularity, it suces to show the following:

Sublemma 9.8. There exists a constant $C$ and depending only $\epsilon$

on and and is independent of $T$ such that if $g_{+},g-$ and be non-

$C_{0}$ $\lambda$
$h$

negative functions satisfying (9.7.1)-(9.7.3) and

(9.7.4) $|g\pm(t)|<\epsilon$ , $|h(t)|<\epsilon$ ,

(9.7.5) $h(0)=0$ ,

then

(9.9) $|g\pm(t)|$ , $|h(t)|<Ce^{-\lambda\beta_{T}(t)}$ .

Proof First we replace the assumption (9.7.5) by $|h(0)|<\delta$ , and

prove
$|g\pm(t)|$ , $|h(t)|<C(e^{-\lambda\beta_{T}(t)}+\delta)$ .

when ,$\delta^{2}T<\mu_{0}$ $\epsilon T<\mu_{0}$

for some $\mu_{0}$ depending only on $C_{0}$
and $\lambda$

. For
this purpose we prove

$|h|<C_{0}(\epsilon^{n}+\epsilon e^{-\lambda\beta\tau(t)}+\delta)$
$(9.10.2n)$

$(9.11.2n.\pm)$ $|g\pm|<C_{0}(\epsilon^{7L}+\epsilon e^{-\lambda\beta_{T}(t)}+\delta)$

by an induction on . (Here $n$ $n$ is a half integer.) Assume $(9.10.2n)$ . Let

$t_{0}\in[-T, T]$ . We put

$\hat{g}_{+}(t)=e^{-\lambda(t-t_{0})}g_{+}(t)$
.

Then, by (9.7.1),(9.7.4),(9.10.2n), and $(9.11.2n -1,\pm)$ , we have:

$\epsilon e^{-\lambda(T-t_{O})}\geq\hat{g}+(T)$

$\geq g_{+}(t_{0})-\int_{t_{O}}^{T}C_{0}^{3}e^{-\lambda(t-t_{0})}(\epsilon^{n-1/2}+\epsilon e^{-\lambda\beta_{T}(t)}+\delta)^{2}dt$

.
 Floer Homology for Oriented 3-Manifolds 49

$(9.11.2n,+)$ follows. For the proof of $(9.11.2n,-)$ , we use $=$ $\hat{g}_{-}$

in a similar way.
$e^{\lambda(t-t_{0})}g_{-}(t)$

It is easy to see that $(9.10.2n)$ and $(9.11.2n)$ imply $(9.10.2n+1)$ .

For general $T$ , we proceed as follows. Apply the first step to $T_{0}=$
, and $\delta=0$ . We have $h(3T_{0}/4)<C_{0}e^{-T_{0}\lambda/4}$ . Then we apply the
$\mu_{0}/\epsilon$

first step to $g\pm(t-3T_{0}/4)$ , $h(t-3T_{0}/4)$ and $T=T_{0}$ . We obtain

$\sup_{0<t<4T_{0}/3}|g\pm(t)|<C_{0}e^{-5T_{0}\lambda/12}$

$\sup_{0<t<4T_{0}/3}|h(t)|<C_{0}e^{-5T_{0}\lambda/12}$ ,

if $3T_{0}/2<T$ . And similarly for $-4T_{0}/3<t<0$ . Hence we can apply

the first step to $T$ $=4T_{0}/3$ . Iterating this, we obtain the desired result.
The proof of Lemma 9.1 is now complete.

Lemma 9.12. For each $\delta$

, $C$ , there exists $\epsilon$

such that if $ a\in$

$A_{\ell}(M)$ ,

$|*_{\sigma}F^{a}-grad_{a}f|_{L_{\ell}^{2}}<\epsilon$

$|a|_{L_{\ell}^{2}}<C$ ,

then there exists $c\in Fl$ and $g\in \mathcal{G}_{\ell+1}$

such that

$|g^{*}a-c|_{L_{\ell}^{2}}<\delta$ .

Proof. If not, there exists $a_{i}\in A_{l}(M)$ and $\delta>0$ , such that

(9.13.1) $\lim_{i\rightarrow\infty}|*_{\sigma}F^{a_{i}}-grad_{a_{i}}f|_{L_{\ell}^{2}}=0$ ,

(9.13.2) $|a_{i}|_{L_{p}^{2}}<C$ ,
(9.13.3) $|g_{i}^{*}a_{i}-c|_{L_{\ell}^{2}}>\delta$

for each , $g_{i}\in G_{\ell+1}$ , and $c\in Fl$ . (9.13.2) implies that, by taking
$i$

a subsequence, converges to an element

$a_{i}$ of . Then, $a_{\infty}$ $A_{\ell-1,\delta}(a, b)$

(9. 13. 1) implies that

$|*_{\sigma}F^{a_{\infty}}-grad_{a_{\infty}}f|_{L_{\ell}^{2}}=0$ .

Hence there exists $g_{i}\in \mathcal{G}_{\ell+1}(M)$

and $c\in Fl$ such that converges to $g_{i}^{*}a_{i}$

in $A_{\ell-1}(M)$ . By replacing
$c$ $g_{i}$ if necessary, we may assume that

(9.14) $d_{c}^{*}(g_{i}^{*}a_{i}-c)=0$ .
50 K. Fukaya

(See $FU.$ ) By (9.13.1) we have

(9.15) $\lim_{i\rightarrow\infty}|*_{\sigma}F^{g_{i}^{*}a_{i}}-grad_{g_{i}^{*}a_{i}}f|_{L_{p}^{2}}=0$ .

By (9.14),(9.15), $\lim|g_{i}^{*}a_{i}-c|_{L_{\ell-1}^{2}}=0$ , and an elliptic estimate, we have

(9.16) $\lim_{i\rightarrow\infty}|g_{i}^{*}a_{i}-c|_{L_{\ell}^{2}}=0$ .

(9.16) contradicts (9.13.3).

Using this lemma, we can improve Lemma 9.1 as follows.
Lemma 9.17. There exists , and $C$ , such that if $d+a_{t}$ be $ T_{0},\epsilon,\lambda$

a $su(2)$ -connection on $M\times[-T, T]$ without component, and if $dt$

(9.18.1) $T>T_{0}$

(9.18.2) $\partial a_{t}=*_{\sigma}F^{a_{t}}-grad_{a_{t}}f$

$\overline{\partial t}$

(9.18.3) $|\frac{\partial a_{t}}{\partial t}|_{L_{\ell}^{2}}<\epsilon$ ,

then there exists $c\in Fl$ and $g\in \mathcal{G}_{\ell+1}(M)$ such that

(9.19) $|g^{*}a_{t}-c|_{L_{\ell}^{2}}<Ce^{-\lambda\beta_{T}(t)}$ .

Here is regarded as a gauge

$g$
transformation on $M\times R$ independent of
the $R$ factor. The constants $ C,\epsilon,\lambda$
are independent of $T$ .

ProofLet be the number determined in Lemma 9.1, and be

$\epsilon_{0}$
$S$

a suciently large positive number determined later. Put . $\delta=\epsilon_{0}/2S$

Then we obtain by Lemma 9.12. We may assume that

$\epsilon$

. By $\epsilon<\delta$

Lemma 9.12, we obtain $c\in Fl$ . Replacing by gauge transformation $a_{t}$

independent of , we may assume that

$t$

(9.20.1) $|a_{0}-c|_{L_{\ell}^{2}}<\delta$

(9.20.2) $d_{c}^{*}(a_{0}-c)=0$ .

By (9.20.1),(9.18.3), and $2S\epsilon<\epsilon_{0}$

, we can apply Lemma 9.1 to $ M\times$

$[-S, S]$ , and obtain

$|a_{t}-c|_{L_{p}^{2}}<Ce^{-\lambda\beta_{S}(t)}$ .
Hence by taking $S$
suciently large, we have
(9.21.1) $|a_{3S/4}-c|_{L_{\ell}^{2}}<\epsilon_{0}/K$

(9.21.2) $|a_{-3S/4}-c|_{L_{\ell}^{2}}<\epsilon_{0}/K$ .
 Floer Homology for Oriented 3-Manifolds 51

Here $K$ is a suciently large positive number determined later. There-

fore there exists such that
$g\in \mathcal{G}_{\ell+1}(M)$

$|g-1|_{L_{p}^{2}}<C\epsilon_{0}/K$

$d_{c}^{*}(g^{*}a_{3S/4}-c)=0$

$|g^{*}a_{3S/4}-c|_{L_{p}^{2}}<C\epsilon_{0}/K$ .

Here $C$ depends only on $M$ . Hence we can apply Lemma 9.1 to
$g^{*}a_{t+3S/4}$ , on $M\times[-S, S]$ . By choosing suciently large, we obtain $S$

$|g^{*}a_{t}-c|_{L_{\ell}^{2}}<C\epsilon_{0}/K$ ,

for $t$
$\in[0,4S/3]$ , provided $3S/2<T$ . By taking $K$ suciently large, we
have
$|a_{t}-c|_{L_{\ell}^{2}}<\delta$ ,

for $t$
By using (9.21.2) we have the same estimate for
$\in[0,4S/3]$ . $ t\in$

$[-3S/4,0]$ . Hence we can apply Lemma 9.1 to $M\times[-4S/3,4S/3]$ if

$3S/2<T$ . Repeating this we obtain the lemma.

Lemma 9.22. There exists $\theta>0$ such that, if $[A]\in \mathcal{M}_{\delta}(a, b)$

with $\mu(a)\neq\mu(b)$ , and if $g^{*}A=d+a_{t}$ , where $d+a_{t}$ is a connection

without factor, then we have
$dt$

$\int_{M\times R}|\frac{\partial a_{t}}{dt}|^{2}dxdt>\theta$

.

By [F] pi22, the integral in the lemma is independent of

Proof. $A$

but depends only on and . Hence the lemma follows from (2.8.1).
$a$
$b$

Theorem 7.5. Fix , $b\in Fl$ . Put $k_{0}=\mu(a)-\mu(b)$ . We

Proof of $a$

shall prove that, for each $\mu(a)\geq\mu(c)\geq\mu(c)\geq\mu(b)$ there exists , $K_{c,c}$

such that the conclusion of Theorem 7.5 holds for

$\epsilon=\mathring{\frac{\epsilon(K_{a,c},,K_{c_{k},b})}{2^{k}}}\cdots$

$ T=\frac{T(K_{a,c_{0}},,K_{c_{k},b})}{2^{k}}\cdots$
.

The proof is by induction on . The first step is obvious, since $k$ $\overline{\mathcal{M}}^{J}(c, c)$

is a finite set if $\mu(c)=\mu(c)+1$ . Hence it is enough to show the last

52 K. Fukaya

step of the induction. We assume that the last step is false. Then we
have , such that
$A_{i}\in\overline{\mathcal{M}}(a, b)$

(9.23.1) $\sup|F^{A_{i}}|<\Lambda$ ,
(9.23.2) $[A_{i}]$
is unbounded in $\overline{\mathcal{M}}(a, b)$
,
(9.23.3) non of $A_{i}$
is a standard model.

Let be a gauge transform such that

$g_{i}$
$g_{i}^{*}A_{i}=d+a_{t}^{i}$
has no $dt$
component.
We have
$\frac{da_{t}^{i}}{dt}=*_{\sigma}F^{a_{t}^{i}}-grad_{a_{t}^{i}}f$
.

If

$|\frac{\partial a_{t}^{i}}{dt}|_{L_{\ell}^{2}}<\epsilon$
,

were true for each , then Lemma 9.17 would imply that $t$
for some $a_{t}^{i}=c$

$c\in Fl$ . It would follow that $a=b$ . This is a contradiction. Hence there

exists such that

$t_{i}^{1}$

$|\frac{\partial a_{t_{i}^{1}}^{i}}{dt}|_{L_{\ell}^{2}}>\epsilon$ .

$t_{i}^{l_{1}}$
$t_{i}^{l.-1}$

.
$t_{i}^{1}$

.
$t_{i}^{2}$

$\ldots$

Figure 5.

Lemia 9.24. There exists $L$

independent of $i$
, and there exist $T_{i}$
,
 Floer Homology for Oriented 3-Manifolds 53

$t_{i}^{1}$

, $\ldots,t_{i}^{\ell_{i}}$
, such that

(9.24.1) $\ell_{i}<L$ ,
(9.24.2) $\lim T_{i}=\infty$ ,

(9.24.3) $|\frac{\partial a_{t}^{i}}{\partial t}|_{L_{\ell}^{2}}<\epsilon$

$if|t-t_{i}^{j}|>T_{i}$
for each , $i$

(9.24.4) $|t_{\dot{x}}^{j}-t_{i}^{j}|>T_{i}$
if $j\neq j$ .

Proof.
The existence of the upperbound of independent of $L$ $\ell_{i}$

is the essential part of the statement. Hence, if Lemma 9.24 is not

$i$

true, then, by taking a subsequence, we may assume that there exist

, ,
$t_{i}^{1}$

, such that (9.24.2),(9.24.4) and

$\cdots$ $t_{i}^{\ell_{i}}\in R$ $T_{i}$

(9.24.5) $\lim\ell_{i}=\infty$

(9.24.6) $|\frac{\partial a_{t_{i}^{j}}^{i}}{dt}|_{L_{p}^{2}}>\epsilon$

hold. By $|a_{t}|<\Lambda$ , and by Uhlenbecks theorem [FU] p117, we can find

$g_{i}^{j}\in \mathcal{G}_{\ell+1}(M)$
such that a subsequence of the connection

$t\mapsto g_{i}^{j*}a_{t-t_{i}^{j}}^{i}$ ,

converges to an element $d+a_{j,t}^{\infty}$ of , for fixed , (in topology

$\lambda\Lambda(c_{j}, c_{j})$ $j$
$C^{2}$

on any compact set.) Here $c_{j}$ , $c_{j}\in Fl$ . By (7.24.6), we have . $c_{j}\neq c_{j}$

Hence by Lemma 9.22

$\int_{M\times R}|\frac{\partial a_{j,t}^{\infty}}{dt}|^{2}dt>\theta$

,

for each . Therefore, Fatous lemma implies

$j$

$\lim_{i\rightarrow\infty}\int|\frac{\partial a_{t}^{i}}{\partial t}|^{2}dt$

$\geq\sum_{j=1}^{\infty}\int_{M\times R}|\frac{\partial a_{j,t}^{\infty}}{dt}dt|^{2}dt$

$=\infty$ .
54 K. Fukaya

This contradicts the fact that

$\int|\frac{\partial a_{t}^{i}}{dt}|^{2}dt$

is independent of but depends only on $i$

$a$ and . The proof of the lemma
$b$

is complete.

By Lemma 9.24 and $|F^{a_{i}}|<\Lambda$ , we can take a subsequence such

that the following holds : $\ell_{i}=\ell$
is independent of : let : $i$
$\hat{a}_{t}^{i,j}=a_{t-t_{i}^{j}}^{i}$

there exists such that converges to an element

$g_{i,j}$
$\lim_{i\rightarrow\infty}g_{i,j}^{*}\hat{a}_{t}^{i,j}$ $a_{i}^{\infty,j}$

of uniformly on every compact set, for some , . If $\ell=1$ ,

$\Lambda 4(c_{j}, c_{j})$ $c_{j}$ $c_{j}$

we can easily prove that is bounded in . This contradicts

$A_{i}$ $\mathcal{M}(a, b)$

(9.23.2). On the other hand, by induction hypothesis, is either an $\hat{a}_{t}^{\infty,j}$

element of , or a standard model. Therefore, using Lemma 9.17

$K_{c_{j},c_{j}}$

and (9.24.3), we can prove that is a standard model for large . This $A_{i}$ $i$

contradicts (9.23.3). The proof of Theorem 7.5 is now complete.

\S 10. Local action on the end of moduli space

Using the results in 8,9, we obtain charts : for $\Phi_{c}$

$X(c)\rightarrow\overline{\mathcal{M}}(a, b)$

each . As we pointed out in \S 7 these charts are not compatible. Then

$c$

we have to perturb them. Also, in order to extend bundles to the $\mathcal{L}_{i}^{(2)}$

boundary, we have to examine its behaviour on the image of each chart.

For these purposes, it is useful to use the notion, local action of groups,
which is a generalization of one introduced by Cheeger-Gromov [CG].
They used the local action to study the end of Riemannian manifolds
with bounded curvature. In their case, a special kind of local action,
$F$ -structure, (that is the local action of Torus,) arises, and the direction

of the orbits is the collapsed one. In our case, the curvature is not
bounded from above. (It might be bounded from below.) Hence the
group acting on the end is not necessary Abelian. (The group $SU(2)$
arises as well.) However the end is also collapsed and the collapsed
direction is homogeneous. (For example, in the case we studied in \S 8,
the collapsed direction is parametrized by SO(3) $\times SO(3)/S^{1}$ .
Before stating our result we shall discuss examples. First consider
the case, when $G_{a}=G_{b}=\{\pm 1\}$ , $G_{c}=G_{c}=U(1)$ , $\mu(a)>\mu(c)>$
$\mu(c)>\mu(b)$ . Choose a compact subset of , consisting $K_{c,c}$
$\overline{\mathcal{M}}^{J}(c, c)$

of irreducible connections. Then, by Theorem 7.6, the intersection of

and a neighborhood of $K_{a,c}\times K_{c,c}\times K_{c,b}$ in
$\overline{\Lambda 4}^{J}(a, b)$
is $C\overline{\mathcal{M}}(a, b)$
 Floer Homology for Oriented 3-Manifolds 55

dieomorphic to

$G_{a}\backslash \overline{K}_{a,c}\times_{G_{c}}\overline{K}_{c,c}\times c_{c}\overline{K}_{c,b}/G_{b}\times(T, \infty)^{2}$

.

On this set we can define an action of $U(1)\times U(1)=G_{c}\times G_{c}$ by

$(h, h)([x, y, z], t, s)=([xh, y, hz], t, s)$ .

Note that is a principal $\overline{K}_{a,c}\rightarrow K_{a,c}$

$U(1)$ bundle, hence $U(1)$ acts on
. As in \S 7, we have a map
$\overline{K}_{a,c}$

$\Phi_{(c,c),(c)}$ : $G_{a}\backslash \overline{K}_{a,c}\times_{G_{c}}\overline{K}_{c,c}\times c_{c}\overline{K}_{c,b}/G_{b}\times(T, \infty)^{2}$

$\rightarrow G_{a}\backslash \mathcal{M}(a, c)\times_{G_{c}}\overline{K}_{c,b}/G_{b}\times(T, \infty)$

$\Phi_{(c,c),(c)}$ : $G_{a}\backslash \overline{K}_{a,c}\times_{G_{c}}\overline{K}_{c,c}\times_{G_{c}}\overline{K}_{c,b}/G_{b}\times(T, \infty)^{2}$

$\rightarrow G_{a}\backslash \mathcal{M}(a, c)\times_{G_{c}}\overline{K}_{c,b}/G_{b}\times(T, \infty)$

Let $Z_{2}$
, $Z_{1}$
be inverse images of and $G_{a}\backslash \overline{K}_{a,c}\times c_{c}\overline{K}_{c,b}/G_{b}\times(T, \infty)$

respectively. (See Figure 6.)

$G_{a}\backslash \overline{K}_{a,c}\times c_{c}\overline{K}_{c,b}/G_{b}\times(T, \infty)$

has a $U(1)$ action. This action is iden-

$G_{a}\backslash \overline{K}_{a,c}\times_{G_{c}}\overline{K}_{c,b}/G_{b}\times(T, \infty)$

tified to the action on the second factor of $U(1)\times U(1)$ on . Similarly $Z_{2}$

the $U(1)$ action of is identified to the ac-

$G_{a}\backslash \overline{K}_{a,c}\times_{G_{c}}\overline{K}_{c,b}/G_{b}\times(T, \infty)$

tion of the first factor of $U(1)\times U(1)$ on . This is exactly the situation $Z_{1}$

of -structure defined in [CG].

$T$

$\{1\}\times U(1)-\uparrow$

action

$)(c)(W_{2})$

$\left(c,c & \prime\right)\times\overline{m}(\prime c,b)$

Figure 6
56 K. Fukaya

Next, consider the case, $G_{a}=G_{b}=\{\pm 1\}$ , $G_{c}=SU(2)$ . A neigh-

borhood of $K_{a,c}\times Kc,b$ in $C\overline{\mathcal{M}}^{J}(a, b)$
is dieomorphic to

$G_{a}\backslash \overline{K}_{a,c}\times_{SU(2)}\overline{K}_{c,b}/G_{b}\times(T, \infty)$

.

On this set $SU(2)$ does not has a global action, but has a local ac-
tion in the following sense. Consider the principal $SU(2)$ bundle :
. Let $SU(2)$ act on itself by conjugation, and
$\overline{K}_{a,c}\rightarrow\overline{K}_{a,c}/SU(2)$

be the associated bundle. $P$ has a structure of Lie

$P\rightarrow\overline{K}_{a,c}/SU(2)$

group bundle. $P$ induces a bundle . $P$ has a $P\rightarrow\overline{K}_{a,c}/G_{c}\times G_{c}\backslash \overline{K}_{c,b}$

fibrewise action to

$\overline{K}_{a,c}\times_{G_{c}}\overline{K}_{c,b}\rightarrow\overline{K}_{a,c}/G_{c}\times G_{c}\backslash \overline{K}_{c,b}$

,

induced from the fibrewise action of $P$ to from left. (Note $SU(2)$ $\overline{K}_{a,b}$

act globally on from right.) This fibrewise action defines a local

$\overline{K}_{ab}$

action. If $\mu(c)>\mu(c)>\mu(b)$ , the local action of $G_{c}=SU(2)$ can be

made to be compatible with the local action of . $G_{c}\times G_{c}$

Note that this action is not an action of a sheaf of groups in the

sense of [CG], because the fibre bundle is not $P\rightarrow\overline{K}_{a,c}/G_{c}\times G_{c}\backslash \overline{K}_{c,b}$

flat, in general.
Take a principal bundle and construct a Lie $\overline{K}_{c,b}\rightarrow SU(2)\backslash \overline{K}_{c,b}$

group bundle in a similar way. $Q$ has also a

$Q\rightarrow\overline{K}_{a,c}/G_{c}\times G_{c}\backslash \overline{K}_{c,b}$

fibrewise action on

$G_{a}\backslash \overline{K}_{a,c}\times_{SU(2)}\overline{K}_{c,b}/G_{b}\times(T, \infty)$

.

This action does not coincide to the action of $P$ . But they have the
same orbits. By convention, we use only the action of $P$ .

Definition 10.1. Let $X$ be a $C^{\infty}$

manifold. A local action on $X$

is a collection $(U_{i}, G_{i}, \varphi_{i,j})$ such that

(10.1.1) is an open covering of $X$ .

$U_{i}$

(10.1.2) . : is a smooth action of a Lie group on .

$G_{i}\times U_{i}\rightarrow U_{i}$ $G_{i}$ $U_{i}$

(10.1.3) is and invariant.

$U_{i}\cap U_{j}$ $G_{i}$ $G_{j}$

(10.1.4) Let $Em(G_{i}, G_{j})$ be the set of all injective homomorphisms.

For $i<j$ , there exists a smooth map : such $\varphi_{i,j}$
$\frac{U_{i}\cap U_{j}}{G_{i}}\rightarrow Em(G_{i}, G_{j})$

that
$g(x)=\varphi_{i,j}([x])(g)(x)$
 Floer Homology for Oriented 3-Manifolds 57

holds for each $x\in U_{i}\cap U_{j}$ , $g\in G_{i}$ .

Example 10.2. Let be a principal $G$ bundle. ( $G$ acts on $X\rightarrow N$

$X$ from right.) Let $P=XX_{ad}$ G. $P$ is a Lie group bundle and has a

fibrewise left action on $X$ . This gives a local action on $X$ .

Example 10.3. Let $\overline{X}^{o}(c)$

be as in \S 7. There exists a fibration

$\overline{X}^{o}(c)\rightarrow G_{a}\backslash \overline{K}_{a,c_{0}}/G_{c_{0}}\times\cdots\times G_{c_{k}}\backslash \overline{K}_{c_{k},b}/G_{b}\times(T(c), \infty)^{k+1}$

the fibre of which is $G_{a}\times G_{c_{0}}\times\cdots\times G_{c_{k}}\times G_{b}$ . We have a Lie group
bundle
$P\rightarrow G_{a}\backslash \overline{K}_{a,c_{0}}/G_{c_{0}}\times\cdots\times G_{c_{k}}\backslash \overline{K}_{c_{k},b}/G_{b}\times(T(c), \infty)^{k+1}$

whose fibre is $G_{a}\times G_{c_{0}}\times\cdots\times G_{c_{k}}\times G_{b}$ . The bundle $P$ has a fibrewise
action to $\overline{X}^{o}(c)$
. This gives a local action on $\overline{X}^{o}(c)$
.

Theorem 10.4. There exist a local action on $\overline{\mathcal{M}}^{J}(a, b)$

and maps

$\Psi_{c}$
: $\mathring{X}(c)\rightarrow\overline{\mathcal{M}}(a, b)$
,
$\Psi_{c,c}$ : $U(c, c)\rightarrow X(c)$ ,

such that
(10.4.1) The restriction by of the local action on $\Psi_{c}$
$\mathring{X}(c)$

of the local
action coincides to one in Example 10.3.
(10.4.2) . (The subset $U$ ( , $c)\subset X(c)$ is as
$\Psi_{c}\Psi_{c,c}=\Psi_{c}$ $c$
in\S 7.)
Theorem 7.1 follows immediately from Theorem 10.4. We have also

(10.5) $|\Phi_{c}-\Psi_{c}|(z)<e(S_{1}, \cdots, S_{k})$ .

Here $\Phi_{c}$
is the map constructed in \S 8, $z=([A_{1}, \cdots, A_{k}], S_{1}, \cdots, S_{k})$ and

$e(S_{1}, \cdots, S_{k})=\sum Ce^{-S_{i}/C}$ .

To prove Theorem 10.4, we modify the maps inductively on . $\Psi_{c}$ $c$

First we take which is maximal with respect to the inclusion and put
$c$

. We do not change
$\Psi_{c}=\Phi_{c}$ while modifying with . For $\Psi_{c}$ $\Phi_{c}$ $c\supset c$

simplicity of the notation, we discuss one step of modifications. We

consider the following case. Let $\mu(a)<\mu(c)<\mu(c)<\mu(b)$ , with
$G_{a}=\{\pm 1\}$ , $G_{c}=G_{c}=G_{b}=U(1)$ , and consider the component
58 K. Fukaya

of consisting of irreducible connections. Suppose, by induction

$K_{c,c}$

hypothesis, we have

$\Psi_{(c,c)}$ : $\overline{K}_{a,c}\times_{G_{c}}\overline{K}_{c,c}\times c_{c}\overline{K}_{c,b}\times(T, \infty)^{2}$

$\rightarrow\lambda 4(a, b)$

$\Psi_{(c,c),(c)}$ : $\overline{K}_{a,c}\times_{G_{c}}\overline{K}_{c,c}\times c_{c}\overline{K}_{c,b}\times(T, \infty)^{2}$

$\rightarrow\overline{\mathcal{M}}(a, c)\times_{G_{c}}\mathcal{M}(c, b)\times(T, \infty)$

$\hat{\Psi}_{(c,c),(c)}$
: $\overline{K}_{a,c}\times c_{c}\overline{K}_{c,c}\times c_{c}\overline{K}_{c,b}\times(T, \infty)^{2}$

$\rightarrow\overline{M}(a, c)\times c_{c}\mathcal{M}(c, b)\times(T, \infty)$

,

and a local action on the image of $\Psi_{(c,c)}$ . We shall define $\Psi_{(c)}$ and $\Psi_{(c)}$

such that
$\Psi_{(c)}\Psi_{(c,c),(c)}=\Psi_{(c,c)}$

on
$W_{1}=\Psi_{(c,c),(c)}^{-1}(\overline{K}_{a,c}\times_{G_{c}}\overline{K}_{c,b}\times[T, \infty))$ ,

and
$\Psi_{(c)}\Psi_{(c,c),(c)}=\Psi_{(c,c)}$

on
$W_{2}=\Psi_{(c,c),(c)}^{-1}(\overline{K}_{a,c}\times_{G_{c}}\overline{K}_{c,b}\times[T, \infty))$ .

(See Figure 6.) By induction hypothesis, and pre- $\Psi_{(c,c),(c)}$ $\Psi_{(c,c),(c)}$

serves and actions respectively. (In this case, those

$G_{c}\times G_{b}$ $G_{c}\times G_{b}$

actions are defined globally since the groups are abelian.) The maps
and
$\Psi_{(c)}$ we shall construct must be invariant. Once we obtain
$\Psi_{(c)}$ $G_{b}$

such maps and we can define a local action on their images

$\Psi_{(c)}$ $\Psi_{(c)}$

by pushing out one by those maps. These local actions can be patched
together with one on the image of by the and $\Psi_{(c,c)}$ $G_{c}\times G_{b}$ $G_{c}\times G_{b}$

invariance of the maps and . $\Psi_{(c,c),(c)}$ $\Psi_{(c,c),(c)}$

We begin the construction of . We choose an open coverings , $\Psi_{c}$ $U_{j}^{1}$

, ,
$U_{j}^{2}$
, of $K-a,c/G_{C}$ ,
$U_{j}^{3}$
, $U_{j}^{4}$
, respectively. $K_{c,c}$
$G_{c}\backslash \overline{K}_{c,b}/G_{b},\overline{K}_{a,c}/G_{c}$

Let be an open covering of $U(1)$ . Take maps

$V_{k}$
and as in \S 8. $J_{k}^{1}$ $J_{k}^{2}$

Choose sections : and . As in \S 8, define a map

$s_{j}^{1}$ $U_{j}^{1}\rightarrow A_{\ell}(a, c)$ $s_{j}^{2},s_{j}^{3}.s_{j}^{4}$

$\overline{\Phi}_{j_{1},j_{2},j_{3},k_{1},k_{2}}$
: $U_{j_{1}}^{1}\times U_{j_{2}}^{2}\times U_{js}^{3}\times V_{k_{1}}\times V_{k_{2}}\times(T, \infty)\times R\rightarrow A_{\ell,\delta}(a, b)$

by
$\overline{\Phi}_{j_{1},j_{2},j_{3},k_{1},k_{2}}([A_{1}], [A_{2}], [A_{3}], g_{1}, g_{2}, S_{1}, S_{2}, S)$
 Floer Homology for Oriented 3-Manifolds 59

$=(J_{k_{1}}(g_{1}, \cdot)^{*}A_{1})(x, t-S)$ for $t$

$<S+S_{1}/3$

$=\chi(\frac{t-S-S_{1}/3}{S_{1}/3})g_{1}^{*}A_{1}(x, t-S)$

$+(1-\chi(\frac{t-S-S_{1}/3}{S_{1}/3}))A_{2}(t-S-S_{1})$

for $t$
$\in[S+S_{1}/3, S+2S_{1}/3]$

$\{$ $=A_{2}(t-S-S_{1})$ for $t\in[S+2S_{1}/3, S+S_{1}+S_{2}/3]$

$=\chi(\frac{t-S-S_{1}-S_{2}/3}{S_{2}/3})A_{2}(t-S_{1}-S)$

$+(1-\chi(\frac{t-S-S_{1}-S_{2}/3}{S_{2}/3}))g_{2}^{*}A_{3}(x, t-S-S_{1}-S_{2})$

for $t\in[S+S_{1}+S_{2}/3, S+S_{1}+2S_{2}/3]$

$=(J_{k_{2}}^{2}(g_{2}, \cdot)^{*}A_{3})(s, t -S-S_{1}-S_{2})$ for $t>S+S_{1}+2S_{2}/3$ .

By perturbing this map as in \S 8, we obtain a map

$\overline{\Phi}_{j_{1},j_{2},j_{3},k_{1},k_{2}}$
: $U_{j_{1}}^{1}\times U_{j_{2}}^{2}\times U_{j_{3}}^{3}\times V_{k_{1}}\times V_{k_{2}}\times(T, \infty)\times R\rightarrow\overline{\mathcal{M}}_{\ell,\delta}(a, b)$

which is a lift of the map $\Phi_{(c,c)}$ of Theorem 7.6. By construction in \S 8,

we have
$|\overline{\Phi}_{j_{1},j_{2},j_{3},k_{1},k_{2}}-\overline{\Phi}_{j_{1},j_{2},j_{3},k_{1},k_{2}}|<e(S_{1)}S_{2})$
.
Similarly we have
$\overline{\Phi}_{j_{1},j_{2},k_{1}}^{J(1)}$
: $U_{j_{1}}^{1}\times U_{j_{2}}^{2}\times V_{k_{1}}\times(T, \infty)\times R\rightarrow A_{\ell}(a, c)$

$\overline{\Phi}_{j_{1},j_{2},k_{1}}^{(1)}$
: $U_{j_{1}}^{1}\times U_{j_{2}}^{2}\times V_{k_{1}}\times(T, \infty)\times R\rightarrow\overline{\mathcal{M}}_{\ell}(a, c)$
,

such that .
$\overline{\Phi}^{(1)}$

$\gamma_{1},j_{2},k_{1}$
is a lift of

$\Phi_{(c)}$ : $G_{a}\backslash \overline{K}_{a,c}\times_{G_{c}}\overline{K}_{c,c}/G_{c}\times(T, \infty)\times R\rightarrow\overline{\Lambda 4}(a, c)$

.

Here $\overline{\Phi}_{j_{1},j_{2},k_{1}}^{(1)\prime}$

is obtained by similar patching procedure as

and that
$\overline{\Phi}$

.
$\gamma_{1},j_{2},j_{3},k_{1},k_{2}$

$|\overline{\Phi}_{j_{1},j_{2},k_{1}}^{\prime(1)}-\overline{\Phi}_{j_{1},j_{2},k_{1}}^{(1)}|<e(S_{1})$
.

We may assume that for each $j_{1},j_{2}$

with
$G_{a}\backslash \overline{U}_{j_{1}}^{1}\times_{G_{c}}\overline{U}_{j_{2}}^{2}\times c_{c}\overline{K}_{c,b}\times(T, \infty)^{2}\subset W_{1}$
,
60 K. Fukaya

there exists $j=j(j_{1},j_{2}, k_{1})$ such that

$Im\overline{\Phi}_{j_{1},j_{2},k_{1}}\subset U_{j}^{4}$
.

We have maps

$\overline{\Phi}_{j,j_{3},k_{2}}^{\prime(2)}$

: $U_{j}^{4}\times U_{j_{3}}^{3}\times V_{k_{2}}\times(T, \infty)\times R\rightarrow A_{\ell}(a, b)$

$\overline{\Phi}_{j,j_{3},k_{1}}^{(2)}$
: $U_{j}^{4}\times U_{j_{3}}^{3}\times V_{k_{2}}\times(T, \infty)\times R\rightarrow\overline{\mathcal{M}}_{\ell}(a, b)$

such that $\overline{\Phi}_{j,j_{3},k_{2}}^{(2)}$

is a lift of

$\Phi_{(c)}$ : $\overline{K}_{a,c}\times_{G_{c}}\overline{K}_{c,b}/G_{b}\times(T, \infty)\times R\rightarrow\overline{\mathcal{M}}(a, b)$

,

Here $\overline{\Phi}_{j,j_{3},k_{2}}^{(2)\prime}$

is obtained by a similar patching procedure as $\overline{\Phi}_{j_{1},j_{2},j_{3},k_{1},k_{2}}$

,
and that
$|\overline{\Phi}_{j,j_{3},k_{2}}^{\prime(2)}-\overline{\Phi}_{j,j_{3},k_{2}}^{(2)}|<e(S_{2})$
.

By construction, we can choose lifts $s_{j}^{1}$

e.t.c. so that

$\overline{\Phi}_{j(j_{1},j_{2},k_{1}),j_{3},k_{2}}^{\prime(2)}(\overline{\Phi}_{j_{1},j_{2},k_{1}}^{\prime(1)}([A_{1}], [A_{2}], g_{1}, S_{1}, S),$ $[A_{3}]$ , $g_{2}$ , $S_{2}$

, $S)$

$=\overline{\Phi}_{j_{1},j_{2},j_{3},k_{1},k_{2}}([A_{1}], [A_{2}], [A_{3}], g_{1}, g_{2}, S_{1}, S_{2}, S)$

.

(Here $S$
is determined by $S,S,S_{1}$ and $S_{2}.$
) It follows that

$|\Phi_{(c)}\Phi_{(c,c),(c)}-\Phi_{(c,c)}|<e(S_{1}, S_{2})$ .

Using induction hypothesis (10.5), we obtain

$|\Phi_{(c)}\Psi_{(c,c),(c)}-\Psi_{(c,c)}|<e(S_{1}, S_{2})$ .

Let and
$\overline{\Psi}_{j_{1},j_{2},j_{3},k_{1},k_{2}}$ $\overline{\Psi}_{j_{1},j_{2},k_{1}}^{(1)}$
be the lifts of $\Psi_{c,c}$ and $\Psi_{(c,c)(c)}$ , respec-
tively. Then we have

$|\overline{\Phi}_{j(j_{1},j_{2},k_{1}),j_{3},k_{2}}^{(2)}(\overline{\Psi}_{j_{1},j_{2},k_{1}}^{(1)}([A_{1}], [A_{2}], g_{1}, S_{1}, S),$ $[A_{3}]$ , $g_{2}$ , $S_{2}$

, $S)$

$-\overline{\Psi}_{j_{1},j_{2},j_{3},k_{1},k_{2}}([A_{1}], [A_{2}], [A_{3}], g_{1}, g_{2}, S_{1}, S_{2}, S)|<e(S_{1}, S_{2})$

.

Therefore we can define

$-=\prime j_{1},j_{2},j_{3},k_{1},k_{2}-$
: $U_{j_{1}}\times U_{j_{2}}\times U_{j_{3}}\times V_{k_{1}}\times V_{k_{2}}\times(T, \infty)^{2}\times[0,1]\rightarrow A_{\ell}(a, b)/R$
 Floer Homology for Oriented 3-Manifolds 61

by
$--=\prime j_{1},j_{2},j_{3},k_{1},k_{2}([A_{1}], [A_{2}], [A_{3}], g_{1}, g_{2}, S_{1}, S_{2}, s)=$

$(1-s)\cdot\overline{\Phi}_{j(j_{1},j_{2},k_{1}),j_{3},k_{2}}^{(2)}(\overline{\Psi}_{j_{1},j_{2},k_{1}}^{(1)}([A_{1}], [A_{2}], g_{1}, S_{1}, S)$

, $[A_{3}]$ , $g_{2}$ , $S_{2}$
, $S)$

$+s\cdot\overline{\Psi}_{j_{1},j_{2},j_{3},k_{1},k_{2}}([A_{1}], [A_{2}], [A_{3}], g_{1}, g_{2}, S_{1}, S_{2}, S)$

.

Since gauge transformation is an ane map (namely $g^{*}(sA+(1-s)B)=$

$sg^{*}A+(1-s)g^{*}B$ holds for each connections $A$ , $B$ and gauge transfor-
mation ), it follows from an argument similar to the proof of Lemma
$g$

8.3 that we can perturb so that it defines a map :

$--=\prime j_{1},j_{2},j_{3},k_{1},k_{2}$ $---J$

$W_{1}\times[0,1]\rightarrow B_{\ell}(a, b)$ , which is invariant. Using Taubes method $G_{b}$

as in \S 8, we can perturb this map and $obtain---:W_{1}\times[0,1]\rightarrow \mathcal{M}_{\ell}(a, b)$ .

This map $---is$ an isotopy between and . Take a $\Psi_{(c,c)}$ $\Phi_{(c)}\Psi_{(c,c),(c)}$

small open neighborhood of in $W_{1}$ $W_{1}$

A4 $(a, c)\times c_{c}$ A4 $(c, c)\times c_{c}\mathcal{M}(c, b)\times(T, \infty)^{2}$ .

be extend to
$---can$ $W_{1}$
. Let $\varphi$
: $W_{1}\rightarrow[0,1]$ be a $G_{b}$
-invariant function
such that
$\varphi(x)=0$ if $x\in\partial W_{1}$
, and if $\Psi_{(c,c),(c)}(x)\in X_{(c)}$
$\{$

$\varphi(x)=1$ if $x\in W_{1}$ .

(See Figure 7.) Define $\Psi_{(c)}$ on $\Psi_{(c,c),(c)}(W_{1})$ by

$\Psi_{(c)}(\Psi_{(c,c),(c)}(x))=---(x, \varphi(x))$ .

Figure 7.
62 K. Fukaya

Since
$(x, 0)=\Phi_{(c)}\Psi_{(c,c),(c)}(x)$ ,
we can extend $\Psi_{(c)}$ , by putting $\Psi_{(c)}=\Phi_{(c)}$ outside $\Psi_{(c,c),(c)}(W_{1})$ .
Since
$(x, 1)=\Psi_{(c,c)}(x)$ ,
we have $\Psi_{(c)}\Psi_{(c,c),(c)}=\Psi_{(c,c)}$ , on . The inequality (10.5) holds $W_{1}$

by construction. Using Lemma 8.14, we can prove that is a dif- $\Psi_{(c)}$

feomorphism to its image. Thus the patching argument for the proof of
Theorem 10.2 is completed in our case. The proof of general case is the
same, but the notation will be more complicated.
Remark 10.6. If we can establish rigorously what we suggested in
Remarks 8.15 and 8.20 we might be able to prove Theorem 10.2 using
the center of mass technique in Riemannian geometry. (See [GK].) But
the direct argument we gave above might be simpler.

\S 11. Extension of the line bundle to the boundary

In this section, we shall prove Theorem 7.3. First we consider the
case when none of are reducible. We put $c_{i}$

$C_{1}\overline{\mathcal{M}}(a, b)=c_{0},\cdots,c_{k},G_{c_{i}}=\{\pm 1\}\cup\overline{\mathcal{M}}^{J}(a, c_{0})\times\prod_{i=0}^{k-1}\overline{\mathcal{M}}(c_{i}, c_{i+1})\times\overline{\Lambda 4}(c_{k}, b)$ .

Lemma 11.1. Let $c$ $=(c_{0}, \cdots, c_{k})$ , $\mu(a)>\mu(c_{0})>\cdots>\mu(c_{k})>$

$\mu(b)$ , $G_{c_{i}}=\{\pm 1\}$ , and

$\Psi_{c}$
: $K_{a,c_{0}}\times\prod K_{c_{i},c_{i+1}}\times K_{c_{k},b}\times(T, \infty)^{k}\rightarrow\overline{\vee\wedge\Lambda}^{J}(a, b)$

be the map given in \S 10. Then there exists an isomorphism of line

bundles
$\varphi_{c}^{i}$

: $\Psi_{c}^{*}\mathcal{L}_{i}^{(2)}(a, b)\rightarrow \mathcal{L}_{i}^{(2)}(a, c_{0})\otimes\cdots\otimes \mathcal{L}_{i}^{(2)}(c_{k}, b)$

.

This lemma follows from Theorem 4.9 and the construction of $\Psi_{c}$
.
Hereafter we write
$\mathcal{L}_{i}^{(2)}(c)=\mathcal{L}_{\dot{x}}^{(2)}(a, c_{0})\otimes\cdots\otimes \mathcal{L}_{i}^{(2)}(c_{k}, b)$
.

Similarly, for $c\subset c$

, we have an isomorphism

$\varphi_{c,c}^{i}$
: $\Psi_{c,c}^{*}\mathcal{L}_{i}^{(2)}(c)\rightarrow \mathcal{L}_{\dot{x}}^{(2)}(c_{J}^{\backslash }$
.
 Floer Homology for Oriented 3-Manifolds 63

Lemma 11.2. On
$K_{a.c_{0}}\times\prod K_{c_{i},c_{i+1}}\times K_{c_{k},b}\times\{(s_{1}, \cdots, S_{k})\}$

we have
$||\varphi_{c,c}^{i}o\varphi_{c}^{i}-\varphi_{c}^{i}||<e(S_{1}, \cdots, S_{k})$ .

This lemma follows from the construction of $\varphi_{c}^{i}$

. By Lemma 11.2,
we can perturb , such that $\varphi_{c}^{i}$
$\varphi_{c,c}^{\dot{\iota}}$

$\varphi_{c,c}^{i}o\varphi_{c}^{i}=\varphi_{c}^{i}$
.

Using these isomorphisms, we can patch the bundles and obtain $\mathcal{L}_{i}^{(2)}(c)$

a line bundle over . $C_{1\vee}\overline{\Lambda\Lambda}(a, b)$

Next we consider the case when some are reducible. The following $c_{i}$

three results are used for this purpose.

Theorem 11.3. The local action on $\overline{\sqrt{}\vee l}^{J}(a, b)$

constructed in\S 10,
can be lifted to $\mathcal{L}_{i}^{(2)}(a, b)$
.

Hence, for each , the line bundle $c$

$\Psi_{c}^{*}\mathcal{L}_{i}^{(2)}(a, b)$
on $\overline{K}_{a,c_{0}}\times c_{c_{0}}\cdots\times c_{c_{k}}$

$\overline{K}_{c_{k},b}\times(T, \infty)^{k}$
has a local $G_{a}\times G_{c_{0}}\times\cdots\times G_{c_{k}}\times G_{b}$ action. Therefore
we obtain a bundle $\Psi_{c}^{*}\mathcal{L}_{i}^{(2)}(a, b)$
on

$K_{a,co}^{*}\times\prod K_{c_{i},c_{i+1}}^{*}\times K_{c_{k},b}^{*}\times(T, \infty)^{k}$ .

Here $K_{c_{i},c_{i+1}}^{*}$ denotes the set of reducible connections. As before we put

$\mathcal{L}_{i}^{(2)}(c)=\mathcal{L}_{i}^{(2)}(a, c_{0})\otimes\cdots\otimes \mathcal{L}_{i}^{(2)}(c_{k}, b)$

,

which is a line bundle on

$K_{a,c_{0}}^{*}\times\prod K_{c_{i},c_{i+1}}^{*}\times K_{c_{k},b}^{*}\times(T, \infty)^{k}$
.

Lemma 11.4. There exist isomorphisms

$\varphi_{c}^{i}$

: $\Psi_{c}^{*}\mathcal{L}_{i}^{(2)}(a, b)\rightarrow \mathcal{L}_{i}^{(2)}(c)$

$\varphi_{c,c^{J}}^{i}$
: $\Psi_{c,c^{J}}^{*}\mathcal{L}_{i}^{(2)}(c)\rightarrow \mathcal{L}_{i}^{(2)}(c)$
.

Lemma 11.5. On
$K_{a,c_{0}}^{*}\times\prod K_{c_{i},c_{i+1}}^{*}\times K_{c_{k},b}^{*}\times\{(S_{1}, \cdots, S_{k})\}$
64 K. Fukaya

we have
$||\varphi_{c,c}^{i}o\varphi_{c}^{i}-\varphi_{c}^{i}||<e(S_{1}, \cdots, S_{k})$ .

Using these results, we can prove Theorem 7.3 in a way similar to

the case when none of are reducible. The proof of Lemmas 11.4 and $c_{i}$

11.5 are similar to one of Lemma 11.1 and 11.2 respectively. In the rest
of this section, we prove Theorem 11.3.
First we lift the action on the image . We $\Psi_{c}(\overline{X}^{o}(c))\subset \mathcal{M}(a, b)$

are studying the determinant bundle of the operator defined on $ 6_{A}^{i}+\epsilon$

R. On their ends, these operators are asymptotic

$\Sigma_{i}\simeq S^{1}\times R\subset M\times$

to , for some $a\in Fl$ . Here the operator

$\frac{\partial}{\partial t}+6_{a}^{i}+\epsilon$
is defined on . $6_{a}^{x}$
$S^{1}$

We choose such that the first eigenvalue of

$\lambda_{0}$
is larger $(6_{a}^{i}+\epsilon)^{*}(6_{a}^{i}+\epsilon)$

than for each .

$\lambda_{0}$
$a$

For simplicity, we shall consider the case where $=(c)$ , $G_{c}\neq\{\pm 1\}$ . $c$

In this case, is a perturbation of the map defined below. (See \S 8.)

$\Psi_{c}$ $\Phi$

Choose an open covering

$U_{1}^{1}\cup\cdots\cup U_{N}^{1}\supseteq K_{a,c}$
,
$U_{1}^{2}\cup\cdots\cup U_{N}^{2}\supseteq K_{c,b}$
,
$V_{1}\cup\cdots\cup V_{N}=G_{c}$ ,

and sections $s_{j}^{1}$

: $U_{j}^{1}\rightarrow A_{\ell,.\delta}(a, c)$ , $s_{j}^{2}$
: $U_{j}^{2}\rightarrow A_{\ell,\delta}(c, b)$ . Let $J_{k}$
:
$V_{k}\times R\rightarrow G_{c}$
be a map such that
1 if $t<-1$
$J_{k}(g, t)=\{$
$g$ if $t$
$>0$

Then the map

$\overline{\Phi}_{j_{1},j_{2},k}$
: $U_{j_{1}}^{1}\times V_{k}\times U_{j_{2}}^{2}\times[T, \infty)\times R\rightarrow A_{\ell,\delta}(a, b)$

is defined by
$\overline{\Phi}_{j_{1},j_{2},k}([A_{1}], g, [A_{2}], S, S)$

$(J_{k}(g, \cdot)^{*}A_{1})(x, t-S)$ if $t<S+S/3$ .

$\chi(\frac{t-S-S/3}{S/3})g^{*}A_{1}(x, t-S)$

$=\{$
$+(1-\chi(\frac{t-S-S/3}{S}))A_{2}(t-S-S)$

if $S+S/3<t<S+2S/3$
$A_{2}(t-S-S)$ if $t>S+2S/3$ .
 Floer Homology for Oriented 3-Manifolds 65

Here $\chi$ is the cut function in \S 8. The maps $\overline{\Phi}_{j_{1},j_{2},k}$

induce a map $\Phi$
:
. They satisfy
$\overline{X}^{o}((c))\rightarrow B_{\ell,\delta}(a, c)$

$||(\Psi_{(c)}-\Phi)([A_{1}], g, [A_{2}], S, S)||_{L_{\ell}^{2}}<Ce^{-S/C}$ .

Therefore, there exists an isomorphism $\Psi_{(c)}^{*}\mathcal{L}_{i}^{(2)}(a, b)\rightarrow\Phi^{*}\overline{\mathcal{L}}_{i}^{(2)}(a, b)$

.
We shall lift the local action of $G_{c}$
on $\overline{K}_{a,c}\times c_{c}\overline{K}_{c,b}$
, to a local action
on $\Phi^{*}\mathcal{L}_{i}^{(2)}(a, c)$
.
Replacing $U_{j_{1}}^{1}$
and $U_{j_{2}}^{2}$
by a smaller one if necessary, we can find
positive numbers $\lambda_{j_{1},j_{2}}<\lambda_{0}$
, such that the following holds.

(11.6.1) If $[a_{t}]\in U_{j_{1}}^{1}$

then $\lambda_{j_{1},j_{2}}$
is not an eigenvalue of $(6_{a_{t}}+\epsilon)^{*}(6_{a_{t}}+$

on .
$\epsilon)$ $\Sigma_{i}$

(11.6.2) If $[a_{t}]\in U_{j_{2}}^{2}$

then $\lambda_{j_{1},j_{2}}$
is not an eigenvalue of $(6_{a_{t}}+\epsilon)^{*}(6_{1,t}+$

on .
$\epsilon)$ $\Sigma_{i}$

Then, by Remark 4.6, $\lambda_{j_{1},j_{2}}$

is not an eigenvalue of $(6_{A}+\epsilon)^{*}(6_{A}+\epsilon)$

on , if $\Sigma_{i}$

$[A]\in\Phi(U_{j_{1}}^{1}\times G_{c}\times U_{j_{2}}^{2}\times(T, \infty)\times R)$

for suciently large $T$ . Let $[A_{1}]\in U_{j_{1}}^{1}$ , $[A_{2}]\in U_{j_{2}}^{2}$ , $g\in V_{k}\subset G_{c}$ , and
$A=\overline{\Phi}_{j_{1},k,j_{2}}([A_{1}], g, [A_{2}], S, S)$ , we put

$L(A_{1}, g, A_{2}, S, S)=\lambda<\lambda_{j,j_{2}}\oplus_{1}\{u|(6_{A}+\epsilon)^{*}(6_{A}+\epsilon)u=\lambda u\}$

,

$L(A_{1}, g, A_{2}, S, S)=\lambda<\lambda_{j,j_{2}}\oplus_{1}\{u|(6_{A}+\epsilon)(6_{A}+\epsilon)^{*}u=\lambda u\}$

,

$L=A_{1},g,A_{2},S,S\cup L(A_{1}, g, A_{2}, S, S)$

,

$L=A_{1},g,A_{2},S,S\cup L(A_{1}, g, A_{2}, S, S)$

.

By (11.6.1) and (11.6.2), the dimensions of $L$

and $L$
are constant. By
definition,

$\Phi^{*}(\mathcal{L}_{i}^{(2)}(a, b))|_{([A_{1}],g,[A_{2}],S,S)}$

$\simeq(to\wedge p(L(A_{1}, g, A_{2}, S, S))^{*}\otimes top\wedge L(A_{1}, g, A_{2}, S, S))^{\otimes 2}$

66 K. Fukaya

Lemma 11.7. Let $t\in[S+S/3, S+2S/3]$ , $u\in L(A_{1}, g, A_{2}, S, S)$ .

The
$|u(x, t)|_{C^{\ell}}<Ce^{-\sqrt{\lambda_{0}-\lambda_{j_{1},j_{2}}}\beta(t)}||u||_{L^{2}}$

Here $\beta(t)=d(t, \partial[S+S/3, S+2S/3])$

The proof of the lemma is similar to one of Lemma 4.5.

For $u\in L(A_{1}, g, A_{2}, S, S)$ , , $h\in G_{c}$ with , $hg\in V_{k}$ , we put
$g$ $g$

$I_{1}(h)(u)(t, x)=$

$J_{k}(hg, t-S)J_{k}(g, t -S)^{-1}u(x, t)$ if $<S+S/3$ .

$t$

$\chi(\frac{t-S-S/3}{S/3})hu(x, t)+(1-\chi(\frac{t-S-S/3}{S/3}))u(x, t)$

$\{$

if $S+S/3<t<S+2S/3$ ,
$u(x, t)$ if $>S+2S/3$ .
$t$

Let $I_{2}(h)(u)$ is
the orthonormal projection of $I_{1}(h)(u)$ to
$L(A_{1}, hg.A_{2}, S, S)$ . Lemma 11.7 implies:

Lemma 11.8.
$||I_{2}(h)(u)-I_{1}(h)(u)||_{L^{2}}<Ce^{-S/C}||u||_{L^{2}}$ .

Lemma 11.9. If $g\in V_{k}$ , $hg\in V_{k}$ and $hhg\in V_{k}$ , the

$||I_{2}(hh)(u)-I_{2}(h)I_{2}(h)(u)||_{L^{2}}<Ce^{-S/C}||u||_{L^{2}}$ .

Next we extend to which is defined also for such that $g\in V_{k}$
$I_{2}$ $I_{5}$ $h$

and . Note that $G_{c}=U(1)$ or $=SU(2)$ . Hence, in fact, we

$hg\not\in V_{k}$

need only two charts and to cover

$V_{1}$
. (This fact is not essential
$V_{2}$ $G_{c}$

for the proof but we use it to simplify the notation.) Choose $ g_{0}\in$

. For $g\in V_{1}$ , $hg\in V_{2}$ , we take

$V_{1}\cap V_{2}$
and such that $h_{1}g=g_{0}$ and
$h_{1}$ $h_{2}$

$h_{2}h_{1}=h$ . Then, for $h\in L(A_{1}, g, A_{2}, S, S)$ , the element $ I_{2}(h_{1})(u)\in$
$L(A_{1}, g_{0}, A_{2}, S, S)$ is well defined. We put

$I_{3}(h)(u)=I_{2}(h_{2})I_{2}(h_{1})(u)$ .

Since , $h_{1}g\in V_{2}$ , it follows that

$h_{2}(h_{1}g)$ in the above formula is $I_{2}(h_{2})$

well defined. Choose : such that

$\chi$
$G_{c}\rightarrow[0,1]$

1 if $g\in V_{1}-(V_{1}\cap V_{2})$ .

$\chi(g)=\{$
0 if $g\in V_{2}-(V_{1}\cap V_{2})$ .
 Floer Homology for Oriented 3-Manifolds 67

Put
$I_{2}(h)(u)$ if $hg\in V_{1}-(V_{1}\cap V_{2})$ ,
$\chi(hg)I_{2}(h)(u)+(\acute{1}-\chi(hg))I_{3}(h)(u)$
$I_{4}^{1}(u)=\{$
if $hg\in V_{1}\cap V_{2}$ ,
$I_{3}(h)(u)$ if $hg\in V_{2}-(V_{1}\cap V_{2})$ .

In the case when $g\in V_{2}$ , we define $I_{4}^{2}(h)$

in a similar way. Finally we
put, for $u\in L(A_{1}, g, A_{2}, S, S)$

$I_{4}^{1}(h)(u)$ if $g\in V_{1}-(V_{1}\cap V_{2})$ ,

$\chi(g)I_{4}^{1}(h)(u)+(1-\chi(g))I_{4}^{2}(h)(u)$
$I_{5}(h)(u)=\{$
if $g\in V_{1}\cap V_{2}$ ,
$I_{4}^{2}(h)(u)$ if $g\in V_{2}-(V_{1}\cap V_{2})$ .

Then is defined for every and

$I_{5}$ $h$
$g$ and depends smoothly on them. By
perturbing a bit we obtain $I_{5}$ $I_{6}(h)$ which is a linear isometry
$L(A_{1}, g, A_{2}, S, S)\rightarrow L(A_{1}, hg, A_{2}, S, S)$ .

By construction, we have

(11.10) $||I_{6}(hh)(u)-I_{6}(h)I_{6}(h)(u)||_{L^{2}}<Ce^{-S/C}||u||_{L^{2}}$ .

Next we use the center of mass technique, to perturb and obtain $I_{6}$

I satisfying $I(h)I(h)=I(hh)$ . Namely we use the following:

Lemma 11.11. For each compact Lie group $G$ and , , there $n$ $\epsilon>0$

exists , such that the following holds.

$\delta_{n}(G, \epsilon)>0$

Let : be a hermitian vector bundle of rank , $G$ act on $X$ ,

$\pi$ $L\rightarrow X$ $n$

and : $\varphi$
be a map. Suppose
$G\times L\rightarrow L$

(11.12.1) $\pi(\varphi(g, v))=g(\pi(v))$ ,

(11.12.2) $\varphi$
is a linear isometry on each fifibre,
(11.12.3) $|\varphi(g_{1}, g_{2}, v)-\varphi(g_{1}(\varphi(g_{2}, v))|<\delta_{n}(G, \epsilon)$ .

Then, there exists a lift of the action of $G$ to $L$

, such that

$|\varphi(g, v)-g$ . $ v|<\epsilon$

.

In the case when $X$ is a point, Lemma 11.11 means that an almost
homomorphism $G\rightarrow U(n)$ is approximated by a homomorphism. This
case is proved in [GKR]. The proof of Lemma 11.11 is identical to that
68 K. Fukaya

case and hence is omitted. (See also [BK] p138.) Note that $\delta_{n}(G, \epsilon)$
in
the lemma is independent of . $X$

Now, using Lemma 11.11, we can perturb to obtain a lift I of $I_{6}$

the local action on to the vector bundle $U_{j_{1}}^{1}\times G_{c}\times U_{j_{2}}^{2}\times(T, \infty)\times R$

$L(A_{1}, g, A_{2}, S, S)$ on it. In a similar way, we can lift the action to
$L(A_{1}, g, A_{2}, S, S)$ . Hence we obtain a lift of the action to the restriction

of $\Phi^{*}\overline{\mathcal{L}}_{a,b}^{(2)}$

to $\overline{U}_{j_{1}}^{1}\times c_{c}\overline{U}_{j_{2}}^{2}\times(\dot{T}, \infty)\times R=U_{j_{1}}^{1}\times G_{c}\times U_{j_{2}}^{2}\times(T, \infty)\times R$

.
(Here are the inverse images of
$\overline{U}_{j_{1}}^{1}$

and $\overline{U}_{j_{2}}^{2}$
$U_{j_{1}}^{1}$
and $U_{j_{2}}^{2}$
in $\overline{K}_{a,c}$
and
$\overline{K}_{c,b}$
, respectively.) We denote the lift by $I_{j_{1},j_{2}}$
. By construction, we
have, on $(\overline{U}_{j_{1}}^{1}\times_{G_{c}}\overline{U}_{j_{2}}^{2})\cap(\overline{U}_{j_{1}}^{1}\times c_{c}\overline{U}_{j_{2}}^{2})\times(T, \infty)\times R$
,

$d(I_{j_{1},j_{2}}(h), I_{j_{1},j_{2}}(h))<Ce^{-T/C}$ .

Hence using a partition of unity, we can patch them as an almost action.

Therefore, using Lemma 11.11, we obtain a lift of the local action to
$\Phi^{*}\mathcal{L}_{i}^{(2)}(a, b)$
.
In order to lift the local action on , we have to patch those $\lambda\Lambda(a, b)$

lifts we constructed above. By construction, they are compatible mod-

ulo a dierence estimated by $e(S_{1}, \cdots, S_{k})$ on $\times\{(S_{1}, \cdots, S_{k})\}\times R$ . $\cdots$

Hence we can apply a similar patching procedure as above. The proof

of Theorem 11.3 is now complete.

\S 12. Boundary operators

In this section, we define the boundary operators
$\partial$

: $C_{k}^{0}\rightarrow C_{k-1}^{0}$

$\partial_{\gamma}$

: $C_{k}^{0}\rightarrow C_{k-3}^{0}$

$\partial_{\gamma_{1},\gamma_{2}}$
: $C_{k}^{0}\rightarrow C_{k-5}^{0}$
.

The definition of is the same as Floers. Let , $b\in Fl$ , with $\partial$
$a$

$\mu(a)=\mu(b)+1$ . Then, consists of finitely many points each $\overline{\mathcal{M}}(a, b)$

of which is given an orientation $+or-$ . We let be the number $\langle\partial a, b\rangle$

of the points with $+orientation$ minus the number of points with -

orientation. Put
$\partial[a]=\sum\langle\partial a, b\rangle[b]$ .

Next we define . For a closed loop on $M$ we obtain a line bundles $\partial_{\gamma}$
$\gamma$

, over . We choose sections

$\mathcal{L}_{\gamma}^{(2)}(c, c)$
to , such $\overline{\mathcal{M}}(c, c)$
$s_{\gamma}(c, c)$
$\mathcal{L}_{\gamma}^{(2)}(c, c)$

that the following holds.

 Floer Homology for Oriented 3-Manifolds 69

(12.1.1) For each , $b\in Fl$ , the collection of the sections

$a$

$s_{\gamma}(a, c_{0})\otimes\cdots\otimes s_{\gamma}(c_{k}, b)$

to
$\mathcal{L}_{\gamma}^{(2)}(a, c_{0})\otimes\cdots\otimes \mathcal{L}_{\gamma}^{(2)}(c_{k}, b)$

can be patched together to give a smooth section on . (We use $C\overline{\mathcal{M}}^{J}(a, b)$

the symbol also for this extension.)

$s_{\gamma}(a, b)$

(12.1.2) The zeros of are transversal and transversal to each $s_{\gamma}(c, c)$

other.

Since we restrict ourselves to the case when $s<3$ if $H_{1}(M;Z)$ is

torsion free, and when $=0$ otherwise, then we need only to study $s$

the case when $\mu(a)<\mu(b)+8$ , $H_{1}(M;Z)$ is torsion free and and $a$

$b$
are irreducible. In this case, if $\mu(a)\geq\mu(c)\geq\mu(c)\geq\mu(b)$ , and if
, , then
$\mathcal{M}(a, c)\neq\emptyset$
does not contain a reducible
$\mathcal{M}(c, b)\neq\emptyset$ $\mathcal{M}(c, c)$

connection. Also in our case, Lemma 5.8 implies that bubbling o of

instanton does not happen. Hence (7.1.6) implies that the set $C\overline{\mathcal{M}}(a, b)$

is compact. The later fact is not really necessary for the argument. (We
can discuss as in Donaldson [D4], in case when and are irreducible.) $a$ $b$

However the former point is essential. We discuss it at the end of this

section.
Now, let $\mu(a)=\mu(b)+3$ . Set

$\Sigma_{\gamma}(a, b)=\{x\in C\overline{\mathcal{M}}(a, b)|s_{\gamma}(a, b)(x)=0\}$ .

Dimension counting, the compactness of $C\overline{\mathcal{M}}^{J}(a, b)$

and the transversal-
ity (12.1.2) imply

$\Sigma_{\gamma}(a, b)\cap\partial C\overline{\mathcal{M}}(a, b)=\emptyset$

$\#\Sigma_{\gamma}(a, b)<\infty$ .

The orientation of $\overline{\lambda\Lambda}(a, b)$

induces an orientation of each point of $\Sigma_{i}$
.
We define by $\langle\partial_{\gamma}a, b\rangle$

$\langle\partial_{\gamma}a, b\rangle=\#\Sigma_{\gamma}$
.
Here and hereafter stands for the number of points $with+orientation$
$\#$

minus the number of points withas orientation. We set

$\partial_{\gamma}[a]=\sum_{b}\langle\partial_{\gamma}a, b\rangle[b]$
.
70 K. Fukaya

For $\mu(b)=\mu(a)+5$ , and loops $\gamma_{1}$

and $\gamma_{2}$
, we put

$\Sigma_{\gamma_{1},\gamma_{2}}(a, b)=\{x\in C\overline{\mathcal{M}}(a, b)|s_{\gamma_{1}}(a, b)(x)=s_{\gamma_{2}}(a, b)(x)=0.\}$

,

and define
$\langle\partial_{\gamma_{1},\gamma 2}a, b\rangle=\#\Sigma_{\gamma_{1},\gamma 2}(a, b)$

$\partial_{\gamma_{1},\gamma_{2}}[a]=\sum_{b}\langle\partial_{\gamma_{1},\gamma_{2}}a, b\rangle[b]$
.

Now we prove Theorem 1.10. For simplicity, we discuss the case

, and prove
$\alpha=\{\gamma\}$ . Let , $b\in Fl$ with $\mu(a)=\mu(b)+4$ .
$\partial_{\gamma}\partial+\partial\partial_{\gamma}=0$ $a$

The line bundle can be extended to by

$\mathcal{L}_{\gamma}^{(2)}(a, b)\rightarrow\overline{\mathcal{M}}(a, b)$ $C\overline{\mathcal{M}}(a, b)$

Theorem 7.3. Since , the set $dim\overline{\mathcal{M}}^{J}(a, b)=3$

$\Sigma_{\gamma}(a, b)=\{x\in C\overline{\mathcal{M}}^{J}(a, b)|s_{\gamma}(a, b)(x)=0\}$

is one dimensional oriented manifold. And

$\partial\Sigma_{\gamma}(a, b)=\Sigma_{\gamma}(a, b)\cap\partial\overline{\mathcal{M}}^{J}(a, b)$
.

By transversality and dimension counting we have

$\partial\Sigma_{\gamma}(a, b)=\{(x, y)\in\overline{\mathcal{M}}(a, b)\times\overline{\Lambda 4}(c, b)|$

$s_{\gamma}(a, c)(x)\cdot s_{\gamma}(c, b)(y)=0$ , $c$

is irreducible.}.
$=\prod_{\mu(c)=\mu(b)+1}\Sigma_{\gamma}(a, b)\times\overline{\mathcal{M}}(c, b)\cup$

$\prod_{\mu(c)=\mu(b)+2}\overline{\Lambda 4}(a, c)\times\Sigma_{\gamma}(c, b)$

.

The orientations are also compatible. Therefore we have

$\sum_{c}\langle\partial_{\gamma}a, c\rangle\langle\partial c, b\rangle+\sum_{c}\langle\partial a, c\rangle\langle\partial_{\gamma}c, b\rangle=0$

.

Hence $\partial_{\gamma}\partial+\partial\partial_{\gamma}=0$
, as required.
The proof of $\partial_{\gamma_{1},\gamma_{2}}\partial+\partial_{\gamma_{1}}\partial_{\gamma_{2}}+\partial_{\gamma_{2}}\partial_{\gamma_{1}}+\partial\partial_{\gamma_{1},\gamma_{2}}=0$
is similar.

Now put
$C_{k}^{s}=\oplus_{s}S^{p}H_{1}\ell\leq(M, Z)\otimes C_{k-2\ell}^{0}$
,
 Floer Homology for Oriented 3-Manifolds 71

and define $\hat{\partial}:C_{k}^{s}\rightarrow C_{k-1}^{s}$

, by

$\hat{\partial}(\gamma_{\alpha}\otimes[a])=\sum_{\alpha^{1}\cup\alpha^{2}=\alpha}\gamma_{\alpha^{1}}\otimes\partial_{\alpha^{2}}[a]$
.

(Here we fix a basis $\gamma_{1}$

, $\cdots$
, $\gamma_{d}$
of the first homology group and put

$\partial_{\alpha}=\sum_{j_{1},\cdot\cdot,jp}\prod_{i}C_{i,j_{i}}\partial_{\gamma_{j_{1}}\gamma_{jp}}\cdots$

if $\alpha=(\sum_{j_{1}}C_{1,j_{1}}[\gamma_{j_{1}}], \cdots, \sum_{jp}C_{\ell,jp}[\gamma_{jp}])$ . Later, in Lemma 12.10, we

shall prove that $\partial_{\gamma}$
are additive with respect to $\gamma.$
) Theorem 1.10 implies
$\hat{\partial}\hat{\partial}=0$

As we pointed out in \S 1, the boundary operator itself does depend $\hat{\partial}$

on the choice of the sections , because the spaces have $s_{\gamma}(c, c)$
$C\overline{\mathcal{M}}(c, c)$

boundaries. Next we prove that the chain complex is independent $(C^{s}.,\hat{\partial})$

of the choice of the section.

Theorem 12.2. Suppose $H_{1}(M;Z)$ is torsion free and $s<3$ . Let

$s_{\gamma}(a, b)$
and are the sections satisfying (12.1.1) and (12.1.2). Let
$s_{\gamma}(a, b)$

and
$(C^{s},\hat{\partial})$ $(C^{s},\hat{\partial})$
be the corresponding chain complexes. Then there
exist maps , $\psi$
$\varphi$
: $C^{s}\rightarrow C^{s}$
such that

(12.2.1) $\hat{\partial}\varphi=\varphi\hat{\partial}$

(12.2.2) $\hat{\partial}\psi=\psi\hat{\partial}$

(12.2.3) $\varphi\psi=\psi\varphi=identity$ .

Proof. For each loop $\gamma$

and , $c\in Fl$ , we choose a section
$c$ $\overline{s}_{\gamma}(c, c)$

to $\mathcal{L}_{\gamma}^{(2)}(c, c)\times[0,1]\rightarrow\overline{\mathcal{M}}^{J}(c, c)\times[0,1]$

such that
$\overline{s}_{\gamma}(c, c)(x, 0)=s_{\gamma}(c, c)(x)$

(12.1.1)
$\overline{s}_{\gamma}(c, c)(x, 1)=s_{\gamma}(c, c)(x)$

(12.3.2) For each , $b\in Fl$ , the collections of sections

$a$

$\overline{s}_{\gamma}(a, c_{0})\otimes\cdots\otimes\overline{s}_{\gamma}(c_{k}, b)$

can be patched together to give a smooth section on . $C\overline{\mathcal{M}}^{J}(a, b)\times[0,1]$

(12.3.3) The zeros of are transversal and are transversal to each $\overline{s}_{\gamma_{i}}$

other.
72 K. Fukaya

Now, let $\mu(a)=\mu(b)+3$ , and put

$\overline{\Sigma}_{\gamma}(a, b)=\{(x, t)\in C\overline{\mathcal{M}}(a, b)\times[0,1]|\overline{s}_{\gamma}(a, b)(x, t)=0\}$

.

Then $dim\overline{\Sigma}_{\gamma}(a, b)=1$

. Note that (12.3.2) implies that

$\overline{\Sigma}_{\gamma}(a, b)\cap(\overline{\mathcal{M}}^{J}(a, c)\times\overline{\mathcal{M}}^{J}(c, b)\times[0,1])\neq\emptyset$

only if $c$
is irreducible and $\mu(c)=\mu(b)+1$ or 2. Therefore

(12.4)
$\partial\overline{\Sigma}_{\gamma}(a, b)=$

$\{(x, 0)|\overline{s}_{\gamma}(a, b)(x, O)=0\}\cup\{(x, 1)|\overline{s}_{\gamma}(a, b)(x, 1)=0\}\cup$

$\prod_{c}\{(x_{1}, x_{2}, t)|\overline{s}_{\gamma}(c, b)(x_{1}, t)\cdot\overline{s}_{\gamma}(a, c)(x_{2}, t)=0\}$

.

$\partial_{\gamma}$

$c)x\overline{m}(\prime c,b)$

$\times\overline{m}(\prime cb\prime,)$

$\overline{m}(\prime a,c)x\overline{m}^{J}(c,b)$

Figure 8.
 Floer Homology for Oriented 3-Manifolds 73

For each , $c\in Fl$ , with $a$ $\mu(a)=\mu(c)+2$ , we put

$\langle\varphi_{\gamma}a, c\rangle=\#\{(x, t)\in\overline{\mathcal{M}}^{J}(a, c)\times[0,1]|\overline{s}_{\gamma}(x, t)=0\}$

.

Note the set in the right hand side is a finite set, by (12.3.3) and dimen-
sion counting. Define : by $\varphi_{\gamma}$
$C_{k}^{0}\rightarrow C_{k-2}^{0}$

$\varphi_{\gamma}[a]=\sum\langle\varphi_{\gamma}a, c\rangle[c]$
.

Then (12.4) implies

(12.5) $\partial_{\gamma}-\partial_{\gamma}+\partial\varphi_{\gamma}-\varphi_{\gamma}\partial=0$
.

Now define $\varphi$

, $\psi$
: $C^{1}\rightarrow C^{1}$
by

$\varphi(1\otimes[a])=1\otimes[a]$

$\varphi(\gamma\otimes[a])=\gamma\otimes[a]+1\otimes\varphi_{\gamma}[a]$ ,
$\psi(1\otimes[a])=1\otimes[a]$

$\psi(\gamma\otimes[a])=\gamma\otimes[a]-1\otimes\varphi_{\gamma}[a]$ .

Then using (12.5), it is easy to verify (12.2.1),(12.2.2), and (12.2.3).

Next we consider the case $s=2$ . Let $\mu(a)=\mu(b)+5$ . Put

$\overline{\Sigma}_{\gamma_{1},\gamma_{2}}(a, b)=\{(x, t)\in C\overline{\mathcal{M}}(a, b)\times[0,1]|\overline{s}_{\gamma_{1}}(x, t)=\overline{s}_{\gamma_{2}}(x, t)=0\}$

.
74 K. Fukaya

We have
(12.6)
$\partial\overline{\Sigma}_{\gamma_{1},\gamma 2}(a, b)=$

$\{(x, 0)|s_{\gamma 1}(a, b)(x)=s_{\gamma_{2}}(a, b)(x)=0\}$

$\cup\{(x, 1)|s_{\gamma_{1}}(a, b)(x)=s_{\gamma_{2}}(a, b)(x)=0\}$

$\cup\prod_{\mu(c_{1})=\mu(b)+1}\{(x, y, t)\in\overline{\mathcal{M}}(a, c_{1})\times\overline{\Lambda 4}(c_{1}, b)\times[0,1]|$

$\overline{s}_{\gamma 1}(a, c_{1})(x, t)=\overline{s}_{\gamma 2}(a, c_{1})(x, t)=0\}$ ,

$\cup\prod_{\mu(c_{4})=\mu(b)+4}\{(x, y, t)\in\overline{\mathcal{M}}(a, c_{4})\times\overline{\mathcal{M}}(c_{4}, b)\times[0,1]|$

$\overline{s}_{\gamma_{1}}(c_{4}, b)(x, t)=\overline{s}_{\gamma_{2}}(c_{4}, b)(y, t)=0\}$ ,

$(x, y, t)\in\overline{\mathcal{M}}^{J}(a, c_{2})\times\overline{\lambda\Lambda}(c_{2}, b)\times[0,1]$

$\cup\prod_{\mu(c_{2})=\mu(b)+2}\{(x, y, t)| \overline{s}_{\gamma_{1}}(a, c_{2})(x, t)=0or=\overline{s}_{\gamma_{2}}(c_{2}, b)(y, t) \}$

$\overline{s}_{\gamma_{1}}(c_{2}, b)(x, t)=0=\overline{s}_{\gamma 2}(a, c_{2})(y, t)$

$(x, y, t)\in\overline{\mathcal{M}}^{J}(a, c_{3})\times\overline{\mathcal{M}}(c_{3}, b)\times[0,1]$

$\overline{s}_{\gamma_{1}}(a, c_{3})(x, t)=0=\overline{s}_{\gamma_{2}}(c_{3}, b)(y, t)$

$|$
$\}$
.
$\cup\prod_{\mu(c_{3})=\mu(b)+3}\{(x, y, t)$
or
$\overline{s}_{\gamma_{1}}(c_{3}, b)(x, t)=0=\overline{s}_{\gamma_{2}}(a, c_{3})(y, t)$

Let , , $\Lambda_{0}$ $\Lambda_{5}$ $\Lambda_{1}$

, $\Lambda_{4}$
, $\Lambda_{2}$
, $\Lambda_{3}$
be the sets in the above formula, respectively.
We have

(12.7.1) $\beta\Lambda_{0}=\langle\partial_{\gamma_{1},\gamma 2}a, b\rangle$

,
(12.7.1) $\beta\Lambda_{5}=-\langle\partial_{\gamma_{1},\gamma_{2}}a, b\rangle$
.

For , $c\in Fl$ with

$a$ $\mu(a)=\mu(c)+4$ , we put

$\langle\varphi_{\gamma_{1},\gamma_{2}}a, c\rangle=\#\{(x, t)\in\overline{\mathcal{M}}(a, c)\times[0,1]|\overline{s}_{\gamma_{1}}(x, t)=\overline{s}_{\gamma_{2}}(x, t)=0\}$

.

Then we have

(12.7.3) $\beta\Lambda_{1}=\sum_{c_{1}}\langle\varphi_{\gamma_{1},\gamma_{2}}a, c_{1}\rangle\langle\partial c_{1}, b\rangle$

,

(12.7.4) $\#\Lambda_{4}=-\sum_{c_{4}}\langle\partial a, c_{4}\rangle\langle\varphi_{\gamma_{1},\gamma_{2}}c_{4}, b\rangle$

.
 Floer Homology for Oriented 3-Manifolds 75

To examine and , we remark that the sections $\#\Lambda_{2}$

can be $\#\Lambda_{3}$ $\overline{s}_{\gamma}(c, c)$

defined by an induction on $\mu(c)-\mu(c)$ . Then, we can assume the

following conditions (12.8). For , $c\in Fl$ with $\mu(c)=\mu(c)+2$ , we put $c$

$T(c, c)=\sup\{t|\exists x(x, t)\in\overline{\Sigma}_{\gamma}(c, c)\}$

,
$S(c, c)=\inf\{t|\exists x(x, t)\in\overline{\Sigma}_{\gamma}(c, c)\}$
.

(12.8.1) If $\mu(c)=\mu(c)+3=\mu(c^{JJ})+5$ , and if $t>T(c, c)$ then

$\overline{s}_{\gamma}^{f}(c, c)(x, t)=\overline{s}_{\gamma}(c, c)(x, 1)$

(12.8.2) If $\mu(c)=\mu(c)+2=\mu(c)+5$ , and if $t<S(c, c)$ , then

$\overline{s}_{\gamma}^{f}(c, c)(x, t)=\tilde{s}_{\gamma}(c, c)(x, 0)$

Using (12.8.1), we can prove:

$\Lambda_{2}=\prod_{c_{2}}\{x\in\overline{\lambda\Lambda}(a, c_{2})|s_{\gamma 1}(x)=0\}\times$

$\{(y, t)\in\overline{\mathcal{M}}(c_{2}, b)\times[0,1]|\overline{s}_{\gamma_{2}}(y, t)=0\}$

$\cup\prod_{c_{2}}\{x\in\overline{\mathcal{M}}(a, c_{2})|s_{\gamma_{2}}(x)=0\}\times$

$\{(y, t)\in\overline{\Lambda 4}(c_{2}, b)\times[0,1]|\overline{s}_{\gamma_{1}}(y, t)=0\}$

.

Therefore
(12.9.1) $\#\Lambda_{2}=-\sum_{c_{2}}\langle\partial_{\gamma_{1}}a, c_{2}\rangle\langle\varphi_{\gamma_{2}}c_{2}, b\rangle-\sum_{c_{2}}\langle\partial_{\gamma_{2}}a, c_{2}\rangle\langle\varphi_{\gamma_{1}}c_{2}, b\rangle$
.

Similarly, using (12.8.2), we can prove:

(12.9.2) $\#\Lambda_{3}=\sum_{c_{3}}\langle\varphi_{\gamma_{1}}a, c_{3}\rangle\langle\partial_{\gamma_{2}}c_{3}, b\rangle+\sum_{c_{3}}\langle\varphi_{\gamma_{2}}a, c_{3}\rangle\langle\partial_{\gamma_{1}}c_{3}, b\rangle$

.

By (12.6.1),(12.7),(12.9), we have
(12.10)
$\partial_{\gamma_{1},\gamma_{2}}+\varphi_{\gamma_{1}}\partial_{\gamma_{2}}+\varphi_{\gamma_{1}}\partial_{\gamma_{2}}+\varphi_{\gamma_{1},\gamma_{2}}\partial=\partial_{\gamma_{1},\gamma_{2}}+\partial_{\gamma_{1}}\varphi_{\gamma_{2}}+\partial_{\gamma_{2}}\varphi_{\gamma_{1}}+\partial\varphi_{\gamma_{1},\gamma_{2}}$
.

Now we put
$\varphi(\gamma_{1}\gamma_{2}\otimes[a])=\gamma_{1}\gamma_{2}\otimes[a]+\gamma_{1}\otimes\varphi_{\gamma 2}[a]+\gamma_{2}\otimes\varphi_{\gamma 1}[a]+1\otimes\varphi_{\gamma_{1},\gamma 2}[a]$

$\psi(\gamma_{1}\gamma_{2}\otimes[a])=\gamma_{1}\gamma_{2}\otimes[a]-\gamma_{1}\otimes\varphi_{\gamma 2}[a]-\gamma_{2}\otimes\varphi_{\gamma_{1}}[a]$

$-1\otimes(\varphi_{\gamma_{1},\gamma 2}+\varphi_{\gamma 1}\varphi_{\gamma 2}+\varphi_{\gamma 2}\varphi_{\gamma_{1}})[a]$

.
76 K. Fukaya

Formulas (12.2.1),(12.2.2),(12.2.3) follow immediately from (12.5) and

(12.10). The proof of Theorem 12.2 is now complete.

Next we shall prove the following:

Lemma 12.11. Let , , , be closed loops on $M$ with $\gamma_{1}$ $\gamma_{2}$ $\gamma$
$\gamma$ $[\gamma_{1}]+$

in $H_{1}(M;Z)$ . Then we can fifind collections of sections

$[\gamma_{2}]=[\gamma]$

, with (12.1.1), (12.1.2) such that the

$s_{\gamma_{1}}(c, c),s_{\gamma_{2}}(c, c)$ $s_{\gamma}(c, c),s_{\gamma}(c, c)$

corresponding boundary operators satisfy

(12.11.1) $\partial_{\gamma_{1}}+\partial_{\gamma_{2}}=\partial_{\gamma}$

(12.11.2) $\partial_{\gamma_{1},\gamma}+\partial_{\gamma_{2},\gamma}=\partial_{\gamma,\gamma}$
.

Let $\mu(a)=\mu(b)+3$ . Consider

Proof. . (We do not divide $C\overline{\mathcal{M}}(a, b)$

it by the $R$ action.) Let be a surface on $M\times R$ which is asymptotic $\Sigma$

to as , and to
$(\gamma_{1}\cup\gamma_{2})\times R$
as . Using the Dirac $ t\rightarrow-\infty$ $\gamma\times R$ $t$ $\rightarrow\infty$

operator on , we can define a line bundle $\Sigma$

on $\mathcal{L}_{\Sigma}^{(2)}(a, b)$
$C\overline{\mathcal{M}}(a, b)=$

R. We put
$C\overline{\mathcal{M}}(a, b)\times$

$CC\overline{\mathcal{M}}(a, b)=C\overline{\mathcal{M}}(a, b)\times[-\infty, \infty]$

.

By construction and Theorem 4.9, the bundles $\mathcal{L}_{\Sigma}^{(2)}(a, b)$

on $C\overline{\mathcal{M}}(a, b)$
,
and on , and on
$\mathcal{L}_{\gamma_{1}}^{(2)}(a, b)\otimes \mathcal{L}_{\gamma_{2}}^{(2)}(a, b)$ $C\overline{\mathcal{M}}(a, b)\times\{-\infty\}$ $\mathcal{L}_{\gamma}^{(2)}(a, b)$

can be patched together to give a line bundle over

$C\overline{\mathcal{M}}(a, b)\times\{\infty\}$

. We extend the sections

$CC\overline{\mathcal{M}}(a, b)$
and to $s_{\gamma_{1}}(a, b)\otimes s_{\gamma_{2}}(a, b)$ $s_{\gamma}(a, b)$

a section on . Then, by an argument similar to the proof of

$CC\overline{\mathcal{M}}(a, b)$

Theorem 12.2, we can find such that $\varphi_{\gamma}$

$\partial_{\gamma}-(\partial_{\gamma_{1}}+\partial_{\gamma 2})=\partial\varphi_{\gamma}-\varphi_{\gamma}\partial$
.

Using this map , we can modify the section $\varphi_{\gamma}$ $s_{\gamma}$ such that (12.11.1) is
satisfied. The proof of (12.11.2) is similar.

Finally, we discuss what happens when $s\geq 1$ in case $H_{1}(M;Z)$ has

a torsion, and when in case $H_{1}(M;Z)$ is torsion free. $s$ $\geq 3$

Suppose first that $H_{1}(M;Z)$ has a torsion, and $\mu(a)=\mu(b)+5$ .

In this case, there may be reducible connections and such that $c$
$c$

$G_{c}=G_{c}=U(1)$ and that $\mu(c)=\mu(c)+1=\mu(b)+2$ . Then

$dim\overline{\mathcal{M}}(a, c)=dim\overline{\mathcal{M}}(c, c)=dim\overline{.\wedge\Lambda}(c, b)=0$

.
 Floer Homology for Oriented 3-Manifolds 77

The set may have a 0 dimensional orbit

$\overline{\sqrt{}\vee l}(c, c)$
which $\overline{\mathcal{M}}_{red}^{J}(c, c)$

consists only of reducible connections. (See Theorem 5.6.) A neighbor-

hood of each point of , in
$\overline{\mathcal{M}}(a, c)\times\overline{\mathcal{M}}_{red}(c, c)\times\overline{\mathcal{M}}(c, b)$ $C\overline{\mathcal{M}}^{J}(a, b)$

is identified to $\times(0, \infty]\times U(1)/\sim$

$(0, \infty]$
, where $(t, s, g_{1})\sim(t, s, g_{2})$

if and only if $=\infty$ or $ s=\infty$ . $t$

Here $\{\infty\}\times(0, \infty)\times U(1)/\sim$
and $\times\{\infty\}\times U(1)/\sim are$ identified to
$(o, \infty)$
and $\overline{\lambda\Lambda}(a, c)\times\overline{\lambda\Lambda}^{J}(c, b)$

respectively. The bundle

$\overline{\mathcal{M}}(a, c)\times\overline{\mathcal{M}}(c, b)$
is extended out- $\mathcal{L}_{\gamma}^{(2)}(a, b)$

side $\infty\times\infty\times U(1)/\sim=point$ . The neighborhood of this point is a cone

of . (It may be more natural to regard that this
$S^{2}$
has two singular $S^{2}$

points.)
Using the basis of $H_{1}(M;Z)$ , chosen at the beginning of \S 2, we
$[\ell_{i}]$

can find such that $\ell_{i_{0}}$

(12.12.1) $c(\ell_{i})=c(\ell_{i})$ if $i\neq i_{0}$ .

(12.12.2) $c(\ell_{i_{0}})=1$ , $c(\ell_{i_{O}})=-1$ .

In this case we can prove that the restriction of the line bundle $\mathcal{L}_{p_{i_{0}}}^{(2)}(a, b)$

to this is nontrivial. (Its chern number is

$S^{2}$ $\pm 1.$
) (See the proof of
Lemma 12.13 below.) Then the formula

$\partial_{\gamma}\partial+\partial_{\gamma}\partial=0$

does not hold in general.

Next suppose that $H_{1}(M;Z)$ is torsion free. Let and be reducible $c$
$c$

connections such that $G_{c}=G_{c}=SU(2)$ , , $G_{A}=U(1)$ , $A\in\overline{\mathcal{M}}^{J}(c, c)$

$\mu(c)=\mu(c)+3$ . Then, if , $b\in Fl$ and if , $(c, b)$ , $a$ $\mathcal{M}(a, c)\neq\emptyset$ $\neq\emptyset$

then $\mu(a)\geq\mu(c)+4,\mu(b)\leq\mu(c)-1$ . Hence, the first case we are to

examine is the case when $\mu(a)=\mu(b)+8=\mu(c)+7=\mu(c)+4$ . In this
case,
$dim\overline{\mathcal{M}}(a, c)=dim\overline{\mathcal{M}}_{red}(c, c)=dim\overline{\mathcal{M}}(c, b)=0$
.

Here is the component of $[A]$ , which consists of one point.

$\overline{\mathcal{M}}_{red}(c, c)$

By Theorem 7.1 a neighborhood of each point of

$\overline{\mathcal{M}}(a, c)\times\overline{\Lambda 4}_{red}(c, c)\times\overline{\mathcal{M}}(c, b)$

in $C\overline{\mathcal{M}}(a, b)$
is

$(\frac{SO(3)\times SO(3)}{U(1)}\times(0, \infty]^{2})/\sim$

78 K. Fukaya

as in (7.1.4). In other words, it is a cone of

$where\sim is$ $CP^{3}/Z_{2}=X$ .
(See the proof of Lemma 12.13.) Here acts by $Z_{2}$

$\tau[z_{0}, z_{1}, z_{2}, z_{3}]=[z_{0}, z_{1}, -z_{2}, -z_{3}]$ .

The fixed points set of this action has two components. The fixed points
correspond to the singular points of $X$ . Those singular locus are identi-
fied to

$(\frac{SO(3)\times SO(3)}{U(1)}\times\{\infty\}\times(0, \infty))/\sim$

$\subset\overline{\mathcal{M}}^{J}(a, c)\times\overline{\mathcal{M}}^{J}(c, b)$

,

and

$(\frac{SO(3)\times SO(3)}{U(1)}\times(0, \infty)\times\{\infty\})/\sim$

$\subset\overline{\mathcal{M}}^{J}(a, c)\times\overline{\mathcal{M}}(c, b)$

,

respectively. We can find $\ell_{i_{0}}$

such that (12.12.1) and (12.12.2) are satis-
fied.
Lemma 12.13.

$\int_{X}c^{1}(\mathcal{L}_{\ell_{i_{0}}}^{(2)}(a, b))^{3}=\pm 4$
.

Let
Proof. be a representative of $a_{t}^{0}$
, (used in \S 8.) $\overline{\mathcal{M}}(c, c)=point$

On , converges to the trivial connection as goes to

$\ell_{i_{0}}\times R$ $a_{t}^{0}$
, and, $t$
$-\infty$

as goes to
$t$
, it converges to a flat connection -1 whose holonomy,
$\infty$

: $Z=\pi_{1}(S^{1})\rightarrow SU(2)$ is given by $\rho_{-1}(1)=-1$ .

$\rho_{-1}$

Sublemma 12.14.
$Index(6_{a_{t}^{0}}+\epsilon)=-1$ .

Proof. We put $S^{1}=R/2\pi Z$ . Let $x$ be the coordinate of $S^{1}$

. We
have
$6_{trivial}=\frac{\partial}{\partial t}+i\frac{\partial}{\partial x}$
.

We can perturb $a_{t}^{0}$

so that it is a connection with holonomy

$\left(\begin{array}{ll}e^{\pi it} & 0\\0 & e^{-\pi it}\end{array}\right)$

.
 Floer Homology for Oriented 3-Manifolds 79

( is a trivial connection and

$a_{0}^{0}$
$a_{1}^{0}=-1.$ ) Then the spectral flow cor-
responding to the operator $ 6_{a_{t}^{0}}+\epsilon$ is as in Figure 9. (Here we take
$\epsilon>0.)$

eigenvalue

-1 1

Figure 9.

The sublemma follows.

Remark 12.15. In our case, the half spin $bund1e\otimes C^{2}$ together with
connection splits to the direct sum of two complex line bundles. The
$a_{t}^{0}$

dotted lines in Figure 9 correspond to the second factor and the others
to the first factor.
The group $U(1)=I_{a_{t}^{0}}$ acts on the eigenspaces, and the index in
Sublemma 12.14 can be regarded as an element of the representation
ring $R(U(1))\sim Z[t, t^{-1}]$ . Here be the representation corresponding to
$t$

$z\mapsto z$ and to$t^{-1}$

, where we identify $U(1)=\{z||z|=1\}$ . By
$z\rightarrow z^{-1}$

Figure 9, The index is equal to $-t^{-1}$ .

If we choose then the index is .
$\epsilon<0$ $t$

Now we consider the map $\pi$

: con-
$SU(2)\times SU(2)\rightarrow \mathcal{M}(c, c)$

structed in Theorem 5.4. Let $\mathcal{L}_{i}(c, c)$

be the line bundle defined in \S 7.
(We have not yet divided it by $G_{c}\times G_{c}.$ ) is trivial.
$\pi^{*}\mathcal{L}_{i}(c, c)$
80 K. Fukaya

On $SU(2)\times SU(2)$ , the group $U(1)=I_{a_{t}^{0}}$ acts by

$h(g_{1}, g_{2})=(g_{1}h, h^{-1}g_{2})$ .

This action lifts to . The quotient is identified to the restric-

$\pi^{*}(\mathcal{L}_{i}(c, c))$

tion of to the image of , which is dieomorphic to $ SU(2)\times$

$\mathcal{L}_{i}(c, c)$ $\pi$

$SU(2)/U(1)$ . By Sublemma 12.14 and Remark 12.15, the action of $U(1)$

on is given by
$\pi^{*}(\mathcal{L}_{i}(c, c))$

(12.16) $h((g_{1}, g_{2}),$ $v)=((g_{1}h^{-1}, hg_{2})$ , $hv)$ ,

(in both cases $\epsilon>0$

and $\epsilon<0.$
)
We put
$\hat{X}=\underline{SU(2)\times SU(2)\times[0,1]}\sim$

where
$(g_{1}, g_{2},0)=(g_{1}, g_{2},0)$ ,
$(g_{1}, g_{2},1)=(g_{1}, g_{2}, 1)$ .

$\hat{X}$

is dieomorphic to $S^{7}$
. By Theorem 7.1,
$\hat{X}$

$X=\overline{U(1)\times Z_{2}}$
.

Here $h\in U(1)$ and $\tau=-1\in Z_{2}$ acts on $\hat{X}$

by
$h([g_{1}, g_{2}, t])=[g_{1}h, h^{-1}g_{2}, t]$ ,
$\tau([g_{1}, g_{2}, t])=[-g_{1}, g_{2}, t]$ .

Hence $\hat{X}/U(1)\simeq CP^{3}$

. By (12.16), the bundle $\mathcal{L}_{i}(a, b)$
on $\hat{X}/U(1)\subset$

is isomorphic to the canonical bundle on

$C\mathcal{M}(a, b)$ $CP^{3}$
. Hence, its
Chern class is equal to the generator, . Therefore, $u$

$\int_{X}c^{1}(\mathcal{L}_{i}^{(2)}(a, b))^{3}=\int_{CP^{3}}(2u)^{3}/2=4$ .

The proof of Lemma 12.13 is now complete.

Using Lemma 12.13, we can discuss as in the proof of Theorem 1.10,
to show

$\sum_{\alpha_{1}\cup\alpha_{2}=\alpha}\partial_{\gamma_{\alpha_{1}}}\partial_{\gamma_{\alpha_{2}}}=4\sum_{c,c}\#\overline{\mathcal{M}}(a, c)$
. $\Downarrow\overline{\mathcal{M}}(c, b)$
,
 Floer Homology for Oriented 3-Manifolds 81

in the case when . $\alpha=(\ell_{i_{0}}, \ell_{i_{0}}, \ell_{i_{0}})$

It might be possible to define an invariant $mod 4$ using the above

formula. But the author does not try to do it here, because he suspects
if it is a correct way.
From the above observation, it seems that we need to examine the
reducible connections more seriously when we generalize the invariant
for larger . $s$

\S 13. Independence of the metrics and the perturbations

The proof of Theorem 1.14 is based on an argument similar to one
in 7-12 and [F]. Let , be two metrics on $M$ and be two $\sigma_{1}$ $\sigma_{2}$
$f_{1},f_{2}$

perturbations as in 2,3. Let and be the set of solutions of $Fl_{1}$ $Fl_{2}$

$*_{\sigma_{1}}F^{a}-grad_{a}f_{1}=0$ ,

and
$*_{\sigma_{2}}F^{a}-grad_{a}f_{2}=0$ ,
respectively. Let and be corresponding complexes
$(C_{(1)}^{s}, \partial^{1})$ $(C_{(2)}^{s}, \partial^{2})$

constructed in \S 12. Choose a family of metrics such that $g_{t}$

(13.1.1) $\sigma_{t}=\sigma_{1}$ for $t$

$<-1$ .
(13.1.2) $\sigma_{t}=\sigma_{2}$ for $t>1$ .

Choose $\chi$ such that

$\chi(t)=1$ for $t>1$ ,
$\chi(t)=0$ for $t<0$ .

Let $\sigma_{t}$
be the metric $\sigma_{t}\oplus dt^{2}$
on $M\times R$ . We consider the equation

(13.2) $F^{A}-\overline{*}_{\sigma_{t}}F^{A}-\chi(-t)(grad_{a_{t}}f_{1}\wedge dt-*_{\sigma_{t}}grad_{a_{t}}f_{1})$

$-\chi(t)(grad_{a_{t}}f_{2}\wedge dt-*_{\sigma_{t}}gradf_{2})=0$ ,

for $A\in A_{p\delta},(a, b)$ . (Compare (3.6).) Here $a\in Fl_{1}$ and $b\in Fl_{2}$ . The
linearization of (13.2) is given by
$0=D_{A}(u, \varphi)=$

$-\frac{\partial u}{\partial t}+(*_{\sigma_{t}}d_{a_{t}}-\psi_{t}-\chi(-t)Hess_{a_{t}}f_{1}-\chi(t)Hess_{a_{t}}f_{2})\wedge u+d_{a_{t}}\varphi$

Here , e.t.c are the same as in (3.8). Let

$u$ $\varphi$
and be the operators $D_{A}^{1}$ $D_{A}^{2}$

in (3.8) for , and $f=f_{1}$ , , respectively.

$\sigma=\sigma_{1}\oplus dt^{2}$ $\sigma_{2}\oplus dt^{2}$ $f_{2}$
82 K. Fukaya

Lemma 13.3. If $A\in A_{l,\gamma}(a, b)$ with $a\in Fl_{1}b\in Fl_{2}$ , then

$dimCoker$ $D_{A}<\infty$ .

Proof. If not we have $(u_{i}, \varphi_{i})$

such that

$D_{A}^{*}(u_{i}, \varphi_{i})=0$ ,
$<(u_{i}, \varphi_{i})$
, $(u_{j}, \varphi_{j})>=\delta_{i,j}$ .

Then, by elliptic regularity, we have $|t_{i}|\rightarrow\infty$

such that

$|$

$(u_{i}(x_{0}, t_{i})$ , $\varphi_{i}(x_{0}, t_{i}))|>C_{0}>0$ .

We may assume that . Put $u_{i}(t, x)=u_{i}(t-t_{i}, x)$ , $ t_{i}\rightarrow\infty$ $\varphi_{i}(t, x)=$

. By taking a subsequence we may assume that

$\varphi_{i}(t-t_{i}, x)$ $(u_{i}, \varphi_{i})$

converges to with respect to the topology on each compact

$(\text{{\it \^{u}}}, \hat{\varphi})$
$C^{\infty}$

set. Then we have

$D_{b}^{(2)*}$ $(\text{{\it \^{u}}}, \hat{\varphi})=0$

$(\text{{\it \^{u}}}, \hat{\varphi})\neq 0$

.

This contradicts (2.6).

Using Lemma 13.3, we can apply the argument of [D3] to obtain a

perturbation , such that the linearized operator of
$Q(\cdot)$ $D_{A}$

(13.4) $F^{A}-\overline{*}_{\sigma_{t}}F^{A}-\chi(-t)(grad_{a_{t}}f_{1}\wedge dt-*_{\sigma_{t}}grad_{a_{t}}f_{1})$

$-\chi(t)(grad_{a_{t}}f_{2}\wedge dt-*_{\sigma_{t}}grad_{a_{t}}f_{2})+Q(A)=0$ .

is surjective. Here $Q(A)$ depends only on a restriction of $A$ to $ M\times$

$[-1,1]$ and its support is also contained in it. Let be the set $\overline{\sqrt{}\vee 1}(a, b)$

of solutions of (13.4) divided by gauge transformations. Let $\overline{\mathcal{M}}_{(1)}(a, b)$

and be the set of solutions of (3.6) for

$\overline{\mathcal{M}}_{(2)}(a, b)$
, $f=f_{1}$ and $\sigma=\sigma_{1}$

, $f=f_{2}$ , divided by the gauge transformations and $R$ action,

$\sigma=\sigma_{2}$

respectively.

Theorem 13.5. For $a\in Fl_{1}$ and $b\in Fl_{2}$ , let $C\overline{\mathcal{M}}(a, b)$
be the
 Floer Homology for Oriented 3-Manifolds 83

disjoint union of
$\overline{\mathcal{M}}(a, b)$
,

$\overline{\mathcal{M}}(a, c_{0})\times\prod_{i=0}^{k-1}\overline{\Lambda 4}_{(2)}(c_{i}, c_{i+1})\times\overline{\mathcal{M}}_{(2)}^{J}(c_{k}, b)$

,

$\overline{\mathcal{M}}_{(1)}^{J}(a, c_{0})\times\prod_{i=0}^{k-1}\overline{\lambda\Lambda}_{(1)}(c_{i}, c_{i+1})\times\overline{\lambda\Lambda}(c_{k}, b)$

,

$\overline{\mathcal{M}}_{(1)}^{J}(a, c_{0})\times\prod_{i=1}^{k_{0}-1}\overline{\mathcal{M}}_{(1)}(c_{i}, c_{i+1})\times\overline{\mathcal{M}}(c_{k_{0}}, c_{k_{0}+1})$

$\times\prod_{i=k_{0}+1}^{k-1}\overline{\mathcal{M}}_{(2)}^{J}(c_{i}, c_{i+1})\times\overline{M}_{(2)}^{J}(c_{k}, b)$ .

Then $C\overline{\mathcal{M}}(a, b)$

has a smooth structure with properties similar to (7.1.1)
$-(7.1.7)$ .

The proof is similar to the proof of Theorem 7.1 and is omitted.

We remark here the reason why we need to fix a basis of $H_{1}(M;Z)$ .
Let be the maps defined in Theorem 5.1 for metrics
$\mu_{1},\mu_{2}$ and $\sigma_{1},\sigma_{2}$

let $f_{1}$
and be functions we used in sections 2 and 3. If we use the
$f_{2}$

same basis of ( $M$ ; Z) (or more precisely $H_{1}(M;Z)\otimes Z_{2}$ ), then we have
$H_{1}$

$\mu_{1}(c)=\mu_{2}(c)$ for each reducible connection . This fact is essential for $c$

the argument of the rest of this section. In fact, suppose, for example,
there exists reducible such that $c$

$\mu_{1}(c)=\mu_{2}(c)-10$ .

Then for some $a\in Fl_{1},b\in Fl_{2}$ with $\mu_{1}(a)=\mu_{2}(b)+1$ , the space $\overline{\mathcal{M}}(a, b)$

may have an end described by

$\overline{\Lambda 4}_{(1)}(a, c)\times\overline{\mathcal{M}}(c, c)\times\overline{\mathcal{M}}_{(2)}(c, b)$

.

And $\mu_{1}(a)-\mu_{1}(c)$ can be greater than 7. Therefore, in the compactifica-

tion of the end we discussed at the end of \S 12 can appear. These
$\overline{\mathcal{M}}_{(1)}$

ends can cause serious problem for the argument of the well definedness.
The point is that the virtual dimension of is -10 but we can $\overline{\mathcal{M}}(a, b)$

not find perturbation to make it empty

The author has no explicit example which shows that our invariant
does depend on the choice of the basis of $H_{1}(M;Z)$ . But it seems quite
unlikely that it is independent.
84 K. Fukaya

We return to the proof of invariance. For $\gamma\simeq S^{1}\subset M$

, we define
bundles
$\mathcal{L}_{\gamma,1}^{(2)}(a, a)$
on $\overline{\sqrt{}\backslash \Lambda}_{(1)*}(a, a)$
,
$\mathcal{L}_{\gamma,2}^{(2)}(b, b)$
on $\overline{\mathcal{M}}_{(2)*}(b, b)$
,
$\mathcal{L}_{\gamma}^{(2)}(a, b)$
on $\overline{\mathcal{M}}(a, b)$
.

Theorem 13.6. The tensor products of $\mathcal{L}_{\gamma,1}^{(2)},\mathcal{L}_{\gamma,2}^{(2)}$

, and $\mathcal{L}_{\gamma}^{(2)}$

can
be patched together to give a line bundle on $C\overline{\mathcal{M}}_{*}(a, b)$
.

The proof is the same as the proof of Theorem 7.3.

Now we define $\varphi$
: $(C_{(1)}^{s}, \partial^{1})\rightarrow(C_{(2)}^{s}, \partial^{2})$ . We put

$<\varphi_{0}(a)$ , $b>=\beta\overline{\mathcal{M}}(a, b)$

if $\mu(a)=\mu(b)$ . (Here $\#$

is the same as in 12.) Set

$\varphi[a]=\sum_{b}<\varphi_{\emptyset}a$
, $b>[b]$ .

This defines the map $\varphi$

: $C_{(1)}^{0}\rightarrow C_{(2)}^{0}$ .
Next we fix sections , , to , $s_{\gamma}(a, b)$ $s_{\gamma,1}(a, a)$ $s_{\gamma,2}(b, b)$
$\mathcal{L}_{\gamma}^{(2)}(a, b)$

, such that (12.1.2) holds and that they can be

$\mathcal{L}_{\gamma,1}^{(2)}(a, a)$ $\mathcal{L}_{\gamma,2}^{(2)}(b, b)$

patched together to give a section of the line bundle obtained in Theorem

13.6. Now, for $\mu(a)=\mu(b)+2$ , we put
$<\varphi_{\gamma}a$ , $b>=\#\{x\in\overline{\lambda\Lambda}(a, b)|s_{\gamma}(x)=0.\}$ .

For $\mu(a)=\mu(b)+4$ , we put

$<\varphi_{\gamma_{1},\gamma_{2}}a$ , $b>=\#\{x\in\overline{\mathcal{M}}(a, b)|s_{\gamma_{1}}(x)=s_{\gamma 2}(x)=0\}$ .

Set

$\varphi_{\gamma}[a]=\sum_{b}<\varphi_{\gamma}a$
, $b>[b]$ ,

$\varphi_{\gamma_{1},\gamma 2}[a]=\sum_{b}<\varphi_{\gamma_{1},\gamma_{2}}a$
, $b>[b]$ .

Lemma 13.7. $If|\alpha|<3$ , the

$\sum_{\alpha_{1}\cup\alpha_{2}=\alpha}\partial_{\alpha_{1}}^{2}\varphi_{\alpha_{2}}=\sum_{\alpha_{1}\cup\alpha_{2}=\alpha}\varphi_{\alpha_{1}}\partial_{\alpha_{2}}^{1}$
.
 Floer Homology for Oriented 3-Manifolds 85

( $If|\alpha|>0$ we assume that $H_{1}(M;Z)$ is torsion free.)

The proof is the same as the proof of Theorem 1.10 in \S 12. Put

$\varphi(\gamma_{\alpha}\otimes a)=\sum_{\alpha_{1}\cup\alpha_{2}=\alpha}\gamma_{\alpha_{1}}\otimes\gamma_{\alpha_{2}}a$
.

Lemma 13.7 implies that $\varphi$

: $(C_{(1)}^{s}, \partial^{1})\rightarrow(C_{(2)}^{s}, \partial^{2})$ is a chain map.

Lemma 13.8. The chain map modulo chain homotopy is inde- $\varphi$

pendent to the choice of the homotopy of the metrics and the pertur- $\sigma_{t}$

bation $Q$ in (13.4).

Let
Proof be the homotopies and perturbations and
$\sigma_{t}^{1},\sigma_{t}^{2},Q_{1},Q_{2}$

, be corresponding chain maps. Choose homotopies

$\varphi_{1}$ $\varphi_{2}$
and $\sigma_{t}^{u}$ $Q_{u}$

among them. Let be the set of solutions of (13.4) for

$\overline{\mathcal{M}}_{u}(a, b)$
$\sigma_{t}=\sigma_{t}^{u}$

and $Q=Q_{u}$ . Let be the disjoint union of $C\overline{\mathcal{M}}_{u}(a, b)$

$\overline{\mathcal{M}}_{u}(a, b)$

$\overline{\mathcal{M}}_{u}(a, c_{0})\times\prod_{i=0}^{k-1}\overline{\mathcal{M}}_{(2)}(c_{i}, c_{i+1})\times\overline{\lambda\Lambda}_{(2)}(c_{k}, b)$

,

$\overline{\mathcal{M}}_{(1)}(a, c_{0})\times\prod_{i=0}^{k-1}\overline{\mathcal{M}}_{(1)}^{J}(c_{i}, c_{i+1})\times\times\overline{\mathcal{M}}_{u}(c_{k}, b)$

,

$\overline{\mathcal{M}}_{(1)}(a, c_{0})\times\mathring{\prod_{i=0}^{k-1}}\overline{\mathcal{M}}_{(1)}(c_{i}, c_{i+1})\times\overline{\mathcal{M}}_{u}(c_{k_{0}}, c_{k_{0}+1})$

$\times\prod_{i=k_{O}+1}^{k-1}\overline{\mathcal{M}}_{(2)}(c_{i}, c_{i+1})\times\overline{\mathcal{M}}_{(2)}(c_{k}, b)$

.

(Here we do not assume that $\mu(a)>\mu(c_{0})>\cdots>\mu(c_{k})>\mu(b).$ ) (Note

$that\mathcal{M}_{(1)}Put(a, b)\neq \mathcal{M}_{1}(a, b).)$

$\mathcal{H}\overline{\mathcal{M}}(a, b)=\cup\overline{\mathcal{M}}_{u}(a, b)\times\{u\}$

,

$CH\overline{\mathcal{M}}(a, b)=\cup C\overline{\mathcal{M}}_{u}(a, b)u\times\{u\}$

.

Theorem 13.9. We can take and such that $\sigma_{t}^{u}$ $Q_{u}$ $C7\{\overline{\mathcal{M}}(a, b)$

hasa smooth structure which has properties similar to (7.1.1)-(7.1.7).

86 K. Fukaya

The proof of Theorem 13.9 is a bit more dicult than that of The-
orem 7.1. The reason is that we can not assume that the operator
obtained by linearizing (13.4) is surjective for every , (even if we
$D_{A}^{(u)}$
$w$

choose and to be generic.) Then we have to use the Kuranishi

$\sigma_{t}^{u}$ $Q_{u}$

map as in [T2], [D2]. For simplicity we prove the case $\mu(a)=\mu(b)$ . Here
$a\in Fl_{1},b\in Fl_{2}$ . Then $dimH\mathcal{M}(a, b)=1$ . In this case, Theorem 13.9

follows immediately from the following two lemmas.

Lemma 13.10. Suppose that the sequence is $(A_{i}, u_{i})\in H\overline{\mathcal{M}}(a)b)$

unbounded. Then, by taking a subsequence if necessary, there exist either

$c\in Fl_{1}$ , , , $t_{i}$
with $\mu(c)=\mu(a)+1$ or
$B\in\overline{\mathcal{M}}_{u}(a, c)$ $C\in\overline{\mathcal{M}}_{(2)}(c, b)$ $ c\in$

, ,
$Fl_{2}$ $t_{i}$
, with $\mu(c)=\mu(a)-1$ such that
$B\in\overline{\mathcal{M}}_{(1)}(a, c)$ $C\in\overline{\mathcal{M}}_{u}(c, b)$

the Conditions (13.10.1) $-(13.10.3)$ or (13.10.1) $-(13.10.3)$ below hold.

(13.10.1) $u_{i}\rightarrow u$

(13.10.2) $|A_{i}(x, t)-B(x, t)|\rightarrow 0$

(13.10.3) $|A_{i}(x, t-t_{i})-C(x, t)|\rightarrow 0$

(13.10.2) $|A_{i}(x, t+t_{i})-B(x, t)|\rightarrow 0$

(13.10.3) $|A_{i}(x, t)-C(x, t)|\rightarrow 0$ .

(See Figure 10.) Note that for generic . $\overline{\mathcal{M}}_{u}(a, c)=\emptyset=\overline{\mathcal{M}}_{u}(c, b)$ $u$

(The virtual dimension of them is-l.) But -parameter family of-l-

dimensional spaces is a finite set. Hence by a generic choice of and $\sigma_{t}^{u}$

there exist a finite number of

$Q_{u}$

, for which or is $w$ $s$ $\overline{\mathcal{M}}_{u}(a, c)$ $\overline{\mathcal{M}}_{u}(c, b)$

nonempty.

$|$
 Floer Homology for Oriented 3-Manifolds 87

Lemma 13.11. Let $B\in\overline{\mathcal{M}}_{u}(a, c)$

, Then there
$C\in\overline{\mathcal{M}}_{(2)}^{J}(c, b)$
.
exist $u(v)$ : $(0, \infty)\rightarrow 0,1$ , $A(v)\in\overline{\mathcal{M}}_{u(v)}(a, b)$ and $t(v)$ , $t(v)\in R$ , such
that

(13.11.1) $\lim_{v\rightarrow\infty}u(v)=u$

(13.11.2) $\lim_{v\rightarrow\infty}|A(v)(x, t -t(v))-B(x, t)|=0$

(13.11.3) $\lim_{v\rightarrow\infty}|A(v)(x, t+t(v))-C(x, t)|=0$ .

Moreover, if satisfifies (13.10.1) (13.10.3) then

$A_{i}$ -
$[A_{i}]=[A(v_{i})]$ for
large . A similar statement holds for .
$i$ $c$

The proof of Lemma 13.10 is similar to the proof in \S 9 and is omitted.

Before proving Lemma 13.11 we complete the proof of Lemma 13.8 in
the case when $s=0$ .
In this case, Theorem 13.9 implies
$\partial H\overline{\mathcal{M}}(a, b)-\overline{\mathcal{M}}_{1}(a, b)-\overline{\mathcal{M}}_{2}(a, b)$

$=\cup\overline{\mathcal{M}}_{u}(a, c)u,c\times\overline{\mathcal{M}}_{(2)}(c, b)\cup u,c\cup\overline{\lambda\Lambda}_{(1)}(a, c)\times\overline{\mathcal{M}}_{u}(c, b)$

.

We put

$<\Phi a$ , $c>=\sum\beta\overline{\mathcal{M}}_{u}(a, c)$

$<\Phi c$
, $b>=\sum Q\overline{\mathcal{M}}_{u}(c, b)$ ,

and

$\Phi[a]=\sum_{c}<\Phi a$ , $c>[c]$

$\Phi[c]=\sum_{b}<\Phi c$
, $b>[b]$ .

Then we have
$\varphi_{1}-\varphi_{2}=\partial\Phi-\Phi\partial$
.

Here and are the chain maps constructed using ,

$\varphi_{1}$ $\varphi_{2}$
and , , $\sigma_{t}^{1}$
$Q_{1}$ $\sigma_{t}^{2}$
$Q_{2}$

respectively. This proves Lemma 13.8 when $s=0$ . The case when $s>0$
can be proved by combining the methods of \S \S 7- 12 and Theorem 13.9.
(In fact, the case $>0$ is simpler, because we do not have to use
$s$

Kuranishi map in that case.)

88 K. Fukaya

Proof of Lemma Let be the operator obtained by lin-

13.11. $D_{A}^{u}$

earizing the equation (13.4) for and $Q=Q_{u}$ . By the generic

$\sigma_{t}=\sigma_{t}^{u}$

choice of and we have $dimCoker$

$\sigma_{t}^{u}$
$D_{B}^{u}=1$ . We consider the set $X$
$Q_{u}$

of the connections which is a standard form of type $(\{B\}, \{C\}, \epsilon, T)$ . By
Remark 4.6, there exists a positive number , such that, if $A\in X$ and if $\lambda_{0}$

$|u-u|<\epsilon$ , then, there is exactly one eigenvalue of smaller than $D_{A}^{u}D_{A}^{u*}$

. Let
$\lambda_{0}$
be the orthonormal projection to this eigenspace, (which is
$\Pi_{I}$

isomorphic to $R$ ). Put $\Pi_{II}=identify$ . For $A\in A(a, b)$ , $u\in[0,1]$ $-\Pi_{I}$

we consider the equation

(13.12)
$\square _{II}(F^{A}-\overline{*}_{\sigma_{t}^{u}}F^{A}-\chi_{u}(-t)(grad_{a_{t}}f_{1}\wedge dt-*_{\sigma_{t}^{u}}grad_{a_{t}}f_{1})$

$-\chi_{u}(t)(grad_{a_{t}}f_{2}\wedge dt-*_{\sigma_{t}^{u}}grad_{a_{t}}f_{2})+Q_{u}(A))=0$ .

$|u$

Figure 11.

The set of solutions of (13.12) divided by gauge transformations

consists a 2-dimensional family Y. Let $Z$ be the set of solutions of (13.12)
for $A\in A(a, c)$ and $u\in[0,1]$ . $(dimZ=1.)$ Then, using the method of
the proof of Theorem 7.1, we can compactify $Y$ by adding $Z\times\{C\}$ . Put
$CY$ $=Y\cup(Z\times\{C\})$ . A neighborhood of $((B, u),$ $C)$ in $CY$ is identified

to $[0, 1)$ $\times(0,1)$ , where {0} $\times(0,1)\subset Z\times\{C\}$ . (See Figure 11.) For
 Floer Homology for Oriented 3-Manifolds 89

$(A, u)$ , we put

$f(A, u)=\Pi_{I}(F^{A}-\overline{*}_{\sigma_{t}^{u}}F^{A}-\chi_{u}(-t)(grad_{a_{t}}f_{1}\wedge dt-*_{\sigma_{t}^{u}}grad_{a_{t}}f_{1})$

$-\chi_{u}(t)(grad_{a_{t}}f_{2}\wedge dt-*_{\sigma_{t}^{u}}grad_{a_{t}}f_{2})+Q_{u}(A))$ .

We identify the image of to $R$ and regard as a function. Using the $\Pi_{I}$ $f$

decay estimate in \S 9 we can extend the function to a smooth function $f$

on $CY$ . The set of zeros of is identified to a neighborhood of $((B, u),$ $C)$ $f$

in . We consider the restriction of to {0} $\times(0,1)\subset Z$ . If

$CH\overline{\mathcal{M}}(a, b)$ $f$

we choose and generic, we may assume that the derivative of

$g_{t}^{u}$ $Q_{u}$

this restriction is nonzero at $((B, u)$ , $C)\in\{0\}\times(0,1)$ . It follows from

implicit function theorem that the zero of in $CY$ is dieomorphic to $f$

$[0, 1)$ where $0\in[0,1)$ corresponds to $((B, u),$ $C)$ . Lemma 13.11 follows
immediately.

The proof of Lemma 13.8 is now complete.

Next we take another metric and another perturbation . $\sigma_{3}$

$f_{3}$

Choose homotopies and from to and from to . $\sigma_{t}^{1,2}$ $\sigma_{t}^{2,3}$

$\sigma_{1}$ $\sigma_{2}$ $\sigma_{2}$ $\sigma_{3}$

Choose also perturbations and . Let and be the chain $Q_{1,2}$ $Q_{2,3}$ $\varphi_{1,2}$ $\varphi_{2,3}$

maps obtained by them, respectively.

Lemma 13.12. We can fifind homotopy of metric $\sigma_{t}^{1,3}$

from $\sigma_{1}$
to
$\sigma_{3}$
and a perturbation such that the chain map $Q_{1,3}$ $\varphi_{1,3}$ : $C_{(1)}^{s}\rightarrow C_{(3)}^{s}$

satisfifies
$\varphi_{3,2}\varphi_{1,2}=\varphi_{1,3}$ .

Proof. We put

$\sigma_{t}^{s}=\chi(-t-s)\sigma_{t+2s}^{1,2}+\chi(t-s)\sigma_{t-2s}^{2,3}$ .

We shift the perturbation by to the negative direction and shift $Q_{1,2}$ $2s$

by
$Q_{2,3}$ to the positive direction. Let
$2s$ be the sum of them. We $Q_{1,3}^{s}$

consider the equation

(13.13)
$F^{A}-\overline{*}_{\sigma_{t}^{s}}F^{A}-\chi(-t-s)(grad_{a_{t}}f_{1}\wedge dt-*_{\sigma_{t}^{s}}grad_{a_{t}}f_{1})$

$-\chi(t+s)\chi(s-t)(grad_{a_{t}}f_{2}\wedge dt-*_{\sigma_{t}^{s}}grad_{a_{t}}f_{2})$

$-\chi(t-s)(grad_{a_{t}}f_{3}\wedge dt-*_{\sigma_{t}^{s}}grad_{a_{t}}f_{3})+Q_{1,3}(A)=0$

Let be the set of solutions of (13.13) divided by gauge trans-

$\overline{\mathcal{M}}(s;a, e)$

formations. Let and be the moduli spaces used in

$\overline{\mathcal{M}}_{1,2}(a, b)$ $\overline{\mathcal{M}}_{2,3}(b, e)$
90 K. Fukaya

the definitions of $\varphi_{1,2}$ and $\varphi_{2,3}$ respectively. (Here $a\in Fl_{1}$ , $b\in Fl_{2}$ ,
$e\in Fl_{3}.)$

By using Remark 4.6, we can prove that the linearized equation for
(13.13) is surjective for suciently large . Consider the disjoint union $s$

of
$C\overline{\lambda\Lambda}(s;a, e)\times\{s\}$ $s$ $\in[s_{0}, \infty)$

and

$\prod_{i=-1}^{k_{0}-1}\overline{\mathcal{M}}_{(1)}^{J}(c_{i}, c_{i+1})\times\overline{\mathcal{M}}_{1,2}(c_{k_{O}}, c_{k_{0}+1})$

$\times\prod_{i=k_{0}+1}^{k_{1}-1}\overline{\mathcal{M}}_{(2)}(c_{i}, c_{i+1})\times\overline{\mathcal{M}}_{2,3}(c_{k_{1}}, c_{k_{1}+1})$

$\times\prod_{i=k_{1}+1}^{k_{2}}\overline{\Lambda 4}_{(3)}(c_{i}, c_{i+1})\times\{\infty\}$

.

(Here we put ) The later one is a compactification of

$a=c_{-1}$ , $e=c_{k_{2}+1}.$

. Let be the union. Using this mod-

$\bigcup_{b}\overline{\mathcal{M}}_{1,2}(a, b)\times\overline{\mathcal{M}}_{2,3}(b, e)$ $CC\overline{M}(a, e)$

uli space, the proof of the lemma goes in a way similar to the argument
of \S \S 7- 13.
Now we are in the position to complete the proof of Theorem 1.14.
Suppose , in Lemma 13.12. Then we can take a trivial homotopy
$\sigma_{1}=\sigma_{3}$

and $Q_{1,3}=0$ . In this case, it is easy to see that the corre-

$\sigma^{1,3}=\sigma_{1}$

sponding chain map is the identity map. Therefore by Lemma 13.12

and Lemma 13.8, is chain homotopic to identity. (In this case
$\varphi_{2,3}\varphi_{1,2}$

Thus the chain map

$\varphi_{2,3}=\varphi_{2,1}.)$ we constructed gives an iso- $\varphi_{1,2}$

morphisms on the homology groups. Also the isomorphism is canonical

because of Lemma 13.8. The proof of Theorem 1.14 is now complete.
The proof of the independence of the exact sequence 1.15 is similar.

References
[Ar] N. Arondazajn, A unique continuation theorem for solutions of elliptic
partial dierential equations or inequalities of the second order, J.
Math. Pures Appl., 36(9) (1957), 235-249.
[A] M. Atiyah, New invariants of three and four dimensional manifolds, in
The Mathematical Heritage of Herman Weyl, Proc. Sym. Pure.
Math. 48, American mathematical Society, 1988.
[AHS] M. Atiyah, N. Hitchin, I. Singer, Self-duality in four dimensional topol-
ogy, Proc. R. Soc. London A, 362 (1978), 425-461.
 Floer Homology for Oriented 3-Manifolds 91

[APS] M. Atiyah, N. Patodi, I. Singer, Spectral asymmetry and Riemannian

geometry III, Proc. Camb. Philos. Soc., 79 (1976), 71-99.
[AS] M. Atiyah, I. Singer, Index of Elliptic operators IV, Ann. of Math., 93
(1971), 97-118.
[BK] P. Buser, H. Karcher, Gromovs almost flat manifolds, Asterisque, 81
(1981), 1-148.
[CG] J. Cheeger, M. Gromov, Collapsing Riemannian manifolds while
keeping their curvatures bounded I, J. Dierential Geom., 23 (1986),
309-346.
[D1] S. Donaldson, An application of gauge theory to the topology of -man- $4$

ifolds, J. Dierential Geom., 18 (1983), 269-316.

[D2] , Connection, cohomology, and intersection forms of -manifolds,
$4$

J. Dierential Geom., 24 (1986), 275-341.

[D3] , The orientation of Yang-Mills moduli spaces and -manifold
$4$

topology, J. Dierential Geom., 26 (1986), 397-428.

[D4] , Polynomial invariants for smooth -manifolds, Topology, 29
$4$

(1990), 257-315.
[DFK] S. Donaldson, M. Furuta, D. Kotshick, Floer homology groups in
Yang-Mills theory, in preparation.
[DK] S. Donaldson, P. Kronheimer, The geometry of four manifolds, Oxford
Science Publications, Oxford, (1990).
[F] A. Floer, An instanton invariant for three manifolds, Commun. Math.
Phys., 118 (1988), 215-240.
[FU] D. Freed, K. Uhlenbeck, Instantons and Four manifolds, Springer,
New York, 1984.
[GK] K. Grove, H. Karcher, How to conjugate close actions, Math. Z.,
$C^{1}-$

132 (1973), 11-20.

[GKR] K. Grove, H. Karcher, E. Ruh, Group actions and curvature, Invent.
Math., 23 (1974), 31-48.
[LM] R. Lockhart, R. McOwen, Elliptic operators on noncompact manifolds,
Ann. Sci. Ec. Norm. Sup. Pisa, IV-12 (1985), 409-446.
[T1] C. Taubes, Self-dual connections on non-selfdual manifolds, J. Dier-
ential Geom., 17 (1982), 139-170.
[T2] , Self dual connections on manifolds with indefinite intersection

matrix, J. Dierential Geom., 19 (1984), 517-560.

[T3] , Gauge theory on asymptotically periodic -manifolds, J. Dif-
$4$

ferential Geom., 25 (1987), 364-430.

[T4] , Cassons invariant and gauge theory, J. Dierential Geom., 31

(1989), 547-599.
[W1] E. Witten, Supersymmetry and Morse theory, J. Dierential Geom.,
17 (1982), 661-692.
[W2] , Topological field theory, Comm. Math. Phys., 117 (1988),
353-386.
92 K. Fukaya

Department of Mathematics
Faculty of Science
University of Tokyo
Hongo, Bunkyo-ku, Tokyo 113
Japan
Advanced Studies in Pure Mathematics 20, 1992
Aspects of Low Dimensional Manifolds
pp. 93-112

Polyhedral Decomposition of
Hyperbolic -Manifolds $3$

with Totally Geodesic Boundary

Sadayoshi Kojima

Dedicated to Professor Kunio Murasugi

on his sixtieth birthday

\S 1. Introduction

A hyperbolic manifold will be a riemannian manifold with constant

sectional curvature-l. It is shown by Epstein and Penner [1] that every
noncompact complete hyperbolic manifold of finite volume, hence having
cusps, is decomposed by ideal polyhedra. The decomposition supplies a
quite convenient block to study several geometries of the cusped manifold
especially in dimension three. See [4] for instance.
A variant of the construction by Epstein and Penner would estab-
lish a decomposition of a compact hyperbolic manifold with nonempty
geodesic boundary by truncated polyhedra as well, which we plan to
discuss in a forthcoming paper [3]. However the process will be rather
unseen in the manifold.
In this paper, taking advantage of working only in dimension three,
we give a more visible construction of this decomposition. In fact we
directly show

Theorem. Let $N$ be a compact hyperbolic 3-manifold with non-

empty totally geodesic boundary. Then the topological decomposition of
$N$ dual to the cut locus
of modulo boundary is homotopic by straight-
$\partial N$

ening to a polyhedral decomposition.

The visible process is expected to lead us to the deep understanding

of geometry of those manifolds. We apply it for example to find the
minimum of their volumes in [2].

Received February 23, 1990.

94 S. Kojima

We describe the rule of the decompostion in the next section with

some detailed accounts of truncated polyhedra. We study the cut locus
of the boundary and its topological dual decomposition in \S 3. Then we
show in the subsequent sections that the straightening of the dual along
its internal edges yields the final polyhedral decomposition. The proof
of Proposition 5.2 thus finishes the proof of the theorem.
I am grateful to Tomoyoshi Yoshida for showing his idea to decom-
pose cusped manifolds by ideal polyhedra.

\S 2. Ikuncated polyhedra
We start with describing a basic piece of truncated polyhedra, called
truncated tetrahedra. An ideal tetrahedron is a hyperbolic polyhe-
dron identified with a finite volume region in the hyperbolic 3-space
bounded by four geodesic planes, every two of which intersect each
$H^{3}$

other, and every three of which intersect at infinity. An ultra ideal tetra-
hedron is one identified with a similar region bounded by four planes,
every two of which intersect each other again but no three of which in-
tersect even at infinity. If we are in the projective model, an ultra ideal
tetrahedron is one whose vertices are located outside of the model disk.
An ultra ideal tetrahedron is of infinite volume. The truncation is
the device to cut o its thick end by a geodesic plane which intersects
three planes towards the end perpendicularly. Such truncation is always
uniquely possible since

Lemma 2.1. For any three metric disks on the euclidean plane
which have no points in common but each two of which have a common
region, there is a unique circle intersecting their boundaries perpendicu-
larly.

Let us name three disks by $A$ , $B$ and $C$ . By conformal

Proof
change, we may assume that one of the intersection points of and $\partial A$

is located at infinity. Then

$\partial B$
and are the lines intersecting say
$\partial A$ $\partial B$

at the origin. By the assumption on the position of disks, $C$ does not
contain the origin. Hence we have a circle centered at the ori-
$ur_{\perp}ique$

gin intersecting perpendicularly. This circle automatically intersects

$\partial C$

both and perpendicularly.

$\partial A$ $\partial B$
Q.E.D.

Regard the boundaries of these disks as the ends of the geodesic

planes which make up a thick end of an ultra ideal tetrahedron. The
circle obtained in Lemma 2.1 will be the boundary of the plane for
truncation. This plane intersects three planes perpendicularly. Cutting
o each thick end by truncation, we get a compact polyhedron. This is a
 Polyhedral Decomposition of Hyperbolic 3-Manifolds 95

truncated tetrahedron. The surface of a truncated tetrahedron consists

of four right angle hexagons on the planes to bound the region, and four
triangles produced by the truncation.

Fig. 1.

A convex truncated polyhedron can be described in a similar man-

ner. Start with a finite set of geodesic planes in , no three of which
$H^{3}$

intersect even at infinity. Assume that it bounds a noncompact convex

region each thick end of which admits truncation. Then cutting o each
end of the region by truncation, we get a compact polyhedron. This
is a convex truncated polyhedron. The surface of a convex truncated
polyhedron consists of right angle polygons on the planes, which we call
internal faces, and the other polygons produced by the truncation, which
we call external faces. The union of internal faces is connected, while
external faces are mutually disjoint.
A tetrahedron is a basic piece of a polyhedron even in this situation.

Lemma 2.2. A convex truncated polyhedron is decomposed by

truncated tetrahedra without producing vertices in the interior.

Proof. Choose an external face and introduce the shortest geo-

$\tau$

desic paths from the face to the other external faces. Such a path
uniquely exists for each face. It lies on the boundary if and the ter-
$\tau$

minal face are joined by just one face. Obviously it lies on this joining
face then. Otherwise the paths go through interior of the polyhedron.
The internal faces touching are now subdivided into right angle
$\tau$

hexagons. Subdivide then the other internal faces by a geodesic path

into right angle hexagons arbitrarily. Each geodesic path introduced
96 S. Kojima

here joins two external faces. It together with the shortest paths as-
signed to the terminal faces span a right angle hexagon in the interior
of the polyhedron. Because two paths determine a geodesic plane in-
tersecting three external faces involved perpendicularly and hence this
plane must contain the last path. The collection of these hexagons di-
vides the original polyhedron into truncated tetrahedra. Q.E.D.

A polyhedral decomposition of a hyperbolic 3-manifold with totally

geodesic boundary is a geometric cellular decomposition by (convex)
truncated polyhedra so that their external faces form the boundary.
This justifies our naming for faces. We also call an edge internal if it is
an intersection of two internal faces, and external otherwise. Notice in
this decomposition that every internal edge is a geodesic path from the
boundary to the boundary.
Let us describe a parametrization of isometry classes of labelled
truncated tetrahedra to show its variation, though the result is not
needed for the proof of the theorem. The isometry class of a trun-
cated tetrahedron is determined by the mutual position of the internal
faces, since the truncation is unique. Label the internal edges as in Fig-
ure 1.1 and denote the dihedral angle along the edge by . s are $j$ $\theta_{j}$ $\theta_{j}$

quantities to describe mutual position. The sum of three dihedral angles

having a common external face must be less than because otherwise $\pi$

three planes towards the end meet in the real world. Thus we have a
necessary condition,
$\theta_{1}+\theta_{2}+\theta_{3}<\pi$

$\theta_{1}+\theta_{5}+\theta_{6}<\pi$

$\{$

$\theta_{2}+\theta_{4}+\theta_{6}<\pi$

$\theta_{3}+\theta_{4}+\theta_{5}<\pi$
.

Conversely,

Lemma 2.3. For , , satisfying the above inequalitites, there

$\theta_{1}$
$\ldots$
$\theta_{6}$

is a unique labelled truncated tetrahedron with these dihedral angles.

Proof. Make four geodesic triangles using , , which would $\theta_{1}$

$\ldots$
$\theta_{6}$

form external faces. They have twelve edge lengths as data we can use.
Choose a triple from these twelve lengths that would be assigned to the
external edges of an internal face we expect to make. Then there is a
unique right angle hexagon having these as non adjacent edge lengths,
which is a candidate of the internal face. Applying the same for the
other triples, we get four right angle hexagons.
 Polyhedral Decomposition of Hyperbolic 3-Manifolds 97

The expected truncated tetrahedra should be obtained by gluing

these faces in , and what we need to show now is that the length of a
$H^{3}$

common internal edge for each pair of hexagons made are the same. We
do this for the internal edge 1. By the hyperbolic cosine rule, we have

$(^{*})$ $\cosh\ell_{ij}=\frac{\cos\theta_{i}\cos\theta_{j}+\cos\theta_{k}}{\sin\theta_{i}\sin\theta_{j}}$
,

where $\{i, j, k\}$ corresponds to 3 angles of a triangle and is the length $\ell_{ij}$

of the external edge connecting edges and . We made two hexagons $i$
$j$

having the edge 1. By the hexagon rule [4], the length of the edge 1 $\ell_{1}$

computed in the hexagon having the edges 2 and 6 is given by

$\cosh\ell_{1}=\frac{\cosh\ell_{12}\cosh\ell_{16}+\cosh\ell_{26}}{\sinh\ell_{12}\sinh\ell_{16}}$
,

and the same having the edges 3 and 5 is given by

$\cosh\ell_{1}=\frac{\cosh\ell_{13}\cosh\ell_{15}+\cosh\ell_{35}}{\sinh\ell_{13}\sinh\ell_{15}}$
.

It is then easy to check by substitution of $(^{*})$

that right hand sides of
both identities are the same. Q.E.D.

\S 3. Cut locus
Studying several properties of the cut locus of the boundary in this
section, we will find a topological cellular decomposition of a hyperbolic
manifold with totally geodesic boundary. It is dual to the cut locus
modulo boundary and turns out to be equivalent to the final one. The
decomposition will be denoted by $K$ .
Here we start with making a few conventions used throughout the
sequel. Let $N$ be a compact hyperbolic 3-manifold with totally geodesic
boundary . Let
$\partial N$
: be the universal covering of $N$ . We
$\pi$
$\overline{N}\rightarrow N$

use the symbol to denote the preimage of a subspace $X$ of $N$ in .

$\overline{X}$ $\overline{N}$

We always identify the universal cover with a subspace in . Then $\overline{N}$

$H^{3}$

the boundary of the universal cover or the preimage

$\partial\overline{N}$

of the $\overline{N}$ $\overline{\partial N}$

boundary is formed by geodesic planes in

$\partial N$
. We often identify a $H^{3}$

cell complex with its underlying polyhedron. The symbol will be $Y^{(k)}$

used to denote the -skeleton of a cell complex $Y$ as usual.

$k$

We define three terminologies for our convenience. To each pair of

components of , associated is a unique shortest path connecting them.
$\partial\overline{N}$

We call this path a short cut. Also there is an associated bisectorial

98 S. Kojima

geodesic plane to the short cut in . We call this plane a middle fence.
$H^{3}$

A short cut descends to the geodesic path in $N$ from the boundary to
the boundary. We call such a path a return path. Though it may come
back to a dierent component, we wish to emphasize by this name that
it comes back to the boundary anyway. These are the terminologies we
shall use frequently.
The cut locus $C$ of in $N$ is a subset in int $N$ which consists of the
$\partial N$

points that admit at least two distinct shortest paths to . Obviously

$\partial N$

a point on $C$ lifts to a point on the middle fence of some short cut.
$C$ is canonically stratified by grouping
the points which have the same
number of shortest paths to the boundary. This stratification is quite
nice in our case since

Proposition 3.1. The stratification defines a convex cellular de-

composition of the cut locus C.

A point on $C$ is in a 2-cell if it admits precisely two shortest paths

to the boundary, however the number of shortest paths the point admits
is rather unrelated with the dimension of the cell in the other case. To
see this proposition, we need a few preliminaries.

Lemma 3.2. Suppose that $A$ and $B$ are ultra parallel planes of
distance in
$d$
. Then the orthogonally projected image of $A$ to $B$ is
$H^{3}$

an open metric disk of radius arccosh(coth ). $d$

Proof. This is an easy consequence of length calculus for a hyper-

bolic rectangle with one ideal vertex and three vertices of right angle as
in Figure 2.

$d$
$\{$

$F$

Fig. 2.
 Polyhedral Decomposition of Hyperbolic 3-Manifolds 99

The hyperbolic cosine rule shows the identities;

$\cosh d=\frac{1}{\sin\alpha}$
,

$\cosh r=\frac{1}{\sin(\pi/2-\alpha)}=\frac{1}{\cos\alpha}$ .

Then we are done by solving the relation between $d$

and $r$
in terms of
$r$
. Q.E.D.

Corollary 3.3. There exist only finitely many return paths with
bounded length.

Choose a component of
Proof. , and project the other bound- $S$
$\partial\overline{N}$

ary components orthogonally to . Then we get an open disk packing on $S$

$S$
invariant under the action of the covering transformations preserving
. Hence $\pi(S)\subset N$ is packed by open balls. It is obvious by definition
$S$

that the packing on $\pi(S)$ does not depend on the choice of a component
$S$
of $\pi^{-1}(\pi(S))$ . Applying the same process to all the other components,
we get a ball packing on . The radius of each ball is related to the $\partial N$

length of the associated return path by Lemma 3.2. Since is com- $\partial N$

pact, the number of balls packing with bounded radius away from $\partial N$

zero is obviously finite. Hence there are only finitely many return paths
of bounded length. Q.E.D.

Choose a component $U$ of the comple-

Proof of Proposition 3.1.
ment of in and let be its boundary in . $U$ is invariant under
$\overline{C}$ $\overline{N}$
$S$
$\partial\overline{N}$

the action of covering transformations preserving . We are interested $S$

in the internal boundary of the closure of $U$ not meeting . It is a $\overline{U}$

$\partial\overline{N}$

part of and formed by a part of middle fences. Since $N$ is compact,

$\overline{C}$

its diameter is bounded, and the points on $C$ have bounded distance to
. The shortest arc from a point on $C$ to
$\partial N$
is lifted to an arc in $\partial N$

. In particular, the distance between

$\overline{Y}$

and any point on the internal $S$

boundary of is bounded. Hence the middle fences involved in this

$\overline{U}$

boundary are associated with the short cuts of bounded lengh.

By Corollary 2.3, there are only finitely many return paths with
bounded length. Hence the middle fences involved in the internal bound-
ary of belong to only finitely many orbits of middle fences by the
$\overline{U}$

action of covering transformations preserving . The internal boundary $S$

of thus gets a locally finite invariant cellular decomposition induced

$\overline{U}$

by the intersection of middle fences involved. It descends to a cellular

decomposition of the internal boundary of . $\pi(\overline{U})$
100 S. Kojima

We can apply the same argument to the other component. It is

an exercise to check that the cell structures for the common part of
dierent internal boundaries are identical. Hence we get an invariant
cellular decomposition of and hence a cell complex structure of C. $\overline{C}$

Each 2-cell of $C$

is convex since it lifts to a convex polygon on some
middle fence bounded by the intersections with a finite number of the
other middle fences. Q.E.D.

From now on, let us mean by $C$ not only the cut locus itself but
endowed with this cellular decomposition by virtue of the proposition.
In the universal cover, we say a 2-cell of faces a component of if $\overline{C}$ $\partial\overline{N}$

the cell can be projected orthogonally to the component by the shortest

paths to . Each 2-cell faces two boundary components associated
$\partial\overline{N}$

to the middle fence containing it. The set of orthogonal projections

for each 2-cell to these components gives rise to an equivariant one-to-
finite orthogonal projection : . The number of the image of $\overline{C}\rightarrow\partial\overline{N}$

is equal to the number of the shortest paths from to

$p\in\overline{C}$
. The $p$
$\partial\overline{N}$

cellular decomposition of is conveyed to an invariant convex polygonal $\overline{C}$

decomposition of . In particular, the cellular decomposition of $C$

$\partial\overline{N}$

induces a convex polygonal decomposition of . $\partial N$

Now, we would like to build up a topological cellular decomposition

$K$ of $N$ dual to $C$ modulo boundary. Start with defining a compact
3-cell, which we call a block, in the universal cover. Its interior will
be a 3-cell in the precise definition of the cell complex $K$ . Take an in-
variant graph $G$ on under the action of $\pi_{1}(N)$ which is dual to the
$\overline{C}$

1-skeleton . Here we mean by dual, the 1-dimensional subcomplex

$\overline{C}^{(1)}$

of the barycentric like subdivision of spanned by vertices not in . $\overline{C}$ $\overline{C}^{(0)}$

Then project it by the one-to-finite orthogonal projection to . The $\partial\overline{N}$

$trace$ of the projection determines a fence which divides into equiv- $\overline{N}$

ariant pieces homeomorphic to a ball. This is a block to built up $K$ .

Let us next define a compact cell which we call a face, an edge or
a vertex according to its dimension. The intersection of two blocks is
the $trace$ of the star subgraph of a vertex of $G$ on a 1-cell of by the $\overline{C}$

orthogonal projection. Hence take it as a dual face to the 1-cell of $C$ on

which the center is located, and call it an internal face. We also take
a component of the intersection of a block and as an external face. $\partial\overline{N}$

A face will be either an internal or external face. The inters$-ection$ of

two internal faces is the $trace$ of a vertex of $G$ on a 2-cell of C. Hence
take it as a dual edge to the 2-cell containing the vertex, and call it
an internal edge. We also take a component of the intersection of an
 Polyhedral Decomposition of Hyperbolic -Manifolds $S$
101

internal face and as an external edge. An edge will be either an $\partial\overline{N}$

internal or external edge. Finally a vertex will be a terminal point of an

edge.
Then let be a cellular decomposition of by the interior of blocks,
$\overline{K}$ $\overline{N}$

faces, edges and vertices. Since it is invariant under the action of $\pi_{1}(N)$ ,
it determines a cellular decomposition $K=\pi(\overline{K})$ of $N$ . This is what
we call a dual to $C$ modulo boundary. Notice that is dual to the $\partial K$

convex polygonal decomposition of induced by the cut locus. $\partial N$

We describe the compact cells of more locally to visualize the $\overline{K}$

situation. Each block contains a unique 0-cell of . We call this a center. $\overline{C}$

Choose a block with the center and let us describe its combinatorial $\sigma$
$p$

structure of the boundary by identifying with the origin of the 3- $p$

dimensional Poincar\e disk, has the shortest rays to finitely many $p$

components of , say , , , . can be identified with a regular

$\partial\overline{N}$
$S_{1}$ $S_{2}$
$\ldots$
$S_{m}$ $\sigma$

neighborhood of the union of these rays. The ray extends and terminates
in the sphere at infinity . The terminal point is the center of the $S_{\infty}^{2}$
$q_{j}$

metric circle on with respect to the canonical spherical metric,

$\partial S_{j}$ $S_{\infty}^{2}$

where $j=1,2$ , , $m$ . Notice that the radii of circles are the same because
$\ldots$

the distances from the origin are the same.

Take the cut locus $D$ of the point set on . $D$ consists $\{q_{1}, \ldots, q_{m}\}$ $S_{\infty}^{2}$

of the points on which admit at least two shortest paths to the $S_{\infty}^{2}$

set . $D$ is unit tangentially equivalent to $C$ at and hence

$\{q_{1}, \ldots, q_{m}\}$ $p$

determines a convex polygonal decomposition on . $S_{\infty}^{2}$

A topological dual decomposition of $D$ on with vertices $D^{*}$ $S_{\infty}^{2}$

, ,
$q_{1}$ is identified with one obtained from the cellular decomposition
$\ldots$ $q_{m}$

of by collapsing each external face to . Notice by the definition of

$\partial\sigma$

$q_{j}$

the cut locus that the vertices of a face of have the same distance to $D^{*}$

the vertex of $D$ in this face. This fact will be used later.
We may assume that each edge of is straight at least in the disks $D^{*}$

bounded by s. Replacing the part of in each disk by

$\partial S_{j}$
, we get $D^{*}$ $\partial S_{j}$

a cellular decomposition on . is equivalent to . $D^{**}$ $S_{\infty}^{2}$ $D^{**}$ $\partial\sigma$

There are several immediate correspondence by the identification

of and . The external faces correspond to the faces bounded
$\partial\sigma$ $D^{**}$

by s, and the internal faces do to the others. The external edges

$\partial S_{j}$

correspond to the edges on

, while the internal edges do to the $\partial S_{j}$ $s$

others. The vertices on the circle correspond to 2-cells of $C$ which $\partial S_{j}$

touches and faces . Both are arranged in the same order.

$p$ $S_{j}$

The final decomposition is obtained by straightening each edges of

. The straightening here is the device first to replace each internal
$\overline{K}$

edges by homotopic short cuts, and then to replace external edges by

geodesic paths using their end points. The straight map we get is sup-
102 S. Kojima

$D$ $D^{*}$ $D^{**}$

Fig. 3.

ported on the 1-skeleton at the beginning and there is no obvious

$K^{(1)}$

reasons why it creates something good. The rest of this paper is to check
the reason why it does.
The first step is to observe that the image of by the straight $\partial K^{(1)}$

map, which will be denoted by , turns out to be a 1-skeleton of

$\partial\Delta^{(1)}$

a convex polygonal decomposition of , denoted by . This will be $\partial N$ $\partial\Delta$

done in the next section. The second step starts by showing that the
map can be straightened over the 2-skeleton . The main step is $K^{(2)}$

then to observe that the straightened image of , denoted by , $K^{(2)}$ $\Delta^{(2)}$

turns out to be a 2-skeleton of a convex polyhedral decomosition of , $N$

denoted by . Since we define the final decomposition from the lower

$\Delta$ $\Delta$

dimensional skeletons, the accessories for we use is not appropriate in $\Delta$

fact, but will be justified by the end of the paper.

\S 4. Polygonal decomposition
We study the eect of straightening on the boundary in this section,
and prove that the straightening defines a convex polygonal decomposi-
tion of equivalent to
$\partial N$
. The argument will be given mainly in the
$\partial K$

universal cover.
An internal edge of bridges two components of
$\overline{K}$

. Hence to $\partial\overline{N}$

each internal edge, assigned is a unique middle fence and a unique short
cut. Recall that an internal edge is a dual to a 2-cell of which lies $\overline{C}$

on this middle fence. The number of orbits of short cuts associated to -

2-cells of $C$ by the action of $\pi_{1}(N)$ was finite. Let $R$ be the set of these
short cuts, and $R=\pi(\overline{R})$ be the set of descending return paths in $N$ .
$R$ is a finite set.
 Polyhedral Decomposition of Hyperbolic -Manifolds $S$
103

The straightening device at this stage is precisely to connect vertices

of by geodesic paths if the corresponding two vertices in
$\overline{R}$

are $\partial\overline{K}^{(0)}$

joined by an external edge. We denote the resultant geodesic 1-complex

by in the abstract sense. The accessories in the notation should
$\partial\overline{\Delta}^{(1)}$

be ignored for the moment. The definition does not immediately tell
us that is an embedded 1-complex. What we obviously know by
$\partial\overline{\Delta}^{(1)}$

definition is that is invariant under the action of $\pi_{1}(N)$ , and that $\partial\overline{\Delta}^{(1)}$

there is an equivariant graph isomorphism : . $h$ $\partial\overline{K}^{(1)}\rightarrow\partial\overline{\Delta^{(1)}.}$

Since the connection rule to build up was followed by the $\partial\overline{\Delta}^{(1)}$

rule for , should be very similar to . The claim to

$\partial\overline{K}^{(1)}$ $\partial\overline{\Delta}^{(1)}$ $\partial\overline{K}^{(1)}$

be proved is that is in fact a 1-skeleton of an invariant convex $\partial\overline{\Delta}^{(1)}$

polygonal decomposition of , and extends to an equivariant $\partial\overline{\Delta}$ $\partial\overline{N}$

$h$

cellular isomorphism of . The statement in is hence $\partial\overline{K}$

$\partial N$

Proposition 4.1. $\partial\Delta^{(1)}=\pi(\partial\overline{\Delta}^{(1)})$

turns out to be 1-skeleton
$a$

of a convex polygonal decomposition $\partial\Delta=\pi(\partial\overline{\Delta})$

of $\partial N$
equivalent to
$\partial K$
.

To see this, we need a few observations about local structure of edges

in . The first one is about the image of the boundary of a face of
$\partial\overline{\Delta}^{(1)}$

$\partial\overline{K}$

Lemma 4.2. The image of the boundary of a face of

$\partial\overline{K}$

by $h$

bounds a convex polygon on S. The canonical extension of $h$

to the face
preserves the orientation.

Choose a face of
Proof. and assume that it lies on a block $\tau$
$\partial\overline{K}$

with the center . The cellular decomposition of

$\sigma$
was described by $p$
$\partial\sigma$

. The external face is identified with a face bounded by a metric

$D^{**}$ $\tau$

circle on . The center of is the terminal point of an extension

$\partial S$ $S_{\infty}^{2}$
$q$
$\partial S$

of the shortest path from to . $p$

$S$

Label the vertices of by with $j=0,1$ , , $-1$ in counter- $\tau$

$v_{j}$ $\ldots$
$n$

clockwise order. Each vertex is a projected image of a dual vertex to a

2-cell in touching and facing . Hence we also label the 2-cell of
$\overline{C}$

$p$
$S$
$\overline{C}$

corresponding to by . $v_{j}$
$F_{j}$

Each is on the middle fence of a short cut from a point on$F_{j}$ $S$

since faces . Hence we let its starting point on by

$F_{j}$
. Because $S$ $S$
$w_{j}$

of the definition of labeling, any adjacent s are joined by an edge in $w_{j}$

. is then a 1-complex formed by geodesic paths $w_{j}w_{j+1}$ with

$\partial\overline{\Delta}^{(1)}$
$h(\partial\tau)$

$j=0,1$ , , $n-1$ , where counts modulo as usual. $\ldots$

$j$ $n$
104 S. Kojima

We show that the vertices , , span a triangle , $w_{j_{0}}$ $w_{j_{1}}$ $w_{j_{2}}$

$\triangle w_{j_{o}}w_{j_{l}}$
$w_{j_{2}}$

and its orientation assigned by how the vertices round induces the coun-
terclockwise orientation on as long as $j_{0}<j_{1}<j_{2}$ up to cyclic per- $S$

mutation. Then using this property, we will get the conclusion by con-
tradiction.
Identify with the 2-dimensional Poincar\e disk and with the ori-
$S$ $q$

gin. The middle fences containing , , respectively are orthog- $F_{jo}$ $F_{j_{1}}$ $F_{j_{2}}$

onally projected to three open metric disks , , on including $B_{j_{0}}$ $B_{j_{1}}$ $B_{j_{2}}$ $S$

the origin. The vertices , , are the centers of these disks. The $w_{j_{0}}$ $w_{j_{1}}$ $w_{j_{2}}$

outside of is reflected into the inside by the orthgonal projection to

$\partial S$

. The picture of the projection is shown in Figure 4.

$S$

$S$

Fig. 4.

By the convexity of $D$ , , and are arranged in counter- $B_{j_{0}}$ $B_{j_{1}}$ $B_{j_{2}}$

clockwise order as in the second picture in Figure 4. We named the

intersections of the boundary of balls as in the figure. Then , , $\alpha_{j_{0}}$ $\alpha_{j_{1}}$

and determine the oriented triangle

$\alpha_{j_{2}}$ inducing the coun- $\triangle\alpha_{j_{0}}\alpha_{j_{1}}\alpha_{j_{2}}$

terclockwise orientation on . $S$

Here is an elementary geometry. Let , , be the bisectors to $\gamma_{j_{o}}$ $\gamma_{j_{1}}$ $\gamma_{j_{2}}$

the segments , and on respectively. These three

$\alpha_{j_{2}}\alpha_{j_{0}}$ $\alpha_{j_{o}}\alpha_{j_{1}}$ $\alpha_{j_{1}}\alpha_{j_{2}}$
$S$

lines meet at the center of the circumscribed circle of the triangle $\beta$

. is on , where $j=j_{0}$ , , . Since

$\triangle\alpha_{jo}\alpha_{j_{1}}\alpha_{j_{2}}$
, ,
$w_{j}$ do $\gamma_{j}$
$j_{1}$ $j_{2}$ $B_{jo}$ $B_{j_{1}}$ $B_{j_{2}}$

not contain , , respectively, the direction of the vector

$\alpha_{j_{1}}$ $\alpha_{j_{2}}$ $\alpha_{j_{o}}$
$\beta w_{j}$

is the same as that of the outward vector from the triangle $\triangle\alpha_{j_{0}}\alpha_{j_{1}}\alpha_{j_{2}}$

along . Hence the centers ,

$\gamma_{j}$, are arranged in counterclock- $w_{j_{o}}$ $w_{j_{1}}$ $w_{j_{2}}$

wise order from the viewpoint , and determines an oriented triangle $\beta$

inducing the counterclockwise orientation on .

$\triangle w_{j_{0}}w_{j_{l}}w_{j_{2}}$ $S$

Suppose now that the union of geodesic paths $w_{j}w_{j+1}$ with $j=$
 Polyhedral Decomposition of Hyperbolic 3-Manifolds 105

0, , $-1$ does not bound a convex polygon on . Then since any

$\ldots$
$n$ $S$

three vertices determines a nondegenerate triangle, there are suces

$0\leq j_{3}$ , $j_{4}\leq n-1$ so that meets the biinfinite extension of $w_{j_{3}}w_{j_{3}+1}$

at an interior point. Then the induced orientations on by the

$w_{j_{4}}w_{j_{4}+1}$
$S$

triangles and are dierent each other.

$\triangle w_{j_{3}}w_{j_{4}}w_{j_{4}+1}$ $\triangle w_{j_{3}+1}w_{j_{4}}w_{j_{4}+1}$

This contradicts what we have proved.

Since the vertices s of and the corresponding vertices s of $v_{j}$
$\tau$
$w_{j}$

the convex polygon mapped by are both arranged in counterclockwise $h$

order, a canonical extension of preserves the orientation. Q.E.D. $h$

We need one more observation about the structure around the vertex
of . Label all the edges coming to the vertex of
$\partial\overline{K}$

. Then by the $\partial\overline{K}$

connection rule of , this labeling is canonically conveyed to the $\partial\overline{\Delta}^{(1)}$

labeling of the edges of which terminate at the corresponding $\partial\overline{\Delta}^{(1)}$

vertex.

Lemma 4.3. The counterclockwise orders of the labeling at a ver-

tex of and the corresponding vertex of
$\partial\overline{K}^{(1)}$ $\partial\overline{\Delta}^{(1)}$

are the same up to

cyclic permutation.

Choose a vertex of
Proof. and assume that it lies on a block $v$
$\partial\overline{K}$

. The cellular decomposition of

$\sigma$
is described by . is identified $\partial\sigma$ $D^{**}$ $v$

with a vertex on the metric circle . $\partial S$

Choose an adjacent vertex to on the same circle . and $v^{J}$

$v$
$\partial S$
$v$
$v$

correspond to 2-cells $F$ and in facing and touching the center of $F^{J}$
$\overline{C}$

$S$

. Recall that the adjacency is reflected by the property that these 2-

$\sigma$

cells $F$ and $F$ have a common 1-cell. Denote by $w$ and $w$ the vertices of
corresponding to and
$\partial\overline{R}$

respectively. Here is a geometric relation $v$

$v$

between the adjacency of , and $w$ , $w$ . The middle fence containing $v$ $v$ $L$

$F$ has an intersection line with the middle fence containing . $l$ $L^{l}$ $F^{/}$

$F$ is orthogonally projected to a convex polygon $P$ on and is to a $S$ $l$

geodesic which is an biinfinite extension of an edge of . The plane

$l_{S}$ $P$

determined by short cuts from $w$ and $w$ is orthogonal to both and $L$

, and in particular to . Hence the geodesic path connecting $w$ and

$L^{J}$ $l$ $w^{J}$

extends to a biinfinite path orthogonal to . $\omega$ $l_{S}$

What we have seen is that to each pair of and , and hence to $v$
$v^{/}$

each edge coming to , associated is an biinfinite extension of an edge $v$ $l_{S}$

of $P$ , and that lies on the geodesic through $w$ and orthogonal to $w^{/}$ $\omega$

. Furthermore, though the vertices $w$ and $w$ may not be separated by
$l_{S}$

, the vector from $w$ to

$l_{S}$
is directed towards the component of $S-l_{S}$ $w^{J}$

not containing , as the vector from to obviously is.

$P$ $v$
$v^{/}$

Now identify with the 2-dimensional Poincar\e disk. The biinfinite $S$
106 S. Kojima

extensions of the edges of $P$ determine a line configulation on . Each $S$

line inherits a label from the associated edge coming to . To each $v$

labelled line, we assign the orthogonal ray from endowed with the $v$

same label. The counterclockwise order of the labeling for orthogonal

rays is the same as that for edges of coming to . $\partial\overline{K}$

$v$

Then for each point on , draw orthogonal rays to the geodesic $S$

lines again keeping the outward direction from $P$ . Then the assignment
of the counterclockwise order of the labeling of rays is a continuous
function on to the set of cyclic orders possibly with singularities. The
$S$

singularity occurs only if two rays coincide. This may happen when two
geodesic lines are ultra parallel. However in this case, the direction of
associated two rays must be opposite since the region bounded by such
lines contains a convex polygon $P$ . Hence this continuous function has
no singularities with discrete image. In particular, the order at $w$ is the
same as one at . Q.E.D. $v$

Proposition 4.1. By Lemma 4.2, extending a graph iso-

Proof of
morphism : , we get a map (still using the same
$h$
$\partial\overline{K}^{(1)}\rightarrow\partial\overline{\Delta}^{(1)}$
$h$

notation) of by assigning to each face of a polygon bounded

$\partial\overline{K}$
$\partial K$

by corresponding edges of . Here is a local homeomorphism on $\partial\overline{\Delta}^{(1)}$

$h$

the interior of faces. Since preserves the orientation for each face, it $h$

must be a homeomorphism also around edges. Lemma 4.3 shows that

the corners of convex polygons fill up a neighborhood of the vertices.
Hence is a local homeomorphism also around the vertices. It is easy
$h$

to see that is surjective. Since the image is simply connected, is a

$h$ $h$

global homeomorphism.
now determines a convex polygonal decomposition
$\partial\overline{\Delta}^{(1)}$

of .
$\partial\overline{\Delta}$ $\partial\overline{N}$

The decomposition is invariant under the action of $\pi_{1}(N)$ , and the map $h$

can be chosen to be equivariant. Hence it determines a convex polygonal

decomponsition of with a descending equivalence from $\partial\Delta=\pi(\partial\overline{\Delta})$ $\partial N$

to .
$\partial K$ $\partial\Delta$
Q.E.D.

\S 5. Polyhedral decomposition

In this section, we study the eect of straightening in the interior

and finish to prove that the straightening determines a convex polyhedral
decomposition of $N$ , which we promised to denote by . The argument $\Delta$

will be given again mainly in the universal cover.

The map : we had at the beginning was
$h$
$\partial\overline{K}^{(1)}\rightarrow\partial\overline{\Delta}^{(1)}\subset\partial\overline{N}$

a graph isomorphism. The main claim in \S 4 was that extends to a $h$

cellular map on to . It obviously further extends as a cellular $\partial\overline{K}$ $\partial\overline{\Delta}$

$h$
 Polyhedral Decomposition of Hyperbolic 3-Manifolds 107

isomorphism to : . We then will see first $h$

$\partial\overline{K}\cup\overline{K}^{(1)}\rightarrow\partial\overline{\Delta}\cup\overline{R}\subset\overline{N}$

that the map extends as a straight map over the 2-skeleton , $h$
$\overline{K}^{(2)}$

showing that the image of the boundary of each internal face of spans $\overline{K}$

a geodesic polygon. Namely, extends to a geodesic 2-complex $\partial\overline{\Delta}\cup\overline{R}$

in in the abstract sense.

$\overline{\Delta}^{(2)}$ $\overline{N}$

Lemma 5.1. The image of the boundary of an internal face of

$\overline{K}$

by bounds a right angle polygon on a geodesic plane in .

$h$
$\overline{N}$

Choose an internal face and assume that it lies on a block

Proof. $\tau$

. The cell decomposition of

$\sigma$
was described in by identifying the $\partial\sigma$ $D^{**}$

center of with the origin of the 3-dimensional

$p$ Poincar\e
$\sigma$
disk. There
are metric circles , , on which are the boundaries of the $\partial S_{1}$
$\ldots$
$\partial S_{m}$ $S_{\infty}^{2}$

nearest components of from . The centers , , of these metric $\partial N$

$p$ $q_{1}$ $\ldots$ $q_{m}$

circles are also the endpoints of the rays extending the shortest path
from the origin to the component . The circles , , , having $S_{j}$ $\partial S_{1}$
$\ldots$
$\partial S_{m}$

the same radius, lie in the complement of the cut locus $D$ of $\{q_{1}, \ldots, q_{m}\}$

on . $S_{\infty}^{2}$

The face is identified with a face not bounded by

. We
$\tau$ $\partial S_{j}$ $s$

rearrange s so that passes through , , ,

$\partial S_{j}$
in counter- $\partial\tau$
$\partial S_{1}$ $\partial S_{2}$
$\ldots$
$\partial S_{k}$

clockwise order, contains a vertex of a cut locus D. Recall as we $\tau$ $u$

noted in the description of $D$ and that every has the same $D^{*}$ $\partial S_{j}$

distance from . In particular, there is a circle on , bounding

$u$
$\partial H$ $S_{\infty}^{2}$

a geodesic plane $H$ in the 3-dimensional Poincar\e disk, that intersects

orthogonally to each , , simultaneously. Moreover, passes $\partial S_{1}$
$\ldots$
$\partial S_{k}$ $\partial H$

through , , in counterclockwise order also.

$\partial S_{1}$
$\ldots$
$\partial S_{k}$

is a piecewise geodesic whose bent occurs only at the end of

$h(\partial\tau)$

external and hence internal edges. Each internal edge is mapped to the
short cut between and . It must lie on the plane $H$ since it $S_{j}$ $S_{j+1}$

intersects both and orthogonally. In particular, the image of $S_{j}$ $S_{j+1}$

internal edges is on a geodesic plane $H$ . The image of external edges

is on s and on $H$ since the intersection of
$S_{j}$
and $H$ is a geodesic $S_{j}$

passing two end points of the short cuts. It is then obvious by the order
of intersections to s that bounds a convex polygon on $H$ . $\partial S_{j}$ $h(\partial\tau)$

Q.E.D.

Denote by the collection of the straight image of each internal $\overline{\Delta}^{(2)}$

faces by Lemma 5.1 and . The accessories in this notation should $\partial\overline{\Delta}\cup\overline{R}$

be ignored for the moment. The definition does not immediately tell
us that is an embedded 2-complex. What we obviously know by
$\overline{\Delta}^{(2)}$

definition is that is invariant under the action of , and that $\overline{\Delta}^{(2)}$

$\pi_{1}(N)$
108 S. Kojima

there is an equivariant cellular isomorphism : which $h$

$\overline{K}^{(2)}\rightarrow\overline{\Delta}^{(2)}$

extends the original . The claim to be proved is that is in fact a $h$

$\overline{\Delta}^{(2)}$

2-skeleton of an invariant convex polyhedral decomosition of , and $\overline{\Delta}$ $\overline{N}$

extends to an equivariant cellular isomorphism of . The statement

$h$
$\overline{K}$

in $N$ is our final goal.

Proposition 5.2. $\Delta^{(2)}=\pi(\overline{\Delta}^{(2)})$

turns out to be a 2-skeleton of
a convex polyhedral decomposition of $N$ equivalent to $K$ .
$\Delta=\pi(\overline{\Delta})$

We have shown so far that if we restrict the map to the set of $h$

external faces or to each internal face, then is an embedding. What $h$

we still do not know is if the image of some internal faces intersect. To

see our final proposition, we proceed further to a local study.

Lemma 5.3. The image of the boundary of a block of

$\overline{K}$

by $h$

bounds a convex polyhedron in . $\overline{N}$

Choose a block and recall that the cell decomposition of

Proof. $\sigma$

is described by
$\partial\sigma$

on . Assigned to each external face was

$D^{**}$ $S_{\infty}^{2}$

a geodesic boundary , and assigned to each internal face now by$S_{j}$ $\tau_{i}$

Lemma 5.1 is a geodesic plane in . Using this description, we $H_{i}$

$H^{3}$

will define a continuous deformation of a restriction of to , $\{h_{t}\}$ $h$ $\partial\sigma$

, so that it eventually pushes the image of internal faces out

$h|_{\partial\sigma}=h_{0}$

to . Then by referring to the fact that

$S_{\infty}^{2}$
is a homeomorphism, we $h_{\pi/2}$

will establish the stable cellularity of to conclude the claim. $h_{t}$

For each internal face , a neighborhood of in is con- $\tau_{i}$

$h_{0}(\tau_{i})$ $h_{0}(\partial\sigma)$

tained in one side of separated by . We call the other side of $H^{3}$ $H_{i}$ $H_{i}$

outwards. The outside of s is similarly defined using the image of $S_{j}$

external faces. Let be the equidistant surface outside of with the

$H_{i}^{t}$ $H_{i}$

distance dO. This is not a geodesic plane but is a surface which

$\int_{0}^{t}\sec\theta$

intersects $H_{i}=H_{i}^{0}$ at with dihedral angle . It can be seen also $S_{\infty}^{2}$ $t$

as an intersection of an euclidean metric sphere with the Poincar\e disk

meeting the unit sphere with dihedral angle . The angle varies $S_{\infty}^{2}$ $t$ $t$

from 0 to $\pi/2$ . As increases, is gradually pushed out towards

$t$
. $H_{i}^{t}$ $S_{\infty}^{2}$

To define the image of an internal face , let us rearrange s $\tau_{0}=\tau$ $\tau_{i}$

in such a way that passes through , , , , ,

$\partial\tau$
$h_{0}(\tau_{1})$ $S_{1}$ $h_{0}(\tau_{2})$ $S_{2}$
$\ldots$
$h_{0}(\tau_{k})$

and in cyclic order.

$S_{k}$
and meet on the intersection of $h_{0}(\tau)$ $h_{0}(\tau_{i})$

$H_{0}^{0}=H^{0}$ and . Take two internal faces and

$H_{i}^{0}$
having a $h_{0}(\tau)$ $h_{0}(\tau_{i_{0}})$

common internal edge, and identify the edge with a segment on the z-
axis in the upper half space model so that it meets at the bottom end. $S_{i_{O}}$

See Figure 5 which shows the situation locally. is a geodesic plane $H_{i_{O}}^{0}$
 Polyhedral Decomposition of Hyperbolic 3-Manifolds 109

Fig. 5.

containing . and are described in the second picture. In

$h_{0}(\tau_{i_{0}})$ $H_{0}^{t}$ $H_{i_{0}}^{t}$

this coordinate, they are euclidean hyperplanes through the origin.

The original is a convex polygon on bounded by the in- $h_{0}(\tau)$
$H^{0}$

tersections with s and s where , $j=1$ , , . As increases, this$H_{i}^{0}$ $S_{j}$

$i$
$\ldots$
$k$ $t$

region is gradually slided to again a convex region on bounded by the $H^{t}$

intersections with s and s, and eventually reaches to a circular $H_{i}^{t}$ $S_{j}$

polygon on . This bounded region on each is the image of by

$S_{\infty}^{2}$ $H^{t}$ $\tau$

. We have not ruled out the possibility that

$h_{t}$
intersects for some $H^{t}$
$S_{j}$

$j>k$ , but it will turn out that this never happen.

We next describe how to map the external faces. The $trace$ of the
deformation of s on the external boundary viewed from the above
$H_{i}^{t}$ $S_{i_{0}}$

is described in Figure 6. The image of an external face on by is $S_{i_{O}}$ $h_{t}$

a convex region on bounded by the intersections with s. As $S_{i_{0}}$

$H_{i}^{t}$ $t$

increases, the region is getting enlarged keeping convexity and finally

fills up . $S_{i_{O}}$

is obviously a continuous deformation for $0\leq t<\pi/2$ , and is

$h_{t}$

still continuous at $t=\pi/2$ if we topologize as a 3-ball. What $H^{3}\cup S_{\infty}^{2}$

is saved in this deformation is the property that is an embedding on $h_{t}$

the set of external faces or on each internal face.

Modify a bit to : by pushing each
$h_{\pi/2}$ outward $\hat{h}_{\pi/2}$ $\partial\sigma\rightarrow S_{\infty}^{2}$
$S_{j}$

to the disk on bounded by . We claim that and hence

$S_{\infty}^{2}$
, $\partial S_{j}$
$\hat{h}_{\pi/2}$
$h_{\pi/2}$

and moreover with near $\pi/2$ is a homeomorphism. is a local

$h_{t}$ $t$ $\hat{h}_{\pi/2}$

homeomorphism on the interior of each faces of by the definition. It $\partial\sigma$

is also a local homeomorphism around edges and around vertices by the

definition of (see Figures 5, 6). Hence it is a local homeomorphism to
$h_{t}$

. Since
$S_{\infty}^{2}$
is compact and the image is simply connected, it must be
$\partial\sigma$
110 S. Kojima

Fig. 6.

a homeomorphism. $h_{\pi/2}$ is $nt$ quite dierent from $\hat{h}_{\pi/2}$

and is clearly a
homeomorphism since so is . The bent of the image by is mild for
$\hat{h}_{\pi/2}$ $h_{t}$

near
$t$
$\pi/2$ , and therefore, is also necessarily to be a homeomorphism $h_{t}$

up to some moment.
For any $o\leq t<\pi/2$ , each separates into a convex inward $H_{i}^{t}$ $H^{3}$

region and its complementary outward region. The intersections of any

two of s look quite simple and are classified by the intersection of
$H_{i}^{t}$

their boundaries on . If the intersection on is nonempty and

$S_{\infty}^{2}$ $S_{\infty}^{2}$

transversal, then surfaces intersect transversely for all . If the intersec- $t$

tion on is empty, then as decreases, the intersection of surfaces is

$S_{\infty}^{2}$ $t$

gradually changed from a circle, a point of contact to an empty set. It

may be empty from the beginning. If the boundaries of surfaces on $S_{\infty}^{2}$

are the same, then the intersection is empty for $0<t<\pi/2$ unless they
are the same surface. The transversality of the intersections of s is $H_{i}^{t}$

missed only when either dierent surfaces without intersections for $t>0$
coinside at $t=0$ , or surfaces with circular intersection at the beginning
contact at some moment.
We thus have a family of very visible stratifications of defined $H^{3}$

by the intersections of s and s. The intersection of their convex

$H_{i}^{t}$ $S_{j}$

inward regions in is a compact convex stratum. The convex stratum

$H^{3}$

bounded by s and s is certainly nonempty for near $\pi/2$ . On

$H_{\dot{x}}^{t}$
$S_{j}$
$t$

the other hand, by the continuity of the deformation, bounds a $h_{t}(\partial\sigma)$

locally convex and hence a convex region in also for near $\pi/2$ . It $H^{3}$ $t$

is the same as the stratum bounded by s and s because of its $H_{i}^{t}$ $S_{j}$

convexity. Hence is a cellular map : $h_{t}$

with respect to the $\partial\sigma\rightarrow h_{t}(\partial\sigma)$

stratification of for close enough to

$H^{3}$
$\pi/2$ . $t$
 Polyhedral Decomposition of Hyperbolic 3-Manifolds 111

In this stratification, every surface, that is any one of s and $H_{i}^{t}$

s, plays a role to determine a face of the convex stratum for near

$S_{j}$
$t$

$\pi/2$ . A consequence to this stable property is that the map is cellular $h_{t}$

with respect to the stratification of on an open interval in $[0, \pi/2]$ $H^{3}$

including $\pi/2$ . It also concludes that s are dierent each other for $H_{i}^{t}$

all $0<t<\pi/2$ .
As decreases form $\pi/2$ , this compact convex stratum is continu-
$t$

ously compressed. If the stratum does not degenerate and the structure
of the stratification on the boundary is kept in the deformation up to
$t=0$ , then we are done since turns out to be an embedding and the $h_{0}$

image bounds a convex polyhedron.

Otherwise, there is the first moment at which fails to be $t_{0}\geq 0$ $h_{t}$

cellular since the cellularity is open. Then by continuity of , $h_{t}$ $h_{t_{0}}(\partial\sigma)$

either still bounds a convex region, which is the convex stratum bounded
by s and s, or degenerates to a convex set on some geodesic plane
$H_{i}^{t_{O}}$
$S_{j}$

in . In the first case, the surfaces still in fact intersect transversely

$H^{3}$

at , but some edge of the stratification on the boundary of the convex

$t_{0}$

stratum degenerates. Then two vertices must be close each other if is $t$

near . However the vertices of the stratification on

$t_{0}$
for $t>t_{0}$ is $h_{t}(\partial\sigma)$

the image of the vertices of by the definition of , and hence their $\partial\sigma$
$t_{0}$

mutual distance is bounded away from zero by the definition of . This $h_{t}$

is contradiction. In the second case, the faces of are mapped on $\partial\sigma$

the same geodesic plane by . Hence three vectors from a vertex of $h_{t_{0}}$

to adjacent vertices in the image of

$\partial\sigma$

must be linearly dependent. $h_{t_{0}}$

However they are always independent by the definition of . This is $h_{t}$

also a contradiction. Q.E.D.

5.2 and Theorem. Assigning to each block of

Proof of Proposition
a polyhedron bounded by the image of its boundary, we get a map
$\overline{K}$

from extending : . It is a local homeomorphism on the $\overline{K}^{(2)}\rightarrow\overline{\Delta}^{(2)}$

$\overline{K}$
$h$

interior of blocks. We have already seen that it is a homeomorphism

on the boundary. Hence it is a local homeomorphism everywhere since
there is no vertices in the interior and every cell meets the boundary.
The surjectivity is obvious. Since the image is simply connected, it must
be a homeomorphism.
now determines a convex polyhedral decomposition
$\overline{\Delta}^{(2)}$

of $N$ . $\overline{\Delta}$

The decomposition is invariant under the action of $\pi_{1}(N)$ and the map
can be chosen to be equivariant. Hence it determines a convex polyhedral
decomposition on $N$ with a descending equivalence from $K$ to .
$\Delta$ $\Delta$

Q.E.D.
112 S. Kojima

References
[1] Epstein, D. B. A. and Penner, R., Euclidean decomposition of noncom-
pact hyperbolic manifolds, J. Dierential Geom., 27 (1988), 67-80.
[2] Kojima, S. and Miyamoto, Y., The smallest hyperbolic -manifolds with
$3$

totally geodesic boundary, J. Dierential Geom., 34 (1991), 175-192.

[3] Kojima, S., Polyhedral decomposition of hyperbolic manifolds with
boundary, preprint, Tokyo Inst. Tech. (1991).
[4] Thurston, W. P., The Geometry and Topology of -Manifolds, Lecture
$3$

Note, Princeton University, 1977.

Department of Information Science

Tokyo Institute of Technology
Tokyo 152
Japan
Advanced Studies in Pure Mathematics 20, 1992
Aspects of Low Dimensional Manifolds
pp. 113-124

Behavior of Knots under Twisting

Masaharu Kouno, Kimihiko Motegi and Tetsuo Shibuya

\S 1. Introduction
This paper is a continuation of [6] in the study of the twist move
of knots. First we recall some notations. Let $K$ be an unoriented
smooth knot in the oriented 3-sphere , and $V$ a solid torus endowed
$S^{3}$

with a preferred framing which contains $K$ in its interior and satis-
fies $w_{V}(K)\geq 2$ . ( $w_{V}(K)$ denotes the geometric intersection number
of $K$ and a meridian disk of $V.$ ) Let be an orientation preserv- $f_{n}$

ing homeomorphism of $V$ satisfying (meridian) $=$ (meridian) and

$f_{7\iota}$

(longitude)=(longitude)+n(meridian) in
$f_{n}$
. (We shall not $H_{1}(\partial V)$

distinguish notationally between a homeomorphism and an isomorphism

on a homology group induced by it.) We denote the knot $f_{n}(K)$ in $S^{3}$

by $K_{V,n}$ . If there exsists an orientation preserving homeomorphism of

carrying
$S^{3}$
to , then we write $K_{1}\cong K_{2}$ . Note that $K_{1}\cong K_{2}$ is
$K_{1}$ $K_{2}$

the same as saying that and are ambient isotopic in

$K_{1}$ $K_{2}$ . We note $S^{3}$

that for a given knot , a solid torus

$K$ $V$ and an integer determine a $n$

unique knot type. For a given knot , we have an abundant solid tori
$K$

which contain $K$ to carry out a twist move. Sect.2 is directed towards
the following question : for a given knot $K$ , is it possible to obtain the
same knot by twistings along distinct solid tori from $K$ ? Concerning the
case when an original knot is trivial, we give Example 2.1 and Theorem
2.2. In the case when both solid tori are knotted, we shall give Theorem
2.6 and Examples (see Figures 4, 5). In Sect.3, the behavior of Gromov
invariants under twistings will be studied. In Sect.4, we study the eects
of twistings on primeness of knots. Throughout this paper $N(X)$ , $\partial X$

and int $X$ denote the tubular neighborhood of $X$ , the boundary of $X$
and the interior of $X$ respectively.

\S 2. On twistings along distinct solid tori

Let and be solid tori containing a knot K. We write
$V_{1}$ $V_{2}$ $V_{1}\cong V_{2}$

provided that there exists an orientation preserving homeomorphism $f$

Received June 20, 1990.

114 M. Kouno, K. Motegi and T. Shibuya

of such that
$S^{3}$
$f(V_{1})=V_{2}$ , $f(K)=K$ . Note that $K_{V_{1},n}\cong K_{V_{2},n}$ holds
for any integer $n$ when . To begin with, we give an example as
$V_{1}\cong V_{2}$

follows.
Example 2.1. In Figure 1, because the winding number
$V_{1}\not\cong V_{2}$

of O in equals 2 and that of O in

$V_{1}$ $V_{2}$
equals 3. But $O_{V_{1},-1}\cong O_{V_{2},-1}$ .

$\searrow$ $\swarrow$

Fig. 1.

For twistings of the unknot, we prove the following theorem.

Theorem 2.2. Let $O$ be the unknot and $V_{i}(i=1,2)$ a solid torus
containing $O$ with $w_{V_{\iota}}(O)\geq 1$ . If $O_{V_{1},n_{j}}\cong O_{V_{2},n_{j}}$ holds for infinitely
many integers , then $n_{j}$ .
$V_{1}\cong V_{2}$

To prove this, we prepare some lemmas. Let $V$ be a solid torus

containing a knot $K$ in its interior with $w_{V}(K)\geq 1$ . Then $V-$ int $N(K)$
is a boundary irreducible Haken manifold. Consider the torus decom-
position of $V-$ int $N(K)$ in the sense of Jaco-Shalen [3] and Johannson
[4]. Combining Thurstons uniformization theorem [7], they assert that
$V-$ int $N(K)$ is uniquely decomposed by a family of tori into pieces each
of which is Seifert fibred or admits a complete hyperbolic structure of
finite volume in its interior. Moreover each Seifert piece is one of torus
knot spaces, cable spaces and composing spaces (see [3]). We denote the
piece which contains by , and the piece containing
$\partial V$
$P_{0}$
by $P$ . $\partial N(K)$
 Behavior of Knots under Twisting 115

If $V$ is an unknotted solid torus in which contains $K$ , then $S^{3}-$ int $V$ $S^{3}$

is also a solid torus, and we denote it by . When we perform $(-1/n)-$ $V_{J}$

Dehn surgery on the unknot (the core of ), then the result is also $J$ $V_{J}$

and the image of

$S^{3}$
$K$ becomes a new knot . The next lemma is an $K_{n}^{*}$

interpretation of a twisting.
Lemma 2.3. $K_{V,n}\cong K_{n}^{*}$ .

It follows that $S^{3}-$

int $N(K_{V,n})$ is homeomorphic to $V_{J}\bigcup_{m_{J}=\ell m^{-n}}(V$

$-$ int $N(K))$ .

Lemma 2.4 ([6]). If is a cable space in which a regular fibre $P_{0}$

is presented by $\ell^{p}m^{q}(p\geq 2)$ , then is a Seifert fibred $V_{J}\bigcup_{m_{J}=\ell m^{-n}}P_{0}$

manifold with two exceptional fibres of indices , $|pn+q|$ . The dual knot $p$

of , in
$J$ $J_{n}^{*}$
is a fibre of index $|pn+q|$ .
$V_{J}\bigcup_{m_{J}=lm^{-n}}P_{0}$

Lemma 2.5 ([6]). is hyperbolic, then there exists $N_{V,K}$ such

If $P_{0}$

that $V_{J}\bigcup_{m_{J}=\ell m^{-n}}P_{0}$ is also hyperbolic for $|n|\geq N_{V,K}$ . Moreover for
any $\epsilon>0$ , there such that
$exs\dot{\iota}tsN_{V,K}(\epsilon)$ is a closed geodesic of $J_{n}^{*}$

length< $\in inV_{J}\bigcup_{m_{J}=\ell m^{-n}}P_{0}$

for $|n|\geq N_{V,K}(\in)$ .

Theorem 2.2. If $w_{V_{1}}(O)=1$ (resp. $w_{V_{2}}(O)=1$ ), then by

Proof of
the assumption and Theorem 4.2 in [6], $w_{V_{2}}(O)=1$ (resp. $w_{V_{1}}(O)=1$ )
must hold. In this case $O$ is a core of both and , so we have $V_{1}$ $V_{2}$

. Assume
$V_{1}\cong V_{2}$ $w_{V_{l}}(O)\geq 2$ and consider the torus decomposition of
$V_{i}-$ int $N(O)$ . Let be the piece containing $P_{i}$
. Since $O$ is trivial, $\partial V_{?}.$

can not be a composing space. We remark that

$P_{i}$
is necessarily $V_{\dot{\iota}}$

unknotted by the assumption (see [9]), and $S^{3}-$

int . is also a solid $V_{?}$

torus . Then we can characterize the core of

$V_{J_{i}}$
in $E(O_{V_{\dot{x}},n})=$ $V_{J_{i}}$

( $V_{i}-$ int $N(O)$ ), which is denoted by

$V_{J_{i}}\bigcup_{m_{J_{i}}=\ell_{\dot{z}}m_{i}^{-n}}$ , as follows. $J_{\dot{0},n}^{*}$

There exists a constant such that is an exceptional fibre of

$N_{V_{i},O}$ $J_{i,n}^{*}$

unique maximal index or a unique shortest closed geodesic in $E(O_{V_{i},n})$

by Lemmas 2.4 and 2.5 for $|n|\geq N_{V_{i},O}$ . Now we take as above. Let $n$

$f$
be an orientation preserving homeomorphism of sending to $S^{3}$
$O_{V_{1},n}$

. Then by an ambient isotopy, we may assume

$O_{V_{2},n}$ maps $N(O_{V_{1},n})$ $f$

to $N(O_{V_{2},n})$ and maps to (see also [8]). From this, we see that
$J_{1,n}^{*}$ $J_{2,n}^{*}$

is an orientation preserving homeomorphism from

$f|_{V_{1}}$
to with $V_{1}$ $V_{2}$

$f|_{V_{1}}(O)=O$ . Moreover maps to . This

$f|_{V_{1}}$ $\ell_{1}m_{1}^{-n}$ $\ell_{2}^{\in}m_{2}^{-\in n}(\in=\pm 1)$

implies that maps to . By extending

$f|_{V_{1}}$
to , we get a
$\ell_{1}$ $\ell_{2}^{\Xi}$ $f|_{V_{1}}$
$S^{3}$

required homeomorphism. This completes the proof of Theorem 2.2.

Q.E.D.

If we require both $V_{1}$

and $V_{2}$
are knotted, the following result holds.
116 M. Kouno, K. Motegi and T. Shibuya

Theorem 2.6. Let $K$ be a knot in and a knotted solid torus $S^{3}$ $V_{i}$

containing K. Suppose that and the core of satisfies

$V_{1}\subset V_{2}$ $C_{1}$ $V_{1}$

$w_{V_{2}}(C_{1})\geq 2$ and $w_{V_{1}}(K)\geq 2$ . Then for any pair $K_{V_{1},m}\not\cong K_{V_{2},n}$

$(m, n)\neq(0,0)$ (Figure 2).

Fig. 2.

Proof. Let : $f_{m}$

and :
$V_{1}\rightarrow V_{1}$
be twist homeo-
$g_{n}$
$V_{2}\rightarrow V_{2}$

morphisms with $m$ twist and -twist respectively. By Theorem 2.1 in $n$

[6], for any integer

$g_{n}(C_{1})\not\cong C_{1}$ . Meanwhile $f_{m}(C_{1})\cong C_{1}$ for
$n$ $\neq 0$

any integer $m$ . So the composition : sends

$g_{n}of_{m}^{-1}$ $V_{1}\rightarrow g_{n}(V_{1})$ $C_{1}$

to . We remark that
$g_{n}(C_{1})\not\cong C_{1}$ and are knotted in
$C_{1}$
, $g_{n}(C_{1})$ $S^{3}$

because they are geometrically essential in the knotted solid torus . $V_{2}$

Also satisfies $g_{n}of_{m}^{-1}(K_{V_{1},m})=K_{V_{2},n}$ . Using Theorem [5], we

$g_{n}of_{m}^{-1}$

can conclude , if . In the case of $=0$ , $K_{V_{2},n}\cong K$

$K_{V_{1},m}\not\cong K_{V_{2},n}$ $n$ $\neq 0$ $n$

but $K_{V_{1},m}\cong K$ holds only when $m=0$ by Theorem 2.1 [6]. It follows
that for any pair $(m, n)\neq(0,0)$ .
$K_{V_{1},m}\not\cong K_{V_{2},n}$ Q.E.D.

Remark. In the above theorem, the condition $w_{V_{2}}(C_{1})\geq 2$ excludes

the following trivial example.
Also in general, if both solid tori and are knotted then by $V_{1}$ $V_{2}$

Schuberts Satz 1 ([12]), we may assume one of the following occurs by

 Behavior of Knots under Twisting 117

Fig. 3.

an ambient isotopy of $S^{3}$

which leaves $K$ fixed. (1) or ,
$V_{1}\subset V_{2}$ $V_{2}\subset V_{1}$

(2) $V_{1}\cup V_{2}=S^{3}$

, and (3) there exists a solid torus $W$ in int $V_{1}\cap int$ $V_{2}$

such that $w_{V_{1}}(C_{W})=w_{V_{2}}(C_{W})=1$ for the core of$C_{W}$of $W$ .

Theorem 2.6 corresponds to the case (1). As for cases (2) and (3),
there exist inessential examples as in Figure 4 and Figure 5 respectively.

Fig. 4.
118 M. Kouno, K. Motegi and T. Shibuya

Fig. 5.

\S 3. Gromov invariants

The notion of the Gromov invariant of closed manifolds was intro-

duced by Gromov [1]. In the 3-dimensional case, Thurston defined the
Gromov invariant of compact 3-manifolds whose boundaries consists of
tori [14]. In this section we shall study the Gromov invariant of the
exterior of a knot $K$ in which we simply call the Gromov invariant of
$S^{3}$

$K$ and we denote it by $||K||$ . For the definition of the Gromov invariant,

the reader is referred to [1], [14] and [13].

First we prove the following.

Theorem 3.1. Let $K$ be a knot in and $V$ a knotted solid torus
$S^{3}$

containing K. Then $||K_{V,n}||=||K||$ holds for any integer . $n$

Proof. If then $K_{V,n}=K$ for any integer . So we

$w_{V}(K)\leq 1$ , $n$

assume $w_{V}(K)\geq 2$ . The exterior of $K_{V,n}(K_{V,0}\cong K)$ is described as

( $S^{3}-$ int $V$ ) ( $V-$ int $N(K)$ ) for some gluing homeomorphism
$\bigcup_{h_{n}}$
. $h_{n}$

Since $V$ is knotted, ( $S^{3}-$

int $V$ ) is an incompressible torus. Also
$\partial$

is $\partial V$

an incompressible torus in $V-$ int $N(K)$ because $w_{V}(K)\geq 2$ . Hence we

have the following equality independent of by Somas theorem [13]. $n$

$||K_{V,n}||=||E((K_{V,n})||=||$ ( $S^{3}-$ int $V$ ) $II$

( $V-$ int $N(K)$ ) . $||$

It follows that $||K_{V,n}||=||K||$ . Q.E.D.

 Behavior of Knots under Twisting 119

Hence, in Theorem 2.6, and have the same Gromov

$K_{V_{1},m}$ $K_{V_{2},n}$

invariants for any pair $(m, n)$ .

The following is straightforward from Theorem 3.1.
Corollary 3.2. Suppose that and are knots with $||K_{1}||\neq$
$K_{1}$ $K_{2}$

. Then
$||K_{2}||$ can not be obtained by a sequence of twistings along
$K_{2}$

knotted solid tori from . $K_{1}$

On the other hand, if $V$ is unknotted we have:

Proposition 3.3. Let $O$ be the unknot in . For any real num- $S^{3}$

$berr$ , there exists an unknotted solid torus $V$ containing $O$ such that
$||O_{V,1}||>r$ .

Proof. Consider a solid torus $V$ as in Figure 6. Then in the exterior

of , there exist incompressible tori which decompose it into figure
$O_{V,1}$ $k$

eight knot spaces, 1 Whitehead link space and 1 composing space. Hence
$||O_{V,1}||=1/v_{3}(k$ Vol(figure eight knot complement) $+Vol(Whitehead$

link complement)), where is the volume of the regular ideal simplex

$v_{3}$

(see [14] [13]). Thus the result holds for some integer $k>0$ . Q.E.D.

$\rightarrow$

$0_{\vee 1}$

Fig. 6.

This also shows that for any knot $K$ and any real number , there $r$

exists an unknotted solid torus $V$ such that $K_{V,1}>r$ .

But the Gromov invariants behave as follows once $V$ is fixed.
Proposition 3.4. Let $K$ be a knot in and $V$ an unknotted solid $S^{3}$

torus containing K. Then $||K_{V,n}||$ is less than a constant $C_{V,K}$ for any
integer . $n$
120 M. Kouno, K. Motegi and T. Shibuya

Proof.We may assume $w_{V}(K)\geq 2$ . If is a cable space, $||K_{V,n}||$ $P_{0}$

is constant for all but at most two integers such that a regular fibre is $n$

presented by for some . If

$\ell^{p}m$
is a composing space, then twisting
$p$ $P_{0}$

along is reduced to that along a knotted solid torus $W$ bounded by the
$V$

torus which separates $K$ and

$(\subset\partial P_{0})$
(see Sublemma 3.7 [6]). Hence $\partial V$

Theorem 2.1 in [6] implies the result. Suppose that is hyperbolic, by $P_{0}$

Lemma 2.3 is also hyperbolic for $|n|\geq N_{V,K}$ . Then

$V_{J}\bigcup_{m_{J}=\ell m^{-n}}P_{0}$

we have $Vol(int(V_{J}\bigcup_{m_{J}=\ell m^{-n}}P_{0}))<Vol(intP_{0})$ by Thurstons theorem

(6.5.6 Theorem [14]), and from this we have the following inequality for
$|n|\geq N_{V,K}$ ,

$||K_{V,n}||=1/v^{3}(\sum_{i\neq 0P_{i}:hyperbo1ic}Vol(intP_{i})+Vol(int(V_{J}\cup P_{0})))m_{J}=\ell m^{-n}$

$<1/v^{3}(\sum_{i\neq 0P_{i}:hyperbo1ic}Vol(intP_{i})+Vol(intP_{0}))$

$=||KII$ $J||$ .

Now we set $C_{1}=\max\{||K_{V,n}|| : and we take

|n|<N_{V,K}\}$ $C_{V,K}=$
$\max\{C_{1}, ||KII J||\}$ , then $C_{V,K}$ is the required constant. Q.E.D.

Example 3.5 (Thurston [14]).

$\rightarrow\cdots$

$||0||=0$ $||0_{\vee,\uparrow}||=2$

Fig. 7.

The Gromov invariants of these knots tend from below to a finite

limit $(=. 3.6)$ .

\S 4. Primeness of knots under twistings

In this section, we investigate the eects of twistings on primeness
of knots. To begin with, we consider the case when a twisting solid torus
is knotted.
 Behavior of Knots under Twisting 121

Theorem 4.1. Let $K$ be a knot in and $V$ a knotted solid torus $S^{3}$

containing K. Then $K$ is prime if and only if $K_{V,n}$ is for any $p\dot{r}ime$

integer $n$ .

Proof. We may assume $w_{V}(K)\geq 2$ . Consider the torus decom-

position of $V$ int $N(K)$ and denote the piece containing

by $\partial N(K)$

$P$ . Suppose that $K$ is a prime knot, then it turns out $P$ is not a com-
posing space. Now we consider the torus decomposition of $E(K_{V,n})=$
( $S^{3}-$ int $V$ ) ( $V-$ int $N(K)$ ). In $E(K_{V,n})$ , $P$ is also a decomposing
$\bigcup_{h_{n}}$

piece. It follows that $K_{V,n}$ is also prime for any integer . Q.E.D. $n$

If $V$ is unknotted, then the following example exists.

Example 4.2. In Figure 8, K is a prime knot, but $K_{V,n}$ is a
composite knot for any nonzero integer n.

$\rightarrow$

$K_{\vee}|\cap$

Fig. 8.

In this example $K$ has a locally knotted arc in $V$ (i.e. there is a
3-ball $B\subset V$ such that $(B, B\cap K)$ is a knotted ball pair). If $K$ does
not have a locally knotted arc in $V$ , then we get the following.
Theorem 4.3. Let $V$ be an unknotted solid torus containing $K$
without a locally knotted arc. Then $K_{V,n}$ is prime for all but at most
finitely many integers . $n$

Proof. Consider the torus decomposition of $V-$ int $N(K)$ , and let
$P$ be a piece containing and a piece containing .
$\partial N(K)$ $P_{0}$ $\partial V$

Sublemma. Suppose that K $\subset V$

does not have a local knot. Then
P can not be a composing space.

Proof of Sublemma. Suppose that $P$ is a composing space. Let $T$ be

a component of $\partial P$
which does not separate and . Note that $T$
$\partial V$
$\partial N(K)$
122 M. Kouno, K. Motegi and T. Shibuya

bounds a nontrivial knot exterior $E$ , and a regular fibre of $P$ coincides
a boundary of a meridian disk of $N(K)$ . Hence we have a saturated
annulus $A$ which joins $T$ and . Then $D=A\cup D$ becomes a
$\partial N(K)$

meridian disk of $W=S^{3}-$ int $E$ . Since $K\cap D^{J}$ and $K\cap D$ consist of
one point, $K$ has a locally knotted arc in $V$ . This is a contradiction.
Q.E.D.

If is a cable space,
$P_{0}$
is a (nontrivial) torus knot
$V_{J}\bigcup_{m_{J}=\ell m^{-n}}P_{0}$

exterior except for at most only two integers by Lemma 2.4. If $n$ is $P_{0}$

a -fold composing space, then

$k$
is a $(k-1)$ -fold com-
$V_{J}\bigcup_{m_{J}=\ell m^{-n}}P_{0}$

posing space for any integer . Finally we consider the case when
$n$ is $P_{0}$

hyperbolic. By Lemma 2.5, we see that is also hyper-

$V_{J}\bigcup_{m_{J}=\ell m^{-n}}P_{0}$

bolic except for at most finitely many integers . It follows that in any $n$

case, is boundary irreducible Haken manifold. Now we

$V_{J}\bigcup_{m_{J}=\ell m^{-n}}P_{0}$

divide into two cases depending upon whether $P=P_{0}$ or not. If $P=P_{0}$ ,
then $V_{J}\bigcup_{m_{J}=\ell m^{-n}}P=V_{J}\bigcup_{m_{J}=\ell m^{-n}}P_{0}$ can not be a composing space
by Sublemma and the above, and it becomes a decomposing piece in
$E(K_{V,n})$ . Thus $K_{V,n}$ is prime except for at most finitely many integers

. If $P\neq P_{0}$ , then it turns out that $P$ is still a decomposing piece in
$n$

$E(K_{V,n})$ . Since $P$ is not a composing space, $K_{V,n}$ is prime except for at

most finitely many integers . $n$ Q.E.D.

Remark 4.4. Even if $K$ does not have a locally knotted arc in $V$ ,
there is an example such that $K_{V,n}$ is a composite knot for some integer
(see Figure 9).
$n$

$\rightarrow$

$K_{\bigvee_{\iota}1}$

Fig. 9.

When an original knot is trivial, Scharlemann-Thompson [11],

Eudave-Munoz and Gordon have shown the following result, which is
a generalization of the theoremUnknotting number one knots are
prime [10].
 Behavior of Knots under Twisting 123

Theorem 4.5 ([11]). Let $V$be a solid torus containing the unknot
$O$ with $w_{V}(O)\leq 2$ . Then $O_{V,n}$ is prime for any integer . $n$

Since the unknot can not have a locally knotted arc, as an applica-
tion of Theorem 4.3, we have the following.

Corollary 4.6. Let $V$ be a solid torus containing the unknot $O$ .
Then $O_{V,n}$ is prime for all but at most finitely many integers . $n$

We conclude this paper with the following question.

Question. Is the result of twisting of the unknot always prime $.p$

Acknowledgement. Authors wish to thank K. Miyazaki for sug-

gesting that the local knottedness is essential in Theorem 4.3. They also
wish to thank the referee for helpful comments.

References
[1] M. Gromov, Volume and bounded cohomology, Inst. Hautes Etudes Sci.
Publ. Math., 56 (1983), 213-307.
[2] W. Jaco, Lectures on three manifold topology, Conference board of
Math. Science, Regional Conference Series in Math. 43. Amer. Math.
Soc., 1980.
[3] W. Jaco and P. Shalen, Seifert fibered spaces in -manifolds, Mem. $3$

Amer. Math. Soc. 220, 1979.

[4] K. Johannson, Homotopy equivalences of -manifolds with boundaries,
$3$

Lecture Notes in Math., Vo1.761. Springer-Verlag, 1979.

[5] M. Kouno, On knots with companions, Kobe J. Math.,2 (1985), 143-148.
[6] M. Kouno, K. Motegi and T. Shibuya, Twisting and knot types, preprint.
[7] J. Morgan and H. Bass, The Smith conjecture , Pure and Applied
Math. Academic Press, 1984.
[8] K. Motegi, Homology -spheres which are obtained by Dehn surgeries
$3$

on knots, Math. Ann., 281 (1988), 483-493.

[9] D. Rolfsen, Knots and links, Mathematics Lecture Series, No.7. Pub-
lish or Perish, Berkeley, Calif., 1976.
[10] M. Scharlemann, Unknotting number one knots are prime, Invent. Math. ,
82 (1985), 37-55.
[11] M. Scharlemann and A. Thompson, Unknotting number, genus, and
companion tori, Math. Ann., 280 (1988), 191-205.
[12] H. Schubert, Knoten und Vollringe, Acta Math., 90 (1953), 131-286.
[13] T. Soma, The Gromov invariant of links, Invent. Math., 64 (1981),
445-454.
[14] W. Thurston, The geometry and topology of -manifolds, Lecture$3$

Note Princeton Univ., 1978.

124 M. Kouno, K. Motegi and T. Shibuya

Masaharu Kouno
College of Liberal Arts
Kobe University
Kobe 657, Japan

Kimihiko Motegi
Department of Mathematics
Kyushu University 33
Fukuoka 812, Japan

Tetsuo Shibuya
Department of Mathematics
Osaka Institute of Technology
Osaka 535, Japan
Advanced Studies in Pure Mathematics 20, 1992
Aspects of Low Dimensional Manifolds
pp. 125-145

Polynomial Invariants
of -Bridge Links through 20 Crossings
$2$

Taizo Kanenobu* and Toshio Sumi

In this paper, we calculate the homfly polynomial $P_{L}(v, z)$ , Kau-

man polynomial $F_{L}(a, z)$ , Jones polynomial $V_{L}(t)$ , $Q$ polynomial $Q_{L}(z)$ ,
2-variable Conway polynomial , and reduced Conway polyno- $\nabla_{L}(t_{1}, t_{2})$

mial of a 2-bridge link with crossing number $\leq 20$ and list all
$\overline{\nabla}_{L}(z)$ $L$

the pairs sharing the same polynomial invariants (Table 2). This paper
is a continuation of [9], where these polynomial invariants except for the
2-variable Conway polynomial for 2-bridge knots through 22 crossings
are calculated and all the pairs having the same polynomial invariants
are listed. The total number of the links is 44,118, where we ignore the
orientations of both a link and its ambient space. If we consider them,
this amounts 175,788. The program is written in Turbo Pascal for the
NEC $PC$ -9801 Series as before.
We observe the following for 2-bridge links through 20 crossings:

Fact 1. $P_{L}(v, z)=P_{L}(v, z)$ i $V_{L}(t)=V_{L}(t)$ and $\overline{\nabla}_{L}(z)=$

$\overline{\nabla}_{L}(z)$
.

Fact 2. If $P_{L}(v, z)=P_{L}(v, z)$ and $P_{L}\wedge(v, z)=P_{L}\wedge(v, z)$ , then

$\nabla_{L}(t_{1}, t_{2})=\nabla_{L}(t_{1}, t_{2})$ .

Fact 3. The number of links having the same homfly or Kauman

polynomial is at most two.

Fact 4. $P_{L}(v, z)=P_{L^{\wedge}}(v, z)$

i $\nabla_{L}(t_{1}, t_{2})=\nabla_{L}\wedge(t_{1}, t_{2})(=$

$-\nabla_{L}(t_{1}^{-1}, t_{2}))$ .

Here is a 2-bridge link obtained from by reversing the orien-

$L^{\wedge}$
$L$

tation of one of the 2 components. Facts 1 and 3 are the same as those
in [9]. For Fact 3, we do not consider the pair of 2-bridge links and $L$

Received June 29, 1990.

$*This$ work was supported in part by Grant-in-Aid for Encouragement of

Young Scientist (No. 01740057), Ministry of Education, Science and Culture.

126 T. Kanenobu and T. Sumi

which share the same Kauman polynomial and have linking number
$L$

zero (such as 3800 1669, 2429 of 19 crossing links in Table ). For these
links, it holds $ F_{L}=F_{L}=F_{L}\wedge=F_{L}\wedge$ . The only if part of Fact 4
cannot be deduced from only Table 2. We must check the 2-variable
Conway polynomials. The example as in Fact 4 is constructed in [5,
Theorem 9].

For the pair and $L$ , where $L\neq L$ ,

$L$
, sharing the same $Q$ $L^{\prime\wedge}$

polynomial, the cases given in Table 1 occur, where the last column gives
an example for each case from Table 2. For example, Case 5 indicates the
pair such that $V_{L}=V_{L}(V_{L}\wedge=V_{L}\wedge)$ , $P_{L}=P_{L}$ , , $F_{L}=F_{L}$ $ P_{L^{\wedge}}\neq P_{L^{\gamma}}\wedge$

,
$(F_{L^{\wedge}}=F_{L}\wedge)$ .
$\nabla\Gamma-\nabla_{L}(\nabla_{L^{\wedge}}\neq\nabla_{L^{J\wedge}}),\overline{\nabla}_{L}=\overline{\nabla}_{L},\overline{\nabla}_{L^{\wedge}}\neq\overline{\nabla}_{L^{l\wedge}}$

Cases 3-5 explain Fact 2. Relating to Cases 2 and 3, we can construct

the following examples:

(i) Arbitrarily many skein equivalent fibered 2-bridge links with the
same 2-variable Conway polynomial ([5, Theorem 7]).

(ii) Arbitrarily many skein equivalent 2-bridge links which have mutu-
ally distinct 2-variable Conway polynomials ([6, Theorem 2]).

Relating to Cases 4-6, we can construct the following examples:

(iii) A pair of skein equivalent 2-bridge links with the same Kauman
polynomial but distinct 2-variable Conway polynomials ([6, Theo-
rem 7]).

(iv) Arbitrarily many skein equivalent fibered 2-bridge links which have
the same Kauman and 2-variable Conway polynomial ([8, Theo-
rem 2]).

Relating to Case 7, we can construct the following example:

(v) Arbitrarily many 2-bridge links which have the same $Q$ and 2-
variable Conway polynomial, but distinct Jones polynomials ([7,
Theorem]).
 Polynomial Invariants of 2-Bridge Links 127

Table 1

\S 1. $2$
-bridge link
The 2-bridge links are classified in Schuberts normal form $S(p, q)$

$[10]$ , where $p>0,$

$-p<q<p$ , and and are coprime integers. $p$ $q$

Proposition 1. $S(p, q)$ and $S(p, q)$ are isotopic as $ori$ ented (resp.
unoriented) links if and only if:

$p=p$ , $q^{\pm 1}\equiv q$

$mod 2p$ (resp. $mod p$ ).

The following properties are easily seen from Schubert s normal form
(cf. [1, Proposition 12.5]):
Proposition 2. (2) A 2-bridge link $L=K_{1}\cup K_{2}$ is interchange-
able, that is, there is an isotopy of such that $\varphi(K_{i})=K_{j}$ , $i\neq j$ .
$\varphi$
$S^{3}$

(2) A 2-bridge link $L=K_{1}\cup K_{2}$ is invertible, that is, there is an

isotopy of such that $\psi(K_{i})=-K_{i}$ , $i=1,2$ .
$\psi$
$S^{3}$

Let be an oriented 2-bridge link. Then we denote by

$L$
a 2-bridge $L^{\wedge}$

link obtained by reversing the orientation of one of the two components

of , and by
$L$
a mirror image of . So if $L=S(p, \pm q)$ , $q>0$ , then
$\overline{L}$
$L$

$L^{\wedge}=S(p, \pm(q-p))$ and $\overline{L}=S(p, \mp q)$ . Note that $\overline{(L^{\wedge})}=(\overline{L})^{\wedge}=$

which we denote by
$S(p, \pm(p-q))$ , Thus according as the isotopy
$\overline{L}^{\wedge}$

types of the four oriented 2-bridge links , , , and , $L=S(p, q)$ ,

$\overline{L}^{\wedge}$

$L$ $L^{\wedge}$ $\overline{L}$

there are three types for the 2-bridge link:

Type $A:L=\overline{L}^{\wedge}\neq\overline{L}=L^{\wedge}$
, that is, $q(p-q)\equiv 1mod 2p$ .
Type $B:L=L^{\wedge}\neq\overline{L}=\overline{L}^{\wedge}$
, that is, $q(p-q)\equiv-1mod 2p$ .
128 T. Kanenobu and T. Sumi

Type $C$
: No two of $L$
, $\overline{L}$

, $L^{\wedge}$
, and
$\overline{L}^{\wedge}$

are isotopic, that is, $q(p-q)\not\equiv\pm 1$

$mod 2p$ .

Given a 2-bridge link in Schuberts normal form $S(p, q)$ , it can be

put in Conways normal form $C(a_{1}, a_{2}, \ldots, a_{k})$ , (cf. [9, Fig. 3]), where
$\underline{1}$
$\underline{p}=a_{1}+\underline{1}$
(1) .
$q$ $a_{2}+\cdots+a_{k}$

Note that this is a normal form for an unoriented 2-bridge link.

Let , $q>0$ and , ,
$p$ , $a_{k}>0$ . Since $a_{k}=(a_{k}-1)+1/1$ , if we
$a_{1}$ $a_{2}$ $\ldots$

suppose $a_{k}>1$ or fix the parity of , this expression is unique and the $k$

crossing number is $a_{1}+a_{2}+\cdots+a_{k}$ .

Proposition 3. Every 2-bridge link $S(p, q)$ , $q>0$ , of Type can $A$

be expressed as $C(a_{1}, a_{2}, \ldots, a_{n}, a_{n}, \ldots, a_{2}, a_{1})$ , $a_{i}>0$ , and vice versa.

Proof Suppose that

$\frac{p}{q}=b_{1}+\frac{1}{b_{2}}+\cdots+\frac{1}{b_{\ell}}$ ,

where $b_{i}>0$ and $\ell$

is even. Then we have

$\left(\begin{array}{ll}s & q\\r & p\end{array}\right)=$ $\left(\begin{array}{ll}0 & 1\\1 & b_{1}\end{array}\right)\left(\begin{array}{ll}0 & 1\\1 & b_{2}\end{array}\right)\ldots\left(\begin{array}{ll}0 & 1\\1 & b_{\ell}\end{array}\right)$
,

where $p>q>s>0$ , $p>r>s>0$ , and $ps-rq$ $=1$ (cf. [11]). Since

$(q^{2}+1)/p\in \mathbb{Z}$ ,

$\frac{r^{2}+1}{p}=\frac{r^{2}+(ps-qr)^{2}}{p}=\frac{q^{2}+1}{p}r^{2}-2qrs+ps^{2}\in \mathbb{Z}$ .

Let $x=(q^{2}+1)/p$ and $y=(r^{2}+1)/p$ . Since xp-qq $=1$ and sp-rq $=1$ ,
there exists an integer such that $s-x=aq$ and $r-q=ap$ . Since $a$

yp-rr $=1$ and sp-qr $=1$ , there exists an integer such that $s-y=br$ $b$

and $q-r$ $=bp$ . Then we have $a=b=0$ and $q=r$ . From the uniqueness
of the continued fraction, we have , $b_{2}=b_{\ell-1}$ , , $b_{\ell/2}=b_{\ell/2+1}$ . $b_{1}=b_{\ell}$
$\ldots$

The converse is easy, and the proof is complete.

Proposition 4. Every 2-bridge link $S(p, q)$ , $q>0$ , of Type $B$ can
be expressed as $C(a_{1}, a_{2}, \ldots, a_{n}, 2a-1, a_{n}, \ldots, a_{2}, a_{1})$ , $a_{i}>0$ , $a>0$ ,
and vice versa.

Proof Since $q(p-q)\equiv-1mod 2p$ , there is an integer $b$

such that

(2) $q^{2}-1=p(q+2b)$ .
 Polynomial Invariants of 2-Bridge Links 129

First we show that there exist positive integers $x$ , , , $w$ satisfying:
$y$ $z$

(3) $\left(\begin{array}{ll}w & y\\z & x\end{array}\right)\left(\begin{array}{ll}0 & 1\\1 & -1\end{array}\right)\left(\begin{array}{ll}w & z\\y & x\end{array}\right)=\left(\begin{array}{lll}q & +2b & q\\ & q & p\end{array}\right)$

and

(4) $xw-yz=\xi j=\pm 1$ , $x>y$ .

From (2), we have

$\frac{q+1}{2}\frac{q-1}{2}=\frac{p}{4}(q+2b)$ ,

which is even, and thus is also even, that is, $p\equiv 0mod 8$ . Let $p/4$

$p\pm=g.c.d.(p/4, (q\pm 1)/2)>0$ . Since g.c.d. $((q+1)/2, (q-1)/2)=1$ ,

we have $p_{+}p-=p/4$ . Let $z=p_{+}+p_{-}$ , which is an odd integer. Let
$(2p_{+}, (q-1)/2p-)$ if $p_{+}<p_{-}$ ,
$(x, y)=\{$
$(2p_{-}, (q+1)/2p_{+})$ if $p_{-}<p_{+}$ .

Since $\frac{q+1}{2p+}\frac{q-1}{2p-}=q+2b$ is odd, both $(q+1)/2p_{+}$ and $(q-1)/2p-$ are

odd, so let . Then , , , $w$ satisfy (3) and (4). Since
$w=\frac{1}{2}(\frac{q+1}{2p+}+\frac{q-1}{2p-})$ $x$ $y$ $z$

$z>x>0$ , there are integers and such that $z=ax+u$ , $a>0$ and $a$ $u$

$x>u>0$ . Let $v=w-ay$ . Then xv-yu $=\in$ , and there exist positive
integers , , ,
$a_{1}$such that $a_{2}$ $\ldots$ $a_{n}$

$\left(\begin{array}{ll}v & y\\u & x\end{array}\right)=$ $\left(\begin{array}{ll}0 & 1\\1 & a_{1}\end{array}\right)\left(\begin{array}{ll}0 & 1\\1 & a_{2}\end{array}\right)\ldots\left(\begin{array}{ll}0 & 1\\1 & a_{n}\end{array}\right)$
.

$Then\in=(-1)^{n}$ and

$\left(\begin{array}{ll}v & y\\u & x\end{array}\right)\left(\begin{array}{lll}0 & & 1\\1 & 2a & -1\end{array}\right)\left(\begin{array}{ll}v & u\\y & x\end{array}\right)=\left(\begin{array}{lll}q & +2b & q\\ & q & p\end{array}\right)$
,

and so $S(p, q)$ can be expressed as $C(a_{1}, a_{2}, \ldots, a_{n}, 2a-1, a_{n}, \ldots, a_{2}, a_{1})$ .
Conversely, if is odd, a rotation through about the axis $E$ as shown in
$n$ $\pi$

Fig. 1, where and ,

$\alpha=S_{2}^{a_{1}}S_{1}^{-a_{2}}\ldots S_{2}^{a_{n}}S_{1}^{1-a}$ $\alpha=S_{1}^{1-a}S_{2}^{a_{n}}\ldots S_{1}^{-a_{2}}S_{2}^{a_{1}}$

gives an isotopy of which reverses the orientation of one of the two $S^{3}$

components. If is even, we have a similar isotopy of . This completes

$n$
$S^{3}$

the proof.
Let be the set of the unoriented 2-bridge links
$\mathcal{L}_{n}$
$C(a_{1}, a_{2}, \ldots, a_{k})$ s
satisfying the following:

(5) $a_{1}$ , $a_{k}\geq 2$ , $a_{2}$ , $\ldots$ , $a_{k-1}\geq 1$ .

130 T. Kanenobu and T. Sumi

$E$

Fig. 1.

(6) Either $a_{i}=a_{k-i+1}$ for all $i\geq 1$ or $a_{1}=a_{k}$ , $a_{2}=a_{k-1}$ , $\ldots$ , $a_{i-1}=$
$a_{k+2-i}$ , $a_{i}>a_{k+1-i}$ for some $i\geq 1$ .

(7) $a_{1}+a_{2}+\cdots+a_{k}=n$ .

In other words, this is the set of representatives of unoriented 2-bridge

links with crossings up to chirality. Let
$n$ and be the $A\mathcal{L}_{n}$ $B\mathcal{L}_{n}$

subsets of consisting of the unoriented 2-bridge links of the form

$\mathcal{L}_{n}$

and $C(c_{1}, \ldots, c_{\ell}, 2c-1, c_{\ell}, \ldots, c_{1})$ , respectively.

$C(b_{1}, \ldots, b_{\ell}, b_{\ell}, \ldots, b_{1})$

There is a bijective mapping

$\psi:A\mathcal{L}_{2m}\rightarrow B\mathcal{L}_{2m-1}$

defined by

$\psi(C(b_{1}, \ldots,b_{\ell}, b_{\ell}, \ldots, b_{1}))$

$C(b_{1}, \ldots, b_{\ell}-1,1, b_{\ell}-1, \ldots, b_{1})$ if $b_{\ell}>1$

$=\{$
$C(b_{1}, \ldots, b_{\ell-2},2b_{\ell-1}+1, b_{\ell-2}, \ldots, b_{1})$ if $b_{\ell}=1$ .

The explicit numbers of and are given by Ernst and Sumners $\mathcal{L}_{n}$ $A\mathcal{L}_{n}$

[3], in which they are denoted by $TL_{n}$ and ATLn. Thus we can know
the number of , which equals ATLn. Let denote the number
$B\mathcal{L}_{n}$ $TL_{n}^{**}$

of oriented 2-bridge links of crossings up to isotopy. Since $n$

$TL_{n}^{**}=4TL_{n}-2ATL_{n+1}-2ATLn$ ,

we have:
 Polynomial Invariants of 2-Bridge Links 131

Proposition 5.

$TL_{n}^{**}=\{$
$(2^{n-2}+2^{\frac{n+2}{2}}-2^{\frac{n-2}{2}}+2)/3$
if $n\equiv 0$ $mod 2$ ,

$(2^{n-2}-2)/3$ if $n\equiv 1$ $mod 2$ .

Remark. $TL_{0}^{**}=2$ and $TL_{1}^{**}=0$ . $\sum_{n=0}^{20}TL_{n}=44,118$ and

$\sum_{n=0}^{20}TL_{n}^{**}=175,788$ .

\S 2. Conway polynomial
Let be a 2-component link and
$L$
its Conway $\nabla_{L}(t_{1}, t_{2})\in \mathbb{Z}[t_{1}^{\pm 1}, t_{2}^{\pm 1}]$

polynomial, where the components correspond to the labels and . $t_{1}$ $t_{2}$

This is a uniquely determined invariant of the isotopy type of an oriented

link and is related to the 2-variable Alexander polynomial by $\triangle(x_{1}, x_{2})$

$\triangle(t_{1}^{2}, t_{2}^{2})=\pm t_{1}^{n_{1}}t_{2}^{n_{2}}\nabla(t_{1}, t_{2})$

,

is a unit (cf. [2,4]). Let

$where\pm t_{1}^{n_{1}}t_{2}^{n_{2}}$ , , be the 2-component $L_{n}$ $n$ $\in \mathbb{Z}$

links with labels and , which contain a 2-braid and are identical
$t_{1}$ $t_{2}$ $\sigma_{1}^{n}$

except near the 2-braid. Let be the Conway polynomial of $\nabla_{n}(t_{1}, t_{2})$

$L_{n}$
.
(8) Suppose that the 2-braid consists of the dierent components with
orientation not parallel. Then
$\nabla_{2}$
( ,
$t_{1}$ $t_{2}$
) $+\nabla_{-2}(t_{1},$ $t_{2})=(t_{1}t_{2}^{-1}+t_{1}^{-1}t_{2})\nabla_{0}(t_{1}$ , $t_{2})$
.

(9) Suppose that the 2-braid consists of the same component having
label and parallel orientation. Then
$t_{i}$

$\nabla_{1}(t_{1}, t_{2})=\nabla_{-1}(t_{1}, t_{2})+(t_{i}-t_{i}^{-1})\nabla_{0}(t_{1}, t_{2})$ .

(10) Let $L\# L$ be the connected sum of two 2-component links and $L$ $L$

such that the connection takes place between the components with the
same label . Then $t_{i}$

$\nabla_{L\beta L}=(t_{i}-t_{i}^{-1})\nabla_{L}\nabla_{L}$
.

(11) For the split 2-component link , $\nabla_{L}=0$ . $L$

(12) For the Hopf link with linking $number\pm 1$ , $L$ $\nabla_{L}=\pm 1$ .
Let be the Conway polynomial of the 2-bridge link
$\nabla(b_{1}, b_{2}, \ldots, b_{m})$

$D(b_{1}, b_{2}, \ldots, b_{m})$ , $m$ odd (cf. [9, Fig. 2]). Hartley [4, (6.4)] shows that
$\nabla(b_{1}, b_{2}, \ldots, b_{m})$
is an integral polynomial in $f=t_{1}t_{2}+t_{1}^{-1}t_{2}^{-1}$ and
$g=t_{1}t_{2}^{-1}+t_{1}^{-1}t_{2}$
. More precisely we have:
132 T. Kanenobu and T. Sumi

Proposition 6.

$\nabla(b_{1}, b_{2}, \cdots, b_{m})=(1,0)A^{b_{m}}B^{b_{m-1}}\cdots A^{b_{3}}B^{b_{2}}A^{b_{1}}$ $\left(\begin{array}{l}0\\1\end{array}\right)$

where
$A=\left(\begin{array}{ll}g & -1\\1 & 0\end{array}\right)$
and $B=\left(\begin{array}{ll}1 & 0\\f-g & 1\end{array}\right)$
.

Note that , where

$z^{-1}\overline{\nabla}(z)$ $\overline{\nabla}(z)$
is the reduced Conway polynomial
[2, p.340], is obtained from $\nabla(b_{1}, b_{2}, \ldots, b_{m})$ by substituting $f=z^{2}+1$
and $g=2$ .

Proof. Apply (9) and (10) to one of the crossings in the 2-braid
with $2b_{m-1}$ crossings. Then
$\nabla(b_{1}, \ldots, b_{m-2}, b_{m-1}, b_{m})=\nabla(b_{1}, \ldots, b_{m-2}, b_{m-1}-1, b_{m})$

$+(t_{1}-t_{1}^{-1})(t_{2}-t_{2}^{-1})\nabla(b_{1}, \ldots, b_{m-2})\nabla(b_{m})$ .

So by induction on $b_{m-1}$ , we have:

$\nabla(b_{1}, \ldots, b_{m-2}, b_{m-1}, b_{m})=\nabla(b_{1}, \ldots, b_{m-2}+b_{m})$

$+b_{m-1}(t_{1}-t_{1}^{-1})(t_{2}-t_{2}^{-1})\nabla(b_{1}, \ldots, b_{m-2})\nabla(b_{m})$ .

Apply (8) to the 2-braid with $2b_{m}$ crossings. Then

$\nabla(b_{1}, \ldots, b_{m-1}, b_{m})+\nabla(b_{1}, \ldots, b_{m-1}, b_{m}-2)$

$=g\nabla(b_{1}, \ldots, b_{m-1}, b_{m}-1)$ ,

and so we have

$(_{\nabla(b_{1},,b_{m}-1)}^{\nabla(b_{1},,b_{m})}\ldots\cdots)=A$ $\left(\begin{array}{llll}\nabla(b_{1} & \cdots & b_{m} & -1)\\\nabla(b_{1} & \cdots & b_{m} & -2)\end{array}\right)$
.

Then we have

$(_{\nabla(b_{1},,b_{m-3},b_{m-2}+b_{m}-1)}^{\nabla(b_{1},,b_{m-3},b_{m-2}+b_{m})}\ldots\cdots)$

$=A^{b_{m}}(_{\nabla(b_{1},,b_{m-3},b_{m-2}-1)}^{\nabla(b_{1},,b_{m-3},b_{m-2})}\ldots\cdots)$
,

and
$\left(\begin{array}{l}\nabla(b_{m})\\\nabla(b_{m}-1)\end{array}\right)=A^{b_{m}}$ $\left(\begin{array}{l}0\\1\end{array}\right)$

,
 Polynomial Invariants of 2-Bridge Links 133

since $\nabla(0)=0$ by (11) and $\nabla(-1)=1$ by (12). Therefore

$(_{\nabla(b_{1},,b_{m-1},b_{m}-1)}^{\nabla(b_{1},,b_{m-1},b_{m})}\ldots\cdots)$

$=A^{b_{m}}(_{\nabla(b_{1},,b_{m-2}-1)}^{\nabla(b_{1},,b_{m-2})}\ldots\cdots)$

$+b_{m-1}(t_{1}-t_{1}^{-1})(t_{2}-t_{2}^{-1})\nabla(b_{1}, \ldots, b_{m-2})A^{b_{m}}$ $\left(\begin{array}{l}0\\1\end{array}\right)$

$=A^{b_{m}}(_{b_{m-1}(t_{1}}-t^{\frac{1}{1}1})(t_{2}-t_{2}^{-1})$ $01)(_{\nabla(b_{1},,b_{m-2}-1)}^{\nabla(b_{1},,b_{m-2})}\ldots\cdots)$

$=A^{b_{m}}B^{b_{m-1}}(_{\nabla(b_{1},,b_{m-2}-1)}^{\nabla(b_{1},,b_{m-2})}\ldots\cdots)$

$=A^{b_{m}}B^{b_{m-1}}\cdots B^{b_{2}}A^{b_{1}}$ $\left(\begin{array}{l}0\\1\end{array}\right)$

and we have the desired formula.

\S 3. Computational process
From [9, Sect.2, Step 1], we have the set . Let $C(a_{1}, a_{2}, \ldots, a_{k})$ $\mathcal{L}_{n}$

$\in \mathcal{L}_{n}$
and , be the integers obtained from the continued fraction (1).
$p$ $q$

Let
$\frac{p}{q}=2b_{1}+\frac{1}{2b_{2}}+\cdots+\frac{1}{2b_{m}}$

and
$\frac{p}{q-p}=2c_{1}+\frac{1}{2c_{2}}+\cdots+\frac{1}{2c_{\ell}}$ ,

where are odd. If let $L=D(b_{1}, b_{2}, \ldots, b_{m})$ , then

$m$ and $\ell$ $L^{\wedge}=$

. We denote these 2-bridge links by $T(p, q)$ and $T(p,$ $q-$

$D(c_{1}, c_{2}, \ldots, c_{\ell})$

$p)(=T(p, q)^{\wedge})$ . Then $T(p, q)$ is isotopic to either $S(p, q)$ or $S(p, q-p).*$

We first compute the homfly polynomials $P_{L}=P(b_{1},$ , , , $b_{2}$

$\ldots$
$b_{m)}^{\backslash }$ $P_{L^{\wedge}}=$

, the Kauman polynomials $F_{L}=F(b_{1}, b_{2}, \ldots, b_{m})$ , and

$P(c_{1}, c_{2}, \ldots, c_{\ell})$

the Conway polynomials using [9, Propositions $\nabla_{L}=\nabla(b_{1}, b_{2}, \ldots, b_{m})$

1 and 4] and Proposition 6. Then we compute: , , , , , $P_{\overline{L}}$

$ P_{L}\wedge$ $ F_{L}\wedge$ $F_{\overline{L}}$ $ F_{\overline{L}}\wedge$

, , and
$\nabla_{L}\wedge$
, using the following:
$\nabla_{\overline{L}}$ $\nabla_{\overline{L}}\wedge$

$P_{\overline{L}}(v, z)=P_{L}(v^{-1}, z)$ ,

$*Note$ added in proof. T. Kanenobu and Y. Miyazawa proved that $T(p, q)=$
$S(p, q-p)$ .
134 T. Kanenobu and T. Sumi

$F_{L}\wedge(a, z)=a^{4\lambda}F_{L}(a, z)$ ,

$F_{\overline{L}}(a, z)=F_{L}(a^{-1}, z)$ ,
$\nabla_{L^{\wedge}}(t_{1}, t_{2})=-\nabla_{L}(t_{1}, t_{2}^{-1})$
,
$\nabla_{\overline{L}}(t_{1}, t_{2})=-\nabla_{L}(t_{1}, t_{2})$ ,
where $\lambda=-b_{1}-b_{2}-\cdots-b_{m}=c_{1}+c_{2}+\cdots+c_{\ell}$ is the linking number
of . $L$

Next we compute the Jones, $Q$ , and reduced Conway polynomials

by suitable substitutions. Finally we search all the pairs of 2-bridge
links through 20 crossings having the same homfly, Kauman, Jones,
and $Q$ polynomials as in [9, Sect.3, Step 3]. For the Conway and re-
duced Conway polynomials, we examine for the pairs having the same
$Q$ polynomials.

\S 4. Computational results

In Table 2, the three numbers , represent the pair of the

$ p$ $q$ $r$

2-bridge links $\{T(p, q), T(p, r)\}$ sharing the same $Q$ polynomial. If there
is an entry
$ V$
(resp. , ,
$ P$
, ), they also share the same
$ F$ $ A$ $C$

Jones (resp. homfly, Kauman, 2-variable Conway, reduced Conway)

polynomial. We do not list the pairs and having the linking number
$L$ $L^{\wedge}$

zero if they are not contained in Cases 1-5. These links have the same
Kauman polynomial.
The two numbers $ pg$ represent the pair of the 2-bridge links
$\{T(p, q), T(p, q)^{\wedge}\}$ sharing the same homfly and 2-variable Conway poly-

nomials (cf. Fact 4). Note that we do not list the pair sharing only the
same 2-variable Conway or reduced Conway polynomial. The entries $ a$

and indicate that the links are of types A and , respectively.

$ b$ $B$

Table 2
242 -177,87 P 370 153,-207 F
9 crossing 13 crossing
248 109 370 -217,163 F
24 5,11 b 110 19,51 25695 380 137,-167 F
124 39,23 264115 380 -243,213 F
11 crossing
132 25,29 280123
15 crossing
7817,35 132 25,59
14 crossing
8419,25 132 29,59 120 29,19 b
9829,-55P 138 31,43 188 35,59 186 41,83
98-69,43V 162 37,73 P 196 69,-155 V 192 43,61
12847 162 -125,-89 V 196 -127,41 V 228 59,47
196 57,-111 P 196 45,37 234 101,43
12 crossing
196 -139,85 V 220 61 , 39 238 109,75
60 11,19 200 61 252 71,55 A 242 111,-197 V

130130 -73,4757,-47 FaFa 232 101

240 71 , 89
242 65 ,-155
b
V
264 71,49
324 127,-233
324 -197,91
V
V
242 -131,45
252 115,47
252 115,79
P
 Polynomial Invariants of 2-Bridge Links 135

Table 2 (continued)
25247,79648-395,181V 728 215,327 V 602 -439,-411 V
26061,49648-395,-467PA 728 -513,-401 V 616 279,113
294127,-209V722305,-455P 742 303,515 V 636 167,151
294 -167,85 P 722 -417,267 V 742 -439,-227 V 638 135,-525 V
304 79 ,63 748 159,317 638 -503,113 V
16 crossing
308 65,87 972 271,-593 V 644 141,153
308 83,97 25681,49V 972 -701,379 V 672 209,239
324 73,145 $P$
256 -175,-207 $V$ 1016397,-651 A 672 211,197
324 -251,-179 V 296 47, 137 1032379,-661A 676 209 ,-519 P
336 89,103316 59,99 1130437,-467 Fa 676 -467,157 V
338 79,-233 V 322 71,57 1130 -693,467 Fa 686 181,209 V
338 -259,105 P 324 77,61 686 -505,-477 V
17 crossing
350 93,-243 V 352 163,291 V 714 155,127
350 -257, 107 V 352 -189,-61 V 240107,53 720 317,133
352 161354 73,163 24655,79 722 151,-533 P
352161,63374167,303V 33853,157P 722 -571,189 V
368169374-207,-71V 338 -285,-181 V 726 263 ,-529 P
37469,169378 137,67 342 53,109 726 -463,197 V
380103,87396 73,91 370 89,59 728 333
384143402 83, 113 380 119,71 728333,229
38885,8940673,143 380 119,61 736337
392139,83V456107,125 380 71,61 738137,331
392139,-309PA462127,83 388 73,93 742233,339
392 -253,83 $PA$ 462 127,97 390 $f01,61$ 744325
392 -253,-309 V 462 83,97 392 113,-223 P 748141,163
400 121484 109,197 V 392 -279,169 V 752345
400 183484 -375,-287 V 406 187,-277 V 754199,225
402125,143506137,93 406 -219,129 V 756235,163P
40693,121 $P$
508 135,119 A 462 79, 101 756-521,-593V
406-313,-285V510107,233 464 101,-379 V 760 333
408121,127516121,223 464 -363,85 V 760 349
416191516121,113A 472 221 764203,179
418111,89516223,113 476 151,-257 V 770137,277
434177,115564245,131 476 -325,219 V 772181,177
448137,201572125,333V 484 221 ,-395 V 774 349 ,-167 P
450 97,133 P 572 -447,-239 V 484 -263,89 P 774 -425,607 V
450 -353,-317 $V$ 572 155,131 486 217,109 P 776 355
468 101,-211 $P$
576 107,125 486 -269,-377 V 776 339
468 -367,257 $V$ 588 209,-463 $V$ 488 229 784 359
476 109,277 P 588 -379,125 V 496 157,405 V 784 279,167 V
476 -367,-199 V 594 163,-413 V 496 -339,-91 V 784 279,-617 PA
484 131,-309 $V$ 594 -431, 181 $V$ 512 191 784 -505,167 PA
$484$ -353,175 $P$
620 253,-347 $F$
512 161,97 V 784-505,-617V
488 213620 -367,273 F 512 161,-415 PA 786163,361
494105,131624145,175 512 -351,97 PA 792347
504221630193,-227V 512 -351,-415 V 798251223
504 221,-115 P 630 -437,403 V 528 163,427 V 798143,283
504 221,389 V 630 193,277 V 528 -365,-101 V 800367
504 -283,-115 P 630 -437,-353 V 536 251 808 371
504 -283,389 V 630 227,-277 F 53693,85 814 173 ,-663 V
504181,197 Ab630 -403,353 $F$
552259 814 -641,151 V
512223638139,371V 560 107,-437 V 834 233,173
520 227638 -499,-267 V 560 -453, 123 V 840 379,181 b
522 119,155 P 644 289,473 V 564 179,197 846 193,-371 P
522 -403,-367 V 644 -355,-171 V 570 181169 846 -653,475 V
536 235666 241,-203 V 578 203,-477 P 854 153,181
574 181159 P 666 -425,463 V 578 -375,101 V 864 269,197 P
574 -443,-415 V 676 287,183 V 594 271,107 864 -595,-667 V
578 169,237 V 676 -389,-493 V 598 113,425 V 868 353,-639 V
578 -409,-341 $P$
702 197,-487 $V$ 598 -485,-173 V 868 -515,229 V
648 253, 181 PA 702 -505,215 V 600 181 874 245 ,-675 V
648 253,-467 V 704 149,299 602 163,191 V 874 -629,199 V
136 T. Kanenobu and T. Sumi

Table 2 (continued)
894247,187 1102-851,309 V 44483,139 900 -691,389 V
896205,261P 1104257,479 452109,85 900 247,-617 V
896-691,-635V 1120 297,457 532109,137 900 -653,283 V
924 415,283 1122245,-811 V 544 93,189 936 295,-329 V
930 421,601 C 1122-877,311 V 558 131 ,-301 V 936 -641,607 V
936 205,277 P 1134509,347 558 -427,257 V 942 287,221
936 205,-659 V 1140 241 ,301 576 119,263 V 948 289,199
936 -731,277 V 1144309,243 576 119,-313 V 952 345,-775 V
936 -731,-659 V 1156339,475 V 576 -457,263 V 952 -607,177 V
938 409,275 1156-817,-681 P 576 -457,-313 V 952 205,171
942 431,197 1164271,325 684 145,107 956 227,251
942 203,215 1190321,349 686 141,-531 V 964 229,221
950 199,249 1216 257,321 686 -545,155 V 976 181,213 V
954 427,-209 P 1242379,343 PA 688 123,307 976 -795,-763 V
954 -527,745 V 1242-863,-899 PA 702 163,-557 V 980 209 ,-631 V
966 409,745 V 1250451,551 F 702 -539,145 V 980 -771,349 V
966 -557,-221 P 1250-799,-699 PF 704 161 , 129 V 994 275,-789 V
968 395,219 PA 1278391,-461 PA 704 -543,-575 V 994 -719,205 V
968 395,-749 V 1278-887,817 PA 704 127,193 996 209,455
968 -573,219 V 1292295,-929 V 720 169,151 1002235 ,433
968 -573,-749 PA 1292-997,363 V 732 337, 151 1008 187,-653 V
990 223,-437 P 1296 505,361 PA 736 135,503 V 1008-821,355 V
990 -767,553 V 1296 505,-935 V 736 -601,-233 V 1008 187,691 V
992 447,639 C 1296-791,361 V 738 173,-583 V 1008-821,-317 V
994 303,-549 V 1296-791,-935 PA 738 -565,155 V 1008355,-317 V
994 -691,445 V 1298349,-1015 V 748 203,137 1008-653,691 V
994 431,767 V 1298-949,283 V 760 159,121 1010313,293 PFAa
994 -563,-227 P 1314401,-475 PA 764 183,199 A 1010 -697,-293 PFAa
996 227,449 1314-913,839 PA 768 241,145 V 1020239,271 A
996 275,233 1316543,355 768 -527,-623 V 1024225,289 V
1008227,299 P 1330389,579 770 159 ,-541 V 1024225,-735 V
1008-781,-709 V 1350377,413 PA 770 -611,229 V 1024-799,289 V
1022 285,313 1350-973,-937 PA 772 185,169 A 1024-799 ,-735 V
1022 285,-751 V 1352365,573 V 780 161,239 1032 185,271
1022-737,271 V 1352365,-779 PA 782 135,169 1040 197,717 V
1022 285,299 1352-987,573 PA 784 141,-475 V 1040-843 ,-323 V
1022313,271 1352-987,-779 V 784 -643,309 V 1044329 ,-751 V
1022313,-723 V 1372405,-995 V 800 153,553 V 1044-715,293 V
1022-709,299V 1372-967,377 V 800 -647,-247 V 1062233,197
1022271,299 1444533,-835 V 812 151,-633 V 1064277,221
1024447 1444-911,609 P 812 -661,179 V 1072235,203 V
1026215,269 1456 393,407 832 191,159 V 1072-837,-869 V
1034285,219 1458 541,433 P 832 -641,-673 V 1078493,885 V
1036317,303 1458-917,-1025 V 836 217,-543 V 1078-585,-193 V
1036 317, 275 1528 549,-931 A 836 -619,293 V 1078475 ,-225 V
1036317,-747 V 1544555,571 A 858 301,-635 V 1078-603,853 V
1036-719,289 V 1682637,-1219V 858 -557,223 V 1100203,603 V
1036303,-761 V 1682-1045,463P 858 181,233 1100-897,-497 V
1036-733,275 V 1784653 868 179,-381 V 1102235,293
1036303,289 1800659 868 -689,487 V 1106 197,239
1036 275,289 1922805,-1179V 870 353,-487 F 1118245 ,-787 V
1056241,463 b 1922-1117,743P 870 -517,383 F 1118-873,331 V
1056 247,457 2024 741 880 317,-387 F 1148241,-935 V
1058459,-737 P 2040781,749Ab 880 -563,493 F 1148-907,213 V
1058-599,321 V 2056755 882 199,163 1156307,-645 V
1058231,415 V 2296843 882 205,-479 V 1156-849,511 V
1058-827,-643 P 2312885 882 -677,403 V 1158269,503
1064299,243 P 896 375,-185 V 1162263 ,-409 V
18 crossing 896 375,711 V 1162-899,753 V
1064299,-821 V
1064-765,243 V 21029,41 896 -521 ,-185 V 1162417,207
1064-765,-821 V 400139,-341V 896 -521,711 V 1164515,527
1102 251 ,-793 V 400 -261,59 V 900 209,-511 V 1164433,343
 Polynomial Invariants of 2-Bridge Links 137

Table 2 (continued)
1166 303,-841 V 1484409,-1159 V 1998 557,-1387 V 754353,111
1166-863,325 V 1484-1075,325 V 1998-1441,611 V 800119,279V
1188 211,545 1488439,409 2014845,-1131 V 800 119,-521 PA
1190431,-269 V 1536689,335 2014-1169,883 V 800 -681,279 PA
1190-759,921 V 1536 359,425 2028 859,547 V 800 -681,-521 V
1190423,213 1542349,679 2028-1169,-1481V 800 241
1218 559,253 1582 345,-1111 V 2040797,-1307 A 808167,127
1218373,283 1582-1237,471 V 2056 7391288 A 816139,173
1232 229,555 1590473,587 2116 873,-1335 V 836151,381
1246 223,265 1596 6911378 2116-1243,781 V 840 193,263
1260 263,-913 V 16206311188 V 2142 593,-1675 V 850 133,303
1260-997,347 V 1620-989,451 V 2142-1549,467 V 858 389,157
1266347,377 1634347,433 2198 957,649 V 882 211,-713 V
1266 263,353 1634617,579 2198-1241,-1549V 882 -671,169 P
1272 571,277 1644611,485 2210863,-837 Fa 884 139,309
1274537,269 1644 713,383 2210 -1347,837 Fa 892 215,231
1274279,-561 V 1650463,373 2212933,1565 V 900 197,217
1274-995,713 V 1666377,-1303 V 2212-1279,-647 V 928 163,-733 V
1278 343,361 1666-1289,363 V 2318 1017,-1339 V 928 -765,195 V
1288269,381 V 1666699 ,-365 V 2318-1301,979 V 936 149,427
1288269,-907 V 1666-967, 1804 V 2500 1051,-1549 F 942 329,299
1288-1019,381 V 1690 759,359 2500-1449,951 F 944 221 ,-771 V
1288-1019,-907 V 1704397,475 2546935 ,-1117 V 944 -723,173 V
1296397,-827 V 1716 727,703 2546-1611,1429 V 948 295 ,301
1296-899,469 V 1758 523,493 2610719,701 968 351,-705 P
1316 571,403 V 1782389,1037 V 30641133,-1899A 968 -617,263 V
1316-745,-913 V 1782-1393,-745 V 30801131,-1941A 976 457
1320 371,349 1786661,-783 V 976 179,667 V
19 crossing
1320371,389 1786-1125,1003 V 976 -797,-309 V
1320349,389 1804477,-1163 V 294131,65 984 461
1326 277,367 1804-1327,641 V 30097,67 988173,211
1330353,733 V 1804391,479 V 42683,59 992173,-787V
1330-977,-597 V 1804-1413,-1325V 438 67,61 992 -819,205 V
1342377,-987 V 1826679,-1313 V 450 61 ,-239 P 1000469
1342-965,355 V 1826-1147,513 V 450 -389,211 V 1000437
1342 379,291 V 1846391,495 472 215,73 1008473
1342-963,-1051 V 1848 773,-403 V 490 211,-349 V 1014235,-701V
1374311,377 1848 773,1445 V 490 -279,141 P 1014-779,313P
1374629,287 1848-1075,-403 V 508159,95 1014161,265
1378 513,293 1848-1075,1445 V 516125,97 1016445
1380 379,301 1860 1081,841 C 592281 1024 193,321 V
1414635,411 V 1870 763,-1097 F 608 277,-107 V 1024 193,-703 PA
1414-779,-1003 V 1870-1107,773 F 608 -331,501 V 1024-831,321 PA
1428401,311 1876823,-409 V 608 289 1024-831,-703 V
1428 583,991 V 1876-1053,1467 V 620149,99 1032451
1428-845,-437 V 1880 737,-767 F 630101,121 1040487
1430607,303 1380-1143,1113 F 640239 1048459
1434529,427 1890523,-1313 V 648289,145P 1048491
1452 329,593 V 1890-1367,577 V 648-359,-503V 1050487,163
1452-1123,-859 V 1904557,837 V 666101,137 1058183,275P
1456431,319 V 1904-1347,-1067V 672319 1058-875,-783V
1456431,-1137 V 19188361188 V 67699,177 1062229,337 P
1456-1025,319 V 1918-1083,-535 V 676105,313P 1062-833,-725 V
1456-1025,-11374 1926517,743 676-571,-363V 1064337,167
1462 575,-309 V 1926 695,-589 V 688 327 1064337,489
1462-887, 1166 V 1926-1231,1337 V 722 115 ,-493 V 1064167,489
1462 607, 1014 V 1936 747,-1365 V 722 -607,229 P 1064499
1462-855 ,-447 V 1936-1189,571 V 726 133,-395 P 1072503
1470 617,-307 V 1962 599,-709 V 726 -593,331 V 1072 205,741 V
1470-853,1163V 1962-1363,1253V 726251,233 1072-867,-331 V
1474313,625 1962527,769 728113,153 1078501 ,893 V
1482335,653 1984895,1151C 732235,217 .
1078 -577,-185 $V$
138 T. Kanenobu and T. Sumi

Table 2 (continued)
1080233,-487P 1364245,-1075 V 1554355,439P 17084811528
1080-847,593V 1364-111288 V 1554-1199,-1115V 17084811488
1092 491,251 1386 317,401 P 1554277,557 1708481,-1255V
1100 189,589 V 1386-1069,-985 V 1562337,359 1708-1227,453 V
1100-911,-511 V 1394487,241 1562 295,863 V 1708523,-1213 V
1102 175,347 14143811826 1562-1267,-699 V 1708-1185,495 V
1104211,-845 V 1422 295,331 V 1564 703,473 1708523,453
1104-893,259 V 1422-1127,-1091V 1566487,343 P 1708495,453
1120209,239b 1422 421,313 P 1566-1079,-1223V 1708521,451
1120257,513 1422-1001,-1109V 1568281 ,617 PA 1710401 ,781
1120513 1426 255,301 1568 281,-951 V 1722457,527
1136 521 1440643,-317 P 1568-1287,617 V 1722457,-1223 V
1148275,299 1440-797, 1120 V 1568-1287,-951 PA 1722-1265,499 V
1152239 1444379,-1141 V 1568487,-297 V 1722457 ,485
1152 239,527 V 1444-1065,303 P 1568487, 1276 V 1722527,499
1152239,-625 V 1456317,-1123 V 1568-1081,-297 V 1722527,-1237 V
1152-913,527 V 1456-1139,333 V 1568-1081,1271 V 1722-1195,485 V
1152-913,-625 V 1456 317,-619 V 1576 723 1722499,485
1152 527 1456-1139,837 V 1582563,283 17343191458
1136203,-749 V 1456333,837 P 1582419,363 1734713,509 V
1156-953,407 P 1456-1123,-619 V 1584709,1285V 1734-1021,-1225P
1156 265,277 1458 593,269 V 1584-875,-299 V 1748367,459
1156 531,251 1458-865,-1189 P 1584 709,-347 P 1758487,367
1162475,267 1458 305,341 V 1584-875,1237 V 1758385,787
1162 341,369 V 1458-1153,-11176 1584 299,-1237 V 1760373,483
1162-821,-793V 1472337,273 V 1584-1285,347 V 1768315,485
1162337,365 1472 337,-1199 V 1586 329,277 1778545,405
1168535 1472-1135,273 V 1586331,733 1786407,467
1176251,419 V 1472-1135,-1199V 1600303,367 V 1792389,333 V
1176251,-757 PA 1476653,-331 P 1600303,-1233 V 1792-1403,-1459V
1176-925,419 PA 1476-823,1136 V 1600-1297,367 V 1800419,779 V
1176-925,-757V 14846411148 V 1600-1297,-1233V 1300419 ,-1021 PA
1178245,207 1484-843,-339 P 1602 733,373 V 1800-1381,779 PA
1184543 1484471,415 V 1602-869,-1229 V 1300-1381 ,-1021V
1188271,-521P 1484-1013,-1069V 1610 507,367 1804767,381
1188-917,667V 1488 277,-1163 V 1610493,723 1806377,827
1190377,547 1488-1211,325 V 1616371,-1261 V 1806479,737
1206383,275 P 1494335,443 P 1616-1245,355 V 1812397,391
1206-823,-931 V 1494-1159,-1051V 16244711708 1812553,379
1206 551,283 1496 533,269 1628309,727 1820817,557
1206 223,533 1496685 1638 349,293 1820543,-1433 V
1218 557,383 1498267,323 1644341 , 503 1820-1277,387 V
1224553,281 1504279,-473V 1650343,757 1826773,389
1224547,227 1504279,1080 V 1652 379,505 1848491,421
1232 563,387 1504-1225,-473 V 1656 373,-731 P 1360401,419
1276 335,303 1504-1225,1031 V 1656 373,925 V 1860389,851
1278 587,299 V 1512479,409 1656-1283,-731 V 1862519,491 V
1278-691,-979 V 1518 703,263 1656-1283,925 V 1862-1343,-1371V
1284301,305 1520411,1171 V 1672 299,365 1862 519,393
1288 $363_{\rangle}405$ 1520-1109,-349 V 1672439,351 1862519,421
1288409,-695 V 1520477,1232 V 1674521,377 P 1862491,393
1288-879,593 V 1520-1043,-283 V 1674-1153,-1297V 1862491 ,421
1288345,-931 V 1528 701 1680 733 ,-1187 V 1862491,421 PA
1288-953,367 V 1536671 1680-947,493 V 1862-1469,-1441PA
1300383,583 V 1540283,-1213 V 1682 753,521 V 1870507,397
1300-917,-717 V 1540-1257,327 V 1682-929,-1161 P 1872281,617 PA
1312229,-603 V 1544707 1682 737,365 187212881386PA
1312-1083,709 V 1550461,411 1692 355,731 1908671,-601 PA
1330607,303 1552293,1069V 1698353,473 1908-1237, 1300 PA
1358625,-927 V 1552-1259,-483 V 1700467,297 1914431,787
1358-733,431V 1552 421,1184 V 1702363,-1385 V 1914431,863
1360613,237 1552-1131,-355 V 1702-1339,317 V 1914787,863
 Polynomial Invariants of 2-Bridge Links 139

Table 2 (continued)
1922 557,433 V 2108951,375 2358697,733P 2632-1907,557 V
1922-1365 ,-1489P 2114873,485 2358-1661,-1625V 2632-1907,-2075V
1922 683,-1053 V 2116919,-1473P 2366 851,1831 PA 2646769 ,-1919 V
1922-1239,869 P 2116-1197,643 V 2366-1515,-535 PA 2646-1877,727 V
1924515,-1357 V 2116 829,461 V 2398 519,1007 2646737,809 PA
1924-1409,567 V 2116-1287,-1655P 2412 751 ,-857 P 2646-1909,-1837PA
1932359,695 V 2128 395,451 V 2412-1661,1555 V 2660737,793 P
1932-1573 ,-1237V 2128-1733,-1677V 2436 743,1091 2660-1923,-1867V
1936439,791 PA 2132 577,-1503 V 2448 761 ,-871 P 2660787,563 P
1936439,-1145 V 2132-1555,629 V 2448 761,1577 V 2660-1873,-2097V
1936-1497,791 V 2136481,487 2448-1687,-871 V 2676 1201,583
1936-1497,791145PA 2136 925,499 2448-1687, 1577 V 2678 1107,-1961 V
1938409,511 2142 871 ,-1577 V 2450 531,1371 P 2678-1571,717 V
1944 757,541 PA 2142-1271,565 V 2450-1919,-1079V 2680821
1944 757,-1403 V 2156 601,657 V 2450 687,-1777 V 2684581,1127
1944-1757,541 V 2156-1555,-1499V 2450-1763,673 V 2686727,795V
1944-1187,-1403PA 2166 913,-1367 P 2450 911,-1889 V 2686-1959,-1891P
1946 579,411 P 2166-1253,799 V 2450-1539,561 P 2704729,1145 V
1946-1367,-1535V 2170647,-1213 P 2464667,723 V 2704729 ,-1559 PA
1946593,-1075 V 2170-1523,957 V 2464-1797,-1741V 2704-1975,1145 PA
1946-1353,871 V 2170883,573 2464 1103,751 V 2704-1975,-1559V
1958449,797 2178925,-1715 P 2464 1103,-1713 V 2724809,587
1958427,361 2178-1253,463 V 2464-1361,751 V 2724761,803
1960573,517 V 2184509,947 2464-1361,-1713V 2728 1587,-1525 C
1960 573,-1443 V 2198 907,593 2482 725,-1723 V 2730739,-1601 V
1960-1387,517 V 2210467,597 2482-1757,759 V 2730-1991,1129 V
1960-1387,-1443V 2212 957,641 2494659,975 2738591,1035 P
1968449,431 2238 971,521 2496 673,737 2738-2147,-1703V
1976451,413 2240473,-983 V 2500 1101,901 F 2738815,-1997 V
1976 535,-1545 V 2240473,1257 P 2500-1399,-1599PF 2738-1923,741 P
1976-1441,431 V 2240-1767,-983 V 2502 779,743 P 2744811,755 P
1978613,-1107 V 2240-1767, 1257 V 2502-1723,-1759V 27448111938 V
1978-1365,871 V 2244625,523 2508 679,-1961 V 2744-1933,755 V
1980 623,-697 PA 2254687,659 V 2508-1829,547 V 2744-1933,-1989V
1980-1357,1283 PA 2254-1567,-1595V 2530 899,711 2738 1237,813 V
1984415,353 2254993,1777 V 2538 775,703 PA 2756-1519,-1943V
1988 705,369 V 2254-1261,-477 P 2538-1763,-1835PA 2738587,421 P
1988-1283,-1619V 2268 925,517 2546 673,581 2758-2161,-1937V
1990617,-1383 PFA 2268 949,1938 V 2546 689,-1991 V 2758625,1605 P
1990-1373,607 PFA 2268-1319,-479 P 2546-1857,555 V 2758-2133,-1153V
1992 587,421 2296 $947_{\backslash }523$
2552 917,-1555 A 2772 1031,733
2010 583,1387 PFA 2298 703,535 2552 917,675 2778 767,827
2010-1427,-623 PFA 2310 521,499 2552-1555,675 2782 1179,589
2014435,-1473 V 2312613,1021 PA 2562 709,541 P 2784769,607
2014-1539,561 V 2312613 ,-1291 V 2562-1853,-2021V 27846012828
2016563,635 PA 2312-1699,1021 V 2568 757,923 2786 1215,817
2016-1453,-1381PA 2312-1699,-1291PA 2568 757,955 2794993,751
2022 905,461 2314645,-1695 V 2568923,955 A 2800821,1221
2022 617,473 2314-1669,619 V 2574787,-929 PA 2828613,837 P
2028599,-1585 V 2320913,607 2574-1787, 1645 PA 2828-2215,-1991V
2028-1429,443 V 2332 641,-1823 V 2592 793,937 PA 2332 791 ,617
2048577,449 PA 2332-1691,509 V 2592 793,-1655 V 2842797,-2003 V
2048 577,-1599 V 2336989,1958 2592-1799,937 V 2842-2045,839 V
2048-14712448 V 2338533,2048 2592-1799,-1655PA 2842643,615 PA
2048-1471,-1599PA 2338697,-1307 P 2600 567,-1097 V 2842-2199,-2227PA
2072601,559 2338-1641,1031 V 2600-2033,1503 V 2352 1841,1988 C
2074439,371 2338659,631 2610 943,-797 PA 2856835,1243 V
2076485,575 2338829,1809 PA 2610-1667, 1813 PA 2856835 ,-1613 V
2080453,-1523 V 2338-1509,-529 PA 2622 1183,565 2856-2021,1243 V
2080-1627,557V 2352 533,701 P 2626 555,1109 2356-2021,-1613P
2082901,487 2352-1819,-1651V 2632 725,557 P 2856619,787 P
2082919,469 2356537,499 2632 725,-2075 V 2856619,-2069 V
140 T. Kanenobu and T. Sumi

Table 2 (continued)
2856-2237,787 V 3232 1407 836153,265
2856-2237,-2069V 3234 1357,2533 V 3850-2257,-2189PF 858155,131 A
2886 1301,623 3234-1877,-701 P 38701039,1489 884415,129
2886653,797 3248 1425,-703 V 3872 1495,1143 V 890 131 ,409
2892 853,799 3248 1425,2545 V 3872 1495,-2729 PA 918 157 ,-659 V
2914 1315,1879 C 3248-1823,-703 P 3872-2377, 1243 PA 918 -7613268 V
2916865,1081 P 3248-1823,2545 V 3872-2377,-2729V 946 259,171
2916-2051,-1835V 3268917,-2523 V 39061049,1525 956 183,175
2924 1273,865 V 3268-2351,745 V 39361729 990 1813268
29241651,-2059P 3270973,913 3952 1737 1020247,263 A
2924 1285,619 3298 1009,-2323 V 3984 1751 1024 159,415 V
2926 1283,2263 V 3298-2289,975 V 4064 1489 1024-865,-609 V
2926-1643 ,-663 P 3332 755,923 P 4088 1617 1028249,225 A
2940869,671 3332-2577,-2409V 40881517,1549 A 1064339,899 V
2940641,1319 33461467,-2357V 40881549 1064-725,-165 V
2946877, 1737 3346-1879,989P 4088 1503 10784814208
2994677, 1319 3352 1243 4104 1507 1084343,207
2996809,683 3360991 41041507,1523 A 1092337,209
3014895,807 V 3362901 , 1729 V 41041523 1100509,191
3014-2119 ,-2207P 3362-2461,-2133P 4120 1219 1102345,925 V
3038 1275,2395 V 3362 1311,1475 P 4192 1825 1102-757,-177 V
3038-1763 ,-643 P 3362-2051,-1887V 4208 1333 1110 169,229
3038883,-2141 V 3362 985 ,-1967 P 4232 1563,1747 V 1110 169,511
3038-2155,897 V 3362-2377, 1397 V 4232 1563,-2485 PA 1110229,511
3038 1761,-1699 C 3364927,-2089 P 4232-2669,1747 PA 1118 165,295
3040853,-2347 V 3364-2437, 1243 V 4232-2669,-2485V 1120453 ,-627 F
3040-2187,693 V 33741423,941 4240 1847 1120-667,493 F
3042 1327,-1793 V 33881475,991 42461261,1173V 1130407,-497 F
3042-1715,1249 P 34401049 4246-2985,-3073P 1130-723,633 F
3048899, 1333 34441411,1459 4328 1605 1158365,239
3052 853,895 34561519 4336 1591 1184217,281 V
3054691,1345 3458971,-2669 V 43441595 1184-967,-903 V
3054697, 3388 3458-2487,789V 4360 1651 1200419,-1021 V
3058663, 1975 V 34681421,1433 44181693,-3195P 1200-781 , 179 V
30582886 ,-1283V 35761309 4418-2725,1223V 1206331 ,-821 V
3074 1105,909 35861065,2297V 4464 1961 1206-875,385 V
3078833,-2407 V 3586-2521,-1289P 4480 1969 1216249,313
3078-2245,671 V 3592 1333 4496 1975 1216385,993 V
3080901,859 3600 1319 45541223,1259 1216-831 ,-223 V
3094863,-2257 V 36501609,1509F 46001749,1701 Ab 1240567,193
3094-2231,837 V 3650-2041,-2141PF 46001707 1242431,-397 V
3094863,-2189 V 36741357,2533P 4616 1749 1242-811 , 845 V
3094-2231,905 V 3674-2323,-109IV 4624 1769 1248 199,329
3094837,905 V 3696 1609 47362063 1256509,195
3094-2257 ,-2189P 36981547,1031P 48021863,-3037V 1258327,191 Va
3102 707, 1961 3698-2151,-2667V 4802-2939,1765P 1258-931,-191 Va
3102 1145,923 3700 1329,-1631 F 4888 1851 1276219,241
3122 845,873 V 3700-2371,2069 PF 50002101,1901PFA 1280401,241 V
3122-2277,-2249V 3710 1027,803 P 50002101,-3099F 1280-879,-1039 V
3122859,-1817 P 3710-2683 ,-2907V 5000-2899,1401F 1292593,-223 V
3122-2263,1305 V 3712 1617 5000-2899,-3099PFA 1292-699,1069 V
3162 715,883 3728 1623 50242207 1304205,531
3172 841,-2487 V 38001669,-1371F 1308607,269
20 crossing 1340323,347
3172-23312886V 38001669,2429F
3198677,859 3800-2131,-137IF 484 67, 155 V 1344415,1087 V
3200879 3800-2131,2429 PF 484 -417,-329 V 1344-929 ,-257 V
3216733,955 3816 1397 572 179,107 1348309,325
32201347,2523V 3824 1401 580 141 , 109 1350431 ,-469 V
3220-1873,-697 P 3832 1421 610 159,89 1350-919,881 V
3222865,901 38441487,-2233P 642 125 , 89 1368253,307
3230737,-2323 V 3844-2357,1611V 654 103,91 1372293 ,-883 V
3230-2493,907 V 38481427 744 115,131 1372-1079,489 V
 Polynomial Invariants of 2-Bridge Links 141

Table 2 (continued)
1376 263,327 V 1778753,-367 V 1888-1403,469 V 21424611566 V
1376-1113,-1049V 1778-1025,1411 V 2000 373,437 V 2142-1681,587 V
1386443,-997 V 1802 335,505 2000-1627,-1563V 2144503,471 V
1386-943,389 V 1804823,1479 V 2000 371 , 1376 V 2144503,-1673 V
1390 211,361 1804-981,-325 V 2000-1629,-629 V 2144-1641,471 V
1406 221,297 1812 571,421 2002 523,607 P 2144-16412178V
1422 293 ,-1003 V 1816 317,285 2002-1479,-1395V 2156457,915
1422-1129,419 V 1826827,1482 V 2002 613,865 V 2166493,-1508 V
1428 335,449 1826-999,-335 V 2002-1389,9911376 2166-1673,607 V
1434451,301 1832 841 , 287 2016 535 ,-473 V 2178455 ,-1888 V
1444341,645 V 1836 793 ,-431 V 2016 535,1543 V 2178-1723,509 V
1444-1103,-799 V 1836-1043, 1406 V 2016-1481 ,-473 V 2178511,457
1444265,417 V 1836 379 ,-1421 V 2016-1481,1543 V 21785111886 V
1444-1179 ,-1027V 1836-1457,415 V 2016 535,-905 V 2178-1667,493 V
1482 677,235 1854565 ,-1271 V 2016 535,1111 V 2178511 ,475
1482677,311 1854-1289,583 V 2016-1481,-905 V 2178457,493
1482 235,311 1854853 ,-383 V 2016-1481,1111 V 2178457,-1508 V
1488349,643 1854-1001,1471 V 2016473,905 V 2178-1721,475 V
1488313,233 1862 333,1313 V 2016473,-1111 V 2178493,475
1508267,-661 V 1862-1529,-549 V 2016-1543,905 V 21904811978
1508-1241 ,847 V 1876 527,-1489 V 2016-1543,-1111V 2190457,607
1532 367,399 A 1876-1349,387 V 2030433,363 2196497,479
1540369,361 A 1886 335,499 V 2032 637,1653 V 2196497,-1735 V
1548 713,-319 V 1886-1551,-1387V 2032-1395,-379 V 2196-1699,461 V
1548-835,1229 V 1888331,395 V 2034929,-427 V 2196497,515
1548341,287 1888-1557,-1493V 2034-1105,1607 V 2196479 ,461
1554289,275 1890337,407 2034623 ,-1393 V 2196479 ,-1681 V
1562271,723 1890863,-397 V 2034-1411,641 V 2196-1717,515 V
1564245,279 1390-1027, 1482 V 2050443,607 V 2196461,515
1566 341,-1243 V 1904333,299 2050-1607,-1443V 2200779,581
1386-1225,323 V 1924865,-695 V 2052431 ,-1441 V 2200933,467
1568275,851 V 1924-1059,1229 V 2052-1621,611 V 2204657,387
1568-1293,-717 V 1926 677,-607 V 2054929,449 2204933,1009
1576247,327 1926-1249,1319 V 2058631,547 V 2222599,-1645 V
1584301,707 1836 331 ,419 V 2058-1427,-1511V 2222-1623,577 V
1586713,-899 V 1936-1605,-1517V 2064643,1675 V 2238 1025,467
1586-873,687 V 1946429,359 2064-1421,-389 V 2238629,695
1586419,379 1946361,767 2068555,731 V 2240513 ,417 V
1598 733 ,-287 V 1946347,697 2068-1513,-1337V 2240-1727,-1823V
1598-865,1311 V 1950 581,529 2076 569,647 2254 1021,685 V
1600281,681 V 1952 3411406 V 2030 553,-487 V 2254-1233,-1569V
1600-1319 ,-919 V 1952-1611,-1547V 2030553,1593 V 2266609,389 V
1652 341 ,-1171 V 1952425,457 V 2080-1527,-487 V 2266-1657,-1877V
1652-1311 ,481 V 1952425,-1495 V 2030-1527, 1598 V 2268883 ,-1637 V
1692383,-913 V 1952-1527,457 V 2082 955,433 2268-1385,631 V
1692-1309,779 V 1952-1527,-1495V 2080359,579 V 2272397,1229 V
1710353,-787 V 1862691,-617 V 2080-1731,-151]V 2272-1875,-1043V
1710-1357,923 V 1962-1271,1345 V 2096459,395 V 2272423,1559 V
1712 299,771 1968 365,461 V 2096-1637,-1701V 2272-1849,-713 V
1728 791,359 V 1968-1603,-1507V 2096651,1699 V 2278399,801
1728791,-1369 V 1972 557,-1619 V 2096-1445,-397 V 2282827,421
1728-937,359 V 1972-1415,353 V 2100607,943 V 2288525 ,931
1728-937,-1369 V 1974367,773 2100-1493,-11576 2288809 ,-647 V
1734373,-1259 V 1974613,703 2112 595,485 2288809,1641 V
1734-13612076 V 1974613,353 2114647,1386 V 2288-1479,-647 V
1748515,1435 V 1974703,353 2114-1467,-863 V 2288-1479,1641 V
1748-1233 ,-313 V 1978 537,365 2128635,373 2292 515,629
1750361,-759 V 1980409,-1391 V 2128499,403 V 2298505,631
1750-1389,991 V 1980-1571,589 V 2128-1629,-1725V 2304529,625 V
1752407,761 1998467,-1081 V 2130443,593 2304-1775,-1679V
1764 799,463 V 1998-1531,917 V 2142 445,-1643 V 2304535,679 V
1764-965,-1301 V 1998595,-1529 V 2142-1697,499 V 23045351826 V
142 T. Kanenobu and T. Sumi

Table 2 (continued)
2304-17692678 V 2530-1643,657 V 2744-2171,-1163V 2882565,829 V
2304-1769 ,-1625V 2530669,449 V 2744-2171,1581 V 2882-2427,-2163V
2322 685,-1835 V 2530-1861,-2081V 2744601 ,-1191 V 2882653,685 V
2322-1637,487 V 2534711,-2005 V 2744-2143,1553 V 2882-2339,-2307V
2322 541,-1259 V 2534-1823,529 V 2744839,727 V 2882653,1741 V
2322-1781,1063 V 2538 1097,-595 V 2744-1905,-2017V 2882-2339,-125IV
2336 441,1609 V 2538-1441,1943 V 2754 1243,-593 V 2882685,1741 V
2336-1895,-727 V 2546677,543 2754-1511,2161 V 2882-2307,-1251V
2340 529,-1271 V 2550 703,533 2758 579,509 2994653,1343
2340-1811 , 3088 V 2552 895,-1889 V 2760859,509 3000679,1321
2352 491 ,-1021 V 2552-1657,663 V 2772 599,-2257 V 3010873,2077 PFA
2352-1861 , 1331 V 2552453,717 V 2772-2173,515 V 3010-2137,-933 PFA
2352 733,835 2552453,-1835 V 2772 599,-1993 V 3014 1117,2213 V
2352 421,-1427 V 2552-2099,717 V 2772-2173,779 V 3014-1897,-801 V
2352-1931 ,925 V 2552-2099,-1835V 2772491,601 3014 1239,679
2356 1039,-1393 V 2352 575,1049 2772 515,779 V 3038687,-2393 V
2356-1317,963 V 2562 785,755 2772-2257,-1993V 3038-2351,645 V
2358 553,-1019 V 2562 745 ,-2027 V 2778 863,989 3042655 ,-2189 V
2358-1805,1339 V 2562-1817,535 V 2778 605,851 3042-2387,853 V
2366439,425 2562 563,1145 2782 1257,829 3042 707 ,-2137 V
2368447,543 V 2568 577,1135 2784853,1003 3042-2335,905 V
2368-1921,-1825V 2616 725,611 2784775,649 3048689,1343
2376 701,557 V 2616 815,929 2784 519,739 V 3054931,901
2376 701,-1819 V 2616 779,797 2784-2275,-2055V 3056701 ,2229 V
2376-1675,557 V 2620 1083,-1557 F 2784641,1137 3056-2355,-827 V
2376-1675 ,-1819V 2620-1537, 3088 F 2826 613,649 3038667,1779 V
2380673,503 2622 1177,571 2832 641,863 3058-2391,-1279V
2398 677,853 2626 783,1187 2834837,-2215 V 3060661,-2219 V
2398 1095,431 2628 1159,1087 2834-1997,619 V 3060-2399,841 V
2400 1061,539 2630 1037,-1067 F 2838 865,619 3064 1197,-1963 A
2406 871,733 2630-1593,2568 F 2844895,-2057 V 3072673,865 V
2412 1057,-551 V 2640 553,1097 2844-1949,787 V 3072673 ,-2207 V
2412-1355,1861 V 2652 575,745 2844661,-1235 V 3072-2399,865 V
2420 549,989 V 2664 1195,619 V 2844-2183,1609 V 3072-2399,-2207V
2420-1871,-143IV 2664 1195,-2045 V 2352 1009,1561 V 3078 1351 ,-701 V
2422447,853 2664-1469,619 V 2852-1843,-1291P 3078-1727,2377 V
2430523,-1097 V 2664-1469,-2045V 2860787,-2117 V 3080 1147,-1973 A
2430-1907,1333 V 2676601,751 2860-2073,743 V 3038707,2251 V
2442 1105,449 2676 1159,625 2862665,-2215 V 3088-2381,-837 V
2448 1103,-529 V 2676 817,559 2862-2197,647 V 3088851 ,707 V
2448 1103,1919 V 2678 799,1203 2380 1013,-907 V 30888512838 V
2448-1345 ,-529 V 2682 623 ,-1165 V 2880-1867, 1973 V 3088-2245,707 V
2448-1345,1919 V 2682-2059,1517 V 2882 1073,765 3088-2245,-2389V
2448 965,-1339 V 2686 579,477 2882 901,811 3102 725,947
2448-1483,1109 V 2698 565,707 2912 1045,-2371 V 3104 1297, 2398
2450 1107, 743 V 2700629 ,-1531 V 2912-1867,541 V 3104919,951
2450-1343,-1707V 2700 -2071, 1169 V 2914 1037, 1601 V 3108577 ,-2447 V
2454689,539 2702 821,709 P 2914-1877,-1313P 3108-2531,661 V
2460 511,1129 2702-1881,-1993V 2924 1035,519 3122663 ,-2193 V
2484 541,-1907 V 2704571 ,987 V 2826 851,-2285 V 3122-2459,929 V
2484-1943,577 V 2704-2133,-1717V 2826-2075,641 V 3130 1277,-1227 F
2484 1135,-521 V 2718 745,-2099 V 2844899,853 3130-1853,1227 Fa
2484-1349,1963 V 2718-1973,619 V 2852 929,-1039 V 3136953,1289 V
2486 571,659 2724 1181,635 2852-2023, 1913 V 3136-2183,-1847V
2502 1087,-581 V 2724 1247,569 2954647,-2125 V 3136 1329,-687 V
2502-1415,1921 V 2728615,769 2954-2307,829 V 3136 1329,2449 V
2508 767,521 2736 625,769 V 2968 551 ,-1913 V 3136-1807,-687 V
2508 767,653 2736625 ,-1967 V 2968-2417, 1055 V 3136-1807,2449 V
2508 521,653 2736-2111,769 V 2838 941,-2155 V 3146593,681 V
2514691,781 2736-2111,-1967V 2988-2047,833V 3146-2553,-2465V
2516441,543 2744573,-1163 V 2880907,927PFA 3152691,723 V
2530887,-1873 V 2744573,1581 V 2880-2083,-2063PFA 3152-2461 ,-2429V
 Polynomial Invariants of 2-Bridge Links 143

Table 2 (continued)
31621369,739 3484 759,2007 V
31681127,841 3484-2725,-1477V
31781397,921V 3498 1261,-2555 V
3178-1781,-2257V 3498-2237,943 V
3182 845,-2251 V 3498 799,755 A
3182-2337,931 V 3514795 ,-1249 V
3186 973,-2303 V 3514-2719,2265 V
3186-2213,883 V 3520 763,653
3182 1381 ,-899 V 3520 931,-2269 V
3182 1381,2293 V 3520-2589,1251 V
3182-1811 ,-899 V 3542 801 ,-1271 V
3182-1811,2293 V 3542-2741,2271 V
3210 1157,983 3562 961,753
3212 597,685 3600 781,1381 V
3216 985,727 3600-2819,-2219V
3222985,-1163 V 3612 767,-1585 V
3222-2237,2059 V 3612-2845,2027 V
3230847,677 3626 1597,-783 V
3256 689,711 3626-2029,2843 V
3258 1177,-995 V 3640773,-1523 V
3258-2081,2263 V 3640 773,2117 V
3264995,1181 3640-2867,-1523V
3268691,863 3640-2867,2117 V
3278 871,3178 V 3668 1587, 1083 V
3278-2407,-2099V 3668-2081,-2585V
3294 1007,-2377 V 3682 1563,2615 V
3294-2287,917 V 3682-2119,-1067V
3300623,887 V 3710809,-1571 V
3300-2677,-2413V 3710-2901,2139 V
3306 1295,-2533 V 3738 1621,-815 V
3306-2011,773 V 3738-2117,2923 V
3318 1439,773 3762 1055,-2905 V
3322 881,749 3762-2707,857 V
3328981,3498 V 3782 2197, 4708 C
3328-2347 ,-1835V 3800 1003,803
3330 1193, 3378 3838 815,1017
3358 623,715 3844 1673,1425 PA
3378 1033,787 3844-2171,-2419PA
3380 911,1431 V 3850871,1579 A
3330-2469,-1949V 3838 1189,-2483 V
3384731 ,-1525 V 3888-2699,1405 V
3384731,1859 V 3894889,1589 A
3384-2653,-1525V 3906 2267, 1763 C
3384-2653,1859 V 3918 1453,1159
3402 775,-1493 V 3934849,-2763 V
3402-2627,1909 V 3934-3085,1171 V
3404987, 1033 3952 1061,1165 V
3406 735,1423 3952-2891,-2787V
3420 781,-1499 V 3952 1733,-1459 V
3420-2639,1921 V 3952-2219,2493 V
3430 1511,-729 V 3962 1719,3168 V
3430-1919,2701 V 3962-2243,-2803V
3432 1051,1021 3972 1477, 1171
3444 1003,-2693 V 3976 1115,-1725 V
3444-2441,751 V 3976 1115,2251 V
3458927, 1031 3976-2861,-1725V
3468 749,-2515 V 3976-2861,2251 V
3468-2719,953 V 3978 1681,841
3468 1531,-2549 V 3982 1119,1053
3468-1937,919 V 3982 1119,1075
3484 1029,-2723 V 3982 1119,-2885 V
3484-2455,761 V 3982-2863,1097 V
144 T. Kanenobu and T. Sumi

Table 2 (continued)
4712-2979,2645 V 4978-3641,1075 V 5434-3187,-3967V5828 2131,3259 C
4722 1751,1397 5018 1399,1347 V 54342247,-3473 V 5850 1571,2281
4756 1315,12575018-3619,-3671V 5434-3187, 1961 V 59662467,-3461 V
4782 1333,14235050 1969,-2071 F 5434 1467,-3473 V 5966-3499,2505 V
4784 1411,2147 V 5050-3081,2979 F 5434-3967, 1961 V 6094 1677,-4395 V
4784-3373,-2637V 5074 1981,1803 5456 1521,2017 V 6094-4417, 4898 V
4796 1337,-3503 V 5112 1933,-3211 A 5456 1521,-3439 V 61362325,-3859 A
4796-3459,1293 V 5128 1907,-3245 A 5456-3935,2017 V 61522275,-3869 A
4802 2015,-1037 V 5150 2161,-3039 F 5456-3935,-3439V 63482345,-3727 V
4802-2787,3765 V 5150-2989,2111 F 5456 1181,14676348-4003,2621 V
4836 1435,13575166 1441,1387 5546 2339,-3113 V 6498 1799,1745
48502039,-2861 F 51722119,2191 5546-3207,2433 V 65782569,-4035 F
4850-2811,1989 F 5196 2129,2153 57043189,2085 C 6578-4009,2543 F
48982149,3097 P 5278 1139,-3817 V 5776 1595,2203 V 66042501,-2579 F
4898-2749,-1801V 5278-4139,1461 V 5776-4181,-3573V 6604-4103,4025 F
48502029,-1931 F 5302 1425,1469 5798 1565,1617 V 69002899,2851 A
4850-2921,3019 F 5336 1969,1411 5798-4233,-4181V 75003151,-4649 F
4850 1073,1337 V 5382 1445,2225 V 5808 2243,-4093 V 7500-4349,2851 F
4850-3877,-3613V 5382-3937,-3157V 5808-3565,1715V
4978 1337,-3903 V 54342247, 1467 V 58141561,2255

References
[1] G.Burde and H. Zieschang, Knots, de Gruyter, Berlin and New York,
1986.
[2] J. H. Conway, An enumeration of knots and links, in Computational
Problems in Abstract Algebra , (ed. J. Leech) Pergamon Press, New
York, 1969, pp. 329-358.
[3] C. Ernst and D. W. Sumners, The growth of the number of prime knots,
Math. Proc. Cambridge Philos. Soc., 102 (1987), 303-315.
[4] R. Hartley, The Conway potential function for links, Comment. Math.
Helv., 58 (1983), 365-378.
[5] T. Kanenobu, Examples on polynomial invariants of knots and links,
Math. Ann., 275 (1986), 555-572.
[6] T. Kanenobu, Examples on polynomial invariants of knots and links II,
Osaka J. Math., 26 (1989), 465-482.
[7] T. Kanenobu, Jones and $Q$ polynomials for -bridge knots and links,
$2$

Proc. Amer. Math. Soc., 110 (1990), 835-841.

[8] T. Kanenobu, Kauman polynomials for -bridge knots and links, Yoko-
$2$

hama Math. J, 38 (1991), 145-154.

[9] T. Kanenobu and T. Sumi, Polynomial invariants of -bridge knots
$2$

through 22 crossings, preprint.

[10] H. Schubert, Knoten mit zwei Br\"ucken, Math. Z., 65 (1956), 133-170.
[11] L. Siebenmann, Exercices sur les nceuds rationels, preprint.

Taizo Kanenobu and Toshio Sumi

Department of Mathematics
Kyushu University 33
Fukuoka 812, Japan
 Polynomial Invariants of 2-Bridge Links

Current address of Taizo Kanenobu

Department of Mathematics
Osaka City University
Osaka 558, Japan
Advanced Studies in Pure Mathematics 20, 1992
Aspects of Low Dimensional Manifolds
pp. 147-166

Invariants of Spatial Graphs

Jun Murakami

\S 1. Introduction

The purpose of this paper is to construct invariants of spatial graphs

from regular isotopy invariants of non-oriented link diagrams of knit trace
type. Kaumans bracket polynomial [4], which is a version of the Jones
polynomial, is of knit $trace$ type. The Dubrovnik polynomial [5], which
is used in the definition of the Kauman polynomial, is also of knit
$trace$ type [6]. Hence these two invariants are generalized to invariants

of spatial graphs by our method. The Yamada polynomial introduced in

[10] is the non-trivial simplest one of our invariants. A similar invariants
are introduced in [9] for ribbon graphs. They use quasi-triangular Hopf
algebras. But we use representations of knit semigroups or braid groups
instead of Hopf algebras.
To introduce regular isotopy invariants of link diagrams of knit trace
type, we need notion of a Markov knit sequence. Let be the field of $\mathbb{C}$

complex numbers. Knit semigroups , $(n=J, 2, \cdots)$ are introduced in $K_{n}$

[6] defined by the following generators and relations.

$K_{n}=\langle\tau_{1}$
, $\cdots$
, $\tau_{n-1}$ , $\tau_{1}^{-1}$
, $\cdots$
, $\tau_{n-1}^{-1},$
$\in_{1}$ , $\cdots,$ $\in_{n-1}$
$|$

$\tau_{i}\tau_{i}^{-1}=\tau_{i}^{-1}\tau_{i}=1$
, $\tau_{i}\tau_{j}=\tau_{j}\tau_{i}(|i-j|\geq 2)$ ,
$\tau_{i}\tau_{i+1}\tau_{i}=\tau_{i+1}\tau_{i}\tau_{i+1}$ , $\tau_{i}\in_{j}=\in_{j}\tau_{i}(|i-j|\geq 2)$ ,
$\epsilon_{i}\in_{i\pm 1}\in_{i}==8_{i}$ , $\mathcal{E}_{i}\Xi_{j}=\in_{j}\in_{i}(|i-j|\geq 2)$ ,
$\epsilon_{i}\tau_{i\pm 1}=\epsilon_{i}\epsilon_{i\pm 1}\tau_{i}^{-1}$
, $\epsilon_{i}\tau_{i\overline{\pm}^{1}1}=\epsilon_{i}\epsilon_{i\pm 1^{\mathcal{T}}i}$
,
$\tau_{i\pm 1}\in_{i}=\tau_{i}^{-1}\in_{i\pm 1}\in_{i}$
, $\tau_{i\pm 1}^{-1}\in_{i}=\tau_{i}\in_{i\pm 1}\in_{i}\rangle$

The generators of are presented graphically as in Figure 1. In the

$K_{n}$

graphical presentation, the product of two elements of corresponds $K_{n}$

to the composite of two diagrams as in the case of braid groups. Let

Received July 6, 1991.

148 J. Murakami

$|1\iota 1|$ $|/||$ Ill $|$

$|1\prime 1||_{1}^{1}|\iota\iota\iota|$ $|I1I|$ $u\Uparrow|$ Ill $|$

1... i-l $ii+1i+2\ldots$ $n$

1... i-l $ii+1i+2\ldots$ $n$
1... i-l $ii+1i+2\ldots$ $n$

$\tau_{i}- 1$
$\epsilon$
$\tau_{i}$
$i$

Fig. 1. Generators of $K_{n}$

.

be the semigroup algebra of

$\mathbb{C}K_{n}$
over . We regard the braid group $K_{n}$ $\mathbb{C}$

as a subsemigroup of
$B_{n}$
generated by , , , . $K_{n}$ $\tau_{1}$ $\tau_{2}$
$\cdots$
$\tau_{n-1}$

Let be a non-zero complex number. Knit semigroup algebra with

$\gamma$

writhe factor , denoted by , is a quotient algebra of

$\gamma$
defined $K_{n}(\gamma)$ $\mathbb{C}K_{n}$

by the following.

$K_{n}(\gamma)=\mathbb{C}K_{n}/(\tau_{i}^{\pm 1}\in_{i}-\gamma^{\pm 1}\in_{i}, \in_{i}\tau_{i}^{\pm 1}-\gamma^{\pm 1}\in_{i} (1\leq i\leq n-1))$ .

Let $A$ be a semisimple -algebra. Let be the set of equivalence $\mathbb{C}$

$\hat{A}$

classes of irreducible representations of . A -linear map $T$ from $A$ to

$A$ $\mathbb{C}$

is called a $trace$ if $T$ is a linear combination of irreducible characters

$\mathbb{C}$

of $A$ , i.e.

(1.1) $T(x)=\sum_{\rho\in\hat{A}}a_{\rho}$
Hace(\rho (x)) $(a_{\rho}\in \mathbb{C})$

The $traceT$ is called faithful if all the coecients are not equal to $a_{\rho}$

0. A sequence , , , , of semisimple -algebras are called

$A_{1}$ $A_{2}$ $\cdots$ $A_{n}$ $\cdots$
$\mathbb{C}$
$a$

knit type sequence if they satisfy the following.

(1) There is an algebra epimorphism from to and $p_{n}$
$K_{n}(\gamma)$ $A_{n}$

monomorphism from to $A_{n+1}$ such that

$j_{n}$ $A_{n}$ $j_{n}\circ p_{n}=p_{n+1}\circ i_{n}$

for $n=1,2$ , , where is an inclusion from

$\cdots$
to $K_{n+1}(\gamma)$
$i_{n}$ $K_{n}(\gamma)$

which sends to $\tau_{i}^{\pm 1}\in K_{n}(\gamma)$

to $\tau_{i}^{\pm 1}\in K_{n+1}(\gamma)and\in_{i}\in K_{n}(\gamma)$

for $1\leq i\leq n-1$ .

$\mathcal{E}_{i}\in K_{n+1}(\gamma)$

(2) There are a complex number and a faithful $trace$ from to $\mu$
$T_{n}$ $A_{n}$

which satisfy the following. For any $x\in A_{n}$ , $T_{n+1}(j_{n}(x))=$

$\mathbb{C}$

$\mu T_{n}(x)$ , $T_{n}(x)=\gamma^{\pm 1}T_{n+1}(j_{n}(x)p_{n+1}(\tau_{n}^{\pm 1}))$ and $T_{n}(x)=$

$T_{n+1}(xp_{n+1}(\in_{n}))$ .
 Invariants of Spatial Graphs 149

For $x\in K_{n}$ , let denote the link diagram obtained from the closure of
$\hat{x}$

(Figure 2). A regular isotopy invariant $X$ of link diagrams is called of

$x$

knit trace type if there is a Markov knit sequence and $X$ is obtained by
the traces of it, i.e. $X(\hat{x})=T_{n}(p_{n}(x))$ for $x\in K_{n}$ . Kaumans bracket
polynomial [4] is of knit $trace$ type (see Section 3 of [7]). The Dubrovnik
polynomial is also of knit $trace$ type [6].

Fig. 2. Closure of $x\in K_{n}$ .

Remark. Let $X$ be a regular isotopy invariant of knit $trace$ type

with writhe factor . For an oriented link diagram , there are a positive
$\gamma$ $x$

integer and $y\in K_{n}$ such that

$n$ is equal to without orientation. $\hat{y}$
$x$

Let $w(x)$ be the sum of signatures of the crossings of . Let $X(x)=$ $x$

. The $X$ is an invariant of links.

$\gamma^{w(x)}X(\hat{y})$

Now we define spatial graphs in . Let is a set of 2-disks and $S^{3}$ $\mathcal{V}$

be a set of edges homeomorphic to

$\mathcal{E}$
$[0, 1]$ in . Each edge has an $S^{3}$

orientation induced by the orientation of $[0, 1]$ . The terminal points of

an edge corresponding to 0 and 1 are called the initial point and the final
point of the edge respectively. The pair is called an oriented $\Gamma=(\mathcal{V}, \mathcal{E})$

spatial graph if it satisfies the following. The disks in are mutually $\mathcal{V}$

disjoint and the edges in are mutually disjoint. Also assume that the $\mathcal{E}$

interiors of the disks in and edges in are mutually disjoint. Terminal $\mathcal{V}$ $\mathcal{E}$

points of edges in are contained in the boundaries of disks in . Two

$\mathcal{E}$ $\mathcal{V}$

spatial graphs and are called equivalent if there is an isotopy of

$\Gamma$ $\Gamma$

which sends to . A spatial graph is called an embedding of a

$S^{3}$ $\Gamma$ $\Gamma$ $\Gamma$
150 J. Murakami

RI

RII $\sim\vee()-)(-)_{\sim}($

RIII $\backslash |\sqrt{|\backslash \backslash }$

$\backslash \sqrt{\backslash |_{\backslash }}^{1}$

RIV $\frac{A}{//_{||\grave{|}}}-\overline{\Lambda_{|||}}-$

RV

Fig. 3. Reidemeister moves.

-valent graph if the degree of all the vertices of are equal to 3. A

$tri$ $\Gamma$

diagram of a spatial graph is defined as in the case of a link.

Proposition 1. Two spatial graphs and $\Gamma$ $\Gamma$

are equivalent if and

only if there is a sequence of Reidemeister moves of types $(SRI)-(SRV)$
sending a diagram of to a diagram of . $\Gamma$ $\Gamma$

For a spatial graph , we define a diagram of as in the case of

$\Gamma$ $\Gamma$

links. Let $A_{1}$

, , be a Markov knit sequence. For each edge $E$ of ,
$A_{2}$ $\cdots$
$\Gamma$

we associate a non-negative integer $N(E)$ , an irreducible representation

$R(E)\in\hat{A}_{n(E)}$ and a signature $S(E)$ . The triple $(N, R, S)$ is called a
coloring of if it satisfies the following. For a vertex of , let
$\Gamma$

be a $v$
$\Gamma$ $\mathcal{E}_{v}$

set of edges with terminal point Then $v$

(1.2) $\sum_{E\in \mathcal{E}_{v}}N(E)=even$

and $2N(E)\leq,\sum_{E\in \mathcal{E}_{v}}N(E)$
for all $E\in \mathcal{E}_{v}$
.
 Invariants of Spatial Graphs 151

We construct an invariant of spatial graphs colored as above. First,

we generalize link invariants of braid trace type to invariants of colored
oriented tri-valent graph embeddings in in \S 2. And then we generalize $S^{3}$

invariants of knit trace type to invariants of colored spatial graphs in \S 3.

By attaching the same color to all the edges of graphs, we get invariants
of spatial graphs. In \S 4, we give some examples.

\S 2. Invariants of colored oriented tri-valent graphs

In this section, we generalize link invariants of braid $trace$ type to
invariants of embeddings of colored oriented -valent graphs in . To $tri$ $S^{3}$

introduce link invariants of braid $trace$ type, we need notion of a Markov

braid sequence.

Definition. Asequence , , , $(A_{n}, T_{n})$ , of $(A_{1}, T_{1})$ $(A_{2}, T_{2})$ $\cdots$ $\cdots$

pairs of a semisimple -algebra and its $trace$ are calleda Markov braid
$\mathbb{C}$

sequence if they satisfy the following.

(1) There is an algebra homomorphism from to and $p_{n}$
$\mathbb{C}B_{n}$ $A_{n}$ $j_{n}$

from to $A_{n+1}$ such that $j_{n}\circ p_{n}=p_{n+1}\circ i_{n}$ for $n=1,2$ ,

$A_{n}$
, $\cdots$

where is an inclusion from

$i_{n}$
to which sends $\mathbb{C}B_{n}$ $\mathbb{C}B_{n+1}$ $\sigma_{i}\in$

to $\mathbb{C}B_{n}$
for $1\leq i\leq n-1$ .
$\sigma_{i}\in \mathbb{C}B_{n+1}$

(2) There is a faithful $traceT_{n}$ from to and , which $A_{n}$ $\mathbb{C}$

$\mu$
$c\in k\backslash \{0\}$

satisfy $\mu T_{n}(x)=T_{n+1}(j_{n}(x))$ , $T_{n}(x)=cT_{n+1}(xp_{n+1}(\sigma_{n}))$ and

$T_{n}(x)=c^{-1}T_{n+1}(xp_{n+1}(\sigma_{n}^{-1}))$ for any $x\in A_{n}$ .

From a Markov braid sequence, we get a -valued link invariant. $\mathbb{C}$

For a braid , let $w(b)=\sum_{i=1}^{r}\in(i)$ .

$b=\sigma_{i(1)^{\epsilon(1)}}\sigma_{i(2)^{\in(2)}}\cdots\sigma_{i(r)^{\in(r)}}\in B_{n}$

Then $w(b)$ is a sum of signatures of all the crossings of . Fora braid , $b$ $b$

let denote the link obtained from the closure of . Let

$ b\wedge$

$b$

$X(\hat{b})=c^{-w(b)}T_{n}(p_{n}(b))$ .

Then Alexanders theorem and Markovs theorem ([1], Theorem 2.1 and
2.2) implies that $X$ is an invariant of links. Link invariant obtained
from a Markov braid sequence as above is called of braid trace type.
Jones polynomial, HOMFLY polynomial and Kauman polynomial are
all of braid $trace$ type and the associated braid type sequences are Jones
algebras, Iwahoris Hecke algebras and a -analogue of Brauers algebras $q$

respectively ([2], [3], [6], [8]).

From now on, fix an invariant $X$ of braid $trace$ type and let , $(A_{1}, T_{1})$

, be the Markov braid

$(A_{2}, T_{2})$ $\cdots$
sequence of $X$ . Since is a semisim- $A_{n}$
152 J. Murakami

pie algebra, we have

$A_{n}=\bigoplus_{\rho\in\hat{A}_{n}}M_{d(\rho)}(\mathbb{C})$

where $d(\rho)$ is the degree of . Let $\rho$ $q_{\rho}$ be an element of $A_{n}$

such that
$\iota/(q_{\rho})=\delta_{\iota/\rho}id\in M_{d(\nu)}(\mathbb{C})$ for $iJ$
$\in\hat{A}_{n}$
.

Let be an element of
$\tilde{q}_{\rho}$
such that . Note that is not
$\mathbb{C}B_{n}$ $p_{n}(\tilde{q}_{\rho})=q_{\rho}$ $\tilde{q}_{\rho}$

unique. Let . We call the

$h_{n}=\sigma_{1}\sigma_{2}\cdots\sigma_{n-1}\sigma_{1}\cdots\sigma_{n-2}\cdots\sigma_{1}\sigma_{2}\sigma_{1}$ $h_{n}$

half twist of . Let $f_{n}=h_{n}^{2}$ and we call

$B_{n}$ the full twist of . It is $f_{n}$ $B_{n}$

known that commute with every element of $f_{n}$

and so is $B_{n}$ $\rho(p_{n}(f_{n}))$

a scalar matrix, i.e. $\rho(p_{n}(f_{n}))=\alpha_{\rho}id$ .

A formal -linear combination of link diagrams are called a virtual
$\mathbb{C}$

link diagram. We generalize the link invariant $X$ to a function from

virtual link diagrams to formally as follows. For a virtual link diagram $\mathbb{C}$

$L=\sum_{i=1}^{r}a_{i}L_{i}$ ( $a_{i}\in k$ , is alink diagram), let $X(L)=\sum_{i=1}^{r}a_{\dot{\iota}}X(L_{i})$ .

$L_{i}$

As in the case of links, we define a diagram of an oriented -valent $tri$

graph embedded in . Let $G$ be an oriented -valent graph. We define $S^{3}$ $tri$

a coloring of $G$ . For each edge $E$ of $G$ , associate a non-negative inte-
ger $N(E)$ , an irreducible representation $R(E)\in\hat{A}_{n(E)}$ and a signature
$S(E)=\pm 1$ . The triple $(N, R, S)$ is called a coloring of $G$ if it satisfies
the following. For a vertex of $G$ , let be a set of edges with end $v$ $E_{v}^{-}$

point and a set of edges with start point . Then

$v$ $E_{v}^{+}$ $v$

$\sum_{E\in E_{v}^{-}}N(E)=\sum_{E\in E_{v}^{+}}N(E)$

.

Let be a diagram of an embedding of an oriented -valent graph

$\Gamma$
$tri$

$G$ colored by $(N, R, S)$ . We identify the edge sets of and $G$ . For an $\Gamma$

edge $E$ of , let $\Gamma$

. Replace $\beta(E)=\frac{1}{2}\tilde{q}_{R(E)}(1+S(E)\alpha_{R(E)}^{-1/2}h_{n})\in \mathbb{C}B_{N(E)}$

every vertices and edges as in Figure 4, we get a virtual link diagram

. For a edge $E$ of , let $c(E)=S(E)\alpha_{R(E)}^{1/2}$ .
$\Gamma^{(N,R,S)}$ $\Gamma$

Theorem 2. Let and be equivalent embeddings of an $ori$ ented $\Gamma$ $\Gamma$

-valent graph $G$ colored by $(N, R, S)$ . Then, for every edge $E$ of $G$ ,
$tri$

there is an integer $d(E)$ such that

(2.1) $X(\Gamma^{(N,R,S)})=\prod_{E\in \mathcal{E}}c(E)^{d(E)}X(\Gamma^{\prime(N,R,S)})$ .

We check (2.1) for Reidemeister moves $(SRI)-(SRV)$ . Let

Proof $\Gamma$

and be diagrams of embeddings of $G$ . We identify the sets of edges

$\Gamma$

of and with that of $G$ .

$\Gamma$ $\Gamma$
 Invariants of Spatial Graphs 153

$N_{1}$ $N_{1}$
$N$

$(N,RE\downarrow S)\rightarrow$

$N$

$N_{3}$
$N_{3}$

Fig. 4. Replace vertices and edges.

Case 1. Assume that and are regular isotopic, . . there is a $\Gamma$ $\Gamma$ $i$
$e$

sequence of Reidemeister moves of types (SRII), (SRIII), (SRIV) sending

to . Then the associated virtual link diagrams
$\Gamma$ $\Gamma$
and $\Gamma^{(N,R,S)}$ $\Gamma^{\prime(N,R,S)}$

are equivalent. Hence we have

(2.2) $X(\Gamma^{(N,R,S)})=X(\Gamma^{J(N,R,S)})$ .

Case 2. In this and the next cases, we check (2.1) for (SRI) moves.
Assume that and are identical except within a ball where they are
$\Gamma$ $\Gamma$

as shown in Figure 5. Let $E$ be the edge of $G$ embedded dierently by $\Gamma$

and . Let $n=N(E)$ , $\rho=R(E)$ , $s=S(E)$ and $\beta=\beta(E)$ . Then there

$\Gamma$

are positive integer $N$ and a braid such that the associated $b\in \mathbb{C}B_{N}$

link diagrams and are equivalent to the closures of

$\Gamma^{(N,R,S)}$ $\Gamma^{J(N,R,S)}$

$b_{1}=b\eta(\beta)$ and $b_{2}=b\eta(\beta)f_{n}$ where is an algebra homomorphism $\eta$

from to defined by
$\mathbb{C}B_{n}$ $\mathbb{C}B_{N}$
for $1\leq i\leq n-1$ . Since $X$ $\eta(\sigma_{i})=\sigma_{i}$

is an invariant of $trace$ type, there is an algebra homomorphism from $J$

to
$A_{n}$
such that $p_{N}\circ\eta=J\circ p_{n}$ . From the definition of $trace$ type
$A_{N}$

invariants, we have
$X(\hat{b}_{2})=T_{N}(p_{N}(b_{2}))=T_{N}(p_{N}(b\eta(\beta)f_{n}))$ .

The definitions of $q_{\rho}$ and $\beta$

imply that $p_{n}(\beta h_{n}^{\pm 1})=(s\alpha_{\rho}^{1/2})^{\pm 1}p_{n}(\beta)$
.
Hence we have
$T_{N}(p_{N}(b\eta(\beta)f_{n}))=T_{N}(p_{N}(b)J(p_{n}(\beta h_{n}^{2})))$
154 J. Murakami

$=T_{N}(p_{N}(b)J(\alpha_{\rho}p_{n}(\beta)))=\alpha_{\rho}T_{N}(p_{N}(b)J(p_{n}(\beta))))$

and so we get
$X(\hat{b}_{2})=\alpha_{\rho}X(\hat{b}_{1})$
.

In other words,

(2.3) $X(\Gamma^{(N,R,S)})=\alpha_{\rho}X(\Gamma^{\prime(N,R,S)})$
.

$\Gamma$
$\Gamma^{t}$

Fig. 5.

Case 3. Let and be diagrams of colored -valent graphs iden-

$\Gamma$ $\Gamma$
$tri$

tical except within a ball where they are as shown in Figure 6. Then, as
in Case 2, we have

(2.4) $X(\Gamma^{(N,R,S)})=\alpha_{R(E)}^{-1}X(\Gamma^{1(N,R,S)})$ .

$\Gamma$ $\Gamma^{\dagger}$

Fig. 6.
 Invariants of Spatial Graphs 155

Case 4. To check (SRV), it is suce to verify the theorem for moves

illustrated in Figures 7-10. Assume that and are identical except $\Gamma$ $\Gamma$

within a ball where they are as shown in Figure 7. Let $n(i)=N(E_{i})$ ,

$\rho(i)=R(E_{i})$ , , $h_{i}=h_{n(i)}$
$s(i)=S(E_{i}),\tilde{q}_{i}=\tilde{q}_{\rho(i)}$ , $q_{i}=q_{\rho(i)}$ , $p_{i}=p_{n(i)}$

and $\beta_{i}=\beta(E_{i})$ for $i=1,2,3$ . Then there are positive integer $N$ and
such that the associated link diagrams
$b\in \mathbb{C}B_{N}$
$\Gamma^{(N,R,S)}$
and $\Gamma^{\prime(N,R,S)}$

are equivalent to the closures of

$b_{1}=b\eta_{1}(\beta_{1})\eta_{2}(\beta_{2})\eta_{3}(\beta_{3})$ ,
$b_{2}=b\eta_{1}(\beta_{1})\eta_{2}(f_{n(2)}\beta_{2})\sigma_{n(1),n(2)}\eta_{3}(\beta_{3})$ ,

where $\sigma_{n(1),n(2)}=\sigma_{n(1)}\sigma_{n(1)+1}\cdots\sigma_{n(1)+n(2)-1}\sigma_{n(1)-1}\cdots\sigma_{n(1)+n(2)-2}$

$\ldots\sigma_{1}\sigma_{2n(2)}\ldots\sigma$ and
are algebra homomorphisms from
$\eta_{1}$ , $\eta_{2}\eta_{3}$ , $\mathbb{C}B_{n(1)}$

defined by the following.

$\mathbb{C}B_{n(2)}\mathbb{C}B_{n(3)}$ to $\mathbb{C}B_{N}$
for $\eta_{1}(\sigma_{i})=\sigma_{i}$ $ 1\leq$

$i\leq n(1)-1$ , for $1\leq i\leq n(2)-1$ and

$\eta_{2}(\sigma_{i})=\sigma_{n(1)+i}$ for $\eta_{3}(\sigma_{i})=\sigma_{i}$

$1\leq i\leq n(3)-1$ . We know that $\eta_{1}(h_{n(1)})\eta_{2}(h_{n(2)}\sigma_{n(1),n(2)})=\eta_{3}(h_{n(3)})$ .

Hence we have
$b_{2}=b\eta_{1}(\beta_{1}h_{1}^{-1})\eta_{2}(\beta_{2}h_{2})\eta_{3}(h_{3}\beta_{3})$
.

Since is an invariant of $trace$ type, there are algebra homomorphisms

$X$

$J_{1}$
, $J_{2}$
and from $A_{n(1)}$ , $A_{n(2)}$ and $A_{n(3)}$ to
$J_{3}$
such that $p_{N}o\eta_{s}=$ $A_{N}$

$J_{s}\circ p_{n(s)}$ for $s=1,2,3$ . From the definition of the $trace$ type, we have

$X(\hat{b}_{2})=T_{N}(p_{N}(b_{2}))$

$=T_{N}(p_{N}(b\eta_{1}(\beta_{1}h_{1}^{-1})\eta_{2}(\beta_{2}h_{2})\eta_{3}(h_{3}\beta_{3}))$

$=T_{N}(p_{N}(b)J_{1}(p_{1}(\beta_{1}h_{1}^{-1}))J_{2}(p_{2}(\beta_{2}h_{2}))J_{3}(p_{3}(h_{3}\beta_{3})))$ .

The definition of $q_{R}$ and $\beta(E)$ implies that

$p_{t}(\beta(t)h_{t}^{\pm 1})=S(t)\alpha_{\rho(t)}^{\pm 1/2}p_{t}(\beta_{t})$ $(t=1,2,3)$ .

Hence we have
$T_{N}(p_{N}(b)J_{1}(p_{1}(\beta_{1}h_{1}^{-1}))J_{2}(p_{2}(\beta_{2}h_{2}))J_{3}(p_{3}(h_{3}\beta_{3})))$

$=(\prod_{t=1}^{3}s(t))\alpha_{\rho(1)}^{-1/2}\alpha_{\rho(2)}^{1/2}\alpha_{\rho(3)}^{1/2}T_{N}(p_{N}(b)J_{1}(p_{1}(\beta_{1}))J_{2}(p_{2}(\beta_{2}))J_{3}(p_{3}(\beta_{3})))$
,

and so we get

$X(\hat{b}_{2})=s(1)\alpha_{\rho(1)}^{-1/2}s(2)\alpha_{\rho(2)}^{1/2}s(3)\alpha_{\rho(3)}^{1/2}X(\hat{b}_{1})$
.
156 J. Murakami

In other words,

(2.5) $X(\Gamma^{(N,R,S)})=s(1)\alpha_{\rho(1)}^{-1/2}s(2)\alpha_{\rho(2)}^{1/2}s(3)\alpha_{\rho(3)}^{1/2}X(\Gamma^{\prime(N,R,S)})$
.

$E_{1}$ $E_{1}$

$E_{2}$ $E_{3}$
$E_{2}$ $E_{3}$
$\Gamma_{1}$ $\Gamma_{2}$

Fig. 7.

Case 5. Assume that and are identical except within a ball

$\Gamma$ $\Gamma$

where they are as shown in Figure 8. Then, as in Case 4, we have

(2.5) $X(\Gamma^{(N,R,S)})=s(1)\alpha_{\rho(1)}^{1/2}s(2)\alpha_{\rho(2)}^{-1/2}s(3)\alpha_{\rho(3)}^{-1/2}X.(\Gamma^{J(N,R,S)})$ .

$E_{1}$ $E_{1}$

$E_{2}$ $E_{3}$ $E_{2}$ $E_{3}$

$\Gamma_{1}$ $\Gamma_{2}$

Fig. 8.

Case 6. Assume that and are identical except within a ball

$\Gamma$ $\Gamma$

where they are as shown in Figure 9. Then, as in Case 4, we have

(2.7) $X(\Gamma^{(N,R,S)})=s(1)\alpha_{\rho(1)}^{1/2}s(2)\alpha_{\rho(2)}^{1/2}s(3)\alpha_{\rho(3)}^{-1/2}X(\Gamma^{\prime(N,R,S)})$
.
 Invariants of Spatial Graphs 157

$E_{1}$ $E_{2}$ $E_{1}$ $E_{2}$

$r_{1}$ $r_{2}$

Fig. 9.

Case 7. Let and be diagrams of colored -valent graphs iden-

$\Gamma$ $\Gamma$
$tri$

tical except within a ball where they are as shown in Figure 10. Then,
as in Case 4, we have

(2.8) $X(\Gamma^{(N,R,S)})=s(1)\alpha_{\rho(1)}^{-1/2}s(2)\alpha_{\rho(2)}^{-1/2}s(3)\alpha_{\rho(3)}^{1/2}X(\Gamma^{\prime(N,R,S)})$ .

$E_{1}$ $E_{2}E_{1}$ $E_{2}$

$\Gamma_{1}$ $r_{2}$

Fig. 10.

The above formulas (2.2)-(2.8) implies Theorem 2. Q.E.D.

\S 3. Invariants of non-oriented spatial graphs

Let $X$ be a regular isotopy invariant of link diagrams of knit $trace$
type with writhe factor . Let $G$ be an abstract graph. For each edge
$\gamma$

$E$ of $G$ , we attach a non-negative integer $N(E)$ , an irreducible represen-

tation $R(E)\in\check{A}_{N(E)}$ and a signature $S(E)=\pm 1$ . If these data satisfy

158 J. Murakami

(1.2) in \S 1, they are called a coloring of $G$ and denoted by $(N, R, S)$ .
Let be the subset of edges of $G$ with a terminal point .
$\mathcal{E}_{v}$
$v$

From now on, fix an invariant $X$ of knit $trace$ type and let , $(A_{1}, T_{1})$

, be the Markov knit sequence of . Since

$(A_{2}, T_{2})$ $X$
$\cdots$
is a semisimple $A_{n}$

algebra, we have
$A_{n}=\bigoplus_{\rho\in\overline{A}_{n}}M_{d(\rho)}(\mathbb{C})$

where $d(\rho)$ is the degree of . Let $\rho$

$q_{\rho}$ be an element of $A_{n}$
such that

$\nu(q_{\rho})=\delta_{\nu\rho}id\in M_{d(\nu)}(\mathbb{C})$ for $iJ$

$\in\check{A}_{n}$
.

Let be an element of
$\tilde{q}_{\rho}$
such that . Note that is $\mathbb{C}K_{n}$ $p_{n}(\tilde{q}_{\rho})=q_{\rho}$ $\tilde{q}_{\rho}$

not unique. Let . We call the

$h_{n}=\tau_{1}\tau_{2}\cdots\tau_{n-1}\tau_{1}\cdots\tau_{n-2}\cdots\tau_{1}\tau_{2}\tau_{1}$ $h_{n}$

half twist of . Let $f_{n}=h_{n}^{2}$ and we call the full twist of . It is

$K_{n}$ $f_{n}$ $K_{n}$

known that commute with every element of $f_{n}$

and so is $K_{n}$ $\rho(p_{n}(f_{n}))$

a scalar matrix, i.e. $\rho(p_{n}(f_{n}))=\alpha_{\rho}id$ .

Let $G$ be an abstract graph colored by $(N, R, S)$ . Let be a colored $\Gamma$

non-oriented spatial graph equal to $G$ as an abstract graph. We identify

the sets of edges of and $G$ . Let be a vertex of . Let , , , $\Gamma$
$v$
$\Gamma$
$E_{1}$ $E_{2}$ $\cdots$ $E_{r}$

be the edges with a terminal point . Let , , , be the terminal $v$ $\xi_{1}$ $\xi_{2}$ $\cdots$ $\xi_{r}$

points of , , , on the boundary of and $N(i)=N(E_{i})$ for

$E_{1}$ $E_{2}$ $\cdots$ $E_{r}$ $v$

$i=1,2$ , , . Replace these points by $\cdots$

,
$r$
, , , , , $\zeta_{1}^{(1)}$ $\zeta_{1}^{(2)}$
$\cdots$
$\zeta_{1}^{(N(1))}$ $\zeta_{2}^{(1)}$
$\cdots$

$\zeta_{2}^{(N(2))}$
, $\cdots$
, $\zeta_{r}^{(1)}$

, $\cdots$
, $\zeta_{r}^{(N(r))}$
as in Figure 11. Let $n_{v}=(\sum_{i=1}^{r}N(i))/2$ .
A diagram $D$ on $v$ is a set of mutually disjoint $n_{v}$ curves connecting $\gamma_{i(1)}^{j(1)}$

to . Two diagrams $D$ and $D$ on are called equivalent if there is an

$\gamma_{i(2)}^{j(2)}$
$v$

isotopy of sending $D$ to $D$ which fixes the boundary of . A diagram

$v$ $v$

$D$ on is called essential if $D$ satisfies the following.

$v$

$(^{*})$
Let $\gamma_{i(1)}^{j(1)}$
and $\gamma_{i(2)}^{j(2)}$
be distinct boundary points of a curve of $D$ .
then $i(1)\neq i(2)$ .

We denote by the set of equivalence classes of essential diagrams on

$D_{v}$

. If the valency of is equal to 3, then

$v$ has only one element. If $v$ $D_{v}$

the valency of is equal to 4 and $N(E_{i})=2$ for $i=1$ , , 4, then

$v$ $\cdots$ $D_{v}$

consists of 3 elements as in Figure 12.

Let $\beta(E)=\frac{1}{2}\tilde{q}_{R(E)}(1+S(E)\alpha_{R(E)}^{-1/2}h_{n})\in \mathbb{C}B_{N(E)}$ . Let $\Gamma^{(N,R,S)}$

be the virtual link diagram obtained by replacing each vertex by a $v$

sum of the all elements of and each edge $E$ by as in the case $D_{v}$ $\beta(E)$

of embeddings of oriented -valent graphs. For a edge $E$ of , let $tri$ $\Gamma$

$c(E)=S(E)\alpha_{R(E)}^{1/2}$ .
 Invariants of Spatial Graphs 159

$2(1)$

Fig. 11. Replace $\xi_{1}$

, $\cdots$
, $\xi_{r}$
by $\zeta_{1}^{(1)},$
$\cdots$
, $\zeta_{1}^{(N(1))}$
, $\zeta_{2}^{(1)}$
, $\cdots$
,
$\zeta_{2}^{(N(2))}$
, $\cdots$
, $\zeta_{r}^{(1)}$

, $\cdots$
, $\zeta_{r}^{(N(r))}$
.

Fig. 12. Elements of $D_{v}$

.

$N$

$N_{i}$

$N_{1}$ $N_{2}$ $N_{1}$ $N_{2}$

$N$
Qi, $R$ , S)

$\rightarrow$
$E|$

Fig. 13. Replace edges and vertices.

Theorem 3. Let $\Gamma$

and $\Gamma$
be colored spatial graphs isomorphic to
160 J. Murakami

a graph $G$ colored by $(N, R, S)$ as abstruct graphs. Identify the sets of
edges of and with that of G. If and
$\Gamma$

are equivalent as spatial

$\Gamma$ $\Gamma$ $\Gamma$

graphs, then there are integers and $d(E)$ for every edge $E$ of $G$ such $d$

that

(3.1) $X(\Gamma^{(N,R,S)})=\gamma^{d}\prod_{E\in \mathcal{E}}c(E)^{d(E)}X(\Gamma^{J(N,R,S)})$ .

We check (3.1) for Reidemeister moves $(SRI)-(SRV)$ . Let

Proof
and
$\Gamma$

be diagrams of colored spatial graphs isomorphic to $G$ . We

$\Gamma$

identify the sets of edges of and with that of $G$ . $\Gamma$ $\Gamma$

Case 1. Assume that and regular isotopic, . . there is a $\Gamma$ $\Gammaare^{-}$ $i$
$e$

sequence of reidemeister moves of types (SRII), (SRIII), (SRIV) sending

to . Then the associated virtual link diagrams
$\Gamma$ $\Gamma$
and $\Gamma^{(N,R,S)}$ $\Gamma^{;(N,R,S)}$

are equivalent and we have

(3.2) $X(\Gamma^{(N,R,S)})=X(\Gamma^{\prime(N,R,S)})$
.

Case 2. In this and the next cases, we check (2.1) for (SRI) moves.
Assume that and are identical except within a ball where they are as
$\Gamma$ $\Gamma$

shown in Figure 5. Let $n=N(E)$ , $\rho=R(E)$ , $s=S(E)$ and $\beta=\beta(E)$ .

Then there are positive integer $N$ and such that the associated $b\in \mathbb{C}K_{N}$

link diagrams and are equivalent to the closures of

$\Gamma^{(N,R,S)}$ $\Gamma^{\prime(N,R,S)}$

$b_{1}=b\eta(\beta)$ and where is an algebra homomorphism

$b_{2}=b\eta(\beta)h_{n}^{2}$ $\eta$

from to defined by
$\mathbb{C}K_{n}$
for $1\leq i\leq n-1$ . Since
$\mathbb{C}K_{N}$ $\eta(\sigma_{i})=\sigma_{i}$

$X$ is a regular isotopy invariant of knit $trace$ type, there is an algebra

homomorphism from to such that $p_{N}\circ\eta=J\circ p_{n}$ . From the

$J$ $A_{n}$ $A_{N}$

definition of $trace$ type invariants, we have

$X(\hat{b}_{2})=T_{N}(p_{N}(b_{2}))=T_{N}(p_{N}(b\eta(\beta)h_{n}^{2}))$ .

The definition of $\beta$

implies that
$p_{n}(\beta h_{n}^{\pm 1})=s\alpha_{\rho}^{\pm 1/2}p_{n}(\beta)$
.

Hence we have
$T_{N}(p_{N}(b\eta(\beta)h_{n}^{2}))=T_{N}(p_{N}(b)J(p_{n}(\beta h_{n}^{2})))$

$=T_{N}(p_{N}(b)J(\alpha_{\rho}p_{n}(\beta)))$

$=\alpha_{\rho}T_{N}(p_{N}(b)J(p_{n}(\beta)))$ ,

and so we get
$X(\hat{b}_{2})=\alpha_{\rho}X(\hat{b}_{1})$
.
 Invariants of Spatial Graphs 161

In other words,
$X(\Gamma^{(N,R,S)})=\alpha_{\rho}X(\Gamma^{J(N,R,S)})$ .

Case 3. Assume that and are identical except within a ball $\Gamma$ $\Gamma$

where they are as shown in Figure 6. Then, as in Case 2, we have

(3.3) $X(\Gamma^{(N,R,S)})=\alpha_{\rho}^{-1}X(\Gamma^{\prime(N,R,S)})$
.

Case 4. Assume that and are identical except within a ball $\Gamma$ $\Gamma$

where they are as shown in Figure 14. Let , , , be edges $E_{1}$ $E_{2}$ $\cdots$ $E_{r}$

around the verjbex . Let $n(i)=N(E_{i})$ for $i=1,2$ ,

$v$ , and $n=$ $\cdots$ $r$

$\sum_{i=1}^{r}n(i)$ . $Let\in_{1,n}=\in_{1}\in_{3}\cdots\in_{2n-1}\in K_{n}$ .

around vertex $v$

around vertex $v^{1}$

$E_{1}E_{2}\ldots E_{r}$

$\Gamma$
$\Gamma^{1}$

Fig. 14.

Let , and
$h_{v}$
be the element of
$e_{v}$
$e_{v}$
corresponding to the diagram $K_{n}$

in Figure 15. Let $\eta_{i,j,k}(i, j>0, k\geq 0, i+k\leq j)$ be a semigroup

homomorphism from to which sends $K_{i}$
to
$K_{j}$
$\tau_{i}^{\pm 1},$
$\in_{i}\in K_{i}$ $\tau_{i+k}^{\pm 1},$
$\in_{i+k}\in$

and $\phi_{i,j}=\eta_{n(i),j,n(1)+n(2)+}.+n(i-1)$ . Note that

$K_{j}$

$h_{n}e_{v}=\gamma^{n}e_{v}$ ,

$h_{n}e_{v}=e_{v}\phi_{1,n}(h_{n(1)})\phi_{2,n}(h_{n(2)})\cdots\phi_{r,n}(h_{n(r)})$ ,

and so we have

(3.4) $e_{v}=\gamma^{n}e_{v}\phi_{1,n}(h_{n(1)}^{-1})\phi_{2,n}(h_{n(2)}^{-1})\cdots\phi_{r,n}(h_{n(r)}^{-1})$ .

Let $\rho(i)=R(E_{i})$ , $s(i)=S(E_{i})$ and $\beta(i)=\beta(E_{i})$ for $i=1,2$ , , $\cdots$

$r$
. Then there are an integer $N$ and an element such that the $b\in \mathbb{C}K_{n}$
162 J. Murakami

$n$
strlngs

$h_{V}=$

$n$ sngs
$\cup\cup$ $\cup$
$\cdots$
$n(1)$ $n(rt-1)n(r)$

$e$

Fig. 15. Diagrams of $h_{v}$

, $e_{v}$ and $e_{v}$
.

associated link diagrams $\Gamma^{(N,R,S)}$

and $\Gamma^{J(N,R,S)}$
are equivalent to the
closures of

$b_{1}=b\eta_{n,N,0}(e_{v})\phi_{1,N}(\beta(1)\tilde{q}_{\rho(1)})\phi_{2,N}(\beta(2)\tilde{q}_{\rho(2)})\cdots\phi_{r,N}(\beta(r)\tilde{q}_{\rho(r)})$ .
$b_{2}=b\eta_{n,N,0}(e_{v}h_{v})\phi_{1,N}(\beta(1)\tilde{q}_{\rho(1)})\phi_{2,N}(\beta(2)\tilde{q}_{\rho(2)})\cdots\phi_{r,N}(\beta(r)\tilde{q}_{\rho(r)})$ .

From (3.4), we have

(3.5) $b_{2}=\gamma^{n}b\eta_{n,N,0}(e_{v})\phi_{1,N}(h_{n(1)}^{-1}\beta(1)\tilde{q}_{\rho(1)})\cdots\phi_{r,N}(h_{n(r)}^{-1}\beta(r)\tilde{q}_{\rho(r)})$ .

Recall that the definition of $q_{R(E)}$ and $\beta(E)$ implies that

$q_{\rho(t)}p_{n(t)}(\beta(t)h_{n(t)}^{\pm 1})=s(t)\alpha_{\rho(t)}^{\pm 1/2}q_{\rho(t)}p_{n(t)}(\beta(t))$

for $t=1,2$ , $\cdots$

, . Hence formula (3.5) implies
$r$

(3.6) $X(\hat{b}_{2})=\prod_{i=1}^{r}S(t)\alpha_{\rho(t)}^{-1/2}X(\hat{b}_{1})$ ,

because $X$ is of knit $trace$ type.

 Invariants of Spatial Graphs 163

Case 5. Assume that and are identical except within a ball

$\Gamma$ $\Gamma$

where they are as shown in Figure 16. Then, as in Case 4, we have

(3.7) $X(\hat{b}_{2})=\prod_{i=1}^{r}s(t)\alpha_{\rho(t)}^{1/2}X(\hat{b}_{1})$ .

$E_{1}E_{2}\ldots E_{r}$

$\Gamma^{\dagger}$

$\Gamma$

Fig. 16.

Formulas (3.2), (3.3), (3.6), (3.7) show Theorem 3. Q.E.D.

Let $N$ be a positive even number. Let $R$ be an irreducible repre-

sentation of the algebra associated with the link invariant $X$ . Let
$A_{N}$ $S$

be 1 or -1. For a spatial graph , let $(N, R, S)$ be the coloring of $\Gamma$ $\Gamma$

defined by $N(E)=E$ , $R(E)=R$ and $S(E)=S$ for every edge $E$ of . $\Gamma$

Let $X^{(N,R,S)}(\Gamma)=X(\Gamma^{(N,RS)},)$ . Then $X^{(N,R,S)}$ is a regular isotopy

invariant of diagrams of spatial graphs.

Corollary 4. Let and be diagrams of the same spatial graph

$\Gamma$ $\Gamma$

G. Then, there are integers and such that $d$ $d$

$X^{(N,R,S)}(\Gamma)=\gamma^{d}\alpha_{R^{d}}X^{(N,R,S)}(\Gamma)$
.

The proof is similar to that of Theorem 2.

\S 4. Examples

KaumanUs bracket polynomial is a regular isotopy invariant $\langle. \rangle$

of knit $trace$ type and the Jones polynomial is obtained from as $\langle. \rangle$

in Remark in \S 1. To fix the notation, we give the definition of the

164 J. Murakami

bracket polynomial . Let which is not equal to any

$\langle. \rangle[4]$ $A\in \mathbb{C}\backslash \{0\}$

roots of unity. The bracket polynomial with parameter $A$ is a regular

isotopy invariant of non-oriented link diagrams defined by the following
relations.

$\langle L_{O}\rangle=1$
,
$\langle L_{x}\rangle=A\langle L_{||}\rangle+A^{-1}\langle L_{\infty}\rangle$
,

where is a trivial knot and

$L_{O}$
, , are link diagrams identical $L_{x}$ $L_{||}$ $L_{\infty}$

except within a ball where they are as shown in Figure 17.

$L_{\infty}$
$L_{0}$ $L_{||}$

Fig. 17. Diagrams of $L_{x}$

, $L_{||}$ , $L_{\infty}$
.

Let $A$ be a non-zero complex number which is not equal to any roots
of unity. Let $J_{n}(A)$ be the Jones algebra defined over by the following. $\mathbb{C}$

$J_{n}(A)=\langle e_{1}$ , $e_{2}$ , $\cdots$

, $e_{n-1}|e_{i}e_{i\pm 1}e_{i}=e_{i}$ , $e_{i}e_{j}=e_{j}e_{i}(|i-j|\geq 2)$ ,
$ e_{i}^{2}=-(A^{2}+A^{-2})e_{i}\rangle$ .

The Markov knit sequence of KaumanUs bracket polynomial is $\langle. \rangle$

$J_{1}(A)$ , $J_{2}(A)$ , . The algebra homomorphism

$\cdots$
from to $J_{n}(A)$ $p_{n}$
$\mathbb{C}K_{n}$

is defined by $p_{n}(\in_{i})=e_{i}$ , $p_{n}(\tau_{i})=A+A^{-1}e_{i}$ and $p_{n}(\tau_{i}^{-1})=A^{-1}+Ae_{i}$ .

Let be the linear representation of $J_{n}(A)$ sending , ,
$\rho_{n}$ , to $e_{1}$ $e_{2}$
$\cdots$
$e_{n-1}$

0. Since $\rho_{n}(p_{n}(\tau_{i}))=A$ , we have

(4.1) $\rho_{n}(h_{n})=A^{n(n-1)/2}$ .

Let $\alpha_{n}=A^{n(n-1)}$ and $\sqrt{\alpha_{n}}=A^{n(n-1)/2}$ . The Yamada polynomial in

[10] is coming from as in Corollary 4 with $N=2$ , $R=\rho_{2}$ and $S=1$ .
$\langle. \rangle$

Let and be two diagrams of spatial graphs as in Figure 18.

$\Gamma_{1}$ $\Gamma_{2}$

The diagrams and are colored as in the figure. Let

$\Gamma_{1}$
, $\Gamma_{2}$ $C_{1}$ $C_{2}$

denote the above coloring for and respectively. Since $p_{2}(1+(A^{2}+$ $\Gamma_{1}$ $\Gamma_{2}$
 Invariants of Spatial Graphs 165

$(2, p_{2},1)$ $(2, p_{2},1)$

$\Gamma_{1}$

$(1, p_{1},1)$ $(1, p_{1},1)$ $(1, p_{1},1)$

Fig. 18. Diagrams of spatial graphs $\Gamma_{1}$

and $\Gamma_{2}$
.

Fig. 19. Virtual link diagrams $\Gamma_{1}^{C_{1}}$

and $\Gamma_{2}^{C_{2}}$
.

, the virtual diagrams

$A^{-2})^{-1}\in_{1})=q_{R_{2}}$ $\Gamma_{1}^{C_{1}}$

and $\Gamma_{2}^{C_{2}}$

associated to the
colorings are given in Figure 19.
Hence we have
$\langle\Gamma_{1}\rangle^{C_{1}}=-\frac{A^{8}+A^{4}+1}{A^{2}(A^{4}+1)}$

and
$\langle\Gamma_{2}\rangle^{C_{2}}=-\frac{-A^{32}+A^{28}+A^{20}+A^{8}+1}{A^{13}(A^{4}+1)}$ .
166 J. Murakami

By (4.1) and Theorem 3, we know that and are not equivalent as

$\Gamma_{1}$ $\Gamma_{2}$

spatial graphs.
To investigate the invariants associated with the Jones polynomial
more closely, Section 4 of [7] may be helpful.
The HOMFLY polynomial $P$ is an oriented link invariant of $trace$
type. Hence we get invariants of colored oriented -valent graph em- $tri$

beddings from the HOMFLY polynomial.

The Kauman polynomial $F$ is an oriented link invariant obtained
from the Dubrovnik polynomial [5], which is a regular isotopy invariant
of unoriented link diagrams. It is shown in [2], [7], [8] that the Dubrovnik
polynomial is of knit $trace$ type. Hence we get invariants of spatial
graphs from the Dubrovnik polynomial. To investigate properties of
these invariants, Section 5 of [7] may be helpful.

References
[1] Birman, J.S., Braids, Links and Mapping Class Groups, Princeton
University Press, Princeton, 1975.
[2] Birman, J.S. and Wenzl, H., Braids, link polynomials and a new algebra,
Trans. Amer. Math. Soc., 313 (1989), 249-273.
[3] Jones, V.F.R., A polynomial invariant for knots via von Neumann alge-
bras, Bull. Amer. Math. Soc., 12 (1985), 103-111.
[4] Kauman, L.H., State models and the Jones polynomial, Topology, 26
(1987), 395-407.
[5] , State models and the Jones polynomial An introduction,

preprint, University of Illinois at Chicago, 1987.

[6] Murakami, J., The Kauman polynomial of links and representation
theory, Osaka J. Math, 24 (1987), 745-758.
[7] , The parallel version of polynomial invariants of links, Osaka J.

Math., 26 (1989), 1-55.

[8] , The representations of the -analogue of Brauer,s centralizer al-
$q$

gebras and the Kauman polynomial of links, Publ. Res. Inst. Math.
Sci., Kyoto Univ., 26 (1990), 935-945.
[9] Reshetikhin, N.Yu. and Turaev, V.G., Ribbon graphs and their in-
variants derived from quantum groups, Commun. Math. Phys., 127
(1990), 1-26.
[10] Yamada, S., An invariant of spatial graphs, J. Graph Theory, 13 (1989),
537-551.

Department of Mathematics
Osaka University
Toyonaka, Osaka 560
Japan
Advanced Studies in Pure Mathematics 20, 1992
Aspects of Low Dimensional Manifolds
pp. 167-261

Foundations of
Flat Conformal Structure

Shigenori Matsumoto

Dedicated to Professor Masahisa Adachi

Introduction

A flat conformal structure on an -dimensional manifold $N$ is a

$n$

maximal system of local charts taking values on , with transition

$S^{n}$

functions Moebius transformations. In short it is a geometric structure

modelled on (A4 ,
$(S^{n})$ ), where A4
$S^{n}$
denotes the group of Moebius
$(S^{r\iota})$

transformations on . Equivalently, it is a conformal equivalence

$S^{n}$

class of conformally flat Riemannian metrics on $N$ if . See $n$ $\geq 3$

\S 1 for Liouvilles theorem. By certain abuse we denote a flat conformal

structure by the same letter as the underlying manifold.
In dimension 2, flat conformal structures are usually called projec-
tive structures and have been extensively studied by various authors in
the field of function theory. Analytic methods such as the theory of
quasiconformal maps often play crucial roles there. In dimension , $\geq 3$

however, the situation is quite dierent. Topology, instead of analysis,

provides major tools of study.
The concept of flat conformal structures was first introduced by
Kuiper ([35],[36],[37]) around 1950. Thereafter it had been forgotten
for some time, until it was revived by Kulkarni ([40], [41], [42], [43]), re-
lated with his study of discrete group actions in general. Then came
an important turning point when Fkied ([13]) established a remarkable
theorem concerning closed similarity manifolds. It solved a fundamental
and annoying problem which one encounters in the primary stage of the
theory, thereby making it possible to have a good grip on elementary
flat conformal structures, with Goldman ([15]) and Kamishima ([25])
contributing significantly to this direction.

Received July 9, 1990.

Revised June 13, 1991.
168 S. Matsumoto

At the same time various interesting examples have been piled up by

many authors including Thurston [56], Bestvina-Cooper [4], Freedman-
Skora [10], Gromov-Lawson-Thurston [19], Kuiper [38] and, quite re-
cently,
Kapovich-Potyagailo [32], making the field even more active.
This article has two objectives. One is to provide the basic knowl-
edge of flat conformal structures and to serve as an introductory guide
of the field. The other is to show some new pieces of knowledge. \S 1\sim \S 3
are devoted to the former purpose, where the reader can find exposition
of fundamental properties of Moebius transformations and flat confor-
mal structures. No original results are included in these early sections.
However for the full understanding of later sections, they are helpful, or
even indispensable.
\S 4 and \S 5 are also mainly expository, though they include some
slightly improved (new) results. Hereafter let $N$ be a connected closed
flat conformal manifold of dimension . In \S 4, we prove the following
$\geq 3$

version of Frieds theorem.

Theorem (4.4). If the holonomy group of $N$ has fifixed point $a$

in , then $N$ is either

$S^{n}$
, an Euclidean space form or a Hopf
$S^{n}$

manifold.
Unlike the original theorem ([13]), we no longer postulate that the
developing map misses the fixed point. This yields clearer understand-
ing of the limit set (5) and a wider range of applications. Using Theo-
rem (4.4), various results (mostly known) can be proved by elementary
and straightforward arguments. Although the proof of Theorem (4.4)
is nothing but a small modification of the argument in [13], it might be
worth while to record it. The same result was obtained independently
by R. Miner [58], who mainly worked in the context of spherical $CR$
structures.
In \S 5, we define the limit set $L(N)$ of a flat conformal manifold $N$ .
Five dierent ways are possible and in Theorem (5.18), they are shown
to coincide eventually. Especially we get that the limit set defined by
means of the holonomy group is identical to the one obtaind by looking
at the behaviour of the developing map. (Most of these facts are already
known to Kulkarni-Pinkall [43].) As immediate corollaries we have the
followings.
Corollary (5.23). If the developing $rr_{v}ap$
of $N$ is not onto $S^{n}$
,
then it is a covering map onto its image.
Corollary (5.24). Suppose the following (1) and (2).
 Flat Conformal Structure 169

(1) $S^{n}\backslash L(N)$ is connected and the fundamental group $\pi_{1}(S^{n}\backslash $

$L(N))$ is fifinitely generated.

(2) For any point $x\in L(N)$ , there exists an arbitrarily small
neighbourhood $U$
of $x$ such that $U\backslash L(N)$ is connected.
Then the developing map is a covering map onto $S^{n}\backslash L(N)$ .

In dimension 2, Corollary (5.23) is well known and easy to show

using hyperbolic metric. For higher dimension, it was first proved by
Kamishima. Again our method is short and straightforward. Corol-
lary (5.24) can be found in Kulkarni-Pinkall [45], where condition (2)
is mistakingly dropped. In \S 5, we also characterize those flat conformal
manifolds whose developing maps are covering maps (onto the images)
and whose holonomy groups are indiscrete. (Theorem (5.26).) In dimen-
sion 3, this was first obtained by Kamishima ([24]) and independently
by $Gusevski\dot{i}$-Kapovich ([20]) in dimension 3.
$N$ is called elementary if the limit set is finite. $N$ is called a
-structure if it is a connected sum of elementary structures and is not
$C$

itself elementary. In dimension 3, we have the following result.

Theorem (6.12). Suppose $dim(N)=3$ . Then $N$ is a $C$

-structure
if and only if the limit set $L(N)$ is a tame Cantor set.

Recall that a Cantor set in is called tame if there exists a

$\wedge r$ $S^{n}$

self homeomorphism of which carries

$S^{n}$
into . Otherwise it is $\prime r$ $S^{1}$

called wild.
The above theorem is proved along the argument of Kulkarni ([43]),
in which Stallings theorem ([54], [55]) concerning ends of groups plays
a central part. The theory of ends are summarized in the appendix for
the convenience of the reader.
After preparing Poincar\es polyhedral theorem in \S 7 (in the frame-
work of flat conformal manifolds), we shall show the following theorem
in \S 8.

Theorem (8.1). There exists $a$

flflat conformal manifold $N$
of
dimension 3 whose limit set $L(N)$ $\dot{\iota}s$

a wild Cantor set.

This theorem is an improvement of the work of Bestvina-Cooper

([4]) who constructed such examples for open 3-manifolds. Our example
in Theorem (8.1) is compact.
Literature concerning flat conformal structures is extensively col-
lected in the reference, though not complete, of course.
170 S. Matsumoto

The author wishes to express his thanks to Y. Kamishima, S.

Kojima, W.M. Goldman, $N.A$ . $Gusevski\dot{i}$ , $L.D$ . Potyagailo and $M.E$ .
Kapovich for helpful conversations.

CONTENTS

\S 1. Conformal map and Liouvilles theorem $\ldots$ $\ldots\ldots\ldots.$ . 170

\S 2. More on Moebius transformation $\ldots$ $\ldots\ldots\ldots\ldots$ $\ldots\ldots$ 177

\S 3. Flat conformal structure $\ldots\ldots$ $\ldots\ldots\ldots$ $\ldots\ldots\ldots$ $\ldots.$ . 189

\S 4. Closed similarity manifolds $\ldots$ $\ldots\ldots$ $\ldots\ldots\ldots\ldots$ $\ldots\ldots$ 202

\S 5. Limit set $\ldots\ldots$ $\ldots\ldots\ldots\ldots\ldots\ldots\ldots\ldots$ $\ldots\ldots$ $\ldots..$

,. . 215

\S 6. Elementary structure and $C$

structure $\ldots\ldots\ldots$ $\ldots\ldots$ . 229

\S 7. Poincar\e polyhedron theorem $\ldots\ldots$ $\ldots\ldots\ldots\ldots\ldots J\ldots$ 234

\S 8. Wild Cantor set as limit set $\ldots$ $\ldots\ldots\ldots$ $\ldots\ldots\ldots$ $\ldots.$ . 239

Appendix End $\ldots\ldots$ $\ldots\ldots\ldots\ldots$ $\ldots\ldots\ldots\ldots$ $\ldots\ldots$ $\ldots.$ . 256

\S 1. Conformal map and Liouvilles theorem

In this section, we give definitions of a conformal map and a Moebius
transformation of the -sphere. After providing fundamental properties,
$n$

we show that a locally defined conformal map is the restriction of a

Moebius transformation if $n\geq 3$ . (Liouvilles theorem.)

Definition (1.1). A real matrix $A$ is called a

$n\times n$
conformal
matrix if $A=\lambda P$ for $\lambda>0$
and an orthogonal matrix $P$ .

Thus $A$ is conformal precisely when $A$ preserves the angle of given
two vectors. Notice that the products and the inverses of conformal
matrices are again conformal.
Let $\hat{R}^{n}=R^{n}\cup\{\infty\}$ be the one point compactification of . $R^{n}$

Points in is indicated by letters

$\hat{R}^{n}$

, and so forth. For $x=$ $a$ $x$

$(x_{1}, \ldots, x_{n})\in R^{n}$ ,

$|x|=(\sum_{\dot{\iota}=1}^{n}x_{i}^{2})^{1/2}$
 Flat Conformal Structure 171

denotes the Euclidean norm of . To endow $x$

$\hat{R}^{n}$

the structure of an
oriented manifold, the following local charts $(U_{\dot{x}}, q_{i})$
are commonly used
$(i=1,2)$ .
$U_{1}=R^{n}$ , $q_{1}=id:U_{1}\rightarrow R^{n}$ ,
$\{$

$U_{2}=\hat{R}^{n}\backslash \{0\}$
, $q_{2}$ : $U_{2}\rightarrow R^{n}$ ,
where $q_{2}$ is defined by

$q_{2}(x_{1}, \ldots, x_{n})=\frac{1}{|x|^{2}}(x_{1}, \ldots, x_{n-1}, -x_{n})$ .

In the above definition and in all that follows, if the image of by a $\infty$

map is clear by the continuity, we do not explicitly state it. An important

property of is that the dierential matrix
$q_{2}$ at any point $D_{a}q_{2}$

is a conformal matrix. Verification is left to the reader.

$a\in R^{n}\backslash \{0\}$

Let $U$ be a domain (i.e. a connected open subet) of $\hat{R}^{n}$

Definition (1.2). A map : is called a conformal $C^{1}$

$f$
$U\rightarrow\hat{R}^{n}$

map if the following condition is satisfied. For any $a\in U$ , if $a\in U_{i}$
and $f(a)\in U_{j}$ , then the dierential is a conformal $D_{q_{i}(a)}(q_{j}\circ f\circ q_{i}^{-1})$

matrix.

Since for any , is a conformal matrix,

$b\in R^{n}\backslash \{0\}$ $D_{b}(q_{2}\circ q_{1}^{-1})$

Definition (1.2) is invariant under possible changes of local charts around

$a$and $f(a)$ . A conformal map is a submersion and thus has a local
inverse, which is again a conformal map. Also the composite of two
conformal maps is conformal.

Lemma (1.3). Suppose $f$ : $U\rightarrow\hat{R}^{n}$

is a $C^{1}$
submersion, where
$U$ is a domain of $\hat{R}^{n}$

If $D_{a}f$ is a
conformal matrix for any $ a\in$

$U\cap R^{n}\cap f^{-1}(R^{n})$ , then $f$ is a conformal map.

Proof. This follows at once from the fact that the conformal ma-
trices form a closed subset in the general linear group. Q.E.D.

Let us give examples of conformal maps. Let $0<p<n$ . By

a dimension sphere in , we mean either a dimension metric
$p$
$\hat{R}^{n}$

$p$

sphere in or a dimension plane in

$R^{n}$
plus . A dimension $p$
$R^{n}$ $\{\infty\}$

sphere is sometimes called a codimension $-p$ sphere.

$p$ $n$

Definition (1.4). Let $\sigma$

be a codimension one sphere in $\hat{R}^{n}$

The inversion at $\sigma$

$J_{\sigma}$
: $\hat{R}^{n}\rightarrow\hat{R}^{n}$
172 S. Matsumoto

is defined as follows.
(1) If $\sigma$
is the sphere of radius $r$
centered at $a$ , then for any
,
$x\in R^{n}\backslash \{a\}$

$J_{\sigma}(x)=\frac{r^{2}}{|x-a|}(x-a)+a$ .

(2) If contains a codimension one plane,

$\sigma$
$J_{\sigma}$
is the reflexion at
that plane.

See Figure (1.1). The inversion is an orientation reversing involution

with the fixed point set . $\sigma$

$)=r^{2}$

Figure (1. 1)

Definition (1.5). Composite of inversions is called a Moebius

transformation. The group of all the Moebius transformations of
$\hat{R}^{n}$

is denoted by . $\mathcal{M}(\hat{R}^{n})$

Proposition (1.6). Moebius transformation is a conformal map

and cani es a sphere in $\hat{R}^{n}$

to a sphere of the same dimension.

Proof Computaion shows that an inversion is a conformal map.

Also it is well known, very easy to show by Euclidean geometry, that an
inversion maps a codimension one sphere to a codimension one sphere.
Therefore a sphere of arbitrary dimension, the intersection of several
 Flat Conformal Structure 173

codimension one $sphereS_{)}$ is mapped to a sphere of the same dimension.

The proposition follows from this. Q.E.D.
Proposition (1.7). The following maps are Moebius transforma-
tions.
(a) Translation by $a$ , $x\mapsto x+a$ .

(b) Magnifification by $\lambda>0$

, $x\mapsto\lambda x$
.
(c) Orthogonal transformation by $P\in O(n)$ , $x\mapsto Px$ .

Proof. Translation is the composite of two inversions at parallel

planes. This shows (a). Likewise positive magnification is the composite
of two inversions at concentric spheres and orthogonal transformation is
the composite of several inversions at planes through 0, showing (b) and
(c). Q.E.D.

Lemma (1.8). Let $f$ : be a Moebius

$\hat{R}^{n}\rightarrow\hat{R}^{n}$

transformation.
If $f(0)=0$ , $ f(\infty)=\infty$ , $D_{0}f=E$ , then $f=id$ .

Proof. Moebius transformations carry circles to circles. Since $f$

keeps 0 and fixed, preserves the (singular) dimension one

$\infty$ $f$

foliation formed by the straight lines through 0. Since

$\mathcal{L}$

is a $f$

conformal map, also preserves the codimension one foliation

$f$ of $\mathcal{L}^{\perp}$

spheres centered at 0. See Figure (1.2). Notice also that keeps the $f$

leaf of invariant, since $D_{0}f=E$ . Thus we obtain

$\mathcal{L}$

$f(x)=\frac{R}{r}x$ .

on the sphere $|x|=r$ . The conformality of $f$

implies

$\frac{dR}{dr}=\frac{R}{r}$
.

Therefore we have $R=ar$ . But $a=1$ since $D_{0}f=E$ . This shows

$f=id$ . Q.E.D.

Proposition (1.9).
(1) $f$ is a Moebius transformation such that $ f(\infty)=\infty$
if and
only if
$f(x)=Ax+b$ .
(2) $f$
is a Moebius transformtion such that $ f(\infty)\neq\infty$
if and only
if
$f(x)=AJ(x-b)+c$ .
174 S. Matsumoto

Figure (1.2)

Here $A$ is a conformal matrix, $b$

and $c$
are points of $R^{n}$
and $J$

is the inversion at the unit sphere $\{|x|=1\}$ .

Proof. It is a direct consequence of Lemma (1.7) that the trans-

formations of the above expressions are Moebius transformations. Con-
versely suppose that $f$
is a Moebius transformation with $ f(\infty)=\infty$ .
Let $f(0)=b$ and $D_{0}f=A$ . Define $g(x)=Ax+b$ . Then $g^{-1}of$
satisfies the hypothesis of Lemma (1.8). Thus $g=f$ . This completes
the proof of (1). On the other hand, suppose that is a Moebius $f$

transformation with $ f(\infty)\neq\infty$ . Let $ f(b)=\infty$ . Define by $h$

$h(x)=J(x-b)$ . Then is a Moebius transformation which

$f\circ h^{-1}$

keeps fixed. By (1), we have

$\infty$

$f\circ h^{-1}(x)=Ax+c$ .

This completes the proof of (2). Q.E.D.

We shall finish this section with the following celebrated theorem of

Liouville.
Theorem (1.10). Let $n\geq 3$ . Suppose $f$ : $U\rightarrow\hat{R}^{n}$
is $a$

conformal map, where $U$ isa domain of Then is the

$\hat{R}^{n}$
$f$

restriction of a Moebius transformation.

 Flat Conformal Structure 175

As a matter of fact, this theorem does not hold for n $=2$ . In

fact the Riemann mapping theorem asserts the abundance of conformal
maps which are not restrictions of Moebius transformations.
Theorem (1.10) was first proved by J. Liouville in his 1850 paper
([46]), under the additional assumption that be of class . Since $f$
$C^{3}$

then, it had been an open problem, astonishingly dicult, to weaken

the dierentiability assumption, until at last in 1969, P. Hartman gave
a complete proof for maps ([21]). $C^{1}$

Independently, $F.W$ . Gehring, among others, developed the the-

ory of quasiconformal maps in dimension . Specifically he defined $\geq 3$

1-quasiconformal maps, which is a genaralization of conformal maps,

where no dierentiability assumption is made. In [14], Gehring showed
that a locally defined 1-quasiconformal map is the restriction of a Moe-
bius transformation.
However these results need involvement in deep general treatment
and cannot be collected here. Instead, we give a simple elementary
proof essentially due to R. Nevanlinna ([49]) assuming that the given
conformal map is . (Nevanlinna postulated that
$f$
$C^{3}$
is ) $f$ $C^{4}.$

Proof of Theorem (1.10). We use the following convention. $x_{i}$

denotes the -th coordinate of

$i$
and for : , , $R^{n}$ $f$
$U\rightarrow\hat{R}^{n}$
$f_{x_{i}}$ $f_{x_{i}x_{j}}$

and so forth denote the first and the second partial derivatives and so
forth. They are vectors of . In the first place, since is conformal,
$R^{n}$ $f$

we have
, $(f_{x_{i}}, f_{x_{j}})=r^{2}\delta_{ij}$

where $r(x)=||D_{x}f||$ is the mapping norm of the Jacobi matrix. Dif-

ferentiating by , we get for $i=j$ ,
$x_{k}$

$(f_{x_{i}x_{k}}, f_{x_{i}})=rr_{x_{k}}$

and for $i\neq j$ ,

$(f_{x_{i}x_{k}}, f_{x_{j}})+(f_{x_{i}}, f_{x_{j}x_{k}})=0$ .

For mutually distinct indices $i,j$ and $k$

, by permuting the indices, we
have
$(f_{x_{i}x_{k}}, f_{x_{j}})=0$ .

Since can be any index except

$j$ $i$
and $k$
and $f_{x_{1}}$
, $\ldots$ , $f_{x_{n}}$
are mutually
orthogonal, we have
$f_{x_{i}x_{k}}=\mu f_{x_{i}}+l/f_{x_{k}}$ ,
where
$\mu=(f_{x_{i}x_{k}}, f_{x_{i}})/r^{2}=r_{x_{k}}/r$
176 S. Matsumoto

$lJ$ $=r_{x_{i}}/r$ .

Letting $\rho=1/r$ , we have

$\rho f_{x_{i}x_{k}}+\rho_{x_{i}}f_{x_{k}}+\rho_{x_{k}}f_{x_{i}}=0$ .

Dierentiating by $x_{j}$ , we obtain

$\rho f_{x_{i}x_{j}x_{k}}+\rho_{x_{j}}f_{x_{i}x_{k}}+\rho_{x_{i}}f_{x_{j}x_{k}}+\rho_{x_{k}}f_{x_{i}x_{j}}$

$+\rho_{x_{i}x_{j}}f_{x_{k}}+\rho_{x_{j}x_{k}}f_{x_{i}}=0$ .

By permutation of the indices, we obtain for $j\neq k$ ,

$\rho_{x_{j}x_{k}}=0$ .

By rotating the coordinates by 45 degrees in the $(x_{j}, x_{k})$ -plane, we have

$\rho_{x_{j}x_{j}}=\rho_{x_{k}x_{k}}$ .

Now since for any $k\neq j$ is constant on the hyperplane

$\rho_{x_{j}x_{k}}=0$ $\rho_{x_{j}}$

$\{x_{j}=c\}$ . Thus it follows that is constant on $\{x_{j}=c\}$ . That $\rho_{x_{j}x_{j}}$

is, . . is constant in $U$ .

$\rho_{x_{1}x_{1}}=\cdots$ $=\rho_{x_{n}x_{n}}$

By composing with a suitable Moebius transformation if neces-

$f$

sary, we may assume that $O\in U$ and $ f(0)=\infty$ . Then the image by
$f$
of an arbitrarily small ball $|x|<\in$ contains $|x|>K$ for some large
$K>0$ . By the volume formula, this implies that for some $\rho(a_{m})\rightarrow 0$

sequence . On the other hand, since

$a_{m}\rightarrow 0$ for some $\rho_{x_{i}x_{j}}=2\alpha\delta_{ij}$

$\alpha>0$ , is a quadratic function on

$\rho$ , with the leading term $U\backslash \{0\}$

. Since
$\alpha|x|^{2}$
is positive valued on
$\rho$ and , we have $U\backslash \{0\}$ $\rho(a_{m})\rightarrow 0$

$\rho(x)=\alpha|x|^{2}$ .

Notice that the same value of $\rho$ is also attained by the inversion $g$

which is defined by
$g(x)=\frac{x}{\alpha|x|^{2}}$ .

Thus by the chain rule, the composite $h=gof^{-1}$ : satisfies $f(U)\rightarrow\hat{R}^{n}$

$||D_{p}h||=1$ for any $p\in f(U)\backslash \{\infty\}$ . That is, is an isometry with $h$

respect to the Euclidean metric on . This implies that $h(x)=Px+b$ $R^{n}$

for some orthogonal matrix $P$ and $b\in R^{n}$ . In fact, all that needs
proof is that is an ane transformation. But since
$h$

$(h_{x_{i}}, h_{x_{k}})=\delta_{ij}$ ,
 Flat Conformal Structure 177

by dierentiating we get

$(h_{x_{i}x_{j}}, h_{x_{k}})=0$ ,

showing that $h_{x_{i}x_{j}}=0$ . This implies that is an ane transformation. $h$

Thus and hence $h$

are the restrictions of Moebius transformations,
$f$

as is required. Q.E.D.

\S 2. More on Moebius transformation

Denote by $\mathcal{M}(\hat{R}^{n})$
the group of Moebius transformations of $\hat{R}^{n}$

Lemma (2.1). Let $f\in \mathcal{M}(\hat{R}^{n})$

and let $\sigma\subset\hat{R}^{n}$
be a codimen-
sion one sphere. Then,

$foJ_{\sigma}of^{-1}=J_{f(\sigma)}$ .

Clearly $g=foJ_{\sigma}of^{-1}oJ_{f(\sigma)}$ is an orientation preserving

Proof.
Moebius transformation which keeps points in fixed. Thus for an $f(\sigma)$

arbitrary Moebius transformation such that $h(f(\sigma))=\{x_{n}=0\}$ , we $h$

have that $k=h\circ g\circ h^{-1}$ keeps $\{x_{n}=0\}$ pointwise fixed. Especially
we obtain that $k(0)=0$ , $ k(\infty)=\infty$ and $D_{0}k=E$ since is $k$

orientation preserving. Therefore by (1.8), we obtain $k=id$ . This

shows (2.1). Q.E.D.

Let be the standard embedding, i.e.,

$\hat{R}^{n}\rightarrow\hat{R}^{n+1}$
$\iota$

:
$\iota(x_{1}, \ldots, x_{n})=(x_{1}, \ldots, x_{n}, 0)$ .

As usual is considered to beasubset of

$R\wedge n$

by . Let $\hat{R}^{n+1}$
$\iota$ $\sigma$

be an $(n -1)$ -dimensional sphere in Then the inversion $\hat{R}^{n}$

$J_{\sigma}$
:
$\hat{R}^{n+1}\rightarrow\hat{R}^{n+1}$
can be extended to the inversion
$\hat{R}^{n}\rightarrow\hat{R}^{n}$
:
$J_{\tau}$
at
the -dimensional sphere
$n$ orthogonal to $\tau$
$\hat{R}^{n}$

such that $\hat{R}^{n}\cap\tau=\sigma$

.
This yields an injection.

$ i:\mathcal{M}(\hat{R}^{n})\rightarrow$
A4 $(\hat{R}^{n+1})$
.

Again is considered to be a subgroup of

$\mathcal{M}(\hat{R}^{n})$ $\mathcal{M}(\hat{R}^{n+1})$
by $i$
.
On the other hand let

$S^{n}=\{x\in R^{n+1}| |x|=1\}$ .

178 S. Matsumoto

Let be an -dimensional sphere in which is perpendicular to

$\hat{R}^{n+1}$
$\tau$ $n$

. Since inversions are conformal maps which send spheres to spheres,

$S^{n}$

is a transformation which keeps

$J_{\tau}$
invariant. Composites of such $S^{n}$

inversions constitute a Lie group of Moebius transformations $\mathcal{M}(S^{n})$

of . Denote the inclusion by

$S^{n}$

$j$
: $\mathcal{M}(S^{n})\rightarrow \mathcal{M}(\hat{R}^{n+1})$
.

Define by
$v\in\lambda\Lambda(\hat{R}^{n+1})$
. where
$v=T\circ J_{2}\circ J_{1}$ is the reflexion $J_{1}$

at the plane $x_{n+1}=-1/2$ , $J_{2}$

is the inversion at the sphere $|x|=2$
and $T$ is the translation by $(0, \ldots, 0, 1)$ . See Figure (2.1).

Figure (2.1)

Notice that $v(\hat{R}^{n})=S^{n}$

. Define

$c_{v}$ : $\mathcal{M}(\hat{R}^{n+1})\rightarrow \mathcal{M}(\hat{R}^{n+1})$

by
$c_{v}(f)=vofov^{-1}$ .
 Flat Conformal Structure 179

Proposition (2.2). $c_{v}$ maps the subgroup $\mathcal{M}(\hat{R}^{n})$

isomorphi-
cally onto the subgroup $\mathcal{M}(S^{n})$
.

Proof. maps an -sphere perpendicular to

$v$ $n$to the n- $\tau$
$\hat{R}^{n}$

sphere which is perpendicular to

$v(\tau)$ . On the other hand it $S^{n}$

follows from (2.1) that . This shows (2.2). Q.E.D.

$c_{v}(J_{\tau})=J_{v(\tau)}$

Let
$D^{n+1}=\{x\in R^{n+1}||x|<1\}$ ,

$H^{n+1}=\{x\in R^{n+1}|x_{n+1}>0\}$ .

Proposition (2.3). We have

A4 $(\hat{R}^{n})=\{f\in \mathcal{M}(\hat{R}^{n+1})|f(H^{n+1})=H^{n+1}\}$ ,
$\mathcal{M}(S^{n})=\{f\in \mathcal{M}(\hat{R}^{n+1})|f(D^{n+1})=D^{n+1}\}$
.

Proof. By virtue of (2.2), it suces to show the statement only

for $\hat{R}^{n}$

(Notice that $v(H^{n+1})=D^{n+1}.$ ) The inclusion is clear. $\subset$

Conversely, suppose that satisfies that $f(H^{n+1})=$

$f\in \mathcal{M}(\hat{R}^{n+1})$

$H^{n+1}$
. First of all, consider the case where $ f(\infty)=\infty$ . Then by
(1.9), $f(x)=\lambda Px+b$ , where $\lambda>0$ , $P\in O(n+1)$ and $b\in R^{n+1}$ .
Since $f(R^{n})=Rn$ , we have that $b\in R^{n}$ . Further since preserves $f$

$H^{n+1}$ ,
we also obtain that

$P=\left(\begin{array}{ll}Q & 0\\0 & 1\end{array}\right)$

,

where $Q\in O(n)$ . Thus it follows from (1.7) that . The $f\in\lambda\Lambda(\hat{R}^{n})$

remaining case can easily be reduced to this case. Details are left to the
reader. Q.E.D.
We need some standard terminologies in geometry.
Definition (2.4). Two Riemannian metrics and on a $g_{1}$ $g_{2}$

manifold $M$ are said to be conformally equivalent, if there exists a

positive valued function on $M$ such that . $\mu$ $g_{2}=\mu g_{1}$

Definition (2.5). A map : of Rieman-$C^{1}$ $f$ $(M_{1}, g_{1})\rightarrow(M_{2}, g_{2})$

nian manifolds is called a Riemannian conformal map if the induced

metric is conformally equivalent to
$f^{*}g_{2}$
. $g_{1}$

Riemannian conformal maps are usually called conformal maps in

the literature. However in order to avoid confusion with Definition (1.2),
180 S. Matsumoto

we call them Riemannian conformal maps in this article. The following

three Riemannian metrics are important in what follows.
Definition (2.6). Denote by the Euclidean metric on , $g_{E}$
$R^{n}$

i.e. , by $gs$ the spherical metric on

$g_{E}=\sum_{i=1}^{n}dx_{i}^{2}$ , that is, the $S^{n}$

restriction of the Euclidean metric on $R^{n+1}$

to the submanifold $S^{n}$

and by the hyperbolic metric on

$g_{H}$ , i.e., $D^{n}$

$\underline{4\sum_{i=1}^{n}dx_{i}^{2}}$

$g_{H}=(1-|x|^{2})^{2}$
.

It is well known that has constant sectional curvature 1 and that$g_{S}$

is a complete Riemannian manifold with constant sectional

$(D^{n}, g_{H})$

curvature -1.
Proposition (2.7). Let $U$ be a domain in A map $\hat{R}^{n}$
$C^{1}$

:
$f$ is a conformal map in the sense of Definition (1.2) if and
$U\rightarrow\hat{R}^{n}$

only if $v\circ fov^{-1}$ : is a Riemannian conformal map w.r.t.

$v(U)\rightarrow S^{n}$

the spherical metric.

Proof. First notice that for a domain $U\subset R^{n}$ ,

$f$
: $(U, g_{E})\rightarrow(R^{n}, g_{E})$

is a Riemannian conformal map if and only if $D_{a}f$ is a conformal matrix

for any $a\in U$ . On the other hand, the following two maps
$v:R^{n}\rightarrow S^{n}$
,
$v\circ q_{2}$ : $R^{n}\rightarrow S^{n}$

are Riemannian conformal maps from $(R^{n}, g_{E})$ to , where $(S^{n}, g_{S})$

is the coordinate chart of

$q_{2}$ defined in \S 1. (2.7) follows from $\hat{R}^{n}$

this. Q.E.D.
Thus Liouvilles theorem can be rephrased as follows.
Let $(n\geq 3)$ be a domain. Then a Riemannian conformal
$U\subset S^{n}$

map $w.r.t$ . the spherical metric is the restriction

: $f$ $U\rightarrow S^{n}$
of $a$

transformation in . $\mathcal{M}(S^{n})$

Hereafter we focus our attention to the action of on $\lambda\Lambda(S^{n})$ $S^{n}$

and $D^{n+1}$
. Thus Moebius transformations are considered primarily as
acting on . However there are some occasions where the coordinates
$S^{n}$

of is more convenient. In what follows, frequent use will be made

$\hat{R}^{n}$

of the following lemma, which is a special case of (2.1). As before $ J\in$

$\mathcal{M}(\hat{R}^{n+1})$
denotes the inversion at $S^{n}$
.
 Flat Conformal Structure 181

Lemma (2.8). For $f\in \mathcal{M}(S^{n})$ , we have $J\circ f=f\circ J$ .

To study the action of , the transformations are classified

$\mathcal{M}(S^{n})$

according to whether they preserve or not. In the first place, we $\infty$

have the following proposition.

Proposition (2.9). For $f\in \mathcal{M}(S^{n})$ , the following statements

are equivalent.
(1) $ f(\infty)=\infty$ .
(2) $f(0)=0$ .
(3) $f$induces an isometry of . $(S^{n}, g_{S})$

(4) $f(x)=Px$ for some $P\in O(n+1)$ .

Proof.
By virtue of (2.8), We have $(1)\Leftrightarrow(2)$ . $(1)\Rightarrow(4)$ follows
from the expression of (1.9), and $(4)\Rightarrow(1)$ $(4)\Rightarrow(3)$ is clear $and(3)Q.ED$
follows from the next lemma.
$\Rightarrow(4)$

Lemma (2.10). Suppose that a Lie group $G$ acts on a connected

-dimensional Riemanniln an manifold $N$ transitively and isometrically.
$n$

Suppose also that the fifirst derivative gives an isomorphism $G_{x}\cong O(n)$ ,
where is the isotropy subgoup at some $x\in N$ .
$G_{x}$
Then $G$ is
precisely the group of all the isometries of $N$ .

For any isometry , there exists a unique element $g\in G$

Proof. $f$

such that $g^{-1}\circ f(x)=x$ and $D_{x}(g^{-1}\circ f)=E$ . Then $g^{-1}of$ keeps
any point on any geodesic ray at fixed. That is, $g^{-1}\circ f=id$ . $x$

Q.E.D.

Next for with $ f(\infty)\neq\infty$ , we define the isometric sphere and

$f$

use it to describe a geometric decomposition of . For an matrix $f$ $n\times n$

$A$
, $||A||$ denotes the mapping norm. In particular if $A$
is a conformal
matrix, then we have $||A||=(\det A)^{1/n}$ .

Definition (2.11). For a transformation $f\in \mathcal{M}(\hat{R}^{n+1})$

with
, the isometric sphere $I(f)$ of
$ f(\infty)\neq\infty$ is defined by $f$

$I(f)=\{x\in R^{n+1}|||D_{x}f||=1\}$ .

The isometric sphere cannot be defined for transformations which

keep fixed. Recall that by (1.9),
$\infty$
can be expressed as $f$

$f(x)=\lambda PJ(x-b)+c$ ,
182 S. Matsumoto

where $\lambda>0$ , $P\in O(n+1)$ and , $c\in R^{n+1}$ . $b$

Note that $ f(b)=\infty$

and $f(\infty)=c$ . For $x\in R^{n+1}$ , we have

$||D_{x}f||=\frac{\lambda}{|x-b|^{2}}$ .

Thus the isometric sphere $I(f)$ is the codimension one sphere of radius
, centered at
$\lambda^{1/2}$
. We summarize fundamental properties of
$f^{-1}(\infty)$

isometric sphere in the following proposition. The proof is left to the

reader.

Proposition (2.12). For $f\in \mathcal{M}(\hat{R}^{n+1})$

such that $ f(\infty)\neq\infty$ ,
we have the following.
(1) The center of the isometric sphere $I(f)$ is the point . $f^{-1}(\infty)$

(2) carries $I(f)$ to $I(f^{-1})$ and induces an isometry there. In

$f$

particular, $I(f)$ and $I(f^{-1})$ have the same radius.

(3) cames the interior of $I(f)$ to the exterior of $I(f^{-1})$ .
$f$

(4) The interior of the isometric sphere $I(f)$ consists precisely of

those points for which $||D_{x}f||>1$ holds.
$x$

Proposition (2.13). For $f\in \mathcal{M}(S^{n})$ such that $ f(\infty)\neq\infty$ , the

isometric sphere $I(f)$ is perpendicular to $S^{n}$
.

Since the action of

Proof. on is not an isometry, there
$f$ $S^{n}$

are points in where the norms of the derivatives of

$S^{n}$
are less $f$

than or greater than 1. This implies that $I(f)$ intersects in an $S^{n}$

$(n-1)$ sphere. induces an isometry from $I(f)$ to $I(f^{-1})$ which

$f$

sends the sphere $I(f)\cap S^{n}$ to the sphere $I(f^{-1})\cap S^{n}$ Thus for
$x\in I(f)$ , the spherical distance in $I(f)$ between and $I(f)\cap S^{n}$ $x$

coincides with the spherical distance in $I(f^{-1})$ between $f(x)$ and

$I(f^{-1})\cap S^{n}$ . That is, for $x\in I(f)$ , we have $|x|=|f(x)|$ and
consequently $||D_{x}J||=||D_{f(x)}J||$ . See Figure (2.2). Dierentiating
the equation $J\circ f=f\circ J$ , we obtain that $||D_{x}f||=1$ implies
$||D_{J(x)}f||=1$ . That is, $J(I(f))=I(f)$ . This shows (2.13). Q.E.D.

Proposition (2.14). A transformation $f\in \mathcal{M}(S^{n})$ such that

$ f(\infty)\neq\infty$ can be decomposed as

$f=J_{\pi(f)}oJ_{I(f)}oP(f)$ ,

where $P(f)$ is a transformation in $O(n+1)$ which preserves $I(f)$ and

$\pi(f)$ is the bisector of the centers of $I(f)$ and $I(f^{-1})$ if $I(f)\neq I(f^{-1})$
 Flat Conformal Structure 183

$x)$

$I(f^{-1})$

Figure (2.2)

and an arbitrary hyperplane which passes through the center of $I(f)$ and
0 if $I(f)=I(f^{-1})$ . See Figure (2.3).

Proof.The transformation $g=J_{\pi(f)}\circ J_{I(f)}$ clearly carries $I(f)$ to

$I(f^{-1})$ and there the norm of the dierential is 1. That is, $I(g)=I(f)$
and $I(g^{-1})=I(f^{-1})$ . It follows that $g^{-1}of$ preserves the sphere
$I(f)$ and is an isometry there. Notice also that preserves the $g^{-1}\circ f$

interior of $I(f)$ . Applying (2.9) to a transformation of $I(f)$ , it follows

that $g^{-1}of=P(f)$ keeps fixed. Since $P(f)$ preserves$\infty$
, $P(f)$ $S^{n}$

is a transformation in $O(n+1)$ . Q.E.D.

It is a well known fact that is a Lie group of dimension $\mathcal{M}(S^{n})$

$\frac{1}{2}(n+1)(n+2)$ with two connected components.

Definition (2.15). Let be a sequence of elements of $\{f_{k}\}_{k=1,2},\ldots$

. We say
$\mathcal{M}(S^{n})$
if and only if for any compact subset $C$ of
$ f_{k}\rightarrow\infty$

, there exists $k_{0}>0$ such that

$\mathcal{M}(S^{n})$
for . $f_{k}\not\in C$ $k\geq k_{0}$

Thus $ f_{k}\rightarrow\infty$
if and only if $f_{k}$
has no subsequence which converges
to an element of $\mathcal{M}(S^{n})$
.
For $f\in \mathcal{M}(S^{n})$ , we define
$||Df||_{S^{n}}=\sup\{||D_{x}f|||x\in S^{n}\}$ .
184 S. Matsumoto

(f)

Figure (2.3)

Proposition (2.16). For a sequence $\{f_{k}\}$ in $\mathcal{M}(S^{n})$

, the fol-
lowing conditions are equivalent.
(1) $ f_{k}\rightarrow\infty$
.
(2) $||Df_{k}||_{S^{n}}\rightarrow\infty$
.
(3) Except for fifinite k, $ f_{k}(\infty)\neq\infty$ and radius $I(f_{k})\rightarrow 0$ .

Proof. First we shall show the equivalence of (2) and (3). Assume
for simplicity that for any . Let
$ f_{k}(\infty)\neq\infty$ $k$

$f_{k}(x)=r_{k}^{2}P_{k}J(x-b_{k})+c_{k}$ .

We have
$||D_{x}f_{k}||=\frac{r_{k}^{2}}{|x-b_{k}|^{2}}$
,

where $r_{k}=$ radius $I(f_{k})$ . Since $I(f_{k})$ is perpendicular to $S^{n}$

, we
obtain
$||Df_{k}||_{S^{n}}=\frac{r_{k}^{2}}{(\sqrt{1+r_{k}^{2}}-1)^{2}}=\frac{(\sqrt{1+r_{k}^{2}}+1)^{2}}{r_{k}^{2}}$ .

See Figure (2.4). From this follows the equivalence of (2) and (3).
Next, is obvious. To show the converse, we assume that
$(2)\Rightarrow(1)$

(2), hence (3), does not hold and will show that (1) fails, that is, has $f_{k}$
 Flat Conformal Structure 185

$)$

Figure (2.4)

a subsequence which converges in . Thus in the course of the $\lambda\Lambda(S^{n})$

proof, we are free to pass to a subsequence, if necessary. If $ f_{k}(\infty)=\infty$

for infinitely many , then such belongs to a compact subgroup

$k$ $f_{k}$

$O(n+1)$ of , showing that (1) does not hold. Therefore we may

$\mathcal{M}(S^{n})$

assume (passing to a subsequence) that for any $k\geq 1$ $ f_{k}(\infty)\neq\infty$

and for some $ 0<\rho\leq\infty$ .

$ r_{k}\rightarrow\rho$

Assume for a while that $ 0<\rho<\infty$ . Then in the decomposition

of (2.14), the sphere may be assumed to converge. That is, the
$I(f_{k})$

inversion converges in
$J_{I(f_{k})}$ . Likewise we may assume that $\mathcal{M}(S^{n})$

and $P(f_{k})$ also converge in A4

$J_{\pi(f)}k$ . This shows that (1) does $(S^{n})$

not hold.
Next consider the case where . Notice that if and $\rho=\infty$ $\rho=\infty$

only if , since the sphere

$ f_{k}^{-1}(\infty)\rightarrow\infty$
centered at $I(f_{k})$ $f_{k}^{-1}(\infty)$

is always perpendicular to the fixed sphere . Take an arbitrary $S^{n}$

transformation of such that $ g(b)=\infty$ for some

$g$ $\mathcal{M}(S^{n})$ $ b\neq\infty$

and consider the sequence $f_{k}og$ . Then . That is, $g^{-1}of_{k}^{-1}(\infty)\rightarrow b$

radius $I(f_{k}\circ g)\rightarrow r(0<r<\infty)$ . Therefore this case can be reduced

to the former case. Q.E.D.

Next we shall show that a Moebius transformation in in- $\mathcal{M}(S^{n})$

duces an isometry of $(D^{n+1}, g_{H})$ . The key step is the following lemma.
186 S. Matsumoto

Lemma (2.17). Let $f\in \mathcal{M}(S^{n})$ and let $x\in R^{n+1}\backslash S^{n}$ . Then

$||D_{x}f||=\frac{1-|f(x)|^{2}}{1-|x|^{2}}$

Both hand sides decompose as products when

Proof. decom- $f$

poses as a composite. Thus it is sucient to show (2.17) only for the

inversion at an -dimensional sphere $\tau=\{|x-a|=r\}$ which is
$J_{\tau}$
$n$

perpendicular to . We have $S^{n}$

$J_{\tau}(x)=r^{2}\frac{(x-a)}{|x-a|^{2}}+a$

and
$||D_{x}J_{\tau}||=\frac{r^{2}}{|x-a|^{2}}$ .

Since the sphere $\tau$

is perpendicular to $S^{n}$
, we have

$|a|^{2}=1+r^{2}$ .

Then it is easy to show by calculation that

$|J_{\tau}(x)|^{2}-1=\frac{r^{2}}{|x-a|^{2}}(|x|^{2}-1)$ .

This shows (2.17). Q.E.D.

Corollary (2.18). An element $f\in \mathcal{M}(S^{n})$ induces an isometry

of $(D^{n+1}, g_{H})$ .

The converse can also be shown using (2.10), once we establish the
following lemma.

Lemma (2.19). For any point $a\in D^{n+1}$ , there exists a trans-
formation $f\in \mathcal{M}(S^{n})$ such that $f(O)=a$ .

Let
Proof. be the radius through . For any $x\in l$ , let
$l$
$a$ $\sigma_{x}$

be the codimension one sphere perpendicular to at and orthogonal $l$

$x$

to . Then
$S^{n}$
sends 0 to some point in . Clearly we
$J_{\sigma_{x}}\in \mathcal{M}(S^{n})$ $l$

have
, , $\lim_{x\rightarrow 0}J_{\sigma_{x}}(0)=0$ $\lim_{x\rightarrow b}J_{\sigma_{x}}(0)=b$

where $b$
is the end point of . By the continuity of $l$
$J_{\sigma_{x}}(0)$
, we obtain
a point $x$ in such that $J_{\sigma_{x}}(O)=a$ .
$l$

Q.E.D.
 Flat Conformal Structure 187

Theorem (2.20). A4 $(S^{n})$ is precisely the group of isometrics of

$(D^{n+1}, g_{H})$ .

Theorem (2.21). In , the geodesies are the circles

$(D^{n+1}, g_{H})$

that are orthogonal to $S^{n}$

. Denoting the distance in $(D^{n+1}, g_{H})$ by
, we also have for
$d_{H}$ $a\in D^{n+1}$

$d_{H}(0, a)=\log\frac{1+|a|}{1-|a|}$ .

Proof First let us find the shortest path combining 0 and $ a(a\neq$
. Let
$0)$ be an arbitrary smooth arc such that $\gamma(0)=0$ and
$\gamma(t)$

$\gamma(1)=a$ . Schwartzs inequality yields

$||\gamma(t)||\leq|\gamma(t)|$ .

Thus we have

$1ength(\gamma)=\int_{0}^{1}\frac{2|\gamma(t)|dt}{1-|\gamma(t)|^{2}}\geq\int_{0}^{1}\frac{2||\gamma(t)||dt}{1-|\gamma(t)|^{2}}$

$\geq\int_{0}^{|a|}\frac{2ds}{1-s^{2}}=\log\frac{1+|a|}{1-|a|}$
.

This shows the last part of (2.21) and that the geodesic through 0 and
$a$are the radius.
Now consider the general case. Let , $b\in D^{n+1}$ . By (2.19), there $a$

exists such that $f(O)=a$ . Since

$f\in \mathcal{M}(S^{n})$ is an isometry, $f^{-1}$ $f^{-1}$

maps the geodesies to the geodesies. Further since is a Moebius $f^{-1}$

transformation, maps the diameter through $f(b)$ to the circle

$f^{-1}$

through and which is orthogonal to

$a$ .
$b$
Q.E.D. $S^{n}$

Finally we shall classify transformations in according to

$\mathcal{M}(S^{n})$

its dynamics on $C1(D^{n+1})$ . By (2.3), they keep $C1(D^{n+1})$ invariant,

where Cl denotes the closure.

Proposition (2.22). Let $f\in\sqrt{}\vee 1(S^{n})$ . For the induced transfor-

mation
: $f$ $C1(D^{n+1})\rightarrow C1(D^{n+1})$ ,
we have the followings.
(1) $f$has at least one fifixed point in $C1(D^{n+1})$ .
(2) If has three or more fifixed points in
$f$ , then $S^{n}$ $f$ has $a$
fifixed
point in $D^{n+1}$
.
188 S. Matsumoto

(1) follows from Brouwers fixed point theorem. To show

Proof.
(2), coordinates of and $H^{n+1}$ are more convenient. By conjugating
$\hat{R}^{n}$

$g=c_{v}^{-1}(f)\in\Lambda 4(\hat{R}^{n})$

by a suitable element of , we may assume that keeps fixed 0,

$\mathcal{M}(\hat{R}^{n})$
$g$

and another point . By (1.9), we have for $x\in R^{n}$ , $g(x)=\lambda Px$ ,
$\infty$ $a$

where and
$\lambda>0$

$P=\left(\begin{array}{ll}Q & 0\\0 & 1\end{array}\right)$

,

where $Q\in O(n)$ . Since also keeps fixed, it follows that .

$g$ $a$ $\lambda=1$

Thus for example, $(0, y)\in H^{n+1}(y>0)$ is fixed by . This completes $g$

the proof of (2). Q.E.D.

Definition (2.23). (resp. ) is called elliptic

$f\in \mathcal{M}(S^{n})$ $\mathcal{M}(\hat{R}^{n})$

if has fixed points in

$f$ $D^{n+1}$
(resp. $H^{n+1}$
), loxodromic if is $f$

not elliptic and has exactly two fixed points in (resp. ) and $S^{n}$
$\hat{R}^{n}$

parabolic otherwise.
Notice that by (2.22), a parabolic transformation has precisely one
fixed point in (resp. ). $S^{n}$
$\hat{R}^{n}$

Next we shall describe the standard forms of conjugacy classes of

these three types of transformations. For elliptic transformations, it is
convenient to work with the coordinates of and to conjugate so that $S^{n}$

0 is the fixed point. However for the other types, the coordinates of $\hat{R}^{n}$

is preferable. Notice that parabolic (resp. loxodromic) transformations

can be conjugated so that they keep (resp. and 0) fixed. $\infty$ $\infty$

Proposition (2.24).
(1) Let be an elliptic transformation such that $f(0)=$
$f\in \mathcal{M}(S^{n})$

. Then we have $f(x)=Px$ for some $P\in O(n+1)$ .

$0$

(2) Let be a loxodromic transformation such that

$f\in \mathcal{M}(\hat{R}^{n})$

$ f(\infty)=\infty$ and $f(0)=0$ . Then we have $f(x)=\lambda Px$ for

some $>0$ and $P\in O(n)$ .
$\lambda\neq 1,$

(3) Let be a parabolic transformation such that

$f\in \mathcal{M}(\hat{R}^{n})$ $f(\infty)$

$=\infty$ . Then by conjugating with a translation of , we have $R^{n}$

$f(x)=Px+b$ for some $P\in O(n)$ and such $b\in R^{n}\backslash \{0\}$

that $Pb=b$ .

To show (2), notice that

Proof. since otherwise would $\lambda\neq 1$ $f$

fix points of the straight line perpendicular to which passes through $R^{n}$

0, contrary to the hypothesis that is loxodromic. $f$

 Flat Conformal Structure 189

To prove (3), let $f(x)=\lambda Px+b$ . Since cannot have a fixed point $f$

in , we have
$R^{n}$
and $b\not\in Image(P I)$ . But $b=(P-I)a+b$ ,
$\lambda=1$

for some $a\in R^{n}$ and Image(P -I). It is a standard exercise in

$ b\perp$

linear algebra to show $Pb=b$ . Conjugating by the translation by $f$

, we obtain the transformation

$a$ $x\mapsto Px+b$ , as is required. Q.E.D.

Definition (2.25). For a loxodromic transformation , $f\in \mathcal{M}(S^{n})$

the geodesic which combines the two fixed points of is called an axis $f$

of . $f$

Definition (2.26). A codimension one sphere in $C1(D^{n+1})$ which

is tangent to at $a\in S^{n}$ is called a horosphere at .
$S^{n}$ $a$

Proposition (2.27). A loxodromic transformation of $\mathcal{M}(S^{n})$

preserves its axis. A parabolic transformation preserves the horospheres

at the fifixed point.

To prove the first part, notice that the standard form (2)
Proof
of (2.24) preserves the $x_{n+1}$ axis in $H^{n+1}$ . The transformation $ v\in$

(defined just before (2.2)) maps $x_{n+1}$ -axis to a diameter in

$\mathcal{M}(\hat{R}^{n+1})$

$D^{n+1}$
. Any transformation of A4 maps a diameter to a geodesic of $(S^{n})$

$D^{n+1}$
. Therefore by conjugating the standard form, we get the desired
result. The latter part can be shown likewise. Notice that the standard
form (3) of (2.24) preserves the plane $\{x_{n+1}=c\}$ $(c>0)$ , which is
mapped by to a horosphere. $v$ Q.E.D.

\S 3. Flat conformal structure

In this section we define a flat conformal structure, its developing
map and holonomy homomorphism. We study their fundamental prop-
erties.
In the first place, we define a $(G, X)$ -structure in general circum-
stances. Let $X$ be a real analytic manifold and let $G$ be a Lie group
acting real analytically, transitively and eectively on $X$ . In this study,
all the group actions are to be on the left, unless otherwise specified. Let
$N$ be a connected topological manifold of the same dimension as $X$ .

Definition (3.1). A collection $\mathcal{U}=\{(U_{\alpha}, q_{\alpha})\}_{\alpha\in\Lambda}$

is called a
$(G, X)$ -atlas if

(1) is an open covering of $N$ .

$\{U_{\alpha}\}$

(2) :
$q_{\alpha}$
is an embedding.
$U_{\alpha}\rightarrow X$

(3) For each component $V$ of $U_{\alpha}\cap U_{\beta}$

, there exists $g\in G$ such
that $q_{\beta}(x)=gq_{\alpha}(x)$ , $x\in V$ .
190 S. Matsumoto

An element $(U_{\alpha}, q_{\alpha})$

is called a $\mathcal{U}$

-chart.
Definition (3.2). A maximal $(G, X)$ -atlas is called a $(G,X)-$
structure on $N$ or a geometric structure vaguely. A manifold equipped
with a $(G, X)$ -structure is called a $(G, X)$ manifold
Let $p$ : $M\rightarrow N$
be a covering map.
Definition (3.3). Let be a $(G, X)$ -atlas on $N$ $\{(U_{\alpha}, q_{\alpha})\}_{\alpha\in\Lambda}$

for a $(G, X)$ structure such that is homeomorphic to an -ball. $\mathcal{U}$

$U_{\alpha}$
$n$

Let be a connected component of

$V_{\alpha}^{i}$
. Then is a $p^{-1}(U_{\alpha})$ $\{(V_{\alpha}^{i}, q_{\alpha}\circ p)\}$

$(G, X)$ -atlas on $M$ . The $(G, X)$ -structure which contains $\{(V_{\alpha}^{i}, q_{\alpha}\circ p)\}$

is called the lift of by and is denoted by . Especially when $\mathcal{U}$

$p$
$p^{*}\mathcal{U}$

$p$is a homeomorphism, and are called isomorphic. $p^{*}\mathcal{U}$

$\mathcal{U}$

Given a $(G, X)$ structure on $N$ , the associated developing map $\mathcal{U}$

and holonomy homomorphism are defined as follows.

Let : be the universal covering space with the base
$p$
$\overline{N}\rightarrow N$

point . Let be the fundamental group at the base point

$x_{0}\in\overline{N}$
$\pi_{1}(N)$

$p(x_{0})$ . As usual, is identified via , with the group of deck

$\pi_{1}(N)$ $x_{0}$

transformations of . Denote by the lift of by . Fix once $\overline{N}$ $\overline{\mathcal{U}}$

$\mathcal{U}$

$p$

and for all a chart around .

$\overline{\mathcal{U}}$

$(U_{0}, q_{0})$ $x_{0}$

Definition (3.4). A sequence $((U_{i}, q_{i}),$ $g_{i})$

, $(1 \leq i\leq r)$ is called
a chart chain from $(U_{0}, q_{0})$ if for $1\leq i\leq r$ , we have
(a) , $g_{i}\in G$ ,
$(U_{i}, q_{i})\in\overline{\mathcal{U}}$

(b) $U_{i-1}\cap U_{i}$ is nonempty and connected,

(c) $q_{i-1}(x)=g_{i}q_{i}(x)$ , $x\in U_{i-1}\cap U_{i}$ .

Given a chart chain as above, it is possible to extend the base map

$q_{0}$ to a continuous map $D$ : by $U_{0}\cup U_{1}\rightarrow X$

$D(x)=g_{1}q_{1}(x)$ , $x\in U_{1}$ .

Successively $D$ can be extended to $U_{0}\cup U_{1}\cup U_{2}$ by

$D(x)=g_{1}g_{2}q_{2}(x)$ , $x\in U_{2}$ .

See Figure (3.1). This motivates the following definition.

Definition (3.5).
(1) The developing map $D$ : $\overline{N}\rightarrow Xw.r.t$
. the base chart $(U_{0}, q_{0})$

is defined by
$D(x)=g_{1}g_{2}\cdots g_{r}\cdot q_{r}(x)$ , $x\in\overline{N}$
,
 Flat Conformal Structure 191

Figure (3. 1)

where , , $(1 \leq i\leq r)$ is a chart chain from

$((U_{i}, q_{i})$ $g_{i})$ $(U_{0}, q_{0})$

such that $x\in U_{r}$ .

(2) The holonomy homomorphism : $\pi_{1}(N)\rightarrow Gw.r.t$ . the base $\varphi$

chart is defined by $(U_{0}, q_{0})$

$\varphi(\xi)=h_{1}h_{2}\cdots h_{S}$ , $\xi\in\pi_{1}(N)$

where $((V_{j},p_{j})$ , $h_{j})$

, $1\leq j\leq s$ is a chart chain from $(U_{0}, qo)$

such that
$(V_{s},p_{s})=(\xi U_{0}, q_{0}\circ\xi^{-1})$ .

$D$ and are well defined since

$\phi$
is simply connected. The proof $\overline{N}$

is routine and is omitted. Also it is clear that $D$ is a submersion (or

immersion).

Definition (3.6). A pair $(D, \varphi)$ is called a $DH$ pair if the

following is satisfied.
(1) $D$ : is a submersion.
$\overline{N}\rightarrow X$

(2) $\varphi$
: is a homomorphism.
$\pi_{1}(N)\rightarrow G$

(3) $D(\xi x)=\varphi(\xi)D(x)$ , $\xi\in\pi_{1}(N)$ ,

$x\in\overline{N}$
.
192 S. Matsumoto

Proposition (3.7). Let $D$ and be the developing map and $\varphi$

the holonomy homomorphism for a base chart $(U_{0}, qo)$ . Then $(D, \varphi)$

is a $DH$ pair.

Proof. To show
$D(\xi x)=\varphi(\xi)D(x)$ , $\xi\in\pi_{1}(N)$ , $x\in\overline{N}$
,

let
$\sigma=\{((U_{i}, q_{i}), g_{i})\}_{1\leq i\leq r}$

be a chart chain from $(U_{0}, q_{0})$ such that $x\in U_{r}$ and let
$\tau=\{((V_{j},p_{j}), h_{j})\}_{1\leq j\leq s}$

be a chart chain from $(U_{0}, q_{0})$ such that

$(V_{s},p_{s})=(\xi U_{0}, q_{0}\circ\xi^{-1})$ .

Let
$\xi_{\phi}\sigma=\{((\xi U_{i}, q_{i}\circ\xi^{-1}), g_{i})\}$ .

$\xi_{\beta}\sigma$
is a chart chain from $(\xi U_{0}, q_{0}\circ\xi^{-1})=(V_{s},p_{s})$ . Thus followed $\tau$

by is a chart chain from

$\xi_{\#}\sigma$ $(U_{0}, qo)$ to the point . That is, we $\xi x$

have
$D(\xi x)=h_{1}h_{2}\cdots h_{s}\cdot g_{1}g_{2}\cdots g_{r}\cdot q_{r}o\xi^{-1}(\xi x)=\varphi(\xi)D(x)$ .
Finally let us show that $\varphi$
is a homomorphism. We have
$\varphi(\xi_{1}\xi_{2})D(x)=D(\xi_{1}\xi_{2}x)=\varphi(\xi_{1})D(\xi_{2}x)=\varphi(\xi_{1})\varphi(\xi_{2})D(x)$ .

It follows that , since the action of $G$ on $X$ is

$\varphi(\xi_{1}\xi_{2})=\varphi(\xi_{1})\varphi(\xi_{2})$

eective and real analytic. (Note that Image(D) is a domain since $D$
is a submersion.) Likewise we have $\varphi(1)=1$ . Q.E.D.
Definition (3.8). Two $DH$ pair and are said to $(D, \varphi)$ $(D, \varphi\prime)$

be equivalent if there exists $g\in G$ such that $D(x)=gD(x)$ and $\varphi(\xi)$

$=g\varphi(\xi)g^{-1}$ for and $\xi\in\pi_{1}(N)$ . $x\in\overline{N}$

Proposition (3.9). The correspondence of (3.7) gives a bijection

between the set of $(G, X)$ -structures on $N$ and the set of the equivalence
classes of $DH$ pairs.
Let Proof. (resp. ( $D$ , )) be the $DH$ pair associated to
$(D, \varphi)$ $\varphi$

the base chart $(U_{0}, qo)$ (resp. ( , ))of a given $(G, X)$ -structure. $U_{0}$ $q_{0}$
 Flat Conformal Structure 193

Consider a chart chain

$((U_{i}, q_{i})$ , $g_{i})$

, $1\leq i\leq r$

from such that

$(U_{0}, q_{0}\prime)$ $(U_{r}, q_{r})=(U_{0}, q_{o})$ . Let $g=g_{1}g_{2}\cdots g_{r}$ . Then
it is easy to show that

$D(x)=gD(x)$ , $\varphi(\xi)=g\varphi(\xi)g^{-1}$
.

Conversely given an equivalence class of $DH$ pairs, one can get a

$(G, X)$ -structure on $N$ by restricting the developing map to small
domains of and projecting down $hy$
$\overline{N}$

. Q.E.D. $p:\overline{N}\rightarrow N$

By certain abuse, $(G, X)$ -structures are sometimes denoted by their

$DH$ pairs as . $[D, \varphi]$

Definition (3.10). For a $(G, X)$ -structure $[D, \varphi]$

on $N$ ,

$H=$ Image(\varphi )\subset G

is called the holonomy group of $[D, \varphi]$

.

By (3.9), the holonomy group of a $(G, X)$ -structure is unique up to

conjugations in $G$ .
Let be a discrete group which acts on $N$ .
$\Gamma$

Definition (3.11). is said to act discontinuously on $N$ , if for

$\Gamma$

any $x\in N$ , there exists a neighbourhood $U$ of such that $x$

Card $\{\gamma\in\Gamma |\gamma U\cap U\neq\phi\}<\infty$ .

The proof of the following proposition is left to the reader.

Proposition (3.12). acts freely and discontinuously on $N$ if $\Gamma$

and only if for any $x\in N$ , there exists a neighbourhood $U$ such that
if , then $\gamma U\cap U=\phi$ .
$\gamma\neq 1$

Suppose $N\rightarrow P$ be a regular covering with the group of deck

trasformations F. Then the action of on $N$ is free and discontinuous. $\Gamma$

Conversely, if acts freely and discontinuously on a manifold $N$ , then

$\Gamma$

the canonical projection : is a regular covering with the

$\pi$ $ N\rightarrow N/\Gamma$

group of deck transformations . $\Gamma$

 194 S. Matsumoto

Proposition (3.13). Suppose that $\Gamma$

acts on $N$ freely and dis-

continuously. Then
$\overline{\Gamma}=$

{ $\overline{\gamma}$

: $\overline{N}\rightarrow\overline{N}|\overline{\gamma}$
is a lift of $\gamma$
, $\gamma\in\Gamma$
}
acts on $\overline{N}$

freely and discontinuously. is the group $\overline{\Gamma}$

of deck trans-
formations of the following universal covering.
$\pi op:\overline{N}\rightarrow N\rightarrow N/\Gamma$
.

We have the following exact sequence;

$1-\pi_{1}(N)\rightarrow\overline{\Gamma}\rightarrow\Gamma\rightarrow 1$
.

We only show that the action of

Proof. is free and discontinuous. $\overline{\Gamma}$

The rest is left to the reader. Let . Take a small neighbourhood $x\in\overline{N}$

of such that
$\overline{U}$

$x$

(1) $U=p(\overline{U})$ is evenly covered by and $p$

(2) $\gamma U\cap U=\phi$ if , . $\gamma\neq 1$ $\gamma\in\Gamma$

Suppose for . Then we have $\gamma(U)\cap U\neq\phi$ , where

$\overline{\gamma}(\overline{U})\cap\overline{U}\neq\phi$ $\overline{\gamma}\in\overline{\Gamma}$

is a lift of . This shows that $\gamma=1$ by (2). Thus

$\overline{\gamma}$

is a deck $\gamma$
$\overline{\gamma}$

transformation of . But by (1), we have . Q.E.D. $p$

$\overline{\gamma}=1$

Let $\mathcal{U}$

be a (G,$ X)$ -structure on N.

Definition (3.14). An action of $\Gamma$

on $N$ is called a $\mathcal{U}$

action
if and only if for any $\gamma\in\Gamma$
, we have $\gamma^{*}\mathcal{U}=\mathcal{U}$
.

Suppose that an action of $\Gamma$

on N is a free and discontinuous

$\mathcal{U}$

-action. As before, :N $\pi$ $\rightarrow N/\Gamma$

is the canonical projection.
Definition (3.15). A $(G, X)$ -strucure , called the projection $\pi_{*}\mathcal{U}$

of , is defined as follows. Let

$\mathcal{U}$

be the $DH$ pair associated to $(D, \varphi)$

a base chart . Since the action of the lift is a action $(U_{0}, q_{0})$
$\overline{\Gamma}$ $\overline{\mathcal{U}}$

we have that is a -chart for any . Thus as in $(\overline{\gamma}U_{0}, q_{0}\circ\overline{\gamma}^{-1})$

$\overline{\mathcal{U}}$
$\overline{\gamma}\in\overline{\Gamma}$

Definition (3.5) (2), we can define a homomorphism

$\psi$
: $\overline{\Gamma}\rightarrow G$

by using a chart chain to $(\overline{\gamma}U_{0}, q_{0}\circ\overline{\gamma}^{-1})$

. Then $(D, \psi)$ is a $DH$ pair
for . is defined to be the
$ N/\Gamma$ $\pi_{*}\mathcal{U}$
$(G, X)$ -structure corresponding to
this $DH$ pair.
Clearly $\psi$
: $\overline{\Gamma}\rightarrow G$

is an extension of the holonomy homomorphism

$\varphi$
: $\pi_{1}(N)\rightarrow G$ .
 Flat Conformal Structure 195

As is shown later, there are many examples of pair $(G, X)$ such
that the isotropy subgroup
$G_{x}=\{g\in G|gx=x\}$

is compact for any $x\in X$ . Then the corresponding $(G, X)$ -structures
have the following striking feature.
Proposition (3.16). Let $N$ be a closed $(G, X)$ -manifold. Sup-
pose the isotropy subgroup is compact for $ x\in$ X. Then the $G_{x}$

developing map $D$ : is a covering map onto X. In particular,

$N\rightarrow X$

if $X$ is simply connected, then $D$ is a homeomorphism.

Since
Proof. is compact, there exists a
$G_{x}$ -invariant, positive $G_{x}$

definite, symmetric, bilinear form on the tangent space $T_{x}X$ . Dis-

tributing it by the action of $G$ , we obtain a $G$ -invariant Riemannian
metric of $X$ . Since
$g$ is $\pi_{1}(N)$ -invariant, it projects down
$\overline{g}=D^{*}g$

to a Riemannian metric on $N$ . Therefore is complete. $\overline{g}$

For small $\in>0$ , we have that $D$ maps any -ball in $ 2\in$
$\overline{N}$

isometrically onto a -ball in $X$ . Then clearly any -ball in $X$ is

$ 2\in$ $\in$

evenly covered by $D$ . Q.E.D.

We shall raise some examples of $(G, X)$ -structures.

Example (3.17). Denote by $Isom(S^{n})$ , $Isom(R^{n})$ or $Isom(D^{n})$

the group of isometries of the Riemannian manifold $(S^{n}, gs)$ , $(R^{n}, g_{E})$
or $(D^{n}, g_{H})$ . The corresponding $(G, X)-$ structure (resp. manifold) is
called spherical, Euclidean or hyperbolic structure (resp. manifold).
Specifically, closed spherical or Euclidean manifold is called spherical
or Euclidean space form.
Notice that $Isom(S^{n})=O(n+1)$ and $Isom(D^{n})=\mathcal{M}(S^{n-1})$ .
Isom(Rn) consists of transformations, called Euclidean motions,
$x\mapsto Px+b$ , $(P\in O(n), b\in R^{n})$ .

All the three satisfy the hypothesis of (3.16). Therefore if the mani-
folds are compact, their universal covering spaces can be identified with
(if
$S^{n}$
$>1$ ),$n$ or $D^{n+1}$ . A spherical space form is isomorphic to
$R^{n}$

if $>1$ , where
$ S^{n}/\Gamma$ $n$ is a finite group of SO(n+l). The following
$\Gamma$

theorem is due to Bieberbach ([5]). A neat proof, quite short, is found

in P. Buser ([6]).
Theorem (3.18). An Euclidean space form has -torus as $n$ $a$
fifinite
covering.
The main object of our study is the following $(G, X)$ -structure.
196 S. Matsumoto

Definition (3.19). A -structure (resp. manifold) is $(\mathcal{M}(S^{n}), S^{n})$

called a flflat comformal structure (resp. manifold). A group action

preserving a flat conformal structure is called a conformal action.

There is another way to get to the same concept.

Definition (3.20). A Riemannian manifold $(N, g)$ of dimension

is called conformally flflat if for any point $x\in N$ , there exist a
$n$

neighbourhood $U$ and an embedding : such that is $f$ $U\rightarrow R^{n}$ $f^{*}g_{E}$

conformally equivalent to . $g|_{U}$

Notice that the above definition does not change if we use as a model
space $(S^{n}, gs)$ instead of $(R^{n}, g_{E})$ . In fact, they are conformally
equivalent as we saw in \S 2.
Now let be a flat conformal structure on $N$ . For each
$\mathcal{U}$ $\mathcal{U}-$

chart , there is the induced Riemannian metric

$(U_{\alpha}, q_{\alpha})$
on . $q_{\alpha}^{*}g_{S}$ $U_{\alpha}$

In a component $V$ of , we have for some $U_{\alpha}\cap U_{\beta}$

$q_{\beta}=g\circ q_{\alpha}$

. Since
$g\in \mathcal{M}(S^{n})$ is a conformal map w.r.t. , and
$g$ $g_{S}$ $q_{\alpha}^{*}g_{S}$ $q_{\beta}^{*}g_{S}$

are conformally equivalent on $V$ . Take a locally finite partition of unity

associated with the covering
$\{t_{\alpha}\}$
of -charts. The Riemannian $\{U_{\alpha}\}$ $\mathcal{U}$

metric
$g=\sum_{\alpha}t_{\alpha}q_{\alpha}^{*}g_{S}$

is a conformally flat metric.

Conversely suppose $n\geq 3$ . Let be a conformally flat metric $g$

on an -dimensional manifold $N$ . Then we have a family

$n$ $\{(U_{\alpha}, f_{\alpha})\}$

such that is an open covering of $N$ , that is an embedding

$\{U_{\alpha}\}$ $f_{\alpha}$

of into and that

$U_{\alpha}$
is conformally equivalent to . Thus
$S^{n}$ $f_{\alpha}^{*}g_{S}$
$g$

for any component $V$ of , $U_{\alpha}\cap U_{\beta}$

$f_{\beta}of_{\alpha}^{-1}|_{f\alpha}(V)$
: $f_{\alpha}(V)\rightarrow f_{\beta}(V)$

is a Riemannian conformal map in $(S^{n}, g_{S})$ . Thus by Liouvilles theo-

rem, we have that
$f_{\beta}of_{\alpha}^{-1}|_{f\alpha}(V)$ $\in \mathcal{M}(S^{n})$
.
We obtain a flat conformal structure. In summary, we have;
Proposition (3.21). Flat conformal structure on a manifold $N$
yields a conformally equivalence class of conformally flflat metrics. Fur-
ther if $n\geq 3$ , this correspondence is bijective.

For $n=2$ , the above two concepts are in fact dierent. In this
dimension, flat conformal structure is often called (complex) projective
 Flat Conformal Structure 197

structure since
$(\mathcal{M}(S^{2}), S^{2})=(PGL(2:C), CP^{1})$ ,
while conformally flat Riemannian metric corresponds to complex struc-
tures.
If $(G, X)\subset(G, X)$ and the $G$ -action
, that is, $G\subset G$ , $X\subset X$

on $X$ is the restriction of the $G$ -action on $X$ , then, as a matter of fact,
a $(G, X)$ -structure is naturally considered as a $(G, X)$ -structure. Thus
spherical manifolds, Euclidean manifolds and hyperbolic manifolds are
considered to be flat conformal manifolds. In fact we have the following
inclusions of $(G, X)$ -pairs.
(Isom(5r1), $ S^{n}$
) $\subset(\mathcal{M}(S^{n}), S^{n})$ .
$(Isom(R^{n}), R^{n})\subset(\mathcal{M}(\hat{R}^{n}),\hat{R}^{n})\rightarrow c_{v}\approx(\lambda\Lambda(S^{n}), S^{n})$ .

$(Isom(D^{n}), D^{n})\subset(\lambda\Lambda(S^{n-1}), D^{n})\subset(\lambda\Lambda(\hat{R}^{n}),\hat{R}^{n})\rightarrow c_{v}\approx(\mathcal{M}(S^{n}), S^{n})$ .

A significant feature of these examples is that the developing maps are

homeomorphisms onto their images (except the case of $(Isom(S^{1}), S^{1})$
. However for a point $a\in S^{n}$ , the isotropy group
$)$
is $\mathcal{M}(S^{n})_{a}$

not compact. (Compare that is compact for $a\in D^{n+1}.$ ) $\mathcal{M}(S^{n})_{a}$

Therefore flat conformal manifolds in general do not enjoy this kind of

good properties. In fact there are many such examples as we shall show
in what follows. We make the following definition.
Definition (3.22). Let be a flat conformal structure $\mathcal{U}=[D, \varphi]$

and let $H=$ Image(\varphi ) be the holonomy group. is said to be of $\mathcal{U}$

type 1 if $D$
is a covering map onto its image and $H$ is discrete, of
type 2 if $D$ is a covering map but $H$ is indiscrete, of type 3 if $H$ is
discrete but $D$ is not a covering map and type 4 otherwise.
Before starting the study of type 1 flat conformal structures, we
need some preparations. Let be a subgroup of At . $\Gamma$
$(S^{n})$

Definition (3.23). A subset $A\subset S^{n}$

is called $\Gamma$

-invariant if
$\gamma(A)=A$ for any . $\gamma\in\Gamma$

Definition (3.24). Let be the set of points $x\in S^{n}$ such $\Omega_{\Gamma}$

that there exists a neighbourhood $U$ of such that $\gamma U\cap U=\phi$ but $x$

for finitely many . is called the domain of discontinuity of

$\gamma\in\Gamma$ $\Omega_{\Gamma}$

$\Gamma$

is the maximal
$\Omega_{\Gamma}$ $\Gamma$

-invariant open subset of $S^{n}$

on which $\Gamma$

acts discontinuously.
198 S. Matsumoto

Definition (3.25). $\Gamma$

is called a Kleinian group if $\Omega_{\Gamma}\neq\phi$

.

Clearly we have:
Proposition (3.26). A Kleinian group is discrete in $\Lambda 4(S^{n})$ .

It is known that the converse does not hold. However we have:

Proposition (3.27). If $\Gamma$

is $ discrete\rangle$ then $\Gamma$

acts on $D^{n+1}$

discontinuously

Proof. is infinite and let

Assume . Since
$\Gamma$

is $\Gamma=\{\gamma_{n}\}$ $\Gamma$

discrete, , that is, $\gamma_{n}\not\in O(n+1)$

$\gamma_{n}\rightarrow\infty$ for but finitely many $n$

and radius . It follows that any compact subset of $D^{n+1}$ is

$I(\gamma_{n})\rightarrow 0$

outside except finite . (Recall

$I(\gamma_{n})$
) (3.27) follows $n$ $I(\gamma_{n})\cap S^{n}\neq\phi.$

from (2.12)(3). Q.E.D.

The following fact is helpful in our study of flat conformal structures
of type 1. The proof is more or less the same as (3.27). The reader will
find it in \S 5, after the definition of limit set is made.
Corollary (5.16). Suppose a discrete group admits an invari- $\Gamma$

ant open set such that is neither empty nor a singleton.

$\Omega$
$ S^{n}\backslash \Omega$

Then acts on
$\Gamma$

discontinuously $\Omega$

Flat conformal structure of type 1 is constructed as follows. Let $\Gamma$

be a Kleinian group in A4( which acts freely and discontinuously $S^{n}\grave{)}$

on a -invariant domain . The action is of course a conformal action

$\Gamma$ $\Omega$

on a flat conformal manifold . Hence the quotient manifold $\Omega$

$\Omega/\Gamma$

admits a flat conformal structure . The developing map $D$ is the $\mathcal{U}$

universal covering followed by the inclusion;

$D:\overline{\Omega}\pi\rightarrow\Omega\subset S^{n}$

and the holonomy group is . Concrete examples of this construction $\Gamma$

will be given in later sections.

Definition (3.28). The flat conformal structure (manifold) con-
structed as above is called a Kleinian structure (manifold).
Definition (3.29). Two flat conformal manifolds are called com-
mensurable if they have isomorphic finite coverings.
Proposition (3.30). Any type 1 flflat conformal compact manifold
$N$ is commensurable to a Kleinian manifold.
The proof needs the following theorem due to Selberg. See e.g.
([53]).
 Flat Conformal Structure 199

Theorem (3.31). Any fifinitely generated subgroup of $GL(n, R)$

has a torsion free subgroup of fifinite index.
As is well known, At is isomorphic to the projectivised Lorentz $(S^{n})$

group $PO(n+1,1)$ . Thus (3.31) is applicable to a subgroup of . $\mathcal{M}(S^{n})$

If the developing map $D$ is onto

Proof of (3.30). , then $D$ $S^{n}$

is a homeomorphism and $N$ is isomorphic to a spherical space form.

Likewise if $D$ misses only one point, then $N$ is isomorphic to an
Euclidean space form. Otherwise, by (5.16), the holonomy group $H$
acts on Image(D) discontinuously. Let
$\Omega=$
be a torsion free finite $\Gamma$

index subgroup of H. acts on freely. We have the following two

$\Gamma$ $\Omega$

covering maps.
, $p:\overline{N}/\varphi^{-1}(\Gamma)\rightarrow N$

$\overline{D}$

: $\overline{N}/\varphi^{-1}(\Gamma)\rightarrow\Omega/\Gamma$
.

is a finite covering since

$p$ is a finite index subgroup of $\pi_{1}(N)$ . $\varphi^{-1}(\Gamma)$

Therefore is compact and is also a finite covering.

$\overline{N}/\varphi^{-1}(\Gamma)$ $\overline{D}$

Q.E.D.

One can show by examples that Proposition (3.30) cannot be sharp-

ened in general.
Next an example of type 2 flat conformal structure is in order.

Example (3.32). Let be a conformal linear trans-

$P(x)=\lambda R_{\theta}x$

formation on ( $R^{2}$ $\lambda>0$ , $R_{\theta}$

; the rotation by ). For $t\in R$ , let $\theta$

$P^{t}(x)=\lambda^{t}R_{t\theta}(x)$ .

Let $Q$ be another conformal transformation which keeps 0 fixed such

that $Q\neq P^{t}$ for any $t\in R$ .

Let $R^{2}/Z^{2}=T^{2}$ . Define $\varphi$

: $Z^{2}\rightarrow \mathcal{M}(\hat{R}^{2})$
by $\varphi(l, m)=P^{l}Q^{m}$

and
$D:R^{2}\rightarrow\hat{R}^{2}$
for some by . Since
$D(x, y)=P^{x}Q^{y}a$ $a\in R\backslash \{0\}$

, we have is a pair. $D$ is clearly a covering map

$PQ=QP$ $DH$ $(D, \varphi)$

onto . But often $H=$ Image(\varphi ) is not discrete, for example

$R^{2}\backslash \{0\}$

when and .
$\lambda=1$ $\theta\not\in Q$

See Figure (3.2). This example cannot be generalized to higher

dimensions, since is simply connected if . However,
$R^{n}\backslash \{0\}$ $n$ $\geq 3$

in \S 5, we give examples of type 2 flat conformal compact manifolds of

dimension\geq 3 and give a characterization of such manifolds.
The following is an example of type 3 flat conformal structure.
200 S. Matsumoto

$a$

Figure (3.2)

Example (3.33). Let be a closed Riemann surface of genus $\Sigma$

, that is, a hyperbolic manifold of dimension 2. The developing map

$\geq 2$

$D$ is a homeomorphism onto a disk in . We shall alter $D$ without $S^{2}$

changing the holonomy homomorphism . Let be a simple closed $\varphi$

$\alpha$

geodesic in and let $V$ be $the\in$ -neighbouhood of

$\Sigma$

for small $\in>0$ . $\alpha$

Then a lift of $V$ in the universal covering

$\overline{V}$

is the mutually $\overline{\Sigma}\cong D^{2}$

disjoint -neighbourhood of a lift of . See Figure (3.3). $D$ is altered

$\in$ $\alpha$

inside to a new map $D$ in such a way that it coincides with $D$
$\overline{V}$

near the boundary of and it goes extra once around . Clearly

$\overline{V}$
$S^{2}$

$D$
can be constructed so that is a $DH$ pair. See Figure (3.4). $(D, \varphi)$

It is easy to show that $D$ is onto . Thus it is not a covering map. $S^{n}$

For more detail, see Goldman ([16]). The same construction is possible
for higher dimension if we start with a compact hyperbolic manifold
which admits a totally geodesic closed submanifold of codimension 1.
See Kourouniotis ([33]).
Finally an example of type 4.
Example (3.34). Prepare two copies of type 2 flat conformal man-
ifolds and constructed in Example (3.34). Inside an atlas
$N_{1}$ $N_{2}$

of
$(U_{i}, q_{i})$
, take a small disk which is mapped by
$N_{i}$
to a $V_{i}$
$q_{i}$

metric disk in . There exists an element such that

$S^{2}$ $g\in\lambda\Lambda(S^{2})$ $g$

maps to the exterior of

$V_{1}$
. Consider the connected sum $V_{2}$

$ N_{1}\# N_{2}=(N_{1}\backslash IntV_{1})\cup(N_{2}\backslash IntV_{2})/\sim$ .

 Flat Conformal Structure 201

$\tilde{\Sigma}$

Figure (3.3)

Figure (3.4)

If we chose the above identification appropriately, we obtain a continuous

map
: $(U_{1}-IntV_{1})\cup(U_{2}-IntV_{2})/\sim\rightarrow S^{2}$ .
$(g\circ q_{1})\cup q_{2}$

Using this we get in an obvious way a flat conformal structure on $N_{1}QN_{2}$ .

It is not dicult to show that the developing map of this structure is
onto and therefore is not a covering map. The holonomy group is
$S^{2}$

indiscrete since we started with type 2 examples.

The above operation, called connected sum of the structure, will

be described in more detail in \S 6.
202 S. Matsumoto

\S 4. Closed similarity manifolds

In this section we assume $n\geq 3$ and mainly work with , instead $\hat{R}^{n}$

of . As before
$S^{n}$
denotes the group of Moebius transforma-
$\mathcal{M}(\hat{R}^{n})$

tions of As shown in \S 1, the isotropy subgroup

$\hat{R}^{n}$

at $\mathcal{M}(\hat{R}^{n})_{\infty}$
$\infty$

consists of transformations
$f(x)=\lambda Px+b$ , $\lambda>0$ , $P\in O(n)$ , $b\in R^{n}$ .
$\lambda$

) is called the $no7m$ (resp. orthogonal part) of

(resp. $P$ and is $f$

denoted by (resp. $P(f)$ ). Clearly a transformation

$||f||$
$f\in \mathcal{M}(\hat{R}^{n})_{\infty}$

induces a transformation of . When viewed as a transformation of $R^{n}$

,
$R^{n}$
is called an Euclidean similarity. The group of Euclidean
$f$

similarities is denoted by ES(Hn). We have an isomorphism

$ES(R^{n})\approx \mathcal{M}(\hat{R}^{n})$
.

Definition (4.1). An $(ES(R^{n}), R^{n})$ -structure (manifold) is called

a similarity structure (manifold).
Euclidean space forms are examples of similarity manifolds. Other
examples are Hopf manifolds to be defined below.

Definition (4.2). A closed similarity manifold $N$ is called a Hopf

manifold if the developing map $D$ is a homeomorphism onto . $R^{n}\backslash \{0\}$

Then the holonomy group $H$ is discrete and is contained in the

isotropy subgroup $ES(R^{n})_{0}$ . By taking norm and orthogonal part, we
obtain the isomorphism
$ES(R^{n})_{0}\cong R_{+}\times O(n)$ .

$||H||=\{||f|||f\in H\}$ is nontrivial since $N$ is closed, and is discrete

since $O(n)$ is compact. Therefore it is infinite cyclic. Let $||h||(h\in H)$
be a generator. Since the kernel $\{||h||=1\}$ is finite, is a finite $\langle h^{2}\rangle$

index subgroup of $H$ . Clearly is homeomorphic to $(R^{n}\backslash \{0\})/\langle h^{2}\rangle$

$S^{n-1}\times S^{1}$
. Thus we have;

Proposition (4.3). Hopf manifold has $a$

fifinite covering which is
homeomorphic to $S^{n-1}\times S^{1}$
.

In [13], Fried has shown that these two examples of similarity mani-
folds are the only examples. That is, an arbitrary similarity manifold is
isomorphic to either an Euclidean space form or a Hopf manifold. See
 Flat Conformal Structure 203

also Kuiper ([36]). The purpose of this section is to give an improved

version of Frieds theorem. Instead of confining ourselves to similarity
manifolds, we consider flat conformal manifolds in general.

Theorem (4.4). Let $N$ be a closed flflat conformal manifold of

dimension such that the holonomy group $H$ is contained in the
$\geq 3$

subgroup . Then $N$ is isomorphic to either , $\hat{R}^{n}$

$\lambda\Lambda(\hat{R}^{n})_{\infty}$
$\dot{\iota}sotropy$

a Hopf manifold or an Euclidean space form.

One can state the original Frieds theorem as a corollary.
Corollary (4.5). Closed similarity manifold of dimension $\geq 3$
is
isomorphic to a Hopf manifold or an Euclidean space form.
What is new in Theorem (4.4) is that the developing map is allowed
to cover the point , while in the original Frieds theorem (Corollary $\infty$

(4.5) it is postulated to miss

$)$
. Although the dierence is apparently $\infty$

not significant and the proof is in fact almost the same, Theorem (4.4)
brings forth a far wider range of applications in practice (as far as flat
conformal structures are concerned). To the best knowledge of the au-
thor, (4.4) cannot be found in the literature. Therefore it is obviously
worth while to give a complete proof of (4.4).

The rest of this section is devoted to the proof of (4.4). In way of

contradiction, we assume that $N$ is isomorphic to neither of the three
structures in (4.4). Denote by $D$ the developing map, by the $\varphi$

holonomy homomorphism and by $H$ the holonomy group. The proof

consists of three steps.
Step 1. Clearly $D^{-1}(\infty)$ is discrete and invariant by the deck
transformation. Thus $N(\infty)=\pi(D^{-1}(\infty))$ is a finite set. Then
$N^{*}=N\backslash N(\infty)$ is a similariry manifold.

Definition (4.6). A domain $U^{*}\subset\overline{N}^{*}=\pi^{-1}(N^{*})$

is called a
copy of $U\subset R^{n}$ if $D|_{U^{*}}$ : $U^{*}\rightarrow U$
is a homeomorphism.

Points in are denoted by , and so forth and their images

$\overline{N}^{*}$
$a^{*}$ $x^{*}$

by $D$ by , and so forth. Thus, $B^{*}(a^{*}, r)$ denotes a copy containing

$a$ $x$

of $B(a, r)$ , the open ball of radius

$a^{*}\in\overline{N}^{*}$
centered at . We call $r$ $a$

and
$a^{*}$
the center and radius of
$r$ $B^{*}(a^{*}, r)$ .

Definition (4.7). A closed subset $l^{*}\subset\overline{N}^{*}$

is called a complete
half line if for any copy of ball , $B^{*}\subset\overline{N}^{*}$ $B^{*}\cap l^{*}$
is mapped by $D$
to $B\cap k$ , where is a complete half line in $k$ $R^{n}$
.
204 S. Matsumoto

$R^{n}$

Figure (4.1)

See Figure (4.1).

By certain abuse, the parametrization of a complete half line $l^{*}$
is
denoted by the same letter, as
$l^{*}$
: $[0, \infty)\rightarrow l^{*}$
.

Notice that given any point and a tangent vector at ,

$x^{*}\in\overline{N}^{*}$
$v$ $x^{*}$

there exists a unique complete half line such that $l^{*}(0)=x^{*}$ and $l^{*}$

tangent to . Clearly deck transformation carries a complete half line

$v$

to a complete half line.

Definition (4.8). A complete half line $l^{*}$

is called short if $D(l^{*})$

is not a complete half line in . $R^{n}$

Claim (4.9). Given a short complete half line , there exists $l^{*}$
$a$

neighbourhood U $\pi(l^{*})\cap U=\phi$ .

of $N(\infty)$ such that

Proof. For any point $c_{i}\in N(\infty)$ , choose a compact neighbour-

hood $U_{i}$
such that
(a) ,
$\pi(l^{*}(0))\not\in U_{i}$

(b) is evenly
$U_{i}$
covered by , $\pi$

(c) For any component of , there exist $a\in R^{n}$ and

$E^{*}$ $\pi^{-1}(U_{i})$

$R>0$ such that the following map is a homeomorphism.

$D|_{E^{*}}$ : $E^{*}\rightarrow E=E(a, R)=\{|x-a|\geq R\}\cup\{\infty\}$

 Flat Conformal Structure 205

Notice that in (c), if one component of is mapped to $E(a, R)$ , $\pi^{-1}U_{i}$

then all the other is also mapped to some $E(a, R)$ . Thus (c) is attained
if one chooses small and appropriate.
$U_{i}$

Let us show that is empty. If not, the image $l\cap E$ is a

$l^{*}\cap E^{*}$

half line starting at . $( l =D(l^{*}).)$ Since is short,

$\partial E$
is not a $l^{*}$ $l$

complete half line of . Then one can choose a ball $R^{n}$ $B\subset E\backslash \{\infty\}$

centered at the point . Then $B$ has a copy in .

$\lim_{t\rightarrow\infty}l(t)$ $B^{*}$ $E^{*}$

But is not the intersection of $B$ with a complete half line in

$B\cap l$
. $R^{n}$

See Figure (4.2). This contradicts the hypothesis that is complete. $l^{*}$

Let . We have
$U=\bigcup_{i}U_{i}$$\pi(l^{*})\cap U=\phi$ . Q.E.D.

Figure (4.2)

For any , let $r(x^{*})$ be the maximal radius of a copy of

$x^{*}\in\overline{N}^{*}$

ball centered at $x^{*}$

. See Figure (4.3).

Claim (4.10). $ r(x^{*})<\infty$ .

Proof.If not, is contained in a copy of

$x^{*}$
, say $P$ . If $R^{n}$

, then $N$ would be an Euclidean space form, contradicting

$P=\overline{N}$

the hypothesis. Suppose . Take a point Fr(P) and a

$P\neq\overline{N}$ $ y^{*}\in$

sequence such that

$\{y_{n}^{*}\}\subset P$. Clearly we have . $y_{n}^{*}\rightarrow y^{*}$ $ D(y_{n}^{*})\rightarrow\infty$

It follows from the continuity of $D$ that $ D(y^{*})=\infty$ . Therefore there is

a neighbourhood $Q$ of which is mapped by $D$ homeomorphically
$y^{*}$

onto $E(0, R)$ for some large $R>0$ . Then $D$ : is a $P\cup Q\rightarrow\hat{R}^{n}$
206 S. Matsumoto

$\overline{N}^{*}$

$R^{n}$

Figure (4.3)

homeomorphism. We have

$N\cong\overline{N}=P\cup Q\cong\hat{R}^{n}$

Again a contradiction. Q.E.D.

Definition (4.11). The Fried metric is a continuous Riemannian

metric on defined by $\overline{N}^{*}$

$g_{F}=\frac{D^{*}g_{E}}{r(x^{*})^{2}}$
on $T_{x^{*}}\overline{N}^{*}$
.

Let $\xi$
be a deck transformation of $\overline{N}$

and $x^{*}\in\overline{N}^{*}$
. We have

$\xi(B^{*}(x^{*}, r))=B^{*}(\xi x^{*}, ||\varphi(\xi)||r)$ .

This shows $r(\xi x^{*})=||\varphi(\xi)||r(x^{*})$ . That is, the deck transformation is $\xi$

an isometry for the Fried metric . Thus induces a Riemannian $g_{F}$ $g_{F}$

metric of , which is also called the Fried metric. The distance

$N^{*}$

functions of Fried metrics both on and on are denoted by $\overline{N}^{*}$

$N^{*}$

$d_{F}$
.
The following is the aim of Step 1.

Claim (4.12). Let $B^{*}=B^{*}(a^{*}, r(a^{*}))$ be the maximal copy of

ball centered at . Then there exists a copy of half space
$a^{*}\in\overline{N}^{*}$ $H^{*}$

such that $B^{*}(a^{*}, r(a^{*}))\subset H^{*}$ .

Proof. For simplicity let us assume $r(a^{*})=1$ and $D(a^{*})=e_{n}=$

$(0, \cdots, 0, 1)$ . By $D$ , we identify with $B=\{|x-e_{n}|<1\}$ . By this $B^{*}$
 Flat Conformal Structure 207

identifidation, we consider the function $r$

or the Fried metric to $g_{F}$

be defined on $B$ . By the maximality of $B^{*}$

, there exists a radius $l^{*}$

which is a complete half line. Assume

$l$
$=D(l^{*})=\{x_{1}=0, \cdots, x_{n-1}=0,0<x_{n}\leq 1\}$ .

See Figure (4.4).

$n=0$

Figure (4.4)

Let us study for a while the Fried metric on $B$ . First of all for any
$x_{0}\in B$ , we have $r(x_{0})\leq|x_{0}|$ . In fact if not, the origin 0 is contained
in
$A=B\cup\{|x-x_{0}|<r(x_{0})\}$ .

has a copy containing

$A$ $a^{*}$
. This contradicts the completeness of $l^{*}$
.
Thus we have on$g_{F}\geq g_{G}$ $B$ , where

$g_{G}=\frac{g_{E}}{|x|^{2}}$ .

For any $x\in B$ , let $\theta=\theta(x)$ be the angle of the vector $\vec{0x}$

and $l$

.
We have
Subclaim (4.12.1). $ d_{F}(x, l)\geq d_{G}(x, l)=\theta$ .

Proof. Let be a smooth path in $B$ combining

$\gamma(t)$ $x$ and a
point in $l$

. Denote by lengthy(7) the length of w.r.t. $\gamma$

$d_{G}$
. Let
$\gamma(t)=|\gamma(t)|p(t)$ .
208 S. Matsumoto

We have
$|\gamma(t)|\geq|\gamma(t)||p(t)|$ .
In fact, since $|p(t)|=1$ , we have $(p(t),p(t))=0$ and
$\gamma(t)=|\gamma(t)|p(t)+|\gamma(t)|p(t)$ .

Therefore we obtain the following equality.

$ 1ength_{G}(\gamma)=\int_{0}^{1}\frac{|\gamma(t)|}{|\gamma(t)|}dt\geq\int_{0}^{1}|p(t)|dt\geq\theta$
.

On the other hand it is easy to show that for a suitable choice of , one $\gamma$

has lengthy= . $\theta$

Q.E.D.
Now by Claim (4.9), There exist a compact submanifold $N_{C}=$
$N-IntU$ which contains and a sequence such that for $\pi(l^{*})$ $ t_{i}\uparrow\infty$

some $c\in N_{C}$ ,

$d_{F}(\pi(l^{*}(t_{i})), c)\downarrow 0$ .
Also assume that $d_{F}(\pi(l^{*}(t_{1})), c)$ is suciently small. Then by (4.12.1),
there exists a point such that $c=\pi(b^{*})$ and
$b^{*}\in B^{*}$

$d_{F}(l^{*}(t_{1}), b^{*})=d_{F}(\pi(l^{*}(t_{1})), c)$ .

Now there exists a sequence $\{\xi_{i}\}$

of deck transformations such that
$d_{F}(l^{*}(t_{i}), \xi_{i}b^{*})\downarrow 0$ .

See Figure (4.5).

Thus passing to the model $B\subset R^{n}$ , we may assume the following.
Let $f_{i}=\varphi(\xi_{i})\in ES(R^{n})$ and $b=D(b^{*})\in B$ .

(1) $f_{i}(b)\in B$ .
(2) $\theta(f_{i}(b))\rightarrow 0$ .
(3) $f_{i}(b)\rightarrow 0$ .
(4) $P(f_{i})\rightarrow P_{0}\in O(n)$ .
(5) $||f_{i}||\rightarrow 0$
.
Notice that (5) follows from (3) since

$||f_{i}||=\frac{r(f\cdot(b))}{r(b)}.\leq\frac{|f_{i}(b)|}{r(b)}\rightarrow 0$
.

See Figure (4.6).

Now for $i>>1$ , taking $j>>i$ , we may assume
(6) $P(f_{i}f_{j}^{-1})$ is very near $E$ ,
(7) $||f_{i}f_{j}^{-1}||$
is very large.
 Flat Conformal Structure 209

Figure (4.5)

$e_{n}$

$x_{n}=0$
$0$

Figure (4.6)

Next by (2), is almost parallel to and is almost perpen-

$\vec{0f_{j}(b)}$ $\vec{e_{n}}$

dicular to . Applying , we still have that

$\partial B$
$f_{i}f_{j}^{-1}$

$\frac{f_{i}f_{j}^{-1}(0)f_{i}(b)\vec{}}{f_{i}f_{j}^{-1}(0)f_{i}(b)}isalmostpara11e1to\vec{e_{n}}isalmostperpendicu1ar tof_{i}f_{j}^{-1}(\partial B)$

$(8)(9)$

In fact (8) follows from (6) and (9) from the fact that $f_{i}f_{j}^{-1}$
is an
210 S. Matsumoto

$f_{i}(b)$

Figure (4.7)

$f_{i}(b)$

Figure (4.8)

Euclidean similarity. See Figure (4.7).

On the other hand, notice that is nonempty and $\xi_{i}\xi_{j}^{-1}B^{*}\cap B^{*}$

is a copy of
$\xi_{i}\xi_{j}^{-1}B^{*}\cup B^{*}$ . Therefore by the completeness
$f_{i}f_{j}^{-1}B\cup B$

of , we have that
$l^{*}$

(10) $f_{i}f_{j}^{-1}(0)\not\in B$ ,
(11) $0\not\in f_{i}f_{j}^{-1}(B)$ .
Let
$f_{i}f_{j}^{-1}(0)=(\alpha_{1}, \cdots, \alpha_{n})$

and for $M>>1$ and $0<\in<<1$ , let

$D=\{|x_{i}-\alpha_{i}|\leq M(1\leq i\leq n-1), |x_{n}-\alpha_{n}|\leq\in\}$ .

Then by (5), (8) and (9), (taking $j>>i>>1$ even greater) we have
$\partial(f_{i}f_{j}^{-1}B)\cap\partial D=\partial(f_{i}f_{j}^{-1}B)\cap\partial_{v}D$ ,

where $\partial_{v}$

denotes the vertical boundary. See Figure (4.8).

 Flat Conformal Structure 211

We have by (11) that $\alpha_{n}>-\in and$ by (10) that . It also $\alpha_{n}<\in$

follows that is very near 0.

$f_{i}f_{j}^{-1}(0)$

This shows that any in the half space $\{x_{n}>0\}$ is in

$x$ $f_{i}f_{j}^{-1}B$

for some $j>>i>>1$ . Since has a copy containing , the

$f_{i}f_{j}^{-1}B\cup B$ $a^{*}$

proof of (4.12) is now complete. Q.E.D.

Step 2. In Step 1, for any point , we have found a $a^{*}\in\overline{N}^{*}$

copy of half space containing $B^{*}(a^{*}, r(a^{*}))$ . We have;

Claim (4.13). A copy of half space $H^{*}$

containing $B^{*}(a^{*}, r(a^{*}))$

is unique.

Clearly $H$ is tangent to $B(a, r(a^{*}))$ and the radius to the

Proof
point of tangency is the developing image of a short complete line. In
other words, there exists a unique short complete line in which is $N^{*}$

contained in $B^{*}(a^{*}, r(a^{*}))$ . This shows the uniqueness of . Q.E.D. $H^{*}$

Definition (4.14). of (4.13) is denoted by $H^{*}(a^{*})$ and its im-

$H^{*}$

age by $D$ by $H(a^{*})$ . The point of tangency of $H(a^{*})$ and $B(a, r(a^{*}))$
is denoted by $p(a^{*})$ .
Notice that maximal copy of half space containing $a^{*}$
may not be
unique. Since $D$ is a submersion,
$D|C1H^{*}(a^{*})$ : $C1H^{*}(a^{*})\rightarrow R^{n}$

is injective and $D(FrH^{*}(a^{*}))$ is an open subset of $\partial H(a^{*})$

.

Definition (4.15). For $a^{*}\in\overline{N}^{*}$

, denote
$L(a^{*})=\partial H(a^{*})\backslash D(FrH^{*}(a^{*}))\subset R^{n}$ .

In other words, $x\in L(a^{*})$ if and only if for some $x=\lim_{t\rightarrow\infty}l(t)$

short complete line such that $l^{*}(0)=a^{*}$ . See Figure (4.9).

$l^{*}$

For $b\in C1H(a^{*})\backslash L(a^{*})\subset R^{n}$ , we denote by the unique point $b^{*}$

in $C1H^{*}(a^{*})\subset N^{*}$ such that $D(b^{*})=b$ .

Claim (4.16). For $ b\in$ $ClH(a^{*})\backslash L(a^{*})$ , $\partial H(b^{*})$
passes through
$p(a^{*})$ .

ProofSuppose not. We have $a\not\in H(b^{*})$ since $H(a^{*})\cup H(b^{*})$

has a copy in . Consider the transformation

$N^{*}$

$f_{j}f_{i}^{-1}=(f_{i}f_{j}^{-1})^{-1}\in ES(R^{n})$
212 S. Matsumoto

$a$

$\partial H(a^{*})$

Figure (4.9)

of Step 1. Recall that is very small, is very near

$||f_{j}f_{i}^{-1}||$ $P(f_{j}f_{i}^{-1})$

$E$ and has a fixed point near

$f_{j}f_{i}^{-1}$ $p(a^{*})$ . Thus is $f_{j}f_{i}^{-1}(\partial H(b^{*}))$

almost parallel to and much near $p(a^{*})$ . Clearly

$\partial H(b^{*})$

$H(a^{*})\cup H(b^{*})\cup f_{j}f_{i}^{-1}(H(b^{*}))$

has a copy in $\overline{N}^{*}$

. This contradicts that $p(b^{*})\in L(b^{*})$ . See Figure

(4.10). Q.E.D.

Figure (4.10)
 Flat Conformal Structure 213

Claim (4.17). $L(a^{*})$ is an ane subspace of $R^{n}$

.

Proof. Let, $y\in L(a^{*})$ . Clearly $x=p(c^{*})$ for some $c^{*}\in H(a^{*})$ .
$x$

Likewise for . If a point on the line passing

$y$ and does not $b$
$x$ $y$

belong to $L(a^{*})$ , apply (4.16) to . Then passes through $b$ $\partial H(b^{*})$ $x$

and , that is, through . A contradiction.

$y$ Q.E.D. $b$

Claim (4.18). The correspondence $a^{*}\mapsto L(a^{*})$ is locally con-

stant.

Take $b^{*}\in H^{*}(a^{*})$ . Then

Proof. passes through $p(a^{*})$
$\partial H(b^{*})$

by (4.16) and $p(a^{*})\in L(b^{*})$ . Since $ L(b^{*})\cap H(a^{*})=\phi$ , we have

. Likewise,
$L(b^{*})\subset\partial H(a^{*})\cap\partial H(b^{*})$ . It
$L(a^{*})\subset\partial H(a^{*})\cap\partial H(b^{*})$

follows easily that $L(a^{*})=L(b^{*})$ . Q.E.D.

Since is connected,
$\overline{N}^{*}$

$L(a^{*})$ is independent of the choice of

$a^{*}\in\overline{N}^{*}$
. Denote $L=L(a^{*})$ .

Claim (4.19). The developing map $D$ is a covering map onto $a$

component of $R^{n}\backslash L$ .

Clearly no points of
Proof. are mapped by $D$ into . Also $\overline{N}^{*}$
$L$

we have that points in are evenly covered by $D$ . Let us consider $R^{n}\backslash L$

the point . For $dimL\geq 1$ , $\infty\in C1L$ cannot be in Image(D). For

$\infty$

$dim(L)=0$ (say $L=\{0\}$ ), if Image(D), then one can show $\infty\in$

that
$D:\overline{N}\rightarrow\hat{R}^{n}\backslash \{0\}$

is a homeomorphism. But $H\subset ES(Hn)$ has $\infty$

as a fixed point. A
contradiction. Q.E.D.

Step 3.

Lemma (4.20). Let $\Gamma=\langle f, g\rangle\subset ES(R^{n})\rangle$

where
(1) $||f||\neq 1$ , $f(a)=a(a\in R^{n})$ ,
(2) $g(a)\neq a$ .

Then $\Gamma$

is indiscrete.

Proof. Assume $||f||<1$ .

Then $||h||=||f||$ Let $h=g\circ f\circ g^{-1}$ .
and $h(g(a))=g(a)$ . Let $h_{n}=f^{n}\circ hof^{-n}$ . We have $||h_{n}||=||h||=$
, the fixed point of
$||f||$ is $f^{n}(g(a))$ and $f^{n}(g(a))\rightarrow a(n\rightarrow\infty)$ .
$h_{n}$

That is, . This shows (4.20).

$h_{n}\rightarrow f$
Q.E.D.
214 S. Matsumoto

Now we shall complete the proof of Theorem (4.4). First of all if

$L=\{0\}$ , then by Step 2, $D$ : is a homeomorphism. $N\rightarrow R^{n}\backslash \{0\}$

That is, $N$ is a Hopf manifold. This contradicts our hypothesis.

Consider the case $dimL\neq n-2$ . Suppose for simplicity that
$L=R^{q}$ . By (4.19), $D$ is a homeomorphism onto a component $V$ of

. Thus the holonomy group $H$ must be discrete. By Step 1,

$R^{n}\backslash R^{q}$

there exists $f\in H$ such that $||f||\neq 1$ . Clearly $f(L)=L$ . Assume

$f(0)=0$ . By (4.20), we have $g(0)=0$ for any $g\in H$ . Therefore
$g(R^{n-q})=R^{n-q}$ , where $R^{n-q}$ is the orthogonal complement of . $R^{q}$

When identified by $D$ , Fried metric is given by

$g_{F}=\frac{g_{E}}{|x_{2}|^{2}}$ ,

where $x=(x_{1}, x_{2})(x_{1}\in R^{q}, x_{2}\in R^{n-q})$ . Since $N=N^{*}$ is compact,

$d_{F}$
is totally bounded. That is, there exists $K>0$ such that for any
, $y\in V$ , $d_{F}(x, gy)<K$ for some $g\in H$ . But this is impossible if we
$x$

choose $y\in R^{n-q}\cap V$ and as large as desired. $x_{1}$

Finally suppose $L=R^{n-2}$ This case needs extra care. Since

is not simply connected, $D$ is not a homeomorphism and
$R^{n}\backslash R^{n-2}$

$H$ may not be discrete. Denote by $R_{\theta}\in ES(R^{n})$ the rotation by

angle around $R^{n-2}$ . Let
$\theta$

$Stab(R^{n-2})=\{f\in ES(R^{n})|f(R^{n-2})=R^{n-2}\}$ .

Notice that $R_{\theta}$

commutes with an element of Stab $(R^{n-2})$ . Let

$H^{n-1}=\{x_{n-1}>0, x_{n}=0\}$ .

Define a homeomorphism

$h:H^{n-1}\times S^{1}\rightarrow R^{n}\backslash R^{n-2}$

by $h(x, t)=R_{2\pi t}x$ . The universal covering of is identified $R^{n}\backslash R^{n-2}$

with $H^{n-1}\times R$ . Then as is easily shown, the lift of $Stab(R^{n-2})$

is identified with $ ES(R^{n-2})\times$ R. That is, we have the following
equivariant mapping of $(G, X)$ -pairs

$(ES(R^{n-2})\times R, H^{n-1}\times R)\rightarrow(Stab(R^{n-2}), R^{n}\backslash R^{n-2})$ .

The $DH$ -pair

$(D, \varphi)$ : $(\overline{N}, \pi_{1}(N))\rightarrow(R^{n}\backslash R^{n-2}, Stab(R^{n-2}))$

 Flat Conformal Structure 215

clearly lifts to a $DH$ -pair

$(D, \varphi)$
: $(\overline{N}, \pi_{1}(N))\rightarrow(H^{n-1}\times R, ES(R^{n-2})\times R)$ .

Since $D$ is a homeomorphism, the image $\overline{H}=\varphi(\pi_{1}(N))$

is discrete.
As before, contains $\overline{H}$

$(f, t)\in ES(R^{n-2})\times R$

such that $||f||\neq 1$ . Let $f(0)=0$ . Since $ES(R^{n-2})$ and $R$

commute, the argument of (4.20) is also valid and we have $g(0)=0$ for
any . The rest of the proof is similar.
$(g, s)\in\overline{H}$

\S 5. Limit set
The purpose of this section is to define limit set for flat conformal
manifolds of an arbitrary type. In this section flat conformal manifolds
are to be connected and compact, unless otherwise specified.
First of all consider an arbitrary subgroup of . ( may $\Gamma$
$\Lambda 4(S^{n})$
$\Gamma$

not be discrete. It may not be even finitely generated.) Let us begin by

defining the limit set for the group by looking at its action on . $\Gamma$ $S^{n}$

There are four dierent ways and all of them are natural and useful.

Definition (5.1). Let $L_{F}=L_{F}(\Gamma)$ be the closure of the set of

the fixed points of loxodromic or parabolic elements of . $\Gamma$

Definition (5.2). Let $L_{J}=L_{J}(\Gamma)$ be the set of points $x\in S^{n}$

such that for any neighbourhood $U$ of , the family $x$ $\{f|_{U}\}_{f\in\Gamma}$ is not
equicontinuous.
Definition (5.3). Let $L_{P}=L_{P}(\Gamma)$ be the set of points $x\in S^{n}$
such that for any neighbourhood $U$ of , the set $\{f\in\Gamma|fU\cap U\neq\phi\}$ $x$

is not precompact in . $\mathcal{M}(S^{n})$

By definition , and are closed -invariant subsets of

$L_{F}$ $L_{J}$ $L_{P}$ $\Gamma$

. Of course
$S^{n}$
is an analogy of Julia set in one dimensional complex
$L_{J}$

dynamical system. Notice that if is discrete, then coincides $\Gamma$

$S^{n}\backslash L_{P}$

with the domain of discontinuity defined in (3.24). $\Omega_{\Gamma}$

Definition (5.4). Let be the set of accumulation

$L_{\omega}=L_{\omega}(\Gamma)$

points in of the orbit

$S^{n}$ $\Gamma a$
of a certain point $a\in D^{n+1}$ .
This definition is independent of the choice of a $\in D^{n+1}$ . In fact, for
another point b $\in D^{n+1}$ and for , we have $d_{H}(\gamma_{k}(a), \gamma_{k}(b))=$ $\gamma_{k}\in\Gamma$
216 S. Matsumoto

, where
$d_{H}(a, b)$ denotes the hyperbolic distance. By the dierence
$d_{H}$

between the hyperbolic distance and the Euclidean distance, we have

$\lim_{k\rightarrow\infty}\gamma_{k}(a)=x$
$\Leftrightarrow$
$\lim_{k\rightarrow\infty}\gamma_{k}(b)=x$ .

Also note that is closed and -invariant. In fact if

$L_{\omega}$
$\Gamma$
$\lim_{k\rightarrow\infty}\gamma_{k}(a)$

$=x$ , then we have for . Below we shall

$\lim_{k\rightarrow\infty}\gamma\gamma_{k}(a)=\gamma(x)$ $\gamma\in\Gamma$

prove the minimality of . $L_{\omega}$

Definition (5.5). Let $A$ be a -invariant closed subset of $\Gamma$ $S^{n}$

such that Card(A) . The convex hull of $A$ , denoted by $C(A)$ ,

$\geq 2$

is defined to be the convex hull in $(D^{n+1}, g_{H})$ of all the geodesies

combining two points of $A$ .

Clearly $C(A)$ is a closed $\Gamma$

-invariant subset of $D^{n+1}$

. See Figure
(5.1).
$A$

$A$

$A$

$A$

Figure (5.1)

Lemma (5.6). Let be an arbitrary

$A$ $\Gamma$

-invariant closed set

such that Card(A) $\geq 2$ . Then we have $L_{\omega}(\Gamma)\subset A$
.

Take the point a $\in D^{n+1}$ of (5.4) inside

Proof. $C(A)$ . Then the
orbit of a cannot evade $C(A)$ . This shows (5.6). Q.E.D.
 Flat Conformal Structure 217

Corollary (5.7). If $\Gamma$

has no fifixed point in , then $S^{n}$ $L_{\omega}(\Gamma)$

is the unique minimal $ set\rangle$ $i.e.$

, it is contained in any nonempty closed
-invariant subset of
$\Gamma$ $S^{n}$
.

For , denote by CI(\gamma ) the convex hull in

$\gamma\in \mathcal{M}(S^{n})$ $R^{n+1}$
of
the isometric sphere . $I(\gamma)$

Lemma (5.8). For $\{\gamma_{k}\}\subset\Gamma\subset \mathcal{M}(S^{n})$

such that $\gamma_{k}\rightarrow\infty$ , we
have $d(CI(\gamma_{k}), L_{\omega}(\Gamma))\rightarrow 0$ .

For the properties of isometric spheres, see (2.11)\sim (2.16).

Proof.
We shall prove (5.8) by establishing . Recall that $d(CI(\gamma_{k}^{-1}), L_{\omega})\rightarrow 0$

if and only if radius $I(\gamma_{k}^{-1})=radiusI(\gamma_{k})\rightarrow 0$ and that

$\gamma_{k}\rightarrow\infty$

is always orthogonal to
$I(\gamma_{k})$
. Therefore given a point $a\in D^{n+1}$ , $S^{n}$

we have for large . That is,

$a\not\in CI(\gamma_{k})$ . See Figure $k$ $\gamma_{k}a\in CI(\gamma_{k}^{-1})$

(5.2). Since , it follows that

$d(\gamma_{k}a, L_{\omega})\rightarrow 0$
. $d(CI(\gamma_{k}^{-1}), L_{\omega})\rightarrow 0$

Q.E.D.

(a)

$)$

Figure (5.2)

Definition (5.9). Two points $x$ , $y\in L_{\omega}$

are called dual in case
there exists such that $\gamma_{k}\in\Gamma$ $\gamma_{k}(a)\rightarrow x$
and $\gamma_{k}^{-1}(a)\rightarrow y(a\in D^{n+1})$ .

This is also independent of the choice of . For , let $a$ $x\in L_{\omega}$ $D_{x}$

be the set of points in which are dual to . Diagonal argument $L_{\omega}$

$x$

shows that is a closed subset. Also if

$D_{x}$
and , $\gamma_{k}a\rightarrow x$
$\gamma_{k}^{-1}a\rightarrow y$
218 S. Matsumoto

then for $\gamma\in\Gamma$

, $\gamma_{k}\gamma^{-1}a\rightarrow x$
and $\gamma\gamma_{k}^{-1}a\rightarrow\gamma y$
. That is, $D_{x}$
is
-invariant.
$\Gamma$

Lemma (5.10). If $\Gamma$

has no fifixed point in $S^{n}$

, then any two
points of are dual. $L_{\omega}$

Proof. For , the condition assures that Card(D:c)\geq 2.

$ L_{\omega}\neq\phi$

Since $D_{x}$
is closed and -invariant, we have by (5.6) that
$\Gamma$

. $D_{x}\supset L_{\omega}$

Q.E.D.

Proposition (5.11). If has no fifixed point in then for $\Gamma$

$ S^{n}\rangle$

any pair of distinct points , there exists a loxodromic $x$ $ y\in L_{\omega}(\Gamma)\rangle$

transformation whose two fifixed points are arbitrarily near and . $x$ $y$

Proof. and
By (5.10), we have that are dual. Let $x$ $y$ $\gamma_{k}(a)\rightarrow x$

and . We have clearly

$\gamma_{k}^{-1}(a)\rightarrow y(\gamma_{k}\in\Gamma, a\in D^{n+1})$ . By $\gamma_{k}\rightarrow\infty$

applying the argument of (5.8), we obtain that is suciently $CI(\gamma_{k}^{-1})$

near and $x$ is suciently near . Since $x\neq y$ , we may

$CI(\gamma_{k})$ $y$

assume that $ CI(\gamma_{k}^{-1})\cap CI(\gamma_{k})=\phi$ . It is easy to show that is $\gamma_{k}$

a loxodromic transformation with one fixed point in and the $CI(\gamma_{k}^{-1})$

other in . This shows (5.11).

$CI(\gamma_{k})$ Q.E.D.

Lemma (5.12). $ L_{\omega}(\Gamma)=\phi$

if and only if $\Gamma$

is precompact.

This follows at once from the fact that for any $a\in D^{n+1}$ ,
Proof.
the isotropy subgroup of at is isomorphic to a compact $\lambda\Lambda(S^{n})$ $a$

group $O(n+1)$ . Q.E.D.

Proposition (5.13). A subgroup of is precompact if $\Gamma$

$\mathcal{M}(S^{n})$

and only if it has a common fifixed point in $D^{n+1}$ . In particular, maximal

compact subgroups of A4 are conjugate to $O(n+1)$ . $(S^{n})$

Proof.The if part is trivial. Let us show that a compact subgroup

has a fixed point in $D^{n+1}$ . (Pass to $C1F$ if
$\Gamma$

is noncompact.) $\Gamma$

Choose an arbitrary point $a\in D^{n+1}$ . Let $d=diam_{H}(\Gamma a)$ and let
$d_{H}(a, ga)=d$ $(g\in\Gamma)$ . Let be the middle point of and $ga$ . $a_{1}$ $a$

For any , consider the hyperbolic tetrahedron with vertices ,

$ h\in\Gamma$ $a$

$ga$ , $ha$ and $hga$ . All the edges have length . Easy hyperbolic $\leq d$

trigonometry shows $d(a_{1}, ha_{1})\leq cd$ for some (computable) $c\in(0,1)$ .

That is, $diam_{H}(\Gamma a_{1})\leq cd$ . Likewise construct , etc. Let $a_{2}$ $a_{3}$

. We have $diam_{H}(\Gamma a_{\infty})=0$ . That is,

$a_{\infty}=\lim_{k\rightarrow\infty}a_{k}$
is a fixed $a_{\infty}$

point of . $\Gamma$

Q.E.D.
 Flat Conformal Structure 219

Corollary (5.14). Unless has $\Gamma$

$a$
fifixed point in $D^{n+1}\cup S^{n}$
, $\Gamma$

contains a loxodromic transformation.

The condition implies that
Proof $CardL_{\omega}\geq 2$ . Therefore (5.14)
follows from (5.11). Q.E.D.
Theorem (5.15). For an arbitrary subgroup $\Gamma\subset \mathcal{M}(S^{n})$
, we
have
$L_{F}(\Gamma)\subset L_{J}(\Gamma)=L_{P}(\Gamma)=L_{\omega}(\Gamma)$ .

Moreover unless $ L_{F}(\Gamma)=\phi$ and $L_{\omega}(\Gamma)$

is a singleton, we have
$L_{F}(\Gamma)=L_{\omega}(\Gamma)$ .

Proof $L_{F}\subset L_{J}\cap L_{P}$ : This follows at once from the local models
of loxodromic and parabolic transformations.
: Suppose . Then by (5.8), for small $\in>0$
$L_{J}\cup L_{P}\subset L_{\omega}$ $x\not\in L_{\omega}$

and for a small neighbourhood $U$ of , we have that CI(\gamma )\cap U $=\phi$ if $x$

radius7(7)< $\in and$ . But the set of such that radius\geq $\gamma\in\Gamma$ $\gamma$
$\in$

is precompact by (2.16). It follows from (2.12) that . $x\not\in L_{J}\cup L_{P}$

We shall divide the proof of the remaining part into four cases.
Case 1. $\Gamma$

has no fixed point in $D^{n+1}\cup S^{n}$

.

By (5.14) we have $ L_{F}\neq\phi$ . Therefore it follows from (5.7) that

. Together with the inclusion we have already established, we
$L_{\omega}\subset L_{F}$

obtain that $L_{F}=L_{J}=L_{P}=L_{\omega}$ .

Case 2. $\Gamma$

has a fixed point in $D^{n+1}$

.

By (5.12) and (5.13), this is equivalent to $ L_{\omega}=\phi$ . We have

$ L_{F}=L_{J}=L_{P}=L_{\omega}=\phi$ .

Case 3. $\Gamma$

has a fixed point $y\in S^{n}$ and that $ L_{\omega}\backslash \{y\}\neq\phi$

.

Let . Notice that parabolic and elliptic transformations

$x\in L_{\omega}\backslash \{y\}$

of the isotropy group keep horospheres at invariant. Therefore

$\Gamma_{y}$
$y$

there must exist loxodromic transformations such that $\gamma_{n}\in\Gamma$ $\gamma_{n}a\rightarrow$

$x(a\in D^{n+1})$ . Then . That is, we have and $\gamma_{n}^{-1}a\rightarrow y$ $y\in L_{\omega}$

, showing that
$L_{\omega}\subset L_{F}$ $L_{F}=L_{J}=L_{P}=L_{\omega}$ .

Case 4. $L_{\omega}=\{y\}$ .

This is the only case where we cannot prove . In order to $L_{\omega}\subset L_{F}$

complete the proof of (5.15), it suces to show that $y\in L_{F}\cap L_{P}$ . Since
, there exists a sequence
$ L_{\omega}\neq\phi$
such that . Since $\{\gamma_{k}\}\subset\Gamma$ $\gamma_{k}\rightarrow\infty$

$\gamma_{k}y=y$ and are not loxodromic, we have $y\in

$\gamma_{k}$
CI(\gamma_{k})\cup CI(\gamma_{k}^{-1})$
220 S. Matsumoto

and Hence for any neighbourhood $U$ of

$ CI(\gamma_{k})\cap CI(\gamma_{k}^{-1})\neq\phi$ ,
. $y$

for suciently large $k>0$ . But by (2.12) and (2.16), we

$CI(\gamma_{k})\subset U$

have that is not equicontinuous. That is, $y\in L_{J}$ . Clearly we

$\{\gamma_{k}|_{U}\}$

have $\gamma_{k}(U)\cap U\neq\phi$ . Therefore $y\in L_{P}$ . Q.E.D.

The following corollary was already used in \S 3.

Corollary (5.16). Suppose a discrete group admits an invari- $\Gamma$

ant open set such that is neither empty nor a singleton.

$\Omega$
$ S^{n}\backslash \Omega$

Then acts on discontinuously.

$\Gamma$ $\Omega$

Since
Proof. is discrete, coincides with the domain of $\Gamma$
$S^{n}\backslash L_{P}$

discontinuity. By (5.6), we have . Therefore is $S^{n}\backslash \Omega\supset L_{\omega}=L_{P}$ $\Omega$

contained in the domain of discontinuity. Q.E.D.

We will give an example of for which $ L_{F}(\Gamma)=\phi$ and $\Gamma$

$L_{\omega}(\Gamma)$

is a singleton. The same example can be found in Kulkarni ([44]).

Example (5.17). Let us work with A4 . We shall construct $(\hat{R}^{4})$

a subgroup such that $ L_{F}(\Gamma)=\phi$ and that

$\Gamma$

. Equiva- $L_{\omega}(\Gamma)=\{\infty\}$

lently, the group consists purely of elliptic elements, keeps fixed

$\Gamma$
$\infty$

and does not have a fixed point in . By (1.9) and (2.24), any element $H^{5}$

has the form

$ f\in\Gamma$

$(*)$ $f(x)=Px+b$ $(P\in O(4), b\in R^{4})$ .

Notice that is elliptic if and only if has a fixed point $a\in R^{4}$ . In
$f$ $f$

fact, then, the point $(a, x)\in H^{5}(x>0)$ is kept fixed by the extended
action of . Likewise the group has a fixed point in
$f$ if and $\Gamma$ $H^{5}$

only if it has a fixed point in . Therefore our purpose is to construct $R^{4}$

a group consisting of transformations

$\Gamma$

of $(*)$ such that $f$

$ f\in\Gamma$ has a fixed point in $R^{4}$

.
$\Gamma$

does not have a fixed point in $R^{4}$

.
First of all let us show that there exist $P$, $Q\in SO(4)$ such that
for any nontrivial reduced word $w(P, Q)$ , we have $|w(P, Q)-E|\neq 0$ .
Notice that for a (possibly real) algebraic group $G$ , if $G$ contains
a free group of two generators, then for any nontrivial reduced word
$w(x, y)$ , the equation $w(x, y)=id$ defines a proper subvariety (that is,

a subvariety of positive codimension) of $G\times G$ . The converse also holds

since the complements of subvarieties of positive codimension are open
dense subsets and their countable intersection is nonempty. Therefore
a real algebraic group contains a free subgroup of two generators if and
only if its complexification does. Now it is well known that SO $(2, 1)$
 Flat Conformal Structure 221

has a free subgroup of two generators. Clearly SO $(2, 1)c=SO(3)c$ .

Therefore by the above consideration, SO(3), hence its universal cov-
ering $SU(2)$ , has a free group of two generators also. Considering the
inclusion of $SU(2)$ into (4), we obtain the desired $P$ and $Q$ .
Let
$f$ : $x\mapsto Px$ and $g:x\mapsto Qx+b(b\neq 0)$ .

Now consists purely of elliptic transformations, since any

$\Gamma=\langle f, g\rangle$

element of has the linear part without eigenvalue 1 and hence has a
$\Gamma$

fixed point in . However and have no common fixed points

$R^{4}$ $f$ $g$

in .
$R^{4}$

As a matter of fact, (5.17) implies that $L_{F}=L_{\omega}$ does not hold in

higher dimension. However in low dimension, we have;

Theorem (5.18). For $\Gamma\subset \mathcal{M}(\hat{R}^{n})$

$(n\leq 3)$ , we have
$L_{F}(\Gamma)=L_{J}(\Gamma)=L_{P}(\Gamma)=L_{\omega}(\Gamma)$ .

Proof. All that need proof is that if , then $L_{F}=\{\infty\}$ . $L_{\omega}=\{\infty\}$

Equivalently, if keeps fixed and if does not have a fixed point

$\Gamma$
$\infty$
$\Gamma$

in , then
$R^{n}$
contains nonelliptic transformations.
$\Gamma$

First of all for $n=1$ , there exist no elliptic transformations that

keep fixed and there is nothing to prove.
$\infty$

For $n=2$ , assume that , have no common fixed $f$

$g\in \mathcal{M}(\hat{R}^{2})_{\infty}$

points in . Computation
$R^{3}$
shows that $[f, g]=fgf^{-1}g^{-1}$ is parabolic,
since the linear parts commute.
Finally let $=3$ . It clearly suces to verify for a group
$n$ con- $\Gamma$

sisting of orientation preserving transformations. Orientation preserving

elliptic transformations in are rotations around their axes. $\mathcal{M}(\hat{R}^{3})_{\infty}$

Let us show first that if two rotations , have disjoint axes, then the $f$ $g$

group they generate contains a parabolic transformations. In

$\langle f, g\rangle$

fact, if the axes are parallel, then $[f, g]$ is parabolic. Suppose they are
not parallel and assume for contradiction that $fg^{-1}$ has a fixed point
$x\in R^{n}$ . Then we have $f(x)=g(x)=y$ . By Euclidean geometry, we
have that the bisector of and contains the axes of and . (See $x$ $y$ $f$ $g$

Figure (5.3).) A contradiction.

Therefore if is purely elliptic and have no common
$\Gamma\subset \mathcal{M}(\hat{R}^{3})_{\infty}$

fixed points, then all the axes of transformarions of must lie in a $\Gamma$

plane and all their rotation angles must be . Therefore there exists $\pi$

an index two subgroup of consisting of parabolic transformations. $\Gamma$

This contradiction shows (5.18). Q.E.D.

222 S. Matsumoto

Figure (5.3)

Now let be a connected closed flat conformal manifold modeled

$N$

on . As before denote by $D$ the developing map, by

$(\mathcal{M}(S^{n}), S^{n})$ the $\varphi$

holonomy homomorphism and by $H$ the holonomy group. Hereafter by

certain abuse, we consider a flat conformal manifold $N$ to be equipped
with a particular choice of developing map, holonomy homomorphism
and holonomy groups. Our purpose is to define the limit set of $N$ . So
far, we already had four kinds of limit set in terms of the holonomy group
$H$ . For a flat conformal manifold, they are denoted by $L_{F}(N)=L_{F}(H)$

and so forth. We need one more definition, which is obtained by looking

at the developing map.
Definition (5.19). Let $L_{O}=L_{O}(N)$ be the set of points $x$

such that for any compact neighbourhood of , the inverse image $\overline{U}$

$x$

has a nonempty and noncompact component.

$D^{-1}(\overline{U})$

As is shown easily, is precisely the set of points which are not

$L_{O}$

evenly covered by D.
For general closed $(G, X)$ manifold, Kulkarni-Pinka11([45]) defined
and
$L_{J}$
and showed
$L_{P}$ and . They also showed
$L_{J}\supset L_{P}$ $L_{J}\supset L_{O}$

that $L_{J}=L_{P}$ for closed flat conformal manifolds. The following is an

elaboration of their rerult.
 Flat Conformal Structure 223

Theorem (5.20). For a connected compact flflat conformal mani-

fold $N$ , we have
$L_{F}(N)=L_{\omega}(N)=L_{J}(N)=L_{P}(N)=L_{O}(N)$ .

: If not, we have $ L_{F}=\phi$ , $L_{\omega}=\{a\}$ and $H$ has

Proof $L_{F}=L_{\omega}$

a fixed point $a\in S^{n}$ . But then by (4.4), $N$ is isomorphic to either
, a Hopf manifold or a Euclidean space form. In any case we have
$S^{n}$

$L_{F}=L_{\omega}$ .
: If $CardL_{O}\geq 2$ , then this follows at once from (5.6). If
$L_{\omega}\subset L_{O}$

$L_{O}=\{a\}$ , then is a fixed point of $H$ and again by (4.4), we have

$a$

$L_{\omega}=L_{O}$ . If $ L_{O}=\phi$ , then $D$ is a covering map onto That is, $S^{n}$

$N$ is a spherical space form and we have . $ L_{\omega}=\phi$

: Denote by the closed disk centered at $x\in S^{n}$ of

$L_{O}\subset L_{J}$ $\overline{B}(x, r)$

radius $r>0$ w.r.t. the spherical metric. The proof is by contradiction.

Suppose . That is, we assume$b\in L_{O}\backslash L_{J}$

(1) For some $r_{k}\downarrow 0$

, $D^{-1}\overline{B}(b, r_{k})$
has a noncompact component
$E_{k}$
.
(2) $\{f|_{\overline{B}(b,r_{1})}\}_{f\in H}$ is equicontinuous.
Choose $a_{k}\in E_{k}$ . (Note that $D(a_{k})\rightarrow b.$ ) Then since $N$ is
compact, there exists $\xi_{k}\in\pi_{1}(N)$ such that is in some compact $\xi_{k}a_{k}$

region of . Assume . Choose a compact neighbourhood

$\overline{N}$

$\xi_{k}a_{k}\rightarrow c$

of such that
$\overline{V}$

$c$

$D|_{\overline{V}}$
: $\overline{V}\rightarrow\overline{B}(D(c), 2\in)$

is a homeomorphism for some $>0$ . Assume also $\in$ $ D(\xi_{k}a_{k})\in$

for any $k>0$ . Choose $\delta>0$ so

$\overline{B}(D(c), \in)$

$x$ , $y\in\overline{B}(b, r_{1})$

, $ d(x, y)<2\delta\Rightarrow d(f(x), f(y))<\in$ for any $f\in H$ .

For $ r_{k}<\delta$
, we have
$\overline{B}(b, r_{k})\subset\overline{B}(D(a_{k}), 2\delta)$
,
$\varphi(\xi_{k})(\overline{B}(D(a_{k}), 2\delta))\subset\overline{B}(D(\xi_{k}a_{k}), \in)$
,
$\overline{B}(D(\xi_{k}a_{k}), \in)\subset\overline{B}(D(c), 2\in)$
.

Therefore
$\varphi(\xi_{k})(\overline{B}(b, r_{k}))\subset\overline{B}(D(c), 2\in)$
,
Now is the component containing
$\xi_{k}E_{k}$
of $\xi_{k}a_{k}$ $D^{-1}(\varphi(\xi_{k})(\overline{B}(b, r_{k}))$

and is contained in . Therefore , hence $\overline{V}$

$\xi_{k}E_{k}$ $E_{k}$
, is compact. A
contradiction. Q.E.D.
224 S. Matsumoto

Definition (5.21). The set in (5.20) is called the limit set of $N$

and is denoted by $L=L(N)$ .

We summarize fundamental properties of $L$

in the following propo-
sition.

Proposition (5.22). $L(N)$ is a closed $H$ -invariant subset of . $S^{n}$

Further if $N$ is not isomorphic to a Hopf manifold, then $L(N)$ is

the unique minimal set. In particular, if $L(N)\neq S^{n}$ , then we have
IntL(N)=\phi .

Below we shall give applications of (5.20). The first one (5.23) is

originally due to Kamishima([25]). See also $Gusevski\dot{i}$-Kapovich([20]).

Corollary (5.23). If the developing map $D$ of a connected com-

pact flflat conformal manifold is not onto , then $D$ is a covering
$S^{n}$

map onto its image.

We need only consider the case where $N$ is not a Hopf

Proof.
manifold. Then by (5.22), we have $L=L_{O}$ is contained in the com-
plement of Image(D). That is, Image(D) is evenly covered by $D$ .
Q.E.D.

The next application is found in Kulkarni-Pinkall ([45]), in which

condition(2) below is mistakingly dropped.

Corollary (5.24). Let $N$ be a connected compact flflat conformal

manifold and let $\Omega=S^{n}\backslash L(N)$ . Suppose
(1) is connected and its fundamental group
$\Omega$

is fifinitely $\pi_{1}(\Omega)$

generated.
(2) For any point $x\in S^{n}$ , there exists an arbitrarily small neigh-
bourhood $U$ such that is connected.
$U\backslash L$

Then the developing map $D$ isa covering map onto its image.

Proof.First of all let us prove that is connected. In fact

$D^{-1}(\Omega)$

given any two points , $b\in D^{-1}(\Omega)$

$a$ , choose a path in joining $p$
$\overline{N}$

$a$ and . The path $b$

is covered by a finite union of small open
$p$

set . We may assume by (2) that

$V_{i}$
is
$V_{i}\backslash D^{-1}(L)\approx D(V_{i})\backslash L$

connected. Then we can make a small change of within fixing $p$ $\bigcup_{i}V_{i}$

the boundary points so that is contained in

$p$ . Therefore $D^{-1}(\Omega)$

is connected.
$D^{-1}(\Omega)$

Now by (4.4), we need only consider the case where $H$ has no
fixed points in . We need only show that $ D(\overline{N})\cap L=\phi$ . Suppose
$S^{n}$
 Flat Conformal Structure 225

the contrary. Choose a small compact ball such that $D$ is a $\overline{V}$

homeomorphism on , that and that $ExtD(\overline{V})\cap L\neq$

$\overline{V}$
$IntD(\overline{V})\cap L\neq\phi$

. Since
$\phi$
is finitely generated, it is supported on some compact
$\pi_{1}(\Omega)$

subset $K$ of . By (5.11), there exists a loxodromic transformation

$\Omega$

$f\in H$ with an attracting fixed point in and with a repelling $IntD(\overline{V})$

fixed point outside . We have for some

$D(\overline{V})$ $f^{n}(K)\subset D(\overline{V})$ $n$ $>$

. Therefore
$0$
is supported on . Now $D$ gives
$\pi_{1}(\Omega)$ $ D(\overline{V})\cap\Omega$

ahomeomorphism from onto . This shows $\overline{V}\cap D^{-1}(\Omega)$ $ D(\overline{V})\cap\Omega$

:
$D_{*}$
is an epimorphism. Since points in
$\pi_{1}(D^{-1}(\Omega))\rightarrow\pi_{1}(\Omega)$
are $\Omega$

evenly covered by $D$ , $D$ gives a homeomorphism from onto $D^{-1}(\Omega)$

. However
$\Omega$

has a noncompact component, which is of

$D^{-1}(D(\overline{V}))$

course disjoint from . A contradiction. Q.E.D.

$\overline{V}$

The condition (2) of (5.24) is in fact necessary. For, let be a $\Sigma$

closed flat conformal 2-manifold corresponding to a -group $\Gamma([3])$ . $B$

That is, is connected and simply connected and

$\Omega=S^{n}\backslash L$ is $\Sigma$

isomorphic to . Apply the construction of (3.37) to

$\Omega/\Gamma$
. We obtain $\Sigma$

a flat conformal structure with the same holonomy group and surjective
developing map. All this is of course well known. For more general
treatment, see e.g. Goldman ([16]).

We shall finish this section by studying type 2 flat conformal struc-

tures, i.e., with the developing maps covering maps and with indis-
crete holonomy groups. First we give examples in dimension . (2- $\geq 3$

dimensional examples were already given in (3.32).) For our purpose the
coordinates of is convenient. $\hat{R}^{n}$

Consider As before, denote by

$\hat{R}^{n-2}\subset\hat{R}^{n}$
the $R_{\theta}\in \mathcal{M}(\hat{R}^{n})$

rotation by angle $\theta$

around
$\hat{R}^{n-2}$
Let

$H^{n-1}=\{x_{n-1}>0, x_{n}=0\}$ .

Define
$h:H^{n-1}\times R\rightarrow\hat{R}^{n}\backslash \hat{R}^{n-2}$

by $h(x, t)=R_{2\pi t}x$ . $h$

is a universal covering. By (2.3), we have

A4 $(\hat{R}^{n-2})=\{g\in \mathcal{M}(\hat{R}^{n})|g(H^{n-1})=H^{n-1}\}$ .
Let us define

$S(\hat{R}^{n-2})=\{f\in \mathcal{M}(\hat{R}^{n})|f(\hat{R}^{n-2})=\hat{R}^{n-2}\}$
.
226 S. Matsumoto

An element $f\in S(\hat{R}^{n-2})$

carries $H^{n-1}$
to a half plane bounded by
$\hat{R}^{n-2}$
Clearly is determined by $f(H^{n-1})$ and
$f$ . This shows $f|_{\hat{R}^{n-2}}$

that $f$
commutes with . Therefore we have the epimorphism $R_{\theta}$

$\psi$
: A4 $(\hat{R}^{n-2})\times R\rightarrow S(\hat{R}^{n-2})$

defined by $\psi(g, t)=R_{2\pi t}g$ .

Consider the diagonal action of $\mathcal{M}(\hat{R}^{n-2})\times R$
on $H^{n-1}\times R$ .
Then

$(\psi, h)$ : $(\mathcal{M}(\hat{R}^{n-2})\times R, H^{n-1}\times R)\rightarrow(S(\hat{R}^{n-2}),\hat{R}^{n}\backslash \hat{R}^{n-2})$

is an equivariant mapping of $(G, X)$ -pairs.

Example (5.25). Let be a discrete subgroup $\Gamma\subset \mathcal{M}(\hat{R}^{n-2})$

which acts freely on $H^{n-1}$

. Suppose $ M=H^{n-1}/\Gamma$ is compact. For
any and any homomorphism
$\theta\in R\backslash \{0\}$ : , define $\mu$
$\Gamma\rightarrow R$

$\overline{\varphi}$

: $\Gamma\times Z\rightarrow \mathcal{M}(\hat{R}^{n-2})\times R$

by . Using a triangulation of
$\overline{\varphi}(\gamma, m)=(\gamma, \mu(\gamma)+m\theta)$
, one $ H^{n-1}/\Gamma$

can construct a continuous map : $H^{n-1}\rightarrow R$

such that $u(\gamma x)=$ $u$

$\mu(\gamma)+u(x)$ for and $x\in H^{n-1}$ . Define a homeomorphism

$\gamma\in\Gamma$

$\overline{D}:\overline{M}\times R\rightarrow H^{n-1}\times R$

by , where we identify the universal covering

$\overline{D}(x, t)=(x, u(x)+\theta t)$

$M$ with $H^{n-1}$ Let and $D=h\circ\overline{D}$ . Then is $\varphi=\psi\circ\overline{\varphi}$ $(D, \varphi)$

clearly a $DH$ pair for $M$ . Therefore it defines a flat conformal $\times S^{1}$

structure on $M\times S^{1}$

. Since is a homeomorphism, $D$ is a covering $\overline{D}$

map onto and the holonomy group $H$ is indiscrete e.g.,

$\hat{R}^{n}\backslash \hat{R}^{n-2}$

if we choose . (Moreover for a suitable choice of , the

$\theta\in R\backslash Q$ $\mu$

rotation part of $H$ is not even infinite cyclic.) Thus this is a type 2
flat conformal structure.

Conversely we have the following theorem which was first obtained

by -Kapovich ([20]) in dimension 3.
$Gusevski\dot{i}$

Theorem (5.26). Suppose $N$ is a type 2 connected closed flflat

conformal manifold modeled on , where . Then $(\mathcal{M}(\hat{R}^{n}),\hat{R}^{n})$
$n$ $\geq 3$

by changing the $DH$ pair within the equivalence class, we have $L(N)=$
 Flat Conformal Structure 227

Moreover N is a hyperbolic manifold bundle over the circle

$\hat{R}^{n-2}$

whose holonomy map is an isometry.

Proof Step 1. $L(N)=\hat{R}^{n-2}$

Let $C1H_{0}$ be the identity component of the closure CIH of $H$ .
Since $H$ is indiscrete, we have $C1H_{0}\neq\{1\}$ .
Case 1. $C1H_{0}$ is noncompact. In this case $L\equiv L_{\omega}(C1H_{0})$ is
nonempty. Notice that $C1H_{0}$ is a normal subgroup of CIH. This
implies that is invariant by the action of CIH and hence by $H$ .
$L$

Therefore by (5.22), we have $L(N)\subset L$ . On the other hand, it is easy

to show that

$L=L_{\omega}(C1H_{0})\subset L_{\omega}(C1H)=L_{\omega}(H)=L(N)$ .

Therefore we have $L=L(N)$ .

Also by (4.4), we obtain that CardL . In fact, if for example $\geq 3$

CardL $=2$ , then (4.4) implies that $N$ or its double covering is a Hopf
manifold, contrary to our hypothesis.
Let us show next that there is no fixed point of $C1H_{0}$ in . Suppose $L$

on the contrary that there exists one, say . Then for any $h\in H$ , $x$

$hx$ is also a fixed point of $C1H_{0}$ , since $C1H_{0}$ is a normal subgroup of

CIH. On the other hand, the orbit $Hx$ is dense in and therefore $L$

has cardinality . That is, there exist at least three fixed points of
$\geq 3$

$C1H_{0}$ in . This implies by the argument of (2.22) that $C1H_{0}$ has

$L$

a fixed point in $D^{n+1}$ , contradicting the assumption that $C1H_{0}$ is

noncompact.
By (5.6) this implies that any $C1H_{0}$ orbit $K$ in is dense in $L$

. Notice that $K$ is an injectively immersed submanifold in

$\hat{R}^{n}$
$L$

By (5.11), there exists a loxodromic transformation $f\in H$ . We may

assume for simplicity that $f(x)=\lambda Px$ , $(\lambda>1, P\in O(n))$ and that
$0\in K$ . Clearly $K$ is kept invariant by . Now the smoothness of $K$ $f$

at 0 implies that $K=\hat{R}^{k}$ for some $1\leq k\leq n$ . ( $K=R^{k}$ implies that
is a fixed point of $C1H_{0}$ , contradicting the above observation.) This
$\infty$

shows $L=\hat{R}^{k}$ Since the developing map $D$ is a covering map onto
its image, we have $ D(\overline{N})\cap L(N)=\phi$ . In particular we obtain $k\neq n$ .
Finally we have $k=n-2$ , since otherwise $D$ is a homeomorphism
onto a connected component of $\hat{R}^{n}\backslash \hat{R}^{k}$

and $H$ must be discrete.

Case 2. $C1H_{0}$ is compact. Here the coordinates of is conve- $S^{n}$

nient. First of all by (5.13), we may assume $C1H_{0}\subset O(n+1)$ . If 0

is the unique fixed point of $C1H_{0}$ , then 0 is also a fixed point of $H$ .
228 S. Matsumoto

That is, $H\subset O(n+1)$ . A contradiction. Therefore the fixed point set
of $C1H_{0}$ in is $(0\leq k\leq n)$ .
$S^{n}$
Since $C1H_{0}$ is nontrivial, we
$S^{k}$

have $k\neq n$ . Likewise we obtain $k\neq n-1$ . In fact since $C1H_{0}$ is

connected, we have $C1H_{0}\subset SO(n+1)$ . Therefore if $k=n-1$ , then
$C1H_{0}$ is trivial.
Notice that is $H$ -invariant and therefore $L(N)\subset S^{k}$ . Let us
$S^{k}$

show $L(N)=S^{k}$ . Since $k\leq n-2$ , we obtain as in the proof of (5.24)

that is connected. In way of contradiction take a point
$\overline{N}\backslash D^{-1}(S^{k})$

$x\in S^{k}\backslash L(N)$ . Consider an -neighbourhood $V$ of in such $\in$ $x$ $S^{n}$

that $ V\cap L(N)=\phi$ .

Then we have as well that is connected. That is, $D^{-1}(V\backslash S^{k})$

$D^{-1}(V)$ is connected. This shows that $D$ is a homeomorphism, con-

trary to our hypothesis. Therefore we have $L(N)=S^{k}$ . As before we

obtain $k=n-2$ .
$Sep$ $2$
. We shall show the last part of (5.26). Since $L(N)=\hat{R}^{n-2}$
,
the $DH$ pair $(D, \varphi)$ lifts to $(\overline{D}, \overline{\varphi})$
, where

$\overline{D}:\overline{N}\rightarrow H^{n-1}\times R$
,
$\overline{\varphi}$

: $\pi_{1}(N)\rightarrow$
A4 $(\hat{R}^{n-2})\times R$
.

Denote by the canonical projection to the -th factor. Consider

$p_{i}$
$i$

a small perturbation of such that and $\overline{\varphi}$

$\overline{\varphi}$
$p_{1}o\overline{\varphi}=p_{1}o\overline{\varphi}$

. Let . Then there exists a submersion

$p_{2}o\overline{\varphi}(\pi_{1}(N))\subset Q$ $\varphi=\psi\circ\overline{\varphi}$

$D$ : such that is a $DH$ pair. (See Thurston [56]

$\overline{N}\rightarrow\hat{R}^{n}$
$(D, \varphi\prime)$

Chapt. 5 or Canary-Epstein-Green [9] Chapt. 1.) The limit set of the

new $DH$ pair is also , since we have altered
$\hat{R}^{n-2}$
only
$(D, \varphi\prime)$ $\overline{\varphi}$

in the $R$-direction. Therefore by (5.24), $D$ is a covering map onto

$\hat{R}^{n}\backslash \hat{R}^{n-2}$
That is, $D$ lifts to a homeomorphism

$\overline{D}$

: $\overline{N}\rightarrow H^{n-1}\times$
R.

Since $p_{2}\circ\overline{\varphi}(\pi_{1}(N))\subset Q$
, we have that is infinite
$p_{2}\circ\overline{\varphi}(\pi_{1}(N))$

cyclic with a generator $\theta$

. Let $\Gamma=Ker(p_{2}o\overline{\varphi})$
. We have an exact
sequence
$1\rightarrow\Gamma\rightarrow\pi_{1}(N)\rightarrow\theta Z\rightarrow 1$
.

Correspondingly we have a bundle structure of $N$ with fiber

over $R/\theta Z\cong S^{1}$ . Clearly the monodromy map is
$H^{n-1}/p_{1}\overline{\varphi}(\Gamma)$

an isometry of a hyperbolic manifold. Q.E.D.

 Flat Conformal Structure 229

\S 6. Elementary structure and $C$ -structure

Also in this section flat conformal manifolds are to be connected and
compact unless otherwise specified. The dimension is always . $\geq 3$

Definition (6.1). A flat conformal manifold $N$ is called elemen-

tary if and only if CardL(N)\leq 2.
As applications of (5.20), we have the following characterizations of
elementary flat conformal manifolds.
Proposition (6.2). The following conditions are equivalent.
(1) $ L(N)=\phi$ .
(2) The holonomy group $H$ consists purely of elliptic transforma-
tions.
(3) $N$ is a spherical space form.

Proposition (6.3). The following conditions are equivalent.

(1) $L(N)$ is a singleton.
(2) $H$ contains parabolic transformations and no loxodromic trans-
formations.
(3) $N$ is an Euclidean space form.
All that need proof is
Proof. $(2)\Rightarrow(1)$ . (2) implies .
$ L(N)\neq\phi$

By (5.14), we have ,
$H\subset \mathcal{M}(S^{n})_{a}$ $a\in S^{n}$ . This shows (1). Q.E.D.
Proposition (6.4). CardL(N)=2 if and only if $N$ or its double
covering is a Hopf manifold.

Proposition (6.5). If CardL(N)\geq 3, then $L(N)$ is a perfect

set.
The assumption implies by (5.14) the existence of a loxo-
Proof.
dromic elememt $f\in H$ . Then at least one point of $L=L(N)$ is not
fixed by . Since
$f$
is invariant by , we obtain that
$L$
is an infinite $f$ $L$

set. Therefore the derived set is nonempty. By the minimality of

$L$

$L((5.22))$ , we have $L=L$ . That is, is perfect. Q.E.D. $L$

Theorem (6.6). If the holonomy group $H$ of a connected com-

pact flflat conformal manifold $N$ does not contain a free group of two
generators, then $N$ is elementary.
Suppose on the contrary that $N$ is nonelementary.
Proof.
Then CardL(N)=\infty and therefore by (5.11), there exist two loxo-
dromic transformations , $g\in H$ with disjoint fixed points. Now it
$f$
230 S. Matsumoto

is a well known fact that $f^{n}$

and $g^{n}$
generate a free group for large
$n$ . Q.E.D.

(6.6) was first proved under the hypothesis that $H$ is virtually
nilpotent by Goldman([15]) and then by Kamishima ([25]) when $H$
is virtually solvable. It is well known that for matrix group virtual
solvability is equivalent to the condition of (6.6). See Tits ([57]).
Corollary (6.7). If $\pi_{1}(M)$ does not contain a free group of two
generators and if a compact connected manifold $M$ does not have as $a$

fifinite covering , or , then $M$ does not admit flflat

$S^{n}$ $S^{n-1}\times S^{1}$ $T^{n}$ $a$

conformal structure.
(6.7) forbids many manifolds to admit flat conformal structures, e.g.,
3-manifolds with Nil or Solv geometry.

Given two flat conformal manifolds and , a new flat con- $N_{1}$ $N_{2}$

formal manifold, called connected sum, is obtained in the following way.

This operation was first introduced by Kulkarni ([42]).
Inside a conformal atlas of , choose a closed ball .
$(U_{i}, q_{i})$ $N_{i}$ $B_{i}$

Assume that there exists such that $f(Intq_{1}(B_{1}))=S^{1}\backslash $

$f\in \mathcal{M}(S^{n})$

. See Figure (6.1). (This is always possible e.g., if we choose

$q_{2}(B_{2})$ $B_{i}$

so that is a metric ball.) Define a homeomorphism

$q_{i}(B_{\dot{\iota}})$
: $h$ $\partial B_{1}\rightarrow$

so that $q_{2}\circ h=f\circ q_{1}$ . Then

$\partial B_{2}$

$(f\circ q_{1})\cup q_{2}$ : $(U_{1}\backslash IntB_{1})\cup(U_{2}\backslash IntB_{2})/h\rightarrow S^{n}$

is a well defined embedding. Using $(foq_{1})\cup q_{2}$ and other small charts
in , we can define a flat conformal structure on the connected sum
$N_{i}$

$(N_{1}\backslash IntB_{1})\cup(N_{2}\backslash IntB_{2})/h$ .

Definition (6.8). The flat conformal structure constructed in this

way is called a connected sum of and and is denoted by . $N_{1}$ $N_{2}$ $N_{1}QN_{2}$

is not uniquely determined. For example if we

Notice that $N_{1}\# N_{2}$

fix and make much smaller, then the resultant

$B_{2}\subset N_{2}$ $B_{1}\subset N_{1}$

connected sum would be dierent as a flat conformal structure.

One can also define the operation of connected sum of more than
two structures.
Definition (6.9). A flat conformal structure (manifold) is called
a structure ( -manifold) if it is a connected sum of finitely many
$C$ $C$

elementary structures and is not itself elementary.

 Flat Conformal Structure 231

Figure (6.1)

It is easy to show that as a flat conformal manifold, $N\# S^{n}$ is

isomorphic to $N$ . This is called a trivial connected sum. There is only
one case where a nontrivial connected sum of elementary flat conformal
manifolds becomes again elementary, that is, when $N_{1}=N_{2}=RP^{n}$ ,
the real projective space. In this case is isomorphic to the
$\pi_{1}(N_{1}QN_{2})$

infinite dihedral group $Z_{2}*Z_{2}$ . One can show directly that $N_{1}\# N_{2}$

has a Hopf manifold as a double covering. In all the other

$S^{n-1}\times S^{1}$

cases the fundamental group of a connected sum contains a free group

of two generators and therefore it cannot be elementary.

Definition (6.10). A Cantor set $\wedge r$

$\subset S^{n}$
is called tame if
and only if there exists a homeomorphism $h$
: $S^{n}\rightarrow S^{n}$
such that
$h(1)\subset S^{1}$ . Otherwise is called wild.
$\prime r$

Proposition (6.11). A -structure

$C$ $N$ is of type 1. The limit set
$L(N)$ is a tame Cantor set.

Proof. For the first part of (6.11), it suces to show the following;
If the developing maps of the flat conformal structure and $N_{1}$ $N_{2}$

are injective, then the developing map of their connected sum is also
injective. Let be the (n-l)-sphere on which the connected
$S\subset N_{1}\# N_{2}$

sum is made. splits

$S$
into two parts
$N_{1}\# N_{2}$ such that . $M_{i}$ $M_{i}\subset N_{i}$

Take a base point in IntMi. We have

$\pi_{1}(N_{1}\# N_{2})\cong\pi_{1}(N_{1})*\pi_{1}(N_{2})$ .

The element of $\pi_{1}(N_{1}\# N_{2})$ is represented uniquely as a reduced word

232 S. Matsumoto

of elements of $\pi_{1}(N_{1})$
and $\pi_{1}(N_{2})$ . Consider the universal covering

$\pi$
: $N_{1}\# N_{2}\rightarrow N_{1}\# N_{2}$ .

Choose one component of $\pi^{-1}(M_{1})$ and denote it by . Choose one $\overline{M}_{1}$

boundary component of and denote it by . The component of $\overline{M}_{1}$

$\overline{S}$

$\pi^{-1}(M_{2})$ which has as a boundary component is denoted by . $\overline{S}$

$\overline{M}_{2}$

The boundary of consists precisely of those components of $\pi^{-1}(S)$

$\overline{M}_{1}$

which are of the form . is adjacent to if $\xi\overline{S}(\xi\in\pi_{1}(M_{1}))$ $\xi\overline{M}_{2}$ $\overline{M}_{1}$

and only if $\xi\in\pi_{1}(M_{1})$ . Now by the assumption the developing map

$D$ is injective on and on . and are in the
$\overline{M}_{1}$ $\xi\overline{M}_{2}$ $D(\overline{M}_{1})$ $D(\xi\overline{M}_{2})$

opposit sides of the sphere . This shows that $D$ is injective on $D(\overline{S})$

$\overline{M}_{1}\cup(\bigcup_{\xi\in\pi_{1}(M_{1})}\xi\overline{M}_{2})$
.

Boundary component of except are of the form $\xi\overline{M}_{2}$ $\xi\overline{S}$

$\xi\eta\overline{S}(\eta\in$

. Again
$\pi_{1}(M_{2})\backslash \{1\})$ and are in the opposite $D(\xi\overline{M}_{2})$ $D(\xi\eta\overline{M}_{1})$

sides of the sphere . Therefore $D$ is injective on the union of $D(\xi\eta\overline{S})$

, and
$\overline{M}_{1}$
$( \xi\in\pi_{1}(M_{1}), \eta\in\pi_{1}(M_{2})\backslash \{1\})$
$\xi\overline{M}_{2}$ $\xi\eta\overline{M}^{1}$
. An induction
on the length of the word of yields that the developing $\pi_{1}(N_{1}\beta N_{2})$

map $D$ of a -structure $N$ is injective. We also have that Image(D)

$C$

is contained in the complement of the limit set $L=L(N)$ and the

holonomy homomorphism is an isomorphism onto a discrete group $H$ .
Next we shall show that is totally disconnected. Once this is $L$

established, we have by (6.5) that is a Cantor set. For simplicity, we $L$

prove this only for the connected sum $N$ of two elementary structures
and
$N_{1}$
. We use the same notations as before. Choose a base point
$N_{2}$

and consider the family of disjoint topological spheres

$x_{0}\in D(Int\overline{M}_{1})$

$S$ $=\{\varphi(\zeta)D(\overline{S})|\zeta\in\pi_{1}(N)\}$
.

A point $x\in S^{n}\backslash Image(D)$ is called accessible if there exists a path

$p$ in combining$S^{n}$
and such that intersects finitely many $x_{0}$ $x$ $p$

spheres in . Accessible points consists precisely of the $H$ -orbits of the

$S$

points in $L(N_{1})\cup L(N_{2})$ . See Figure (6.2). (We made the convension
that coincides with the restriction of the developing map of
$D|_{\overline{M}_{i}}$ . $N_{\dot{0}}$

This is always possible if we change the $DH$ pairs of within the $N_{2}$

equivalence classes.)
Therefore accessible points are at most countable in number. Let
$x\in S^{n}\backslash Image(D)$ be a nonaccessible point. Then there are infinitely
many nested spheres which separates from . $\varphi(\zeta_{i})D(\overline{S})(i\geq 1)$ $x$ $x_{0}$
 Flat Conformal Structure 233

Figure (6.2)

Since is distinct and

$\zeta_{i}$ $H$ is discrete, we have $\varphi(\zeta_{i})\rightarrow\infty$
. Therefore
by (5.8), we have
$diam\varphi(\zeta_{i})D(\overline{S})\rightarrow 0$

since . This shows that the component of $S^{n}\backslash Image(D)$

$ D(\overline{S})\cap L=\phi$

at a nonaccessible point is a singleton. Since accessible points are at

$x$

most countable, this shows that $S^{n}\backslash Image(D)$ , hence , is totally $L$

disconnected.
At this point we have obtained that $L=S^{n}\backslash Image(D)$ , since
$S^{n}\backslash Image(D)$ , having no interior, is not evenly covered by $D$ .
Finally to show the tameness of , we have to define a homeomor- $L$

phism : $h$
such that $h(L)\subset S^{1}$ . First of all, define
$S^{n}\rightarrow S^{n}$
on $h$

so that
$C1D(\overline{M}_{1})$
carries all the boundary components to spheres in-
$h$

tersecting and that

$S^{1}$
carries all the accessible points in
$h$ $C1D(\overline{M}_{1})$

into . Next extend

$S^{1}$
to the adjacent components. Proceeding like
$h$

this we can define the homomorphism on the whole of . Details $h$ $S^{n}$

are left to the reader. Q.E.D.

In dimension 3, we have the converse of (6.11).

Theorem (6.12). Let $N$ be a connected compact flflat conformal
manifold of dimension 3 such that the limit set $L(N)$ is a tame Cantor
set. Then $N$ is a $C$-manifold.
employ a method of Kulkarni ([43]) based upon the
Proof. We
study of ends of a group. The necessary parts of the theory of ends are
234 S. Matsumoto

summarized in Appendix.
First of all notice that a tame Cantor set $L(N)$ satisfies the con-
ditions of (5.24). Especially $\Omega=S^{n}\backslash L(N)$ is simply connected.
Therefore the developing map $D$ : is a homeomorphism and $\overline{N}\rightarrow\Omega$

the holonomy group $H$ acts on freely and discontinuously. That $\Omega$

is, we have an isomorphism $N\cong\Omega/H$ of flat conformal structures. By

Selbergs theorem (3.31), $H$ has a torsion free subgroup $H$ of finite
index. $N=\Omega/H$ is a finitely sheeted covering of $N$ and therefore
a compact manifold. Clearly we have $L(N)=L(N)$ . By (A.4) of
Appendix, we have
$Card\mathcal{E}(H)=Card\mathcal{E}(\Omega)=\infty$ .

Therefore by Stallings theorem (A.9), we obtain that $H$ is a non-

trivial free product. Consequently $N$ decomposes as a nontrivial con-
nected sum (as a manifold). See e.g. Hempel ([22]). Thus we have that
$\pi_{2}(N)=\pi_{2}(N)\not\cong 1$ . By sphere theorem, this implies that $N$ is re-
ducible. It follows from (6.6) that $N$ is not homeomorphic to , $S^{2}\times S^{1}$

since $N$ is not an elementary structure. Therefore $N$ is nonprime,

that is, decomposes as a nontrivial connected sum $N=N_{1}QN_{2}$ as a
manifold.
Let $S\subset N$ be the two sphere on which the connected sum is made.
Let $N=M_{1}\bigcup_{S}M_{2}$ and $N_{i}=M_{i}\bigcup_{S}B_{i}$ , where is homeomorphic $B_{i}$

to the closed 3-ball. Let : be the universal covering and let

$\pi$
$\overline{N}\rightarrow N$

be a lift of
$\overline{S}$

$S$
to . Denote by the connected component of
$\overline{N}$ $\overline{M}_{i}$

$\pi^{-1}(M_{i})$
which has as a boundary component. All the boundary
$\overline{S}$

components of is of the form

$\overline{M}_{i}$

. Since is a $\xi\overline{S}(\xi\in\pi_{1}(N_{i}))$
$D|_{\overline{M}_{i}}$

homeomorphism, it extends in an equivariant way to .

. $\overline{N}_{?}$
$=\overline{M_{i}}\bigcup_{\xi\overline{S}}\xi\overline{B}_{1}$

From this we obtain a flat conformal structure on , showing that $N_{i}$

the given structure on $N$ is a connected sum of these two structures.

Clearly we have $L(N_{i})\subset L(N)$ . Therefore either $CardL(N_{i})\leq 2$ or
is again a tame Cantor set. In the latter case, apply the whole
$L(N_{\dot{t}})$

argument once again to . It is well known in 3-manifold theory that

$N_{i}$

this process terminates. We obtain that $N$ is a -structure. Q.E.D. $C$

It is unknown whether (6.12) holds in dimensin . In \S 8, we shall $\geq 4$

give an example of flat conformal 3-manifold whose limit set is a wild

Cantor set. By (6.11), this is not a manifold. $C$

\S 7. Poincar\e polyhedron theorem

This section is devoted to the exposition of a fundamental theorem

 Flat Conformal Structure 235

of Poincar\e. It will be given in its simplest form, which is sucient for

our purpose of the next section. More general treatment is found e.g. in
Maskit [47] in the framework of hyperbolic geometry.
Let , $T_{i}(1\leq i\leq m)$ be metric $(n -1)$ -spheres in
$T_{i}$
. Assume $S^{n}$

that any pair of them either intersect in an $(n-2)$ -sphere or are disjoint
and that any triple do not intersect at all. Let be the family $\mathcal{E}=\{e_{j}\}$

of $(n-2)$ -spheres of the intersections. Let $P$ be a component of the

complement of the union of all and . Assume that any element $T_{i}$ $T_{i}$

of is contained in
$\mathcal{E}$

. See Figure (7.1). $\partial P$

$s_{4}$

$s_{3}$

Figure (7.1)

Let and . They are puctured $(n -1)-$

$S_{i}=T_{i}\cap\partial P$ $S_{i}=T_{i}\cap\partial P$

spheres. Let $=\{S_{i}, S_{i}\}$ . An element of or is called respectively

$S$ $S$ $\mathcal{E}$

a side or an edge of $P$ . Our first hypothesis is this.

(7.1). For each i, there exists $f_{i}\in \mathcal{M}(S^{n})$ such that $f_{i}(S_{i})=$

$S_{i}$
and $ f_{i}(P)\cap P=\phi$ .
Fix once and for all and let $f_{i}$
. An element of $\mathcal{F}=\{f_{i}, f_{i}^{-1}\}$ $\mathcal{F}$

is called a side $pair?.ng$ transformation. A side pairing transformation,

say , sends an edge in
$f_{i}$
to an edge in . We call $e$ $\partial S_{i}$ $e$ $\partial S_{i}$ $e$

and are related. This relation generates

$e$
an equivalence relation in
. is partitioned into equivalence classes, called cycles. Each cycle
$\mathcal{E}$ $\mathcal{E}$

$C$
can be cyclically ordered as
$C$ $=\{e_{1}, \ldots, e_{p-1}, e_{p}=e_{0}\}$
236 S. Matsumoto

in such a way that for each $1\leq l/\leq p$ , there exists $f_{l/}\in \mathcal{F}$
such that
$f_{l/}(e_{\nu-1})=e_{l/}$ . Let
$f_{C}=f_{p}o\cdots\cdot\cdot of_{1}$ .

Clearly $f_{C}(e_{0})=e_{0}$ . For each cycle $C$

, $f_{C}$
is well defined up to inverse
and conjugation. See Figure (7.2).

$e_{1}$

$e_{2}$

$f_{C}$

$e_{3}=e_{0}$

Figure (7.2)

For each edge , the angle of

$e\in \mathcal{E}$ $P$ at $e$
is denoted by $\theta(e)$
.
For a cycle as above, define
$C$

$\theta_{C}=\sum_{1\leq\nu\leq p}\theta(e_{\nu})$
.

Our second hypothesis is;

(H.2). For each cycle $C$

, we have $\theta_{C}=2\pi/q$ and $f_{C}^{q}=id$
for
some $q\geq 1$ .

The relation $f_{C}^{q}=id$ is called a cycle relation. Denote by the $\Gamma$

subgroup of A4 generated by
$(S^{n})$ and let be the abstract group $\mathcal{F}$ $\Gamma^{*}$

with generators the side pairing transformations and with relations the
cycle relations. Clearly we have an epimorphism : . $\psi$
$\Gamma^{*}\rightarrow\Gamma$
 Flat Conformal Structure 237

Definition (7.1). For a subgroup , an open subset $G\subset \mathcal{M}(S^{n})$

$R\subset S^{n}$ is called a fundamental domain of $G$ if and only if the

following two conditions are satisfied. ( denotes the domain of $\Omega_{G}$

discontinuity of G. )

(FD.1) $\Omega_{G}=g\in G\cup g(C1R)$

.

(FD.1) $ g(R)\cap R=\phi$ for any $g\in G\backslash \{1\}$ .

Theorem (7.2). Assume (H.I) and (H.2). Then $\psi$

: $\Gamma^{*}\rightarrow\Gamma$
is
an isomorphism, is a discrete subgroup of $\Gamma$
$\mathcal{M}(S^{n})$
and $P$ is $a$

fundamental domain of . $\Gamma$

Proof. Think of the family of domains. By a side pairing $\{\gamma P\}_{\gamma\in\Gamma}$

transformation , $fP$ is attached to $P$ along a side of $P$ . Next

$f$

$fgP(g\in \mathcal{F}, g\neq f^{-1})$ is attached to $fP$ . If this process is continued,

then around an edge which belongs to a cycle$e_{0}\in \mathcal{E}$

$C$ $=\{e_{1}, \ldots, e_{p-1}, e_{p}=e_{0}\}$

such that $f_{\nu}(e_{\nu-1})=e_{l/}$ , there is a sequence of domains

$P$ , $f_{p}P$ , $f_{p}f_{p-1}P$ , $\ldots\ldots$ , $f_{p}\cdots f_{1}P$ ,
$\ldots$ , $(f_{p}\cdots f_{1})f_{p}P$ , $\ldots\ldots$ , $(f_{p}\cdots f_{1})^{q-1}f_{p}\cdots f_{2}P$ .

They surround the edge . By virture of (H.2), the sum of their angles $e_{0}$

at is just
$e_{0}$ and the last domain $ 2\pi$

$(f_{p}\cdots f_{1})^{q-1}f_{p}\cdots f_{2}P$

is attached to $P$ by . The essential part of the proof is to show that

$f_{1}$

in this way the family forms a tesselation of . However $\{\gamma P\}$ $S^{n}$

in order to be precise, we must argue in a formal way as follows. Start

with abstract copies of $P$ and attach them one by one by side pairing
transformations, thus constructing a replica of the domain of disconti-
nuity of . Next we show the existence of an embedding of the replica
$\Gamma$

into . The abstract group

$S^{n}$
is convenient for this development. $\Gamma^{*}$

Let us embark upon the proof.

Define an equivalence relation $\sim$
in $\Gamma^{*}\times C1P$
generated by the
following.
$(\gamma, x)\sim(\gamma, x)$ if $\gamma=\gamma f$
, x $=f(x)$ for some f $\in \mathcal{F}$
.
238 S. Matsumoto

Let
$\Omega^{*}=\Gamma^{*}\times C1P/\sim$ .

The action of $\Gamma^{*}$

on $\Omega^{*}$
is defined by
$\gamma(\gamma, x)=(\gamma\gamma, x)$ .

Claim 1. $\Omega^{*}$
is $a$
flflat conformal manifold on which $\Gamma^{*}$
acts
conformally.

Choose . Suppose first of all that $x\in P$ . Then

$(\gamma, x)\in\Gamma^{*}\times C1P$

by (H.I), there is no point $x\in C1P$ such that $x=f(x)$ . $(f\in \mathcal{F})$

That is, is equivalent to no other point and therefore it certainly

$(\gamma, x)$

has a neighbourhood homeomorphic to an -ball. Suppose next that $n$

$x\in IntS$ , where is a side of $P$ , with a side pairing map

$S$
: . $f$ $S\rightarrow S$

Then as is shown easily, the only point in which is equivalent $\Gamma^{*}\times C1P$

to is $(\lambda f^{-1}, f(x))$ . Clearly one can construct a neighbourhood

$(\lambda, x)$

of the identified point , homeomorphic to an -ball, in $[(\lambda, x)]$ $n$

$\{\lambda\}\times C1P\cup\{\lambda f^{-1}\}\times C1P/\sim$ .

Finally consider the case where $x\in e_{0}$ $(e_{0}\in \mathcal{E})$

. Let

$C=\{e_{1}, \ldots, e_{p-1}, e_{p}=e_{0}\}$

be a cycle such that $f_{\nu}(e_{\nu-1})=e_{\nu}$ . Then we have

$(\lambda, x)\sim(\lambda f_{p}^{-1}, f_{p}x)\sim\cdots\cdot$ .

$\sim(\lambda f_{2}^{-1}\cdots f_{p}^{-1}(f_{1}^{-1}\cdots f_{p}^{-1})^{q-1}, (f_{p}\cdots f_{1})^{q-1}f_{p}\cdots f_{2}x)$ .

By the definition of the cycle, these are shown to be all the points that
are equivalent to . By (H.2), we can construct a desired neigh-
$(\lambda, x)$

bourhood of . This shows that is a manifold. Since the side

$[(\lambda, x)]$ $\Omega^{*}$

pairing transformations are Moebius transformations, it is easy to endow

a flat conformal structure. Also one can show without diculty that
$\Omega^{*}$

the action of is conformal. $\Gamma^{*}$

Next consider the conformal mapping

$E:\Omega^{*}\rightarrow S^{n}$

defined by $E(\lambda, x)=\psi(\lambda)x$where : is the canonical $\psi$ $\Gamma^{*}\rightarrow\Gamma$

projection. $E$ is well defined and -equivariant, that is, $\psi$

$E(\lambda(\lambda, x))=\psi(\lambda)E(\lambda, x)$ .

 Flat Conformal Structure 239

Claim 2. E is an embedding onto a connected open subet $\Omega\subset$

$S^{n}$
.

For the proof we need hyperbolic geometry of $D^{n+1}$ . Let us extend

first of all $(n -1)$ spheres and used to define $P$ to half n- $T_{i}$ $T_{i}$

spheres , orthogonal to . These are totally geodesic

$\overline{T}_{i}$ $\overline{T}_{i}\subset D^{n+1}$ $S^{n}$

hyperplanes in $(D^{n+1}, g_{H})$ . Using and we can extend the $\overline{T}_{i}$ $\overline{T}_{i}$

domain $P$ to a domain . Define as before $\overline{P}\subset D^{n+1}$

$\overline{\Omega}^{*}=\Gamma^{*}\times\overline{P}/\sim$
,
$\overline{E}:\overline{\Omega}^{*}\rightarrow D^{n+1}$
.

The argument of Claim 1 shows that is a hyperbolic manifold and $\overline{\Omega}^{*}$

that is an isometric immersion. Furthermore one can show that

$\overline{E}$

there exists $\in>0$ such that any point in has a neighbourhood $\overline{\Omega}^{*}$

isometric to hyperbolic -ball. Therefore is complete and thus $\in$

$\overline{\Omega}^{*}$
$\overline{E}$

is a covering map. That is, is a bijective isometry. Since $\overline{E}$

$E\cup\overline{E}:\Omega^{*}\cup\overline{\Omega}^{*}\rightarrow S^{n}\cup D^{n+1}$

is continuous, we obtain Claim 2.

Claim 2 implies that : is an isomorphism and that $\psi$
$\Gamma^{*}\rightarrow\Gamma$

is a discrete subgroup of
$\Gamma$

which acts discontinuously on $\mathcal{M}(S^{n})$

. What is left is to show that $P$ is a fundamental domain.

$\Omega=E(\Omega^{*})$

This is equivalent to the following.

Claim 3. $\Omega$

is precisely the domain of discontinuity $\Omega_{\Gamma}$

of $\Gamma$

We already had . To show the converse, it suces by (5.15)

$\Omega\subset\Omega_{\Gamma}$

to show that . Take a point . Then for any

$S^{n}\backslash \Omega\subset L_{\omega}(\Gamma)$ $ x\in S^{n}\backslash \Omega$

small neighbourhood $U$ of in $S^{n}\cup D^{n+1}$

, we have $x$
$\gamma_{k}\overline{P}\cap U\neq\phi$

for infinitely many . By (5.8), we have for large

$\gamma_{k}\in\Gamma$ $ CI(\gamma_{k})\cap\overline{P}=\phi$

, since
$k$ $C1P\cap L_{\omega}(\Gamma)=\phi$ . This implies

$diam\gamma_{k}\overline{P}\leq diamCI(\gamma_{k}^{-1})\rightarrow 0$
.

That is, $U$ contains infinitely many $\gamma_{k}\overline{P}$

, showing that $x\in L_{\omega}(\Gamma)$ .

Q.E.D.

\S 8. Wild Cantor set as limit set

In this section we shall construct an example of type 1 compact
flat conformal 3-manifold whose limit set is a wild Cantor set. Such
240 S. Matsumoto

an example was first obtained by Bestvina-Cooper [4] for an open 3-

manifold. Our example is a variant of what they constructed. We shall
follow [4] rather closely.
For a while we adopt the coordinates of instead of . First
$\hat{R}^{3}$
$S^{3}$

of all let $K$ be a graph embedded in , depicted in Figure (8.1). $R^{3}$

$z$

Figure (8. 1)

The segment ab, $cd$ , $ef$ , $gh$ and are straight lines and the $ij$

other parts are circular arcs of the same radius. Choose a family of
2-spheres
$A_{1}$
, $\ldots$ , $A_{n}$
, $A_{1}$
, $\ldots$ , $A_{n}B_{1}$ , $\ldots$ , $B_{n}$

$B_{1}$
, $\ldots$ , $B_{n}C$ , $CD$ , $D$
, $E$ , $E$
,

as in Figure (8.2).
We assume the followings.
(P.I) All the spheres have the same radius and have centers in $K$ .
(P.2) The union of balls they bound covers $K$ .
(P.3) The centers of $C$ , $E$ , $D$ , $C$ , $E$ , $D$ are in the -axis. $x$

, $C$ ,
$A_{n}$
and
$A_{n}$
, $E$ , have centeres in straight lines
$A_{1}$ $A_{1}$

parallel to -axis. $z$

, $E$
, and , $D$ ,
$B_{1}$
have centers $B_{1}$ $B_{n}$ $B_{n}$

in straight lines parallel to -axis. $y$

 Flat Conformal Structure 241

$n$

Figure (8.2)

(P.4) Adjacent two spheres intersect at angle $2\pi/28$ .

Let $P$ be the complement in of the union of all the balls.

$\hat{R}^{3}$

Next we shall define side pairing transformations for $P$ .

(8.2) , where
$\alpha_{j}=I_{xy}\circ I_{A_{j}}$ is the reflexion at the $I_{xy}$ $xy$ -plane and
$I_{A_{j}}$
is the inversion at the sphere . $A_{j}$

(8.2) $\beta_{j}=I_{xz}\circ I_{B_{j}}$ .

(5.3) is the rotation by $+90$ de-
$\gamma=R_{y}^{C}\circ I_{\pi(C,C)}\circ I_{C}$
, where $R_{y}^{C}$

grees around the oriented line through the center of $C$ parallel
to the positive direction of -axis and $\pi(C, C)$ is the bisector $y$

of the centers of $C$ and $C$ .

(S.4) $\delta=R_{z}^{D}\circ I_{\pi(D,D)}\circ I_{D}$
.
(@.5) $\epsilon=R_{x}^{E}\circ I_{\pi(E,E)}\circ I_{E}$
.

Denote the side of $P$ by $A_{j}^{*}=A_{j}\cap C1P$ , $B_{j}^{*}=B_{j}\cap C1P$ and so

forth. They satisfy the condition (H.I) of \S 7. That is,

(H. 1) $\alpha_{j}(A_{j}^{*})=A_{\acute{j}}^{*}$
, $\beta_{j}(B_{j}^{*})=B_{\acute{j}}^{*}$
, $\gamma(C^{*})=C^{J}*$ , $\delta(D^{*})=D^{l}*$ ,
$\epsilon(E^{*})=E^{J}*$
and $f(P)\cap P=\phi(f=\alpha_{1}, \ldots, \epsilon)$ .

Next we shall verify the condition (H.2) of \S 7 by listing up the cycles.

242 S. Matsumoto

First of all for any $1\leq j\leq n-1$ , we have the following cycle.

(C.i) $A_{j}\cap A_{j+1}\rightarrow A_{j}\alpha_{j}\cap A_{j+1}\alpha_{j+1}^{-1}\rightarrow A_{j}\cap A_{j+1}$

keeps points in $A_{j}\cap A_{j+1}$ fixed. By (P.4), it is a rotation

$\alpha_{j+1}^{-1}\alpha_{j}$

by $2\pi/14$ around $A_{j}\cap A_{j+1}$ . Therefore (H.2) is satisfied for $q=14$ .

Likewise the following cycle satisfies (H.2).

(C.i) $B_{j}\cap B_{j+1}\rightarrow B_{j}\beta_{j}\cap B_{j+1}\beta_{j+1}^{-1}\rightarrow B_{j}\cap B_{j+1}$

There are two more cycles. The first one is;

(C.i) $A_{1}\cap E\alpha_{1}\rightarrow A_{1}\cap E\epsilon\rightarrow B_{1}\cap E\beta_{1}^{-1}\rightarrow B_{1}\cap E\rightarrow A_{1}\epsilon^{-1}\cap E$

Computation shows that keeps $A_{1}\cap E$ pointwise fixed. It

$\epsilon^{-1}\beta_{1}^{-1}\epsilon\alpha_{1}$

is a rotation by $2\pi/7$ around $A_{1}\cap E$ and thus (H.2) is satisfied for

$q=7$ . Now the last cycle.

(C.i) $E\cap C\rightarrow E\epsilon\cap D\delta^{-1}\rightarrow B_{n}\cap D\beta_{n}^{-1}\rightarrow B_{n}\cap D\delta\rightarrow C\cap D$

$\rightarrow A_{n}\gamma^{-1}\cap C\alpha_{n}^{-1}\rightarrow A_{n}\cap C\gamma\rightarrow E\cap C$

By studying Figure (8.2), we obtain that yields the

$\gamma\alpha_{n}^{-1}\gamma^{-1}\delta\beta_{n}^{-1}\delta^{-1}\epsilon$

translation by $\pi/2$ on the circle $E\cap C$ . We also obtain (H.2) for

$q=4$ .

Thus by (7.2), the group generated by the side pairing transfor- $\Gamma$

mations is discrete, with the domain of discontinuity and $P$ is the $\Omega$

fundamental domain for . That is, we have the followings. $\Gamma$

(FD.1) $\Omega=\gamma\in\Gamma\cup\gamma(C1P)$
.

(FD.2) $\gamma(P)\cap P=\phi$ for any $\gamma\in\Gamma\backslash \{1\}$

.

Let be a torsion free subgroup of

$\Gamma_{0}$
of finite index. Then the $\Gamma$

quotient space is a compact flat conformal manifold and we

$N=\Omega/\Gamma_{0}$

have
$ L=L(N)=L_{P}(\Gamma_{0})=L_{P}(\Gamma)=\hat{R}^{3}\backslash \Omega$
.

The rest of this section is devoted to the proof of the following theorem.
 Flat Conformal Structure 243

Theorem (8.1). The limit set $L$

is a wild Cantor set.
The proof consists of a series of lemmas. The main part is to show
that is a Cantor set. First of all notice the following feature of our
$L$

construction. See Figure (8.3). For any side of $P$ , let $T$ be the $T^{*}$

2-sphere which contains and let be an edge. Since all the $T^{*}$ $e\subset\partial T^{*}$

translates of $P$ which gather at have angle $2\pi/28$ there, the part $e$

of $T$ which is opposite to w.r.t. and near is also a side of a $T^{*}$ $e$ $e$

translate of . That is, the side prolongs in the tesselation.

$P$

So far in the construction of , we have used the coordinates of $\Gamma$

However in the rest, we change the coordinates from to

$\hat{R}^{3}$ $\hat{R}^{3}$

. . $S^{3}$

Thus, distance, radius, etc. are measured in the Euclidean metric of $R^{4}$

which contains as a unit sphere $\{|x|=1\}$ . $S^{3}$

Let be an infinite sequence. Since is discrete, we have

$\{\gamma_{k}\}\subset\Gamma$
$\Gamma$

$\gamma_{k}\rightarrow\infty$ .

Figure (8.3)

Lemma (8.2). For any edge e of $P_{\rangle}$

we have radius $\gamma_{k}(e)\rightarrow 0$
.

Proof This follows at once from (5.8), since we have $ e\cap L=\phi$ .
Q.E.D.
Let
$\Sigma=\{\gamma(T)|\gamma\in\Gamma, T=A_{1}, \ldots, E\}$ ,
$\Lambda=$
{ $\gamma(e)|\gamma\in\Gamma$ , $e$
; an edge of $P$
}.
244 S. Matsumoto

Notice that consists of disjoint circles, while spheres in

$\Lambda$

may in- $\Sigma$

tersect. Furthermore at this point we do not know, for example, whether

or not it so happens that two spheres in are tangent. We have a $\Sigma$

control of since any of its circle is contained in

$\Lambda$

. However this is $\Omega$

not the case with . We only have a rather weak grip on $\Sigma$

. $\Sigma$

For , let$ S\in\Sigma$

$\Lambda_{S}=\{l \in\Lambda|l \subset S\}$ .

Take a base point $x_{0}\in S$ such that $x_{0}$ lies in a translate of the interior
of a side of $P$ .

Lemma (8.3). For $x\in S$ , $x$lies in if and only $\Omega$

if there exists
only fifinitely many circles in $\Lambda_{S}$
which separate from $x$ $x_{0}$ .

Proof. Notice first of all that circles in are mutually disjoint. $\Lambda_{S}$

To show the if part, let be a path in from to which meets

$p$
$S$ $x_{0}$ $x$

circles in at finitely many points. An induction on the number of

$\Lambda_{S}$

points shows that is contained in the union of translates of sides of

$p$

$P$ . In particular, we have . $ x\in\Omega$

For the converse, suppose that for a fixed edge of $P$ , $e$ $\gamma_{k}(e)$

$(1\leq k<\infty)$ separates from . By (8.2), we have that $x$ $x_{0}$

radius $\gamma_{k}(e)\rightarrow 0$
.

Notice that (FD.1) and (FD.2) implies that is disjoint from $\gamma_{k}(e)$

a small neighbourhood of . Since are mutually disjoint, $x_{0}$ $\gamma_{k}(e)$

we obtain that . Therefore the family cannot be

$\gamma_{k}(e)\rightarrow x$ $\{\gamma_{k}^{-1}\}$

equicontinuous on any neighbourhood of . That is, $x\in L_{J}(\Gamma)=L$ . $x$

(See (5.15).) Q.E.D.

Corollary (8.4). A connected component of $L\cap S$ is a singleton.

In particular is open and dense in . $\Omega\cap S$ $S$

Proof. This follows from the fact that $\gamma_{k}(e)\subset\Omega$

. Q.E.D.

A word of caution. In the above corollary, we do not assert that the

component of at a point of is a singleton.
$L$ $S$

In spirit we are going to show the total disconnectedness of in $L$

a way similar to (8.4) using spheres in instead of circles in . $\Sigma$

$\Lambda_{S}$

However as we remarked earlier, we do not have yet a good grip on how

looks like. The main diculty comes from the fact that $ S\cap L\neq\phi$
$\Sigma$

for . In what follows we shall carry out study of

$ S\in\Sigma$
step by step. $\Sigma$
 Flat Conformal Structure 245

Corollary (8.5). We have $ S\cap\gamma(P)=\phi$ for any $\gamma\in\Gamma$

and
$ S\in\Sigma$
. (Recall that $P$ is an open set.)

For contradiction, take a point $x\in S\cap\gamma(P)$ . By (FD.1),

Proof.
we have . Therefore by (8.3), there exist finitely many circles in
$ x\in\Omega$

separating
$\Lambda_{S}$
from . Now the argument of the first part of
$x$ $x_{0}$

(8.3) can be applied to show that is contained in the union of sides $x$

of translates of $P$ . That is, . A contradiction. Q.E.D. $x\not\in\gamma(P)$

Lemma (8.6). Let , $S$ $ S\in\Sigma$

. If $l$
$=S\cap S$ is a circle, then
$l$

. In particular, we have
$\in\Lambda$ $l$ $\subset\Omega$
.

By (8.4), there exists a point

Proof. . By (8.5), we have $x\in l\backslash L$

that for any . Likewise we obtain that

$x\not\in\gamma(P)$ does not
$\gamma\in\Gamma$ $x$

lie in a translate of the interior of a side of $P$ , since otherwise either

$S$
or would intersect some
$S$
. Therefore we have $x\in l$ for $\gamma(P)$

some . Should not coincide with , there would be another

$ l\in\Lambda$ $l$ $l$

sphere such that , and $S$ meet in general position at

$ S\in\Sigma$ $S$ $S$

$x\in\gamma(P)$ . Again one of the three spheres would intersect some $Q.E.D\gamma(P).\cdot$

A contradiction.

Lemma (8.7). Let . Suppose that for some $ S\in\Sigma$ $\gamma_{k}\in\Gamma$

, $\gamma_{k}(S)$

are distinct spheres. Then we have radius . $\gamma_{k}(S)\rightarrow 0$

Proof. Supposethe contrary. We may assume further that $\gamma_{k}(S)$

converges to a 2-sphere . Let us show first that . Take a

$S_{0}$ $ S_{0}\cap L\neq\phi$

point $a\in S$ and assume that . For any neighbourhood $\gamma_{k}(a)\rightarrow b\in S_{0}$

$U$ of , we have $b$

for arbitrary $j>>i>>1$ . That
$\gamma_{j}\gamma_{i}^{-1}(U)\cap U\neq\phi$

is, $b\in L_{P}(\Gamma)=L$ . (See (5.15).) On the other hand, we have . $S_{0}\not\subset L$

In fact, would imply that

$S_{0}\subset L$is not connected. However this is $\Omega$

impossible since the fundamental domain $P$ is connected. (Notice that

by the minimality (5.6) of , we have IntL $=\phi$ . Compare (5.22).) $L$

Consider a path in which combines a point of to a

$S_{0}$ $\Omega\cap S_{0}$

point of $L\cap S_{0}$ . As in the proof of (8.3), one finds a sphere $ S\in\Sigma$

which separates these two points. Clearly $S\cap S_{0}=l$ is a circle.

Since , we have that
$\gamma_{k}(S)\rightarrow S_{0}$ . By (8.6), we $\gamma_{k}(S)\cap S\rightarrow l$

have . Since are all distinct, we may assume

$\gamma_{k}(S)\cap S\in\Lambda$ $\gamma_{k}(S)$

(passing to a subsequence if necessary) that are all distinct. $\gamma_{k}\cap S$

This contradicts (8.2). Q.E.D.

Lemma (8.8). Fix once and for all $x_{0}\in P$ . A point $x\in S^{3}$
belongs to if and only if there exist only fifinitely many spheres in
$\Omega$ $\Sigma$

which separate from . $x$ $x_{0}$

246 S. Matsumoto

Suppose there exist infinitely many

Proof. $ S_{k}\in\Sigma$
. Then by (8.7),
. As in the proof of (8.3), we have
$diamS_{k}\rightarrow 0$ $x\in L$ . (In fact this
part will not be used in the sequel.)
Let us embark upon the proof of the converse. Define a closed subset
inductively as follows.
$Y_{j}\subset S^{3}$

$Y_{0}=C1P$.

$ Y_{j}=\cup\gamma C1P\gamma$
where $\gamma C1P\cap Y_{j-1}\neq\phi$ , for $j>0$ .

Define an open subset $X_{j}$ by

$X_{j}=S^{3}\backslash Y_{j}$
.

The set theoretic frontier is an angular surface (possibly with

$\partial X_{j}$

singularities) composed of the translates of sides of $P$ , which we call

sides of . We have a filtration
$X_{j}$

$(*)$ $C1X_{0}\supset X_{0}\supset C1X_{1}\supset X_{1}\supset C1X_{2}\supset X_{2}\cdots$ .

See Figure (8.4).

Figure (8.4)
 Flat Conformal Structure 247

Since $P$ is the fundamental domain for $\Gamma$

, we have

$L=j\geq 0\cap X_{j}$

.

For , the connected component of

$ S\in\Sigma$
opposite to the base $S^{3}\backslash S$

point is called the inside of

$x_{0}$ and is denoted by Inside(S). It $S$

is an open subset by definition. Let

$\Sigma_{j}=$
{ $S\in\Sigma|S$ contains a side of $X_{j}$
}.
Once we establish the following sublemma, a point $x\in L$ can be
shown to be inside infinitely many spheres in , completing the proof $\Sigma$

of (8.8). Q.E.D.

Sublemma (8.8.1).
(1) For any $S\in\Sigma_{j}$ , we have $S\subset C1X_{j}$ .
(2) We have
$ X_{j}=S\in\Sigma_{j}\cup$
Inside(S).

The following properties of $P$ , very easy to check, play a

Proof.
crucial part in th proof. Denote by a side of $P$ , by the sphere $T_{\alpha}^{*}$ $T_{\alpha}$

containing and by an edge of $P$ .

$T_{\alpha}^{*}$
$e_{\nu}$

(a) If , then we have .

$ T_{\alpha}^{*}\cap T_{\beta}^{*}=\phi$ $ T_{\alpha}\cap T_{\beta}=\phi$

(b) Suppose and let be an arbitrary sphere

$ T_{\alpha}^{*}\cap e_{\nu}=\phi$ $ S\in\Sigma$

which passes through . If , then contains a $e_{\nu}$

$ S\cap T_{\alpha}\neq\phi$ $S$

side of $P$ such that$S^{*}$

and . $ S^{*}\cap T_{\alpha}^{*}\neq\phi$ $e_{\nu}\subset\partial S^{*}$

(c) If , and
$ T_{\alpha}\cap T_{\beta}\neq\phi$ , then two of $ T_{\beta}\cap T_{\gamma}\neq\phi$ $ T_{\gamma}\cap T_{\alpha}\neq\phi$

the three spheres , and must coincide. $T_{\alpha}$ $T_{\beta}$ $T_{\gamma}$

The proof of (8.8.1) is by induction on . For $j=0$ , this is clear $j$

by the construction of $P$ . Let $j>0$ . Assume (8.8.1) for $j-1$ .

Proof of (1). For a given , let be a side of . $S\in\Sigma_{j}$ $S^{*}\subset S$ $X_{j}$

Choose a point in the interior of $x$ . By the filtration $(*)$ , we have $S^{*}$

$x\in X_{j-1}$ . The induction hypothesis implies that $x\in Inside(T)$ for
some $T\in\Sigma_{j-1}$ . Since , there exists a translate having
$S^{*}\subset\partial X_{j}$ $\gamma C1P$

as a side such that

$S^{*}$
$\gamma C1P\cap X_{j}=\phi$ . That is, . By the $\gamma C1P$ $\subset Y_{j}$

definition of , we have $\gamma C1P\cap\partial X_{j-1}\neq\phi$ . Since $x\in Inside(T)$ ,

$Y_{j}$

$7CIP$ must lie in $T\cup Inside(T)$ . Therefore we have

$\gamma C1P\cap T\cap\partial X_{j-1}\neq\phi$ .

248 S. Matsumoto

Clearly $\gamma C1P\cap T\cap\partial X_{j-1}$ is either a side or an edge of $\gamma C1P$ . Since
$S^{*}\subset X_{j-1}$ , we have $ S^{*}\cap(\gamma C1P\cap T\cap\partial X_{j-1})=\phi$ .
If $\gamma C1P\cap T\cap\partial X_{j-1}$ is a side, then it follows from (a) that $ S\cap T\neq\phi$ .
That is,
$S\subset Inside(T)\subset X_{j-1}$ .

Clearly this implies that $S\subset C1X_{j}$ .

Suppose on the contrary that $\gamma C1P\cap T\cap\partial X_{j-1}$ is an edge, that
is a circle . By (b), we obtain the same conclusion except in the
$l$
$\in\Lambda$

case where $T$

contains a side of $7CIP$ such that $ T^{*}\cap S^{*}\neq\phi$
$T^{*}$

and . See Figure (8.5). In this case choose a point $y\in S^{*}\cap T^{*}$ .
$T^{*}\supset l$

As before we obtain that $y\in Inside(T)$ for some $T\in\Sigma_{j-1}$ and that
$l$

. If $ S\cap T\neq\phi$ , then $T$ contains a side

$\subset T$
of $7CIP$ such $T^{l}*$

that $ T^{J}*\cap S^{*}\neq\phi$

and . This contradicts (c). Therefore we
$T^{l}*\supset l$

have $ S\cap T=\phi$ . As before we obtain $S\subset C1X_{j}$ .

Figure (8.5)

Proof of (2). By the construction $Y_{j}$

is connected. Therefore for
 Flat Conformal Structure 249

$S\in\Sigma_{j}$ , $S\subset C1X_{j}$ implies that Inside(S)\subset Xj. This shows

$s\in\Sigma_{j}\cup Inside(S)\subset X_{j}$

.

For the converse, consider a path $p$combining the base point $x_{0}$

to a given point $x\in X_{j}$ . Any such $p$ must intersect and hence $\partial X_{j}$

. Choose so that
$\bigcap_{S\in\Sigma_{j}}S$ $p$

(i) $p$does not pass through the intersection of two distinct spheres
of , $\Sigma_{j}$

(ii) the sum

$\sum_{S\in\Sigma_{j}}$
Card(p\cap S)

is the minimal.
Then for each , we have Card(p\cap S) . In fact, if not,
$S\in\Sigma_{j}$ $\leq 1$

one can find a subarc of such that and $q$ $p$ $\partial q\subset S$ $q\backslash \partial q\subset Inside(S)$

One can push out of Inside(S) in such a way that the numbers of
$q$

intersections of with the other spheres do not change. This contradicts

$p$

the minimality (ii).

We obtain that Card(p\cap S) $=1$ for some . That is, $S\in\Sigma_{j}$

$x\in Inside(5)$ , as is required.

Q.E.D.

At this point we need a concrete picture how $C1X_{1}$ and $C1X_{2}$ look
like. The picture of $C1X_{1}$ near $A_{j}\cap A_{j+1}$ , $A_{1}\cap E$ and $A_{n}\cap C$ are
shown in Figures (8.6)\sim (8.8).
The point is that Figure (8.8) shows that there occurs a separation of
components of $C1X_{1}$ near $A_{n}\cap C$ . As a matter of fact, the same thing
happens near any edge in the cycle of $A_{n}\cap C$ . Furthermore we find
a lot of separation of components of $C1X_{2}$ . In particular in $\epsilon^{-1}(C1P)$

which is inside $E$ , we observe that a component of $C1X_{2}\cap\epsilon^{-1}(C1P)$

which intersets do not intersect . See Figure (8.9).
$\epsilon^{-1}B_{1}$ $\epsilon^{-1}B_{1}$

The same thing happens inside $E$

. In summary we have the fol-
lowing.
Let $T$ , $T$ be any adjacent pair of 2-spheres chosen from , , $E$ . $A_{1}$
$\ldots$

Then a component of $ClX_{2}\backslash (Inside(T)\cup Inside(T))$ which inter-

sects $T$ does not intersect $T$ .
As a matter of fact, much more can be said concerning the smallness
of components of $C1X_{2}$ . However this is all that we need.
250 S. Matsumoto

$A_{j+1}$ $A_{j}$
$P$

Figure (8.6)

$A_{1}$ $E$
$P$

Figure (8.7)

Lemma (8.9). For arbitrary spheres , $ S\in\Sigma$ such that

$S$ $=$ $l$

$S\cap S$
is a circle, let $D$ (resp. $D$ ) be one of the disks in (resp.
$S$

$S)$ which is bounded by 1. Suppose that the angle of $D$ and $D$ at
$l$

is $2\pi/28$ . Let $Q$ be the closure of the component of $S^{3}\backslash (S\cup S)$

bounded by $D$ and $D$ . Then a component of $L\cap Q$ which intersects

Flat Conformal Structure 251

$A_{n}$

Figure (8. $S$

Figure (8.9)
252 S. Matsumoto

$D$ does not intersect $D$ .

Since we have
Proof. by (8.6), the proof reduces to the case
$l$ $\in\Lambda$

where is an edge of $P$ and $D$ and $D$ contains adjacent sides
$l$

of $P$ . Since $L\subset C1X_{2}$ , (8.9) follows from the above observation. See
Figure (8.10). Q.E.D.

Figure (8.10)

Another way to put (8.9) is the following.

Corollary (8.10). For arbitrary spheres , $ S\in\Sigma$ $S$

such that
$=S\cap S$ is a
$l$

the component of at a point

$ circle\rangle$ $L$ $x\in S\backslash S$ does
not intersect . $S$

Proof.
Let be the component of
$\triangle$

at and let be $S\backslash S$ $x$

the closure of either of the components of which contains $S^{3}\backslash (S\cup S)$

in its boundary. Then by (8.9), we obtain that for any

$\triangle$

, $y\in\triangle\cap L$

the component of at does not intersect . See Figure (8.11).

$\cup--\cap L$
$y$
$S$

It is easy to show that (8.10) follows from this. Q.E.D.

Lemma (8.11). Let $S$

be an arbitrary sphere in $\Sigma$

. For any
$x\in S\cap L$ , the component of $L$
at is
$x$
itself.
$\{x\}$
 Flat Conformal Structure 253

Figure (8.11)

By (8.8), there exists an infinite sequence

Proof. such $\{S_{k}\}\subset\Sigma$

that separates
$S_{k}$
from the base point $x$ . Note that . $ x_{0}\in\Omega$ $S_{k}\rightarrow x$

For large , intersects

$k$
at a circle. Therefore by (8.10), the
$S_{k}$ $S$

component of at does not intersect

$L$
. This completes the
$x$ $S_{k}$

proof. Q.E.D.

Corollary (8.12). For any $S\in\Sigma_{\rangle}$

the component of $L$
at $a$

point does not intersect

$x\in L\backslash S$ $S$
.

Corollary (8.13). $L$

is totally disconnected.

Let $x\in L$ . If $x\in S$ for some $ S\in\Sigma$ ,

Proof. then we have
already shown (See (8.11).) that the conponent of at $L$ $x$ is a singleton.
So consider the other case. By (8.8), there exist infinitely many spheres
which separate
$ S_{k}\in\Sigma$
from a base point $x_{0}\in P$ .
$x$ By (8.7), we
have . Therefore (8.12) implies (8.13).
$S_{k}\rightarrow x$
Q.E.D.

By (6.5), this implies that is a Cantor set. Thus we have finished $L$

the proof of the first part of Theorem (8.1). Let us show in the remainder
that is wild. First of all we have the following well known fact, which
$L$

is easy to show.

Proposition (8.14). If $\wedge r$

$\subset S^{n}$
is a tame Cantor set, then $S^{3}\backslash 1$

is simply connected.
254 S. Matsumoto

Thus once we establish that the inclusion : induces an $i$ $ P\rightarrow\Omega$

injection on the fundamental groups, then the proof of Theorem (8.1)

will be complete.
Some readers may have a feeling that this can be solved by looking
at the homomorphism

$\pi_{1}(P)\rightarrow\pi_{1}(i_{*}\Omega)\rightarrow\pi_{1}(p_{*}\Omega/\Gamma)$
.

The computation of , the fundamental group of an orbifold, is $\pi_{1}(\Omega/\Gamma)$

in fact easy. However order to show the nontriviality of the homomor-

in
phism , one is lead to the word problem of
$p_{*}oi_{*}$ , in which an $\pi_{1}(\Omega/\Gamma)$

approach geometric in nature is obviously indispensable. Instead of go-

ing to this direction, we employ the following argument which is totally
geometric and is applicable also to the word problem of . $\pi_{1}(\Omega/\Gamma)$

The key fact is the following lemma.

Lemma (8.15). Let be a side $T^{*}$

of $C1P\rangle$
then the inclusion
induces an injecti on on
$T^{*}\rightarrow C1P$
. $\pi_{1}$

By sliding handles of
Proof. , one obtains that $C1P$ is $S^{3}\backslash P$

a handlebody of genus 2. Therefore $\pi_{1}(C1P)$ is a free group freely

generated by and . If $T^{*}\neq E^{*}$ or $E*$ , then the lemma follows
$\alpha$ $\beta$

easily. For $T^{*}=E^{*}$ , the image of is generated by and $\pi_{1}(E^{*})$ $\alpha$

. It is well known, easy to show using the once puctured

$\alpha\beta\alpha^{-1}\beta^{-1}$

torus model, that they generate free subgroups. Q.E.D.

Now let us embark upon the proof. Let : be a loop such $\alpha$
$S^{1}\rightarrow P$

that in $P$ . Suppose on the contrary that

$\alpha\not\simeq 1$
in . Let $\alpha\simeq 1$ $\Omega$

:
$\beta$
be the extension of . By a small perturbation, one may
$ D^{2}\rightarrow\Omega$
$\alpha$

assume that is smooth and transverse to any circle in

$\beta$
and to any $\Lambda$

sphere in . Their inverse images form a graph

$\Sigma$
$G$ in . ( $G$ may $D^{2}$

contain smooth circles as connected components.) As a matter of fact,

we have $ G\neq\phi$ and $ G\cap S^{1}=\phi$ . See Figure (8.12).
Let us choose so $\beta$

(M.I) the number of vertices of $G$ is the minimal,

(M.2) the number of edges of $G$ is the minimal among those which
satisfy (M. 1).
Let be a connected component of
$\triangle$

which is homeomorphic $D^{2}\backslash G$

to an open disk. Then for some . Since is $\beta(\triangle)\subset\gamma P$ $\gamma\in\Gamma$ $\beta$

transverse to , we have that is a simple closed curve.

$\gamma\partial P$
$\partial\triangle$

Claim (8.16). The number of vertices of $\partial\triangle$

$is\geq 3$ .
 Flat Conformal Structure 255

Figure (8.12)

If not,
Proof. is contained in the union of adjacent two
$\beta(\partial\triangle)$
$\mathcal{T}$

sides of for some . Clearly $T$ is homotopic in $7CIP$ to a

$\gamma\partial P$ $\gamma\in\Gamma$

single side. See Figure (8.2). Since is null homotopic in , $\beta(\partial\triangle)$ $\gamma C1P$

we obtain by (8.15) that is null homotopic in . But then we $\beta(\partial\triangle)$

$\mathcal{T}$

can alter the map so that and eventually push

$\beta$
out of $\beta(\triangle)\subset \mathcal{T}$ $\beta$

$7CIP$ . This contradicts the minimality assumption (M.I) if has a $\partial\triangle$

vertex and (M.2) otherwise. Q.E.D.

Now consider the family of smooth circle components of $G$ . Let $l$

be the innermost one and let $V$ be the open ball bounded by . In $l$

case there is no smooth circles, let $V=IntD^{2}$ . By (8.16), there must

exist components of $G$ in $V$ . Consider $G=G\cap V$ . $G$ has no
longer a smooth circle component. Let , , be the connected $G_{1}$
$\ldots$
$G_{r}$

component of $G$ . Let be the component of which contains $E_{i}$ $V\backslash G_{i}$

and let
$\partial V$

$H(G_{i})=V\backslash E_{?}.$ .

Notice that is a simple closed curve since it is the inverse image

$\partial H(G_{i})$

by of a surface
$\beta$
for some . Therefore $H(G_{i})$ is a closed
$\gamma\partial P$ $\gamma\in\Gamma$

disk. Define a partial order in the set by $\prec$ $\{G_{1}, \ldots, G_{r}\}$

$G_{j}\prec G_{i}\Leftrightarrow H(G_{j})\subset H(G_{i})$ .

256 S. Matsumoto

Let be the minimal element. Then any component of

$G_{j}$ is $H(G_{j})\backslash G_{j}$

an open disk. That is, gives a polyhedral decomposition of $H(G_{j})$ .

$G_{j}$

Let , $f$ and $e$

be the number of faces, edges and vertices of the
$v$

decomposition. By virtue of (8.16), we have

$3f\leq 2e$ .

Notice that by (P.4), exactly 28 edges gather at each vertex of $G_{j}$ .

Therefore we have
$14v$ $=e$ .

The computation of Euler number yields;

$1=f-e+v\leq\frac{2}{3}e-e+\frac{e}{14}<0$ .

This contradiction shows that $i_{*}$

: $\pi_{1}(P)\rightarrow\pi_{1}(\Omega)$
is an injection, as is
requied.

Appendix End

The concept of end of a topological space and of a discrete group was

first introduced in 1931 by Freudenthal ([11]) and was studied, among
others, by Hopf ([23]). See also Freudenthal [12] and Epstein [8]. After
almost 40 years, Stalling ([54], [55]) established a celebrated theorem
concerning finitely generated groups with infinite ends. See Dunwoody
[7] for related topics and a geometric proof of Stallings theorem for
finitely presented groups. All this has a wide range of applications. For
the convenience of the reader, we collect here some parts of the theory,
mostly without proof.
First of all we define the ends of a connected locally finite simplicial
complex $U$ .

Definition (A. ). A sequence $I$

$\{M_{k}\}$ of subsets of $U$ is called
discrete if for any compact subset $C$ of $U$ ,
we have $ M_{k}\cap C=\phi$ for
but finitely many . $k$

Definition (A.2). A point sequence $\{x_{k}\}\subset U$ is called admis-

sible if for any $k>0$ , there exists a path $P_{k}\subset U$ combining and $x_{k}$

$x_{k+1}$ such that the family is discrete. $\{P_{k}\}$

Definition (A.3). Two admissible sequence and are $\{x_{k}\}$ $\{x_{k}\}$

said to be equivalent, (denoted by $\{x_{k}\}\sim\{x_{k}\}$ ) if and only if there

exists a path $(k>0)$ combining
$P_{k}$
and such that the family $x_{k}$ $x_{k}$
 Flat Conformal Structure 257

is discrete. An equivalence class of admissible sequences are called

$\{P_{k}\}$

an end of U. The set of the ends of U is denoted by . $\mathcal{E}(U)$

It is easy to show that the relation in (A.3) is in fact an equivalence

relation. Notice that a subsequence of an admissible sequence is again
admissible and in the same equivalence class.
For applications to flat conformal structures, we need the following.
Proposition (A.4). For an open domain $U$ of $(n\geq 2)$ , $S^{n}$

the set of ends is in one to one correspondence with the set of

$\mathcal{E}(U)$

connected components of $Y=S^{n}\backslash U$ .

Proof. First of all let us define a correspondence of an end to a

connected component. Let be an admissible sequence of $U$ . $\{x_{k}\}$

Then we have $d(x_{k}, Y)\rightarrow 0$ . Let us show furthermore that there exists
a unique connected component of $Y$ such that $Y_{\nu}$ $ d(x_{k}, Y_{\nu})\rightarrow$

. Suppose the contrary. Then there exist subsequences

$0$
and $\{y_{k}\}$

of
$\{z_{k}\}$
such that$\{x_{k}\}$ and for
$d(y_{k}, Y_{\nu})\rightarrow 0$ $d(z_{k}, Y_{\mu})\rightarrow 0$

disjoint components and of $Y$ . Then there

$Y_{\nu}$
exists a compact
$Y_{\mu}$

neighbourhood $B$ of in such that $ B\cap Y_{\mu}=\phi$ and

$Y_{\nu}$ $S^{n}$
. $\partial B\subset U$

Then any path in $U$ combining

$P_{k}$
and must intersect the $y_{k}$ $z_{k}$

compact set . This contradicts the fact that

$\partial B$
$\{y_{k}\}\sim\{z_{k}\}$ . The
same argument shows that the component thus chosen is independent of
the particular choice of an admissible sequence in the equivalence class.
Thus an end corresponds to a connected component.
The converse correspondence is defined as follows. For any con-
nected component of $Y$ , we can find a sequence
$Y_{\nu}$
of compact $\{B_{k}\}$

connected neighbourhoods of in such that and that

$Y_{\nu}$ $S^{n}$ $\partial B_{k}\subset U$

. Furthermore one may assume that

$\bigcap_{k}B_{k}=Y_{\nu}$ is a finite union of $\partial B_{k}$

codimension one connected submanifold. Notice that any codimension

one connected submanifold splits $(n \geq 2)$ into two parts. Since $U$ $S^{n}$

is connected, this shows that is arcwise connected. Choose $IntB_{k}\backslash Y$

an arbitrary point $x_{k}\in intBk-$ Combine and $x_{k+1}$ by a path $x_{k}$ $P_{k}$

in . This shows that

$IntB_{k}\backslash Y$ is an admissible sequence. Q.E.D. $\{x_{k}\}$

The ends of a group is defined by virtue of the following theorem.

Theorem (A.5). Let be fifinitely generated group which acts $\Gamma$
$a$

on a connected locally fifinite simpli cial complex $U$ freely and discontin-
uously such that the quotient is compact. Then the set of ends $ U/\Gamma$

is determined (up to a bijection) only by the group

$\mathcal{E}(U)$
. It does $\Gamma$

not depends upon the particular choice of the space $U$ .

Definition (A.6). The set of ends in (A.5) is called the end set
of the group and is denoted by .
$\Gamma$
$\mathcal{E}(\Gamma)$
258 S. Matsumoto

Theorem (A.7). The end set $\mathcal{E}(\Gamma)$

of $a$
fifinitely generated group
$\Gamma$

is infifinite [in fact uncountably infifinite) if $Card\mathcal{E}(\Gamma)\geq 3$ .

We have the following characterization of the group according to its

end set.
Theorem (A.8). Let $\Gamma$

be $a$
fifinitely generated group.
(1) $\mathcal{E}(\Gamma)=\phi$
if and only if is fifinite group.
$\Gamma$
$a$

(2) $\mathcal{E}(\Gamma)$
consists of two points if and only if $\Gamma$

has the infifinite

cyclic group as fifinite index subgroup.
$Z$ $a$

For a group with infinitely many ends, Stalling obtained a complete

characterization. However for the sake of simplicity we only state the
following partial result.
Theorem (A.9). $A$ fifinitely generated torsion free group has $\Gamma$

infifinite ends if and only if has a nontrivial decomposition as a free

$\Gamma$

product . $\Gamma=\Gamma_{1}*\Gamma_{2}$

As an application, if a torsion free group in (A.9) acts on a $\Gamma$

domain $U\subset S^{n}$ freely and discontinuously and if the complement

has more than two components, then
$S^{n}\backslash U$
is a nontrivial free $\Gamma$

product.

References
[1] B.N. Apanasov, On non-connectedness of the space of uniformized con-
formal structures on hyperbolic manifolds, Institut Mittag-Leer,
Djursholm Report No.1 (1989/90), p. 19.
[2] A.F. Beardon, The geometry of discrete groups, Springer Verlag, 1983.
[3] L. Bers, On boundaries of Teichm\"uller spaces and on Kleinian groups,
Ann. Math., 91 (1970), 570-600.
[4] M. Bestvina and D. Cooper, A wild Cantor set as the limit set of a
conformal group action on , Proc. A.M.S., 99 (1987), 623-626. $S^{3}$

[5] L. Bieberbach, \"Uber die Bewegungsgruppen der Euclidischer R\"aume I:,

Math. Ann., 70 (1911), 297-336; II, Math. Ann., 72 (1912), 400-412.
[6] P. Buser, A geometric proof of Bieberbachs theorems on crystallo-
graphic groups, LEnseignement Math., 31 (1985), 137-145.
[7] M.J. Dunwoody, The accessibility of finitely presented groups, Invent,
math., 81 (1985), 449-457.
[8] D.B.A. Epstein, Ends, in Topology of -Manifolds and Related Topics, $3$

Printice-Hall, 1962, pp. 110-117.

[9] D.B. A. Epstein, R. Canary and D. Green, Notes on notes of Thurston,
in Analytic and geometric aspects of hyperbolic spaces, London
Math. Soc. Lecture Notes 111, Cambridge University Press, 1987.
 Flat Conformal Structure 259

[10] M.H. Freedman and R. Skora, Strange actions of groups on spheres, J.

Dierential Geom., 25 (1987), 75-98.
[11] H. Freudenthal, \"Uber die Enden topologisher R\"aume und Gruppen,
Math. Z., 33 (1931), 692-713.
[12] , \"Uber die Enden diskreter R\"aume und Gruppen, Comment. Math.

Helv., 17 (1945), 1-38.

[13] D. Fried, Closed similarity manifolds, Comment. Math. Helv., 55 (1980),
576-582.
[14] F.W. Gehring, Rings and quasiconformal mappings in spaces, Trans.
A.M.S., 103 (1962), 353-393.
[15] W.M. Goldman, Conformally flat manifolds with nilpotent holonomy
and the uniformization problem for -manifolds, Trans. A.M.S., 278
$3$

(1983), 573-583.
[16] , Projective structures with Fuchsian holonomy, J. Dierential

Geom., 25 (1987), 297-326.

[17] W.M. Goldman and Y. Kamishima, Conformal automorphisms and
closed conformally flat manifolds, to appear in Trans. A.M.S. (1990).
[18] , Topological rigidity of developing maps with applications to flat

conformal structures, Contemporary Math., 74, 199-204.

[19] M. Gromov, H. B. Lawson and W. Thurston, Hyperbolic -manifolds
$4$

and conformally flat -manifolds, Publ. Math. I.H.E.S., 68, 27-45.

$3$

[20] N. A. Gusevski and M. E. Kapovich, Conformal structures on three

dimensional manifolds, Soviet Math. Dokl., 35 (1987), 314-318.
[21] P. Hartman, On isometries and on a theorem of Liouville, Math. Z., 69
(1958), 202-210.
[22] J. Hempel, -manifolds, Ann. Math. Studies, Princeton University
$3$

Press, 1978.
[23] H. Hopf, Enden oener R\"aume und unendliche diskontinuierliche Grup-
pen, Comment. Math. Helv., 15 (1943), 27-32.
[24] D. Johnson and J. J. Milson, Deformation spaces associated to compact
hyperbolic manifolds, Progress in Mathematics, 67 (1987), 48-106.
[25] Y. Kamishima, Conformally flat manifolds whose developing maps are
not surjective, Trans. A.M.S., 294 (1986), 607-623.
[26] M. E. Kapovich, Some properties of developments of conformal struc-
tures on -manifolds, Soviet Math. Dokl., 35 (1987), 146-149.
$3$

[27] M.E. Kapovich, Flat conformal structures on -manifolds.

$3$

[28] , Deformation of group representations and uniformization of

$3$ -

manifolds, to appear in Proceedings of International Conference Au-

tomorphic forms and related topics, Khabarovsk.
[29] , On absence of Sullivans cusp finiteness theorem in higher dimen-

sions, in Abstracts of International Algebraic Conference Dedicated

to the Memory of A.I.Maltsev, Nobosibirsk, 1989.
[30] , Deformation space of flat conformal spaces, to appear in Answers

and Questions in General Topology.

[31] , On absence of Sullivans finiteness theorem, Preprint, Khabarovsk.
260 S. Matsumoto

[32] M.E. Kapovich and L. Potyagailo, On absence of Ahlfors finiteness the-

orem for Kleinian groups in dimension 3, To appear in Topology and
Applications.
[33] C. Kourouniotis, Deformations of hyperbolic structures on manifolds of
several dimensions, Dissertations, Kings College London.
[34] S.L. Krushkal, B. N. Apanasov and N. A. Gusevski , Kleinian groups
and uniformization in examples and problems, Transl. Math. Mono.
A.M.S. vol 62, 1986.
[35] N.H. Kuiper, On conformally flat spaces in the large, Ann. Math., 50
(1949), 916-924.
[36] , On compact conformally Euclidean spaces of dimension $>2$ ,

Ann. Math., 52 (1950), 478-490.

[37] , On convex locally projective surfaces, in Convegno Interna-
tionale di Geometria Dierenziale, Rome, 1953.
[38] , Hyperbolic manifolds and tesselations, IHES Publ. Math., 68

(1989), 76-97.
[39] , Fairly symmetric hyperbolic manifolds, Preprint IHES.

[40] R.S. Kulkarni, Conformally flat manifolds, Proc. Nat. Acad. Sci. USA,
69 (1972), 2675-2676.
[41] , Conformal geometry in higher dimensions I, Bull. AMS, 81
(1975), 736-738.
[42] , On the principle of uniformization, J. Dierential Geom., 13

(1978), 109-138.
[43] , Groups with domains of discontinuity, Math. Ann., 237 (1978),

253-272.
[44] , Conjugacy classes in $M(n)$ , in Conformal Geometry (ed. R.S.
Kulkarni and U. Pinkall), Aspects of Math., Vieweg.
[45] R.S. Kulkarni and U. Pinkall, Uniformization of geometric structures
with application to conformal geometry, Springer LNM, 1209, 190-
209.
[46] J. Liouville, Extension au cas des trois dimensions de la questions du
trac\e geographic, in Application de lAnlyse \a la Geometrie, Paris,
1850, pp. 609-616.
[47] B. Maskit, Kleinian groups, Springer-Verlag, 1987.
[48] S. Matsumoto, Topological aspects of conformally flat manifolds, Proc.
Japan Acad., 65 (1989), 231-234.
[49] R. Nevanlinna, On dierentiable mappings, in Analytic Functions,
Princeton University, 1960, pp. 3-9.
[50] L.D. Potyagailo, Schottky groups and three dimensional conformal
manifolds with trivial automorphism groups, Soviet Math. Dokl., 31
(1985), 509-512.
[51] , Kleinian groups in the space which are isomorphic to funda-

mental groups of Haken manifolds, (in Russian), Soviet Math. Dokl.,

1988.
 Flat Conformal Structure 261

[52] R. Schoen and S.T. Yau, Conformally flat manifolds, Kleinian groups
and scalar curvatures, Invent. Math., 92 (1988), 47-71.
[53] A. Selberg, On discontinuous groups in higher dimensional symmetric
spaces, in Contribution to Function Theory, Bombey, Tata Insti-
tute, 1960, pp. 147-164.
[54] J.R. Stallings, On torsion free groups with infinitely many ends, Ann.
Math., 88 (1968), 312-334.
[55] , Group theory and three dimensional manifolds, Yale Math.
Monograph no.4, Yale University Press.
[56] W. Thurston, Geometry and topology of -manifolds, Princeton Uni-
$3$

versity Lecture Notes, 1978.

[57] J. Tits, Free subgroups in linear groups, J. Algebra, 20 (1972), 250-270.
[58] R. Miner, Spherical CR manifolds with amenable holonomy, Interna-
tional J. Math., 1 (1990), 479-501.

Department of Mathematics
College of Science and Technology
Nihon University
Kanda-Surugadai, Tokyo 101
Japan
Advanced Studies in Pure Mathematics 20, 1992
Aspects of Low Dimensional Manifolds
pp. 263-299

Deformation Spaces on Geometric Structures

Yoshinobu Kamishima
and
Ser P. Tan

Contents
0. Introduction
1. Canonical decomposition of conformally flat manifolds
2. Thurston parametrization of projective structure on surfaces
3. Projective structures on surfaces and holonomy function groups
4. $(G, X)$ -structures on manifolds

5. invariant geometric structures

$S^{1}$

0. Introduction

In this note we shall study geometric structures on smooth manifolds

and deformation spaces. In 1981 Thurston gave a lecture on projective
structures on surfaces in which he has established the following structure
theorems (unpublished):
I. There is a canonical decomposition by convex hulls on a hy-
perbolic surface which admits a (one dimensional complex)
$S$

projective structure.
$II$
. There is an isomorphism between the deformation space
$CP^{1}(S_{g})$ and the product . $\mathcal{T}(S_{g})\times A4\mathcal{L}(S_{g})$

Here is the Teichm\"uller space of a closed orientable surface

$\mathcal{T}(S_{g})$
of $S_{g}$

genus $g\geqq 2$ and is the space of measured laminations.

$\Lambda 4\mathcal{L}(S_{g})$

Since there was considerable interest in the argument of proof and the
key idea seems to be generalized in higher dimension, we have decided
to write down an exposition of the above structure theorems (I), (II).
(Complex) projective structure on surfaces is equivalent to con-
formally flat structure on surfaces when we identify $CP^{1}=S^{2}$ and

Received September 25, 1990.

Revised March 22, 1991.
264 Y. Kamishima and S. Tan

$PSL_{2}(C)=PO(3,1)^{0}$ .We shall generalize I to conformally flat struc-

ture on manifolds in arbitrary dimensions.
In Chapter 1, we show that there is a canonical decomposition of a
conformally flat manifold. The above Thurston correspondence will be
proved in Chapter 2. There is no originality concerning the argument
of Chapter 2 except for a certain generalization. It is nothing but our
interpretation of Thurstons lecture. In Chapter 3 we shall describe
various projective structures by using Kleinian groups. In Chapter 4
we review $(G, X)$ -structures and examine the properties of limit sets of
$(G, X)$ -manifolds. The deformation space for $(G, X)$ -structures will be
defined more generally. As an application, we study the deformation
spaces of invariant geometric structures in Chapter 5. Particularly
$S^{1}$

we treat spherical $CR$ structures and conformally flat structures as such

geometric structures.
The authors have been informed from Professor William Goldman,
and Professor Sadayoshi Kojima that Kulkarni-Pinkall also showed the
existence of canonical stratification of conformally flat manifolds (cf.
[34] . $)$

We would like to thank Professor William Goldman for showing us

his note of the Thurstons lecture. And we also thank the referee to
pointing out our mistakes in earlier draft.

1. Canonical decomposition of conformally flat manifolds

A conformally flat structure on a smooth -manifold is a maxi- $n$

mal collection of charts modelled on the standard -sphere whose $n$ $S^{n}$

coordinate changes lie in the group Conf(5n) of conformal transfor-

mations of . The group Conf(5n) is isomorphic to the Lorentz
$S^{n}$

group $PO(n +1,1)$ . If a smooth -manifold $M$ admits a confor- $n$

mally flat structure, then by the monodromy argument there exists

a developing pair $(\rho, dev)$ , where $dev$ : is a conformal im-
$\tilde{M}\rightarrow S^{n}$

mersion and : $\pi_{1}(M)$

$\rho$ Conf(Sn) is a homomorphism such that
$\rightarrow$

. $dev(\gamma\in\pi_{1}(M))$ . Here
$dev\cdot\gamma=\rho(\gamma)$ is the universal covering space
$\tilde{M}$

of $M$ and $\pi_{1}(M)$ is the fundamental group. The map $dev$ is called a de-
veloping map and is called a holonomy homomorphism, both unique
$\rho$

up to an element of $Conf(S^{n})$ . Remark that the term conformal means

in the category $(Conf(S^{n}), S^{n})$ , which is dierent from the usual ter-
minology when $dim=2$ .
 Deformation Space 265

1.1. Maximal balls

Definition 1.1.1. Let $H^{n+1}$ be a (real) hyperbolic space with

boundary $\partial H^{n+1}=S^{n}$ . The dimensional sphere is a conformally $n$ $S^{n}$

flat manifold by stereographic projection. A geometric -sphere is $k$ $S^{k}$

the boundary of a $(k+1)$ dimensional totally geodesic subspace of $H^{n+1}$ .

A geometric ball is a domain of bounded by a codimension one geo- $S^{n}$

metric sphere.

Definition 1.1.2. Let $N$ be a conformally flat manifold. Given

a conformal immersion : , a geometric ball of $N$ is an open $f$ $N\rightarrow S^{n}$

subset $U$ such that : $U\rightarrow f(U)$ is a dieomorphism onto a geometric

$f$

ball of . Then the set of geometric balls of $N$ is partially ordered by

$S^{n}$

inclusions. We call a maximal geometric ball a maximal ball.

The following is a generalization of the proposition due to Thurston.

Proposition 1.1.3. Let : $f$ $N\rightarrow S^{n}$

be a conformal immersion.
Then either one of the following is true.
(i) is conformally equivalent to the standard sphere
$N$ $S^{n}$
, $a$ eu-
clidean space , or a hyperbolic space . $R^{n}$ $H^{n}$

(ii) Every point of $N$ lies in a proper maximal ball.

Suppose that (ii) is false. A point of $N$ lies in some geo-

Proof.
metric ball but not in a maximal ball. And so there exists a sequence
of geometric balls containing . The union
$ U_{1}\subset U_{2}\subset\cdots\subset U_{i}\subset\cdots$ $x$

$W=\bigcup_{i=1}^{\infty}U_{i}$ is not a geometric ball. As $f$

is injective on each $U_{i}$
, $f$
must
map isomorphically onto a euclidean space $R^{n}(\approx S^{n}-\{\infty\})$ . If
$W$

$N\neq W$ then maps the closure isomorphically onto

$f$
. And thus $\overline{W}$
$S^{n}$

it follows that $N=\overline{W}$

. This proves (i). If some maximal ball $U$ is not
proper, then $N=U$ . Since the image $f(U)$ is a geometric ball of , it $S^{n}$

is conformally equivalent to a hyperbolic space . Q.E.D. $H^{n}$

Let : be a conformal immersion. The spherical metric

$f$ $N\rightarrow S^{n}$

on defines a Riemannian metric on $N$ so that is a local isometry.

$S^{n}$ $f$

Let be the metric completion of $N$ . It is easy to see that extends

$\overline{N}$

$f$

to a map . Recall that for a maximal ball $U$ , : is a

$\overline{f}:\overline{N}\rightarrow S^{n}$
$f$ $U\rightarrow B$

dieomorphism onto an dimensional ball . Note that the closure of

$B$ $n$

$U$ in $N$ is not compact by maximality. However we

have

(1.1.4) $\overline{f}:\overline{U}\rightarrow\overline{B}$
266 Y. Kamishima and S. Tan

is a homeomorphism onto a closed ball of . Denote by the bound- $S^{n}$ $\partial U$

ary of in . Let $M^{n}$

be a closed conformally flat manifold and
$\overline{U}$ $\overline{N}$

$(\phi, dev)$ : $(\pi_{1}(M),\tilde{M})\rightarrow(Conf(S^{n}), S^{n})$ be the developing pair as in

1.1 of Chapter 1. Recall that $dev$ is a conformal immersion. We have

the following application.

Lemma 1.1.5. Let $M^{n}$ be a closed conformally flat manifold and

be the set of all maximal balls of the universal covering space
$\mathcal{F}$

. If $\tilde{M}$

the boundaries of all the elements of meet at a common point, then $\mathcal{F}$

the developing map is a covering map.

We put $\tilde{M}=N$ , $\pi_{1}(M)=\Gamma$ , $dev=f$ . Let

Proof. be the $\overline{N}$

metric completion of $N$ and : be the map extending . Let $\overline{f}$

$\overline{N}\rightarrow S^{n}$
$f$

be a common
$x$ point of for all . Note that $x\not\in N$ , otherwise $\partial U$ $U\in \mathcal{F}$

$x$would be an interior point of some maximal ball. Put . $\overline{f}(x)=\infty$

We prove that misses the point . Suppose that there is $y\in N$

$f$ $\{\infty\}$

such that $ f(y)=\infty$ . The point lies in some maximal ball $U$ . Since $y$

is a homeomorphism and
$\overline{f}:\overline{U}\rightarrow\overline{B}$
, it is impossible. Therefore, $x\in\partial U$

as the developing map misses a point, it is a covering map onto its image
(cf. [25] , [34]). Q.E.D.

Proposition 1.1.6. Let $M^{n}$ be a closed conformally flat mani-

fold. Suppose that consists of fifinite elements (possibly empty) or the $\mathcal{F}$

boundaries of all the elements of meet at fifinite number of common $\mathcal{F}$

$a$

points. Then $M$ is conformally equivalent to a spherical space form, $a$

Hopf manifold, euclidean space form, or a hyperbolic space form. $a$

Suppose the latter case. By the above lemma we have that

Proof.
$dev$ : is a covering map. Since the com-
$\tilde{M}\rightarrow dev(\tilde{M})\subset S^{n}-\{\infty\}$

mon points are finite, the fundamental group $\pi_{1}(M)$ has a subgroup $\pi$

of finite index those elements of which leave these points fixed. And
so the holonomy subgroup belongs to the similarity subgroup of $\phi(\pi)$

$Conf(S^{n})$ which is the stabilizer at in . Therefore $dev$ is a home- $\{\infty\}$ $S^{n}$

omorphism of either or $S^{n}-\{0, \infty\}$ (cf. [14]). In our case $dev$ is a $R^{n}$

homeomorphism onto $S^{n}-\{0, \infty\}$ or $M$ is a Hopf manifold.

For the remaining case, if is empty then $M$ is either a spherical space $\mathcal{F}$

form or a euclidean space form. Suppose that consists of finite ele- $\mathcal{F}$

ments. Then is covered by the union of those finite maximal balls

$\tilde{M}$

$U$ . It follows that the number of $dev^{-1}(x)$ is finite for each $x\in S^{n}$ . It

is easy to see that $dev$ : is a finite covering map. Passing $\tilde{M}\rightarrow dev(\tilde{M})$

to a subgroup of finite index in if necessary we can assume that $\pi$ $\pi$

leaves each element $U$ of invariant. Let be the holonomy group to $\mathcal{F}$ $\Gamma$
 Deformation Space 267

$\pi$
. Since the image of misses more than one point, we know that
$dev$

$dev(\tilde{M})\subset S^{n}-L(\Gamma)$ . On the other hand, note that is discrete be- $\Gamma$

cause $dev$ : is a homeomorphism. This implies that $L(\Gamma)\subset S^{n-1}$

$U\rightarrow B$

where we view $\partial B=S^{n-1}$

. Moreover acts properly discontinuously on $\Gamma$

$S^{n}-L(\Gamma)$ by (1.1.5). In particular it follows that either $dev(\tilde{M})=H^{n}$

or $dev(\tilde{M})=S^{n}-L(\Gamma)$ . The former case implies that $M$ is a hyperbolic

space form. In the latter case the set of all maximal balls in $S^{n}-L(\Gamma)$
must be finite. However if we note that $ S^{n-1}-L(\Gamma)\neq\emptyset$ , it is easy to see
that for any point $x\in S^{n-1}-L(\Gamma)$ there are infinitely many maximal
balls containing . This is impossible in this case. $x$ Q.E.D.

Note. If $M$ is a Hopf manifold, every maximal ball of $\tilde{M}$

meets
at exactly two points.

1.2. Decomposition of conformally flat manifolds

Let : $f$
be a conformal immersion and
$N\rightarrow S^{n}$ $\mathcal{F}$

the set of all

maximal balls of $N$ . Let $U$ be an element of . Put $\mathcal{F}$

$U_{\infty}=\overline{U}-N$
.

Then $\partial U$
decomposes into a disjoint union of $\partial U\cap N$
and $U_{\infty}$
.

Definition 1.2.1. The set is called an ideal set of $U$ . The $U_{\infty}$

ideal set is a closed subset of . (For example, if is a closed disk, $\overline{N}$ $\overline{U}$

then may look like a Cantor set and

$U_{\infty}$
is a disjoint union of $\partial U\cap N$

intervals. Since $U$ has the natural Poincar\e metric, corresponds to $U_{\infty}$

a closed subset of points at infinity.)

Recall from (1.1.4) that is conformally equivalent to a closed ball $\overline{U}$

. We may form the convex hull

$\overline{B}$

for inside U. (Note that this $C(U_{\infty})$ $U_{\infty}$

can be defined when contains more than one point.) Let $D^{m+1}=$
$U_{\infty}$

$H^{m+1}\cup S^{m}$
be the compactification of a hyperbolic space $H^{m+1}$ . If $K$
is a closed subset of then we denote by $H(K)$ the convex hull in
$S^{m}$

$H^{m+1}$ . it is easy to check the following.

Lemma 1.2.2. Let $P\subset H^{m+1}$ be a totally geodesic hyperplane

such that either one of the components of $S^{m}-\partial P$ does not meet $K$ .
Then $H(K\cap\partial P)=H(K)\cap P$ .

Using this lemma we define pleats on the boundary of $\partial C(U_{\infty})$

. Given a closed convex set $C$ of

$C(U_{\infty})$ $D^{n+1}$
there is a canonical re-
traction : $D^{n+1}\rightarrow C$
called a closest point mapping. Recall that
$\Phi_{C}$

if $x\in S^{n}-C$ there is a horoball centered at disjoint from C. Then $x$

268 Y. Kamishima and S. Tan

$\Phi_{C}(x)$ is the point of the first contact when we increase the radius of
this horoball continuously until it touches C. See [8] for details. Denote
by the complement of $fU$ in
$fU^{c}$ . If $C=\overline{H}(fU^{c})$ is the closure of $S^{n}$

the convex hull $H(fUc)$ in $H^{n+1}$ , then we have a map : $D^{n+1}\rightarrow C$ . $\Phi_{C}$

Note that is a totally geodesic subspace of $H^{n+1}$ , so we set

$7\{(\partial(fU^{c}))$

$\prime\mu(\partial(fU^{c}))=H^{n}$ . Since , the above map restricts to a

$\overline{H}(\partial(fU^{c}))\subset C$

map
$\Phi_{U}$
: $fU(=B)\rightarrow H^{n}$ .

Note that it is a conformal dieomorphism.

Definition 1.2.3. For a totally geodesic hyperplane $P\subset H^{n}$
we call also a totally geodesic hyperplane in $U$ . Put
$f^{-1}\Phi_{U}^{-1}(P)$

$f^{-1}\Phi_{U}^{-1}(P)=Q$ . If $P^{1}<P^{2}<\cdots<P^{n-2}<P^{n-1}=P$ is a chain

of totally geodesic subspaces, then there exists a dimensional totally $k$

geodesic subspace of $U$ and similarly a chain $Q^{1}<Q^{2}<\cdots<$

$Q^{k}$

$Q^{n-2}<Q^{n-1}=Q$ and so on.

Since and
$\Phi_{U}(C(\overline{f}U_{\infty}\cap\partial P))=H(\overline{f}U_{\infty}\cap\partial P)$ $\prime H(\overline{f}U_{\infty}\cap\partial P)=$

by the above lemma, we have that

$\mathcal{H}(\overline{f}U_{\infty}\cap P)$ $C(\overline{f}U_{\infty}\cap\partial P)=$

$fC(U_{\infty})\cap\Phi_{U}^{-1}(P)$ . Noting that , it $fC(U_{\infty}\cap\partial Q)=C(\overline{f}U_{\infty}\cap\partial P)$

is easy to check that $C(U_{\infty}\cap\partial Q)=C(U_{\infty})\cap Q$ . An iteration of this

argument yields that
(1.2.4) $C(U_{\infty}\cap\partial Q^{k})=C(U_{\infty})\cap Q^{k}(k=1, \cdots, n-1)$ .

Definition 1.2.5. Let $Q^{1}<Q^{2}<\cdots<Q^{n-2}<Q^{n-1}=Q$ be a

chain of totally geodesic subspaces in a maximal ball $U$ . Suppose that
either one of does not meet . If Int
$\partial U-\partial Q$
, (equivalently $U_{\infty}$ $ C(U_{\infty})\neq\emptyset$

Int $C(U_{\infty})\cap U$ is open in ) and

$U$ $C(U_{\infty})\cap Q^{k}$
contains an open subset
in then by (1.2.4) that
$Q^{k}$
is said to be a dimensional $C(U_{\infty}\cap\partial Q^{k})$ $k$

pleat of the boundary $\partial C(U_{\infty})(k=1,2, \cdots, n-1)$ .

Put . Choosing all possible geodesic hyperplanes $Q$ in

$\Lambda_{U}=\partial C(U_{\infty})$

$U$ and
passing to all chains of geodesic subspaces , we obtain all $\{Q^{k}\}$

pleats in . The set is composed of all possible pleats in dimension

$\Lambda_{U}$ $\Lambda_{U}$

less than or equal to $n-1$ . In the case that Int , there exists $ C(U_{\infty})=\emptyset$

a totally geodesic subspace $Q$ such that $C(U_{\infty})=C(U_{\infty})\cap Q$ is open

in $Q$ . We say that is an $m$ dimensional pleat if $dim$ $Q=m$ .

$C(U_{\infty})$

Inductively we can define pleats of unless $m=1$ . Note that $\partial C(U_{\infty})$

is a one dimensional pleat if and only if

$C(U_{\infty})$ consists of a pair of $U_{\infty}$

points.
Let : $f$ $N\rightarrow S^{n}$
be a conformal immersion as before. Using the
spherical metric of $S^{n}$
, $N$ admits a Riemannian metric such that is a $f$
 Deformation Space 269

local isometry. Recall that is a metric completion and is $\overline{N}$

$\overline{f}:\overline{N}\rightarrow S^{n}$

a map extending . Choose a point in . Let $W(x)$ be the union of

$N$ $f$ $x$

all maximal balls containing . $x$

Lemma 1.2.6. If $\overline{W(x)}$

is the closure of $W(x)$ in $\overline{N}$

then $\overline{W(x)}$

is compact.

Proof.
Let (resp. ) be the distance function of
$\rho$ (resp.
$\rho_{0}$ ). $\overline{N}$
$S^{n}$

Since maps $W(x)$ injectively, is a homeomorphism of $W(x)$ onto its

$f$ $f$

image . Let be an arbitrary sequence of

$\Omega$
$\{p_{i}\}$
. Choose a sequence $\overline{W(x)}$

in $W(x)$ such that $\rho(p_{i}, q_{i})<1/i$ . Since is compact, the sequence

$\{q_{i}\}$
$\overline{\Omega}$

$\{f(q_{i})\}$ has an accumulation point and so $\{f(q_{i})\}$ is Cauchy. Given

$\in>0$ , choose such that $ 0<\delta<\in$ . Suppose that
$\delta$
$\rho_{0}(f(q_{i}), f(q_{j}))<\delta$ .
If is suciently small, then there exists a maximal ball in containing
$\delta$ $\Omega$

the points $f(q_{i}),f(q_{j})$ . And so contains a minimizing geodesic between $\Omega$

$f(q_{i})$ and $f(q_{j})$ . It implies that $\rho_{0}(f(q_{i}), f(q_{j}))=\rho(q_{i}, q_{j})$ . In particular

the sequence is Cauchy. Since is complete, the sequence
$\{q_{i}\}$
has $\overline{N}$
$\{q_{i}\}$

a limit point . And thus we have $\lim p_{i}=q$ . Hence

$q$ is compact. $\overline{W(x)}$

Q.E.D.

Theorem 1.2.7. Let : be a conformal immersion and $f$ $N\rightarrow S^{n}$

the nonempty set of all maximal balls. Then every point of $N$ lies in
$\mathcal{F}$

the convex hull for a unique element . $C(U_{\infty})$ $U\in \mathcal{F}$

Proof.
Choose a point in $N$ and let $W(x)$ be as above. Put $x$

$W(x)_{\infty}=\overline{W(x)}-N$ . Note that it contains more than one point for

otherwise there are no maximal balls containing . Since $W(x)_{\infty}$ is a $x$

closed subset of , is compact by the above lemma. And so

$\overline{W(x)}$ $W(x)_{\infty}$

is a closed subset of
$\overline{f}(W(x)_{\infty})$
. If is the convex $S^{n}$ $\prime H$
$=\mathcal{H}(\overline{f}(W(x)_{\infty}))$

hull in $H^{n+1}$
, then we have a closest point mapping : $D^{n+1}\rightarrow H$ . $\Phi_{H}$

Now there is a unique totally geodesic hyperplane $P$ through $\Phi_{\mathcal{H}}(f(x))$

perpendicular to the geodesic from $f(x)$ to $\Phi_{H}(f(x))$ . Then we have

from (1.2.2) that
$\Phi_{H}(f(x))\in H(\overline{f}(W(x)_{\infty}))\cap P=H(\overline{f}(W(x)_{\infty})\cap\partial P)$ .

Let be a geometric ball containing $f(x)$ such that $\partial B=\partial P.The$ set
$B$

$U=f^{-1}(B)$ is a maximal ball containing because $B\subset f(W(x))$ . As $x$

, it follows that
$\overline{f}\partial U=\partial fU=\partial P$

$H(\overline{f}(W(x)_{\infty})\cap\partial P)=\mathcal{H}(\overline{f}(W(x)_{\infty}\cap\partial U))$

$=\mathcal{H}(\overline{f}(U_{\infty}))$
.
270 Y. Kamishima and S. Tan

Since $\Phi_{U}(fC(U_{\infty}))=\Phi_{U}(C(\overline{f}U_{\infty}))=H(\overline{f}U_{\infty})$ and $\Phi_{?t}|fU=\Phi_{U}$ ,

it follows that $\Phi_{U}(f(x))=\Phi_{H}(f(x))\in\Phi_{U}(f(C(U_{\infty})))$ and hence
$x\in C(U_{\infty})$ . The proof of the uniqueness is the converse of the above
argument. Q. E. D.

Corollary 1.2.8.
(1) The family $\{C(U_{\infty});U\in \mathcal{F}\}$ consists of disjoint subsets.
(2) The $set\cup\Lambda_{U}$
is closed in $N$ .

Proof There exists a unique element of such that each point $\mathcal{F}$

of $N$ lies in its convex hull by the above theorem. This implies that
$\{C(U_{\infty});U\in \mathcal{F}\}$ are disjoint. For (2), it suces to show that if a
sequence converges to $x\in N$ , then there exists an
$\{x_{i}\}\in\Lambda_{U_{i}}(U_{i}\in \mathcal{F})$

element such that

$U\in \mathcal{F}$
. Recall that $\Lambda_{U}=C(U_{\infty})-$ int
$x\in\Lambda_{U}$ . $C(U_{\infty})$

There exists such that $x\in

$U\in \mathcal{F}$ C(U_{\infty})$ . If is not contained in , $x$ $\Lambda_{U}$

then $x\in intC(U_{\infty})$ . It follows that for suciently large , meets $i$
$\Lambda_{U_{i}}$

with $C(U_{\infty})$. This contradicts that are disjoint.

$\{C(U_{\infty});U\in \mathcal{F}\}$

Q.E.D.

Let $M^{n}$ be a closed conformally flat manifold and the set of all $\tilde{\mathcal{F}}$

maximal balls of the universal covering space . It is obvious that , $\tilde{M}$ $\mathcal{F}$

the family and the

$\{C(U_{\infty})\}$are invariant under the fundamen-
$set\cup\Lambda_{U}$

tal group . $\pi$

Corollary 1.2.9. Let $M^{n}$ be a closed conformally flat manifold.

Suppose that is not empty. Then the universal covering space
$\mathcal{F}$ $\tilde{M}$

supports a invariant canonical decomposition

$\pi$
. $\{C(U_{\infty});U\in \mathcal{F}\}$

2. Thurston parametrization of projective structure on sur-

faces
In this chapter we shall prove the Thurston isomorphism stated $II$

in Introduction. Recall that (complex) projective structure on sur-

faces is equivalent to conformally flat structure on surfaces when
we identify $(S^{2}, Conf(S^{2})^{0})$ with $(CP^{1}, PSL_{2}(C))$ . As before, given
a projective structure on a surface we have a developing pair$S$

$(\phi, dev)$ : $(\pi_{1}(S),\tilde{S})\rightarrow(PSL_{2}(C), CP^{1})$ up to conjugation by elements

of $PSL_{2}(C)$ .
2.1.Deformation spaces on surfaces
Suppose that is a closed orientable surface of genus $g\geqq 1$ . For
$S_{g}$

the brevity we set $S=S_{g}$ and . A surface is a hyperbolic $\Gamma=\pi_{1}(S_{g})$ $\Sigma$

 Deformation Space 271

surface if the universal covering space is conformally equivalent to a

hyperbolic plane . Consider the subspace (cf. 4.3.4); $H^{2}$ $\Omega^{+}(S)$

$\Omega^{+}(S)=\{(\phi, dev) : (\Gamma,\tilde{S})\rightarrow(PSL_{2}(C), CP^{1})\}/Diffff^{0}(S)$ ,

where $dev$ are orientation-preserving immersions.

The topology on is given by the following subbasis: $\Omega^{+}(S)$

(1) $N(U)=U/\sim where$ $U$ is an open subset in $Map(\tilde{S}, CP^{1})$ with

the compact-open topology.
(2) $N(K)=\{dev\in Map(\tilde{S}, CP^{1})|dev|K$ is an embedding for a
compact subset } . $K\subset\tilde{S}$
$/\sim$

Put
$CP^{1}(S)^{+}=PSL_{2}(C)\backslash \Omega^{+}(S)$ .

Definition 2.1.1. Let be a hyperbolic surface. The space $S_{g}$

$CP^{1}(S_{g})^{+}$ is called the deformation space of projective structures or

$CP^{1}$
structures on . $I(S_{g})$ is the usual Teichm\"uller space. $S_{g}$

Thurston has introduced the notion of geodesic laminations on sur-

faces. (Cf. $[9],[40],[8].$ ) Namely, a geodesic lamination on a hyper-
bolic surface is a closed subset consisting of a disjoint union of simple
$\Sigma$

geodesies. Let be a geodesic lamination on . By a transversal we

$\Lambda$ $\Sigma$

mean an embedding : such that at each where the

$\ell$
$[0, 1]\rightarrow\Sigma$ $t$
$\ell(t)\in\Lambda$

map is transverse to the leaf through

$\ell$

. $\ell(t)$

A transverse measure on is a function which assigns to each transver- $\Lambda$

$\mu$

sal a Radon measure

$\ell$

on $[0, 1]$ supported by $\{t\in[0,1] |\ell(t)\in\Lambda\}$ $\mu(\ell)$

which is compatible under the canonical homeomorphisms between

nearby transversals. We call the pair a measured geodesic lami- $(\Lambda, \mu)$

nation of . $\Sigma$

Definition 2.1.2. is the space of measured geodesic lam- $\lambda\Lambda \mathcal{L}(\Sigma)$

inations on , equipped with the $weak*topology$ .

$\Sigma$

If is a homeomorphism of a closed hyperbolic surface

$f$ onto $\Sigma$ $\Sigma$

then induces a homeomorphism :

$f$ . (See [8], [5].) $f$ $\mathcal{M}\mathcal{L}(\Sigma)\rightarrow \mathcal{M}\mathcal{L}(\Sigma)$

Let be a hyperbolic surface. Note that

$S_{g}$
is homeomorphic to $\mathcal{T}(S_{g})$

$R^{6g-6}$
. In this case the space of measured laminations is also $\lambda\Lambda \mathcal{L}(S_{g})$

homeomorphic to the real vector space of dimension $6g-6$ . Moreover,

$CP^{1}(S_{g})^{+}$ can be identified with the cotangent bundle of $I(S_{g})$ . (Com-

pare [5].) In contrast to this identification we have a new parametrization

on $CP^{1}(S_{g})^{+}$ .
272 Y. Kamishima and S. Tan

Theorem (Thurston). Let $S_{g}$

be a hyperbolic surface. Then there
exists a homeomorphism
$\Theta$
: $CP^{1}(S_{g})^{+}\rightarrow I(S_{g})\times \mathcal{M}\mathcal{L}(S_{g})$
.

See [16],[17],[15],[13] for the related topics. The rest of this section is
devoted to the proof of this theorem.

2.2. Locally convex pleated maps

Let : $N\rightarrow CP^{1}$ be a conformal immersion. We have shown in
$f$

Theorem 1.2.7 that every point of $N$ lies in the convex hull of $x$ $C(U_{\infty})$

a unique maximal ball $U$ . Let be the set of maximal balls. For each $U$ $\mathcal{F}$

there is a closest point mapping : $f(U)\rightarrow H^{2}(\subset H^{3})$ . (See Section $\Phi_{U}$

1.) Set $\Psi(x)=\Phi_{U}(f(x))$ if $x\in C(U_{\infty})$ . By uniqueness it defines a well

defined map

(2.2.1) $\Psi$
: $N\rightarrow H^{3}$
.

It is obvious that is a continuous map. Note that in our case $\Psi$

each is either a region or a (one dimensional) pleat, the im-

$C(U_{\infty})$

age $\Psi(C(U_{\infty}))$ lies on a geodesic or in a totally geodesic hyperplane of

$H^{3}$
.

Definition 2.2.2. Given an arbitrary conformal immersion : $f$

, we have passed from it to a map

$N\rightarrow CP^{1}$
: $N\rightarrow H^{3}$
. The map $\Psi$ $\Psi$

is called a pleated map.

2.3. Assignment of $CP^{1}(S_{g})^{+}$

to $\mathcal{T}(S_{g})$

In general a pleated map is not locally injective. By definition $\Psi$

(2.2.2), is injective on each $\Psi$

for . On the other hand, we $C(U_{\infty})$ $U\in \mathcal{F}$

consider a point $x\in N$ such that there is a sequence converging to $\{x_{i}\}$

$x$such that $\Psi(x_{i})=\Psi(x)$ . Since each lies in a distinct and $x_{i}$

$C(U_{\infty}^{i})$ $\Psi$

is injective on for suciently large , all have the same

$C(U_{\infty}^{i})$ $i$ $C(U_{\infty}^{i})$

dimension equal to 1. The map fails to be injective on the union of $\Psi$

those . Each ideal set is a locally constant pair of points in

$C(U_{\infty}^{i})$ $U_{\infty}^{i}$

$\overline{N}$

Definition 2.3.1. Denote by $B$

the set of those $C(U_{\infty}^{i})$
on which
$\Psi$
fails to be injective.
is locally injective on $N-B$ . Let $N$ be the space obtained from
$\Psi$

$N-B$ by identifying the boundaries of each component of $N-B$ which

have the same image. Let : $N\rightarrow N$ be the resulting collapse
$\Psi$
$\eta$
 Deformation Space 273

map which is clearly a homotopy equivalence. Since each component of

$N-B$ is isometric to a hyperbolic region with boundary composed of
complete geodesies, the image $N$ supports a complete hyperbolic metric.
Moreover if is an conformal automorphism of $N$ then it leaves $N-B$
$g$

invariant. When stabilizes a component, it acts as isometries with

$g$

respect to a hyperbolic metric of that component. Otherwise translates $g$

one component to another component preserving boundary geodesies.

Therefore the map induces a hyperbolic isometry $\theta(g)\in PSL_{2}(R)$ on
$g$

where we put $N=H^{2}$ . The map satisfies that

$H^{2}$
. Now
$\eta$
$\eta\circ g=\theta(g)\circ\eta$

given a projective structure $(\phi, dev)$ in $CP^{1}(S_{g})^{+}$ , we apply the above

argument to $(\phi, dev)$ : $(\Gamma,\tilde{S})\rightarrow(PSL_{2}(C), CP^{1})$ . Then it induces an
equivariant homotopy equivalence : . The
$(\theta, \eta)$ $(\Gamma,\tilde{S})\rightarrow(PSL_{2}(R), H^{2})$

map induces a homotopy equivalence of onto

$\eta$ . Within the $S$ $H^{2}/\theta(\Gamma)$

homotopy class of , there is a dieomorphism : $\eta$ up to $h$ $S\rightarrow H^{2}/\theta(\Gamma)$

an element of Di(S). Hence a projective structure $(\phi, dev)$ defines a

well defined element $[S, h]$ of $I(S_{g})$ .

2.4. Canonical measure on circular lamination

Let be as before. Recall from (1.3.2) that the subset
$f$
: $N\rightarrow CP^{1}$

is closed and consists of a disjoint union of (one

$\tilde{\Lambda}_{1}=\cup\{\Lambda_{U}|U\in \mathcal{F}\}$

dimensional) pleats. In order to define the canonical measure on $\tilde{\mu}_{1}(\ell)$

a transversal for , it suces to specify the nondecreasing function

$\ell$ $\tilde{\Lambda}_{1}$

whose derivative is equal to

$\varphi(t)=\int_{[0,t]}\tilde{\mu}_{1}(\ell)dt$ . For each $\in[0,1]$ , $\tilde{\mu}_{1}(\ell)$ $t$

let be a unique maximal ball such that

$U^{t}\in \mathcal{F}$
. If $\ell(t)\in C(U_{\infty}^{t})$

, $t\in[0,1]$ are suciently close, the balls

$s$
and must intersect. Let $U^{s}$ $U^{t}$

denote the dihedral angle of intersection of the circles

$\Theta(s, t)$ , , $\partial U^{s}$ $\partial U^{t}$

measured inside one ball and outside the other. The function is $\varphi(t)$

then defined as the infimum of all -sum $\Theta$

$\ominus(0, t_{1})+O-(t_{1}, t_{2})+\cdots+\ominus(t_{n}, t)$

over all subdivisions $0<t_{1}<t_{2}<\cdots<t_{n}<t$ of $[0, t]$ . The following is

the elementary calculation of the trigonometry.
Lemma 2.4.1. If $r<s<t$ are suciently close, then $\ominus(r, s)$ ,
and
$\Theta(s, t)$ are defifined and $\Theta(r, s)+\Theta(s, t)\leqq O-(r, t)$ .
$\Theta(r, t)$

With this lemma a nondecreasing function can be defined as

$\varphi(t)=\lim\sum\Theta(0, t_{1})+\Theta(t_{1}, t_{2})+\cdots+\Theta(t_{n}, t)$

where runs over all subdivisions of

$\sum$
and $\lim$ is taken as the mesh $[0, t]$

of subdivisions goes to zero. Therefore the derivative is $\varphi/=\tilde{\mu}_{1}(\ell)$

274 Y. Kamishima and S. Tan

a measure on a transversal for . Finally the compatiblity of the $\ell$ $\tilde{\Lambda}_{1}$

measure on the various transversals is deduced from the following re-

mark. The leaf through determines the ball and the measure $\ell(t)$ $U^{t}$

on is determined by the angles made by the

$\ell$

. Thus corresponding $\partial U^{t}$

transversals determine the same measure.

Let : be a pleated map for an immersion :
$\Psi$ $N\rightarrow H^{3}$
. $f$
$N\rightarrow CP^{1}$

Let be the set as in (2.3.1). It follows that

$B$
. If we $\eta(B)=\eta(\tilde{\Lambda}_{1})$

put , then is a geodesic lamination on $N(=H^{2})$ . More-

$\eta(\tilde{\Lambda}_{1})=\tilde{\Lambda}_{2}$ $\tilde{\Lambda}_{2}$

over let be a transversal to , then

$\ell$

is also a transversal to $\tilde{\Lambda}_{2}$

$\eta^{-1}(\ell)$

. Set
$\tilde{\Lambda}_{1}$

. We have a measure on . And thus

$\tilde{\mu}_{2}(\ell)=\tilde{\mu}_{1}(\eta^{-1}(\ell))$ $\tilde{\mu}_{2}$
$\tilde{\Lambda}_{2}$

is a measured geodesic lamination on $N$ . As before suppose

$(\tilde{\Lambda}_{2},\tilde{\mu}_{2})$

that $(\phi, dev)$ : $(\Gamma,\tilde{S})\rightarrow(PSL_{2}(C), CP^{1})$ is a developing pair. The

above argument implies that there is a measured geodesic lamination
over . It is easy to see that
$(\tilde{\Lambda}_{2},\tilde{\mu}_{2})$
is invariant un-
$(\theta(\Gamma), H^{2})$ $(\tilde{\Lambda}_{2},\tilde{\mu}_{2})$

der the group . That is, and

$\theta(\Gamma)$
. $\theta(\gamma)(\tilde{\Lambda}_{2})=(\tilde{\Lambda}_{2})$ $\tilde{\mu}_{2}(\theta(\gamma)(\ell))=\tilde{\mu}_{2}(\ell)$

It induces a measured geodesic lamination on . There $(\Lambda_{2}, \mu_{2})$ $H^{2}/\theta(\Gamma)$

is a dieomorphism : as above. Then we have a $h$ $S\rightarrow H^{2}/\theta(\Gamma)$

geodesic measured lamination on such that and $(\Lambda, \mu)$ $S$ $h(\Lambda)=\Lambda_{2}$

$\mu(\ell)=\mu_{2}(h(\ell))$ . Hence it defines an element . $(\Lambda, \mu)\in\lambda 4\mathcal{L}(S)$

2.5. Thurston correspondence

We have a well defined map $\Theta$
: $CP^{1}(S_{g})\rightarrow \mathcal{T}(S_{g})\times \mathcal{M}\mathcal{L}(S_{g})$
,

$\Theta((\phi, dev))=([S, h], (\Lambda, \mu))$ .

It is easy to see that is injective, because given two projective struc- $\Theta$

tures which have the same image in . The coincidence $I(S_{g})\times\Lambda 4\mathcal{L}(S_{g})$

on the first summand implies that each developing map coincides out-
side each . But the second summand measures the dierence on and
$B$ $B$

so two developing maps coincide on the whole . $\tilde{S}$

Let be the subspace of such that every lamination

$\lambda 4\mathcal{L}(S_{g}, S)$ $\mathcal{M}\mathcal{L}(S_{g})$

consists of compact leaves. If is the closure in , $\overline{\lambda\Lambda \mathcal{L}_{h}(M,S^{n-1})}$ $\lambda\Lambda \mathcal{L}(S_{g})$

then it is known that . We show that there is a $\overline{\lambda\Lambda \mathcal{L}_{h}(S_{g},S)}=\lambda\Lambda \mathcal{L}(S_{g})$

map

(2.5.1) $\mathfrak{S}$

: $\mathcal{T}(S_{g})\times\Lambda 4\mathcal{L}(S_{g})\rightarrow CP^{1}(S_{g})$

such that . . $\Theta$ $\mathfrak{S}=id$

To prove this we need some preliminaries.

Definition 2.5.2 (cf. [16]). Let $\alpha>0$ be any mumber and $W_{\alpha}=$

$\{z\in C|0\leqq Imz\leqq\alpha\pi\}$ .

 Deformation Space 275

Let be the stereographic projection which maps

$s$
onto $S^{2}-\{\infty\}$ . $C^{*}$

Then we define a map : to be the exponential map $\exp$ : $\xi$

$C\rightarrow S^{2}$ $C\rightarrow$

followed by .
$C^{*}$ $s$

Put . Both and are conformally flat manifolds with

$\xi(W_{\alpha})=C_{\alpha}$ $W_{\alpha}$ $C_{\alpha}$

boundary. We call an -pile and an -crescent. $W_{\alpha}$ $\alpha$ $C_{\alpha}$ $\alpha$

Let $([S, h], (\Lambda, \mu))$ be a representative element of

Proof of (2.5.1).
. We suppose first that
$\mathcal{T}(S_{g})\times \mathcal{M}\mathcal{L}(S_{g})$
. The map $(\Lambda, \mu)\in\lambda\Lambda \mathcal{L}(S_{g}, S)$

maps a
$(\theta,\tilde{h})$
invariant measured geodesic lamination
$\Gamma$

onto a $(\tilde{\Lambda},\tilde{\mu})$

invariant measured geodesic lamination . The map is a $\tilde{h}$

$(\tilde{\Lambda},\tilde{\mu}\prime)$
$\theta(\Gamma)$

homeomorphism of onto where is viewed as the upper hemi-

$\tilde{S}$
$H^{2}$ $H^{2}$

sphere of . Cut along$S^{2}$

and then insert the crescents
$H^{2}$
and glue $\tilde{\Lambda}$

$C_{\alpha}$

them along the boundary components. Here these angles come from $\alpha$

those of the measure . Similarly cut along and insert the piles $\tilde{\mu}$
$\tilde{S}$ $\tilde{\Lambda}$

$W_{\alpha}$

and then glue along the corresponding boundary components by the map
. The resulting manifold
$\tilde{h}^{-1}o\xi$
is invariant under an action of and $\tilde{S}$
$\Gamma$

thus the orbit space is still homeomorphic to . Since both and are $S$
$\tilde{h}$

$\xi$

projective immersions, combined with these maps, we have a well defined

projective immersion $dev$ : and since the group acts as $\tilde{S}\rightarrow S^{2}=CP^{1}$ $\Gamma$

projective transformations with respect to this structure on , there is a $S$

holonomy homomorphism . If we set $\mathfrak{S}([S, h], (\Lambda, \mu))=(\phi, dev)$ , then $\phi$

the map is well defined on such that on $\mathcal{T}(S_{g})\times \mathcal{M}\mathcal{L}(S_{g}, S)$ $\Theta\cdot \mathfrak{S}=id$

. For an element
$I(S_{g})\times\Lambda 4\mathcal{L}(S_{g}, S)$ there is a sequence $(\Lambda, \mu)\in \mathcal{M}\mathcal{L}(S_{g})$

that converges to
$\{(\Lambda_{i}, \mu_{i})\}\in\Lambda 4\mathcal{L}(S_{g}, S)$ . Let $[S, h]$ be an arbi- $(\Lambda, \mu)$

trary element of $I(S_{g})$ and fix it once. The map maps $\mathfrak{S}$
$([S, h], (\Lambda_{i}, \mu_{i}))$

to a sequence of projective structures . Recalling the topol- $\{(\phi_{i}, dev_{i})\}$

ogy of $CP^{1}(S_{g})$ from (2.1) and by the fact that each coincides with $dev_{i}$

the map outside , the sequence of developing maps

$\tilde{h}$
$\tilde{\Lambda}_{i}$

converges $\{dev_{i}\}$

to a map on each compact set of . And so it is easy to see that it con- $\tilde{S}$

verges to a map $dev$ : which is obviously a projective $\tilde{S}\rightarrow S^{2}=CP^{1}$

immersion. The projective immersion $dev$ determines a holonomy ho-

momorphism up to conjugation. Setting $\mathfrak{S}([S, h], (\Lambda, \mu))=(\phi, dev)$ ,
$\phi$

we obtain a continuous map : such $\mathfrak{S}$

$I(S_{g})\times\Lambda 4\mathcal{L}(S_{g})\rightarrow CP^{1}(S_{g})$

$that\ominus\cdot$
. $\mathfrak{S}=id$
Q.E.D.

2.6. Modular space of projective structures

Recall that $CP^{1}(S)^{+}=PSL_{2}C\backslash \Omega(S)^{+}/Diffff^{0}(S)$ . Then the space
is called the modular space of
$\lambda\Lambda CP^{1}(S)^{+}=PSL_{2}C\backslash \Omega(S)^{+}/Diffff^{+}(S)$

projective structures. On the other hand, the Teichm\"uller space $I(S)$ is

defined alternately to be $R(\Gamma, PSL_{2} R)/PGL_{2}$ R. And so $I(S)$ is identi-
276 Y. Kamishima and S. Tan

fied with the quotient space of sense-preserving discrete faithful represen-

tations, $\mathcal{T}(\Gamma)=R^{+}(\Gamma, PSL_{2} R)/PSL_{2}$ R. There is also a similar iden-
tification . Since each element of $Diffff^{+}(S)/Diffff^{0}(S)$
$\Lambda 4\mathcal{L}(S)=\mathcal{M}\mathcal{L}(\Gamma)$

maps onto itself, there is an action of $Out^{+}(\Gamma)$ on

$\Lambda 4\mathcal{L}(S)$
. $\lambda\Lambda \mathcal{L}(\Gamma)$

Corollary 2.6.1. There is a commutative diagram on which

$Out^{+}(\Gamma)$ acts diagonally.

$\rightarrow$

$Diffff^{+}(S)/Diffff^{0}(S)\downarrow$ $Out^{+}\downarrow(\Gamma)$

$\rightarrow 0-$

$CP^{1}(S)^{+}\downarrow$ $I(\Gamma)\times \mathcal{M}\mathcal{L}(\Gamma)\downarrow$

$\rightarrow\ominus\wedge$

$\mathcal{M}CP^{1}(S)^{+}$ $\mathcal{T}(\Gamma)\times\lambda\Lambda \mathcal{L}(\Gamma)/Out^{+}(\Gamma)$

.

If we recall that $Out^{+}(\Gamma)$ acts properly discontinuously on $I(\Gamma)\times$

, it follows that
$\mathcal{M}\mathcal{L}(\Gamma)$

Corollary 2.6.2. $Diffff^{+}(S)/Diffff^{0}(S)$ acts properly discontinu-

ously on $CP^{1}(S)^{+}$ .

3. Projective structures on surfaces and holonomy function

groups

3.1 Subspaces of $CP^{1}(S_{g})^{+}$

As before is a closed orient $S$

surface $S_{g}$
of genus $g\geqq 2$ and
$\Gamma=\pi_{1}(S_{g})$ . Recall that

$CP^{1}(S)^{+}=PSL_{2}(C)\backslash \Omega^{+}(S)$ ,

where is the deformation space of orientation-preserving devel-

$\Omega^{+}(S)$

oping maps. (See Chapter 3.)

Definition 3.1.1. Let $P$ : $\Omega^{+}(S)\rightarrow CP^{1}(S)^{+}$ be the canonical

projection. Let be the subspace of $CP^{1}(S)^{+}$ consisting of in-
$CP^{1}(S)_{0}^{+}$

jective developing maps. And is the subspace of $CP^{1}(S)^{+}$ con- $CP^{1}(S)_{1}^{+}$

sisting of nonsurjective developing maps. Let $\Omega^{+}(S)_{i}=P^{-1}(CP^{1}(S)_{i}^{+})$

$(i=0,1)$ .

Proposition 3.1.2.
(i) $CP^{1}(S)_{0}^{+}$ is a closed subspace of $CP^{1}(S)^{+}$ .
 Deformation Space 277

(ii) $CP^{1}(S)_{1}^{+}$ is a closed subspace of $CP^{1}(S)^{+}$

Proof. Suppose that a sequence in $CP^{1}(S)_{0}^{+}$ converges to $\{dev_{i}\}$

a developing map $\{dev\}$ .

Suppose that $dev$ is not injective and $dev(x)=dev(y)$ for $x\neq y$ in . $\tilde{S}$

There exists a compact neighborhood $K$ of which does not contain $x$

$y$and is mapped homeomorphically onto a closed ball $dev(K)$ . By the

above topology on $CP^{1}(S)^{+}$ , is also an embedding on $K$ for su- $dev_{i}$

ciently large . Let be the spherical metric of $CP^{1}$ and let

$i$
$\rho$ be the $\ell_{i}$

shortest circular arc from $dev_{i}(x)$ to devi(y). Since $dev|K$ is an embed-

ding for suciently large , there is a sequence of points $\{p_{i}\}\in CP^{1}$ each $i$

of which is the first contact of to $\partial dev_{i}(K)=dev_{i}(\partial K)$ . There exists $\ell_{i}$

a sequence of points such that devi(zi)= . Since gives

$\{z_{i}\}\in\partial K$ $p_{i}$
$\ell_{i}$

the distance ( $dev_{i}(x)$ , devi(y)), it follows that $\rho(dev_{i}(x), dev_{i}(z_{i}))\leqq$

$\rho$
$ $

$\rho(dev_{i}(x), dev_{i}(y))$ . The sequence converges to some point $\{z_{?}.\}\in\partial K$

. Then the above inequality yields that

$z\in\partial K$

$\rho(dev(x), dev(z))\leqq\rho(dev(x), dev(y))=0$ ,

while is an interior point of $K$ , which is a contradiction. This proves

$x$

(i).
Consider (ii). Suppose that a sequence converges to an $\{(\phi_{i}, d_{i})\}$

element $\{(\phi, d)\}$ in $CP^{1}(S)^{+}$ . We can assume that the closure is $\overline{\phi(\Gamma)}$

neither a finite group nor a subgroup of the group of similarity trans-

formations $Sim(R^{2})$ , for otherwise would be covered by a sphere or $S$

a torus respectively. In particular contains a loxodromic element. $\phi(\Gamma)$

If is a loxodromic element for some

$\phi(\gamma)$
, there is a point in $\gamma\in\Gamma$ $x$

$CP^{1}$
such that $\phi(\gamma)x=x$ . (Note that there exist exactly two points.)
Let be the limit set for
$L(\phi(\Gamma))$ . (See 4.1 or [11] for the defini- $\phi(\Gamma)$

tion.) It follows that $x\in L(\phi(\Gamma))$ . The $trace$ formula (cf. [4]) implies
that is either elliptic or parabolic if and only if $|tr^{2}g|\in[0,4]$ for
$g$

$g\in PSL_{2}(C)$ . Since and is loxodromic, it follows

$\phi_{i}(\gamma)\rightarrow\phi(\gamma)$ $\phi(\gamma)$

that is also loxodromic for suciently large . And so there exists

$\phi_{i}(\gamma)$ $i$

a point such that $x_{i}$ for each . Note that $\phi_{i}(\gamma)x_{i}=x_{i}$ $i$
$\{x_{i}\}\in L(\phi_{i}(\Gamma))$

and . (See for example [25].)

$ d_{i}(\tilde{S})\cap L(\phi_{i}(\Gamma))=\emptyset$

The sequence has an accumulation point. Since

$\{x_{i}\}$
and $\phi_{i}(\gamma)\rightarrow\phi(\gamma)$

,
$\phi_{i}(\gamma)x_{i}=x_{i}$ fixes that accumulation point. We can assume that
$\phi(\gamma)$

$\lim x_{i}=x$ .

We claim that misses the point . Suppose not. Let $d(p)=x$ for $d$ $x$

some . Choose a compact neighborhood $C$ of in

$p\in\tilde{S}$
and a closed $p$
$\tilde{S}$

ball centered at in $CP^{1}$

$\overline{B}$

so that : is a dieomorphism. We $x$ $d$ $C\rightarrow\overline{B}$

note that for suciently large , and is an embedding. $i$ $x_{i}\in\overline{B}$

$d_{i}|C$
278 Y. Kamishima and S. Tan

Case 1. lies inside $d_{i}(C)$ . Since

$x$ , it implies that $x_{i}\not\in d_{i}(C)$ $\rho(x_{i}, x)\geqq$

dist $(x_{i}, d_{i}(C))$ and let be the sequence of points which attains
$\{a_{i}\}\in\partial C$

the distance dist , i.e., $\rho(x_{i}, d_{i}(a_{i}))=dist(x_{i}, d_{i}(C))$ . The se-

$(x_{i}, d_{\dot{0}}(C))$

quence has a limit point

$\{a_{?}.\}$
. Since it follows that $d(a)\in\partial B$ ,
$a\in\partial C$

we obtain that $\rho(x, d(a))>0$ . On the other hand, the above inequality
yields that $0\geqq\rho(x, d(a))$ , which is a contradiction.
Case 2. lies outside $d_{i}(C)$ . Since $p\in C$ , it follows dist $(x, d_{i}(C))\leqq$
$x$

$\rho(x, d_{i}(p))$ . Note that $\lim$ dist $(x, d_{i}(C))=0$ because $\lim d_{i}(p)=d(p)=$

$x$. Similarly as above we have a sequence of the points $\{b_{i}\}\in\partial C$

such that $\rho(x, d_{i}(b_{i}))=dist(x, d_{i}(C))$ . As $\lim b_{i}=b$ for some point
, it follows that $\lim d_{i}(b_{i})=d(b)\in\partial B$ . And so $0<\rho(x, d(b))=$
$b\in\partial C$

$\lim$
dist $(x, d_{i}(C))$ , being a contradiction. Therefore misses the point $d$

$x$ .
By virtue of the theorem of [25] we have that $d$
is a covering map.
This shows (ii). Q.E.D.

3.2. Description of Kleinian groups by projective

structures
be a Kleinian group, i.e., a finitely generated discrete subgroup
Let $G$

of $PSL_{2}$ C. Put $\Omega=\Omega(G)=S^{2}-L(G)$ . Recall that

$G$ is a function group if there is a component of invariant under $\Omega_{0}$ $\Omega$

$G$ .
$G$ is a quasi-Fuchsian group if (i.e., consists of two compo-
$\Omega=\Omega_{0}\cup\Omega_{1}$

nents). As the special case if is a round circle then $G$ is a

$\partial\Omega_{0}(=\partial\Omega_{1})$

Fuchsian group.
$G$ is a group if has only one invariant simply connected component.
$\Omega$

Let $ S=H^{2}/\Gamma$ be a closed orientable surface.

Definition 3.2.1.
$\mathcal{F}(\Gamma)=$
{ is a function group
$\theta\in Hom(\Gamma, PSL_{2}C)|\theta(\Gamma)$

$ B(\Gamma)=\{\theta\in Hom(\Gamma, PSL_{2}C)|\theta$ : is an isomorphism, is

$\Gamma\rightarrow\theta(\Gamma)$ $\theta(\Gamma)$

a Kleinian group and an invariant component is simply connected.}

$\mathcal{R}_{2}(\Gamma)=\{\theta\in Hom(\Gamma, PSL_{2}C)|\theta(\Gamma)$ is quasi-conformally equivalent to

(i.e., the set of quasi-Fuchsian groups).

$\Gamma.\}$

Let
$P:Hom(\Gamma, PSL_{2}C)\rightarrow Hom(\Gamma, PSL_{2} C)/PSL_{2}$ $C$

be the canonical projection and put

$F(\Gamma)=P(\mathcal{F}(\Gamma))$ , $B(\Gamma)=P(B(\Gamma))$ , and $T_{2}\Gamma=P(\mathcal{R}_{2}(\Gamma))$ .

Note that $Hom(\Gamma, PSL_{2} C)/PSL_{2}C$ is connected but not a Hausdor

space.
 Deformation Space 279

Recall that $H$ : $\Omega^{+}(S)\rightarrow Hom(\Gamma, PSL_{2}C)$ is the map which

assigns to an oriented projective structure its holonomy representa-
tion. The map $H$ induces a holonomy map $hoi$ : $CP^{1}(S)^{+}$ $\rightarrow$

$Hom(\Gamma, PSL_{2} C)/PSL_{2}$ C.

Corollary 3.2.2. The holonomy map $hoi$ maps $CP^{1}(S)_{1}^{+}\rightarrow$

$F(\Gamma)$ . In particular the holonomy groups are discrete.

Proposition 3.2.3. The holonomy map $hoi$

defifines a homeomor-
phism of $CP^{1}(S)_{0}^{+}$ onto $B(\Gamma)$ .

Let Proof. be a representative element of

$\theta\in B(\Gamma)$ with an $B(\Gamma)$

invariant simply connected component . Since $\theta\in Hom(\Gamma, PSL_{2} C)$ , $\Omega_{0}$

there is an orientation-preserving conformal homeomorphism : $f$ $\Omega_{0}\rightarrow$

such that
$H^{2}$
is Fuchsian. Let :
$f\theta(\Gamma)f^{-1}$
be an $\psi$ $\Gamma\rightarrow f\theta(\Gamma)f^{-1}$

isomorhism defined by $\psi(\gamma)=f\theta(\gamma)f^{-1}$ . Then it is well known that

there is a quasi-conformal homeomorphism : which induces $h$ $H^{2}\rightarrow H^{2}$

. Put $dev=f^{-1}\circ h$ . It is easy to see that

$\psi$
is an element of $[\theta, dev]$

$CP^{1}(S)_{0}^{+}$ .
Let $P$ : be the canonical projection of the defor-
$\Omega_{0}^{+}(S)\rightarrow CP^{1}(S)_{0}^{+}$

mation spaces (cf. (3.1.1)). We will show that $H$ maps onto . $\Omega_{0}^{+}(S)$ $B(\Gamma)$

If $(\phi, dev)$ is an element of , then it follows that . If suf-

$\Omega_{0}^{+}(S)$ $\phi\in \mathcal{F}$

fices to check that has an invariant simply connected component of

$\phi(\Gamma)$

$\Omega=S^{2}-L(\phi(\Gamma))$ . Since $dev$ is injective, has a simply connected do- $\phi(\Gamma)$

main which sits in . Let

$dev(\tilde{S})$
be an invariant maximal component
$\Omega$
$\Omega_{0}$

in containing
$\Omega$

. Since acts properly discontinuously and

$dev(\tilde{S})$ $\phi(\Gamma)$

freely on , we can choose a

$\Omega_{0}$
invariant Riemannian metric on .
$\phi(\Gamma)$ $\Omega_{0}$

The map $dev$ is a covering map because is compact. Since : $S$ $\phi$ $\Gamma\rightarrow\phi(\Gamma)$

is an isomorphism, $dev$ must be an isometry. And thus . $dev(\tilde{S})=\Omega_{0}$

We prove that $H$ is injective. For $(\phi_{i}, dev_{i})(i=1,2)$ , suppose that

$H(\phi_{1}, dev_{1})=H(\phi_{2}, dev_{2})$ , i.e., . Then, $dev_{1}(\tilde{S})=dev_{2}(\tilde{S})$ . $\phi_{1}=\phi_{2}$

$Forantcomponentsthisifnot$ $theni.e.$ $\phi(\Gamma)=\phi_{1}(\Gamma)=dev_{1}(\tilde{S}),dev_{2}(\tilde{S})\phi_{2}(\Gamma)hasatleasttwoinvari(cf.[5]).Hence\phi(\Gamma)isquasi-$

, ,
Fuchsian. However since both $dev_{1}$ and $dev_{2}$ are orientation-preserving,
it is impossible. Put . Then it follows that
$\tilde{f}=dev_{2}^{-1}\circ dev_{1}$ $\tilde{f}\circ\gamma=\gamma\circ\tilde{f}$

for . Therefore $f\in

$\gamma\in\Gamma$ Diffff^{0}(S)$ and $[\phi_{2}, dev_{2}]\circ f=[\phi_{1}, dev_{1}]$ .
And hence $H$ is a one-to-one continuous map. Since $H$ is a local home-
omorphism by the Holonomy theorem 4.3.9 (cf. Chapter 4), it follows
that $H$ is a homeomorphism of onto . Since the action of $\Omega_{0}(S)$ $B(\Gamma)$

$PSL_{2}C$ on both and is free, it implies that $hoi$ is a homeo-

$\Omega_{0}(S)$ $B(\Gamma)$

morphism. Q.E.D.
280 Y. Kamishima and S. Tan

Let be the space of quasi-Fuchsian groups in $Hom(\Gamma, PSL_{2}C)$

$\mathcal{R}_{2}(\Gamma)$

as in (3.2.1). If is the closure of in $Hom(\Gamma, PSL_{2} C)$ , then

$\overline{\mathcal{R}_{2}(\Gamma)}$
$\mathcal{R}_{2}(\Gamma)$

we put .
$\partial \mathcal{R}_{2}(\Gamma)=\overline{\mathcal{R}_{2}(\Gamma)}-\mathcal{R}_{2}(\Gamma)$

Definition 3.2.4. Define the following subspaces

$\Omega^{+}(S, qf)=\{(\phi, dev)\in\Omega^{+}(S)|\phi\in \mathcal{R}_{2}(\Gamma)\}$ ,
$CP^{1}(S, qf)^{+}=P(\Omega^{+}(S, qf))$ ,

and

$\Omega^{+}(S, \partial)=\{(\phi, dev)\in\Omega^{+}(S)|\phi\in\partial \mathcal{R}_{2}(\Gamma)\}$

,
$CP^{1}(S, \partial)^{+}=P(\Omega^{+}(S, \partial))$ .

(resp. $CP^{1}$ ( , ) ) is called the deformation space of

$CP^{1}(S, qf)^{+}$ $S$ $\partial$

(oriented) projective structures with quasi-Fuchsian (resp. boundary)

holonomy.

We have the following subspaces of $CP^{1}(S, qf)^{+}$ (resp. $CP^{1}$

( , ) )
$S$ $\partial$

whose developing maps are injective;

$CP^{1}(S, qf)_{0}^{+}=CP^{1}(S, qf)^{+}\cap CP^{1}(S)_{0}^{+}$ ,
(3.2.5)
$CP^{1}(S, \partial)_{0}^{+}=CP^{1}(S, \partial)^{+}\cap CP^{1}(S)_{0}^{+}$
.

The simultaneous uniformization of Bers ([5]) is stated as follows.

Corollary 3.2.6 (Bers). $CP^{1}(S, qf)_{0}^{+}\approx I(\Gamma)\times\triangle$

.
Here is an open cell contained in
$\triangle$
$\Lambda 4\mathcal{L}(S)$
.

3.3. Insertion of annuli and operation on projective

structures with boundary holonomy
An insertion of annuli (more generally, a graft $ing$ ) produces a new
structure from a given projective structure. (See Goldman [16].) Es-
pecially, let be the space of projective structures with quasi-
$\Omega_{0}^{+}(S, qf)$

Fuchsian holonomy groups and with injective developing maps. Let the $C$

set of all isotopy classes of a disjoint collection of homotopically nontriv-

ial simple closed curves on . Let denote the set of measured $S$ $\Lambda 4\mathcal{L}(2Z)$

geodesic laminations supported on a disjoint union of closed geodesies $\mu$

lying in and together with $C$

times positive integer weights. Then, $ 2\pi$

each defines an operation which assigns to a structure of

$\sigma\in \mathcal{M}\mathcal{L}(2Z)$ $\#$

a structure with surjective developing map.

$\Omega_{0}(S, qf)$

Goldman ([16]) has shown that

 Deformation Space 281

Theorem 3.3.1. $CP^{1}(S, qf)^{+}\approx CP^{1}(S, qf)_{0}^{+}\times \mathcal{M}\mathcal{L}(2Z)$

.

It follows also that

(3.3.2) $\Omega^{+}(S, qf)\approx\Omega_{0}^{+}(S, qf)\times\lambda\Lambda \mathcal{L}(2Z)$

.

If $x\in\Omega_{0}^{+}(S, qf)$ and , then is a new structure with $\sigma\in \mathcal{M}\mathcal{L}(2Z)$ $ x\#\sigma$

surjective developing map and with the same holonomy as that of . It $x$

lies in one component of $\Omega^{+}(S, qf)$ dierent from . And so it $\Omega_{0}(S, qf)$

follows that for which

$\Omega^{+}(S, qf)=\bigcup_{\sigma\in \mathcal{M}\mathcal{L}(2Z)}(\Omega_{0}^{+}(S, qf)\#\sigma)$
$\Omega_{0}^{+}(S, qf)\#\sigma$

is one component homomorphic to $\Omega_{0}^{+}(S, qf)$

.

Let be the space of oriented projective structures with

$\Omega_{0}^{+}(S, \partial)$

boundary holonomy and with injective developing maps (cf. (3.2.5)).

The operation can be defined on . We shall prove the similar
$\#$ $\Omega_{0}^{+}(S, \partial)$

result for . $\Omega^{+}(S, \partial)$

Proposition 3.3.3.
$\Omega_{0}^{+}(S, \partial)\times\lambda\Lambda \mathcal{L}(2Z)\approx\Omega^{+}(S, \partial)$
.

In order to prove this proposition, we need the following lemmata.

Lemma 3.3.4. The holonomy map
$H$ : $\overline{\Omega^{+}(S,qf)}\rightarrow\overline{\mathcal{R}_{2}(\Gamma)}$
is locally injective.

Proof. If we note that is discrete in

$\lambda\Lambda \mathcal{L}(2Z)$ $\mathcal{M}\mathcal{L}(S)$
, then
$\overline{\Omega\dagger(S,qf)}\approx\overline{\Omega_{0}^{+}(S,qf)}\times \mathcal{M}\mathcal{L}(2Z)$
. We prove that $H$ : $\overline{\Omega_{0}^{+}(S,qf)}\rightarrow$

is injective. Let , $y\in\partial\Omega_{0}^{+}(S, qf)(=\Omega_{0}^{+}(S, qf)(S)-\Omega_{0}^{+}(S, qf))$

$\overline{\mathcal{R}_{2}(\Gamma)}$
$x$

and suppose $H(x)$ $=H(y)$ . First note that $H(x)$ $=$ $\in\partial \mathcal{R}_{2}(\Gamma)$

since $H$ is a local homeomorphism and by the defini-

$\overline{\mathcal{R}_{2}(\Gamma)}-\mathcal{R}_{2}(\Gamma)$

tion (3.2.4). There are neighborhoods $U,V$ of $x,y$ respectively such that
$H$ : , $H$ : are homeomorphisms where $W$ is a neighbor-
$U\rightarrow W$ $V\rightarrow W$

hood of $H(x)$ . Since is open, there are $a\in U$ , $b\in V$ so

$ W\cap \mathcal{R}_{2}(\Gamma)\neq\emptyset$

that $H(a)=H(b)$ in . Since $H|\Omega_{0}^{+}(S, qf)$ is a homeomorphim,

$W\cap \mathcal{R}_{2}(\Gamma)$

it follows that $a=b$ , , . It implies that $x=y$ . $i.e.$ $ U\cap V\neq\emptyset$

We note that nearby structures outside do not have quasi- $\overline{\Omega_{0}(S,qf)}$

Fuchsian holonomy groups. Namely, for , there is a $x\in\partial\Omega_{0}^{+}(S, qf)$

neighborfood $U$ of such that for any $x$ , . $z\in U-\overline{\Omega_{0}^{+}(S,qf)}$ $H(z)\not\in \mathcal{R}_{2}(\Gamma)$

Q.E.D.
282 Y. Kamishima and S. Tan

Lemma 3.3.5. $\partial\Omega^{+}(S, qf)=\Omega^{+}(S, \partial)$ .

Proof. Using the above lemma, we have for $x\in\partial(\Omega^{+}(S, qf))$ that
. By the definition 3.2.4 it follows that
$H(x)\in\partial \mathcal{R}_{2}(\Gamma)$ $\partial(\Omega^{+}(S, qf))\subset$

. Let
$\Omega^{+}(S, \partial)$
so that
$x\in\Omega^{+}(S, \partial)$ . Let $U$ be any$H(x)\in\partial \mathcal{R}_{2}(\Gamma)$

neighborhood of in . Since $H$ is a local homeomorphism, there

$x$ $\Omega^{+}(S)$

exists a neighborhood $W$ of contained in $U$ such that $H$ : $W\rightarrow H(W)$
$x$

is a homeomorphism. As , we have .
$H(x)\in\partial \mathcal{R}_{2}(\Gamma)$ $ H(U)\cap \mathcal{R}_{2}(\Gamma)\neq\emptyset$

Choose $y\in U$ with . Again by the definition 3.2.4, it

$H(y)\in \mathcal{R}_{2}(\Gamma)$

follows that $y\in\Omega^{+}(S, qf)$ , or $ U\cap\Omega^{+}(S, qf)\neq\emptyset$ . And hence $ x\in$

$\Omega^{+}(S, qf)$ . Since $H(x)$ is not a quasi-Fuchsian group, $x\in\partial(\Omega^{+}(S, qf))$ .

Q.E.D.

Proof of (3.3.3). Since $\overline{\Omega+(S,qf)}\approx\overline{\Omega_{0}^{+}(S,qf)}\times \mathcal{M}\mathcal{L}(2Z)$

, it implies
that

(3.3.6) $\partial\Omega^{+}(S, qf)\approx\partial\Omega_{0}^{+}(S, qf)\times \mathcal{M}\mathcal{L}(2Z)$

.

On the other hand, is a closed subspace of by Proposi-

$\Omega_{0}^{+}(S)$ $\Omega^{+}(S)$

tion 3.1.2. It is noted that . And so we have that $\overline{\Omega_{0}^{+}(S,qf)}\subset\Omega_{0}^{+}(S)$

. In view of (3.3.6), the subspace of

$\partial\Omega_{0}^{+}(S, qf))\subset\Omega_{0}^{+}(S)$ $\partial\Omega^{+}(S, qf)$

consisting of injective developing maps, . $(\partial\Omega^{+}(S, qf))_{0}=\partial\Omega_{0}^{+}(S, qf)$

By Lemma 3.3.5 it follows that and by (3.3.6) $\Omega_{0}^{+}(S, \partial)=(\partial\Omega^{+}(S, qf))_{0}$

that . Q.E.D.
$\Omega^{+}(S, \partial)\approx\Omega_{0}^{+}(S, \partial)\times\Lambda 4\mathcal{L}(2Z)$

4. $(G,X)$ -structures

4.1. Limit sets in $(G, X)$

Recall that a geometric structure on a smooth -manifold is a max- $n$

imal collection of charts modelled on a simply connected dimensional $n$

homogeneous space $X$ of a Lie group $G$ whose coordinate changes are re-
strictions of transformations from $G$ . We call such a structure a $(G, X)-$
structure. A manifold with this structure is called a $(G, X)$ -manifold.
Suppose that a smooth connected -manifold $M$ admits a $(G, X)-$ $n$

structure. Then there exists a developing pair $(\rho, dev)$ , where $dev$ :
is a structure-preserving immersion and
$\tilde{M}\rightarrow X$
: $\pi_{1}(M)\rightarrow G$ $\rho$

is a homomorphism (both unique up to elements of ). The group $G$

$\Gamma=\rho(\pi_{1}(M))$ is called the holonomy group for $M$ .

In particular the developing pair $(\rho, dev)$ is an invariant of the

$(G, X)$ -structure. In fact this developing map and holonomy give us
a powerful tool in understanding the topology of $(G, X)$ -manifolds. The
first question arises when the developing map is a covering map onto its
 Deformation Space 283

image. In order to study this problem we introduce the notion of limit

sets in $(G, X)$ due to Kulkarni [33].
We consider the following sets. Let be a subgroup of $G$ . $\Gamma$

(4.1.1)
$\Lambda_{0}=the$ closure of the set { $x\in X|$ the stabilizer $\Gamma_{x}$
is an infinite
subgroup}.
$\Lambda_{1}=the$ set of cluster points of $\{\gamma y|\gamma\in\Gamma\}$
where $y\in X-\Lambda_{0}$ .
$\Lambda_{2}=the$ set of cluster points of $\{\gamma K|\gamma\in\Gamma\}$
where $K$ is a compact
subset of . $X-\{\Lambda_{0}, \Lambda_{1}\}$

Then the set is said to be the limit set of

$\Lambda=\Lambda(\Gamma)=\Lambda_{0}\cup\Lambda_{1}\cup\Lambda_{2}$

. And the set

$\Gamma$
$\Omega=X-\Lambda$ is called the domain of discontinuity for . $\Gamma$

(Compare also [39] for further results of limit sets.) It is the fundamental
result that
Proposition 4.1.2. If , then $\Omega\neq\emptyset$ $\Gamma$

acts properly discontinuously

on . In particular is discrete in $G$ .
$\Omega$ $\Gamma$

(4.1.3) We examine another limit sets to our use. Let $Y$ be a complete
simply connected Riemannian manifold of nonpositive sectional curva-
ture. Then there is a compactification of $Y$ . The space , $\overline{Y}=\partial Y\cup Y$ $\overline{Y}$

equipped with the cone topology, is homeomorphic to the closed ball and
the boundary is the set of points at infinity consisting of the equiv-
$\partial Y$

alence classes of asymptotic geodesies. The group of isometries Iso(F)

extends to a topological action on its boundary. For example, recall
that the -sphere is viewed as the ideal boundary of the real hyper-
$n$ $S^{n}$

bolic space $H^{n+1}$ . Similarly $S^{2n+1}$ is the ideal boundary of the complex
hyperbolic space . Moreover, when $Y$ is a hyperbolic space
$H_{C}^{n+1}$ $H^{n}$

or , Iso(Y) acts as conformal (resp. $CR$ ) automorphisms of the

$H_{C}^{n+1}$

sphere. We write Iso(F) $=Conf(S^{n})$ or $Aut_{CR}(S^{2n+1})$ respectively.

Definition 4.1.4. For a subgroup of Iso(Y) the limit set $\Gamma$
$ L(\Gamma)\subset$

$\partial Y$
is defined to be the set of cluster points of the orbit . for $\Gamma$
$x$ $x\in Y$ .

As to the relation between the above limit set $\Lambda$

, we have (cf. [25])

Proposition 4.1.5. Let be a discrete subgroup $\Gamma$

of either
$Conf(S^{n})$ or $Aut_{CR}(S^{2n+1})$ . Then it follows that

$\Lambda(\Gamma)=L(\Gamma)$ .

4.2. Application to developing maps

Suppose that $M$ is a closed connected $(G, X)$ -manifold. Let us be
given a invariant closed subset $F$ in $X$ . Suppose that there exist a
$\Gamma$
284 Y. Kamishima and S. Tan

component $Y$ of the complement $X-F$ and a component $N$ of $dev^{-1}(Y)$ .

We then have the restriction of the developing map $dev$ : $N\rightarrow Y$ . We
have proved the following result in [18] (cf. also in [16], [18]).

Lemma 4.2.1. Under the above hypothesis, suppose that $Y$ admits
a $\Gamma$

invariant complete Riemannian metric. Then the developing map

$dev$ : $N\rightarrow Y$ is a covering map.

As an application, we shall prove that;

Proposition 4.2.2. Let $Y$ be a invariant closed subset of $X$ with $\Gamma$

HausdorJJ dimension $k<n-1$ . Suppose that the complement $X-Y$

admits a invariant complete Riemannian metric.
$\Gamma$

(i) if $k<n-2$ , then $dev$ : is a homeomorphism.

$\tilde{M}\rightarrow X-\Lambda$

(ii) for $n-2\leqq k<n-1$ , assume that either $dev^{-1}(Y)=\emptyset$ or

: $\pi_{1}(\tilde{M}-dev^{-1}(Y))\rightarrow\pi_{1}(X-Y)$ is surjective. Then
$dev_{*}$

$dev:\tilde{M}\rightarrow X-Y$
is a covering map, or is $dev:\tilde{M}\rightarrow X-\Lambda$
$a$

homeomorphism.

Proof. Note first that $\tilde{M}-dev^{-1}(Y)$ is connected since the Haus-

,
$X-Yislconnecteddorffffdimensionkisless$ $Wehavethann-fromLemma4.2.lthatl(cf.[l8]).Moreoverif$ $devk<:\tilde{M}-n-2$

$dev^{-1}(Y)\rightarrow X-Y$ is a covering map. As above if $k<n-2$ ,

$dev$ : $\tilde{M}-dev^{-1}(Y)\rightarrow X-Y$ is a homeomorphism. If $n-2\leqq k<n-1$

then according to that $dev^{-1}(Y)=\emptyset$ or $dev^{-1}(Y)\neq\emptyset$ under the sur-

jectivity assumption it follows that $dev$ : $\tilde{M}\rightarrow X-Y$ is a covering
map or $dev$ : $\tilde{M}-dev^{-1}(Y)\rightarrow X-Y$ is a homeomorphism. Since
$dev$ is an immersion and $k<n-1$ , for any point in there exists $x$
$\tilde{M}$

a neighborhood $U$ of in such that $dev(U)\cap(X-Y)\neq\emptyset$ . This

$x$
$\tilde{M}$

implies that $dev$ : is injective. Hence

$\tilde{M}\rightarrow dev(\tilde{M})$
acts properly $\Gamma$

discontinuously on which shows that

$dev(\tilde{M})$
. Since $dev(\tilde{M})\cap\Lambda=\emptyset$ $\Gamma$

acts properly discontinuously on $ X-\Lambda$ by Proposition 4.1.2, it follows

that $dev(\tilde{M})=X-\Lambda$ . Q.E.D.

4.3. Deformation space of $(G, X)$ -structures and the

Holonomy theorem
In this section we shall examine the structure of the deformation
space of $(G, X)$ -structures invariant under Lie groups. Let $H$ be a con-
nected Lie group acting on a smooth closed $(2n+1)$ -manifold $M$ .

4.3.1. The deformation space $I(H, M)$ is a space of $H$ invariant

marked $(G, X)$ -structures on manifolds homeomorphic to M. $\mathcal{T}(H, M)$
 Deformation Space 285

consists of equivalence classes of equivariant dieomorphisms : $f$ $ M\rightarrow$

$M$
from the action $(H, M)$ to $H$ invariant $(G, X)$ -manifolds $M$ . Two
such dieomorphisms : $(i = 1, 2)$ are equivalent if
$f_{\dot{0}}$ $ M\rightarrow$ $M_{i}$

there is an equivariant isomorphism ( , $a(G, X)$ -structure preserv- $i.e.$

ing dieomorphism) $h:M_{1}\rightarrow M_{2}$ such that is isotopic to . $h\circ f_{1}$ $f_{2}$

$\rightarrow f1\simeq$

$ M_{1}\downarrow$
$f_{2}^{\searrow}M$

$h$

$M_{2}$

Note that it is not necessarily assumed to be equivariantly isotopic. On

the other hand if $M$ is a $(G, X)$ -manifold then there is a developing
pair $(\rho, dev)$ : $(Aut(\tilde{M}),\tilde{M})\rightarrow(G, X)$ such that , where $\pi\subset Aut(\tilde{M})$

$\pi=\pi_{1}(M)$ and Aut is a group of $(G, X)$ isomorphism of .

$(\tilde{M})$
$\tilde{M}$

4.3.2. is the space consisting of all possible develop-

$\hat{\Omega}(H, M)$

ing pairs $(\rho, dev)$ which satisfy that $(\rho, dev)$ represents an $H$ invariant
$(G, X)$ -structure on $M$ and such that if one forgets the structure then
the action $(H, M)$ is smoothly equivalent to the original action $(H, M)$ .
That is, the action of each element of is topologically unique $\hat{\Omega}(H, M)$

but geometrically distinct.

The topology on is given by the following subbasis (cf. [8]).
$\hat{\Omega}(H, M)$

(1) $N(U)=\{U\}$ where $U$ is an open subset of $Maps(\tilde{M}, X)$ in the

compact open topology of $Maps(\tilde{M}, X)$ .
(2) $N(K)=\{dev\in\hat{\Omega}(H, M)$ $dev|K$ is an embedding for $|$

$a$

compact subset }. $K\subset\tilde{M}$

4.3.3. We introduce a subgroup Di $(H, M)$ of . Let $Diffff(\tilde{M})$

Di(H, $M$ ) be the group of equivariant dieomorphisms of $M$ onto itselft.

Denote by $Diffff\ovalbox{\tt\small REJECT}$

isotopic to the identity map. Consider the following exact sequences of

the dieomorphism groups, where (resp. ) is the $N_{Difff(\overline{M})}(\pi)$ $C_{Difff(\overline{M})}(\pi)$

normalizer (resp. centralizer) of $\pi$

in $Diffff(\tilde{M})j$

1 $\rightarrow\pi\rightarrow N_{Difff(\overline{M})}(\pi)\rightarrow\eta$
Di(M) $\rightarrow 1$

$\uparrow$ $\uparrow$

$C_{Difff(\overline{M})}(\pi)\rightarrow Diffff^{0}(M)$
286 Y. Kamishima and S. Tan

Put Di $(H, _{M})=\eta^{-1}$ (Di(H, $M$ )) and let $\overline{Diffff}^{0}(H, M)$

be the identity
0
component. It follows easily that 77(Di $(H, M)$ ) $=Diffff^{0}(H, M)$ and
0
Di $(H, M)\subset C_{Difff(\overline{M})}(\pi)$ .

4.3.4. The actions on (H, M). The natural right action of

$\hat{\Omega}$
$ $

Di $(H, M)$ and the left action of $G$ on are defifined by setting $\hat{\Omega}(H, M)$

$(\rho, dev)\circ\tilde{f}=(\rho\circ\mu(\tilde{f}), dev\circ\tilde{f})$

,
go $(\rho, dev)=$ ( $go\rho og^{-1}$ , godev), $ $

where : is an isomorphism defined by

$\mu(\tilde{f})$
$\pi\rightarrow\pi$ . $\mu(\tilde{f})(\gamma)=\tilde{f}\circ\gamma\circ\tilde{f}^{-1}$

Obviously both actions commute.

It is noted that two developing pairs $(\rho_{i}, dev_{i})(i=1,2)$ represent
the same structure on $M$ if and only if there exists an element $g\in G$
such that $g\circ dev$ ) $=dev_{2}$ . Put

$\Omega(H, M)=\hat{\Omega}(H, M)/\overline{Diffff}^{0}(H, M)$

.
The action of $G$ induces an action of $\Omega(H, M)$ . Then it is easy to show
that
Lemma 4.3.5. The elements of $I(H, M)$ are in one-to-one cor-
respondence with the orbits of $G\backslash \Omega(H, M)$ .

Definition 4.3.6. The space equipped with the quo- $G\backslash \Omega(H, M)$

tient topology is called the deformation space $I(H, M)$ of $H$ invariant
$(G, X)$ structure on $M$ .

Note that if one choose the trivial group as $H$ then $\mathcal{T}(M)=$
is the usual deformation space. If
$\mathcal{T}(\{1\}, M)$ : $M\rightarrow M$ is a $f$

representative element of then there is a developing pair

$\mathcal{T}(H, M)$

$(\rho, dev)$ : $(\pi_{1}(M),\tilde{M})\rightarrow(G, X)$ . We have the holonomy representation

0
: up to conjugate by an element of $G$ . Let $\hat{H}\subset Diffff(H, M)$
$\rho\circ f\mathfrak{g}$
$\pi\rightarrow G$

be a closed connected subgroup such that $\eta(\hat{H})=H$ . Note that the

group centralizes and is equivariant (cf. (4.3.1)). The group
$\hat{H}$
$\pi$ $f$

centralizes
$\rho(\hat{H})$
. Here we assume that $\rho\circ f_{\beta}(\pi)$

$(4.3.7 *)$ .there exist a group $K\subset G$ and an isomorphism : $\phi$

$\hat{H}\rightarrow K$
for which every representation satisfifies that $ g\circ\rho\circ g^{-1}=\phi$ $\rho$

for some $g\in G$ .

 Deformation Space 287

It is noted that lies in the centralizer $C_{G}(K)$ up to conjugation.

$\rho\circ f\mathfrak{g}(\pi)$

Let be the subset of

$\hat{\Omega}_{0}(H, M)$
whose holonomy representations $\hat{\Omega}(H, M)$

0
lie in $C_{G}(K)$ and
$\rho$ . Put $\Omega_{0}(H, M)=\hat{\Omega}_{0}(H, M)/Diffff$ $(H, M)$ .
$\rho|\hat{H}=\phi$

The projection $\Omega_{0}(H, M)\rightarrow I(H, M)$ is surjective by $(4.3.7*)$ . More-

over we assume that

$(4.3.7 **)$ . If two

such representations of $\pi$
are conjugate in $G$

then they are conjugate by an element of $C_{G}(K)$ .

We then obtain a map $hoi$ : $\mathcal{T}(H, M)\rightarrow Hom(\pi, C_{G}(K))/C_{G}(K)$ which

assigns to a marked structure its holonomy representation. By the defi-
nition $hoi$ lifts to a map $\overline{ho}1:\Omega_{0}(H, M)\rightarrow Hom(\pi, C_{G}(K))$ which makes
the following diagram commute.

$\overline{ho1}$

$\Omega_{0}(H, M)\rightarrow$ $Hom(\pi, C_{G}(K))$

(4.3.8) $\downarrow$ $\downarrow$

$\rightarrow ho1Hom(\pi, C_{G}(K))/C_{G}(K)$

$I(H, M)$ .

If $H=\{1\}$ then it implies that $K=\{1\}$ and so $C_{G}(K)=G$ . We

have the usual holonomy map $hoi$ : $I(M)\rightarrow Hom(\pi, G)/G$ . It has been
proved by Lok ([37]) (see also [24],[48]) that : $\Omega(M)\rightarrow Hom(\pi, G)$ $\overline{ho}1$

is a local homeomorphism. We prove also that

Holonomy Theorem 4.3.9. $\overline{ho}1$

: $\Omega_{0}(H, M)\rightarrow Hom(\pi, C_{G}(K))$

is a local homeomorphism.

Proof. If we prove that the canonical map $\Omega_{0}(H, M)$ $=$

is injective, then
$\hat{\Omega}_{0}(H, M)/\overline{Diffff}^{0}(H, M)\rightarrow\Omega(M)=\hat{\Omega}(M)/\overline{Diffff}^{0}(M)$

the holonomy map : $\Omega(M)\rightarrow Hom(\pi, G)$ restricts to a holonomy

$\overline{ho}1$

map : $\Omega_{0}(H, M)\rightarrow Hom(\pi, C_{G}(K))$ . And so it is a local homeo-

$\overline{ho}1$

morphism. Now suppose that two elements $(\rho, dev)$ and rep- $(\rho, dev^{J})$

0
resent the same element in $\hat{\Omega}(M)/Diffff(M)$ . There exists an element
0
$\tilde{f}\in Diffff(M)$ such that by $(4.3.7*)$ , it $dev^{J}=dev\circ\tilde{f}$
Since
. $\rho|\hat{H}=\rho|\hat{H}$

follows that $dev=dev\circ(h\tilde{f}h^{-1}\tilde{f}^{-1})$ . As is connected for each $h\in\hat{H}$ $\hat{H}$

and the map $dev$ is a local homeomorphism, this equality implies that
0
for every $h\in\hat{H}$ . It follows that $\tilde{f}\in Diffff(H, M)$ by the
$\tilde{f}\circ h=h\circ\tilde{f}$

definition 4.3.3. Hence the canonical map is injective. Q.E.D.

288 Y. Kamishima and S. Tan

Remark 4.3.10. Two assumptions of (4.3.7) will be satisfied when

we consider semifree circle actions of $H$ over $H$ invariant spherical $CR$
structures and $H$ invariant conformally flat structures. We shall see this
in the next chapter.

5. $S^{1}$
invariant geometric structures

5.1. Description of deformation spaces $\mathcal{T}(S^{1}, M)$

In this section we examine deformation spaces of invariant spher- $S^{1}$

ical $CR$ structures and invariant conformally flat structures. Namely,

$S^{1}$

5.1.1. Let $H=S^{1}$ .

(1) $(G, X)$ $=$ (PU( $n$ The corresponding space

$+1,1$ ), $S^{2n+1}$
).
$I(H, M)=C\mathcal{R}(S^{1}, M)$ is the
deformation space of invariant $S^{1}$

spherical $CR$ structures on $M$ by the defifinition 4.3.1.

(2) $(G, X)=(PO(n+1,1),$ . As before, the corresponding space
$S^{n})$

$I(H, M)=C\mathcal{O}(S^{1}, M)$ is the

deformation space of invariant $S^{1}$

conformally flat structures on $M$ .

5.1.2. Let $M$ be a closed $(2n+1)$ -manifold. We suppose that the

action $(S^{1}, M)$ has the following properties for the $CR$ case.
(i) $M$ has a fixed point.
(ii) The orbit space $M^{*}$
is a complex Kleinian orbifold $D^{2n}-$

$L(\pi^{*})/\pi^{*}$ .

Recall that the complex hyperbolic group PU $(n, 1)$ acts on $D^{2n}$
by biholomorphic transformations of and $CR$ transformations of $H_{C}^{n}$

$S^{2n-1}$
. The group $\pi^{*}\subset PU(n, 1)$ and recall that is the limit set $L(\pi^{*})$

of in $S^{2n-1}$
$\pi^{*}$
. By (i) the fixed point set $F$ is homeomorphic to the ideal
boundary $S^{2n-1}-L(\pi^{*})/\pi^{*}$ . For the conformal case, the action $(S^{1}, M)$
on a closed -manifold $M$ has the same property as (i), but instead of
$n$

(ii) we suppose
(ii) the orbit space $M^{*}$
is a Kleinian orbifold $D^{n-1}-L(\pi^{*})/\pi^{*}$ .

Recall from (4.3.2) that every element of $\Omega(S^{1}, M)$ represents an

$H$ invariant (PU( $n+1$ , 1), $S^{2n+1}$ )-structure on $M$ and the $CR$ action

$(H, M)$ is topologically equivalent to the action $(S^{1}, M)$ of (5.1.2). Since
$M$ has a fixed point, it is noted that a lift of $H$ to is isomorphic $\hat{H}$ $\tilde{M}$

to $\hat{H}=H$
(cf. (4.3.6)).
We have shown the topological rigidity of developing maps (cf. [18] [27]).
 Deformation Space 289

Proposition 5.1.3. Let $M$ be a closed spherical $CR$ (resp. confor-

mally flat) manifold with $CR$ (resp. conformat) circle actions. Suppose
that the action $(S^{1}, M)$ has the property of (5.1.2). Put $\pi=\pi_{1}(M)$ . If
$(\rho, dev)$ is the developing pair
of an invariant $(G, X)$ structure on $M$ $S^{1}$

where $(G, X)=(PU(n+1,1),$ $S^{2n+1})$ , $(PO(n+1,1),$ respectively, $S^{n})$

then the developing map $dev$ maps homeomorphically onto the following
subset of $X$ up to an element of $G\mathfrak{j}$

(1) $(\rho, dev)$ : $(S^{1}, \pi,\tilde{M})\rightarrow(U(1), U(n, 1), S^{2n+1}-L(\rho(\pi)))$ .

(2) $(\rho, dev)$ : $(S^{1}, \pi,\tilde{M})\rightarrow(SO(2),$
SO $(n-1, 1)^{0}\times SO(2)$ , $S^{n}-$
$L(\rho(\pi)))$ .

Fere : $\pi\rightarrow\rho(\pi)\subset U(n, 1)$ (resp. $PO$ (

$\rho$ $n$ $-1,1$ ) $\times SO(2)$ ) is an
isomorphism and $L(\rho(\pi))$ is the limit set of $\rho(\pi)$
lying in $S^{2n-1}$
(resp.
$S^{n-2})$ .

Since $M$ has a fixed point by the condition (i) of (5.1.2),

Proof.
we have a lift of action such that has a fixed point (cf. [6]).
$(S^{1},\tilde{M})$
$\tilde{M}$

Then follows from Proposition 3

it and Note 2 of [27] that

$dev$ : $\tilde{M}\rightarrow S^{2n+1}-L(\rho(\pi))$

is homeomorphic

and

$\rho$ : $(S^{1}, \pi)\rightarrow(\rho(S^{1}), \rho(\pi))\subset(U(n-m), P(U(m+1,1) \times U(n-m)))$ .

is an isomorphism for some $m\leqq n-1$ . Moreover the limit set

$L(\rho(\pi))$ $S^{2m+1}$
and $S^{2m+1}-L(\rho(\pi))$ is the fixed point set of
$\subset$

. In particular we have that $M\approx S^{2n+1}-L(\rho(\pi))/\rho(\pi)$ and

$\rho(S^{1})$

$Fix(S^{1}, M)=S^{2m+1}-L(\rho(\pi))/\rho(\pi)$ . On the other hand the $CR$ action

$(S^{1}, M)$ is topologically equivalent to the action of (5.1.2) which implies

that $Fix(S^{1}, M)\approx S^{2n-1}-L(\rho(\pi^{*}))/\pi^{*}$ . Hence $m=n-1$ . It follows

that $\rho(S^{1})=U(1)$ and $P(U(n, 1)$ $\times U(1))=U(n, 1)$ . The fixed point
set of $U(1)$ is $S^{2n-1}-L(\rho(\pi))$ in this case. The similar result holds for
the conformal case when we note the results of [19], [26]. Q.E.D.

We shall check that the conditions of (4.3.7) are satisfied for

and $CO(S^{1}, M)$ .
$C\mathcal{R}(S^{1}, M)$

Remark 5.1.4.
(1) Let $(\rho, dev)$ be a spherical $CR$ structure on $M$ . Each $(g\circ\rho\circ$

, godev) for $g\in G(=PU(n, 1))$ represents the same structure as

$g^{-1}$

$(\rho, dev)$ by the definition. The structure on $M$ does not depend on the

choice of geometric $(2n-1)$ -sphere $S^{2n-1}$ such that $L(\rho(\pi))\subset S^{2n-1}$ by

290 Y. Kamishima and S. Tan

Proposition 5.1.3. When we choose $K=U(1)$ as $K$ of $(4.3.7*)$ , for every

representation there exists $g\in G$ such that $\rho$: is an $g\circ\rho\circ g^{-1}$ $S^{1}\rightarrow U(1)$

isomorphism. And so the condition $(4.3.7*)$ is satisfied for ( , M-), $C\mathcal{R}$ $S^{1}$

similarly for if we choose $K=SO(2)$ . Then it is easy to see

$C\mathcal{O}(S^{1}, M)$

that the centralizer

$U(n, 1)$for $G=PU(n+1,1)$
$C_{G}(K)=\{$
SO $(n-1, 1)^{0}\times SO(2)$ for $G=PO(n+1,1)$ .

Recall that is the subspace of $\Omega^{CR}(S^{1}, M)$ whose holon-

$\Omega_{0}^{CR}(S^{1}, M)$

omy represntations belong to $U(n, 1)$ (cf. (4.3.7)). It is easy to see that
two such pairs $(\rho, dev)$ , $(\rho, dev)$ represent the same structure if and
only if there is an element $h\in U(n, 1)$ such that $dev=h\circ dev$ and
$ho\rho oh^{-1}=\rho$ . The condition $(4.3.7**)$ is satisfied by this fact. As in

(4.3.7), is surjective. We have the commuta-

$\Omega_{0}(S^{1}, M)\rightarrow C\mathcal{R}(S^{1}, M)$

tive diagram from (4.3.9)

$\Omega_{0}^{CR}(S^{1}, M)\rightarrow\overline{ho1}$
$Hom(\pi, U(n, 1))$

(5.1.5) $\downarrow$ $\downarrow$

$ho1$
$C\mathcal{R}(S^{1}, M)$ $\rightarrow R(\pi, U(n, 1))/U(n, 1)$ ,

similarly for $CO(S^{1}, M)$ .

(2) If is the orbit space of

$\tilde{M}^{*}$

then the action induces an $S^{1}$ $(\pi,\tilde{M})$

action of on . Let $\pi$

be its action. The induced map
$\tilde{M}^{*}$
$(\pi^{*},\tilde{M}^{*})$ $\pi\rightarrow$

is an isomorphism. Let
$\pi^{*}$
$U(1)\rightarrow U(n, 1)\rightarrow PU(n, 1)$ be the exact

sequence for the $CR$ case. The projection $P$ maps isomorphically $\rho(\pi)$

onto its image . The homomorphism induces an isomorphism

$\rho(\pi)^{*}$ $\rho$

:
$\rho^{*}$
such that the diagram is commutative:
$\pi^{*}\rightarrow\rho(\pi)^{*}$

$\rightarrow\rho\rho(\pi)$

$\downarrow$ $\downarrow$

$\rho^{*}$

$\pi^{*}\rightarrow\rho(\pi)^{*}$

Definition 5.1.6. $R_{CR}(\pi^{*})$ is the subspace of Hom( , PU( , 1)) $\pi^{*}$

$n$

such that for each element there exists a homeomorphism :

$\rho^{*}$ $f^{*}$
$D^{2n}\rightarrow$

$D^{2n}$
such that $\rho^{*}(\alpha)=f^{*}\circ\alpha\circ f^{*-1}(\alpha\in\pi^{*})$ and in addition the
restriction is a smooth map. Note that
$f^{*}|H_{C}^{n}$ : is $\rho^{*}$ $\pi^{*}\rightarrow\rho^{*}(\pi^{*})$
 Deformation Space 291

an isomorphism and is discrete in PU(n, 1). $R_{CO}(\pi^{*})$ is defined

$\rho^{*}(\pi^{*})$

similarly to be the subspace of $Hom(\pi^{*}, PO(n-1,1)^{0})$ .

Remark 5.1.7. Given an isomorphism : $\rho^{*}$

$\pi^{*}$ $\rightarrow$ $\rho^{*}(\pi^{*})$

Hom( , PU( , 1)), it does not always exist such a homeomor-

$\subset$
$\pi^{*}$
$n$

phism : $D^{2n}\rightarrow D^{2n}$ . However, for example $n=1$ (PU $(1, 1)$
$f^{*}$ $\approx$

$PO(2,1)\approx PSL_{2}(R))$ , and is type-preserving (cf. [p.302, 23]), then

$\rho^{*}$

it is well known that there exists a quasiconformal homeomorphism

:
$f^{*}$
which induces
$D^{2}\rightarrow D^{2}$
. In this case the space $R_{CR}(\pi^{*})$ is$\rho^{*}$

alternatively defined to be the set of those elements consisting of type-

preserving discrete faithful representations of into PU $(1, 1)$ . Note $\pi^{*}$

that $R_{CR}(\pi^{*})\approx R_{CO}(\pi^{*})$ in this case (cf. [26].)

Definition 5.1.8. Let $R_{CR}(\pi)$ be the subspace of $Hom(\pi, U(n, 1))$

whose elements project down to $R_{CR}(\pi^{*})$ . If we note the exact sequence,
$Hom(\pi, U(1))\rightarrow Hom(\pi, U(n, 1))\rightarrow Hom$ ( , PU(n, 1)), then it follows $\pi^{*}$

that
(5.1.9) $R_{CR}(\pi)=R_{CR}(\pi^{*})\times Hom(\pi, U(1))$ .
Similarly,
(5.1.10) $R_{CO}(\pi)=R_{C0}(\pi^{*})\times Hom$ ( , SO(2)).
$\pi$

Lemma 5.1.11. $\overline{ho}1$

maps $\Omega_{0}^{CR}(S^{1}, M)\dot{\iota}ntoR_{CR}(\pi)$ , similarly

for $\Omega_{0}^{CO}(S^{1}, M)$
.

Proof. Let $(\rho, dev)$ We

be a representative element of $\Omega_{0}^{CR}(S^{1}, M)$
.
know that $(\rho, dev)$ : $(S^{1}, \pi,\tilde{M})\rightarrow(U(1), U(n, 1), S^{2n+1}-L(\rho(\pi)))$ is
homeomorphic. Then $(\rho, dev)$ induces a homeomorphism

$(\rho^{*}$
, $dev*$

Note from (ii) of (5.1.2) that $\tilde{M}^{*}=D^{2n}-L(\rho^{*}(\pi^{*}))$ . In particular

dev* : is homeomorphic. Since dev* is still an
$H_{C}^{n}(=Int\tilde{M}^{*})\rightarrow H_{C}^{n}$

immersion, the complete metric of with $ISO(HQ)=PU(n, 1)$ induces $H_{C}^{n}$

a Riemannian metric such that dev* is a local isometry. And hence

dev* : is an isometry. The space
$H_{C}^{n}\rightarrow H_{C}^{n}$
has a compactification $\tilde{M}^{*}$

$D^{2n}=\tilde{M}^{*}\cup L(\pi^{*})$
. The isometry dev* extends to a homeomorphism
:
$f^{*}$
for which $f^{*}(L(\pi^{*}))=L(\rho^{*}(\pi^{*}))$ and
$D^{2n}\rightarrow D^{2n}$ $\rho^{*}(\alpha)=f^{*}\circ\alpha\circ$

$f^{*-1}(\alpha\in\pi^{*})$ . It follows by the definition 5.1.7 that and $\rho^{*}\in R_{CR}(\pi^{*})$

thus $\rho\in R_{CR}(\pi)$ . Q.E.D.

292 Y. Kamishima and S. Tan

The diagram (5.1.5) reduces to the following commutative one.

$\overline{ho1}$

$\Omega_{0}^{CR}(S^{1}, M)\rightarrow$ $R_{CR}(\pi)$

(5.1.13) $\downarrow$ $\downarrow$

$\rightarrow ho1R_{CR}(\pi)/U(n, 1)$

$C\mathcal{R}(S^{1}, M)$ .

Since $Hom(\pi, U(1))$ is a $k$

dimensional torus for some , it follows from $k$

(5.1.9) that
(5.1.13) $R_{CR}(\pi)/U(n, 1)=R_{CR}(\pi^{*})/PU(n, 1)$ $\times T^{k}$
.
(5.1.14) $R_{CO}(\pi)/SO(n-1, 1)^{0}\times SO(2)=R_{CO}(\pi^{*})/SO(n-1,1)^{0}\times T^{k}$ .

5.2. Structure of deformation spaces $\mathcal{T}(S^{1}, M)$

There is the natural homomorphism : Di Out . $\varphi$

$(S^{1}, M)\rightarrow$ $(\Gamma)$

Note that $Ker\varphi$ contains the subgroup $Diffff\ovalbox{\tt\small REJECT}$

exists a right action of $Diffff(S^{1}, M)/Diffff^{0}(S^{1}, M)$ on . We $\mathcal{T}(S^{1}, M)$

examine the structure of in terms of representation spaces,

$\mathcal{T}(S^{1}, M)$

where $\mathcal{T}(S^{1}, M)=C\mathcal{R}(S^{1}, M)$ or $CO(S^{1}, M)$ .

Proposition 5.2.1. Let
$ho1$ : $C\mathcal{R}(S^{1}, M)\rightarrow R_{CR}(\pi^{*})/PU(n, 1)$ $\times T^{k}$

and

$ho1$ : $CO(S^{1}, M)\rightarrow R_{CO}(\pi^{*})/SO(n-1,1)^{0})/SO(n-1,1)^{0}\times T^{k}$

be the holonomy map respectively. Put $G=Ker\varphi/Diffff^{0}(S^{1}, M)$ . If the

fundamental group is torsionfree, then$\pi$

(1) is surjective.
$hoi$

(2) Each fifiber of $hoi$ consists of the $G$ -orbit.

(3) There exists a neighborhood $U$ for each point of $\mathcal{T}(S^{1}, M)$
such
that $ho1(U)$ is open.

Proof. We prove for the $CR$ case. (1) Given $\rho\in R_{CR}(\pi)$ , $\rho(\pi)$

is discrete in $U(n, 1)$ and $L(\rho(\pi))\subset S^{2n-1}$ . Then the group acts $\rho(\pi)$

properly discontinuously on $S^{2n-1}-L(\rho(\pi))$ . Since is torsionfree, $\rho(\pi)$

it acts freely. We obtain a spherical $CR$ manifold $M(\rho)=S^{2n-1}-$

$L(\rho(\pi))/\rho(\pi)$ . It is noted that $U(1)$ acts on $M(\rho)$ by $CR$ automorphisms.

We then show that $M$ is dieomorphic to $M(\rho)$ . For this, let be $\rho^{*}$

an element of $R_{CR}(\pi^{*})$ induced from . There is a homeomorphism $\rho$

 Deformation Space 293

$f^{*}$
: such that $\rho^{*}(\pi^{*})=f^{*}\pi f^{*-1}$ . If we note that $M(\rho)^{*}=$
$D^{2n}\rightarrow D^{2n}$

$D^{2n}-L(\rho^{*}(\pi^{*}))/\rho(\pi^{*})$ then the map induces a homeomorphism : $f^{*}$ $h^{*}$

. Consider the following diagram (cf. (5.1.1));

$M^{*}\rightarrow M(\rho)^{*}$

$\rightarrow\approx$
$\pi^{*}$

$\downarrow$ $\downarrow$

$S^{1}\rightarrow\tilde{M}-\tilde{F}\rightarrow Int$ $M^{*}(=H_{C}^{n})$

$||$ $\downarrow$ $\downarrow$

$ S^{1}\rightarrow M-F\rightarrow$ $H_{C}^{n}/\pi^{*}$

,

where is the fixed point set of . It follows that $M-F=$

$\tilde{F}\approx F$ $S^{1}$

. The same is true for $M(\rho)$ . Then we can find an equivariant

$H_{C}^{n}/\pi^{*}\times S^{1}$

homeomorphism : $M-F$ $\rightarrow M(\rho)-F(\rho)$ which induces Int .

$h_{1}$ $h^{*}|$ $M^{*}$

Here is the fixed point set of $U(1)$ in $M(\rho)$ . Since

$F(\rho)$
$F\approx\partial M^{*}\rightarrow h^{*}$

, we can choose a homeomorphism

$\partial M(\rho)^{*}\approx F(\rho)$ : which $h_{2}$ $F\rightarrow F(\rho)$

covers . Combining and , it

$h^{*}$
is easy to construct an equivariant
$h_{1}$ $h_{2}$

homeomorphism : $M\rightarrow M(\rho)$ . Therefore $M$ admits an $h$

invariant $S^{1}$

spherical $CR$ structure which is mapped by $hoi$ to . This proves (1). $\rho$

(2) Suppose that $hoi$ $([\rho, dev])=hol$ $([\rho, dev])$ . Then it follows that
$\rho=g\circ\rho og^{-1}$ for some $g\in U(n, 1)$ . Since $dev:\tilde{M}\rightarrow S^{2n+1}-L(\rho(\pi))$ ,

and $dev$ : $\tilde{M}\rightarrow S^{2n+1}-L(\rho(\pi))$ are homeomorphisms, we can put

$\tilde{f}=(dev^{r})^{-1}o$ godev. It is easy to see that induces an element $\tilde{f}$

$f\in Diffff(S^{1}, M)$ such that $\varphi(f)=1$ . Hence $(f)\in G$ . By definition we

have that $[\rho, dev]\circ(f)=[\rho, dev$ .

(3) It follows from the Holonomy theorem 4.3.9 that there ex-
ists a neighborhood in for which is open in
$\tilde{U}$
$\Omega_{0}^{CR}(S^{1}, M)$ $\overline{ho}1(\tilde{U})$

$R(\pi, U(n, 1))$ . Let $U$

be the image of in . Since verti- $\tilde{U}$

$C\mathcal{R}(S^{1}, M)$

cal arrows are open maps in the diagram (5.1.6), we obtain that $ho1(U)$
is open. It can be shown similarly for . Q.E.D. $C\mathcal{O}(S^{1}, M)$

Corollary 5.2.2. Suppose that $\hat{\varphi}$

: $Diffff(S^{1}, M)/Diffff^{0}(S^{1}, M)\rightarrow$

Out(\pi ) is infective. Then $C\mathcal{R}(S^{1}, M)$

is homeomorphic to
$R_{CR}(\pi^{*})/PU(n, 1)$ $\times T^{k}$
(Similarly, $C\mathcal{O}(S^{1}, M)$ is homeomorphic to
$R_{CO}(\pi^{*})/SO(n-1,1)^{0}\times T^{k})$ .

See [26] for examples of this Corollary. (Indeed, $Ker$ $\varphi$ $=$

$Diffff\ovalbox{\tt\small REJECT}$

of $Diffff(S^{1}, M)/Diffff^{0}(S^{1}, M)$ on $\Omega(S^{1}, M)$ (cf. (4.3.4)). Let $G=$

294 Y. Kamishima and S. Tan

$Ker\varphi/Diffff^{0}(S^{1}, M)$ be as before. In order to study the action of $G$

on $\Omega(S^{1}, M)$ , we need the following lemma.

Lemma 5.2.3. Suppose that $\pi$

is not virtually solvable.
(1) $U(n, 1)$acts properly on $R_{CR}(\pi)$ .
(2) SO $(n-1, 1)^{0}\times SO(2)$ acts properly on $R_{CO}(\pi)$ .

We prove (1). Recall that $ R_{CR}(\pi)=R(\pi, U(n, 1))\approx$

Proof.
$R$ ( , PU( , )
$\pi^{*}$
. Let $P:R$ ( , PU(n, )
$n$ $1)$
$\times T^{k}$
( PU(n, 1)) $\pi^{*}$
$1)$
$\times T^{k}\rightarrow R$
$\pi^{*},$

be the projection. Given a compact subset $K$ of $R$ ( , PU( , 1)) , $\pi^{*}$
$n$
$\times T^{k}$

put $K^{*}=P(K)$ . Let $\zeta_{U(n,1)}(K)=\{g\in U(n, 1)|g. K\cap K\neq\emptyset\}$

where those elements of $U(n, 1)$ act by conjugation on $R(\pi, U(n, 1))$ .
Recall that $P$ : $U(n, 1)\rightarrow PU(n, 1)$ is the projection with kernel isomor-
phic to $U(1)$ . Then it follows that $\zeta_{U(n,1)}(K)\subset P^{-1}(\zeta_{PU(n,1)}(K^{*}))\approx$
$\zeta_{PU(n,1)}(K^{*})\times U(1)$ . Since $\zeta_{U(n,1)}(K)$ is a closed subset in $U(n, 1)$ , it
suces to show that $\zeta_{PU(n,1)}(K^{*})$ is compact. By the hypothesis, $\pi\approx\pi^{*}$

is not virtually solvable. Then the set $R$ ( , PU( , 1)) consists of stable $\pi^{*}$
$n$

representations in the sense of Johnson-Millson ([p.53, 24]). And so it

follows from Proposition 1.1 ([24]) that PU(n, 1) acts properly on the
subset $R$ ( , PU( , 1)). Hence $\zeta_{PU(n,1)}(K^{*})$ is compact.
$\pi^{*}$
$n$

(2) follows similarly when we note from Proposition 1.1 ([24]) that the
set $R$ ( , SO( $n-1$ , ) ) consists of stable representations.
$\pi^{*}$ $1$
Q.E.D.

Proposition 5.2.4. Suppose that is not virtually solvable. Let $\pi$

$G=Ker\varphi/Diffff^{0}(S^{1}, M)$ be as
before. Then $G$ acts properly discontin-
uously on where $\mathcal{T}(S^{1}, M)=C\mathcal{R}(S^{1}, M)$ or $CO(S^{1}, M)$ .
$\mathcal{T}(S^{1}, M)$

When $K$ is a compact subset of

Proof. , it has only $\mathcal{T}(S^{1}, M)$

to be shown that $\zeta_{G}(K)=\{(f)\in G|K\circ(f)\cap K\neq\emptyset\}$ is com-

pact. Suppose we have sequences $\{f_{i}\}\in G$ and , $[\rho_{i}, dev_{i}]$ $[\rho_{i}, dev_{i}]\in K$

such that where

$[\rho_{i}, dev_{i}]\circ(f_{i})=[\rho_{i},$and $dev_{i}^{J}|$ $\{[\rho_{i}, dev_{i}]\}$ $\{[\rho_{i}, dev_{i}^{J}]\}$

converge to some $[\rho, dev]$ and in $K$ respectively. Then by $[\rho, dev^{J}]$

the remark (1) of (5.1.5) there exists a sequence $U(n, 1)$ $\{g_{i}\}$ $\in$

(resp. SO( $n-1$ , ) $\times SO(2)$ ) such that (i) $1$

, (ii) $g_{i}\circ dev_{i}=dev_{i}\circ\tilde{f}_{i}$

. Since each lies in $Ker\varphi$ , it follows that

$g_{i}\circ\rho_{i}\circ g_{i}^{-1}=\rho_{i}\circ\mu(\tilde{f_{i}})$ $f_{i}$

(ii) . We note that , $\{\rho_{i}\}\in R_{CR}(\pi)$ (resp.

$g_{i}\circ\rho_{i}\circ g_{i}^{-1}=\rho_{i}$ $\{\rho_{i}\}$

$R_{CO}(\pi))$ , and (resp. ) (resp. ). By Lemma 5.2.3, (ii)

$\{\rho_{i}\}$ $\{\rho_{i}\}$
$\rightarrow\rho$
$\rho$

implies that the sequence converges to some $g\in U(n, 1)$ (resp. $\{g_{i}\}$

SO $(n-1, 1)^{0}\times SO(2))$ .

On the other hand, the maps , induce homeomorphisms $dev_{i}$ $dev_{i}^{J}$

$d\hat{e}v_{i}$
: $M\rightarrow S^{m}-L(\rho_{i}(\Gamma))/\rho_{i}(\Gamma)$ , $d\hat{e}v_{i}$
: $M\rightarrow S^{m}-L(\rho_{i}(\Gamma))/\rho_{i}(\Gamma)$ ,
 Deformation Space 295

where $m=2n+1$ or $n$ . Each $g_{i}$ defines a homeomorphism : $S^{m}-$ $\hat{g}_{i}$

$L(\rho_{i}(\Gamma))/\rho_{i}(\Gamma)\rightarrow S^{m}-L(\rho_{i}(\Gamma))/\rho_{i}(\Gamma)$ . Therefore we obtain from (i)

that $f_{i}=(d\hat{e}v_{i})^{-1}(\hat{g}_{i}od\hat{e}v_{i})$
. Since $M$ is compact, $(d\text{\^{e}} v)^{-1}(\hat{g}\circ d\hat{e}v^{J})$
is
also defined so that . Put $f=$ :
$\{f_{i}\}\rightarrow$
$(d\text{\^{e}} v)^{-1}(\hat{g}od\hat{e}v)$ $(d\text{\^{e}} v)^{-1}(\hat{g}\circ d\hat{e}v)$

$M\rightarrow M$
. Since each , it follows that represents an element
$ f_{i}\in Ker\varphi$ $f$

of $G$ . Hence is compact. $\zeta_{G}(K)$ Q.E.D.

For example, $G=Ker\varphi/Diffff^{0}(S^{1}, M)$ is trivial if $dimM=3$ (cf.

[26] . However in general there are examples in higher dimentions for
$)$

which $G$ is nontrivial. For them we have the following.

Proposition 5.2.5. Suppose that $\pi$

is not virtually solvable.
Then $G$ acts freely on $I(S^{1}, M)$ , where or
$\mathcal{T}(S^{1}, M)=C\mathcal{R}(S^{1}, M)$

$C\mathcal{O}(S^{1}, M)$ .

We prove the case that $\mathcal{T}(S^{1}, M)=C\mathcal{R}(S^{1}, M)$ . Suppose

Proof.
that $[\rho, dev]o(f)=[\rho, dev]$ . Then there exists an element $g\in U(n, 1)$
such that (1) godev , (2) $ g\circ\rho\circ g^{-1}=\rho\circ\mu(\tilde{f})=\rho$ . If
$=dev\circ\tilde{f}$
is $\rho^{*}$

the corresponding element in ( , PU( , 1)) then (2) implies that (3)
$R$ $\Gamma^{*}$
$n$

$g^{*}o\rho^{*}og^{*-1}=\rho^{*}$ for $g^{*}\in PU(n, 1)$ . The group acts invariantly $\rho^{*}(\Gamma^{*})$

in . Suppose that
$H_{C}^{n}$ leaves invariant a totally geodesic subspace
$\rho^{*}(\Gamma^{*})$

of
$H_{C}^{k}$
for 1 $\leqq k\leqq n$ . Then
$H_{C}^{n}$ leaves $S^{2k-1}$ invariant so $\rho^{*}(\Gamma^{*})$

that it belongs to the subgroup $Aut_{CR}(S^{2n-1}, S^{2k-1})=P(U(k, 1)$ $\times$

$U(n-k))$ . Let $Q$ : $P(U(k, 1)\times U(n-k))\rightarrow PU(k, 1)$ be the projection
whose kernel is isomorphic to $U(n-k)$ . We can assume that is the $k$

smallest dimension. And so is Zariski-dense in PU(k, 1). The $Q(\rho^{*}(\Gamma^{*}))$

condition (3) implies that leaves also $S^{2k-1}$ . It implies that $g^{*}$ $ g^{*}\in$

$P(U(k, 1)$ $\times U(n-k))$ . Then the element $Q(g^{*})$ centralizes the group
and so does its algebraic closure. Since the algebraic closure
$Q(\rho^{*}(\Gamma^{*}))$

is PU $(k, 1)$ by the above remark, $Q(g^{*})$ must be the identity map.
In particular we obtain that $g^{*}\in U(n-k)$ . As $U(n, 1)=P(U(n, 1)$ $\times$

$U(1))$ , it follows that $g\in U(n-k)\times U(1)(=P(Z(k, 1)$ $\times U(n-k)\times$

$U(1)))$ where $Z(k, 1)$ is the center of $U(k, 1)$ . On the other hand,
$dev$ : $\tilde{M}\rightarrow S^{2n+1}-L(\rho(\pi))$ is homeomorphic and by (1) it follows that

$\tilde{f}=(dev)^{-1}$ ogodev. It is noted that $L(\rho(\pi))=L(\rho^{*}(\pi^{*}))\subset S^{2k-1}$

and $S^{2k-1}$ is the fixed point set of $U(n-k)$ . We can choose a path in $c$

$U(n-k)$ between and the identity map. By the above remark there
$g^{*}$

is a lift of the path starting at with its endpoint $\tilde{c}(1)\in U(1)$ . Since
$\tilde{c}$

$c$ $g$

$dev$ is equivariant with respect to and $U(1)$ actions, we conclude that $S^{1}$

is isotopic to
$\tilde{f}$

. It is easy to check that is isotopic to the identity

$\tilde{c}(1)$
$\tilde{f}$

map of . Hence belongs to $\tilde{M}$

$f$
$Diffff\ovalbox{\tt\small REJECT}$
296 Y. Kamishima and S. Tan

We can prove similary for the case that $\mathcal{T}(S^{1}, M)=CO(SO(2), M)$ .

Q.E.D.

Corollary 5.2.6. Let $M$ be a closed invariant spherical $CR$

$S^{1}$

manifold of dimension $2n+1$ (resp. a closed invariant confor-$S^{1}$

mally flat manifold.

$n$ Suppose that the orbit space is a complex $M^{*}$

Kleinian orbifold $D^{2n}-L(\pi^{*})/\pi^{*}$ with nonempty boundary (resp. $a$

Kleinian orbifold $D^{n-1}-L(\pi^{*})/\pi^{*}$ with nonempty boundary) and is $\pi^{*}$

torsionfree.
If is
$\pi_{1}(M)$ not virtually solvable, then
(1) $ho1$ : $C\mathcal{R}(S^{1}, M)\rightarrow R_{CR}(\pi^{*})/PU(n, 1)$ is a covering map
$\times T^{k}$

whose fifiber is isomorphic to . $G$

(2) $ho1$ : $CO(S^{1}, M)\rightarrow R_{CO}(\pi^{*})/SO(n-1,1)^{0}\times T^{k}$ is a covering

map whose fifiber is isomorphic to $G$ .

Proof The group $G$ acts properly discontinuously and freely on

$\mathcal{T}(U(1), M)$ by Lemma 5.2.3 and Proposition 5.2.4. Thus there exists a

neighborhood $U$ in $I(U(1), M)$ such that $ Uog\cap U=\emptyset$ if and only if
$g\neq 1$ for $g\in G$ . Then the result follows from Proposition 5.2.1.

Q.E.D.

References

[1] B. Apanasov, Two kinds of deformation of conformal structure which are

distinct from bending, Proceedings of the Thirteenth Rolf Nevalinna
Colloquium and Teichm\"uller theory Conference, Helsinki (1987).
[2] , Deformations of conformal structures on hyperbolic manifolds,

(1990), to appear.
[3] W. Ballmann, M. Gromov and V. Schroeder, Manifolds of nonpositive
curvature, Progress in Math. 61, Birkh\"auser, 1985.
[4] A. Beardon, The geometry of discrete subgoups, Springer, New York,
1983.
[5] L. Bers, On boundaries of Teichm\"uller spaces and on Kleinian groups I,
Ann. of Math., 91 (1970), 570-600.
[6] G. Bredon, Introduction to compact transformation groups, Academic
Press, New York, 1972.
[7] D.M. Burns and S. Schnider, Spherical hypersurfaces in complex mani-
folds, Invent. Math., 33 (1976), 223-246.
[8] R. Canary, D.B.A. Epstein and P. Green, Notes on notes of Thurston,
London Mathematical Society Lecture Notes, 111 (1987), 3-92, Cam-
bridge University Press.
 Deformation Space 297

[9] A. Casson and S. Bleiler, Automorphisms of surfaces after Nielsen

and Thurston, London Mathematical Society Student Texts 9,
Cambridge University Press, 1988.
[10] P.J. Cheeger and D. Ebin, Comparison Theorems in Riemannian Ge-
ometry, North-Holland Publ. Co., Amsterdam, 1975.
[11] S. Chen and L. Greenberg, Hyperbolic Spaces, Contribution to Analysis
(1974), 49-87, Academic Press, New York.
[12] P.E. Conner and F. Raymond, Injective operations of the toral groups,
Topology, 10 (1971), 283-296.
[13] D.B.A. Epstein and A. Marden, Convex hulls in hyperbolic spaces, a
theorem of Sullivan,and measured pleated surfaces, London Mathe-
matical Society Lecture Notes, 111 (1987), 114-253, Cambridge Uni-
versity Press.
[14] D. Fried, Closed similarity manifolds, Comment. Math. Helv., 55 (1980),
576-582.
[15] D. Gallo, W. Goldman and R. Porter, Projective structures with mon-
odromy in $PSL(2, R)$ , Preprint (1989).
[16] W. Goldman, Projective structures with Fuchsian holonomy, J. Di.
Geom., 25 (1987), 297-326.
[17] , Geometric Structures on manifolds and varieties of representa-

tions, Contemporary Math., 74 (1988), Proceedings of the Conference

on Representation Theory.
[18] W. Goldman and Y. Kamishima, Topological rigidity of developing
maps with applications to conformally flat structures, Contemporary
Math. A.M.S., 74 (1988), 199-203.
[18] , Conformal automorphisms and conformally flat manifolds, Trans.

A.M.S., 323 (1991), 797-810.

[20] W. Goldman and J. Millson, The deformation theory of representations
of fundamental groups of compact K\"ahler Manifolds, Publ. I.H.E.S.,
to appear.
[21] W. Goldman and J. Millson, Local rigidity of discrete groups acting on
complex hyperbolic space, Invent. Math., 88 (1987), 495-520.
[22] R. Gunning, Special coordinate coverings of Riemann surfaces,
Math. Ann., 170 (1967), 67-85.
[23] W. Harvey, Discrete groups and automorphic functions, Academic
Press, 1972.
[24] D. Johnson and J. Millson, Deformation spaces associated to com-
pact hyperbolic manifolds, Discrete groups in Geometry and Analysis,
Progress in Math, 67 (1987), 48-106.
[25] Y. Kamishima, Conformally flat manifolds whose development maps are
not surjective, Trans. A.M.S., 294 (1986), 607-623.
[26] , Conformally flat -manifolds with compact automorphism
$3$

groups, J. London Math. Soc. 38(2) (1988), 367-378.

[27] Y. Kamishima and T. Tusboi, $CR$-structures on Seifert manifolds, In-
vent. Math., 104 (1991), 149-163.
298 Y. Kamishima and S. Tan

[28] M. Kapovich, Flat conformal structures on -manifolds, Preprint (1989).

$3$

[29] , Deformation spaces of flat conformal structures, Preprint (1989).

[30] C. Korouniotis, Deformations of hyperbolic manifolds, Math. Proc.

Cambridge Phil. Soc., 98 (1985), 247-261.
[31] I. Kra, A generalization of a theorem of Poincar\e, Proc. of A.M.S., 27
(1971), 299-302.
[32] R. Kulkarni, On the principle of uniformization, J. Di. Geom., 13
(1978), 109-138.
[33] , Groups with domains of discontinuity, Math. Ann., 237 (1978),

253-272.
[34] R. Kulkarni and U. Pinkall, Uniformization of geometric structures
with applications to conformal geometry, Proceedings, Di. Geom.
Pe\~nscola , Lecture notes in Math., 1209 (1986), 140-209,
Springer-Verlag.
[35] , A canonical metric for M\"obius structures and its applications,

Preprint.
[36] R. Kulkarni and K.B. Lee and F. Raymond, Deformation spaces for
Seifert manifolds, Proceedings, Special year in Geometry and Topol-
ogy, Univ. of Maryland, Lecture notes in Math., 1167 (1985),
180-216, Springer-Verlag.
[37] L. Lok, Deformations of locally homogeneous spaces and Kleinian
groups, Doctoral dissertation (1984), Columbia University.
[38] A. Marden, The geometry of finitely generated Kleinian groups, Ann. of
Math., 99 (1974), 383-462.
[39] S. Matsumoto, Foudations of flat conformal structure, in these proceed-
ings.
[40] J. Morgan and P. Shalen, Valuations, trees, and degenerations of hyper-
bolic structures, I, Ann. of Math, 120 (1984), 401-476.
[41] , Degenerations of hyperbolic structures, II: Measured Lamina-

tions,Trees and -Manifolds, Preprint, 1986.

$3$

[42] P. Orlik, Seifert manifolds, Lecture Notes in Math., 291 (1972),

Springer-Verlag.
[43] P. Orlik and F. Raymond, Actions of $SO(2)$ on -manifolds, Proceedings
$3$

of the Conference on Transformation groups, Lecture Notes in Math.,

299 (1968), 297-318, Springer-Verlag.
[44] F. Raymond, Classification of the actions of the circle on -manifolds,
$3$

Trans. Amer. Math. Soc., 31 (1968), 51-78.

[45] G.P. Scott, The geometries of -manifolds, Bull. London Math., 15
$3$

(1983), 401-487.
[46] D. Sullivan and W. Thurston, Manifolds with canonical coordi-
nates:Some exapmles, Lens. Math., 29 (1983), 15-25.
[47] Ser P. Tan, Representations of surfaces into $PSL(2, R)$ and Geometric
structures, Doctoral dissertation (1988), University of Maryland.
[48] W. Thurston, The geometry and topology of three manifolds, preprint.
[49] J. Wolf, Spaces of constant curvature, McGraw-Hill, Inc., 1967.
 Deformation Space

Yoshinobu Kamishima
Department of Mathematics
Kumamoto University
Kumamoto 860
Japan

Ser Peow Tan

Department of Mathematics
National University of Singapore
Lower Kent Ridge Rd
Singpore 0511
Singapore
Advanced Studies in Pure Mathematics 20, 1992
Aspects of Low Dimensional Manifolds
pp. 301-330

On -Manifolds Homotopy Equivalent

$4$

to the -Sphere $2$

Osamu Saeki

Dedicated to Professor Akio Hattori on his 60th birthday

\S 1. Introduction

Let $V$ be a compact topological 4-manifold homotopy equivalent

to the 2-sphere . We say such a 4-manifold is a homotopy
$S^{2}$
. The $S^{2}$

boundary of $V$ is always a closed connected 3-manifold with the same

$\partial V$

integral homology and the same linking pairing as those of the lens space
$L(p, 1)$ for some $p(\geq 0)$ . We say such a 3-manifold is a homology $L(p, 1)$ .

For a fixed homology $L(p, 1)$ , $M$ , the homotopy s bounded by $M$ are
$S^{2}$

classified up to homeomorphism by certain equivalence classes of some

elements of ( $M$ ; Z) ([3]).
$H_{1}$

In this paper, concerning homotopy s, we consider the following

$S^{2}$

problems.
(A) For a fixed homology $L(p, 1)$ , $M$ , how many homotopy s does $M$ $S^{2}$

bound? Furthermore, how many of them admit smooth structures?.

(B) Give a lower bound for the genera of topologically locally flatly
embedded surfaces in $V$ representing the generator of $H_{2}(V;Z)$ . If $\gamma$

$V$ is smooth, what is the necessary condition for to be represented $\gamma$

by a smoothly embedded 2-sphere?

(C) Let $K$ be a tame knot in the boundary of $V$ . Under what condition
does $K$ bound a topologically embedded flat 2-disk in $V$ ? If $V$ and
$K$ are smooth and $K$ bounds such a topologically embedded 2-disk,

does $K$ also bound a smoothly embedded 2-disk in $V$ ?

(D) Does there exist a homotopy admitting more than one smooth
$S^{2}$

structures?.
In [28], we considered problem (A) and showed that if $V$ is a smooth
homotopy satisfying a certain condition on the order of $H_{1}(\partial V;Z)$
$S^{2}$

Received June 27, 1989.

Revised March 4, 1991.
302 O. Saeki

such that the generator of $H_{2}(V;Z)$ is represented by a smoothly embed-

ded 2-sphere, then every smooth homotopy with the same boundary
$S^{2}$

as $V$ is homeomorphic to . Furthermore we gave the exact number of

$V$

homotopy s bounded by the 3-manifold

$S^{2}$
. In this paper we con-
$\partial V$

sider smoothly immersed 2-spheres instead of embedded ones and give a

similar result. More precisely, we show that if $V$ is a smooth homotopy
satisfying a certain condition on the order of $H_{1}(\partial V;Z)$ such that the
$S^{2}$

generator of $H_{2}(V;Z)$ is represented by a smoothly immersed 2-sphere

with relatively few double points, then the same results as above hold
(2). We note that, in some cases, this result can be applied to give
a lower bound for the number of double points of smoothly immersed
2-spheres representing the generator of $H_{2}(V;Z)$ .
In \S 3 we define some topological invariants for homotopy s and $S^{2}$

use them to attack problem (B). First, we define a Casson invariant

for a homotopy using the extension of the usual Casson invariant for
$S^{2}$

homology 3-spheres ([6]) to marked homology lens spaces ([4, 15]). We

show that if the generator of the second homology group of a smooth
homotopy $S^{2}$
is represented by a smoothly embedded 2-sphere, then
its Casson invariant modulo 2 must vanish. In fact, this is proved only
using the well-known theorem of Rohlin. We do not know what essential
properties of a homotopy this Casson invariant reflects. Next we
$S^{2}$

define Casson-Gordon invariants for a homotopy (cf. [7]) and use $S^{2}$

them to give a lower bound for the genera of topologically locally flatly
embedded surfaces representing the generator of the second homology
group of a homotopy . In our case these invariants are the -signatures
$S^{2}$
$p$

of a certain knot in a homology 3-sphere. If a smooth homotopy $S^{2}$

consists of one 0-handle and one 2-handle, then it has already been
known that a lower bound for the genera of such smoothly embedded
surfaces is given by the -signatures of the knot along which the 2-handle
$p$

is attached to the 0-handle ([31]). In this paper, we extend this to general

homotopy s employing a method similar to that in [19].
$S^{2}$

In \S 4, we give a sucient condition for certain tame knots in the

boundary of a homotopy to bound topologically embedded flat 2-
$S^{2}$

disks. We see that almost all knots satisfying a certain homological

condition bound such 2-disks in the homotopy . Furthermore, we give
$S^{2}$

an example of smooth knots in the boundary of a smooth homotopy $S^{2}$

which bound topologically embedded flat 2-disks in the homotopy $S^{2}$

but never bound smooth ones. In the case of knots in the boundary of
the 4-ball, the same examples have been found first by Casson and, after
that, by several authors [8, 9, 18]. Our technique is similar to theirs in
that we use the celebrated theorem of Donaldson [10].
As to problem (D), Akbulut [1] has recently found a compact homo-
 On 4-Manifolds Homotopy Equivalent to the 2-Sphere 303

topy with more than one smooth structures. In this paper we give
$S^{2}$

an example of infinitely many open 4-manifolds homotopy equivalent to

each of which admits at least 3 smooth structures (5).
$S^{2}$

Throughout the paper, all homology groups are with integral coef-
ficients unless otherwise indicated.

\S 2. Smooth homotopy $S^{2}$

and immersed -spheres $2$

Definition. We say that a non-zero integer satisfies property $p$

if-l is not a quadratic residue modulo ; i.e., if

$(\neq)$ $n^{2}\not\equiv-1(mod p)$ $p$

for every integer . $n$

Lemma 2.1. Let $|p|=2^{e}p_{1}^{e_{1}}p_{2}^{e_{2}}\cdots p_{r}^{e_{r}}(e\geq 0, e_{i}\geq 1)$ be the prime

decomposition $of|p|$ . Then $p$
satisfifies property $(\neq)$
if and only if $e\geq 2$
or $p_{i}\equiv 3(mod 4)$ for some $p_{i}$ .

For the proof of Lemma 2.1, see, for example, the proof of Corollary
3.11 in [28]. Note that, by Lemma 2.1, if $|p|\equiv 0,3(mod 4)$ then $p$

satisfies property $(\#)$ .

Let $V$ be a smooth oriented homotopy . Define $k_{+}(V)$ (resp. $S^{2}$

$k_{-}(V))$ to be the minimum number of positive (resp. negative) double

points of smoothly immersed self-transverse 2-spheres representing the

generator of $H_{2}(V)$ . We call $k_{+}(V)$ (resp. $k_{-}(V)$ ) the positive (resp.
negative) kinkiness of $V$ . Then our main theorem of this section is the
following.

Theorem 2.2. Let $V$ be a smooth oriented homotopy bounded $S^{2}$

by a homology $L(p, 1)$ , $M$ , and let $\gamma\in H_{2}(V)$ be a generator. We assume
satisfifies property
$p=|\gamma^{2}|$ . $Let\in be$ the sign of . If
$(\neq)$ $\gamma^{2}$

$(p-6)/4$ ( : even)
$p$

$(p-1)/4$ $(p: odd,$ $\neq 15,21)$

$k_{\Xi}(V)\leq\{$
2 $(p=15)$
4 $(p=21)$ ,

then the following holds.

(1) Every smooth homotopy bounded by $M$ is homeomorphic to $V$ . $S^{2}$

(2) Every homeomorphism $h:M$ acts on $H_{1}(M)$ by the multipli- $\rightarrow M$

cation $of\pm 1$ .
(3) If $(e\geq 0, e_{i}\geq 1)$ is the prime decomposition
$p=2^{e}p_{1}^{e_{l}}p_{2}^{e_{2}}\cdots p_{r}^{e_{r}}$

of , then the number of homeomorphism types of homotopy s

$p$
$S^{2}$
304 O. Saeki

bounded by $M$ is equal to

$2^{r-1}$
$(e=1)$
$\{$ $2^{r}$
$(e=0,2)$
$2^{r+1}$ $(e\geq 3)$ .

Remark. Even when does not satisfy property $(\#)$ , part (2) of
$p$

Theorem 2.2 does hold if the homeomorphism is orientation preserving. $h$

Before we prove Theorem 2.2, we describe some of its consequences.

The proof will be given at the end of this section.
Let $K$ be a smooth knot in the oriented 3-sphere . For an integer $S^{3}$

, we denote by $V(K;p)$ the smooth oriented homotopy

$p$ obtained $S^{2}$

by attaching a 2-handle to the 4-ball along the knot $K$ with the p- $D^{4}$

framing. Furthermore we denote by $M(K;p/1)$ the boundary 3-manifold

of $V(K;p)$ . Note that $M(K;p/1)$ is dieomorphic to the 3-manifold
obtained by performing the $p/1$ -Dehn surgery on the knot $K([26])$ .
We denote by $u(K)$ the unknotting number of a knot $K$ in (for $S^{3}$

example, see [16] . Note that $)$$K$ bounds in the 4-ball a smoothly im-
mersed self-transverse 2-disk with $u(K)$ double points ([9]). Taking the
union of this immersed 2-disk and the core disk of the 2-handle, we
can represent the generator of $H_{2}(V(K;p))$ , for any , by a smoothly $p$

immersed 2-sphere with $u(K)$ double points; i.e., we always have

$k_{\pm}(V(K;p))\leq k_{+}(V(K;p))+k_{-}(V(K;p))\leq u(K)$ .

Then we obtain the following immediately.

Corollary 2.3. Let $K$ be a smooth knot in and let be an $S^{3}$

$p$

integer satisfying property . If $|p|\geq 4u(K)+6$ , then the following

$(\neq)$

holds.
(1) Every smooth homotopy bounded by $M(K;p/1)$ is homeomorphic
$S^{2}$

to $V(K;p)$ .
(2) Every homeomorphism : $M(K;p/1)$ $M(K;p/1)$ acts on
$h$ $\rightarrow$

$H_{1}(M/(K;p/1))$ by the multiplication $of\pm 1$ .

(3) If is the prime decomposition
$|p|=2^{e}p_{1}^{e_{1}}p_{2}^{e_{2}}\cdots p_{r}^{e_{r}}(e\geq 0, e_{i}\geq 1)$

$of|p|$ , then the number of homeomorphism types

of homotopy s $S^{2}$

bounded by $M(K;p/1)$ is equal to

$2^{r-1}$
$(e=1)$
$\{$
2 $r$

$(e=0,2)$
$2^{r+1}$ $(e\geq 3)$ .
 On $A$
-Manifolds Homotopy Equivalent to the 2-Sphere 305

This corollary shows that, for any knot $K$ , if is suciently large $|p|$

and satisfies property $(\#)$ , then $M(K;p/1)$ satisfies the above properties.
This is the major dierence between Theorem 2.2 and the result obtained
in [28]; the latter is applicable only to slice knots.
Let and be relatively prime integers. Then we denote by $T(a, b)$
$a$
$b$

the left-hand $(ab>0)$ or the right-hand $(ab<0)$ torus knot of type

$(a, b)$ . As a corollary of Theorem 2.2 we have the following.

Corollary 2.4. Every smooth homotopy bounded by the lens $S^{2}$

space $L(4n+3,4)$ $(n \geq 0)$ is homeomorphic to the handlebody

$V(T(2,2n+1);-(4n+3))$ .
By Moser [25], $\partial V(T(2,2n+1);-(4n+3))=M(T(2,2n+$
Proof.
$1);-(4n+3)/1)$ is homeomorphic to the lens space $L(4n+3,4)$ . Note
that the integer $p=4n+3$ always satisfies property $(\#)$ . On the other
hand, it is well-known that $u(T(2,2n+1))\leq n$ . Thus by the previous
remark, $k_{-}(V(T(2,2n+1);-p))\leq n=(p-3)/4$ . Since is odd, the $p$

result follows from Theorem 2.2 unless $p=15$ . If $p=15$ , it can be

shown that the number of homotopy s bounded by $L(15,4)$ is 1 (see $S^{2}$

Example 2.8 of [28] . This completes the proof. $)$

In [28], we remarked that there are exactly 1401 homotopy s $S^{2}$

bounded by the lens spaces $L(p, q)$ with $2\leq p\leq 100$ and that among
them there exist at least 701 homotopy s which cannot admit any $S^{2}$

smooth structures and at least 274 homotopy s admitting smooth $S^{2}$

structures. Using the above Corollary 2.4 and the technique used in
the proof of Theorem 2.2, we can find, among the 1401 homotopy s, $S^{2}$

additional 16 non-smoothable homotopy s. We also note that, using $S^{2}$

the result of Maruyama [23], we can find additional 14 homotopy $S^{2}$

$s$

admitting smooth structures.

Next we indicate how Theorem 2.2 can be applied to give lower
bounds for the kinkinesses of a homotopy . $S^{2}$

Definition. Let $V$ be a homotopy and let $S^{2}$

$\delta\in H_{2}(V, \partial V)$

be a generator. Then we define $\beta(V)=\partial\delta\in H_{1}(\partial V)$ , where $\partial$

:
is the boundary homomorphism. Note that
$H_{2}(V, \partial V)\rightarrow H_{1}(\partial V)$ $\beta(V)$

generates and is determined up to the multiplication $of\pm 1$ .

$H_{1}(\partial V)$

Remark. If $H_{1}(\partial V)$ is finite cyclic of order , we always have $p$

$1k(\beta(V), \beta(V))=\pm 1/p$ , where $1k:H_{1}(\partial V)\times H_{1}(\partial V)\rightarrow Q/Z$ is the

linking pairing of $\partial V$

.
Theorem 2.5. Let $V$ be a smooth homotopy with $S^{2}$
$H_{1}(\partial V)$
fifinite
cyclic of order . Suppose there exists a homeomorphism
$p$
$h$
: $\partial V\rightarrow\partial V$
306 O. Saeki

such that $h_{*}(\beta(V))=r\beta(V)$ with $r\not\equiv\pm 1(mod p)$ and $r^{2}\equiv 1(mod p)$ .
Then we have
$(p-4)/4$ ( : even)
$p$

$(p+1)/4$ $(p: odd,$ $\neq 15,21)$

$k_{\Xi}(V)\geq\{$
3 $(p=15)$
5 $(p=21)$ ,

$where\in(=\pm 1)$ is the signature of $V$ .

If satisfies property $(\#)$ , the above Theorem is a direct consequence

$p$

of Theorem 2.2. The general case can be proved by the same method as
in the proof of Theorem 2.2 below.
As a typical example, we can obtain a lower bound for the kink-
inesses of the homotopy s obtained by attaching a 2-handle to the
$S^{2}$

4-ball along some torus knots.

Corollary 2.6. Let and be relatively prime integers greater

$s$ $r$

than 1. Suppose that $s^{2}\not\equiv\pm 1(mod rs+\in)$ and that $ rs+\in$ divides
$s^{4}-1$ , $where\in=\pm 1$ . Then we have

$(rs+\in-4)/4$ ( $ rs+\in$ : even)

$(rs+\in+1)/4$ $(rs+\epsilon : odd\neq 15,21)$
$k_{-}(V(T(r, s);-(rs+\in)))\geq\{$
3 $(rs+\in=15)$
5 $(rs+\in=21)$ .

Proof. By Moser [25], is dieomorphic to

$\partial V(T(r, s);-(rs+\in))$

the lens space $L(rs+\in, s^{2})$ . Since , there exists a

$s^{4}\equiv 1(mod rs+\in)$

homeomorphism : $L(rs+\in,
$h$ s^{2})\rightarrow L(rs+\in, s^{2})$ which acts on the first
homology group by the multiplication of . Now the result follows from $s^{2}$

Theorem 2.5.

Example 2.7. There do exist , $r$ $s$

and $\in satisfying$ the condition
in Corollary 2.6. For example, we have

$k_{-}(T(2,7);-15)\geq 3$ ,

$k_{-}(T(3,5);-16)\geq 3$ ,

$k_{-}(T(3,13);-40)\geq 9$ ,

$k_{-}(T(4,13);-51)\geq 13$ .

We also note that $k_{-}(T(a, b);p)\leq(|a|-1)(|b|-1)/2$ for any . $p$

 On 4-Manifolds Homotopy Equivalent to the 2-Sphere 307

Using Corollary 2.6, we can obtain a lower bound for the unknotting
numbers of certain torus knots. However, this lower bound is worse
than the one obtained in [16]. There is a conjecture that the unknotting
number of the torus knot $T(a, b)$ is equal to $(|a|-1)(|b|-1)/2$ . We
do not know whether there is a smoothly immersed 2-sphere in some
$V(T(a, b);p)$ representing the generator of $H_{2}(V(T(a, b);p))$ with the
number of double points strictly fewer than $(|a|-1)(|b|-1)/2$ . Note
that if $p=\pm 1$ , then there always exists a topologically locally flatly
embedded 2-sphere or torus representing the generator of the second
homology group (see Proposition 3.6).
Now we proceed to the proof of Theorem 2.2. Our method is similar
to that in [28].
Let $X$ be a smooth closed 1-connected oriented 4-manifold. Given
a homology class $\zeta\in H_{2}(X)$ , one can represent by an immersed 2- $\zeta$

sphere whose self-intersections are transverse. Define (resp. ) to $d_{\zeta}^{+}$

$d_{\zeta}^{-}$

be the minimum number of positive (resp. negative) double points of

such immersed 2-spheres representing . The following is a theorem of $\zeta$

Kuga and Suciu which plays a key role in the proof of Theorem 2.2. See
also the remark after Theorem 4 in [16].

Theorem 2.8 (Kuga [22], Suciu [30]).

(1) Let $X$ be a smooth closed oriented 4-manifold homotopy equivalent to
and let and be generators of $H_{2}(X)$ such that $\xi^{2}=\eta^{2}=0$
$S^{2}\times S^{2}$ $\xi$
$\eta$

and . $\eta=1$ . If $\zeta=a\xi+b\eta\in H_{2}(X)$ satisfifies

$\xi$
, then we have $\zeta^{2}\neq 0$

$d_{\zeta}^{\Xi}\geq\min\{(|a|-1)(|b|-1)$ , $[\frac{|ab|+1}{2}]\}$ ,

where the sign of

$\in is$ and, for $x\in Q$ , denotes the largest
$\zeta^{2}$
$[x]$

integer not exceeding . $x$

(2) Let $X$ be a smooth closed oriented 4-manifold homotopy equivalent

to and let
$CP^{2}\#\overline{CP^{2}}$
and be generators of $H_{2}(X)$ such that
$\xi$
$\eta$

$\xi^{2}=1$ , $\eta^{2}=-1$ and . $\eta=0$ . If $\zeta=a\xi+b\eta\in H_{2}(X)$ satisfifies

$\xi$

, then we have
$\zeta^{2}\neq 0$

$d_{\zeta}^{\epsilon}\geq\min\{\frac{(|a|+|b|-2)(||a|-|b||-1)}{2}$ , $[\frac{|a^{2}-b^{2}|+3}{4}]\}$ ,

$where\in is$ the sign of $\zeta^{2}$

.

Theorem 2.2. Changing the orientation of $V$ if necessary,

Proof of
we may assume $V$ is positive definite $(i.e., \in=1)$ . Let $V$ be a smooth
308 O. Saeki

homotopy with dieomorphic to $M$

$S^{2}$
and let :
$\partial V$ $=\partial V$ $h$ $\partial V\rightarrow\partial V$

be a homeomorphism. Changing by an isotopy, we assume is a $h$ $h$

dieomorphism. Furthermore, we orient $V$

so that is an orientation $h$

preserving map. Thus $V$

is not necessarily positive definite. We have,
for some $r\in Z$ , $h_{*}(\beta(V))=r\beta(V)$ . We will show $r\equiv\pm 1(mod p)$ .
Then the part (1) of Theorem 2.2 follows from a result of Boyer [3].
Furthermore the part (2) is also proved if we set $V=V$ . Then the part
(3) follows by the same argument as in [28].
Set $X=V\bigcup_{h}(-V)$ , which is a smooth closed 1-connected 4-
manifold with $H_{2}(X)\cong Z\oplus Z$ . $X$ has the orientation induced from
those of $V$ and $-V$ . By [28], there are generators and of $H_{2}(X)$ $\theta$
$\tau$

with the following properties.

(i) is represented by a smoothly immersed 2-sphere with $k_{+}(V)$ pos-
$\theta$

itive double points.

(ii) . $\theta=p$ and . $\tau=r$ .
$\theta$ $\theta$

Remember that comes from the generator of $H_{2}(V)$ and is de-

$\theta$
$\tau$

fined to be the union of the generator of $H_{2}(V, \partial V)$ and times the $r$

generator of . $H_{2}(V, \partial V)$

Set $ t=\tau$ . . Then the intersection matrix of $X$ with respect to

$\tau$
$\theta$

and is $\tau$

$Q=\left(\begin{array}{ll}p & r\\r & t\end{array}\right)$

.

Since $Q$ is unimodular, $\det Q=pt-r^{2}=\pm 1$ . Hence $r^{2}\equiv\mp 1(mod p)$ .

Since satisfies property $(\#)$ , $r^{2}\equiv 1(mod p)$ . Thus $\det Q=-1$ and
$p$

the intersection form of $X$ is indefinite. Hence, there are generators $\xi$

and of $H_{2}(X)$ with respect to which the intersection matrix of $X$ is

$\eta$

one of the following forms;

(A) $\left(\begin{array}{ll}0 & 1\\1 & 0\end{array}\right)$

or

(B) $\left(\begin{array}{ll}1 & 0\\0 & -1\end{array}\right)$

.

Case (A). Since

must be of even type, is even. Suppose
$Q$ $p$

$\theta=a\xi+b\eta(a, b\in Z)$ . Since $ p=\theta$ . $\theta=2ab$ is positive, we may assume

$a>0$ and $b>0$ , changing the orientations of and if necessary. By $\xi$
$\eta$

Theorem 2.8, we have

(2.1) $d_{\theta}^{+}\geq\min\{(a-1)(b-1)$ , $[\frac{ab+1}{2}]\}$ .

 On 4-Manifolds Homotopy Equivalent to the 2-Sphere 309

On the other hand, by the hypothesis, we have

(2.2) $d_{\theta}^{+}\leq k_{+}(V)\leq\frac{p-6}{4}$

.

Combining (2.2) with (2.1), we have

(2.3) $\frac{ab-3}{2}\geq\min\{(a-1)(b-1)$ , $[\frac{ab+1}{2}]\}$ .

Since $(ab-3)/2<[(ab+1)/2]$ , we have $(ab-3)/2\geq(a-1)(b-1)$ . This

shows that $a\leq 1$ or $b\leq 1$ . Then the same argument as in [28] shows
$r\equiv\pm 1(mod p)$ .
Case (B). Suppose $\theta=a\xi+b\eta(a, b\in Z)$ . We may assume $a\geq 0$
and $b\geq 0$ , changing the orientations of $\xi and/or$ if necessary. Note $\eta$

that . $\theta=a^{2}-b^{2}>0$ . By Theorem 2.8, we have

$ p=\theta$

(2.4) $d_{\theta}^{+}\geq\min\{\frac{(a+b-2)(a-b-1)}{2}$ , $[\frac{a^{2}-b^{2}+3}{4}]\}$ .

If $p$ is even, we have

(2.5) $d_{\theta}^{+}\leq k_{+}(V)\leq\frac{p-6}{4}$

.

Combining this with (2.4), we obtain $a+b\leq 3$ or $a-b\leq 1$ . Since

, $b\geq 0$ and is prime to , we have $a-b=1$ . This contradicts the
$a\geq 0$ $a$
$b$

fact that $p=a^{2}-b^{2}$ is even.

If $p=(a+b)(a-b)$ is odd not equal to 15 or 21, we have, by the
hypothesis,

(2.6) $d_{\theta}^{+}\leq k_{+}(V)\leq\frac{p-1}{4}$

.

Combining this with (2.4), we obtain

(2.7) $(a+b-4)(a-b-2)\leq 3$ .

Note that $a+b$ and $a-b$ are odd and that and are relatively prime. $a$
$b$

Then we have $a-b=1$ or $(a, b)=(4,1)$ or $(5, 2)$ . Since $ p=a^{2}-b^{2}\neq$

$15,21$ , we have $a-b=1$ . When $p=15$ , we obtain

(2.2) $2\geq\min\{\frac{(a+b-2)(a-b-1)}{2}$ , $4\}$

310 O. Saeki

and when $p=21$ ,

(2.9) $4\geq\min\{\frac{(a+b-2)(a-b-1)}{2}$ , $6\}$ .

Both inequalities imply $a-b=1$ . Thus when is odd we always have $p$

$a-b=1$ . Then the same argument as in [28] shows $r\equiv\pm 1(mod p)$ .
This completes the proof.

Remark. When is equal to 15 or 21, we cannot omit the special

$p$

condition that $k_{\in}(V)\leq 2$ and $k_{\in}(V)\leq 4$ respectively. For example, con-

sider $V=V(T(2,7);-15)$ . Then $k_{-}(V)\leq 3$ (cf. Example 2.7). However
$\partial V=L(15,4)$ admits a self-homeomorphism acting on $H_{1}(L(15,4))$ by

the multiplication $of\pm 4$ . When $p=21$ , consider $V=V(T(2,11);-21)$ .

Then $k_{-}(V)\leq 5$ . However $\partial V=L(21,4)$ bounds the smooth homotopy
, $V(T(4,5);-21)$ , which is not homeomorphic to $V$ .
$S^{2}$

\S 3. Topological invariants for homotopy $S^{2}$

and embedded
surfaces

Let $M$ be a homology $L(p, 1)$ $(p>0)$ ; i.e., $M$ is a closed oriented
3-manifold such that $H_{*}(M)\cong H_{*}(L(p, 1))$ and $1k(\alpha, \alpha)\equiv-\in/p(mod$
$Z)$ for some generator of $H_{1}(M)(\in=\pm 1)$ . We call such a pair
$\alpha$ $(M, \alpha)$

a marked homology $L(p, 1)$ . It is well-known that $M$ is obtained by $(\in p)-$

surgery on a knot $K$ in some homology 3-sphere so that the core of $\Sigma$

the surgery torus in $M$ represents . Define $\alpha$

(3.1) $\lambda_{0}(M, \alpha)=p\lambda(\Sigma)+(\in/2)\Delta_{K}(1)$ ,

where is the Casson invariant of the homology 3-sphere

$\lambda(\Sigma)$
and $\Sigma([6])$

is the normalized Alexander polynomial of . For a fixed marked

$\Delta_{K}(t)$ $K$

homology $L(p, 1)$ , and $K$ as above are not uniquely determined. How-
$\Sigma$

ever, by results of Boyer-Lines [4] and Fukuhara [15], is an $\lambda_{0}(M, \alpha)$

invariant of the marked homology $L(p, 1)$ , . We warn the reader $(M, \alpha)$

that here is times the invariant defined in [4] or [15] and that
$\lambda_{0}$
$p$
$\lambda_{0}$

here is integer valued. Note also that if $p=1$ , i.e. if $M$ is a homology 3-
sphere, (3.1) with replaced by is nothing but the surgery
$\lambda_{0}(M, \alpha)$ $\lambda(M)$

formula for the usual Casson invariant. Therefore, agrees with for $\lambda_{0}$
$\lambda$

homology 3-spheres.

Definition. Let $V$ be an oriented homotopy with fi- $S^{2}$

$H_{1}(\partial V)$

nite cyclic of order . Note that is a marked homology $L(p, 1)$ .

$p$ $(\partial V, \beta(V))$
 On 4-Manifolds Homotopy Equivalent to the 2-Sphere 311

Then we define , which we call the Casson invari-

$\overline{\lambda}(V)=\lambda_{0}(\partial V, \beta(V))$

ant of V. is a topological invariant of $V$ .

$\overline{\lambda}(V)$

Remark. If is a homology 3-sphere, then $\partial V$

. $\overline{\lambda}(V)=\lambda(\partial V)$

In general, however, is not an invariant of . For example, denot-

$\overline{\lambda}(V)$ $\partial V$

ing by the trivial knot in

$U$ , we have $\overline{\lambda}(V(U;5))=0$ and ; $S^{3}$ $\overline{\lambda}(V(T(2,3)$

-5))=-1, though $\partial V(U;5)\cong\partial V(T(2,3);-5)\cong L(5,1)$ $([25])$ .

The first result of this section is the following.

Theorem 3.1. Let $V$ be a smooth homotopy with fifi- $S^{2}$

$H_{1}(\partial V)$

nite. If the generator of $H_{2}(V)$ can be represented by a smoothly embed-

ded 2-sphere, then . $\overline{\lambda}(V)\equiv 0(mod 2)$

Remark. If $V$ is dieomorphic to a handlebody $V(K;p)$ for some

knot $K$ in , then agrees with the Arf invariant of $K$ . In
$S^{3}$
$\overline{\lambda}(V)mod 2$

this case, the above result has already been known ([2]).

Proof of Theorem 3.1. Suppose is a smoothly embedded 2-sphere $S$

in IntV which represents the generator of $H_{2}(V)$ . We denote by $N(S)$

the tubular neighborhood of . Then it is easily seen that the 4-manifold $S$

$Y=V-IntN(S)$ is a homology cobordism between and $L(p, 1)$ $\partial V$

$(\cong\partial N(S))$ . Furthermore, if $\alpha\in H_{1}(L(p, 1))$ is the homology class

corresponding to $\beta(V)\in H_{1}(\partial V)$ through the homology cobordism $Y$ ,
then is represented by the core of the surgery torus of $ M(U;p/1)(\cong$
$\alpha$

$L(p, 1))$ , where $U$ is the trivial knot in ; hence, $\alpha)=0$ . $S^{3}$
$\lambda_{0}(L(p, 1),$

Then Theorem 3.1 follows from the following Proposition 3.2.

Proposition 3.2. The Casson invariant modulo 2 for marked $\lambda_{0}$

homology $L(p, 1)$ is a homology cobordism invariant; , if $Y$ is $i.e.$ $a$

smooth homology cobordism between the homology $L(p, 1)$ s and $M_{0}$ $M_{1}$

and if and correspond through this homology

$\alpha_{0}\in H_{1}(M_{0})$ $\alpha_{1}\in H_{1}(M_{1})$

cobordism $Y$ , then . $\lambda_{0}(M_{0}, \alpha_{0})\equiv\lambda_{0}(M_{1}, \alpha_{1})(mod 2)$

Remember that, for homology 3-spheres, the Casson invariant mod-

ulo 2 is the Rohlin invariant, which is a homology cobordism invariant.

Proof of Proposition 3.2. We may assume $p\geq 2$ . Let $K_{i}(i=0,1)$

be a smooth knot in representing . Let $W$ be the 4- $M_{i}$ $\alpha_{i}\in H_{1}(M_{i})$

manifold obtained by attaching two 2-handles and to $Y$ along $h_{0}$ $h_{1}$ $K_{0}$

and in such a way that

$K_{1}$
consists of two disjoint homology 3- $\partial W$

spheres and . Denote by

$\Sigma_{0}$
the knot which is the boundary
$\Sigma_{1}$ $K_{i}\subset\Sigma_{i}$

of the cocore of the 2-handle (Figure 1). Then is obtained by $h_{i}$ $M_{i}$
312 O. Saeki

w :

$M_{0}$ $M_{1}$

Figure 1

the -surgery on the knot

$(\pm p)$ . Furthermore the core of this surgery $K_{i}$

torus in corresponds to
$M_{i}$
which represents $K_{i}$
. Hence $\alpha_{i}\in H_{1}(M_{i})$

. Let $f_{i}\in H_{2}(W, \partial W)(\cong H_{2}(W))$ be

$\lambda_{0}(M_{i}, \alpha_{i})=p\lambda(\Sigma_{i})\pm(1/2)\Delta_{K_{i}}(1)$

the homology class represented by the cocore of the 2-handle . Since $h_{i}$

$Y$ is a homology cobordism, there is a 2-chain in $Y$ with boundary $c$ $K_{0}$

and . Denote by $e\in H_{2}(W)$ the homology class represented by the

$K_{1}$

2-cycle which consists of the cores of and the 2-chain . Then $h_{i}$ $c$ $(f_{i}, e)$

are generators of $H_{2}(W, \partial W)\cong H_{2}(W)\cong Z\oplus Z$ . Furthermore we have

, $f_{0}\cdot f_{1}=0$ , . $e=\pm 1and\pm pe=\pm f_{0}\pm f_{1}$ . Hence,
$f_{i}\cdot f_{i}=\pm p$ $f_{i}$

$e$ . $e=\frac{1}{p^{2}}(\pm f_{0}\pm f_{1})\cdot(\pm f_{0}\pm f_{1})$

$=\frac{1}{p^{2}}(\pm p\pm p)$ .

Since this must be an integer, we have

(A) $e$ . $e=0$ or
$\{$

(B) $e$ . $e=\pm 1$ and $p=2$ .

Case (A). The homology class $f_{0}+f_{1}\in H_{2}(W, \partial W)\cong H_{2}(W)$ is
characteristic and we can represent it by a smoothly embedded annulus
$A$ by
tubing the cocores of and . $h_{0}$ $h_{1}$

When is even, $p$ $W$ is spin and its signature is zero; hence, $\lambda(\Sigma_{0})\equiv$

$\lambda(\Sigma_{1})(mod 2)$ . Thus, for the proof of Proposition 3.2, it suces to

prove

(3.2) $\lambda(\Sigma_{0})+\frac{1}{2}\Delta_{K_{\acute{O}}}(1)\equiv\lambda(\Sigma_{1})+\frac{1}{2}\Delta_{K_{1}}(1)$
$(mod 2)$ .
 On 4-Manifolds Homotopy Equivalent to the 2-Sphere 313

Note that is equal to the Casson invariant of the

$\lambda(\Sigma_{i})+(1/2)\Delta_{K_{i}}(1)$

homology 3-sphere obtained by $(+1)$ -surgery on in

$\Sigma_{i}$
. Let $X$ $K_{i}$ $\Sigma_{i}$

be the 4-manifold obtained by attaching two 2-handles and to $W$ $h_{0}$ $h_{1}$

along $K_{i}$
with the $(+1)$ -framing. Note that consists of the homology $\partial X$

3-spheres and . Denote by

$\Sigma_{0}$
the smoothly embedded 2-sphere in
$\Sigma_{1}$ $S$

$X$ which consists of the annulus $A$ and the cores of and . Note that $h_{0}$ $h_{1}$

$[S]\in H_{2}(X)$ is characteristic and that is equal to the signature of $[S]^{2}$

$X$ . Then using the same method as in [21], one can deduce $\lambda(\Sigma_{0})\equiv$

$\lambda(\Sigma_{1})(mod 2)$ . This shows the equality (3.2) holds.

Case (B). The homology classes , $f_{1}\in H_{2}(W, \partial W)\cong H_{2}(W)$ are $f_{0}$

both characteristic. Let be the 4-manifold obtained by attaching a $X_{i}$

2-handle to $W$ along

$h_{i}$
with the $(+1)$ -framing. Denote by the
$K_{i}$ $S_{i}$

smoothly embedded 2-sphere in consisting of the cocore of the 2- $X_{i}$

handle and the core of the 2-handle . Note that $[S_{i}]\in H_{2}(X_{i})$ is
$h_{i}$ $h_{i}$

characteristic and that is equal to the signature of . Thus, by

$[S_{i}]^{2}$ $X_{i}$

the same argument as in Case (A), we have

$\lambda(\Sigma_{0})\equiv\lambda(\Sigma_{1})$
$(mod 2)$ and
$\lambda(\Sigma_{0})\equiv\lambda(\Sigma_{1})$
$(mod 2)$ .

Hence,

(3.3) $\lambda(\Sigma_{0})+\frac{1}{2}\Delta_{K_{\acute{O}}}(1)\equiv\lambda(\Sigma_{1})$
$(mod 2)$ and

(3.4) $\lambda(\Sigma_{0})\equiv\lambda(\Sigma_{1})+\frac{1}{2}\Delta_{K_{1}}^{JJ}(1)$
$(mod 2)$ .

Adding (3.3) and (3.4) shows

$2\lambda(\Sigma_{0})+\frac{1}{2}\Delta_{K_{\acute{O}}}(1)\equiv 2\lambda(\Sigma_{1})+\frac{1}{2}\Delta_{K_{1}}^{JJ}(1)$
$(mod 2)$ .

This completes the proof of Proposition 3.2 and hence Theorem 3.1.

Remark. There is a smooth homotopy , $V$ , such that $S^{2}$ $\overline{\lambda}(V)\equiv 0$

$(mod 2)$ and yet the generator of $H_{2}(V)$ cannot be represented by a

smoothly embedded 2-sphere. For example, consider $V=V(T(3, -11)$ ;
34). A computation shows . However the generator of $H_{2}(V)$ $\overline{\lambda}(V)=40$

cannot be represented by a smoothly embedded 2-sphere (see [28] or

Example 3.5 below).

Remark. Let be an oriented marked homology $L(p, 1)$ . If

$(M, \alpha)$

$p$ is odd, the Rohlin invariant $\mu(M)(\in Z/16Z)$ of $M$ is defined and it
314 O. Saeki

is a homology cobordism invariant. It can be shown, using the method

similar to that in the proof of Proposition 3.2, that

$8\lambda_{0}(M, \alpha)\equiv\mu(M)-(p-1)p$ . $1k(\alpha, \alpha)$

$(mod 16)$ ,

where $1k(\alpha, \alpha)=\pm 1/p$ .

Next we define the Casson-Gordon invariants for a homotopy . $S^{2}$

In our case, they are essentially the -signatures of a certain knot in a $p$

homology 3-sphere.
Let $K$ be a smooth knot in an oriented homology 3-sphere and $\Sigma$

let be a Seifert matrix of . For a positive integer , set

$L$ $K$ $p$ $\omega_{p}=$

$\exp(2\pi\sqrt{-1}/p)$ . Then we define $\sigma_{K}(\omega_{p}^{r})(r=1,2, \cdots, p-1)$ to be the

signature of the Hermitian matrix $(1-\omega_{p}^{-r})L+(1-\omega_{p}^{r})L^{T}$ , where $L^{T}$

is the transpose of . It is well-known that are invariants of $K$

$L$ $\sigma_{K}(\omega_{p}^{r})$

and they are called -signatures of $K$ . Note that they are independent $p$

of the orientation of $K$ , while they depend on the orientation of in $\Sigma$

general. If we change the orientation of $K$ , its Seifert matrix becomes

and the signature of the corresponding Hermitian matrix does not
$L^{T}$

change.

Lemma 3.3. Let $\Delta_{i}(i=0,1)$ be a compact oriented contractible

topological -manifold and let be a tame knot in the homology 3-sphere
$A$ $K_{i}$

. Let be the oriented -manifold obtained by attaching a 2-handle

$\partial\Delta_{i}$ $V_{i}$ $A$

to along with the

$\Delta_{i}$
-framing $(p\geq 0)$ . If is orientation
$K_{i}$ $(\pm p)$ $V_{0}$

preservingly homeomorphic to , then when $p>0$ , $V_{1}$

$\sigma_{K_{0}}(\omega_{p}^{r})=\sigma_{K_{1}}(\omega_{p}^{r})$
for $1\leq r\leq p-1$

and when $p=0$ ,

$\sigma_{K_{O}}(\omega_{q}^{s})=\sigma_{K_{1}}(\omega_{q}^{s})$
for every $q>0$ and $1\leq s\leq q-1$ .

We prove the case of $p>0$ . When $p=0$ , the proof is

Proof
similar. We may assume are positive definite. Orient arbitrarily $V_{i}$ $K_{i}$

and let be the homology class represented by a meridian

$\alpha_{i}\in H_{1}(\partial V_{i})$

loop of . Let : be a homeomorphism. By the restriction

$K_{i}$ $h$ $V_{0}\rightarrow V_{1}$

, we identify
$h|\partial V_{0}$
and and denote it by $M$ . We may assume $\partial V_{0}$ $\partial V_{1}$

that, by this identification, in $H_{1}(M)$ , changing orientations $\alpha_{0}=\alpha_{1}$

of if necessary. Define the homomorphism

$K_{i}$
: $H_{1}(M)\rightarrow Z/pZ$ by $\varphi$

( $1\in Z/pZ$ is the generator). Let

$\varphi(\alpha_{i})=1$ be a Seifert matrix for $L_{i}$
 On 4-Manifolds Homotopy Equivalent to the 2-Sphere 315

$K_{i}$
. Then by [7, Lemma 3.1],

(3.5) $\sigma_{r}(M, \varphi)=signV_{i}-$ sign((l $-\omega_{p}^{-r}$ ) $L_{i}+(1-\omega_{p}^{r})L_{i}^{T}$ )

$-\frac{2r(p-r)}{p}$ $(i=0,1)$ ,

where is the Casson-Gordon invariant of $M$ associated with

$\sigma_{r}(M, \varphi)$ $\varphi$

and . Note that in [7] everything is assumed to be smooth. However,

$r$

their method is easily extended to the topological category, since the

$G$ -signature
theorem holds also in the topological case (see [32]). Since
signVb $=signVb$ , we have by (3.5) $\sigma_{K_{0}}(\omega_{p}^{r})=\sigma_{K_{1}}(\omega_{p}^{r})$

Definition. Let $V$ be an oriented homotopy with iso- $S^{2}$

$H_{1}(\partial V)$

morphic to $Z/pZ$ $(p\geq 0)$ . Then by [3], $V$ is obtained by attaching a

2-handle to a compact contractible 4-manifold along some tame knot $\Delta$

$K$ in the homology 3-sphere with the -framing. If $p>0$ , we $\partial\Delta$

$(\pm p)$

define
$(1\leq r\leq p-1)$ ,
$\sigma_{V}(\omega_{p}^{r})=\sigma_{K}(\omega_{p}^{r})$

which we call the -signatures of $p$ $V$ . Similarly if $p=0$ , we define, for
every $q>1$ ,
$\sigma_{V}(\omega_{q}^{s})=\sigma_{K}(\omega_{q}^{s})$ $(1\leq s\leq q-1)$ .

By Lemma 3.3, this is well-defined.

As the equation (3.5) in the proof of Lemma 3.3 shows, the p-
signatures of a homotopy are essentially the Casson-Gordon invari- $S^{2}$

ants of the boundary 3-manifold.

Next we use these invariants to attack problem (B) in \S 1.
Definition. Let $V$ be a homotopy . We define $g(V)$ to be the $S^{2}$

minimal genus of topologically locally flatly embedded surfaces repre-

senting the generator of $H_{2}(V)$ .
Remark. Even if $V$ itself is not smooth, IntF admits a smooth
structure (see [14, 8.2]). Thus the generator of $H_{2}(V)$ ( (IntV)) is $\cong H_{2}$

always represented by a locally flatly embedded surface.

Theorem 3.4. Let $V$ be a homotopy with fifinite cyclic $S^{2}$
$H_{1}(\partial V)$

of order $p(p>0)$ . Then for every prime power dividing , we have $d$
$p$

$2g(V)\geq|\sigma_{V}(\omega_{d}^{s})|$ $(s=1,2, \cdots, d-1)$ .

Remark. This lower bound has already been known for the case
that $V$ consists of one 0-handle and one 2-handle and the embedded
316 O. Saeki

surfaces considered are smooth ([31]). In this case, when is $H_{1}(\partial V)$

infinite cyclic, the above inequality holds for all prime power . We do $d$

not know whether Theorem 3.4 also holds for $V$ with infinite $H_{1}(\partial V)$

cyclic.

Theorem 3.4. We may assume $V$ is positive definite. Let

Proof of
$F$ be a topologically locally flatly embedded surface in IntV of genus

$g=g(V)$ representing the generator of $H_{2}(V)$ . There exists a -fold $d$

cyclic branched covering : branched along $F$ such that the $\pi$
$\overline{V}\rightarrow V$

-fold covering
$d$
: corresponds to the homomorphism
$\pi|\partial\overline{V}$
$\partial\overline{V}\rightarrow\partial V$

: $H_{1}(\partial V)\rightarrow Z/dZ$ defined by $\varphi(\beta(V))=1$ . Let

$\varphi$
: be $\tau$
$\overline{V}\rightarrow\overline{V}$

the canonical covering translation. Define to be the $E_{s}\subset H_{2}(\overline{V})\otimes C$

-eigenspace of
$\omega_{d}^{s}$
: $\tau_{*}$
(note that ).
$H_{2}(\overline{V})\otimes C\rightarrow H_{2}(\overline{V})\otimes C$ $\tau_{*}^{d}=id$

Furthermore define to be the signature of the restriction to

$\in_{s}(\overline{V})$
of $E_{s}$

the intersection pairing on . Then by the definition of the $H_{2}(\overline{V})\otimes C$

Casson-Gordon invariants [7],

$\sigma_{s}(\partial V, \varphi)=signV-\in_{s}(\overline{V})-\frac{2ps(d-s)}{d^{2}}$
.

Combining this with [7, Lemma 3.1] and the definition of -signatures, $p$

we have
$\in_{s}(\overline{V})=\sigma_{V}(\omega_{d}^{s})$
.

Since , it suces to show that $dim_{C}E_{S}=2g$ .

$|\in_{s}(\overline{V})|\leq dim_{C}E_{s}$

It is easily verified, using a method similar to that in [19, \S 4], that

$dim_{C}H_{2}(\overline{V})\otimes C=2g(d-1)+1$

(note that $d$

is a prime power by the assumption). It is well-known that
$E_{0}=\pi^{*}(H_{2}(V)\otimes C)$ . Hence,

$\sum_{s=1}^{d-1}dim_{C}E_{s}=2g(d-1)$ .

Using this equation and the linear algebra together with the assumption
that is a prime power, we easily deduce $dim_{C}E_{s}=2g(s=1,2,$
$d$
, $d-$ $\cdots$

. This completes the proof.

$1)$

Example 3.5. Consider $V=V(T(a, b);p)$ , where $|a|$ , $|b|\geq 2$ and

$p\neq 0,$ . Then the generator of $H_{2}(V)$ cannot be represented by a
$\pm 1$

topologically locally flatly embedded 2-sphere since the -signatures of $p$

the torus knot $T(a, b)$ do not vanish. Remember that if is even, the $p$
 On 4-Manifolds Homotopy Equivalent to the 2-Sphere 317

-signature $\sigma_{T(a,b)}(-1)$ is the usual signature of the torus knot $T(a, b)$ .
$p$

Thus the lower bound for the 4-ball genera of torus knots given in [29]
is also valid for $g(V(T(a, b);p))$ if is even. $p$

Definition. For a homotopy , $V$ , we denote by $ KS(V)(\in$ $S^{2}$

$Z/2Z)$ the Kirby-Siebenmann obstruction to extending the product

smooth structure on across $V\times R$ . $\partial V\times R$

If is a homology 3-sphere, Theorem 3.4 gives no restrictions on

$\partial V$

$g(V)$ . In that case, we have the following.

Proposition 3.6. Let $V$ be a homotopy with a homology $S^{2}$ $\partial V$

3-sphere. Then $g(V)=0$ if $\mu(\partial V)=KS(V)$ and $g(V)=1$ if $\mu(\partial V)\neq$

$KS(V)$ , where is the Rohlin invariant of

$\mu(\partial V)$
. $\partial V$

Proof. Let $P=CP^{2}-IntD^{4}$and $Q=Ch-IntD^{4}$ , where $Ch$

is the Chern manifold ([12]). Furthermore let be the contractible 4- $\Delta$

manifold bounded by $\partial V([12])$ . Then by [3] is homeomorphic to $V$ $ P\#\Delta$

if $\mu(\partial V)=KS(V)$ and if $\mu(\partial V)\neq KS(V)$ , where denotes the

$ Q\#\Delta$ $\#$

boundary connected sum. Since $P$ is homeomorphic to a bundle over $D^{2}-$

, the generator of
$S^{2}$
$H_{2}(P\#\Delta)$ is represented by a locally flatly embedded
2-sphere. Furthermore, $Ch$ is homeomorphic to $V(T(2,3);1)\cup\Delta$ , where
is the contractible 4-manifold bounded by $M(T(2,3);1/1)$ . Thus the
$\Delta$

generator of $H_{2}(Q\#\Delta)$ is represented by a locally flatly embedded torus.

Hence, $g(V)=0$ if $\mu(\partial V)=KS(V)$ and $g(V)\leq 1$ if $\mu(\partial V)\neq KS(V)$ .
Next we show that $g(V)\neq 0$ if $\mu(\partial V)\neq KS(V)$ . Suppose $g(V)=0$
and let be a locally flatly embedded 2-sphere in IntV representing
$S$

the generator of $H_{2}(V)$ . By [13], has a neighborhood $N(\subset IntV)$ $S$

which is a 2-disk bundle over . Note that is homeomorphic to

$S$ $\partial N$

. Set
$S^{3}$
$\Delta=(V-IntN)\bigcup_{\partial N}D^{4}$ . Then is equal to the Kirby- $\mu(\partial V)$

Siebenmann obstruction of , which in turn is equal to $KS(V)$ . This

$\Delta$

contradicts the assumption that $\mu(\partial V)\neq KS(V)$ . This completes the
proof.

In Theorem 3.4, only the -signatures of the form with $p$ $\sigma_{V}(\omega_{d}^{s})$ $d$

a prime power are handled. For the general -signatures, we have the $p$

following.

Proposition 3.7. Let $V$ be a smooth homotopy with $S^{2}$

$H_{1}(\partial V)$

fifinite cyclic of order . For an integer $d(>0)$ dividing , suppose

$p$ $p$

, where
$H_{1}(\overline{\partial V};Q)=0$
is the -fold cyclic covering associated
$\overline{\partial V}\rightarrow\partial V$
$d$

with the homomorphism : defifined by $\varphi(\beta(V))=1$ . If

$\varphi$
$H_{1}(\partial V)\rightarrow Z/dZ$
318 O. Saeki

the generator of $H_{2}(V)$ is represented by a smoothly embedded 2-sphere,

then
$\sigma_{V}(\omega_{d}^{s})=0$
for $1\leq s\leq d-1$ .

Remark. Suppose $V$ is obtained (topologically) by attaching a 2-

handle to a contractible 4-manifold along a knot $K$ in the homology $\Delta$

3-sphere . Then if and only if

$\partial\Delta$
for $H_{1}(\overline{\partial V};Q)=0$ $\Delta_{K}(\omega_{d}^{s})\neq 0$

$1\leq s\leq d-1$ , where is the Alexander polynomial of . In$K$ $\Delta_{K}(t)$

particular, if is a prime power, we always have $d$

. $H_{1}(\overline{\partial V};Q)=0$

Proof of Proposition 3.7.

Let be a smoothly embedded 2-sphere $S$

in Inty representing the generator of $H_{2}(V)$ and let $Y=V-IntN(S)$ ,

where $N(S)$ is the tubular neighborhood of . Then $Y$ is a smooth $S$

homology cobordism between and $L(p, 1)$ . Let : $ H_{1}(L(p, 1))\rightarrow$ $\partial V$ $\varphi$

$Z/dZ$ be the homomorphism defined by composing the isomorphism

$H_{1}(L(p, 1))$ induced by $Y$ and the homomorphism
$\rightarrow$ $H_{1}(\partial V)$

:
$\varphi$
. Then by Matic [24] and Ruberman[27],
$H_{1}(\partial V)\rightarrow Z/dZ$

$\sigma_{1}(\partial V, \varphi s)=\sigma_{1}(L(p, 1),$ $\varphi s)$

$(1\leq s\leq d-1)$ .

Since $\sigma_{s}(\partial V, \varphi)=\sigma_{1}(\partial V, \varphi s)$

and , , we
$\sigma_{s}(L(p, 1),$ $\varphi)=\sigma_{1}(L(p, 1)$ $\varphi^{s})\prime$

have $\sigma_{s}(\partial V, \varphi)=\sigma_{s}(L(p, 1),$ $\varphi)$

. Combining this with [7, Lemma 3.1],
we obtain
$\sigma_{V}(\omega_{d}^{s})=0$ $(1\leq s\leq d-1)$

(see also the proof of Theorem 3.1 in this section). This completes the
proof.

In [4], Boyer-Lines extended the Casson invariant for homology 3-

spheres to homology lens spaces. Furthermore, they gave a relation
of this invariant to the Casson-Gordon invariants. Using this result,
Theorem 3.1, and Proposition 3.7, we have the following.

Proposition 3.8. Let $V$ be $an\in- defifinite$ $(\in=\pm 1)$ smooth homo-
topy satisfying the condition on $H_{1}(\partial V;Q)$ as in Proposition 3.7 for
$S^{2}$

$d=p$ . If the generator of $H_{2}(V)$ is represented by a smoothly embedded

2-sphere, then

$12p\lambda(\partial V)\equiv-\in\frac{(p-1)(p-2)}{2}$
$(mod 24)$ ,

where $\lambda(\partial V)$

is the Boyer-Lines Casson invariant for the homology lens
space $\partial V$
.
 On 4-Manifolds Homotopy Equivalent to the 2-Sphere 319

Remark. $12p\lambda(\partial V)$

is always an integer ([4, Theorem 2.8]).
Suppose $V$ is obtained (topologically) by
Proof of Proposition 3.8.
attaching a 2-handle to a contractible 4-manifold along a knot $K$ in $\Delta$ $\partial\Delta$

with -framing. Let

$(\in p)$
be the homology class represented
$\alpha\in H_{1}(\partial V)$

by a meridian of . Then by [4, Proposition 2.23],

$K$

$p\lambda(\partial V)=\lambda_{0}(\partial V, \alpha)+\frac{1}{8}\sigma(K,p)+\frac{1}{8}\tau(\partial V)$

,

where
$\sigma(K, p)=\sum_{r=1}^{p-1}\sigma_{K}(\omega_{p}^{r})$

and
$\tau(\partial V)=\sum_{r=1}^{p-1}\sigma_{r}(\partial V, \varphi)$
.

(Here, : $H_{1}(\partial V)\rightarrow Z/pZ$ is the homomorphism defined by $\varphi(\beta(V))=$

$\varphi$

By Theorem 3.1, we have

$1.)$ and by Proposition
$\lambda_{0}(\partial V, \alpha)\equiv 0(mod 2)$

3.7, $\sigma(K,p)=0$ . Furthermore, by the proof of Proposition 3.7, we have

; hence,
$\sigma_{r}(\partial V, \varphi)=\sigma_{r}(L(p, \in),$ $\varphi)$

$\tau(\partial V)=\tau(L(p, \in))=-\in\frac{(p-1)(p-2)}{3}$ .

Thus,
$8p\lambda(\partial V)\equiv-\in\frac{(p-1)(p-2)}{3}$
$(mod 16)$ .

Multiplying 3/2 gives the result.

\S 4. Knots in the boundary of a homotopy $S^{2}$

Definition. Let $V$ be a homotopy and let $K$ be a tame knot $S^{2}$

in representing
$\partial V$
$\beta(V)\in H_{1}(\partial V)$ . We say $K$ is a boundary knot if
there is a topologically embedded proper flat 2-disk $D$ in $V$ such that
$\partial D=K$ and $D$ represents the generator of . Here, a properly $H_{2}(V, \partial V)$

embedded 2-disk $D$ is flat if it is the core of an embedded open 2-handle

$D\times R^{2}$
in $V$ with $(D\times R^{2})\cap\partial V=\partial D\times R^{2}$ .
Definition. be a tame knot in a homology $L(p, 1)$ , $M$ , with
Let $K$

$1k([K], [K])=\pm 1/p(p>0,p\neq 2)$ . Let be the homology 3-sphere $\Sigma$

obtained by a surgery on $K$ such that the coecient of the corresponding

surgery by which $M$ is obtained from is an integer. Note that $\Sigma$ $\Sigma$
320 O. Saeki

depends only on $M$ and $K$ if $p\geq 3$ . Then we denote by $\mu(K)(\in Z/2Z)$
the Rohlin invariant of . Note that, when $p=1$ , $\mu(K)$ is still
$\mu(\Sigma)$ $\Sigma$

well-defined, though is not. When $p=2$ , $\mu(K)$ cannot be defined.

$\Sigma$

Our first result of this section is the following.

Proposition 4.1. Let $V$ be a homotopy with fifinite $S^{2}$

$H_{1}(\partial V)$

cyclic of order . Suppose $K$ is a tame knot in

$p$ representing $\beta(V)\in$ $\partial V$

. Then we have the following.

$H_{1}(\partial V)$

(1) When is even, $K$ is always a boundary knot.

$p$

(2) When is odd, $K$ is a boundary knot if and only if $\mu(K)=KS(V)$ .

$p$

Proof. First, we construct an embedded flat 2-disk bounded by $K$ .

Set $M=\partial V$ and let $X$ be the 4-manifold obtained by attaching a 2-
handle to $M\times[0,1]$ along $K\times\{1\}$ in such a way that
$h_{0}$
consists $\partial X$

of $M$ and a homology 3-sphere . By [12], there exists a contractible $\Sigma$

4-manifold with . Denote by $V$ the 4-manifold

$\Delta$ $\partial\Delta=\Sigma$
. Note $ X\cup\Sigma\Delta$

that $V$
is a homotopy with $\partial V=M$ . By [3] and the hypothesis on
$S^{2}$

$\mu(K)=\mu(\Sigma)$ , we see that there is a homeomorphism : such $h$ $V\rightarrow V$

that $h|\partial V=id_{M}$ . Since $h(K)$ bounds in $V$

an embedded flat 2-disk
(the core of the 2-handle ), $K$ also bounds one in $V$ . $h_{0}$

Conversely, suppose is odd and $K$ bounds in $V$ a topologically $p$

embedded proper flat 2-disk $D$ in $V$ which represents the generator
of . Then there exists a closed neighborhood $N$ of $D$ in $V$
$H_{2}(V, \partial V)$

homeomorphic to . Denote by $B$ the closure of $V-N$ . Note

$D^{2}\times D^{2}$

that is a homology 3-sphere and that $\mu(K)=\mu(\partial B)$ . Then it

$\partial B$

is easily shown that $B$ is a homology 4-ball. Thus $V$ is obtained by

attaching a 2-handle $N$ to $B$ . By [3], $KS(V)$ is equal to $\mu(\partial B)=\mu(K)$ .
This completes the proof.

Remark. There exists a knot in the boundary of a homotopy , $S^{2}$

$V$ , which is not a boundary knot but bounds in $V$ an embedded

flat
2-disk. For example, consider the knot $K$ in $\partial V(U;p)$ as in Figure 2,
where is the trivial knot and is odd. Then it is easily seen $\mu(K)=1$ .
$U$ $p$

(To see this, observe that $\mu(K)$ is equal to the Rohlin invariant of the
homology 3-sphere obtained by the surgery along the framed link in as $S^{3}$

in Figure 3. Then we can apply a formula of Kaplan [20, Theorem 4.2].)

Since the Kirby-Siebenmann obstruction of $V(U;p)(=V)$ vanishes, $K$
is not a boundary knot by Proposition 4.1. However, $K$ bounds in $V$ a
smoothly embedded 2-disk. This can be constructed as follows. There
is a smoothly immersed 2-disk $D$ in with one self-intersection such $D^{4}$

that $\partial D=K$ . $D$ intersects in $V$ the smoothly embedded 2-sphere
$S$
(the union of the core of the 2-handle and the cone over $U$ in ) $D^{4}$
 On 4-Manifolds Homotopy Equivalent to the 2-Sphere 321

$U$

$KC3V(U_{i}p)$

Figure 2

$p$

Figure 3

transversely in one point. Piping and $D$ along an arc on $D$ which
$S$

connects the self-intersection point of $D$ with the intersection point of

$D$ and as in Figure 4, we obtain the desired embedded 2-disk $D$ . Of
$S$

course, $D$ does not represent the generator of . $H_{2}(V, \partial V)$

Next we give an example of knots in the boundary of a smooth homo-

topy which are boundary knots but never bound smoothly embedded
$S^{2}$

2-disks.

Definition. Let $K$ be a smooth knot in the boundary of a smooth

homotopy , $V$ . Then we denote by $k(K)$ the minimal number of
$S^{2}$

double points of smooth properly immersed self-transverse 2-disks in $V$

bounded by $K$ . Following [17], we call $k(K)$ the kinkiness of $K$ .
322 O. Saeki

$v$

$\downarrow$

$v$

Figure 4

Let $n$ be odd integers with ,

and $l$

and set $p=l(nl+2)$ $n$

$l$
$\geq 3$

and $q=nl$ $+1$ . Note that and are relatively prime integers.
$p$ $q$

Set $V=V(T(p, q);-pq)$ . Note that by [25], is dieomorphic to $\partial V$

$L(p, q)\# L(q,p)$ . Set $r=q^{2}-q-1$ and let , , , be the $m_{1}$ $m_{2}$ $\cdots$
$m_{r}$

$r$
oriented knots in represented by the meridians of
$\partial V$
$T(p, q)$ (see
Figure 5). Here we make the orientation convention $[m_{i}]=-\beta(V)$
in . Let $K$ be any knot in
$H_{1}(\partial V)$ obtained by performing the $\partial V$

oriented band connected sum operations to $m_{1}\cup m_{2}\cup\cdots\cup m_{r}$ be-

tween distinct components $(r-1)$ times. Note that $[K]=-r\beta(V)$
in . Since $q^{2}\equiv 1$ $(mod p)$ , there is a dieomorphism
$H_{1}(\partial V)$ : $h$

$L(p, q)\rightarrow L(p, q)$ which acts on $H_{1}(L(p, q))$ by the multiplication of

. Set $h=h\#id$ : $L(p, q)\# L(q,p)\rightarrow L(p, q)\# L(q,p)$ . It is easily

$q$

seen that acts on$h$

by the multiplication of $-r$ (note that
$H_{1}(\partial V)$

$H_{1}(\partial V)\cong H_{1}(L(p, q)\# L(q,p))\cong Z/pZ\oplus Z/qZ\cong Z/pqZ)$ . Note that
$r^{2}\equiv 1(mod pq)$ . Then we denote by $K$ the smooth knot $h(K)$ in . $\partial V$
 On 4-Manifolds Homotopy Equivalent to the 2-Sphere 323

Note that $K$ represents $\beta(V)\in H_{1}(\partial V)$ .

$-pq$

$m_{1}$

$m_{2}$

.
$.\cdot$

$mr$

Figure 5

Proposition 4.2. The knot $K$ bounds in $V$ a topologically embed-

ded proper flat 2-disk. However,

$k(K)\geq[\frac{n(nl-1)+2}{4}]$ .

In particular, $K$ cannot bound in $V$ any smoothly embedded 2-disks.

Proof. Since $pq$ is a boundary knot by Proposition 4.1.

is even, $K$

Now set \partial V_{1}\rightarrow\partial V_{0})$ , which is

$X=V_{0}\bigcup_{h}(-V_{1})(V_{0}=V_{1}=V, h :

a smooth closed 1-connected 4-manifold with $H_{2}(X)\cong Z\oplus Z$ . Let $D$

be a smoothly immersed self-transverse 2-disk in $V_{0}(\cong V)$ bounded by
$K$ with $k(K)$ double points. Let be the topologically embedded (not
$S_{i}$

locally flat) 2-sphere in which consists of the core of the 2-handle of

$V_{i}$

$V_{i}=V(T(p, q);-pq)$ and the cone over $T(p, q)$ in . Furthermore let $D^{4}$ $S_{2}$

be the smoothly immersed 2-sphere in $X$ which consists of the cocores $r$

of the 2-handle in corresponding to

$V_{1}$ $m_{1}\cup m_{2}\cup\cdots\cup m_{r}$ , the $(r-1)$

bands used to make $K$ , and the immersed 2-disk $D$ . Note that is a $S_{2}$

smoothly immersed 2-sphere with $k(K)$ double points. Furthermore let

$\tau\in H_{2}(X)$ be the homology class represented by the union of the cocore

of the 2-handle of , the cocores of the 2-handle of

$V_{0}$ $r$
and a 2-chain $V_{1}$

in connecting their boundaries (note that $h_{*}(r\beta(V_{1}))=$

$\partial V_{0}=h(\partial V_{1})$

$-\beta(V_{0}))$ . Then by [28], $\theta=[S_{1}]$ and generate $H_{2}(X)$ . Furthermore $\tau$

$[S_{2}]=\tau+j[S_{0}]$ for some integer . $j$

324 O. Saeki

It is easily seen that the intersection matrix of $X$ with respect to

the basis and is
$\theta$
$\tau$

$Q=\left(\begin{array}{lllll}-l(nl+ & 1)(nl+ & 2) & n^{2}l^{2}+nl- & 1\\n^{2}l^{2}+nl- & 1 & & -n(nl-1) & \end{array}\right)$
.

( . $\tau=-n(nl-1)$ is the consequence of the fact that $\det Q=\pm 1$ and

$\tau$

that $|l(nl+1)(nl+2)|>2.)$ Furthermore $Q$ is isomorphic to the form

$\left(\begin{array}{ll}0 & 1\\1 & 0\end{array}\right)$

.

Thus there are generators and of $H_{2}(X)$ with $\xi$

$\eta$
$\xi$ . $\xi=\eta\cdot\eta=0$ and
. $\eta=1$ . Furthermore, we may assume
$\xi$

$\theta=\frac{l(nl+1)}{2}\xi-(nl+2)\eta$ and

$\tau=\frac{1-nl}{2}\xi+n\eta$ .

Then we have
$[S_{0}]=\frac{l(nl+1)}{2}\xi+(nl+2)\eta$ .

Thus
$[S_{2}]=\tau+j[S_{0}]$

$=\{\frac{1-nl}{2}+\frac{l(nl+1)}{2}j\}\xi+\{n+(nl+2)j\}\eta$ .

Set

$\alpha=\frac{1-nl}{2}+\frac{l(nl+1)}{2}j$
and
$\beta=n+(nl+2)j$ .

By Theorem 2.8, the number of double points of $S_{2}(=k(K))$ is greater

than or equal to

$\min\{(|\alpha|-1)(|\beta|-1)$ , $[\frac{|\alpha\beta|+1}{2}]\}$
.

Since both $|\alpha|$

and $|\beta|$
attains its minimum when $j=0$ , we have

$k(K)\geq\min\{(\frac{nl-1}{2}-1)(n-1)$ , $[\frac{n(nl-1)+2}{4}]\}$

$=[\frac{n(nl-1)+2}{4}]$ .
 On 4-Manifolds Homotopy Equivalent to the 2-Sphere 325

This completes the proof.

Remark. The smooth knot $K=h^{-1}(K)$ does bound in $V$ a

smoothly embedded 2-disk. Thus : does not extend to $h$ $\partial V\rightarrow\partial V$

a self-dieomorphism of . In fact, it is easily seen that

$V$ does not $h$

extend even to a self-homeomorphism of $V$ .

We end this section by posing a problem.

Problem. Is there a smooth homotopy , $V$ , such that some $S^{2}$

$l$

cannot be represented by any knot which is the boundary of

$\in\pi_{1}(\partial V)$

a smoothly embedded 2-disk in $V$ ?

admits a handlebody decomposition without 3-handles, every

If $V$

$l$

is so represented ([11]). Thus if the problem above is af-

$\in\pi_{1}(\partial V)$

firmative, the homotopy needs a 3-handle in any of its handlebody

$S^{2}$

decomposition. Note also that if represents $\pm\beta(V)$ in , then, $l$

$H_{1}(\partial V)$

by Proposition 4.1, can be represented by a knot which bounds a topo-

$l$

logically embedded flat 2-disk in $V$ .

\S 5. Exotic open homotopy $S^{2}$

For an integer , let $D(p)$ denote the

$p$
$D^{2}$
bundle over with euler $S^{2}$

number . $D(p)$ is a (compact) homotopy

$p$
$S^{2}$
. Our result of this section
is the following.

Proposition 5.1. Let and $m$ be relatively prime odd integers $l$

greater than 2. Then the open -manifold IntD $(41m)$ admits at least 2 $A$

smooth structures other than the canonical one.

To prove Proposition 5.1, we need the following.

Lemma 5.2. and be relatively prime integers and set

Let $a$
$b$ $\zeta=$

$a\xi+b\eta\in H_{2}(S^{2}\times S^{2})$ , where $\xi=\{*\}\times[S^{2}]$ and $\eta=[S^{2}]\times\{*\}$

are the standard generators. Then can be represented by a smoothly $\zeta$

immersed 2-sphere in with simply connected complement and

$S^{2}\times S^{2}$

with $(|a|-1)(|b|-1)$ double points.

We may assume $a\geq 0$ and $b\geq 0$ . If $(a, b)=(1,0)$ or $(0, 1)$ ,

Proof.
the assertion is trivial. Hence, we may assume $a\geq 1$ and $b\geq 1$ . We
construct a desired immersed 2-sphere by the standard method. Take
distinct $(a+1)$ points , , , $x_{a}\in S^{2}$ and distinct $(b+1)$ points
$x_{0}$ $x_{1}$ $\cdots$

, ,
$y_{0}$ , $y_{b}\in S^{2}$ . Set
$y_{1}$ $\cdots$
and
$R=(\bigcup_{i=1}^{a}\{x_{i}\}\times S^{2})\cup(\bigcup_{j=1}^{b}S^{2}\times\{y_{j}\})$

$X=S^{2}\times S^{2}-R$ . Here, we orient and so that $R$ $\{x_{i}\}\times S^{2}$ $S^{2}\times\{y_{j}\}$
326 O. Saeki

represents . Note that $X=(S^{2}-\{x_{1}, \cdots, x_{a}\})\times(S^{2}-\{y_{1}, \cdots, y_{b}\})$ .

$\zeta$

Set $z=(x_{0}, y_{0})\in X$ and define , , , , , , $\alpha_{1}$

, $\beta_{b}\in\pi_{1}(X, z)$
$\alpha_{2}$
$\cdots$ $\alpha_{a}$
$\beta_{1}$ $\beta_{2}$ $\cdots$

as follows. Connect and a point near to $x_{i}(i=1,2, \cdots, a)$ by an

$x_{0}$

arc in $(S^{2}-\{x_{1}, \cdots, x_{a}\})\times\{y_{0}\}$ as in Figure 6. Then is represented $\alpha_{i}$

by a loop in $(S^{2}-\{x_{1}, \cdots x_{a}\})\times\{y_{0}\}(\subset X)$ which starts at , goes $x_{0}$

along the arc toward , goes around$x_{i}$ once counterclockwise, and $x_{i}$

goes back to along the same arc.

$x_{0}$ can be defined using an arc in $\beta_{i}$

$\{x_{0}\}\times(S^{2}-\{y_{1}, \cdots, y_{b}\})$ in a similar way. Note that $\alpha_{1}\alpha_{2}\cdots\alpha_{a}=$

and that
$\beta_{1}\beta_{2}\cdots\beta_{b}=1$ $\pi_{1}(X, z)\cong<\alpha_{1}$ , , $\alpha_{a}|\alpha_{1}\cdots\alpha_{a}=1>\times<$ $\cdots$

,
$\beta_{1}$
, $\cdots$
.
$\beta_{b}|\beta_{1}\cdots\beta_{b}=1>$

$S^{2}x\{y_{0}\}$

Figure 6

Next we do the smoothing operations to $R$ at , , , $(x_{1}, y_{1})$ $(x_{1}, y_{2})$ $\cdots$

,
$(x_{1}, y_{b})$ ,
$(x_{2}, y_{1})$, ,
$(x_{3}, y_{1})$ as follows. Set
$\cdots$ $(D^{4},
$(x_{a}, y_{1})$ B)=(D^{2}\times$

, $D^{2}\times\{0\}\cup\{0\}\times D^{2})$ $(\partial(D^{4}, B)=$ ( , Hopf link) . Furthermore

$D^{2}$ $S^{3}$ $)$

let $A$ be the annulus embedded in as in Figure 7 and denote by $A$ the
$S^{3}$

properly embedded annulus in which is obtained by pushing $IntA$

$D^{4}$

into $IntD^{4}$
. Note that $\partial(D^{4}, A)\cong\partial(D^{4}, B)$ and that $\pi_{1}(D^{4}-A)\cong$

Z. The smoothing operation at a double point of $R$ means that we $q$

replace $(D^{4}(q), D^{4}(q)\cap R)\cong(D^{4}, B)$ by $(D^{4}, A)$ or $-(D^{4}, A)$ , where
$D^{4}(q)$ is a suciently small 4-ball in centered at . Here we $S^{2}\times S^{2}$
$q$

choose $(D^{4}, A)$ or $-(D^{4}, A)$ so that the orientation is consistent with
that of $R$ . Denote by the immersed oriented surface which results
$S$

from the $(a+b-1)$ smoothing operations. It is easily seen that is $S$

an immersed 2-sphere representing with $(a-1)(b-1)$ double points. $\zeta$

 On 4-Manifolds Homotopy Equivalent to the 2-Sphere 327

Furthermore, by van Kampens Theorem we see that $\pi_{1}(S^{2}\times S^{2}-S, z)$

is isomorphic to $\pi_{1}(X, z)$ with additional relations $((i,j)=$ $\alpha_{i}=\beta_{j}^{-1}$

$(1, 1)$ , $(1, 2)$ , , $(1, b)$ , $(2, 1)$ , $(3, 1)$ ,
$\cdots$
, $(a, 1))$ . Thus $\pi_{1}(S^{2}\times S^{2}-S, z)$
$\cdots$

is generated by and we have the relations

$\alpha_{1}$
. Since $\alpha_{1}^{a}=\alpha_{1}^{b}=1$ $a$

and are relatively prime, we see that the complement of is simply

$b$ $S$

connected. This completes the proof.

A
$\iota$

Figure 7

Proof of Proposition 5.1. Set

$\zeta_{1}=\xi+2lm\eta$

$\zeta_{2}=2\xi+lm\eta$

$\zeta_{3}=2l\xi+m\eta$ .

By Lemma 5.2, is represented by a smoothly immersed 2-sphere

$\zeta_{i}$
in $S_{i}$

with simply connected complement with the number of double

$S^{2}\times S^{2}$

points equal to
0 $(i=1)$
$\{$ $lm$ $-l$ $(i=2)$

$(2l -1)(m-1)$ $(i=3)$ .

Let be a 4-ball in
.
$B_{?}^{4}$
which avoids double points of
$S^{2}\times S^{2}$
such that $S_{i}$

is the standard disk pair. Then by [5],

$(B_{i}^{4}, B_{i}^{4}\cap S_{i})$
is represented $\zeta_{i}$

by a Casson handle which has $S_{i}-IntB_{i}^{4}$ as its first stage core; i.e., $\zeta_{i}$

is represented by $V_{i}=B_{i}^{4}\cup CH_{i}$ , where $CH_{i}$ is a Casson handle which

is attached to along the trivial knot. Since $CH_{i}$ is homeomorphic to
$B_{i}^{4}$

by [12],
$D^{2}\times R^{2}$
is homeomorphic to IntD $(4Zra)$ (note that
$V_{\dot{x}}=IntV_{i}$
328 O. Saeki

$\zeta_{i}^{2}=4lm)$ has the smooth structure as an open set of

. Note that $V_{i}$

and that
$S^{2}\times S^{2}$
is dieomorphic to lntD $(41m)$ .
$V_{1}$

Next we show are mutually non-dieomorphic. Let

$V_{i}$
be $k(V_{i})$

the minimum number of double points of smoothly immersed 2-spheres

representing the generator of . Note that is an invariant of
$H_{2}(V_{i})$ $k(V_{i})$

the smooth manifold . By the construction of , we have

$V_{i}$ $V_{i}$

$k(V_{1})=0$ and
$k(V_{2})\leq lm-1$ .

On the other hand by Theorem 2.8,

$k(V_{2})\geq lm-1$ and

$k(V_{3})\geq\min\{(2l-1)(m-1)$ , $[\frac{2lm+1}{2}]\}$

$=lm$ ,

since are submanifolds of

$V_{i}$
$S^{2}\times S^{2}$
representing . Hence, ,
$\zeta_{i}$
$k(V_{1})$ $k(V_{2})$

and are all distinct. Thus

$k(V_{3})$ $V_{i}$
are not dieomorphic to each other.
This completes the proof.

In [22], Kuga showed that has infinitely many smooth $D^{2}\times R^{2}$

structures. Our method above is similar to Kugas.

We conjecture that for a given positive integer $N$ , there exists an
open homotopy admitting at least $N$ smooth structures. This con-
$S^{2}$

jecture is true if, in Theorem 2.8 in \S 2, $d_{\zeta}^{\in}=(|a|-1)(|b|-1)$ .

Remark. Akbulut [1] has recently found a compact homotopy $S^{2}$

with at least two smooth structures. The generator of the second ho-
mology group of this homotopy can be represented by an embedded $S^{2}$

2-sphere which is smooth with respect to one of the smooth structures,

while it can never be smooth with respect to the other smooth struc-
ture. Thus the interior of this homotopy also has at least two smooth $S^{2}$

structures. We note that this open homotopy is not homeomorphic $S^{2}$

to the interior of a bundle over . $D^{2}$ $S^{2}$

 On 4-Manifolds Homotopy Equivalent to the 2-Sphere 329

References

[1] S. Akbulut, An exotic -manifold, preprint.

$4$

[2] S. Boyer, Shake-slice knots and smooth contractible -manifolds, Math. $4$

Proc. Camb. Phil. Soc., 98 (1985), 93-106.

[3] , Simply-connected -manifolds with a given boundary, Trans.
$4$

Amer. Math. Soc., 298 (1986), 331-357.

[4] S. Boyer and D. Lines, Surgery formulae for Cassons invariant and
extensions to homology lens spaces, J. reine angew. Math., 405 (1990),
181-220.
[5] A.J. Casson, Three lectures on new infinite constructions in -dimen- $4$

sional manifolds, in A la recherche de la topologie perdue,

Birkh\"auser, Boston-Basel-Stuttgart, 1986, pp. 201-244.
[6] , Lecture at MSRI, March 1985.

[7] A.J. Casson and C.McA. Gordon, On slice knots in dimension three,
Proc. Symp. Pure Math., 32 (1978), 39-53.
[8] T.D. Cochran and R. Gompf, An application of Donaldsons theorem
to classical knot concordance, homology -spheres and property P,
$3$

Topology, 27 (1988), 495-512.

[9] T.D. Cochran and W.B.R. Lickorish, Unknotting information from
-manifolds, Trans. Amer. Math. Soc., 297 (1986), 125-142.
$4$

[10] S.K. Donaldson, An application of gauge theory to the topology of

-manifolds, J. Di. Geom., 18 (1983), 269-316.
$4$

[11] R. Fenn, On Dehns lemma in dimensions, Bull. London Math. Soc.,

$4$

3 (1971), 79-81.
[12] M. Freedman, The topology of -manifolds, J. Di. Geom., 17 (1982),
$4$

357-453.
[13] , The disk theorem for four-dimensional manifolds, in Proc. In-

ternat. Cong. Math., Warszawa, 1983, pp. 647-663.

[14] M. Freedman and F. Quinn, Topology of -manifolds, Princeton Math-
$4$

ematical Series 39, Princeton Univ. Press, Princeton, N.J., 1990.

[15] S. Fukuhara, Cassons invariant for homology lens spaces, Lecture at
Tokyo Institute of Technology, January 1989.
[16] S. Fukuhara, Y. Matsumoto and O. Saeki, An estimate for the unknot-
ting numbers of torus knots, Topology Appl., 38 (1991), 293-299.
[17] R. Gompf, Infinite families of Casson handles and topological disks,
Topology, 23 (1984), 395-400.
[18] , Smooth concordance of topologically slice knots, Topology, 25

(1986), 353-373.
[19] W.C. Hsiang and R.H. Szczarba, On embedding surfaces in four-
manifolds, Proc. Symp. Pure Math., 22 (1971), 97-103.
[20] S. Kaplan, Constructing framed -manifolds with given almost framed
$4$

boundaries, Trans. Amer. Math. Soc., 254 (1979), 237-263.

[21] M.A. Kervaire and J.W. Milnor, On -spheres in -manifolds, Proc Nat.
$2$ $4$

Acd. Sci. U.S.A., 47 (1961), 1651-1657.

330 O. Saeki

[22] K. Kuga, On immersed -spheres in

$2$
, in A Fete of Topology (Pa-
$S^{2}\times S^{2}$

pers dedicated to Itiro Tamura), Academic Press, 1988, pp. 525-541.

[23] N. Maruyama, On Dehn surgery along a certain family of knots, J. Tsuda
College, 19 (1987), 261-280.
[24] G. Matic, $SO(3)$ -connections and rational homology cobordisms, J. Di.
Geom., 28 (1988), 277-307.
[25] L. Moser, Elementary surgery along a torus knot, Pacific J. Math., 38
(1971), 737-745.
[26] D. Rolfsen, Knots and links, Publish or Perish Inc., Berkeley, 1976.
[27] D. Ruberman, Rational homology cobordisms of rational space forms,
Topology, 27 (1988), 401-414.
[28] O. Saeki, Simply connected -manifolds of second betti number 1
$4$

bounded by homology lens spaces, J. Math. Soc. Japan, 41 (1989),

593-606.
[29] T. Shibuya, A lower bound of -genus for torus knots, Memoirs of the
$4$

Osaka Inst. of Technology, Series A, 31 (1986), 11-17.

[30] A. Suciu, Immersed spheres in $CP^{2}$ and , Math. Z., 196 (1987),
$S^{2}\times S^{2}$

51-57.
[31] A.G. Tristram, Some cobordism invariants for links, Proc. Camb. Phil.
Soc., 66 (1969), 251-264.
[32] C.T.C. Wall, Surgery on compact manifolds, Academic Press, New
York, 1970.

Department of Mathematics
Faculty of Science
Yamagata University
Yamagata 990
Japan
Advanced Studies in Pure Mathematics 20, 1992
Aspects of Low Dimensional Manifolds
pp. 331-363

On the Deformations of the Geometric Structures

on the Seifert -Manifolds $4$

Masaaki Ue

We call a closed orientable 4-manifold $S$

a Seifert 4-manifold if $S$

has a structure of a fibered orbifold $\pi:S\rightarrow B$

over some 2-orbifold $B$
with general fiber a 2-torus where the total space
$T^{2}$
is a nonsin-
$S$

gular manifold. In [10], [11] we discussed the relations between them

and certain eight geometries in dimension 4 in the sense of Thurston
and also gave their topological classification. Here by a geometric struc-
ture of we mean the structure of the form
$S$
dieomorphic to
$\Gamma\backslash X$ $S$

where $X$ is a 1-connected complete Riemannian 4-manifold and is $\Gamma$

a discrete subgroup of the group $Isom^{+}X$ of all orientation-preserving

isomorphisms of $X$ acting freely on . The purpose of this paper is to
$X$

determine the Teichm\"uller spaces for their geometric structures in the

cases when the base orbifolds are either hyperbolic or euclidean ( and $1$

\S 2). Our results are parallel to [5], [6] in which the Teichm\"uller spaces
for the geometric structures on the Seifert 3-manifolds were discussed.
But a little more arguments are needed for our cases since we should
take account of the nontrivial monodromies. In the meanwhile some
of the Seifert 4-manifolds have complex structures compatible to their
geometric structures ([12]). In these cases we will also give the relations
between the Teichm\"uller spaces and the deformations of the associated
complex structures via the Kodaira Spencer maps. In all cases we treat
here these maps are surjective but not injective in general (and hence
the Teichm\"uller spaces are not eectively parametrized as families of
complex structures \S 3). Finally in \S 4 we also give a remark on the
moduli spaces for the geometric structures when the base orbifolds are
hyperbolic and show that they are defined as Hausdor spaces whereas,
as is well known, the moduli spaces for the complex structures can not
be defined as Hausdor spaces in general. For simplicity in this paper
we only consider the Seifert 4-manifolds over the closed orientable base
orbifolds. All the subjects will be considered in the smooth category.

Received July 2, 1990.

Revised June 22, 1991.
332 M. Ue

We will use the same notations for the geometries as in [12], [13], [10]
and [11]. The 2-dimensional hyperbolic space will be denoted by $H^{2}$

and also by $H$ (as the complex space). Furthermore and will $C^{*}$ $R^{+}$

be the set of nonzero complex numbers and the set of positive numbers
respectively.

\S 1. Definitions of Teichm\"uller spaces and the cases when the

bases are hyperbolic

be a closed orientable Seifert 4-manifold over a 2-orbifold $B$

Let $S$

and be its fiber map with general fiber

$\pi:S\rightarrow B$
. In [10] and [11] $T^{2}$

we proved that has a geometric structure if $S$ $B$ is either euclidean,

spherical or bad and if $B$ is hyperbolic is geometric if and only if $S$ $S$

has a complex structure (and is an elliptic surface). Let $G=Isom^{0}X$

be the identity component of $Isom^{+}X$ . For simplicity we only consider
the geometric Seifert 4-manifolds over either hyperbolic or euclidean
orientable base orbifolds of the form with and the $(G, X)$ $\Gamma\backslash X$ $\Gamma\subset G$

structures on them. We note that any geometric Seifert 4-manifold over

the hyperbolic 2-orbifold (we have assumed that the base orbifold is
orientable) is of the form where $X$ is either or $\Gamma\backslash X$
$H^{2}\times E^{2}$ $\overline{SL}_{2}\times E$

and $\Gamma\subset G=Isom^{0}X([11])$ . Here is the universal covering of $\overline{SL}_{2}$

$SL_{2}R$ .

Definition 1. Let $\mathcal{R}(\Gamma, G)$

be the set of all faithful discrete co-
compact representations from $\Gamma$

to $G$ with compact open topology.

The group Inn $G$ of the inner automorphisms of $G$ and the group
Aut of the automorphisms of act on
$\Gamma$

by . and $\Gamma$
$\mathcal{R}(\Gamma, G)$ $g$
$\rho(\gamma)\cdot g^{-1}$

by . $\rho$ respectively where $\in G$ , , $\phi\in

$\phi(\gamma)=\rho(\phi(\gamma))$ Aut$ $g$ $\gamma\in\Gamma$ $\Gamma$

and . The second action commutes with the first one and
$\rho\in \mathcal{R}(\Gamma, G)$

induces the action of the group Out of the outer automorphisms on $\Gamma$

the quotient Inn . $G\backslash \mathcal{R}(\Gamma, G)$

Definition 2. We call the quotient space

$\mathcal{T}(\Gamma, G)=Inn$ $G\backslash \mathcal{R}(\Gamma, G)$

a Teichm\"uller space of $S=\Gamma\backslash X$ , and the quotient

$\Lambda 4(\Gamma, G)=Inn$ $ G\backslash \mathcal{R}(\Gamma, G)/O_{11}t\Gamma$

a moduli space of $S=\Gamma\backslash X$ .

 Geometric Structures on the Seifert 4-Manifolds 333

The fundamental group $\pi_{1}^{orb}B$

of the base orbifold $B$ has the repre-
sentation of the form
$\{\overline{\alpha}_{1}$

, $\ldots$ , $\overline{\alpha}_{2g}$
, $\overline{q}_{1}$
, $\ldots$ , $\overline{q}_{r}|$

$\overline{q}_{1}^{m_{1}}=\cdots=\overline{q}_{r}^{m_{r}}=\prod_{i=1}^{g}[\overline{\alpha}_{2i-1 },\overline{\alpha}_{2i}]\prod_{j=1}^{r}\overline{q}_{j}=1\}$

where corresponds to a meridian circle around the -th cone point

$\overline{q}_{\dot{\iota}}$
$i$

of cone angle $2\pi/m_{i}$ , and , , form a symplectic base of the $\overline{\alpha}_{1}$

$\ldots$
$\overline{\alpha}_{2g}$

fundamental group of the underlying space $|B|$ of $B$ of genus . Here we $g$

define . Note that $\pi^{orb}B$ is isomorphic to a discrete

$[\alpha, \beta]=\alpha\beta\alpha^{-1}\beta^{-1}$

subgroup in and where if $B$ is hyperbolic and

$\overline{\Gamma}$
$Isom\overline{X}$ $B=\overline{\Gamma}\backslash \overline{X}$ $\overline{X}=H$

if $B$ is euclidean. Then

$\overline{X}=E^{2}$
has the following representation: $\pi_{1}S$

$\{\alpha_{1}$
, $\ldots$ , $\alpha_{2g}$ , $q_{1}$ , $\ldots$ , $q_{r}$ , , $\ell$
$h|$

$[\ell, h]=1$ , $[q_{j}, \ell]=[q_{j}, h]=1$ for $j=1$ , $\ldots$ , , $r$

$(\alpha_{i}\ell\alpha_{i}^{-1}, \alpha_{i}h\alpha_{i}^{-1})=(\ell, h)A_{i}$

for $i=1$ , $\ldots$ , $2g$ ,
$q_{s}^{m_{s}}\ell^{a_{s}}h^{b_{s}}=1$
for $s=1$ , $\ldots$ , , $r$

$\prod_{\dot{x}=1}^{g}[\alpha_{2i-1}, \alpha_{2i}]\prod_{s=1}^{r}q_{s}=\ell^{a}h^{b}\}$ .

Here , are the lifts of

$\alpha_{i}$
and respectively, and form a base
$q_{j}$
$\overline{\alpha}_{i}$
$\overline{q_{j}}$
$\ell$
$h$

of the fundamental group of the general fiber , $A_{i}\in SL_{2}Z$ is $Z^{2}$ $T^{2}$

the monodromy matrix corresponding to with respect to , and $\overline{\alpha}_{i}$

$(\ell, h)$

is the Seifert invariant of the -th multiple fiber of multiplic-

$(m_{s}, a_{s}, b_{s})$ $s$

ity over the -th cone point. The Seifert invariants for such
$m_{s}$ are $s$ $S$

denoted by
$\{A_{1}, \ldots, A_{2g}, (a, b), (m_{1}, a_{1}, b_{1}), \ldots, (m_{r}, a_{r}, b_{r})\}$ .

First consider when is hyperbolic. Then by $S=\Gamma\backslash X$ $B=\overline{\Gamma}\backslash \overline{X}$

the results in [15] the fibration of is unique up to fiber-preserving $S$

dieomorphisms and is uniquely determined by up to group $\pi_{1}^{orb}B=\overline{\Gamma}$ $\Gamma$

automorphisms. Moreover all the monodromy matrices are the powers of

some common periodic matrix $Q$ ([11], Theorem ). If every monodromy $B$

is trivial then the pair $e=(a+\sum a_{j}/m_{j}, b+\sum b_{j}/m_{j})\in Q^{2}mod GL_{2}Z$
is well defined and is called the rational euler class of . The type of $X$ $S$

is if every monodromy is trivial and $e\neq(0,0)$ and $X=H^{2}\times E^{2}$

$\overline{SL_{2}}\times E$

otherwise. Furthermore we can assume that $a+\sum a_{j}/m_{j}=0$ in the

first case by some coordinate change of the fiber ([11]).
334 M. Ue

Proposition 1. Let be a geometric Seifert 4-manifold over $S$ $a$

hyperbolic orbifold B. Then for appropriate choices of the lattices of

the general fifiber and the symplectic basis of the curves on $B$ generat-
ing $H_{1}(|B|, Z)$ we can assume that $A_{i}=I$ for $i\geq 2$ and is either $A_{1}$

$\left(\begin{array}{ll}1 & 1\\-1 & 0\end{array}\right)$

, $\left(\begin{array}{ll}0 & 1\\-1 & -1\end{array}\right)$
, $\left(\begin{array}{ll}0 & 1\\-1 & 0\end{array}\right)$
, $\pm I$
, where I is the identity matrix.

We can assume that all monodromy matrices

Proof. of are $A_{i}$ $S$

powers of a periodic matrix $Q$ . Then by choosing the lattice , ap- $\ell$
$h$

propriately we can suppose that $Q$ is either one of the five matrices
given above since $Q$ is periodic. In particular all are mutually com- $A_{i}$

mutative. If we consider the pullback of by a self-automorphism $S$ $\phi$

of the base $B$ fixing all cone points and the base point of $B$ then $\phi$

induces the symplectic isomorphism on $H_{1}(|B|, Z)$ and the mon- $\phi_{*}$

odromy matrices , , are transformed to , , where $A_{1}$

$\ldots$
$A_{2g}$ $\phi_{*}A_{1}$
$\ldots$
$\phi_{*}A_{2g}$

if for the symplectic bases , ,

$\phi_{*}\overline{\alpha}_{i}=\delta_{1}\overline{\alpha}_{1}+\delta_{2}\overline{\alpha}_{2}+\cdots\delta_{2g}\overline{\alpha}_{2g}$ $\overline{\alpha}_{1}$

$\ldots$
$\overline{\alpha}_{2g}$

of $H_{1}(|B|, Z)$ then . Since all s are powers of $\phi_{*}A_{i}=A_{1}^{\delta_{1}}A_{2}^{\delta_{2}}\cdots A_{2g}^{\delta_{2g}}$ $A_{i}$

$Q$ Euclid algorithm shows that this process simplifies the monodromy

matrices (which are still powers of $Q$ after this process) of the Seifert
fibration of induced from the original one by if is chosen appro- $S$ $\phi$ $\phi$

priately. In fact we can see that some automorphism of $B$ fixing all $\phi$

the cone points induces the isomorphism such that $\phi_{*}A_{i}=I$ for $\phi_{*}$

$i\geq 2$ as follows. First for any $P_{i}\in SL_{2}Z$ , $(i=1, \ldots, g)$ there is an
automorphism of $B$ such that $\psi$

$(\psi_{*}\overline{\alpha}_{2i-1}, \psi_{*}\overline{\alpha}_{2i})=(\overline{\alpha}_{2i-1}, \overline{\alpha}_{2i})P_{i}$

since every symplectic isomorphism can be realized by some orinetation

preserving self-dieomorphism of $B$ fixing all the cone points. Using
such the monodromy matrices can be transformed so that $A_{2i}=I$ for
$\psi$

$i=1$ , , . Next consider the symplectic isomorphism satisfying

$\ldots$ $g$ $\rho$

$\rho(\overline{\alpha}_{2})=\overline{\alpha}_{2}-\overline{\alpha}_{2i}$

$\rho(\overline{\alpha}_{2i-1})=\overline{\alpha}_{1}+\overline{\alpha}_{2i-1}-\overline{\alpha}_{2i}$

$\rho(\overline{\alpha}_{j})=\overline{\alpha}_{j}$
otherwise

which maps $A_{2i-1}$ to $A_{2i-1}A_{1}$ , leaves the other monodromies unchanged

and can be realized by the Dehn twists along the curves representing
and . Then applying the isomorphisms of this type or their
$\overline{\alpha}_{1}\overline{\alpha}_{2i}^{-1}$ $\alpha_{1}^{-1}$

inverses (or those obtained by exchanging the roles of and ) and $\alpha_{1}$ $\alpha_{2i-1}$

the isomorphism mapping to we can see (by Euclid $(\alpha_{1}, \alpha_{2})$ $(\alpha_{2i-1}, \alpha_{2i})$

algorithm) that $A_{2i-1}(i\geq 2)$ is reduced to I with all others except

 Geometric Structures on the Seifert 4-Manifolds 335

for left unchanged. The desired automorphism is obtained by the

$A_{1}$ $\phi$

composition of the above automorphisms. Finally consider if necessary

the symplectic isomorphism satisfying $\sigma$

$\sigma(\overline{\alpha}_{1})=-\overline{\alpha}_{1}$

$\sigma(\overline{\alpha}_{2})=-\overline{\alpha}_{2}$

$\sigma(\overline{\alpha}_{j})=\overline{\alpha}_{j}$
for $i\geq 3$

(which maps to , leaves $A_{i}=I(i\geq 2)$ unchanged and is also

$A_{1}$ $A_{1}^{-1}$

realized by an automorphism of $B$ ) and some further change of the

lattice of the general fiber we obtain the desired representations of the
monodromy matrices since is still a power of some periodic matrix. $A_{1}$

Q.E.D.

In the case of a geometric Seifert 4-manifold over a hyper- $S=\Gamma\backslash X$

bolic base orbifold , $X$ is complex analytically equivalent to

$B=\overline{\Gamma}\backslash \overline{X}$

$H\times C$ ([11], [12] and see the proof of Theorem A below) such that the

lattice of the general fiber of acts on the $C$ -factor as translations.

$\Gamma_{0}$ $S$

Furthermore the lift of the elements of to induce the orientation-

$s$
$\overline{\Gamma}$
$\Gamma$

preserving automorphisms of the $C$ -factor (up to translations) which

do not depend on the choices of the lifts and which preserve the lat-
tice in $C$ defined by . Then we have a homomorphism from to
$\Gamma_{0}$
$\overline{\Gamma}$

$GL^{+}(1, C)=C^{*}$ which we call the monodromy representation of $\Gamma\backslash X$

and denote by . The relation between and the monodromy matri-

$\phi$ $\phi$

ces of is explained in Theorem A and its proof below. To

$S=\Gamma\backslash X$

describe the Teichm\"uller space of we introduce the following extra $\Gamma\backslash X$

notations.
$I(\overline{\Gamma}, Isom^{+}\overline{X})$
the Teichm\"uller space of $B=\overline{\Gamma}\backslash \overline{X}$

$H^{1}(\overline{\Gamma}, C^{\phi})$
the 1st cohomology group of $\overline{\Gamma}$

with coecients $C^{\phi}$

.

Here is $C$ twisted by ,

$C^{\phi}$
where $\phi$ $\mathcal{T}(\overline{\Gamma}, Isom^{+}\overline{X})=R^{2(3g-3+r)}\times Z_{2}$ $g$

is the genus of $B$ and is the number of the cone points of . Let
$B$ $r$ $\mathcal{T}_{g,r}$

be the identity component of . (It is well known that $I(\overline{\Gamma}, Isom^{+}\overline{X})$ $\mathcal{T}_{g,r}$

which is also the Teichm\"uller space of the orbifold $B$ depends only on $g$

and ) $r.$

Theorem A. $\mathcal{T}(\Gamma, G)$

for a Seifert 4-manifold $S=\Gamma\backslash X$ over $a$

hyperbolic 2-orbifold $B=\overline{\Gamma}\backslash \overline{X}$

has a structure of a trivial fifiber bundle
of the following form.
$\mathcal{F}$ $\rightarrow I(\Gamma, G)\rightarrow \mathcal{T}(\overline{\Gamma}, Isom^{+}\overline{X})$
.
336 M. Ue

Here the fifiber $\mathcal{F}$

is isomorphic to $H^{1}(\overline{\Gamma}, C^{\phi})\times T_{1}$

where
$R^{+}\times H\times Z_{2}$ , if $X=H^{2}\times E^{2}$ and $\phi\equiv\pm id$

$T_{1}=\{$ $R^{+}\times Z_{2}$

, if $X=H^{2}\times E^{2}$ and $\phi\not\equiv\pm id$

$H\times Z_{2}$ , if $X=\overline{SL}_{2}\times E$

and corresponds to the deformations of the lattice of the general fifiber

$T_{1}$

generated by and with $c\in R^{+}$ ,

$c$
and . has two
$c\lambda^{-1}$
$\lambda\in C$ $s\propto\lambda\neq 0$ $T_{1}$

components according to the sign of . The monodromy representation $ s^{\propto}\lambda$

satisfifies
$\phi$

if the monodromy matrix corresponding to

$\phi(\overline{\alpha}_{i})=\pm 1$ $A_{i}$ $\overline{\alpha}_{i}$

is . If
$\pm I$
$A_{i}\neq\pm I$
for some , the lattice in $C$ of the general fifiber $i$ $\Gamma_{0}$

is uniquely determined up to scalar multiplication and is also uniquely $\phi$

determined once the sign is fifixed. satisfifies $ of\propto s\lambda$ $H^{1}(\overline{\Gamma}, C^{\phi})$

$H^{1}(\overline{\Gamma}, C^{\phi})=\{$
$C^{2g}$
if $\phi\equiv id$

$C^{2g-2}$
otherwise.

The identity component of $\mathcal{T}_{0}$

$\mathcal{T}(\Gamma, G)$
is $I_{g,r}\times H^{1}(\overline{\Gamma}, C^{\phi})\times T_{0}$
where $T_{0}$

is the connected component of $T_{1}$

and is homeomorphic to $a$ euclidean
space.

Proof. We may assume that satisfies the conditions in Proposi- $S$

tion 1. First suppose that the type $X$ of the geometry of is $S=\Gamma\backslash X$

which is identified with $H\times C$ . In this case

$H^{2}\times E^{2}$ $ G=Isom^{0}H^{2}\times$

$Isom^{0}E^{2}$
. Let be any element for . Then induces
$\rho\in \mathcal{R}(\Gamma, G)$ $S=\Gamma\backslash X$ $\rho$

the representation with $\overline{G}=Isom^{0}H^{2}=PSL_{2}R$ . gives

$\overline{\rho}\in \mathcal{R}(\overline{\Gamma}, \overline{G})$
$\overline{\rho}$

the representation of the base $B$ as the hyperbolic orbifold . $B=\overline{\Gamma}\backslash H$

Then for the coordinates $(z, w)\in H\times C$ , we have

(0) $\rho(\ell)(z, w)=(z, w+c)$

(1) $\rho(h)(z, w)=(z, w+d)$

(2) $\rho(\alpha_{i})(z, w)=(\overline{\rho}(\overline{\alpha}_{i})z, \lambda_{i}w+w_{i})$ $(i=1, \ldots, 2g)$

(3) $\rho(q_{j})(z, w)=(\overline{\rho}(\overline{q}_{i})z, w-(a_{j}c+b_{j}d)/m_{j})$

where , $d\in C$ are linearly independent over $R$ ,

$c$
, $w_{i}\in C$ $\lambda_{i}\in S^{1}\subset C$

which satisfy the following relations. We note that (3) comes from the
relations , $[q_{j}, \ell]=[q_{j}, h]=1$ . Put $c=u+iv$ , $d=u+iv$
$q_{j^{m_{j}}}\ell^{a_{j}}h^{b_{j}}=1$

and $P=$ $\left(\begin{array}{ll}u & v\\u, & v,\end{array}\right)$

$\in GL2R$ . Then we deduce from $\alpha_{i}(\ell, h)\alpha_{i}^{-1}=$

$(\ell, h)A_{i}$ that

$PA_{i}P^{-1}=\left(\begin{array}{ll}cos\theta_{i} & -sin\theta_{i}\\sin\theta_{i} & cos\theta_{i}\end{array}\right)$
 Geometric Structures on the Seifert 4-Manifolds 337

where Thus we have for $i\geq 2$ since

$\lambda_{i}=\exp(\sqrt{-1}\theta_{i})$
. $\lambda_{i}=1$ $A_{i}=I$ . If
$A_{1}=\pm I$ , then and there is no further restriction on $\lambda_{1}=\pm 1$ $(c, d)$ . For

the remaining cases we have $(c, d)=(c, \lambda^{-1}c)$ with

(4) $\lambda_{1}=\lambda=\exp(\pm 2\pi i/6)$ if $A_{1}=\left(\begin{array}{ll}1 & 1\\-1 & 0\end{array}\right)$

(5) $\lambda=\exp(\pm 2\pi i/6)$ , $\lambda_{1}=\lambda^{2}$

if $A_{1}=\left(\begin{array}{ll}0 & 1\\-1 & -1\end{array}\right)$

(6) $\lambda_{1}=\lambda=\pm i$ if $A_{1}=\left(\begin{array}{ll}0 & 1\\-1 & 0\end{array}\right)$

.

The monodromy representation is defined by , $\phi:\overline{\Gamma}\rightarrow C^{*}$ $\phi(\overline{\alpha}_{i})=\lambda_{i}$

. We note that is uniquely determined if $A_{i}=\pm I$ . If $A_{1}\neq\pm I$

$\phi(\overline{q}_{i})=1$ $\phi$

we have two choices of above according or . Hence $\lambda$

$as\propto s\lambda>0$ $s^{\propto}\lambda<0$

the parameter space for the lattice of the general fiber (represented by
and ) has two components and is homeomorphic to
$c$
$\lambda$

. The $C^{*}\times Z_{2}$

monodromy representation depends only on the choice of . If $A_{1}=$ $\phi$

$\lambda$

, (and if we write ( , $d)=(c,$

$\pm I$
) then ranges over $c$ $c\lambda^{-1})$ $(c, \lambda)$ $C^{*}\times H\times$

. In this case we have two components according to the sign

$Z_{2}$
. $of\propto s\lambda$

Finally from the relation we deduce $\prod_{j=1}^{g}[\alpha_{2j-1}, \alpha_{2j}]\prod_{j=1}^{r}q_{j}=\ell^{a}h^{b}$

$\sum(\phi(\overline{\alpha}_{2j-1})g-1)w_{2j}-\sum_{j=1}^{g}(\phi(\overline{\alpha}_{2j})-1)w_{2j-1}$

$j=1$
(7)
$=(a+\sum_{j=1}^{r}a_{j}/m_{j})c+(b+\sum_{j=1}^{r}b_{j}/m_{j})d$ .

Clearly we can find satisfying (7) for any since the right hand $w_{j}$
$\overline{\rho}$

side on (7) is 0 if . Hence defined by $(0)-(3)$ satisfying $(4)-$ $\phi\equiv id$ $\rho$

$(7)$ defines a discrete faithful representation from to $G$ . Conversely $\Gamma$

every has a lift . Hence the natural projection

$\overline{\rho}\in \mathcal{R}(\overline{\Gamma}, \overline{G})$
$\rho\in \mathcal{R}(\Gamma, G)$

is surjective. Next we describe the fiber

$p:\mathcal{R}(\Gamma, G)\rightarrow \mathcal{R}(\overline{\Gamma}, \overline{G})$
of . $\mathcal{F}$

$p$

Pick up and fix a lift of satisfying $(0)-$

$\overline{\rho}_{0}\in \mathcal{R}(\overline{\Gamma}, \overline{G})$ $\rho_{0}\in \mathcal{R}(\Gamma, G)$ $\overline{\rho}$

$(7)$ as a base point of the fiber over . Hereafter the parameters in $\overline{\rho}_{0}$

$(0)-(7)$ for are denoted by the same symbols with sux 0. We can
$\rho_{0}$

choose in (2) for so that $w_{j}^{0}=0$ if

$w_{j}^{0}$
or $w_{j}^{0}=0$ for $j\neq 2$ $\rho_{0}$
$\phi\equiv id$

and $w_{2}^{0}=((a+\sum a_{j}/m_{j})c+(b+\sum b_{j}/m_{j})d)/(\phi(\overline{\alpha_{1}})-1)$ if $\phi\not\equiv id$

under the assumption in Proposition 1. Take such that $\rho\in \mathcal{R}(\Gamma, G)$

, $(c, d)=(c_{0}, d_{0})$ and the monodromy representations for and

$\overline{\rho}=\overline{\rho}_{0}$ $\rho$
338 M. Ue

$\rho_{0}$
are the same. Put $m(\overline{\alpha}_{j})=w_{j}-w_{j}^{0}$ , $m(\overline{q}_{j})=0$ . Then $m(\overline{\alpha}_{j})$
satisfy

(8) $\sum_{j=1}^{g}(\phi(\overline{\alpha}_{2j-1})-1)m(\overline{\alpha}_{2j})-\sum_{j=1}^{g}(\phi(\overline{\alpha}_{2j})-1)m(\overline{\alpha}_{2j-1})=0$

and $m$ can be extended to a crossed homomorphism satisfy- $m:\overline{\Gamma}\rightarrow C$

ing for , . Let be the -th

$m(\overline{\alpha}\overline{\beta})=m(\overline{\alpha})+\phi(\overline{\alpha})m(\overline{\beta})$ $\overline{\alpha}$ $\overline{\beta}\in\overline{\Gamma}$ $C^{i}(\overline{\Gamma}, C^{\phi})$ $i$

cochain of with coecients $C$ twisted by . Then $m$ is contained in the

$\overline{\Gamma}$

$\phi$

kernel of the coboundary map . Conversely if $\delta:C^{1}(\overline{\Gamma}, C^{\phi})\rightarrow C^{2}(\overline{\Gamma}, C^{\phi})$

$ m\in ker\delta$ then since and . Furthermore for $m(\overline{q}_{j})=0$ $\overline{q}_{j}^{m_{j}}=1$ $\phi(\overline{q}_{j})=1$

any such $m$ , we can define a faithful discrete representation satisfying $\rho$

, $(c, d)=(c_{0}, d_{0})$ ,

$\overline{\rho}=\overline{\rho}_{0}$
. The choice of $m$ does not $w_{j}=w_{j}^{0}+m(\overline{\alpha}_{j})$

depend on the choice of if is fixed. Thus the fiber is homeo- $(c_{0}, d_{0})$ $\phi$
$\overline{\mathcal{F}}$

morphic to where if or
$(ker\delta)\times\tilde{T}_{1}$ $\tilde{T}_{1}=C^{*}\times Z_{2}$ $\phi\not\equiv\pm id$ $C^{*}\times H\times Z_{2}$

if . We note that the choices of the parameters of

$\phi\equiv\pm id$ (and the $\tilde{T}_{1}$

choices of in (2) satisfying (7) for the fixed lift

$w_{j}^{0}$
of ) do not $\rho_{0}$ $\overline{\rho}_{0}$

depend on . So the projection gives a product

$\overline{\rho}$
$p:\mathcal{R}(\Gamma, G)\rightarrow \mathcal{R}(\overline{\Gamma}, \overline{G})$

fibration and . Next we check the action

$\mathcal{R}(\Gamma, G)=\mathcal{R}(\overline{\Gamma}, \overline{G})\times ker\delta\times\tilde{T}_{1}$

of Inn on $G$ . The action of Inn $PSL_{2}R$ is nonrivial only on the

$\mathcal{R}(\Gamma, G)$

first factor of and yields a Teichm\"uller space of

$\mathcal{R}(\overline{\Gamma}, \overline{G})$
$\mathcal{R}(\Gamma, G)$ $\mathcal{T}(\overline{\Gamma}, \overline{G})$

the base orbifold B. has just two components which correspond $\mathcal{T}(\overline{\Gamma}, \overline{G})$

to the Teichm\"uller spaces of the hyperbolic structures on $B$ and $B$ with
opposite orientation. Each one is identified with the Teichm\"uller space
$\mathcal{T}_{g,r}=R^{2(3g-3+r)}$ of -pointed Riemann surface of genus . Next we pick $r$ $g$

up $\mu\in Isom^{+}E^{2}$ defined by $\mu(z, w)=(z, \sigma w+w)$ with , $w\in C$ $\sigma\in S^{1}$

(acting trivially on the first coordinate). If satisfying $(0)-$ $\rho\in \mathcal{R}(\Gamma, G)$

$(3)$ , then we have

$\mu\rho(\ell)\mu^{-1}(z, w)=(z, w+\sigma c)$ ,

$\mu\rho(h)\mu^{-1}(z, w)=(w, w+\sigma d)$ ,
$\mu\rho(\alpha_{j})\mu^{-1}(z, w)=(\overline{\rho}(\overline{\alpha}_{j})z, \lambda_{j}w+\sigma w_{j}+(1-\lambda_{j})w)$
,
$\mu\rho(q_{j})\mu^{-1}(z, w)=(\overline{\rho}(\overline{q_{j}})z, w-(a_{i}\sigma c+b_{i}\sigma d)/m_{i})$ .

where , $d=c\lambda^{-1}$ as in $(4)-(6)$ . Then are transformed

$\lambda_{j}=\phi(\overline{\alpha}_{j})$ $(c, \lambda)$

to and is transformed to
$(\sigma c, \lambda)$
. Thus if the
$w_{j}$
$\sigma w_{j}+(1-\phi(\overline{\alpha}_{j}))w$

representative of in Inn is fixed so that $c\in R^{+}$ , then $c$ $Isom^{+}E^{2}\backslash \mathcal{F}$

s are defined modulo the image of :

$m(\overline{\alpha}_{j})$
in $\delta^{0}$
$C^{0}(\overline{\Gamma}, C^{\phi})\rightarrow C^{1}(\overline{\Gamma}, C^{\phi})$

Inn . Therefore we obtain

$Isom^{+}E^{2}\backslash \overline{\mathcal{F}}$

$\mathcal{T}(\Gamma, G)=I(\overline{\Gamma}, \overline{G})\times(T_{1}\times H^{1}(\overline{\Gamma}, C^{\phi}))$

 Geometric Structures on the Seifert 4-Manifolds 339

where is parametrized by ( $mod$ Inn )

$I(\Gamma, G)$ , $\overline{\rho}$
$\overline{G}$
$\in \mathcal{T}(\overline{\Gamma}, \overline{G})$ $(c, \lambda)\in$

$T_{1}=R^{+}\times H\times Z_{2}$ or , and as

$R^{+}\times Z_{2}$ $m(\overline{\alpha}_{j})(mod Im\delta^{0})\in H^{1}(\overline{\Gamma}, C^{\phi})$

desired. is a vector space over $C$

$H^{1}(\overline{\Gamma}, C^{\phi})$
whose dimension is easily
computed by (8) under the assumption of Proposition 1 as indicated in
Theorem A. We note that is the same as $H^{1}(\pi_{1}|B|, C^{\phi})$ since $H^{1}(\overline{\Gamma}, C^{\phi})$

the coecient of the cohomology is torsion free and the monodromies

along the torsion elements in are trivial. $\overline{q}_{j}$
$\overline{\Gamma}$

Next we consider the case when with $X=\overline{SL_{2}}\times E$ $S=\Gamma\backslash X$

and R. In this case $X$ is identified with $H\times C$ with

$ G=\overline{SL_{2}}\times zR\times$

coordinates $(z, w)$ , $z\in H$ , $w\in C$ so that $w$ corresponds to $\log dz([13])$ .
Here the imaginary part of $\log dz$ corresponds to the lift of the unit
tangent vector at $z\in H^{2}$ to the fiber of the natural projection $\pi:\overline{SL}_{2}\rightarrow$

. This projection is defined via the identification of $PSL_{2}R$ with

$H^{2}$

the unit tangent bundle of . The real part of $\log dz$ belongs to $T_{1}H^{2}$ $H^{2}$

the $E$-factor of $X$ . Then in induces the element in $\rho$

$\mathcal{R}(\Gamma, G)$ $\overline{\rho}$
$\mathcal{R}(\overline{\Gamma}, \overline{G})$

where and $\overline{G}=PSX2R$

. Moreover must be of the following
$\overline{\Gamma}=\pi_{1}^{orb}B$
$\rho$

form:

$\rho(\ell)(z, w)=(z, w+c)$

$\rho(h)(z, w)=(z, w+d)$

$\rho(\alpha_{j})(z, w)=\tilde{\alpha}_{j}(z, w)+(0, w_{j})$

$\rho(q_{j})(z, w)=\tilde{q}_{j}(z, w)+(0, y_{j})$

where $\tilde{\alpha}_{j}(z, w)$

is a lift of $\overline{\rho}(\overline{\alpha}_{j}):z\rightarrow(a_{j}z+b_{j})/(c_{j}z+d_{j})$ defined by

$\tilde{\alpha}_{j}(z, w)=(\overline{\rho}(\overline{\alpha}_{j})z, w-2\log(c_{j}z+d_{j}))$ .

Here the imaginary part of the second factor is chosen so that it is con-
tinuous and it coincides with the image of by the parallel translation $s^{\propto}w$

from to $z$
along the axis of the hyperbolic element (which is
$\overline{\rho}(\overline{\alpha_{j}})z$ $\overline{\rho}(\overline{\alpha}_{j})$

defined as the lift of that on via the projection ) if $T_{1}H^{2}$ $\overline{SL}_{2}\rightarrow T_{1}H^{2}$
$z$

lies in this axis. These conditions determine the choice of the branch of
in the image of . A lift
$\log$ of is taken so that in $G$ (cf.
$\tilde{\alpha}_{j}$ $\tilde{q}_{j}$ $\overline{\rho}(\overline{q}_{j})$ $\tilde{q}_{j}^{m_{j}}=1$

[11], [13] . Note that $)$

is uniquely determined since the $R\times R$-factor $\tilde{q}_{j}$

of $G$ lies in the center of $G$ . Then

$y_{j}=-(a_{j}c+b_{j}d)/m_{j}$

from $q_{j}^{m_{j}}\ell^{a_{j}}h^{b_{j}}=1$ so that

. We have chosen $\ell$

, $h$
$a+\sum a_{j}/m_{j}=0$ ,
$b+\sum b_{j}/m_{j}\neq 0$ and hence we also deduce from $\prod[\alpha_{2j-1}, \alpha_{2j}]\prod q_{j}=$
340 M. Ue

$\ell^{a}h^{b}$
that
$d=(2\pi i\chi^{orb}B)/(b+\sum b_{j}/m_{j})\neq 0$

where denotes the orbifold euler characteristic. Therefore the pa-

$\chi^{orb}$

rameters , are fixed, $c=u+iv$ is an arbitrary complex num-

$y_{j}$
$d$

ber with $u\neq 0$ (since and must be linearly independent over $R$ ), $c$ $d$

and are arbitrary complex numbers. Then the natural projection

$w_{j}$

defined by is surjective and the fiber

$p:\mathcal{R}(\Gamma, G)\rightarrow \mathcal{R}(\overline{\Gamma}, \overline{G})$ $p(\rho)=\overline{\rho}$

of is
$\mathcal{F}$

$p$ where $T_{1}=H\times Z_{2}$ which corresponds to

$C^{1}(\overline{\Gamma}, C)\times T_{1}$ $c$

(or equivalently ). has two components according to the sign of

$ic$ $T_{1}$

. Since every translation $(z, w)\rightarrow(z, w+s+ti)$ commutes with

$\Re c$

any element $\in G$ , the action of Inn $G$ on $g$ yields the following $\mathcal{R}(\Gamma, G)$

isomorphism;

$I(\Gamma, G)=\mathcal{T}(\overline{\Gamma}, \overline{G})\times H^{1}(\overline{\Gamma}, C)\times T_{1}$

which proves Theorem A.

\S 2. The cases with euclidean base orbifolds

Suppose that is a Seifert 4-manifold over a closed orientable eu-

$S$

clidean 2-orbifold $B$ . In this case has always a geometric structure of $S$

type $X$ where $X=E^{4}$ , Nil3

paper we restrict our attention to the cases when $\pi_{1}S=\Gamma$ is a subgroup
of $G=Isom^{0}X$ . Then by the results in [10] we have only to consider
the cases when is dieomorphic to one of the followings (note that the
$S$

fibration of is not unique when $B$ is not hyperbolic).

$S$

(I) $B=T^{2}$ .
(1) $S=T^{4}$ , $X=E^{4}$ ;

(2) $S=\{I, I, (a, b)\}$ with $(a, b)\neq(0,0)$ , $X=Ni1^{3}\times E$ ;

(3) $S=$ $\{\left(\begin{array}{l}1\lambda\\ 01\end{array}\right), I; (a, b)\}$
, $\lambda\neq 0$
, $b\neq 0$ , $X=Ni1^{4}$ ;

(4) $S=\{A, I; (a, b)\}$ with $trA\geq 3$ , $X=So1^{3}\times E$ .

(II) $B$ has genus 0.

(1) The rational euler class of $e$ $S$
equals $(0, 0)$ , $X=E^{4}$ ;

(2) $e\neq(0,0)$ , $X=Ni1^{3}\times E$ .

Here is dieomorphic to a hyperelliptic surface in case (II-1), a

$S$

primary Kodaira surface in case (1-2), and a secondary Kodaira surface

 Geometric Structures on the Seifert 4-Manifolds 341

in case (II-2). We note that

$Isom\ovalbox{\tt\small REJECT}$

Nil3 $>\triangleleft SO_{2}\times R$

if $X=Ni1^{3}\times E$ ,
$G=\{$
Sol3 $\times R$
if $X=So1^{3}\times E$ ,
Nil4 if $X=Ni1^{4}$ .

Theorem B. For the Seifert 4-manifold $S=\Gamma\backslash X$ in the above

list we have the list of $I(\Gamma, G)$ as follows.
(1) The cases when $B=T^{2}$ .
$SO_{4}\backslash GL_{4}R$ , in case (I-1),
$I_{1,0}\times F$ , in case (I-2),
$\mathcal{T}(\Gamma, G)=\{$
$(R^{*})^{2}\times R^{2}$ , in case (I-3),
$(R^{*})^{2}\times(Z_{2})^{2}\times R$ , in case (1-4).

Here in case (1-2) $\mathcal{T}_{1,0}=SO_{2}\backslash GL_{2}R=R^{+}\times H\times Z_{2}$ , $\mathcal{F}=R^{2}\times H\times Z_{2}$
.
(2) The cases with the base orbifolds of genus 0. Let $r$
be the number
of the cone points of $B$ ( $r=3$ or 4). Then we have
$\mathcal{T}(\Gamma, G)=I_{0,r}\times T_{1}$

where
$I_{0,r}=\{$
$R^{2(r-3)}\times R^{+}\times Z_{2}$
if $X=Ni1^{3}\times E$
$R^{2(r-3)}\times R^{+}$
if $X=E^{4}$
and
$T_{1}=\{$
$H\times Z_{2}$
if $X=Ni1^{3}\times E$
$R^{+}\times H\times Z_{2}$
if $X=E^{4}$ .

Proof. In case (1-1) we have $\Gamma=Z^{4}$ whose generators are given $\alpha_{i}$

by translations $\alpha_{i}x=x+\ell_{i}$ for , . Here are mutually linearly $x$ $\ell_{i}\in E^{4}$ $\ell_{i}$

independent and hence is parametrized by $GL4R$ . The action of Inn $G$ $\Gamma$

is given by for . Hence

$(\ell_{1}, \ldots, \ell_{4})=(\sigma\ell_{1}, \ldots, \sigma\ell_{4})$ $\sigma\in SO_{4}$ $\mathcal{T}(\Gamma, G)=$

which has two components and each one is homeomorphic

$SO_{4}\backslash GL_{4}R$

to .
$R^{10}$

Next consider case (1-2). In this case $X=Ni1^{3}\times E$ . Here we recall

the structure of Nil3 . The point of $X$ is represented by $(w, z)\in C^{2}$
$\times E$

such that the action of $G=(Ni1^{3}\times R)\lambda S^{1}$ is defined by $(w, z)(w, z)=$
$(w+w-i\overline{z}z, z+z)$ for $(w, z)\in X$ , and $t(w, z)=(w, tz)$ for $t$ $\in S^{1}\subset$

C. We can assume that $S=\{I, I; (a, 0)\}$ for $a\neq 0$ and , , , $h|$ $\Gamma=\{\alpha$ $\beta$
$\ell$
342 M. Ue

$[\ell, h]=[\alpha, \ell]=[\alpha, h]=[\beta, \ell]=[\beta, h]=1$ , $[\alpha, \beta]=\ell^{a}\}$

. The subgroup $\Gamma_{0}$

generated by , is the center of and the projection Nil3

$\ell$ $h$ $\Gamma$ $\times E\rightarrow R^{2}$

defined by $(w, z)\rightarrow z$ induces the structure of a bundle over of the $T^{2}-$ $T^{2}$

form which gives the above fibration

$\Gamma_{0}\backslash R\times E\rightarrow\Gamma\backslash Ni1^{3}\times E\rightarrow Z^{2}\backslash R^{2}$

([12]). Thus must have the following form:

$\rho\in \mathcal{R}(\Gamma, G)$

$\rho(\ell)(w, z)=(w+\ell_{0}, z)$

$\rho(h)(w, z)=(w+h_{0}, z)$

$\rho(\alpha)(w, z)=(a_{0}+w-i\overline{b_{0}}z, b_{0}+z)$

$\rho(\beta)(w, z)=(a_{1}+w-i\overline{b_{1}}z, b_{1}+z)$

where $\ell_{0}$
are linearly independent over R. Since
and $h_{0}$
, $b_{0}$
and $b_{1}$

must be equal to ,
$\rho([\alpha, \beta])(w, z)=(w+i(b_{0}\overline{b_{1}}-\overline{b_{0}}b_{1}), z)$ $\rho(\ell^{a})(w, z)$

we have and since this is a nonzero real number

$\ell_{0}=i(b_{0}\overline{b_{1}}-\overline{b_{0}}b_{1})/a$ $h_{0}$

is an arbitrary number with . Thus we have $s^{\propto}h_{0}\neq 0$

$\mathcal{R}(\Gamma, G)=\overline{\mathcal{F}}\times \mathcal{R}(Z^{2}, \overline{G})$

where which is represented by , , , $\overline{G}=Isom^{0}E^{2}$ ,

$\overline{\mathcal{F}}=H\times Z_{2}\times C^{2}$
$h_{0}$
$a_{0}$ $a_{1}$

and $\mathcal{R}(Z^{2}, G)=GL_{2}R$ represented by and . Next taking the con- $b_{0}$ $b_{1}$

jugation of by $\gamma=(w_{0},
$\rho$
z_{0})$ and $t\in S^{1}$
we can see that the parameters
are transformed as follows:
$a_{0}\rightarrow a_{0}+i(\overline{b_{0}}z_{0}-\overline{z_{0}}b_{0})$

$a_{1}\rightarrow a_{1}+i(\overline{b_{1}}z_{0}-\overline{z_{0}}b_{1})$

$b_{0}\rightarrow tb_{0}$

$b_{1}\rightarrow tb_{1}$
.

Thus to get the representation in $I(\Gamma, G)$ we can assume that $b_{0}\in R^{+}$
(choose $t\in S^{1}$ appropriately). Then and $i(\overline{b_{0}}z_{0}-\overline{z_{0}}b_{0})=-2b_{0}\propto sz_{0}$

hence choosing so that $z_{0}$

then is
$s^{\propto}z_{0}=\Re a_{0}/2b_{0}$ $a_{0}+i(\overline{b}_{0}z_{0}-\overline{z}_{0}b_{0})$

pure imaginary. On the other hand $i(\overline{b_{1}}z_{0}-\overline{z_{0}}b_{1})=2((\propto sb_{1})(\Re z_{0})-$

$(\Re b_{1})(\Re a_{0}/2b_{0}))$ . $Since\propto sb_{1}\neq 0$ by the assumption we can choose $z_{0}$

so that both and are transformed to pure imaginary numbers.

$a_{0}$ $a_{1}$

Consequently we have where

$\mathcal{T}(\Gamma, G)=\mathcal{F}\times \mathcal{T}(\overline{\Gamma}, \overline{G})$ $\mathcal{F}=R^{2}\times H\times Z_{2}$

represented by , , and
$ia_{0}$ $ia_{1}$ $h_{0}$ $I(\overline{\Gamma}, \overline{G})=SO_{2}\backslash GL_{2}R=R^{+}\times H\times Z_{2}$

represented by and . $b_{0}$ $b_{1}$

In case (II) we can assume that

$S=\{(0,0), (m_{1}, a_{1}, b_{1}), \ldots, (m_{r}, a_{r}, b_{r})\}$

 Geometric Structures on the Seifert 4-Manifolds 343

with $r=3$ or 4 and

$\Gamma=\{q_{1}$ , $\ldots$ , $q_{r}$ , ,

$\ell$
$h|[\ell, h]=[q_{i}, \ell]=[q_{i}, h]=1$ ,

$q_{i}^{m_{i}}\ell^{a_{i}}h^{b_{i}}=1$
, $\prod q_{i}=1\}$ .

Here if $r=3$ , then $(m_{1}, m_{2}, m_{3})=(2,4,4)$ , (2, 3, 6), or (3, 3, 3) and
if $r=4$ then $(m_{1}, m_{2}, m_{3}, m_{4})=(2,2,2,2)$ . Furthermore in this case
$X=E^{4}$ or Nil3 . In either case the subgroup
$\times E$
generated by and $\Gamma_{0}$
$\ell$

$h$
are the center of and the exact sequence $\Gamma$ $1\rightarrow\Gamma_{0}\rightarrow\Gamma\rightarrow\overline{\Gamma}\rightarrow 1$

(where ) yields the original Seifert fibration. First suppose

$\overline{\Gamma}=\pi_{1}^{orb}B$

that $X=E^{4}$ , $G=R^{4}>\triangleleft SO_{4}$ . Pick up . Then the holonomy $\rho\in \mathcal{R}(\Gamma, G)$

group of which is the image

$\rho(\Gamma)$
of under the natural map $\overline{\rho(\Gamma)}$
$\rho(\Gamma)$

$G\rightarrow SO_{4}$ is cyclic (since is dieomorphic to a hyperelliptic surface). $S$

Since must be contained in the translation parts of $G$ we can

$\rho(\Gamma_{0})$

assume that there is a decomposition $C\times C$ of such that acts $E^{4}$ $\rho(\Gamma_{0})$

trivially on the first factor. Since commutes with any element in $\rho(\Gamma_{0})$

we can see that any element of

$\rho(\Gamma)$
is contained in $\overline{\rho(\Gamma)}$
$ SO_{2}\times 1\subset$

$SO_{2}\times SO_{2}\subset SO_{4}$ . If we take another , then there exists $\rho\in \mathcal{R}(\Gamma, G)$

$\sigma\in SO_{4}$ such that satisfies the above condition for the same
$\sigma\rho(\Gamma_{0})\sigma^{-1}$

decomposition of and hence the image of $E^{4}$

under the above $\sigma\rho(\Gamma)\sigma^{-1}$

map is also contained in the same subgroup $SO_{2}\times 1\subset SO_{4}$ . Therefore it
suces to consider the representation satisfying the above conditions $\rho$

for the fixed decomposition of . Thus projects to where $E^{4}$

$\rho$
$\overline{\rho}\in \mathcal{R}(\overline{\Gamma}, \overline{G})$

$\overline{G}=Isom^{0}E^{2}$
and we must have

$\rho(\ell)(z, w)=(z, w+\ell_{0})$

$\rho(h)(z, w)=(z, w+h_{0})$

$\rho(q_{i})(z, w)=(\overline{\rho}(\overline{q_{i}})z, w+s_{i})$ , $(1\leq i\leq r)$

where and are linearly independent over $R$ ,

$\ell_{0}$ $h_{0}$
is the image $\overline{q_{b}i}\in\overline{\Gamma}$

of under the projection

$q_{i}$ , C. Then from the relation $\Gamma\rightarrow\overline{\Gamma}$
$ s_{i}\in$

we deduce $s_{i}=-(a_{i}\ell_{0}+b_{i}h_{0})/m_{i}$ . Next we must see

$q_{i}^{m_{i}}\ell^{a_{i}}h^{b_{i}}=1$

exactly when the two representations and of the above forms are $\rho$
$\rho$

in the same orbit under the action of Inn . Suppose that there exists
$G$

$\sigma\in G$
such that with ,
$\sigma\rho\sigma^{-1}=\rho$
, $s_{0}\in R^{4}$ $\sigma x=\overline{\sigma}x+s_{0}$ $\overline{\sigma}\in SO_{4}$

and $x\in R^{4}$ . Then we can see that where , $\overline{\sigma}x=(\sigma_{1}z, \sigma_{2}w)$ $\sigma_{1}$
$\sigma_{2}\in SO_{2}$

or , $\sigma_{2}\in O_{2}-SO_{2}$ and $x=(z, w)$ with , $ w\in$ C. Suppose that

$\sigma_{1}$ $z$

$\sigma(z, w)=(\sigma_{1}z+a_{0}, \sigma_{2}w+b_{0})$ with , , , $b_{0}\in C$ . Then $\sigma_{1}$

$\sigma_{2}\in SO_{2}$ $a_{0}$
$\rho$
344 M. Ue

satisfies
$\rho(\ell)(z, w)=(z, w+\sigma_{2}\ell_{0})$

$\rho(h)(z, w)=(z, w+\sigma_{2}h_{0})$

$\rho(q_{i})(z, w)=(\sigma_{1}\overline{\rho}(\overline{q_{?}.})(\sigma_{1}^{-1}(z-a_{0}))+a_{0}, w+\sigma_{2}s_{i})$

.

Next suppose that $\sigma(z, w)=(\sigma_{1}\overline{z}+a_{0}, \sigma_{2}\overline{w}+b_{0})$

with $\sigma_{1}$
, $\sigma_{2}\in SO_{2}$ ,
, $b_{0}\in C$ . Then
$a_{0}$

$\rho(\ell)(z, w)=(z, w+\sigma_{2}\overline{\ell_{0}})$

$\rho(h)(z, w)=(z, w+\sigma_{2}\overline{h_{0}})$

$\rho(q_{i})(z, w)=(\overline{\rho(q_{i})}(z), w+\sigma_{2}\overline{s_{i}})$

where is in the component of

$\overline{\rho(q_{i})}$
dierent from that contain- $\mathcal{R}(\overline{\Gamma}, \overline{G})$

ing . On the other hand the identity component

$\overline{\rho}$

of $\mathcal{R}^{0}(\overline{\Gamma}, \overline{G})$ $\mathcal{R}(\overline{\Gamma}, \overline{G})$

is homeomorphic to $R^{2(r-3)}\times R^{+}$

where the first factor coincides with
the Teichm\"uller space for the flat structures of area 1 of the base orb-
ifold and the second factor corresponds to the area of $B$ . Thus we have
$B$

where $R^{+}\times H\times Z_{2}=SO_{2}\backslash GL_{2}R$

$\mathcal{T}(\Gamma, G)=R^{2(r-3)}\times R^{+}\times R^{+}\times H\times Z_{2}$

corresponds to the deformations of the lattice of the general fiber.

On the other hand in case (II) with $X=Ni1^{3}\times E$ the natural projec-
tion represented by
$X\rightarrow C$
for the coordinates defined above $(w, z)\rightarrow z$

yields a given fibration for $S=\{(a, b), (m_{1}, a_{1}, b_{1}), \ldots, (m_{r}, a_{r}, b_{r})\}$ .
(In fact we can assume that $a=b=0.$ ) Here we can assume that
$a+\sum a_{i}/m_{i}=0$ , $b+\sum b_{i}/m_{i}\neq 0$ . Take an arbitrary representation

. Then induces a representation

$\rho\in \mathcal{R}(\Gamma, G)$
where $\rho$
$\overline{\rho}\in \mathcal{R}(\overline{\Gamma}, \overline{G})$
$\overline{\Gamma}=$

and .
$\{\overline{q_{1}}, \ldots, \overline{q_{r}}|\overline{q}_{1}^{m_{1}}=\cdots=\overline{q}_{r}^{m_{r}}=\overline{q}_{1}\cdots\overline{q}_{r}=1\}$ $\rho(\overline{\Gamma})\subset\overline{G}=Isom^{0}E^{2}$

Here and $(m_{1}, \ldots, m_{r})$

$r$
satisfy the same conditions as in the case with
$X=E^{4}$ . Each has the following representation: $\overline{\rho}(\overline{q_{i}})$

$\overline{\rho}(\overline{q_{j}})z=\rho_{j}(z-z_{j})+z_{j}$

for , $z_{j}\in C$ such that the order of

$\rho_{j}\in S^{1}$
$\rho_{j}$ is $m_{j}$ and $\prod\rho_{j}=1$ . Thus
must be of the following form:
$\rho$

$\rho(\ell)(w, z)=(w+\ell_{0}, z)$

$\rho(h)(w, z)=(w+h_{0}, z)$

$\rho(q_{j})(w, z)=(w+w_{j}+i\overline{z_{j}}(z-\rho_{j}(z-z_{j})), \overline{\rho}(\overline{q_{j}})z)$

where $\ell_{0}$
and $h_{0}$
are linearly independent over $R$ , $w_{j}\in C$ . Then we have

$\rho(q_{j}^{m_{j}})(w, z)=(w+m_{j}(w_{j}+i|z_{j}|^{2}), z)$ .

 Geometric Structures on the Seifert 4-Manifolds 345

Thus form $q_{j}^{m_{j}}\ell^{a_{j}}h^{b_{j}}=1$

we deduce

(1) $w_{j}=-i|z_{j}|^{2}-a_{j}\ell_{0}/m_{j}-b_{j}h_{0}/m_{j}$

Suppose that $r=3$ . Then from $\prod\overline{q_{j}}=1$

we have
(2) $(1-\rho_{1})z_{1}+(\rho_{1}-\rho_{1}\rho_{2})z_{2}+(\rho_{1}\rho_{2}-\rho_{1}\rho_{2}\rho_{3})z_{3}=0$ .

Thus we can see that

$\rho(q_{1}q_{2}q_{3})(w, z)=(w+\sum_{j=1}^{3}w_{j}+iU, z)$

where
$U=\sum\rho_{j}|z_{j}|^{2}+\overline{z_{2}}z_{3}(1-\rho_{2}-\rho_{3}+\rho_{2}\rho_{3})+\overline{z_{1}}z_{2}(1-\rho_{1}-\rho_{2}+\rho_{1}\rho_{2})+$

$\overline{z_{1}}z_{3}(1-\overline{\rho}_{1}-\overline{\rho}_{3}+\overline{\rho}_{1}\overline{\rho}_{3})$
.
Thus from (1) $\rho$
is well defined if and only if

(3) $iV=(a+\sum a_{j}/m_{j})\ell_{0}+(b+\sum b_{j}/m_{j})h_{0}$

where $V=U-\sum|z_{j}|^{2}$ .

Claim. iV is a nonzero real number.

Easy computation (using (2)) shows that $iV$ is invariant under trans-
lations along the real line for $(z_{1}, z_{2}, z_{3})\rightarrow(z_{1}+\lambda, z_{2}+\lambda, z_{3}+\lambda)$ $\lambda\in R$

and the rotations in the origin. Therefore to prove Claim we can as-
sume that $z_{1}=0$ and is a nonzero real number . Then again by $z_{2}$ $r$

easy computation we can see that $V=r^{2}(\rho_{1}-1)(\rho_{2}-1)/(\rho_{1}\rho_{2}-1)$

and $V$ is a nonzero pure imaginary number. Thus we deduce that
$h_{0}=iV/(b+\sum b_{j}/m_{j})\neq 0$ from the normalization $a+\sum a_{j}/m_{j}=0$ .

Then is an arbitrary number $with\propto s\ell_{0}\neq 0$ . The case with $r=4$

$\ell_{0}$

( $m_{i}=2$ , $\rho_{j}=-1$ for all ) can be treated by a similar computation $j$

and we can see that is some fixed real number and $h_{0}$
is also an ar- $\ell_{0}$

bitrary number with . In any case can be lifted to

$s^{\propto}\ell_{0}\neq 0$ $\overline{\rho}\in \mathcal{R}(\overline{\Gamma}, \overline{G})$

some and we have

$\rho\in \mathcal{R}(\Gamma, G)$
where the $\mathcal{R}(\Gamma, G)=\mathcal{R}(\overline{\Gamma}, \overline{G})\times H\times Z_{2}$

factor $H\times Z_{2}$ corresponds to . The action of Inn $G$ on can $\ell_{0}$ $\mathcal{R}(\Gamma, G)$

be determined as in case (1-2). For any element the conjugation $\gamma\in G$

by acts on the factor

$\gamma$
as the inner automorphism of and acts
$\overline{\rho}(\overline{q_{i}})$
$\overline{G}$

trivially on , . Since these parameters determine the remaining

$\rho(\ell)$ $\rho(h)$

ones uniquely and the natural map Inn $G\rightarrow Inn$ is surjective we have $\overline{G}$

$\mathcal{T}(\Gamma, G)=\mathcal{T}(\overline{\Gamma}, \overline{G})\times H\times Z_{2}$

.
346 M. Ue

The proofs for the other cases are done by similar methods and hence
are omitted.

\S 3. The Teichm\"uller spaces and the complex structures

In this section we consider the Seifert 4-manifolds over closed ori-

entable hyperbolic or euclidean 2-orbifolds which admit complex struc-
tures. The arguments in [12, \S 7] show that any elliptic surface with $S$

$c_{2}=0$ and with $\kappa=0$ or 1 is biholomorphic to where a geometry $\Gamma\backslash X$

$X$ has a complex structure such that any element of $G=Isom^{0}X$ acts

on $X$ as a biholomorphism and . (We need to restrict $G$ to $U(2)$ $\Gamma\subset G$

when $X=E^{4}.$ ) Here we start with such a Seifert 4-manifold $S=\Gamma\backslash X$

with a given compatible complex structure. Let be the genus of $B$ and $g$

$r$
be the number of the cone points (with the prescribed cone angles) of
. In this section let
$B=\overline{\Gamma}\backslash \overline{X}$
and be the $\mathcal{R}=\mathcal{R}^{0}(\Gamma, G)$ $\mathcal{T}=\mathcal{T}^{0}(\Gamma, G)$

connected components of and of containing $\mathcal{R}(\Gamma, G)$ $\mathcal{T}(\Gamma, G)$ $S=\Gamma\backslash X$

with given geometric structure and its equivalence class $\rho_{0}\in \mathcal{R}$ $[\rho_{0}]\in \mathcal{T}$

respectively.
First suppose that $B$ is hyperbolic. In this case $X=H^{2}\times E^{2}$
or each of which is identified with $H\times C$ as in \S 1. Here-
$\overline{SL_{2}}\times E$

after we adopt the same notations for , and as in \S 1 and we $\Gamma$

$\overline{\Gamma}$

$S$

assume that the monodromies of satisfy the conditions in Proposi- $S$

tion 1. Now we will describe as a dierentiable family of the $\mathcal{T}^{0}(\Gamma, G)$

complex structures on . Let be the Seifert 4-manifold with the $S$ $S_{\rho}$ $S$

geometric structure corresponding to and be its equivalence $\rho\in \mathcal{R}$ $[\rho]$

class in respectively. For simplicity put $\mathcal{T}$

and let be the $S_{0}=S_{\rho o}$ $\Theta_{0}$

sheaf of germs of holomorphic tangent vector fields on . First recall $S_{0}$

that . Here is (the connected component

$\mathcal{T}=\overline{I}\times T_{0}\times H^{1}(\overline{\Gamma}, C^{\phi})$
$\overline{\mathcal{T}}$

of) the Teichm\"uller space of or equivalently of the Fuchsian $B=\overline{\Gamma}\backslash H$

group , is the monodromy representation for , and

$\overline{\Gamma}$

$\phi$
is the identity $S$ $T_{0}$

component of in \S 1. Put $T_{1}$

$H$ if $\phi\equiv\pm id$

$\overline{T}_{0}=\{$

1 otherwise

and let . Then denotes the Teichm\"uller space

$\overline{I}=\overline{I}\times\overline{T}_{0}\times H^{1}(\overline{\Gamma}, C^{\phi})$
$\overline{\mathcal{T}}$

of the geometric structures which fix the area of the general fiber (if $X=$
. On the other hand the Teichm\"uller space $T$ for the Fuchsian
$H^{2}\times E^{2})$

group is realized as a bounded domain in $C^{3g-3+r}$ . Moreover there is

$\overline{\Gamma}$

a fiber space over where $D(\tau)$ is a

$\overline{C}=\{(z, \tau)\in C\times\overline{\mathcal{T}}|z\in D(\tau)\}$
$\overline{\mathcal{T}}$

domain in $C$ on which a quasi-Fuchsian group corresponding to acts $\overline{\Gamma}^{\tau}$

$\tau$
 Geometric Structures on the Seifert 4-Manifolds 347

([1]). has a complex structure such that the map

$\overline{C}$

$(z, \tau)\rightarrow(\tau(\alpha)z, \tau)$

for any is holomorphic. Here acts on $z\in D(\tau)$ via

$\alpha\in\overline{\Gamma}$
. It $\tau(\alpha)$
$\overline{\Gamma}^{\tau}$

follows that for any we have a biholomorphism : and $\tau\in\overline{I}$ $h_{\tau}$ $H\rightarrow D(\tau)$

a representation with such that $\overline{\rho}\in\overline{\mathcal{R}}$

$[\overline{\rho}]=\tau$ $\tau(\alpha)\cdot h_{\tau}(z)=h_{\tau}(\overline{\rho}(\alpha)z)$

for any $z\in H$ , . $\alpha\in\overline{\Gamma}$

If $X=H^{2}\times E^{2}$ then is lifted to such $h_{\tau}$ $h_{\tau}\times id:H\times C\rightarrow D(\tau)\times C$

that it commutes with the translations in the -factor. On the other

$C$

hand we can choose the elements , , of which maps $m_{1}$ $\ldots$ $m_{d}$
$C^{1}(\overline{\Gamma}, C^{\phi})$

to the basis of where . Thus we have the

$H^{1}(\overline{\Gamma}, C^{\phi})$ $d=dim_{C}H^{1}(\overline{\Gamma}, C^{\phi})$

family of representations of on as follows. Let $\Gamma$ $\overline{C}\times C\times C^{d}\times T_{0}$ $\tau_{0}\in\overline{I}$

be the equivalence class of the element determined by . Then $\overline{\rho_{0}}\in\overline{\mathcal{R}}$

$\rho_{0}$

is biholomorphic to
$S_{0}$
where is represented (via ) $\overline{\Gamma}^{\tau_{0}}\backslash (D(\tau_{0})\times C)$
$\rho_{0}$
$\overline{\Gamma}^{\tau_{0}}$

as follows;
$\rho_{0}(\ell)(z, w)=(z, w+r_{0})$ ,
$\rho_{0}(h)(z, w)=(z, w+r_{0}h_{0})$ ,
$\rho_{0}(\alpha_{i})(z, w)=(\tau_{0}(\overline{\alpha_{i}})z, \phi(\overline{\alpha_{i}})w+w_{j}^{0})$
,
$\rho_{0}(q_{i})(z, w)=(\tau_{0}(\overline{q_{\dot{\iota}}})z, w-a_{i}r_{0}/m_{i}-b_{i}r_{0}h_{0}/m_{i})$

where $(z, w)\in D(\tau_{0})\times C$ , $r_{0}\in R^{+}$ , $h_{0}\in H$ and is defined as $w_{j}^{0}\in C$

in the proof of Theorem A. Then we have the following representations

$\rho=\rho(\tau, r, h, s)$ from to the group of the biholomorphisms of $D(\tau)\times C$ $\overline{\Gamma}$

where $s=(s_{1}, \ldots, s_{d})\in C^{d}$ , , $(r, h)\in T_{0}$ ( $h=h_{0}$ if ). $\tau\in\overline{I}$
$\phi\not\equiv\pm id$

$\rho(\ell)(z, w)=(z, w+r)$ ,

$\rho(h)(z, w)=(z, w+rh)$ ,

$\rho(\alpha_{i})(z, w)=(\tau(\overline{\alpha_{i}})z, \phi(\overline{\alpha_{i}})w+w_{j}^{0}+\sum_{j=1}^{d}s_{j}m_{j}(\overline{\alpha_{i}}))$ ,

$\rho(q_{i})(z, w)=(\tau(\overline{q_{i}})z, w-a_{i}r/m_{i}-b_{i}rh/m_{i})$ .

Here is identified with . Thus we get the fiber space over

$H^{1}(\overline{\Gamma}, C^{\phi})$ $C^{d}$ $C$

$T$ obtained from by the actions of defined above $\overline{C}\times C\times T_{0}\times H^{1}(\overline{\Gamma}, C^{\phi})$
$\rho$

such that the fiber of over is an elliptic surface corresponding $C$ $\tilde{\tau}\in I$

to . Also we have the fiber space over

$\tilde{\tau}$

by restricting the above $\overline{C}$ $\overline{\mathcal{T}}$

representations to the cases with $r=r_{0}$ (the constant). The above

representation depends holomorphically on all the parameters except for
$r\in R^{+}$ (if has the -factor). Therefore we have the dierentiate
$T_{0}$ $R^{+}$

family and the complex analytic family

$C$
of the complex
$\rightarrow \mathcal{T}$
$\overline{C}\rightarrow\overline{I}$

structures on respectively. $S$

348 M. Ue

If then $X$ is identified with $H\times C$ with coordinates

$X=\overline{SL_{2}}\times E$

$(z, w)$ so that $w$ corresponds to $\log dz$ . Hence in this case : $h_{\tau}$ $H\rightarrow D(\tau)$

is lifted to the biholomorphism : $H\times C\rightarrow D(\tau)\times C$ defined by $h_{\tau}$

where the branch of the

$\overline{h_{\tau}}(z, w)=(h_{\tau}(z), w+\log(\partial h_{\tau}/\partial z)(z))$
is $\log$

chosen so that depends holomorphically on . Again com-

$\overline{h_{\tau}}$
$\tau\in\overline{I}$ $\overline{h_{\tau}}$

mutes with the translation $(z, w)\rightarrow(z, w+c)$ with $c=a$ constant.
On the other hand we can define for $\tilde{\rho}(\alpha):H\times C\rightarrow H\times C$ $\alpha\in\Gamma$

such that (where is hyperbolic), (where

$\overline{\alpha}_{i}=\tilde{\rho}(\alpha_{i})$
$\overline{\alpha}_{i}\in\overline{\Gamma}$
$\overline{q_{i}}=\tilde{\rho}(q_{i})$

is elliptic) satisfy the same conditions as

$\overline{q}_{i}\in\overline{\Gamma}$
stated in the $\overline{\alpha}_{i},\overline{q_{i}}$

proof of Theorem A. Thus for with , we can define $\overline{\rho}\in\overline{\mathcal{R}}$ $[\overline{\rho}]=\tau\in\overline{\mathcal{T}}$

for or
$\tilde{\tau}(\alpha):D(\tau)\times C\rightarrow D(\tau)\times C$which covers such $\alpha=\alpha_{i}$ $q_{i}$
$\tau(\overline{\alpha})$

that where is the image of in . Using this lift we

$\overline{h_{\tau}}\tilde{\tau}(\alpha)=\tilde{\rho}(\alpha)\overline{h_{\tau}}$ $\overline{\alpha}$
$\alpha$
$\overline{\Gamma}$

can define the family of representations parametrized by (in this case $\mathcal{T}$

since there is no $R$-factor in T) as in the case with $X=H^{2}\times E^{2}$ .

$\overline{\mathcal{T}}=I$

Thus we also get the analogous family of complex structures on . $S$

Next we will describe the Kodaira-Spencers infinitesimal deforma-

tion map
$\Phi:T_{0}I\rightarrow H^{1}(S_{0}, \Theta_{0})$

where $T_{0}I$ is the tangent space of at (or equivalently at ). $\mathcal{T}$

$S_{0}$ $[\rho_{0}]$

Since the base orbifold of is hyperbolic in our case here is homeo-

$B$ $T$ $S$

morphic to a euclidean space and is homeomorphic to $T_{0}I$ . To describe

we recall some results in [8]. Let
$H^{1}(S_{0}, \Theta_{0})$
be the complex torus $T^{1}$

of dimension 1. For a holomorphic Seifert fibering with $S=\Gamma_{\rho}\backslash X$

, the base orbifold

$\Gamma_{\rho}=\rho(\Gamma)\subset G$ $B$ is naturally a nonsingular curve $B_{\rho}$

of the form where or $C$ , . (In [8], , , $B$ ,

$\overline{\Gamma}_{\rho}\backslash \overline{X}$
$\overline{X}=H$ $\overline{\Gamma}_{\rho}=\pi_{1}^{orb}B$ $S$ $\overline{X}$ $\overline{\Gamma}$

are denoted by $M$ , $W$, $V$, $N$ respectively.) Then induced by the $\overline{S}\rightarrow\overline{X}$

covering projection is a principal bundle and , $\overline{X}\rightarrow B$ $T^{1}-$ $\overline{S}=\overline{X}\times T^{1}([8$

1, 7] in which is denoted by $B$ ). Let , , be the sheaves over $\overline{S}$

$Z^{2}$ $\mathcal{O}$
$\mathcal{T}^{1}$ $\overline{X}$

of germs of local holomorphic maps from into , $C$ , respectively. $\overline{X}$ $Z^{2}$ $T^{1}$

Then the action of on is defined by the element . $\overline{\Gamma}$

$\overline{S}$

$m\in H^{1}(\overline{\Gamma}, I^{1})$

Here where the coecients

$H^{1}(\overline{\Gamma}, \mathcal{T}^{1})=H^{1}(\overline{\Gamma}, H^{0}(\overline{X}, I^{1})^{\phi})$ $H^{0}(\overline{X}, \mathcal{T}^{1})$

on the right hand side is the space of global holomorphic maps on $\overline{X}$

and is twisted by the monodromy representation C. The el- $\phi:\overline{\Gamma}\rightarrow GL_{1}$

ement $m$ is represented by a crossed homomorphism $m:\overline{\Gamma}\rightarrow H^{0}(\overline{X}, I^{1})$

such that $m(x, \alpha)$ for a fixed is a holomorphic map from to $\alpha\in\overline{\Gamma}$ $\overline{X}$
$T^{1}$

satisfying
$(x\in\overline{X})$

$m(x, \alpha\beta)=\phi(\alpha)m(\alpha^{-1}x, \beta)+m(x, \alpha)$ for $\alpha$

, $\beta\in\overline{\Gamma}$

where $\overline{\Gamma}$

acts on $\overline{X}$

via $\overline{\rho}$

and $\phi(\alpha)$
gives the automorphism of $T^{1}$
since
 Geometric Structures on the Seifert 4-Manifolds 349

it preserves the lattice of the fiber. The action of $\overline{\Gamma}$

on $\overline{S}=\overline{X}\times T^{1}$
is
given by

$\alpha(x, t)=(\alpha x, \phi(\alpha)t+m(\alpha x, \alpha))$ for $\alpha\in\overline{\Gamma}$

, $x\in\overline{X}$
, $t\in T^{1}$

(see [8, 7.2]). From the exact sequence

$0-Z^{2}\rightarrow \mathcal{O}\rightarrow I^{1}\rightarrow 0$

we have the exact sequence

$H^{1}(\overline{\Gamma}, H^{0}(\overline{X}, \mathcal{O})^{\phi})\rightarrow H^{1}\eta(\overline{\Gamma}, H^{0}(\overline{X}, \mathcal{T}^{1})^{\phi})\rightarrow H^{2}(\overline{\Gamma}, Z^{2^{\phi}})c$

where $c(m)$ represents the extension . On the $1\rightarrow Z^{2}\rightarrow\Gamma\rightarrow\overline{\Gamma}\rightarrow 1$

other hand is described by the following exact sequences

$H^{1}(S_{0}, \Theta_{0})$
([8,
\S 2]).
(1) 0 $D\rightarrow H^{1}(S_{0}, \ominus_{0})\rightarrow G\rightarrow 0$

(2) $O\rightarrow F\rightarrow G\rightarrow H\rightarrow 0$

(3) 0 $C\rightarrow D\rightarrow E\rightarrow 0$

where the exact sequence (3) splits ([8, 4]). Here

(4) $E=$ { $a\in C;a\overline{\phi(\alpha)}=\phi(\alpha)a$

for all $\alpha\in\overline{\Gamma}$

}
corresponds to fiber deformations ([8, Theorem 7.10, 4]). Note that in
our cases is a root of unity for any when the base is hyperbolic
$\phi(\alpha)$ $\alpha$

(1) and it suces to consider the cases with trivial monodromies when
the base is not hyperbolic (in the case of euclidean base orbifolds, we
have only to consider the cases (1-1), (1-2), (II-1) and (II-2) in \S 2). Thus
the kernel of has finite index in and then the assumption in [8, \S 7]
$\phi$
$\overline{\Gamma}$

is automatically satisfied. Hence we have

$C$ , if $\phi(\alpha)=\pm 1$ for any $\alpha\in\Gamma$

$E=\{$
0, otherwise.

The subspace $C$ in (3) corresponds to the twist deformations com-

ing from the the complex analytic family of the form $ m+\eta(s\ell)\in$
for , $s\in C([8, \S 3])$ . In our
$H^{1}(\overline{\Gamma}, H^{0}(\overline{X}, \mathcal{T}^{1})^{\phi})$ $\ell\in H^{1}(\overline{\Gamma}, H^{0}(\overline{X}, \mathcal{O})^{\phi})$

case by [8, \S 7] we have

(5) $C=H^{1}(\overline{\Gamma}, H^{0}(\overline{X}, \mathcal{O})^{\phi})$

350 M. Ue

unless $g=1$ , $r=0$ . The subspace $F$ in (2) corresponds to the base
deformations and by [8, \S 7], we have

(6) $F=H^{1}(\overline{\Gamma}, \Theta_{\overline{X}})=H^{1}(B, \Theta_{B|d})$

where $B=B_{\rho o}$ is considered as a nonsingular curve, $d=\sum_{i=1}^{r}p_{i}$ is the

divisor corresponding to the cone point and is the sheaf of germs $p_{i}$ $\Theta_{B|d}$

of holomorphic tangent vector fields on $B$ which vanish on . Finally the $d$

element of the space $H$ represents (if it is not obstructed) a deformation

of which destroys the fiber structure. By [8, Theorem 7.13] we have
$S$

(7) $H=0$ unless $g=1$ , $r=0$ or $g=0$ , $r<3$ .

Now we consider the Kodaira Spencer map for $\Phi:T_{0}\mathcal{T}\rightarrow H^{1}(S_{0}, \Theta_{0})$

the case with $X=H^{2}\times E^{2}$ . The tangent space is homeomorphic $T_{0}\mathcal{T}$

to where and
$T_{0}\overline{\mathcal{T}}\times \mathcal{F}_{0}$
$T_{0}=R^{+}\times H$ if ,
$\mathcal{F}_{0}=T_{0}\times H^{1}(\overline{\Gamma}, C^{\phi})$ $\phi\equiv\pm id$

$T_{0}=R^{+}$ if $\phi\neq\pm id$ . The derivatives of the family of representations

defined above span and the discussions in \S 3-\S 5 in [8] show that
$T_{0}\mathcal{T}$

the Kodaira Spencer map preserves the fiber structure as is indicated

by the following commutative diagram.

$T_{0}\mathcal{T}\rightarrow\Phi H^{1}(S_{0}, \Theta_{0})$

(8) $\downarrow$ $\downarrow$

$T_{0}\overline{\mathcal{T}}\rightarrow\overline{\Phi}$

$F$

Here the vertical maps are the projections of the fibrations and gives $\overline{\Phi}$

the infinitesimal deformation map for the Teichm\"uller space of the r-

pointed Riemann surface of genus at $B=B_{\rho o}$ . We have $g$
$T_{0}\overline{\mathcal{T}}=$

$R^{2(3g-3+r)}([6])$ , $dim_{C}F=3g-3+r$ ( , Lemma 7.3]) and is a $[8$

$\overline{\Phi}$

homeomorphism. Here we note that if two geometric Seifert 4-manifolds

,
$S=\Gamma\backslash X$ (with $X=H\times C$ ) over the hyperbolic 2-orbifolds
$S=\Gamma\backslash X$

$B$ , $B$ are biholomorphic then $B$ and $B$ are isometric. For, any biholo-

morphism is lifted to a biholomorphism from $H\times C$ to itself

$\varphi:S\rightarrow S$ $\tilde{\varphi}$

such that there is an automorphism satisfying $\psi:\Gamma\rightarrow\Gamma$ $\tilde{\varphi}(\gamma(z, w))=$

for $(z, w)\in H\times C$ ,

$\psi(\gamma)(\tilde{\varphi}(z, w))$
. Here and have the exact $\gamma\in\Gamma$ $\Gamma$ $\Gamma$

sequences and such

$1\rightarrow\Gamma_{0}\rightarrow\Gamma\rightarrow\overline{\Gamma}\rightarrow 1$ $1\rightarrow\Gamma_{0}\rightarrow\Gamma\rightarrow\overline{\Gamma}^{J}\rightarrow 1$

that , and , correspond to the fundamental

$\overline{\Gamma}=\pi_{1}^{orb}B$ $\overline{\Gamma}=\pi_{1}^{orb}B$ $\Gamma_{0}$ $\Gamma_{0}$

groups of the general fibers respectively. Moreover induces the isomor- $\psi$

phism between and and also induces the isomorphism

$\Gamma_{0}$
. $\Gamma_{0}$
$\overline{\psi}:\overline{\Gamma}\rightarrow\overline{\Gamma}$

On the other hand and act on $H\times C$ by the translations in the

$\Gamma_{0}$ $\Gamma_{0}$

$C$ -factor since and are geometric (of type $S$

or ). $S$ $H^{2}\times E^{2}$ $\overline{SL}_{2}\times E$
 Geometric Structures on the Seifert 4-Manifolds 351

Hence if we write $\tilde{\varphi}(z, w)=(h(z, w),$ $k(z, w))\in H\times C$ , then $h(z, w)$ is
invariant under the action of in the $w$ -coordinate. Since has rank $\Gamma_{0}$ $\Gamma_{0}$

2, $h(z, w)$ depends only on , i.e., $h(z, w)=h(z)$ which gives a biholo- $z$

morphism from $H$ to itself and descends to an isometry $h:H\rightarrow H$ . $\tilde{\varphi}$

Since we have for $z\in H$ , we can see that $B$

$h(\overline{\gamma}(z))=\overline{\psi}(\overline{\gamma})h(z)$ $\overline{\gamma}\in\overline{\Gamma}$

and $B$ are isometric. Thus by the fact that the action of Aut on the $\overline{\Gamma}$

Teichm\"uller space of $B$ is properly discontinuous we can see directly that

the kernel of induced from above is zero. Thus is an isomorphism
$\overline{\Phi}$
$\Phi$
$\overline{\Phi}$

(compare the dimensions of the spaces in (8)). Moreover induces the $\Phi$

map between the fibers of the projections in (8) of the form

(9) $H^{1}(\overline{\Gamma}, C^{\phi})\times T_{0}\rightarrow C\Phi_{1}\times\Phi_{2}\times E$

.

If then $E=C$ such that small $s\in C$ determines the complex

$\phi\equiv\pm id$

structure of the general fiber whose period matrix is given by $\Omega(s)=$

$(1 +s, h_{0}+s\overline{h_{0}})$ where corresponds to that for the original $\Omega(0)$ $S_{0}$

([8, Lemma 4.5] ). We have the representations parametrized by $\rho$ $s$

in the above family whose -component $h(s)$ satisfies $h(s)=(h_{0}+$ $T_{0}$

$s\overline{h_{0}})/(1+s)$ and corresponds to 1 ([8, 4]). Thus maps

$\Phi(\partial/\partial s)$ $\Phi_{2}$

$T_{0}$
onto $E$ whose kernel is the -component of represented by the $R^{+}$ $T_{0}$

parameter detecting the deformation of the area of the general fiber. On

the other hand and is the map
$C=H^{1}(\overline{\Gamma}, H^{0}(\overline{X}, \mathcal{O})^{\phi})$ $\Phi_{1}$ $ H^{1}(\overline{\Gamma}, C^{\phi})\rightarrow$

induced by the natural inclusion

$H^{1}(\overline{\Gamma}, H^{0}(\overline{X}, \mathcal{O})^{\phi})$
. $C\subset H^{0}(\overline{X}, \mathcal{O})$

Moreover by the naturality of the spectral sequences (used in [8]) we

have the following commutative diagram.

$\rightarrow\Phi_{1}H^{1}(\overline{\Gamma}, H^{0}(\overline{X}, \mathcal{O})^{\phi})$

$H^{1}(\overline{\Gamma}, C^{\phi})$

$\uparrow\varphi_{1}$ $\uparrow\varphi_{2}$

$\Phi$

$ H^{1}(B, C(Q))\rightarrow$ $H^{1}(B, \mathcal{O}(Q))$

Here $Q$ is the flat $C$ -bundle over $B=B_{\rho_{0}}$ determined by the mon-
odromy representation which can be considered as the representation $\phi$

of (see \S 1). The coecient $C(Q)$ (resp.

$\pi_{1}(B_{\rho o})$
) is the sheaf of $\mathcal{O}(Q)$

the germs of locally constant (resp. holomorphic) sections of $Q$ and $\varphi_{1}$

and are the isomorphisms ([8, 7]). The map

$\varphi_{2}$
is the part of the $\Phi$

following exact sequence (in which the base $B$ is omitted)

$0\rightarrow H^{0}(C(Q))\rightarrow H^{0}(\mathcal{O}(Q))\rightarrow H^{0}(\Omega^{1}(Q))$

$\Phi$

$\rightarrow H^{1}(C(Q))\rightarrow H^{1}(\mathcal{O}(Q))\rightarrow H^{1}(\Omega^{1}(Q))$

352 M. Ue

which comes from the exact sequence

$\partial$

0 $C(Q)\rightarrow \mathcal{O}(Q)\rightarrow\Omega^{1}(Q)\rightarrow 0$
.

Here is the sheaf of the germs of holomorphic 1 forms on the non-

$\Omega^{1}$

singular curve $B$ . By the Riemann-Roch theorem (and since $c_{1}(Q)=0$ )

dime $H^{0}(\Omega^{1}(Q))=$ dime equals if $\phi=id$ and equals $H^{1}(\mathcal{O}(Q))$ $g$

$-1$ otherwise (cf. [8, \S 7] ). Comparing this with $dim_{C}H^{1}(C(Q))=$

$g$

which equals $2g$ if

$dim_{C}H^{1}(\overline{\Gamma}, C^{\phi})$
and $2g$ $-2$ otherwise we $\phi\equiv id$

deduce the exact sequence

$\Phi$

$O\rightarrow H^{0}(\Omega^{1}(Q))\rightarrow H^{1}(C(Q))\rightarrow H^{1}(\mathcal{O}(Q))\rightarrow 0$

.

Hence the kernel of is isomorphic to $\Phi_{1}$ $H^{0}(B, \Omega^{1}(Q))$ . The same argu-
ment holds for the case with $X=\overline{SL_{2}}\times E$
except for the fact that the
kernel of is trivial since there is no
$\Phi_{2}$ $R^{+}$
-component in . Thus we $T_{0}$

have
Theorem C-l. The Kodaira Spencer map for the Teichm\"uller $\Phi$

space for the Seifert 4-manifold over the closed orientable hyper-
$\mathcal{T}$
$S$

bolic 2-orbifold $B$ with any given representation $T$ is surjective $\rho\in$

and the kernel of at $S=S_{\rho}$ is homeomorphic to

$\Phi:T_{0}\mathcal{T}\rightarrow H^{1}(S, \Theta)$

or $H^{0}(B, \Omega^{1}(Q))(ifX=\overline{SL_{2}}\times E)$

$H^{0}(B, \Omega^{1}(Q))\times R^{+}(ifX=H^{2}\times E^{2})$

for the base curve $B$ determined by . The subspace of defifined above $\rho$
$\mathcal{T}$ $\mathcal{T}$

gives a locally complete complex analytic family of the complex structures

on . $S$

The last statement comes from [4]. We can see directly that any
deformation in the subspace $H^{0}(B, \Omega^{1}(Q))$ of (which depends on $T_{0}\mathcal{T}$

the choice of ) does not change the complex structure as fol-

$\rho\in \mathcal{T}$

lows. Take any $w\in H^{0}(B, \Omega^{1}(Q))$ . Lift $w$ to the 1-form on $H$ which

is represented as for some holomorphic function

$ d\psi$
on $H$ satisfying $\psi$

$d\psi(\alpha z)=\phi(\alpha)d\psi(z)$ for any , $z\in H$ . Taking the integral we de- $\alpha\in\overline{\Gamma}$

duce that $b(\alpha)=\psi(\alpha z)-\phi(\alpha)\psi(z)$ is a constant. Furthermore we have

$b(\alpha\beta)=b(\alpha)+\phi(\alpha)b(\beta)$ . If is a torsion then we can choose the fixed $\alpha$

point of as and hence $b(\alpha)=0$ . The image of

$\alpha$ $z$
in $b(\alpha)$ $H^{1}(\overline{\Gamma}, C^{\phi})$

maps to 0 in since $b(\alpha)=\psi(z)-\phi(\alpha)\psi(\alpha^{-1}z)$ and

$H^{1}(\overline{\Gamma}, H^{0}(H, \mathcal{O})^{\phi})$

conversely any element in the kernel of can be represented by the $\Phi_{1}$

above way. Then the biholomorphic automorphism of $H\times C$ defined

by $(z, w)\rightarrow(z, w+s\psi(z))$ for $s\in C$ descends to the biholomorphism
between and such that the dierence
$S_{\rho}$ $S_{\rho}$
of the parameters $m_{\rho}-m_{\rho}$

in is and all the other parameters are the same.

$H^{1}(\overline{\Gamma}, C^{\phi})$ $sb$
 Geometric Structures on the Seifert 4-Manifolds 353

Next consider the case when $B$ is euclidean and has a complex $S$

structure. If $g=1$ , $r=0$ ( $B$ is a torus) then we can assume that $S$

is either (a complex torus) or a primary Kodaira surface. In the

$T^{4}$

first case $X=E^{4}$ , $G=Isom^{0}E^{4}$ . However since the complex structure

of $X$ is not preserved by $G$ but is preserved by $G=E^{4}\lambda U(2)$ we
consider $\mathcal{T}=the$ identity component of $I(\Gamma, G)$ in this case. Then
and this is realized by the family of translations
$\mathcal{T}=U(2)\backslash GL_{4}^{+}R$ $\rho$

defined by
$\rho(\alpha_{i})(z, w)=(z+w_{i1}, w+w_{i2})$

for the generators $\alpha_{i}$

, $(i=1, \ldots, 4)$ in $\Gamma=Z^{4}$
such that

$w_{12}=0$ , $w_{11}$ , $w_{22}\in R^{+}$ , $\det(\Re w_{ij}, \propto sw_{ij})>0$

where is the matrix of rank 4 defined by

$(\Re w_{ij}, \propto sw_{ij})$ for $\Re w_{ij},$ $\propto sw_{ij}$

$i,j=1$ , , 4. Thus we have a dierentiate family

$\ldots$ over of the $C$ $\mathcal{T}$

complex structures on S. contains the subfamily (which is complex $\mathcal{T}$

analytic) consisting of the representations with

$w_{11}=w_{22}=1$ , $w_{12}=w_{21}=0$ , $\det(\propto s(w_{ij})_{i,j=3,4})>0$

which is complete and eectively parametrized ([3]). It follows that the

Kodaira Spencer map for at any point is surjective. In the second case
$\mathcal{T}$

$X=Ni1^{3}\times E$
and is homeomorphic to
$\mathcal{T}$

. Here the $R^{+}-$$R^{+}\times H\times R^{2}\times H$

factor, the first $H$ -factor, the last $H$ -factor and the -factor correspond $R^{2}$

to the area of the base, the period of the base, the period of the fiber
(the image of one of the lattices of the fiber is uniquely determined and
not deformed) and the twisting parameters for the fibrations respectively
(see \S 2). Hence we have a dierentiate family of the complex $C$ $\rightarrow \mathcal{T}$

structures. On the other hand in the decomposition of we have $H^{1}(S, \ominus)$

$E=F=C([8])$ . Since the canonical divisor of is trivial there is $K$ $S$

an isomorphism and hence $dim_{C}H^{1}(S, \Theta)=h^{1,1}=2$ since the

$\Theta\cong\Omega^{1}$

Hodge numbers satisfy $h^{0,2}=h^{2,0}=1$ and $b_{2}=h^{2,0}+h^{0,2}+h^{1,1}=4$ .

It follows that $C=H=0$ and as in the cases when $B$ is hyperbolic the
Kodaira Spencer map is surjective with kernel $=R^{+}\times R^{2}$ .
Finally consider the case when $B$ is euclidean of genus 0. In this case
$\mathcal{T}=I_{0,r}\times H\times R^{+}$if $X=E^{4}$ and if $X=Ni1^{3}\times E$ . Here
$\mathcal{T}=I_{0,r}\times H$

$\mathcal{T}_{0,r}=R^{2(r-3)}\times R^{+}$ (with

$r=3,4$ ) denotes the Teichm\"uller space of
the base orbifold $B$ where the first factor corresponds to the Teichm\"uller
space of an -pointed Riemann surface of genus 0 (with fixed area) and
$r$

the -factor corresponds to the area of the base $B$ . The other factor
$R^{+}$

in $T$ corresponds to the deformations of the lattices of the fiber (in the
case with $X=Ni1^{3}\times E$ one of the lattices of the fiber has the fixed
354 M. Ue

image and hence there is no -factor). If $r=3$ then the base $B$ is $R^{+}$

parametrized by the area only and if $r=4$ then the $R^{2(r-3)}$ -factor is
identified with the Teichm\"uller space of the double covering torus of $B$
which is isomorphic to H. In either case we have the dierentiable family
$C$
of the complex structures of as in the arguments in \S 2. (In the
$\rightarrow \mathcal{T}$
$S$

case with $X=E^{4}$ , $=0$ we can choose the representatives for such $g$ $\rho$
$\mathcal{T}$

that the image of lies in and hence we do not need to restrict

$\rho$
$E^{4}>\triangleleft U(1)$

$G$ to the subgroup $U(2).)$ On the other hand in the decomposition $ E^{4}\lambda$

of we have $C=H=0$ , $E=C$ , $F=C^{r-3}$ and we can argue

$H^{1}(S_{0}, \Theta_{0})$

as in the case when $B$ is hyperbolic. Thus we obtain

Theorem C-2. Let be a Seifert 4-manifold over some orientable $S$

hyperbolic or euclidean 2-orbifold $B$ which admits a complex structure.

Then has a geometric structure of type $(X, G)$ with $X=H^{2}\times E^{2}$ ,
$S$

, or Nil3
$\overline{SL_{2}}\times E$
and $G=Isom^{0}$ X. Let I be the identity compo-
$E^{4}$ $\times E$

nent of the Teichm\"uller space $I(\Gamma, G)$ where X. (In the case when $\Gamma=\pi_{1}$

$S=T^{4}$ restrict $G$ to $U(2).)$ Then gives a dierentiable family $ E^{4}\lambda$ $\mathcal{T}$

of complex structures on such that the infifinitesimal deformation map $S$

at any point in I is surjective.

Remark. The statements in Theorem C-2 do not hold in general for

a Seifert 4-manifold over a closed orientable spherical or bad 2-orbifold $S$

$B$ . In this case is either a ruled surface of genus 1 (with $X=S^{2}\times E^{2}$ )
$S$

or a Hopf surface (with $X=S^{3}\times E$ ). In either case not every complex

structure on comes from the geometric one nor every dierentiable
$S$

family of the complex structures containing the geometric one comes

from the Teichm\"uller space of the geometric structures. In general the
dimension of $H^{1}(S, \Theta)$ (which is not constant) can be greater than that
of the Teichm\"uller space (cf. [9], [14], [2], [12]).

\S 4. A remark on the moduli spaces

In this section we give a remark on the moduli space for $\lambda\Lambda(\Gamma, G)$

a geometric Seifert 4-manifold over a closed orientable hyper- $S=\Gamma\backslash X$

bolic base orbifold with $\Gamma\subset G=Isom^{0}X$ . We adopt the

$B=\overline{\Gamma}\backslash \overline{X}$

representations of given in \S 1 and also assume that the monodromy $\Gamma$

matrices , , satisfy the conditions in Proposition 1. In this case

$A_{1}$
$\ldots$
$A_{2g}$

the fibration of is unique and then every element of Aut induces

$S$ $\varphi$
$\Gamma$

the automorphism of and also induces the automorphism of $\overline{\varphi}$

$\overline{\Gamma}$
$Z^{2}$

generated by , . Put $\ell$

. $h$ $\pi_{*}=\prod[\alpha_{2j-1}, \alpha_{2j}]\prod q_{j}$

Proposition 2. Any element $\varphi\in Aut\Gamma$ must be of the following

 Geometric Structures on the Seifert 4-Manifolds 355

form.
$\varphi(\alpha_{i})=\tilde{\varphi}(\alpha_{i})\ell^{s_{i}}h^{t_{i}}$

$\varphi(q_{j})=\tilde{\varphi}(q_{j})\ell^{u_{j}}h^{v_{j}}$

$(\varphi(\ell), \varphi(h))=(\ell, h)P$ .

Here $P\in GL_{2}Z$ and $\tilde{\varphi}(\alpha_{i}),\tilde{\varphi}(q_{j})$

are the words of $\alpha_{1}$
, $\ldots$ , $\alpha_{2g}$ , $q_{1}$ , $\ldots$ , $q_{r}$

satisfying

$\tilde{\varphi}(q_{j})=\mu_{j}q_{\nu(j)}^{\sigma}\mu_{j}^{-1}$

$\tilde{\varphi}(\pi_{*})=\mu\pi_{*}^{\sigma}\mu^{-1}$

where $\sigma=\pm 1$ are some words of , , , , , and

, $\mu$ , $\mu_{j}$
$\alpha_{1}$
$\ldots$ $\alpha_{2g}$ $q_{1}$ $\ldots$ $q_{r}$

$lJ:(1, \ldots, r)\rightarrow(l/(1), \ldots, \nu(r))$ is a permutation. We have further con-

ditions on the above parameters and the words as follows. Let , , $\epsilon_{i}$
$\eta_{j}$ $\eta$

be the exponent sums of in , , respectively. $\alpha_{1}$

$\tilde{\varphi}(\alpha_{i})$
$\mu_{j}$ $\mu$

(0) $m_{\iota/(i)}=m_{i}$

(1) $P^{-1}A_{1}^{\epsilon_{i}}P=A_{i}$

(2) $\left(\begin{array}{l}u_{j}\\v_{j}\end{array}\right)=\sigma A^{\eta_{j}}$ $\left(\begin{array}{l}a_{I/(j)}/m_{\nu(j)}\\b_{\iota,(j)}/m_{\nu(j)}\end{array}\right)$ $-P$ $\left(\begin{array}{l}a_{j}/m_{j}\\b_{j}/m_{j}\end{array}\right)$

(3) $\sigma A_{1}^{\eta}$ $\left(\begin{array}{l}a\\b\end{array}\right)$

$+\sigma\sum A_{1}^{\eta_{j}}$ $\left(\begin{array}{l}a_{\iota/(j)}/m_{\iota/(j)}\\b_{\iota,(j)}/m_{\nu(j)}\end{array}\right)$
$+(A_{1}-I)$ $\left(\begin{array}{l}s_{2}\\t_{2}\end{array}\right)=$

$P$ $\left(\begin{array}{l}a+\Sigma a_{i}/m_{i}\\b+\Sigma b_{i}/m_{i}\end{array}\right)$

Sketch of Proof. The proof is similar to that of [7, \S 5, Lemma 4,

Theorem 5] (1) and (2) are derived from the relations $\alpha_{i}(\ell, h)\alpha_{i}^{-1}=$

,
$(\ell, h)A_{i}$ . (3) comes from (1), (2) and the remaining relsL-
$q_{j}^{m_{j}}\ell^{a_{j}}h^{b_{j}}=1$

tion .
$\prod[\alpha_{2j-1}, \alpha_{2j}]\prod q_{j}=\ell^{a}h^{b}$

The map induces the homomorphism : Aut $\varphi\rightarrow\overline{\varphi}$

$q$
$\Gamma\rightarrow Aut\overline{\Gamma}$

which descends to a homomorphism : Out Out . Let Aut $\overline{q}$

$\Gamma\rightarrow$
$\overline{\Gamma}$

$(\overline{\Gamma}, q)$

and Out be the images of and respectively. Also put $K=$

$(\overline{\Gamma}, \overline{q})$
$q$
$\overline{q}$

(Inn ). Then since maps Inn onto Inn the natural projection
$q^{-1}$
$\overline{\Gamma}$

$q$
$\Gamma$
$\overline{\Gamma}$

: Aut
$\pi$
maps $K$ onto and we have the following
$\Gamma\rightarrow Out\Gamma$ $\overline{K}=ker\overline{q}$

commutative diagram with exact rows and columns.

356 M. Ue

1 1
$\downarrow$ $\downarrow$

1 $\rightarrow Inn\Gamma\rightarrow$ $K$ $\rightarrow$

$\overline{K}$
$\rightarrow 1$

$||$ $\downarrow$ $\downarrow$

1 $\rightarrow Inn$ $\Gamma\rightarrow$

Aut $\Gamma$
$\rightarrow$ Out $\Gamma$
$\rightarrow 1$

$\downarrow$ $\downarrow\overline{q}$

Out $(\overline{\Gamma}, \overline{q})=Out$ $(\overline{\Gamma}, \overline{q})$

$\downarrow$ $\downarrow$

1 1

It is easy to see that the action of Aut (resp. Out ) preserves $\Gamma$ $\Gamma$

the product fibration for , $\overline{\mathcal{F}}\rightarrow \mathcal{R}\rightarrow\overline{\mathcal{R}}$

$\mathcal{R}=\mathcal{R}(\Gamma, G)$
$\overline{\mathcal{R}}=\overline{\mathcal{R}}(\overline{\Gamma}, \overline{G})$

(resp. for I $=I$ ( , ), $G$ ) (given in \S 1) and

$\mathcal{F}\rightarrow I\rightarrow\overline{\mathcal{T}}$ $\Gamma$ $\overline{\mathcal{T}}=\overline{I}(\overline{\Gamma},$ $\overline{G})$

induces the natural action of Aut (resp. Out ( , )) on (resp. ). $(\overline{\Gamma}, q)$
$\overline{\Gamma}$

$q$
$\overline{\mathcal{R}}$ $\overline{I}$

Now we will check the action of on (note that acts trivially $\overline{K}$
$\mathcal{F}$
$\overline{K}$

on ). First suppose that

$\overline{T}$
$X=H^{2}\times E^{2}$ . Define by $\rho_{0}=\rho 0(\ell_{0}, \lambda)\in \mathcal{R}$

$\rho 0(\ell)(z, w)=(z, w+\ell_{0})$

$\rho_{0}(h)(z, w)=(z, w+\ell_{0}\lambda^{-1})$

$\rho_{0}(\alpha_{1})(z, w)=(\overline{\rho}_{0}(\overline{\alpha}_{1})z, \phi(\overline{\alpha}_{1})w+w_{1}^{0})$

$\rho_{0}(\alpha_{i})(z, w)=(\overline{\rho}_{0}(\overline{\alpha}_{i})z, w+w_{i}^{0})$ $(i\geq 2)$

$\rho 0(q_{j})(z, w)=(\overline{\rho}_{0}(\overline{q}_{j})z, w-(a_{j}\ell_{0}+b_{i}\ell_{0}\lambda^{-1})/m_{j})$

where
$\ell_{0}\in R^{+\alpha},s\lambda\neq 0$
,

$w_{j}^{0}=0$ for any $j$

if $A_{1}=I$

and if $A_{1}\neq I$

$\ell_{0}((a+\sum a_{i}/m_{i})+(b+\sum b_{i}/m_{i})\lambda^{-1})/(\phi(\overline{\alpha}_{1})-1)$ if $j=2$

$w_{j}^{0}=\{$

0 if $j\neq 2$ .
 Geometric Structures on the Seifert 4-Manifolds 357

(We can choose such . See the proof of Theorem A in $w_{j}^{0}$

\S 1.) Further-
more by the assumption in Proposition 1 we have

$\lambda=\phi(\overline{\alpha}_{1})=\exp(\pm 2\pi i/6)$ if $A_{1}=\left(\begin{array}{ll}1 & 1\\-1 & 0\end{array}\right)$

$\lambda=\phi(\overline{\alpha}_{1})=\exp(\pm 2\pi i/4)$ if $A_{1}=\left(\begin{array}{ll}0 & 1\\-1 & 0\end{array}\right)$

$\lambda=\exp(\pm 2\pi i/6)$ , $\phi(\overline{\alpha}_{1})=\lambda^{2}$

if $A_{1}=\left(\begin{array}{ll}0 & 1\\-1 & -1\end{array}\right)$
.

There is no further restriction on if $A_{1}=\pm I$ . Then the image $\lambda$

$[\rho_{0}]\in I$

of belongs to the fiber

$\rho_{0}$ of over the image $(\cong \mathcal{F}=T_{1}\times H^{1}(\overline{\Gamma}, C^{\phi}))$ $\mathcal{T}$

of . Its -coordinates are detected by

$[\overline{\rho}_{0}]\in\overline{I}$
and it corre-
$\overline{\rho}_{0}$ $T_{1}$ $(\ell_{0}, \lambda)$

sponds to 0 in the -component. Now we take the subfamily $H^{1}(\overline{\Gamma}, C^{\phi})$

of with a fixed image

$\mathcal{R}_{0}$
in whose elements
$\mathcal{R}$

are $\overline{\rho}_{0}$
$\overline{\mathcal{R}}$

$\rho=\rho(\ell_{0}, \lambda, m)$

defined by
$\rho(\ell)(z, w)=(z, w+\ell_{0})$

$\rho(h)(z, w)=(z, w+\ell_{0}\lambda^{-1})$

$\rho(\alpha_{1})(z, w)=(\overline{\rho}_{0}(\overline{\alpha}_{1})z, \phi(\overline{\alpha}_{1})w+w_{1})$

$\rho(\alpha_{i})(z, w)=(\overline{\rho}_{0}(\overline{\alpha}_{i})z, w+w_{z})$ $(i\geq 2)$

$\rho(q_{j})(z, w)=(\overline{\rho}_{0}(\overline{q}_{j})z, w-(a_{j}\ell_{0}+b_{j}\ell_{0}\lambda^{-1})/m_{j})$ .

Here $(\ell_{0}, \lambda)$

satisfies the same conditions as before and
$w_{i}=w_{i}^{0}+m(\overline{\alpha}_{i})$

where $m$ is a crossed homomorphism from $\overline{\Gamma}$

to $C^{\phi}$
(with $m(\overline{q}_{j})=0$ )
satisfying
if $m(\overline{\alpha}_{1})=m(\overline{\alpha}_{2})=0$ $A_{1}\neq I$ .

(There are no restrictions on $m(\overline{\alpha}_{j})$

if $A_{1}=I.$ ) Hence we have

$w_{1}=0$ , $w_{2}=w_{2}^{0}$ , $w_{j}=m(\overline{\alpha}_{j})(j\geq 3)$ if $\phi\not\equiv id$

$w_{j}=m(\overline{\alpha}_{j})$
for any $j$
if $\phi\equiv id$ .

We note that if $A_{1}\neq I$ then $w_{2}=w_{2}^{0}$ is fixed once is fixed by $(\ell_{0}, \lambda)$

the relations (7), (8) in the proof of Theorem A (1) and any crossed
homomorphism with for $j\geq 2$ is contained in the
$n:\overline{\Gamma}\rightarrow C^{\phi}$
$n(\overline{\alpha}_{j})=0$

image of . Therefore the $m$ s satisfying the

$\delta:C^{0}(\overline{\Gamma}, C^{\phi})\rightarrow C^{1}(\overline{\Gamma}, C^{\phi})$

above conditions descend isomorphically onto . Taking these $H^{1}(\overline{\Gamma}, C^{\phi})$

358 M. Ue

facts into account we can see that the family generated by $\mathcal{R}_{0}$
$\rho(\ell_{0}, \lambda, m)$

whose parameters satisfy the above conditions is the subfamily of the

fiber over whose -components give the representatives of . To
$\overline{\rho}_{0}\in \mathcal{R}$
$\mathcal{F}$ $\mathcal{F}$

check the action of on , it suces to find the element $\mu\in Inn$ $G$ for $\overline{K}$ $\overline{\mathcal{F}}$

given , ( may depend on and ) such that

$\rho\in \mathcal{R}_{0}$ .
$\varphi\in K$ $\mu$ $\rho$ $\varphi$ $\mu\cdot\rho$
$\varphi\in \mathcal{R}_{0}$

and examine the action of . . (which is independent of in the $\mu$ $\rho$ $\varphi$ $\overline{\rho}_{0}$

$w$ -coordinate). Since the action of $K$ commutes with that of Inn $G$ , it
suces to consider the element of the following form. $\varphi\in K$

$\varphi(\alpha_{\dot{0}})=\alpha_{i}\ell^{s_{i}}h^{t_{i}}$

(4) $\varphi(q_{j})=q_{j}\ell^{u_{j}}h^{v_{j}}$

$(\varphi(\ell), \varphi(h))=(\ell, h)P$

where $s_{i}$ , $t_{i}$

, $u_{i}$ , $v_{i}\in Z$ , $P\in GL_{2}Z$ satisfy

(5) $PA_{1}P^{-1}=A_{1}$

(6) $\left(\begin{array}{l}u_{j}\\v_{j}\end{array}\right)=(I-P)$ $\left(\begin{array}{l}a_{j}/m_{j}\\b_{j}/m_{j}\end{array}\right)$

(7) $(A_{1}-I)$ $\left(\begin{array}{l}s_{2}\\t_{2}\end{array}\right)=(P-I)$ $\left(\begin{array}{l}a+\Sigma a_{i}/m_{i}\\b+\Sigma b_{i}/m_{i}\end{array}\right)$

.

For such $\varphi\in K$ and denote the first and the second factors of
$\rho\in \mathcal{R}_{0}$

$\rho$
. $\varphi(\alpha)(z, w)$ by $\rho$
. $\varphi(\alpha)(z, w)_{i}(i=1,2)$ respectively for . Then $\alpha\in\Gamma$

we have
$\overline{\rho}_{0}(\overline{\alpha})z$
if $\alpha=\alpha_{i}$ or $q_{j}$

(8) $\rho$
. $\varphi(\alpha)(z, w)_{1}=\{$
$z$
if $\alpha=\ell$
or $h$

and
$\rho\cdot\varphi(\alpha_{1})(z, w)_{2}=\phi(\overline{\alpha}_{1})(w+\ell_{0}(s_{1}+t_{1}\lambda^{-1}))+w_{1}$

$\rho$
. $\varphi(\alpha_{i})(z, w)_{2}=w+w_{i}+\ell_{0}(s_{i}+t_{i}\lambda^{-1})(i\geq 2)$

(9) $\rho\cdot\varphi(q_{j})(z, w)_{2}=w-\ell_{0}(a_{j}+b_{j}\lambda^{-1})/m_{j}+\ell_{0}(u_{j}+v_{j}\lambda^{-1})$

$\rho\cdot\varphi(\ell)(z, w)_{2}=w+\ell_{0}$

$\rho\cdot\varphi(h)(z, w)_{2}=w+h_{0}$

where . Let be the subgroup of $K$ consisting

$(\ell_{0}, h_{0})=(\ell_{0}, \ell_{0}\lambda^{-1})P$ $K_{1}$

of the elements satisfying $(4)-(7)$ with $P=I$ and let be its image in $\overline{K}_{1}$

. For any
$\overline{K}$

we deduce $u_{j}=v_{j}=0$ and if $A_{1}\neq I$ , $s_{2}=t_{2}=0$ .

$\varphi\in K_{1}$

Also let be the subgroup of generated by the elements of the

$K_{0}$ $K_{1}$

above forms with $s_{j}=t_{j}=0$ for $j\geq 2$ and be its image in . Note $\overline{K}_{0}$ $\overline{K}$
 Geometric Structures on the Seifert 4-Manifolds 359

that and $K_{1}\cap Inn$

$\overline{K}_{1}\cong K_{1}/(K_{1}\cap Inn\Gamma)$
, $\overline{K}_{0}\cong K_{0}/(K_{0}\cap Inn\Gamma)$ $\Gamma=$

$K_{0}\cap Inn$ $\Gamma=InnZ^{2}$ where is the subgroup of generated by and $Z^{2}$ $\Gamma$ $\ell$

. (This comes from the fact that is centerless.) The element

$h$
$\overline{\Gamma}$

$\varphi\in K_{0}$

comes from Inn $Z^{2}$

if $\left(\begin{array}{l}s_{1}\\t_{1}\end{array}\right)=(A_{1}^{-1}-I)$ $\left(\begin{array}{l}s\\t\end{array}\right)$

and hence

$\overline{K}_{0}$

is finite if $A_{1}\neq I$ and moreover $\overline{K}_{0}=1$

if $A_{1}=\left(\begin{array}{ll}1 & 1\\-1 & 0\end{array}\right)$
.

Given and we take an inner automorphism by the

$\rho\in \mathcal{R}_{0}$ $\varphi\in K_{1}$ $\mu$

element if $A_{1}\neq I$ .
$(z, w)\rightarrow(z, w+\phi(\overline{\alpha}_{1})(s_{1}\ell_{0}+t_{1}\ell_{0}\lambda^{-1})/(\phi(\overline{\alpha}_{1})-1))$

Then we can see by the conditions on above for (8), (9) that the $w_{i}$

correspondence . (if $A_{1}=I$ ) or . (if $A_{1}\neq I$ ) gives a

$\rho\rightarrow\rho$ $\varphi$ $\rho\rightarrow\mu\cdot\rho$ $\varphi$

map from to itself such that the parameters are changed as follows.
$\mathcal{R}_{0}$

$(\ell_{0}, \lambda)\rightarrow(\ell_{0}, \lambda)$

$m(\overline{\alpha_{i}})\rightarrow m(\overline{\alpha_{i}})+s_{i}\ell_{0}+t_{i}\ell_{0}\lambda^{-1}$

.
for $i\geq 1$ if $A_{1}=I$ and for if $A_{1}\neq I$ . (We can see from (7) that $?$ $\geq 3$

the -parameter of
$w_{2}$ is the same as that for if $A_{1}\neq I.$ ) Since $\mu\cdot\rho\cdot\varphi$ $\rho$

,
$s_{?}.$
( $i\geq 1$ if $A_{1}=I$ and $i\geq 3$ if $A_{1}\neq I$ ) are arbitrary integers this
$t_{i}$

gives the action of on . Hence $\overline{K}_{1}$ $\mathcal{F}$

$\mathcal{F}/\overline{K}_{1}\cong T_{1}\times H^{1}(\overline{\Gamma}, C^{\phi})/H^{1}(\overline{\Gamma}, Z^{2^{\phi}})$

with
$(T^{1})^{2g}$ if $A_{1}=I$
$H^{1}(\overline{\Gamma}, C^{\phi})/H^{1}(\overline{\Gamma}, Z^{2^{\phi}})\cong\{$

$(T^{1})^{2g-2}$ if $A_{1}\neq I$

where $T^{1}\cong C/Z^{2}$ is the complex torus of dimension 1 whose lattice is

generated by and . Here we note that if $A_{1}\neq I$ then
$\ell_{0}$
is $\ell_{0}\lambda^{-1}$ $\overline{K}_{0}$

the subgroup of which acts trivially on . Hence (or $\overline{K}_{1}$ $\mathcal{F}$ $\overline{K}_{1}$ $\overline{K}_{1}/\overline{K}_{0}$

if $A_{1}\neq I$ ) acts eectively and properly discontinuously on . Next $\mathcal{F}$

consider the action of not decending to . $\varphi\in K$ $\overline{K}_{1}$

Case (1). $A_{1}=\pm I$ . Put $P=\left(\begin{array}{ll}p & q\\r & s\end{array}\right)$ $\in GL_{2}Z$ for $\varphi\in K$ defined
in (4). Here , , , for must be defined as elements in
$u_{j}$ $v_{j}$ $s_{2}$ $t_{2}$ $\varphi$
$Z$

according to (6) and (7) (if $A_{1}=I$ then (7) becomes obvious since the
right hand side of (7) is 0 in case $X=H^{2}\times E^{2}$ ). Then considering (8),
(9) for and we can take an inner automorphism of the form
$\rho\in \mathcal{R}_{0}$ $\varphi$ $\mu$

$(z, w)\rightarrow(z, \sigma w+c)$ for some $c\in C$ with $\sigma=|p+r\lambda^{-1}|/(p+r\lambda^{-1})$ such
360 M. Ue

that the the correspondence $\rho\rightarrow\mu$

. $\rho$
. $\varphi$
gives a map from $\mathcal{R}_{0}$
to itself
of the form

$\lambda\rightarrow(p\lambda+r)/(q\lambda+s)$

$\ell_{0}\rightarrow|p+r\lambda^{-1}|\ell_{0}$

$m(\overline{\alpha}_{i})\rightarrow\sigma(m(\overline{\alpha}_{i})+s_{i}\ell_{0}+t_{i}\ell_{0}\lambda^{-1})$

for $i\geq 1$ if $A_{1}=I$ and for $i\geq 3$ if $A_{1}=-I$ .

Case (2). $A_{1}\neq\pm I$ . In this case we deduce from the conditions on

$\lambda$

above that

$(\ell_{0}, \ell_{0}\lambda^{-1})A_{1}=(\lambda\ell_{0}, \ell_{0})$

for $A_{1}=\left(\begin{array}{ll}1 & 1\\-1 & 0\end{array}\right)$

if $A_{1}=\left(\begin{array}{ll}0 & 1\\-1 & -1\end{array}\right)$

and

$(\ell_{0}, \ell_{0}\lambda^{-1})A_{1}=(\lambda\ell_{0}, \ell_{0})$

otherwise. Furthermore for the presentation of $\varphi$

in (4) we must have

$P=A_{1}^{k}$ (in case $A_{1}=$ $\left(\begin{array}{ll}0 & 1\\-1 & -1\end{array}\right)$
) or $P=A_{1}^{k}$ (otherwise) for some
$k\in Z$ by the condition (5) and

$u_{j}$ , $v_{j}$ , $s_{2}$ , $t_{2}\in Z$

where these numbers are defined by (6) and (7) for the above $P$ (we
have assumed that $P\neq I$ since ) Then in the presentation (9) $\varphi\not\in K_{1}.$

for , we have
$\rho\in \mathcal{R}_{0}$ $\varphi\in K$

$(\ell, h)=(\ell_{0}, \ell_{0}\lambda^{-1})P=(\lambda^{k}\ell_{0}, \lambda^{k-1}\ell_{0})$

for $P=A_{1}^{k}$ or $P=A_{1}^{k}$ as above. Hence taking an inner automorphism

by the element $(z, w)\rightarrow(z, \lambda^{-k}w+c)$ where
$\mu$

$c=\lambda^{-k}\phi(\overline{\alpha}_{1})(s_{1}\ell_{0}+t_{1}\ell_{0}\lambda^{-1})/(\phi(\overline{\alpha}_{1})-1)$

we can see that the correspondence $\rho\rightarrow\mu\cdot\rho\cdot\varphi$ gives a map from $\mathcal{R}_{0}$

to itself of the form

$(\ell_{0}, \lambda)\rightarrow(\ell_{0}, \lambda)$

$m(\overline{\alpha}_{i})\rightarrow\lambda^{-k}(m(\overline{\alpha}_{i})+s_{i}\ell_{0}+t_{i}\ell_{0}\lambda^{-1})$ $(i\geq 3)$ .

 Geometric Structures on the Seifert 4-Manifolds 361

In the cases when $A_{1}=\pm I$ the above correspondence shows that the
action of on preserves the product fibration of the form $\overline{K}$
$\mathcal{F}$
$ H^{1}(\overline{\Gamma}, C^{\phi})\times$

(where $T_{1}=R^{+}\times H\times Z_{2}$ ) which induces the

$R^{+}\rightarrow \mathcal{F}\rightarrow H\times Z_{2}$

properly discontinuous action on $H\times Z_{2}$ (which is identified with $\{\lambda\in$

$C|s^{\propto}\lambda\neq 0\})$ of the form $\lambda\rightarrow(p\lambda+r)/(q\lambda+s)$ for some matrix

$\left(\begin{array}{ll}p & r\\q & s\end{array}\right)$
$\in GL_{2}$ Z. In the cases when $A_{1}\neq\pm I$ we have at most finite
number of possible choices for $P$ in the presentation of above since $\varphi$
$A_{1}$

(or ) is periodic. Hence by the above correspondences and the action

$A_{1}$

of (and taking the fact that

$\overline{K}_{1}$

is finite if $A_{1}\neq I$ into account) we can $\overline{K}_{0}$

easily see that (or if ) acts properly discontinuously on $\overline{K}$

$\overline{K}/\overline{K}_{0}$ $\phi\not\equiv id$

in either case. Finally consider the action of Out on . The group

$\mathcal{F}$ $\Gamma$ $\mathcal{T}$

Out acts on so that it preserves the product fibration

$\Gamma$ $\mathcal{T}$ $\mathcal{F}$
$\rightarrow \mathcal{T}\rightarrow\overline{\mathcal{T}}$

and induces the action of Out on which is properly discontinuous $(\overline{\Gamma}, \overline{q})$
$\overline{\mathcal{T}}$

since the action of Out on has, as is well known, the same property. $(\overline{\Gamma})$
$\overline{\mathcal{T}}$

Since (which is the subgoup of Out which induces the identity on

$\overline{K}$
$\Gamma$

T) acts properly discontinuously on the fiber of as above we can see $\mathcal{T}$

that Out acts also properly discontinuously on . The cases with

$\Gamma$ $\mathcal{T}$

(in this case $A_{1}=I$ ) can be treated similarly and we omit

$X=\overline{SL}_{2}\times E$

the details. Thus we have

Proposition 3. Let be a geometric Seifert 4-manifold $S=\Gamma\backslash X$

over a closed orientable hyperbolic orbifold $B$ with $\Gamma\subset G=Isom^{0}X$ .

Then Out (or Out in case the monodromy representation $\Gamma$

of $\Gamma/\overline{K}_{0}$ $\phi$

$S$
is not trivial where is fifinite subgroup of defifined above) acts $\overline{K}_{0}$
$a$
$\overline{K}$

on properly discontinuously and the muduli space

$\mathcal{T}(\Gamma, G)$
is $\mathcal{M}(\Gamma, G)$

Hausdor.
On the other hand if then we must have of the form $\overline{K}_{1}\neq\overline{K}$
$\varphi\in K$

(4) satisfying $(5)-(7)$ with $P\neq I$ . In particular , , , defined by $s_{2}$ $t_{2}$

$u_{j}$ $v_{j}$

(5) and (6) for some appropriate $P\neq I$ satisfying (4) must be integers.
From these conditions we can deduce some extra conditions on the Seifert
invariants of and derive the following proposition. Here we omit the $S$

details of the computations.

Proposition 4. Let be a geometric Seifert 4-manifold $S=\Gamma\backslash X$

as in Proposition 3. Then if the monodromies of satisfy the con- $S$

ditions in Proposition 1 and if the Seifert invariants of do not sat- $S$

isfy the conditions below then and is a Seifert fifibra- $\overline{K}=\overline{K}_{1}$

$\Lambda 4(\Gamma, G)$

tion over (which is defifined above) with general fifiber

$\overline{\mathcal{T}}(\overline{\Gamma}, \overline{G})/Out(\overline{\Gamma}, \overline{q})$

$T_{1}\times H^{1}(\overline{\Gamma}, C^{\phi})/H^{1}(\overline{\Gamma}, Z^{2^{\phi}})$

.
362 M. Ue

(I) $A_{1}=\left(\begin{array}{ll}1 & 1\\-1 & 0\end{array}\right)$

.

(1) There are no multiple fifibers;

(2) ,
$(m_{i}, a_{i}, b_{i})=(3, \epsilon_{i}, \epsilon_{i})$ $\epsilon_{i}=\pm 1$
for any ; $i$

(3) $m_{i}=2$ for any . $i$

(II) $A_{1}=\left(\begin{array}{ll}0 & 1\\-1 & -1\end{array}\right)$

.

(1) There are no multiple fifibers and $a\equiv bmod 3$ ;

(2) $(m_{i}, a_{i}, b_{i})=(3, \epsilon_{i}, \epsilon_{i})$

with $\epsilon_{i}=\pm 1$
for any $i$
and $\sum\epsilon_{i}\equiv 0$

$mod 3$ ;
(3) $m_{i}=2$ for any $i$
and $\sum a_{i}-2a\equiv\sum b_{i}-2b$ mod 3. $ $

(III) $A_{1}=\left(\begin{array}{ll}0 & 1\\-1 & 0\end{array}\right)$

.

(1) There are no multiple fifibers;

(2) $m_{i}=2$ for every and $\sum a_{i}\equiv\sum b_{\dot{0}}mod 2$ . $i$

(IV) $A_{1}=\pm I$ .

References
[1] L. Bers, Fiber spaces over Teichm\"uller spaces, Acta Math., 130 (1973),
89-126.
[2] K. Dabrowsky, Moduli spaces for Hopf surfaces, Math. Ann., 259 (1982),
201-226.
[3] K. Kodaira, D. C. Spencer, On deformations of complex analytic struc-
tures, I, II, Ann. Math., 67 (1958), 328-466.
[4] , A theorem of completeness for complex analytic fiber spaces,

Acta. Math., 100 (1958), 281-294.

[5] R. Kulkarni, K. B. Lee, F. Raymond, Deformation spaces for Seifert
manifolds, in Geometry and Topology, Lect. Notes in Math. vol.
1167, Springer, 1985, pp. 180-216.
[6] K. Ohshika, Teichm\"uller spaces of Seifert fibered manifolds with infinite
, Topology and its appl., 27 (1987), 75-93.
$\pi_{1}$

[7] P. Orlik, Seifert Manifolds, Springer, Berlin-Heidelberg-New York,

1972.
[8] T. Suwa, Deformations of Holomorphic Seifert Fiber Spaces, Invent.
math., 51 (1979), 77-102.
[9] , On ruled surfaces of genus 1, J. Math. Soc. Japan, 21 (1969),

291-311.
[10] M. Ue, Geometric -manifolds in the sense of Thurston and Seifert $4$

-manifold I, J. Math. Soc. Japan, 42 (1990), 511-540.

$4$

[11] , Geometric -manifolds in the sense of Thurston and Seifert $4$

-manifolds II, J. Math. Soc. Japan, 43 (1991), 149-183.

$4$
 Geometric Structures on the Seifert 4-Manifolds 363

[12] C. T. C. Wall, Geometric structures on compact complex surfaces, Top-

ology, 25 (1986), 119-153.
[13] , Geometries and geometric structures in real dimension 4 and

complex dimension 2, in Geometry and Topology. Lect. Notes in

Math., 1167, Springer, 1985, pp. 268-292.
[14] J. Wehler, Versal deformations for Hopf surfaces, J. Reine Angew. Math. ,
328 (1982), 22-32.
[15] H. Zieschang, On toric fiberings over surfaces, Math. Notes, 5 (1967),
341-345.

Institute of mathematics
Yoshida College
Kyoto University
Kyoto 606
Japan
Advanced Studies in Pure Mathematics 20, 1992
Aspects of Low Dimensional Manifolds
pp. 365-376

Homologically Trivial Smooth Involutions

on $K3$ Surfaces
Takao Matumoto

Dedicated to Professor Sh\^or\^o Araki on his 60th birthday

Abstract.
We will show that any smooth involution on a K3 surface induces
a non-trivial action on its homology. In fact, a closed spin 4-manifold
$M$ with $H_{1}(M;Z_{2})=0$ and sign $M$ will be shown to admit no
$\neq 0$

homologically trivial locally linear involutions. The proof uses only

the $G$ -signature theorem and the sublattices and branched coverings
arguments.

\S 1. Introduction
Some complex surfaces including K3 surfaces admit no homologi-
cally trivial holomorphic involutions. There posed a question in [12;11.8]
whether the same is true for the smooth involutions or not. This paper
answers the question armatively at least for the smooth involutions on
K3 surfaces. Note that a smooth involution is locally linear.
Theorem 1. Let $M$ be a closed connected oriented spin 4-manifold
with . Suppose that there is an orientation preserving
$H_{1}(M\cdot, Z_{2})=0$

locally linear involution on $M$ which operates as identity on

$\sigma$
$(M; Q)$ .$H_{2}$

Then, sign $M$ $=0$ .

Since a K3 surface is a simply-connected spin 4-manifold with sig-

nature-16, it admits no homologically trivial locally linear involutions.
According to Edmonds [5] Theorem 1 in the case that $M$ is simply-
connected is already proved by D. Ruberman.
The author thanks Dr. M. Masuda for informing of Edmonds paper
and Dr. M. Sekine for the discussions about Lemma 2.4. Some results on
the homologically antipodal locally linear involutions are also obtained

Received July 9, 1990.

Revised July 5, 1991.
366 T. Matumoto

with the collaboration of Y. Matsumoto and A. Kawauchi, which will

be published elsewhere.

\S 2. Preliminary lemmas

We prepare some lemmas which will be used later and may be useful
for the other purposes. We begin with a lemma to construct a double
covering from two 2-sheet branched coverings.

Lemma 2.1. Let be a locally linear involution on a connected $\sigma$

manifold $M$ with fixed point set F. Suppose there is a subunion of con-
nected components with a non-trivial element of $H^{1}(M/\sigma-$ $F\subseteq F$ $e_{\tau}$

$F$
; ) which takes non-zero value on the image of $H_{1}(\partial N(x)/\sigma;Z)$ for
$Z_{2}$

any of $F$ , where$x$ stands for the orbit space and $N(x)$ is a fiber at $-/\sigma$

$x$
of an equivariant normal disk bundle $N(U_{x})$ for a neighborhood of $U_{x}$

$x$ in F. Then, there is a locally linear -action with generators $Z_{2}\times Z_{2}$ $\tilde{\sigma}$

and on a double ( $=connected2$ -sheet unbranched) covering manifold

$\tilde{\tau}$

of $M$ such that the orbit space

$\overline{M}$

is canonically homeomorphic to $\overline{M}/\tilde{\tau}$

$M$ and induces with this identification.

$\tilde{\sigma}$

$\sigma$

unbranched
$M$ $\rightarrow$ $\overline{M}/\tilde{\tau}=M$

covering
$\downarrow$ $\downarrow$

$\overline{M}/\tilde{\sigma}=M$ $ M/\sigma$
$\rightarrow$

The projection
Proof : $M-F\rightarrow M/\sigma-F$ is a covering $\pi$

map induced from a non-trivial element of $H^{1}(M/\sigma-F;Z_{2})=$ $e_{\sigma}$

$Hom(H_{1}(M/\sigma-F;Z), Z_{2})=Hom(\pi_{1}(M/\sigma-F), Z_{2})$ which takes non-

zero value on $H_{1}(\partial N(x)/\sigma;Z)$ for any of $F$ . Let : $M/\sigma-F\rightarrow M/\sigma-$ $x$ $j$

$F$
be the inclusion. Then, we have , since takes zero value $j^{*}e_{\tau}\neq e_{\sigma}$ $e_{\tau}$

on $H_{1}(\partial N(x)/\sigma;Z)$ for any of $F-F$ . So, we get a covering $x$ $Z_{2}\times Z_{2}$

of $M/\sigma-F$ associated to : $H_{1}(M/\sigma-F;Z)\rightarrow Z_{2}\times Z_{2}$ . $(j^{*}e_{\tau}, e_{\sigma})$

Consider the base change : $ H_{1}(M/\sigma-F;Z)\rightarrow$ $(j^{*}e_{\tau}, j^{*}e_{\tau}+e_{\sigma})$

. The completed 2-sheet branched coverings

$Z_{2}\times Z_{2}$ : $ M\rightarrow M/\sigma$ $\pi$

and : $\pi$
(resp.) induced by and
$ M^{JJ}\rightarrow M/\sigma$ (resp.) have $j^{*}e_{\tau}$ $j^{*}e_{\tau}+e_{\sigma}$

the disjoint branch loci $F$

and $F-F$ (resp.). So, the completed $2\times 2-$

sheet branched covering : , induced by :

$\overline{\pi}$ $\overline{M}\rightarrow M/\sigma$ $(j^{*}e_{\tau}, j^{*}e_{\tau}+e_{\sigma})$

$H_{1}(M/\sigma-F;Z)\rightarrow Z_{2}\times Z_{2}$ , has the locally linear involutions and $\overline{\sigma}$ $\overline{\sigma}$

so that : and
$\overline{\pi}$

: are the 2-
$\overline{M}\rightarrow\overline{M}/\overline{\sigma}=M$ $\overline{\pi}$ $\overline{M}\rightarrow\overline{M}/\overline{\sigma}=M$

sheet branched coverings with branch loci $(\pi)^{-1}(F-F)$ and $(\pi)^{-1}(F)$

 Homologically Trivial Smooth Involutions 367

respectively. By the definition and commute outside . Since $\overline{\sigma}$ $\overline{\sigma}$

$\overline{\pi}^{-1}(F)$

is dense in
$\overline{M}-\overline{\pi}^{-1}(F)$
, and commute also on whole .
$\overline{M}$
$\overline{\sigma}$ $\overline{\sigma}$
$\overline{M}$

Put . Then, has no fixed point either in

$\overline{\tau}=\overline{\sigma}o\overline{\sigma}$ $\overline{\tau}$ $\overline{M}-\overline{\pi}^{-1}(F)$

or in $\overline{\pi}^{-1}(F)=(\overline{\pi})^{-1}(\pi)^{-1}(F-F)\cup(\overline{\pi})^{-1}(\pi)^{-1}(F)$ and hence in

whole . Moreover,
$\overline{M}$

is the branched covering induced by

$\overline{M}/\overline{\tau}\rightarrow M/\sigma$

, that is, equivalent to $M$

$j^{*}e_{\tau}+j^{*}e_{\tau}+e_{\sigma}=e_{\sigma}$ . $\rightarrow M/\sigma$

Since $M$ is connected, is connected. If $ F=\emptyset$

, the cover- $ M/\sigma$

ing associated to the non-trivial element of $H^{1}(M/\sigma;Z_{2})$ is connected.

Otherwise the branch locus of is non-empty and $M$ is con- $ M\rightarrow M/\sigma$

nected. Then, since the branch locus of is non-empty, is $\overline{M}\rightarrow M$ $\overline{M}$

connected. Q.E.D.

We recall and define some notions about lattices now. A -free mod- $Z$

ule of finite rank with non-degenerate symmetric bilinear form

$L$
: $\langle, \rangle$

is called a lattice. Let

$L\times L\rightarrow Z$
denote the dual module Homz $(L, Z)$ $L^{*}$

and we have a canonical embedding defined by . The $L\subset L^{*}$ $ x\mapsto\langle, x\rangle$

factor group $L^{*}/L$ is finite abelian and its order divides discr where $|$

$L|$

discr for some basis . Let be a prime. For a finite

$ L=\det\langle e_{i}, e_{j}\rangle$ $\{e_{i}\}$ $p$

abelian group we denote the minimal number of generators of $A$ and

$A$

by
$A\otimes Z_{p}$ and respectively. A lattice is called unimodular or
$\ell(A)$ $\ell_{p}(A)$

-unimodular if $L^{*}/L=0$ or $\ell_{p}(L^{*}/L)=0$ respectively. A submodule

$p$

$S$
of is called primitive or -primitive if $L/S$ is -free or contains no
$L$ $p$
$Z$

-torsion respectively. Define the orthogonal complement $S^{\perp}=\{y\in L$ ;

$p$

for any $x\in S$ }. If is unimodular and is a primitive sub-

$\langle y, x\rangle=0$ $L$ $S$

lattice, i.e., primitive and the pairing is non-degenerate not only on $\langle, \rangle$

$L$
but also on , we have a natural isomorphism $S$
. (See $S^{*}/S\cong S^{\perp}*/S^{\perp}$

[3;1.2.5] and [10] for example.) Moreover, we can prove

Lemma 2.2. Let be a prime. Let be a -unimodular lattice $p$

$L$
$p$

and a -primitive sublattice.

$S$
Then, the orthogonal complement
$p$

$K=S^{\perp}is$ also a sublattice and the -torsion part $(S^{*}/S)_{(p)}$ of $S^{*}/S$ is $p$

isomorphic to the -torsion part of $(K^{*}/K)_{(p)}$ of $K^{*}/K$ . $p$

Proof.
Take an element of . Then, can be considered $\ell$
$L$ $\ell*=\langle, \ell\rangle$

as an element of in if and only if . If we consider

$S^{*};$ $\ell_{1}^{*}=\ell_{2}^{*}$ $S^{*}$ $\ell_{1}-\ell_{2}\in K$

also as an element in
$\ell^{*}$

, we get a homomorphism $Im(L\rightarrow $K^{*}$

S^{*})/S\rightarrow$

$K^{*}/K$ . That is -primitive implies $(S^{*}/Im(L^{*}\rightarrow S^{*}))_{(p)}=0$ . Since

$S$
$p$

$(L^{*}/L)_{(p)}=0$ by the assumption, we have $(S^{*}/S)_{(p)}=(Im(L^{*}\rightarrow$

$S^{*})/S)_{(p)}=(Im(L\rightarrow S^{*})/S)_{(p)}$ and we get a correlation homomor-

phism $(S^{*}/S)_{(p)}\rightarrow(K^{*}/K)_{(p)}$ . By the definition it is easy to see that

$K$ is a primitive sublattice of and is a minimal primitive sublattice $L$ $K^{\perp}$
368 T. Matumoto

of containing . So, $(K^{\perp}/S)_{(p)}=0$ by the assumption. Then, we get

$L$ $S$

also a homomorphism $(K^{*}/K)_{(p)}\rightarrow(K^{\perp}*/K^{\perp})_{(p)}=(S^{*}/S)_{(p)}$ which

is an inverse of the homomorphism above. Q.E.D.

Next we give a sucient and nearly necessary condition to get a

branched covering in some cases.

Lemma 2.3. Let be a prime. Let , , be disjointly em- $p$

$S_{1}^{2}$
$\ldots$
$S_{\ell}^{2}$

bedded 2-spheres in a closed orientable -manifold $M$ with normal disk $A$

bundles , , . $N(S_{1}^{2})$
$\ldots$
$N(S_{\ell}^{2})$

(1) Suppose that the homology classes , , are linearly de- $[S_{1}^{2}]$
$\ldots$
$[S_{\ell}^{2}]$

pendent in $H_{2}(M;Z_{p})$ . Then, there is a non-trivial element of $H^{1}(M$ $-$

; ) which takes non-zero value on

$\bigcup_{i=1}^{\ell}S_{i}^{2}$
$Z_{p}$
for some . $H_{1}(\partial N(S_{i}^{2});Z)$ $i$

(2) Suppose that , , are linearly independent in $H_{2}(M;Z)$$[S_{1}^{2}]$

$\ldots$
$[S_{\ell}^{2}]$

and generate a submodule of $L=H_{2}(M;Z)/tor$ . Let be the minimal $S$ $\overline{S}$

primitive submodule of containing , that is, is -free. Then, $L$ $S$ $L/\overline{S}$ $Z$

is a finite (possibly zero) abelian group and we have an isomorphism

$\overline{S}/S$

$\overline{S}/S\cong Ker(H_{1}(M-\bigcup_{\dot{x}=1}^{\ell}S_{i}^{2}; Z)\rightarrow H_{1}(M;Z))$

.

Note that the torsion part of $L/S$ is . So, if $L/S$ contains a non- $\overline{S}/S$

trivial -torsion, there is a non-trivial element of

$p$ $H^{1}(M-\bigcup_{i=1}^{\ell}S_{i}^{2}; Z_{p})$

which takes non-zero value on for some . Moreover, $H_{1}(\partial N(S_{i}^{2});Z)$ $i$

when $H_{1}(M;Z)\otimes Z_{p}=0$ , the converse is also true, that is, if there is $a$

non-trivial element of which takes non-zero value $H^{1}(M-\bigcup_{i=1}^{\ell}S_{i}^{2}; Z_{p})$

on for some , $L/S$ contains a non-trivial -torsion.

$H_{1}(\partial N(S_{i}^{2});Z)$ $i$
$p$

(3) Suppose $[S_{i}^{2}]^{2}\equiv 0mod p$ for every and $2\ell>b_{2}(M)$ . Then, $i$

either , , are linearly dependent in $H_{2}(M;Z_{p})$ or linearly in-

$[S_{1}^{2}]$
$\ldots$
$[S_{\ell}^{2}]$

dependent in $H_{2}(M;Z_{p})$ and $L/S$ contains a non-trivial -torsion, where $p$

$L=H_{2}(M;Z)/tor$ and is a submodule generated by , , in $S$ $[S_{1}^{2}]$

$\ldots$
$[S_{\ell}^{2}]$

L. Note that $b2(M)=dim$ H2 $(M;Q)=rank$ L.

(1) Put
Proof. and $N=M$ Int $N(F)$ . Un- $F=S_{1}^{2}\cup\cdots\cup S_{\ell}^{2}$

der the hypothesis we have a non-zero element of $a_{1}[S_{1}^{2}]+\cdots+a_{\ell}[S_{\ell}^{2}]$

$H_{2}(F;Z_{p})=H_{2}(N(F);Z_{p})$ which sends to zero in $H_{2}(M;Z_{p})$ in the

 Homologically Trivial Smooth Involutions 369

following commutative diagram:

$\partial$

$ H_{3}(M, N(F);Z_{p})\rightarrow$ $H_{2}(N(F);Z_{p})$ $\rightarrow H_{2}(M;Z_{p})$

$ PD\uparrow\cong$ $ PD\uparrow\cong$

$H^{1}(N;Z_{p})$ $\rightarrow$ $H^{2}(M, N;Z_{p})$

$\downarrow$ $\downarrow\cong$

$\delta$

$H^{1}(\partial N(F);Z_{p})$ $\rightarrow H^{2}(N(F), \partial N(F);Z_{p})$

Here the horizontal sequences are natural and exact. So, there is an ele-
ment of $H_{3}(M, N(F);Z_{p})$ such that
$\alpha$
. By the Poincar\e duality $\partial\alpha\neq 0$

we get an element $\alpha\in H^{1}(N;Z_{p})=H^{1}(M-F;Z_{p})$ such that . $\delta\alpha\neq 0$

Since , takes non-zero value on

$\partial N(F)=\bigcup_{i=1}^{\ell}\partial N(S_{i}^{2})$ $\alpha$ $H_{1}(\partial N(S_{i}^{2});Z)$

for some . $i$

(2) Note first that there is an isomorphism , where $\overline{S}/S\cong S^{*}/\overline{S}^{*}$ $A^{*}$

stands for the dual $Hom_{Z}(A, Z)$ . Consider the following commutative
diagram whose horizontal sequences are exact and the coecient is : $Z$

$ H_{2}(M, N)\rightarrow\partial$ $\rightarrow j_{*}H_{1}(M)$

$H_{1}(N)$

$ PD\uparrow\cong$ $ PD\uparrow\cong$ $ PD\uparrow\cong$

$\rightarrow i^{*}H^{2}(N(F))\rightarrow\delta H^{3}(M, N(F))\rightarrow j^{*}H^{3}(M)$

$H^{2}(M)$

$||$ $||$

$ L^{*}\oplus tor\rightarrow$ $S^{*}$

Since is torsion free, $Im$ $i^{*}=Im$ . Moreover since

$S^{*}$ $L^{*}$ $L$
is unimod-
ular, $ImL^{*}$ is by the definition of . So,
$\overline{S}^{*}$ $\overline{S}$

$\overline{S}/S\cong S^{*}/\overline{S}^{*}=Cokeri^{*}\cong Im\delta=Kerj^{*}$

By the Poincar\e duality we get $Kerj^{*}\cong Ker(j_{*}$ : $H_{1}(N;Z)=H_{1}(M-$

$F;Z)\rightarrow H_{1}(M;Z))$ .
(3) We may assume that the homology classes , , are lin- $[S_{1}^{2}]$
$\ldots$
$[S_{\ell}^{2}]$

early independent in $H_{2}(M;Z_{p})$ and in particular linearly independent

in $H_{2}(M;Z)$ . We divide into two cases : (i) the case that for $[S_{i}^{2}]^{2}\neq 0$

every , and (ii) otherwise.

$i$

In case (i) the pairing on is non-degenerate and $\ell_{p}(S^{*}/S)=\ell$ . $\langle, \rangle$ $S$

On the other hand rank $ S^{\perp}=b_{2}(M)-\ell$ implies $\ell_{p}(S^{\perp}*/S^{\perp})\leq b_{2}(M)-\ell$ .

370 T. Matumoto

So, if is -primitive i.e.,

$S$
$p$ , then by Lemma 2.2 we have $\ell_{p}(\overline{S}/S)=0$

$\ell\leq b_{2}(M)-\ell$ , which contradicts our hypothesis.

In case (ii) we may assume $[S_{i}^{2}]^{2}=0(1\leq i\leq k)$ and $\neq 0(k+1\leq$
$i\leq\ell)$ . Put $\xi_{i}=[S_{?}^{2}. ]\in H_{2}(M;Z)(1\leq i\leq\ell)$ . Assume that is $S$

-primitive. Then, we have a homology class $\eta_{1}\in H_{2}(M;Z)p$-dual to

$p$

, that is,
$\xi_{1}$
. Now, we put
$\langle\xi_{1}, \eta_{1}\rangle=mp+1$ $\xi_{i}=(mp+1)\xi_{i}-\langle\xi_{i}, \eta_{1}\rangle\xi_{1}$

for $ 2\leq i\leq\ell$ so that , $\xi_{i}^{\prime 2}=0(2\leq i\leq k)$ and

$\langle\xi_{\dot{x}}, \eta_{1}\rangle=\langle\xi_{i}, \xi_{1}\rangle=0$

$\neq 0(k+1\leq i\leq\ell)$ and , , , are also linearly independent. $\xi_{1}$ $\xi_{2}$
$\ldots$
$\xi_{\ell}$

Let be a sublattice generated

$U_{1}$
by and . Since $\ell_{p}(U_{1}^{*}/U_{1})=0$ , $\xi_{1}$
$\eta_{1}$

$L_{1}=\{x\in L : \langle x, \xi_{1}\rangle=\langle x, \eta_{1}\rangle=0\}$ is a -unimodular lattice by $p$

Lemma 2.2. Let be the submodule of generated by ,

$S_{1}$
, . $L_{1}$ $\xi_{2}$
$\ldots$
$\xi_{\ell}$

Recall we assume that $L/S$ contains no -torsion. Then, it is equivalent $p$

to say that contains no -torsion, because $(U_{1}\oplus L_{1})/S\cong Z\oplus$

$L_{1}/S_{1}$ $p$

$L_{1}/S_{1}$ and in the exact sequence

$L/(U_{1}\oplus L_{1})\subset U_{1}^{*}/U_{1}\oplus L_{1}^{*}/L_{1}$ $ 0\rightarrow$

$(U_{1}\oplus L_{1})/S\rightarrow L/S\rightarrow L/(U_{1}\oplus L_{1})\rightarrow 0$ .

By an induction argument we get a -unimodular lattice of $p$ $L_{k}$

rank=rank $L-2k$ containing modified linearly independent homology

classes , , . If we define
$\xi_{k+1}$ by the submodule of
$\ldots$ generated
$\xi_{\ell}$ $S_{k}$ $L_{k}$

by these modified , , , then on is non-degenerate and $\xi_{k+1}$ $\ldots$

$\xi_{\ell}$ $\langle, \rangle$ $S_{k}$

contains no
$L_{k}/S_{k}$ -torsion, that is, is a -primitive sublattice of the$p$ $S_{k}$
$p$

-unimodular lattice
$p$ . Then, by Lemma 2.2 , $L_{k}$ $\ell_{p}(S_{k}^{*}/S_{k})=\ell_{p}(K_{k}^{*}/K_{k})$

where denotes the orthogonal complement of

$K_{k}$ in . So, by an $S_{k}$ $L_{k}$

argument as in the case (i) $\ell-k\leq(b_{2}(M)-2k)-(\ell-k)$ or equiva-

lently $2\ell\leq b_{2}(M)$ , which contradicts our hypothesis. This means that, if
, ,
$[S_{1}^{2}]$
are linearly independent in $H_{2}(M;Z_{p})$ , then $L/S$ contains
$\ldots$
$[S_{\ell}^{2}]$

a non-trivial -torsion. $p$ Q. E. D.

We want to estimate the first Betti number $b_{1}(\overline{M})=dimH_{1}(\overline{M};Q)$

of the 2-sheet branched covering of $M$ . $\overline{M}$

Lemma 2.4. Let be a locally linear involution acting on a com- $\sigma$

pact connected manifold with fixed point set $F$ and orbit space $M$ . $\overline{M}$

Suppose that $(M; Q)=0$ , $F$ admits an equivariant normal disk bun-
$H_{1}$

dle in and one of the following three conditions is satisfied:

$\overline{N}(F)$
$\overline{M}$

(1) , (2) contains neither codimension one nor codimension

$ F=\emptyset$
$F$

two component, or (3) $F$ contains no codimension one component and

any connected component of codimension two part is simply-connected.
Then,
$b_{1}(\overline{M})\leq\ell_{2}(H_{1}(M-F;Z))-1$ .

Here $\ell_{2}(A)$ stands for the number of minimal generators of $A\otimes Z_{2}$ .
 Homologically Trivial Smooth Involutions 371

Sekine [13; 1] gives a proof in case $M=S^{4}$ and $F$ has

Proof.
codimension two. Put Int . The natural projection $\overline{N}=\overline{M}-$ $\overline{N}(F)$

:
$\pi$
induces a double covering :
$\overline{M}\rightarrow M$
of compact manifolds. $\pi$
$\overline{N}\rightarrow N$

We define a chain complex by the exact sequence: $\hat{C}_{*}$

$0-\hat{C}_{*}\rightarrow C_{*}(\overline{N};Z)\rightarrow C_{*}(N;Z)\pi_{*}\rightarrow 0$

.

Let be a generator of
$t$
. Then, $\hat{C}_{*}=(1-t)C_{*}(\overline{N};Z)$ . So, is
$Z_{2}$
$\hat{C}_{*}\otimes Z_{2}$

isomorphic to $(1+t)C_{*}(\overline{N};Z_{2})\cong C_{*}(N;Z_{2})$ as chain complex.

Since is also exact, we
$0\rightarrow\hat{C}_{*}\otimes Q\rightarrow C_{*}(\overline{N};Q)\rightarrow C_{*}(N;Q)\rightarrow 0$

consider the exact sequence:

$H_{1}(\hat{C}_{*}\otimes Q)\rightarrow H_{1}(\overline{N};Q)\rightarrow H_{1}(N;Q)\rightarrow H_{0}(\hat{C}_{*}\otimes Q)\rightarrow 0$
.

Put $d=dimH_{1}(\overline{N};Q)-dimH_{1}(N;Q)$ . Then, $d\leq dimH_{1}(\hat{C}_{*}\otimes Q)-$

$dimH_{0}(\hat{C}_{*}\otimes Q)$
.
Because and is finitely generated, we have
$H_{0}(\hat{C}_{*}\otimes Z_{2})=Z_{2}$ $H_{0}(\hat{C}_{*})$

two cases: (i) is finite and $\ell_{2}(H_{0}(\hat{C}_{*}))=1$ and (ii)

$H_{0}(\hat{C}_{*})$ $ H_{0}(\hat{C}_{*})\cong$

$Z\oplus$
(odd torsion). In case (i) we have $H_{0}(\hat{C}_{*})*Z_{2}=Z_{2}$ and $ H_{1}(\hat{C}_{*}\otimes$

$Z_{2})=(H_{1}(\hat{C}_{*})\otimes Z_{2})\oplus Z_{2}$

by the universal coecient theorem. So,

$d\leq dimH_{1}(\hat{C}_{*}\otimes Q)\leq dim_{Z_{2}}H_{1}(\hat{C}_{*})\otimes Z_{2}=dim_{Z_{2}}H_{1}(\hat{C}_{*}\otimes Z_{2})-1$

.

In case (ii) we have $H_{0}(\hat{C}_{*})*Z_{2}=0$ . So,

$d\leq dimH_{1}(\hat{C}_{*}\otimes Q)-1\leq dim_{Z_{2}}H_{1}(\hat{C}_{*})\otimes Z_{2}-1=dim_{Z_{2}}H_{1}(\hat{C}_{*}\otimes Z_{2})-1$

.

Note that $H_{1}(\hat{C}_{*}\otimes Z_{2})\cong H_{1}(N;Z_{2})=H_{1}(M-F;Z_{2})=H_{1}(M-$

$F;Z)\otimes Z_{2}$ . If $ F=\emptyset$

, then $H_{1}(N;Q)=H_{1}(M;Q)=0$ . Hence, the
result follows from the condition (1).
Under the condition (2) or (3) the natural maps $ H_{0}(\partial\overline{N}(F))\rightarrow$

and $H_{0}(\partial N(F))\rightarrow H_{0}(N)\oplus H_{0}(N(F))$ are injective

$H_{0}(\overline{N})\oplus H_{0}(\overline{N}(F))$

with coecient in $Q$ due to the condition that $F$ has no codimension
one component. Hence, we have the following commutative diagram of
Mayer-Vietoris exact sequences with coecient in $Q$ :
$(_{\overline{J}*},\overline{\iota}_{*})$

$H_{1}(\partial\overline{N}(F))\rightarrow H_{1}(\overline{N})\oplus H_{1}(\overline{N}(F))\rightarrow H_{1}(\overline{M})\rightarrow 0$

$\pi_{*}\downarrow$ $\pi_{*}\oplus\downarrow\pi_{*}$ $\pi_{*}\downarrow$

$(j_{*},i_{*})$

$H_{1}(\partial N(F))\rightarrow H_{1}(N)\oplus H_{1}(N(F))\rightarrow H_{1}(M)\rightarrow 0$ .

372 T. Matumoto

Note that : $H_{1}(\overline{N}(F))\rightarrow H_{1}(N(F))$ is an isomorphism in any coef-

$\pi_{*}$

ficient because they are canonically equal to $H_{1}(F)$ . If $F$ has no codi-
mension two component, we have an exact sequence of groups $Z_{2}\rightarrow$

$i_{*}$

$\pi_{1}(\partial N(F))$
$\pi_{1}(N(F))\rightarrow 0$ . So, $i_{*}$
: $H_{1}(\partial N(F);Q)\rightarrow H_{1}(N(F);Q)$

is onto. Since $\tilde{\iota}_{*}$

: $\pi_{1}(\partial\overline{N}(F))\cong\pi_{1}(\overline{N}(F))$
, $i_{*}$
: $H_{1}(\partial N(F);Q)=$

$H_{1}(\partial\overline{N}(F);Q)^{\sigma_{*}}c_{->}H_{1}(\partial\overline{N}(F);Q)\cong H_{1}(\overline{N}(F);Q)=H_{1}(N(F);Q)$
is
injective. Hence, : $H_{1}(\partial N(F);Q)\rightarrow H_{1}(N(F);Q)$
$i_{*}$
is also an isomor-
phism. So, the condition (2) implies $dimH_{1}(M;Q)-dimH_{1}(M;Q)=$
$dimH_{1}(\overline{N};Q)-dimH_{1}(N;Q)=d$ , which implies the result as before.

Let be a connected component of codimension two. Assume the

$F_{2}$

condition (3). Then, there is an exact sequence . $Z\rightarrow\pi_{1}(\partial N(F_{2}))\rightarrow 0$

If is finite, then $H_{1}(\partial\overline{N}(F_{2});Q)=H_{1}(\partial N(F_{2});Q)=0$ .

$\pi_{1}(\partial N(F_{2}))$

Otherwise is injective or zero if and

$\tilde{J}*:H_{1}(\partial\overline{N}(F_{2});Q)\rightarrow H_{1}(\overline{N};Q)$

only if : $H_{1}(\partial
$j_{*}$ N(F_{2});Q)\rightarrow H_{1}(N;Q)$ is injective or zero respectively.
So, the condition (3) also implies $dimH_{1}(\overline{M};Q)-dimH_{1}(M;Q)=$
$dimH_{1}(\overline{N};Q)-dimH_{1}(N;Q)=d$ , which completes a proof. Q.E.D.

Remark. Probably we need not to assume the existence of equiv-

ariant normal disk bundle; it suces that $F\times CP^{2}$ has a compact
invariant manifold neighborhood in so that $\overline{N}(F\times CP^{2})$ $\overline{M}\times CP^{2}$

$F\times CP^{2}c_{-\succ}\overline{N}(F\times CP^{2})$

is a homotopy equivalence and $\partial\overline{N}(F\times$

is a spherical homotopy fibration.

$CP^{2})\rightarrow\overline{N}(F\times CP^{2})$

The following lemmas are not new but we list them up to quote in
the proof of Theorem.

Lemma 2.5. Let be an orientation preserving locally linear in- $\sigma$

volution on an oriented closed 4-manifold $M$ with fixed point set F. Let
$F^{2}$
denote the 2-dimensional part of $F$ .
(1) Any isolated point of $F$ can be blow up, that is, there is $x$ $a$

locally linear involution on $M^{*}=M\#\overline{CP}^{2}=(M-x)\cup CP^{1}$ such

$\sigma$

that $\sigma|M^{*}-CP^{1}=\sigma|M-x$ and $\sigma|CP^{1}=id$ . In particular, op- $\sigma$

erates as identity on the newly introduced homology class represented by

$CP^{1}$
and $\pi_{1}(M^{*}/\sigma)=\pi_{1}(M/\sigma)$ . We may take also M#CP2 instead
of ; this comes from that we have an orientation reversing dif-
$M\#\overline{CP}^{2}$

feomorphism of $RP^{3}$ .
(2) (Freedman-Quinn) admits an equivariant normal disk bundle $F^{2}$

$N(F^{2})$ in $M$ .
 Homologically Trivial Smooth Involutions 373

(3) ( $G$ -signature theorem)

sign $(-1, M)=e(F^{2})$ ,

where $e(F^{2})$ denotes the total Euler number of the normal bundle of $F^{2}$

and-l stands for the involution concerned.

Proof (1) Since is locally linear, we have a local complex co-

$\sigma$

ordinate in a disk neighborhood $U$ of so that $x=(0,0)$ and

$(z_{1}, z_{2})$ $x$

$\sigma(z_{1}, z_{2})=(-z_{1}, -z_{2})$ . Take a homogeneous coordinate of $CP^{1}$ $[\zeta_{1}, \zeta_{2}]$

and consider on the product space $U\times CP^{1}$ the subset defined $U^{*}$

by $z_{1}\zeta_{2}-z_{2}\zeta_{1}=0$ . It is easy to see that is a complex surface $U^{*}$

in $U\times CP^{1}$
, the projection : gives an identification of
$\pi$ $U^{*}\rightarrow U$

$U^{*}-\pi^{-1}(0,0)$ with $U-(0,0)$ , the preimage $(0, 0)$ of $(0, 0)$ is $\times CP^{1}$

isomorphic to $CP^{1}$ . Consider a holomorphic involution $(\sigma|U)\times id$ on

$U\times CP^{1}$
. Then, we get a holomorphic involution on such $\sigma|U^{*}$ $U^{*}$

that $\sigma|U^{*}-\pi^{-1}(0,0)=\sigma|U-(0,0)$ and $\sigma|(0,0)\times CP^{1}=id$ . Define

$M^{*}=(M-U)\cup U^{*}$ and $\sigma|M^{*}-U=\sigma|M-U$ . Then, $M^{*}-CP^{1}=M-x$

and $M^{*}$
is dieomorphic to because $[CP^{1}]^{2}=-1$ . Since
$M\#\overline{CP}^{2}$

$\partial U^{*}/\sigma=\partial U/\sigma=RP^{3}$ and $\pi_{1}(U^{*}/\sigma)=\pi_{1}(U/\sigma)=0$ , we have

$\pi_{1}(M^{*}/\sigma)=\pi_{1}(M/\sigma)$ by the van Kampen theorem.

(2) Since is a manifold near

$ M/\sigma$ and is a locally flat subman-
$F^{2}$ $F^{2}$

ifold, $F^{2}$
admits a normal disk bundle due to Freedman-Quinn [6;9.3].
So, a lifting gives an equivariant normal disk bundle.
(3) In the smooth case $G$ -signature theorem is due to Atiyah-Singer
[2] but has many elementary proofs at least in our case of dimension 4
and semi-free, for example, in Gordon [8]. These elementary proofs can
apply also to a locally linear involution, because it admits an equivariant
tubular neighborhood of by (2). See also the comments in Edmonds
$F^{2}$

[5;\S 4]. Q.E.D.

Lemma 2.6 (Edmonds $[5;Prop$ . 3.1&3.2]). Let $M$ be a connected
oriented spin 4-manifold and a locally linear involution that preserves
$\sigma$

orientation and some spin structure. Then, the fixed point set $F$ , if
non-empty, consists either of isolated points or of orientable surfaces.
In the smooth case the codimension homogeneity modulo 4 is proved
by Atiyah-Bott [1] and the orientability of surfaces has many proofs
including Edmonds [4]. The proof in the locally linear case is given in
Edmonds [5].

\S 3. Proof of Theorem 1
Since $H_{1}(M;Z_{2})=0$ , the spin structure on $M$ is unique and we may
374 T. Matumoto

assume that preserves the spin structure. Lemma 2.6 implies that the
$\sigma$

fixed point set $F$ consists either of isolated points or of orientable sur-
faces. If $F$ consists of isolated points, then by the $G$ -signature theorem
described as Lemma 2.5 (3) sign $(-1, M)=0$ . Hence, sign $M$ $=0$ be-
cause operates as identity on $H_{2}(M;Q)$ . So, we may assume that
$\sigma$

$F$ consists of orientable surfaces. In particular, is also a mani- $ M/\sigma$

fold. Note that has an equivariant normal disk bundle $N(F)$ in $M$ by
$F$

Lemma 2.5 (2).

Since $H_{*}(M/\sigma;Q)=H_{*}(M;Q)^{\sigma_{*}}$ , $H_{1}(M;Q)=0$ and $\sigma_{*}|H_{2}(M;Q)$
$=id$ , we have the equality $\chi(M/\sigma)=\chi(M)$ of Euler numbers. Put
$\chi=\chi(M)$ . Then, from the formula $\chi(M)=2\chi(M/\sigma)-\chi(F)$ we
get also $\chi(F)=\chi$ . So, $F$ contains at least numbers of components $\chi/2$

of . Note that $S^{2}$

$M$ has an even intersection form : $ H_{2}(M;Z)/tor\times$ $q_{M}$

$H_{2}(M;Z)/tor$ and hence $\chi=\chi(M)$ is even. Let

$\rightarrow Z$
$F=S_{1}^{2}$ , , $\ldots$ $S_{\chi/2}^{2}$

be the subset of $F$ consisting of numbers of . Since $H_{1}(M/\sigma;Q)=$ $\chi/2$ $S^{2}$

$H_{1}(M;Q)^{\sigma_{*}}=0$ , we have $\chi=2+b_{2}(M/\sigma)>b_{2}(M/\sigma)$ . Taking account

of and Lemma 2.5 (2), we can apply Lemma 2.3 (3)

$[S_{i}^{2}]_{M/\sigma}^{2}=2[S_{i}^{2}]_{M}^{2}$

for $p=2$ and $ F\subset M/\sigma$ . So, by Lemma 2.3 (1) and (2) there is a sub-
union $F$ of connected components of $F$ such that we have a branched
covering of with branch locus $F$ , that is, $(M, \sigma, F\subset F)$ sat-
$ M/\sigma$

isfies the condition of Lemma 2.1 except $F\neq F$ . Note here that
$H_{1}(\partial N(x);Z)\rightarrow H_{1}(\partial N(S_{i}^{2});Z)$ is a surjection for any of . If $x$ $S_{i}^{2}$

$F\neq F$ , then Lemma 2.1 implies that there is a connected 2-sheet

unbranched covering of $M$ . But this contradicts the condition that

$H^{1}(M;Z_{2})=Hom(H_{1}(M;Z), Z_{2})=Hom(\pi_{1}(M), Z_{2})=0$ . This means
$F=F$ . Hence, $F=F$ , that is, $F$ consists of numbers of . $\chi/2$
$S^{2}$

Since the intersection form of $M$ is even, we can also apply $q_{M}$

Lemma 2.3 (3) for $p=2$ and $F\subset M$ . By Lemma 2.3 (1) and (2) there is
a non-trivial element of $H^{1}(M-F;Z_{2})$ which takes non-zero value on
for some . This means that there is a branched covering
$H_{1}(\partial N(S_{i}^{2});Z)$ $i$

: $M$
$\tilde{\pi}$

with branch locus

$\rightarrow M$
; a locally linear involution on $F_{1}\subset F$ $\tau$

with fixed point set . So, there is a non-trivial element of $H^{1}(M$

$\overline{M}$
$F_{1}$ $-$

; ) which takes non-zero value on

$F_{1}$ $Z_{2}$
for every . $H_{1}(\partial N(S_{i}^{2});Z)$ $S_{i}^{2}\subset F_{1}$

Because $H^{1}(M;Z_{2})=0$ , this implies that (i) the homology classes of the
connected components of are linearly dependent in $H_{2}(M;Z_{2})$ or (ii) $F_{1}$

they are independent and generate a submodule of $L=H_{2}(M;Z)/tor$ $S$

so that contains a non-trivial 2-torsion according to the last part

$\overline{S}/S$

of Lemma 2.3 (2). Assume that $F_{1}\neq F$ . In case (i) the homology
classes of the connected components of are also linearly dependent $F_{1}$

in $H_{2}(M/\sigma;Z_{2})$ and this leads to a contradiction with $H^{1}(M;Z_{2})=0$

 Homologically Trivial Smooth Involutions 375

through Lemma 2.3 (1) and Lemma 2.1 as before. In case (ii) notice that
is the submodule generated by the homology classes of the connected
$\pi_{*}S$

components of in $H_{2}(M/\sigma;Z)/tor$ for the projection : $M$

$F_{1}$
. $\pi$ $\rightarrow M/\sigma$

Since is an isomorphism,
$\pi_{*}|S$ is isomorphic to . Note
$\pi_{*}\overline{S}/\pi_{*}S$ $\overline{S}/S$

also that . Then, contains a non-trivial

$\pi_{*}\overline{S}/\pi_{*}S\subset\overline{\pi_{*}S}/\pi_{*}S$ $\overline{\pi_{*}S}/\pi_{*}S$

2-torsion. We can apply Lemma 2.3 (2) for $p=2$ and and we $ F_{1}\subset M/\sigma$

get the same contradiction with $H^{1}(M;Z_{2})=0$ by applying Lemma 2.1

for $(M, \sigma, F_{1}\subset F)$ since we have assumed $F_{1}\neq F$ . Hence, $F_{1}=F$ , that
is, the branch locus for : is also $F$ and $\chi(\overline{M})=\chi(M)$ .
$\tilde{\pi}$
$\overline{M}\rightarrow M$

We will show that $\ell_{2}(H_{1}(M-F;Z))=1$ . Since $H^{1}(M;Z_{2})=0$ ,

it is equivalent to say $\ell_{2}(Ker(H_{1}(M-F;Z)\rightarrow H_{1}(M;Z)))=1$ . Put
$N=M-$ Int $N(F)$ and consider the following commutative diagram:
$H_{1}(\partial N(F);Z)\rightarrow H_{1}(N;Z)\rightarrow H_{1}(N, \partial N(F);Z)$

$\downarrow$ $\downarrow\cong$

$H_{1}$
( $M$ ; Z) $\rightarrow H_{1}(M, N(F);Z)$

Since the horizontal sequence is exact, any element of $Ker(H_{1}(N;Z)=$

$(M -F;Z)\rightarrow H_{1}(M;Z))$ comes from $H_{1}(\partial N(F);Z)$ .
$H_{1}$
We know
that there is an element of $Hom(H_{1}(M-F;Z), Z_{2})$ which takes non- $\alpha$

zero value on for every

$H_{1}(\partial N(S_{i}^{2});Z)$ in $F$ . Now we assume that $S_{i}^{2}$

$\ell_{2}(Ker(H_{1}(M-F;Z)\rightarrow H_{1}(M;Z)))\geq 2$ . Then, we have some element

of $Hom(H_{1}(M-F;Z), Z_{2})$ which is dierent from , that is, takes zero

$\beta$ $\alpha$

value on for at least one . Note that we used here the

$H_{1}(\partial N(S_{i}^{2});Z)$ $i$

special property of . Let $F$ be the subset of $F$ removed such

$Z_{2}$
o. $S_{i}^{2}$

Since $F\neq F$ , the same argument as the above paragraph can be applied
again and get a contradiction with the condition $H^{1}(M;Z_{2})=0$ .
Now since $H_{1}(M;Q)=0$ and $F$ consists of numbers of , $\chi/2$
$S^{2}$

$\ell_{2}(H_{1}(M-F;Z))=1$ implies by Lemma 2.4. So, $b_{1}(\overline{M})=0$ $\chi(\overline{M})=$

$\chi(M)$ implies $b_{2}(\overline{M})=b_{2}(M)$ . Hence, $H_{2}(M;Q)=H_{2}(\overline{M};Q)^{\tau_{*}}$ implies

, that is, $\tau_{*}=id$ on

$H_{2}(\overline{M};Q)^{\tau_{*}}=H_{2}(\overline{M};Q)$
. Therefore, $H_{2}(\overline{M};Q)$

sign $(-1, \overline{M})=sign\overline{M}$ . Recall that sign $(-1, M)=sign$ $M$ and the G-
signature theorem says that

sign $(-1, M)=\sum_{i=1}^{x/2}[S_{i}^{2}]_{M}^{2}=\sum_{i=1}^{x/2}2[S_{i}^{2}]_{\overline{M}}^{2}=2$

sign $(-1, \overline{M})$
.

On the other hand sign $M$ because $=sign\overline{M}$ $H_{2}(M;Q)=H_{2}(\overline{M};Q)^{\overline{\tau}_{*}}$

$=H_{2}(\overline{M};Q)$
. Hence, sign $M=0$ . This completes a proof of Theorem 1.
376 T. Matumoto

References
[1] M. F. Atiyah and R. Bott, A Lefshetz fixed point formula for elliptic
complexes II. Applications, Ann. Math., 88 (1968), 451-491.
[2] M. F. Atiyah and I. M. Singer, The index of elliptic operators: III, Ann.
Math., 87 (1968), 546-604.
[3] W. Barth, C. Peters and A. Van de Ven, Compact Complex Surfaces,
Springer-Verlag, Berlin Heiderberg New York Tokyo, 1984.
[4] A. Edmonds, Orientability of fixed point sets, Proc. Amer. Math. Soc.,
82 (1981), 120-124.
[5] , Aspects of group actions on four-manifolds, Topology Appl., 31

(1989), 109-124.
[6] M. Freedman and F. Quinn, Topology of -manifolds, Princeton Math.
$4$

Ser. 39, Princeton Univ. Press, Princeton, 1990.

[7] R. H. Fox, Covering space with singularities, in Algebraic Geometry
and Topology, Princeton Univ. Press, Princeton, 1957, pp. 243-257.
[8] C. McA. Gordon, On the $G$-signature theorem in dimension four, in
A la Recherche de la Topologie Perdue, Birkha\"user, Boston, 1986,
pp. 159-180.
[9] R. C. Kirby and L. C. Siebenmann, Normal bundles for codimension
2 locally flat imbeddings, in Geometric Topology, Lecture Notes
in Math. 438, Springer-Verlag, Berlin Heidelberg New York, 1975,
pp. 310-324.
[10] V. V. Nikulin, Integral symmetric bilinear forms and some of their ap-
plications, Math. USSR Izvestia, 14 (1980), 103-167.
[11] V. A. Rokhlin, Two-dimensional submanifolds of four-dimensional man-
ifolds, Functional Anal. Appl., 5 (1971), 39-48.
[12] R. Schultz, Problems submitted to the A.M.S. summer research confer-
ence on group actions, in Group Actions on Manifolds, Contemp.
Math. 36, Amer. Math. Soc., Providence, 1984, pp. 513-568.
[13] M. Sekine, On homology of the double covering over the exterior of a
surface in -sphere, Hiroshima Math. J., 21 (1991), 419-426.
$4$

Department of Mathematics
Faculty of Science
Hiroshima University
Higashi-Hiroshima 724
Japan