Jennifer Schultens

Graduate Studies
in Mathematics

Volume 151
Introduction
to 3-Manifolds

Jennifer Schultens

Graduate Studies
in Mathematics

Volume 151

American Mathematical Society

Providence, Rhode Island
 EDITORIAL COMMITTEE
David Cox (Chair)
Daniel S. Freed
Rafe Mazzeo
Gigliola Staffilani

2010 Mathematics Subject Classification. Primary 57N05, 57N10, 57N16, 57N40, 57N50,
57N75, 57Q15, 57Q25, 57Q40, 57Q45.

For additional information and updates on this book, visit

www .ams.org/bookpages/gsm-151

Library of Congress Cataloging-in-Publication Data

Schultens, Jennifer, 1965-
Introduction to 3-manifolds / Jennifer Schultens.
pages cm - (Graduate studies in mathematics; v. 151)
Includes bibliographical references and index.
ISBN 978-1-4704-1020-9 (alk. paper)
1. Topological manifolds. 2. Manifolds (Mathematics) I. Title. II. Title: Introduction to
three-manifolds.

QA613.2.S35 2014
514'.34-dc23
2013046541

Copying and reprinting. Individual readers of this publication, and nonprofit libraries
acting for them, are permitted to make fair use of the material, such as to copy a chapter for use
in teaching or research. Permission is granted to quote brief passages from this publication in
reviews, provided the customary acknowledgment of the source is given.
Republication, sy stematic copying, or multiple reproduction of any material in this publication
is permitted only under license from the American Mathematical Society. Requests for such
permission should be addressed to the Acquisitions Department, American Mathematical Society,
201 Charles Street, Providence, Rhode Island 02904-2294 USA. Requests can also be made by
e-mail to reprint-permission©ams. org.

© 2014 by the author.

Printed in the United States of America.

§ The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http: I /www.ams.org/

10 9 8 7 6 5 4 3 2 1 19 18 17 16 15 14
Dedicated to Misha and Esther and our extended family
Contents

Preface ix
Chapter 1. Perspectives on Manifolds 1
§1. 1 . Topological Manifolds 1
§1.2. Differentiable Manifolds 7
§1 .3. Oriented Manifolds 10
§1.4. Triangulated Manifolds 12
§1 .5. Geometric Manifolds 21
§1.6. Connected Sums 23
§1 .7. Equivalence of Categories 25
Chapter 2. Surfaces 29
§2. 1 . A Few Facts about !-Manifolds 29
§2.2. Classification of Surfaces 31
§2.3. Decompositions of Surfaces 39
§2.4. Covering Spaces and Branched Covering Spaces 41
§2.5. Homotopy and Isotopy on Surfaces 45
§2.6. The Mapping Class Group 47
Chapter 3. 3-Manifolds 55
§3.1. Bundles 56
§3.2. The Schonflies Theorem 62
§3.3. 3-Manifolds that are Prime but Reducible 71

-
v
vi Contents

§3.4. Incompressible Surfaces 72

§3.5. Dehn's Lemma * 75
§3.6. Hierarchies * 80
§3.7. Seifert Fibered Spaces 87
§3.8. JS J Decompositions 96
§3.9. Compendium of Standard Arguments 98
Chapter 4. Knots and Links in 3-Manifolds 101
§4. 1 . Knots and Links 101
§4.2. Reidemeister Moves 106
§4.3. Basic Constructions 108
§4.4. Knot Invariants 113
§4.5. Zoology 1 18
§4.6. Braids 122
§4.7. The Alexander Polynomial 126
§4.8. Knots and Height Functions 128
§4.9. The Knot Group * 137
§4. 10. Covering Spaces * 139
Chapter 5. Triangulated 3-Manifolds 143
§5. 1 . Simplicial Complexes 143
§5.2. Normal Surfaces 148
§5.3. Diophantine Systems 155
§5.4. 2-Spheres * 162
§5.5. Prime Decompositions 166
§5.6. Recognition Algorithms 169
§5.7. PL Minimal Surfaces ** 172
Chapter 6. Heegaard Splittings 175
§6. 1 . Handle Decompositions 175
§6.2. Heegaard Diagrams 180
§6.3. Reducibility and Stabilization 182
§6.4. Waldhausen's Theorem 188
§6.5. Structural Theorems 193
§6.6. The Rubinstein-Scharlemann Graphic 196
§6.7. Weak Reducibility and Incompressible Surfaces 200
Contents vii

§6.8. Generalized Heegaard Splittings 202

§6.9. An Application 208
§6. 10. Heegaard Genus and Rank of Fundamental Group * 212
Chapter 7. Further Topics 215
§7. 1 . Basic Hyperbolic Geometry 215
§7.2. Hyperbolic n-Manifolds** 220
§7.3. Dehn Surgery I 226
§7.4. Dehn Surgery II 232
§7.5. Foliations 238
§7.6. Laminations 243
§7. 7. The Curve Complex 248
§7.8. Through the Looking Glass ** 252
Appendix A. General Position 261
Appendix B. Morse Functions 269
Bibliography 275
Index 283
Preface

This book grew out of a graduate course on 3-manifolds taught at Emory

University in the spring of 2003. It aims to introduce the beginning graduate
student to central topics in the study of 3-manifolds. Prerequisites are kept
to a minimum but do include some point set topology (see [ 109]) and some
knowledge of general position (see [ 128]) . In a few places, it is worth our
while to mention results or proofs involving concepts from algebraic topology
or differential geometry. This should not stop the interested reader with
no background in algebraic topology or differential geometry from enjoying
the material presented here. The sections and exercises involving algebraic
topology are marked with a *, those involving differential geometry with
a ** .
This book conveys my personal path through the subject of 3-manifolds
during a certain period of time (roughly 1990 to 2007) . Marty Scharlemann
deserves credit for setting me on this path. He remains a much appreciated
guide. Other guides include Misha Kapovich, Andrew Casson, Rob Kirby,
and my collaborators.
In Chapter 1 we introduce the notion of a manifold of arbitrary dimen­
sion and discuss several structures on manifolds. These structures may or
may not exist on a given manifold. In addition, if a particular structure
exists on a give n manifold, it may or may not be presented as part of the
information given. In Chapter 2 we consider manifolds of a particular di­
mension, namely 2- manifolds, also known as surfaces. Here we provide an
overview of the classification of surfaces and discuss the mapping class group.
Chapter 3 gives examples of 3-manifolds and standa rd techniques used to
study 3-manifolds. In Chapter 4 we catch a glimpse of the interaction of
pairs of manifolds, specifically pairs of the form ( 3-manifold, 1-manifold) . Of
-

ix
x Preface

particular interest here is the consideration of knots from the point of view
of the complement ( "Not Knot" ) . For other perspectives, we refer the reader
to the many books, both new and old, mentioned in Chapter 4, that provide
a more in-depth study. In Chapter 5 we consider triangulated 3-manifolds,
normal surfaces, almost normal surfaces, and how these set the stage for
algorithms pertaining to 3-manifolds. In Chapter 6 we cover a subject near
and dear to the author's heart: Heegaard splittings. Heegaard splittings are
decompositions of 3-manifolds into symmetric pieces. They can be thought
of in many different ways. We discuss key examples, classical problems,
and recent advances in the subject of Heegaard splittings. In Chapter 7 we
introduce hyperbolic structures on manifolds and complexes and provide a
glimpse of how they affect our understanding of 3-manifolds. We include two
appendices: one on general position and one on Morse functions. Exercises
appear at the end of most sections.
I wish to thank the many colleagues and students who have given me
the opportunity to learn and teach. I also wish to thank the institutions
that have supported me through the years: University of California, Emory
University, Max-Planck-Institut fiir Mathematik Bonn, Max-Planck-Institut
fiir Mathematik Leipzig, and the National Science Foundation.
 Chapter 1

Perspectives
on Manifolds

Manifolds are studied in a variety of categories. We will introduce sev­

eral of these categories and highlight the advantages of the different points
of view. We will focus on topological manifolds (TOP), differentiable (or
smooth) manifolds (DIFF) and triangulated (or combinatorial) manifolds
(T RIANG) . To see these definitions in context, consult introductory Chap­
ters in [ 93), [ 100] , [48] , [ 148) , [ 63) , [ 67] , [ 59) , or [ 37] .

1 . 1 . Topological Manifolds

Definition 1.1.1. A topological n-manifold is a second countable Hausdorff

space M for which there exists a family of pairs { ( M011¢0) } with the follow­
ing properties:
• Va, M 0 is an open subset of M and M = LJ0M o;
• Va, ¢0 is a homeomorphism from M 0 to an open subset of ]Rn .
We often refer to a topological manifold simply as a man ifold.
A pair (M 0,¢0) is called a chart of M. The family of pairs {(M 0,¢0) }
is called an atlas for M . (We sometimes write only {M 0} rather than
{(M 0,¢0) } when the maps ¢0 are not part of the discussion.)
The dimension of an n-manifold is n. Two n-manifolds are considered
equivalent if they are homeomorphic.
Remark 1.1.2. Every subset of ]RN is second countable and Hausdorff.
Thus to show that a subset M of JRN is an n-manifold it suffices to show
that every point in M has a neighborhood homeomorphic to ]Rn .
-

1
2 1. Perspectives on Manifolds

Requiring these homeomorphisms to map onto Rn is equivalent to requir­

ing them to map ont o open subsets of Rn . In exhibiting homeomorphisms
we often opt for the latter, in the interest of simplicity.
Example 1.1.3. One immediate example of an n-manifold is Rn and open
subsets of Rn .
Example 1.1.4. The set §n = {x E Rn+l : llxll = 1} is an n-dimensional
manifold called the n-sphere. Stereographic projection provides a homeo­
morphism h : §n \ { (0, . . . , 0, 1)} --+ Rn . Thus any point x E §n such that
x =I- (0, . . . ' 0, 1) has the neighborhood sn \ { (0, . . . ' 0, 1) } that is homeo­
morphic to Rn . To exhibit a neighborhood of (0, . . . , 0, 1) that is home­
omorphic to Rn we compose the reflection in Rn x {O} with h to obtain
h' : §n \{ (0, . . . , 0, -1)} --+ Rn . Thus §n \{ (0, . . . , 0, - 1) } is a neighborhood
of (0, . . . , 0, 1) homeomorphic to Rn . See Figure 1 . 1 .

------

Figure 1.1. The 2-dimensional sphere.

Example 1.1.5. The set 'll'n = § 1 x x § 1 (n factors) is called the n­

· · ·

torus. The n-torus is a quotient space obtained as follows: In Rn , consider

the group G generated by translations of distance 1 along the coordinate
axes. Then identify two points x, y E Rn if and only if there is a g E G such
that g(x) = y. Denote this quotient map by q: Rn --+ 'll'n .
To see that 'll'n is an n-manifold, let [x] E 'll'n and let U be the sphere
of radius ! centered at x E Rn . Note that g(U) n U = 0 Vg E G. It
follows that q - 1 1q(U) : q(U) --+ U is a homeomorphism. The set of all such
homeomorphisms provides an atlas for Rn . See Figures 1.2 and 1.3.
Example 1.1.6. The result of identifying antipodal points on §n is an n­
manifold. It is called n-dimensional real projective space and is denoted by
RPn . To verify that RPn is an n-manifold, consider a point [x] in RPn . Since
1.1. Topological Manifolds 3

b
Figure 1.2. The 2-dimensional torus is obtained from a squai·e via identifications.

Figure 1.3. The 2-dimensional torus.

sn is an n-manifold, there is a neighborhood U of x and a homeomorphism

h : U -t !Rn . Set -U = a(U), for a : sn -t sn the antipodal map. Then
-U is a neighborhood of -x and -h =ho a is a homeomorphism between
-U and !Rn . After shrinking U, if necessary, a(U) n U = 0. Thus we have
a ho meomorphism [h] : [U] -t !Rn . The set of all such homeomorphisms
defines an atlas for IRPn . See Figure 1 .4.

a
Figure 1.4. The projective plane obtained from a bigon via identifications.

Definition 1 . 1 . 7. Let M be an n-manifold. A p-dimensional submanifold

of M is a closed subset L of M for which there exists an atlas { (Ma, <l>a)}
4 1. Perspectives on Manifolds

of M such that 'Vx E L there is a chart in the atlas with x E Ma and

</Ja (L n Ma ) = {O} x JRP c Rn .
Remark 1 . 1 .8. A submanifold is itself a manifold.
Example 1 . 1 .9. The equatorial circle in the 2-sphere indicated in Figure
1 . 1 is a submanifold of the 2-sphere.
Definition 1 . 1 . 10. Let L, M be manifolds. A map f : L -+ M is an
embedding if it is a homeomorphism onto its image f(L) and f (L) is a
submanifold of M.
Example 1 . 1 . 1 1 . If L is a submanifold of M, then the inclusion map i :
L -+ M of an abstract copy L of L to L c M is an embedding.
We will also consider a slightly larger class of objects:
Definition 1 . 1 . 12. Set Hn = {(x 1 , . . . , xn ) E Rn : x 1 2 O} . An n-manifold
with boundary is a second countable Hausdorff space M with an atlas such
that Va, </>a is a homeomorphism from Ma to an open subset of Rn or Hn .
The boundary of M is the set of all points in M that have a neighbor­
hood homeomorphic to Hn but no neighborhood homeomorphic to Rn . The
boundary of M is denoted by 8M. Points not on the boundary are called
interior points. Two n-manifolds with boundary are considered equivalent
if they are homeomorphic.
Example 1 . 1 . 13. The set Rn = {x E Rn : ll x l l :S 1} is an n-dimensional
manifold with boundary called the n-ball. For interior points, there is noth­
ing to check (because the identity map on Rn provides the required home­
omorphism) . For boundary points, an extension of the map obtained by
stereographic projection provides the required homeomorphism. See Fig­
ure 1.5.

Figure 1 . 5 . The 2-ball is also called the disk.

1. 1. Topological Manifolds 5

Example 1 . 1 . 14. The pair of pants is a 2-manifold with boundary. See

Figure 1.6.

Figure 1 . 6 . The pair of pants.

Example 1 . 1 . 15. The 1-holed torus is a 2- manifold with boundary. See

Figure 1 .7.

0 'CY

Figure 1 . 7. The I-holed torus.

Definition 1 . 1 . 16. Let M be an n-manifold with boundary. A p-dimen­

sional submanifold of M is a closed subset L of M for which there is an atlas
{ (Ma, ef>a)} of M and p E {O, . . . , n} such that \::I x E L in the interior of M
there is a chart in the atlas such that x E Ma and
=
ef>a(L n Ma) {O} x JR.P c IR.n
and \::I x E L in the boundary of M there is a chart in the atlas such that
x E Ma and

and such that

ef>a (x) E {O} x ()HP C aHn .
Remark 1 . 1 . 17. The boundary, aM, of an n-manifold M is not a subman­
ifold of M, though it is an (n - 1)-dimensional manifold that is contained
in M.
Example 1 . 1 . 18. The diameter of the disk pictured in Figure 1.8 is a
submanifold of a manifold with boundary.
6 1. Perspectives on Manifolds

Figure 1.8. A submanifold of the disk.

Definition 1 . 1 . 19. We say that the n-manifold M is closed if M is compact

and 8M = 0 .
Example 1 . 1 .20. Spheres and tori are examples of closed manifolds.

In the TOP category, that is, when we study manifolds from the point of
view in this section, we are interested in continuous maps between manifolds.
Example 1 . 1.21. Projection from 1£'2 = § 1 x § 1 onto the second factor is a
continuous map between manifolds.
To catch a glimpse of intriguing topological manifolds browse [78].

Exercises
Exercise 1. Convince yourself that the statements in Remark 1.1.2 are
true.
Exercise 2. Prove that the product of two manifolds is a manifold. What
can you say about its dimension? (This provides an alternate proof of the
fact that ']['n is a manifold.)
Exercise 3. Show that the boundary of an n-manifold with boundary is
an (n - 1)-manifold without boundary. See for instance the 1-holed torus in
Figure 1 .7.
Exercise 4. By drawing pictures, convince yourself that the surface with
boundary called the pair of pants is aptly named.
Exercise 5. Argue that the I-manifolds pictured in Figure 1.9 are equiva­
lent. (You needn't give a formal proof.)
1.2. Differentiable Manifolds 7

Figure 1.9. Homeomorphic I-manifolds.

1 . 2 . Differentiable Manifolds

Recall the following notion from calculus:

Definition 1.2.1. A map f : IR.n -+ IR.m is said to be Cq if it has continuous
partial derivatives of order q. A map is said to be smooth (or C00) if it has
partial derivatives of all orders.
Definition 1.2.2. A Cq_manifold, for q E
[O, oo] , is a topological mani­
fold M with an atlas that satisfies the additional requirement of being Cq,
meaning that for any pair of charts (M00 </>a), (M13, </>13) in this atlas, the
map </>13 o </>� 1 (where it is defined) is Cq. A C00-manifold is also called a
differentiable, or smooth, manifold.
Given an atlas for a manifold, a map of the form </>13 o </>� 1 is called a
transition map and is denoted by <l>a/3 ·
Example 1 .2.3. One immediate example of a smooth manifold is again IR.n
or open subsets of IR.n . These manifolds admit an atlas with a single chart.
So the condition on transition maps is vacuously true.
Example 1 .2.4. In Example 1 . 1 .4 we saw that §n is an n-manifold by
exhibiting an atlas with two charts:
(§n \(O, . . . , 0, 1), h) , (§n \(O, . . . , 0, - 1 ) , h' )
where

and
h1(x 1 , . . . , Xn + 1 ) = 1 + 1 + (xi , . . . , xn ) ·
Xn I
Thus to show that §n is a smooth manifold, we need to show that the
transition maps h' o h -1 and h o (h') -1 are smooth. We will only check that
8 1. Perspectives on Manifolds

h' o h-1 is smooth. (The case ho ( h' )-1 is analogous.) Here

h - 1 (y1 , . . . , Yn ) =
- 1 + Yf + . . · + y; )
(
2y 1 2yn
. . .
1 + Yf + . . · + Y� ' ' 1 + Yf + . . + Yt 1 + Yf + . . · + Y� '
·

h 'o h - 1 (yi , . . . , Yn ) = l (y , . . Y
Y12 + · · · + Yn2 1 · ' n ) ·
It follows that h'o h - 1 is smooth except at the origin where the composition
of maps is not defined. Thus §n is a smooth manifold.
Example 1.2.5. In the exercises you proved that the product of manifolds
is a manifold. Since the product of smooth maps is smooth, the product
of smooth manifolds is a smooth manifold. It follows that 'Jl'n is a smooth
manifold.
In calculus we learn about differentiable maps from ]Rn to ]Rm. Some
concepts extend to manifolds.
Definition 1.2.6. Let M be a manifold with atlas { (Ma, <Pa) } and let N
be a manifold with atlas { (N13, 1/113) } . We say that the map f : M -+ N is
Cq if 'i::fa , /3, the map 1/113 o fo¢-;;1 (where it is defined) is Cq.
Definition 1.2.7. A Cq-map between Cq-manifolds with a cqnverse is
called a Cq-diffeomorphism. A C00-diffeomorphism is simply called a diffeo­
morphism.
Remark 1.2.8. The map f: JR -+ JR given by f(x) = x3 is a C00-map but
is not a diffeomorphism because its derivative is singular at 0. (In fact, it is
not even a C 1 -diffeomorphism.)
Definition 1.2.9. Two Cq-manifolds are considered equivalent if there is a
Cq-diffeomorphism between them.
In the exercises, you will extend the notion of submanifold, manifold
with boundary, and submanifold of a manifold with boundary to the DIFF
category, that is, to the setting in which manifolds are considered in this
section.
In the DIFF category we are interested in smooth maps between mani­
folds.
Example 1.2.10. Projection from 11'2 = § 1 x § 1 onto the second factor is a
smooth map between manifolds.
In Appendix A, we introduce the notion of transversality in the category
of DIFF manifolds. Another concept that is best described in this category is
that of a Morse function. We discuss the concept in more detail in Appendix
B but provide the basic definition here.
Exercises 9

Definition 1.2.11. Let M be a Cq-manifold for q 2: 1 , x E M, and (Ma, ef>a)

a chart with x E Ma. We say that x is a critical point of a function f :
M -t IR. if it is a critical point of f o¢;1 .
Definition 1.2.12. A critical point of a function g : IR.n -t IR. is non­
degenerate if the Hessian of g is non-singular at x . For M, x , (Ma, ef>a), f
as above, we say that x is a non-degenerate critical point of f if it is a
non-degenerate critical point of f o¢;1 .
Remark 1.2.13. In the exercises, you will verify that these two definitions
do not depend on the chart used.
Definition 1.2.14. A Morse function on a manifold M is a smooth function
f : M -t IR. that satisfies the following:
• f has only non-degenerate critical points;
• distinct critical points of f take on distinct values.

a•••• • • • • • • • • • •
----""'···�··· ···················································

Figure 1 . 10. Projection onto the z-axis.

Example 1.2.15. The torus pictured in Figure 1 . 10 sits in IR.3 and projec­
tion onto the third coordinate defines a function. This function has four
critical points: (1) a minimum; (2) two saddle points; (3) a maximum. You
will verify in the exercises that all these critical points are non-degenerate.
Since they occur at distinct levels, this function is a Morse function.

Exercises

Exercise 1. Verify that the definitions of critical point and non-degenerate

critical point for a function f : M -t IR. are independent of the chart used.
10 1. Perspectives on Manifolds

Exercise 2. Verify that the critical points of the function given in Example
1.2.15 are non-degenerate.
Exercise 3. Find other Morse functions on the torus by drawing other
pictures representing the torus in IR.3 .
Exercise 4. Compute the quantity #minima + #maxima- #saddles for the
Morse function in Example 1.2.15 and the Morse functions from Exercise 3.
Exercise 5. Extend the notions of submanifold, manifold with boundary,
and submanifold of a manifold with boundary to the DIFF category.

1 . 3 . Oriented Manifolds
The notion of orientation allows us to partition manifolds into two types:
orientable and non-orientable. We introduce the notion here in the DIFF
category, though analogous notions exist in some other categories, most
notably TOP (discussed above) and TRIANG (discussed below) .
Definition 1.3.1. A C00-manifold M with boundary is orientabl e if it has
minant. Otherwise M is non-orientable. An orientation of M is such an
an atlas such that the Jacobians of all transition maps have positive deter­
atlas. We often write ( M, {<Pa}) to denote an oriented manifold.
Example 1.3.2. The annulus A = § 1 x (-1, 1) is orientable. It can be
covered by two charts:
{ ( { (eix, t): -� < x < 5: , - 1 < t < 1 } , ¢1 ) ,
( { (eix, t) : < x < : - 1 < t < 1 } , ¢2) }
: 3 9 ,

where ¢>i((eix, t )) (x, t ) . The transition maps are

=

¢2 o ¢>1 1 ( (x, t)) {

=
��'�) 27r, t ) �; �1r!f <<xx <<s}'
and
'f'l 'f'2 ((x, t))
,/,. o ,/,._1
_

-
{ ((x -t)27r, t ) ifif 3;7r << xx << ;;
x,
1 9
s
,
.
4
Thus the Jacobians of all transition maps (where they are defined) have
positive determinant.
Example 1.3.3. The Mobius band is the surface with boundary obtained
by identifying two sides of a rectangle, as pictured in Figure 1 . 1 1 . It is non­
orientable. Intuitively speaking, this is because in an orientable surface,
there is a well-defined notion of, for instance, "above" , or "being to the
1.3. Oriented Manifolds 11

right". (You can make sense out of this i n IR.2 . The condition on Jacobians
guarantees that the notion patches together correctly on overlapping charts.)
Now consider the core curve of the Mobius band, as pictured in Figure 1 . 12,
and imagine a curve "above" or "to the right" of the core curve. Such a

------------1
curve does not exist, and so the Mobius band is non-orientable.
a

J
l ____________ J
Figure 1 . 1 1 . The Mobius band.

Figure 1 . 1 2 . The core of the Mobius band.

Let M be a differentiable manifold with two orientations (M, { ef>a}) and

(M, {1/113 } ) . The subset of M on which ef>a o1f{i 1 is defined and has a Jacobian
with a positive determinant is the subset where the orientations of (M, { ef>a})
and (M, {1/113 }) are said to coincide. The subset where ef>a o 1/lfi 1 is defined
and has a Jacobian with a negative determinant is the subset where the
orientations of ( M, { ef>a}) and ( M, {1/113}) are said to differ. These two sets
are open. Thus for a connected manifold M, two orientations either coincide
on all of M or differ on all of M.
Given an oriented manifold (M, { ef>a } ) , we can create an oriented man­
ifold ( M, { 1/ia}) for which the orientations differ on all of M. Indeed, for
each ef>a : Ua --t IR.n , defined by ef>a (x) = (x 1 , x 2 , . . . , xn ) , we substitute
1/ia : Ua --t IR.n , defined by 1/ia (x) = (-x 1 , x 2 , . . . , xn ) · The resulting orien­
tation is called the opposite orientation of the original one. We denote the
resulting oriented manifold by -M.
If M is an oriented manifold with boundary, then there is a natural
orientation on 8M, called the induced orientation of 8M. See for instance
[48] . We will not discuss the technicalities here.
On a related note, given an oriented manifold M, we sometimes assign
an orientation to an orientable submanifold. When the dimension of the
submanifold is one less than the dimension of the manifold in which it lies,
I2 1. Perspectives on Manifolds

then choosing an orientation on the submanifold is equivalent to choosing a

smoothly varying normal direction. For details see [48] .
Definition 1.3.4. For oriented C00-manifolds (M, {4>a } ) , (N, {1/J.a}) of the
same dimension, a smooth map h : M -+ N is orient at ion-preserving if
the Jacobians of the maps 1/J.a o h o 4>� 1 ( where they are defined ) all have
positive determinant. If they all have negative determinant it is orient at ion­
reversing.
Example 1.3.5. The map f : § 1 -+ § 1 given by f ( e21rix ) = e41rix is
orientation-preserving.
Example 1 .3.6. The map f § 1 -+ § 1 given by f ( e21rix ) = e-21rix is
orientation-reversing.
In a non-orientable manifold M each chart defines a local orientation. If
c is a closed I-dimensional submanifold of M, then the transition maps for
those charts of M that meet c may or may not have Jacobians with positive
determinants. If there is an atlas for M for which all charts that meet
c have transition maps whose Jacobians have positive determinants, then
we say that c is an orient at ion-preserving closed I-dimensional submanifold
of M. If there is no such atlas, then c is an orient at ion-reversing closed
I-dimensional submanifold of M.
In the exercises, you will prove the following:
Lemma 1.3. 7. The manifold M is non-orient able if and only if M cont ains
an orient at ion-reversing closed I-dimensional submanifold.

Exercises

Exercise 1 . Show that §n is an orientable manifold.

Exercise 2. Show that the product of two orientable manifolds is an ori­
entable manifold.
Exercise 3. Show that 'fn is an orientable manifold.
Exercise 4. Prove Lemma 1.3.7.
1 .4. Triangulated Manifolds

In this section we consider manifolds in the TRIANG category. The defini­

tions that lay the foundation for this study ( simplicial complexes and related
notions ) are discussed in greater detail in [148] . See also [102] .
Note that in Definitions l .4. I2 and l .4.I3 we make a subtle departure
from the traditional definitions. This is in line with a contemporary view of
1 .4. Triangulated Manifolds 13

triangulations. For more detail, see Chapter 5, where we will reconcile the

�k+l
two notions.
Definition 1.4.1. Denote the ( k + 1 ) -tuple in with i-th entry 1 and
all other entries 0 by Vi· The set

{aovo akvk : ao + }
� 0, . . . , ak � 0, t ai = 1

[vo, ..., vk]

+ · · ·

i=O
is called the standard (closed} k -simplex and is denoted by or
simply by [s] . The dimension of the standard k-simplex is k.

{aovo akvk : ao , ak t ai }
The set

+ + > 0, . . . > 0, = 1

vo, ..., vk)

· · ·

i=O

[]
is called the standard open k-simplex and is denoted by (
by (s). We also call (s) the interior of s .
or simply

Example 1.4.2. The standard 0-simplex is the point 1 E R This is also

the standard open 0-simplex.

: [s] []
Definition 1.4.3. A k-simplex in a topological space X is a continuous map

]
f ---* X such that s is the standard k-simplex and the restriction f l ( s)
is a homeomorphism onto its image. An open k-simplex is the restriction of
a k-simplex to the interior (s) of [s .
Abusing notation slightly, we will often refer to the image of f as a
k-simplex.
Example 1.4.4. The standard 1-simplex is homeomorphic to an interval.
See Figures 1 . 13 and 1. 14.

Figure 1 . 13. The standard I-simplex.

14 1. Perspectives on Manifolds

• •

Figure 1 . 14. A I-simplex.

Figure 1 . 15. A 2-simplex.

Example 1 .4.5. The standard 2-simplex is a "triangle" with vertices at

(1, 0, 0) , (0, 1 , 0) , and (0, 0, 1). Figure 1 . 15 depicts a 2-simplex.
Example 1 .4.6. The standard 3-simplex is a tetrahedron with vertices at
(1, 0, 0, 0) , (0, 1, 0, 0) , (0, 0, 1 , 0) , and (0, 0, 0, 1 ) . Figure 1 . 16 depicts a 3-
simplex.

Figure 1 . 16. A 3-simplex.

Definition 1.4. 7 (Faces) . For j = 0, . . . , k, a face of the standard k-simplex

[s] is a subset of [s] of the form
{ aovo + + akvk : ai1 = 0, . . . , ai; = O}.
· · ·

The dimension of the face is k - j. The faces of the k-simplex f: [s] -+ X

are the restriction maps fl[t] for [t ] a face of the standard k-simplex [s] . A
0-dimensional face of a simplex is also called a vert ex. A I-dimensional face
of a simplex is also called an edge.
1 .4. Triangulated Manifolds 15

Example 1.4.8. A 0-simplex has only itself as a face.

Example 1.4.9. The standard 1-simplex [vo, v1 ] (and hence all 1-simplices)
has itself as a 1-dimensional face. It also has two 0-dimensional faces, [vo]
and [v1 ] .
Example 1.4.10. A 2-simplex has one 2-dimensional, three 1-dimensional,
and three 0-dimensional faces.
Example 1.4.11. A 3-simplex has one 3-dimensional, four 2-dimensional,
six 1-dimensional, and four 0-dimensional faces.
Definition 1.4.12 (Simplicial complex) . A simplicial complex based on the
topological space X is a set of simplices
K = {! [s] ---+ X}
:

in the topological space X such that

(1) V simplices f E K, all faces of f are in K;
(2) V simplices Ji, h E K, im(f1 i(s1)) nim( f2 1 (s2)) =/: 0 ===> im( fil(s1)) =
im( f2 i (s2)) ·
The dimension of a simplicial complex K is the supremum of the di­
mensions of the simplices in K. We denote the union of the images of the
simplices in K by IKI and call IK I the underlying space of K.
For examples see Figures 1. 17, 1. 18, and 1 . 19. Traditionally, one also
requires that each closed simplex be embedded and that two distinct closed
simplices meet in at most one face. We do not make these assumptions here!

Figure 1 . 17. A I -dimensional simplicial complex.

Definition 1.4.13. A triangulated n-manifold is a pair (M, K) , where M

is a topological n-manifold and K is a simplicial complex based on M such
that
• IK l = M ;
• K is locally finite, i.e. , for every compact subset C of M, the set
{f E K : C n im f =/: 0} is finite;
16 1. Perspectives on Manifolds

Figure 1 . 18. A 2-dimensional simplicial complex.

Figure 1 . 19. A 3-dimensional simplicial complex.

• for f, g E K, restricted to open simplices, the map g -1 of is affine

on its domain.
Here K is called the triangulation of M. For f : [ s] -t M an n-simplex
in K, the pair ( im f, 1 -1 ) is called a simplicial chart of the triangulation.
In this context, we often write K Ua} and denote the collection of all
=

charts on M by { ( im f J; 1 ) } . We also refer to f as a simplex of M.

°', °'

There are related categories of manifolds ( piecewise linear (PL) , combi­

natorial ) that we will not discuss here. Suffice it to say that some distinctions
are subtle.
Example 1 .4. 14. Figure 1.20 depicts a triangulation of §2 .

Example 1 .4.15. Figure 1.21 depicts a triangulation of 1'2 with two 2-

simplices.

Example 1 .4. 16. Figure 1.22 depicts a triangulated cube. It can be inter­
preted as a portion of a triangulation of 1'3 . We obtain a triangulation of
1'3 by identifying eight appropriately chosen reflections of this cube. The
resulting triangulation contains forty 3-simplices.

Theorem 1.4.17. Every compact I -manifold admits a triangulation.

1 .4. Triangulated Manifolds 17

-------

,,.
,,. ....
'
'

Figure 1 . 20. A triangulation of §2 •

a a

Figure 1 . 2 1 . A triangulation of1'2 •

Proving the above theorem is an easy exercise in understanding compact

1-manifolds. Analogous theorems for 2- and 3-manifolds also hold but are
much harder to prove.
Theorem 1.4.18 (T. Rado, B. Kerekjarto) . Every compact 2-manifold ad­
mits a triangulation.
A proof can be found in [4] .
18 1 . Perspectives on Manifolds

'
'
'
'
'
'

'
'
'
'
'
'
....

Figure 1 . 22. A portion of a triangulation of'll'3 .

Theorem 1.4.19 (R. H. Bing, E. Moise) . Every compact 3-manifold admits

a triangulation.
A proof can be found in [101] .
This theorem has been put to extensive use in the study of 3-manifolds.
In view of Lemma 1 .4.28 below, it allows us to build 3-manifolds by identi­
fying 3-simplices along faces. In Chapter 5 we will see a brief introduction
to normal surface theory (after Wolfgang Haken) where this point of view
bears fruit.
Definition 1.4.20. Let K and L be simplicial complexes. We say that a
continuous map </> : IKI -+ ILi is a simplicial map if for every simplex f in
K, there is a simplex g in L such that </> o f g. =

Informally speaking, a simplicial map maps simplices to simplices, but

the latter might be of lower dimension than the former.
Example 1.4.2 1. Figure 1.23 depicts a simplicial map from the top sim­
plicial complex (containing four triangles) to the bottom simplicial complex
(containing two edges) : The two vertices on the left go to the left vertex
below, the two vertices in the middle go to the middle vertex below, and the
two vertices on the right go to the right vertex below. Extend this map of
vertices linearly over the 2-simplices to obtain a simplicial map.
Non-Example. A map that is not simplicial is depicted in Figure 1.24. It
maps the top simplicial complex (containing four triangles) to the bottom
simplicial complex (containing one edge) via nearest point projection (in JR2 ) .
1 .4. Triangulated Manifolds 19

Figure 1 . 23. A simplicial map.

Figure 1 . 24. A map between simplicial complexes that is not simplicial.

Definition 1.4.22. Let Ki , K2 be simplicial complexes and let <P !Ki I -+ :

IK2 I be a map. We say that <P is a simplicial isomorphism if it is simplicial

and a homeomorphism. Two simplicial complexes Ki , K2 are isomorphic if
there is a simplicial isomorphism <P : !Ki l -+ I K2 I ·
Definition 1.4.23. Let (Mi , Ki) and (M2 , K2 ) be triangulated n-manifolds.
The two triangulated n-manifolds are considered equivalent if there is a
simplicial isomorphism¢: Mi-+ M2 .
Definition 1.4.24. A subcomplex of a simplicial complex K is a simplicial
complex L such that f EL implies f EK. We write L c K.
Example 1.4.25. The unshaded area in Figure 1.25 depicts a subcomplex
of the simplicial complex depicted in Figure 1 . 18.
20 1 . Perspectives on Manifolds

Figure 1 . 25. A subcomplex of a simplicial complex.

One concept that is easily defined in the TRIAN G category is the fol­
lowing:
Definition 1.4.26. The Euler characteristic of a finite simplicial complex
K of dimension k is computed via the following formula:

k
x (K) = I.) -l) i #{simplices of dimension i in K} .
i=O
Remark 1.4.27. In fact, Definition 1.4.26 is well-defined for the underlying
space of a simplicial complex. This can be established via a computation
involving the notion of homology from algebraic topology.

The standard k-simplex lies in JR.k+1 and is thus a topological space.

Suppose that (M, K) is a triangulated k-manifold and consider a collection
C that is the disjoint union of copies of standard k-simplices, one copy for
each k-simplex f in K. Now construct a quotient space from C as follows:
We identify points in the collection C if and only if the corresponding points
x E [s] and y E [t] for f : [s] -+ M and g : [t] -+ M in K satisfy f ( x ) = g (y) .
We leave the proof of the following lemma as an exercise.
Lemma 1.4.28. The quotient space M' (in the quotient topology) is home­
omorphic to the topological manifold M.

The observation in Lemma 1.4.28 allows us to think of triangulated

n-manifolds as objects that are constructed out of standard simplices by
identifying faces.
Example 1.4.29. Figure 1.26 describes a triangulation of §3 . The labeling
gives the identification of the edges. 2-dimensional faces are identified if all
three of their edges are identified.
1 . 5 . Geometric Manifolds 21

d d

Figure 1 .26. A triangulation of §3 .

Exercises

Exercise 1. Prove that a simplicial isomorphism has an inverse that is also

a simplicial isomorphism.
Exercise 2. Prove Lemma 1 .4.28. (Hint: The simplices in M piece together
to give a continuous bijection h : M --+ M'. To prove that h-1 is also
continuous, use local finiteness of K to establish that h-1 is a closed map.)
Exercise 3. Find a triangulation of the projective plane.
Exercise 4. How many triangulations of the 2-torus can you find (e.g. , 0,
1, oo)?
Exercise 5. Prove that every compact 1-manifold admits a triangulation.
Exercise 6. Consider the triangulation of §2 in Figure 1.20. Try to add ver­
tices and edges to create a triangulation such that any pair of 2-dimensional
simplices meets in at most one edge.
Exercise 7. Compute the Euler characteristic of the sphere and of the torus
by choosing triangulations.
1 . 5 . Geometric Manifolds
Geometry has evolved since Euclid's time. Euclidean geometry and the other
structures now also considered geometries provide insight into manifolds. We
will assume a basic familiarity with Euclidean geometry in this section. In
Chapter 7, we will develop the basics of hyperbolic geometry. While in this
section we focus on Euclidean manifolds, the definitions carry over verbatim
with the term "Euclidean" replaced by the term "hyperbolic" in order to
22 1. Perspectives on Manifolds

define hyperbolic manifolds. For more on geometric 3-manifolds, see [ 145] ,

[ 1 54] , [ 1 53] , or [126] .
Definition 1.5. 1. An isometry is an invertible map between metric spaces
that preserves distances. A Euclidean isometry, i.e., an isometry f : JRn -+
]Rn , is also called a rigid motion.
Definition 1.5.2. A Euclidean manifold is a topological manifold M with
an atlas that satisfies the additional requirement that the transition maps
are restrictions of Euclidean isometries. We often write (M, {<Pa}) to denote
a Euclidean manifold.
The more general term is that of a geometric manifold. There the ad­
ditional requirement is that the transition maps are isometries in a given
geometry.
Example 1.5.3. An immediate example of a Euclidean manifold is again
]Rn . This manifold admits an atlas with a single chart. So the condition on
transition maps is vacuously true.
Example 1.5.4. It follows from the description of 'Il'n given in Example
1 . 1.5 that 'Il'n is a Euclidean manifold.
Definition 1.5.5. Given two Euclidean manifolds (M, <Pa) , (N, </>13), a map
f : M -+ N is an isometry if Va, f3 the maps ¢�1 o f o '1/J13 are restrictions of
Euclidean isometries.
Definition 1.5.6. Two Euclidean manifolds M, N are considered equivalent
if there is an invertible isometry h : M -+ N.
Remark 1.5.7. On a connected Euclidean manifold there is a well-defined
metric. Metric concepts such as completeness, diameter, volume, etc., are
thus defined for Euclidean manifolds.

Exercises

Exercise 1. Show that among Euclidean 2-manifolds there are infinitely

many inequivalent 2-tori.
Exercise 2. The definition of Euclidean manifolds extends to manifolds
with boundary. Show that among Euclidean manifolds with boundary there
are infinitely many inequivalent annuli.
Exercise 3. Show that §1 is a Euclidean manifold.
Exercise 4 *. Show that §n is not a Euclidean manifold for n > 1.
1 . 6. Connected Sums 23

1 . 6 . Connected Sums

The following two notions are fundamental to the study of manifolds.

Definition 1 .6 .1 . Two continuous maps Jo, Ji : M ---+ N are homotopic
if there is a continuous map H : M x [O, 1] ---+ N such that H(x, 0) Jo =

and H(x, 1) fi(x) Vx E M. The map H is called a homotopy between Jo

=

and Ji .
Definition 1 .6 .2 . Two embeddings Jo, Ji : M ---+ N are isotopic if there
is a continuous map H : M x [O, 1] ---+ N such that H(x, 0) Jo and
=

H(x, 1) fi(x) Vx E M and such that Vt E [O, 1] , the map ft defined by

=

H( , t) is an embedding. The map H is called an isotopy between Jo and Ji .

Two submanifolds So, Si of M are isotopic if their inclusion maps are
isotopic.
The following two theorems are due to Gugenheim (see [47] ) in the
TRIANG category and are also a consequence of the Isotopy Uniqueness of
Regular Neighborhoods Theorem due to Rourke and Sanderson (see [128] ) .
They are fundamental theorems in the study of n-manifolds. We will need
these theorems in the discussion of Definition 1 .6.5.
Theorem 1 .6 .3 . Every orientation-preserving homeomorphism of an n-ball
or n-sphere is isotopic to the identity.
We will prove a special case of Theorem 1 .6.3, known as the Alexander
Trick, in Section 2.5.
Theorem 1 .6 .4. If Bi , B2 are n-balls in the interior of a connected n­
manifold M, then there is an isotopy f : M x I ---+ M such that f( O) J B1 is
,

the identity and f( , l ) J B1 is a homeomorphism onto B2 .

The proof of Theorem 1.6.4 is left as an exercise. Theorem 1 .6.4 ensures
that Definition 1 .6.5 is well-defined. Orientations also play a subtle role in
Definition 1.6.5.
Definition 1 .6.5. Let Mi , M2 be n-manifolds. Delete small open n-balls
Bi from Mi and B2 from M2 . Identify Mi and M2 along the resulting
(n - 1)-sphere boundary components. This results in an n-manifold called
a connected sum of Mi and M2 and is denoted by Mi #M2 . In the case that
Mi and M2 are oriented, we further require that the identification of the
boundaries of Bi and B2 be via an orientation-reversing homeomorphism
(with respect to the induced boundary orientations on 8Bi , 8B2 ) .
The connected sum Mi #M2 is trivial if either Mi or M2 is §n. Other­
wise, Mi #M2 is non-trivial.
24 1. Perspectives on Manifolds

In deleting the small open n-balls, we are making a choice, but Theorem
1.6.4 ensures that this choice is inconsequential. If Mi , M2 are oriented,
then Theorem 1.6.3 tells us that any two choices of identification of 8Bi
and -8B2 are isotopic and it follows that the manifolds obtained via this
identification are homeomorphic. Thus for oriented manifolds Mi , M2 , there
is a unique connected sum (i.e., connected sum is well-defined in the oriented
category) .
For orientable (but not oriented) manifolds it is possible to have two non­
homeomorphic connected sums of Mi and M2 . Specifically, endow Mi , M2
with orientations and consider Mi #M2 versus Mi # (-M2 ) . For examples
of oriented 3-manifolds Mi , M2 for which Mi #M2 =/:- Mi # (-M2 ) , see [63] .
If at least one of Mi , M2 , say Mi , is non-orientable, then there is,
in fact, a unique connected sum Mi #M2 . Indeed, this is because every
non-orientable n-manifold M contains an orientation-reversing closed 1-
dimensional submanifold (by Lemma 1 .3.7) . Isotoping a small n-ball along
such a 1-dimensional submanifold reverses the local orientation on an open
neighborhood of the n-ball Bi and hence also the induced boundary orien­
tation of (Mi \Bi) \8Mi . This isotopy extends to an isotopy between the
two possibilities for connected sums of Mi and M2 , thereby demonstrating
that they are homeomorphic.
In the case of surfaces, the situation. is simpler. Every oriented sur­
face 8 admits an orientation-reversing homeomorphism h : 8 -t 8. Thus
for any pair 8i , 82 of oriented surfaces, 8i #82 8i #h(82 ) 8i # - 82 .
= =

Hence for orientable surfaces, the choice of identifying homeomorphism is

inconsequential and the connected sum is unique.
A word on notation: We denote repeated connected sums by M =

Mi # · · · #Mn . We leave it as an exercise to show that connected sum of

manifolds is associative and commutative. (An example to consider is the
following: The connected sum of four 3-manifolds Mi , . . . , M4 can be ob­
tained in several ways. For instance, take the connected sum by removing
three small 3-balls from Mi , one small 3-ball from M2 , M3 , M4 and identify­
ing the resulting boundary components of Mi with those of M2 , M3 , M4 . On
the other hand, take the connected sum by removing one small 3-ball from
Mi and M4 , two small 3-balls from M2 and M3 and identifying the result­
ing boundary component of Mi with one of M2 's, the resulting boundary
component of M4 with one of M3 's, and the remaining resulting boundary
components of M2 and M3 with each other. Figure 1.27 hints at why the
repeated connected sum is nevertheless well-defined.)
Definition 1 .6 .6 (Prime n-manifold) . An n-manifold M is prime if M =

Mi #M2 implies that either Mi or M2 is the n-sphere, i.e. , that the connected
sum M Mi #M2 is trivial.
=
1 . 7. Equivalence of Categories 25

Figure 1 . 27. The spheres in a prime decomposition.

Remark 1 .6 .7 . An n-manifold is prime if and only if it contains no sepa­

rating essential ( n 1 )-sphere.
-

Exercises

Exercise 1 . Prove Theorem 1.6.4.

Exercise 2. Compute the Euler characteristic of '1!'2 #1R.P2 .
Exercise 3 . Prove that if S1 , 82 are surfaces, then

Exercise 4. Prove that connected sum is associative and commutative.

Exercise 5. Is §2 x § 1 a non-trivial connected sum?

1 . 7. Equivalence of Categories
We have discussed manifolds from several points of view. The key idea,
except in the TRIANG category, is the following: Choose a collection G of
invertible maps of IR.n that is closed under group operations and require the
transition maps to belong to G. The PL category is similar to the TRIANG
category, but there are subtle differences that we do not wish to discuss here.
There are natural maps from all the categories considered here to TOP
(in each case, it is the forgetful functor) . Amazingly, for 2-manifolds and
3-manifolds, these maps are isomorphisms in the case of DIFF and TRI­
ANG! This means that every topological 3-manifold admits a triangulation.
26 1 . Perspectives on Manifolds

Furthermore, any two triangulations of a topological 3-manifold admit a

common subdivision. Much of the work establishing the equivalence of the
categories TOP and TRIANG is due to E. Moise and R. H. Bing. See [102]
and [101] .
It deserves to be mentioned that this equivalence of categories is not
true for higher-dimensional manifolds. This was first shown by Milnor. He
exhibited examples of 7-spheres with distinct differentiable structures, thus
demonstrating that DIFF f:. TOP for 7-dimensional manifolds. To date, it
is unknown whether there are smooth structures on 84 that are not diffeo­
morphic. A candidate for an "exotic" smooth structure on the 4-sphere was
given by Scharlemann. See [132] . Decades later it was proved by Akbulut
that this smooth structure is in fact diffeomorphic to the standard smooth
structure on the 4-sphere. See [5] .
The advantage of the equivalence of the categories TOP, DIFF, and
TRIANG in the case of 3-manifolds is that we can introduce a concept or
prove a theorem in the category that best suits the concept or theorem. For
instance, we introduced the notion of a Morse function in the DIFF cate­
gory and we introduced the notion of Euler characteristic in the TRIANG
category. The notions can be introduced in the other categories as well,
but the definitions are necessarily more cumbersome, so we will not do so
here. We will be moving back and forth between categories when this is
convenient. When categories are equivalent, it is acceptable, though not es­
thetically pleasing, to prove theorems using concepts introduced in distinct
categories.
Recall, for instance, the definition of Euler characteristic in Section 1.4.
We will use the notion of Euler characteristic in the other categories as well.
For instance, given a smooth surface S, we can compute the Euler character­
istic by triangulating the surface and taking the specified alternating sum.
If S contains a finite graph r such that S\r consists of a finite number of
disks, then the Euler characteristic can be calculated directly from r and
S\r. You will prove this in the exercises.
We can compute the Euler characteristic not only for closed surfaces but
also punctured surfaces. Specifically, since points have Euler characteristic
1, the result of removing n points from a surface S is x ( S) - n. One theorem
we will use repeatedly, stated here in the case of 2-manifolds, is the following:
Theorem 1 .7.1 (Poincare-Hopf Index Theorem) . Let S be a surface and
let h : S � IR. be a Morse function. Then

x (S) = # (minima of h) + # (maxima of h) - #(saddles of h) .

For a proof of this therem, see [48] .

Exercises 27

Exercises
Exercise 1. Show that, even in the case of closed 2-manifolds, the categories
DIFF and GEOM are not equivalent.
Exercise 2. Prove that if a surface S contains a finite graph r with v
vertices and e edges and S\r consists of r disks, then
x (S) = v - e + r.

Exercise 3 . Draw pictures to convince yourself that Theorem 1.7. 1 is true.

Exercise 4. Explore how distinct structures on manifolds interact. For
example, read [126] .
 Chapter �

S urfaces

Topological 2-manifolds, also known as surfaces, are well understood. See

[93] . We here give a brief overview of what is known. There are additional
structures on surfaces that add interest and provide tools for the study of
3-manifolds. See Sections 2.5 and 2.6 of this chapter. For an in-depth
treatment of the topics in Sections 2.5 and 2.6, see [2 5] and [39] .

2 . 1 . A Few Facts about !-Manifolds

It is not hard to show that the only compact connected !-manifolds are the
circle and the interval. (You will prove this in the exercises.) In this chapter
we will often consider !-dimensional submanifolds of surfaces. In subsequent
chapters we will consider !-dimensional and 2-dimensional submanifolds of
3-manifolds.
Definition 2 .1 .1 . A simple closed curve in a surface S is a compact con­
nected !-dimensional submanifold of S without boundary. A simple arc in
a surface S is a compact connected !-dimensional submanifold of S with
non-empty boundary.
Remark 2 .1 .2 . The only compact connected 1-manifold without boundary
is the circle. So we also think of a simple closed curve in a surface S as
an embedding of the circle into S. The only compact connected 1-manifold
with non-empty boundary is the interval. So we also think of a simple arc
in S as an embedding of the interval into S.
One of the fundamental theorems about simple closed curves in surfaces,
which we will not prove but which we state here from a topologist's point
-
29
30 2. Surfaces

of view, is the following:

Theorem 2.1.3 (Jordan Curve Theorem) . If c is a simple closed curve in
IR2 , then !R2 \c consists of two components: one open annulus and one open
disk. If c is a simple closed curve in §2 , then §2 \c consists of two open disks.
All I-manifolds are orientable. (See Exercise 2 below.) In the case of a
simple closed curve, we can think of an orientation as a direction in which
the circle is traversed. In what follows we will be interested in assigning
orientations to points of intersection of I-dimensional submanifolds in a
surface.

Figure 2 . 1 . Two points of intersection with the same orientation.

Figure 2.2. Two oppositely oriented points of intersection.

Definition 2 . 1 .4. Let a, b be simple closed curves in a surface S. We say

that a, b intersect transversely and write a rh b if \:Ix E a n b, there is a chart
(U-p <P'Y) with <P'Y(U'Y) IR2 such that <P'Y maps x to (0, 0) , a to the x-axis,
=

and b to the y-axis.

It is not hard to see that in an oriented surface S, the chart (U'Y, <P'Y)
can be chosen to be in the orientation of S. (Postcompose with a reflection
in IR2 . ) It is harder to see that for any pair (a, b) of simple closed curves in
a surface S, a can be isotoped to a simple closed curve a' so that a' rh b.
2.2. Classification of Surfaces 31

Indeed, this is a consequence of Theorem A.0.19. For the more general

definition of transversality, see Appendix A.
Definition 2 .1 .5. Let a, b be oriented 1-dimensional submanifolds in the
oriented surface S such that a m b. An orientation for a point x E a n b
is the assignment ±1 obtained as follows: Let (U'Y, </>'Y) be a chart in the
orientation of S with </>'Y(U'Y) JR2 such that </>'Y maps x to (0, 0) , a to the
=

x-axis, and b to the y-axis. If </>'Y l a and </>"f ib are both orientation-preserving
or both orientation-reversing, then the orientation on x is + 1 . Otherwise,
it is -1. See Figures 2.1 and 2.2.
Definition 2 .1 .6 . Let a, b be oriented simple closed curves in the oriented
surface S such that a m b. The sum of the orientations on points in a n b is
called the oriented intersection number of a and b.
A straightforward but technical result that can be found for instance in
[48] is the following:
Theorem 2 .1 .7. Suppose that a, b, a', b' are oriented simple closed curves
in an oriented surface S. Suppose further that a is homo topic to a', b is
homotopic to b', a m b, and a' m b' . Then the oriented intersection number
of a and b is equal to the oriented intersection number of a' and b' .
Definition 2 .1 .8. For homotopy classes [a] , [b] , the oriented intersection
number, [a] [b] , is the oriented intersection number of transverse represen­
·

tatives of the classes.

In particular, for a homotopy class [c] , the oriented intersection number
of [c] with itself is the oriented intersection number of representatives c, c'
of [c] that intersect transversely.
Example 2 .1 .9. In 11'2 , the simple closed curve m §1 x {O} is homotopic
=

to a disjoint simple closed curve m' . Thus [m] [m] is 0.

·

Exercises

Exercise 1 . Show that the only compact connected 1-manifolds are the
circle and the interval.
Exercise 2 . Prove that all 1-manifolds are orientable.
Exercise 3 . Generalize Example 2.1.9 to arbitrary oriented surfaces.
2 . 2 . Classification of Surfaces

In Chapter 1 we discussed several examples of surfaces, among them the

sphere, the torus, the projective plane, the disk, and the annulus. The
32 2. Surfaces

sphere and the torus are orientable, whereas the projective plane is not.
Two more examples of surfaces deserve to be pointed out: (1) The Klein
bottle. The Klein bottle can be obtained by identifying the sides of a square
as in Figure 2.3. We will denote the Klein bottle by OC2 . The Klein bottle
contains a Mobius band (see Figure 2.4) and is hence non-orientable. (2)
The connected sum of g tori is called the genus g surface.
b

a II 11 a

Figure 2.3. The Klein bottle.

"a

Figure 2.4. A Mobius band in the Klein bottle.

Definition 2 .2 .1 . Let S be a connected surface. A simple closed curve c in

S is separating if S\ c has two components. Otherwise it is non-separating.
A simple closed curve c in a surface S is inessential if c is separating and a
component of S\c is a disk or annulus. The curve c is essential if it is not
inessential.
A curve c is non-separating if and only if its complement is connected.
Since connectedness is equivalent to path connectedness for manifolds, it
2.2. Classification of Surfaces 33

I
I

0 0

Figure 2.5. The genus 2 surface as a connected sum of two 2-tori.

follows that c is non-separating if and only if there is some other simple

closed curve that intersects c exactly once.
Lemma 2 .2 .2 . The closed surface S is prime if and only if S contains no
essential separating simple closed curve.

Proof. If S = S1 #82 is a non-trivial connected sum, then it contains a

copy of the curve along which S1 \ (disk) , S2 \(disk) were identified. See the
curve pictured in Figure 2.5. Conversely, if a simple closed curve c in S is
essential and separating, then neither component of S\c is a disk. Capping
off (i.e. , taking the union of) these two components with closed disks creates
S1 /: §2 , S2 /: §2 such that S = S1 #S2 . D

Example 2 .2 .3 . The sphere is a prime surface. This is a consequence of the

Jordan Curve Theorem, which tells us that an essential curve as in Lemma
2.2.2 can't exist.
The Klein bottle is not a prime surface. The curve pictured in Figure 2.6
cuts the Klein bottle into two open Mobius bands. Attaching disks to the
remnants of the curve in these Mobius bands produces two projective planes.
Hence the Klein bottle is the connected sum of two projective planes. The
annulus is not a prime surface. It is the connected sum of two disks. More
generally, a compact surface with boundary is not prime. It is the connected
sum of a closed surface and a finite number of disks.
Triangulations have proved useful in the study of surfaces. Most im­
portantly, they allow us to consider a compact surface as the identification
space of a finite number of triangles; see Lemma 1 .4.28. For many applica­
tions, it is expedient to consider not only triangles but other polygons. In
this context, a polygon is a convex disk in JR 2 whose boundary is partitioned
into edges. Specifically, it is often preferable to first identify the triangles
to a single polygon. The surface is obtained from this polygonal representa­
tion by identifying appropriate pairs of edges. In Chapter 1 we encountered
34 2. Surfaces

- -

a 111 I
a

-
- -

polygonal representations of the torus and the projective plane. Above, we

encountered a polygonal representation of the Klein bottle.
Lemma 2 .2 .4. Every compact surface admits a polygonal representation.
Proof. Recall from Chapter I that every surface S admits a triangulation
(S, K) . The dual graph of K is the graph obtained by associating a vertex
to each 2-simplex of K and an edge to each I-simplex of K such that the
vertices incident to this edge are the vertex or vertices associated with the
2-simplex or 2-simplices in which the I-simplex lies.
A spanning tree of a graph is a tree that contains all vertices of the
graph. Let T be a spanning tree of the dual graph of K. (It exists because
K and hence also its dual graph are finite.) In the exercises, you will use
induction on the number of vertices in T to construct a polygon P (in �2 )
from a finite collection of standard 2-simplices (in �3 ) , corresponding to the
2-simplices of K, such that the result of identifying appropriate edges of P
is the surface S. This provides a polygonal representation of S. 0

Lemma 2 .2 .5. The sphere, the projective plane, and the torus are prime
surfaces.
Proof.
Case 1 . The sphere is prime. See Example 2.2.3.
Case 2 . The projective plane.
We argue by contradiction. If �P2 is not prime, then it contains a
separating essential simple closed curve c guaranteed by Lemma 2.2.2. We
analyze this curve in the polygonal representation of �P2 given in Figure 2.7.
General position, a consequence of Theorem A.0. I9, allows us to assume that
c misses the vertices of the polygonal representation and meets the edges
transversely.
2.2. Classification of Surfaces 35

Figure 2. 7. A simple closed curve in JRP2 .

If a subarc, a, of c in this polygonal representation of JRP 2 cobounds a

disk D together with a subarc, b, of an edge of the boundary of the polygon,
then we can assume that a is an outermost such arc, i.e. , that the disk D
cobounded by a and b has interior disjoint from c. The disk D describes
an isotopy that moves a across D to coincide with b and then off of b just
beyond b. A disk cut off by an outermost arc is shaded in Figure 2. 7. The
effect of the isotopy is pictured in Figure 2.8.
a

Figure 2.8. The same simple closed curve in JRP2 •

Proceeding in this manner, we can remove all subarcs of c cobounding

disks with subarcs of edges of the polygon. After this isotopy, c consists en­
tirely of subarcs that connect distinct edges of our polygonal representation
of JRP2 • If c consists of only one such arc, then c is non-separating (see Fig­
ure 2.10 where the dashed arc represents a simple closed curve intersecting
c once) , so this is impossible.
If c consists of more than one arc, consider the two outermost such arcs
as in Figure 2.9. The way in which the edges of our polygonal representative
of JRP2 are identified forces the endpoints of these two arcs to be identified
pairwise. Thus c consists of exactly these two arcs. In this case c cuts
IRP2 into a Mobius band and a disk and is hence not essential, so this is
also impossible. Hence there is no essential simple closed curve, whence the
projective plane is prime.
36 2. Surfaces

Figure 2.9. A simple closed curve in !RP2 •

'
'

Figure 2 . 10. A non-separating simple closed curve in !RP2 •

Case 3 . The torus is prime. We leave this case as an exercise.

D

Lemma 2 .2 .6 . The sphere, the torus, and the projective plane are the only
closed prime surfaces.

Proof. Given a surface S, we consider a polygonal representation of S with

polygon P. It may happen that P has two adjacent edges as pictured in
Figure 2.11. In this case we modify P by identifying the two edges to obtain
a polygon with fewer edges. We continue to denote this polygon by P.
It may also happen that P has two adjacent edges as pictured in Figure
2. 12. If there are no other edges, then S is a projective plane. If there are
other edges, then the curve represented by the dotted arc in Figure 2.12
must cut off a disk to the other side. (Otherwise S would not be prime.)
Hence S is again R.P2 •
We may assume, therefore, in the following, that pairs of edges of P that
are identified are not adjacent. Let c be a simple closed curve in S that is
realized by an arc in P connecting a pair of edges that are identified. (In
particular, c intersects no other remnants of edges of P in S.) Suppose first
that c is separating. Since S is prime, one of these components, say 82 , must
2.2. Classification of Surfaces 37

\
\ I
I

\ I
I

\
' - - - - - - - - - - - - _ ,

Figure 2 . 1 1 . Redundant edges in a polygonal representation of S.

\ I

\ I

, _ _ _ _ _ _ _ _ _ _ _ _ _ ,

Figure 2.12. A polygonal representation of S containing a projective plane.

be a disk. This disk meets remnants of edges of P. This implies that there
is a polygonal representation of S with fewer edges.
Now suppose that c is non-separating. Let a be a simple closed curve
that intersects c once; see Figure 2. 13. Furthermore, let C(c U a) be a collar
neighborhood (see Appendix A and Figure 2. 13) of c U a.
We leave it as an exercise to show that C(c U a ) is one of three things:
a torus minus a disk, a Klein bottle minus a disk, or a Mobius band minus
a disk. In each case, the boundary curve( s) of C ( c U a) must bound disks in
S. (Otherwise S would be a connected sum.) The case of a Klein bottle can
not occur as the Klein bottle is not prime. Thus S is a torus or a projective
plane.
By choosing a polygonal representation of S with the smallest number
of edges we establish the lemma. D
38 2. Surfaces

I
I
I
I
I
I

, _ _

Figure 2.13. The curve c and arc a.

Theorem 2 .2 .7 (The Classification of Surfaces) . Every closed connected

surface is homeomorphic to either a sphere or a connected sum of tori or a
connected sum of projective planes.
Definition 2 .2.8. For an orientable surface, the number of tori in this
connected sum is called the genus. For a non-orientable surface the number
of projective planes in this connected sum is called the genus.
The proof of Theorem 2.2. 7 duplicates the reasoning in the proof of
Lemma 2.2.6. However, instead of concluding, at certain points, as we could
there, that S is a torus or a projective plane, we split off summands (IR.P2 or
1I'2 ) . The fact that this strategy works, i.e. , that the procedure terminates,
follows from considering the number of vertices in the polygonal representa­
tion of the surface as summands are split off. In Exercise 3 below, you will
prove that the number of vertices decreases as tori and projective planes are
split off.
The fact that an orientable surface is built from torus summands (and
not projective planes) is immediate. The fact that a non-orientable surface
can be built from projective planes follows from Exercise 2 below.
Corollary 2 .2 .9. For S an orientable surface of genus g, x(S) = 2 - 2g .
For S a non-orientable surface of genus g, x(S) 1 - g.
=

The Euler characteristic and orientability of a closed connected surface

are two examples of topological invariants (quantities that are well-defined
2.3. Decompositions of Surfaces 39

for a topological manifold) . Thus Corollary 2.2.9 tells us that Euler char­
acteristic and orientability are a complete set of topological invariants for
surfaces.
Our goal is to understand 3-manifolds. We hope to discover and prove a
theorem analogous to Theorem 2.2. 7 for 3-manifolds. Two of the most basic
problems toward this endeavor are the following:
Problem 2 .2 .10. List all possible 3-manifolds.
Problem 2 .2 .11 . Decide whether or not two given 3-manifolds are home­
omorphic.
In Chapter 3, we will discuss prime factorization of 3-manifolds. In the
context of 3-manifolds, 2-spheres will take the place of the circles used with
surfaces. There are other decompositions of 3-manifolds, involving surfaces
other than 2-spheres that are also of interest. We will say more about this
in Section 3.4.
Exercises

Exercise 1 . Show that the torus is prime.

Exercise 2 . Let (S, K) and T be as in the proof of Lemma 2.2.4. Use
induction on the number of vertices in T to construct a polygon P in R2
from a finite collection of standard 2-simplices in R3 corresponding to the
vertices in K such that the result of identifying appropriate edges of P is
the surface S.
Exercise 3. Show that in the proof of Lemma 2.2.6, C(c U a ) is either a
punctured torus, punctured Klein bottle, or punctured Mobius band.
Exercise 4. Prove that 1!'2 #RP2 RP2 #RP2 #RP2 .
=

Exercise 5. Prove that if P is a polygonal representation of a surface 8 and

8 81 #82 is a non-trivial connected sum, then 81 and 82 admit polygonal
=

representations whose polygons have fewer vertices than P.

2 . 3 . Decompositions of Surfaces

In the preceding section we saw that every surface admits a decomposition

into its prime summands. Note that the prime decomposition of a surface
given by Theorem 2.2. 7 is not unique up to isotopy, though for orientable
surfaces the summands are unique up to homeomorphism. See Figures 2. 14
and 2.15. Specifically, for non-orientable surfaces there are other prime
decompositions than those in Theorem 2.2. 7, namely those involving tori,
e.g. , 11'2 #RP2 RP2 #RP2 #RP2 .
=
40 2. Surfaces

0 0

Figure 2. 14. A prime decomposition of the genus 2 surface.

Figure 2 . 1 5 . Another prime decomposition of the genus 2 surface.

There are other natural decompositions of surfaces. Here we will consider

decompositions into pairs of pants. The thrice-punctured sphere, or pair of
pants, forms an important building block for orientable surfaces. Of interest
here is the fact that a pair of pants contains no essential simple closed curves.
You will establish this property in the exercises.
Definition 2 .3 . 1 . A pants decomposition of a compact surface S is a col­
lection of pairwise disjoint simple closed curves { c1 , . . . , Cn} such that each
component of

is a pair of pants. Two pants decompositions are equivalent if they are

isotopic.

Pants decompositions of surfaces are not unique. See Figures 2.16 and
2.17. In the exercises you will show that every compact orientable surface of
negative Euler characteristic of genus g with b boundary components admits
a pants decomposition and each pants decomposition contains exactly 3g -
3 + b simple closed curves.
2.4. Covering Spaces and Branched Covering Spaces 41

- - -
- ' .... - - .....

Figure 2.16. A pants decomposition of the genus 2 surface.

Figure 2. 17. Another pants decomposition of the genus 2 surface.

Exercises

Exercise 1 . Prove that a pair of pants contains no essential simple closed

curves.
Exercise 2. Show that every compact orientable surface of negative Eu­
ler characteristic of genus g with b boundary components admits a pants
decomposition and that every such pants decomposition contains exactly
3g - 3 + b simple closed curves.
Exercise 3*. Parametrize the isotopy classes of simple closed curves on the
torus by pairs of integers.
Hint: Consider the curve § 1 x (point) , called the meridian, and the curve
(point) x § 1 , called the longitude.
2.4. Covering Spaces and Branched Covering Spaces

When we discussed the torus in Chapter 1 , we introduced it as a quotient

space of JRn . Underlying this description of the torus is a more general
notion:
Definition 2.4. 1 . Let p : E -t B be a continuous map. The open set
U c B is said to be evenly covered if p - 1 (U) is a disjoint union of open sets
{Ua} such that Va Pl ucr : Ua -t U is a homeomorphism.
42 2. Surfaces

Let E be a manifold, B a connected manifold, and p : E ---+ B a contin­

uous map. The triple (E, B, p) is a covering if Vb E B there is an open set
U c B with b E U that is evenly covered.
The map p is called the covering map.
Example 2.4.2. The triple (R, § 1 , p) , where p : R ---t § 1 is given by p(x) =
e 21Tix, is a covering. See Figure 2. 18.

'
1

,,,.. - - ...
-

- - - - - - _ .,.
I

'
\
I
I

Figure 2.18. R covers §1 .

Example 2.4.3. The triple (R2 , 'Il'2 , p) , where p : R2 ---t 11'2 is given by
p(x, y) = (e 21Tix , e 21TiY) , is a covering (cf. Section 1.1).
Example 2.4.4. The triple (§n , RPn , p) , where p : §n ---t �pn is defined by
p(x) = [x] = [-x] (the equivalence class containing x and -x) , is a covering
(cf. Section 1 . 1 ) .
Note that if p : E ---+ B is a covering map, then p is a local homeomor­
phism, but not conversely. For example, consider p : JR + ---t § 1 given by
p (x) = e 21Tix . Then p is a local homeomorphism but not a covering map.
Definition 2.4.5. Let (E, B , p) be a covering space. A homeomorphism
t : E ---+ E is a covering transformation if p o t = p.
Example 2.4.6. For n E Z the map t : R ---+ R given by t ( x) = x + n is a
covering transformation for the covering in Example 2.4.2.
2.4. Covering Spaces and Branched Covering Spaces 43

Example 2.4.7. For n, m E Z the map t : IR2 ---+ IR2 given by t(x, y) =

(x + n, y + m ) is a covering transformation for the covering in Example

2.4.3.
Example 2.4.8. The antipodal map of sn is a covering transformation for
the covering in Example 2.4.4.
You will show in the exercises that the covering transformations for
a given covering space form a group. It is called the group of covering
trans!ormations.
Example 2.4.9. The group of covering transformations for the covering
space in Example 2.4.4 is Z /2. Indeed, only the identity map and the an­
tipodal map are covering maps. To see this, note that if t is a covering
transformation, then 'r:/x E §n , there are only two options: either t (x ) x =

or t ( x ) -x. Since t is continuous, these two options extend continuously

=

over §n to either the identity map or the antipodal map.

You will show in the exercises that the group of covering transforma­
tions for the covering spaces in Examples 2.4.2 and 2.4.3 are Z and Z2 ,
respectively.
Lemma 2.4.10. If (E, B, p) is a covering, B is a compact connected sim­
plicial complex, and for b E B, p - 1 (b) is finite, then
x (E) # {p - 1 (b) } x (B) .
= ·

The proof of this lemma requires showing that #{p - 1 (b) } is constant
and then counting simplices.
There is a more general notion:
Definition 2.4. 1 1 . Let E, B be manifolds, E' a submanifold of E, B'
a submanifold of B, and p : E ---+ B a continuous map. The quintet
( E, E', B, B', p) is a branched covering if
•PI E\ E' : E\E' ---+ B\B' is a covering map;
•PIE' : E' ---+ B' is a covering map.
Here B' is called the branch locus and E' is called the ramification locus.
Example 2.4. 12. The quintet (C, 0, C, 0, f) , where f(z) zn , is a branched
=

covering.
Example 2.4. 13. The quintet (11'2 , {a, b, c, d} , §2 , {a', b', d , d'} , p) can be
made into a branched covering by appropriately defining p. See Figure
2.19. In Figure 2.19 an axis intersects the torus in four points, a, b, c, d. An
order 2 rotation of the torus around the axis identifies the upper half of the
torus with the lower half. In this identification, points are identified pairwise
44 2. Surfaces

except for the four points of intersection with the axis. The quotient is a
sphere. We denote the quotient map by p and denote p( a) a' , p(b) b', = =

p(c) d , p(d) d'. Then

= =

PI T\{a,b,c ,d} : 1'\{a, b, c, d} --+ §2 \{a' , b' , c' , d' }

is a 2-fold cover and
Pl {a,b,c,d} : {a, b, c, d} -t {a' , b' , c' , d' }
is a homeomorphism.

E>

Figure 2.19. 'f 2 as a branched cover of §2 .

Lemma 2.4. 14 (Riemann-Hurwitz) . If (E, E', B, B', p) is a branched cov­

ering, B is a compact connected surface, E', B' are finite sets, and for
b E B\B', p - 1 (b) is finite, then
x (E) # {p - 1 (b) } . ( x (B) - x (B' ) ) + #{E' } .
=

Exercises

Exercise 1 . Show that the covering transformations for a given covering

space form a group.
Exercise 2. Show that the group of covering transformations for the cov­
ering space in Example 2.4.2 is Z.
Exercise 3. Show that the group of covering transformations for the cov­
ering space in Example 2.4.3 is Z2 .
2. 5 . Homotopy and Isotopy on Surfaces 45

Exercise 4. Show that there is a covering map p : 11'2 -+ JK2 and that
the group of covering transformations of the corresponding covering space
is Z/2.
Exercise 5. Prove the Riemann-Hurwitz formula.

2 . 5 . Homotopy and Isotopy on Surfaces

Isotopies are always homotopies. The converse is not true in general. In

this section we will see that on surfaces, homotopy classes of simple closed
curves correspond to isotopy classes of simple closed curves.
Definition 2.5. 1 . Let F be a surface and let C { c1 , . . . , en } be a collection
=

of simple closed curves in F that have been isotoped to intersect in a minimal

number of points. We say that C is filling if F\ (c1 U U en ) is a union of
· · ·

disks.
Example 2.5.2. The pair of curves on 11'2 pictured in Figure 2.20 is filling.

Figure 2.20. A pair of tilling curves in T2 •

Lemma 2.5.3 (Alexander Trick) . Suppose that f : ID>n -+ ID>n is a homeo­

morphism such that fl8JDn is the identity. Then f is isotopic to the identity.
Proof. The isotopy H : ID>n x I -+ ID>n is given by
H(x ' t) =
x
{
tf ( x / t) if 0 ::; llxl l < t,
if t ::; ll x ll ::; 1.
See Figure 2.21. D

Theorem 2.5.4. Suppose that F is a closed orientable surface and h : F -+

F is a homeomorphism. If h is homotopic to the identity, then h is isotopic
to the identity.
The key ingredient in this proof is that homotopic simple closed curves
in F are isotopic. We provide a sketch of the argument in a special case.
46 2. Surfaces

Figure 2.21. Schematic for the Alexander 'Irick.

Proof in the case F = 1I'2 • Let h : F ---+ F be a homeomorphism that is

homotopic to the identity. Let c be a simple closed curve in 1I'2 and compare
c and h(c) . We may assume that c rh h(c) . See Figure 2.22.

Figure 2 . 2 2 . Curves c and h ( c) .

If there is a disk that is cobounded by an arc in c and an arc in h(c) ,

then we can use an innermost such disk to perform an isotopy that reduces
the number of points of intersection between c and h( c) . We continue to
perform isotopies reducing the number of points of intersection until there
are no disks cobounded by an arc in c and an arc in h(c) . See Figure 2.23.

Figure 2.23. An innermost disk D.

2. 6. The Mapping Class Group 47

Now every component of h(c)\c is either entirely contained in 1r2 \c or

crosses the open annulus 1r2 \c. In particular, the orientations of the points
in c n h(c) are the same. Recall that [c] · [c] = 0. By Theorem 2 . 1 . 7,
[c] · [h(c)J 0. It follows that c n h(c) 0. Thus h(c) is contained in the
= =

annulus 1r2 \c. Therefore, h(c) is isotopic to c, whence hie is isotopic to id le·
Consider the pair of filling curves { m, l} in Example 2.5.2. The above
argument allows us to assume that, after isotopy, h( m) coincides with m.
You will prove in the exercises that there is an isotopy fixing m after which
l and h(l) coincide, i.e. , that hit is isotopic to id It ·
Now consider the polygonal representation of 1r2 with four sides that
identify to m U l. Then h defines a map on this square Q and hlaQ idaQ · =

By the Alexander Trick, the map h lQ is isotopic to the identity. Thus h

is isotopic to the identity. (This completes the proof in the case of the
torus.) 0

Exercises

Exercise 1 . For m, l, h as in the proof of Theorem 2.5 .4, prove that there
is an isotopy fixing m after which l and h(l) coincide.
Exercise 2. Find a filling pair of curves for the genus 2 surface.
Exercise 3. Show that on any surface, homotopic simple closed curves are
isotopic.

2 . 6 . The Mapping Class Group

Self-diffeomorphisms of surfaces are interesting in their own right but are

of particular interest here because of their role in certain constructions on
3-manifolds (e.g. , mapping tori (see Chapter 3) and the curve complex (see
Chapter 7) ) . More specifically, it is not self-diffeomorphisms of a surface
but isotopy classes of self-diffeomorphisms of a surface that play a role. For
this reason we are interested in the mapping class group defined below.
In the exercises, you will show that the set of all self-diffeomorphisms
of a topological space forms a group. In the case of an oriented surface, we
denote this group by Diffeo( S) . You will also show that the set of orientation­
preserving self-diffeomorphisms and the set of self-diffeomorphisms that are
homotopic to the identity form normal subgroups. In the case of an oriented
surface, we denote these groups by Diffeo+ (S) and Diffeoo (S) , respectively.
We will define the mapping class group in terms of diffeomorphisms.
Homeomorphisms of surfaces are isotopic to diffeomorphisms; hence our
definitions could equally well be made in terms of homeomorphisms. In
48 2. Surfaces

particular, our definition of Dehn twist given below yields homeomorphisms

rather than diffeomorphisms. However, these homeomorphisms can easily
be smoothed.
Definition 2.6. 1 . Let S be a compact orientable surface. The mapping
class group of S, denoted by MCQ(S) is Diffeo+ (S)/ Diffeoo (S) , the group of
orientation-preserving self-diffeomorphisms of S modulo the subgroup con­
sisting of self-diffeomorphisms of S homotopic to the identity.
Lemma 2.6.2. The mapping class group of the torus is isomorphic to
SL(2, Z) , the group of 2 x 2 matrices with integer coefficients and deter­
minant 1 .
Proof. Given a simple closed curve 'Y on 1r2 , we define an integer vector
i ('Y) = ( ['Y] [m] , ['Y] [b] ) .
· ·

As we saw in the proof of Theorem 2.5.4, the vector i ('Y) completely de­
termines the isotopy class of 'Y on 'lr2 . Then, given a homeomorphism
f E Homeo('lr2 ) , we define the matrix At , whose columns are the vectors
i ( J (l) ) and i ( J (m) ) .
The proof of Theorem 2.5.4 shows that the matrix At completely de­
termines the isotopy class of f. In the exercises you will verify that the
map
1J : f i---+ At
is a homomorphism from Homeo('lr2 ) to the group of integer matrices. In
particular, the image of 7J is contained in GL(2, Z) .
Since for homotopic curves a, a' we have
i (a) = i ( a' ) ,
it follows that for g E Homeo0(1r2 ) ,
A9 = I ,
the identity matrix in GL(2, Z) . Therefore, the homomorphism 1J descends
to a homomorphism
H : MCQ('lr2 ) -+ GL(2, Z) .
Below, we will verify that H is an isomorphism onto SL(2, Z) . As a first
step, given a matrix A E SL(2, Z) ,

we construct a homeomorphism f fA E Homeo('lr2 ) such that At = A

=

as follows: Recall the covering map p : JR2 -+ 1r2 from Example 2.4.3 and
consider the matrix A as a linear transformation of JR 2 . Then A maps
2. 6. The Mapping Class Group 49

horizontal lines p - 1 (m) to lines of slope � and these project to simple closed
curves a on 'Il'2 so that
i (a) = [�].
Likewise, A maps vertical lines p - 1 (l) to lines of slope � and these project
to simple closed curves /3 on S so that
i (/3) = [�].
We define a map f = fA : 'Il'2 --+ 'Il'2 as follows: Let x E 'Il'2 and let x be a
point in JR2 such that p(x ) = x . Set f(x) = p o A(x) . You will verify in the
exercises that the map f : 'Il'2 --+ 'Il'2 is well-defined and is a homeomorphism.
By construction,
i (f(l)) = [�], i(f(m) ) = [�].
Therefore, At = A. Note that, in the charts of 'Il'2 implicitly defined by the
covering map p, the Jacobian matrix of f is the matrix A. Thus, if A E
SL(2, Z) , then f = fA is orientation-preserving. Therefore, it immediately
follows that
SL(2, Z) c im(H) .
Furthermore, if A = Ag, g E Homeo('Il'2 ) , then for f = fA , we obtain
A = At = Ag
and, hence, the maps fA and g are isotopic to each other. Therefore, the
map

satisfies
J o H = id .
Hence, the homomorphism H is injective. It remains to show that im(H) c
SL(2, Z) . Note that each diffeomorphism f = fA for which det(A) = - 1
is orientation-reversing. Thus, if g E Homeo('Il'2 ) is such that det(Ag) =
- 1 , then g is isotopic to the orientation-reversing diffeomorphism f = fA ·
Since an orientation-preserving diffeomorphism cannot be isotopic to an
orientation-reversing one, it follows that for each g E Homeo+ ('Il'2 ) ,
Ag E SL(2, Z) .
This concludes the proof that H is an isomorphism MCQ('Il'2 ) --+ SL(2, Z) .
D
50 2. Surfaces

Remark 2.6.3. The reader familiar with homology will recognize that the
map i defines coordinates on the homology group H1 (11'2 , Z) , which is isomor­
phic to Z 2 . Furthermore, A I is the marix of the automorphism of H1 (11'2 , Z)
determined by f.
A simple closed curve c in a surface S has a regular neighborhood (for the
technical definition of regular neighborhood, see Appendix A) . When c is
orientation-preserving, this means that there is an embedding of an annulus
f : § 1 x [O, 1] ---+ S such that c f(§ 1 x { 1/2}) whose image is the regular
=

neighborhood. The regular neighborhood can be taken to be open or closed.

If it is open, we write ry(c) (for f(§ 1 x (0, 1))) ; if it is closed, we write N(c)
(for f(§ 1 x [O, 1]) ) . If c is orientation-reversing, an analogous statement holds
with the annulus replaced by a Mobius band.
Definition 2.6.4. Let c be an orientation-preserving simple closed curve
in a compact surface S and let N(c) be a regular neighborhood of c that is
oriented via the parametrization i § 1 x [O, 1] ---+ ry(c) . A map f : S ---+ S is
:

called a left Dehn twist around c if

• f ls\N( c) is the identity map; and
• f lN( c) is the map of the annulus given by
f( e 2i1r 9 , t) ( e 2i7r ( 9 +t ) , t ) V ( e 2i1r 9 , t) E § 1 x [O, 1] .
=

See Figure 2.24. Likewise, a map f : S ---+ S is called a right Dehn twist
around c if
• f ls\ N(c) is the identity map; and
• f I N( c) is the map of the annulus given by
f( e 2i7r9 , t) ( e 2i7r ( 9 - t) ' t) V ( e 2i1r9 , t) E § 1 x [O, l] .
=

I
I
'
'
'

f(a) '
'

Figure 2.24. Local effect of a left Dehn twist.

We can think of a (left/right) Dehn twist as doing the following: cutting

the surface along the core of the annulus, performing a full twist (to the
left/to the right) , and regluing. Theorem 2.6.5 shows that Dehn twists are
the basic building blocks for all surface diffeomorphisms.
Theorem 2.6.5 (Dehn, 1938; Lickorish, 1962) . Every surface diffeomor­
phism is isotopic to a composition of Dehn twists. In other words, the map­
ping class group is generated by Dehn twists.
2. 6. The Mapping Class Group 51

Theorem 2.6.5 was proved by M. Dehn and discussed in his Breslau

Lectures in 1922. It was published in 1938; see [33] . The theorem was
rediscovered (via a rather clever argument that we outline below) by W. B.
R. Lickorish in 1962; see [88] . We will discuss some of the insights that go
into the proof of this theorem.
Lemma 2.6.6. If the simple closed curves a, /3 in the compact surface S
intersect exactly once, then there is a pair Ji , f2 of Dehn twists such that
h · fi(a) is isotopic to /3 .
Proof. If we choose Ji to be a left Dehn twist along /3 and f2 to be a right
Dehn twist along a, then indeed f2 · fi (a) /3. See Figures 2.25, 2.26, and
=

2.27. D

Figure 2.25. a and /3 .

Figure 2.26. fi (a) .

Lemma 2.6. 7. If a, /3 are oriented simple closed curves in the orientable

surface S, then there is a series of Dehn twists Ji , /2, . . . , fk such that fk · · . . ·
f2 · Ji (a) is either disjoint from /3 or intersects /3 in exactly two oppositely
oriented points.
Sketch of proof. The proof is by induction on the number of points in
the intersection of the two curves. Suppose that there are two points of
intersection with the same orientation that are adjacent along /3. See Figure
2.28. Then denote the subarc of a that connects these two points by c. By
52 2. Surfaces

Figure 2.27. h · fi (cx) .

taking the union of the subarc (or one such subarc if there are two) of f3
along which these two points are adjacent (either will do if there are two)
with c, we construct a simple closed curve. After a small isotopy, this curve
intersects a in exactly one point. Denote this curve by c. In Exercise 3 you
will show that for Ji a right Dehn twist along c, Ji (a) has fewer points of
intersection with f3 than a.
If there are no points of intersection with the same orientation that are
adjacent along f3 and if there are at least three points of intersection, then
a similar construction allows a reduction of the number of points in a n /3.
You will show this in Exercise 4. D

Figure 2.28. Adjacent points of intersection of ex and /3.

To prove Theorem 2.6.5, let S be a closed orientable surface. There is

a standard collection of curves ai , bi , . . . , a9, b9 that cuts S into a disk; see
Figure 2.29. A self-diffeomorphism of S takes this fixed collection of curves
to another such collection of curves. Using Lemmas 2.6.6 and 2.6. 7 allows
us to perform a series of Dehn twists that reverses the effect of the surface
diffeomorphism on the specified collection of curves. The collection of Dehn
twists can be seen as "undoing" the effect of the surface diffeomorphism on
2. 6. The Mapping Class Group 53

the specified collection of curves. The map on the complementary disk is

isotopic to the identity by the Alexander Trick.

/ I ht

Figure 2.29. Standard collection of curves ai , bi , a2 , b2 in a genus 2 surface.

Definition 2.6.8. A diffeomorphism h : S --+ S is called periodic if hn = ids

for some n E N . It is called reducible if 3 an essential simple closed curve
C c S such that </>(C) = C (setwise) .
Example 2.6.9. Let h 1 : 'lr2 --+ 'lr2 be defined by h 1 ( ei9 , eief> ) = ( e -ief> , ei9 ) .
Then h 1 is periodic; in fact, hf = id.
Let h2 : 'lr2 --+ 'lr2 be defined by h2 ( ei9 , eief> ) = ( ei9 , ei ( 9 + ef>) ) . Then h2 is
reducible, and it preserves the essential simple closed curve { (1, eief> ) I 0 :S
</> < 211"}. More generally, for any surface S a Dehn twist along an essential
simple closed curve is reducible.
Let h3 : 'lr2 --+ 'lr2 be defined by h3 ( ei6 , eief> ) = ( ei (WH) , ei (e + ef>) ) . Then h3
is neither periodic nor reducible.

More generally,
Definition 2.6. 10. An element of the mapping class group of S is periodic
if it has a periodic representative. An element of the mapping class group
of S is reducible if it has a reducible representative.

It is a deep theorem due to W. Thurston that elements of the map­

ping class group are classified as being either periodic, reducible, or pseudo­
Anosov. We discuss what it means to be pseudo-Anosov in Chapter 7, when
we discuss laminations.
54 2. Surfaces

Exercises
Exercise 1 . Let X be a topological space. Show that the set of self­
diffeomorphisms of X forms a group.
Exercise 2. Let S be an oriented surface. Show that Diffeo+ (S) <J Diffeo(S)
and Diffeoo (S) <J Diffeo(S) .
Exercise 3. Show that the map fA : ']['2 -+ ']['2 constructed in the proof
of Lemma 2.6.2 is well-defined and is a self-homeomorphism of the torus.
(Hint: First, show that for every covering transformation r E Z2 of the
covering p, there exists another covering transformation r' so that
A o r = r1 o A.
This will imply that the map fA is well-defined. Then, for B = A - 1 E
SL(2, Z) verify that fA = fs 1 . )
Exercise 4. Let a, /3 be as in Lemma 2.6.7, except that now there is one
point of a n /3 with the opposite orientation between the two points with the
same orientation. Prove that a similar construction still reduces the number
of points in a n /3. See Figure 2.30.

C-----

Figure 2.30. Adjacent points of intersection of a and /3.

Exercise 5. Complete the proof of Theorem 2.6.5 in the case of the torus.
(Hint: Let m, l be the curves considered in the proof of Lemma 2.6.2. Let
h : ']['2 -+ ']['2 be a homeomorphism. Construct a homeomorphism h, realized
by a product of Dehn twists, such that h, - 1 o h(m) = m. Conclude that
h, - 1 o h corresponds to the matrix
A= [� �]
and is hence a Dehn twist around m.)
 Chapter 3

3- Manifolds

One of our goals in this chapter is to discuss prime factorization of 3-

manifolds. This prime factorization is almost unique, but the 2-sphere
submanifolds of a 3-manifold M along which M decomposes can be po­
sitioned in more than one way. We will also discuss other decompositions of
3-manifolds. For more on these subjects, see [63] , [67] , [71] , [73] , and [72] .
In this chapter and beyond, we will assume familiarity with the con­
cept of general position. For details, see Appendix A. One of the insights
gleaned from general position is that compact 2-dimensional submanifolds
of a 3-manifold M can always be isotoped slightly to intersect in a com­
pact !-dimensional submanifold. See Figure 3. 1 . A compact !-dimensional
submanifold and a compact 2-dimensional submanifold of M can always be
isotoped slightly to intersect in a finite number of points. See Figure 3.2.
Compact !-dimensional submanifolds of M can be isotoped to be disjoint. A
point, i.e. , a 0-dimensional submanifold of M, can be assumed to be disjoint
from 2-, 1-, or other 0-dimensional submanifolds of M.

, _ _ _ _ _ _ _ _ _ _ _ _ _

Figure 3 . 1 . Local picture of general position for two 2-submanifolds of

a 3-manifold.

55
56 3. 3-Manifolds

Figure 3.2. Local picture of general position for a 1-submanifold and

a 2-submanifold of a 3-manifold.

More generally, let M be an n-manifold containing a k-dimensional sub­

manifold K and an l-dimensional submanifold L that are in general position.
Then K n L is a ( possibly empty ) submanifold of M of dimension ( k + l - n) .
Furthermore, general position can always be guaranteed and maintained by
small isotopies.
Convention 3.0. 1 . To simplify our statements, we will, in this section and
beyond, make the assumption that, for k :::; n, a manifold of dimension k
in a manifold of dimension n is a submanifold. In particular, simple closed
curves and simple arcs lying in a surface or 3-manifold and surfaces lying in
a 3-manifold are submanifolds.
An interesting 2-sphere contained in JR3 that is not a submanifold in the
sense defined in Chapter 1 was exhibited by J. W. Alexander. It is called the
Alexander horned sphere. ( For an artistic rendition, see for instance [127] .)
An interesting fact concerning the Alexander horned sphere is that it does
not bound a 3-ball. Convention 3.0. 1 now eliminates this type of example
from our discussion.
3 . 1 . Bundles
Recall two of the examples of surfaces with boundary discussed in Chapter
2: the annulus and the Mobius band. See Figures 3.3, 3.4, 3.5, and 3.6.

· { - ---------- -------------- 1 ·

Figure 3.3. The annulus.

3. 1 . Bundles 57

'
'

Figure 3.4. The annulus.

·f-------------------------f.
Figure 3.5. The Mobius band.

Figure 3.6. The Mobius band.

Definition 3.1.1. A bundle is a quartet (M, F, B, n') where M, F, and B

are manifolds and 7r : M -+ B is a continuous map such that the following
hold:
• for every b E B, tr - 1 (b) is homeomorphic to F;
• there is an atlas {Ua} for B such that 'Va 3 a homeomorphism
ha : 7r - 1 (Ua) -+ Ua X F;
• the following diagram commutes (the right arrow is projection onto
the first factor) :

id
----+ Ua .
58 3. 3-Manifolds

Here M is called the total space, F is called the fiber, B is called the
base space, and 7r is called the projection. A pair (U00 ha) is called a bundle
chart. The family { (Ua , ha) } is called a bundle atlas.
We sometimes refer to (M, F, B, 7r ) simply as an F-bundle over B.
In the context of 3-manifolds, we sometimes use the word fibering instead
of bundle.
Example 3.1.2. Any product manifold X x Y is a bundle. The total
space is X x Y, the fiber is Y (or X, respectively) , the base space is X (or
Y, respectively) , and the projection is projection onto the first (or second,
respectively) factor.
The annulus is an example of this type of bundle.
Example 3.1 .3. The Mobius band is a bundle with fiber I and base space
§ 1 . The projection can be described by considering the rectangle in Figure
3.5. The projection consists in vertical projection onto the core curve, that
is, the marked curve. We leave it as an exercise to show that the map thus
obtained is continuous, as required.
Example 3.1.4. Let S be a closed connected n-manifold and f : S --+ S a
homeomorphism. The mapping torus of f is the (n + 1)-manifold obtained
from S x [- 1, 1] by identifying the points (x, -1) and (f(x), 1), Vx E S. The
mapping torus of f is a bundle with fiber S and base space § 1 . The total
space is often denoted by ( S x I)/ ,....,f .
The role of the homeomorphism f in the above example is an interest­
ing one. One crucial fact is that the isotopy class of this homeomorphism
determines the (n + 1)-manifold.
The Mobius band is an example of a mapping torus, with S [- 1 , 1] =

and f : [- 1, 1] --+ [- 1, 1] given by f(x) -x.

=

Definition 3.1.5. Suppose that (M, F, B, 7r ) and (M', F, B', 11"1 ) are bundles.
An isomorphism between the bundles is a pair of homeomorphisms
h: M -+ M'
and
f : B -+ B'
such that the following diagram commutes:
M � M'

B � B'.
3. 1 . Bundles 59

We say that two bundles are isomorphic or equivalent if there is an

isomorphism between them.
The annulus and the Mobius band are inequivalent bundles. To see this,
note, for instance, that the total space of the annulus has two boundary
components whereas the total space of the Mobius band has only one. Thus
there can be no bundle homeomorphism between the two.
There is a more intuitive approach to understanding the difference be­
tween the annulus and the Mobius band: Cut the total spaces along their
core curves. The core curve of the annulus is marked in Figure 3.3; that of
the Mobius band is marked in Figure 3.5. If we cut along the core curve
of the annulus, we obtain two annuli. If we cut along the core curve of the
Mobius band, we obtain only one annulus. Both the annulus and the Mobius
band are I-bundles over the circle, in fact, they are the only two I-bundles
over the circle.
Example 3.1 .6. The annulus and the Mobius band are inequivalent bun­
dles.
Definition 3.1 .7. A bundle that is isomorphic to a product bundle is called
a trivial bundle. A bundle that is not a trivial bundle is called a non-trivial
bundle.
Example 3.1 .8. The annulus is a trivial bundle. The Mobius band is a
non-trivial bundle.
Definition 3.1 .9. Given a bundle E = (M, F, B, 7r) and a submanifold B'
of B, the restriction of E to B' is a bundle with total space M' = 7r -1 (B') ,
fiber F, base space B', and projection 7r l M' · We denote this restriction by
E IB' ·
Example 3 . 1 . 10. We construct a non-trivial I-bundle over the Mobius
band as follows: Consider the rectangle in Figure 3.5. Points in the rectangle
lie in [- 1, 1] x [- 1, l] . To form the Mobius band, we identify (-1, x ) with
(1, -x ) . Now consider [- 1, 1] x [- 1, 1] x [- 1, 1] and identify (-1, x, y) with
(1, - x , -y) . The result is a non-trivial bundle over the Mobius band. We
leave it as an exercise to show that the total space of this bundle is a solid
torus. See Figure 3. 7.
The above example can be visualized as follows: First we visualize the
Mobius band. We start with a square and identify the left and right sides
so that we obtain a non-trivial bundle over the core circle. We do so by
stretching and bending the square so that the left and right sides of the
square come up out of the page. We then twist the right side by 180 degrees
and attach the two sides.
60 3. 3-Manifolds

Figure 3.7. The solid torus containing the Mobius band.

Now we visualize the non-trivial bundle over the Mobius band. We start
with the trivial I-bundle over the square as in Figure 3.5. This is a cube.
We identify the left and right sides of the cube so that we obtain a non­
trivial bundle. We do so by stretching and bending the cube so that the two
squares come up out of the page. We must identify these squares so that (if
we put our head between the two and look to one side and then the other)
the left side of one is identified to the left side of the other (and the right
side of one to the right side of the other) and the top of one is identified to
the bottom of the other.
The trivial bundle B 2 x § 1 and the bundle in Example 3. 1 . 10 are in­
equivalent bundles. Their total spaces are homeomorphic, but their bundle
structures differ.
Definition 3. 1 . 1 1 . A section of a bundle is a continuous map u : B --+ E
such that 7r o u = ids (where ids is the identity map on B) .
For a comprehensive treatment of bundles, see [65] or [150] .
Definition 3 . 1 . 12. Let M be an n-manifold. Let S be a submanifold of
M of dimension m. A regular neighborhood of S is a submanifold N ( S)
of M of dimension n that is the total space of a bundle over S with fiber
nn- m . A regular neighborhood of a I-manifold in a 3-manifold is also called
a tubular neighborhood. A regular neighborhood of an ( n - 1 )-dimensional
submanifold of an n-manifold that is a trivial bundle is also called a collar
(or, if n = 2, a bicollar) . An open regular neighborhood of S is a submanifold
'r/(S) of M of dimension n that is the total space of a bundle over S with
fiber i nteri or ( Bn m ) .
-

Figures 3.8 and 3.9 exhibit the local picture of regular neighborhoods of
simple closed curves, simple arcs, and surfaces in a 3-manifold. For more
3. 1 . Bundles 61

() ()
Figure 3.8. Local picture of a I-manifold and its regular neighborhood
in a 3-manifold.

Figure 3.9. Local picture of a surface and its regular neighborhood in

a 3-manifold.

details see Appendix A. In the exercises you will show that for simple closed
curves there are only two possible total spaces for regular neighborhoods.
In particular, if c is an orientation-preserving simple closed curve in an
orientable 3-manifold M, then there is a solid torus, f : § 1 x ID>2 --+ M, such
=
that c /(§ 1 x {(O, O) } ), whose image is the regular neighborhood of c in M.
When c is an orientation-reversing simple closed curve in a non-orientable
3-manifold M, then there is an embedding of a twisted ID>2 -bundle over § 1 ,
=
f : § 1 xID>2 --+ M, such that c f (§ 1 x { (O, O) }), whose image is the regular
neighborhood of c in M.
The following theorem summarizes results discussed in Appendix A.
Theorem 3.1.13. Let M be an n-manifold and let S be a k-manifold in
M. There exists a regular neighborhood N ( S) for S in M. Furthermore,
any two regular neighborhoods of S in M are isotopic.

A related, but more specific, theorem is the following:

Theorem 3.1.14. If S is a surface in the 3-manifold M and both M and
S are orientable, then N(S) is a trivial I-bundle.
Definition 3.1.15. Let S be a surface in the 3-manifold M. We say that S
is a 2-sided surface if a regular neighborhood of S in M is a trivial I-bundle.
We say that S is a I -sided surface if a regular neighborhood of S in M is a
non-trivial, or twisted, I-bundle.
62 3. 3-Manifolds

Exercises

Exercise 1 . Prove that the projection mentioned in Example 3.1.3 is con­

tinuous (as required) .
Exercise 2 . Prove that the total space of the bundle described in Example
3. 1 . 10 is the solid torus.
Exercise 3. Use the core curves of the annulus and of the Mobius band to
describe sections of the bundles described in Examples 3.1.2 and 3.1 .3.
Exercise 4. Prove that there are only two bundles with base space § 1 and
fiber I.
Exercise 5. Prove that there are only two bundles with base space § 1 and
fiber lBln .

3 . 2 . The Schonflies Theorem

The Schonfiies Theorem generalizes the Jordan Curve Theorem. One of the
key notions in this context is Definition 3.2. 1:
Definition 3.2. 1. A 3-manifold M is irreducible if every 2-sphere in M
bounds a 3-ball. A 3-manifold is reducible if it contains a 2-sphere that does
not bound a 3-ball.
Example 3.2.2. The 3-manifold §2 x § 1 is reducible.

Before stating and proving the Schonflies Theorem, we prove a lemma.

We will use this lemma in the proof of the Schonflies Theorem. It will also
serve as a warm-up for our discussion of 3-manifolds.
Lemma 3.2.3. Let S1 , S2 be surfaces in a 3-manifold M. Suppose that
S1 nS2 contains a simple closed curve c such that, for i = 1, 2, one component
of Si \c is a disk Di such that D2 is disjoint from S1 . If c U Di U D2 bounds
a 3-ball in M, then
• (S1 \D 1 ) U D2 is isotopic to S1 ;
• there is an isotopy of S1 that eliminates the curve c from S1 n 82
and introduces no new components of intersection.

Note that we are considering S1 before and after the isotopy. It would
make sense to write Sf to distinguish the positioning of S1 at various times
during the isotopy, but this is usually unnecessary, as context makes it clear
which positioning is meant. So we simply write S1 .
3.2. The Schonflies Theorem 63

Figure 3.10. Surfaces intersecting.

Proof. A neighborhood of the 3-ball in M bounded by D i U D2 is homeo­

morphic to a neighborhood of the upper hemisphere of the standard 3-ball
in JR3 . See Figure 3. 10, where D i is the disk above the xy-plane.
To obtain the first conclusion, note that straight line projection of D i c
8i onto the xy-plane is an isotopy that replaces 8i with (8i \D i ) U D2 .
To obtain the second conclusion, note that straight line projection of the
portion of 8i lying above y = -E (this includes D i ) onto the plane y = -E is
an isotopy of 8i that eliminates the component c from 8i n 82 . See Figure
3. 1 1 . D

�
/ __ �/

Figure 3 . 1 1 . After the projection and isotopy.

Definition 3.2.4. Let X be an n-manifold and x a subset of X (typically

a submanifold of dimension n - 1 ) . To cut X along x means to consider
X\ 'fJ( x) . Likewise, x cuts X into a given set of manifolds means that X\ 1J ( x)
consists of this set of manifolds.
Theorem 3.2.5 (Schonfiies Theorem) . Any 2-sphere in JR3 bounds a 3-ball.
A beautiful elementary proof of this theorem was given by Morton Brown
in 1960; see [18] . Students are encouraged to read this proof in the original.
The following proof is based on one that appeared in a lecture series given
by Andrew Casson in China in 2002. We include it here because it illustrates
64 3. 3-Manifolds

some of the techniques employed in the contemporary study of 3-manifolds.

It relies on the the Poincare-Hopf Index Theorem mentioned in Chapter 1.
If you don't understand the proof below in its entirety on a first reading,
focus on the intuition behind it and the point of view espoused. You will
see the concepts and techniques used to prove this theorem over and over
again. We are working in the DIFF category.
We will use the concept of a "height function" . A height function is a
Morse function with at most two critical points, namely a maximum and
a minimum. This is a strong requirement. In fact, there are only three
connected 3-manifolds that admit height functions: §3 (requires a maximum
and a minimum, by compactness), JB3 (requires a maximum and a minimum,
by compactness) , and R3 (requires no critical points) . For more information
on Morse functions, see Appendix B.
We will also use induction on an ordered pair of numbers. Recall that
induction can be performed on any countable ordered set. When we perform
induction on an ordered pair of numbers, both entries must come from a
countable ordered set. The product of the countable sets is countable and
is ordered via the dictionary order.
Proof. Let S c R3 be a 2-sphere submanifold. We isotope S so that the
height function h : R3 ---+ R given by projection onto the third coordinate
restricts to a Morse function h i s ·
By the Poincare-Hopf Index Theorem,
#maxima + #minima - #saddles = x(S) = 2.
Note that since S is compact, h i s has at least one maximum and one
minimum. Let n be the number of saddle points. If n = 0, then S has one
maximum, y2 , and one minimum, YI · We may assume, by multiplying h
with an appropriate constant, that y2 - YI = 2. We may further assume, by
translating h upwards or downwards, that Y2 = 1 and hence Y I = -1. Each
r such that - 1 < r < 1 corresponds to a level curve in S that is a simple
closed curve in the plane z = r. It follows from the Jordan Curve Theorem
that each such curve can be isotoped within the plane z = r into the circle
such that x 2 + y2 = 1 - r 2 . It is not too hard to see (though harder to prove
rigorously) that these isotopies can be chosen so as to vary smoothly with
r. Thus S is isotopic to the standard 2-sphere in R3 and hence bounds a
3-ball.
Note that if n = 1, then there are two minima (or one minimum, resp.),
one maximum (or two maxima, resp.), and one saddle. There are two pos­
sibilities: a "non-nested" saddle or a "nested" saddle. See Figures 3.12 and
3. 13. In either case, there is a plane z = c (containing the saddle) that meets
3.2. The Schonflies Theorem 65

S in a figure 8. Call the plane H. Then H n S is a figure 8 that cuts S into

three disks, Di , D2 , D3 . Each of these disks contains exactly one critical
point ( either a maximum or a minimum ) . Furthermore, the figure 8 cuts
H into two disks and one unbounded component. At least one of these two
disks, call it D, shares its boundary with one of Di , D2 , D3 . Attach 8D to
the boundary of the appropriate Di , D 2 , D3 , say Di , to obtain a piecewise
smooth 2-sphere D U Di .
After a small isotopy ( that introduces only one new critical point ) , DUDi
has exactly two critical points ( a maximum and a minimum) . Since it has
no saddles, the case above applies and it bounds a 3-ball. By Lemma 3.2.3,
(S\Di ) U D is isotopic to S. After a small isotopy, (S\Di ) U D is smooth
and has no saddles. Thus the situation reduces to the case n = 0.

Figure 3.12. A sphere with one (non-nested) saddle.

Figure 3.13. A sphere with one (nested) saddle.

More generally, suppose that n 2:: 2 and that every 2-sphere in JR3 with
less than n saddles bounds a 3-ball. Then there is a plane H', given by
z = r, for a regular value r, such that there are saddle points both above
and below H' and such that the number of components of H' n S is minimal.
Denote the number of components of H' n S by # I H' n S I . We argue by
induction on (n, # I H' n S I ) .
66 3. 3-Manifolds

Here H' n S is a compact 1-manifold (without boundary) , so H' n S is

a finite union of disjoint simple closed curves in H'. Let a be an innermost
component of H' n S in H'. It follows that there is a disk D c H' that meets
S only in a. Here a separates S into two disks, Di , D2 . Set 81 = D U D 1
and 82 = D U D2 . Then both 81 and 82 are piecewise smooth 2-spheres in
IR.3 . We isotope 81 and 82 slightly (in a way that introduces only one new
critical point) to be smooth. See Figures 3. 14 and 3.15.

Figure 3. 14. A choice of a .

Figure 3 . 1 5 . The spheres 81 and 82 .

At least one of 81 , 82 , say 82 , has fewer saddles than S and hence bounds
a 3-ball. Thus by Lemma 3.2.3, S is isotopic to 81 . Note that 81 has no
more saddles than S and, after a further isotopy as in Lemma 3.2.3, it has at
least one fewer component of intersection with H'. (Note that the isotopies
in Lemma 3.2.3 do not increase the number of saddles in 81 .) Thus, by
induction on ( n , # I H' n SI ) , S bounds a 3-ball. D

Corollary 3.2.6. IR.3 , 18\3 , and §3 are irreducible.

3.2. The Schonfl.ies Theorem 67

Proof. Let S be a 2-sphere submanifold in JR 3 . The Schonflies Theorem

tells us that S bounds a 3-ball. Thus JR3 is irreducible. Let S' be a 2-sphere
submanifold in the 3-ball. lll\3 is a subset of JR3 that can be stretched (via
an appropriate isotopy) to be arbitrarily large without stretching S'. The
proof of the Schonflies Theorem takes place in a compact subset of JR3 and
thus can be adapted to take place in lll\ 3 . Hence S' bounds a 3-ball in lll\ 3 .
Whence the 3-ball is irreducible. Finally, let S" be a 2-sphere submanifold
of §3 . Here §3 = JR3 U oo. We may assume, after an isotopy, that S" misses
oo; then S" c JR3 c §3 . Thus the Schonflies Theorem furnishes a 3-ball
bounded by S" (in JR3 c §3 ) , whence §3 is irreducible. 0

We can say even more about §3 : The 2-sphere S" in §3 bounds a 3-ball
to one side. If we remove a point (that we think of as oo) from the 3-ball
bounded by S", we can repeat the argument in the proof of Corollary 3.2.6.
This produces another 3-ball bounded by S"-to the other side!
The ideas used in the proof of the Schonflies Theorem carry over to other
settings. We prove Alexander's Theorem via an argument analogous to that
used in the proof of the Schonflies Theorem.
Theorem 3.2.7 (Alexander) . Let 'f be a torus submanifold in §3 , then one
of the components oj§3 \ 'f has closure homeomorphic to a solid torus § 1 x lll\ 2 .
We will use the notion of "cobounding" . The manifolds L 1 , . . . , Ln are
said to cobound the manifold L if aL = L 1 u . . . u Ln . For instance, below,
we will see two simple closed curves in a surface that cobound an annulus.
Proof. Let 'f c JR3 be a torus submanifold. We isotope 'f so that the height
function h : JR3 -+ JR given by projection onto the third coordinate restricts
to a Morse function h l 'll' · By the Poincare-Hopf Index Theorem,
#maxima + #minima - #saddles = x('f) = 0.
Denote the number of saddles by n . Note that since 'f is compact, it
contains at least one maximum and one minimum. The Poincare-Hopf Index
Theorem then implies that 'f has at least two saddles.
If n = 2, then 'f has one maximum and one minimum. A level curve of
hl'll' near the maximum bounds a disk in the corresponding level surface of
h. At the first saddle, this disk is either
(1) pinched into two disks (see Figure 3. 16) or
(2) turns into a pinched annulus (see Figure 3.17) .
Between the two saddles a level surface intersects 'f correspondingly either
(1) in two non-nested circles bounding disks or
(2) in two nested circles cobounding an annulus.
68 3. 3-Manifolds

Below the second saddle ( near the minimum) , a level curve of h lir again
bounds a single disk. Thus at the second saddle, either
(1) the two non-nested circles bounding disks are wedged together or
(2) the annulus is pinched.

Figure 3.16. A pinching into two disks.

Figure 3. 17. Pinching into an annulus.

3.2. The Schonflies Theorem 69

In both cases, these disks and annuli stack on top of each other to form
a solid torus. See Figure 3.18 for an illustration of option (1). Note that
in option (2) , there is already a solid torus (annulus) x I between the two
saddles.

... - - - - .... - - - -

Figure 3.18. Between the critical points.

Suppose now that n � 3, and suppose that every torus in �3 with less
than n saddles bounds a solid torus. Choose a regular level surface H, given
by z = r, such that there are saddle points both above and below H and
such that the number of components of H n 1l' is minimal. We now argue by
induction on ( n , # I H n 1!' 1 ) .
Here H n 1l' is a compact I-manifold (without boundary) , so H n 1l' is
a finite union of disjoint simple closed curves in H. Let a be an innermost
component of H n 1l' in H. It follows that there is a disk D c H that meets
S only in a. There are two cases:
Case 1. a is separating in 'll' .
Let P1 , P2 be the surfaces obtained by cutting 1l' along a. Since x (�) is
odd (an orientable surface with one boundary component) , x (Pi) � 2, and
x (P1 ) + x (P2 ) = x('ll') = 0, it must be the case that either P1 or P2 , say
70 3. 3-Manifolds

Pi , is a disk and that the other component, P2 , is a punctured torus. We

proceed as in the proof of the Schonflies Theorem:
Set 8i = D U Pi . Then 8i is a piecewise smooth 2-sphere in JR3 that can
be isotoped to be a smooth 2-sphere. It thus bounds a 3-ball. By Lemma
3.2.3, '][' is isotopic to ('1r\Pi ) U D. Now ('1r\Pi) U D is a piecewise smooth
torus with at most n saddles. It can be isotoped into a smooth torus with at
most n saddles. By Lemma 3.2.3, ('1r\Pi ) U D can be isotoped to intersect H
in one fewer level curve than '1r n H. (Note that the isotopies in Lemma 3.2.3
do not increase the number of saddles of ('1r\Pi ) U D.) Thus, by induction,
('1r\Pi) U D and hence '][' bounds a solid torus.
Case 2. a is non-separating in '1r.
Here '1r\a is an annulus A. We may color the portion of this annulus
that lies above H red and the portion that lies below H blue. Then every
point in A is either colored red or colored blue or lies on one of the circles
in H n '1r. Just above a, A is colored red. Just below a, A is colored blue.
Thus there must be at least one other component a' of H n '][' that is non­
separating in '1r. We may assume that a' is innermost in H. Indeed, if it is
not, then it bounds a disk fJ containing other components of '][' n H; if an
innermost component of '][' n H in fJ is separating, proceed as in Case 1 ; if
an innermost component of '][' n H in fJ is non-separating, replace a' with
this component.
Now '1r\(a U a') consists of two annuli Ai , A 2 . Denote the disk in H\'1r
bounded by a by D and that bounded by a' by D'. Set 8i = Ai U D U D'
and 82 = A 2 U D U D'. Both 8i and 82 are piecewise smooth 2-spheres.
Thus each of them bounds a 3-ball on either side. Two of these 3-balls can
be identified along D and D' to form a solid torus. 0

The following notion is related to irreducibility for manifolds with non­

empty boundary.
Definition 3.2.8. A 3-manifold M is boundary irreducible if every simple
closed curve c in aM that bounds a disk in M cuts M into two 3-manifolds
one of which is a 3-ball.
Example 3.2.9. A boundaryless 3-manifold is necessarily boundary irre­
ducible.
Example 3.2. 10. The 3-ball is boundary irreducible.
The solid torus V = § i x !D>2 is not boundary irreducible. Indeed, consider
the curve m = {p} x aID>2 c av. Then m is non-separating in av; hence
the disk D = {p} x ID>2 is also non-separating.
3.3. 3-Manifolds that are Prime but Reducible 71

Definition 3.2. 1 1 . For V, m, D as above, m is called a meridian and D

is called a meridian disk. A simple closed curve in DV that intersects m in
one point is called a longitude.

Exercises

Exercise 1 . Alexander's Theorem does not generalize to surfaces of genus

greater than or equal to 2, in the sense that not every connected orientable
2-dimensional submanifold of §3 is the boundary of a regular neighborhood
of a graph in §3 . To convince yourself of this, draw a picture of a piecewise
smooth closed orientable surface S of genus 2 in §3 so that neither component
of §3 \S is a regular neighborhood of a graph in §3 .
Exercise 2*. Let M be a covering space of a 3-manifold M. Prove that
M is irreducible only if M is irreducible. (The converse is true but is more
difficult to prove.)
Exercise 3. Use Exercise 2 to deduce that the 3-torus 1l'3 is irreducible.
Exercise 4. Prove that the solid torus is irreducible.

3 . 3 . 3-Manifolds that are Prime but Reducible

There are two compact connected 3-manifolds that deserve attention in the
context of prime and reducible 3-manifolds. These two 3-manifolds are both
§2 -bundles over § 1 • In fact, they are both mapping tori with S = §2 . One
of these 3-manifolds is the product manifold §2 x § 1 ( = (§2 x I)/ ""ids2 ) .
The other is also an (albeit twisted) mapping torus over §2 •
Definition 3.3. 1 . We denote by §2 x § 1 the 3-manifold (§2 x I)/ "'I where
f : §2 --+ §2 is the antipodal map. We also call §2 x § 1 the twisted product of
§2 over § 1 .
We already discussed separating and non-separating simple closed curves
in surfaces. The more general definition is analogous:
Definition 3.3.2. A submanifold A of a connected manifold X is separating
if X\A has at least two components; otherwise it is non-separating.
Remark 3.3.3. Connected manifolds are path connected. It follows that
a submanifold A of a connected manifold X is non-separating if and only if
there is a simple closed curve in X that intersects A transversely in a single
point. Moreover, it follows that a submanifold A of a connected manifold
X is non-separating if and only if there is a simple closed curve in X that
intersects A transversely in an odd number of points.
72 3. 3-Manifolds

The 3-manifolds § 2 x § i and §2 x§ i contain non-separating 2-spheres.

Theorem 3.3.4 below shows that this is a rare property. The proofs of many
theorems in low-dimensional topology, and Theorem 3.3.4 is no exception,
rely on the existence of regular neighborhoods. For more details see Appen­
dix A.
Theorem 3.3.4. An irreducible closed connected 3-manifold is prime. An
orientable closed connected prime 3-manifold is either irreducible or §2 x § i .
Remark 3.3.5. More generally, a closed connected prime 3-manifold is
either irreducible or §2 x § i or § 2 x § i , but we will not prove this fact here.
Proof. A non-trivial connected sum of 3-manifolds contains a sphere that
does not bound a 3-ball. Hence an irreducible 3-manifold is prime.
Suppose M is prime and let S be a 2-sphere in M. If S is separating,
then M S has two components, Ni , N2 . If neither Ni nor N2 is a 3-ball,
-

then M = Ni #N2 and M is not prime. Thus either Ni or N2 is a 3-ball;

i.e. , S bounds a 3-ball.
If S is non-separating, let a be a simple closed curve in M that intersects
S once transversely. Furthermore, let N ( S) be a regular neighborhood of S
in M and let N (a) be a regular neighborhood of a. Both N (a) and N ( S)
are trivial I-bundles, so, in particular, 8N(S) consists of two copies of S.
Transversality of S and a guarantees that N(a) intersects N(S) in a solid
cylinder.
Let M be the submanifold of M obtained by removing the interior of
N ( S) U N (a) . Then {)M is a 2-sphere S in M. Moreover, S is a separating
2-sphere. To one side of S is N(S) U N(a) . Hence there must be a 3-ball to
the other side of S. Therefore M = N(S) U N(a) U (3-ball) = §2 x § i . 0
In Section 5.5, we will prove that prime decompositions of 3-manifolds
exist and that they satisfy an appropriately defined notion of uniqueness.
Exercises
Exercise 1. Construct a 2-to-1 covering map from §2 x
§ i to §2 x § i .
Exercise 2. Consider the connected sum of 'Il'3 and §2 x § i . How many
non-isotopic 2-spheres decompose this manifold as a connected sum?
Exercise 3. Show that § 2 x § i is non-orientable.

3.4. Incompressible Surfaces

In the study of surfaces, essential simple closed curves lying on the surface
play an important role. They allow for the use of cut and paste techniques
3.4. Incompressible Surfaces 73

that suffice to classify surfaces. It is natural to try to proceed analogously

in higher dimensions. This line of investigation has been pursued with a
certain amount of success. We discuss the results in Chapter 5.
Some surfaces contained in a 3-manifold are more interesting than others.
We here discuss one useful class of surfaces. This class of surfaces can be
thought of as a generalization of the notion of "essential simple closed curve"
in a 2-manifold.
Definition 3.4. 1 . A submanifold S in a compact n-manifold M is proper
if aS = S n aM.
Convention 3.4.2. In what follows we will assume that a submanifold S
of M is proper and write simply S (contained) in M or S c M, unless as
is explicitly said to lie elsewhere.

inessential arc

boundary essential arc

Figure 3.19. An inessential arc and an essential arc in the punctured torus.

Definition 3.4.3. A simple arc a in a surface F is essential if there is no

simple arc /3 in aF such that a U /3 is a closed I-manifold that bounds a disk
in F. See Figure 3. 19.
Definition 3.4.4. Let M be a 3-manifold. A surface S in M is compressible
if either
• S is a 2-sphere bounding a 3-ball in M or
• there is a simple closed curve c in S that bounds a disk D with
interior in M\S but that bounds no disk with interior a component
of S\c.
A surface that is not compressible is called incompressible. See Figures
3.20, 3.21, and 3.22. The disk D is called a compressing disk for S, or simply
a compressing disk if the context is clear.
Example 3.4.5. If the surface S c JIB3 is incompressible, then S is a union
of disks. Indeed, each component of as is a simple closed curve in aJIB3 .
Such a curve bounds a disk to either side in aJIB3 and hence bounds a disk
74 3. 3-Manifolds

Figure 3.20. A compressing disk for a surface.

- - - - - - - - - - - - - - - - ·
'
'
'
'
'
'
'
'

Figure 3.21. A compressible 2-torus in the 3-torus.

- - - - - - - - - - ,- - - - 1I ,
,,
I ',
', ' "'
''
I
_ _ _ _ _ _ _ _ _ _ _ _ _ _ J_ _ _ _ _ _ _ _ :,
I
- - - - - - - - - - - - - - - _,
'

Figure 3.22. An incompressible 2-torus in the 3-torus.

in B. Thus each component of 88 must bound a disk in S as well. You will

show in the exercises that there are no closed incompressible surfaces in JIB 3 .
Definition 3.4.6. Let M be a 3-manifold. A surface S c M is boundary
compressible, or a-compressible, if there is an essential simple arc a in S and
an essential simple arc f3 in f)M such that a U f3 is a closed I-manifold that
3. 5 . Dehn 's Lemma 75

bounds a disk D in M with interior disjoint from S. See Figure 3.23. A

surface that is not boundary compressible is boundary incompressible.

Figure 3.23. A boundazy compressing disk.

Definition 3.4.7. Let M be a connected 3-manifold. A 2-sphere S c M is

essential if it does not bound a 3-ball. A surface F c M is boundary parallel
if it is separating and a component of M\F is homeomorphic to F x I . A
surface F in a 3-manifold M is essential if it is incompressible, boundary
incompressible, and not boundary parallel.
The following definition honors Wolfgang Haken, who pioneered the
study of incompressible surfaces in 3-manifolds via a machinery we will dis­
cuss in Chapter 5.
Definition 3.4.8. An orientable irreducible 3-manifold that contains a
proper essential surface is called a Haken 3-manifold.
Remark 3.4.9. In subsequent sections and chapters we will always assume,
unless stated otherwise, that submanifolds are proper.

Exercises
Exercise 1 . Prove that the 2-torus in 'Il'3 pictured in Figure 3.22 is incom­
pressible.
Exercise 2. Prove that a torus in §3 is necessarily compressible.
Exercise 3. Suppose that F is a surface and <P is a homeomorphism of
F. Show that each copy F x { point } of F in the mapping torus of <P is
incompressible.
Exercise 4 *. Show that there are no closed incompressible surfaces in JIB3
or §3 . (Hint: Adapt the proof of Alexander's Theorem.)
3 . 5 . Dehn's Lemma *
Max Dehn was among the first to investigate knots with a view towards
3-manifolds. One of his aims was to answer a fundamental question in knot
theory that we will encounter again in Chapter 4: how to decide whether
or not a given simple closed curve in §3 bounds a disk. In 1910, in Uber
76 3. 3-Manifolds

die Topologie des dreidimensionalen Raumes (see [32] ) , he believed he had

(and was believed to have) established that this is the case if and only if the
fundamental group of the complement of such a submanifold is abelian. But
his proof hinged on a lemma he referred to as simply "the lemma" . This
lemma has since become known as "Dehn's Lemma" .
Theorem 3.5.1 (Dehn's Lemma) . Suppose that M is a 3-manifold and f :
D --+ M is a continuous map from the disk into M such that f (aD) c aM .
If for some neighborhood U of aD, f l u is an embedding, then floD extends
to an embedding.

Dehn's original argument for his lemma was found to be incomplete, as

pointed out by Kneser in 1929. In Geschlossene Flachen in dreidimension­
alen Mannigfaltigkeiten (see [80] ) , Kneser believed he had extended Dehn's
results. Then as his article went to press, he realized that Dehn's argument
concerning the removal of double curves of an immersed disk was incom­
plete. He added a brief note to this effect at the end of his article [80,
p. 260] . Dehn never fixed the proof of his lemma. It deserves to be men­
tioned that the teaching position he took on after fleeing Germany left him
little time for research.
A proof of Dehn's Lemma was finally given by Christos Papakyriakopou­
los in 1957; see [119] . In this proof, Papakyriakopoulos employed what he
termed a "tower construction" . Since then, this proof of Dehn's Lemma has
undergone various revisions. For instance in Jaco and Rubinstein's PL min­
imal surfaces in 3-manifolds (see [69] ) , the authors introduce a PL theory
of minimal surfaces and use it, among other things, to provide an alter­
nate proof of Dehn's Lemma. In their proof, Papakyriakopoulos's universal
covers are replaced by 2-fold covers. Here is a sketch of an argument:
Let f : D --+ M be a map as specified in Dehn's Lemma. It is then pos­
sible, though not obviously so, to assume that the self-intersections of f(D)
occur in the interior of f (D) and are modeled on the two types indicated in
Figures 3.24 and 3.25.
The "easy" case (the argument presented by Dehn covered this case) :
If the self-intersections of f(D) occur along double curves and there are no
triple points, then the preimage of such a double curve consists of two simple
closed curves in D. Let a be an innermost simple closed curve in D that
is the preimage of a double curve and let b be another simple closed curve
such that f (a) = f(b) . Here a and b bound disks Da and Db in D. Two
cases need to be considered: (1) Da and Db are disjoint; (2) Da c Db.
If Da and Db are disjoint, then we can alter f by swapping f(Da) and
f(Db) (the image of Da under f is replaced with f (Db) and vice versa; see
Lemma 3.2.3) . After isotoping the resulting map slightly, all assumptions of
3. 5 . Dehn 's Lemma 77

Figure 3.24. A double curve.

Figure 3.25. A triple point.

Dehn's Lemma are still met, but there are fewer curves of self-intersection
in the image of the new map.
If Da C Db, then we can alter f by replacing f(Db) by f(Da) · The result
is a map satisfying all assumptions of Dehn's Lemma but whose image has
fewer self-intersections. Thus the curves of self-intersection of f(D) can be
removed to produce the required embedding.
The "hard" case: If the self-intersections of f(D) include triple points,
we must proceed with more caution. Note that f(D) is not a submanifold.
Nevertheless, it is possible to construct an analog of a regular neighborhood
of f(D) in M. We denote this regular neighborhood by Vi . See Figure 3.26.
If Vi has only 2-sphere boundary components, then f(8D) (a simple
closed curve on 8Vi ) bounds an embedded disk in M, as required. If Vi
has other surfaces as boundary components, then it has positive first Betti
number. So it is the base space for many covering spaces. Let (M1 , Vi , p 1 )
be a 2-fold cover where Mi is connected.
For each point in f ( D) there are exactly two points in p! 1 (f ( D) ) . More­
over, Mi is homeomorphic to a regular neighborhood of P1 1 (f(D) ) . Thus
P1 1 (f(D)) is connected. At the same time, P1 1 (f(D)) is the lift of the image
of the continuous map f. Since D is simply connected, f has two lifts to
78 3. 3-Manifolds

boundary of M

Figure 3.26. A "cross-section " of f (D) .

M1 . Thus P1 1 (f(D)) is the image of a continuous map Ji from the disjoint

union of two disks into Mi . Here P I maps double curves and triple points
of p! 1 (f(D)) to double curves and triple points of f(D) , but not vice versa.
Suffice it to say that a complexity argument allows the conclusion that there
are fewer double curves in the image of at least one of these disks; call it
D1. Note that we can identify D with Di so that (p o fi) I Di = f.
Now Ji, Di , Mi are as required in the hypotheses of Dehn's Lemma.
Thus we can repeat the process above. We conclude that either fi (8D 1 )
bounds an embedded disk in M1 or there is a connected 2-fold cover
(M2 , V2 , p2 ) with Vi a regular neighborhood of fi (D 1 ) in Mi and a con­
tinuous function h : D 2 --+ M2 satisfying the hypotheses of Dehn's Lemma
but such that f2(D 2 ) has fewer double curves than fi (D 1 ). If there are triple
points in f2(D 2 ) , we repeat the process. After some number of iterations,
we reach a 2-fold cover (Mn , Vn , Pn ) and a map fn : Dn --+ Mn such that
fn is an embedding. (This alternating process of taking submanifolds and
2-fold covers is known as a "tower construction" .)
Note that since fn is an embedding, Pn lfn (Dn ) is at most 2-to-l . Thus
since (Pn o fn ) I Dn = fn- l i fn-1 (Dn- i } has no triple points. Points at which
fn- 1 (Dn-1 ) is not embedded come from pairs of points of fn (Dn ) and are
hence double points. It follows from the argument in the "easy" case that
fn - 1 (8Dn - 1 ) bounds an embedded disk in Mn- I · This gives an embedding
of Dn- l into Mn - I that we continue to denote by fn- 1 : Dn-1 --+ Mn-1 ·
Then since (Pn - 1 o fn-1 ) I Dn -i = fn-2 , fn - 2 (Dn - 2 ) has no triple points, etc.
By induction, we can conclude that f(8D) bounds an embedded disk in M.
3. 5 . Dehn 's Lemma 79

The following quintuplet is attributed to John Milnor:

The perfidious Lemma of Dehn,
put many a good man to shame.
But Christos Pap-
akyriakop-
oulos proved it without any pain.
The tower construction also allowed Papakyriakopoulos to prove two
other theorems that he called the Loop Theorem and the Sphere Theorem.
Below is a generalization of the Loop Theorem formulated by John Stallings;
see [149] :
Theorem 3.5.2 (The Loop Theorem) . Let M be a 3-manifold and F a
connected surface in 8M. If N is a normal subgroup of 7r1 (F) and if
ker( 7r1 ( F) -t 7r 1 ( M)) /N i= 0, then there is a proper embedding g : ( D, 8D)
-t (M, F) such that [glaD] is not in N.

Corollary 3.5.3. Let M be a 3-manifold and F a connected surface in

8M . If ker(7r 1 (F) -t 7r1 (M)) i= {1}, then there is a proper embedding
g : (D, 8D) -t (M, F) such that [gl aD ] is non-trivial in 7r 1 (F) .

The geometric interpretation of the corollary above is that if the funda­

mental group of F does not inject into the fundamental group of M, then
F is compressible.
Theorem 3.5.4 (The Sphere Theorem) . Let M be an orientable 3-manifold
and N a 7r1 (M) -invariant subgroup of 7r2 (M) . If 7r2 (M)/N i= 0, then there
is an embedding g : §2 -t M such that [g] is not in N.

The common theme to Dehn's Lemma and the results mentioned above
is that of promoting an "immersed" disk (a disk with double curves and
triple points) to an embedded disk. In the setting of Dehn's Lemma this
is achieved for a disk while fixing its boundary. In the setting of the Loop
Theorem this is achieved for a disk while fixing the homotopy class of its
boundary (not the boundary itself) . In the setting of the Sphere Theorem
this is achieved for a sphere. This common theme recurs quite frequently in
the subject of 3-dimensional topology.
You will prove the following lemma in the exercises.
Lemma 3.5.5. Let M be a 3-manifold. A surface F C M is incompressible
if and only if 7r1 ( F) -t 7r 1 ( M) is injective.
80 3. 3-Manifolds

Exercises
Exercise 1. The contrapositive of the Loop Theorem with N = (1) states
that if a surface F in a 3-manifold M is incompressible, then 71'I (F) -+ 11' 1 (M)
is injective. Show that the converse is true.
Exercise 2. Prove Lemma 3.5.5.
Exercise 3. Let F be a compact surface without boundary. List all, up
to isotopy, incompressible boundary surfaces (those with and those without
boundary) in M = F x I .

3.6. Hierarchies *
The idea of a hierarchy parallels the process of cutting a surface along sim­
ple closed curves and simple arcs until only disks remain. Surfaces can be
classified by reversing this procedure. This approach was not used in Chap­
ter 2 but is discussed, for instance, in [63, Chapter 13) . See Figure 3.27.
Codimension 1 submanifolds of a surface are curves and arcs. Codimen­
sion 1 submanifolds of a 3-manifold are surfaces. A hierarchy provides a
sequence of surfaces that cuts a 3-manifold into 3-balls. The idea goes back
to Wolfgang Haken, whose work we will discuss in Chapter 5.

Figure 3.27. The codimension 1 submanifolds in a 2-dimensional ana­

log of a hierarchy.

Definition 3.6. 1. Let M be a compact 3-manifold. A hierarchy for M is a

finite sequence of pairs (MI , FI ) , . . . , (Mn , Fn ) such that
(1) Fi is a 2-sided incompressible and a-incompressible surface in Mi;
(2) Mi = Mi - I \77 (.Fi - 1 ) , for 77 (.Fi - I ) an open regular neighborhood of
Fi - I (we also say that Mi is obtained by cutting Mi - I along Fi-d i
(3) MI = M;
(4) each component of Mn+ I is a 3-ball.
3. 6. Hierarchies 81

As an example, we illustrate a hierarchy for the 3-torus: Here (Mi , Fi)

consists of the 3-torus Mi and an incompressible torus Fi contained in Mi .
See Figure 3.28. The second pair, (M2 , F2 ) , consists of torus x I and an
incompressible annulus therein. See Figure 3.29. The third pair, (M3 , F3 ) ,
consists of annulus x I and an essential disk therein. See Figure 3.30. Cut­
ting M3 along F3 yields a 3-ball. So this is a hierarchy.

'
'

Figure 3.28. An incompressible torus in the 3-torus.

Figure 3.29. An incompressible annulus in torus x I.

Recall that an orientable irreducible 3-manifold is Haken if it contains

a 2-sided incompressible surface. Haken 3-manifolds played a major role in
the study of 3-dimensional topology in the 1960s, 1970s, and 1980s. We are
restricting our attention to the case of orientable 3-manifolds, though many
of the results discussed here hold, in a more general form, for non-orientable
3-manifolds. The main theorem concerning hierarchies is the following:
Theorem 3.6.2. A compact orientable irreducible 3-manifold M is Haken
if and only if M has a hierarchy.
82 3. 3-Manifolds

'
, '
, '
,
, \
\
F3

Figure 3.30. A disk in annulus x I.

Theorem 3.6.2 tells us that a given Haken manifold can be constructed

by taking an appropriate collection of 3-balls and identifying appropriate
portions of their boundaries.
A key ingredient in proving this theorem is the following lemma:
Lemma 3.6.3. Suppose M is a compact orientable 3-manifold such that {)M
contains a surface of positive genus. Then M contains a properly embedded,
2-sided, incompressible, a-incompressible surface F such that {1} '::/: [BF] E
tr1 (8M) .
The proof of this lemma is not hard but relies on an understanding of
homology. It can be found in [67, Theorem 111.10] or [63, Lemma 6.8] .
Definition 3.6.4. The length of the hierarchy (Mi , Fi) , . , (Mn , Fn ) is n . . .

Haken manifolds admit hierarchies of a very special type:

Theorem 3.6.5. Let M be a Haken manifold. Then M has a hierarchy of
length 4.
Specifically, a Haken manifold admits a hierarchy
(Mi , F1 ), (M2 , F2 ), (M3 , F3 ), (M4 , F4 )
where F1 consists of disjoint closed surfaces in M, F2 consists of disjoint
surfaces with non-empty boundary in M2 , F3 consists of disjoint annuli, and
F4 consists of disjoint disks.
This is [67, Theorem IV. 19] .
Remark 3.6.6. In Theorem 3.6.5, F4 consists of disks and Ms consists of
3-balls. It follows that M4 is a disjoint union of handlebodies.
The existence of the hierarchy of length 4 provided by Theorem 3.6.5
is proved by employing important existence and maximality results. For
3. 6. Ilierarchies 83

instance, Fi is a maximal collection of disjoint non-parallel closed incom­

pressible surfaces in Mi . If Mi is closed, then a collection of disjoint non­
parallel closed incompressible surfaces in Mi exists because Mi is Haken.
The fact that a finite maximal such collection exists follows from Theorems
5.4.3 and 5.4.5 (together known as Kneser-Haken finiteness) . The collec­
tions of surfaces F2 , Fa exist by Lemma 3.6.3, and Kneser-Haken finiteness
again establishes the fact that there are finite maximal collections of disjoint
non-parallel incompressible surfaces of the specified type.
The cut and paste techniques that provide the basis for understanding
2-manifolds have a chance of succeeding in the context of Haken manifolds.
This can be seen as a motivating principle behind some of Haken's work and
also for the theorem of Waldhausen mentioned below. However, the extra
dimension does add difficulties. In addition, note that many (some would
say "most" ) 3-manifolds are not Haken.
Definition 3.6. 7. Two topological spaces X, Y are homotopy equivalent if
there are continuous maps f : X --+ Y and g Y --+ X such that g o f is
:

homotopic to idx and f o g is homotopic to idy .

A key result on Haken manifolds is due to Friedhelm Waldhausen; see
[156] .
Theorem 3.6.8 (Waldhausen's Theorem) . Homotopy equivalent closed
Haken manifolds are homeomorphic.
This theorem does not hold for 3-manifolds in general. Consider for
instance the following construction: Let Ti , T2 be solid tori. Here 'n =
IIB 2 x § i . We denote the simple closed curve ( 8IIB 2 ) x {point} by mi and we
denote the simple closed curve p x § i , where p E 8IIB2 , by li . Now identify
Ti and T2 along their boundaries via the element [h] of the mapping class
group of o'n (see Lemma 2.6.2) corresponding to the matrix

where p, q are relatively prime integers and r, s are such that ps + qr = 1 .

(Thus m i is identified with a simple closed curve in 8T2 that meets m2 in q
(equivalently oriented) points and h in p (equivalently oriented) points.)
While Haken manifolds constitute what might look like a small subset
of 3-manifolds, recent work of Agol shows that if M is an irreducible 3-
manifold, then M is finitely covered either by §3 or by a Haken manifold.
See [2] .
Definition 3.6.9. A 3-manifold that is constructed as above is called a lens
space . It is denoted by L(p, q) .
84 3. 3-Manifolds

As it turns out, L(7, 1) and L(7, 2) are homotopy equivalent but not
homeomorphic. (See [63, Lemma 3.23] and [127, p. 235] and the exercises.)
Theorem 3.6.10 (The Poincare Conjecture) . If M is a closed 3-manifold
that is homotopy equivalent to §3 , then M is homeomorphic to §3 .
The Poincare Conjecture was an open conjecture for over a century. It
was finally proved by G. Perelman in the early 21st century using geometric
and analytic methods.
We now describe the Whitehead manifold. It is another manifold that is
of interest in the context of Waldhausen's Theorem and the Poincare Con­
jecture. The Whitehead manifold is an example of an irreducible 3-manifold
that is open, connected, contractible, and hence homotopy equivalent to the
open 3-ball, yet it is not homeomorphic to the open 3-ball. We construct
the Whitehead manifold as follows: Consider the solid torus T c §3 as in
Figure 3.31. Map T into itself via the map h : T --+ T so that h(T) lies
inside of T as in Figure 3.31.

Figure 3.31. The solid torus h(T) within the solid torus T.

Now consider the solid torus h(T) and map h(T) into itself in the same
way to obtain h2 (T). Continuing in this fashion yields an infinite sequence
of nested solid tori T, h(T), h 2 ( T) , . . . . Set X = n hn (T) . The Whitehead
manifold is W = §3 \X.
Proposition 3.6. 1 1 . The Whitehead manifold is open, irreducible, con­
nected, and simply connected.

Proof. Here we set

and
3. 6. Hierarchies 85

Then Wn is open,

and
W = Wo U W1 U W2 U . . . .
In particular, W is open. Note that in §3 , h(T) is isotopic to T; see Fig­
ure 3.32. (The isotopy is accomplished by "untwisting" the "inner" portion
of h(T) as in Figure 3.32 and then shortening the same portion.) Thus, Vn
Wn is homeomorphic to Wo.

Figure 3.32. The solid torus h(T) is isotopic to T.

To see that W is irreducible, consider a 2-sphere submanifold S in W.

Since S is compact and {Wn } is an open cover for W, 3n such that S c
Wn . Now Wn c §3 , so S c §3 where it is separating and bounds 3-balls to
either side. Since Xn is connected, it lies in only one of these 3-balls; thus
the other 3-ball bounded by S lies in Wn and hence in W.
To see that W is connected, note that a manifold is connected if and only
if it is path-connected. So suppose that x , y E W. By the same reasoning
as above, 3m such that x , y E Wm . Since Wm is path-connected, there is a
path from x to y lying in Wm and hence in W.
To see that W is simply connected, consider an embedded curve c in W.
Since c is compact, by the same reasoning as above, 3l such that c C Wz .
It hence suffices to show that 7r 1 (W1) is trivial in 7r1 (W) . We will show
that 11"1 (W1) is trivial in 11"1 (W1 +i ) by establishing that 11" 1 (Wo) is trivial in
7r1 (W1 ) .
Consider the curve J in Figure 3.31. Here J is the generator for 7r1 (Wo).
The isotopy in Figure 3.32 affects J as pictured in Figure 3.33. After a
homotopy, J is as in Figure 3.34 and thus bounds a disk in W1 . 0

Remark 3.6.12. In fact, the Whitehead manifold is contractible. This

is a non-trivial consequence of the fact that it is irreducible and simply
connected: By the Sphere Theorem, 7r2 (W) = {1} since W is irreducible.
It follows from a theorem of Whitehead that a connected open 3-manifold
86 3. 3-Manifolds

Figure 3.33. The curve J.

Figure 3.34. The curve J.

with trivial fundamental group and second homotopy group is homotopic to

a point.

The following definition is of interest in this context. We will not use it

otherwise.
Definition 3.6.13. A topological space W is said to be simply connected
at infinity if for every compact subset K C W there is a compact set K'
such that K c K' c W and the inclusion map W \ K' <---+ W \ K induces the
trivial map on fundamental groups.
Theorem 3.6.14 (Brin-Thickstun) . An open, connected, contractible 3-
manifold that is simply connected at infinity is homeomorphic to the open
3-ball.
3. 7. Seifert Fibered Spaces 87

Exercises

Exercise 1 *. Prove that £(7, 1) and £(7, 2) are homotopy equivalent but
not homeomorphic. (This is challenging. See [63 , Lemma 3.23] for a proof
that £(7, 1) and L(7, 2) are homotopy equivalent. To see that they are
not homeomorphic, see the discussion in [127, p. 235] and the references
therein.)
Exercise 2. Prove that the Whitehead manifold is not simply connected
at infinity.
Exercise 3. Prove that the Whitehead manifold is not homeomorphic to
the open 3-ball.

3 . 7. Seifert Fibered Spaces

Seifert fibered spaces constitute an important class of 3-manifolds. For de­

tails, see [146] . Not all of them are Haken, but many are. We discuss them
here to provide more examples of 3-manifolds in general and of Haken man­
ifolds in particular. Much can be said about Seifert fibered spaces. We will
only touch on some of the highlights. One of the reasons for the continued
interest in Seifert fibered spaces is the fact that they project in a natural
way onto a 2-dimensional "base space" . This fact makes them particularly
amenable to computations.
Definition 3.7. 1 . The fibered solid torus of type (l, m) is a solid torus parti­
tioned into circles obtained as follows: Consider the cylinder 1!» 2 x [- 1, l] . It
is partitioned into intervals of the form {point} x [- 1, l] . Identify 1!»2 x { -1}
to 1!»2 x {1} by setting (rei21T8 , 1) equal to (rei27T (8+ .!f' ) , -1), for m E N U {O}
and l E N . A circle formed by intervals of the form {point} x [- 1 , 1] is called
a fiber of the fibered solid torus.
The core of a. fibered solid torus is the fiber obtained from {O} x I . If
l > 1, then we call the fibered solid torus T of type (l, m) an exceptionally
fibered solid torus. In this case the core of T is called an exceptional fiber
and all other fibers of T are called regular fibers . If l = 1, then we call a
fibered solid torus T of type (l, m) a regularly fibered solid torus. In this case
all of the fibers of T are called regular fibers. See Figure 3.35. In addition,
when l = 1, the disk D = 1!»2 x {O} c T intersects each fiber once, so there
is a bundle (T, § 1 , D, p) , where p : T -+ D send each fiber to its point of
intersection with D .
Suppose T1 and T2 are fibered solid tori. A fiber-preserving homeomor­
phism between T1 and T2 is a homeomorphism h : T1 -+ T2 that takes fibers
to fibers.
88 3. 3-Manifolds

identify after twist

- - -

Figure 3.35. A fibered solid torus.

Remark 3.7.2. We may assume that 0 :::; m < � · Indeed, any fibered solid
torus is homeomorphic via a fiber-preserving homeomorphism to a fibered
solid torus of type (l, m) with l, m satisfying this constraint. With this
assumption, the existence of a fiber-preserving homeomorphism between a
fibered solid torus of type (li , m 1 ) and a fibered solid torus of type (l 2 , m2 )
necessitates li = l2 and m 1 = m2 .
Definition 3. 7.3. A fibered neighborhood of a simple closed curve f in a
3-manifold M that consists of simple closed curves is
• a closed regular neighborhood homeomorphic to a fibered solid
torus via a fiber-preserving homeomorphism when f is in the inte­
rior of M or
• a closed regular neighborhood homeomorphic to a product (half-disk)
x § 1 partitioned into fibers {point} x § 1 when f c 8M.

Definition 3.7.4. An orientable 3-manifold M is a Seifert fibered space if

M is the union of pairwise disjoint simple closed curves called fibers such
that each fiber in M has a closed regular neighborhood that is a fibered
neighborhood. The description of M in terms of fibers is called the Seifert
fibration of M.
Denote the quotient space of M obtained by identifying each fiber to
a point by B and denote the corresponding quotient map by p : M -t B.
Here B is called the base space of M.
Analogous to fibered solid tori, there are also fibered solid Klein bot­
tles. These allow an extension of the notion of Seifert fibered space to
non-orientable 3-manifolds, but we will not pursue this more general topic
here.
3. 7. Seifert Fibered Spaces 89

Remark 3.7.5. Note that an § 1 -bundle over a surface is a Seifert fibered

space (with no exceptional fibers, whether or not the bundle is trivial) . How­
ever, a Seifert fibered space is not necessarily an § 1 -bundle over a surface.
(The complement of the exceptional fibers of a Seifert fibered space is an
§ 1 -bundle over a surface.)
Lens spaces are examples of Seifert fibered spaces. To see the fibers of
L(p, q) , recall that L(p, q) is constructed by identifying two solid tori Ti , T2
along their boundaries via a homeomorphism corresponding to
Ah = [: :].
Each of the two solid tori is a fibered solid torus in many different ways.
We need to describe T1 , T2 as fibered solid tori in such a way that the
fibers match up along 8T1 = 8T2 . There are several ways to do this. One
possibility is the following: Using the notation in the discussion preceding
Definition 3.6.9, choose a simple closed curve c on 8T1 = 8T2 that intersects
m 1 once. Any simple closed curve that intersects m2 in r # 0 (equivalently
oriented) points and l 2 in s (equivalently oriented) points will do. Now there
is exactly one way in which � is a fibered solid torus with c as a fiber. This
decomposes L(p, q) into fibers as required.
Remark 3. 7.6. The above examples illustrate the lack of uniqueness of a
Seifert fibration. There are infinitely many ways to fiber a solid torus. This
translates into infinitely many different Seifert fibrations for a lens space.
Interestingly, Seifert fibrations for more complicated Seifert fibered spaces
are unique up to isotopy. See Theorem 3.7. 19 below.
The 3-sphere arises as a lens space, specifically, §3 = L( 1, n ) . The
exceptional fibers of the Seifert fibration of §3 are pictured in Figure 3.36.

Figure 3.36. The exceptional fibers of a Seifert fibration of §3 •

90 3. 3-Manifolds

Example 3.7.7. Recall the non-trivial I-bundle over the Mobius band de­
scribed in Example 3. 1 . 10. The total space of this bundle is a solid torus,
which can be viewed as a fibered solid torus in many different ways and is
hence a Seifert fibered space.
Analogously, we can define a non-trivial I-bundle over the Klein bottle,
called the twisted I-bundle over the Klein bottle. The total space of this
bundle can be constructed from a cube by appropriately identifying the
sides (front to back and left to right) . The cube consists of intervals running
front to back and also consists of intervals running left to right. In both
cases, the intervals yield fibers of a Seifert fibration for the resulting total
space. Thus the twisted I-bundle over the Klein bottle is a Seifert fibered
space and it admits distinct Seifert fibrations.
Definition 3. 7.8. Let M be a 3-manifold with non-empty boundary. The
double of M is the 3-manifold obtained from two copies of M that have been
identified (via the identity map) along their boundaries.
Example 3. 7.9. The double of a Seifert fibered space is also a Seifert fibered
space.
Lemma 3.7.10. If Ti and T2 are fibered neighborhoods of the fiber f, then
f is an exceptional fiber of Ti if and only if f is an exceptional fiber of T2 .

Proof. This is a consequence of the uniqueness of regular neighborhoods;

see Theorem 3. 1 . 13 and Remark 3.7.2. D

Lemma 3.7.10 allows us to refer to the fibers of a Seifert fibered space

as exceptional fibers or regular fibers unambiguously.
Remark 3.7. 1 1 . The base space B of a Seifert fibered space is a surface.
This can be seen as follows: The quotient of a fibered solid torus is a disk. Let
q be a point in B. Then p- i (q) is a fiber f of M. The fibered neighborhood T
of f hence yields a neighborhood p(T) of q that is homeomorphic to a disk. If
f is an exceptional fiber, then p(f) is called an exceptional point (also called
a cone point) . Sometimes we attach a number to an exceptional point.
This number represents the number l of the (exceptionally) fibered solid
torus neighborhood of f. Thus we think of B as a surface with exceptional
points.
We can describe several Seifert fibered spaces in terms of their base space.
One class of Seifert fibered spaces is called prism manifolds. Prism manifolds
are Seifert fibered spaces with base space a 2-sphere and three exceptional
fibers. Some prism manifolds, albeit a minority, are Haken manifolds. For
more on Seifert fibered spaces and Haken manifolds see Theorem 3.7.18.
3. 7. Seifert Fibered Spaces 91

---

Figure 3.37. The base space of a lens space.

• •

Figure 3.38. The base space of a prism manifold.

Remark 3.7. 12. The interiors of the fibered neighborhoods of fibers form
an open cover for a Seifert fibered space. It follows that in a compact Seifert
fibered space, there will be only finitely many exceptional fibers.
Definition 3. 7.13. A subset Z of X is saturated with respect to p : X -t Y
if Z = p -1 (p(Z)).

In the construction of lens spaces, two solid tori were identified along
their boundaries. The torus resulting from the identification of the boundary
tori is a saturated torus with respect to the quotient map. It corresponds
to a circle in the base space. See Figure 3.37.
The Seifert fibration of a Seifert fibered space provides a means of de­
scribing and isotoping surfaces. Incompressible surfaces in Seifert fibered
spaces can be described completely. (See Theorem 3.7. 16.) As a warm-up,
we will prove a lemma.
Lemma 3.7. 14. A properly embedded connected orientable incompressible
surface F in a solid torus V is either a disk or an annulus.
92 3. 3-Manifolds

Proof. We first show that a solid torus V is irreducible. Here V is homeo­

morphic to the solid torus
{(x, y, z ) E IR3 : (Jx 2 + y 2 - 2 ) 2 + z 2 = l } .
Thus we may assume that V c IR3 c §3 . Now given a 2-sphere S E V,
S E §3 , where it bounds 3-balls to either side. Since aV is connected and
disjoint from S, one of these 3-balls lies entirely in V. Hence V is irreducible.
Let V be a solid torus and suppose that F c V is incompressible. Let
D be a meridian disk of V. The proof is by induction on the number of
components of F n D. If F n D is empty, then F is an incompressible
surface in the 3-ball V\D and is hence a disk. (See Example 3.4.5.) Simple
closed curves in F n D can be eliminated via a standard innermost disk
argument; see Lemma 3.2.3 and Example 3.9. 1. Thus F n D consists of arcs.
For C a component of aF, either C is inessential or essential in aV. In
the case of the former, C bounds a disk in aV c V. Since F is incompress­
ible, this implies that F is a disk. Thus we may assume, in what follows,
that all components of aF are essential in aV. It follows that the compo­
nents of aF are parallel essential simple closed curves in aV. We isotope F
to minimize the number of components of F n D.
Suppose that F n D # 0. Let a be an outermost arc in D. You will
show in the exercises that the endpoints of a lie on distinct components
of aF. Let b, c be the two components of aF containing the endpoints of
a. Let N(a U b U c) be a regular neighborhood of a U b U c in F. It is a
thrice-punctured sphere. You will show in the exercises that the boundary
component of N(a U b U c) that lies in the interior of F bounds a disk in
M. It hence bounds a disk in F. Thus F = N(a U b U c) U (disk) is an
annulus. 0

To further understand surfaces lying in Seifert fibered spaces, the fol­

lowing definitions prove useful:
Definition 3. 7. 15. A surface in a Seifert fibered space is vertical if it sat­
urated, i.e. , if it consists of fibers. A surface in a Seifert fibered space is
horizontal if it is everywhere transverse to the fibration.
Saturated annuli and tori are examples of vertical surfaces in Seifert
fibered spaces. Covering spaces of the base space of a Seifert fibered space
can sometimes be embedded in the Seifert fibered space. (This fact accounts
for those prism manifolds that are Haken.) When this is possible, such a
surface is an example of a horizontal surface in a Seifert fibered space.
Theorem 3.7.16. An orientable essential surface in an orientable Seifert
fibered space can be isotoped so that it is either horizontal or vertical.
3. 7. Seifert Fibered Spaces 93

The proof of this theorem is lengthy; see [67, Theorem Vl.34] .

We conclude this section with a few well-known theorems that provide
a brief overview of further results on Seifert fibered spaces.
Theorem 3. 7 .17. An orientable Seifert fibered space is either irreducible
or homeomorphic to §2 x § 1 or JRIP3 #JRIP3 .

Proof. Let M be an orientable Seifert fibered space. Two cases need to be

considered:
Case 1. fJM i= 0.
Let B be the base space of M. Then fJB i= 0. The proof is by induction
on the pair ( - x (B) , # exceptional points) (in the dictionary order) . Suppose
first that B is a disk with at most one exceptional point. Then M is a solid
torus. The first paragraph in the proof of Lemma 3.7. 14 established that a
solid torus is irreducible.
If there are at least two exceptional points in B, then there is an arc
a in B that cuts off a disk containing exactly one exceptional point. The
arc a corresponds to an essential saturated annulus p - 1 (a) . Similarly, if
- x (B) 2'.: 0, then there is an essential arc in B corresponding to an essential
saturated annulus in M. Denote the essential saturated annulus by A.
Let S be a 2-sphere in M. If S is disjoint from A, then S lies in M\A,
the interior of a Seifert fibered space with boundary. By the inductive
hypothesis, this Seifert fibered space is irreducible. Hence S bounds a 3-ball
in M\A. Thus S bounds a 3-ball in M.
Suppose that AnS i= 0. Let c be a component of Ans that is innermost
in S. Then one component of S\c is a disk D that is disjoint from S. In
the exercises, you will prove a result that implies that A is incompressible.
Since A is incompressible, c is inessential in A and hence one component,
D', of A\c is a disk. Here D U D' U c is a 2-sphere that, after a small isotopy,
lies in one component of M\A and hence bounds a 3-ball. Proceeding as in
Lemma 3.2.3 provides an isotopy of S reducing the number of components
in A n S. Thus if we isotope S so that the number of components of A n S
is minimal, then A n S 0. Above we proved that if S is disjoint from A,
=

then S bounds a 3-ball.

Case 2. fJM = 0.
If M contains a saturated incompressible torus T, then the argument
above can be applied to show that a 2-sphere S c M can be isotoped to be
disjoint from T. Thus S lies in M\T, the interior of a Seifert fibered space
with boundary. It follows that S bounds a 3-ball in M\T and hence in M.
94 3. 3-Manifolds

It remains to consider the cases where B is either a 2-sphere with at most

three exceptional fibers or a projective plane with at most one exceptional
fiber. In the exercises you will prove that in all other cases M contains a
saturated incompressible torus.
The case of lens spaces (B = § 2 with up to two exceptional points)
is left as an exercise. As it turns out, a lens space is either irreducible
or homeomorphic to § 2 x § 1 . In addition, in the case of prism manifolds
(B = §2 with three exceptional points) we refer to a theorem of Waldhausen,
which states that prism manifolds are irreducible.
To treat the remaining case, suppose that B is a projective plane with at
most one exceptional point. Let t be a simple closed curve in B such that one
component of B\t is a disk containing the exceptional point if there is one.
Let T = p- 1 (t) be the corresponding saturated torus. Since t is separating,
so is T. T cuts M into two components: (1) M', a circle bundle over the
Mobius band with bundle structure coming from the Seifert fibration and
(2) V, a fibered solid torus. Since M, and thus M', is orientable, M' must
be the twisted circle bundle over the Mobius band.
Suppose S is a 2-sphere in M. If S is disjoint from T, then it lies in
either M' or V and hence bounds a 3-ball. We may thus assume, in what
follows, that S n T is not empty. Let c be a component of S n T that is
innermost in S. Then at least one component of S\c is a disk D that is
disjoint from T. If c is inessential in T, then a component of T\c is a disk
D' and D U D' U c is a 2-sphere that, after isotopy, lies in a Seifert fibered
space with boundary where it bounds a 3-ball. By Lemma 3.2.3, there is an
isotopy reducing the number of components of S n T. Thus, if the number
of components of S n T is minimal, then S n T consists of curves essential
in T.
We make the following observations (the proofs are left to the reader) :
We assume that the number of components of S n T is minimal. Under this
assumption:
(1) there are no components of S n M' or S n V that are boundary
compressible via a boundary compressing disk that meets more than one
component of S n T;
(2) 8M' is incompressible in M' ;
(3) a properly embedded orientable incompressible surface with non­
empty boundary in M' that is not boundary compressible via a boundary
compressing disk that meets more than one component of S n T must be an
annulus that double covers the Mobius band.
Exercises 95

It now follows that S must consist of an annulus in M' that double covers
the Mobius band together with two disks in V. Thus S cuts M into two
identical pieces, each consisting of one of the two twisted /-bundles over a
Mobius band coming from M' and one of the 3-balls coming from V. Each
of these pieces is homeomorphic to a once-punctured JRIP3 . 0

Theorem 3. 7. 18. A closed orientable Seifert fibered space M is either a

Haken manifold, a lens space {including §2 x § 1 , §3 ), JRIP3 #JRIP3 , or a prism
manifold. In the latter case, M is Haken if and only if H1 (M) is infinite .
See [67, Theorem VI. 15] .
Theorem 3. 7. 19. The only Seifert fibered spaces with non-unique fiberings
are:
(a) lens spaces {including §2 x § 1 , §3 };
(b) prism manifolds (only in some cases);
( c) the solid torus;
( d) the twisted I-bundle over the Klein bottle;
( e) the double of the twisted I-bundle over the Klein bottle {this 3-mani­
fold also fibers over §2 with four exceptional fibers for which l 1 = l2 = l3 =
l4 = 2).
See [67, VI. 16] .
The following theorem was long known as the Topological Seifert Con­
jecture. It was finally proved by a series of results due to Tukia, Mess,
Gabai, Casson-Jungreis, and Scott.
Theorem 3. 7.20. Let M be a compact orientable irreducible 3-manifold
with 11" 1 ( M) infinite. Then M is a Seifert fibered space if and only if 11" 1 ( M)
has a normal subgroup isomorphic to Z.

Exercises

Exercise 1 . Prove Remark 3.7 . 2 .

Exercise 2. Let V, D, a, F be as in Lemma 3.7.14 and assume that F has
been isotoped so that the number of components of F n D is minimal. Show
that the endpoints of a lie on distinct components of oF.
Exercise 3. Let V, D, a, F, b, c, N(a U b U c) be as in Lemma 3.7. 14 and
assume that F has been isotoped so that the number of components of
F n D is minimal. Observe that a cuts off a disk fJ from D. Here b U c cuts
8V into two annuli. One of these annuli, call it A, meets fJ. Use A together
96 3. 3-Manifolds

with b to show that the component aN(a U b U c)\(b U c) bounds a disk

in M.
Exercise 4. Prove the observations near the end of the proof of Theorem
3.7. 17.
Exercise 5. Prove that lens spaces are irreducible.
Exercise 6. Prove that a saturated annulus in a Seifert fibered space that
cuts off an exceptionally fibered solid torus to one side and a Seifert fibered
space that is not fibered over a disk or that has at least two exceptional
fibers to the other side is incompressible.
Exercise 7. What hypotheses guarantee that a saturated annulus or torus
is incompressible?
3 . 8 . JSJ Decompositions

JSJ decompositions provide a geometric decomposition of 3-manifolds. The

acronym stands for Jaco, Shalen, and Johannson, whose work served to
define and prove the existence and uniqueness of JSJ decompositions. See
[71] and [72] .
Definition 3.8. 1 . A 3-manifold is atoroidal if it contains no essential torus
submanifold.
Theorem 3.8.2. A closed orientable irreducible 3-manifold M contains a
collection of disjointly embedded incompressible tori 7 such that each com­
ponent of the 3-manifold obtained by cutting M along 7 is either atoroidal
or Seifert fibered. Moreover, a minimal such collection of tori is unique up
to isotopy.
A similar theorem holds for non-orientable 3-manifolds so long as one
also includes Klein bottles in the collection 7.
Definition 3.8.3. The collection of tori 7 as above is called the JSJ de­
composition of M.
The diagram in Figure 3.39 describes a large class of 3-manifolds with
similar JSJ decompositions. Each edge corresponds to a torus, each vertex to
either a Seifert fibered space or an atoroidal 3-manifold. For instance, vertex
1 might correspond to an atoroidal 3-manifold with two torus boundary
components. Vertices 2 and 3 might correspond to Seifert fibered spaces
with at least three torus boundary components. A pair of the boundary
components of each of these two Seifert fibered spaces is identified; another
boundary component is identified to a boundary component of the atoroidal
3-manifold corresponding to vertex 1.
3.8. JSJ Decompositions 97

2
1 ______

Figure 3.39. Schematic for a JSJ decomposition.

Definition 3.8.4. If the result of cutting M along 7 yields only Seifert

fibered spaces, then M is called a graph manifold.

Figure 3.40. A JSJ decomposition of a graph manifold.

Figure 3.40 can be interpreted as a graph manifold. For instance, imag­

ine two Seifert fibered spaces, Mi , M2 , each with base space a disk and two
exceptional fibers glued along their boundaries. The great circle indicated
corresponds to a saturated torus. If the gluing of Mi and M2 is performed
in such a way that fibers do not match up with fibers, even up to isotopy,
then the great circle indicated corresponds to the unique torus in the JSJ
decomposition. If the gluing is such that fibers do match up with fibers,
then the result is a Seifert fibered space with base space the sphere and four
exceptional fibers; hence the JSJ decomposition is empty.
Theorem 3.8.2 can be rephrased as follows: "A closed orientable irre­
ducible 3-manifold M has a Seifert fibered submanifold with atoroidal com­
plement. The Seifert fibered submanifold with the least possible number of
boundary tori is unique up to isotopy."
98 3. 3-Manifolds

Definition 3.8.5. The Seifert fibered submanifold of a closed 3-manifold

M that has the least possible number of boundary tori is called the charac­
teristic submanifold of M.
Consider the graph manifolds described above. In the first case, the
characteristic submanifold has two components: ( shrunk versions of) Mi
and M2 . In the second case, the characteristic submanifold is the entire
manifold.
Exercises

Exercise 1. Construct a graph manifold with three tori in its JSJ decom­
position.
Exercise 2. By considering the 3-torus, show that the minimality assump­
tion in Theorem 3.8.2 is necessary.
3.9. Compendium of Standard Arguments

There are several constructive arguments that have become standard in low­
dimensional topology. We have used some of these in this chapter.
Example 3.9.1. Standard innermost disk argument: Building on Lemma
3.2.3, one can isotope incompressible surfaces in an irreducible 3-manifold so
that their intersection contains only simple closed curves that are essential
in the two incompressible surfaces. Indeed, suppose there are simple closed
curves in the intersection of two incompressible surfaces, Si and S2 , in the
3-manifold M, that are inessential in, say, Si . Let c be an inessential simple
closed curve in Si n S2 that is innermost in Si . Then at least one component
of Si \c is a disk D that is disjoint from S2 . Since c bounds a disk in M, it
also bounds a disk D' in S2 . Since M is irreducible, the 2-sphere D U D' U c
bounds a 3-ball in M. Thus by Lemma 3.2.3 there is an isotopy reducing
the number of components of Si n S2 . See Figure 3.41.
In particular, if one of the incompressible surfaces is a sphere or disk,
then a succession of such isotopies eliminates all closed curves of intersec­
tion. We used this argument in the proofs of Theorem 3.2.5, Theorem 3.2.7,
Lemma 3.7. 14, and Theorem 3.7. 17.
Lemma 3.9.2. Let Si , S2 be surfaces in a 3-manifold M. Suppose that
Si n S2 contains a simple arc a such that one component of Si \a is a disk
Di such that D2 is disjoint from Si . If Di U D2 bounds a 3-ball in M, then:
• (Si \Di ) U D2 is isotopic to Si ;
• there is an isotopy of 8i that eliminates the curve c from 8i n 82
and introduces no new components of intersection.
3. 9. Compendium of Standard Arguments 99

innermost disks

Figure 3.41. Innermost disks.

outermost arcs

Figure 3.42. Outermost arcs.

The proof of this lemma is analogous to the proof of Lemma 3.2.3.

Example 3.9.3. Standard outermost arc argument: Building on Lemma
3.9.2, one can isotope a pair of incompressible and boundary incompressible
surfaces in an irreducible 3-manifold so that their intersection contains no
inessential arcs. We did not use the standard outermost arc argument in
this pure form.
Occasionally, this argument can be applied under weaker hypotheses to
eliminate arcs of a specified type or to describe isotopies. We did not use this
weaker form of the standard outermost arc argument in this chapter. Echos
of the standard outermost arc argument can be seen for instance toward the
end of the proof of Lemma 3.7. 14.
Example 3.9.4. Cut and paste arguments: These are not as standard as
the other two types of arguments. Recall the usage of the word cut in
Definition 3.2.4. Likewise, paste refers to taking a union of two manifolds
along their boundary or frontier. Specifically, we create an identification
space from two spaces X, Y by identifying pairs of points x = [x, y] , where
x, y lie in specified subsets of X, Y. We also say that we glue X and Y along
100 3. 3-Manifolds

the specified subsets, especially if there is a function f : X -+ Y such that

y = f(x), when we say that we glue X to Y via f.
The ideas of cutting and pasting are implicit in constructions such as,
e.g. , (81 \D 1 ) U D 2 in Lemmas 3.2.3 and 3.9.2. Standard innermost disk
and outermost arc arguments are collectively referred to as cut and paste
arguments. Cut and paste arguments can involve the Euler characteristic.
Here it is important to note that, for instance, the replacement of 81 by
(81 \D 1 ) U D2 in Lemmas 3.2.3 and 3.9.2 has no impact on the Euler char­
acteristic of the surface under consideration.
 Chapter 4

Knots and Links

in 3- Manifolds

Knots provide an excellent starting point for the understanding and appreci­
ation of 3-manifolds. My generation of 3-manifold theorists found challenge
and joy in working through Rolfsen's inspired and inspiring book [127] .
Many of the standard constructions used in the topological study of knots
and presented in Rolfsen's book rely on underlying general principles per­
taining to 3-manifolds. They also represent key examples illustrating these
principles. For this reason we include a brief introduction to the beautiful
world of knots in 3-manifolds.
For a gentle introduction to knot theory, see [1] or [89] . For a more
rigorous treatment, see [19] or [87] .

4. 1 . Knots and Links

In general, a knot is an equivalence class of submanifolds. Our definitions

(of submanifold, simple closed curve, knot, and link) rule out what are called
"wild knots" . For example, see Figure 4.1.

Figure 4. 1 . A wild knot.

-

101
102 4. Knots and Links in 3-Manifolds

Definition 4.1.1. A knot in §3 is a smooth isotopy class of smooth embed­

dings of § 1 into §3 . We write K to denote such an equivalence class.
We think of a representative of a knot as a simple closed curve k in §3 .
Two simple closed curves k, k' represent the same knot, i.e., are equivalent,
if there is an isotopy of pairs between (§3 , k) and (§3 , k') . Note that in
Definition 4. 1 . 1 we are not considering orientations on § 1 or k . In particular,
k and -k represent the same knot.
Definition 4.1.2. An n-component link in §3 is a smooth isotopy class of
smooth embeddings of the union of n copies of § 1 into §3 . We write L to
denote such an equivalence class.
See the examples in Figures 4.2, 4.3, 4.4, and 4.5.
More generally, we consider smooth isotopy classes of simple closed
curves in a given 3-manifold. Indeed, we will do so in the applications in later
chapters of this book. However, for the purposes of this chapter, we restrict
our attention to knots and links in §3 . Moreover, since JR3 = §3 \{point},
this is equivalent to considering knots in IR3 .

Figure 4.2. The unknot.

Definition 4. 1.3. The unknot is the isotopy class that contains the unknot­
= {(x, y, z ) I x 2 + y 2 = 1, z = O} c JR 3 .
ted circle § 1
Henceforward, we will blur the distinction between a knot and its rep­
resentatives.
In depicting knots, we consider projections of knots in JR3 onto a plane.
An application of transversality guarantees that in this projection we need
only worry about two points projecting to the same point (rather than three
or more) and that there will be only a finite number of such double points.
At such double points, our convention is for instance to draw the arc con­
taining the points above and closer to the plane as a broken line and the
arc containing the points above and further away from the plane as a solid
arc. See Figures 4.3 and 4.4. In this fashion, we capture all information
about the knot in the planar projection. In other words, the projection can
4. 1 . Knots and Links 103

Figure 4.3. The trefoil.

Figure 4.4. The figure 8 knot.

Figure 4.5. The Whitehead link.

be interpreted as a depiction of the knot in §3 in which all but a few short

arcs lie in the plane. We call this type of projection a knot diagram.
Since a knot is an isotopy class and due to the fact that there are many
possible projections of a knot in §3 , diagrams of knots that look different
might very well depict the same knot. Indeed, to create very complicated
diagrams of the unknot, proceed as follows: Take a pencil and start drawing
a curve. When you return to the curve, pass under the curve drawn so far to
create an undercrossing in ·the projection. Continue in this fashion, passing
under the (broken) curve each time you return. You can pass under the
(broken) curve any number of times and wind your way around the page
104 4. Knots and Links in 3-Manifolds

any number of times. Then return to the starting point to close up and
create a knot projection. See Figure 4.6.

Figure 4 . 6 . A complicated diagram of the unknot.

To see that the above description yields the unknot, imagine that your
pen is moving farther and farther back as you draw. The process is consistent
with this interpretation. At the very end, you add a short arc from way far
back to the point where you started that is perpendicular to the planar
projection you are considering. Now imagine pulling the former part of the
knot tight, so that it becomes parallel to the latter arc. Then you see that
the closed curve drawn is the unknot.
The two main questions in knot theory are the following:
Question 4. 1 .4. How do you tell whether or not two knot diagrams repre­
sent the same knot?
Question 4.1.5. How do you generate a list of all possible knots?
All knots are homeomorphic. They are all circles. What is important
is the precise way that a circle sits in §3 . Note that if instead we were to
consider smooth isotopy classes of smooth embeddings of § 1 into §2 , then
our task would be rather easy and unrewarding. In this case the Jordan
Curve Theorem tells us that for 1-dimensional knots in §2 there is only one
knot, namely the unknot. Likewise, if we were to consider smooth isotopy
classes of smooth embeddings of §2 into §3 , then our task would also be
rather easy and unrewarding. In this case the Schonfiies Theorem tells us
that for 2-dimensional knots in §3 there is only one possibility, namely the
standard 2-sphere sitting in §3 .
Interestingly enough, the same phenomenon occurs for different reasons
if we consider smooth isotopy classes of smooth embeddings of § 1 into §4 .
Consider for instance the diagram of the trefoil in Figure 4.3. This dia­
gram represents a knot that sits in a 3-dimensional subspace (in fact, in a
small 3-dimensional neighborhood of a 2-dimensional subspace) of §4 • At a
crossing, imagine pushing the subarc of the undercrossing off into the fourth
4. 1 . Knots and Links 105

dimension. Then in a parallel 3-dimensional subspace we isotope the arc into

an overcrossing arc. Now we isotope it back into our original 3-dimensional
subspace. This way we can switch crossings so that an overcrossing becomes
an undercrossing and vice versa.
By changing overcrossings to undercrossings we can turn any given knot
diagram into one obtained by the above prescription for drawing a diagram
of the unknot. In other words, the diagram is changed into one representing
the unknot. This means that any smooth embedding of § 1 into §4 represents
the unknot. Thus for !-dimensional knots in §4 there is also only one knot,
again the unknot.
Finally, we introduce an infinite but fairly well-understood class of knots.
These knots live on the torus and for this reason are called torus knots.

Figure 4.7. The (3, 2)-torus knot.

Definition 4.1.6. Recall the definition of the n-torus in Example 1 . 1 .5. In

IR2 , consider the line y = �x for p, q E Z and (p, q) = 1 (i.e. , p and q are
relatively prime) . The covering map 7r : IR 2 � 'll'2 maps this line to a simple
closed curve on the torus. We call this simple closed curve a torus knot,
more specifically, the (p, q) -torus knot. See Figure 4.7.
The 2-torus can be embedded in §3 in such a way that it bounds a solid
torus on either side. This embedding is called the unknotted solid torus.
Given a torus knot, it then determines a knot in §3 that is also called a
torus knot and more specifically the (p, q) -torus knot.
In the context of 3-manifolds we are often interested in the knot com­
plement C(K) = §3 ry(K) , where ry(K) is an open regular neighborhood
-

of K . One of the most fundamental theorems in the subject of knot theory

(and one that was rather difficult to prove) is the following (see [46]) :
Theorem 4.1.7 (Gordon-Luecke) . Knots are determined by their comple­
ment. (I. e., if C(K) and C(K') are homeomorphic, then K and K' are
isotopic.)
106 4. Knots and Links in 3-Manifolds

The boundary of C(K) is a single torus. It is the boundary of a closed

regular neighborhood of the knot. A simple closed curve on 8C(K) that
bounds a disk in 17(K) is called a meridian. Any simple closed curve that
intersects the meridian once is called a longitude. All meridians are isotopic.
All longitudes are not isotopic. In the case of the unknot the preferred
longitude is the curve that bounds a disk in the complement C(K) of the
unknotted solid torus 17(K) . More generally, the preferred longitude is the
curve that bounds a compact connected orientable surface with a single
boundary component. Such a surface always exists and is called a Seifert
surface. We will define and discuss these in Section 4.3.
Note that the meridian and the preferred longitude for the unknot, (m, l) ,
provide a coordinate system for torus knots: Specifically, given a represen­
tative k of the (p, q)-torus knot, p [m] [k] and q [l] · [k] .
= · =

Exercises
Exercise 1 . Construct a knot out of a piece of string and experiment by
moving it around in space, tugging first on one portion and then another.
Exercise 2. Draw several diagrams based on projections of the knot in
Exercise 1.
Exercise 3. Contemplate criteria to distinguish the unknot from the trefoil.
Exercise 4. Exhibit the trefoil as a torus knot by finding appropriate p
and q.
Exercise 5. Prove that the complement of a torus knot is a Seifert fibered
space.

4.2. Reidemeister Moves

Manipulating knot diagrams can be a daunting task. It requires some

amount of visual acuity. The work of Reidemeister and independently
Alexander and Briggs in the 1920s streamlined this task significantly. The
idea was to formalize the procedure of showing that two diagrams represent
the same knot. This was achieved by isolating three moves. These are now
called the Reidemeister moves. See Figures 4.8, 4.9, and 4. 10.
These moves represent isotopies that take place in a specified region of
the knot diagram. Inside the region the diagram looks as pictured on the
left-hand side (or the right-hand side, respectively) of one of the Figures 4.8,
4.9, or 4. 10) . After the move, the inside of the region looks as pictured on
the right-hand side (or the left-hand side, respectively) of the same figure.
Everything outside of the specified region remains the same.
4.2. Reidemeister Moves 107

Figure 4.8. Type I Reidemeister move.

l -)(
Figure 4.9. Type II Reidemeister move.

Figure 4.10. Type III Reidemeister move.

Definition 4.2. 1. A planar isotopy of a knot diagram is an isotopy of the

arcs in the knot diagram that does not affect the crossings.
What makes the Reidemeister moves so effective is the following theo­
rem:
Theorem 4.2.2 ( Reidemeister ) . Two knot diagrams represent the same
knot if and only if one diagram can be transformed into the other by a
sequence of Reidemeister moves and planar isotopies.
In many areas of knot theory the Reidemeister moves provide the tool to
prove that a property of knots or a quantity is well-defined. This is especially
true for many of the modern knot invariants. We will not pursue this here
but will provide one illustration of how to use the Reidemeister moves in
this way.
Definition 4.2.3. A knot diagram is 3-colorable if the arcs in the diagram
can be colored in such a way that the following hold: (1) Each arc in the
diagram is colored with a single color; (2) at each crossing, either all arcs
meeting at that crossing are colored with the same color or all three colors
are used; (3) all three colors are used in the diagram.
Theorem 4.2.4. The property of being 3-colorable is well-defined for knots.
Proof. We need to show that if one diagram of a knot is 3-colorable, then
so is any other diagram. So suppose a diagram of the knot is 3-colorable.
108 4. Knots and Links in 3-Manifolds

Any other diagram of the knot is obtained from this diagram by a sequence
of Reidemeister moves and planar isotopies. Now observe that if a diagram
is 3-colorable, then it is also 3-colorable after a Type I Reidemeister move
because only one color can occur on a strand and also at the crossing of the
loop. The same is true after a Type II Reidemeister move. Indeed, either
all strands are the same color before and after the Reidemeister move or all
three colors occur in the strands on the left-hand side and only two colors
occur on the right-hand side of Figure 4.9. If this is the case, then since two
colors occur on the right-hand side and since the knot is connected, there
must be a crossing involving all three colors. The situation for a Type III
move is even simpler. 0

The property of being 3-colorable provides a means of distinguishing

between knots based on their diagrams. If we compare two knot diagrams
and find that one is 3-colorable whereas the other is not, then we know that
the two diagrams represent distinct knots.
Exercises

Exercise 1. Find a sequence of Reidemeister moves that transforms the

diagram of the unknot pictured in Figure 4.6 into the round unknot.
Exercise 2. Prove that the unknot and the trefoil are distinct using Theo­
rem 4.2.4.
4.3. Basic Constructions

Lay people confronted with knot theory often ask what knots could possibly
have to do with mathematics, since they equate mathematics with algebra.
Their peace of mind can be restored by an appeal to the semigroup properties
enjoyed by knots. Indeed, there is an addition operation for knots. In the
definition of this addition operation we will think of a knot as an oriented
I-dimensional submanifold of the oriented 3-manifold §3 , with orientation
inherited from the embedding of § i with the standard orientation.
Definition 4.3. 1 . Let (§3 , Ki ) , (§3 , K2 ) be oriented knots. Their connected
sum is the oriented pairwise connected sum (§ 3 , Ki ) # (§3 , K2 ) .

Remark 4.3.2. Abusing notation slightly, we refer to our knots as Ki and

K2 and simply write Ki #K2 .
Figure 4. 1 1 depicts the connected sum of the trefoil (see Figure 4.3) and
the figure 8 knot (see Figure 4.4) . In the oriented pairwise connected sum
we remove from each pair a 3-ball of §3 containing a subarc of Ki · We then
identify the resulting 2-sphere boundary components in such a way that the
4.3. Basic Constructions 109

Figure 4.11. The connected sum of the figure 8 knot and the trefoil.

endpoints of the deleted arcs are identified. A priori there are two choices,
up to isotopy, for this identification, but we decree that the orientations
match up. In particular, the endpoints of the arcs must be identified in such
a way that the induced orientations on the subarcs match up.
In the unoriented category this operation is not well-defined. See Figures
4.12 and 4. 13. Moreover, this addition operation is not well-defined for links.

Figure 4.12. The square knot.

Figure 4.13. The granny knot.

The unknot acts as the identity in the semigroup defined by this addition
operation. Note that the operation is commutative. There are no inverses
for this operation. This is a non-trivial fact that we leave as an exercise. The
definitions below and Theorem 4.3.6 provide tools that suffice to establish
this fact.
1 10 4. Knots and Links in 3-Manifolds

Definition 4.3.3. A Seifert surface of a knot K is a compact connected

orientable surface S in §3 such that 88 = K .

Figure 4. 14. A Seifert surface for the trefoil knot.

Theorem 4.3.4. Every knot possesses a Seifert surface.

Proof (Seifert's Algorithm) .
Step 1. Give the knot an orientation.
Step 2. Examine a diagram of the knot. At each crossing, alter the diagram
as in Figure 4.15.

x -

Figure 4.15. Replacement at each crossing.

The result is a collection of disjoint oriented simple closed curves.

Step 3. Consider disks in IR2 bounded by the simple closed curves. Orient
these disks with a + or a - depending on whether their boundaries are
oriented counterclockwise or clockwise, respectively. If curves are nested,
then the disks can be made disjoint by raising them out of the plane of the
diagram with the innermost one raised to be highest.
Step 4. Connect the disks at the old crossings via half-twisted strips as in
Figure 4.16.
4.3. Basic Constructions 111

'
'

Figure 4.16. Twisted band for each crossing.

,'�'-, '
, ,, '
'

,, '
'

,, '

�
'

/ +

�
. '
I

'
I I
.
I
' I

' , ,,
+ , '

... ... _ _ _ _ _

Figure 4.1 7. Twisted band for each crossing.

If a twisted band connects non-nested disks, then one disk is oriented

with a + and the other with a -. So attaching the twisted band is consistent
with the orientations of the disks. See Figure 4.16.
If a twisted band connects nested disks, then either both disks are ori­
ented with a + or both are oriented with a -. Here too, attaching the
twisted band is consistent with the orientations of the disks. See Figure
4. 17. (Note that the band lies above the lower disk and the - side of the
band connects to the underside of the lower disk.) 0
1 12 4. Knots and Links in 3-Manifolds

Seifert surfaces capture some of the quality of the knottedness exhibited

by a knot. They also provide an important tool in many investigations of
knot complements. In some cases, they help us distinguish between knots
by means of the following definition:
Definition 4.3.5. The genus of a knot K, denoted by g(K) , is the smallest
possible genus of a Seifert surface of K.
Theorem 4.3.6. Let Ki , K2 be knots. Then g(Ki#K2 ) = g(Ki ) + g(K2 ) .
Proof. To prove this equality, we prove two inequalities. First we prove the
following inequality:

Consider minimal genus Seifert surfaces Si and S2 for Ki and K2 , respec­

tively, and take the connected sum of triples for (Ki , Si , §3 ) and (K2 , S2 , §3 ) .
In doing so we remove from Si a small disk with boundary partitioned into
an arc in 8Si and an arc in the interior of Si and glue the surfaces along the
two interior arcs. As the result, we obtain a Seifert surface of Ki #K2 of
genus g (Ki ) + g(K2 ) . This gives an upper bound for the genus of Ki #K2 .
Now we prove the other inequality:
g (Ki#K2 ) 2 g(Ki ) + g(K2 ) .
Consider a minimal genus Seifert surface S for Ki #K2 . We wish to
use S to construct Seifert surfaces for Ki and K2 . As we will see, this is
achieved by the cut and paste techniques typical of 3-manifold topology.
It follows from the definition of connected sum of knots that there is a 2-
sphere Z in §3 that separates Ki #K2 into its two summands. In particular,
the knot intersects Z in exactly two points. The intersection of S with Z,
after an isotopy making S and Z transverse, consists of simple closed curves
and simple arcs. Moreover, since the endpoints of arcs correspond to points
of intersection with the knot, there must be exactly one arc of intersection.
The complement of this arc of intersection in Z is a disk D containing only
closed components of intersection.
Suppose first that there are no closed components of intersection. Then
the arc of intersection cuts S into two components, one, Si , a Seifert surface
for Ki , the other, S2 , a Seifert surface for K2 . Furthermore, g(Si ) + g(S2 ) =
g(S) , so the above inequality holds.
You will prove in the exercises that the complement of a knot in §3 is
irreducible and that a minimal genus Seifert surface is incompressible. Thus
closed components of S n Z can be removed via a standard innermost disk
argument. Now the above construction yields the desired inequality. 0
4.4. Knot Invariants 113

Exercises
Exercise 1 *. Prove that K is the unknot if and only if g (K) = 0. (It is
not hard to see that if K is the unknot, then g (K) = 0. The converse is
challenging. Hints: If g(K) = 0, consider a regular neighborhood N(K) and
the Seifert surface of K that is a disk D. Use these to describe §3 as a lens
space. Use the fact that §3 has trivial fundamental group to show that K
must be the unknot.)
Exercise 2. Prove that if Ki and K2 are non-trivial knots, then Ki #K2
can never be the unknot. Deduce that there are no inverses for the operation
of connected sum. (Hint: Use Exercise 1 and Theorem 4.3.6.)
Exercise 3. Prove that every knot has a prime decomposition. (This prime
decomposition is unique up to reordering of summands, but this is harder
to prove.)
Exercise 4. Construct Seifert surfaces that show that the genus of the
(p, q)-torus knot is at most � (p - l) (q - 1) . (In fact, the genus of the (p, q)­
torus knot is � (p - l) (q - 1) , though this is harder to prove.)
Exercise 5. Prove that a minimal genus Seifert surface of a knot in §3 is
incompressible.
Exercise 6. Prove that the complement of a knot in §3 is irreducible.

4.4. Knot Invariants

The genus of a knot is an example of a knot invariant. More generally, a
knot invariant is any quantity that is well-defined on the isotopy class of the
knot. The property of 3-colorability is another example of a knot invariant.
In recent decades, the study of knot invariants has taken on a very different
flavor than the study of 3-manifolds. Many beautiful algebraic insights have
been obtained, for instance, in the study of topological quantum field theory
and more recently in the study of Heegaard Floer homology. Discussing these
results would take us too far afield from our goals here. Suffice it to say that
recent years have witnessed a resurgence by the more algebraically inclined
to understand the topological implications of these insights and to interpret
such insights for a more topologically inclined audience.
We wish to discuss some of the more classical knot invariants. We have
already seen the property of 3-colorability and the genus of a knot. The
following definition gives us a very natural invariant that relies heavily on
1 14 4. Knots and Links in 3-Manifolds

the representation of knots via diagrams:

Definition 4.4. 1 . The crossing number of a knot K, denoted by c(K) , is
the least number of crossings in a diagram of the knot.
Example 4.4.2. The crossing number of the trefoil is 3.

To prove that the crossing number of the trefoil is indeed 3, we must

prove two things: (1) that the crossing number of the trefoil is at most 3;
(2) that the crossing number of the trefoil is no smaller than 3.
To prove the first statement, we simply observe that the diagrams of the
trefoil exhibited in Figures 4.3 and 4. 14 both have crossing number 3. So
the crossing number of the trefoil is at most 3.
Proving the second statement requires more work. Why can't the cross­
ing number of the trefoil be 0, 1 , or 2? Fortunately, we can rule out these
options one by one, though each requires a small amount of work. The
crossing number of the trefoil cannot be 0 because a diagram with crossing
number 0 necessarily depicts the unknot (this is the Jordan Curve Theo­
rem) . To show that it can't be 1 , draw one crossing and try to connect up
the endpoints of the arcs without introducing any other crossings. In all
cases we obtain an unknot that can be transformed into the round unknot
after one Type I Reidemeister move and planar isotopies. An analogous
analysis shows that the crossing number of a non-trivial knot cannot be 2.
More generally, Murasugi proved the following (see [110] ) :
Theorem 4.4.3 (Murasugi) . The crossing number of the (p, q) -torus knot
is
min{p(q - 1 ) , q(p - 1 ) } .
Knot tables employ the crossing number as an ordering scheme. Knots
with crossing number 3 are listed first, then knots with crossing number 4,
then with crossing number 5, etc. Incidentally, there is only one knot, the
trefoil, with crossing number 3 and only one knot, the figure 8 knot (see
Figure 4.4) , with crossing number 4. We leave the proofs of these facts as
exercises.
The number of knots with a given crossing number is of interest. The
sequence starts with the number of knots of crossing number 3 and continues
on up. The first few terms are 1 , 1 , 2, 3, 7, 21, 49, 165, 552, 2176, 9988,
46972, . . . . This explosive growth shows that for some purposes, the crossing
number is not a good ordering scheme.
Some interesting facts are known about the crossing number. We give
one example below. But .many natural questions remain.
4.4. Knot Invariants 115

Definition 4.4.4. A knot diagram is alternating if the over or under nature

of the crossings alternates as one travels along the knot. A knot is alternating
if it has an alternating diagram.
Definition 4.4.5. A nugatory crossing in a knot diagram is a crossing such
that there is a circle that intersects this crossing and is disjoint from the
rest of the knot diagram. See Figure 4.18.

'

'

I
'
'
'
'

'
,

- - -

Figure 4.18. A nugatory crossing.

Note that a nugatory crossing can be removed via an isotopy (in §3 and
hence also by a series of Reidemeister moves) .
Definition 4.4.6. A knot diagram is reduced if it contains no nugatory
crossings.
A major result concerning crossing number was obtained independently
by Kauffman, Murasugi, and Thistlethwaite:
Theorem 4.4.7 (Kauffman, Murasugi, and Thistlethwaite) . A reduced al­
ternating diagram of K realizes the minimal crossing number of K.
Many questions concerning the crossing number remain. For instance,
it is unknown how the crossing number behaves under the operation of
connected sum. It is not too hard to show that the crossing number of a
composite knot is at most the sum of the crossing numbers of its summands.
We leave the proof of this fact as an exercise. Consider the implications of
this fact in light of the number of knots of a given crossing number. The
number of knots with a given crossing number grows explosively. One upshot
is the interesting fact that the number of prime knots with a given crossing
number is far greater than the number of composite knots with that crossing
number. On the other hand, from the probabilistic point of view pursued,
for instance, in [36] , it follows, by [35] , that a generic knot is composite,
more specifically, has a trefoil summand.
116 4. Knots and Links in 3-Manifolds

Conjecture 4.4.8. Crossing number is additive under connected sum of

knots.
Another classical knot invariant grows out of the following: Consider a
diagram of K and change one of the crossings either from an overcrossing to
an undercrossing or vice versa. This yields a knot K1 , typically distinct from
K. Now consider a diagram of Ki (not necessarily related to the diagram
of K) and continue the process. For every knot, this process can be chosen
in such a way that the process terminates with the unknot. To see this,
recall our construction of complicated diagrams for the unknot. Using this
idea, any knot diagram can be transformed into a diagram of the unknot
by a finite number of crossing changes. Specifically, if the diagram has n
crossings, we need to change at most � crossings.
Definition 4.4.9. The unknotting number of a knot is the minimal number
n such that Kn (in the above process) is the unknot. (The unknotting
number is also called the Gordian number.) We denote the unknotting
number of a knot K by u(K) .

Figure 4.19. A crossing change on a diagram of the figure 8 knot yield­

ing the unknot.

The discussion preceding the definition ensures that the unknotting num­
ber is well-defined. It also proves the following theorem:
Theorem 4.4. 10. For any knot K, u(K) :S ! c(K) .
The unknotting number appears to be related to 4-dimensional phenom­
ena. Consider the process of unknotting a knot K in §3 . Over time, the
knot sweeps out a surface in §3 x I C llll4 • The questions and techniques
studied in 4-dimensional topology differ greatly from those in 3-manifold
topology. For instance, gauge theory has been an effective tool in the study
of 4-manifolds. Using gauge theory, Kronheimer and Mrowka proved the
following theorem (see [82] ) :
Theorem 4.4. 1 1 (Kronheimer-Mrowka) . The (p, q) -torus knot has unknot­
ting number ! (P - l ) (q - 1) .
4.4. Knot Invariants 117

Proving that the unknotting number of the (p, q)-torus knot is at most
� (p - 1) ( q - 1) is comparatively easy. One need only consider an appropriate
diagram of the (p, q)-torus knot. Kronheimer and Mrowka's co:ntribution lay
in proving the reverse inequality.
Despite its intuitive appeal, understanding the unknotting number re­
mains tantalizingly out of reach. The big open question parallels the above
question on crossing number.
Question 4.4. 12. How does the unknotting number behave under the con­
nected sum of knots?
Some amount of progress on this question was made by Scharlemann,
who proved the following theorem:
Theorem 4.4. 13 (Scharlemann) . Unknotting number 1 knots are prime.

Yet another invariant for links is the linking number. To compute it,
we must first orient the components of our link. Once each component is
oriented, we can distinguish between positive and negative crossings. See
Figure 4.20.

x +
x
Figure 4.20. A positive and a negative crossing.

Now given a link, we can consider one-half of the difference of the number
of positive and negative crossings.
Definition 4.4. 14. The linking number between two components L 1 , £2 of
a link is one-half of the number of positive crossings minus the number of
negative crossings. It is denoted by lk(L 1 , L 2 ) .
The linking number of the two components of the Hopf link is ±1 de­
pending on the orientations chosen. See Figure 4.21.
Recent years have seen rapid progress in understanding invariants of
knots and 3-manifolds growing out of homology theories. These subjects go
beyond the scope of this book. The reader is invited to consult [76] , [116] ,
[117] , and [118] .
1 18 4. Knots and Links in 3-Manifolds

Figure 4.21. The Hopf link.

Exercises
Exercise 1. Show that the only knot with crossing number 3 is the trefoil.
Exercise 2. Show that the only knot with crossing number 4 is the figure
8 knot.
Exercise 3. Show that the crossing number of a composite knot is at most
the sum of the crossing numbers of its summands.
Exercise 4. Show that the crossing number of the (p, q)-torus knot is at
most p(q - 1).
Exercise 5. Show that the crossing number of an unknotting number 1
knot can be arbitrarily large.
Exercise 6. Show that the unknotting number of the (p, q)-torus knot is
at most ! (P l ) (q 1).
- -

Exercise 7 . Show that the unknotting number of a composite knot is at

most the sum of the unknotting numbers of its summands.
Exercise 8. Prove Conjecture 4.4.8 for alternating knots. (Hint: A litera­
ture search is appropriate.)

4 . 5 . Zoology
We mentioned above that the crossing number can be an ineffective tool for
ordering knots due to the explosive growth of the number of knots with a
given crossing number. Instead, we often focus on certain classes or types of
knots. We have already discussed one such class of knots, namely the torus
knots. This is an infinite family of knots. Not only is this class of knots
infinite, but it also includes knots of arbitrarily high crossing number and
4. 5 . Zoology 1 19

of arbitrarily high unknotting number. In this section we introduce a few

more rich classes of knots.
A pretzel knot is a knot modeled on Figure 4.22. Here the boxes con­
taining the letters p, q, r represent that number of half-twists of the appro­
priate strands of the knot. In Figure 4.23 we illustrate, as an example, the
(-2, 3, 7)-pretzel knot.

p q r

Figure 4.22. Model for the (p, q, r )-pretzel knot.

Figure 4.23. The ( - 2 , 3, 7)-pretzel knot.

Pretzel knots are a subclass of a larger class called Montesinos knots.

Montesinos knots, in turn, are a subclass of the class of arborescent knots.
Arborescent knots have been classified by Bonahon and Siebenmann. The
class of arborescent knots was also studied in detail by Gabai. (See [42] .)
Among other things, he found that for knots in this class the genus is the
genus realized by a Seifert surface obtained from a "standard" diagram; in
the case of a pretzel knot this would be Figure 4.23, via Seifert's algorithm.
A 2-bridge knot is a knot that can be isotoped to have exactly two
maxima with respect to a height function on §3 . As it turns out, every
2-bridge knot is modeled on Figure 4.24. A box labeled ti represents ti half­
twists of the appropriate strands of the knot. The trefoil and the figure 8
knot are examples of 2-bridge knots. 2-bridge knots also belong to the larger
120 4. Knots and Links in 3-Manifolds

t'

'

u
Figure 4.24. Model for 2-bridge knots.

class of arborescent knots. The repeated fraction expansion of a 2-bridge

knot diagram as in Figure 4.24 is the rational number

p 1
q

The repeated fraction expansion of a rational number � is not unique, but

as it turns out, 2-bridge knot diagrams correspond to the same knot if and
only if they give repeated fraction expansions of the same rational number
� · (In Conway's notation, the 2-bridge knot is simply the � -rational knot.
See [28] .) We will occasionally refer to 2-bridge knots in the chapters that
follow. In certain settings 2-bridge knots and 2-bridge knot complements
provide interesting examples of pathological behavior. In Section 4.8 below,
we will discuss bridge numbers of knots, another knot invariant.
Given two knots, the operation of connected sum produces a new knot
from the two given knots. In fact, the operation of connected sum is a special
4. 5 . Zoology 121

case of a more general binary operation:

Definition 4.5.1 (Satellite construction) . Let A be a non-trivial knot in
§3 and let V be a closed regular neighborhood of A. Furthermore, let B
be a knot in the solid torus V. The pair (V, B) is called the pattern. The
minimal number of times that B intersects a meridian disk of V is called
the index or wrapping number. Suppose that the index of B is at least 1 .
We denote the image of B under a homeomorphism of V into V by K . Here
K is called a satellite knot. The knot A is called a companion of K. The
boundary of V is called a companion torus.

Figure 4.25. A satellite knot and its companion torus.

The satellite construction is interesting for many reasons. One such

reason is that the boundary of V is an essential torus in the complement of
the satellite knot.
To reconstruct the connected sum of Ki and K2 from the satellite con­
struction, set A = Ki and choose a solid torus containing K2 with wrapping
number 1 as the pattern. Then the resulting satellite knot will be Ki #K2 .
In this case the resulting essential torus is called a swallow-follow torus. (It
"swallows" K2 and "follows" Ki .) The roles of Ki and K2 can be reversed.
This yields a different swallow-follow torus. (This one "swallows" Ki and
"follows" K2 .)
Remark 4.5.2. A knot can have many companion knots and companion
tori. In the 1950s, Horst Schubert investigated the finiteness and relative
positioning of companion tori; see [141] . In essence, he proved the existence
and uniqueness of JSJ decompositions for knot complements.
122 4. Knots and Links in 3-Manifolds

Exercises
Exercise 1. Show that the plane in Figure 4.26 separates the figure 8 knot
into trivial arcs. Specifically, draw a pair of disks above the plane and a pair
of disks below the plane such that the pair exhibits an isotopy of the two
arcs into the plane.

Figure 4.26. A plane separating the figure 8 knot into trivial arcs.

Exercise 2. Draw a swallow-follow torus for the connected sum of the

trefoil and the figure 8 knot.
Exercise 3. Suppose that two non-isotopic swallow-follow tori have been
isotoped to intersect in a minimal number of curves. What can you say
about their intersection?

4.6. Braids

We now briefly discuss braids. For the purposes here, braids will merely be a
special positioning for knots and links. But in fact, braids represent a point
of view. Many fascinating results about knots and links have been obtained
by using the structure given by the representation of knots and links as
braids. Moreover, braids naturally give rise to a group. For this reason,
braids have been studied not only by topologists, but also by algebraists
and even mathematical physicists. For more on braids and related topics,
see [10] .
Definition 4.6.1. An n-string braid is a collection of disjoint arcs in a box
that monotonically connect the top to the bottom. See Figure 4.27. Two
braids are considered equivalent if they are isotopic in the box relative to
their endpoints (i.e. , via an isotopy that fixes the endpoints) .
4. 6. Braids 123

Figure 4.27. A braid.

Definition 4.6.2. Denote by CTi the braid with a single positive crossing
such that the i-th strand crosses over the (i + 1)-th strand.

Figure 4.28. The braid 0"3 .

Two n-string braids can be multiplied by stacking the first on top of

the other. Under this operation, braids form a group called the braid group.
This group has the presentation
(cri , . . . , CTn- 1 I CTi efj = CTj CTi if I i - J I � 2, CTi efi +Wi = CTi+Wi CTi +1 ) .
Definition 4.6.3. The closure of a braid is the diagram of a knot (or link)
obtained from a braid by connecting the top endpoints of the arcs to the
bottom endpoints of the arcs with disjoint arcs (without crossings) . See
Figure 4.29. Any knot (or link) in this form is called a closed braid.
Definition 4.6.4 (Alternate definition of closed braid) . An axis in §3 is
an unknot A in §3 together with a product structure (open disk) x § 1 on
§3 \A. A closed braid is a link L in §3 for which there is an axis such that L
intersects each (open disk) x {point} in the product structure transversely.
We leave it as an exercise to show how the two definitions of closed braid
are equivalent. Observe that the depiction of the trefoil in Figure 4.3 is a
closed braid whereas the depiction of the trefoil in Figure 4.30 is not.
124 4. Knots and Links in 3-Manifolds

Figure 4.29. A closed braid.

Figure 4.30. The trefoil.

Theorem 4.6.5 (Alexander) . Every knot {or link) can be isotoped to be a

closed braid.

Sketch of proof; for details see [7] . Let L be an oriented link in §3 . We

will work in the TRIANG category in the sense that we will think of L as a
finite collection of edges joined end to end. Consider a diagram of L in the
plane.
Choose a point a in JR.2 (representing an axis in §3 orthogonal to JR.2 )
that is not collinear with any edge of L. Connect the endpoints of the
edges of L with a to form triangles. See Figure 4.32. The orientation of L
provides an orientation of the edges of L. Thus each such triangle inherits
an orientation. We leave it as an exercise to show that L is a braid (in the
4.6. Braids 125

Figure 4.31. The 6 1 knot.

sense of Definition 4.6.4) if and only if all triangles coming from the same
component of L have the same orientation.

I
I

Figure 4.32. A triangle.

It now suffices to show that we can perform isotopies of L after which an

arbitrary such triangle has the appropriate orientation. If a triangle has the
wrong orientation and the edge [AB] of L in the triangle contains no crossing
points, choose a point C E JR2 as in Figure 4.33 such that the triangle [ABC]
contains a. Then replace the edge [AB] with the two edges [AC] , [BC] . This
process may introduce new crossings of L, but as long as [AC] , [BC] always
cross over (or always cross under) any other portions of L, the exchange can
be accomplished via an isotopy of L. See Figure 4.33.
If the edge contains one overcrossing ( undercrossing, respectively) , an
identical replacement can be made so long as at the new edges [AC] , [BC]
always cross over (under, respectively) any other portions of L. If the edge
contains multiple crossings, subdivide it into shorter edges, each containing
exactly one crossing, and proceed analogously. D
126 4. Knots and Links in 3-Manifolds

Figure 4.33. A replacement.

Exercises

Exercise 1. Write down the group element for the braid in Figure 4.27.
Exercise 2. Write down the group element of the braid whose closure is
the (p, q)-torus knot.
Exercise 3. Find a closed braid representation of the figure 8 knot.
Exercise 4. Show that the two definitions of closed braid are equivalent.
(Hint: Given a closed braid satisfying the alternate definition, "straighten"
the arcs on the left and "squeeze" the portion containing crossings into a
box.)
4. 7. The Alexander Polynomial

The Alexander polynomial is the oldest known knot polynomial. We discuss

it briefly as an exercise in understanding the ambient space of a knot, a
3-manifold. The definition is a lengthy one. We treat it only informally.
Given a knot, there are several ways to compute a polynomial according
to specific instructions. These several ways correspond to several differ­
ent knot polynomials. In the last two decades, the Jones polynomial has
eclipsed the Alexander polynomial in importance. The Jones polynomial is
computed from a knot diagram. The connection to the ambient 3-manifolds
is obscured. For this reason we omit a discussion of the Jones polynomial
here. For the interested reader, Lickorish's book (see [87]) provides a rich
introduction to the topic. In recent years, a generalization of the Jones
polynomial, called the colored Jones polynomial, has been used in the for­
mulation of the "Kashaev Conjecture" . This conjecture postulates that a
certain growth rate of the coefficients of the colored Jones polynomial gives
the hyperbolic volume of the knot complement.
4. 7. The Alexander Polynomial 127

To compute the Alexander polynomial of a knot K, we first consider

a Seifert surface S of K. Here S is a once-punctured compact connected
orientable surface. As such it is homeomorphic to the surface pictured in
Figure 4.34.

Figure 4.34. The once-punctured compact orientable surface.

We consider the collection of simple closed curves in this surface pictured

in Figure 4.35. Those familiar with algebraic topology will recognize that
the curves correspond to a set of generators for the first homology of the
surface. The homeomorphism between S and the surface pictured in Figure
4.34 defines a collection of simple closed curves in S. (They will depend on
the homeomorphism.)

Figure 4.35. A collection of oriented circles in the once-punctured com­

pact orientable surface.

Because S is an oriented submanifold of the oriented 3-manifold §3 ,

Theorem 3. 1 . 14 tells us that a regular neighborhood of S is a trivial bundle
S x [- 1, 1 ] . Consider a copy of one of the circles ai . We denote a copy of ai
that has been isotoped slightly off of S towards S x { 1} by at. We call ai
a push-off of ai . Figure 4.36 depicts the push-offs in the case of the trefoil.
Define a matrix A with ij-th entry equal to (lk(ai , aJ )). The Alexander
polynomial is the polynomial t::. K = det ( t A - At ) , where At is the transpose
of A. It is a non-trivial fact that this is well-defined up to sign, despite the
numerous choices made.
128 4. Knots and Links in 3-Manifolds

+1

Figure 4.36. The circles a i , a2 and their push-offs at , at for the trefoil.

[ -01 -11 ] '

Example 4. 7. 1 . In the case of the trefoil, the computation is as follows:
A=

t [
tA - At = [ � � t ] - �l �1 ] = [ -�� l -t �1 ] '
AK= t 2 - 2t + 1 + t = t 2 - t + 1 .
If we had made other choices, we could have ended up with -t2 + t - 1 .

Exercises

Exercise 1. Convince yourself that every once-punctured compact con­

nected orientable surface is homeomorphic to a surface as in Figure 4.34.
Exercise 2. Compute the Alexander polynomial for the figure 8 knot.
Exercise 3. Prove that if Ki and K2 are knots, then f),.Ki #K2 = AK1 AK2 •

4.8. Knots and Height Functions

Recall that a height function is a Morse function with at most two critical
points. A height function h on §3 has a maximum and a minimum. We
may assume that a knot K in §3 misses the maximum and the minimum.
It then meets only regular level surfaces of h. Each regular level surface is a
4. 8. Knots and Height Functions 129

2-sphere. Locally, we visualize K in JR3 with h given by projection onto the

z-coordinate. When we do so, we refer to this height function as the natural
height function.
In the 1950s Schubert used height functions for some of his ground­
breaking results on knots and knot complements. See [142] and [141] . In
the process, he introduced several notions, for instance the notion of bridge
number. The following is a reformulation of the notion of bridge number:
Definition 4.8.1. The bridge number of a knot K, denoted by b (K) , is the
least number of maxima for K with respect to a height function on §3 .
To motivate the terminology, consider the following alteration: Isotope
all maxima of the knot K upward so that they all have the same height.
Then consider a level surface L just below the maxima. Here L cuts K
into two collections of arcs: those above L that each contain exactly one
maximum and no other critical points and those below L that each contain
exactly one minimum and no other critical points.
Consider the portion of K lying below L. It consists of subarcs that can
be isotoped into L relative to their endpoints. The result of this isotopy is
a collection of arcs r in L. Moreover, this isotopy can be performed in such
a way that the resulting arcs r in L are disjoint. Now we consider a copy of
K that consists of r along with the small arcs containing the maxima. Each
of the latter arcs is a "bridge" , hence the terminology. See Figure 4.37.

Figure 4.37. Classical bridge position for the figure 8 knot.

Definition 4.8.2. If K is embedded in §3 in such a way that all maxima

occur above all minima of K, then K is said to be in bridge position.
In Section 4.5 we discussed 2-bridge knots. These knots derive their
name from the fact that they are precisely the knots that have bridge num­
ber 2. (They include the trefoil and the figure 8 knot.) The embedding
indicated in Figure 4.24 is in bridge position with respect to the natural
height function. Note how the crossing number of a 2-bridge knot can be
130 4. Knots and Links in 3-Manifolds

arbitrarily large. To see this, recall Theorem 4.4.7. With appropriate choices
of bi , . . . , bn , a 2-bridge knot is reduced and alternating so the diagram re­
alizes the crossing number.
Knots can have arbitrarily high bridge number. This is implicit in the
following theorem that was proved by Schubert:
Theorem 4.8.3 (Schubert) . The (p, q) -torus knot has bridge number
min{p, q} .
Schubert introduced the bridge number for a specific purpose: to show
that a knot can have at most finitely many companions. He first proved vital
results on the relative positioning of tori in knot complements. Specifically,
he showed that tori are either disjoint or intersect in a controlled way. The
following theorem, also due to Schubert, implies that there can be only
finitely many disjoint non-isotopic companion tori:
Theorem 4.8.4 (Schubert) . If K is a satellite knot with companion A and
pattern of index r, then b(K) 2'.: rb(A) . Furthermore, in the case r = 1,
when the satellite construction produces the connected sum of knots Ki and
K2 ,

In the 1980s Gabai introduced the notion of thin position of a knot.

It can be considered either the opposite of bridge position or a refinement
thereof. The idea here is to isotope maxima of a knot to lie below minima
of a knot whenever possible. See Figure 4.38.
Definition 4.8.5. Let h : 83 -t [O, 1) be a height function. Let k c 83 be
a representative of a knot K and let ci , . . . , Cn be the critical values of hl k
listed in increasing order, i.e. , so that ci < · · · < Cn · Choose ri , . . . , rn - i
so that Ci < ri < Ci+ l · Set I4 = h - i (ri) · The width of k relative to h,
denoted by w(k, h) , is Li l k n !4 1 . The width of K, denoted by w(K) , is
the minimum of this relative width over all representatives of K. We say
that K is in thin position with respect to h (or merely in thin position if
the context is clear) if K is embedded so as to minimize width with respect
to h.
The width of the figure 8 knot is 8, as can be seen by its depiction in
Figure 4.39. Exhibiting this picture proves that its width can't be more
than 8. You will prove in the exercises that only the unknot can have width
strictly less than 8.
It is conjectured that for a schematic diagram such as that in Figure 4.38,
the braids bi , b2 , and b3 can be chosen so that the schematic represents a
knot in thin position. If this is indeed the case, then, with appropriate
choices of bi , b2 , and b3 , the knot in Figure 4.38 has width 26.
4.8. Knots and Height Functions 131

Figure 4.38. Schematic of a knot in thin position.

A result relating bridge position and thin position was obtained by

Thompson:
Theorem 4.8.6 (Thompson) . If K admits a thin position that is not a
bridge position, then the complement of K, C(K) , contains a closed essential
surface.

For a proof of this theorem, see [151] . The complement of a torus

knot contains no closed essential surfaces. This follows from our discussion
of essential surfaces in Seifert fibered spaces. It follows that thin position
is bridge position for torus knots. Moreover, since the bridge number of
torus knots is min{p, q}, there are torus knots with arbitrarily large bridge
number. Hence there are also torus knots with arbitrarily large width.
Definition 4.8.7. A regular level surface L such that the first critical point
of K above L is a maximum and the first critical point of K below L is a
minimum is called a thick level.
132 4. Knots and Links in 3-Manifoldt

Figure 4.39. The only regular levels considered in computing the width
of the figure 8 knot.

A regular level surface L such that the first critical point of K above
L is a minimum and the first critical point of K below L is a maximum is
called a thin level.
Example 4.8.8. If K is in bridge position, then the plane separating the
maxima from the minima is a thick level and there are no other thick or
thin levels.
Given a knot in thin position, we really only need to know the number of
intersection points of K with the thick and thin levels in order to compute
the width.
Theorem 4.8.9. Suppose that K is a knot in thin position that intersects
the thick levels (in order) ao, . . . , am times and the thin levels (in order)
bi , . . . , bm times. Then

Proof. If thin position is bridge position for the knot K, then this result is
due to McCrory. In this case m = 0, ao is twice the bridge number, and we
proceed as follows: In computing the width of K, R1 will intersect K in two
points, R2 will intersect K in four points, and so forth, down to the thick
level that intersects K in ao points, the regular level below the thick level
will intersect K in ao 2 points, and so forth, and Rn - l will intersect K in
-

two points. Here I K n � I = I K n � - 1 1 ± 2.

4.8. Knots and Height Functions 133

•
• •
• • •
• • • • .....__ thick level
• • •
• •
•

Figure 4.40. A schematic for computing the width of a knot in bridge position.

We keep track of one-half the number of points of intersection with the

regular levels via a diamond as in Figure 4.40. The top row of the diamond
contains one point representing ! I K n Ri l · The next row of the diamond
contains one more point, representing ! IK n R2 1 · If the bridge number is
l, then there are l maxima. Thus, the thick level will have 2l points of
intersection with the knot, represented by l dots on the thick level of the
diamond. Furthermore, n = 2l; hence the corresponding diamond will have
l rows. The diamond is equivalent to the l x l geometric square in the plane
with dots on the integer lattice. Thus, there are l 2 dots in the diamond.
Hence in this case the width of the knot is 2l 2 = � .
2

When thin position is not bridge position, the analogous schematic,

rather than being a diamond, becomes more complicated but still shows
!w(K) dots. See Figure 4.41. The dots in the n rows of the schematic cor­
respond to ! IK n R1 1 , . . . , ! I K n .Rn - 1 1 , respectively. Furthermore, IK n � I
= I K n � - 1 1 ± 2.

•
• •
• • •
• • • • ....__ thick level
thin level ____. • • •
• • • •
• • • • • ....__ thick level
• • • •
thin level ____. • • •
• • • • ....__ thick level
• • •
• •
•

Figure 4.41 . A schematic for computing the width of a knot.

134 4. Knots and Links in 3-Manifolds

The numbers of dots in the thick levels of the schematic are ' , . . . , �
and the numbers of dots in the thin levels of the schematic are pt, . . . , � . At
each thick level ti for i = 1, . . . , n in the schematic we can inscribe a "thick"
diamond Di in the schematic whose diagonal is the thick level. See Figure
4.42. Then Di is equivalent to the � x � geometric square and contains '1-
2

dots.
•
• •
• • •
• • • • .....__ thick level
thin level � • • •
0 • • 0
0 0 • 0 0 .....__ thick level
0 0 0 0
thin level � 0 0 0
0 0 0 0 .....__ thick level
0 0 0
0 0
0

Figure 4.42. A thick diamond indicated by black dots.

There will be exactly m + 1 thick diamonds inscribed in the schematic.

Together, they contain all dots in the schematic. However,n the thick dia-
monds will overlap. So the width of K will be less than ! E at .
i=O
Analogously, we can inscribe m "thin" diamonds Ei, . . . , Em in the
schematic. Each of these will be equivalent to a � x � geometric square
and contain ¥ dots. Now note that the successive thick diamonds Di, Di+t
overlap in the thin diamond Ei. See Figure 4.43.
We leave it as an exercise to show that for each dot d, if d lies in Ek n
Ek+I n · · nE1, then d also lies in Dk 1 n Dk n n D1 Thus, for I Dil, I Eil the
· - · · · .

number of dots in Di, Ei, respectively, the number of dots in the schematic
is given by
m m
Di
L I l - L I Ei l·
i=O i=l
The width of the knot is twice the number of dots in the schematic; hence
the formula holds. D

In Figure 4.44 we count the dots indicated in black once, we count the
dots indicated in white twice and subtract them once, we count the dot
indicated in gray three times and subtract it twice.
4.8. Knots and Height Functions 135

•
• •
• 0 •
• o o • ..--- thick level
thin level ---.. o o o
• 0 0 •
• • o • • ..--- thick level
• • • •
thin level ---.. • • •
• • • • ..--- thick level
• • •
• •
•

Figure 4.43. A thin diamond indicated by white dots .

•
• •
• 0 •
• 0 0 •
0 0 0
• 0 0 •
• • • • •
• 0 0 •
0 0 0
• 0 0 •
• 0 •
• •
•

Figure 4.44. A schematic for computing the width of a knot.

The behavior of the width under the connected sum of knots is erratic.
It is not hard to prove the following proposition:
Proposition 4.8.10.
w(K1 #K2 ) :::; w(K1 ) + w(K2 ) - 2.
Proof. To provide an embedding of Ki #K2 of width w(K1 ) + w(K2 ) 2, -

we isotope K1 to lie above K2 . Then we form the connected sum by deleting

a small arc containing the minimum of K1 and a small arc containing the
maximum of K2 · See Figure 4.45. D

In some cases, this inequality is sharp.

136 4. Knots and Links in 3-Manifolds

K,

K2

Figure 4.45. An embedding for K1 #K2 .

Definition 4.8. 1 1 . A knot is small if its complement contains no closed

essential surfaces.
Theorem 4.8.12 ( Rieck-Sedgwick ) . For small knots

The following result, due to Scharlemann and the author, gives us some
indication of what to expect:
Theorem 4.8. 13 ( Scharlemann-Schultens ) .

The central technique of the proof of this theorem can be used to con­
struct candidates for knots that realize this inequality. This strategy was
pursued by Scharlemann and Thompson and was accomplished by Blair and
Tomova, who exhibit knots for which this inequality is sharp.

Exercises

Exercise 1 . Prove that only the unknot has width strictly less than 8.
Exercise 2. Prove the "easy inequality" in Schubert's Theorem concerning
torus knots; specifically, prove that the bridge number of the (p, q ) -torus
knot is at most min {p, q } .
4. 9. The Knot Group 137

Exercise 3. Prove the statement made in the proof of Theorem 4.8.9 that
was left as an exercise: "For each dot d, if d lies in Ek n Ek+ l n n E1 , then
· · ·

d also lies in Dk - 1 n Dk n · · · n D1 .11

Exercise 4. Provide an example of a knot that is not small.

4.9. The Knot Group *

To every knot we can associate a group as described below. This can serve
to distinguish knots if their associated gro.ups are known to be distinct. In
general, distinguishing groups based on their generators and relations can
be rather tricky. Nevertheless, the insights of an algebraic point of view are
sometimes complementary to the insights from a topological point of view.
The knot group and invariants derived from the knot group are an active
area of study today.
The knot group is an example of a more general construction: To every
connected topological space one can associate the "fundamental group" . The
knot group is the fundamental group of the complement of the knot. For
this reason it is well-defined. We will not cover this fundamental result here.
Rather, we will focus on computing knot groups.
The Wirtinger presentation provides an algorithm for computing knot
groups from knot diagrams:
Step 1 . Choose a point far from the knot. See Figure 4.46 .

Figure 4.46. Choosing a point.

Step 2. To each arc in the knot diagram associate a generator. See Figure
4.47.

Step 3. To each crossing associate a relation as in Figure 4.48, e.g. , x 2 x 1· ·

x2 1 x4 1 for the bottom right crossing.

·
138 4. Knots and Links in 3-Manifolds

Figure 4.47. Generators of the knot group.

; 4
1 1
Figure 4.48. The relation for a crossing: x2 x 1 x x .

Step 4. Simplify if possible.

For the knot group of the figure 8 knot we obtain

Substituting

and
4. 1 0. Covering Spaces 1 39

from the first and fourth relations, we obtain

( X 1 , X 2 I X 1 X 2 X -1 1 X 2- 1 X -1 1 X 2 X 1 X 2- 1 X -1 1 X 2- 1 ,
X 22 X -1 1 X 2- 1 X 1 X 2 X 1 X 2- 1 X 2 X 1 X 2- 1 X 2 X -1 1 X 2- 1 X -1 1 X 2 X 1 X 2- 1 )
= (X i , X2 I X 1 X2 X - 1 -1 -1 -1 -1 -1 2 -1 -1 -1 -1 -1
1 X 2 X 1 X2 X 1 X 2 X 1 X2 , X 2 X 1 X2 X 1 X2 X 1 X2 X l X2 X 1 X 2 )
= (x 1 , x2 I X 1 X2 X - 1 -1 -1 -1 -1 -1 -1 -1 -1 -1
1 X2 X 1 X2 X 1 X 2 X 1 X2 , X2 X 1 X2 X 1 X2 X 1 X2 X 1 X2 X 1 )
= (X i , X2 I X 1 X2 X-
1 -1 -1 -1 -1 -1
1 X2 X l X2 X 1 X2 X l X 2 )
(since in the preceding line the second relation is just a conjugate of the
inverse of the first relation) .
Exercises

Exercise 1 . Calculate the knot group of the trefoil knot. (Hint: The answer
should be ( a, bja2 b - 3 ) . )
Exercise 2. Calculate the knot group of your favorite knot.
Exercise 3. Prove Dehn's Theorem, which states that a knot is the unknot
if and only if the fundamental group of its complement is cyclic. (Hint: Use
Exercise 1 from Section 4.3.)
4.10. Covering Spaces *

Recall our discussion of covering spaces in Section 2.4. In the case that X
is a manifold, a covering space is just a bundle over X that has a discrete
fiber. There is a strong connection between covering spaces and subgroups
of the fundamental group:
Theorem 4.10. 1 . Let X be a connected topological space. There is a 1-
to-1 correspondence between connected covering spaces of X and conjugacy
classes of subgroups of 7r 1 ( X ) .
Definition 4.10.2. A cover X' of X is universal if it is simply connected.
The universal cover of a space corresponds to the trivial subgroup of
the fundamental group of the space. You will prove in the exercises that
the universal cover of a manifold is unique up to homeomorphism. In the
context of knots, we are particularly interested in the infinite cyclic covering
space. It corresponds to the commutator subgroup [7r 1 (C(K) ) , 7r 1 (C(K))]
(i.e., the subgroup generated by elements xyx - 1 y - 1 , for pairs (x, y) of el­
ements of 7r 1 (C(K))) of the knot group, since H1 (C(K)) � Z and hence
7r1 (C(K))/[7r 1 (C(K)) , 7r 1 (C(K)) J � Z) .
The infinite cyclic cover of a knot complement is constructed as follows:
Let S be a Seifert surface for the knot K c §3 . Abusing notation slightly, we
140 4. Knots and Links in 3-Manifolds

also denote S n C(K) by S. If we cut C(K) along S, we obtain a manifold

M with (connected) boundary. Recall that by Theorem 3. 1 . 14, S is 2-sided.
Thus in the boundary of M, we see two remnants of S. We denote one by s+
and the other by s - . Now consider a countably infinite collection of copies
of M, indexed by Z. Specifically, for each n E Z, we have Mn such that
8Mn cont�ins two marked copies of the Seifert surface, S;t and S;; . For all
n we identify S;t in Mn with S�+l in Mn +I to obtain M'. See Figure 4.49.

s�-1

Figure 4.49. A portion of the infinite cyclic cover of a knot complement.

There is a natural map f : M' -+ M'. It takes each point x E Mn to the

corresponding point in Mn + l · Let (f) be the group generated by f (i.e. , the
group consisting of powers of the map !) . It corresponds to a covering space
(M', C(K) , p) , where p : M' -+ C(K) is the quotient map M' -+ M' / (!) .
Here f and powers of f are covering transformations. (The covering
transformation f played an implicit role in our computation of the Alexander
polynomial.) The infinite cyclic cover factors through k-fold cyclic covers of
the form (Mk = M' / (f k ) , C(K) , pk ) , where pk : M k -+ C(K) is the map
induced by p. Thus we obtain a tower of covers:
M'

1
M'/ (J k ) = Mk

lpk
M'/ (f) = C(K) .
Recall also our discussion of branched covering spaces. To obtain the
k-fold branched cover of §3 over K, first construct the k-fold cover
(Mk , C(K) , pk ) of C(K) . Then glue a solid torus V to {)M k in such a way
that the copies of S meet {)V in longitudes. We thus obtain a manifold i/fk .
Let c be the core of V, that is, the simple closed curve given by { 0} x § 1 .
The map qk given by z -+ z k on the first factor and the identity on the
second factor defines a map V -+ V such that (V, c , V, c, qk ) is a branched
Exercises 141

cover. Note that by attaching a solid torus to C(K) in such a way that S
meets the solid torus in a longitude, we obtain §3 . Thus by concatenating
p k and q k to a map r k we obtain the branched cover (M k , c, §3 , K, r k ) .
One case of special interest is that of a 2-fold branched cover. Note that
when k is 2, then a covering transformation is an involution (i.e. , its square
is the identity) . One reason why we are interested in covering spaces and
branched covers in particular are the two theorems below:
Theorem 4.10.3 (Alexander) . Every 3-manifold can be obtained as a
branched cover over a link in §3 .
See [6] . For an updated proof, see [40] .
In fact, more is true:
Theorem 4. 10.4 (Thurston, Hilden-Lozano-Montesinos) . There is a knot
K in §3 (in fact there are many) such that every closed orientable 3-manifold
can be realized as a branched cover over K.
See [64] .
Exercises

Exercise 1. Construct at least three distinct 3-manifolds that have a bundle

structure with the genus 2 surface as their base space and the circle as their
fiber.
Exercise 2. Prove that the universal cover of a manifold is unique up to
homeomorphism.
Exercise 3. Consider a 2-bridge knot K and the 2-fold branched cover
of §3 branched over K. §3 contains sphere submanifolds that separate the
2-bridge knot into two pairs of unknotted arcs. What is the preimage of
such a 2-sphere under the covering map? (Hint: Consider Figure 4.50.)

Figure 4.50. The quotient o f a surface by an involution.

 Chapter 5

Triangulated
3- Manifolds

Triangulating a 3-manifold allows us to focus on tetrahedra rather than on

the 3-manifold as a whole. For more on the origins of this approach, see
[102] . Many beautiful results have come out of this theory and there are
surely many more to come. Normal surface theory is a sophisticated tool
in the theory of 3-manifolds. It was envisioned and developed by Wolfgang
Haken who used it to describe algorithms detecting certain properties of
3-manifolds; see [53]. This theory is very natural, the sophistication lies in
the details. For further reading on normal surface theory, see [61] or [96] .

5 . 1 . Simplicial Complexes

Recall the basic definitions for triangulated manifolds discussed in Section

1 .4. We here expand on these definitions for a more substantial discussion
of triangulated 3-manifolds. As we will see, the notion of barycentric subdi­
vision interpolates between the more traditional definition of a triangulation
of a manifold, where every simplex is required to be embedded and pairs of
simplices are allowed to meet in at most one simplex, and the more modern
definition used here. The triangulations defined can be subdivided to obtain
triangulations in the traditional sense.
In what follows we will consider only locally finite simplicial complexes.
The reasoning used in Lemma 1 .4.28 then allows us to view our simplicial
complexes as quotient spaces of standard simplices.
Definition 5.1.1. Let K be a k-simplex. Then for r :::; k, the r-skeleton
Kr of K is the collection Kr = { [s] E K; dim [s] :::; r }.
-

143
144 5 . Triangulated 3-Manifolds

Example 5.1.2. Figure 5.1 depicts the 1-skeleton of the simplicial complex
depicted in Figure 1. 18.

Figure 5 . 1 . The I-skeleton of a simplicial complex.

Remark 5 . 1 .3. The r-skeleton of a simplicial complex K is a subcomplex

of K.
Definition 5.1.4. For a point v in the standard k-simplex [s] =
[vo, . . . , vk ] ,
there is a vector (ao , . . . , ak ) with ai E [O, 1 ] such that v = aovo + · · · + ak vk .
The ( k + 1 ) -tuple (ao , . . . , ak ) is called the barycentric coordinates of v.
Remark 5.1.5. Elementary linear algebra shows that the barycentric co­
ordinates of a point v E [s] are unique.
In Rk+1 , most collections wo, . . . , wk of k + 1 points span a k-simplex
[wo, . . . , wk ] · Note that there will always be an affine map from the standard
k-simplex to [wo, . . . , wk ] · Below, we define a 0-simplex called the barycenter
of a simplex and a collection of simplices called the barycentric subdivision
of a simplex.
Definition 5 . 1 .6. The point k�l vo + · · · + k� Vk is called the barycenter of
l
[vo, . . . , vk ] and is denoted by b([vo, . . . , vk ] ) .
Remark 5.1 .7. Note that b([vo] ) = vo.
Definition 5.1.8. Let K be a simplicial complex. Define a partial ordering
on K by [s 1 ] � [s 2 ] if and only if [s 1 ] is a face of [s 2 ] . We write [s 1 ] < [s 2 ]
when [s 1 ] � [s 2 ] and [s 1 ] # [s 2 ] .
Below we define the barycentric subdivision of a simplicial complex K.
Subdivisions can be defined more generally, but we will not do so here.
The idea is to take each standard simplex in the simplicial complex and
subdivide it into smaller simplices. This can be achieved by adding the
barycenters of edges, thereby subdividing each edge into two edges, then
adding the barycenters of 2-dimensional faces and adding six edges in each
face between the barycenter and the vertices of these ( subdivided ) edges of
the 2-dimensional face and proceeding inductively adding simplices spanned
by barycenters to obtain a simplicial complex.
5 . 1 . Simplicial Complexes 145

Definition 5.1.9. Let [s] be the standard k-simplex. Let [so] , . . . , [st] be
faces of [s] such that [so] < [s 1 ] < · · < [st] . Consider the simplex [b([so] ) ,
·

b([s 1 ] ) , . . . , b([sz] )] . The first barycentric subdivision of [s] is the simplicial

complex consisting of all simplices of this form:
{ [b( [so] ) , b([s 1 ] ) , . . . , b([sz])] :
[so] , [s 1 ] , . . . , [st] faces of [s] such that [so] < [s 1 ] < · · · < [sz] }.
Let K be a simplicial complex. Then I KI is the quotient space obtained
by identifying standard simplices. The first barycentric subdivision of K,
denoted by K( l ) , is the simplicial complex consisting of the union of first
barycentric subdivisions of the standard simplices in K.
Abusing terminology, we continue to think of K( l ) as a quotient space
obtained from standard simplices.
Definition 5 . 1 . 10. The simplicial complex K(n ) = (((K( 1 ) )( 1 ) ) . . . )< 1 > is
called the n-th barycentric subdivision of K.
Caution: Kn =/= K(n ) .
Example 5 . 1 . 1 1 . Figure 5.2 depicts the first barycentric subdivision of the
simplicial complex depicted in Figure 1 . 18.

Figure 5.2. The fi.rst barycentric subdivision o f the simplicial complex

in Figure 1 . 18.

Recall that the modern definition of triangulation allows a simplex in a

manifold to share more than one face with another simplex. It also allows
distinct faces of a simplex to coincide. You will prove in the exercises that the
second barycentric subdivision of a modern triangulation is a triangulation
in the traditional sense: Each simplex is embedded and any two simplices
share at most one face.
In studying simplicial complexes, it is of interest to understand the local
picture near a vertex. The local picture of a simplicial complex near a vertex
I46 5. Triangulated 3-Manifolds

is captured by the following concept:

Definition 5.1.12. The star of v, for v a vertex of a simplicial complex K,
is the set
St(v) = { (s) E K : v E [s] } .
More generally, the star of a subcomplex K' of K is the set
St(K') = { (s) E K : K' n [s] � 0} .
Remark 5.1.13. Here St(v) and St(K') are not themselves simplicial com­
plexes since they are unions of open simplices. We will nevertheless denote
the set of points in St(v) by ISt(v) I .
Recall the definition of a simplicial map in Section I.4. A related notion
is the following:
Definition 5.1.14. Let K, L be simplicial complexes and let f : I K I -+
ILi be a continuous map. A simplicial map <P : I KI -+ ILi is a simplicial
approximation to f if f(ISt(v) I ) C ISt(<jJ(v)) I for each vertex v of K.
Theorem 5.1.15. Let K, L be finite simplicial complexes and let f : IK I -+
ILi be a continuous map. For sufficiently large n, there exists a simplicial
map <P : I K(n ) I -+ I Li that is a simplicial approximation of f .
A proof of this theorem can be found in [148] . The definitions there
differ slightly from the definitions used here. However, with appropriate
modifications, the proof still carries over.
Example 5.1.16. Consider the continuous map of a I-dimensional simpli­
cial complex into a 2-dimensional simplicial complex pictured in Figure 5.3.
To construct a simplicial approximation, we first need to consider the con­
dition on stars. Note that there is no vertex in the 2-dimensional simplicial
complex that could be the image of the vertex v3 of the I-dimensional sim­
plicial complex and satisfy the star condition. Thus we must first take the
barycentric subdivision of the I-dimensional simplicial complex to construct
a simplicial approximation. See Figure 5.4. The simplicial approximation is
pictured in Figure 5.5.

Definition 5.1.17. Let K be a simplicial complex. For v E K0, consider

the star St(v; K( l ) ) of v in the first barycentric subdivision of K. Take
St(v) = { [s] : (s) E St(v) }.
The link of v, denoted by link(v) , is defined by
link(v) = St(v; K( 1 ) )\St(v; K ( l ) ) .
 5 . 1 . Simplicial Complexes 147

Figure 5.3. A continuous map of a I-dimensional simplicial complex

into a 2-dimensional simplicial complex.

Figure 5.4. The barycentric subdivision.

Figure 5.5. A simplicial approximation.

The subject of general position is a rather intricate one and is briefly

treated (in the DIFF category) in Appendix A. We require a special case of
general position in the TRIANG category that we introduce here.
Definition 5 . 1 . 18. Let v be a point in IR.n and let A be a subset of IR.n .
The pair (v, A) is in general position if v � A and, for each ai , a2 E A with
ai =I= a2 , [v, a1] n [v, a2 ] = {v}.
148 5. Triangulated 3-Manifolds

Definition 5 . 1 . 19. Let (v, A) be in general position. The set

LJ [v, a]
aEA
is called the cone of v over A and is denoted by v A. *

Example 5.1 .20. Figure 5.6 depicts the cone of a pair consisting of a point
v and a line segment A in R2 .

Figure 5.6. The cone of the pair ( v, A) where A is a line segment in R2 •

Exercises
Exercise 1. Let (M, K) be a triangulated manifold. Prove that the second
barycentric subdivision of K is a triangulation of M in the traditional sense.
Exercise 2. Let [s] = [vo, . . . , vk ] be the standard k-simplex. Prove that
(vo, [vi , . . . , vk ]) is in general position and vo [vi , . . . , vk] = [s] .
*

Exercise 3. Prove that for a simplicial complex K, the collection {St( v )}veKo
is a covering of I KI ·
Exercise 4. Let f : IKI -+ ILi be a continuous map and let </> : IKI -+ ILi
be a simplicial approximation. Prove that </> : I K I -+ I Li is homotopic to f
(via a "straight line homotopy" ) .
5 . 2 . Normal Surfaces

Normal surfaces, defined below, are a natural class of surfaces to consider

in the context of triangulated 3-manifolds. We shall see that incompress­
ible surfaces can be isotoped to be normal surfaces. In this context, neither
incompressible surfaces nor normal surfaces are subcomplexes of the trian­
gulations of the 3-manifolds under consideration. Nevertheless, via the tool
5.2. Normal Surfaces 149

of normal surfaces, the triangulation of the manifold helps us understand

the incompressible surface.
Definition 5.2.1. A (properly embedded) arc on a 2-dimensional face [ !]
of a 3-simplex [s] is a normal arc if its endpoints lie on distinct edges of [f] .
A simple closed curve c on the boundary of a 3-simplex [s] is a normal curve
if every component of intersection of c with a 2-dimensional face [!] of [s] is
a normal arc. See Figures 5.7 and 5.8
Normal arcs lie on the 2-dimensional faces of a 3-simplex [s] . Here [s]
inherits a geometry. This allows us, for simplicity, to assume that normal
arcs are straight line segments.

Figure 5.7. A normal curve.

Figure 5.8. A curve that is not normal.

Definition 5.2.2. The length of a normal curve c on the boundary of a

3-simplex [s] is the number of points in c n l [s] 1 J .
Example 5.2.3. The length of the normal curve depicted in Figure 5 . 7 is 4.
Lemma 5.2.4. A closed normal curve on the boundary of a 3-simplex either
has length 3 or 4 or it meets some edge more than once.
150 5. Triangulated 3-Manifolds

Proof. Let [s] be a 3-simplex and let c be a closed normal curve on Bl [s] I .
Here 81 [s] 2 1 is homeomorphic to a sphere. Thus c is a Jordan curve on the
sphere. It separates 8i [sJ 2 1 into an "inside" (disk) and an "outside" (disk) .
Consider l [s]0 i . It consists of four vertices. These vertices are distributed
among the inside and outside disks.
Claim. Each disk contains at least one vertex.
Suppose a disk contains no vertices. How do the edges of [s] intersect
this disk? They intersect it in arcs. An outermost arc cuts off a subarc
of c that has both endpoints on the same edge. This implies that c is not
normal, a contradiction. Thus the disk is disjoint from i [sJ 1 1 and is contained
in a 2-dimensional face. This is also a contradiction, as a 2-dimensional face
contains no closed normal curve.
It follows that, up to renaming, there are only two possibilities: The
"inside" of c contains either one or two points of l [s]0 i . Suppose that the
"inside" of c contains only the vertex v. Let ei , e2 , e3 be the edges incident
to v . Then c must intersect e1 , e 2 , e 3 because the other vertices to which
ei , e 2 , e 3 are incident lie "outside" of c. In fact, c must intersect each of
ei , e 2 , e 3 an odd number of times. Similarly, it must meet the other three
edges an even number of times. Thus if c meets no edge more than once,
then it has length 3.
Suppose now that the "inside" of c contains the vertices v 1 , v2 . Then by
analogous reasoning, there are four edges that are met an odd number of
times and two edges that are met an even number of times. If the former
edges each meet c once and the latter are disjoint from c, then c has length
4. Otherwise, c meets some edge more than once. 0

Figure 5.9. A normal curve of length 3.

Definition 5.2.5. A normal triangle in a 3-simplex [s] is a (properly em­

bedded) disk whose boundary is a normal curve of length 3. A normal
5 . 2. Normal Surfaces 151

Figure 5 . 1 0 . A normal curve of length 1 2 .

quadrilateral in a 3-simplex [s] is a (properly embedded) disk whose bound­

ary is a normal curve of length 4. A normal disk is a normal triangle or
quadrilateral.
Example 5.2.6. The triangle spanned by the normal curve of length 3 in
Figure 5.9 is a normal triangle. The quadrilateral spanned by the normal
curve of length 4 in Figure 5. 7 is a normal quadrilateral.
Non-Example. The normal curve in Figure 5 . 10 does not span a normal
disk.
Definition 5.2.7. Let (M, K) be a triangulated 3-manifold. A normal
surface in (M, K) (or simply in M when it is clear what triangulation is
meant) is a surface S c M such that for every 3-simplex [s] in K, S n l [s] I
consists of disjoint normal disks in [s] .
Example 5.2.8. Recall that the 3-sphere can be obtained by identifying
the faces of two 3-simplices as in Figure 5 . 1 1 . The two shaded 2-simplices
in Figure 5 . 1 1 identify to a normal 2-sphere.

d d

Figure 5 . 1 1 . A normal 2-sphere in the 3-sphere obtained by identifications.

152 5. Triangulated 3-Manifolds

Example 5.2.9. The 3-torus is obtained by identifying eight appropriately

chosen reflections of the cube pictured in Figure 5. 12. A normal 2-torus in
this triangulated 3-torus is obtained by choosing four of these to contain
reflections of the square pictured.

/ / /

---
'
' I ,
/

'
'
'
'
'
'

Figure 5 . 1 2 . A portion of a normal 2-torus in the 3-torus.

Definition 5.2. 10. The weight of a surface S in a triangulated 3-manifold

(M, K) is the number of components of S n I K 1 1 . It is denoted by w ( S) .
Similarly, the measure of S, m(S) , is the number of components of S n
( I K2 l \ I K 1 1 ) .
Example 5.2. 1 1 . The weight of the normal 2-sphere in the 3-sphere pic­
tured in Figure 5.11 is 3. Its measure is also 3.
Example 5.2.12. The weight of the normal 2-torus in Example 5.2.9 is 12.
Its measure is 36.
In the context of normal surfaces we often work with the following, more
restricted, notion of isotopy.
Definition 5.2.13. A normal isotopy is an isotopy through normal surfaces.
Theorem 5.2. 14. Let M be a closed irreducible 3-manifold containing an
incompressible surface S. Then for any triangulation (M, K) of M there is
an isotopy that takes S to a normal surface in ( M, K ) .
The analogous result holds if M is merely compact but S is both incom­
pressible and boundary incompressible.
5 . 2. Normal Surfaces 153

Proof. Let (M, K) be a triangulation of M and let S be an incompressible

surface in M. Isotope S so that (w(S) , m(S)) is minimal (in the dictionary
order) . (We assume that S is in general position with respect to the trian­
gulation. In particular, S is disjoint from vertices, intersects edges in a finite
number of points, and intersects 2-simplices in a finite number of arcs.)
Let [s] be a 3-simplex of K and let [!] be a 2-dimensional face of [s] .
Since M is irreducible and S is incompressible, a standard innermost disk
argument (see Section 3.9) shows that simple closed curves in S n l [ !J I can
be removed. Removing a simple closed curve lowers (w(S) , m(S)) by (0, 1).
Minimality of (w(S) , m(S)) thus ensures that S n l [J] I contains no closed
components. (See Figures 5.13 and 5. 14.)
A standard outermost arc argument (see Section 3.9) shows that simple
arcs in S n l [J] I with both endpoints on the same edge can be removed.
Removing such an arc lowers (w(S) , m(S)) by (2, 1). Thus minimality of
(w(S) , m(S)) ensures that, after normal isotopy, S n l [J] I consists of normal
arcs. Since this is true for all 2-dimensional faces of [s] , S n 8 1 [s] I consists of
normal curves.

Figure 5.13. A closed component of intersection of S with K2 •

Let c be a normal curve in S n al [s] I . We need to show that c bounds

a disk in l [s] I . Since c bounds a disk E in the 3-ball l [s] I and since S is
incompressible, c must in fact bound a disk § in S. A priori the interior
of E may not be disjoint from S, but a standard innermost disk argument
shows how to eliminate simple closed curves of intersection in E n S. Thus
154 5. Triangulated 3-Manifolds

Figure 5. 14. After the isotopy.

E U S is a 2-sphere in an irreducible 3-manifold and hence bounds a 3-ball

B. If S does not lie entirely in l [s] I , then either w(int( S )) or m(int( S )) or
both are positive and hence B describes an isotopy lowering (w(S) , m(S))
by this amount, a contradiction. Thus c bounds the disk S = S n l [s] I .
We now need to show that S is a normal disk. By Lemma 5.2.4 it
suffices to show that 8 S does not meet any edges of [s] more than once.
Suppose that 8 S meets the edge [e] of [s] more than once. Here c = 8S
partitions 81 [s] I into two disks, D, D' . One of these disks, say D, meets
(e) in a subarc a connecting two adjacent points of intersection in S n l [e] I .
Furthermore, S is isotopic to D . This guarantees the existence of an arc /3
in the interior of S and a disk E with 8E = a U /3 and interior in (s) . The
disk E can be used to describe an isotopy analogous to the isotopy used in
the standard outermost arc argument. This isotopy lowers w(S) by 2. Thus
the minimality of (w(S) , m(S)) guarantees that 8S does not meet any edges
of [s] more than once and hence, by Lemma 5.2.4, S is a normal disk.
Applying this reasoning to all 3-simplices in K shows that S is a normal
surface. D

Exercises

Exercise 1. Choose a triangulation (§3 , K) of §3 and give an example of a

normal surface in (§3 , K) .
Exercise 2. Is the converse of Theorem 5.2. 14 true? (I.e. , is a normal
surface in a triangulated 3-manifold necessarily incompressible?)
 5 . 3. Diophantine Systems 155

5 . 3 . Diophantine Systems

One advantage of working with normal surfaces lies in the fact that local be­
havior is controlled. Given a normal surface S in a triangulated 3-manifold
(M, K) , consider a 3-simplex [s] in K. There are exactly four normal iso­
topy types of normal triangles in [s] : Indeed, a normal curve c of length
3 partitions the vertices of [s] into one vertex on one side of c and the re­
maining three vertices on the other side of c. Thus, up to isotopies through
normal curves of length 3, there are exactly four such curves. Moreover,
a normal triangle is determined by its boundary. Likewise, you will prove
in the exercises that there are exactly three normal isotopy types of normal
quadrilaterals in [s] and that representatives of distinct types must intersect.
Thus S n [s] determines a vector with seven entries, (t 1 , t 2 , t 3 , t4 , q 1 , q2 , q3 ) .
The first four entries correspond to the number of normal triangles of a given
normal isotopy type, the last three entries correspond to the number of nor­
mal quadrilaterals of a given normal isotopy type, and only one of the latter
three entries can be non-zero. A vector of this form (seven non-negative
integral entries, with only one of the last three entries non-zero) is said to
satisfy the square restriction. Thus if t is the number of 3-simplices in a
triangulated 3-manifold (M, K) , then S determines a vector with 7t entries
that satisfies the square restriction.

Figure 5.15. A normal surface in adjacent 3-simplices.

Now consider two adjacent 3-simplices [s 1 ] and [s 2 ] (see Figure 5. 15)

with vectors (tl , t�, tr, tf, q} , q�, qD, (t�, t�, t�, t�, q� , qi, q�) . The normal tri­
angles and normal quadrilaterals in [s 1 ] and [s 2 ] must match up along any
2-simplices in which [s 1 ] and [s 2 ] meet. This places restrictions on the nature
and number of normal triangles and quadrilaterals that can occur in [s 1 ] and
[s 2 ] . To be precise, there are three normal isotopy types of normal arcs in
each 2-simplex. For each such normal arc, there is one normal isotopy type
of normal triangle and one normal isotopy type of normal quadrilateral that
156 5. 'Ii:iangulated 3-Manifolds

give rise to this normal arc. Thus there are three equations of the form
. .
t i + l1i = t k2 + q2l
associated with a 2-simplex along which [ s 1 ] and [s 2 ] meet. Hence there are
6t gluing equations that must be satisfied by the vector determined by S.
We wish to use linear algebra to help us find normal surfaces in a tri­
angulated 3-manifold. To this end we need to make sense out of the sum
of two normal surfaces. More specifically, given two normal surfaces S1 , S2
and corresponding vectors V'1 , v2 , we wish to define the sum S1 + S2 corre­
sponding to V'1 + v2 . This is possible, but we must proceed with caution:
In a 3-simplex, the normal triangles and quadrilaterals corresponding to S1
can intersect the normal triangles and quadrilaterals corresponding to S2 .
In the 3-simplex, we may be able to isotope these apart, but note that the
surface connects up to other normal disks in adjacent 3-simplices. The iso­
topy might not extend beyond the 3-simplex to make progress globally. This
will certainly be the case if S1 and S2 can't be made disjoint in M.

Figure 5.16. A regular switch.

Figure 5. 17. An irregular switch.

The operation in Figure 5.16 is called a regular switch. (Compare this

with the operation in Figure 5. 17 that is not a regular switch.) If, in a given
3-simplex [s] , two normal triangles intersect, then the regular switches on the
2-dimensional faces of [ s ] extend into ( s ) to define a cut and paste operation
on the normal triangles resulting in two disjoint normal triangles of the same
normal isotopy types as the original pair. Similarly, if a normal triangle and
5 . 3. Diophantine Systems 157

a normal quadrilateral intersect in [s] , then the regular switches on the 2-

dimensional faces of [s] extend into (s) to define a cut and paste operation
resulting in a disjoint normal triangle and a normal quadrilateral of the
same normal isotopy types as the original pair. If two normal quadrilaterals
intersect in [s] , the regular switches on the 2-dimensional faces of [s] extend
into ( s) only if the two quadrilaterals are of the same normal isotopy type.
If the two quadrilaterals are of different normal isotopy types, the regular
switches on the 2-dimensional faces of [s] do not extend into (s) (the cut
and paste operation is not defined) . We leave this as an exercise for the
reader. Two normal surfaces 81 , 82 in the triangulated 3-manifold M are
said to satisfy the square restriction if for every 3-simplex [s] in M, at most
one normal isotopy type of quadrilateral occurs in (81 U 82 ) n l [sJ I . Note
that the surfaces in question need not be connected.
Definition 5.3. 1 . Let 81 , 82 be a pair of normal surfaces in the 3-manifold
M that satisfy the square restriction. We denote the result of performing
the cut and paste operations discussed above by 81 + 82 .

I '

'
'

'

I '
' I

Figure 5.18. The surfaces S1 and S2 .

Given a normal surface 8, the normal triangles and quadrilaterals that

constitute 8 match up along the faces of adjacent 3-simplices. Thus the
vector v determined by S satisfies the square restriction and the 6t gluing
equations; i.e. , Av = 0 for some matrix A with non-negative integer entries.
Conversely, given a vector with 7t non-negative integral entries that sat­
isfy the gluing equations and the square restriction, we can· build a normal
surface from the corresponding number of normal triangles and quadrilat­
erals. Note that, up to normal isotopy, there is only one way that a given
158 5. Triangulated 3-Manifolds

i.
I

' '

'
- - - - �- - - 2_,_
'
- - - - - -·
'
'

'

'

Figure 5.19. The surface S1 + S2 .

collection of normal triangles and normal quadrilaterals satisfying the square

restriction can be embedded in a 3-simplex. See Figure 5.20.

Figure 5.20. Si n [s] .

The normal triangles and quadrilaterals match up to form a surface be­

cause the gluing equations are satisfied. Furthermore, consider two surfaces,
S1 , S2 corresponding to the same vector. In each simplex [s) , S1 n l [s] I must
be properly isotopic to S2 n l [s] I . This isotopy can be realized, for instance,
by taking the straight line homotopy on the edges of [s] (which intersect
S1 and S2 in ordered sets of points of the same cardinality) and extending
across the faces and the interior of [s] . Moreover, these normal isotopies in
the simplices piece together to a normal isotopy between S1 and S2 . Hence
5 . 3. Diophantine Systems 159

there is a 1-to-1 correspondence between normal surfaces in a triangulated

3-manifold, considered up to normal isotopy, and non-negative integral so­
lutions to the system of equations described above. By considering the
non-negative solutions of a system of equations we are really considering
a system of equations and inequalities. A system of linear equations and
inequalities with integral coefficients is called a Diophantine system.
Diophantine systems have been studied extensively. We wish to describe
their solution sets to further our understanding of normal surfaces. The
following theorem is our key to understanding. It gives us a finite generating
set for all our solutions. This provides a finite set of normal surfaces that
generate all of our normal surfaces up to normal isotopy!
Theorem 5.3.2. Let Ax = 0, Xi 2: 0 Vi be a Diophantine system. There is a
finite set of integral solutions that generates the full set of integral solutions
to the Diophantine system under the + operation.
Proof. Let A be an ( m x n ) -matrix with non-negative integer coefficients.
We consider zn c JR.n . Set
X = { x = ( x 1 , . . . , xn ) E 1Rn : Xi 2: 0 and Ax = O } .
Note that each of the rows of A defines a hyperplane through the origin.
Thus X lies in the intersection of these m hyperplanes. The intersection
of these hyperplanes is a subspace V of JR.n of dimension d at least n m . -

So X is the intersection of this subspace with the first quadrant of JR.n . In

particular, X is convex. Now consider the hyperplane H defined by
X I + . . . + Xn = 1 .
Then H n X is a polyhedron that is the convex hull of a finite set of points
C, where C consists of the vertices of H n X.
Let y E C. It is not hard to see ( by considering Gauss-Jordan elimina­
tion ) that the entries in y are rational. Thus if we multiply y by the least
common multiple of the denominators of its entries, we obtain an integral
point x. Let C' be the collection of integral points thus obtained. Note that
C' represents a set of vectors w 1 , . . . , wz such that every vector in X can be
written as I::!=l aiwi with ai 2: 0 Vi.

{ 2: 1]}.
Consider the "parallelogram"
P' = txxl tx E [O,
x EC'
Then P' is compact. Thus the set L = P' n zn is a finite number of points.
Here L is a finite set of non-negative solutions of Ax = 0. It remains to show
that L generates the full set of non-negative integral solutions of Ax = 0
under the + operation.
160 5 . Triangulated 3-Manifolds

Let z E X n '!IP . Then z = ai w 1 + · · · + alwl for some ai , . . . , al E �+ ·

But then

where LaJ denotes the greatest lesser integer of a. Furthermore, all entries
of the vector z - ( LaiJ w 1 + + Lad wt) are differences of integers, hence
· · ·

themselves integers. In particular, z - ( LaiJ w 1 + + Lad wt) E L. Now · · ·

Since LaiJ W 1 + . . . + Lad Wt is generated by L under the + operation, x n zn

is generated by L under the + operation. D

Definition 5.3.3. A set of fundamental solutions to a Diophantine system

is a finite set of solutions F such that any solution of the system is a linear
combination of elements of F with positive integral coefficients and such that
no element of F is a positive integral linear combination of other elements
of F. The elements of F are called fundamental solutions.
Remark 5.3.4. A set of fundamental solutions can be chosen from the finite
set L described in Theorem 5.3.2 by eliminating redundancies.

Given a 3-manifold M, we can triangulate M and write down the corre­

sponding Diophantine system. Theorem 5.3.2 provides a finite set of solu­
tions that yields a set of fundamental solutions after eliminating redundan­
cies. Each fundamental solution corresponds to a fundamental surface. The
( finitely many ) fundamental surfaces generate all normal surfaces. In other
words, every normal surface is a finite sum of fundamental surfaces under the
+ operation. In the exercises you will show that a normal surface is a fun­
damental surface if and only if it is not the sum of two ( non-empty ) normal
surfaces. Note that a fundamental surface is thus necessarily connected.
Many properties of surfaces can be traced back to the analogous prop­
erties for the fundamental surfaces of which they are the sum. See, for
instance, the results below.

Lemma 5.3.5. Suppose that F is a connected normal surface (considered up

to normal isotopy) in the 3-manifold M. Suppose further that F = G + H,
for G, H normal surfaces in M, and that G, H are chosen (up to normal
isotopy) so that the number of components of G n H is minimal among all
pairs of summands of F. Then G and H are connected.
Exercises 161

Proof. Suppose that H = H1 LJ H2 for H1 , H2 # 0. Set G' = G + H1 . Then

F = G + H = G + H1 + H2 = G' + H2 . Here the components of GnH consist
of the components of G n H1 together with the components of G n H2 . Since
F is connected, neither of the latter two sets are empty. Thus the number
of components of G' n H2 is less than the number of components of G n H.
But this contradicts the minimality of the number of components of G n H
among all pairs of summands of F. Thus G and H are connected. 0

How incompressibility features into this equation is a more difficult ques­

tion, but it was answered by Jaco and Oertel (see [68] ) :
Theorem 5.3.6 (Jaco-Oertel) . Suppose that S is a 2-sided incompressible
normal surface that is positioned so that w (8) is minimal within the isotopy
class of 8. Suppose further that S = 81 + 82 Then both 81 and 82 are
.

incompressible.
Often we are interested in Haken manifolds and hence whether or not
a 3-manifold contains an incompressible surface. It follows from Jaco and
Oertel's Theorem that a 3-manifold contains an incompressible surface if
and only if it contains an incompressible fundamental surface. In the next
chapter we will see how incompressibility can be expressed algebraically.
Existence of an incompressible fundamental surface can then be checked
algorithmically.
Corollary 5.3.7 (Haken) . There is an algorithm to detect whether or not
a 3-manifold contains an incompressible surface.

Exercises

Exercise 1. Prove that there are exactly three types of normal quadrilat­
erals in a 3-simplex. Prove also that any two quadrilaterals of distinct types
must intersect.
Exercise 2. Show that the + operation on normal surfaces is commutative
and associative.
Exercise 3. Show that for normal surfaces P, Q, the Euler characteristic
is additive:
x (P + Q ) = x (P) + x (Q) .

Exercise 4 . Show that a normal surface is fundamental if and only if it is

not the sum of two (non-empty) normal surfaces.
162 5. THangulated 3-Manifolds

Exercise 5*. Let Qi , Q 2 be normal quadrilaterals in the 3-simplex [s] .

Show that regular switches on the 2-dimensional faces of [s] extend to a cut
and paste operation on Qi U Q 2 if and only if Qi and Q 2 are of the same
normal type.

5.4. 2-Spheres *
One application of normal surface theory lies in identifying 2-spheres in 3-
manifolds. Understanding 2-spheres in 3-manifolds provides a foundation
for understanding prime decompositions of 3-manifolds. We here discuss a
result of Kneser, proved in 1929, and a result of Haken that is obtained by
an analogous argument. Combined, the two theorems are known as Kneser­
Haken finiteness. These results are interesting in their own right. As we will
see, Kneser's result is also a crucial step in obtaining prime decompositions.
Definition 5.4. 1 . Let M be a 3-manifold. A punctured M is a 3-manifold
homeomorphic to M\(finite union of pairwise disjoint 3-balls) .
Let S = Si U · · · U Sk be a disjoint union of 2-spheres in a 3-manifold M.
We say that S is an independent set of 2-spheres if no component of M\S
is a punctured 3-sphere.

Example 5.4.2. Let M = Mi # . . . #Mk, where Mi i= 83 for all i . Then

M contains an independent set S = Si U · · · U Sk - i of k 1 2-spheres.
-

Note that a 3-manifold can contain any number of 2-spheres. Simply

look at a regular neighborhood of a point. Its boundary is a 2-sphere. Such
2-spheres and other inessential 2-spheres are not interesting in this context.
Another type of 2-sphere that is uninteresting in this context is the following:
Let Si , 82 be disjoint essential 2-spheres in a connected 3-manifold. Since
the 3-manifold is path-connected, there is a simple arc a connecting Si and
82 . The boundary of a regular neighborhood of Si U 82 U a consists of three
2-spheres, one parallel to Si , one parallel to 82 , and one additional 2-sphere,
S. If Si and 82 are essential and non-parallel, then S is also an essential
2-sphere; it is not parallel to Si or 82 , but Si U 82 U S is not independent!
In 1929, Kneser shed light on how many independent 2-spheres can sit in a
3-manifold (see [80] ) .
Theorem 5.4.3 (Kneser's Theorem) . Let (M, K ) be a triangulated 3-mani­
fold containing no punctured JR.P3 . Suppose that the number of 3-simplices
in K is t . If M contains an independent set of 2-spheres with k components,
each of which is separating, then k < 6 t .
5.4. 2-Spheres 163

More generally, Kneser proved that if (M, K) is a triangulated 3-manifold

and the number of 3-simplices in K is t, then an independent set of 2-spheres
in M has strictly less than 6t + 2 dim H2 ( M; Z2 ) components.
We first prove a lemma:
Lemma 5.4.4. If M contains an independent set of 2-spheres with k com­
ponents, each of which is separating, then for any triangulation (M, K) of
M, (M, K) contains an independent set of k normal 2-spheres.
The proof of this lemma is very similar to the proof of Theorem 5.2. 14.
One important difference is that here M is allowed to be reducible. This
is why Theorem 5.2. 14 describes an isotopy of a surface (that rests on the
irreducibility of the 3-manifold in question) , whereas here we merely state
an existence result. Another difference is that we here use the independence
of the 2-spheres in place of incompressibility.
Proof. Let (M, K) be a triangulation of M. Let S = Si Li · · · u Sk be a
set of independent 2-spheres in M chosen so as to minimize (w(S) , m(S) )
among all sets of k independent 2-spheres. We assume that S is in general
position with respect to the triangulation.
For each 3-simplex [s] E K and each face (J) of [s] , S n l (J) I consists of
simple arcs and simple closed curves. Suppose that S n I (J) I contains simple
closed curves and let c be an innermost such curve. Then c bounds a disk
D in l (J) I that meets S only in its boundary. Let Si be the component of S
containing c. Call the component of M\Si not containing D the "outside"
of Si and the component containing D the "inside" of Si. Here c cuts Si
into two disks, D' and D".
Set s: = D U D' and S:' = D U D". Abusing notation slightly, we also
denote small disjoint push-offs of s: and s:' to the inside of Si by s: and s:'.
See Figures 5.21 and 5.22. Set S' = (S\Si) u s: and S" = (S\Si) u sr,
Claim. Either S' or S" is independent.
Suppose that neither S' nor S" is independent. Then s: lies in the
frontier of a punctured 3-sphere B' and s:' lies in the frontier of a punctured
3-sphere B". Suppose that Si is contained in, say, B'. Then by the Schonflies
Theorem, the 2-sphere Si cuts the punctured 3-sphere B' into two punctured
3-spheres, Bi , B2 . Note that s: lies in the frontier of one of these punctured
3-spheres, say Bi . The other, B2 , does not meet s: and hence is a component
of M \S. (This is where we use the hypothesis that Si is separating.) But
this contradicts the independence of S. Thus B' does not contain Si. It
follows that B' lies on the "inside" of Si. Likewise, B" lies on the "inside"
of Si.
164 5. Triangulated 3-Manifolds

Figure 5 . 2 1 . An innermost disk.

CD

Figure 5.22. The two resulting spheres.

It now follows from the construction that B' and B" meet along D.
Hence B' U B" U D forms a punctured 3-sphere B in M\S. But this also
contradicts the fact that S is independent. Thus either S' or S" is indepen­
dent.
Suppose S' is independent. Then, after a small isotopy that eliminates
c U D from S' n J (f) J , ( w (S'), m(S')) < ( w (S) , m(S) ) . But this contradicts
our choice of S so as to minimize ( w (S) , m(S) ) . Thus for each 3-simplex
[s] E K and each face (!) of [s] the intersection S n J (f) J consists of simple
arcs. The rest of the proof is identical to the proof of Theorem 5.2. 14. 0
We now prove Kneser's Theorem. As we will see, the proof relies heavily
on understanding how a collection of spheres intersects a given 3-simplex.
5.4. 2-Spheres 165

Proof of Kneser's Theorem. Suppose M contains a set of k independent

2-spheres, each of which is separating, with k ;::: 6t. By Lemma 5.4.4, M
contains a set S = 81 LJ • • • LJ Sk of k independent normal 2-spheres, each of
which is separating.
Let [s] be a 3-simplex in K. A component of 81 [s] l\S is "good" if it is an
annulus that contains no vertex in l [s]0 j . At most six components of 81 [s] l \S
are "bad" : Indeed, Sn l [s] I contains at most one type of normal quadrilateral.
If, in addition, it contains all types of normal triangles, there will be six
"bad" components. Otherwise, there will be fewer "bad" components. See
Figure 5.23.

good

Figure 5.23. Good components and bad components.

A component X of M\S is "good" if, for every 3-simplex [s] E K, every

component of X n 81 [s] I is good. At most 6t components of M\S are "bad" .
Since k 2: 6t, M\S has at least 6t + 1 components. Thus there is at least
one "good" component. A "good" component, X, is the union of regions
homeomorphic to {triangle) x I and (quadrilateral) x I. It is not hard to see
that then X = §2 x I. Specifically, the (triangle) x I and (quadrilateral) x I
regions constitute an /-bundle over a surface B and this /-bundle is bounded
by 2-spheres. Thus B is covered by the 2-sphere and must hence be either
a projective plane {in the case of a twisted /-bundle) or a 2-sphere {in the
case of a product) .
We leave it as an exercise to show that the twisted I-bundle over IRP2 is
a punctured IRP3 . It follows that X = §2 x I, but this contradicts the fact
that the set S is independent. Hence k < 6t. 0

To recover the stronger form of Kneser's Theorem, we need to consider

the case where X is a twisted /-bundle over the projective plane. In this
case X is bounded by a 2-sphere. Hence it is a summand of M. Moreover,
166 5. Triangulated 3-Manifolds

its homology is a direct summand of the homology of M and contributes to

dim H2 (M; Z2 ) .
An analogous argument also proves the following theorem:
Theorem 5.4.5 (Haken's Theorem) . Let (M, K) be a triangulated 3-mani­
fold. Suppose that the number of 3-simplices in K is t. If M contains a set of
surfaces F = F1 U · · · LI Fk such that no component of M\F is homeomorphic
to (surface) x l, then k ::; 6t + 2 dim H2 (M; Z 2 ) .
Actually, Haken gave a bound involving 61t. Many arguments for this
theorem have been given and the bound has improved over time. Combined,
the above two theorems are known as Kneser-Haken finiteness.

Exercises

Exercise 1. Show that the twisted I-bundle over IR.P 2 is a punctured IR.P3 .
Exercise 2. How many non-isotopic incompressible surfaces are contained
in the 3-torus? How large can a collection of disjoint non-isotopic incom­
pressible surfaces in the 3-torus be?
Exercise 3. Generalize Haken finiteness to the case of incompressible sur­
faces with boundary properly embedded in 3-manifolds with boundary. Cau­
tion: There is a hypothesis that must be added to obtain a generalization.

5 . 5 . Prime Decompositions

We will show that every orientable 3-manifold has a prime decomposition

that is unique up to a reordering of its factors. (A prime 3-manifold is con­
sidered to have a prime decomposition with just itself as a factor.) Beware
that there are subtleties concerning the "uniqueness11 of this prime decom­
position. The prime decomposition is "unique" in the sense that the factors
are unique up to reordering. The decomposing spheres for the factoriza­
tion are far from unique (even up to isotopy) . The existence of a prime
decomposition can be established using Kneser's Theorem:
Theorem 5.5. 1 . Every compact 3-manifold can be expressed as a connected
sum of a finite number of prime factors.
Proof. Let M be a compact 3-manifold and let (M, K) be a triangulation
of M. Let S be an independent set of k normal 2-spheres, each of which is
separating, chosen so that k is maximal among all such sets. This is possible
by Kneser's Theorem. Then
M\ S = M1 LI . . . U Mk + l ·
5.5. Prime Decompositions 167

Denote the compact 3-manifold obtained by attaching 3-balls along the fron­
tier of Mi by Mi . We need to show that Mi is prime. Indeed, suppose to
the contrary that there is a separating essential 2-sphere S in Mi . Then
S LI S is a set of k + 1 normal 2-spheres, each of which is essential and
separating. Since S was chosen to have a maximal number of components
among independent sets of 2-spheres, each of which is separating, S LI S is
not independent. Thus a component of Mi \ S is a punctured 3-sphere. But
then the corresponding component of Mi \ S is a 3-ball. Since S is essential,
this is a contradiction. Hence Mi is prime. Thus
M = Mi # · · · # Mk+ l
is as required. D

Definition 5.5.2. A component of an independent set of separating 2-

spheres is called a decomposing sphere.
Recall the discussion of the operation of connected sum in Chapter 1.
In considering prime decompositions of 3-manifolds, we must take into ac­
count that there can be several independent sets of 2-spheres providing a
prime decomposition of a given 3-manifold. I.e., collections of decomposing
spheres are not unique. Nevertheless, prime decompositions of 3-manifolds
are unique in the sense expressed in the following theorem:
Theorem 5.5.3 (Uniqueness of prime decompositions ) . Let M be a compact
orientable 3-manifold. If M = Mi # · · · # Mk = Ni # · · · # Nz are prime
decompositions, then k = l and, after reordering, Mi is homeomorphic to
Ni, for i = 1, . . . , l .
Before proving the uniqueness of prime decomposition, we prove a lemma:
Lemma 5.5.4. Let E be a non-separating 2-sphere and let S be a separating
2-sphere in the 3-manifold M. Let c be an innermost component of E n S
in S. Let D be the component of S\c that is disjoint from E and let D', D"
be the components of E\c. Set E' = D U D' U c and E" = D U D" U c. Then
either E' or E" is non-separating.
Proof. Recall Remark 3.3.3. Since E is non-separating, there is a simple
closed curve a in M that meets E exactly once and is transverse to E U S.
The number of points in a n (E' U E") is equal to the sum of the number
of points in a n E' = a n (D U D') and a n E" = a n (D U D") . Modulo 2,
this sum is equal to the sum of the number of points in a n D' and a n D"
which is equal to a n (D' U D") and hence equal to the number of points in
a n E. Thus there are an odd number of points in either a n E' or a n E",
say in a n E'. The existence of the simple closed curve a that meets E' in
an odd number of points shows that E' is non-separating. D
168 5. Triangulated 3-Manifolds

Proof of Theorem 5.5.3.

Case 1 . M contains no non-separating 2-spheres.
In this case, each Mi and each Nj is irreducible. Let S = Si U U Sk - i
· · ·

be a set of 2-spheres such that M\S = Mi LI · · · U Mk , where Mt is a

punctured copy of Mi for i = 1, . . . , k, and let :E be a 2-sphere such that
M\:E = Ni LI M', for Ni a once-punctured Ni .
We may assume that :E and S have been chosen so that the number of
components of :E n S is minimal over all spheres isotopic to :E and all sets of
2-spheres that decompose M into 3-manifolds homeomorphic to Mi , . . . , Mk .
Suppose that :E n s #- 0. Let c be an innermost component of :E n S in :E. Let
D be the component of :E\c that is disjoint from S. In particular, D c Mj
for some j . Let Si be the component of S containing c.
Note that Si\c consists of two disks, D', D". Set S' = D U D', S" =
D U D". Recall that Mj is irreducible. Thus S', S" bound 3-balls B', B" in
Mj . If the interiors of B', B" are disjoint, then Si = 8(B' U B") , but this
is impossible, since Si is a decomposing sphere. Thus, since Mj is not §3 ,
it must be the case that either B' c B" or B" c B', say B' c B". Now
we construct a new set of 2-spheres S* by setting s; = S" and Sj = Si
for j #- i . Then S* = Si LI · · · LI S'k - i also has the property that M\S*
is the disjoint union of punctured copies of Mi , . . . , Mk but the number of
components of :EnS* is strictly less than the number of components of :EnS.
This contradicts our choice of :E and S. Thus :E n S = 0.
Suppose now that a component of S, say Sc, lies in Ni - Since Ni is
irreducible, Sc bounds a 3-ball in Ni . On the other hand, since Sc is essential
in M, it does not bound a 3-ball in M and hence does not bound a 3-ball
in Ni . It follows that Sc bounds a punctured 3-ball in Ni with frontier a
subset of 8Ni n :E. A component, M; , of M\S adjacent to Sc then contains
a punctured copy of Ni . Likewise, if no component of S lies in Ni , then
Ni C M; for some c. It follows that for j = 1, . . . , l, a punctured copy of
Nj lies in Mc; for some Cj . Furthermore, since Mc; is prime, Mc; does not
contain punctured copies of other factors. By a symmetric argument, for
i = 1, . . . , k, a punctured copy of Mi lies in NPi for some Pi and NPi does not
contain punctured copies of other factors. Thus k = l and, after reordering,
Mi = Ni .
Case 2. M contains a non-separating 2-sphere S .
Let S be as above. We may assume that S is chosen so that the number of
components of S n S is minimal over all separating 2-spheres in M. Suppose
that S n S #- 0. Let c be an innermost component of S n S in S. Proceeding
as in Lemma 5.5.4, we obtain S', S" such that S' or S", say S', is non­
separating. However, the number of components of S'nS is strictly less than
5. 6. Recognition Algorithms 169

the number of components of S n S, contradicting the assumed minimality.

Thus S n S = 0. Let Mt be a component of M\S containing the non­
separating 2-sphere S. By Theorem 3.3.4, Mi = §2 x § 1 . Similarly, we show
that for some j , Nj = §2 x § 1 . Thus after reordering, Mi = Ni. Splitting
off summands equal to §2 x § 1 reduces the argument to the case where M
contains no non-separating 2-sphere. D

For non-orientable 3-manifolds prime factorizations are not unique. More

specifically, if M is non-orientable and M = M1 # (§2 x § 1 ) , then it is also
the case that M = Mi # (§2 x § 1 ) and vice versa. However, this prime fac­
torization becomes unique if we decree that reducible prime summands for
non-orientable 3-manifolds always be §2 x § 1 • See [63] .
Recall that a 3-manifold is non-orientable if and only if it contains a
submanifold homeomorphic to (Mobius band) x I. Twisted I-bundles are
interesting when considering the question of orientation. The Mobius band
can be thought of as a twisted I-bundle over the circle. In particular, the
circle is orientable but the Mobius band is not. The projective plane is not
orientable, but the twisted I-bundle over the projective plane is orientable.
Its boundary is a 2-sphere.

Exercises

Exercise 1. Which prime 3-manifold contains the twisted I-bundle over

the projective plane?
Exercise 2. How many different decomposing spheres does (§2 x § 1 ) #1f3
contain (up to isotopy)?

5 . 6 . Recognition Algorithms
Corollary 5.3. 7 establishes the existence of an algorithm to determine wheth­
er an irreducible 3-manifold contains an incompressible surface. In [52] ,
Haken discusses how to use normal surface theory to decide whether or not
two Haken 3-manifolds are homeomorphic. See also [156] , [61] , or [96] .
Since the 3-sphere is not a Haken manifold, that discussion does not address
the question as to whether it is possible to recognize the 3-sphere. It is
in fact possible to recognize the 3-sphere, but this was proved much later
by Rubinstein. Rubinstein sketched an argument that was completed by
Thompson; see [152] . Matveev gave an alternative treatment; see [96]. The
algorithm relies on an extension of normal surface theory called "almost"
normal surface theory. A recognition algorithm for all closed orientable
3-manifolds was given by Sela (see [147]) and Manning (see [92]).
170 5. Triangulated 3-Manifolds

Theorem 5.6. 1. There is an algorithm to determine whether or not a closed

3-manifold is §3 .
Definition 5.6.2. A disk in a 3-simplex that is bounded by a normal curve
of length 8 is called an almost normal disk. A surface in a triangulated
3-manifold is almost normal if it is normal away from exactly one almost
normal disk.
Remark 5.6.3. In some contexts, almost normal surfaces are defined in a
broader sense. There, they meet exactly one 3-simplex in normal disks and
either one almost normal disk or two normal disks that are joined by a tube.
We will not use this more general definition here.
Remark 5.6.4. Note that the set of almost normal surfaces is disjoint from
the set of normal surfaces!
A crucial step in the proof of Theorem 5.6. 1 establishes that a closed
triangulated 3-manifold that contains an almost normal 2-sphere and no
normal 2-spheres is necessarily §3 . As it turns out, if an irreducible trian­
gulated 3-manifold (M, K) contains a normal 2-sphere, then M admits a
triangulation (M, K') with fewer 3-simplices. Conversely, a triangulation of
§3 contains either an almost normal 2-sphere or a normal 2-sphere. Whether
or not a 3-manifold contains an almost normal 2-sphere can be determined
via an algorithm similar to Haken's algorithm.
The following theorem and corollary rely heavily on Theorem 5.6. 1.
Theorem 5.6.5. If the closed triangulated 3-manifold (M, K) contains an
incompressible 2-sphere or projective plane, then it contains an incompress­
ible fundamental 2-sphere or projective plane.
Proof. Let F be the 2-sphere or projective plane. By isotoping F as in
Theorem 5.2.14 we arrange for F to be normal. We may assume that F
has been chosen so that w(F) is minimal among all incompressible normal
2-spheres and projective planes in M and that F is in general position with
respect to K.
Suppose F is not fundamental. By Lemma 5.3.5 there are connected
normal surfaces G, H such that F = G + H, w(G) > 0, w(H) > 0, w(F) =
w(G) + w(H) , and x (F) = x (G) + x (H) . Moreover, by Theorem 5.3.6, G
and H are incompressible.
Case 1 . x (F) = 2; i.e. , F is a 2-sphere.
Up to renaming of G, H, there are only two possibilities: ( 1) x ( G) = 2
and x (H) = 0 or (2) x (G) = 1 and x (H) = 1. In both of these cases, G
is either a 2-sphere or projective plane and w(G) < w(F) . This contradicts
the minimality of w(F) .
5. 6. Recognition Algorithms 171

Case 2. x(F) = 1; i.e. , F is a projective plane.

Up to renaming of G, H, there are only two possibilities: (1) x(G) = 2 and
x( H ) = - 1 or (2) x(G) = 1 and x( H ) = 0. Thus G is either a 2-sphere or
projective plane and w(G) < w(F) , contradicting the assumed minimality
of w(F) .
Thus F is not the sum of other incompressible surfaces under the +
operation. It follows that F is fundamental. In particular, M contains a
fundamental 2-sphere or projective plane. 0

Corollary 5.6.6. There is an algorithm to decide whether or not a closed

3-manifold contains an incompressible 2-sphere or projective plane.
The idea for the proof of this corollary is the following: Solve the Dio­
phantine system corresponding to the 3-manifold as in Theorem 5.3.2. Re­
call that this gives a finite set of solutions that generates all solutions of
the Diophantine system. Moreover, this finite set of solutions contains the
set of fundamental solutions. These can then be found by discarding those
elements of this finite set that are linear combinations with non-negative
integral coefficients of other elements. Now for each fundamental solution,
consider a representative of the corresponding normal isotopy class of fun­
damental surfaces and compute its Euler characteristic. If the Euler char­
acteristic of a fundamental surface is 1 , then the fundamental surface is a
projective plane. If it is 2, then it is a 2-sphere. In each such case, check
whether or not the projective plane or 2-sphere is incompressible. In the case
of the 2-sphere, this latter step requires the 3-sphere recognition algorithm,
Theorem 5.6. 1.
In a similar vein, Haken proved the following theorem:
Theorem 5.6.7 (Haken) . If the triangulated 3-manifold (with boundary)
M contains an essential disk with specified boundary and if it contains no
projective planes, then it contains a fundamental disk with boundary isotopic
to the boundary specified.
The system of Diophantine equations employed in the proof of this the­
orem must incorporate the information concerning the specified boundary.
Specifically, if we are concerned with disks with specified boundary c, we
find a triangulation for M such that each 2-dimensional face of a 3-simplex
lying in the boundary of M meets c in at most one normal arc. This can
easily be accomplished by choosing any given triangulation of M, isotoping
c, and taking barycentric subdivisions of the triangulation if necessary. Ev­
erything else remains the same. Note that the fundamental solutions of this
"relative" system of Diophantine equations will correspond to fundamental
172 5. Triangulated 3-Manifolds

surfaces that are either closed surfaces or have boundary components that
lie in the boundary specified.
The application Haken had in mind is the following:
Corollary 5.6.8. There is an algorithm to decide whether or not a knot
K c §3 is the unknot.
Proof. Let 17(K) be an open regular neighborhood of K and set C(K) =
§3 \17(K) . By Exercise 1 of Section 4.3, we know that a knot is the unknot
if and only if C(K) contains an essential disk with longitudinal bound­
ary. Since the complement of a knot in §3 contains no projective planes,
Haken's Theorem implies that containing an essential disk with longitudinal
boundary is equivalent to containing a fundamental disk with longitudinal
boundary. (Since a disk with longitudinal boundary is non-separating, it
is necessarily essential.) The existence of a fundamental disk with longitu­
dinal boundary can be established or ruled out by triangulating the knot
complement appropriately and finding the fundamental solutions of the cor­
responding Diophantine system. 0

In [85) and [84] , Li used the theory of almost normal surfaces and other
techniques to prove a conjecture of Waldhausen: An irreducible atoroidal
3-manifold admits only finitely many distinct Heegaard splittings of a given
genus.
Exercises
Exercise 1 . Show that if F = G + H and H is an inessential sphere, then
F is isotopic to G.
Exercise 2. Devise an algorithm to generate all Seifert surfaces of a knot
complement.
Exercise 3. Devise an algorithm to calculate the genus of a knot K (that
is, the smallest possible genus of a Seifert surface for K) .
5 . 7. PL Minimal Surfaces * *

Triangulations can be used to endow manifolds with metric structures sim­

ilar to those constructed in the context of differentiable manifolds. For in­
stance, the sophisticated analytic theory of minimal surfaces has an analog
in the context of triangulated 3-manifolds. All that is needed here is some
understanding of metric structures on surfaces and in fact only on triangles.
Because this provides a more accessible version of minimal surface theory,
we include a brief overview here. PL minimal surface theory was developed
by Jaco and Rubinstein. For more details, see [70] , [69] , and [38) .
5. 7. PL Minimal Surfaces 173

Given a compact triangulated 3-manifold (M, K) , the 2-simplices can

be endowed with a metric g of negative curvature with geodesic edges and
with vertices at infinity. Moreover, this can be accomplished in such a
way that the metrics on these simplices match up along 1-simplices where
they meet and the resulting path-metric on K 2 is complete. Call such a
triple (M, K, g) a 3-manifold with metric triangulation. Jaco and Rubinstein
employ hyperbolic metrics on the 2-simplices of the triangulation but point
out that this is not necessary.
Recall that the weight, denoted by w(F) , of a surface F in (M, K, g) is
the number of points in F n K 1 . The measure of F, denoted by l(F) is the
total length l (F n K2 ), that is, the sum of the lengths of all the arcs in which
F meets the 2-simplices of the triangulation. (Do not confuse length with
measure.) The surface F is PL-minimal if it minimizes l for small isotopies
of F. It is PL least area if it minimizes (w, l) in its isotopy class (according
to the dictionary order) .
We list several theorems about PL least area surfaces. All are variations
of theorems in [69] . We have formulated them in terms of isotopy classes
of embedded surfaces, though the theory has traditionally been formulated
in terms of homotopy classes of immersed surfaces. (The arguments carry
over verbatim to the context of isotopy classes of embedded surfaces.)
The following existence and uniqueness result provides a canonical choice
of representative for an isotopy class of surfaces. This type of canonical
representative proves useful in many arguments in 3-dimensional topology.
For a short proof of the uniqueness part of this theorem, see [113] .
Theorem 5 . 7. 1 . Let E be a compact orientable irreducible 3-manifold en­
dowed with a metric triangulation and let F be a compact incompressible
surface in E. Then there exists a unique PL least area representative in the
isotopy class of F.
The theorem below tells us that lifts of PL least area surfaces from a
3-manifold to a covering space are PL least area and also that projections
of PL least area surfaces from the covering space to the 3-manifold are PL
least area, provided that the projection is injective on the surface.
Theorem 5.7.2. Let E be a compact orientable irreducible 3-manifold en­
dowed with a metric triangulation and let M be any covering space of E.
Let FE be a 2-sided compact incompressible surface in E whose lift FM to
M is also compact. Then FE is PL least area if and only if FM is PL least
area.
Minimality arguments are used frequently in 3-dimensional topology,
most notably arguments relying on surfaces that have been isotoped to in­
tersect in a minimal number of components. The two results below establish
174 5. TI:iangulated 3-Manifolds

the fact that PL least area surfaces necessarily intersect in the smallest num­
ber of components possible in their isotopy classes.
Theorem 5. 7.3. Let E be a compact orientable irreducible 3-manifold en­
dowed with a metric triangulation and let F1 and F2 be PL least area incom­
pressible surfaces in E with disjoint representatives in their isotopy classes.
Then either Fi and F2 are disjoint or they coincide.
Theorem 5. 7 .4. If F1 , F2 are PL least area incompressible surfaces in the
compact orientable irreducible 3-manifold E endowed with a metric triangu­
lation, then their number of components of intersection is minimal over all
representatives of the isotopy classes of the surfaces.

Exercises
Exercise 1 *. Use PL minimal surface theory to prove that a covering space
of a 3-manifold M is irreducible if and only if M is irreducible.
Exercise 2. Prove that the double curve sum of two minimal genus Seifert
surfaces contains two minimal genus Seifert surfaces.
Exercise 3. Suppose that the knot K has only finitely many isotopy classes
of minimal genus Seifert surfaces. Prove that a PL least area minimal genus
Seifert surface for K whose area is strictly less than that of all other Seifert
surfaces for K is disjoint from at least one non-isotopic Seifert surface of K.
 Chapter 6

Heegaard Splittings

A Heegaard splitting is, roughly speaking, a splitting of a 3-manifold into

two simple pieces called handlebodies. Below, we will state a theorem of
Bing that establishes that every 3-manifold admits such a splitting. It turns
out, however, that "the whole is more than the sum of its parts" . Specifi­
cally, though the two handlebodies are well understood as 3-manifolds, their
relative positioning remains a worthy object of study. For the origins of the
subject, see [60] . For an overview, see [133] .
In this chapter, we will be interested in Heegaard splittings and gener­
alized Heegaard splittings. We will see basic theorems concerning Heegaard
splittings, structural and classification theorems for Heegaard splittings, and
generalized Heegaard splittings and applications of such theorems.

6 . 1 . Handle Decompositions

A handle decomposition of a 3-manifold is a particular way to build the

3-manifold. Handle decompositions exist for manifolds of any dimension.
For smooth manifolds, this is a consequence of the fact that every manifold
admits a Morse function. For more information, see Appendix B.
Definition 6.1.1. In the 3-dimensional setting, a k-handle is a 3-ball,
thought of as [O, 1] 3 , that is attached (to some preexisting submanifold)
along [O, 1 ] 3 k x 8[0, l] k . (More generally, in the n-dimensional setting, a
-

k-handle is an n-ball, thought of as [O, l]n, that is attached (to some preex­
isting submanifold) along [O, l]n- k x 8[0, l] k .) See Figure 6. 1 .
Concretely, a 0-handle is a 3-ball attached to the empty set. We think
of this as a 3-ball appearing out of nowhere. A 1-handle is a 3-ball attached
-

175
176 6. Heegaard Splittings

along [O, 1] 2 x 8 [0, 1] . We think of this as a solid cylinder attached along its
two bounding disks. A 2-handle is a 3-ball attached along [O, 1] x 8[0, 1] 2 .
We think of this as a solid cylinder attached along the annular portion of its
boundary. Finally, a 3-handle is a 3-ball attached along 8 [0, 1] 3 , i.e. , along
its entire boundary. We think of this as a 3-ball that is being used to fill an
existing hole.

Figure 6 . 1 . Two 0-handles and two 1-handles.

Definition 6 . 1 .2. A handle decomposition of a 3-manifold M is a sequence

of 0-handles, I-handles, 2-handles, and 3-handles whose union is M.
In Appendix B we state a theorem that guarantees the existence of Morse
functions on a manifold. We also explain how this guarantees the existence
of handle decompositions of the manifold. In addition, it follows from tech­
nical results on Morse functions that manifolds have handle decompositions
in which all 0-handles are attached before all I-handles, all I-handles are at­
tached before all 2-handles, all 2-handles are attached before all 3-handles,
and so on. We will sometimes, but not always, have such handle decompo­
sitions in mind. The subtle issue of which handle decompositions describe
the same manifold is treated in [ 77] .
Definition 6.1 .3. The core of a 3-dimensional k-handle is { ! } 3 - k x [O, l] k .
The cocore is [O, 1] 3 - k x 0 } k . See Figures 6.2 and 6.3.

Remark 6.1.4. In the context of 3-manifolds, a k-handle is dual to a (3-k)­

handle in the following sense: A k-handle in a closed 3-manifold is a 3-ball
attached to some preexisting submanifold along a portion of its boundary.
6. 1 . Handle Decompositions 177

Figure 6.2. The core of a 3-dimensiona.l 1-ha.ndle.

Figure 6.3. The cocore of a 3-dimensiona.l 1-ha.ndle.

The remaining portion of the 3-manifold will be attached to the boundary

of the resulting 3-manifold.
Given a handle decomposition of the 3-manifold M (thought of as a
procedure for building M) , we can "reverse the procedure" to obtain the
dual handle decomposition: We attach what were formerly 3-handles first,
but they are now attached to nothing and hence are 0-handles. We then
attach what were formerly 2-handles, but these are now attached along two
disks and hence are 1-handles, etc. The core of a k-handle in the original
handle decomposition is the cocore of the corresponding (3 - k )-handle when
we reverse the procedure.
Definition 6.1.5. A handlebody is a compact connected orientable 3-mani­
fold with boundary that possesses a handle decomposition consisting of
0-handles and 1-handles. The genus of a handlebody is the genus of its
boundary.
A collection of meridian disks for a handlebody is a collection of disks
that cut the handlebody into 3-balls.
I78 6. Heegaard Splittings

Example 6.1.6. The handlebody pictured in Figure 6. I is a genus I han­

dlebody. We also refer to this handlebody as a solid torus.
Every handle body has a collection of meridian disks: Start with a handle
decomposition consisting of only 0-handles and I-handles and take the set
of cocores of the I-handles in this handle decomposition. See Figure 6.4.

Figure 6.4. A set of meridian disks for a handlebody.

Definition 6.1.7. A Heegaard splitting of a closed 3-manifold M is a de­

composition M = V U s W such that
• V, W are handlebodies and
• S = aV = aW.
Here S is called the splitting surface of M = V Us W.
Two Heegaard splittings are considered equivalent if their splitting sur­
faces are isotopic. The genus of a Heegaard splitting is the genus of S.
We think of the handlebodies V, W being glued along S to create M.
Example 6.1 .8. The 3-sphere has a genus 0 Heegaard splitting: We think of
§3 as consisting of those points in JR4 that have distance I from the origin.
The subspace given by w = 0 intersects §3 in a 2-sphere. This 2-sphere
separates §3 into two 3-balls, each a handlebody of genus 0.
Example 6.1 .9. If two solid tori T1 and T2 are glued together in such a
way that a meridian of T1 is identified to a longitude of T2 , then the result
is §3 . This describes a genus I Heegaard splitting of §3 .
6. 1 . Handle Decompositions I79

Example 6.1.10. As above, consider two solid tori T1 , T2 . If the two solid
tori are glued together in such a way that a meridian of T1 is identified to
a (p, q ) -torus knot of T2 , then the result is a lens space, more specifically, a
genus I Heegaard splitting of the lens space L (p, q ) .
Example 6. 1 . 1 1 . We can describe a Heegaard splitting of the 3-torus § 1 x
§ 1 x § 1 as follows: We think of the 3-torus as a quotient space obtained by
identifying opposite sides of a cube. Denote a regular neighborhood of the
I-skeleton of the cube by V and denote the closure of its complement by W.
Then both V and W are handlebodies. The two handlebodies meet in the
surface S of genus 3. See Figure 6.5.

Figure 6 . 5 . A Heegaard splitting of the 3-torus.

The following theorem is due to E. Moise; see [101] .

Theorem 6.1.12 ( Moise ) . Every closed orientable 3-manifold admits a Hee­
gaard splitting.
We offer two proofs of this theorem:
Proof (Version 1 ) . This theorem follows from two facts: ( I ) Every 3-
manifold admits a handle decomposition in which all 0-handles are attached
before all I-handles, which in turn are attached before all 2-handles, which
in turn are attached before all 3-handles; ( 2 ) 2-handles are dual to I-handles
and 3-handles are dual to 0-handles.
The 0-handles and I-handles in a handle decomposition of this type
provide one handlebody, and the 2-handles and 3-handles, dually, provide
the other. 0

Proof (Version 2) . Let M be a closed 3-manifold. Then M admits a

triangulation (M, K) . Set V = N(K 1 ) . Then V is a handlebody. It is
180 6. Heegaard Splittings

not too hard to see that the closure, W , of the complement of V is also a
handlebody. See Figure 6.6. D

Figure 6.6. Local picture of Heegaard splitting.

For a broader view of Heegaard splittings and how they relate to other
structures on 3-manifolds, see [83] and [90] .
Exercises
Exercise 1 . Prove that §3 is the only 3-manifold with a genus 0 Heegaard
splitting.
Exercise 2. Prove that all handlebodies of genus g are homeomorphic.
Exercise 3 . Prove that, for g > 0, there are handlebodies of a genus g in
§3 that are not isotopic.
Exercise 4. Prove that the construction in Version 2 of the proof of Theo­
rem 6. 1 . 12 describes a Heegaard splitting of the 3-manifold.
6 . 2 . Heegaard Diagrams

You showed in the exercises for the previous section that all handlebodies
of a given genus are homeomorphic. It follows that, given a Heegaard split­
ting M = V Us W of genus g, we can visualize the genus g handlebody
V in IR.3 . To reconstruct M together with its Heegaard splitting we need
only know how to attach W to the outside of V. This reduces to under­
standing how a collection of meridian disks for W is attached to aV. (By
Alexander's Theorem, there is then only one way to attach the 3-handles
W\ (meridian disks) .)
Note that the embedding of V in IR.3 specifies a collection of meridian
disks for V (though not uniquely) . These, in turn, specify (and are specified
6.2. Heegaard Diagrams 181

by ) , a collection of curves on aV . This gives us a framework in which to

represent the curves along which the collection of meridian disks for W is
attached along aV .

Figure 6.7. A Heegaard diagram for the 3-sphere.

Definition 6.2. 1 . A Heegaard diagram is a closed orientable surface S of

genus g that is equipped with two collections of essential simple closed curves
{ V 1 ' ' Vg } and { Wi , . . . ' Wg } .
. . •

Remark 6.2.2. We think of obtaining a 3-manifold M by attaching 2-

handles to the inside of S x I along { v1 , . . . , v9 } and to the outside of S x
I along { w1 , . . . , w9 } . Attaching 2-handles to the inside of S x I along
{ v1 , . . . , v9 } yields a handlebody V . An embedding of V in JR.3 specifies a
collection of curves { v� , . . . , v� } , though not uniquely. However, attaching
2-handles to the inside of aV x I = S x I along any such collection of curves
and 2-handles to the outside of S x I along curves { w1 , . . . , w9 } yields the
same 3-manifold. For this reason we only indicate one collection of curves,
{ w i , . . . , w9 } , in Figures 6.7 and 6.8.
Example 6.2.3. Figure 6.7 gives a Heegaard diagram of §3 .
Example 6.2.4. Figure 6.8 gives a Heegaard diagram of the lens space
£(2, 1 ) .

Figure 6.8. A Heegaard diagram for real projective 3-space.

A more traditional way to exhibit a Heegaard diagram is to obtain a

punctured surface by cutting along the curves v1 , . . . , v9 and presenting this
182 6. Heegaard Splittings

punctured surface along with the remnants of w1 , . . . , w9 • The boundary

components of the resulting punctured surface are then numbered so as to
allow a reconstruction of the genus g surface.
Example 6.2.5. Figure 6.9 gives a traditional Heegaard diagram of the
3-torus.
'
·' · , ,·
,·' ,. ·
i ;
- - - ...
-
i ,. ...

! !
ii .'/ ,,' : ...
I I1
-
-
-
- ...
...
' ,
' ' \

d>ii J;� G
I I \ \
I I ,.-., ,

1
CD\ \ \
- Ij / I I
ii G
II,'
I I

\-::����:=�����:.:.>/ \ : - - - - - - -; : /
--
...
... - .. - - ... ,

Figure 6.9. A traditional Heegaard diagram for the 3-torus.

Exercises

Exercise 1 . Draw a Heegaard diagram representing §3 .

Exercise 2. Draw a Heegaard diagram representing a manifold obtained
by Dehn surgery on the figure 8 knot.
Exercise 3. Prove that IR.P3 is homeomorphic to L(2, 1 ) .
6 . 3 . Reducibility and Stabilization

Given two 3-manifolds with Heegaard splittings, we can find a Heegaard

splitting of the connected sum of the two 3-manifolds. We do this by consid­
ering the pairwise connected sum of the 3-manifolds relative to their splitting
surfaces. Furthermore, given one 3-manifold with a Heegaard splitting, we
can find other Heegaard splittings for this manifold. This is accomplished,
for instance, by the process, described below, known as stabilization.
Definition 6.3. 1. Let M = V Us W be a Heegaard splitting and let §3 =
V' Ur W' be the standard genus 1 Heegaard splitting of §3 . The pairwise
connected sum (M, 8) # (§3 , T) defines a Heegaard splitting M = V U 3 W
called an elementary stabilization of M = V Us W. A Heegaard splitting is
called a stabilization of M = V Us W if it is obtained from M = V Us W by
performing a finite number of elementary stabilizations.
6.3. Reducibility and Stabilization 183

� - -

0 0

Figure 6.10. A stabilization.

Theorem 6.3.2 (Reidemeister-Singer) . Any two Heegaard splittings of a

3-manifold M become equivalent after a finite number of stabilizations.
This theorem can be proved by considering Morse functions. In Appen­
dix B, we discuss how Heegaard splittings correspond to Morse functions.
As it turns out, any two Morse functions are related by a sequence of two
moves, namely stabilization and exchanging levels of critical points. The lat­
ter can be understood via handle decompositions: The order in which two
handles in the handle decomposition are attached is interchanged. Thus this
latter move changes the handle decomposition corresponding to the Morse
function but not the corresponding Heegaard splitting. Morse functions were
studied extensively by J. Cerf in his analysis of smooth real-valued functions
on smooth manifolds in [26] .
We will consider two important properties for Heegaard splittings. These
properties will allow us to establish two theorems. One consequence will be
that a Heegaard splitting of a connected sum of 3-manifolds factors into
Heegaard splittings of the summands.
Definition 6.3.3. A Heegaard splitting M = V U s W is reducible if there
is an essential simple closed curve c c S and disks D c V, E c W with
aD = 8E = c. Alternatively, M V Us W is reducible if there is a 2-sphere
=

E in M such that E n S is an essential simple closed curve. A Heegaard

splitting is irreducible if it is not reducible.
A Heegaard splitting M = V Us W is weakly reducible if there are disks
D c V, E c W such that aD, {)E are essential in S and aD n aE = 0. A
Heegaard splitting is strongly irreducible if it is not weakly reducible.
Note that a reducible Heegaard splitting is weakly reducible since aD can
be isotoped to be disjoint from 8E. It follows that a strongly irreducible
Heegaard splitting is irreducible. Strong irreducibility has proved to be
a useful concept in the study of Heegaard splittings. In many ways, the
184 6. Heegaard Splittings

splitting surface of a strongly irreducible Heegaard splitting behaves like an

incompressible surface.
Let us briefly consider the Heegaard splittings we encountered in the
previous section. The Heegaard splitting of §3 of genus 0 is irreducible.
The Heegaard splitting of §3 of genus 1 is also irreducible. The Heegaard
splittings described for lens spaces are irreducible unless the 3-manifold in
question turns out to be §2 x § 1 . In the latter case the Heegaard splitting is
reducible. The genus 3 Heegaard splitting of the 3-torus is irreducible but
weakly reducible. See Figure 6. 1 1 .

Figure 6 . 1 1 . The genus 3 Heegaard splitting of the 3-torus is weakly reducible.

Interestingly, the Heegaard splitting of genus 1 of §3 is stabilized, though

it is irreducible. It is the unique example of a stabilized Heegaard splitting
that is irreducible. The standard Heegaard splitting of §2 x § 1 , on the other
hand, is unstabilized, though it is reducible. It is the unique example of a
reducible Heegaard splitting that is not stabilized. You will establish these
two facts in the exercises.
Lemma 6.3.4. An incompressible and boundary incompressible surface in
a handlebody is a disk.
Note that the statement is false if we drop the assumption of boundary
incompressibility. (For instance, handlebodies of genus at least one contain
incompressible boundary parallel annuli.)
Proof. Let H be a handlebody of genus g . We proceed by induction on g.
Note that an incompressible surface in a 3-ball must be a disk. See Example
3.4.5. So suppose g > 0. Let S be an incompressible surface in H and let D
be a component of a set of meridian disks for H. Then S can be isotoped to
be disjoint from D by standard innermost disk and outermost arc arguments.
Thus our incompressible and boundary incompressible surface sits in H\D,
6.3. Reducibility and Stabilization 185

a handlebody of lower genus, where it is still incompressible and boundary

incompressible. D

The following theorem is due to Haken. It is one of the fundamental the­

orems concerning Heegaard splittings. One consequence of this theorem is
that a Heegaard splitting of a connected sum of 3-manifolds can be factored
into Heegaard splittings of the summands.
Theorem 6.3.5 (Haken) . Suppose M is a reducible 3-manifold and M =
V Us W is a Heegaard splitting. Then M = V Us W is reducible.
Proof. Let S be an essential 2-sphere in M and suppose that S is chosen
so that the number of components, # I S n S I , of S n S is minimal over all
essential 2-spheres in M. You will prove in the exercises that handlebodies
are irreducible; thus #I S n S I > 0. Assuming # IS n S I > 0, we reason as
follows:
Claim 1. S n V is incompressible in V and S n W is incompressible in W.
Suppose that S n V, say, is compressible in V. Then there is a simple
closed curve c E S that does not bound a disk in S n V but bounds a
disk D c V that is disjoint from S. In particular, S \ c = S1 Li S2 . Set
Si = Si U D U c .
Note that
#I S n S I = # I S1 n S I + # I S2 n s1 .
Furthermore, since c does not bound a disk in S n V,
#I S1 n s1 > o, # I S2 n S I > o .
If S1 is inessential (in M) , then S2 is isotopic to S; see the proof of
Lemma 3.2.3. This implies that S2 is essential in M; yet
# I S2 n S I < # I S n S I ,
so this violates our choice of S, chosen so as to minimize # IS n S I over all
essential 2-spheres in M. Thus S1 is essential, but
#I S1 n S I < #I S n S I ,
again violating our choice of S. This proves the claim.
By Lemma 6.3.4, S n V is either boundary compressible in V or is a
disk. Likewise, S n W is either boundary compressible in W or is a disk.
Suppose that a component, Q, of, say, S n V is boundary compressible in
V. The boundary compressing disk describes an isotopy of S. See Figures
6.12 and 6. 13.
186 6. Heegaard Splittings

Figure 6.12. A boundary compressing disk.

�:: : : : - - : : -�
Figure 6.13. An isotopy through a boundary compressing disk.

This isotopy has the effect of either producing one component, Q ' , out
of Q, where
x ( Q') = x(Q) + 1 ,
or of producing two components, Qi , Q 2 , out of Q, where
x (Q 1 ) + x (Q 2 ) = x(Q) + L
See Figures 6 . 14 and 6 . 1 5 to see how the boundary compressing disks meet
S n V. We call the former non-separating and the latter separating.
If we choose our essential 2-sphere in M so that x (S n V) is maximal,
while retaining the minimality of # IS n SI among essential 2-spheres in
M, then it follows from the Euler characteristic computation above that
S n V consists of disks. Under these assumptions, the argument in Claim 1
still shows that components of S n W are incompressible. (Indeed, the two
spheres created in the proof of Claim 1 would also intersect V in disks. )
Suppose that the number of components / disks in S n V is n. Then
S n W is a planar surface with n boundary components. Lemma 6.3.4 tells
us that each component of S n W is either boundary compressible or a disk.
A non-separating boundary compression corresponds to an arc that has its
endpoints on two distinct components of o(S n W ) . A separating boundary
compression corresponds to an arc that has its endpoints on one component
of o(S n W ) . Arcs corresponding to successive boundary compressions can
be simultaneously embedded in S n W. Note that there can be at most
6.3. Reducibility and Stabilization 187

- x (S n W) non-parallel separating essential arcs in S n W corresponding to

separating boundary compressions. Thus there can be at most - x (Sn W) =
n - 2 such arcs.

Figure 6. 14. Non-separating.

o�
0
Figure 6 . 1 5 . Separating.

The effect of doing a boundary compression of the non-separating type

is to lower the number of components of S n V by one and to leave the
number of components of S n W unchanged. The effect of doing a boundary
compression of the separating type is to leave the number of components of
S n V unchanged and to raise the number of components of S n W by one.
We now perform boundary compressions on Sn W in an effort to produce
a 2-sphere S isotopic to S such that S n V is a connected planar surface
and S n W consists of disks. As we do so, we need perform at most n - 2
separating boundary compressions. Since S n W is connected and since
each boundary compression of the separating type increases the number of
components of the portion of the 2-sphere lying in W by one, it follows that
S n W has at most n - 1 components. In particular,
#IS n S I = #IS n Sl - L
This contradicts our assumption that # IS n S I is minimal over all essential
2-spheres in M.
188 6. Heegaard Splittings

It follows that there are no boundary compressions. In particular, S n V

and S n W are disks and S n S is an essential simple closed curve. Therefore
M = V Us W is reducible. 0

Exercises
Exercise 1. Prove that any two Heegaard splittings resulting from ele­
mentary stabilizations of a Heegaard splitting M = V Up W are equivalent.
(This property is called the uniqueness of stabilization.)
Exercise 2. Prove that a stabilized Heegaard splitting of a 3-manifold not
equal to §3 is reducible.
Exercise 3. Prove that a reducible Heegaard splitting of a prime 3-manifold
not equal to §2 x § I is stabilized.
Exercise 4. Prove that handlebodies are irreducible.

6.4. Waldhausen's Theorem

One of the first theorems proved about Heegaard splittings is due to Wald­
hausen. See [155] . He analyzed the Heegaard splittings of §3 and found that
there is only one Heegaard splitting of §3 of any given genus. Moreover, the
Heegaard splittings of §3 arise via stabilization of the genus 0 splitting.
Other proofs of this theorem have been given since then; see for in­
stance [136] . But Waldhausen's proof remains of interest. The strategy
is very natural and may generalize to prove similar results in situations
where the techniques of [136] fail to apply. This is particularly relevant as
recent years have seen a resurgence of interest in issues pertaining to stabi­
lization of Heegaard splittings. Waldhausen's strategy lies in applying the
Reidemeister-Singer Theorem to compare a given Heegaard splitting of §3
to the genus 0 Heegaard splitting. The two collections of stabilizing pairs of
disks are compared and played off against each other. Below, we provide a
sketch of the argument. We will follow Waldhausen's strategy, but not his
terminology.
Definition 6.4. 1 . Let M = V Up W be a Heegaard splitting. A good system
of n disks in V is the union of n disjoint disks v = V I U U Vn in V and n
· · ·

disjoint disks w = WI U U Wn in W such that:

· · ·

(1) 8vj n 8wj consists of exactly one point and

(2) 8vi n 8wj = 0 when i > j .
If, in addition, 8vi n 8wj = 0 when i < j , then the system of disks is
called a stabilizing system of disks.
6.4. Waldhausen 's Theorem 189

-------

Figure 6.16. A Heegaard diagram of §3 .

Figure 6 . 1 7. Another Heegaard diagram of §3 .

Figure 6.18. A good system of disks.

In Figure 6. 18, a disk slide of w1 over w2 turns a good system of disks

into a stabilizing system of disks. This is always possible; that is, for any
good system of disks v U w there is a sequence of disk slides of components
of w over components of w that yields a stabilizing system of disks v U w ' .
Recall that an elementary stabilization of a Heegaard splitting can be
expressed via a pairwise connected sum (M, F) # (§3, T) , where T is the
unknotted torus in §3 . Here T divides §3 into two solid tori. The meridian
of one of these solid tori is a longitude in the other. Since the meridian and
190 6. Heegaard Splittings

longitude of a given solid torus intersect in one point, so do the boundaries of

the two meridian disks. This pair of disks persists in the Heegaard splitting
obtained by the elementary stabilization.
If a stabilized Heegaard splitting is the result of n elementary stabiliza­
tions, then there will be n pairs of disks. Moreover, the pairs will be disjoint
from each other, and in each pair, the boundaries of the two disks will inter­
sect exactly once. In other words, the stabilized Heegaard splitting comes
with a system of stabilizing disks. Cutting along one disk per pair recre­
ates the Heegaard splitting from which the stabilized Heegaard splitting was
obtained.
Remark 6.4.2. In the exercises, you will prove two facts about a good
system of disks v U w for a Heegaard splitting M = V UF W: ( 1 ) Cutting V
along v yields a handlebody whose complement is also a handlebody. So this
yields a Heegaard splitting. Likewise, cutting W along w yields a Heegaard
splitting. (2) These two Heegaard splittings are equivalent.
Definition 6.4.3. A Heegaard splitting obtained from the Heegaard split­
ting M = V Us W by cutting along a component in a good system of disks
is called a destabilization. If the good system of disks has 2n components ( n
in v and n in w ) , then cutting along all components in v is called an n-fold
destabilization (along v ) .

/{bl
I
, \
I
I
I
'
.... - - �
;'
0 0

Figure 6.19. A stabilizing pair of disks.

Remark 6.4.4. If follows from Remark 6.4.2 and the exercises that if both
vU w and v' U w are good systems of disks, then cutting along v yields the
same Heegaard splitting as cutting along v' .
The following lemma sets the stage for the proof of Waldhausen's The­
orem. Among other things, it establishes that, given two good systems of
disks, a sequence of slides of components of one over components of the
other guarantees a certain disjointness.
6.4. Waldhausen 's Theorem 191

Lemma 6.4.5. If M = VUF W and M = V' UF' W' are Heegaard splittings,
then there is a Heegaard splitting M = X Us Y and good systems x U y and
x' U y' of disks such that the following hold:
(1) M = V UF W is an n-fold destabilization of M = X Us Y along x;
(2) M = V' UF' W' is an m-fold destabilization of M = X Us Y along x';
(3) x n x' = 0;
(4) y n y ' = 0.

Proof. By the Reidemeister-Singer Theorem, M = V U F W and M =

V'U F' W' have a common stabilization. We denote this common stabilization
by M = X Us Y. It remains to show that M = X Us Y satisfies the required
properties. Properties (1) and (2) follow immediately from the definition of
stabilization.
To show that properties (3) and (4) are satisfied, it suffices to show that
we may alter x, y, x', y' so that x n x' = 0 and y n y' = 0. We first consider
closed components of intersection and note that these may be removed via
an innermost disk argument. Next we consider an arc of intersection, say
between Xi c x and xj c x'. Furthermore, we assume that this arc is
outermost in xj. The outermost arc cuts off a disk D from xj. Cutting
Xi along 8D n Xi and attaching copies of D to the resulting boundary arcs
yields two disks. One of these contains the point of intersection between Xi
and Yi · Denote this latter disk by fh We replace Xi with fh By Remark
6.4.4, the veracity of properties ( 1 ) and (2) is unaffected, but the number of
arcs of intersection between x n x' has b�en reduced. 0

Below you will find a sketch of Waldhausen's argument. Many (re)proofs

of Waldhausen's Theorem have been given. The proof below follows Wald­
hausen's original proof closely. See [139] for background information. A
proof that deserves special mention is that given by Scharlemann and Thomp­
son in [136] .
Theorem 6.4.6 (Waldhausen) . The 3-sphere has a unique Heegaard split­
ting of any given genus.

Proof. It follows from the Schonflies Theorem that §3 has a unique Hee­
gaard splitting of genus 0. Let §3 = V UF W be a Heegaard splitting of
genus greater than 0. We wish to show that §3 = V UF W is stabilized. The
theorem will then follow from the uniqueness of stabilization.
Choose a Heegaard splitting §3 = X Us Y as in Lemma 6.4.5 such that
both §3 = V U F W and the genus 0 Heegaard splitting §3 = V' U F' W' are
destabilizations of §3 = X Us Y. As above, we denote the corresponding
192 6. Heegaard Splittings

systems of disks by x U y and x' U y'. We will also assume that §3 = X Us Y

has been chosen to be minimal genus subject to these conditions.
Case 1 . Xn n y' = 0.
Then 8xn survives in the boundary of the result of cutting Y along y'.
But the result of cutting Y along y' is a 3-ball B. Thus OXn c 8B bounds
a disk in this 3-ball. Since this 3-ball is entirely contained in Y, it follows
that 8xn bounds a disk Dy in Y. Thus Xn U Dy forms a 2-sphere in §3 .
By Alexander's Theorem, this 2-sphere is separating. It follows that 8xn
is separating on S. This is a contradiction since 8xn intersects OYn exactly
once.
Case 2. Xn n y' i= 0 and consists of exactly one point.
We may assume that the point of intersection lies in, say, yj . Then
cutting Y along yj yields a Heegaard splitting isotopic to the Heegaard
splittings obtained by cutting X along Xn or xj , respectively (see Remark
6.4.4) . It follows that this resulting Heegaard splitting is a common stabi­
lization of both §3 = V U F W and the genus 0 Heegaard splitting of §3 . This
contradicts the minimality assumption on the genus of §3 = X Us Y.
Case 3. Xn n y' i= 0 and consists of more than one point but no more than
one point in any one component of y.
In this case we may alter y' by disk slides to reduce the number of points
in Xn n y' 'f: 0.

Figure 6.20. Setting for a disk slide of yj over Yn ·

Case 4. Xn meets a component of y' in more than one point.

We may assume that Xn meets, say, y� , at least two times. This allows
us to alter y' by a disk slide over Yn to reduce the number of points of
intersection between X n and y'. See Figure 6.20. D
6. 5. Structural Theorems 193

Exercises

Exercise 1. Show that given a good system of disks v U w, there is a

v
' '
stabilizing system of disks U w , where w is obtained from w by disk
slides.
Exercise 2. Suppose v U is a good system of disks for the Heegaard
w

splitting M = V Us W. Show that cutting V along v yields the same

Heegaard splitting as cutting along w .
6.5. Structural Theorems

Waldhausen's Theorem concerning Heegaard splittings of §3 makes a state­

ment about the structure of all Heegaard splittings of §3 . It is the first of
a sequence of theorems describing the structure of Heegaard splittings of
3-manifolds and classes of 3-manifolds. F. Bonahon and J.-P. Otal proved
the following:
Theorem 6.5.1 (Bonahon-Otal) . All Heegaard splittings of a lens space
M =/= §3 are stabilizations of a unique genus 1 Heegaard splitting of M.
For a proof of Theorem 6.5. 1, see [14] .
Definition 6.5.2. The genus of a 3-manifold M is the smallest possible
genus of a Heegaard splitting for M.
Recall that lens spaces are obtained by identifying the boundaries of
two solid tori. This is a description of lens spaces in terms of Heegaard
splittings. This description makes it clear that the genus of a lens space is 1 .
(It can't b e 0 since you showed above that only the 3-sphere has genus 0.)
The theorem of Bonahon and Otal tells us that each lens space has a unique
Heegaard splitting of genus 1 and every other of its Heegaard splittings is a
stabilization of this Heegaard splitting of genus 1 .
Theorem 6.5.3 (Boileau-Otal) . All Heegaard splittings of the 3-torus are
stabilizations of the genus 3 Heegaard splitting described above.
For a proof of Theorem 6.5.3, see [13] . The theorem of Boileau and
Otal explicitly describes the structure of Heegaard splittings of the 3-torus.
The construction can be generalized to manifolds homeomorphic to (closed
orientable surface) x § 1 .
Definition 6.5.4. Let Q be a closed orientable surface. Let p be a point
in Q and let al , . . . , a2g be a collection of arcs based at p that cut Q into
a disk. Let t be a point in § 1 . Denote a closed regular neighborhood of
al x t U· U a2g x t U p x § 1 by V. Denote the closure of the complement of
· ·
194 6. Heegaard Splittings

V by W. Set S = aV and M = Q x § 1 . You will show in the exercises that

M = V U s W is a Heegaard splitting. This Heegaard splitting is called the
standard Heegaard splitting of M.

Figure 6.21. The standard Heegaard splitting of Q x §1 •

Theorem 6.5.5 (Schultens) . All Heegaard splittings of ( closed orientable

surface) x § 1 are stabilizations of the standard Heegaard splitting.
For a proof of Theorem 6.5.5, see [143] . The standard Heegaard splitting
of (closed orientable surface) x § 1 is described in terms of a spine for V, i.e. ,
a graph whose regular neighborhood is V. This is a useful strategy in many
settings. Consider the following example:
Definition 6.5.6. Let M be a prism manifold, i.e. , a Seifert fibered space
with base orbifold the 2-sphere with three exceptional points a, b, c. Denote
the base orbifold of M by 0. Let / be an arc in M whose projection is a
simple arc connecting the exceptional points a and b. Denote / together with
the exceptional fibers that project to a and b by r. See Figure 6.22. Denote
a closed regular neighborhood of r by V, the closure of the complement of
V by W and the surface V n W by S. You will show in the exercises that
M = V Us W is a Heegaard splitting for the prism manifold M.
Any Heegaard splitting of a prism manifold of this form is called a ver­
tical Heegaard splitting.
The notion of a vertical Heegaard splitting can be generalized to more
complicated Seifert fibered spaces. This can then be seen as the canonical
construction of Heegaard splittings for Seifert fibered spaces. There is a less
canonical construction that yields Heegaard splittings for some, but not all,
Seifert fibered spaces. We will not describe it here. Suffice it to say that
this type of Heegaard splitting is called a horizontal Heegaard splitting. It
Exercises 195

Figure 6.22. A vertical Heegaard splitting of a prism manifold.

was proved by Y. Moriah and the author that under certain orientability
assumptions, all Heegaard splittings of Seifert fibered spaces are horizontal
or vertical. See [106] .
Another structural theorem for Heegaard splittings for a class of mani­
folds is that of D. Cooper and M. Scharlemann for solvmanifolds; see [29] . A
solvmanifold is a 3-manifold that is the mapping torus of an automorphism
</J : torus � torus such that the absolute value of the trace of (the matrix
corresponding to) </J is strictly greater than 2.
A solvmanifold possesses a standard Heegaard splitting that is con­
structed analogously to the standard Heegaard splitting of the product man­
ifolds above. It is of genus 3. Typically, this is the unique unstabilized Hee­
gaard splitting of this manifold. However, in those cases where (the matrix
corresponding to) </J is conjugate to a matrix of the form

the standard Heegaard splitting of genus 3 is stabilized. In this case there

is a genus 2 Heegaard splitting that is the unique unstabilized Heegaard
splitting of the solvmanifold.

Exercises

Exercise 1 . Show that the splitting described in Definition 6.5.4 is indeed

a Heegaard splitting.
Exercise 2. The standard Heegaard splitting of (closed orientable surface)
x § 1 was described in terms of a spine for V. Describe a spine for W.
196 6. Heegaard Splittings

Exercise 3. Show that the splitting described in Definition 6.5.6 is indeed

a Heegaard splitting.
Exercise 4. The standard Heegaard splitting of a prism manifold was
described in terms of a spine for V. Describe a spine for W.

6.6. The Rubinstein-Scharlemann Graphic

In the 1990s Rubinstein and Scharlemann decided to investigate the relative
positioning of two Heegaard splittings of the same manifold with respect
to each other. To do so, they reinterpreted and reformulated the work of
Cerf. The specific tool they used has become known as the Rubinstein­
Scharlemann graphic. See [130] , [129] , and [131] .
Most important here is the fact that the handlebody V is a regular
neighborhood of the spine r. This means that V\r is homeomorphic to
8V x [O, 1). Consequently, if M = V U s W is a Heegaard splitting with rv
a spine of V and r a spine of W, then M\ (rv U r ) is homeomorphic to
w w

S x ( - 1 , 1). Furthermore, the image of S x { t} under this homeomorphism

is isotopic to S, for all t E (-1, 1), and hence defines the same Heegaard
splitting as S. In this section we denote the interval [- 1 , 1] by I .
Definition 6.6. 1 . Let M = V U s W, rv, and r be as above. A continuous
w

map h : S x (I, 8!) -+ (M, rv u r ) that is a homeomorphism on S x (-1, 1)

w

and maps S x 8{ -1} to rv and S x 8{ 1} to r is called a sweepout of M.

w

We denote the image of (S, t) be St .

Suppose that M = X U Q Y is also a Heegaard splitting. Then for r a
x

spine of X and ry a spine of Y there is also a sweepout g Q x (I, al) -+

:

(M, r U ry ) · Rubinstein and Scharlemann were interested in analyzing the

x

intersections of Qr and St .
Suppose that the two sweepouts are transverse. We consider ( r, t) E
I x I . If ( r, t) is generic, then Qr m St . As it turns out (for details see
[125] ) , there are four types of points in the interior of the square I x I =
{ ( r , t ) : 0 :S r :S 1 , 0 :S t :S 1}:
(1) those where Qr and St meet transversely;
(2) those where Qr and St meet transversely except at a single non­
degenerate tangent point (i.e. , a point of tangency modeled on the
point of tangency of z = x 2 ± y2 and z = O) ;
(3) those where Qr and St meet transversely except at two non-degen­
erate tangent points;
6. 6. The Rubinstein-Scharlemann Graphic 197

(4) those where Qr and St meet transversely except at a single de­

generate critical point modeled on Qr = { (x, y, z ) l z = O} and
St = { (x, y, z ) l z = x 2 + y 3 } .
The last two types of points are isolated; the third type is called cross­
ing vertices, the fourth birth-death vertices. The points of the second type
occur in codimension 1 strata that we call edges. The points of the first
type make up the rest of the interior of the square and thus constitute con­
nected open sets that we call regions. The edges and vertices make up a
graph that is called the Rubinstein-Scharlemann graphic and is denoted by
r. The Rubinstein-Scharlemann graphic naturally extends to the closed
square, where we continue to call it the Rubinstein-Scharlemann graphic.

Figure 6.23. The Rubinstein-Scharlemann graphic.

Consider a point (r, t) E (J x J)\r and denote the region it lies in by

R . The corresponding surfaces Qr and St intersect transversely. We are
interested in the curves of intersection that are essential in Qr . If there is
such a curve that bounds a disk in X (or Y, respectively) , then we label
the region R with an X (or Y, respectively) . We are also interested in the
curves of intersection that are essential in St. If there is such a curve that
bounds a disk in V (or W, respectively) , then we label the region R with a
V (or W, respectively) .
If we imagine a path within a region, then each point on this path
corresponds to a pair of surfaces that intersect transversely. Moreover, since
all these pairs of surfaces intersect transversely, their curves of intersection
are remaining constant up to isotopy. In particular, the labeling is well­
defined on the region!
Our labeling has many consequences. The lemma below follows imme­
diately from these definitions.
Lemma 6.6.2. If any region is labeled both X and Y, then M = X UQ Y is
weakly reducible. Analogously, if any region is labeled both V and W, then
M = V Us W is weakly reducible.
198 6. Heegaard Splittings

Part of the strength of this labeling scheme is that we can say something
about the labeling of regions that are adjacent along an edge. To this end,
we imagine a short path in I x I that connects two such regions. We are
especially interested in the point at which the path crosses the edge. See
Figures 6.24 and 6 . 25.

- - ·- -

Figure 6.24. Surfaces corresponding to an edge in the Rubinstein­

Scharlemann graphic.

Figure 6.25. Surfaces corresponding to an edge in the Rubinstein­

Scharlemann graphic.

If the path crosses the edge in a point corresponding to surfaces meeting

as in Figure 6.24, then only inessential curves of intersection are created or
destroyed. From the point of view of our labeling, this is of no interest at
all. On the other hand, if the path crosses the edge in a point corresponding
to surfaces meeting as in Figure 6.25, then essential curves of intersection
may be created or destroyed. See Figure 6.26 .
. -- · - · - · - · - · - · - · - · - · - .

----"."I"':::"�"."::"-----'·\
I
/

(.0 '·
'

Figure 6.26. Possible effect of crossing and edge.

6. 6. The Rubinstein-Scharlemann Graphic 199

Lemma 6.6.3. Suppose that regions R1 and R2 are adjacent along an edge.
Suppose further that R1 is labeled X and R2 is labeled Y. Then M = X UQ Y
is weakly reducible.
Proof. If R1 is also labeled Y or if R2 is also labeled X, then the result
follows from Lemma 6.6.2. So we will assume that R1 is not labeled Y and
R2 is not labeled X.
Consider a short path in I x I that begins in R i and ends in R2 and
intersects the edge between these two regions in a single point. The point of
intersection (r, t) corresponds to a pair of surfaces Qr, St such that Qr n St
contains a figure 8 ( cf. Figure 6.25) . If we think of Qr as the splitting surface
Q and let t vary ( exhibiting the horizontal surfaces St stacked neatly on top
of each other) , then to one side (say in Q n St -E ) of this figure 8 there is
an essential curve that bounds a disk in X. To the other side ( in Q n St +E )
of this figure 8 there is an essential curve that bounds a disk in Y. Thus
M = X UQ Y is weakly reducible. 0

More technical arguments can be employed to say more about the label­
ing of the regions. The following results deserve particular mention:
Lemma 6.6.4. If all four labelings X, Y, V, W appear in the quadrants of a
crossing vertex, then either two opposite quadrants are unlabeled or one of
M = X UQ Y or M = V U s W is weakly reducible or M = §3 .
Lemma 6.6.5. There is an unlabeled region.
Proposition 6.6.6. For one of the pairs of labels X, Y or V, W there is
a (generic) path that traverses only unlabeled regions and begins at an edge
labeled X (or V, respectively) and ends at an edge labeled Y (or W, respec­
tively).
One of the most impressive results obtained by Rubinstein and Scharle­
mann using this general setup is the following:
Theorem 6.6. 7. Suppose M = X U Q Y and M = V U s W are strongly
irreducible Heegaard splittings of genus q and s, respectively. Suppose also
that q ::; s . Then there is a genus 8q + 5s - 1 1 Heegaard splitting of M that
is a stabilization of both M = X UQ Y and M = V Us W.
Using the additional machinery of generalized Heegaard splittings, de­
scribed in subsequent sections, Rubinstein and Scharlemann obtained a qua­
dratic bound on the number of stabilizations required in a generalized version
of Theorem 6.6.7 to the case where the Heegaard splittings are not strongly
irreducible.
200 6. Heegaard Splittings

Exercises
Exercise 1. Prove Lemma 6.6.5.
Exercise 2. Let M be a prism manifold, i.e. , a Seifert fibered space that is
fibered over the sphere and has three exceptional fibers. Construct a spine
for a handlebody V by connecting the three exceptional fibers together by
arcs that project to embedded arcs. The closure, W, of the complement
of this handlebody is also a handlebody. Show that the Heegaard splitting
M V Us W thus defined is stabilized.
=

Exercise 3. Prism manifolds have up to three distinct vertical Heegaard

splittings. Show that these become isotopic after one stabilization. (Hint:
See the exercise above.)

6 . 7. Weak Reducibility and Incompressible Surfaces

In this section we prove a theorem of Casson and Gordon; see [24] . The the­
orem establishes a connection between a Heegaard splitting being weakly re­
ducible and the existence of an incompressible surface. This theorem proved
to be seminal. Eventually, it engendered the concept of a thin manifold de­
composition of a 3-manifold pioneered by Scharlemann and Thompson. This
concept in turn gave rise to the notion of a generalized strongly irreducible
Heegaard splitting, a structure possessed by every compact 3-manifold .

....
/_ .___ � _ /
Figure 6.27. Local picture of a portion of a weakly reducible Heegaard splitting.
6. 7. Weak Reducibility and Incompressible Surfaces 201

Theorem 6. 7. 1 . Suppose M is a closed orientable 3-manifold and M =

V Us W is a weakly reducible Heegaard splitting. Then either M contains

an incompressible surface or M V Us W is reducible.
=

Proof. Let V be a non-empty disjoint union of non-parallel essential disks

in V and let E be a non-empty disjoint union of non-parallel essential disks
in W such that a'D n aE = 0. Consider S\ (V U E) . For each component of
this surface, take the union with appropriate components of V U E to obtain
a closed surface S* . After a small isotopy, S * is embedded in M. (For every
disk D in V U E, there will be two remnants of D in S* . )
Since M V U s W is weakly reducible, we can choose V and E to be
=

non-empty. More importantly, we will assume, in what follows, that V U E

is chosen so that x (S* ) is maximal.
Case 1 . A component, Q, of S * has positive genus.
In this case it follows from our maximality assumption that Q is incom­
pressible.
Case 2. All components of S* are 2-spheres.
Let V be the components of S* that meet the interior of V and let W
be the components of S * that meet the interior of W.
Claim 1. V n W:;f 0.
If Vn W 0, then reversing the cut and paste operations performed above
=

would connect components in V with components in V and components in

W with components in W and this would result in at least two components.
Since S is connected, this is impossible.
Let S be a component of V n W. Then S lies mostly in S. Furthermore,
Sn V and S n W are non-empty and consist of disks. Let c be a simple closed
curve in S that separates the components of S n V from the components of
S n W. Note that c is also a simple closed curve in S.
Claim 2. c is an essential curve in S.
There are essential curves to either side of c; thus c can't bound a disk
in S.
Now the disk in S\c that meets V can be isotoped slightly to one side of
S to lie entirely in V and the other can be similarly isotoped to lie entirely
in W. This shows that M V Us W is reducible.
= D

In the case in which M V Us W is reducible, two things can happen:

=

(1) The 2-sphere constructed can be essential; (2) the 2-sphere constructed
202 6. Heegaard Splittings

can be inessential. In the first case, M contains an incompressible surface,

namely the 2-sphere. In the second case, the 2-sphere splits off an §3 sum­
mand with a Heegaard splitting of positive genus. It then follows from the
theorem of Waldhausen discussed in Section 6.4 that M = V Us W is in fact
stabilized.
So far, we have discussed Heegaard splittings in the context of closed
3-manifolds. The notion can be generalized in more than one way. The
most common such generalization involves the notion defined below.
Definition 6.7.2. A compression body is a compact 3-manifold W that can
be obtained from a closed surface Q and 0-handles by attaching 1-handles
that don't meet Q x {O} c Q x [O, l] .
We denote Q x {O} by c'L W and oW\(L W by 8+ W ·
Note that handlebodies form a subset of compression bodies.
Definition 6.7.3. Let M be a compact 3-manifold. A Heegaard splitting
of M is a decomposition M V Us W, where V, W are compression bodies
=

and S = 8+ V = 8+ W.
The terminology introduced for Heegaard splittings of closed 3-manifolds
(equivalence, genus, etc.) carries over to this more general setting.
Exercises
Exercise 1. Consider the standard genus 3 Heegaard splitting of the 3-
torus. Show that it is weakly reducible by exhibiting a pair of disks that
satisfy the definition. Then use these disks as in Theorem 6.7. 1 to produce
an incompressible surface.
Exercise 2. Generalize Theorem 6.7. 1 to manifolds with boundary.

6 . 8 . Generalized Heegaard Splittings

Section 6.6 provided a taste of the utility of strongly irreducible Heegaard
splittings. Another crucial feature of strongly irreducible Heegaard splittings
revolves around the fact that their splitting surfaces behave like incompress­
ible surfaces in many ways. Recall, for instance, that two incompressible
surfaces can be isotoped to intersect only in curves that are essential in
both surfaces. Analogously:
Lemma 6.8.1. Let M be an irreducible 3-manifold. Let F be an incom­
pressible surface (possibly with boundary) and let M = V U s W be a strongly
irreducible Heegaard splitting. Then S can be isotoped so that all components
of F n S are essential in both F and S.
6.8. Generalized Heegaard Splittings 203

Proof. Let rv be a spine of V and let r be a spine of W. Let h be a

w

sweepout of M = V Us W. For t close to - 1 , St n F consists of small

simple closed curves bounding regular neighborhoods of the points r n F, v

that is, disks in V. Note that these curves of intersection are essential in
St. Likewise, for t close to 1, St n F consists of small simple closed curves
bounding regular neighborhoods of the points r n F, that is, disks in W.
w

Note that these curves of intersection are essential in St.

e V thin

o W thin

-·-·-·- intermediate

/
/

- -
- - -
� -·-·
/

Figure 6.28. St n F, for three different values of t.

The subset of [- 1, 1] consisting of t such that St n F contains simple

closed curves that are essential in St and bound disks in Vt n F is closed,
as is the subset of [- 1, 1] consisting of t such that St n F contains simple
closed curves that are essential in St and bound disks in Wt n F. Thus
these subsets either overlap or are disjoint. Since M = V Us W is strongly
irreducible, they can't overlap. Thus they are disjoint. Therefore there exist
t such that St n F contains no simple closed curves that are essential in St
and bound disks in F. (See Figure 6.28.) Simple closed curves in St n F
that are inessential in St can be removed via Lemma 3.2.3. 0

Not every 3-manifold possesses a strongly irreducible Heegaard splitting.

Consider, for instance, the 3-torus. By Theorem 6.5.3 of Boileau and Otal, it
has a unique unstabilized Heegaard splitting. We have seen that this genus
3 Heegaard splitting is weakly reducible. See Figure 6.29.
Heegaard splittings correspond to handle decompositions of a 3-manifold.
But not every handle decomposition of a 3-manifold corresponds to a Hee­
gaard splitting. This is because in a handle decomposition, I-handles need
not be attached prior to 2-handles.
204 6. Heegaard Splittings

Figure 6.29. The standard Heegaard splitting of the 3-torus is weakly reducible.

One motivation for requiring the I-handles to be attached prior to 2-

handles in a Heegaard splitting is to attain a certain symmetry: For a han­
dle decomposition of a closed 3-manifold, a surface, the Heegaard surface,
separates the I-handles from the 2-handles. Since I-handles are dual to 2-
handles, the submanifolds on either side of the surface are handlebodies of
the same genus and hence homeomorphic.
Breaking symmetry, we drop the requirement of attaching all I-handles
prior to all 2-handles. Instead of having a single surface that captures all
relevant information concerning our splitting, we now have several surfaces
deserving attention. More specifically, we attach some number of I-handles,
followed by some number of 2-handles, then more I-handles, then more 2-
handles, and so on. We are now interested in the surface bounding the
result of attaching the first collection of I-handles, the surface bounding the
result of attaching the first collection of 2-handles, that bounding the result
of attaching the second collection of I-handles, and so on. See Figure 6.30.
Note that successive surfaces of the type described cobound compression
bodies. This point of view lies at the heart of Scharlemann and Thompson's
notion of thin position for 3-manifolds; see [137] .
Definition 6.8.2. A generalized Heegaard splitting is a decomposition
M = ( Vi Us1 W1 ) UF1 (l/2 Us2 W2 ) UF2 • • • UFn-l (Vn Usn Wn ) ,
where each Vi and each Wi is a compression body,
Si = &+Vi = &+Wi,
and
� = o_ Wi = 8-Vi+ I ·
Two generalized Heegaard splittings
M = ( Vi Us1 W1 ) UF1 (l/2 U s2 W2 ) UF2 · • • UFn- l (Vn U sn Wn )
6.8. Generalized Heegaard Splittings 205

2-handles

s)

I -handles

F2
2-handles

sl
I -handles

F,
2-handles

s,
I -handles

Figure 6.30. Schematic for a generalized Heegaard splitting.

and
M = ( V,1' Us' W{) U F' (V� Us'2 W� ) UF.'2 . . . U Fn-
l l
' 1 (V� Us1n W� )
are considered equivalent if the collections of surfaces
S1 U Fl U S2 U F2 . . U Fn- 1 U Sn
·

and
s� u F{ u s2 u F� . . . u F� _ 1 u s�
are isotopic.
The schematic diagram in Figure 6.30 can be misleading in terms of
connectedness of the surfaces Si and Fi. We do not, in fact, require these
surfaces to be connected. However, it is common to assume that each Vi and
each Wi has only one "active component" , i.e. , only one component that is
not a trivial compression body.
The diagram in Figure 6.31 provides a schematic of a generalized Hee­
gaard splitting that indicates the number of components of each Si and Fi .
Concerning this topic, a series of lecture notes by T . Saito, M . Scharle­
mann, and the author is in progress.
Remark 6.8.3. A Heegaard splitting is also a generalized Heegaard split­
ting, so every 3-manifold possesses a generalized Heegaard splitting.
Definition 6.8.4. A generalized Heegaard splitting
M = ( Vi Us1 W1) Up1 ( Vi Us2 W2 ) Up2 • • • Upn _ 1 (Vn Usn Wn )
is strongly irreducible if every Fi is incompressible in M and every Vi U si Wi
is a strongly irreducible Heegaard splitting of Mi = Vi U Wi.
206 6. Heegaard Splittings

F,

s,

Figure 6.31. Schematic for a generalized Heegaard splitting.

Theorem 6.8.5. Every 3-manifold possesses a strongly irreducible general­

ized Heegaard splitting.

Sketch of proof. Let M be a 3-manifold. By the theorem of Moise, M

possesses a Heegaard splitting M = V Us W . If this Heegaard splitting
is strongly irreducible, then there is nothing to prove. If it is weakly irre­
ducible, then there are essential disks Dv c V and Dw c W that are disjoint.
Now Dv is the cocore of a 1-handle and Dw is the core of a 2-handle. It is a
deep fact, one we will not address here, that the disjointness of Dv and Dw
implies that the 2-handle corresponding to Dw can be "lowered" so as to be
attached before the 1-handle corresponding to Dv .
Assume, then, that our handle decomposition is constructed by attaching
!-handles only until one of the 2-handles can be attached; continuing by
attaching as many 2-handles as possible; attaching more 1-handles but only
until the next 2-handle can be attached; and so forth. You will show in the
exercises that Vi Usi Wi must be strongly irreducible. (This completes our
sketch of the proof of Theorem 6.8.5.) 0

Definition 6.8.6. A rearrangement of the order of attachment of the 1-

handles and 2-handles in a Heegaard splitting that results in a strongly
irreducible generalized Heegaard splitting is called an untelescoping.
Example 6.8.7. Consider the Heegaard splitting of the 3-torus described
in Section 6.1. It has genus 3. It is obtained by attaching three 1-handles
and then three 2-handles. The order of attachment can be rearranged: We
first attach two 1-handles. This gives us Vi . See Figure 6.32.
6.8. Generalized Heegaard Splittings 207

'
'
'

Figure 6.32. Vi .

We then attach one 2-handle. This gives us Vi U s1 W1 . The bounding

surface of Vi U s1 W1 , F1 , consists of two parallel tori. See Figure 6.33.

I ''
,
- --------- ,
I '
I '
...

'
'
'
'
'
'
'
'
'
'

Figure 6.33. F1 .
We have filled up half of the 3-torus. To fill up the remainder we proceed
along dual lines: The third I-handle is attached (a spine of Vi is depicted
in Figure 6.34) and then the remaining two 2-handles.

'

- - - _:,
'

1
- - - - - - -
'
'
'
'
'
'
'
'
'

Figure 6.34. A spine of V2 .

208 6. Heegaard Splittings

Exercises
Exercise 1. Let V be a compression body. Prove that {)_ V is incompress­
ible in V.
Exercise 2. Describe ( in vague terms, invoking the deep fact mentioned in
the sketch of a proof of Theorem 6.8.5) why the assumption about the order
of attachment of I-handles and 2-handles in the sketch of a proof of Theorem
6.8.5 forces the Heegaard splittings Vi U si Wi to be strongly irreducible. See
[136] .

6 . 9 . An Application

One application of the notion of a generalized Heegaard splitting relates to

the behavior of the tunnel number of a knot under the operation of connected
sum.
Definition 6. 9 . 1 . A tunnel system for a knot K in §3 is a collection of simple
arcs u 1 U U Un in C(K) = §3\ry(K) such that C(K)\17(u 1 U U Un ) is
· · · · · ·

a handlebody. The tunnel number of K, t(K) , is the minimum number of

components required for a tunnel system.
The tunnel number of a knot is closely related to the Heegaard genus of
the knot complement: A tunnel system defines a Heegaard splitting C(K) =
V U s W by setting W = C(K)\17(u 1 U U Un ) and setting V to be the
· · ·

closure of 17(8C(K)) U17(u 1 U U un ) The genus of this Heegaard splitting is

· · ·
·

t(K) + 1 . Conversely, given a Heegaard splitting, we obtain a tunnel system

by appropriately manipulating a spine of the compression body containing
8C(K) . It then follows that the genus of C(K) is exactly t(K) + 1 .
As it turns out, the tunnel number of a knot behaves rather erratically
under the operation of connected sum of knots. Morimoto, Sakuma, and
Yokota exhibited examples of two knots with tunnel number 1 whose sum
has tunnel number 3; see [108] . Morimoto exhibited examples of knots
having tunnel numbers 1 and 2, respectively, whose sum has tunnel number
2; see [107] . Kobayashi extended these examples to show that the difference
t(K1 # K2 ) - t(K1 ) - t(K2 ) can be arbitrarily large; see [81] . The following
theorem, due to Scharlemann and the author ( see [135] ) , thus came as a
surprise:
Theorem 6.9.2.

We will not discuss the proof here, as it is rather technical. But we

will discuss the key ingredients of the proof of a related theorem, also by
6. 9. An Application 209

Scharlemann and the author (see [134] ) :

Theorem 6.9.3.
t(K1 # · · · #Kn ) � n.
We observed above that for a knot K to have tunnel number at least n,
C(K) must have Heegaard genus at least n + 1. We prove a specialized form
of Theorem 6.9.3:
Proposition 6.9.4. Let K = Ki# · · · #Kn . (I. e., K is a connected sum of
n non-trivial knots.) Suppose that C(K) = V U s W is a strongly irreducible
Heegaard splitting of the complement of K. Then the genus of C(K) =
V Us W is at least n + 1 .

Figure 6.35. A decomposing sphere.

Figure 6.36. Visualizing the knot complement.

210 6. Heegaard Splittings

Proof. First a general fact about a connected sum of n knots: The comple­
ment of such a knot will contain a collection of n 1 decomposing spheres,
-

S = 81 Li · · · LJ Bn- 1 , characterized by the fact that

C(K)\S
consists of punctured copies of C(K1 ) Li · · · Li C(Kn) · Each of the spheres
intersects K in two points and C(K) in an annulus called a decomposing
annulus. The decomposing annuli cut C(K) into LJ� 1 C(Ki) · See Figures
6.35 and 6.36.
Because C(K) = V U s W is strongly irreducible, S can be isotoped so
that it intersects each decomposing annulus in curves that are essential in
both S and in the decomposing annulus. Let A be an arbitrarily chosen
decomposing annulus. Then S n A consists of curves that are essential in
A and that are hence meridians of 8C(K) . You will prove in the exercises
that the complement of a non-trivial knot can't be a proper submanifold
of a compression body. Hence S must meet each of the knot complements
C(Ki) ·
Now consider cutting S along all curves of intersection with the decom­
posing annuli. Let Q be a component of the resulting surface. We argue
that Q can't be an annulus: Recall that the boundary components of Q are
meridians of 8C(K) . Hence if Q is an annulus, then it is either boundary
parallel or it is a decomposing annulus. We assume the following: (1) that S
intersects the collection of decomposing annuli in the fewest possible num­
ber of curves (subject to the condition that the curves of intersection are
essential) ; (2) that each summand Ki is prime. (Otherwise we in fact have
a connected sum of more than n knots.) Under these assumptions, if Q is
an annulus, then it is parallel to one of the decomposing annuli. In partic­
ular, there must be some other component of S\S in the knot complement
containing Q.
It follows that each knot complement contains a non-annular component
of S\S . Since S is orientable, a non-annular subsurface of S that has (non­
empty) boundary consisting of curves essential in S can't be a sphere, disk,
or torus. Hence it must have strictly negative Euler characteristic. In the
exercises you will show that it also must have even Euler characteristic.
Summing over the components of C(K)\S, this tells us that x(S) ::::; 2n - .

Thus g (S) 2: n + 1 . D

How does this specialized result help us prove the theorem mentioned?
It provides the key idea for a more formal proof. If our Heegaard splitting
is weakly reducible, then we untelescope it to obtain a strongly irreducible
generalized Heegaard splitting. The key idea helps guide us through a proof
6. 9. An Application 211

of the theorem that relies on strongly irreducible generalized Heegaard split­

tings. In a generalized Heegaard splitting, we need to consider more than
just one surface; we need to consider (LJi Fi) U (LJi Si) · The following defi­
nition is crucial:
Definition 6.9.5. The index J(V) of a compression body V is defined by
the formula
J(V) = x(8_ V) - x(8+ V) .
Lemma 6.9.6. If
M = ( Vi Us1 W1 ) Up1 ( Vi Us2 W2 ) Up2 • • • Up,._1 (Vn Us,. Wn )
is a strongly irreducible generalized Heegaard splitting that is an untelescop­
ing of the Heegaard splitting
M = V Us W,
then
-x(S) = -x(8- Vi ) + x L J(Vi) .
i
You will prove Lemma 6.9.6 in the exercises. Now consider the decom­
posing annulus as it winds its way through a strongly irreducible generalized
Heegaard splitting. See Figure 6.37.

Figure 6.37. Schematic for a decomposing annulus and a generalized

Heegaard splitting.

In the warm-up case considered above, we captured a portion of the

splitting surface with negative Euler characteristic in each knot complement.
212 6. Heegaard Splittings

Now, we proceed similarly, but compute indices. We can do this by relying

on the the following auxiliary lemma:
Lemma 6.9. 7. If A is a collection of annuli in a compression body V and
X is a component of V\A, then
x ( 8_ v n X) - x ( 8+ v n X) � o .
Furthermore, if
x ( 8_ v n X) - x ( 8+ v n X) = o ,
then v n x is a product.
You will prove Lemma 6.9.7 in the exercises. Consequently, each knot
complement, since it is not a product, captures positive index. A counting
argument then establishes the desired inequality genus(S) � n + 1 . This in
turn establishes the stated bound on the tunnel number.

Exercises
Exercise 1 . Prove that a separating surface (possibly with boundary) in a
knot complement has even Euler characteristic.
Exercise 2. Prove that the complement of a non-trivial knot can't be a
proper submanfold of a compression body.
Exercise 3. Prove Lemma 6.9.6.
Exercise 4. Prove Lemma 6.9.7.

6 . 10. Heegaard Genus and Rank of Fundamental Group *

The description of a 3-manifold via a Heegaard splitting gives a natural way

of computing the fundamental group of a 3-manifold. In this section we
consider two distinct notions, the Heegaard genus of a 3-manifold and the
rank of the fundamental group of a 3-manifold. The insight here translates
into an inequality for these invariants.
Definition 6.10. 1 . The Heegaard genus of a 3-manifold M, denoted by
g ( M ) , is the least possible genus of a splitting surface of a Heegaard splitting
for M.
E.g. , g(§3 ) = 0, g(lens space) = 1, g(prism manifold) = 2.
Definition 6. 10.2. The rank of a 3-manifold M, denoted by r(M) , is the
least number of generators required for 7r 1 (M) .
Theorem 6. 10.3. r(M) � g ( M ) .
6. 1 0. Heegaard Genus and Rank of Fundamental Group 213

Proof. Given a Heegaard splitting M =

V U s W that realizes g(M) , we
may compute the fundamental group of M as follows: We consider M to
be built from V in g(M) + 1 steps. At each of the first g(M) steps, we add
an open neighborhood of a 2-handle that can also be thought of as an open
neighborhood of a meridian disk for W. In the final step, we add an open
neighborhood of the 3-handle.
This description translates into a computation of 7r 1 (M) . Here 7r 1 (V)
is the free group on g(M) generators. Adding an open neighborhood of a
disk (whose fundamental group is trivial) adds a relation. Hence we add a
relation at each of the first g(M) steps. In the final step, a 3-ball (also with
trivial fundamental group) is added along its boundary 2-sphere. Thus the
fundamental group is unchanged.
We obtain a ( "balanced" ) presentation
D

The converse is not true; i.e. , there are 3-manifolds for which the in­
equality is strict. See [12] .
Theorem 6. 10.4 (Schultens-Weidmann) . Given any n E N, there is a 3-
manifold Mn such that g(Mn ) - r(Mn ) � n.

0 •

•
'
'
'
'

0 0
'
• '
'
'
'

0
Figure 6.38. A base orbifold considered in the proof of Theorem 6.10.4
(case n = 4) .
214 6. Heegaard Splittings

For a proof of Theorem 6. 10.4, see [144] . The examples constructed in

the proof are graph manifolds. A graph manifold is a 3-manifold modeled
on a graph such that each vertex corresponds to a Seifert fibered space and
each edge corresponds to the identification of two boundary components of
these Seifert fibered spaces. The examples in question are modeled on star
shaped graphs with 2n + 3 vertices and n edges. In these examples, ;���
is roughly i . The minimum possible value for this expression is unknown.
For more on this topic, see Li's treatment in [86] .
Rank and genus are examples of "classical" invariants of 3-manifolds.
Recent years have seen rapid progress in understanding "modern" invariants
of 3-manifolds, growing out of homology theories. These topics go beyond
the scope of this book. The reader is invited to peruse [66] and [115] .
Exercises
Exercise 1 . Assume that ;��� 2:: i for all 3-manifolds M and prove that
r(M) = 0 implies M = §3 .
Exercise 2. Design a sufficiently complicated genus 2 Heegaard splitting
and calculate the fundamental group of the 3-manifold using that Heegaard
splitting.
 Chapter 7

Further Topics

7. 1 . Basic Hyperbolic Geometry

Hyperbolic n-space can be realized in a variety of ways. We here discuss the

upper half-space model. See also [8] , [25] , [74] , [75] , and [124] . Consider
the set
u;n = {(x 1 , . . . , X n ) E �n : Xn > O}.

The element of hyperbolic arc length on IUn is

2 2
ds = Jdx 1 + + dxn · · ·
·

Xn
and the element of hyperbolic volume is given by

dV = dx 1 xn dxn .
· · ·

n
This tells us how to calculate arc length and volume in hyperbolic space.
See the 2-dimensional computations below.
Example 7 . 1 . 1 . To calculate the arc length L of the horizontal path from
(0, 1) to (1, 1) we parameterize the path by x 1 (t) = t, x2 (t) = 1 for t E [O, 1 ]
.

1 1 ds = 1 1 Jdx21 + dx22 = 1 1 J12 + 0 dt = l.

Then,

L=
o o x2 o 1
Example 7. 1.2. For n = 2, to calculate the arc length L of the vertical path
from (O, a) to (O, b) , for b > a > 0, we parameterize the path by x 1 (t) = 0,
-

215
216 7. Further Topics

x 2 (t) = t. Then,
l b lb Jdx21 + dx2
ab
L = ds = 2

l J0+12
a X2 ·

= dt = [ln t] �
at
b
= ln b - ln a = ln - .
a

In Exercise 1 of this section, you will show that a vertical arc minimizes
the length among all arcs between (0, a ) and (0, b) .
Example 7. 1 . 3 . The subarc of the unit circle from (0, 1 ) to (1, 0) not in­
cluding this second endpoint has infinite length. To see this, we parameterize
this arc as x 1 (t) = sin t, x 2 (t) = cos t and compute the following improper
integral:
1 1
ds = 1 � v. rdx21 + dx22
0 0 X2
,-__2___-�2-
r� cos t + sin t
v
=
}0 cos t

lo
� 1
= - __ dt
cos t
= lim [ln(sec t + tan t)]g = oo .
a-t �
Example 7. 1 .4. To calculate the area of the region A between the two
vertical rays given by x 1 = ± 1 and above the semicircle { x : lxl = 1 } (see
Figure 7 . 1 ) , we compute the following integral:
-1 1
Area - dV -
dx 1 dx 2
A X 22
1 100
A
1 dx 2 dx 1
-

1 [ ] 00
- 1 � x�
1 1
= -- dx 1
- 1 X2 �
= 1 1 1
- 1 J1-=xr
dx 1
= [arcsin x 1 ] � 1 = 1r.

Example 7. 1 . 5 . To calculate the area of the region A in the half-plane

x 1 2: 0, above the semicircle { x : lxl = 1 } and below the semicircle
7. 1 . Basic Hyperbolic Geometry 217

Figure 7. 1 . A region with area 1r.

{x Ix - ll = 2} ( see Figure 7.2) , we compute the following integral:

Area - - 1A dV - 1A dxX1 dx22 2 -- 1 1 1�

J4 - (x1 - 1 ) 2 dx 2 dx 1
X 22
l ] J4 - (x1 - 1 } 2
0

=
1 1
-
[- X2 � dx1
O

= !.' ( - ./4 - (�1 - 1) 2 + h.) dx,

= - [arcsin ( xi ; )]: + [arcsin x1]i

1

7r 7r 7r
= -6 + 2 = 3·
In what follows, arc length will always be computed as above using the
element of hyperbolic arc length. To summarize:
Definition 7. 1.6. The upper half-space model for hyperbolic space is a
metric space obtained as follows: Let un = { ( X 1 ' . . . ' Xn ) E IR.n I Xn > 0 } .
The distance between two points is the minimal length of an arc connecting
the points. We denote this metric space by (ll.Jn , dun ) or simply by un .
218 7. Further Topics

Figure 7.2. A region with area �·

Definition 7. 1 . 7 . Given a metric space (X, d) , a geodesic in (X, d) is a

path 'Y that is locally distance minimizing. I.e. , for each x E X, there is a
neighborhood U such that for any two points in 'Y n U the distance between
these two points is the length of the subarc of 'Y n U connecting them.
Definition 7. 1 .8. A complete geodesic is a bi-infinite geodesic.
Theorem 7. 1.9. The complete geodesics in vn are vertical rays and half­
circles that are orthogonal to !Rn- l x {O} c !Rn . (The geodesics in vn are
subsegments of the complete geodesics.)
For a proof of this fact see for instance John Radcliffe's book Foundations
of Hyperbolic Manifolds, [124] . In attempting to understand a metric space,
the symmetries of the space are of particular interest. Recall the notion of
isometry from Chapter 1 . It generalizes the notion of symmetry.
Example 7. 1 . 10. Self-homeomorphisms of vn of the form
(x 1 , . . . , Xn ) -t (ax 1 , . . . , axn )
are called dilations. Since
ds =
J(dx 1 ) 2 + · · · + (dxn ) 2
...._
.:.._ _______

Xn
-t
J( daxi) 2 + · · · + (daxn ) 2 =
Ja2 (dxi) 2 + · · · + a2 (dxn ) 2 =
�
axn axn
under this self-homeomorphism, dilations are isometries.
Example 7. 1 . 1 1 . Self-homeomorphisms of vn of the form
(x 1 , . . . , xn ) -t (x 1 + bi , . . . , Xn + bn )
7. 1 . Basic Hyperbolic Geometry 219

are called translations. Since

-Idx2 + . . . + dx 2
ds = V I n
Xn
2
-t Jd(x I + b I ) + · · · + d(x n + bn )
2 Jdxr + · · · + dx�
= = ds
Xn Xn
under this self-homeomorphism, translations are isometries.
Example 7. 1 . 12. The self-homeomorphism of l!Jn given by
(
(x I , . . . , xn ) -t 2 .X. I. 2 , . . . , 2 .X.n. 2
X I + + Xn X I + + Xn
)
is an example of an inversion. In the exercises, you will prove that this
inversion is an isometry.
Example 7. 1 . 13. The self-homeomorphisms of l!Jn of the form

for A E SO( n 1) are called rotations. In the exercises, you will prove that
-

rotations are isometries.

The group generated by translations acts transitively on vertical rays.
The group generated by dilations, rotations, and translations acts transi­
tively on half-circles orthogonal to JRn - I x {O} c IRn . Also note that the in­
version in the third example maps the vertical ray limiting on (1/2, 0, . . . , 0 )
to a half-circle orthogonal to ]Rn - I x {O} C ]Rn and limiting on (0, . . . , 0) and
(2, 0, . . . , 0 ) . Thus the group generated by dilations, translations, and the
given inversion acts transitively on geodesics. In fact, you will show in the
exercises that this group is the full group of isometries of l!Jn . This group is
denoted by either Isom(l!Jn ) or Mob(l!Jn ) .
Definition 7. 1 . 14. A convex subspace of vn is a subset X c vn such that
X contains all geodesic segments between pairs of points in X.
Remark 7. 1 . 15. The space sn - I = (JRn - I x {O}) U {oo} is called the sphere
at infinity. Thus, for instance, the top "end" of a vertical ray is said to lie
on the sphere at oo. Likewise, the two "ends" of a complete geodesic lie (at
two distinct points) on the sphere at oo.
Definition 7. 1 . 16. A hyperbolic triangle is a 2-dimensional convex subset of
vn that is bounded by three geodesics (either geodesic segments or complete
geodesics) that connect three points each of which lies either in vn or on
the sphere at oo. If the hyperbolic triangle is bounded by three complete
geodesics (and all "vertices" lie on the sphere at infinity) , then the triangle
is called an ideal triangle.
220 7. Further Topics

Figure 7.3. An ideal triangle.

In Example 7. 1 .4 we computed the area of a particular ideal triangle. It

is 1T'. In the exercises, you will show that all ideal triangles are isometric. A
consequence of this is that the area of any ideal triangle is 1T' .

Exercises

Exercise 1. Show that the distance in 1U2 between (0, a) and (0, b) is ln � ·
Exercise 2. Show that the inversion in Example 7. 1 . 12 is an isometry.
Exercise 3. Show that rotations are isometries.
Exercise 4. Show that dilations, along with translations, as well as the one
inversion given in Example 7. 1 . 1 2 generate the full group of isometries of
hyperbolic space, Isom(1Un ) .
Exercise 5. Use compositions of dilations, translations, and the inversion
given in Example 7. 1 . 12 to show that all ideal triangles are isometric.

7. 2 . Hyperbolic n-Manifolds**

In Chapter 1, we discussed examples of manifolds with additional structure.

Hyperbolic manifolds are another such example. For more on hyperbolic
manifolds, see [8] , [74] .
7.2. Hyperbolic n-Manifolds 221

Definition 7.2.1. A hyperbolic n-manifold is a manifold M with an atlas

{ (Ma, <Pa)} that satisfies the following additional requirements:
• for each a, </>a : Ma -+ lUn ;
• for each pair a, /3, the transition map </>13 o ¢-;; 1 is the restriction of
an isometry of un .
Example 7.2.2. Open subsets of un are hyperbolic n-manifolds.
A hyperbolic manifold M inherits the pull-back metric from its charts.
The fact that transition maps are isometries guarantees that the metric
is well-defined. Thus metric notions such as distances and geodesics are
defined locally on M. Moreover, angles between two geodesics intersecting
in a point can be measured by measuring the Euclidean angle between their
tangents at the point.
From now on, unless stated otherwise, we will always assume that hy­
perbolic manifolds are connected and complete. If M is a compact manifold,
then we say that M is hyperbolic if its interior is a hyperbolic manifold.
Below, we will be interested in a triangle with angles ( i , i • i ) · To
exhibit such a triangle, proceed as follows: First consider the ideal triangle
bounded by the upper half of the unit circle and the vertical rays limiting
on ( - 1 , 0) and ( 1 , 0) as in Figure 7 . 1 .
The angles in this triangle are all 0. Now replace the vertical rays by
large semicircles, orthogonal to IR.n - l x {O} that limit on ( - 1 , 0) and ( 1 , 0)
and pass through (0, t) , for t > 1. See Figure 7.4. Then note that t can be
chosen so that the angle at (0, t) is i ·
Next, we replace the upper half of the unit circle by a semicircle (centered
at the origin) that passes through (0, s ) , for 1 < s < t. See Figure 7.5. As
s increases continuously, the equal angles in the isosceles triangle increase
continuously from 0 to 37r /8. (In the limit, when s = t, the triangle is an
infinitesimally small Euclidean triangle.) Thus for some intermediate value
of s , we obtain angles of 7r /8. This provides a hyperbolic triangle with angles
( 47r , 87r • 87r ) .
A similar argument can be used to show that for any a, /3, 'Y � 0 such
that a + /3 + 'Y < 7r, there is a hyperbolic triangle with angles a, /3, 'Y· See
the exercises.
Before discussing other examples, we wish to state the relation between
hyperbolic n-manifolds and subgroups of Isom(lUn ) . To do so, we must define
the following notion:
Definition 7 .2.3. Let r be a group of homeomorphisms acting on a topolog­
ical space X. A group r acts properly discontinuously if for any compact sub­
set K c X, KngK is non-empty for only finitely many g E G. Furthermore,
222 7. Further Topics

Figure 7.4. Constructing the isosceles triangle.

Figure 7.5. Achieving the desired angles.

r acts freely if for every x E x the stabilizer of x r = {g E r I gx = x }

I I x I

is trivial.
Theorem 7.2.4. M is a closed hyperbolic n-manifold if and only if it is
the quotient of vn by a subgroup r of isometries of vn that acts freely and
properly discontinuously on vn .
The proof of this theorem is not difficult if one accepts the (non-trivial)
fact that a simply connected hyperbolic n-manifold must be isometric to vn .
7.2. Hyperbolic n-Manifolds 223

Example 7 .2.5. The closed orientable surface of genus 2 can be given

a structure of a hyperbolic manifold. To see this, glue eight hyperbolic
triangles with angles ( � , � , �) by isometries in such a way that the eight
angles of � match up. This yields an octagon with angles of � .

Figure 7.6. An octagon with edge identifications.

When we identify opposite edges in this octagon, as in Figure 7.6, we

obtain a genus 2 surface with a hyperbolic structure.
Example 7 .2.6. Higher genus closed orientable surfaces are also hyperbolic
manifolds. Indeed, they can be realized as covering spaces of the genus 2
surface. The hyperbolic structure lifts to define a hyperbolic structure on
the higher genus surface.
Example 7.2.7. The complement of the figure 8 knot.
In the exercises you will be invited to study Thurston's description of
a hyperbolic structure for the figure 8 knot; see [154] . The study of low­
dimensional hyperbolic manifolds was profoundly shaped by the work of
William Thurston. In 1982 he was awarded the Fields Medal for his work.
(He received the medal in 1983, due to the rescheduling of the ICM because
of martial law in Poland.) Part of Thurston's vision was communicated in
his book (see [154] ) and lecture notes (see [153] ) . See also [8] .
The contemplation of hyperbolic structures raises three questions: (1)
Do hyperbolic structures exist on a given manifold? (2) If so, are they
unique? (3) How do they help us to understand low-dimensional manifolds?
224 7. Further Topics

In partial answer to these questions, we saw in the previous section that

closed orientable surfaces of genus at least 2 admit hyperbolic structures.
Thus hyperbolic structures exist on these 2-dimensional manifolds. Note,
however, that neither the sphere nor the torus admits a hyperbolic structure.
In the case of 3-dimensional manifolds, theorems of Thurston and Perelman,
stated below, provide necessary and sufficient conditions for 3-manifolds to
possess a hyperbolic structure. Thus we have a complete understanding of
the existence of hyperbolic structures in dimensions 2 and 3.
Definition 7.2.8. A 3-manifold M is anannular if it contains no essential
annuli. It is atoroidal if every 7r 1 -injective map of the 2-torus into M is
homotopic into 8M.
Theorem 7.2.9 (W. Thurston) . An anannular atoroidal Haken 3-manifold
with (possibly empty) toroidal boundary admits a hyperbolic structure on its
interior.
The following theorem is known as the Hyperbolization Theorem. It
was conjectured by Thurston and proved by Perelman. See [121] , [123] ,
and [122] .
Theorem 7.2.10 (Perelman) . If M is a compact anannular atoroidal 3-
manifold with (possibly empty) toroidal boundary whose universal cover is
contractible, then M is hyperbolic.
In Theorem 7.2. 10, all the assumptions are necessary; this theorem does
not hold for 3-manifolds in general. For instance, Seifert fibered spaces pro­
vide counterexamples if we drop the assumption of being atoroidal. The
theorem below generalizes Theorem 7.2. 10. It is known as the Geometriza­
tion Conjecture of Thurston.
Theorem 7.2. 1 1 (Geometrization Conjecture) . Let M be a closed prime
3-manifold that does not contain 2-sided projective planes. Then M admits
a decomposition along disjoint incompressible tori and Klein bottles into
pieces Mj such that each Mj is geometric,. i. e., each Mi is either hyperbolic
or Seifert fibered.
Perelman's outline of the proof of Theorem 7.2. 1 1 can be found in [121] ,
[122] , [123] . For details, see [9] , [23] , [27] , [79] , [103] , and [104] . For
a larger context, see [97] . Perelman's work built on and realized Richard
Hamilton's vision of the Ricci flow on manifolds; see [54] , [55] . Perelman's
methods and results provided profound insight into many issues pertaining
to Riemannian 3-manifolds. In particular, he proved the Poincare Conjec­
ture. For this achievement, Perelman was awarded, but declined to accept,
the Fields Medal in 2006. He was also awarded, but declined to accept,
7.2. Hyperbolic n-Manifolds 225

$1 ,000,000 awarded him by the Clay Institute for solving a Millennium Prize
Problem, the Poincare Conjecture.
In fact, hyperbolic structures not only exist on many manifolds but can
be abundant. Closed orientable surfaces of genus � 2 , for instance, admit
infinitely many distinct hyperbolic structures. The set of all hyperbolic
structures on a given surface can be endowed with a suitable topology and
is known as the Teichmiiller space of the surface. In contrast, a (relatively)
classical result in the study of hyperbolic manifolds is the following:
Theorem 7.2.12 (The Mostow Rigidity Theorem) . If M1 and M2 are ho­
motopy equivalent hyperbolic n-manifolds (of finite volume) for n � 3, then
Mi and M2 are isometric. In particular, they are homeomorphic. Further­
more, the homotopy equivalence is homotopic to an isometry.
In the context of low-dimensional manifolds, we are interested in the
topological implications of hyperbolic structures. Specifically, what does the
fact that a 3-manifold possesses a hyperbolic structure tells us about the 3-
manifold? In fact, current research often considers the space of all hyperbolic
structures on a manifold. See, for instance, [17] . We now consider several
applications of hyperbolic geometry to problems in topology.
We already saw an application of hyperbolic geometry in Section 2.6
when we alluded to an important theorem of Thurston. It deserves to be
referenced again in this context: the classification of elements of the mapping
class group into periodic, reducible, and pseudo-Anosov. We will discuss this
classification in more detail in Section 7.6. The classification was inspired
by work of Dehn and it was Nielsen who first used hyperbolic structures for
its study. In the end, Thurston proved the Classification Theorem. In fact
he provided more than one proof. One of his proofs employed hyperbolic ge­
ometry, specifically, the hyperbolic structures possessed by closed orientable
surfaces of genus at least 2.
Another example of the use of hyperbolic geometry to solve problems
in low-dimensional topology is the Smith Conjecture. It asks the following:
Given a finite-order orientation-preserving diffeomorphism of the 3-sphere
with non-empty fixed set, must the fixed set be the unknot? That this is
indeed the case emerged from the work of several mathematicians, most
notably Thurston, who employed hyperbolic structures on 3-manifolds. See
[105] .
Our final application of the use of hyperbolic geometry to study prob­
lems in low-dimensional topology is the more recent solution to the Tameness
Conjecture. This line of investigation goes back to Waldhausen, who proved
that the universal cover of a Haken 3-manifold is R3 . See [156] . As a gener­
alization of this result, Simon conjectured that if M is a compact 3-manifold
226 7. Further Topics

and M is a covering space of M with finitely generated fundamental group,

then M is tame. (Recall that a 3-manifold is tame if it is homeomorphic to
the interior of a compact 3-manifold.) This conjecture was proved recently
by a combination of the work of several people, most notably Perelman,
Calegari-Gabai (see [20] ) , and Agol (see [3] ) . The crux of the matter was
to handle the case of hyperbolic 3-manifolds. In this case it is not necessary
to consider covering spaces. As proved by Agol and independently Calegari­
Gabai, a hyperbolic 3-manifold with finitely generated fundamental group
is tame. This fact was also known as the Marden Conjecture.
Exercises
Exercise 1 . Let a, (3, 'Y > 0 with a + f3 + "/ < 1r. Construct a hyperbolic
triangle with angles a, (3, 'Y · (Hint: Proceed as in Example 7.2.2. )
Exercise 2. Show that a closed orientable surface of genus n > 2 is a
hyperbolic manifold.
Exercise 3. Understand Thurston's example of a hyperbolic structure for
the figure 8 knot complement. (This is described on pages 39 through 42 in
his book.)
Exercise 4. Watch the movl.e Not Knot.
7.3. Dehn Surgery I

The idea behind Dehn surgery is simple: Given a knot (or link, respectively)
in K c §3 , set C(K) = §3 \ry(K) . Now create a new 3-manifold by attaching
a solid torus (or solid tori, respectively) to the components of 8C(K) . More
needs to be said concerning the specifics of the regluing. The goal here is to
obtain a new 3-manifold, not to simply reconstruct §3 . For this reason we
introduce a coordinate system. To do so, we must first prove a lemma.
Lemma 7.3.1. Let K c §3 be a knot. Set C(K) = §3 \ry(K) . Let Si , S2 be
Seifert surfaces for K. Then S1 n aC(K) is parallel to S2 n 8C(K) .
Proof. Isotope S1 and S2 so that S1 n C ( K) and S2 n C ( K) intersect in arcs
and simple closed curves. Note that if Si n C(K) is compressible, then we
may perform compressions to obtain an incompressible surface. The result
is a surface that is not homeomorphic to Si but that has the same boundary
as Si n C(K) . Hence, as we are only interested in a statement about the
boundaries of S1 n C(K) and S2 n C(K) , we may assume that S1 n C(K)
and S2 n C(K) are incompressible.
Now if there are simple closed curves in the intersection of S1 n C(K)
and S2 n C(K) , then they may be removed by a standard innermost disk
7. 3. Dehn Surgery I 227

argument. Further isotope 81 n C(K) and 82 n C(K) so that there are

as few components of intersection as possible. Then consider an arc of
intersection. Recall that Seifert surfaces are oriented. Along an arc of
intersection, the right-hand rule induces an orientation on this arc. This
orientation determines an initial + or - endpoint and a terminal - or +
endpoint of the arc. In particular, the number of initial points is the same
as the number of terminal points.
Now consider how the two torus knots 81 n {)C(K) and 82 n 8C(K)
intersect on the torus {)C(K) . Recall that they have been isotoped to inter­
sect in as few points as possible. By marking the plus and minus sides of
81 n {)C(K) and 82 n aC(K) on 8C(K) , we can keep track of whether the
points of intersection are initial or terminal points. Note, however, that as
the torus knots intersect in a minimal number of points, they are either all
initial points or all terminal points. Thus there can be no such intersections
and the curves in 81 n 8C(K) and 82 n {)C(K) are parallel D

Definition 7.3.2. Let K c §3 be a knot. Set C(K) = §3 \17(K) . Denote by

m the curve on {)C(K) that bounds a disk in N(K) . We call this curve the
meridian. Any curve on 8C(K) that intersects m exactly once is called a
longitude. Let 8 be a Seifert surface for K. Denote by l the curve 8n8C(K) .
We call this curve the preferred longitude.
The process of removing 17(K) from §3 and attaching a solid torus to
the resulting 3-manifold in such a way that a meridian goes to a curve of
slope m/l ( i.e. , wraps m times around the meridian and l times around the
preferred longitude) on 8C(K) is called m/l-Dehn surgery. The case r = 1/0
is also called co-surgery or trivial surgery.
If M is a compact 3-manifold with a specified torus boundary component
on which coordinates have been fixed, then the process of attaching a solid
torus to M in such a way that the meridian goes to a curve of slope r on
the specified torus boundary component of M is called a Dehn filling of
M and is denoted by M(r) . Similarly, if M has several torus boundary
components with specified coordinates, we write M(r 1 , . . . , rk ) for the 3-
manifold resulting from the appropriate sequence of Dehn fillings.
As an example, let us draw the preferred longitude of the trefoil. A
Seifert surface of the trefoil is pictured in Figure 7. 7.
Thus the preferred longitude is pictured in Figure 7.8.
Equivalently, the preferred longitude of the trefoil knot is pictured in
Figure 7.9. In other contexts, the preferred longitude is called the natural
framing. It is usually distinct from the so-called blackboard framing which
derives its name from the fact that it is the closed curve that remains on the
front side of a regular neighborhood of the knot as depicted on a blackboard.
228 7. Further Topics

Figure 7. 7. The Seifert surface of the trefoil knot.

Figure 7.8. The preferred longitude of the trefoil knot.

Being able to construct a given surface homeomorphism from a sequence

of Dehn twists has profound ramifications in the study of 3-manifolds. As
we will see, Heegaard splittings along with this description of surface home­
omorphisms allows us to obtain a given 3-manifold via Dehn surgery on a
knot or link.
7.3. Dehn Surgery I 229

Figure 7.9. The preferred longitude of the trefoil knot.

Lemma 7.3.3. Let M be the connected sum of g factors of §2 x § i . Then

M is obtained by (I/O, . . . , I/0) -Dehn surgery on the g component unlink
in §3 .
Proof. We consider first the case in which g = In this case I/0-Dehn
1.
surgery involves removing a regular neighborhood of the unlink, which cre­
ates a solid torus V, and then attaching a solid torus W to the resulting
boundary component in such a way that a meridian of W goes to a meridian
of V. This yields §2 x § i .
More generally, consider the g component unlink. Separate the g compo­
nents by a disjoint collection S of g - I 2-spheres in §3 . On each component
of the unlink, perform I/0-Dehn surgery. Now S is a set of decomposing
spheres that factors the resulting 3-manifold into g factors, each homeomor­
phic to §2 x § i . D

Lemma 7 .3.4. Let S be a closed orientable surface of genus g. Let c be

a simple closed curve in S. Let f be a Dehn twist around c. Let Mi be
the 3-manifold obtained by identifying two genus g handlebodies along their
boundaries via f . Let M2 be the 3-manifold obtained by identifying S with the
splitting surface of the standard genus g Heegaard splitting of the connected
sum of g factors of §2 x § i and then performing I/I -Dehn surgery along c.
Then Mi is homeomorphic to M2 .

Proof. Both 3-manifolds in question have genus g Heegaard splittings. In

the case of Mi , this follows from the construction. In the case of M2 ,
230 7. Further Topics

Figure 7.10. Before the Dehn surgery.

Figure 7 . 1 1 . After the Dehn surgery.

Figure 7.12. The curve c.

see Figure 7. 1 1 . In M2 , we can isotope c to lie just below the splitting

surface and then consider the Dehn surgery to be taking place entirely in
one handlebody. Denote this handlebody by H.
Let A be an annulus between c and the boundary of H. Denote a slightly
truncated version of A lying in H\17(c) by A'. Now consider the effect of
the Dehn surgery on c in H. It is the same as the effect of removing 17(c)
from H, cutting along A' , performing a full twist, regluing along A' , and
replacing 17(c) .
Exercises 231

Figure 7.13. Dehn surgery as Dehn twist.

A curve on 8H will have changed by a Dehn twist along c. In particular,

the curves that specify how to attach the 2-handles of the complementary
compression body will have changed by a Dehn twist along c. I.e. , Mi is
homeomorphic to M2 . 0

We now prove the main theorem:

Theorem 7 3 5 Every closed orientable 3-manifold can be obtained by
. . .

Dehn surgery on a link in §3 .

Proof. Let M be a closed orientable 3-manifold. Then M has a Heegaard
splitting M = V Us W. Let g be the genus of this Heegaard splitting. By
Lemma 7.3.3 we obtain the connected sum of g copies of §2 x § 1 by Dehn
surgery on a link in §3 . Here the Heegaard splitting of the connected sum
of g copies of §2 x § 1 has two copies of the handlebody V identified along
their boundaries via the identity map.
By Theorem 2.6.5 we can factor f into Dehn twists, i.e., f = Ji o · · · o f n,

where each fi is a Dehn twist around a curve Ci · Consider now the 3-

manifold that is the connected sum of g factors of §2 x § 1 . Further consider n
parallel copies S1 , . . . , Sn of the splitting surface of this 3-manifold. Cutting
along each surface Si and reidentifying via the Dehn twist fi yields M. By
Lemma 7.3.4, this goal may also be attained via 1/1-Dehn surgeries along
the curves Ci . 0

Exercises

Exercise 1. Draw the longitude for the figure 8 knot.

Exercise 2. Generalize the notion of Dehn surgery on a knot to Dehn
surgery on links.
Exercise 3. Describe how to obtain L(2, 1) by surgery on a link in §3 .
Exercise 4. Describe how to obtain the 3-torus § 1 x §1 x § 1 by surgery on
a link in §3 .
232 7. Further Topics

7.4. Dehn Surgery II

In the 1980s Gordon and Luecke began approaching the study of 3-manifolds
using Dehn surgery. Their school of thought caught on quickly and continues
to produce results. See [4 6] and [30] .
Definition 7.4. 1 . For M a compact 3-manifold with torus boundary and
fixed coordinates on the torus boundary, the result of doing a Dehn filling
along r is denoted by M(r) . If M(r) has cyclic fundamental group, then the
Dehn filling is called a cyclic surgery.
This definition is motivated by the example of knot complements. Knot
complements have a single torus boundary component. As we saw above,
the torus boundary component inherits a natural coordinate system. We
are interested in those slopes for which Dehn surgery on the knot yields a
3-manifold with cyclic fundamental group.
Definition 7.4.2. Let r, s be torus knots that have been isotoped to inter­
sect as few times as possible. Then we denote the number of intersections
of r and s by b.. ( r, s) .
You will show in the exercises that b.. (mi/li , m2 /l 2 ) = m 1 l2 - m2 li .

Figure 7.14. D. ( 1/2, 0/ 1) = 1 .

Theorem 7 . 4 . 3 ( The Cyclic Surgery Theorem by Culler-Gordon-Luecke­

Shalen ) . Suppose that M is not a Seifert fibered space. If n 1 (M(r)) and
n 1 (M(s) ) are cyclic, then l b.. ( r, s) I :S 1 .
Theorem 7.4.4 ( Gordon-Luecke ) . If two knots have homeomorphic com­
plements, then they are equivalent.
One of the fundamental strategies employed by Gordon and Luecke was a
combinatorial analysis of intersection patterns of surfaces. We will consider
a few of the definitions involved and prove a lemma that illustrates the
strength of these techniques.
7.4. Dehn Surgery II 233

Definition 7 .4.5. Let M be a knot complement. Suppose P and Q are

properly embedded compact connected planar incompressible surfaces in M
that are isotoped to intersect in as few components as possible. Denote the
slope represented by 8P by 7r and the slope represented by 8Q by 'Y ·
Denote a capped off copy of P, i.e. , a copy of P with a disk attached
to each boundary component, by P. Denote a capped off copy of Q by Q.
We construct graphs in P and Q: Each disk that has been adjoined to P to
form P is a "fat" vertex. Each arc of intersection of Q with P is an edge.
This defines a graph in P. We denote this graph by fp. Analogously, we
obtain a graph in Q. We denote this graph by rQ .
Remark 7.4.6. The minimality assumption on the number of components
of intersection of P and Q guarantees that no arc of P n Q is parallel to
either 8P or 8Q.
Remark 7.4.7. There is a l-to-1 correspondence between edges in f p and
rQ .

Label the components of 8P by 1 , 2, . . . , np, likewise for 8Q. Then

it follows from our minimality assumptions on the number of components
of intersection of P n Q (and hence also 8P n 8Q) that on a boundary
component of 8Q we encounter intersections with components of 8P labeled
1 , 2, . . . , np, 1, 2, . . . , np, . . . , 1, 2, . . . , np and on a boundary component of
8P we encounter intersections with components of 8Q labeled 1, 2, . . . , nQ ,
1 , 2, . . . , nQ, . . . , 1, 2, . . . , nQ in that order. See Figure 7. 15.

Figure 7. 15. Labels on rp .

234 7. Further Topics

Definition 7.4.8. An orientation of P ( or Q , respectively ) induces orien­

tations on the components of 8P. We say that two components of 8P are
parallel if they inherit the same orientation ( on 8M) and we say they are
antiparallel if they inherit opposite orientations (on 8M) . We call vertices in
fp (or fQ , respectively ) parallel if they correspond to parallel components
of {)P ( or 8Q , respectively ) . We call vertices in fp ( or fQ , respectively )
antiparallel if they correspond to antiparallel components of {)P (or 8Q,
respectively) .
For an oriented edge, we denote the component of oe that has orientation
- 1 by {)_e, and the component that has orientation + 1 by 8+ e. For an
oriented vertex x, we denote the same vertex with the opposite orientation
by x .
The arcs of P n Q are arcs of intersection of orientable surfaces in an
orientable manifold. This leads to the following parity rule: If e is an edge
of r p that connects parallel vertices of r p , then the labels at its endpoints
represent antiparallel vertices of rQ . If e is an edge of r p that connects
antiparallel vertices of r p , then the labels at its endpoints represent parallel
vertices of rQ .

- - - - - - - - - - - - - - - - - - - - ,
'
'
'
'
'
'
'
'
'
'
'
'

' .

!.. - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - _,

Figure 7 . 1 6 . Parity rule for an edge.

Definition 7.4.9. An x-cycle in f p or fQ is a sequence of edges e i , . . . , en

such that the following hold for a suitable choice of orientations:
(1) the edges are all distinct;
(2) {)_ei+l = 8+ ei for i = 1 , . . . , n - 1 and {)_e 1 = 8+ en;
(3) the vertices oei are all parallel;
(4) the label at {)_ei is x for i = 1, . . . , n .
7.4. Dehn Surge1y II 235

Definition 7.4.10. A Scharlemann cycle is an x-cycle in rp (or rQ , respec­

tively) such that the interior of one of the disks in P (or Q, respectively)
that the cycle bounds is disjoint from rp (or rQ , respectively) .

y x

Figure 7. 17. A Scharlemann cycle.

Definition 7.4. 1 1 . The length of a cycle is the number of edges it contains.

Scharlemann cycles play an important role in the combinatorics of in­
tersection patterns of pairs of surfaces. Most importantly, they provide a
strategy for simplifying such patterns. The following two lemmas give two
immediate applications of Scharlemann cycles.
Lemma 7.4. 12. Neither rp nor rQ contains a Scharlemann cycle of length
1.

Proof. (This is [30, Lemma 2.5.l] . ) Consider a cycle of length one in rp .

It corresponds to an arc of intersection of P n Q with both ends on the same
component of 8P. If it is a Scharlemann cycle, then this arc of intersection
cobounds, together with an arc in 8M, a disk in P. Furthermore, the
interior of this disk is disjoint from r p and hence from Q. It therefore
defines a boundary compression. This in turn defines a compression, but
this is a contradiction because P was assumed to be incompressible. D

Lemma 7.4.13. If rp contains a Scharlemann cycle, then M('Y) has. a lens

space summand.

Proof. (This is [138, Proposition 5.6] .) Denote by D the subdisk of P

bounded by the Scharlemann cycle whose interior is disjoint from rp . A
posteriori, D lies in P since it is disjoint from r p and hence does not meet
the components of P\P as these correspond to "fat" vertices of rp . The
236 7. Further Topics

boundary of D is partitioned into subarcs that lie alternately in 8M and

P n Q.
All edges in the Scharlemann cycle have the same label at their initial
point. Recall the I-to-I correspondence between edges of rp and rQ . The
fact that the edges of the Scharlemann cycle in r p all have the same labels
at their initial point means that they all correspond to edges in r Q beginning
at the same "fat" vertex of rQ. Denote this "fat" vertex of rQ by e_ .
Because the interior of D is disjoint from rp, an edge in the Scharlemann
cycle connects e_ to a vertex that corresponds to a component of 8Q that
is adjacent to the one corresponding to e_ . Recall that all vertices in a
Scharlemann cycle are parallel. Recall also that if vertices are parallel in
rp, then they are antiparallel in rQ. Thus an edge in the Scharlemann
cycle in r p corresponds to an edge in rQ that connects e_ to an antiparallel
vertex. There are at most two candidates for such a vertex, but since Q
is orientable (so 8D meets only one side of Q) , there is exactly one such
vertex. We denote it by e + .
We can embed a copy of Q in M("f) . Consider the solid torus J'Y that
has been attached to M to obtain M('Y) . There are two meridian disks in J.'Y
bounded by the curves corresponding to e_ and e + , respectively. We denote
these by E_ and E+ , respectively.
Denote the 3-ball (solid cylinder) of [)J'Y cobounded by [)E_ and 8E+
that meets 8D by B. Then V = B U N(Q) is a solid torus. (Here Q is
a 2-sphere and B is a I-handle attached to N(Q) .) Thus V U N (D) is a
punctured lens space. 0

Note, in particular, that the fundamental group of the punctured lens

space obtained above is Z/pZ, where p is the number of edges in the Scharle­
mann cycle.
Definition 7.4. 14. A knot K is said to satisfy Property P if any non-trivial
Dehn surgery on K yields a 3-manifold that has non-trivial fundamental
group.
In [1 1] , Bleiler and Scharlemann show that this definition is equivalent
to the following alternate definition:
Definition 7.4. 15 (Alternate definition) . A knot K is said to satisfy Prop­
erty P if ±I-surgery on K yields a 3-manifold that has non-trivial funda­
mental group.
Theorem 7.4. 16. Every non-trivial knot satisfies Property P.
The proof of Theorem 7.4. I6, due to Kronheimer and Mrowka, can be
found in [82] . Earlier contributions can be found in [16] , [34] , [31] , and
Exercises 237

[41] . A related, but more specific, question is whether or not Dehn surgery
on a knot in §3 can result in §2 x § 1 .
Definition 7 .4. 17. A knot K is said to have Property R if no Dehn surgery
on K yields §2 x § 1 •
Somewhat more generally, one can ask about Dehn surgeries on knots
that produce manifolds with an §2 x § 1 summand. Poenaru conjectured
that no non-trivial such knots exist:
Conjecture 7 . 4 . 18 (Poenaru) . No surgery on a non-trivial knot yields a
manifold with an §2 x § 1 summand.
The following theorems were proved by Gabai; see [43] , [44] .
Theorem 7 . 4 . 19 . Every non-trivial knot satisfies Property R.
Theorem 7 . 4 . 20 . The Poenaru Conjecture is true.
The 271'-Theorem states conditions under which the result of Dehn filling
a 3-manifold results in a hyperbolic 3-manifold. It was proved by Gromov
and Thurston and improved by Agol and, from a different point of view, by
Lackenby. We state it here in abbreviated form:
Theorem 7 . 4 . 21 (The 271'-Theorem) . Let M be a hyperbolic 3-manifold that
is the complement of a knot in §3 . With the exception of at most 12 slopes,
the result of Dehn filling M is hyperbolic.
More generally, one can ask what conditions allow the result of surgeries
on a knot, or a class of knots, to have a given attribute, e.g. , reducible,
atoroidal, Haken, or hyperbolic. See the work of Rieck, Sedgwick, and
others. Using Definition 7.4.22, graphic renditions of the results of such
investigations have been given by Hall, Schleimer, and Segerman.
Definition 7.4.22. Let M be a compact 3-manifold with 8M a torus. The
set {M(r) I r E Q} is called the Dehn surgery space of M.

Exercises

Exercise 1. Show that !::;. ( mi /Li , m 2 /l 2 ) = m 1 l2 - m2 li .

Exercise 2. What does the Cyclic Surgery Theorem tell us about the
number of slopes such that Dehn surgery along that slope is a cyclic surgery?
Exercise 3. Describe how Properties P and R relate to the Poincare Con­
jecture.
238 7. Further Topics

7.5. Foliations
Product manifolds enjoy a special place among manifolds due to their natu­
ral quotient maps. A question pertaining to a given product manifold often
reduces to an analogous question on a manifold of lower dimension, where it
is easier to solve. Of course, not every manifold is a product manifold. Foli­
ations provide a local product structure that can be used in lieu of a global
product structure. For example, recall how the product structure on R3 was
used in our proof of the Schonfl.ies Conjecture. Analogous arguments can be
used on manifolds with appropriate foliations. For more on foliations, see
[21] , [22] , [44] , [50] , [49] , [51] , and [1 14] .
Definition 7.5. 1 . For q E [O, oo] , a Cq-codimension p foliation of an n­
manifold M is a Cq-atlas with the following additional properties:
(1) for each chart (M00 <Pa) , Ma factors into RP x Rn -p ;
(2) Vr E RP :ls E RP such that the transition maps <Pa/3 : Ma n M13 -+
Ma n M13 satisfy <l>af3 : { r} x Rn-p -+ { s} x Rn-p.
We often denote a foliation by F.
A maximal connected ( n p)-dimensional submanifold F such that each
-

non-empty intersection F n Ma has the form (constants) x Rn -p is called a

leaf of the foliation.
Example 7.5.2. In Section 3.7 we discussed Seifert fibered spaces. Seifert
fibrations constitute a special case of codimension 2 foliations of 3-manifolds;
in fact, they provide all foliations of 3-manifolds whose leaves are circles.
For a specific example, consider the foliation of §3 = L( l , 1) discussed there.
This foliation is also known as the Hopf fibration of §3 .
We will be particularly interested in codimension 1 foliations of surfaces
and 3-manifolds. On a surface, a codimension 1 foliation decomposes the
surface into leaves that are I-manifolds. On a 3-manifold, a codimension 1
foliation decomposes the 3-manifold into leaves that are surfaces.
Example 7 .5.3. The set of parallel lines of a fixed slope in R2 defines a
codimension 1 foliation of R2 • Likewise, the set of parallel planes in R3
defines a codimension 1 foliation of R3 .
Example 7.5.4. The 2-torus 11'2 = § 1 x § 1 admits a codimension 1 foliation
with leaves x x § 1 .
Example 7.5.5. The 3-torus 11'3 = § 1 x § 1 x § 1 admits a codimension 1
foliation with leaves x x § 1 x § 1 .
If a manifold M', covers another manifold, M, and M' has a foliation
F' with charts that factor through the covering map, then F' projects to a
7. 5 . Foliations 239

foliation F of M. The leaves F' cover the leaves of F. Examples 7.5.4 and
7.5.5 provide illustrations.
To construct many more examples, consider the covering p : !Rn ---+ 'Il'n
given by p(x 1 , . . . , Xn ) = (ei21Txi , , e i21Txn ) and foliations of !Rn by parallel
• . .

codimension k subspaces. Note that the leaves of the foliations induced on

the n-torus can be compact or non-compact! E.g. , in the case of 1r2 , the
leaves are circles if the slope of the lines in the cover are rational and they
are lines if the slope of the lines in the cover are irrational.
Example 7.5.6. The infinite strip JR x [- 1 , 1] admits a foliation as pictured
in Figure 7. 18. The map p : JR x [- 1 , 1] ---+ § 1 x [- 1, 1] given by p(x, y) =
( e i21Tx, y) is a covering that defines a foliation of the annulus. The foliation
on the annulus thus obtained is called the Reeb foliation of the annulus.

Figure 7.18. A foliation of the infinite strip.

Example 7.5. 7. Rotating the above infinite strip around its core {O} x JR
sweeps out an infinite solid cylinder and defines a foliation. The infinite
solid cylinder covers the solid torus and defines a foliation of the solid torus.
This foliation is called the Reeb foliation of the solid torus. See Figure 7.19.

Example 7.5.8. A codimension 1 foliation for the 3-sphere can be built

from the genus 1 Heegaard splitting. The splitting surface is one leaf of the
foliation and it separates the 3-sphere into two open solid tori foliated with
Reeb foliations.
The only closed (connected) orientable surface to admit a codimension
1 foliation is the torus. (The technical reasons that preclude other closed
orientable surfaces from having a codimension 1 foliation are that such a
foliation induces a non-vanishing section of the projectivized tangent bundle
of the surface S. This section lifts to a non-vanishing section of the tangent
bundle of a 2-fold cover S' of S, i.e. , a non-vanishing vector field on S. By
the Poincare-Hopf Index Theorem, the existence of such a vector field forces
the Euler characteristic of S' to be 0. Since the Euler characteristic of a
p-fold cover of S is p x(S) , we see that x(S) = O; hence S is the torus.)
·
240 7. Further Topics

Figure 7.19. The Reeb foliation of the solid torus.

Recall that for a closed 3-manifold M, x (M) = 0 and there is no obvious

obstruction to the existence of codimension 1 foliations. In fact, we have:
Theorem 7.5.9 (Lickorish-Zieschang) . Every closed orientable 3-manifold
admits a codimension 1 foliation.

Sketch of proof. Recall two of the theorems of Alexander: (1) Every closed
orientable 3-manifold is a branched cover o/ §3 over a link; (2) every link can
be realized as a closed braid. Also recall the foliation F of §3 given above.
Let M be a closed orientable 3-manifold and suppose that M is a
branched cover of §3 over the link L. Isotope L to be a closed braid that
lies in a small neighborhood of the core curve of one of the solid tori fo­
liated by the Reeb foliation, so that each component L' of L intersects F
transversely. We need only lift the foliation of §3 to a foliation of M. There
is no obstruction to doing so: Away from L, the foliation lifts via a local
homeomorphism; near L, it also lifts, in this case because L is transverse to
F For more details, see [44] . D

The Reeb foliation plays an important part in the argument above for
a good reason: There are 3-manifolds whose foliations necessarily contain
a solid torus that is foliated by the Reeb foliation. However, for many
applications, we are interested in precisely the foliations that contain no
such solid torus.
7. 5 . Foliations 241

Definition 7.5.10. A foliation of a 3-manifold is Reebless if it contains no

solid torus foliated by the Reeb foliation.
Definition 7.5. 1 1 . Let M be a 3-manifold with foliation F. A transversal
to F is a simple arc "'( that meets F transversely. A closed transversal to F
is a simple closed curve "Y that meets F transversely.
The following theorem has been fundamental in the application of foli­
ations to the study of 3-manifolds. It is a deep theorem and we will only
sketch a proof of one of its parts.
Theorem 7.5.12 (S. Novikov [114] ) . Let M be a closed orientable 3-
manifold with a Reeb less foliation F. Then the following hold:
(i) if F is a leaf of F, then F is incompressible in M;
(ii) if "Y is a closed transversal to F, then no power of "Y is trivial in
tr1 (M) ;
(iii) either tr2 (M) = 0 or F is the product foliation on M = §2 x § 1 ;
(iv) M is not homeomorphic to S3 .
Idea of proof of (i) . Suppose that the simple closed curve c on the leaf F
of F bounds a disk D in M but not in F. Consider the intersection of the
leaves of F with D. See Figure 7.20.

Figure 7.20. Induced foliation on a disk.

It is a non-trivial fact, necessitating the assumption that F be Reebless,

that D can be isotoped so that away from a finite number of components
of intersection, this intersection is a foliation FD of D by closed curves.
Moreover, the remaining components of intersection must be figure 8 curves
and points.
If the number of such figure 8 curves is non-zero, we replace D by an
innermost subdisk of D cut out by a figure 8 curve. Now D is a disk with
242 7. Further Topics

a codimension 1 foliation away from a finite number of points. Recall the

application of the Poincare-Hopf Index Theorem discussed above. If D is
a disk with a codimension 1 foliation away from a finite number of points,
then the foliated subsurface of D must have Euler characteristic 0, hence
must be an annulus. Specifically, D meets F in simple closed curves parallel
to f)D and one point. See Figure 7.21 for a cross-section of the disk relative
to the foliation.
Locally, leaves of F are level sets of a Morse function f with no critical
levels. Near the critical point of f lv , say a maximum, D intersects leaves
of F in inessential circles. Thus D can be isotoped downward until it lies
in F.

Figure 7.21. Schematic for F near D.

Arguing by induction on the number of figure 8 curves in the intersection

of D with the leaves of F now provides an isotopy ( relative to fJD) of D into
F, contradicting our initial assumption. 0

Definition 7.5.13. A foliation F of M is taut if it has a closed transversal

'Y that intersects every leaf of F.
Lemma 7.5. 14. A taut foliation must be Reebless.

Proof. A closed transversal would not be able to enter and escape a solid
torus foliated by the Reeb foliation. 0

Novikov's Theorem has consequences:

Corollary 7.5.15. If M is a closed orientable 3-manifold that admits a taut
foliation, then 11" 1 ( M) is infinite.

Proof. By Lemma 7.5. 14, the taut foliation, F, is Reebless and admits a
closed transversal. Thus M is a closed orientable 3-manifold that admits
a Reebless foliation F and a closed transversal to F; hence, by Novikov's
Theorem, 11' 1 (M) is infinite. 0
7.6. Laminations 243

The above is j ust one of several useful properties enj oyed by 3-manifolds
that admit taut foliations. The work of Candel, Lawson, Thurston, Sulli­
van, and others yields many more insights into foliations, especially their
geometric structure. The work of Gabai provides specific applications of
foliations to the study of knots and 3-manifolds.

Exercises

Exercise 1 . Draw pictures of the foliations discussed in Examples 7.5.2

and 7.5.3. What is the closure of a leaf in Example 7.5.3?
Exercise 2. Prove that 'll'3 admits infinitely many taut foliations.
Exercise 3. Prove that §3 does not admit a taut foliation.
Exercise 4. Describe several foliations of 'll'5 .

7 . 6 . Laminations

We will focus our discussion of laminations mainly on the case of surfaces,

that is, 2-dimensional manifolds, though the definitions apply in general. As
we will see, the results for surfaces have direct applications in the world of
3�manifolds. For more detail, see [25] , [39] , [45] , and [12 0] .
Definition 7.6. 1 . A Cq-codimension p lamination of an n-manifold M is
a Cq-codimension p foliation of a closed subset of M. The leaves of the
lamination are the leaves of the foliation.
Example 7.6.2. A simple closed Cq-smooth curve in a surface S defines a
Cq-codimension I lamination of S.
Example 7.6.3. Consider the annulus A = § 1 x I and let C c I be the
Cantor set. Then {§ 1 x { x } I x E C } defines a codimension 1 lamination
of A.
Example 7.6.4. Figure 7.22 shows a codimension I lamination of the genus
2 surface. ( Note how some leaves wind around others. )

Definition 7.6.5. Let S be a surface with a codimension 1 lamination A

and let J be an arc in S that is transverse to A. A transverse isotopy of J
is an isotopy of J such that J remains transverse to F and such that the
endpoints of J either remain in the complement of A or remain in the same
( respective ) leaves during the entire isotopy.
A transverse invariant measure for A is a function f from arcs in S
transverse to A into the set of measures on the arcs, f(J) = µJ , such that
244 7. Further Topics

Figure 7.22. A lamination on a genus 2 surface.

the following hold:

• f is invariant under transverse isotopies of arcs;
• if I c J, then M = µ J l 1 ;
• supp(µ J ) = J n A.
A measured lamination on S is a lamination together with a transverse
invariant measure.
The condition on the support of µ is not necessary, but it will simplify
our discussion considerably. In particular, it guarantees that the topology
defined below is Hausdorff.
Example 7 .6.6. Let l be a simple closed curve in S and let c E JR + . Then
l is a lamination. We can define a function f that takes arcs transverse to
l and assigns them the number c n, where n is the number of times that
·

the given arc intersects l. This turns l into a measured lamination called a
weighted curve. Here c is called the weight of the simple closed curve.
Example 7.6. 7. Let F be the foliation of JR2 by parallel lines. We define a
transverse measure to F by projecting onto a line orthogonal to the leaves
of F and taking the length of the projection times some number c > 0.
Example 7.6.8. Consider the foliation of 1f2 given in Example 7.5.4. We
can define a function f that takes arcs transverse to this foliation and assigns
to a given arc J the number c r, where c E JR and r is the length of the arc
·

obtained by projecting J onto the first factor in § 1 x § 1 .

There is a natural operation of multiplication of a measured lamination
A by a positive real number k, denoted kA: The lamination itself does not
change while the transverse measure is multiplied by k .
From now on, we will assume that S is a hyperbolic surface (with a fixed
hyperbolic structure) and that our laminations are geodesic, that is, each
7. 6. Laminations 245

leaf is a geodesic. The set of all measured (geodesic) laminations on S is

denoted by MC(S) .
A closed geodesic c on S provides a function c : MC(S) --+ JR: For a
given geodesic measured lamination A on S, either c is transverse to A or it
is contained in a leaf of A. In the first case, c(A) is simply the measure of A
evaluated on c, where we can think of c as partitioned into subarcs. In the
second case, we decree that c(A) = 0 . We topologize MC(S) by giving it
the weakest topology in which all functions c obtained from closed geodesics
in this way are continuous. Thurston showed that weighted simple closed
geodesics (as in Example 7.6.6) are dense in MC(S) .
Definition 7.6.9. For weighted simple closed geodesics (c, w ) , (c', w' ) in the
closed hyperbolic surface S, the function i is defined by i ( (c, w ) , (c', w' )) =
w w · n, where n is the number of times that c intersects c' (transversely).
'
·

The function i is called the weighted intersection number.

Theorem 7.6. 10 (Penner) . The function i extends continuously to MC(S) .
The extension of i to MC(S) is called the intersection number between
measured laminations and is denoted
i : MC(S) x MC(S) --+ JR + .
Theorem 7.6. 1 1 (Thurston) . Let S be a closed orientable surface of genus
g 2:: 2 . There is a homeomorphism of MC(S) onto JR6Y - 6 . Moreover, this
homeomorphism preserves multiplication by positive real numbers.
Definition 7.6.12. We say that the measured laminations A and A' are
projectively equivalent if they are equal as laminations and their transverse
measures are positive scalar multiples of each other. The equivalence class
[A] of a (non-empty) measured lamination A under the equivalence relation
defined by projective equivalence is called a projective measured lamination.
The set of all projective measured laminations is denoted by P MC(S) .
Corollary 7.6. 13. Let S be a closed orientable surface of genus g 2:: 2 .
Then PMC(S) is homeomorphic to a sphere of dimension 69 - 7 .
A closed hyperbolic surface S of genus g > 1 can b e obtained from a 4g­
gon by identifying opposite sides. If the 4g-gon is regular, then a rotation of
the 4g-gon through an angle of, say, 21T /4g induces a periodic automorphism
of S of order 4g.
Recall from Section 2.6 that every surface homeomorphism is a product
of Dehn twists. A Dehn twist fixes the simple closed curve along which the
Dehn twist is performed and is hence reducible. Likewise, a product of Dehn
twists along a disjoint collection of simple closed curves is reducible.
246 7. Further Topics

Recall our brief discussion of the mapping class group in Section 7.3.
Note that a mapping class ( isotopy class of self-homeomorphisms of S ) is
periodic if it has a periodic representative and it is reducible if it has a
reducible representative.
For the following definition, note that measured foliations are a special
case of measured laminations.
Definition 7.6.14. Let S be a closed orientable surface and suppose h :
S --+ S is a homeomorphism. If there are measured foliations F+ , F- of S
such that F+ rh F- and such that h ( F+ ) = kF+ and h ( F- ) = k F- for
some k > 1 , then h is called Anosov. A mapping class is called Anosov if it
has an Anosov representative.
Example 7.6.15. Recall the covering of 1I'2 by IR2 discussed in Section 7.5.

[� �]
The matrix
Ah =
acts on IR2 as a linear transformation and induces a self-homeomorphism h
of 1I'2 via the covering map IR2 --+ 1I'2 . The eigenvalues of A h are
k = 3 / 2 + ../5/ 2 and � = 3/2 - ../5/ 2
and the corresponding eigenvectors are
( 1 , - 1 / 2 + ../5/ 2) and ( 1 , - 1 / 2 - ../5/ 2) .
We can foliate IR2 by lines of slope - 1 / 2 + .../5/ 2 to obtain f:+ and by lines
of slope - 1 / 2 - .../5/ 2 to obtain J:- . We endow these foliations with the
transverse measures as in Example 7.6.7. Then j:± proj ects to a measured
foliation F± on 1I'2 . Moreover, F+ and F- are transverse ( in fact, their
leaves are perpendicular ) . Finally, note that h ( F+ ) = kF+ and h ( F- ) =
k F- . Thus h is Anosov.
Due to the lack of foliations for closed oriented surfaces other than the
torus, only maps of the torus have a chance of being Anosov. For this reason,
the more general notion of pseudo-Anosov homeomorphisms given below is
of interest.
Definition 7.6.16. A measured lamination A is arational if it contains no
closed leaves. A lamination A is called uniquely ergodic if it admits a unique
( up to scaling) transverse measure.
It is easy to see that every uniquely ergodic measure is arational ( unless
it consists of a single closed curve ) , while the converse is false.
Definition 7.6.17. A lamination A on a closed orientable surface S is said
to be filling if S \ A consists of disks.
7. 6. Laminations 247

Definition 7.6.18. Let S be a closed orientable surface and suppose h :

S --+ S is a homeomorphism. If there are filling measured laminations
A + , A- of S such that A + m A- and such that h(A + ) = kA + and h(A- ) =
!A- for some k > 1 , then h is called pseudo-Anosov. Here A + is called the
unstable lamination and A - is called the stable lamination.
A mapping class is called pseudo-Anosov if it has a pseudo-Anosov rep­
resentative.

Theorem 7.6.19 ( Thurston) . The stable and unstable laminations, A± , of

a pseudo-Anosov homeomorphism have the following properties:
• A± are arational;
• A± have no proper sublaminations; moreover, A± are uniquely er­
godic;
• V[A'] E PM.C(S) \ {A- } the sequence { [hn (A')] } converges to [A + ] .

Theorem 7.6.20 ( Thurston ) . Let S be a closed orientable surface and let h :

S --+ S be an (orientation-preserving) homeomorphism. Then the mapping
class [h] is either periodic, reducible, or pseudo-Anosov.

Proof in the case that S is a torus. First note that [h] is determined
by its action on 7r 1 ( 1r ) . Moreover, this action is determined by a matrix, Ah ,
in S L (2, Z) . Understanding the structure of the matrix facilitates our un­
derstanding of the homeomorphism. Note that the characteristic polynomial
for such a matrix is t 2 - trace ( Ah ) t + 1 .
There are three cases: ( 1 ) I trace ( Ah ) I < 2; (2) I trace ( Ah ) I = 2; (3)
I trace ( Ah ) I > 2. We consider each of these in turn.
( 1 ) I trace ( Ah ) I < 2.
Since the entries in Ah are integers, the only possibilities are trace ( Ah ) =
0, ± 1 ( see Section 2.6) . In each of these three subcases, we can compute the
eigenvalues of Ah explicitly. They are ±i if trace ( Ah ) = 0, 1 / 2 ± i( ./3/ 2) if
trace ( Ah ) = 1 , and - 1 / 2 ± i( ./3/ 2) if trace ( Ah ) = - 1 . It follows that Ah is
idempotent and hence that h is periodic.
(2) I trace ( Ah ) I = 2.
Since the entries in Ah are integers, the only possibilities are trace ( Ah ) =
±2. In both of these subcases, we can compute the eigenvalues of Ah explic­
itly; in fact, they are either both 1 or both - 1 . In the first case, Ah has an
eigenvector corresponding to a curve on 1£' 2 that is fixed, so h is reducible.
In the second case, Ah has an eigenvector corresponding to a curve on 1£'2
that is fixed setwise, but reversed, so again h is reducible.
248 7. Further Topics

(3) I trace(Ah) I > 2.

Here, too, Ah has two eigenvalues, but there are infinitely many pos­
sibilities for trace(Ah) , so we cannot calculate them explicitly. Consider­
ation of the characteristic polynomial of Ah (together with the fact that
det(Ah) = 1) tells us only that they are distinct real numbers .X, .x-1 with,
say, I .X I > 1 > 1.x-1 1 . The eigenvector corresponding to .X determines a folia­
tion of 1l' as does the eigenvector corresponding to .x-1 (cf. Example 7.6. 15) .
These two foliations satisfy the properties required for h to be an Anosov
homeomorphism. D

Exercise

Exercise 1. Prove the following: All leaves of A are dense {:::::::;> A has no
proper sublaminations.

7. 7. The Curve Complex

The curve complex, introduced by Harvey (see [57] ) , initially served to study
topics related to surfaces and their automorphisms. In the late 1990s it
reemerged, this time as a tool in the study of 3-manifolds, spurred on by a
point of view proposed by J. Hempel; see [62] . We here provide a glimpse of
those aspects of the curve complex most directly related to the topological
study of 3-manifolds.
The curve complex exists on a surface. Through Heegaard diagrams
and Heegaard splittings it allows us to describe and interpret characteristic
features of 3-manifolds.
Definition 7.7. 1 . For any surface S, define the complexity of S, c(S) , by
c(S) = 3 genus(S) + # l 8SI 4.
-

Let S be a surface with c(S) > 0. The curve complex of S is a simplicial

complex with simplices as follows: The 0-simplices consist of isotopy classes
of essential simple closed curves in S. The 1-simplices consist of pairs of 0-
simplices that have disjoint representatives. More generally, the n-simplices
consist of n-tuples of 0-simplices that have pairwise disjoint representatives.
We denote the curve complex of S by C(S) .
The distance between two vertices of the curve complex is the minimal
number of edges in an edge-path between the two vertices.

Note that for the excluded surfaces, the above definition would yield a
highly disconnected complex. For this reason, one defines the curve complex
slightly differently in these cases. For instance, on the torus no two non­
isotopic essential curves are disjoint. There 1-simplices are defined to consist
7. 7. The Curve Complex 249

of pairs of curves that have representatives intersecting exactly once. The

resulting curve complex has the Farey graph as its I-skeleton.

Definition 7. 7.2. Let a, /3 be two simple closed curves in the surface S. The
geometric intersection number of a and /3, I(a, /3) , is the minimal number
of times a curve isotopic to a intersects a curve isotopic to /3.
Lemma 7.7.3. Let S be a surface with c(S) > 0. For vertices x, y of C(S)
and representatives ax, ay of x, y, respectively, with !(ax , ay ) > 0, we have

Proof. Our proof proceeds by induction on !(ax , ay ) · If !(ax , ay ) = 1 , then

ax and ay can be isotoped to intersect once. Consider a collar of ax U ay ,
C(ax U ay ) · Here 8C(ax U ay ) is a curve that is disjoint from both ax and
ay . This curve does not bound a disk in C(ax U ay ) and also does not bound
a disk in the complement of C(ax U ay ) (indeed, this would force S to be a
torus) . Thus d(x, y) = 2. Since 2 :::; 2 + 0, the inequality holds.
Now suppose that !(ax , ay ) 2: 2. (Here it does not help to consider the
collar of ax U ay , because its boundary consists of several curves, all of which
could be inessential. ) Consider a subarc a of ax that has its endpoints on
ay and interior disjoint from ay . We may assume that ax and ay have been
isotoped to intersect in a minimal number of points. It then follows that a
is not parallel to ay .
The endpoints of a cut ay into two subarcs. Let b be the subarc that
meets ax no more times than the other (i.e. , fewer, if possible) . Now a
push-off of a U b, which we denote by c (see Figure 7.23) and whose isotopy
class we denote by z, intersects ay at most once and ax at most �I (ax, ay )
times. Thus, by induction,

d(x, y) :::; d(x, z ) + d(z, y) :::; 2 + (2 + 2 log2 ( / ( c , ay )))

:::; 4 + 2 log 2 ( � !(ax , ay ) ) = 4 + 2(-1 + log 2 !(ax , ay ))

= 2 + 2 log2 !(ax, ay ),

as required. D

Theorem 7.7.4. For a surface S with x (S) :::; -2, the curve complex, C(S) ,
is connected.

Proof. This follows immediately from Lemma 7. 7.3. D

250 7. Further Topics

Figure 7.23. The curves a., , ay , and c.

We list two poignant results that provide structural information con­

cerning the curve complex. The second requires a few definitions:
Theorem 7.7.5 (Hempel) . For h : S --+ S a pseudo-Anosov automorphism,
n
--+oo d(x, h (x)) = oo .
nlim
Consequently, the diameter of C(S) is infinite.
The proof of this theorem relies on measured laminations and their rela·
tion to pseudo-Anosov automorphisms. The idea of the proof is to argue by
contradiction and show that boundedness of the set {d(x, hn (x)) I n E N}
contradicts key properties concerning the behavior of laminations with re­
spect to pseudo-Anosov automorphisms. We mention these key properties
as needed.
Proof. Suppose that {d(x, hn (x)) I n E N} is bounded. By passing to a
subsequence, if necessary, we may assume that
n
--+oo d(x, h (x)) = m
nlim
and
d(x, hn (x)) = m Vn.
This assumption guarantees that for each n, there is a sequence Y8, . . . , y::i
(corresponding to a row in the array below) such that Y8 = x, y::i = hn (x),
and d(yj , yj+l ) = 0. Thus for each j , we obtain a sequence {yj I n } ( corre­
sponding to columns in the array below) :
x Ym21 -1
x Ym -1
x Ym - 1
3

Note that a curve is also a lamination. Recall Theorem 7.6. 1 1 which tells
us that M.C(S) is homeomorphic to JR.6Y - 6 . Thus by adding appropriate
7. 7. The Curve Complex 251

weights, we can ensure that all the laminations under consideration lie in
the unit sphere of JR.69 - 6 :
(x, w ) (yi , w D ( Yin - I , win - I ) (h(x) , w I )
(x, w ) (y�, w n ( Y! - I , w! - I ) (h 2 (x) , w 2 )
(x, w ) (yf , w f) ( Y� - I ' w� _ I ) (h3 (x) , w 3 )

Since the unit sphere of JR.69 - 6 is compact, we may assume, by passing

to a subsequence, if necessary, that the measured laminations forming the
columns in this array converge to limits. Denote the limiting lamination of
the sequence { (yj , wj ) I n} by (Lj , Wj ) · The curves yj , YJ'+ I are disjoint;
i.e. , i ( (yj , wj ) , ( YJ'+ I , wj+ l ) ) = 0 and the function i is continuous. Hence
i ( (Lj , Wj ) , (Lj + l , Wj +I ) ) = 0.
The measured lamination obtained as a limit of iterated powers of h,

is the stable lamination of h. It has special properties. One key property is

that for any projective measured lamination µ, either µ = A or i (µ, A) =/: 0.
Since i((Lm - I , Wm - I ) , (A, w00 )) = 0, we have
(Lm - I , Wm - I ) = (A, w00 ) .
In particular, Lm - I has the same properties as A . This in turn implies that

(Lm - 2 , Wm - 2 ) = (Lm - I , Wm - I ) · · · (L I , W I ) = (x, w ) .

However, another key property is that it is not possible for the stable lami­
nation of a pseudo-Anosov automorphism to be a weighted curve. Thus we
obtain a contradiction. 0

H. Masur and Y. Minsky studied the curve complex extensively; see

[94] and [95] . They used a variety of methods, most notably those from
geometric group theory. One of their most striking results relies on the
following definition, due to M. Gromov:

Definition 7.7.6. A triangle in a metric space is o-thin if each of its sides

lies within a o-neighborhood of its other two sides. A metric space is called
geodesic if every two points are connected by a distance-minimizing geodesic.
A geodesic metric space is Gromov hyperbolic if for some o < oo , every
triangle is o-thin.

Theorem 7.7.7 (Masur-Minsky) . If S is a surface with c(S) > 0, then

C(S) is Gromov hyperbolic.
252 7. Further Topics

The proof of this theorem is lengthy and requires several techniques from
geometric group theory and Teichmi.iller theory that go beyond the scope of
this book. Bowditch has a more accessible proof. See [ 15 ] .
An element of the mapping class group of a surface S takes essential
curves to essential curves. Moreover, it takes disjoint essential curves to
disjoint essential curves. Thus it induces a map on C(S) . In this way the
mapping class group acts not only on surfaces but also on their associated
curve complexes. The converse is partially true:
Theorem 7.7.8 (Luo) . Let S be a surface not equal to a twice-punctured
torus and suppose that 3 genus(S) +# I BSl -4 2:: 1 . Then every automorphism
of C(S) is induced by an automorphism of S.
For S the twice-punctured torus, there is an automorphism of C(S) that
is not induced by an automorphism of S.
For a proof of Theorem 7.7.8 see [9 1 ] .
Exercises

Exercise 1 . Draw a picture of a genus 2 surface and a pair of curves of

distance 2.
Exercise 2. Prove that hyperbolic n-space is 8-thin for 8 = 1 .
Exercise 3. Prove that the dimension of C(S) is 3 genus(S) + # I BS I - 4.
1.8. Through the Looking Glass * *

I n this section we explore some of the ideas, results, and ramifications of

Hempel's work in "3-manifolds viewed from the curve complex" ; see [62]
and [ 15 ] . Here the curve complex can be thought of as a looking glass that
allows for a view of 3-manifolds complementary to the view from within the
3-manifold.
Definition 7.8 . 1 . Let S be a surface and X a simplex of C(S) . We denote
by Vx the result of attaching 2-handles to S x I along the curves c x 1
for c E X and attaching 3-handles whenever possible. Denote by Kx the
subcomplex of C(X) consisting of all simplices Y such that Vy = Vx .
The distance between subcomplexes of C(S) is the minimum distance
between vertices in the subcomplexes.
Given a Heegaard splitting M = V U s W, denote the simplex in C(S)
spanned by a given set of meridian disks of V by X and the simplex spanned
by a given set of meridian disks of W by Y. The distance of the Heegaard
splitting M = V Us W, d(M = V U s W) , is the distance, in C(S) , between
the subcomplexes Kx and Ky .
7.8. Through the Looking Glass 253

In effect, the distance of a Heegaard splitting is the smallest possible

distance between a meridian disk for one handlebody and a meridian disk
for the other handlebody.

Figure 7.24. A Heegaard diagram for the standard Heegaard splitting

of T3 .

Example 7.8.2. Earlier, we discussed the standard Heegaard splitting of 1!'3

and a diagram (see Figure 7.24) for this Heegaard splitting. Recall that this
Heegaard splitting is weakly reducible. Specifically, note that, for instance,
the disk in V corresponding to the disk labeled 2, that V was cut along,
and the disk in W, corresponding to the solid curve, are disjoint. The
former corresponds to a vertex in Kx , the latter to a vertex in Ky . Thus
d(M = V Us W) = 1 .
Consider what it means for a Heegaard splitting to have distance greater
than 2. Suppose that a is a curve that bounds a disk in one handlebody
and b is a curve that bounds a disk in the other handlebody. A distance
greater than 2 means that these curves meet and that the complement of
their union contains no essential curves. This makes the definition below
relevant.
Definition 7 .8.3. Two curves in a surface, isotoped to realize their inter­
section number, are said to fill the surface if their complement consists of
disks.

The following conjectures and theorems show how the curve complex
provides a lens through which to see and understand 3-manifolds.
Conjecture 7.8.4 (Schleimer) . A 3-manifold has only finitely many Hee­
gaard splittings of distance greater than 2 .
Theorem 7.8.5 (Hempel) . A Heegaard splitting of a Seifert fibered space
has distance at most 2 .
25 4 7. Further Topics

Proof. We prove this theorem in the special case of prism manifolds. As

discussed in Section 6.5, these manifolds have only vertical and horizontal
Heegaard splittings. Recall Definition 6.5.6. The cocore of a regular neigh­
borhood of the arc 'Y is a disk D. The annulus in the prism manifold that
projects to 'Y contains a disk E. These two disks intersect twice along their
boundaries. Hence our conclusion follows from the argument in the proof of
Lemma 7.7.3. (This argument holds in the more general setting of vertical
Heegaard splittings for arbitrary Seifert fibered spaces. )
Recall also the informal discussion o f horizontal Heegaard splittings. A
horizontal Heegaard surface comes from two copies of F in the surface bundle
F x I/</> away from some exceptional fiber e . Because F x I/</> is Seifert
fibered, </> must be periodic.
Consider an arc a in F. It gives rise to two disks (a x [O, 1/2] and
a x [1/2, 1]) that extend to disks in the two handlebodies of the Heegaard
splitting. The corresponding disks can be made disjoint in F x 1/2 (where
they are parallel) but will intersect along F x 0 = F x 1 . Because </> is
periodic, it does not matter that they intersect, because they do not fill F.
This gives us a curve on F that is disjoint from both disks. 0

Theorem 7.8.6 (Hartshorn) . A Heegaard splitting of a toroidal 3-manifold

has distance at most 2 .

The proof of Theorem 7.8.6 can b e found in [56] . It can b e generalized to

Heegaard splittings of 3-manifolds containing incompressible surfaces. The
following idea is key:

Lemma 7.8. 7. A connected surface 8 in a compression body W with 88 c

8+ W is either a disk or it is compressible or it is boundary compressible.

Proof. Denote the incompressible surface by 8 and the compression body

by W. Choose a set V of meridian disks of W. We may assume that
8 intersects V in as few components as possible. A standard innermost
disk argument then shows that all closed components of intersection define
compressing disks. Furthermore, a standard outermost arc argument shows
that all arcs of intersection define boundary compressions. Hence we need
only consider the case in which 8 is disjoint from V.
Here 8 lies in (closed orientable surface) x I union 3-balls. You will
prove in the exercises that if it lies in (closed orientable surface) x I, then it
is either compressible or boundary parallel. In particular, it is either a disk
or is compressible or is boundary compressible. If 8 lies in a 3-ball, then 88
is a circle on the boundary of a 3-ball. Hence here, too, 8 is either a disk or
is compressible. 0
7. 8. Through the Looking Glass 255

We can now outline a proof of Hartshorn's Theorem: Let M be a toroidal

3-manifold with Heegaard splitting M = V Us W. Denote an essential torus
in M by T. We may assume that the number of components of S n T is
minimal. Then any component SnT that is inessential in T must be essential
in S. An innermost such component provides a curve on S bounding a
meridian disk in either V or W.
If there is no such component bounding a meridian disk in, say, V,
then T n V is boundary compressible. The boundary compression raises
the Euler characteristic of T n V by 1 . Furthermore, all resulting curves
of intersection between T and S can be made disjoint from the original
curves of intersection and hence are distance 1 from all such curves. At this
point a series of classical, but technical, results guarantees that under our
minimality assumption, there must be a meridian disk for V in T n V after
the boundary compression.
Conjecture 7.8.8 (Hempel) . If the distance of a Heegaard splitting is suf­
ficiently large, then the 3-manifold represented by the Heegaard splitting is
hyperbolic.
In his dissertation, Hossein Namazi considered this conjecture together
with related questions. This led to a collaboration with Juan Souto. The
duo proved several striking results, some of which we mention below. Many
of these theorems can be interpreted as descriptions of "generic" behavior
for Heegaard splittings.
Theorem 7.8.9, Theorem 7.8. 1 1 , and Corollary 7.8.12 use the notion of
distance to interpolate between topology and geometry. They make exten­
sive use of Riemannian geometry. Corollary 7.8.12, in addition, requires
familiarity with algebraic topology.
Given a Heegaard splitting M = V Us W, we consider MCQ(S) of
(an abstract copy of) S. We then consider the subgroup of MCQ(S) that
extends to both V and W (i.e. , the intersection of the kernels of the maps
(iv) * : MCQ(S) Y MCQ(V) and (iw) * : MCQ(S) Y MCQ (W) induced by
the inclusion maps iv : S Y V and iw : S Y W). The following theorem
concerns this group.
Theorem 7.8.9 (Namazi) . If the distance of a Heegaard splitting M =
V Us W is sufficiently large, then the subgroup of the mapping class group
of S that extends to both V and W is finite.
For a proof of Theorem 7.8.9, see [1 1 1] . In the preceding section we
saw that V defines a subcomplex Kx of C( 8+ V) . Note that C( 8+ V) C
PM.C( 8+ V) . Consider the following setup: Vi , Vi are compression bodies
with 8+ Vi and 8+ V2 homeomorphic hyperbolic surfaces identified with the
abstract surface S.
256 7. Further Topics

Definition 7.8.10. The map h : 8+ Vi -+ 8+ Vi is a generic pseudo-Anosov

if h E MCQ(S) is pseudo-Anosov, the stable lamination of h is not contained
inKv1 , and the unstable lamination of h is not contained in Kv2 •

Theorem 7.8. 1 1 ( Namazi-Souto ) . Let Vi , Vi be compression bodies with

8+ Vi and 8+ Vi homeomorphic hyperbolic surfaces identified with an abstract
surface S and let h : 8+ Vi -+ 8+ Vi be a generic pseudo-Anosov. For every
e > 0, there exists n€ such that if n ;:::: n€, then Mn = Vi Uhn Vi admits a
Riemannian metric Pn , € with all sectional curvatures pinched between - 1 - e
and - 1 + e . Moreover, Pn ,€ has a lower bound for its injectivity radius
independent of n and e .
Corollary 7.8. 12. Under the assumptions above, 7r 1 ( Mn ) is infinite and
word hyperbolic.
Perelman's work then implies that M is hyperbolic.
Considering sequences of manifolds raises questions concerning limiting
behavior. There are several ways to define limits of such a sequence. One
such way grows out of the notion of Hausdorff distance and was formulated
by Gromov:
Definition 7.8.13. We say that the Hausdorff distance between the Rie­
mannian manifolds M, N is less than e if there is a diffeomorphism q : M -+
N such that
dM(x, y) € ::; dN(q(x) , q(y)) ::; dM ( X , y) + €.
-

A map such as q is called a (1, e ) -quasi-isometry.

The notion of Hausdorff distance ( and the related notion of Hausdorff

topology ) is well adapted to the study of compact spaces. For instance, it
enables a discussion of convergent sequences. Gromov's generalization takes
the notion a step further and thereby adapts it to the setting of non-compact
spaces.
Definition 7.8. 14. A sequence of pairs (Mn , Pn ), where Mn is a manifold
and Pn is a point in Mn , converges in the Gromov-Hausdorff topology if for
every R > 0, the balls
Bn = {x E Mn I dMn (X, Pn ) ::; R}
converge in the Hausdorff topology.

For example, the sequence of circles in �2 with increasing radii and based
at ( 0, 0 ) pictured in Figure 7 . 25 will converge to the x-axis in the Gromov­
Hausdorff topology. For details on the results of Namazi and Souto, see
[112] .
7. 8. Through the Looking Glass 257

Figure 7.25. A sequence of circles converging to R.

Theorem 7.8.15 (Namazi-Souto) . Consider the sequence sequence

(Mn , Pn , e ) , where Mn , Pn , e are as in Theorem 7.8. 1 1 , in the Gromov-Hausdorff
topology. Choose any set of base points for this sequence. The limit of any
convergent subsequence of this sequence is a hyperbolic 3-manifold homeo­
morphic to either Vi , Vi, or S x JR, where S = &+ \ti .
The following result concerns fundamental topological questions:
Theorem 7.8.16 (Namazi-Souto) . Let Vi , V2 , S, h be as above. If n is suf­
ficiently large, then S is the unique minimal genus Heegaard surface for
Vi uhn Vi .
There are several other complexes that are more or less related to the
curve complex. The most immediate is the arc complex:
Definition 7.8. 17. Let S be a compact surface with non-empty boundary.
The arc complex of S, denoted by A(S) , is defined analogously to the curve
complex: 0-simplices are isotopy classes of essential arcs and closed curves.
Edges correspond to distinct arcs and curves, etc.

The arc complex serves as a natural generalization of the curve com­

plex to the setting of surfaces with non-empty boundary. However, the
usual curve complex is also considered in the context of surfaces with non­
empty boundary with the caveat that vertices correspond to essential non­
peripheral (i.e. , not boundary parallel) simple closed curves. That is the
complex used by Masur and Minsky in their notion of subsurface projec­
tion.
Definition 7.8. 18. Suppose that X is a subsurface of S. The subsurface
projection map 7r : C(S) -+ C ( X ) assigns to vertices of C(S) simplices of
C ( X ) as follows: Let a be a simple closed curve in S that represents a vertex
of C(X) . Denote the collection of non-peripheral curves in the boundary of a
258 7. Further Topics

regular neighborhood of a U 8X by ax . The components of ax are disjoint,

so ax corresponds to a simplex cr ( ax ) in C(X) . We set 7r ( a ) = cr ( ax ) .

Figure 7.26. Before the projection.

Figure 7.27. After the projection.

It is interesting to observe how distances behave under this projection

map. In general, distance can go up under projection, but Masur and Minsky
provide a characterization of this phenomenon. S. Schleimer cleverly exploits
this characterization in his proof of the theorem below concerning the "end
of the curve complex" ; see [140] .
Definition 7.8. 19. Let X be a metric space and let B = B(x, R) c X be
a closed metric R-ball. We define the number c(B) to be the number of
unbounded components of X \ B. Fix a point x E X. The number of ends
of a metric space X is
E(X) = sup c(B(x, R) ) .
R;::: O

Note that E(X) is independent of x.

Theorem 7.8.20 (Schleimer) . Let S be a surface of genus at least 2 with
exactly one boundary component. Then C ( S) has only one end.
Recall the notion of pants decomposition from Section 2.3. Given a com­
pact orientable surface, A. Hatcher constructed a 2-dimensional simplicial
7.8. Through the Looking Glass 259

complex P ( S ) as follows:
• The vertices of P ( S ) are the equivalence classes of pants decompo­
sitions of S.
• The edges of P ( S) correspond to two types of alterations of pants
decompositions of S in which one curve is replaced by another. See
Figures 7.28 and 7.29.
• The 2-simplices of P ( S ) correspond to a finite set of possible types
of alterations ( that we will not discuss explicitly here ) involving
three or five curves or certain combinations of moves as in Figures
7.28 and 7.29.

I
I

® '.
I

I
I
I
I

Figure 7.28. The S-move.

Figure 7.29. The A-move.

Theorem 7.8.21 ( Hatcher ) . The pants complex is simply connected.

See http: // www.math.cornell.edu / hatcher / Papers / pantsdecomp.pdf.

Remark 7.8.22. The pants complex is related to the dual of the curve
complex.
260 7. Further Topics

Exercises
Exercise 1. Show that a Heegaard splitting is reducible if and only if its
distance is 0. Furthermore, show that it is weakly reducible if and only if
its distance is at most 1 .
Exercise 2. Suppose that M is a toroidal 3-manifold. Show that the
distance of any Heegaard splitting M = V U s W is at most 2.
Exercise 3. Show that if S is an incompressible surface in ( closed orientable
surface ) x I with 88 c ( closed orientable surface ) x { 1 } , then S is boundary
parallel.
Exercise 4. Exhibit an example of a 3-manifold with infinite mapping class
group.
Exercise 5 . To contrast with Theorem 7.8. 16, exhibit 3-manifolds with
distinct minimal genus Heegaard splittings and explain which hypotheses
are not satisfied.
 Appendix A

General Position

The numerous technicalities involved in describing and discussing general

position in the TOP category often stand in the way of clarity. We opt to
discuss the analogous concept from the DIFF category. This is the notion of
transversality. In our discussion we will need several basics from the DIFF
category. We will barely touch on the key issues. For more details, see [100] ,
[48] , or [128] .
In this appendix we consider only manifolds embedded in Euclidean
space. The following theorem tells us that this in no way limits our scope:
Theorem A.0. 1 ( Whitney ) . Every compact n-dimensional manifold admits
an embedding into �2n .
Calculus teaches us how to find the tangent space to a manifold X
embedded in Euclidean space at a point x in X: It is the best linear approx­
imation to the manifold at the given point and is denoted by Tx (X). See
Figures A. l and A.2.
Fact 1 . The tangent space to an n-manifold at a given point is isomorphic
to �n .
Fact 2. The collection of all tangent spaces at points in the n-manifold X
forms an �n -bundle over X.
Definition A.0.2. The bundle described above is called the tangent bundle
of X and is denoted by T(X) .
Given a differentiable map f : X --+ Y and a point x E X, its derivative
at dfx , maps Tx (X) to Tt (x ) (Y) . Recall that a point x E X is critical if
x,
dfx fails to be surjective. A value y E Y is critical if its preimage contains
-
261
262 A. General Position

(1,1)

Figure A . 1 . The tangent line to a curve at the point ( 1 , 1 ) .

Figure A.2. A portion of the tangent plane to a torus at a specified point.

critical points. The following theorem lies at the heart of transversality:

Theorem A.0.3 (Sard) . The set of critical values of a differentiable map
from one manifold to another has Lebesgue measure 0.
We provide some illustrative examples:
Example A.0.4. The inclusion map l : JR -+ IR.2 from lR onto the y-axis has
as its critical values the y-axis, a set of Lebesgue measure 0 in IR.2 •
Example A.0.5. The map f : lR -+ lR given by f (x) = x 2 has exactly one
critical point, the point 0. This point is mapped to the value O; hence 0 is
the only critical value. This is a set of measure 0 in R
Example A.0.6. The projection map 71'1 : IR.2 -+ lR from IR.2 onto the x-axis
has no critical values.
A. General Position 263

Example A.0.7. The map f : IR.2 ---+ IR.2 given by f (x, y) = (x 2 y, y) has as
its critical points the two coordinate axes. These are mapped to the values
(0, y) , that is, the y-axis. This is a set of measure 0 in IR.2 .

We now define the notion of transversality, first for submanifolds of a

single manifold and then in more generality. Transversality provides the
basis for many arguments in the theory of 3-manifolds. It guarantees nice
intersections of submanifolds. For instance, every time we assume that a
curve and a surface in a 3-manifold intersect in points (rather than, say, line
segments) we are invoking general position/ transversality. Note that for Y
a submanifold of X and p a point in Y c X, Tp(Y) is a subspace of Tp(X) .
Definition A.0.8 (Transversality for submanifolds) . Two submanifolds
Y, Z of the manifold X are transverse at the point p E Y n Z if Tp(Y)
and Tp(Z) together span Tp(X) . Two submanifolds of X are transverse if
they are transverse at each point of intersection. We write Y rh Z.

Figure A.3. A transverse intersection of two I-manifolds in IR2 •

Example A.0.9. In IR.2 , the x-axis and the y-axis are transverse.
Non-Example. In IR.3 , the x-axis and the y-axis are not transverse.
Example A.0.10. In IR.3 , the x-axis and the line parallel to the y-axis
passing through ( 1 , 0, 1) are transverse.
Example A.0. 1 1 . In IR.3 , the xy-plane and the z-axis are transverse.

Non-Example. In IR.3 , the xy-plane and the y-axis are not transverse.
Non-Example. In IR.4 , the xy-plane and the z-axis are not transverse.
Example A.0.12. In IR.3 , the xy-plane and the xz-plane are transverse.
264 A. General Position

'
I

J
Figure A.4. A transverse intersection of a I-manifold and a 2-manifold
in JR 3 .

Non-Example. In JR.2 , the graph of the function f JR. � JR. given by

f ( x) = x 3 and the x-axis are not transverse.
More generally, transversality can be formulated for maps. This for­
mulation may appear unnatural at first glance; but it allows us to build a
coherent theory.
Definition A.0.13 (Thansversality in general) . Let f : Z � X be a differ­
entiable map and let Y be a submanifold of X. Let y E Y and let z E Z be
such that f(z) = y. Then f is transverse to Y at z if the image of dfz and
Ty (Y) together span Ty (X) . The map f is transverse to Y if for every point
z E Z such that f(z) E Y, f is transverse to Y at z.
This condition is sometimes expressed as

Image( dfz ) + Ty (Y) = Ty (X) ,

but note that this equality must be interpreted in the appropriate sense.
Remark A.0. 14. Thansversality for submanifolds is a special case of trans­
versality in general. Indeed, two submanifolds of X are transverse if the
inclusion map of one submanifold is transverse to the other submanifold.
Remark A.0. 15. Let X be a manifold of dimension nx with submanifolds
Y, Z of dimensions ny , nz . If Y m Z, then it is a consequence of the definitions
that the dimension of Y n Z is ny + nz - nx . More generally, for f : Z � X
a differentiable map transverse to Y c X, the codimension of 1- 1 (Y) in Z
is the same as the codimension of Y in X.
Example A.0.16. Two !-dimensional submanifolds of a 3-manifold must
be disjoint. (The proof is left as an exercise. )
Example A.0. 17. A !-dimensional submanifold and a 2-dimensional sub­
manifold of a 3-manifold intersect in points.
A. General Position 265

Example A.0.18. Two 2-dimensional submanifolds of a 3-manifold inter­

sect in a 1-dimensional submanifold.

Theorem A.0. 19 below tells us that transversality is attainable via small

isotopies called perturbations.
Theorem A.0.19 ( The TI:ansversality Homotopy Theorem) . For any smooth
map f : Z -+ X and any submanifold Y of X there exists a smooth map
g : Z -+ X homo topic to f that is transverse to Y. Moreover, suppose that
H : Z x I -+ X is a homotopy between f and g. Then Ve > 0, the map
9e : Z -+ X defined by 9e (z) = H(z, c ) is also transverse to Y.

Figure A.5. A non-transverse intersection for the x-axis and a curve.

Figure A.6. A perturbation resulting in transverse intersection.

Recall the non-examples to transversality mentioned above: ( a) In R.3 ,

the x-axis and the y-axis are not transverse, but the x-axis and a small
perturbation of the y-axis ( for instance, a translation onto a parallel axis )
are transverse; ( b ) in R.3 , the xy-plane and the y-axis are not transverse,
but the xy-plane and a small perturbation of the y-axis ( for instance, a
266 A. General Position

Figure A. 7. Another perturbation resulting in transverse intersection.

translation onto a parallel axis not lying in the xy-plane) are transverse;
(c) in R4 , the xy-plane and the z-axis are not transverse, but the xy-plane
and a small perturbation of the z-axis (for instance, a translation onto an
axis not lying in the xy-plane) are transverse; (d) in R2 , the graph of the
function f : R -t R given by f (x) = x 3 and the x-axis are not transverse,
but a translation of the graph (one that raises the point of inflection of the
graph above or below the x-axis) and the x-axis are transverse.
Remark A.0.20. In fact, in the Transversality Homotopy Theorem, Z is
allowed to have boundary. In this case the conclusion can be strengthened:
Denote the restriction of the function h : Z -t X to 8Z by ah. Then we can
conclude that, aside from being homotopic to f and {) f, respectively, both
g and 8g are transverse to Y.

Manifolds are metrizable spaces. Thus if Y is a submanifold of X, we

can look at all points within a distance c of Y. The following theorem gives
us an explicit description of this set.
Theorem A.0.21 (The €-Neighborhood Theorem) . Let Y be a compact k­
dimensional submanifold of the n-manifold X . Let y e denote the set of all
points in X with distance less than c from Y. If c is sufficiently small, then
ye is a Bn- k _bundle over Y .

For a proof of this theorem, see [48] . This theorem reminds us that
£-neighborhoods exist and it explicitly describes them as bundles.
Remark A.0.22. The term general position is the TOP category and TRI­
ANG category equivalent of the DIFF category term transversality. The
term regular neighborhood is the TOP category equivalent of the DIFF cat­
egory term €-neighborhood.
Exercises 267

()
Figure A.8. An €-neighborhood of a I -manifold in a 3-manifold.

Figure A.9. An €-neighborhood of a surface in a 3-manifold.

Theorem A.0.23 (Uniqueness of Regular Neighborhoods). Let Ni , N2 be

regular neighborhoods of Y in X . There exists an isotopy of X fixed on Y
carrying Ni onto N2 .
For a proof of this theorem, see [128] .
Exercises

Exercise 1. Show that transverse I-dimensional submanifolds of a 3-

manifold must be disjoint.
Exercise 2. Draw and imagine several examples of transverse and non­
transverse intersections.
 Appendix B

Morse Functions

Morse functions provide for descriptions of manifolds in terms of certain

basic building blocks. Moreover, they exist on all smooth manifolds. We
will describe the 1-, 2-, and 3-dimensional scenarios. For more details, see
(99] and (48] .
Definition B.0. 1 . Let f : Rn -+ R be a smooth function. A critical point
x of f is non-degenerate if the Hessian matrix of second partial derivatives
is non-singular at x.
Let M be a smooth n-manifold and let f : M -+ R be a smooth function.
A point x E M is a critical point of f if there is a chart <Pa near x such that
<Pa(x) = 0 and 0 is a critical point of f o ¢� 1 . It is non-degenerate if 0 is a
non-degenerate critical point of f o ¢� 1 . A critical value of f is a value c
such that f -1 (c) contains at least one critical point.
In fact, the above definition is independent of charts. If a point in M
is a critical point for one chart, then it is a critical point for all charts.
Likewise, if a point in M is a non-degenerate critical point for one chart,
then it is a non-degenerate critical point for all charts. As it turns out, non­
degenerate critical points can be described rather informatively in terms of
local coordinates; namely, if x is a non-degenerate critical point of f : M -+
R, then there are local coordinates in which
k n
f(x 1 , . . . , xn ) = - I: r + L xT .
x
i= l i =k+ I

This result is known as the Morse Lemma. The number k is called the index
of the critical point.
-

269
2 70 B. Morse Functions

Definition B.0.2. Let M be a manifold. A smooth proper map f : M --+ IR.

is a Morse function if all of its critical points are non-degenerate and distinct
critical points correspond to distinct critical values.
Theorem B.0.3. If M is a smooth manifold, then there exist Morse func­
tions on M.
In fact, for any smooth manifold M, Morse functions are dense in
C00 ( M ) . ( Here C00 ( M ) is the set of smooth real-valued functions on M
topologized via the notion of C00-convergence on compacts. ) Theorem B.0.3
is a consequence of several properties of smooth manifolds including Sard's
Theorem. For details, see [48] .
A Morse function f : M --+ IR. decomposes the manifold M into regular
and singular level sets. If there are no critical values between two regular
values, then their level sets are diffeomorphic. Furthermore, the regular
values that lie strictly between two consecutive critical values provide a
product structure on the corresponding level sets. More precisely,
f - 1 ( ( a , b))
is diffeomorphic to
f - 1 (c) x (a, b)
whenever ( a , b) is an interval containing no critical values and c E (a, b) .
Hence to understand the manifold, we need to understand how level sets
change as we pass through a critical value. Specifically, we wish to observe
how the submanifold f - 1 ( ( - oo , y] ) changes as we pass through a critical
value YO · The expression of f in terms of local coordinates provides a com­
plete picture of this change. We will consider the scenario in dimensions 1 ,
2, and 3.
Consider a Morse function on the closed I-dimensional manifold § 1 . The
regular level sets are discrete; hence the regular values that lie strictly be­
tween two consecutive critical values correspond to arcs in the manifold. A
critical point of index 0 is modeled on x 2 . Thus a new point appears and
and breaks into two as we move from level sets below to level sets above
the critical point. A critical point of index 1 is modeled on -x 2 • Thus two
points of f - 1 (y) merge into one as we move from level sets below to level
sets above the critical point.
For a Morse function on a closed 2-dimensional manifold, the regular
level sets are disjoint unions of circles. Hence preimages of intervals con­
taining no critical values are cylinders in M. A critical point of index 0 is
modeled on x� + x� . Thus a point appears and expands into a circle as we
move from level sets below to those above the critical point. A critical point
of index 1 is modeled on -x� + x� and is also known as a saddle singularity.
B. Morse Functions 271

Figure B . 1 . A Morse function on the circle with critical levels indicated.

One of two things happens as we move from level sets below to those above
the critical value: Either one circle is pinched and splits into two circles or
two circles are wedged at a point and merge into one circle. A critical point
of index 2 is modeled on -xi - x�. Thus a circle is collapsed into a point
and disappears.

Figure B.2. A Morse function on the torus with critical levels indicated.

For a Morse function on a closed 3-dimensional manifold, the regular

level sets are surfaces. Hence the regular values that lie strictly between two
consecutive critical values correspond to products of the form (surface) x I. A
critical point of index 0 is modeled on xi + x� + x5. Thus a point appears and
expands into a 2-sphere as we move from level sets below to level sets above
the critical value. A critical point of index 1 is modeled on -xi + x� + x5.
Thus the two branches of a hyperboloid of two sheets meet at two points
and merge into a hyperboloid of one sheet as we move from level sets below
to level sets above the critical value. A critical point of index 2 is modeled
on -xi - x� + x5. Thus a hyperboloid of one sheet collapses by pinching
along a circle and breaks into a hyperboloid of two sheets as we move from
level sets below to level sets above the critical value. See Figure B.3.
272 B. Morse Functions

Figure B.3. A hyperboloid of two sheets collapsing to a hyperboloid

of one sheet or vice versa (x 1 is the vertical coordinate).

A critical point of index 3 is modeled on - x i - x � - x � . Thus a 2-sphere

collapses to a point and disappears as we move from level sets below to level
sets above the critical value.

I -p_-�7
Figure B.4. Level surface below an index 1 critical value.

Figure B . 5 . Level surface at an index 1 critical value.

More generally, let M be a closed n-manifold, f : M ---+ � a Morse func­

tion, and xo a critical point of f. The critical point xo has a neighborhood
homeomorphic to [- 1 , l] n . By the Morse Lemma there is a k E N such that
there are local coordinates about xo in which
k n
f ( x 1 , . . . , xn ) = - :L r + :L
x
i =l
B. Morse Functions 273

Figure B.6. Level surface above an index 1 critical value.

Let Yo be the critical value corresponding to xo . Observe that the homeo­

morphism type of 1 -1 ((-00, y]) changes as we pass through YO · Specifically,
for a, b such that a < Yo < b and such that Yo is the only critical value
between a and b, 1 - 1 ((-00, b]) is homeomorphic to the result of attaching a
copy of [-1, l] n to 8f - 1 ((-oo, a] ) along 8[-1, l] k x [-1, 1 1 n- k . (In fact, the
word "homeomorphism" can be replaced with the word "diffeomorphism"
in the discussion above, though this is harder to see.)
Definition B.0.4. For Ma an n-manifold with boundary, a k-handle is
[-1, l] n attached to Ma along 8[-1, l] k x [-1, l] n - k .
It follows that every closed manifold can be built from handles starting
with a 0-handle. For instance, an n-sphere can be built from one 0-handle
and one n-handle. (Simply attach the n-handle to the boundary of the
0-handle.) More generally, to build a given manifold M, choose a Morse
function h : M --+ JR and then mimic the growth of h - 1 ((-oo, y]) by attach­
ing a k-handle when y passes through a critical value of index k.

lfYJ
0 0 0 0

Figure B.7. Attaching a handle of index 1 to a 3-manifold.

Remark B.0.5. Given a Morse function h : M --+ JR on the n-manifold

M, the function -h : M --+ JR is also a Morse function. Critical points
of index k for h correspond to critical points of index n - k for -h. This
means that in the context of n-manifolds, a k-handle is dual to an (n - k)­
handle in the following sense: If we can build an n-manifold M by attaching
handles h 1 , h2 , . . . , ht, then we can also build M by attaching these handles
"upside down" and in reverse order. I.e. , for each k-handle hi , rather than
attaching the k-handle hi to the submanifold h -1 ((-oo, Yo - £]) of M, we
instead attach an (n - k)-handle to h -1 ([y0 + £, oo)) .
274 B. Morse .Functions

Note that a 0-handle is attached to nothing and hence is a copy of

[- 1 , l] n appearing out of thin air. Dually, an n-handle is attached along
8[-1, l] n and hence is a topological n-ball that caps off a spherical boundary
component.
Definition B.0.6. A description of a manifold M in terms of handles is
called a handle decomposition.
Conversely, under the right smoothness assumptions, a handle decom­
position corresponds to a Morse function.
Definition B.O. 7. A Morse function f : M -+ � is self-indexing if, for all k,
critical points of index k correspond to smaller critical values than critical
points of index k + 1.
Let M be a closed 3-manifold and let h : M -+ � be a self-indexing
Morse function. Consider a regular value r such that all critical points of h
of index 0 and 1 occur below r and all critical points of index 2 and 3 occur
above r . Then 1 - 1 ((-00, r]) is constructed from 0-handles and 1-handles
and is hence a handlebody. Dually, 1 - 1 {[r , oo)) is also a handlebody. Hence:
Remark B.0.8. A self-indexing Morse function defines a Heegaard split­
ting.
Conversely, a Heegaard splitting can be used to define a Morse function,
though the Morse function is not unique.
Exercises

Exercise 1. Suppose that a closed n-manifold M admits a Morse function

with exactly two critical points. Prove that M is homeomorphic to §n . (In
general, M need not be diffeomorphic to §n . This fact was used by Milnor in
[9 8] to construct examples of 7-dimensional topological spheres with exotic
smooth structures.)
Exercise 2. Suppose that a closed 3-manifold M admits a Morse function
with exactly 4 critical points. Prove that M is a lens space or §3 .
Exercise 3. Explore how Morse functions interact with other structures on
manifolds. For example, read [ 5 8 ] .
Bibliography

1. Colin C. Adams, The knot book. An elementary introduction to the mathematical

theory of knots, W. H. Freeman and Company, New York, 1994. Reprinted with
corrections, Amer. Math. Soc., Providence, RI, 2004.
2. Ian Agol, The Virtual Haken Conjecture, arXiv:1204.2810 (math. GT] .
3. Ian Agol and Yi Liu, Presentation length and Simon's conjecture, J. Amer. Math.
Soc. 25 (2012), no. 1, 151-187. MR2833481
4. Lars Ahlfors and Leo Sario, Riemann surfaces, Princeton Mathematical Series,
vol. 26, Princeton University Press, Princeton, NJ, 1960.
5. Selman Akbulut, Scharlemann's manifold is standard, Ann. of Math. (2) 149 (1999),
no. 2, 497-510. MR1689337 (2000d:57033)
6. James Waddell Alexander, Note on Riemann spaces, Bull. Amer. Math. Soc. 26
(1920), no. 8, 370-372.
7. ___ , A lemma on systems of knotted curves, Proc. Nat. Acad. Sci. U.S.A. 9
(1923), no. 3, 93-95.
8. Riccardo Benedetti and Carlo Petronio, Lectures on hyperbolic geometry, Universi­
text, Springer-Verlag, Berlin, 1992. MR1219310 (94e:57015)
9. Laurent Bessieres, Gerard Besson, Sylvain Maillot, Michel Boileau, and Joan Porti,
Geometrisation of 3-manifolds, EMS Tracts in Mathematics, vol. 13, European Math­
ematical Society (EMS), Zurich, 2010. MR2683385 (2012d:57027)
10. Joan S. Birman, Braids, links, and mapping class groups, Princeton University
Press, Princeton, NJ, 1974, Annals of Mathematics Studies, No. 82. MR0375281
(51 #11477)
11. Steven Bleiler and Martin Scharlemann, A projective plane in R4 with three critical
points is standard. Strongly invertible knots have property P, Topology 27 (1988),
no. 4, 519-540.
12. Michel Boileau, Donald Collins, and Heiner Zieschang, Scindements de Heegaard des
petites varietes de Seifert, C. R. Acad. Sci. Paris Ser. I Math. 305 (1989), no. 12,
557-560.
13. Michel Boileau and Jean-Pierre Otal, Sur les scindements de Heegaard du tore T3 ,
J. Differential Geom. 32 (1990), no. 1, 209-233.
-

275
276 Bibliography

14. Francis Bonahon and Jean-Pierre Otal, Scindements de Heegaard des espaces lentic­
ulaires, C. R. Acad. Sci. Paris Ser. I Math. 294 (1982), no. 17, 585-587.
15. Brian H. Bowditch, Intersection numbers and the hyperbolicity of the curve complex,
J. Reine Angew. Math. 598 (2006), 105-129.
16. Mark Brittenham and Ying-Qing Wu, The classification of exceptional Dehn surg­
eries on 2-bridge knots, Comm. Anal. Geom. 9 (2001), no. 1, 97-113. MR1807953
(2001m:57008)
17. Jeffrey F . Brock, Kenneth W. Bromberg, Richard D. Canary, and Yair N. Minsky,
Local topology in deformation spaces of hyperbolic 3-manifolds, Geom. Topol. 1 5
(2011), no. 2 , 1169-1224. MR2831259
18. M. Brown, A proof of the generalized Schoenfties Theorem, Bull. Amer. Math. Soc.
66 (1960), 74-76.
19. Gerhard Burde, Heiner Zieschang, and Michael Heusener, Knots, third ed.,
de Gruyter Studies in Mathematics, vol. 5, Walter de Gruyter & Co., Berlin, 2013.
20. Danny Calegari and David Gabai, Shrinkwrapping and the taming of hyperbolic 3-
manifolds, J. Amer. Math. Soc. 19 (2006), no. 2, 385-446.
21. Alberto Candel and Lawrence Conlon, Foliations. I, Graduate Studies in Mathe­
matics, vol. 23, American Mathematical Society, Providence, RI, 2000. MRI 732868
(2002f:57058)
22. ___ , Foliations. II, Graduate Studies in Mathematics, vol. 60, American Mathe­
matical Society, Providence, RI, 2003. MR1994394 (2004e:57034)
23. Huai-Dong Cao and Xi-Ping Zhu, A complete proof of the Poincare and Geometriza­
tion Conjectures-application of the Hamilton-Perelman Theory of the Ricci flow,
Asian Journal of Mathematics 10 (2006), no. 2, 165-492.
24. A. J. Casson and C. McA. Gordon, A loop theorem for duality spaces and fibred
ribbon knots, Invent. Math. 74 (1983), no. 1, 119-137.
25. Andrew J. Casson and Steven A. Bleiler, Automorphisms of surfaces after Nielsen
and Thurston, London Mathematical Society Student Texts, vol. 9, Cambridge Uni­
versity Press, Cambridge, 1988. MR964685 (89k:57025)
26. Jean Cerf, La stratification naturelle des espaces de fonction differentiables reelles et
le theoreme de la pseudo-isotopie, Inst. Hautes Etudes Sci. Puhl. Math. 39 (1970),
5-173.
27. Eric Charpentier, Etienne Ghys, and Annick Lesne (eds.), The scientific legacy of
Poincare, History of Mathematics, vol. 36, American Mathematical Society, Prov­
idence, RI, 2010, Translated from the 2006 French original by Joshua Bowman.
MR2605614 (2011b:00005)
28. J. H. Conway, An enumeration of knots and links, and some of their algebraic prop­
erties, Computational Problems in Abstract Algebra (Proc. Conf., Oxford, 1967),
Pergamon, Oxford, 1970, pp. 329-358. MR0258014 (41 #2661)
29. Daryl Cooper and Martin Scharlemann, The structure of a solvmanifold 's Heegaard
splittings, Proceedings of 6th Gokova Geometry-Topology Conference, vol. 23, 1999,
pp. 1-18. MR1701636 (2000h:57034)
30. Marc Culler, C. McA. Gordon, J. Luecke, and Peter B. Shalen, Dehn surgery on
knots, Ann. of Math. (2) 125 (1987), no. 2, 237-300.
31. Oliver T. Dasbach and Tao Li, Property P for knots admitting certain Gabai disks,
Topology Appl. 142 (2004), no. 1-3, 113-129.
32. M. Dehn, Uber die Topologie des dreidimensionalen Raumes, Math. Ann. 69 (1910),
137-168.
Bibliography 277

33. ___ , Die Gruppe der A bbildungsklassen. (German) Das arithmetische Feld auf
Flachen, Acta Math. 69 (1938), no. 1, 135-206.
34. Charles Delman and Rachel Roberts, Alternating knots satisfy strong Property P,
Comment. Math. Helv. 74 (1999), no. 3, 376-397.
35. Y. Diao, J. C. Nardo, and Y. Sun, Global knotting in equilateral random polygons,
JKTR 10 (2001), no. 4, 597-697.
36. Yuanan Diao, Nicholas Pippenger, and De Witt Sumners, On random knots. Random
knotting and linking, JKTR 3 (1994), no. 3, 419-429.
37. Manfredo Perdigao do Carmo, Riemannian geometry, Mathematics: Theory & Ap­
plications, Birkhiiuser Boston Inc., Boston, MA, 1992, Translated from the second
Portuguese edition by Francis Flaherty. MR1138207 (92i:53001)
38. Cornelia Drutu and Michael Kapovich, Lectures on geometric group theory, preprint.
39. A. Fathi, F . Laudenbach, and V. Poenaru, Travaux de Thurston sur les surfaces,
Asterisque 66-67 (1979), 1-284.
40. M. Feighn, Branched covers according to J. W. A lexander, Collect. Math. 37 (1986),
no. 1, 55-60.
41. David Gabai, Foliations and the topology of 3-manifolds, J. Differential Geom. 18
(1983), no. 3, 445-503. MR723813 (86a:57009)
42. __ , Genera of the arborescent links, Mem. Amer. Math. Soc., vol. 59, American
Mathematical Society, Providence, RI, 1986.
43. ___ , Foliations and the topology of 3-manifolds. III, J. Differential Geom. 26
(1987), no. 3, 479-536. MR910018 (89a:57014b)
44. ___ , 3 lectures on foliations and laminations on 3-manifolds, Laminations and
foliations in dynamics, geometry and topology (Stony Brook, NY, 1998), Contemp.
Math., vol. 269, Amer. Math. Soc., Providence, RI, 2001, pp. 87-109. MR1810537
(2002g:57032)
45. David Gabai and Ulrich Oertel, Essential laminations in 3-manifolds, Ann. of Math.
(2) 130 (1989), no. 1, 41-73. MR1005607 (90h:57012)
46. C. McA. Gordon and J. Luecke, Knots are determined by their complements, Bull.
Amer. Math. Soc. (N.S.) 20 (1989), no. 1.
47. V.K.A.M. Gugenheim, Piecewise linear isotopy and embedding of elements and
spheres. I, II, Proc. LMS 3 (1953), 29-53.
48. V. Guillemin and A Pollack, Differential topology, Prentice Hall, Inc., Englewood
Cliffs, NJ, 1974.
49. Andre Haefiiger, Varietes feuilletees, Ann. Scuola Norm. Sup. Pisa (3) 16 (1962),
367-397. MR0189060 (32 #6487)
50. __ , Feuilletages riemanniens, Asterisque (1989), no. 177-178, Exp. No. 707,
183-197, Seminaire Bourbaki, Vol. 1988/89. MR1040573 (9le:57047)
51. ___ , Travaux de Novikov sur les feuilletages, Seminaire Bourbaki, Vol. 10, Soc.
Math. France, Paris, 1995, Exp. No. 339, 433-444. MR1610457
52. Wolfgang Haken, Ein Verfahren zur Aufspaltung einer 3-Mannigfaltigkeit in irreduz­
ible 3-Mannigfaltigkeiten, Math. Z. 76 (1961), 427-467. MR0141108 (25 #4519c)
53. __ , Theorie der Normalftachen, Acta Math. 105 (1961), 245-375. MR0141106
(25 #4519a)
54. Richard S. Hamilton, A compactness property for solutions of the Ricci flow, Amer.
J. Math. 117 (1995), no. 3, 545-572. MR1333936 (96c:53056)
278 Bibliography

55. ___ , The formation of singularities in the Ricci flow, Surveys in differential ge­
ometry, Vol. II (Cambridge, MA, 1993), Int. Press, Cambridge, MA, 1995, pp. 7-136.
MR1375255 (97e:53075)
56. Kevin Hartshorn, Heegaard splittings of Haken manifolds have bounded distance, Pa­
cific J. Math. 204 (2002), no. 1, 61-75. MR1905192 (2003a:57037)
57. William James Harvey, Geometric structure of surface mapping class groups, Homo­
logical group theory (Proc. Sympos., Durham, 1977), London Math. Soc. Lecture
Note Ser., vol. 36, Cambridge Univ. Press, Cambridge-New York, 1979, pp. 255-269.
58. Joel Hass, Paul Norbury, and J. Hyam Rubinstein, Minimal spheres of arbitrarily
high Morse index, Comm. Anal. Geom. 11 (2003), no. 3, 425-439. MR2015753
(2004i:53082)
59. Allen Hatcher, Notes on basic 3-manifold topology, preprint.
60. P. Heegaard, Sur l ' "Analysis situs ", Bull. Soc. Math. France 44 (1916), 161-242.
MR1504754
61 . Geoffrey Hemion, The classification of knots and 3-dimensional spaces, Oxford Sci­
ence Publications, The Clarendon Press Oxford University Press, New York, 1992.
MR1211184 (94g:57015)
62. John Hempel, 3-manifolds as viewed from the curve complex, Topology 40 (2001),
no. 3, 631-657.
63. ___ , 3-manifolds, AMS Chelsea Publishing, Providence, RI, 2004, reprint of the
1976 original.
64. H. Hilden, M. T. Lozano, and J . M. Montesinos, Universal knots, Bull. Amer. Math.
Soc. (N.S . ) 8 (1983), no. 3, 449-450.
65. Dale Husemoller, Fibre bundles, third ed., Graduate Texts in Mathematics, vol. 20,
Springer-Verlag, New York, 1994. MR1249482 (94k:55001)
66. Michael Hutchings, Floer homology of families. I, Algebr. Geom. Topol. 8 (2008),
no. 1, 435-492. MR2443235 (2009h:57046)
67. William Jaco, Lectures on three-manifold topology, CBMS Regional Conference Series
in Mathematics, vol. 43, American Mathematical Society, Providence, RI, 1980.
68. William Jaco and Ulrich Oertel, An algorithm to decide if a 3-manifold is a Haken
manifold, Topology 23 (1984), no. 2, 195-209.
69. William Jaco and J. Hyam Rubinstein, PL minimal surfaces in 3-manifolds, J. Dif­
ferential Geom. 27 (1988), no. 3, 493-524.
70. ___ , PL equivariant surgery and invariant decompositions of 3-manifolds, Adv.
in Math. 73 (1989), no. 2, 149-191.
71. William Jaco and Peter B. Shalen, Seifert fibered spaces in irreducible, sufficiently­
large 3-manifolds, Bull. Amer. Math. Soc. 82 (1976), no. 5, 765-767. MR0415623
(54 #3706)
72. Klaus Johannson, On exotic homotopy equivalences of 3-manifolds, Geometric topol­
ogy (Proc. Georgia Topology Conf., Athens, GA, 1977), Academic Press, New York,
1979, pp. 101-111. MR537729 (80m:57002)
73. ___ , Topology and combinatorics of 3-manifolds, Lecture Notes in Mathematics,
vol. 1599, Springer-Verlag, Berlin, 1995. MR1439249 (98c:57014)
74. Michael Kapovich, Hyperbolic manifolds and discrete groups, Modern Birkhauser
Classics, Birkhauser Boston Inc., Boston, MA, 2009, reprint of the 2001 edition.
MR2553578 (2010k:57039)
Bibliography 279

75. Svetlana Katok, Fuchsian groups, Chicago Lectures in Mathematics, University of

Chicago Press, Chicago, IL, 1992. MRl l 77168 (93d:20088)
76. Mikhail Khovanov, Link homology and categorification, International Congress of
Mathematicians. Vol. II, Eur. Math. Soc., Zurich, 2006, pp. 989-999. MR2275632
( 2008f:5 7009)
77. Robion Kirby, A calculus for framed links in 83 , Invent. Math. 45 (1978), no. 1,
35-56. MR0467753 (57 #7605)
78. Robion Kirby and Paul Melvin, The Ea -manifold, singular fibers and handlebody
decompositions, Proceedings of the Kirbyfest (Berkeley, CA, 1998), Geom. Topol.
Monogr., vol. 2, Geom. Topol. Puhl., Coventry, 1999, pp. 233-258. MR1734411
{2000j :57051)
79. Bruce Kleiner and John Lott, Notes on Perelman's papers, Geom. Topol. 12 (2008),
no. 5, 2587-2855. MR2460872 {2010h:53098)
80. Helmut Kneser, Geschlossene Flachen in dreidimensionalen Mannigfaltigkeiten, Jber.
d. D. Math. Verein. 38 (1929), 248-260.
81. Tsuyoshi Kobayashi, A construction of arbitrarily high degeneration of tunnel num­
bers of knots under connected sum, J. Knot Theory Ramifications 3 (1994), no. 2,
179-186. MR1279920 (95g:57011)
82. P. B. Kronheimer and T. S. Mrowka, Witten's conjecture and property P, Geom.
Topol. 8 (2004), 295-310 (electronic). MR2023280 (2004m:57023)
83. Marc Lackenby, Heegaard splittings, the virtually Haken conjecture and property (r) ,
Invent. Math. 164 (2006), no. 2, 317-359. MR2218779 (2007c:57030)
84. Tao Li, Heegaard surfaces and measured laminations. I. The Waldhausen conjecture,
Invent. Math. 167 (2007), no. 1, 135-177. MR2264807 (2008h:57033)
85. ___ , An algorithm to determine the Heegaard genus of a 3-manifold, Geom. Topol.
15 (2011), no. 2, 1029-1106. MR2821570 (2012m:57033)
86. ___ , Rank and genus of 3-manifolds, J. Amer. Math. Soc. 26 {2013), no. 3, 777-
829. MR3037787
87. W. B. Raymond Lickorish, An introduction to knot theory, Graduate Texts in Math­
ematics, vol. 175, Springer-Verlag, New York, 1997.
88. W. B. R. Lickorish, A representation of orientable combinatorial 3-manifolds, Ann.
of Math. 76 (1962), no. 2, 531-540.
89. Charles Livingston, Knot theory, Carus Mathematical Monographs, vol. 24, Mathe­
matical Association of America, Washington, DC, 1993. MR1253070 (94m:57021)
90. D. D. Long, A. Lubotzky, and A. W. Reid, Heegaard genus and property r for hy­
perbolic 3-manifolds, J. Topol. 1 (2008), no. 1, 152-158. MR2365655 (2008j :57036)
91. Feng Luo, Automorphisms of the complex of curves, Topology 39 (2000), no. 2, 283-
298. MRl 722024 (2000j :57045)
92. Jason Manning, Algorithmic detection and description of hyperbolic structures on
closed 3-manifolds with solvable word problem, Geom. Topol. 6 (2002), 1-26.
93. William S. Massey, Algebraic topology: An introduction, Springer-Verlag, New
York, 1977, reprint of the 1967 edition, Graduate Texts in Mathematics, Vol. 56.
MR0448331 (56 #6638)
94. H. A. Masur and Y. N. Minsky, Geometry of the complex of curves. I. Hyperbolicity,
Invent. Math. 138 {1999), no. 1, 103-149.
95. ___ , Geometry of the complex of curves. JI. Hierarchical structure, Geom. Funct.
Anal. 10 (2000), no. 4, 902-974.
280 Bibliography

96. Sergei Matveev, Algorithmic topology and classification of 3-manifolds, Algorithms

and Computation in Mathematics, vol. 9, Springer-Verlag, Berlin, 2003. MR1997069
(2004i:57026)
97. Curtis T. McMullen, The evolution of geometric structures on 3-manifolds, Bull.
Amer. Math. Soc. (N.S.) 48 (2011), no. 2, 259-274. MR2774092 (2012a:57024)
98. John Milnor, On manifolds homeomorphic to the 7-sphere, Ann. of Math. (2) 64
(1956), 399-405. MR0082103 (18,498d)
99. ___ , Morse theory. Based on lecture notes by M. Spivak and R. Wells, Annals of
Mathematics Studies, vol. 51, Princeton University Press, Princeton, NJ, 1963.
100. John W. Milnor, Topology from the differentiable viewpoint, Princeton Landmarks
in Mathematics, Princeton University Press, Princeton, NJ, 1997, based on notes by
David W. Weaver, revised reprint of the 1965 original. MR1487640 (98h:57051)
101. Edwin E. Moise, Affine structures in 3-manifolds. V. The triangulation theorem and
Hauptvermutung, Ann. of Math. 56 (1952), no. 2, 96-114.
102. __ , Geometric topology in dimensions 2 and 3, Springer-Verlag, New York, 1977,
Graduate Texts in Mathematics, Vol. 47. MR0488059 (58 #7631)
103. Frank Morgan, Manifolds with density and Perelman's proof of the Poincare conjec­
ture, Amer. Math. Monthly 1 16 (2009), no. 2, 134-142. MR2478057 (2010a:53164)
104. John Morgan and Gang Tian, Ricci flow and the Poincare conjecture, Clay Mathe­
matics Monographs, vol. 3, American Mathematical Society, Providence, RI, 2007.
MR2334563 (2008d:57020)
105. John W. Morgan and Hyman Bass (eds.), The Smith conjecture, Pure and Applied
Mathematics, vol. 112, Academic Press Inc., Orlando, F L, 1984, Papers presented at
the symposium held at Columbia University, New York, 1979. MR758459 (86i:57002)
106. Yoav Moriah and Jennifer Schultens, Irreducible Heegaard splittings of Seifert fibered
spaces are either vertical or horizontal, Topology 37 (1998), no. 5, 1089-1112.
MR1650355 (99g:57021)
107. Kanji Morimoto, There are knots whose tunnel numbers go down under connected
sum, Proc. Amer. Math. Soc. 123 (1995), no. 11, 3527-3532. MR1317043 (96a:57022)
108. Kanji Morimoto, Makoto Sakuma, and Yoshiyuki Yokota, Examples of tunnel number
one knots which have the property "1 + 1 3 ", Math. Proc. Cambridge Philos. Soc.
=

119 (1996), no. 1, 113-118. MR1356163 (96i:57007)

109. James R. Munkres, Topology: A first course, Prentice-Hall Inc., Englewood Cliffs,
NJ, 1975. MR0464128 (57 #4063)
110. Kunio Murasugi, On the braid index of alternating links, Trans. Amer. Math. Soc.
326 (1991), no. 1, 237-260.
111. Hossein Namazi, Heegaard splittings and hyperbolic geometry, ProQuest LLC, Ann
Arbor, MI, 2005, Thesis (Ph.D . )-State University of New York at Stony Brook.
MR2707466
112. Hossein Namazi and Juan Souto, Heegaard splittings and pseudo-Anosov maps,
Geom. Funct. Anal. 19 (2009), no. 4, 1195-1228. MR2570321 (2011a:57035)
113. Y. Ni, Uniqueness of PL minimal surfaces, Acta Math. Sin. (Engl. Ser.) 23 (2007),
no. 6, 961-964.
114. S. P. Novikov, The topology of foliations, Trudy Moskov. Mat. Obsc. 14 (1965),
248-278. MR0200938 (34 #824)
115. Peter Ozsvath, Andras I. Stipsicz, and Zoltan Szabo, Combinatorial Heegaard Floer
homology and nice Heegaard diagrams, Adv. Math. 231 (2012), no. 1, 102-171.
MR2935385
Bibliography 281

116. Peter Ozsvath and Zoltan Szabo, Knot Floer homology and the four-ball genus, Geom.
Topol. 7 (2003) , 615-639. MR2026543 (2004i:57036)
117. ___ , Knot Floer homology, genus bounds, and mutation, Topology Appl. 141
{2004), no. 1-3, 59-85. MR2058681 {2005b:57028)
118. Peter S. Ozsvath and Zoltan Szab6, Knot Floer homology and rational surgeries,
Algebr. Geom. Topol. 11 (2011), no. 1, 1-68. MR2764036 (2012h:57056)
119. C. D. Papakyriakopoulos, On Dehn's lemma and the asphericity of knots, Ann. of
Math. 66 {1957), no. II, 1-26.
120. R. C. Penner and J. L. Harer, Combinatorics of train tracks, Annals of Mathematics
Studies, vol. 125, Princeton University Press, Princeton, NJ, 1992. MR1144770
(94b:57018)
121. G. Perelman, The entropy formula for the Ricci flow and its geometric applications,
http: //arxiv.org/ abs/math/0211159.
122. ___ , Finite extinction time for the solutions to the Ricci flow on certain three­
manifolds, http://arxiv.org/abs/math/0307245.
123. ___ , Ricci flow with surgery on three-manifolds, http://arxiv.org/abs/math/
0303109.
124. John G. Radcliffe, Foundations of hyperbolic manifolds, second ed., Graduate Texts
in Mathematics, vol. 149, Springer Verlag, New York, 2006.
125. Yo'av Rieck and Eric Sedgwick, Thin position for a connected sum of small knots,
Algebr. Geom. Topol. 2 {2002), 297-309 {electronic) .
126. Igor Rivin, Euclidean structures on simplicial surfaces and hyperbolic volume, Ann.
of Math. (2) 139 (1994), no. 3, 553-580. MR1283870 {96h:57010)
127. Dale Rolfsen, Knots and links, Mathematics Lecture Series, vol. 7, Publish or Perish
Inc., Houston, TX, 1990, corrected reprint of the 1976 original.
128. C. Rourke and B. Sanderson, Introduction to piecewise-linear topology, Springer
Study Edition, Springer-Verlag, Berlin-New York, 1982, reprint of the 1972 origi­
nal.
129. Hyam Rubinstein and Martin Scharlemann, Comparing Heegaard splittings of non­
Haken 3-manifolds, Topology 35 (1996), no. 4, 1005-1026. MR1404921 (97j :57021)
130. __ , Transverse Heegaard splittings, Michigan Math. J. 44 {1997), no. 1, 69-83.
MR1439669 (98c:57017)
131. ___ , Comparing Heegaard splittings-the bounded case, Trans. Amer. Math. Soc.
350 (1998), no. 2, 689-715. MR1401528 (98d:57033)
132. Martin Scharlemann, Constructing strange manifolds with the dodecahedral space,
Duke Math. J. 43 (1976), no. 1, 33-40. MR0402760 (53 #6574)
133. ___ , Heegaard splittings of compact 3-manifolds, Handbook of geometric topology,
North-Holland, Amsterdam, 2002, pp. 921-953. MR1886684 (2002m:57027)
134. Martin Scharlemann and Jennifer Schultens, The tunnel number of the sum of n
knots is at least n, Topology 38 {1999), no. 2, 265-270. MR1660345 (2000b:57013)
135. ___ , Annuli in generalized Heegaard splittings and degeneration of tunnel number,
Math. Ann. 317 (2000), no. 4, 783-820.
136. Martin Scharlemann and Abigail Thompson, Thin position and Heegaard splittings
of the 3-sphere, J. Differential Geom. 39 (1994), no. 2, 343-357.
137. ___ , Thin position for 3-manifolds, Geometric topology (Haifa, 1992), Contemp.
Math., vol. 164, Amer. Math. Soc., Providence, RI, 1994, pp. 231-238.
282 Bibliography

138. Martin G. Scharlemann, Unknotting number one knots are prime, Invent. Math. 82
(1985), no. 1, 37-55. MR808108 (86m:57010)
139. Saul Schleimer, Waldhausen's theorem, Workshop on Heegaard Splittings, Geom.
Topol. Monogr., vol. 12, Geom. Topol. Pub!., Coventry, 2007, pp. 299-317.
MR2408252
140. ___ , The end of the curve complex, Groups Geom. Dyn. 5 (2011), no. 1, 169-176.
MR2763783
141. Horst Schubert, Knoten und Vollringe, Acta Math. 90 (1953), 131-286.
142. ___ , Uber eine numerische Knoteninvariante, Math. Z. 61 (1954), 245-288.
143. Jennifer Schultens, The classification of Heegaard splittings for (compact orientable
surface) x 8 1 , Proc. London Math. Soc. (3) 67 ( 1993), no. 2, 425-448.
144. Jennifer Schultens and Richard Weidman, On the geometric and the algebraic rank
of graph manifolds, Pacific J. Math. 231 (2007), no. 2, 481-510. MR2346507
(2009a:57030)
145. Peter Scott, The geometries of 3-manifolds, Bull. London Math. Soc. 15 (1983),
no. 5, 401-487. MR705527 (84m:57009)
146. Herbert Seifert and William Threlfall, Seifert and Threlfall: A textbook of topology,
Pure and Applied Mathematics, vol. 89, Academic Press Inc. (Harcourt Brace Jo­
vanovich Publishers], New York, 1980, translated from the German edition of 1934
by Michael A. Goldman, with a preface by Joan S. Birman, with "Topology of 3-
dimensional fibered spaces" by Seifert, translated from the German by Wolfgang
Heil. MR575168 (82b:55001)
147. Zlil Sela, The isomorphism problem for hyperbolic groups I, Ann. Math. 141 (1995),
217-283.
148. I. M. Singer and J. A. Thorpe, Lecture notes on elementary topology and geome­
try, second ed., Undergraduate Texts in Mathematics, Springer-Verlag, New York­
Heidelberg-Tokyo, 1976.
149. John Stallings, On the loop theorem, Ann. of Math. (2) 72 (1960), 12-19. MR0121796
(22 #12526)
150. Norman Steenrod, The topology of fibre bundles, Princeton Landmarks in Mathe­
matics, Princeton University Press, Princeton, NJ, 1999, reprint of the 1957 edition,
Princeton Paperbacks. MR1688579 (2000a:55001)
151. Abigail Thompson, Thin position and bridge number for knots in the 3-sphere, Topol­
ogy 36 (1997), no. 2, 505-507.
152. ___ , Algorithmic recognition of 3-manifolds, Bull. Amer. Math. Soc. (N.S.) 35
(1998), no. 1, 57-66.
153. William P. Thurston, The geometry and topology of three-manifolds, http://www.
msri.org/publications/books/gt3m/.
154. ___ , Three-dimensional geometry and topology, vol. 1, Princeton Mathematical
Series, vol. 35, Princeton University Press, Princeton, NJ, 1997.
155. F. Waldhausen, Heegaard-Zerlegungen der 3-Sphiire, Topology 7 (1968) , no. 2, 195-
203.
156. Friedhelm Waldhausen, On irreducible 3-manifolds which are sufficiently large, Ann.
of Math. (2) 87 (1968), 56-88. MR0224099 (36 #7146)
Index

27r-Theorem, 237 branch locus, 43

2-bridge, 1 1 9 branched covering, 43
2-fold branched cover, 141 bridge number, 129
3-colorable, 107 bundle, 57
81 + 82 , 157 bundle atlas, 58
a-thin, 251 bundle chart, 58
k-handle, 175, 273
k-simplex, 13 characteristic submanifold, 98
n-manifold, 1 chart, 1
n-torus, 2 classification of surfaces, 37
r-skeleton, 143 closure of a braid, 123
cocore, 176
Ago!, 237 collar, 60
Alexander Trick, 45, 53 companion, 121
Alexander's Theorem, 67, 71 complete, 218
almost normal, 170 complexity, 248
alternating, 1 15 compressible, 73
anannular, 224 compression body, 202
Anosov, 246 cone, 148
arational, 246 connected sum, 23, 108
arborescent knot, 1 1 9 convex, 219
arc complex, 257 core, 176
atlas, 1 covering, 42
atoroidal, 96, 224 covering space, 139
crossing number, 1 1 4
barycentric coordinates, 144 curve complex, 248
base space, 58 cut, 63
bicollar, 60 cut and paste argument, 99
blackboard framing, 227 cyclic surgery, 232
boundary, 4
boundary incompressible, 74 decomposing annulus, 210
boundary irreducible, 70 decomposing sphere, 167
braid, 122 Dehn filling, 227
-

283
284 Index

Dehn surgery, 227, 232 Heegaard splitting, 178, 202, 228

Dehn surgery space, 237 height function, 64, 128
Dehn twist, 50, 245 hierarchy, 80
Dehn's Lemma, 76 homotopy, 23
Dehn's Theorem, 139 homotopy equivalent, 83
destabilization, 190 Hopf link, 117
dilation, 218 horizontal, 92, 194, 254
dimension, 1 hyperbolic arc length, 215
Diophantine system, 159 hyperbolic n-manifold, 221
distance, 248 hyperbolic volume, 215
distance of the Heegaard splitting, 252
double, 90 ideal triangle, 219
incompressible, 73
embedding, 4 independent, 162
ends, 258 index, 211, 269
equivalent, 19, 22, 59, 102, 178, 205 inessential, 32
essential, 32, 73, 75 infinite cyclic cover, 139
Euler characteristic, 20 innermost disk argument, 98
exceptional fiber, 87 intersection number, 31
inversion, 219
face, 14 irreducible, 62, 183
fiber, 58, 87 isometry, 22
fibering, 58 isomorphic, 59
fill, 253 isomorphism, 58
filling, 45, 246 isotopy, 23
foliation, 238
fundamental, 160 Jones polynomial, 126
fundamental group, 139 Jordan Curve Theorem, 30, 64
Gauss-Jordan elimination, 159 Kneser-Haken finiteness, 162
general position, 55, 147 knot, 102
generalized Heegaard splitting, 204 knot diagram, 103
generic, 256 knot invariant, 113
genus, 38, 112, 177, 193
geodesic, 218 lamination, 243
geometric intersection number, 249 leaf, 238
geometric manifold, 22 leaves, 243
Geometrization Conjecture, 224 length, 82, 149, 235
glue, 99 lens space, 83
glued, 178 link, 102, 146
good system, 188 linking number, 117
granny knot, 109 longitude, 106, 227
graph manifold, 97 Loop Theorem, 79
Gromov, 237
Gromov hyperbolic, 251 mapping class group, 48, 255
Gromov-Hausdorff topology, 256 measure, 152
meridian, 106, 227
Haken 3-manifold, 75 meridian disks, 177
handle decomposition, 274 metric triangulation, 173
handlebody, 177 Mobius band, 10
Hausdorff distance, 256 Montesinos knot, 119
Heegaard diagram, 181 Morse function, 9, 64, 270
Heegaard genus, 212 Mostow Rigidity Theorem, 225
Index 285

natural framing, 227 Ricci flow, 224

negative curvature, 173 Riemann-Hurwitz Theorem, 44
non-degenerate, 269 rotation, 219
non-separating, 71
normal curve, 149 satellite knot, 121
normal disk, 151 Scharlemann cycle, 235
normal isotopy, 152 Schonfties Theorem, 63, 67, 104, 163
normal surface, 151 section, 60
normal triangle, 150 Seifert fibered space, 88, 238, 253
nugatory, 1 1 5 Seifert surface, 1 10, 1 19, 226
Seifert's Algorithm, 1 10
open regular neighborhood, 60 self-indexing, 274
opposite, 1 1 separating, 32, 71
orientable, 10 simple arc, 29
orientation, 31 simple closed curve, 29
orientation-preserving, 12 simplices, 248
orientation-reversing, 12 simplicial complex, 15, 19
outermost arc argument, 99 simplicial isomorphism, 19
simplicial map, 18
pair of pants, 5 simply connected at infinity, 86
pants decomposition, 40 small, 136
partial ordering, 144 sphere, 2
pattern, 121 Sphere Theorem, 79
Perelman, 224 spine, 194
periodic, 53 square knot, 109
PL least area, 173 square restriction, 157
Poincare Conjecture, 84, 224, 237 stabilization, 182
Poincar&-Hopf Index Theorem, 26, 64, stable, 247
67, 239, 242 standard Heegaard splitting, 194
pretzel knot, 1 1 9 standard innermost disk argument, 92
prime, 24 star, 146
prime decomposition, 162, 166 strongly irreducible, 183, 205
prime factorization, 55 subcomplex, 19
prism manifolds, 90 subdivision, 145
projection, 58 submanifold, 3
projective measured lamination, 245 subsurface projection, 257
projectively equivalent, 245 swallow-follow torus, 121
proper, 73 sweepout, 196
properly discontinuously, 221
Property P, 236 tangent bundle, 261
pseudo-Anosov, 53, 247, 250 taut, 242
punctured, 162 thick level, 131
thin level, 132
rank, 212 thin position, 130
real projective space, 2 torus knot, 105
reduced, 1 1 5 total space, 58
reducible, 5 3 , 6 2 , 183 translation, 219
Reeb foliation, 239 transversal, 241
Reebless, 241 transversality, 102
regular neighborhood, 50, 60 transverse, 263, 264
Reidemeister moves, 106 transverse invariant measure, 243
restriction, 59 transverse isotopy, 243
286 Index

transversely, 123
trivial, 23
tunnel system, 208

uniquely ergodic, 246

universal, 139
unknot, 102
unknotting number, 116
unstable, 247
untelescoping, 206
upper half-space model, 217

vertical, 92, 194, 254

Waldhausen's Theorem, 83
weight, 152
weighted intersection number, 245
Whitehead manifold, 84
width, 130
wild knot, 101
Wirtinger presentation, 137
T h i s b o o k grew out of a grad uate c o u rse on 3 - m a n i fo l d s a n d is i nte n d e d fo r a math ­
e m ati cally experienced a u d i e n c e that is n ew to l ow- d i m e n s i o n a l to pol ogy.

The expos ition begi ns with the d efi n iti o n of a m a n i fo l d , exp l o res poss i b l e additional
structu res o n m a n ifo l d s , d i scusses the c l ass ifi c ati o n of s u rfaces, i ntro d u ces key fou n ­
d ati o n a l res u lts fo r 3 - man ifo l d s , a n d p rovi de s an ove rvi ew o f knot theo ry. It th e n
conti n u e s w i t h m o re special ized to pics by b r i efly c o n s i d e r i ng triangu l ati o n s of
3 - ma n ifo l d s , n o rmal s u rface theo ry, and H e egaard s p l itti ngs. T h e book fi n i s h es with a
d i s c u s s i o n of top i cs re l evant to viewing 3 - m a n i fo l d s via the c u rve co m p l ex.

With about 250 figu res a n d m o re than 200 exe rc i ses, th i s book can se rve as an excel­
lent ove r v i ew a n d starting p o i nt fo r the study of 3 - m a n i fo l d s .

I SBN 978-1 -4704-1 020-9

9 781 470 41 0209

G S M/ 1 5 1