Knots, Molecules, and The Universe - An Introduction To Topology (Flapan)

KNOTS, MOLECULES,
AND THE UNIVERSE:

AN INTRODUCTION TO TOPOLOGY
https://doi.org/10.1090//mbk/096
KNOTS, MOLECULES,
AND THE UNIVERSE:
AN INTRODUCTION TO TOPOLOGY
ERICA FLAPAN
with
Maia Averett David Clark Lew Ludwig

Lance Bryant Vesta Coufal Cornelia Van Cott
Shea Burns Elizabeth Denne Leonard Van Wyk
Jason Callahan Berit Givens Robin Wilson
Jorge Calvo McKenzie Lamb Helen Wong
Marion Moore Campisi Emille Davie Lawrence Andrea Young
AMERICAN MATHEMATICAL SOCIETY

2010 Mathematics Subject Classification. Primary 57M25, 57M15,
92C40, 92E10, 92D20, 94C15.
For additional information and updates on this book, visit

www.ams.org/bookpages/mbk-96
Library of Congress Cataloging-in-Publication Data

Flapan, Erica, 1956–
Knots, molecules, and the universe : an introduction to topology / Erica Flapan ; with Maia
Averett [and seventeen others].
pages cm
Includes index.
ISBN 978-1-4704-2535-7 (alk. paper)
1. Topology—Textbooks. 2. Algebraic topology—Textbooks. 3. Knot theory—Textbooks.
4. Geometry—Textbooks. 5. Molecular biology—Textbooks. I. Averett, Maia. II. Title.
QA611.F45 2015
514—dc23
2015031576
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 select pages 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, systematic copying, or multiple reproduction of any material in this publication
is permitted only under license from the American Mathematical Society. Permissions to reuse
portions of AMS publication content are handled by Copyright Clearance Center’s RightsLink
service. For more information, please visit: http://www.ams.org/rightslink.
Send requests for translation rights and licensed reprints to reprint-permission@ams.org.
Excluded from these provisions is material for which the author holds copyright. In such cases,
requests for permission to reuse or reprint material should be addressed directly to the author(s).
Copyright ownership is indicated on the copyright page, or on the lower right-hand corner of the
first page of each article within proceedings volumes.

c 2016 by the American Mathematical Society. All rights reserved.
The American Mathematical Society retains all rights
except those granted to the United States Government.
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://www.ams.org/
10 9 8 7 6 5 4 3 2 1 21 20 19 18 17 16
Contents
Preface xi
Acknowledgments xv
Part 1. Universes
Chapter 1. An Introduction to the Shape of the Universe 3

1. To Infinity and Beyond 3
2. A. Square and His Universe 5
3. Straightest Paths in Flatland 8
4. Exploring the Shape of a Cave 11
5. Creating Universes by Gluing 13
6. Games on a Gluing Diagram for a Cylinder and a Torus 15
7. Extended Diagrams 17
8. Introducing the 3-Dimensional Torus and Sphere 19
9. Distinguishing a 2-Sphere from a 2-Torus 22
10. Distinguishing a 3-Sphere from a 3-Torus 24
11. Exercises 26
Chapter 2. Visualizing Four Dimensions 29

1. Teaching A. Square About the Third Dimension 29
2. Projections and Perceptions 31
3. Movies and Movement 33
4. Inductively Defining Cubes of Dimensions 0, 1, 2, and 3 36
5. A. Square Learns About a 3-Dimensional Cube 37
6. A 4-Dimensional Cube 40
7. Tetrahedra in Various Dimensions 42
8. Exercises 44
Chapter 3. Geometry and Topology of Different Universes 49

1. Intrinsic and Extrinsic Properties of a Space 49
2. Intrinsic and Extrinsic Geometry 52
3. Straightest Paths and Geodesics 53
4. The Definition of a Triangle 59
5. The Sum of the Angles of a Triangle 60
6. Triangles on a Flat and Curved Torus 61
7. Triangles on a Flat and a Curved Cylinder 62
8. Extrinsic and Intrinsic Topology 64
9. Using Loops to Understand Intrinsic Topology 67
10. Local and Global Properties 69
v
vi CONTENTS
11. Manifolds 71
12. Assumptions about Universes 72
13. A 2-Dimensional Sphere with a Point Removed 74
14. A 3-Dimensional Sphere with a Point Removed 77
15. Curves on a Torus 78
16. How to Draw a (p, q)-Curve on a Flat Torus 80
17. Lines in the Extended Diagram of a Flat Torus 83
18. Exercises 84
Chapter 4. Orientability 87
1. The Möbius Strip 87
2. Orientation Reversing Paths 89
3. The Klein Bottle 90
4. The 3-Dimensional Klein Bottle 92
5. Tic-Tac-Toe on a Torus and Klein Bottle 93
6. The Projective Plane 96
7. The Projective Plane with a Disk Removed 98
8. Projective 3-Dimensional Space 100
9. 1-Sided and 2-Sided Surfaces 101
10. Non-orientability and 1-Sidedness 103
11. Exercises 106
Chapter 5. Flat Manifolds 109

1. Flat Surfaces 109
2. Flat Gluing Diagrams 110
3. The Point of a Cone 114
4. Using Extended Diagrams to Understand Cone Points 116
5. Anti-cone Points 119
6. We Show that the 3-Torus Is Flat 120
7. A Method to Determine if a Glued Up Cube Is Flat 122
8. Extended Diagrams of Glued Up Cubes 123
9. Other Types of Gluings of a Cube 127
10. Exercises 128
Chapter 6. Connected Sums of Spaces 135

1. Einstein-Rosen Bridges 135
2. Connected Sums of Surfaces 136
3. Arithmetic Properties of the Connected Sum 137
4. Gluing Diagrams for nT 2 and nP 2 138
5. The Classification of Surfaces 140
6. Dividing a Surface into Vertices, Edges, and Faces 142
7. The Euler Characteristic of a Surface 144
8. The Euler Characteristic of Connected Sums 146
9. The Genus of a Surface 147
10. The Genus of nT 2 and nP 2 150
11. Connected Sums of 3-Manifolds 151
12. Exercises 153
Chapter 7. Products of Spaces 157

1. Products of Sets 157
CONTENTS vii
2. Products of Spaces 159

3. A. Square and A. Pentagon Are Products 160
4. Some Products where One Factor Is a Circle 162
5. Examples of Spaces Incorrectly Expressed as Products 164
6. The Topological Uniqueness of Products 166
7. The Dimension of Product Spaces 167
8. Visualizing S 2 × S 1 and nT 2 × S 1 168
9. Geometric Products 170
10. Geometric Products of Flat Spaces 171
11. A Flatland-Friendly Geometric S 1 × I 173
12. Flat 3-Dimensional Spaces as Geometric Products 174
13. A Geometric nT 2 × S 1 177
14. A Geometric S 2 × S 1 179
15. Isotropic and Non-isotropic Spaces 180
16. Exercises 181
Chapter 8. Geometries of Surfaces 185

1. Euclid’s Axioms 185
2. Flat Surfaces and Euclidean Geometry 191
3. Some Alternative Axioms 192
4. Spherical Trigonometry 193
5. The Area of a Disk in a Sphere 196
6. Maps of the Earth 199
7. Hyperbolic Geometry 200
8. A. Square Learns to Draw a Hyperbolic Plane 202
9. Homogeneous Geometries for all nT 2 with n ≥ 2 205
10. A Homogeneous Geometry for P 2 207
11. Uniqueness of Homogeneous Geometries for Surfaces 208
12. Exercises 209
Part 2. Knots
Chapter 9. Introduction to Knot Theory 215

1. 1-Dimensional Universes 215
2. When Are Two Knots Equivalent? 217
3. The Mirror Image of a Knot or Link 219
4. The Connected Sum of Two Knots 220
5. A Brief History of Knot Theory 221
6. Reidemeister Moves 223
7. Coloring Knots with Three Colors 226
8. Tricolorability and Knot Equivalence 229
9. Oriented Knots and Invertibility 232
10. Connected Sums of Non-invertible Knots 233
11. Exercises 235
Chapter 10. Invariants of Knots and Links 239

1. What’s an Invariant? 239
2. Crossing Number, Tricolorability, and Number of Components 241
3. Positive and Negative Crossings 242
viii CONTENTS
4. Writhe 244
5. Linking Number 245
6. Nugatory Crossings and Alternating Knots and Links 247
7. Tait’s Conjectures about Alternating Knots 248
8. What Proportion of Knots Are Alternating? 250
9. Seifert Surfaces 252
10. The Genus of a Knot 254
11. Using Euler Characteristic to Compute Genus 257
12. A Potpourri of Knot Invariants 258
13. Exercises 263
Chapter 11. Knot Polynomials 269

1. An Introduction to Polynomial Invariants 269
2. The Rules for the Bracket Polynomial 270
3. The Bracket Polynomial and Reidemeister II Moves 272
4. The Bracket Polynomial with Only One Variable 274
5. The Bracket Polynomial and Reidemeister I Moves 275
6. The X-Polynomial 277
7. The Jones Polynomial 279
8. The State Model for Computing the Bracket Polynomial 282
9. Exercises 285
Part 3. Molecules
Chapter 12. Mirror Image Symmetry from Different Viewpoints 291

1. Mirror Image Symmetry 291
2. Geometric Symmetry 293
3. Geometric Chirality and Achirality 294
4. Chemical Chirality and Achirality 295
5. Chemical Achirality and Geometric Chirality of Figure 8 298
6. Euclidean Rubber Gloves 301
7. Geometrically Chiral and Achiral Knots 302
8. A Topological Rubber Glove 304
9. Topological Chirality 306
10. Exercises 308
Chapter 13. Techniques to Prove Topological Chirality 311

1. Topological Chirality 311
2. Molecular Knots 312
3. Technique 1: Knot Polynomials 313
4. Technique 2: 2-Fold Branched Covers 315
5. Technique 3: Chiral Subgraphs 318
6. Some Graph Theory 320
7. Technique 4: A Combinatorial Approach 322
8. Exercises 325
Chapter 14. The Topology and Geometry of DNA 329

1. Synthetic versus Biological Molecules 329
2. The Biology of DNA 330
3. The Problem of Packing DNA into a Cell 332
CONTENTS ix
4. Supercoiling 333
5. Visualizing DNA as a Circular Ribbon 335
6. The Linking Number of the Backbones 337
7. The Average Writhe of the Axis 339
8. The Twist of a Backbone around the Axis 340
9. The Relationship between Linking, Twisting, and Writhing 343
10. Representing the Supercoiling with a Single Number 345
11. Replication 346
12. Site Specific Recombination 348
13. Knotted and Linked Products of Recombination 349
14. An Introduction to Tangles 350
15. Rational Tangles 352
16. Operations on Tangles 354
17. The Tangle Model of Site Specific Recombination 355
18. Applying the Model to the Enzyme Tn3 Resolvase 357
19. Exercises 358
Chapter 15. The Topology of Proteins 361

1. An Introduction to Proteins 361
2. Trapping Knots in Proteins 362
3. Which Knots Occur in Proteins? 364
4. Examples of Knotted and Linked Proteins 366
5. Non-planar Graphs in Metalloproteins 369
6. The Protein Structure Nitrogenase 373
7. Möbius Ladders in Metalloproteins 376
8. Möbius Ladders in Small Proteins 378
9. Goodbye for Now 380
10. Exercises 381
Index 383
Preface
There are a quite a few excellent topology textbooks written for undergraduate
mathematics majors in their junior or senior year of college. Such books usually
begin with an introduction to point set topology, followed by geometric topics such
as homotopy, covering spaces, knot theory, the classification of surfaces, or some
area of applied topology. While not all undergraduate topology texts follow this
precise structure, almost all presuppose experience with proofs and exposure to a
rigorous definition of continuity. For math departments that do not have a large
number of majors or have an inordinate burden of service courses to teach, offering
a topology course which presupposes such a background is almost impossible. As a
result, many mathematics majors graduate without having the opportunity to take
a topology class.
In addition to textbooks written for math majors, there are a number of note-
worthy intuitive introductions to topology intended for a wider audience. Such
books are great for curious students and for the general public, but can be difficult
to teach out of because they rarely include definitions, statements of theorems, or
a sufficient number of elementary exercises (without solutions in the back).
What seems to be absent from the available topology books is one that is easy
to teach from, includes a wide range of topics, has no prerequisites, and makes no
assumptions about the mathematical sophistication or motivation of the reader.
Such a book could be the basis for a course that would attract students to the
math major who might not be excited by the analytical and algebraic arguments
that they typically see in their math courses during the first two years of college.
This book is intended to fill this gap. It is an elementary introduction to
geometric topology and its applications to chemistry, molecular biology, and cos-
mology. It does not assume any particular mathematical or scientific background,
sophistication, or even motivation to study abstract mathematics. It is meant to be
fun and engaging while at the same time drawing students in to learn about funda-
mental topological and geometric ideas. Though the book can be read and enjoyed
by non-mathematicians, college students, or even eager high school students, it is
intended to be used as an undergraduate textbook.
With this in mind, all of the concepts are introduced with explicit definitions
and the theorems are clearly stated so that it is easy for an instructor to develop
lectures and for students to refer to when doing the homework. Some theorems
are proved formally, others are proved intuitively, and for those results that aren’t
proved, students are told in which advanced course they might expect to see a
proof. In addition, each chapter concludes with numerous exercises which are at a
level appropriate for students in their first two years of college, and whose solutions
are only available to instructors.
xi
xii PREFACE
In contrast with most topology textbooks, the style of the book is informal
and lively. Though all of the definitions and theorems are explicitly stated, they
are given in an intuitive rather than a rigorous form. For example, rather than
introducing the formal definition of an isotopy, two knots are defined to be equiva-
lent if one can be deformed to the other. This style of presentation allows students
to develop intuition about topology and geometry without getting bogged down
in technical details. In addition, the topics in the book were chosen for their vi-
sual appeal, which is highlighted by the abundance of illustrations throughout.
In order to make the text more fun and engaging, early on it introduces a cast
of characters who reappear in multiple chapters. This includes the 1-dimensional
character A. Dash, the 2-dimensional characters A. Square and B. Triangle, and
the 3-dimensional character A. 3D-Girl, each of whom is learning about geometry
and topology in order to understand his or her own universe. The characters are
illustrated with amusing pictures that entice readers to think about the experience
from the character’s point of view.
The book is divided into three parts corresponding to the three areas referred to
in the title. Part 1, consisting of Chapters 1–8, concerns two and three dimensional
universes, though there is a brief foray into the fourth dimension in Chapter 2. The
goal of Part 1 is to develop techniques that enable creatures in a given space to
visualize possible shapes for their universe, and to use topological and geometric
properties to distinguish one such space from another.
The second part of the text, consisting of Chapters 9–11, provides an introduc-
tion to knots and links. Building on the ideas developed in Part 1, the emphasis in
Part 2 is on deformations of knots and links, and the use of invariants to distinguish
inequivalent knots and links. Tricolorability is introduced in Chapter 9 as an in-
variant which is relatively simple to understand and apply but does not distinguish
many knots. Then Chapter 10 surveys a collection of invariants which are not hard
to understand but in some cases are hard to compute, and in the exercises we ask
the reader to figure out what invariants should be used to distinguish a given pair
of knots or links. Finally, Chapter 11 introduces the Kauffman bracket and Jones
polynomial, and provides a step-by-step explanation of how to compute them, as
well as some theorems that illustrate their power.
The third part of the text, consisting of Chapters 12–15, presents applications of
topology and geometry to chemistry and molecular biology. In particular, Chapter
12 compares the concepts of mirror image symmetry from geometric, topological,
and chemical viewpoints. It also introduces the idea of using embeddings of graphs
in 3-dimensional space as models of non-rigid molecules. Then Chapter 13 presents
techniques that can be used to show that some non-rigid molecules are topologically
distinct from their mirror images. Chapter 14 explores the topology and geometry
of DNA, developing material about rational tangles as a way to model DNA recom-
bination. Finally, Chapter 15 describes topologically complex protein structures,
giving examples of knotted and linked proteins, as well as proteins containing non-
planar graphs. While Chapter 13 depends on Chapter 12, Chapters 14 and 15 are
independent of one another and of Chapters 12 and 13. Thus an instructor can
choose which of the applied topics they would like to cover.
The three parts of the book make it usable as a textbook for several different
types of courses. For example, Part 1 of the book could be the basis for a seminar
for first year college students on a topic such as “Dimensions and the Shape of
PREFACE xiii
the Universe.” Since the book is written with students in mind, the instructor
could give daily reading assignments and then lead the class in a discussion of the
material followed by breaking the class into small groups to work through some
of the exercises. Such a seminar would also lend itself well to open ended writing
projects on diverse topics: Describe a sport that could take place in a 2-dimensional
universe. Pick a well-known game (for example chess, checkers, Othello, Hex, or
Go), and explain how the strategy for the game would change if it were played on
a flat torus or a flat Klein bottle. Design a 4-dimensional version of such a game,
and explain how the strategy compares to the usual 2- or 3-dimensional version.
Imagine the form of a 4-dimensional person, and explain what we would see if the
person passed through our space in different ways. If we had a spaceship that could
travel in any direction as fast and far as we would like, what evidence should the
spaceship seek to help us figure out the shape of our universe? These are a few of
the topics.
The book could serve as the text for a topology course for math majors with or
without prerequisites. The instructor could either cover all three parts of the text,
or pick and choose topics from each of the three parts and supplement with point set
topics from a more traditional topology text. Another option for a course for math
majors would be for the instructor to give lectures to introduce each chapter, and
then have students take turns giving lectures on each section within a chapter. The
sections were intentionally made short to enable students to read and understand
a single section at a time.
The text could also be used for an interdisciplinary course linking topology
with molecular biology and chemistry. In this case, the course would focus on
Part 2 and Part 3 (comprising Chapters 9–15). Part 2 would give students enough
background in knot theory to understand the applications to molecular symmetry
and the topology of DNA and proteins developed in Part 3. Note that it is not
necessary for students to read Part 1 in order to understand Part 2, nor is it
necessary for students to have a background in biology or chemistry in order to
understand Part 3.
A geometry course for future teachers could also be based on this text. Such
courses often consist of an axiomatic approach to Euclidean and non-Euclidean ge-
ometry. However, approaching geometry this way can be tedious and may actually
kill any natural interest the student has in geometry. By contrast, a course for fu-
ture teachers based on this text would focus on developing geometric thinking and
visual intuition together with problem solving and mathematical reasoning. Such
a course would cover Part 1, culminating in Chapter 8 which presents Euclid’s five
Axioms and alternatives to the Fifth Axiom that are valid on a sphere and a hy-
perbolic plane. The instructor for such a course would want to present the material
slowly in order to emphasize visualization and logical deduction throughout. The
nearly 450 illustrations in the text are designed to help students develop their visual
intuition, and the many exercises at the end of each chapter would give students
necessary practice writing mathematical arguments. Furthermore, the exercises
are designed to encourage students to explain their reasoning in complete sentences
and integrate illustrations into their explanations whenever possible. This course
could culminate with a project such as designing a museum exhibit featuring the
student’s choice of the ten most important images from the text with an introduc-
tion to the exhibit and short explanatory descriptions associated with each image.
xiv PREFACE
The students could also be asked to include some 3-dimensional models constructed
themselves.
Finally, because the text is readable by students and is self-contained, the book
would be well suited for an independent study or senior year project. Also, because
of its lack of prerequisites and its division into three parts and the subdivision of
each chapter into short sections, it would work well as the textbook for a four-week
winter term or a summer course.
We hope that at least in some small way this book contributes to one or more
of the following large goals: it motivates some undergraduates to major in math; it
motivates some math majors to study topology; it motivates some math students
to read their textbooks; it convinces some future K–12 teachers that geometry is
fun; or it convinces some future scientists, administrators, or government officials
that even the purest areas of mathematics can have important applications. At the
very least, we hope that you and your students enjoy reading it.
Acknowledgments
As is apparent from the eighteen co-authors listed on the cover, this book has
an unusual history. It began as lecture notes I developed for an intensive course I
taught at the Summer Mathematics Program (SMP) at Carleton College for eleven
summers between 2000 and 2014. The SMP was a very successful four-week, NSF-
funded program for women students in their first and second year of college, with
the goal of motivating, encouraging, and mentoring these students to get PhD’s in
mathematics. The students in the program had some prior exposure to proofs, but
there was no specific prerequisite material. The aim of my course was to introduce
them to the beauty and applications of topology and geometry, with the hope that
the visual types of thinking they would experience might motivate those who were
not as excited by algebraic and analytical arguments to continue in mathematics.
My course was originally inspired by three books. I used ideas from Jeffrey
Weeks’ wonderful book The Shape of Space to introduce the geometry and topology
of 2- and 3-manifolds; I used ideas from Colin C. Adams’ excellent book The Knot
Book to expose students to knot theory; and I used ideas from my own book
When Topology Meets Chemistry to explore applications of topology to molecular
structures. However, over the fourteen years that I taught the course, my lectures
took on a life of their own, veering away from these books. In particular, my
approach to 2- and 3-manifolds became more formal than that of Weeks, though
some of the arguments and organization in Chapters 1 through 8 are similar to his.
For example, the proof in Chapter 4 that a surface in an orientable 3-manifold is 1-
sided if and only if it is non-orientable is essentially the same as his, simply because
I couldn’t imagine a better proof. My approach to knot theory became focused
on invariants in contrast with Adams’ comprehensive introduction to knot theory,
and my approach to applications of topology became less rigorous and more light-
hearted than that of When Topology Meets Chemistry, and included the topology
of proteins which was not in my earlier book.
While my lecture notes worked well for the intensive course that I taught at
SMP, I did not originally foresee them becoming a book that could be used more
widely. However, my ideas changed when I led the Undergraduate Faculty Program
(UFP) at the Park City Mathematics Institute (PCMI) in 2011. As the leader of
the UFP, I was charged with guiding a group of faculty to collaboratively produce
materials that could be used at a wide range of institutions. I helped select the 18
UFP participants largely based on their interest in developing an undergraduate
topology course with few or no prerequisites. These faculty members, who ranged
from post-docs to full professors, came from a great variety of institutions. Some of
their institutions were small while others were large, some were private and others
public, and some had significant numbers of students headed for graduate school
xv
xvi ACKNOWLEDGMENTS
in mathematics, while others had few math majors, almost none of whom would go
on to graduate school.
During the UFP, I divided my lectures notes into twelve parts that would
become twelve chapters of the manuscript. I then divided the eighteen participants
into six groups of three and assigned each group two parts of my lecture notes to
make into two chapters, one in the first half of the book and the other in the second
half. Starting with their parts of the notes, the groups added topics, examples,
exercises, figures, and more complete explanations. In addition, all eighteen UFP
participants helped to write solutions to the exercises that were in the 2011 draft
of the manuscript. After the program was over, some of the participants continued
working on the manuscript in various capacities including editing chapters and
adding more exercises and solutions. Ultimately, I revised, edited, rewrote, and
added to the final manuscript so that it would all be in the same style and level,
and would flow well.
Below, I list the eighteen UFP participants together with their individual con-
tributions to the manuscript during and after the program.
Maia Averett co-wrote Chapters 2 and 9, and edited Chapter 8.
Lance Bryant co-wrote Chapters 3 and 10, and edited solutions for ex-
ercises in Chapter 3.
Shea Burns co-wrote Chapters 3 and 10.
Jason Callahan co-wrote Chapters 3 and 10, edited Chapters 6, 7, and
11, co-created the table of contents and index, clarified throughout the
book which proofs would be included in the manuscript and which would
be seen in more advanced courses, standardized numbering of results
throughout, and contributed the book’s title.
Jorge Calvo co-wrote Chapters 2 and 9, edited Chapter 5, created some
of the most difficult figures, and formatted and standardized word usage
in Chapters 1–7.
Marion Moore Campisi co-wrote Chapters 4 and 11, and added and
edited figures throughout.
David Clark co-wrote Chapters 2, 9, 10, and 14, revised and added exer-
cises to Chapters 9, 10, and 11, edited solutions for exercises in Chapters 4
and 14, added and edited figures throughout, ensured all figures complied
with AMS graphics guidelines, helped develop ideas for the graphic on the
cover, and compiled this list of contributions.
Vesta Coufal co-wrote Chapters 5 and 13, edited Chapters 1 and 10,
added exercises to Chapters 6 and 7, checked lengths of sections, cre-
ated common preambles for the chapters, co-created the table of contents
and index, created a driver file, and worked out the technical details of
formatting the final manuscript.
Elizabeth Denne co-wrote Chapters 6, 7, and 12, edited Chapters 6, 7,
12, and 14, edited solutions for exercises in Chapters 6, 7, and 12, added
exercises to Chapter 9, added references and URLs, and proofread the
entire manuscript twice.
Berit Givens co-wrote Chapters 5 and 13, edited Chapters 1 and 3, and
edited solutions for exercises in Chapters 6, 7, and 13.
McKenzie Lamb co-wrote Chapters 4 and 11, and edited solutions for
exercises in Chapter 8.
ACKNOWLEDGMENTS xvii
Emille Davie Lawrence co-wrote Chapters 1 and 8, edited the pref-

ace, introduction, and Chapters 5–15, added exercises to Chapter 3, and
designed the graphic for the cover.
Lew Ludwig co-wrote Chapters 6, 7, and 12, edited Chapter 9, formatted
the manuscript as a book, and compiled a list of URLs used in the book.
Cornelia Van Cott co-wrote Chapters 1 and 8, edited Chapters 10 and
11, and edited solutions for exercises in Chapters 9, 10, and 11.
Leonard Van Wyk co-wrote Chapters 6, 7, and 12.
Robin Wilson co-wrote Chapters 4, 11, and 14, and created figures for
the solutions.
Helen Wong co-wrote Chapters 5, 8, and 13, edited Chapters 1, 3, 7, 8,
14, and 15. Helen has also been my unofficial sounding board and helper
for so many aspects of this book that I cannot list them all.
Andrea Young co-wrote Chapters 1 and 8, edited Chapter 4, and edited
solutions for exercises in Chapters 1 and 8.

In addition to the work of the UFP participants, several undergraduate and
graduate students contributed to the development of the manuscript and solutions.
I list them below together with their affiliation at the time when they worked on
the project.
Bryan Brown (undergraduate at Pomona College) proofread a draft of
the entire manuscript and the solutions.

Dwayne Chambers (PhD student at Claremont Graduate University)
provided technical support during PCMI, TEXed solutions and created

figures for the solutions during PCMI.
Gabriella Heller (undergraduate at Pomona College) wrote the first
version of Chapter 15 and created the protein drawing using the program
Pymol.
Indra Elizabeth Kumar (undergraduate at Scripps College) created
and edited figures throughout the manuscript. In particular, the pictures

of A. 3D-Girl were designed and created by her.
Becky Patrais (PhD student at University of Minnesota) proofread and
edited solutions to exercises.

Benjamin Russell (undergraduate at Carleton College) wrote and edited
some solutions to exercises in Chapters 8 and 15.

I also want to take this opportunity to thank Jeff Weeks and Colin Adams.
Much of the manuscript was inspired by their outstanding books. In addition, I
am grateful to Jeff Weeks for encouraging me to pursue this project. When I was
afraid this manuscript would be too similar to his book The Shape of Space, he
insisted that what is important is bringing this material to more students, not who
was first to do so. I am also indebted to Colin Adams for his suggestions along the
way, and for helpful comments on the figures throughout the manuscript. I hope
this book is anywhere near as clear and inspiring as the many wonderful books that
Colin has written.
Finally, I want to thank Grace Hibbard for generously sharing her mother Helen
Wong with me for the last year of this project. Last, but certainly not least, I want
to thank Francis Bonahon and Laure Flapan, whose love and support sustained me
throughout this project and always.
Erica Flapan
Part 1
Universes
https://doi.org/10.1090//mbk/096/01
CHAPTER 1
An Introduction to the Shape of the Universe
Topics:
• 2-dimensional universes
• Straightest paths in different spaces
• Gluing diagrams
• The flat torus and the curved torus
• Games on a flat cylinder and a flat torus
• The extended diagram for a flat torus
• The 3-torus
• The 3-sphere
1. To Infinity and Beyond
People normally assume that our universe goes off infinitely in all directions,
but the true shape and size of our universe is not known. It is hard to believe
that we really don’t know something as basic as the shape of our universe, and it’s
hard to imagine how the universe could not go on forever in all directions. If the
universe were finite, would that mean that the universe has an end that you can’t
go beyond?
Let’s imagine the following scenario, featuring an astronaut whose name is
A. 3D-Girl. Like her name, she is extraordinary and super-duper in every way.
She’s smart, funny, charming, adventurous, brave, and anything else you’d want
in an interstellar explorer. She’s been sent off deep into space to investigate a
potential colonization site that was identified by astronomers through super-duper
technologically advanced telescopes. The astronomers were pretty excited when
they saw this planet, because it seemed to have many of the same characteristics
as Earth. They even named it “Earth 2”. A. 3D-Girl’s spaceship, aptly named
A. Spaceship, has a super-duper hyper-warp drive that allows it to zoom off in
any direction, going as fast as she wants and as far as she wants. After months
of planning and training for the mission, A. 3D-Girl blasts off in her spaceship,
traveling in a straight line past Mars, Jupiter, Saturn, Uranus, Neptune, and Pluto,
beyond Alpha Centauri, in fact beyond all known stars and planets, until Earth 2
finally comes into view.
As the spaceship gets close to Earth 2, A. 3D-Girl notices more similarities
with our earth. Earth 2 is a blue and green planet with a single moon. It’s also
the third planet from its sun. After landing, she cautiously descends from her
spaceship and looks around. To her great relief, the place feels comfortable and
3
4 1. AN INTRODUCTION TO THE SHAPE OF THE UNIVERSE
Figure 1. A. 3D-Girl waving to the crowd as her spaceship blasts off.
familiar. After looking around a bit more, she does a double take. This place is
too familiar. There’s even a doughnut shop that looks exactly like the one that
A. 3D-Girl stopped at before she boarded her spaceship. She checks and rechecks
her flight log. Nope, she didn’t make a mistake in navigation, and didn’t make
any U-turns. In fact, she didn’t make any turns at all in getting here from Earth.
She set A. Spaceship to full throttle and went along the straightest, shortest path
from Earth to the exact coordinates where the astronomers had told her to go.
As crowds of people gather around her, A. 3D-Girl boldly remarks, “Huh? What
happened?”
Although it is hard to imagine ever developing a spaceship as technologically
advanced as A. 3D-Girl’s, this story is not completely unbelievable. Many elemen-
tary schools teach children that people in the middle ages thought the earth was
flat, and it was not until 1492 when Christopher Columbus sailed the ocean blue
that people found out that the earth is round. In fact, this is a myth that was put
forth in 17th-century works of fiction including Washington Irving’s fictionalized
biography The Life and Voyages of Christopher Columbus. But what is noteworthy
about this myth is that to this day people accept it as completely reasonable. We
can imagine ourselves in the middle ages, looking out in all directions and conclud-
ing that either the Earth must have an edge way off in the distance or it must be
a plane that goes on forever.
The story about A. 3D-Girl’s voyage isn’t that different from the myth that
Christopher Columbus discovered that the earth is round. When we look out into
space in all directions, we can’t imagine how space could wrap around and come
back to where it started. So we assume that either the universe must end way off
in the distance or it must go on forever. But just because we don’t know how to
visualize a universe that closes back on itself like the Earth, doesn’t mean it isn’t
possible.
Though the actual shape of the universe is not known, in this and the next
seven chapters, we’ll develop our intuition about some of the possibilities for the
shape of our universe. We’ll learn how to visualize universes of different dimensions,
and how we might distinguish one such universe from another. While you may end
2. A. SQUARE AND HIS UNIVERSE 5
up with more questions than you started with about our own universe, we hope
you’ll find the voyage intriguing.
2. A. Square and His Universe
Sometimes, when you’re faced with a hard math problem, it’s a good idea to
start by solving an easier version of the same problem. So instead of thinking
about the shape of our universe, we begin one dimension down, by considering
how 2-dimensional creatures living in a 2-dimensional universe might think about
possible shapes for their universe.
You might have heard of the short novel titled Flatland: A Romance of Many
Dimensions which was written in 1884 by the English school teacher Edwin A. Ab-
bott. The book was meant to be a commentary on Victorian society, but it can
give us some insight into the world of 2 dimensions. Whether or not you’ve read
Flatland does not matter here (though if you would like to read it, you can down-
load a copy from the internet for free). We will simply introduce you to the main
character, whose name is A. Square. His portrait is given in Figure 2.
Figure 2. A portrait of A. Square.
A. Square lives in a 2-dimensional universe. He and his space have no thickness,

like a shadow or an image projected on a screen. In the book Flatland, A. Square’s
universe is a plane. However, there are other possibilities for his 2-dimensional
universe. For example, his universe could be a sphere, a torus (that is the surface
of a doughnut), a 2-holed torus, a 3-holed torus, etc. (see Figure 3). Continuing in
this way, we find that there are infinitely many 2-dimensional universes.
Figure 3. A. Square on the surfaces of a sphere, a torus, and a

2-holed torus.
While you might think of these surfaces as 3-dimensional objects in our space,
keep in mind that we are only interested in the surfaces of the doughnuts (or the
doughnut hole when he’s on a sphere) and not the cake inside. This means that
a 2-dimensional being like A. Square could live in one of these universes, but a
3-dimensional person could not. Furthermore, if A. Square lived in one of these
universes, he could only move back and forth and up and down, staying in the icing
of the doughnut with no way to get to (or even known about) the cake underneath.
We say that A. Square’s universe is 2-dimensional because he can only move in

two perpendicular directions. Back and forth is one direction, and up and down
is another direction. Of course he can combine these two directions to go in any
diagonal direction he wants.
In general, the number of dimensions of a space is the number of perpendicular
directions in which creatures in the space can move. A line and a circle each have
only 1 dimension because a creature living inside one of these spaces can only move
backward and forward. This is true no matter how the line or circle is situated
in our space. In particular, even though we need 3 dimensions to tie a knot, if
we consider a knotted circle as a space by itself, it would only be 1-dimensional.
Figure 4 illustrates A. Dot who lives in a circle and her cousin A. Dash who lives in
a knotted circle. Each of the cousins is only able to move backwards and forwards
in her space. They seem happy, but imagine what boring lives they must have.
They talk on the phone on a daily basis to report how many laps around their
space they run. However, they have no idea that their spaces are different because
the difference can only been seen with our 3-dimensional perspective.
Figure 4. A. Dot lives in a circle and A. Dash lives in a knotted circle.
Notice that even though the surfaces in Figure 3 have no boundaries, they all
have finite area. Figure 5 illustrates some 2-dimensional universes with infinite
area. Since there is only a finite amount of space on a page, we use dotted lines
to mean that the spaces go on forever. The illustration on the left is an infinite
cylinder. The illustration next to it has two planes joined with a tube and a handle
on the lower plane. The illustration on the bottom has a tube and two such handles.
Keep in mind that just like the surfaces in Figure 3, these universes consist only
of surfaces. In particular, A. Square and his friends can’t get to the 3-dimensional
space inside of the tubes or handles.
Observe that we could create a 2-dimensional universe with any number of
planes joined together with any number of tubes, and each plane could have any
number of handles stuck to it. Pushing this idea even further, we could have
infinitely many planes joined with infinitely many tubes, and each plane could have
infinitely many handles. This gives us infinitely many 2-dimensional universes all
with infinite area.
Of course, A. Square can’t picture any of these spaces. He thinks there is only
one possible universe, and it’s a plane that goes on forever in all directions. We
denote such an infinite plane by R2 , since we can think of it as the coordinate space
with an x-axis and a y-axis. Similarly, we use R3 to denote the coordinate space
2. A. SQUARE AND HIS UNIVERSE 7
Figure 5. Some 2-dimensional universes with infinite area.
with x, y, and z axes. Knowing that there are infinitely many finite and infinite 2-
dimensional universes gives us the idea that perhaps our own 3-dimensional universe
could be something besides R3 . But we will have to wait until later in the chapter
to see some possibilities.
Maybe we should consider the disk illustrated in Figure 6 as a possible 2-
dimensional universe. The problem is that a disk has a boundary. If A. Square
reached the boundary, he would hit a wall and could go no further. He could not
fall off the edge of the universe, because the universe is all there is. He can’t step
outside his universe. He also couldn’t go around the edge of the disk and come back
on the bottom because, since the disk is only 2-dimensional, it doesn’t have a top
and a bottom. Imagine if there was a similar wall in our 3-dimensional universe.
Suppose we sent out a spaceship and it smashed into a wall at the edge of our
universe. This seems neither pleasant nor likely.
Figure 6. If A. Square lived in a disk, when he reached the edge

he could go no further.
As we think about possible universes for A. Square and for ourselves, we do

not want to consider the possibility that universes might have walls that you could
smash into. This just seems too implausible to be true. Thus, from now on whenever
we use the word universe, we will mean a space that has no boundary. Note that
this does not mean that we require every universe to be infinite. Keep in mind that
there are universes like those in Figure 3 which have finite area but no boundary.
By contrast, when we use the word space, we won’t make any assumptions about
whether or not the space has a boundary. For example, we say that a disk like the
one in Figure 6 is a 2-dimensional space but not a 2-dimensional universe.
Now that we know that there are so many possibilities for 2-dimensional uni-
verses, we want to help A. Square learn about his own universe. Ideally, he would
like to create a list of all possible 2-dimensional universes. Even though such a
list would be infinite, if it was set up in a systematic way, it would give him an
idea what the possibilities were. Even if his list isn’t complete, he would like to
determine some property of his universe that would enable him to distinguish it
from other spaces on the list or at least to reduce the number of possibilities. For
example, if he could determine that his universe had finite area, he could eliminate
all of the spaces with planes and tubes and handles.
You can think about properties of a space the way we think about character-
istics of people. For example, suppose you had a list of all of the students in your
math class, and you wanted to find out the name of the girl who sits next to you.
Knowing that she’s a girl might help you to reduce the number of possibilities, but
you still wouldn’t know which one was her. If the list included major and year of
graduation, and you were able to find out this information about her, you could
reduce the possibilities even further.
We’d like to do the same thing with properties of a universe. For example,
suppose that A. Square somehow determines that his universe is one of the infinite
surfaces illustrated in Figure 5. The number of planes, tubes, and handles that the
surface has is a property which would enable us to distinguish them. However, if
A. Square were living in one of these universes, he would not be able to see the
planes, tubes, and handles like we can. So we need to think of properties that
A. Square could use to distinguish among these universes. If we can succeed at
that, perhaps it will give us some insight about properties we could use to learn
about our own universe.
3. Straightest Paths in Flatland
One of the tools that A. Square can use in investigating his universe is the
notion of a “straight line” between two points. You’ve probably heard people say
“the shortest path between two points is a straight line”. This is true if the two
points are in a plane. However, if A. Square is on a sphere, what does “straight
line” even mean? There are no straight lines or even straight line segments on a
sphere. A path which appears to A. Square to be the shortest path between two
points will appear to us as a curve since every path on a sphere is a curve.
One way to think about what “straight” means to A. Square is to imagine that
he has a 2-dimensional flashlight that he is shining in front of him. We could picture
his flashlight like the lamp that’s attached to a miner’s helmet. A 2-dimensional
ray of light from his flashlight stays within his universe, illuminating the shortest
path to get to a nearby point. So if he follows the beam of his flashlight, his path
will appear to him to be a straight line. We will call the path of a light ray within
his universe a straightest path, even if it doesn’t actually follow a straight line in
our 3-dimensional universe.
3. STRAIGHTEST PATHS IN FLATLAND 9
Another approach that A. Square could take to finding the straightest path
between two points would be to attach a piece of string to one of the points in his
space and then go to the other point and pull the string tight. Then if he walks
along the string, he knows he will be going as straight as he can.
If A. Square lives in a plane, then the straightest path between two points is
a line segment. But if he lives in a sphere, then the straightest path between two
points is an arc on what’s known as a great circle. The great circles on a sphere
are circles whose radius is the same as the radius of the sphere. An equator is an
example of a great circle. Great circles can be thought of as the intersection of a
sphere with a plane that divides the sphere into two equal halves. Note that if the
plane does not divide the sphere exactly in half, then the circle of intersection is
not a great circle. Paths which follow great circles curve as little as possible while
staying within the surface of the sphere. Airplanes and ships follow great circles
around the earth because these are the shortest paths between two points on earth.
If you have looked at flight paths in airline magazines, you may have noticed that
the flight paths are drawn as curves rather than as straight lines. These flight paths
appear to be longer than necessary, but that’s because a flat map of our spherical
earth will necessarily be distorted.
Let’s consider how A. Square might use the idea of straightest paths to distin-
guish between a plane and a sphere. Suppose his universe is a plane, and he leaves
home to take a trip following a straightest path (making use of his headlamp to go
straight). Much to the disappointment of his family, if he continues following this
straightest path, he will never return home. However, if A. Square lived on a sphere
and left on a similar voyage following a straightest path, after some time he would
happily return to his awaiting loved ones (see Figure 7). At this point he would
know that his universe could not be a plane. It’s important in this experiment that
A. Square use a taut string or a light ray to follow a straightest path. Otherwise,
he could just walk in a circle on the plane and return to his starting point.
Figure 7. Straightest paths on a plane and a sphere.
In order to see if a similar method would enable A. Square to differentiate

between a sphere and a torus, consider the path illustrated in Figure 8. A string
could be pulled taut along this path so that A. Square would know that it’s indeed
a straightest path. If A. Square travels along this path, he will eventually return
to his starting point as he did on the sphere. Thus finding a straightest path that
returns to its starting point will enable A. Square to distinguish between a torus
and a plane, but will not give him enough information to distinguish between a
torus and a sphere.
In search of some other property that a torus has and a sphere doesn’t have,
A. Square tries to find a different type of straightest path on the torus. The problem
Figure 8. A straightest path on a torus.
is that finding such paths on the torus is not as straightforward as it seems. For
example, take a look at the grey circle on the left of Figure 9. At first glance, we
might think that it is a straightest path. It’s nice and round and rests just at the
top of the surface. However, if A. Square marks points p and q on the grey circle
and pulls a string tight between these points, the string will not lie on the grey
path. The black string between points p and q on the left in Figure 9 is just a bit
shorter than the grey path going between the points. So the grey path on the left
of Figure 9 isn’t a straightest path on the torus after all. While it’s hard to identify
a straightest path on the top of a torus, an inner circle going around the hole of the
torus is a straightest path because the circle cannot be made any shorter. Such a
shortest path is illustrated on the right of Figure 9
Figure 9. The grey circle on the left is not a straightest path, but
the grey circle on the right is.
When looking at the illustration on the left of Figure 9, it’s important to keep
in mind that to us the black string is curved, though not as curved as the grey
string. A truly straight segment would cut through the inside of the torus, but
A. Square’s string could not leave the surface of the torus to follow such a path.
Now let’s imagine that A. Square lives on a torus and has two balls of string,
one black and the other one grey. He pins down the end of the black string, then
follows the straightest path illustrated on the left in Figure 10 pulling the string
taut as he goes. After returning to his starting point, he ties the two ends of the
black string together. Then he repeats this process with the ball of grey string
going around the inner circle of the torus, again as illustrated on the left in Figure
10. He notices that the black and grey circles are both straightest paths, and they
intersect at precisely one point (indicated by a dark grey dot).
On the other hand, if A. Square lives on a sphere, any two straightest paths
will intersect in two points as illustrated by the black and grey circles on the right
4. EXPLORING THE SHAPE OF A CAVE 11
Figure 10. Two straightest paths can meet in exactly one point
on a torus, but not on a sphere.
in Figure 10. Thus this experiment with colored string will allow him to distinguish
between a torus and a sphere.
Could A. Square use additional balls of colored string to distinguish between
tori with different numbers of holes? He would have to begin by finding some
straightest paths on a torus with two holes. You may want to experiment with this
yourself. Get a bagel, a giant pretzel, some pins, and a few skeins of colored yarn,
and use them to try to distinguish the bagel from the pretzel. (How many holes
does a pretzel have?) Afterwards, when you eat the bagel and pretzel make sure
you first remove the yarn and the pins.
4. Exploring the Shape of a Cave
Having watched A. Square use straightest paths to compare different universes,

we would like to use similar ideas to compare 3-dimensional universes. But before
considering such a hard problem, let’s check on A. 3D-Girl, who has been suffering
from nightmares and insomnia ever since her confusing voyage to Earth 2.
During a restless night of sleep, A. 3D-Girl thinks she’s woken up in a deep,
dim, dank cave. Feeling her way along the wall, she discovers that the cave has
no exit and no entrance to any other passageway. While exploring the entire cave,
she starts thinking about a book she read a long time ago about a square trying
to figure out the shape of his universe, and suddenly a ball of grey string appears
in her hand. She places one end of the string under a big rock so it can’t move,
and trails the string behind her as she walks. Eventually she returns to the big
rock where she started without ever retracing her steps. From this exploration she
concludes that the cave is like the inside of a doughnut. The cave has some nooks
and crannies as all caves do, but the overall shape is as illustrated in Figure 11.
Depressed at the thought of a life walking around in circles trailing grey string, she
decides to go back to sleep hoping to wake up in her own comfortable bed.
Unfortunately, the next thing she knows she’s in a different cave, again with
no exit and again with a ball of grey string in her hand. After exploring for a while
in the dark, she discovers that this cave has a single four-way intersection. She
winds the string around both circular passageways making it into a figure-eight.
She pictures this cave as illustrated in Figure 12.
She becomes agitated as she dreams that she is waking up in more and more
complicated caves. What if the cave had two three-way intersections? The cave
could look like two circular loops connected by a passageway (as illustrated on the
left of Figure 13), or it could be one circular loop with a passageway cutting across
Figure 11. A cave which looks like the inside of a doughnut.
Figure 12. A cave with one four-way intersection.
Figure 13. These caves each have two three-way intersections.
the middle (as illustrated on the right of Figure 13). How would she know which
one it was? Could she use a ball of string to help figure it out?
After some thought, A. 3D-Girl decides that she could distinguish these two
caves by seeing if there is a circular path from one three-way intersection back to
itself that doesn’t pass near the other three-way intersection, as there would be in
the cave on the left in Figure 14. There would be no such path if she were in the
cave on the right.
But what if she were in a cave that had many different types of intersections,
and included some branches which seemed to go on forever and others which looped
back on themselves? The thought of listing all the possibilities and coming up
with a plan to distinguish them makes her head spin so much that she passes out.
Luckily, the next time she wakes up she is actually back in her own bed. But her
night adventures leave her wondering what could help her answer these sorts of
questions. Perhaps if she checks back with A. Square, she can get some ideas from
him.
5. CREATING UNIVERSES BY GLUING 13
Figure 14. On the left there’s a path from a three-way intersec-

tion back to itself that doesn’t go near the other three-way inter-
section.
5. Creating Universes by Gluing
A. Square was able to use the notion of straightest paths to distinguish between
a plane (denoted by R2 ), a sphere (denoted by S 2 ), and a torus (denoted by T 2 ).
Note that we use the exponent 2 whenever we want to indicate that a space is 2-
dimensional. This is in contrast with ordinary 3-dimensional space which is denoted
by R3 . The terms “ball” and “sphere”, and “disk” and “circle” are often used inter-
changeably in normal life. But when talking about spaces of different dimensions,
we want to use these terms more precisely. A sphere S 2 is the 2-dimensional surface
that is the boundary of a 3-dimensional ball B 3 . A circle S 1 is the 1-dimensional
boundary of a 2-dimensional disk D2 .
We will now introduce the idea of gluings and instant transport to give A. Square
a way to visualize some of the possibilities for his universe. Later, we will use sim-
ilar techniques to enable us to visualize some analogous 3-dimensional universes.
We begin by trying to help A. Square visualize a sphere. Drawing the usual picture
of a sphere (like the one on the right side of Figure 10) won’t help A. Square at all,
since he doesn’t have the 3-dimensional perspective necessary to understand that
the dotted arcs are behind the rest of the picture. Instead we want to break up the
sphere into pieces that A. Square can understand.
As 3-dimensional creatures we know that a sphere can be obtained by gluing
two hemispheres together along their boundaries. To draw a picture of two hemi-
spheres being glued together, we put arrows on the boundaries of each hemisphere
to indicate how they should be matched up (as illustrated in Figure 15). In partic-
ular, the head of the arrow on one hemisphere is glued to the head of the arrow on
the other hemisphere. A diagram which uses arrows to indicate how the boundaries
of shapes are glued together is called a gluing diagram.
We can think of each hemisphere as being a disk that has been curved, like
a disk made of play dough that you mold around half of a ball or even around
your fist. Of course, A. Square cannot visualize disks being curved around a 3-
dimensional object, because that would require him to have 3-dimensional vision.
However, he can draw two flat disks with arrows on their boundaries. In order
to help him understand that the boundaries of the two disks are glued together,
we explain to him that when he arrives at the boundary of one disk, he will be
instantly transported to the boundary of the other disk (see Figure 16). We show
him the gluing diagram in Figure 16 to help him understand. A. Square reads a lot
Figure 15. A sphere consists of two hemispheres that are glued

together along their boundaries.
of science fiction so he accepts the idea that instant transport can take him from
one disk to the other.
Figure 16. When A. Square arrives at the boundary of one disk,

he is instantly transported to the boundary of the other.
Another way to think about gluing diagrams is as pages of a road atlas. When
you drive off the right side of the map on one page of an atlas, there is a number
or a letter to indicate that you drive onto the left side of a map on another page.
If there were a road that went all the way around the earth, you could follow the
pages of the atlas going from the map on one page to a map on another page and
another and another, until eventually you would come back to the page where you
started.
Now that A. Square can visualize a sphere as two disks glued together, we
would like to use a gluing diagram to help him visualize a torus. Let’s consider
a 2-dimensional universe consisting of a square where opposite sides of the square
are glued together. We put arrows on the sides of the square to indicate how they
are attached. The head of the single arrow is glued to the head of the other single
arrow, and the head of the double arrow is glued to the head of the other double
arrow, as illustrated on the left side of Figure 17.
Figure 17. Gluing up opposite sides of a square.

6. GAMES ON A GLUING DIAGRAM FOR A CYLINDER AND A TORUS 15
As 3-dimensional people, we can physically glue up opposite sides of the square

(if the square is made of a sufficiently flexible material). After gluing together one
pair of sides we get a cylinder. Then if we glue together the double arrows on the
ends of the cylinder, we get a torus as illustrated on the right side of Figure 17. You
might want to check for yourself that it doesn’t matter which pair of sides we glue
together first—we’ll still end up with a torus after both pairs of sides are glued.
A. Square can’t visualize gluing up the sides of the square. So we have to
explain to A. Square that if he goes through the top edge of the square, he is
instantly transported to the bottom edge, and if he goes through one side, he is
instantly transported to the other side (see Figure 18). It’s just like a video game
where a character who goes off the screen on one side reappears on the opposite
side. When A. Square is not reading science fiction, he likes to play video games.
So this explanation also makes sense to him.
Figure 18. When A. Square exits from the right edge, he returns
on the left edge.
If we want to physically glue up opposite sides of a square, the square needs to

be made out of a flexible material. In particular, if the square is made of paper, we
can only glue together one pair of opposite sides at a time (you should try this to
see for yourself what happens). In fact, there is no way to make a nice round torus
out of paper. For this reason, it is more convenient for us to think of the torus the
way A. Square does, as an abstractly glued up square. We use the word “abstractly”
here to mean that we’re thinking about it in terms of instant transport rather than
as a square with sides that are physically glued together. When we think of a torus
this way, we call it a flat torus; whereas when we imagine a torus as the surface of
a doughnut or a bagel, we call it a curved torus.
A flat torus is a “better” universe than a curved torus because a flat torus is
the same everywhere. For example, if A. Square watches his friend B. Triangle
move around on a flat torus, she will always have the same area even if she goes out
one edge and comes in another (as illustrated on the left side of Figure 19). This
is not the case if she moves around on a curved torus. For example, on the right
side of Figure 19 we see that B. Triangle might get a stomach ache if she eats a big
dinner at a restaurant near the outer circle of a torus and then walks towards the
inner circle. It wouldn’t be very pleasant to live in a place where you would get a
stomach ache every time you walked in a certain neighborhood.
6. Games on a Gluing Diagram for a Cylinder and a Torus
One way to check your understanding of abstract gluings is to play games on a

gluing diagram. For example, you can play Connect Four on a gluing diagram for
a cylinder (see Figure 20). The normal game of Connect Four is like tic-tac-toe but
Figure 19. B. Triangle changes her shape when she moves around
on a curved torus.
with gravity. One player has black tokens and the other player has grey tokens.
The players take turns dropping their tokens into a vertical plastic grid which has
seven columns and six rows. As you drop in a token, it falls down due to gravity.
The game ends when one player has four in a row either horizontally, vertically, or
diagonally.
Figure 20. The game of Connect Four on a cylinder.
While normal Connect Four is played in a vertical rectangular grid, we can also
play Connect Four on such a grid where the pair of vertical sides have been glued
together as indicated by the arrows in Figure 20. This gives us a cylindrical grid.
There still is gravity which causes the tokens to drop down to the bottom of the
cylindrical grid as illustrated in Figure 20. As with normal Connect Four, you win
if you are the first player to get four in a row horizontally, vertically, or diagonally.
For example, consider the game in progress in Figure 20. It’s Grey’s turn, and Grey
can win in one move. Can you see how?
You should find a partner and play Connect Four a few times on the gluing
diagram for a cylinder. In 1988, Victor Allis “solved” Connect Four in the sense
that he determined a strategy wherein the first player can always win. How might
your strategy for Connect Four on a cylinder be different from your strategy in a
normal Connect Four game?
We can also play games on a flat torus. In Figure 21, we see a game of tic-tac-
toe on a torus where X has won. This game would be much harder to draw (and
to think about) on a curved torus. You should stop reading now to play a game of
torus tic-tac-toe with a friend.
7. EXTENDED DIAGRAMS 17
Figure 21. X has won this game of torus tic-tac-toe.
There are many other games that you can play on a flat torus. Could you play
Connect Four on a flat torus? You should check out the website
http://www.geometrygames.org/TorusGames/
where you can play tic-tac-toe, mazes, crossword puzzles, word search puzzles,
jigsaw puzzles, chess, pool, and gomoku all on a flat torus.
7. Extended Diagrams
One way to plan your strategy when playing games on a gluing diagram is
to draw an extended diagram to help you see where you go when you go off one
side of the board. In order to construct an extended diagram for the flat torus,
let’s first imagine that A. Square lives all by himself in a flat torus. To keep track
of his experience, let’s imagine that curtains are hung at the edges of the gluing
diagram. This gives him the impression that his neighborhood is a square, but he
can walk through the curtains. He puts on his hat to go out for a walk. When he
walks through the curtains on the right side of the gluing diagram he is instantly
transported to the left side. However, he thinks he has entered a second square
neighborhood where there is another guy with a hat just like his, who is also taking
a walk at the same speed that he is, and that guy is going through the curtain on
the right just as A. Square enters through the curtain on the left (see Figure 22).
In fact, the square that he sees taking a walk ahead of him is really just A. Square
himself.
Figure 22. A. Square sees a copy of himself up ahead.
A. Square continues his walk in the same direction, believing he is leaving the
second square and entering a third square, and so on. Continuing in this fashion,
he believes that there are infinitely many squares in front of him, one after another.
Similarly, he believes that there are infinitely many squares on top of him stacked
one above the other, and infinitely many squares going off on a diagonal. Altogether,
Figure 23. A portion of an extended diagram of a flat torus with

A. Square in it.
he imagines there is an entire plane full of squares glued together left to right and
top to bottom, containing other creatures who look just like him, wearing hats like
his, and moving at exactly the same rate as he is.
The space he imagines he lives in is called the extended diagram of the flat torus.
We see a portion of the extended diagram of the flat torus in Figure 23. Keep in
mind that all of the square gluing diagrams in the extended diagram are really just
a single flat torus, and all of the guys with hats on are really just A. Square himself.
Figure 24. The three black X’s in an extended diagram represent

three in a row in flat torus tic-tac-toe.
8. INTRODUCING THE 3-DIMENSIONAL TORUS AND SPHERE 19
It’s a bit like when you are in a dressing room with opposing mirrors and it seems
like there are infinitely many copies of you, though one is blocking another so you
can’t really see them all. You know that really you are the only you, and the copies
of you in the mirror are just your images.
If we draw an extended diagram of a flat torus tic-tac-toe board, then it is
easier to see all the possible ways you can get three in a row. For example, the row
of X’s that we saw in Figure 21 looks like an ordinary three in a row in the extended
diagram (see the three black X’s in Figure 24). In fact, there are infinitely many
copies of these three X’s in the extended diagram and we could choose any three
adjacent ones to represent the 3 in a row on the tic-tac-toe board from Figure 21.
You should try to see if an extended diagram can help you find a winning
move for grey in the game of Connect Four illustrated in Figure 20. Note that
the extended diagram for a cylinder will look like a strip of paper that goes on
indefinitely to the left and the right but not on the top and the bottom.
8. Introducing the 3-Dimensional Torus and Sphere
Now that we have some experience helping A. Square understand some possi-
bilities for his 2-dimensional universe, let’s consider some analogous possibilities for
our 3-dimensional universe. It is natural for us to assume that our universe goes off
infinitely in all directions, just as it was natural for A. Square to assume that his
universe was an infinite plane. We saw that while it is easy for us to think of many
different possibilities for A. Square’s universe, it is hard for him to visualize these
possibilities. We used the idea of gluing up disks and squares in order to help him
understand what it would be like if his universe were a sphere or a torus. This gives
us the idea that our own 3-dimensional universe might be an analogously glued up
3-dimensional shape.
We begin by exploring a 3-dimensional version of the flat torus. Suppose that
you live in a large cubical room with pairs of opposite walls glued together and
the floor and the ceiling glued together. We imagine this space “abstractly” just
as A. Square imagines a flat torus. We don’t assume that the room is flexible and
exists in a higher dimension where opposite walls curve around so that they can be
glued together. Rather, we imagine that a person is instantly transported from one
wall to the same point on the opposite wall. This 3-dimensional universe is called
a 3-torus or a 3-dimensional torus and is denoted by T 3 .
In order to help us visualize this, let’s imagine that A. 3D-Girl lives in a 3-torus.
If she floats through the ceiling, she comes back through the floor (see Figure 25).
This is similar to what we saw in Figure 18 when A. Square travelled through the
right edge of the square and reappeared on the left edge.
Even though we visualize the 3-torus as a glued up cube, there is no boundary
to this space. We never have to worry about walking “off the edge” or bumping into
a wall. Figure 26 illustrates a gluing diagram for a 3-torus where pairs of identical
arrows are used to indicate that opposite walls are glued together. Notice that the
head of an arrow is glued to the head of the similar arrow with the colored part
glued to the colored part and the white part glued to the white part. We use three
different types of arrows because there are three pairs of opposite walls that are
glued together.
Figure 25. If we glue up opposite faces of a cube, we get a 3-torus.
Figure 26. The gluing diagram of a 3-torus.
A very, very large version of the 3-torus could actually be the shape of our
universe. If this were the case and we had a spaceship with a super-duper hyper-
warp drive that allows it to zoom off as fast as we want and as far as we want, then
we could send it off in one direction and it would eventually come back to Earth
from the opposite direction. In reality, though, such a spaceship would probably
take billions of years to return to Earth. In fact, if our universe were a 3-torus and
we had a really powerful telescope, we could see a copy of our own Earth (though
the image we would see could be billions of years old since it would take the light
billions of years to traverse the entire universe to get back to us).
Another possibility for the shape of our universe is a 3-dimensional sphere (also
known as a 3-sphere), which is denoted by S 3 . It is very important to remember
that S 3 is NOT a solid 3-dimensional ball (which we denote by B 3 ). Rather S 3 is
the 3-dimensional analog of a 2-dimensional sphere S 2 . The ball B 3 cannot be a
universe because it has a boundary, and we are assuming that no universe can have
a boundary.
There are several ways to visualize S 3 , each of which is analogous to an approach
that A. Square could use to visualize S 2 . The first way to think about S 3 is to
formally define it in terms of coordinates using an equation. With this approach,
we would define
S 2 = {(x, y, z)|x2 + y 2 + z 2 = 1}
as a subset of R3 . Keep in mind that S 2 is 2-dimensional even though it’s a subset
of R3 . This definition of S 2 makes it easy for 3-dimensional creatures to picture
S 2 as the outside of a round ball with radius 1. By analogy, we could define S 3 in
8. INTRODUCING THE 3-DIMENSIONAL TORUS AND SPHERE 21
terms of coordinates as
S 3 = {(x, y, z, w)|x2 + y 2 + z 2 + w2 = 1}
as a subset of R4 . However, this doesn’t give us much intuition about how to
visualize S 3 since it’s hard enough to picture R4 . A. Square has the same problem
if he tries to picture S 2 = {(x, y, z)|x2 + y 2 + z 2 = 1} as a subset of R3 .
For a more visual approach, recall that we helped A. Square visualize a sphere
by telling him that a sphere is made of two disks which have been glued together
along their boundaries (see Figure 16). In order to construct S 3 in an analogous
way, we need to first find the 3-dimensional analogue of a disk. One way of defining
a disk is as the set of points in R2 whose distance from the origin is 1. We would
express this symbolically as
D2 = {(x, y)|x2 + y 2 ≤ 1}
as a subset of R2 . Thus the 3-dimensional analogue of a disk is the set of points in
R3 whose distance from the origin is 1. We express this symbolically as
B 3 = {(x, y, z)|x2 + y 2 + z 2 ≤ 1}
as a subset of R3 . Hence the analogue of a disk is just a solid ball. Now the 3-
dimensional analogue of constructing S 2 by gluing two disks together along their
boundaries is constructing S 3 by gluing two solid balls together along their bound-
aries. If A. 3D-Girl lives in a 3-sphere, then she can travel through the boundary
of one of the solid balls into the other solid ball as illustrated in Figure 27.
Figure 27. S 3 consists of two balls that are glued together along
their boundaries.
No matter how we imagine contorting or smashing the balls in Figure 27, it

is impossible for us to visualize the boundaries of the balls actually being glued
together. This is because we would need four dimensions in order to do the gluing.
This is the same problem that A. Square would have if he tried to visualize gluing
together the boundaries of the two disks in Figure 16. The best way for us to
imagine the gluing is to think about it point by point, so that each point on the
boundary of one ball is glued to the corresponding point on the boundary of the
other ball. The gluing arrows on the spheres can help us remember how one ball is
glued to the other.
Now we know that in addition to R3 , our universe could be a 3-torus T 3 or
a 3-sphere S 3 . As soon as we discover new possibilities for our universe, we want
to know how we could determine whether our universe has one of these forms. If
we could somehow map out every cubic inch of our universe, we would know that
our universe was finite. So we could conclude that it wasn’t R3 . But how could we
distinguish between T 3 and S 3 ? As usual, it’s a good idea to start by answering
the analogous question for 2-dimensional universes before we attempt to answer it
for our own universe.
9. Distinguishing a 2-Sphere from a 2-Torus
In Figure 10, we saw that A. Square could figure out whether he was living in
S 2 or T 2 by considering different straightest paths. In particular, he observed that
two straightest paths can meet in exactly one point on T 2 , but not on S 2 .
Another way to distinguish between S 2 and T 2 is to consider what would
happen if a wall were built along a straightest path. For example, on the left side of
Figure 28 we see that A. Square is in the northern hemisphere of S 2 and B. Triangle
is in the southern hemisphere. The Great Circular Wall which was built along the
equator divides S 2 into two pieces preventing A. Square from visiting B. Triangle.
On the other hand, as we see on the right side of Figure 28, a wall along a straightest
path on T 2 might make it inconvenient for A. Square to visit B. Triangle, but if
he’s willing to walk around the long way he will eventually get there.
Of course, walls can be built on a torus which separate it into two regions as
we see in Figure 29. However, such a wall does not follow a straightest path.
Figure 28. A wall along a straightest path in S 2 divides the space

into two regions, but it doesn’t in T 2 .
Figure 29. This wall is not along a straightest path.

9. DISTINGUISHING A 2-SPHERE FROM A 2-TORUS 23
In order to explain to A. Square how he can use walls along straightest paths
to distinguish between a torus and a sphere, we need to use gluing diagrams rather
than the illustrations in Figure 28, since he can’t understand 3-dimensional illus-
trations. Also, it’s easier to draw straightest paths in the gluing diagram than on
the curved torus, since straightest paths in the gluing diagram just consist of line
segments. To show him a wall along a straightest path in T 2 , we draw Figure 30.
It’s then easy for him to check that the wall doesn’t divide T 2 into two separate
regions.
Figure 30. This wall doesn’t separate T 2 into two regions.
Next we want to show A. Square that any wall along a straightest path in
S 2 will separate it into two regions. The problem is it’s not so easy to draw a
straightest path in the gluing diagram for S 2 , because the gluing diagram for S 2
flattens the hemispheres into disks which distorts distances. This means that if
we draw a line segment in one of the disks, it won’t necessarily correspond to a
great circle on S 2 . Instead, let’s start out with a great circle which goes between
A. Square and B. Triangle (illustrated as a thick black circle in the left image of
Figure 31). This great circle will be our wall. To make it easier to draw this circle
in the gluing diagram, we divide the sphere into two hemispheres so that half of the
black circle is in each hemisphere, as illustrated in the second image of Figure 31.
Then we flatten out the hemispheres into disks so that the black semi-circles are
now diameters of the disks, as illustrated in the right image of Figure 31. We only
show A. Square the picture of two disks on the right, since the other pictures in the
figure wouldn’t make sense to him anyway. We then explain to A. Square that even
if he goes from one disk to the other (as indicated by the dotted lines and arrows)
he will still be on the opposite side of the wall from B. Triangle.
Figure 31. This wall separates S 2 into two regions.

Figure 32. The grey circle will become a diameter in each disk if
we cut along this great circle.
You may be concerned that we have only shown A. Square that this particular
wall separates a sphere into two regions. But given any great circle, we can cut
the sphere into hemispheres so that half of the great circle is in each hemisphere,
and hence each half of a great circle becomes a diameter of one of the disks. For
example, if we want to build a wall along the great circle which is grey in Figure 32,
we can cut the sphere into hemispheres along the black circle, giving us the same
picture of disks that we had in Figure 31. So we have in fact shown that any wall
along a straightest path on the sphere separates it into two regions.
10. Distinguishing a 3-Sphere from a 3-Torus
Now let’s see if we can use walls along straightest paths to distinguish between
T 3 and S 3 . Since we are inhabitants of our space, we have to use gluing diagrams,
just as we did for A. Square. To build the wall, we lay a row of bricks along
a straightest path, expanding the wall by laying more bricks up, down, left, and
right, just as you would if you were building a brick wall to keep deer out of your
vegetable garden. Recall that for the walls we built in T 2 and S 2 , one side of
the wall met the other side so the wall was circular. Since we now have one more
dimension, not only do the sides of the wall meet, but the top and bottom of the
wall also meet.
Figure 33 illustrates such a wall in T 3 . We didn’t put the arrows on opposite
faces of the cube, but you should remember that opposite sides of the cube are
glued together. This means that the top of the wall is indeed glued to the bottom
of the wall and the left side of the wall is glued to the right side of the wall. If
A. 3D-Girl walks with her back to the wall, she can go visit her friend on the other
side of the wall. Thus this wall does not separate T 3 into two regions, just as the
wall in Figure 30 didn’t separate T 2 into two regions.
We have to be a little more careful with the gluing diagram for S 3 , since the two
balls have been “flattened” to fit in three dimensions just as the two hemispheres
of S 2 had to be flattened to fit in two dimensions. Recall that we saw in Figures 31
and 32 that given any wall along a straightest path in S 2 , we can divide S 2 into
two disks so that the wall cuts each disk exactly in half. By analogy we now see
that given any wall along a straightest path in S 3 , we can divide S 3 into two balls
so that the wall splits each ball exactly in half. In Figure 34 we illustrate a wall
along a straightest path in S 3 which splits each ball exactly in half. As we can see
in the figure, even if A. 3D-Girl walks away from the wall into the next ball, she
10. DISTINGUISHING A 3-SPHERE FROM A 3-TORUS 25
Figure 33. A. 3D-Girl is going to visit her friend on the other

side of this wall in T 3 .
Figure 34. No matter where A. 3D-Girl goes, she can’t visit her
friend on the other side of the wall.
will remain on the same side of the wall. In fact, no matter where she goes in S 3 ,
her friend will always be on the other side of the wall.
Thinking about 3-dimensional universes like T 3 and S 3 reminds A. 3D-Girl

about her nightmare with the caves, and she starts to worry that she’ll have the
same nightmare when she goes to sleep later. She has a momentary panic about
losing her ball of string, and forces herself to take a few deep breaths. Once she’s
a bit more calm, she starts thinking about whether she could build walls along
straightest paths to distinguish different caves.
Recall that in Section 4, A. 3D-Girl was faced with the problem of distinguishing
between the two caves in Figure 35. To solve this problem, she had figured out
that for the cave on the left she could lay out a loop of string from one three-way
intersection back to itself that didn’t go near the other three-way intersection. This
wouldn’t be possible for the cave on the right of Figure 35.
Now she realizes that she could also distinguish between the two caves in Fig-
ure 35 by building walls along straightest paths. In particular, for the cave on the
left in Figure 36, she could build a wall that separates it into two doughnut-shaped
regions which she couldn’t do for the cave on the right.
A. 3D-Girl reasons that even if she didn’t want to go to the trouble of building
a complete wall, she could use rocks to mark where the wall should go. Knowing
two different techniques to distinguish the caves reassures A. 3D-Girl enough that
Figure 35. On the left, there’s a path from a three-way intersec-

tion back to itself that doesn’t go near the other three-way inter-
section.
Figure 36. The wall on the left separates the cave into two
doughnut-shaped regions.
she can go to sleep, during which has the most wonderful, peaceful, blissful sleep
she’s had since her voyage to Earth 2.
11. Exercises
1. Suppose A. Square lives alone in a small flat torus and each of his sides is
a different color. Imagine that he can see in all directions at once. Draw
an extended diagram representing how A. Square might imagine his universe.
Explain your drawing.
2. Suppose A. Cube lives alone in a small 3-torus and each of her sides is a
different color. Imagine that she can see in all directions at once and has depth
perception. Draw an extended diagram of A. Cube’s universe. Explain your
drawing.
3. Suppose A. Square lives alone in a small sphere. What direction would he look
if he wanted to see the top of his head?
4. How could A. Square detect the difference between a plane, an infinite cylinder,
and an infinite cone? Is an infinite cone or an infinite cylinder a reasonable 2-
dimensional universe for A. Square to live in? Why or why not?
5. How do we create an infinite cylinder and an infinite cone by gluing up a planar
surface? What could be the 3-dimensional analogue of an infinite cylinder?
What about an infinite 3-dimensional cone?
11. EXERCISES 27
6. Suppose we tell A. Square that his universe is either a flat torus, a sphere, or
a plane. What information would help him to determine which it really is?
7. Suppose we were told that our universe was either a 3-torus or R3 . What
information would we look for to determine which it really was?
8. Draw a circular path on a torus which intersects both the thick black and the
thick grey circle in Figure 10 in just one point. Draw a second circular path
also with this property, but which can’t be deformed on the torus to the first
one you drew. Your circular paths do not have to be straightest paths.
9. Draw a circular path on the torus in Figure 10 which intersects the thick black
curve in two points and intersects the thick grey curve in three points.
10. In the Connect Four game on a cylinder illustrated in Figure 37, can either
player win in one move if that player goes next?
Figure 37. Illustration for Exercise 10.
11. In tic-tac-toe, we say two moves are equivalent if the strategy for the rest of
the game after one of the moves is analogous to the strategy for the rest of the
game after the other move. For example, in regular tic-tac-toe, a first move in
one corner is equivalent to a first move in any of the other three corners; and a
first move in the middle of one side is equivalent to a first move in the middle
of any side. Thus in regular tic-tac-toe, there are three inequivalent moves for
the first player: the center, a corner, or the middle of a side.
How many inequivalent first moves are there for tic-tac-toe on a cylinder
(see Figure 38).
12. In torus tic-tac-toe how many inequivalent moves are there for the first player?
13. In torus tic-tac-toe, how many inequivalent moves are there for the second
player?
14. Is either player guaranteed to win torus tic-tac-toe if both players use optimal
strategies? Justify your conclusion.
15. Is it possible to have a tie in torus tic-tac-toe if both players wanted to? Justify
your conclusion.
16. If you play torus chess, with the pieces set in the usual starting positions,
explain how the first player can win in just one move. Now, design a better
starting position that does not lead to either checkmate or a stalemate in just
one move.
17. The wall built along a straightest path in Figure 30 does not divide T 2 into two
regions. Suppose you built two walls in T 2 both along straightest paths. How
many regions would the walls divide T 2 into? (Hint: Your answer may depend
on whether the walls intersect or not.)
18. It is possible to build three intersecting walls along straightest paths in T 3
that still do not separate T 3 into different regions. Explain or draw a picture
showing how this can be done.
19. Explain how A. 3D-Girl can use walls to distinguish between a doughnut-shaped
cave (see Figure 11) and a cave with one four-way intersection (see Figure 12).
20. In Figure 32, we started with a wall along the thick black circle on the sphere,
then cut along a perpendicular equator to get the gluing diagram on the right.
What would the thick black circle look like in a gluing diagram if we cut along
an equator that was not perpendicular to it? Draw a sketch and explain.
CHAPTER 2
Visualizing Four Dimensions
Topics:
• How A. Square can visualize 3 dimensions
• Projections and perception
• Movies and movement
• Inductively defining n-dimensional cubes
• Inductively defining n-dimensional tetrahedra
1. Teaching A. Square About the Third Dimension
To help us think about 4-dimensional space, let’s begin by imagining how Flat-
landers might think about 3-dimensional space. A. Square is probably convinced
that 3-dimensional space can’t possibly exist. To prove to him that it does exist we
could play some 3-dimensional tricks on him. Suppose A. Square has a flower in his
garden. We could remove the flower from his 2-dimensional space just by pulling
it up into the third dimension, as we’ve done in Figure 1. To A. Square, the flower
would seem to instantly disappear into thin air, leaving him mystified. Similarly, if
a 4-dimensional creature pulled my desk out of 3-dimensional space, it would seem
to me that my desk had disappeared into thin air.
Figure 1. If we lifted A. Square’s flower into the third dimension,

it would seem to disappear.
If I were evil, I could lift A. Square’s heart right out of his body. But if I did
that, a 4-dimensional creature might remove my heart from my body. So let’s not
do that. On the other hand, if A. Square has a heart condition, we could perform
“closed heart surgery” to repair one of his heart valves without making any cuts in
his skin just by reaching into his chest from the third dimension.
We can even help A. Square perform non-invasive surgery on himself. In Fig-
ure 2, we see an unhappy A. Square suffering from appendicitis and in desperate
need of an appendectomy. Unfortunately, the nearest Flatland hospital is far out
of town. He’s got a hand, of course, but he wouldn’t want to cut himself with
29
30 2. VISUALIZING FOUR DIMENSIONS
a scalpel. Certainly, A. Square can’t reach into himself to remove his appendix
without cutting through his skin.
Figure 2. A. Square needs to have his appendix removed.
But we can help him by pulling his hand into the third dimension (something
he certainly couldn’t do by himself), as in Figure 3. With our help, he can then
grab and remove his own appendix, just like I could grab and remove his heart. If
another Flatlander were observing this appendectomy happening, what would she
see?
Figure 3. We could pull A. Square’s arm into the third dimension

and help him remove his own appendix.
We could play tricks on A. Square that don’t involve surgery. Suppose we

lifted A. Square up into the third dimension, turned him over, and put him back
in Flatland (see Figure 4). Could he tell what happened? What would the other
Flatlanders think?
Figure 4. A. Square has been turned over.
What would happen if a 4-dimensional creature turned me over? Would I know

what happened? Would my friends notice any changes in me? If a 4-dimensional
creature turned a clock over, what would it look like?
2. PROJECTIONS AND PERCEPTIONS 31
Of course playing tricks is not the best way to teach A. Square about the third
dimension. Instead, let’s help him imagine what some 3-dimensional objects might
look like.
Before we discuss objects in various dimensions, we want to clarify the differ-
ences between the words “universe”, “space”, and “object”. You may remember
from Chapter 1, that while a space may or may not have a boundary, a universe has
no boundary. For example, in Chapter 1 we saw that A. Square’s universe could
be a sphere, a torus, an infinite plane, or any other finite or infinite surface that
doesn’t have a boundary. On the other hand, a finite cylinder with two circular
boundaries is a 2-dimensional space but not a 2-dimensional universe. We use the
word object to refer to something which is contained in a space. My pencil and
paper and even my friends are objects in my space. In fact, I’m an object in my
own space. Figure 5 illustrates some objects that A. Square might find in his space.
Figure 5. A. Square, a flower, C. Circle, and H. Hexagon are all

objects in Flatland.
The dimension of an object can be smaller than the dimension of the space (for
example, a shadow which has no 3-dimensional thickness is an object in our space).
Similarly, a sphere with no 3-dimensional thickness, like a holograph, could be an
object in our space. Even though such a sphere curves around in 3 dimensions,
the sphere itself is still 2 dimensional because creatures in it can only move in two
perpendicular directions. We saw in Chapter 1 that a 2-dimensional sphere could be
the universe for Flatland. This means that Flatland could be an object inside of our
3-dimensional universe. Analogously perhaps our universe is a 3-dimensional object
with no 4-dimensional thickness living inside some 4-dimensional universe. In fact,
perhaps there is a 1-dimensional universe living inside a 2-dimensional universe
which lives inside our 3-dimensional universe which lives inside a 4-dimensional
universe, and so on with the dimensions getting bigger and bigger forever. We
could even imagine that all of these different dimensional universes live inside a
single infinite-dimensional universe. But even if this were true, we would have no
way of knowing.
2. Projections and Perceptions
When A. Square encounters another Flatlander, what does he actually see?

If every object appeared to him as a line segment, how could A. Square tell the
difference between B. Triangle (who he’s infatuated with) and any other shape out
for a walk? Perhaps his brain uses shadows to help him determine distances. Or
perhaps he even has a second eye which gives him a sense of depth and perspective,
just as our two eyes give us depth perception (see Figure 6).
Figure 6. How does A. Square perceive B. Triangle?
In fact, our visual abilities as 3-dimensional creatures are not so different from
those of Flatlanders. When viewing a 3-dimensional object, our eyes create a 2-
dimensional projection of the object, very much like a photograph. You probably
don’t think about it this way when you look around, but that’s because your brain
has lots of tricks to turn a 2-dimensional projection into 3-dimensional perception.
Without conscious thought, your brain synthesizes a great deal of information from
light, shadows, motion, and relative size. Also, your two eyes together give a sense
of depth and distance to an otherwise flat image. This is the difference between
what we see and what we perceive.
If our brains did not have the ability to perceive depth and distance, we would
find all of the 2-dimensional images that we see somewhat confusing. Walking on
a snowy, icy path is difficult no matter what. But if you want a real challenge, the
next time you encounter an icy path, try walking on it with one eye shut. With
only one eye you lose a lot of your depth perception, making it harder to anticipate
the little bumps and valleys in the ice, which in turn makes it harder to navigate.
Those of us who live in warm climates are sadly deprived of the opportunity to
try walking with one eye closed on an icy path. So let’s consider another example.
If we had no depth perception, one coil of a spring might look to us like one of the
projections in Figure 7. If we only had one of these projections and we had never
seen a spring before, it might be difficult to go from this flat image to knowing the
actual form of a spring.
Figure 7. Without any depth perception a coil of a spring might

look like one of these images.
In the image on the right, the spring seems to be intersecting itself. But our
depth perception allows us to understand that one part of the spring is actually
behind the other. When we are making a sketch of an object in 3 dimensions, we
often use a gap to indicate that one part of the object passes behind another part.
For example, we could redraw the image on the right of Figure 7 as in Figure 8. Of
course, there is no actual gap in the spring, but the drawing with the gap is easier
for our brains to interpret than either of the drawings in Figure 7.
3. MOVIES AND MOVEMENT 33
Figure 8. A gap in the drawing helps our brains realize that one
part of the spring passes behind the other.
3. Movies and Movement
A. Square has just had his first date with B. Triangle and is on his way home
humming softly to himself, while thinking peaceful, happy thoughts about what a
great day he had. Meanwhile, as we can see in Figure 9, A. Ball is lurking just
above him in the third dimension preparing to pay him a surprise visit.
Figure 9. A. Ball preparing to visit A. Square.
A. Ball pushes herself through Flatland from above so that A. Square sees only
slices of her. Figure 10 shows what A. Square might perceive is happening at each
second from t = 0 when she first touches Flatland through t = 4 when only a cross
section of the bow on her head is in Flatland. Notice that we’ve drawn t = 0 at the
bottom of the picture and t = 4 at the top of the picture. Drawing the figure from
bottom to top may seem backwards. However, we draw it this way so that if we put
all of the slices of A. Ball together, we will see a right-side-up 3-dimensional image
of her. Note that A. Ball is solid, and the disks that A. Square sees are actually
cross sections of her body. We draw them as ovals in order to make the picture
look more 3 dimensional.
As you can imagine, A. Square is more than a bit disturbed to notice a black
dot appear, grow into a larger and larger circle, then shrink into a smaller and
smaller circle, turn into two rounded triangles glued together at a dot, and then
disappear. This all seems quite spooky and mysterious to him.
In Figure 11, we’ve put together a sequence of frames (like the frames of an an-
imated movie) to illustrate what A. Square sees as A. Ball passes through Flatland.
As with Figure 10, the images start at the bottom and go upward so that if the
images are stacked one on top of the other we get a right-side-up picture of A. Ball.
Since we are now viewing the scene from above, the ovals that we drew in Figure 10
now appear as circles. Note that if we were really drawing all of the images that
Figure 10. A. Ball passing through Flatland.
A. Square would see, there would be many more frames. In this way, Flatlanders
can view a 3-dimensional object passing through Flatland as an animated movie of
gradually changing 2-dimensional shapes.
These examples show us that Flatlanders can get a sense of 3-dimensional ob-
jects by watching 2-dimensional slices of them pass through Flatland. In an anal-
ogous way, we can get a sense of 4-dimensional objects by watching 3-dimensional
slices of them pass through our space. For example, in Chapter 1 we learned that
by analogy with the definition of a 2-dimensional sphere in R3 , the 3-dimensional
sphere is defined by
S 3 = {(x, y, z, w)|x2 + y 2 + z 2 + w2 = 1}
as a subset of R4 .
In order to help A. Square visualize S 2 , we told him that S 2 can be obtained by
gluing two disks together along their boundaries. However, in order to physically
glue the disks together, we would first have to stretch them into two hemispheres
in R3 , which would be impossible for A. Square to picture. Similarly, we can think
3. MOVIES AND MOVEMENT 35
Figure 11. Frames of a movie of A. Ball passing through Flatland.
of S 3 as a gluing of two 3-dimensional balls along their boundaries. However, if we

wanted to physically glue the balls together, we would need to stretch them into
two “hemiballs” in R4 . Since this is impossible for us to picture, let’s instead think
about what we would see if S 3 passed through our space.
It’s always a good idea to begin by considering the same situation one dimension
down. What would A. Square see if a 2-sphere passed through Flatland? It would
be similar to how A. Square perceived A. Ball passing through Flatland, except
that a 2-dimensional sphere would be hollow (like a holograph) and wouldn’t have
a bow on its head. In particular, A. Square would see a point appear out of thin
air, and then he would see it grow into a small hollow circle. The circle would grow
until it had radius 1, which is the radius of
S 2 = {(x, y, z)|x2 + y 2 + z 2 = 1}.
Then the circle would shrink, finally becoming a single point again before disap-
pearing altogether.
In a similar way, if S 3 passed through our 3-dimensional universe, we would
see a point appear out of thin air, and then grow into a small hollow sphere, like a
round soap bubble. The hollow sphere would continue to grow until it had radius
1, and then it would shrink back down until it finally became a single point again,
after which it would disappear altogether.
While we can’t put the slices of S 3 together physically, watching the slices
change over time enables us to visualize S 3 in a 4-dimensional space made up of
three physical dimensions together with time as the fourth dimension. Similarly,
A. Square can visualize A. Ball in a 3-dimensional space made up of two physical
dimensions together with time as the third dimension. In fact, no matter how many
spatial dimensions you live in, you can always add time as the next dimension.
At first this seems like a satisfying way of visualizing objects in higher dimen-
sions. However, our relationship with time is very different from our relationship
with spatial dimensions. We can move forwards and backwards in space at var-
ious speeds. But we can’t go back in time, stay still in time, or move faster in
time. We can only move forward in time at a fixed “speed”. Since mathematicians
look for patterns and symmetry, they prefer all dimensions to be interchangeable.
There is no particular dimension which is the first dimension, and another which
is the second dimension, and so on. Rather, dimensions 1, 2, and 3 are all inter-
changeable. Therefore, from a mathematician’s point of view, dimension 4 should
be interchangeable with the others as well, and hence should not be time.
Nonetheless, time is useful because it enables us to use movies and slicing as
one way to help us visualize higher-dimensional objects. Also, time is a necessary
coordinate. If I want to meet a friend for dinner, then I have to specify not only
three spatial coordinates (e.g., on the second floor of the building at the corner of
Broadway and 95th street), but also a time coordinate (e.g., at 7:00 PM on June
29th ). A mix-up in any one of these four coordinates will prevent us from meeting.
4. Inductively Defining Cubes of Dimensions 0, 1, 2, and 3
In this section we will explain how to define cubes of different dimensions

inductively. This means that we will define a 0-dimensional cube, and explain
a method to construct a 1-dimensional cube from a 0-dimensional cube. Then we
will use the same method to construct a 2-dimensional cube from a 1-dimensional
cube. Continuing on, we again use this method to construct a 3-dimensional cube
from a 2-dimensional cube. And so on.
We begin by defining a 0-dimensional cube to be a single point (we don’t have
much choice here; any 0-dimensional object is a single point). We construct a
1-dimensional cube by dragging the 0-dimensional cube along a straight path of
unit length in a perpendicular direction. Since all directions are perpendicular to
a point, we can choose any direction for our path. In Figure 12, we have dragged
our 0-dimensional cube in the horizontal direction, so that our 1-dimensional cube
is a horizontal line segment of unit length.
Figure 12. A 1-dimensional cube.
Now we use the same method to construct a 2-dimensional cube, starting with
a 1-dimensional cube. This means we drag our 1-dimensional cube along a straight
path of unit length in a direction which is perpendicular to the 1-dimensional cube.
We see from Figure 13 that a 2-dimensional cube is a square. Note that even though
we have not colored the interior of the square, every point inside the square is part
of our 2-dimensional cube.
Continuing with this method, we construct a 3-dimensional cube by dragging

a 2-dimensional cube along a straight path of unit length in a direction which is
5. A. SQUARE LEARNS ABOUT A 3-DIMENSIONAL CUBE 37
perpendicular to the 2-dimensional cube. In particular, by dragging Figure 13 in a

direction which is perpendicular to the page of the paper we obtain a 3-dimensional
cube, which is what we usually call a cube (see Figure 14). Again, remember that
this is a solid cube and not just a wire frame or a cardboard box in the shape of a
cube.
5. A. Square Learns About a 3-Dimensional Cube
Before we construct a 4-dimensional cube, we want to check in on A. Square

who was so disturbed by his visit from A. Ball that he has gone to bed. However,
A. Ball is not ready to leave him alone yet. She has the idea that perhaps she can
teach him about the third dimension without scaring him if she talks to him in a
soothing voice while he is dozing off. This works well because A. Square assumes
her voice is part of a dream, and he is quite used to having bizarre dreams.
In order to help A. Square visualize a 3-dimensional cube, A. Ball is planning
to describe Figure 14 to him. However, she suddenly realizes that A. Square would
see the figure as a hexagon with a bunch of strange dotted line segments inside. So
instead, A. Ball takes the approach of describing what A. Square would see if cubes
of dimensions one, two, and then three passed through Flatland.
First, she tells him that if a line segment (a 1-dimensional cube) pierced through
Flatland end-on he would see a 1-dimensional cross section, which would appear,
stay for a while, and then disappear suddenly. Because he can picture cross sections
of a line segment in his own space, he can understand when she explains that this is
actually a continuous progression of distinct points which are cross sections of the
line segment. Similarly, if a square (a 2-dimensional cube) passed through Flatland
edge first, A. Square would see the cross sections of the square as a single segment
which appeared out of nowhere, stayed for a while, and then disappeared. Again
because he can visualize cross sections of a square, A. Square can understand that
this is actually a continuous progression of distinct segments which are “stacked
together” to form a square.
At this point, A. Ball tells A. Square that if a 3-dimensional cube passed
through Flatland, he would see the cross sections of the cube as a single square
which appeared out of nowhere, stayed for a while, and eventually disappeared.
A. Square thinks that this sounds reasonable after having thought about the cross
sections of the line segment and square. But it’s still hard for him to visualize how
a continuous progression of squares could be put together to create a 3-dimensional
cube. A. Ball considers showing him the pictures of cross sections in Figure 15.
But then she realizes that these pictures would be as incomprehensible to him as
Figure 14 would have been.
Figure 15. Two cross sections of a 3-dimensional cube.
One reason that the pictures appearing in Figures 14 and 15 are incomprehen-
sible to A. Square is that they contain segments which pass behind other segments.
We may have trouble seeing what goes behind what in the picture, but A. Square
doesn’t have the perspective to be able to do this at all. We can avoid drawing
such segments if we are willing to use size to represent distance from the viewer.
For example, in the illustration of a room in Figure 16, the grey square in the back
of the room is drawn smaller than the black square (which we are looking through)
in order to indicate that the back wall is farther away from the viewer than the
front wall. In art, this type of drawing is called one-point perspective and is often
described in “learn-to-draw” books.
Figure 16. A drawing of a 3-dimensional room with one-point perspective.
Figure 17 is a drawing of a 3-dimensional cube using one-point perspective. At

first glance, this figure just looks like an arrangement of line segments in the plane.
However, if you think about the picture in three dimensions as you did the room
in Figure 16, you can force your mind to see it as a cube. To see the six faces of
the cube, observe that the center square is the back face of the cube and the four
trapezoids around it are the four sides of the cube. The large outer square is the
front face of the cube that we are looking through. Notice that the front face is
actually superimposed on the other five faces.
A. Square cannot see Figure 17 as we see it. However, he can understand what
a square is, and he can understand what it means for one square to be smaller
than another. He can even understand what A. Ball means if she describes the
drawing as a smaller square inside of a larger square with the corners connected.
She then explains to A. Square that the smaller square is not actually inside the
larger square, but rather it is behind the larger square in the third dimension. She
also tells him that both squares are actually the same size, and that one just looks
smaller than the other because it’s farther away in the third dimension. Even
though he doesn’t understand the notions of “behind in the third dimension” or
5. A. SQUARE LEARNS ABOUT A 3-DIMENSIONAL CUBE 39
Figure 17. A one-point perspective drawing of a cube.
“farther away in the third dimension”, he is willing to accept this picture. A. Ball
decides to call pictures of this type “Flatland-friendly” pictures, because they don’t
require 3-dimensional perspective in order to understand them.
Returning to her earlier description of what A. Square would see if a 3-dimen-
sional cube passed through Flatland, A. Ball now draws the Flatland-friendly pic-
tures of slices of a cube in Figure 18. When we look at this figure, we can imagine
a vertical plane moving from the front to the back of a cube, creating identical grey
slices at each stage. By contrast, when A. Square looks at this figure he sees a grey
square in the left image that is bigger than the grey square in the middle, which
in turn is bigger than the grey square on the right. Nonetheless, A. Square seems
willing to accept the idea that this drawing represents slices of a 3-dimensional
cube.
Figure 18. A Flatland-friendly picture of slices of a cube as it

passes through Flatland.
At this point, A. Ball feels like she’s on a roll (which, after all, is one of her
favorite feelings). So she decides to give A. Square even more insight into a 3-
dimensional cube by drawing projections of the cube rotating around a vertical axis
in 3-dimensional space (see Figure 19). In order to avoid confusing A. Square, she
does not use gaps to indicated when one segment passes behind another. However,
this makes the drawings somewhat more confusing for a 3-dimensional person.
If we set aside our 3-dimensional intuition for the moment, we can see each
image of Figure 19 through A. Square’s eyes as a collection of quadrilaterals. If
we track the grey shape from one image to the next, we will observe that its
edges stretch and shrink, and the angles between them increase and decrease. In
particular, the grey square in the center of the first image becomes a trapezoid
starting in the third image, since its left edge shrinks while its right edge stretches.
Similarly, the white trapezoid that is on the left side of the image at time t = 0
will slowly morph into a square. Most interesting of all, the white trapezoid on the
right side of the figure at t = 0 first becomes thinner, then when t = 2, its short
side switches from the left to the right, and finally it gets bigger and more square.
Figure 19. A cube rotating around a vertical line.
A. Square is a bit uncomfortable with these images of a rotating cube. However

he says he accepts them so that A. Ball will shut up and let him dream peacefully
about B. Triangle.
6. A 4-Dimensional Cube
Now that A. Square has learned about the 3-dimensional cube, it’s time for us
to take the leap and learn about the 4-dimensional cube. Continuing the inductive
construction that we began in Section 4, we build a 4-dimensional cube by dragging
a 3-dimensional cube along a straight path of unit length in a direction which is
perpendicular to the cube itself. Now, it’s pretty tough for us to understand how
something could be perpendicular to all three of our dimensions, just as it was
pretty tough for A. Square to understand how something could be perpendicular to
his two dimensions. But we can take our inspiration from A. Square’s experience
and draw a “3-dimension-friendly” picture of a 4-dimensional cube by drawing two
3-dimensional cubes, one inside the other, and connecting analogous vertices in the
outside and inside cubes with line segments (see Figure 20). Remember that the
inner cube is not smaller than the outer cube, it’s just “farther away in the fourth
dimension”. Notice that this visualization is analogous to the one-point perspective
of a 3-dimensional cube that we illustrated in Figure 17.
Figure 20. A “3-dimension-friendly” drawing of a 4-dimensional cube.

6. A 4-DIMENSIONAL CUBE 41
Let’s examine Figure 20 more carefully using A. Square’s visualizations of a 3-

dimensional cube as a guide. A 4-dimensional cube has eight 3-dimensional “solids”
that are the analogues of the six 2-dimensional faces of a 3-dimensional cube. These
solids are actually 3-dimensional cubes, though most of them look distorted in the
figure. This is analogous to the distortion of the faces of the 3-dimensional cube
that we saw in Figure 17. One of these solids is the small cube in the center of the
figure, another one is the large cube containing the figure, and the other six solids
are the prisms that connect each face of the small cube to the corresponding face
of the large cube.
By analogy with our discussion of slicing a 3-dimensional cube, we can under-
stand the 4-dimensional cube as a continuous progression of 3-dimensional cubical
cross sections. For example, the grey cubes in the middle of the images in Figure 21
are 3-dimensional cubical cross sections of the 4-dimensional cube. Note that each
cross section consists not only of the six faces of a cube, but of the cubical volume
enclosed by those faces. Just as A. Square had to accept that the cross sections
of a cube in Figure 18 were not nested, though they appeared to him as if they
were, we have to accept that the cubical cross sections of the 4-dimensional cube
in Figure 21 are not nested though they appear to us as if they are.
Figure 21. Two 3-dimensional cross sections of a 4-dimensional cube.
You can find animations of a rotating 4-dimensional cube on the internet. How-
ever, when you look at such animations you should remember the distortions that
A. Square saw in Figure 19. We were able to understand the projections in Fig-
ure 19 because of our experience with solid cubes. In particular, we knew that all
of the faces of the cube were really squares, that all of the edges have the same
length, and that every angle is a right angle. The same is true for animations of
a rotating 4-dimensional cube. The various edges appear to stretch and shrink,
while the angles between pairs of adjacent edges seem to change as the edges pivot
about the vertices. As a consequence, the faces and solids appear to rotate and flex.
Some of the solids even look like they are turning inside out during the rotation.
However, we have to keep in mind that the faces are always flat squares and the
solids are always cubical boxes, no matter how they look.
Don’t worry if you are uncomfortable with Figure 21 or the animations of a
rotating 4-dimensional cube that you can find on the internet. A. Square knows
exactly how you feel.
7. Tetrahedra in Various Dimensions
Cubes are not the only geometric figure that can be constructed inductively in
different dimensions. We will use a similar procedure to construct a 4-dimensional
tetrahedron. Again, we start with a single point which we call a vertex. This is
a 0-dimensional tetrahedron (remember a point is the only 0-dimensional shape).
To get a 1-dimensional tetrahedron, we add a new vertex one unit away from
the previous one and connect the two vertices with a line segment. As we see in
Figure 22, a 1-dimensional tetrahedron is the same as a 1-dimensional cube.
Figure 22. A 1-dimensional tetrahedron is the same as a 1-

dimensional cube.
Next, we add a third vertex which is one unit from each of the previous two
vertices. Notice that we cannot put the new vertex on the same line as the previous
two. To get a 2-dimensional tetrahedron, we add line segments that connect the
new vertex with the vertices of the 1-dimensional tetrahedron. This creates an
equilateral triangle (including its 2-dimensional interior) as seen in Figure 23.
Figure 23. A 2-dimensional tetrahedron.
Continuing in a similar fashion, we build a 3-dimensional tetrahedron by adding

a new vertex that is one unit from each of the previous three vertices. In order to
do this, we are forced to place our new vertex outside the plane of the paper. We
connect this new vertex to the vertices of the 2-dimensional tetrahedron by means
of line segments, thus creating a solid triangular pyramid like the one shown in
Figure 24(a). This is what we usually think of as a tetrahedron.
Figure 24. Two projections of a 3-dimensional tetrahedron.
Flatlanders prefer the image in Figure 24(b), which shows the tetrahedron from
above and does not have one segment behind another. They think of this as an
equilateral triangle with a vertex in the center and line segments connecting it
7. TETRAHEDRA IN VARIOUS DIMENSIONS 43
to the other three vertices. We explain to the Flatlanders that even though the
segments on the inside of the triangle look shorter than the sides of the triangle,
the central vertex is actually in a different plane in the third dimension so that all
of the segments have exactly the same length.
To give the Flatlanders more intuition, we explain that if a 3-dimensional tetra-
hedron passed through Flatland bottom first, they would see a 2-dimensional tetra-
hedron (an equilateral triangle) which appeared out of nowhere and gradually got
smaller until it eventually shrank down to a point. In Figure 25, we illustrate two of
the cross sections that Flatlanders would see. Again they prefer the images in Fig-
ure 25(b) over those in Figure 25(a) since they cannot understand the 3-dimensional
perspective in Figure 25(a). However, we still have to remind them that the two
grey triangles in Figure 25(b) are not nested, even though they appear that way.
Figure 25. Cross sections of a tetrahedron, viewed from two dif-

ferent perspectives.
Finally, to construct a 4-dimensional tetrahedron, we add a new vertex which

is one unit away from the previous four vertices. This is impossible for us to
accomplish in three dimensions, but poses no problems in 4-dimensional space.
If we momentarily suspend our disbelief and connect this new vertex to all of
the vertices in the 3-dimensional tetrahedron using line segments, then we will
create a 4-dimensional tetrahedron. From our 3-dimensional perspective, we can
construct a 4-dimensional tetrahedron by placing the new vertex in the center of
the 3-dimensional tetrahedron, as in Figure 26. However, we should realize that
the central vertex is not actually inside the tetrahedron; rather, it is poking out of
our 3-dimensional space into the fourth dimension. Furthermore, all of the edges in
the 4-dimensional tetrahedron are the same length even though the ones containing
the central vertex look shorter.
If a 4-dimensional tetrahedron passed through our 3-dimensional space bottom
first, we would see a 3-dimensional tetrahedron which appeared out of nowhere and
gradually got smaller until it eventually shrank down to a point. One of these cross
sections is illustrated in Figure 27. Remember that the tetrahedra are not nested,
even though the picture makes it look like the grey tetrahedron is inside of the
white tetrahedron.
Figure 26. A 3-dimensional drawing of a 4-dimensional tetrahedron.
Figure 27. A 3-dimensional cross section of a 4-dimensional tetrahedron.
8. Exercises
1. Another way to explain a 3-dimensional cube to a Flatlander uses the gluing

technique we discussed in the last chapter. We cut open the cube along some
of the edges (like collapsing a box), and then label the cut-open edges with
numbers so that we know how to reassemble the box. Thus, edge 1 is glued to
the other edge 1, edge 2 is glued to the other edge 2, and so on (as illustrated
on the left side of Figure 28).

8. EXERCISES 45
Now consider the unfolded 4-dimensional cube illustrated on the right side of
Figure 28. Fill in the missing numbers to show how the faces are glued together
to produce a 4-dimensional cube.
2. What does A. Square perceive if a 3-dimensional person slowly puts his/her

hand into Flatland? What might we perceive if a 4-dimensional person slowly
stuck his/her hand in our space?
3. A 3-dimensional ball is defined by
B 3 = {(x, y, z)|x2 + y 2 + z 2 ≤ 1}
as a subset of R3 . By analogy we define a 4-dimensional ball by
B 4 = {(x, y, z, w)|x2 + y 2 + z 2 + w2 ≤ 1}
as a subset of R4 . Describe the slices that we would see if B 4 passed through

our 3-dimensional space.
4. What would A. Square see if a line segment passed through Flatland at an

angle other than 90◦ ?
5. What would A. Square see if a square passed through Flatland corner first?
6. Suppose a cube passes through Flatland in various different ways. What are
all the possible 2-dimensional shapes that a Flatlander might see?
7. Consider the frame sequences in Figure 29, and imagine that A. Square is
experiencing these movies frame-by-frame (from bottom to top). In each case,
describe the shape of the object that is passing through Flatland.

8. What if a solid torus (in other words, a doughnut) passes through Flatland,
as in Figure 30(a)? What shapes would A. Square see, and how would they
change? What sequence of frames would he perceive if the solid torus was
oriented as in Figure 30(b)?
9. Suppose that A. Square is watching a circle move in a straight line away from
him as in Figure 31(a). Would A. Square perceive the same thing if a slanted
cylinder passed through Flatland as in Figure 31(b)? Could he tell the difference
between these two scenarios? If so, how?
10. Design a 2-dimensional room with a lock on the door so that the Flatlanders
inside can lock themselves in so that nobody will rob them while they are
sleeping. Note that gravity and magnets don’t exist in Flatland, so you cannot
use them in designing your lock.
11. Design a digestive system for A. Square that does not cause him to fall apart.
For the sake of health, A. Square’s digestion must have a separate entrance
for food and exit for waste, both of which might be open at the same time in
case of illness. Also you cannot assume that food is simply absorbed through
a membrane.
8. EXERCISES 47
12. If we lift A. Square above Flatland and hold him perpendicular to it, what will
he see?
13. If Flatland and everything in it had the exact same amount of 3-dimen-
sional thickness (including A. Square’s eyes), could A. Square detect it? If
everything in our universe had the same small 4-dimensional thickness, could
we detect it?
14. What would we see if another 3-dimensional universe intersected ours perpen-
dicularly? How would we see the 3-dimensional creatures of that space?
15. How many vertices, edges, faces, and solids does a 4-dimensional cube have and
why? What about a 4-dimensional tetrahedron? Find inductive formulae for
the number of vertices, edges, faces, and solids of an n-dimensional cube and an
n-dimensional tetrahedron in terms of these numbers for an (n−1)-dimensional
cube and (n − 1)-dimensional tetrahedron. Justify your conclusions.
16. An n-dimensional pyramid consists of an (n − 1)-dimensional cube with all
the vertices attached to a single additional vertex in the nth dimension (see
Figure 32). Determine how many vertices, edges, faces, and 3-dimensional
solids a 4-dimensional pyramid and 5-dimensional pyramid have. Justify your
conclusions.

CHAPTER 3
Geometry and Topology of Different Universes
Topics:
• Intrinsic and extrinsic geometry
• Geodesics
• Triangles and geometric properties
• Intrinsic and extrinsic topology
• Loops and topological properties
• Local and global properties
• Homogeneity
• Manifolds
• Assumptions about universes
• 2- and 3-dimensional spheres with a point removed
• Curves on a torus
1. Intrinsic and Extrinsic Properties of a Space
Early one Saturday morning your phone rings. Half-asleep, you grab the phone
and mumble “hello”. You don’t catch the person’s name, but she claims to be your
fourth cousin twice removed, or something like that. “Oh, yes, nice to hear from
you” you say, though you actually have no idea what it means to be a fourth cousin
twice removed. She says she just found out that she’s related to you and she’s eager
to get to know you. She wants to know everything about your life, starting with
where you live. You tell her you live in New York City. But she asks where New
York City is. This seems strange since everyone in the US (and in a large part of
the world) knows where New York City is. Things get even stranger when you tell
her it’s in the Northeastern part of the US and she doesn’t seem to know what you
mean. You find yourself describing the geography of the US, while wondering how
such an ignorant person could possibly be any kind of a cousin of yours no matter
how many times removed. Eventually, after you have explained that the earth is a
planet in the universe she asks “what’s the shape of the universe?” At this point,
you are wide awake and decide this is a prank call. You slam the phone down in
exasperation. The problem is that her last question keeps going around and around
in your head. What is the shape of the universe?
In Chapter 1, we learned that in addition to R3 , our universe could be T 3 or S 3 .
But it’s unlikely that there are only three possibilities for the shape of our universe.
Especially since in Chapter 1 we saw that there are infinitely many 2-dimensional
49
50 3. GEOMETRY AND TOPOLOGY OF DIFFERENT UNIVERSES
universes. However, coming up with other possibilities for our 3-dimensional uni-
verse seems much harder than coming up with more 2-dimensional universes. Of
course, for A. Square it’s much easier to think of possibilities for 1-dimensional uni-
verses than 2-dimensional universes. In general, it’s easier for an external observer
in a higher dimension to understand a space than for inhabitants to understand
their own space. To help us keep track of the distinction between properties of a
space that only an outsider can observe and properties that inhabitants can figure
out themselves we make the following definition.
Definition. A property of a space is said to be extrinsic if it can be detected
by an observer outside of the space but it can NOT be detected from within the
space. Otherwise, the property is said to be intrinsic.
Recall from Chapter 1 that properties of spaces are like characteristics that we
use to tell people apart. For example, suppose that I am talking on the phone to
A. Person who is in A. Faraway Place and has never heard of Santa Claus or the
Easter Bunny. If I said that Santa always wears a red suit and the Easter Bunny is
naked except for its fur, A. Person would agree that they must be different beings.
Thus how they are dressed or not dressed is a characteristic that distinguishes them.
Properties of spaces play the same role. They are useful because if you know that
one space has a property and another space doesn’t, then you can conclude that
they must be different.
Extrinsic properties have to do with how a space is situated in a larger or higher-
dimensional space. For example, suppose that A. Square is meeting B. Triangle for
lunch. Whether they live on a flat plane or a plane that is bent like the one
illustrated on the right in Figure 1, A. Square has to walk the same distance to
meet her. In fact, there is no way that he can tell the difference between the two
planes illustrated in Figure 1. On the other hand, to us he seems much closer to
her if the plane is bent than if it’s flat. But he cannot jump through 3-dimensional
space to go from one side of the bend to the other. So the fact that the two of
them are closer in 3-dimensional space is irrelevant (and meaningless) to him. Thus
whether or not a plane is bent is an extrinsic property.
Figure 1. A. Square has to walk the same distance to get to

B. Triangle whether his plane is flat or bent.
Now suppose that A. Square somehow knows that Flatland is either a sphere
or a curved torus. From our 3-dimensional perspective we have no trouble telling
these spaces apart since we can see that the curved torus has a hole in it while
the sphere does not. Because we see the hole easily from the outside, we might be
tempted to say that the existence of such a hole is an extrinsic property. However,
for a property to be extrinsic it must be undetectable by the inhabitants of the
1. INTRINSIC AND EXTRINSIC PROPERTIES OF A SPACE 51
space. You might recall from Chapter 1 that A. Square could in fact distinguish
a sphere and a torus either by using straightest paths or by using walls. Thus the
existence of the hole in the torus is actually an intrinsic property, though A. Square
certainly wouldn’t use the word “hole” in his explanation of the difference between
a torus and a sphere.
This example illustrates that even though a property may be easier for creatures
outside of the space to detect, if the creatures in the space can detect it at all (even
with great difficulty and without real understanding or insight into what’s going
on), then the property is intrinsic.
Definition. We say that two spaces are intrinsically the same if a being
in one space would have no way of knowing if he or she were suddenly transported
from one space to the other. We say that say two spaces are extrinsically the
same if external beings would be unable to distinguish the two spaces.
In other words, two spaces are intrinsically the same if all of their intrinsic prop-
erties are the same. Since external beings can observe both intrinsic and extrinsic
properties of a space, two spaces are extrinsically the same if all of their intrinsic
and extrinsic properties are exactly the same. This means that if two spaces are
extrinsically the same, they are automatically intrinsically the same.
For example, the flat plane and the bent plane in Figure 1 are intrinsically
the same but extrinsically different. On the other hand, a sphere and a curved
torus are intrinsically different (and extrinsically different) because of the intrinsic
properties involving straightest paths and walls that we used to distinguish them
in Chapter 1.
The language can be a bit confusing here, and it’s worth taking some time
to carefully think through the definitions to make sure you understand them. In
particular, keep in mind the following important observations.
• A property of a space cannot be both intrinsic and extrinsic, since it is
only extrinsic if it is not intrinsic.
• Two spaces (like a sphere and a curved torus) can be both intrinsically
and extrinsically different.
• Two spaces that are identical (like two spheres of the same size) are both
intrinsically and extrinsically the same.
• If two spaces are intrinsically different, then one has an intrinsic property
that the other does not have. Hence they are also extrinsically different.
• If two spaces are extrinsically different (like a flat plane and a bent plane),
they may or may not be intrinsically different.
The goal of this chapter is to develop ways to determine if two spaces are the
same or different, both from the outside and from the inside. To do this we will
need to introduce a lot of definitions which will build on one another. In order to
absorb all of this material, you should read the chapter slowly, and at the end of
each section you should stop and check that you understand and remember earlier
definitions before you continue reading. For example, right now you should check
that you understand the definitions in this section before starting the next section.
2. Intrinsic and Extrinsic Geometry
One way to distinguish spaces is to use some kind of measuring device like a
ruler to measure lengths or a protractor to measure angles. More formally, we have
the following definition.
Definition. A property of a space that can be numerically measured is said to
be a geometric property.
Recall from the last section that whether a plane is flat or bent like in Figure 1
is an extrinsic property. We can now be more specific and say that whether or
not a plane is bent is an extrinsic geometric property because we can measure the
distance between A. Square and B. Triangle in 3-dimensional space and see that
they are closer together when they are on the bent plane than when they are on
the flat plane.
Definition. We say that two spaces have the same intrinsic geometry if
there is no measurement that would enable a being in one space to know if he or
she were suddenly transported to the other space. We say that two spaces have the
same extrinsic geometry if there is no measurement which would enable external
beings to distinguish the two spaces.
This definition means that two spaces have the same intrinsic geometry if all
of their intrinsic geometric properties are the same, and two spaces have the same
extrinsic geometry if all of their intrinsic and extrinsic geometric properties are the
same.
For example, let’s consider the two spheres illustrated in Figure 2. These
spheres have different intrinsic geometry because if A. Square were transported from
one to the other, he could measure the total area of his space to know that it had
changed. Since total area is an intrinsic geometric property, they automatically also
have different extrinsic geometry. However, this is not the only way to distinguish
their extrinsic geometry. For example, the two spheres enclose different volumes in
R3 , which is an extrinsic geometric property that distinguishes them.
Figure 2. These spheres have different intrinsic and extrinsic geometry.
Suppose that A. Square and B. Triangle live on a rigid steel sheet. If we moved
the sheet without bending it, would the distance between them be any different?
What if we rotated it or turned it over and then remeasured it? These types of
movements of the steel sheet are examples of rigid motions. Any motion of an
3. STRAIGHTEST PATHS AND GEODESICS 53
object in space that doesn’t involve stretching, shrinking, or bending is said to be

a rigid motion. If one object can be taken to another by a rigid motion, then the
two objects are identical. In particular, this means both their extrinsic geometry
and their intrinsic geometry are the same. However, the following theorem (whose
proof you might see in a more advanced geometry class) tells us even more.
Theorem 3.1. Two objects in R3 have the same extrinsic geometry if and only
if there is a rigid motion of R3 taking one object to the other.
It’s not surprising that if there is a rigid motion taking one object to another,
then the two objects are identical and hence have the same extrinsic geometry.
What’s more surprising is that if two objects have the same extrinsic geometry,
then there is actually a rigid motion taking one to the other. In other words,
saying that two objects have the same extrinsic geometry is the same as saying
they are identical in every respect. So if you can find any aspect of them that is
different, then you can conclude that they have different extrinsic geometry.
3. Straightest Paths and Geodesics
Almost every geometric property has something to do with straightest paths.

We need straightest paths to be able to measure length, area, volume, and even
angles. In Chapter 1, we defined straightest paths in a practical way, saying that
light travels along a straightest path and a taut string follows a straightest path.
But now that we are going to be using straightest paths to help us distinguish the
geometry of different spaces, we need to be a little clearer about exactly what we
mean by a straightest path. Thus we introduce a new word for a straightest path
and a more formal definition of what it means.
Definition. A path in a space is said to be a geodesic, if for any small enough
region which intersects the path, and points p and q in the small region on the same
segment of the path, then the shortest path from p to q is on that segment.
As a first example, suppose we have a cylinder made out of paper and a path
on the cylinder which is a geodesic. Let’s see what we can learn about the path.
According to the definition, we need to focus on a “small enough region” of the
cylinder which intersects the path. So let’s choose our region to be a really, really,
really small disk which intersects the path in a really, really, really small arc, and
let p and q be points on the really, really, really small arc. Because the disk is so
very small and the path is a geodesic, we know that the shortest path from p to q
on the cylinder goes along the arc in the disk (see Figure 3).
Now let’s cut the disk out of the cylinder and lay it flat on a table. Because
paper is not stretchy, the lengths of paths in the disk are the same whether the
disk is part of the cylinder or lying flat on the table. This means that the geodesic
arc from p to q in the disk will also be the shortest path from p to q after the disk
has been cut out and is lying flat on the table. But we know that for the flat disk
the shortest path from p to q is a line segment. Thus the geodesic arc in the disk
becomes a line segment when the disk is lying flat on the table.
Since this is true for every really, really, really small disk that intersects the
geodesic, it follows that if we cut open the entire cylinder and lay it flat on the
table, then the entire geodesic will become one or more line segments on the flat
Figure 3. p and q are points on the really really really small arc
in the really really really small disk.
Figure 4. When we cut open the cylinder along the thick black
line, this geodesic becomes three line segments.
page. In Figure 4, we cut open the cylinder along the thick black line and see that
our geodesic becomes three line segments.
We can make a similar argument going in the reverse direction. That is, starting
with a line segment on a piece of paper, if we roll the paper into a cylinder, then
the line segment will become a geodesic on the cylinder. This is exactly what we
will do in our next example, except we will roll the paper into a cone rather than
a cylinder.
In this case, we will start with a piece of paper in the shape of a wedge whose
top angle is 120◦ . Then we draw a horizontal line segment from one edge of the
paper to the corresponding point on the other edge (see the left side of Figure 5).
Now when we roll the wedge into a cone, the ends of the grey segment will touch.
In Figure 5, we glue the edges of the paper in the front of the picture to make it
easier to see how the ends of the grey segment come together at the junction.
Now we draw a really, really, really small disk on the cone which intersects the
grey segment as illustrated in Figure 6, and we let p and q be points on the segment
in the disk. As we saw with the cylinder, we can cut out the disk and lay it flat,
and it won’t change the length of any path. Since the grey arc is a line segment
on the flat disk, we know that the grey arc is the shortest path from p to q on the
cone.
Next let’s consider a small disk centered around the point where the endpoints
of the grey arcs meet, and let p and r be points on the grey segment as illustrated
in Figure 7. In this case, we can see that the shortest path from p to r would
go above the above the grey arcs rather than along them. However, this does not
Figure 5. When we attach the sides of the cone, the endpoints of

the grey segment will come together at the junction.
Figure 6. The shortest path from p to q is along the grey arc in

the disk.
contradict our definition of a geodesic since the points p and r are on different arcs
in the disk.
Since we only have to consider pairs of points which are on the same arc in the
disk, we can use the same argument as we did for Figure 6 for any really, really,
really small disk intersecting the grey segment. Hence it follows that the grey path
is in fact a geodesic on the cone. Thus for both the cone and the cylinder, a line
segment in the gluing diagram becomes a geodesic once we glue up the appropriate
edges.
Returning to the cone, we see that the geodesic in Figure 7 doesn’t end when
the endpoints of the grey segment come together. To see where it goes, let’s see
what happens when A. Square takes a walk on the gluing diagram following the
grey segment as it goes off to the right. Since the right and left edges of the wedge
are glued together, we know A. Square will re-enter the wedge on the left. But we
have to be careful about what direction he’s facing when he re-enters.
Figure 7. The points p and r are on different arcs in the disk.
In Figure 8, we can see that as he is exiting on the right, the right edge of the
diagram cuts across him from just in front of the feather in his hat to his mouth.
So when he re-enters the diagram on the left, the left edge of the diagram has to
cut across him in the exact same way, from just in front of the feather in his hat
to his mouth. Also, after he exits on the right, there is a 30◦ angle between his
eye and the gluing edge. Thus after he re-enters on the left, there must also be
30◦ angle between his eye and the gluing edge. Since 180◦ − 30◦ − 30◦ = 120◦ , we
can now conclude that when A. Square re-enters in the left, he has been rotated by
120◦ , as illustrated in Figure 8.
Figure 8. When A. Square goes off the right side of the diagram,

he re-enters on the left rotated by 120◦ .
A similar thing happens if A. Square takes a walk on the gluing diagram fol-
lowing the grey segment as it goes off to the left. Thus in the gluing diagram, we
can extend the grey segment on both sides as illustrated on the left in Figure 9.
In particular, suppose A. Square starts on the grey segment marked 1, walking
in the direction of the arrow. This is the reverse of the way he walked in Figure 8.
Now, when he crosses the gluing edge, he continues on the grey segment marked 2,
again walking in the direction of the arrow. Then after again crossing the gluing
edge, this time as he did in Figure 8, he continues on the grey segment marked 3
Figure 9. The grey segment can be extended on both sides.
Figure 10. The points p and q are on the same arc in the disk,
but r is on a different arc.
again in the direction of the arrow. His path looks more normal when we look at
the grey segments on the cone.
Now that we’ve extended the grey segment on both ends, let’s return to the
really really really small disk cutting across the gluing line on the cone that we saw
in Figure 7. As we see in Figure 10, this disk intersects the grey path in two arcs,
and the points p and q are on one arc while the point r is on the other arc.
In order to see what the disk in Figure 10 looks like in the gluing diagram, we
need to put the two half-disks together. To do this, we first split the wedge down
the middle, as illustrated in the first image in Figure 11. Then we move the left
half-wedge to the right side and rotate both half-wedges so that the thin black edges
are lined up as in the second image. Finally, we glue the half-wedges together along
the thin black edges as in the third image. Thus we see that the grey segments
form an X and the disk from Figure 10 becomes a disk in the gluing diagram.
Figure 11. We cut apart and reattach the gluing diagram to see
the disk from Figure 10 in the gluing diagram.
Now because the length of the grey arc from p to q on the cylinder is the same
as it is in the gluing diagram, the shortest path between p and q on the cylinder
must go along the grey arc. Thus the grey path on the cone is indeed a geodesic.
Before we end this section, let’s revisit a sphere using our new word “geodesic”
in place of “straightest path”. In Chapter 1 we saw that on a sphere the great
circles are “straightest paths”. In particular, for any two points on a sphere, there
is a great circle containing them, and the shortest route from one point to the other
will be along a great circle. Thus the geodesics on a sphere are the great circles.
Observe that if two points are not opposite one another on the sphere, then there is
only one great circle containing them and one arc on this great circle will be shorter
than the other. For example, in Figure 12 we see that there is only one shortest
path from point a to point b. By contrast, for points x and y, which are exactly
opposite one another on the sphere, there are infinitely many geodesics between
them, and the lengths of all of these geodesics are the same.
Figure 12. There is only one geodesic from a to b, but infinitely

many geodesics from x to y.
4. THE DEFINITION OF A TRIANGLE 59
4. The Definition of a Triangle
In addition to measuring distances and size, measuring the angles of a triangle

is a good way to study the intrinsic geometry of a space. In a plane, a triangle is a
finite region bounded by three line segments. Using the idea of a geodesic, we can
now talk about triangles in any space.
Definition. We define a triangle in a space as a finite region bounded by

three geodesic segments such that the region can be deformed to a disk.
For example, in Figure 13, the grey region bounded by three grey segments is
finite, but the white region bounded by the same three segments is infinite. Thus
the white region can’t be a triangle.
Figure 13. The white region is not a triangle because it is infinite.
Also notice that we can deform the grey region to a disk as illustrated in
Figure 14. Hence we can conclude that the grey region is in fact a triangle and
the white region is not. This is important, since the sum of the angles of the grey
region is 180◦ , but the sum of the angles of the white region is much more.
Figure 14. The grey region is a triangle because we can deform

it to a disk.
Next observe that in Figure 15, the three arcs on the torus on the left do not
form a triangle, since they don’t bound a region on either side. On the other hand,
for the torus on the right, the grey region is a triangle, but the white region is not
a triangle since it cannot be deformed to a disk.
Figure 15. On the left, the arcs don’t form a triangle. On the
right, the grey region is a triangle but the white region isn’t.
5. The Sum of the Angles of a Triangle
We can measure the angle between a pair of intersecting geodesics by projecting

the geodesics onto a plane which is tangent to the surface at the point of intersection.
For example, Figure 16 illustrates two triangles whose sides are segments of great
circles on a sphere. For the large triangle, the sum of the angles is 270◦ , since all
three angles are 90◦ . In fact, on a sphere the sum of the angles of any triangle is
greater than 180◦ , and the bigger the triangle relative to the area of the sphere, the
greater the sum of its angles. As we can see in Figure 16, a really small triangle
looks almost flat, and hence its angle sum is almost equal to 180◦ . In fact, the
complement of each of these triangles is also a triangle, with an even bigger angle
sum.
Figure 16. On a sphere, the bigger a triangle is, the greater the
sum of its angles will be.
However, a sphere is not the only surface where the sum of the angles of a
triangle is not equal to 180◦ . Figure 17 illustrates a saddle surface where any
triangle has angle sum which is less than 180◦ , and the bigger the triangle is, the
smaller the angle sum will be. Notice that on the sphere in Figure 16, each of the
groups of three geodesic segments bound two triangles—the one that’s shaded grey
and the one on the outside of the grey region. By contrast in Figure 17, the three
geodesic segments only bound one triangle, since only one of the two regions is
finite.
Since A. Square can measure angles with his protractor, he can tell that the
sums of angles of triangles in a plane, a sphere, and a saddle surface are all different.
From this he can conclude that these three surfaces have different intrinsic geometry.
By using Theorem 3.1, we also know that these three surfaces have different extrinsic
geometry.
6. TRIANGLES ON A FLAT AND CURVED TORUS 61
Figure 17. A triangle on a saddle surface whose angle sum is less

than 180◦ .
6. Triangles on a Flat and Curved Torus
Let’s use triangles to compare the intrinsic geometry of a flat torus and a curved
torus. On a flat torus, any triangle that does not cross the edges of the gluing
diagram is an ordinary planar triangle, and hence the sum of its angles is 180◦ . In
order to check that triangles that do cross the edges of the gluing diagram also have
angle sum 180◦ , we use the idea of an extended diagram which we introduced when
we were studying tic-tac-toe on a flat torus in Chapter 1. Figure 18 illustrates an
extended diagram of A. Square in a flat torus. The grid lines and the arrows are
merely to indicate how the space is constructed; A. Square would not perceive them
at all. In fact, depending on how big the flat torus is, he may or may not realize
that the beings he sees are all just copies of himself.
Figure 18. An extended diagram of A. Square in a flat torus.
A triangle in the flat torus behaves just as A. Square does. Since the extended
diagram is a plane, any triangle in the gluing diagram corresponds to infinitely
many identical triangles in the plane (see Figure 19). In particular, the angles of
the triangle in the flat torus are the same as they are for each of the triangles in the
extended diagram. Since planar triangles have an angle sum of 180◦ , every triangle
in the flat torus must also have an angle sum of 180◦ , regardless of whether or not
it crosses the edges of the gluing diagram.
Now let’s consider the two triangles in the curved torus illustrated in Figure 20.
First observe that the outer region of the torus seems to curve like a sphere, while
Figure 19. A triangle in the flat torus corresponds to infinitely

many triangles in the extended diagram.
the region near the hole seems to curve like a saddle. As a result, triangles in the
outer region have angle sums that are greater than 180◦ , while triangles in the
inner region have angle sums that are less then 180◦ . If you look carefully enough,
you might even find some triangles whose angle sums are precisely 180◦ . So, there
is no general statement that we can make about angle sums of triangles on the
curved torus. The fact that not all triangles on a curved torus have angle sums
equal to 180◦ implies that the flat torus and the curved torus have different intrinsic
geometry.
Figure 20. The triangle on the outside has an angle sum greater
than 180◦ , while the one near the hole has an angle sum less than
180◦ .
Note we can’t compare the extrinsic geometry of a flat torus and a curved torus
because a flat torus is not contained in R3 . It’s an abstractly glued up surface that
exists in and of itself, rather than as a subset of a larger space. Hence it has
no extrinsic geometry. So we can’t say that its extrinsic geometry is the same or
different from that of the curved torus. Similarly, the 3-torus T 3 is an abstractly
glued up space, which is not contained in a larger space, and hence has no extrinsic
geometry.
7. Triangles on a Flat and a Curved Cylinder
Next let’s compare the intrinsic geometry of a flat infinite cylinder with that of
a curved infinite cylinder. We illustrate these two surfaces in Figure 21, where we
use dotted boundaries to indicate that the surfaces go on indefinitely. Just like on
7. TRIANGLES ON A FLAT AND A CURVED CYLINDER 63
the flat torus, the geodesics on the flat infinite cylinder are ordinary line segments
(or infinite lines) and every triangle has angle sum 180◦ , even those that cut across
the gluing line. You should check this by drawing an extended diagram of the flat
infinite cylinder. Note that just as we saw with the flat torus, the flat cylinder has
no extrinsic geometry since it’s an abstractly glued up surface.
Figure 21. Triangles on a flat and a curved infinite cylinder.
Next we want to consider triangles on a curved infinite cylinder. To think

about the geodesics on a curved infinite cylinder, you should make a curved (finite)
cylinder by rolling up a piece of paper and taping a pair of opposite sides together.
Draw two points on your cylinder and then draw what seems like a geodesic segment
P between the points. Now cut out a little region around your geodesic and lay
it flat on a table. Since paper is not stretchy, flattening out your region doesn’t
lengthen or shrink the path P that you drew. Thus if P was a geodesic on your
curved cylinder, then P will be a line segment when the region is flattened out.
Now that we know what a geodesic is on the curved cylinder, let’s consider
triangles on the curved cylinder. Observe that if we draw a triangle on the curved
cylinder, we can cut it out and lay it flat on a table as we did with the region
around our geodesic. Since each of the edges of the triangle becomes a line segment
when it’s flattened out, the cut-out triangle becomes a triangle in the plane when
it’s flattened out. Thus we know that its angles add up to 180◦ . Just as geodesics
don’t lengthen or shrink when we cut them out and lay them flat, angles don’t get
bigger or smaller when we cut out our triangle and lay it flat. Thus the angles of
any triangle on a curved cylinder must also add up to 180◦ .
In fact, since distance and angles are the same in the flat and curved cylinder,
any intrinsic measurement that we make in a curved cylinder will be the same as
the corresponding measurement within a flat cylinder. Thus the intrinsic geometry
of the flat cylinder is the same as that of the curved cylinder.
This may seem strange since the flat and curved torus have different intrinsic
geometries. The reason is that a curved cylinder can be made out of a piece of paper
while a curved torus cannot. If you try to make a curved torus from a piece of paper,
you can start by making a cylinder without any problem. But it’s impossible to
make the cylinder into a torus without some stretching or shrinking. We can see
this in the third picture of Figure 22 where the grey arc that will become a circle
going around the outside of the torus has to become longer than the black arc that
will become the circle going around the hole of the torus. Thus a curved torus can’t
be made from a piece of paper. You should try it yourself with a piece of paper
and some tape to see that indeed it can’t be done.
Figure 22. We can’t make a curved torus out of a piece of paper

because paper can’t stretch or shrink.
As we have seen in these examples when we stretch or shrink an object, we

change its intrinsic geometry. But, as we’ll see in the next section, geometry isn’t
everything.
8. Extrinsic and Intrinsic Topology
Since A. Square has been spending a lot of time with B. Triangle, he’s been
going out to eat a lot. His friends see that he has put on a little extra weight in the
rear and tease him by making comments about noticing a change in his extrinsic
geometry (see Figure 23). He can’t deny that his intrinsic geometry has changed
a little because his clothes have been feeling tight lately. But he insists that his
shape is essentially the same as it always was.
Figure 23. A. Square has changed his geometry.
In order to explain what he means by “essentially the same” he reminds his

friends of what happened to their old classmate B. Rectangle. She was sitting on
her porch using super glue to repair a broken vase, when a huge dog ran at her
growling aggressively. In the shock of seeing it coming right at her, she squeezed
the tube of glue in her hand as she was bending over the vase. The dog’s owner
grabbed its leash just in time. But B. Rectangle’s head had touched the hand
that held the glue, attaching her head and hand together forever. A. Square even
brought out the article with “Before” and “After” photos that he had cut out and
saved from the local newspaper (see Figure 24). A. Square concluded his story by
saying, “Now that’s what I would call an essential change in shape”. The memory
of this sad story led to an awkward silence lasting several minutes. Nobody ever
teased A. Square about his shape again.
Comparing the types of bodily changes that happened to A. Square and B. Rec-
tangle motivates us to introduce a word to refer to the “essential” shape of an object
or space.
Definition. Two objects living in the same space have the same extrinsic
topology if one can be deformed to the other.
Deforming an object includes any kind of stretching, shrinking, twisting, bend-
ing, sliding, rotating, etc., but it does not include tearing, cutting, or gluing two
parts together, like what happened to B. Rectangle. Thus when A. Square put on
8. EXTRINSIC AND INTRINSIC TOPOLOGY 65
Figure 24. Headline News: Local rectangle gets bent out of shape
as a result of an out-of-control dog.
some weight he still had the same topology, but B. Rectangle’s accident changed
her topology because it attached her hand to her head.
We illustrate a doughnut and a coffee cup in Figure 25. Though it’s hard to
see at first, one can be deformed to the other. You can watch a video of this
deformation at http://www.youtube.com/watch?v=dwrhCSORERA
Figure 25. A doughnut has the same extrinsic topology as a cof-

fee cup.
Thus a doughnut and a coffee cup have the same extrinsic topology. This is why
some people say “topologists are people who can’t tell the difference between a
doughnut and a coffee cup”.
Let’s consider the knotted and unknotted tori illustrated in Figure 26. Try as
we might, we cannot deform the knotted torus on the left into the unknotted torus
on the right. This is because the two tori have different extrinsic topology.
Figure 26. These tori have different extrinsic topology.
On the other hand, Flatlanders living in the surface of one of these tori would
have no way of knowing whether their torus was knotted or not. This is because the
knotting has to do with how the torus is situated in space, rather than the surface
itself. Hence knotting is an extrinsic property of a torus. In fact, a Flatlander
would not even understand the concept of a knot. Imagine trying to explain what
a knot is to A. Square. This motivates our next definition.
Definition. Two objects living in the same space have the same intrinsic
topology if one can be obtained from the other by some combination of deforming
the object and cutting it apart then deforming and regluing it so that the same points
are glued together.
To clarify this definition, imagine that you have a bed frame made of wooden
boards that fit neatly together into slots. In order to move to a new apartment,
you have to take apart your bed so that it will fit out the door. You label each pair
of pieces that fit together, so that when you try to reassemble your bed in the new
apartment it will be easy to know what pieces go where. We do the same thing
when we cut apart an object, deform it in space, and then glue it back together. In
particular, we need to make sure that the pieces go together exactly the way they
were before.
We saw that a knotted and unknotted torus have different extrinsic topology.
However, in Figure 27 we cut open a knotted torus, undo the knot, and then reglue
the ends exactly as they were before to get an unknotted torus. This illustrates
that a knotted and unknotted tori have the same intrinsic topology.
Figure 27. A knotted torus can be cut apart, deformed, and then
reglued to get an unknotted torus.
In general, whether we’re talking about topology or geometry, an outsider has

more perspective on a space than an inhabitant of the space has. Thus if two
objects are situated in the same space in such a way that they are extrinsically the
same, then they are necessarily intrinsically the same. This is true regardless of
whether we’re talking about topology or geometry. We summarize the relationships
between extrinsic and intrinsic topology and geometry as follows:
Relationship between intrinsic and extrinsic sameness:

• If two objects have the same extrinsic geometry, then they are identical.
Hence they must have the same intrinsic geometry, intrinsic topology, and
extrinsic topology.
• If two objects have the same extrinsic topology, then they have the same
intrinsic topology.
You might learn the proofs of the above assertions in a more advanced geometry
or topology class.
9. USING LOOPS TO UNDERSTAND INTRINSIC TOPOLOGY 67
9. Using Loops to Understand Intrinsic Topology
If we could understand the extrinsic topology or geometry of our universe, we

would be able to understand its intrinsic topology or geometry. But unfortunately,
as inhabitants of the universe, we can only detect intrinsic properties. There is no
way for us to know if our universe is contained in a larger or higher-dimensional
space, let alone to understand how it might be situated in such a space (or for that
matter to know if some 4-dimensional beings are currently discussing how to teach
us something about our space).
Thus we have no choice but to think about our space intrinsically. For example,
we can think about what it might be like to live in a 3-torus. Astronomers might
even be able to design an experiment to try to to determine if our universe is indeed
a 3-torus. But even if it is a 3-torus, we would have no way of knowing whether or
not it is glued up in some 4-dimensional space. Thus if we want to have any hope
of understanding the geometry or topology of our own universe, we need to develop
intrinsic methods to gather information.
In the last section, we said that two objects in the same space have the same
intrinsic topology if one can be obtained from the other by some combination of
deforming the object and cutting it apart, then deforming and regluing it so that
the same points are glued together. This definition is fine for 3-dimensional beings
to determine if two surfaces are intrinsically the same or different. But it doesn’t
help the inhabitants of a surface to determine if their own surface is intrinsically
different from some other surface, since there is no way for them to know if their
surface can be cut apart, deformed in three dimensions, and reglued to get another
surface. It similarly doesn’t help us learn about the intrinsic topology of our own
space. The following definition can help us.
Definition. Any property of a space that can be detected without making any
measurements or making use of geodesics is said to be a topological property.
Such a property is intrinsic if it can be detected by inhabitants of the space and
extrinsic if it cannot be detected by inhabitants of the space.
Topological properties are useful because if one space has a particular topolog-
ical property and another space does not, then the two spaces cannot be topolog-
ically the same. In this section, we will see that the types of loops in a space is
a topological property that enables us to compare the intrinsic topology of many
different spaces.
We saw in Chapter 1 that A. Square and B. Triangle can distinguish a sphere
from a torus by using geodesics. Now we’ll see that they can do so even more easily
with loops that aren’t necessarily geodesics. In particular, suppose that A. Square
Figure 28. A. Square and B. Triangle can pull back any loop.
Figure 29. On a torus a loop can get stuck around the hole.
holds one end of a string while B. Triangle carries the ball of string with her as
she takes a walk around their space. When she returns to where she started, the
two of them simultaneously pull on the string. If their universe is a sphere, then
no matter where the string went, together they can pull it back as illustrated in
Figure 28
But if their universe is a torus and B. Triangle has walked around the hole, as
in Figure 29, the string will get stuck, so they can’t pull it back.
Definition. If a loop can be pulled in, we say the loop is shrinkable. Other-
wise, we say the loop is unshrinkable.
As we see in Figure 30, there are loops on a curved torus, a knotted torus,
and a flat torus that are unshrinkable. The existence of such loops is an intrinsic
topological property because it does not involve measurements, does not require
the loops to be geodesics, and can be detected by creatures within the space. Since
every loop in a sphere is shrinkable, a sphere is topologically different from a torus
(whether knotted, curved, or flat).
Figure 30. The curved torus, knotted torus, and flat torus all
have unshrinkable loops.
On a torus we can divide all of the loops into two distinct types: those that
are unshrinkable and those that bound a region. We can see in Figure 31 that
A. Square and B. Triangle can trap a 2-dimensional rabbit in a region bounded by
a loop. But an unshrinkable loop on the torus can’t trap anything.
Definition. If a loop in a surface bounds a region, then we say the loop is
separating. Otherwise we say the loop is non-separating.
In contrast with the torus, on a 2-holed torus there are three different types
of loops. In Figure 32, we see that a loop of type 1 is unshrinkable and is non-
separating, a loop of type 2 is separating and shrinkable, and a loop of type 3 is
separating and unshrinkable. We can see in Figure 32 that the third type of loop
separates the surface into two pieces each containing one of the holes. The existence
10. LOCAL AND GLOBAL PROPERTIES 69
Figure 31. On a torus some loops bound a region and others do not.
of separating unshrinkable loops is a topological property that a 2-holed torus has

but a 1-holed torus does not. This shows that a 1-holed torus and a 2-holed torus
have different intrinsic topology.
Figure 32. A 2-holed torus has three types of loops.
10. Local and Global Properties
In this section, we will consider geometry as well as topology and explain how
to divide properties into those that are local and those that are global.
Let’s begin by considering a point on a sphere. In a small region around this
point, we can draw various triangles and add up the sums of their angles. No
matter how small the region is, and no matter what triangles we choose, this sum
will always be greater than 180◦ . Thus we say that the property of having a triangle
with angle sums greater than 180◦ is a local property of the sphere. By contrast,
the total surface area of a sphere cannot be measured in a small region around a
point. Hence, we say that total surface area is a global property of the sphere. More
generally, we have the following definition.
Definition. A property of a space is said to be local if it can be detected
in arbitrarily small regions around points. Otherwise, the property is said to be
global.
We would like to be able to combine the concepts of local and global with the
concepts of geometry and topology and intrinsic and extrinsic to help us distinguish
spaces. For example, consider two flat tori whose gluing diagrams are of different
sizes. Their total area is different, and total area is a measurement that cannot be
detected in a small region around a point. Furthermore, this difference is intrinsic
because we aren’t considering the flat torus inside of a larger space. Thus we can
conclude that the two tori have different intrinsic global geometry. On the other
hand, any small region around a point on either a big or small flat torus will have
the same geometry as a similar small region of R2 . Thus both the big and small
flat torus have the same local intrinsic geometry as R2 .
In the last section, we showed that a torus is topologically different from a
sphere because the torus has unshrinkable loops but every loop on a sphere is
shrinkable. Let’s think about whether this loop property is local or global. If the
property is local, then we should be able to detect it in any region around a point.
However, no matter where we are on the torus, if we pick a small enough region,
it will not include the hole. Thus any loop in such a small region is shrinkable (see
Figure 33). This leads us to conclude that having unshrinkable loops is a global
intrinsic topological property.
Figure 33. Every loop in the dotted region is shrinkable.
On the other hand, if we look at a very small region around a point in either
a sphere or a torus we can deform the region to a disk. It follows that the sphere
and the torus have the same local intrinsic topology, but different global intrinsic
topology.
In order to compare the local intrinsic geometry of a sphere with that of a
curved torus, we consider the sums of angles of triangles. However, we have to
be careful not to refer to any specific triangle because a specific triangle cannot
be detected in an arbitrarily small region. For example, we saw that on a sphere
there is a triangle with three right angles. But such a triangle does not occur in a
small region of the sphere. Thus the existence of this particular type of triangle is
a global property of the sphere.
In order to consider the local geometry of a sphere and a torus, we want to
consider sums of angles of triangles, no matter how small. On a sphere the sum of
the angles of any triangle is always greater than 180◦ , no matter where we are on
the sphere and no matter how small the triangle is. On the curved torus, the story
is quite different. As we saw in Figure 20, in some regions the sum of the angles
of a small triangle is greater than 180◦ , but in other regions the sum of the angles
of a small triangle is less than 180◦ . This means that the local intrinsic geometry
of a curved torus is different in different regions. It follows that the sphere and the
curved torus have different local intrinsic geometry.
The local intrinsic geometry of the curved torus is difficult to study precisely
because it changes as we move from one point to another. We would like to focus
on spaces where the geometry doesn’t depend on where you are. In particular, we
make the following definition.
11. MANIFOLDS 71
Definition. A space is said to be homogeneous if its local intrinsic geometry

is the same at all points.
The word “homogeneous” may remind you of homogenized milk. In fact, dairies
homogenize their milk so that the cream will be evenly mixed and milk from dif-
ferent cows will be evenly mixed. This means that every sip you take of the milk
will taste the same, from one carton to the next and from the top of the carton to
the bottom of the carton. Thus the words “homogeneous” for spaces and “homoge-
nized” for milk are actually closely related, though their pronunciation is somewhat
different.
The curved torus is not homogeneous, because its geometry is not the same at
every point. On the other hand, a sphere, a flat torus, a flat infinite cylinder, and
Rn are all homogeneous spaces. Even the 3-torus that we constructed in Chapter 1
by gluing opposite sides of a cube is homogeneous.
Homogeneous spaces are easier to study than non-homogeneous spaces because
we only have to look at the local intrinsic geometry in regions around one point
in order to understand the local intrinsic geometry everywhere. In particular, if a
space is homogeneous, the sum of the angles of a triangle of a given size is the same
no matter where the triangle is.
11. Manifolds
Though cosmologists don’t know the global topology of our universe, it is rea-
sonable to assume that in any small region, the local topology is the same as in a
small region of R3 . This is certainly true near the Earth and in the areas of space
that have been explored.
Definition. A space is said to be an n-manifold if its local intrinsic topology
is the same as Rn at every point.
This means that a 2-manifold is a space where any point has a region around
it that can be deformed to a disk with that point at the center. In Figure 34
we illustrate some 2-manifolds. On the curved torus, around any point there is a
curved disk that can be deformed to an ordinary disk. In the flat torus the disks
may be split by the gluing wall, but they are still disks. The illustration on the
right is a plane with the origin removed. This is also a 2-manifold, but we will have
to choose smaller and smaller disks as we consider points closer and closer to the
missing origin. Note that because the origin is not in the space, you can’t choose
it as one of the points whose local intrinsic topology you’re considering.
Figure 34. A curved torus, a flat torus, and a plane with the
origin removed are all 2-manifolds.
Figure 35. Neither a disk nor two intersecting planes is a manifold.
A region around a point in R3 can be deformed to a ball with that point at the
center. So a space is a 3-manifold if every point in the space has a region around it
that can be deformed to a ball with the point at the center. For example, a 3-torus
is a 3-manifold, though some balls may be split by the gluing wall, just as some
disks on a flat torus were.
Figure 35 illustrates two spaces that are not manifolds. The disk on the left is
not a 2-manifold because a region around p on the boundary will look like a half of
a disk. While this region could be deformed to a round disk, the point p will still
be on the boundary of the region rather than in the center. So the local intrinsic
topology around this point is not the same as on a plane. The space on the right
in the figure consists of two intersecting planes. This space is not a 2-manifold
because a region around the point q will look like two intersecting disks rather than
a single disk.
A 2-manifold is often referred to more simply as a surface. Unfortunately, there
is no such convenient word to refer to a 3-manifold. So we have to call a 3-manifold
a 3-manifold.
12. Assumptions about Universes
While A. Square cannot know for certain the shape of his universe, there are
some assumptions that seem reasonable for him to make. For instance, it would
be very strange if his universe had two or more disconnected pieces, preventing
him from ever visiting some parts of his universe without the magical powers of
teleportation. Thus, it seems reasonable for A. Square to assume that his universe
consists of a single connected space such that given enough time, sufficiently super-
duper technologically advanced transportation, and the appropriate visa, he could
travel to any part of it. Furthermore, it seems unlikely that his universe has an edge
which forms an invisible wall that nobody can pass through. So it’s safe to assume
that his universe isn’t a disk or any other 2-dimensional shape with a boundary.
It also seems unlikely that Flatland contains two intersecting planes, because at
the line of intersection a vertical shape could cut through the middle of a horizontal
shape. For example, in Figure 36, A. Square has a terrible stomach ache because
C. Oval is passing through his stomach. Perhaps all of our undiagnosed aches and
pains are caused by other 3-dimensional creatures passing through our insides in
this way. But this seems unlikely. So it’s reasonable to assume that Flatland is a
2-manifold.
12. ASSUMPTIONS ABOUT UNIVERSES 73
Figure 36. A. Square has a stomach ache because C. Oval is

passing through him.
Figure 37. When B. Triangle takes a walk on a curved torus her

angles change.
If Flatland were a curved torus, other problems could occur. In particular,

as we’ve seen, if B. Triangle took a walk, her area and angles could dramatically
change causing serious internal pain, as illustrated in Figure 37 . It would be pretty
strange to live in a universe where every time you walked in a certain area you got
a stomach ache. In order to avoid this kind of situation, it also seems reasonable
to assume that Flatland is homogeneous.
The above assumptions about Flatland make sense for any universe. This gives
us the following list of assumptions that we will make about all universes, including
our own.
List of Assumptions about Universes:

(1) Every universe is connected.
(2) Any universe where n-dimensional beings live is an n-manifold.
(3) All universes are homogeneous.
Recall that the creatures in a universe can only study properties of their space
that are intrinsic, since they cannot leave their universe to get an extrinsic per-
spective. For this reason, we focus on trying to discover the intrinsic topology and
geometry of our universe, rather than assuming our space is contained in a higher-
dimensional space and trying to imagine what its extrinsic topology and geometry
could be.
By our second assumption about universes every n-dimensional universe has the
local intrinsic topology of Rn . Thus, what remains to discover is its global intrinsic
topology. While it would be nice to also know about the global intrinsic geometry
of our universe, often such information is hard to find because it requires measuring
something very large like the total volume of the universe. By contrast, since we’re
assuming our universe is homogeneous, understanding the local intrinsic geometry
at one point will tell us the local intrinsic geometry at every point. Thus our efforts
might be better spent exploring the local intrinsic geometry of our universe rather
than trying to determine its global intrinsic geometry.
In summary, we are most interested in learning about the global intrinsic
topology and the local intrinsic geometry of our universe.
13. A 2-Dimensional Sphere with a Point Removed
Now that we have developed our vocabulary, we would like to explore the
topology of some specific spaces in more detail.
As our first example we will show that a sphere with a point removed has the
same extrinsic topology as a plane when considered as objects in R3 . We begin by
removing a point from a sphere in Figure 38. We’ve stretched open the hole a little
bit to make it easier to see, and we use a dotted boundary to indicate that there
is no actual boundary after removing the point. Next, we stretch out the surface,
and flatten it so it lies in a plane (as illustrated in the center and right side of
Figure 38). In this way, we get a disk whose boundary is not included. We call this
an open disk. We can further stretch or shrink the open disk so that it has radius
1.
Figure 38. A sphere with a point removed can be stretched out

to get a flat disk with no boundary.
The next step is to subdivide the open disk into rings with radius 12 , 34 , 78 , and
so on forever (see the left side of Figure 39). This makes the open disk look like a
target with an infinite number of rings that get thinner and thinner as we approach
but never reach the boundary. In particular, the bulls-eye has radius 12 , the first
ring has width 14 , the next ring has width 18 , etc.
We will now stretch the open disk so that it covers the whole plane. Leaving
the bulls-eye fixed, we stretch the first ring so that its width is 12 , which is the
same as the radius of the bulls-eye. Then we stretch the second ring so it has the
same width and so on, stretching one ring after another so that each ring has the
same width as the previous one. We do this forever to get a target with an infinite
number of concentric rings all of the same width as illustrated on the right side of
Figure 39. Since there are an infinite number of rings with the same width, the
target will now fill up the entire plane.
To summarize, we have removed a point from a sphere, deformed it into an
open disk, and then stretched the open disk to cover the entire plane. It follows
13. A 2-DIMENSIONAL SPHERE WITH A POINT REMOVED 75
…
… …
…
Figure 39. We subdivide an open disk into concentric rings, then
stretch the rings so that they fill up the plane.
that a sphere with a point removed has the same extrinsic topology as a plane.
This implies that they have the same intrinsic topology as well.
In order to see this another way, we introduce the method of stereographic
projection. We start with a sphere tangent to a plane R2 so that it is only touching
the plane at the South Pole. Then we extend a line from the North Pole through
each point on the sphere to a point on the plane as illustrated in Figure 40. This
assigns to each point on the sphere other than the North Pole, a point in the plane.
We can imagine that we have a very bright light at the North Pole and each point
on the sphere casts a shadow on the plane. All of the circles on the sphere that
are are parallel to the equator cast circular shadows in the plane. For example, the
equator in Figure 40 casts the circular shadow shown. The closer such a circle is
to the North Pole, the larger its circular shadow will be.
Figure 40. Every point in S 2 except the North Pole can be paired
with a point in R2 .
We can’t extend this projection to the whole sphere because there’s no point
on the plane for the North Pole to go. To solve this problem, we send the North
Pole to a point outside of the plane that we call the “point at ∞”. You get to this
point if you go infinitely far in any direction in R2 . In this way, we have assigned
every point on the sphere to a point in the plane together with ∞. We write this
Figure 41. Gluing together these two regions gives us a sphere

with a point removed.
as S 2 = R2 ∪ {∞}. Thinking about S 2 this way allows us to imagine that R2 is

contained in S 2 , because if we remove the point at the North Pole from S 2 , we are
left with R2 .
We would like to explain to A. Square that a sphere with a point removed
has the same intrinsic topology as a plane. He can understand Figure 39 together
with the argument about how we stretch an open disk into a plane, but he can’t
understand Figure 38. So we’ll have to use the gluing diagram for a sphere that we
introduced in Chapter 1. Figure 41 illustrates the two disks that are glued together
to get a sphere, but now we have removed a single point from the center of one
of the disks. A. Square remembers the gluing diagram for a sphere, so he has no
trouble understanding Figure 41.
We can deform the light grey region in R3 as follows. We pull the hole up to
get an inverted funnel, stretch open the hole, and then we push it down into the
plane to get an ring with no outer boundary, as illustrated in Figure 42.
Figure 42. We deform the disk with a point removed to get a

ring with no outer boundary.
Finally, we glue the inner boundary of the ring to the outer boundary of the
darker disk as illustrated in Figure 43. Since the darker disk fills the hole of the
ring, we end up with an open disk.
A. Square can understand the gluing in Figure 43, but the deformation in
Figure 42 took place in 3-dimensional space, and hence A. Square can’t follow it.
So instead we tell him that we turn the ring inside out in 3-dimensional space so
that the inside and outside are switched. In order to understand this he tries to
think about something analogous that is one dimension down. He has the idea that
if a line in Flatland has an arrow on it, he can turn the line around so that the arrow
will switch directions (see Figure 44). But the idea of switching the direction of an
arrow would seem impossible to a 1-dimensional creature like A. Dash. Keeping
this in mind, A. Square is willing to accept the idea that the inside and outside of
a ring can be switched in 3-dimensional space.
14. A 3-DIMENSIONAL SPHERE WITH A POINT REMOVED 77
Figure 43. We glue the disk into the hole in the ring to get an
open disk.
Figure 44. An arrow can be turned around in Flatland, but not in Lineland.
This convinces A. Square that a sphere with a point removed has the same
intrinsic topology as an open disk. We then use Figure 39 to show him that an
open disk has the same intrinsic topology as a plane. Thus he agrees that a sphere
with a point removed has the same intrinsic topology as a plane.
Note that it would be pointless to try to convince A. Square that in R3 the
sphere with a point removed has the same extrinsic topology as a plane, since there’s
no way that he can know anything about the extrinsic topology of a 2-dimensional
space.
14. A 3-Dimensional Sphere with a Point Removed
Using ideas similar to those in the last section, we now see as follows that S 3
with a point removed has the same intrinsic topology as R3 . To do this, we need
to imagine 3-dimensional analogues of Figures 39, 41, 42, and 43. First recall,
that S 3 is obtained by gluing two balls together along their boundaries, just as
S 2 is obtained by gluing together two disks along their boundaries. The topology
of these glued up balls is the same as S 3 , but the geometry is different since the
balls had to be deformed to fit into our 3-dimensional space. But we are currently
only interested in the topology, so the geometric differences are not important to
us here.
We remove a point from the center of one of the gluing balls and stretch open
the hole so it’s easier to see, as we did for the disks in Figure 41. You can visualize
this as an avocado with nothing where the pit would normally be. We now visualize
the gluing space for S 3 with a point removed as one solid ball and one hollow ball
that looks like an avocado which has an outer boundary (where the skin is) but no
inner boundary (where the pit would begin).
Next we accept the fact that the hollow ball can be turned inside out in 4-
dimensional space so that it has no outer boundary but has an inner boundary.
This deformation is as impossible for us to picture just as it was impossible for
A. Square to picture how we turned the ring inside out in Figure 42. But it seems
reasonable that such a deformation can take place in 4-dimensional space since we
saw how to turn a ring inside out in 3-dimensional space. Finally, we glue the
solid ball to the inner boundary of our hollow ball, as we did with the disk and the
ring in Figure 43. This gives us an open solid ball, analogous to the open disk in
Figure 43. In particular, we now have a ball whose boundary is not included.
Next we subdivide our solid ball into a ball of radius 12 , and concentric hollow
balls of smaller and smaller widths like the concentric rings of smaller and smaller
widths on the left side of Figure 39. Finally, we stretch these hollow balls so that
they all have width 12 , as we did with the concentric rings in Figure 39. Since there
are infinitely many concentric hollow balls all of width 12 surrounding the inner ball
of radius 12 , together they cover all of R3 just as the bulls-eye and rings on the
right side of Figure 39 covered all of R2 . This shows us that the space we get by
removing a point from the gluing diagram of S 3 has the same intrinsic topology as
R3 .
Recall that in the last section we observed (with the help of stereographic
projection) that in addition to thinking of R2 as S 2 with a point removed, we can
think of S 2 as R2 together with a point at infinity. We wrote this as S 2 = R2 ∪{∞}.
We can now think of both R3 as S 3 with a point removed, and S 3 as R3 together
with a point at infinity. We write this as S 3 = R3 ∪ {∞}. You get to the point
at ∞ in S 3 by going infinitely far in any direction. This way of visualizing S 3
is convenient because it enables us to imagine that R3 is contained in S 3 , just as
visualizing S 2 in the analogous way enabled us to imagine that R2 is contained in
S 2.
15. Curves on a Torus
In this section, we want to use loops to explore the topology of a torus in more
detail. Recall that in Figure 31 we saw that a torus has two types of loops, those
which are separating and those which are non-separating. It may seem like this is
all there is to know about loops on a torus. However, it turns out that on a torus,
there are actually infinitely many non-separating loops whose extrinsic topology
are all distinct.
In order to analyze loops on a torus, we begin by picking two special loops that
we’ll refer to. A circle that runs once around the curved torus the long way (as
in Figure 45) is called a longitude, while one that runs once around the torus the
short way is called a meridian. We are thinking of a longitude and meridian as
topological objects in the torus. So any loop which can be deformed to a longitude
is also a longitude, and any loop which can be deformed to a meridian is also a
meridian.
In order to remember which circle on the curved torus is a meridian and which is
a longitude, you can use the following trick. If you want to make a bagel sandwich,
you first have to cut the bagel open the long way, along an inner and outer longitude
so that you can put meat, cheese, lox, or tofu together with whatever condiments
you like between the two halves. Finally, after closing up your sandwich, you cut
15. CURVES ON A TORUS 79
Figure 45. A meridian and a longitude on a curved and flat torus.
it in half by slicing along two meri dians which makes you merry because you can
now eat your sandwich.
In the flat torus pictured on the right of Figure 45, the sides of the square
represent loops, but both have the same length. We designate the loop on the left
and right sides of the square as the meridian and the loop on the top and bottom
sides as the longitude. However, it doesn’t actually matter which is which, since we
can rotate the square by 90◦ to interchange the pairs of sides. This is an important
difference between the curved torus, which is an object in R3 , and the flat torus
which has no extrinsic topology.
We are now interested in looking at loops on the torus which don’t intersect
themselves. We’ll call such loops on the torus curves. We begin by considering
the grey curves illustrated in Figure 46. These curves wrap once around the torus
meridionally and once around the torus longitudinally. However, by following the
arrows we can see that one curve goes around the back of the torus from the outside
longitude to the inside longitude, while the other curve goes around the back of
the torus from the inside longitude to the outside longitude. Because one of these
curves cannot be deformed to the other on the torus, the two curves have different
extrinsic topology.
Figure 46. Two topologically distinct loops on a torus.
We orient the longitude and meridian, and then we refer to the curve whose
orientation agrees with the orientations of both the meridian and the longitude as
a (1, 1)-curve and refer to the curve whose orientation agrees with the longitude
but not the meridian as a (−1, 1)-curve. Of course, which is which depends on how
we choose our arrows.
More generally, for a given pair of positive or negative numbers p and q, we
would like to define a (p, q)-curve on the torus as a curve which goes p times around
the torus meridionally and q times around the torus longitudinally. For example,
consider the grey curve illustrated in Figure 47. If you trace it out with your
finger, you can see that it winds around the torus three times meridionally while
Figure 47. A (3, 2)-curve on a torus.
winding two times around the torus longitudinally. Thus this curve is referred to
as a (3, 2)-curve.
You should notice that the grey curve intersects the meridian in two points and
intersects the longitude in three points (marked with dots in the picture). Keep
in mind that the dashed arcs are on the under side of the torus and hence are not
intersecting the longitude. Furthermore, there is no way to deform the grey curve
to intersect the meridian or longitude in fewer points. In general, if a curve on the
torus intersects the meridian in p points and the longitude in q points and can’t be
deformed to intersect either in fewer points, then the curve must be a (p, q)-curve.
16. How to Draw a (p, q)-Curve on a Flat Torus

You may find that it’s not so easy to draw a specific (p, q)-curve on the curved
torus. For example, you might pause here and try to draw a (3, 4)-curve on a curved
torus. By contrast, we will now learn an easy method for drawing (p, q)-curves on
a flat torus.
We begin by explaining how to draw a (3, 2)-curve on a flat torus. We will use
the observation that a (3, 2)-curve intersects a longitude three times and a meridian
two times. Thus in our flat torus the (3, 2)-curve will have three points of intersec-
tion along the top and bottom sides of the square (representing the longitude) and
two points of intersection along the left and right sides of the square (representing
the meridian). In order to draw the (3, 2)-curve on the flat torus, we first label the
points on the edges of the square where we want the (3, 2)-curve to intersect the
meridian and longitude as illustrated in Figure 48.
Since a (3, 2)-curve never intersects itself, it must be a collection of non-
intersecting paths that start and end at the labeled points on the top, bottom,
Figure 48. A flat torus with points labeled to indicate where the
(3, 2)-curve intersects a meridian and a longitude.
16. HOW TO DRAW A (p, q)-CURVE ON A FLAT TORUS 81
and sides of the gluing square. Hence the paths making up our (3, 2)-curve must
adhere to the following rules:
Rule 1: Every labeled point must be an endpoint of a path.

Rule 2: No pair of paths intersect.
Since every labeled point is an endpoint of a path and together the paths make
a loop, it doesn’t matter which labeled point we begin with. So let’s say we choose
a2 on the bottom edge as a starting point for our first path. The first path must
go from a2 on the bottom edge to some labeled point either on the top or on one
of the sides.
First, let’s see what happens when the path from a2 on the bottom edge goes to
a3 on the top edge. Since a3 on the top edge is glued to a3 on the bottom edge, the
second path must start at a3 on the bottom edge. Any path from a3 to a labeled
point to the left of a3 would intersect the first path, as we see in the left image of
Figure 49. Thus, the ending point of the second path must be either b1 or b2 on
the right edge, as illustrated on the right in Figure 49.
Figure 49. Problems that could arise if we choose the wrong end-
ing point for the second path.
However, if b1 is the ending point of the second path, then any path starting at
b2 on the right side would have nowhere to go without intersecting the path from a3
to b1 . A similar problem would occur if b2 were the ending point of the second path.
Thus, once we choose a3 as the ending point of the first path, we are eventually
forced to isolate one of our labeled points so that it cannot be an endpoint of any
path (as illustrated on the right of Figure 49) This contradicts Rule 1. So a3 cannot
be the ending point of the first path.
An analogous argument on the left edge of the square shows that a1 cannot be
the ending point of the path from a2 . Furthermore, if we chose a2 on the top edge
as the ending point of the path from a2 on the bottom, we would again violate one
of the rules. Would it violate Rule 1 or Rule 2?
Thus, starting at a2 on the bottom edge, the first path must end at one of the
labeled points on the side of the square, since every other possibility will result in
a violation of Rule 1 or Rule 2. Moreover, if the path from a2 ended at b2 on either
side, then we would isolate one of the points on the bottom edge. (Which one in
each case?) Therefore, we have no choice but to make b1 be the ending point of the
first path, though it doesn’t matter if it’s the b1 on the right edge or the b1 on the
left edge.
We choose b1 on the right edge as the ending point of the first path (see the first
image in Figure 50). Then b1 on the left edge is the starting point of the second
path. In order to avoid isolating any labeled points or intersecting the path from
a2 , we must then choose a1 on the top edge as the ending point of the second path
(see the second image in Figure 50). This means that the third path will start at a1
on the bottom edge. Again, keeping with our strategy of not isolating any point or
intersecting any existing paths, a3 on the top edge must be the ending point of the
third path (see the third image in Figure 50). What would have happened if b2 on
the left edge were chosen as the ending point of the third path? Continuing in this
way, the fourth path has starting point a3 on the bottom edge and ending point b2
on the right edge (see the fourth image in Figure 50). The fifth and final path has
starting point b2 on the left edge and ending point a2 on the top edge, and we have
come full circle (see the fifth image in Figure 50). In this way, we obtain the set of
paths making up a (3, 2)-curve.
Figure 50. How we construct the paths that make up a (3, 2)-curve.
Note that if we had started with b1 on the left edge as the ending point of the
first path and continued from there, we would end up with the mirror image of
the last drawing in Figure 50. Also if we had made our labeled points equidistant
on each edge, we could make the paths in the square parallel line segments as in
Figure 51. What would the slope of these parallel line segments be?
Try to use the above method to draw a (3, 4)-curve on a flat torus. What
happens if you try to draw a (2, 2)-curve on the flat torus? What about a (4, 6)-
curve? Explore these questions by placing the appropriate number of points along
the edges of a square and follow a procedure similar to the one above. Can you
deduce anything about the relationship between the integers p and q in a (p, q)-curve
on the torus?
17. LINES IN THE EXTENDED DIAGRAM OF A FLAT TORUS 83
Figure 51. A (3, 2)-curve drawn on the flat torus with lines of
equal slope. What is the common slope of these lines?
17. Lines in the Extended Diagram of a Flat Torus
Another way to explore curves on a flat torus is to start with a line in an

extended diagram and then see what curve it represents on the flat torus. For
example, consider the line with slope 21 illustrated on the left in Figure 52. When we
draw this line in the flat torus it becomes just two segments, because the intersection
of the grey line in Figure 52 with the square in the top and bottom rows of the
extended diagram represent the same segment on the flat torus. The two segments
in the flat torus together form a (2, 1)-curve because the loop intersects a longitude
of the flat torus in two points and a meridian of the flat torus in one point. Notice
that the points in the lower left corner and the top right corner of the flat torus
count as both one intersection point with a meridian and one intersection point
with a longitude.
Figure 52. A line with slope 21 in the extended diagram represents

a (2, 1)-curve on the flat torus.
In general, we can consider any line in the extended diagram with slope pq where
p and q are integers with no common factors. Even though our line is infinitely
long, in the flat torus it has only a finite number of segments, and together these
segments form a (p, q)-curve.
But what happens if we consider a line in the extended diagram whose slope
cannot be written as pq where p and q are integers. In this case, no pair of points on
the line will represent the same point in the flat torus. So the line in the extended
diagram will give us infinitely many different segments in the flat torus. These
segments form an infinitely long path rather than a closed loop in the torus. Thus,
even though a torus has finite area, it contains lines that never end just like lines
in the plane never end. What would happen if A. Square went for a walk on such
a line? Would he ever return to his starting point?
18. Exercises
1. Consider the space R2 − {(0, 0)} illustrated in Figure 53. Is there a geodesic
between the points p and q? If so, find it. If not, why not?
2. Draw the extended diagram for a flat infinite cylinder and explain how we know
that the sum of the angles of every triangle in the flat infinite cylinder is 180◦ .
3. Make a conical party hat and draw two points on it. Does the distance between
the points change when the hat is cut open into a flat surface? What does a
geodesic segment on the hat look like when the hat is cut open? Is a circle
parallel to the circular base of the hat a geodesic?
4. Using one or more pieces of paper and some tape, can you make a surface
containing a triangle whose angles add up to more than 180◦ ?
5. Can you make a surface out of paper and tape with a triangle whose angles
add up to less than 180◦ ?
6. Can you make a surface out of paper and tape that contains some triangles
whose angles add up to 180◦ , some which add up to less than 180◦ , and some
which add up to more than 180◦ ? Explain your constructions.
7. Prove that the ratio of the perimeter of a hemisphere to its area is different
from the ratio of the perimeter of a disk to its area. Conclude from this that
the intrinsic geometry of a hemisphere is different from that of a disk.
8. In Figure 40, the shadows cast by latitudinal circles are circles in the plane.
What kinds of shadows are cast by longitudinal circles?
9. If the sphere in Figure 40 were the Earth, compare the sizes of the shadows
cast on the plane by Antarctica, Hawaii, Greenland, and the polar ice caps at
the North Pole.
18. EXERCISES 85
10. Is the property of being homogeneous global or local? Is the property of being
an n-manifold global or local?
11. Consider the circular strips in Figure 54, which you should imagine were created
using three identical rectangles of paper. Compare their extrinsic geometry and
topology. Then make a table comparing their intrinsic properties (geometric
and topological, local and global).
12. Compare the local and global intrinsic geometry of a disk and a conical party
hat. Then compare their local and global intrinsic topology. How could Flat-
landers distinguish between these two spaces?
13. Give examples (different from the ones given in the chapter) of pairs of spaces
with the following properties.
(a) The same local intrinsic topology, but different local intrinsic geometry.
(b) The same local intrinsic geometry, but different global intrinsic topology.
(c) The same global intrinsic topology, but different global extrinsic topology.
(d) The same local intrinsic geometry, but different global intrinsic geometry.
(e) The same global intrinsic geometry, but different extrinsic geometry.
14. Is the surface of a cube homogeneous? Why or why not?
15. Compare the local and global intrinsic geometry and topology of a 3-torus made
from a room whose length is longer than its width and one made from a room
whose length is equal to its width.
16. Do two spheres of different sizes have different global intrinsic geometry? Do
two spheres of different sizes have different local intrinsic geometry? Explain
your conclusions.
17. How would you define a 1-manifold? Give two examples of 1-manifolds and
explain how you know they are different.
18. Do the two objects in Figure 55 have the same extrinsic topology in R3 ? Do
they have the same extrinsic geometry in R3 ?
19. Show that every capital letter in our alphabet can be deformed to one of the
capital letters: A, B, or C. Thus topologically, there are only three capital
letters.
CHAPTER 4
Orientability
Topics:
• Möbius strips
• Orientation reversing paths
• The Klein bottle
• Tic-Tac-Toe on a Klein bottle
• The 3-dimensional Klein bottle
• The projective plane
• Cross-caps
• Projective 3-dimensional space
• Sidedness and orientability of surfaces in 3-manifolds
1. The Möbius Strip
The person you see in the mirror each morning looks quite similar to you, and
yet when you raise your right hand the person in the mirror raises his or her left
hand. Is it possible to move into a position so that when you raise your left hand
your mirror image does the same? What if the mirror is on the ceiling? What if
you are standing on your head?
If you look in the mirror while wearing a t-shirt that says “Math is Cool”, you
will see yourself with a t-shirt that says “looC si htaM”. If you leave your t-shirt
on and you take a walk, no matter where you go, your t-shirt will never change
from saying “Math is Cool” to saying “looC si htaM”.
We want to understand what it would be like to live in a universe where you
could go on a trip and come back as your mirror image. We begin by considering
what happens to A. Square when he takes a long walk around Flatland. He puts
on his walking hat and leaves town. He never crosses the road or changes direction.
After walking for longer than ever before, he sees that he is approaching his home-
town from the opposite direction. However, when he gets close to the doughnut
shop with his favorite cinnamon twists, the sign in front of the shop looks different
than usual (see Figure 1).
Figure 2 shows a picture of the road that A. Square walked along. A. Square
does not feel like he has changed at all, but he thinks that everything else has
become its mirror image. On the other hand, his friends think that he has become
mirror reversed. It’s not just that he is upside down. He could turn around, but
then he would be facing the opposite direction. The only way for him to fix himself
87
88 4. ORIENTABILITY
Figure 1. The sign in front of A. Square’s favorite doughnut shop.
Figure 2. Walking along this path reverses A. Square.
is to go for a second walk all the way around Flatland. But he is certainly too tired
for that.
The road that A. Square has walked along is what is called a Möbius strip
(also known as a Möbius band ). You can make a Möbius strip out of a belt by
adding a half-twist and then buckling the inside to the outside. We illustrate a
gluing diagram for a Möbius strip in Figure 3. Notice that the bottom corner of
the gluing diagram is attached to the top corner of the other side. As a result, the
boundary of the Möbius strip is a single circle. You should check this for yourself
by tracing your finger along the boundary in Figure 2.
Figure 3. Gluing diagram of a Möbius strip.
Once we glue the two edges of the strip together, the “seam” that we glued
along disappears completely just as it does on the flat torus. For the Möbius strip
this means that there is no specific location where the mirror-flip happens all of a
sudden. A. Square had to make a complete circle around the Möbius strip for this
phenomenon to occur. If he travelled part of the way or even all of the way and
then retraced his steps, he would not be reversed.
2. ORIENTATION REVERSING PATHS 89
2. Orientation Reversing Paths
In order to talk about paths like the one A. Square traveled along and the kinds
of spaces that contain them, we introduce the following definition.
Definition. We say a circular path in an n-dimensional space is orientation
reversing if an n-dimensional being can walk along it to become mirror reversed.
We illustrate the importance of the dimension in the definition as follows. The
path that A. Square followed in Figure 2 was an orientation reversing path in
Flatland. In Figure 4, A. 3D-Girl is taking a walk along a Möbius strip in R3 . Her
left hand is swollen because it was stung by a bee. Her right hand is OK, but she’s
keeping it in her pocket to protect it from any bees she might encounter. Notice
that she comes back upside down and underneath her starting point, but she is not
reversed. In particular, her left hand is still swollen and her right hand is still in
her pocket. Thus the orientation reversing path that A. Square walked along in the
Möbius strip is not orientation reversing for 3-dimensional people like A. 3D-Girl.
Figure 4. A. 3D-Girl walks along a Möbius strip but is not mirror reversed.
We have a special name for spaces like a Möbius strip that contain orientation
reversing paths.
Definition. An n-dimensional space is said to be non-orientable if it con-
tains a path which is orientation reversing for n-dimensional creatures. If it con-
tains no such path, then it is said to be orientable.
90 4. ORIENTABILITY
When we look at a gluing diagram for a surface, it is generally hard to tell

what the surface would look like if we glued it up. Nonetheless, it is possible to
check if the surface is non-orientable just by looking for an orientation reversing
path. Any path that does not cross the boundary of a gluing diagram cannot be
orientation reversing since such a path would be contained in a region of R2 . This
means that we just have to check whether a path that goes out one edge and in
another edge is orientation reversing. To find such a path, we look for the gluing
diagram of a Möbius strip within our gluing diagram. Any Möbius strip will contain
an orientation reversing path. For example, in Figure 5 we see that the grey strip
is a a gluing diagram for a Möbius strip, and hence we know the surface containing
it is non-orientable.
Figure 5. A non-orientable gluing diagram containing a Möbius strip.
In what follows we see that there is no good way to define left and right in
a non-orientatible surface. In Figure 6, A. Jester takes the same walk around the
Möbius strip that A. Square did. At his starting position, he is shown with his
left arm and left leg painted grey and his right arm and right leg painted black.
However, when he comes back to his starting position after one trip around the
Möbius strip, his left arm and left leg are painted black and his right arm and right
leg are painted grey. Since left and right can morph into one another in this way,
it is meaningless to define left and right in a surface containing a Möbius strip.
Figure 6. When A. Jester returns home, his left and right are reversed.
3. The Klein Bottle
Though a Möbius strip is a non-orientable space, it’s not a 2-dimensional uni-

verse because it has a boundary. In order to get rid of the boundary of a Möbius
strip, we can glue together the top and bottom sides of the gluing diagram in Fig-
ure 3. The surface that we obtain is called a Klein bottle, named after Felix Klein
3. THE KLEIN BOTTLE 91
who introduced this surface in 1882. This surface, denoted by K 2 , is illustrated in

Figure 7.
Figure 7. A Klein bottle is a non-orientable 2-manifold.
Notice that the gluing diagram for the Klein bottle in Figure 7 is a square,
whereas the gluing diagram for a Möbius strip in Figure 5 is a long thin rectangle.
If we had used a rectangle as a gluing diagram for the Klein bottle, its intrinsic
topology and local intrinsic geometry would be the same as in Figure 7. However,
its global intrinsic geometry would be different. Whether defined on a square or a
rectangle, since the Klein bottle contains a Möbius strip, it contains an orientation
reversing path and hence is non-orientable. Also, since the Klein bottle has no
boundary, it’s a 2-manifold. Thus, Flatland could theoretically be a Klein bottle.
While there is nothing wrong with defining the Klein bottle intrinsically as
a glued up square, you may have the feeling that you don’t really understand it
because you haven’t seen a picture of it glued up in 3-dimensional space. But before
attempting to glue up the Klein bottle, let’s look at Figure 8 to remember how we
glued up a torus.
Figure 8. This is how we glued up a torus.
We start to glue up the Klein bottle in Figure 9 by gluing the two arrows that
point in the same direction as we did for the torus in Figure 8. Next, we would
like to glue together the circles on the ends of the cylinder in the central drawing
of Figure 9. However, when we try to do this we run into a problem because the
arrows don’t match up as they did in Figure 8.
?
Figure 9. Our attempt to glue up a Klein bottle.
In order to align the arrows, the surface would have to pass through itself so
that one circle could approach the other from the back. (You might try this for
yourself with a slinky.) But if the surface intersects itself, then it would not be
a 2-manifold and hence not a legitimate universe. We can solve this problem if
92 4. ORIENTABILITY
the Klein bottle is in R4 rather than R3 . In this case, we can bring one end of
the cylinder up into the fourth dimension so that it can pass into the inside of the
cylinder without intersecting the cylinder (just as a 4-dimensional creature could
remove my heart without piercing my skin). We cannot draw an accurate picture
of this in 3-dimensional space, let alone on this 2-dimensional page. The best we
can do is to draw Figure 10 where the Klein bottle appears to intersect itself.
Figure 10. Gluing up a Klein bottle in 4-dimensional space.
When looking at Figure 10, you should keep in mind that there really is no
“circle of intersection”. Analogously, imagine holding two crisscrossing pencils up
to a light so that their shadows on the wall look like they intersect when in reality
the two pencils don’t intersect at all. When we draw the Klein bottle in Figure 10,
we are discarding one dimension’s worth of information, just like when we look at
the shadow of two crisscrossing pencils.
Notice that the local intrinsic geometry of the glued up Klein bottle in Figure 10
is non-homogeneous just like that of the curved torus. On the other hand, the
gluing diagram of the Klein bottle in Figure 7 is nice and flat and homogeneous
just like the flat torus. So although trying to imagine a curved Klein bottle sitting
in 4-dimensional space is an interesting mental exercise, we prefer to use the gluing
diagram of the flat Klein bottle as in Figure 7. So from now on whenever we refer to
the Klein bottle, you should think of the gluing diagram rather than of the picture
illustrated in Figure 10.
4. The 3-Dimensional Klein Bottle
Now that we understand the gluing diagram for the 2-dimensional Klein bottle,
we would like to create a gluing diagram for a 3-dimensional Klein bottle K 3 . We
start with a rectangular room where the left wall is glued to the right wall and the
front wall is glued to the back wall, just as we did for the 3-torus. But in this case,
we glue the ceiling to the floor with a left-right flip, rather than gluing it straight as
we did for the 3-torus. In Figure 11, we indicate how the different pairs of walls are
5. TIC-TAC-TOE ON A TORUS AND KLEIN BOTTLE 93
Figure 11. The gluing diagram for a 3-dimensional Klein bottle.
glued with different types of arrows. Note that the arrows on the top and bottom
are mirror images of each other.
Let’s try to imagine what it would be like to live in a 3-dimensional Klein bottle.
First recall that in a 3-torus T 3 when A. 3D-Girl floated through the ceiling, she
come back through the floor. Now in K 3 when she floats through the ceiling and
comes back through the floor, she is mirror reversed. For example, in Figure 12,
when A. 3D-Girl comes back through the floor she is still facing forward, but her
left and right sides have switched. Also, notice that she floats through the left part
of the top face and comes back through the right part of the bottom face.
Figure 12. When A. 3D-Girl goes through the ceiling, her left
and right sides are reversed.
It’s pretty strange to think about living in a non-orientable space. There would
be no way to distinguish left and right since your left and right hands could switch
as they did for A. 3D-Girl in Figure 12. Imagine that you take a clock with you
when you walk along an orientation reversing path. What will the clock look like
when you return? In what direction would the hands be moving?
5. Tic-Tac-Toe on a Torus and Klein Bottle
We can play tic-tac-toe on a flat Klein bottle, as we did on a flat torus. But now
it’s a little trickier. It’s useful to draw an extended diagram to see all the different
ways to have three in a row. To check that you’ve drawn the extended diagram
correctly, check that the same squares are adjacent in the extended diagram and are
94 4. ORIENTABILITY
Figure 13. A Klein bottle tic-tac-toe board together with an ex-

tended diagram.
adjacent in the gluing diagram. For example, in both the gluing diagram and the
extended diagram of Figure 13, if you go up from square 3 you enter square 7, and
if you go down from square 9 you enter square 1. Although the extended diagram
continues infinitely in all directions, we only draw nine boards because the pattern
repeats.
Before we analyze Klein bottle tic-tac-toe, let’s remember what we learned
about tic-tac-toe and torus tic-tac-toe in Chapter 1. In particular, we learned that
two moves are said to be equivalent if the strategy for the rest of the game after one
of the moves is analogous to the strategy for the rest of the game after the other
move. For example, in regular tic-tac-toe, a first move in one corner is equivalent
to a first move in any of the other three corners; and a first move in the middle of
one side is equivalent to a first move in the middle of any side. Thus in regular tic-
tac-toe, there are exactly three inequivalent moves for the first player: the center,
a corner, and the middle of a side.
What’s important in any kind of tic-tac-toe is which squares are adjacent to
which. If you did the exercises on torus tic-tac-toe in Chapter 1, then you probably
came up with an argument explaining why all first moves of torus tic-tac-toe are
equivalent. As one step in your argument, you might have observed that if we
rotate the numbers on the board for torus tic-tac-toe by 90◦ without changing the
gluing arrows, then we don’t change which squares are adjacent to which.
For example, we can see from Figure 14 that after rotating the numbers in the
gluing diagram for torus tic-tac-toe by 90◦ , square 1 is still adjacent to squares 2,
4, 3, and 7, square 4 is still adjacent to squares 1, 7, 5, and 6, and so on. Since
this rotation takes 1 to 7, 7 to 9, 9 to 3, and 3 to 1, whatever strategy we have
beginning with square 1, can be rotated to get an analogous strategy beginning
with square 7, 9, or 3. Thus we can conclude that a first move in one corner square
is equivalent to a first move in any other corner square.
What happens if we rotate the numbers on a Klein bottle tic-tac-toe board by
90◦ without changing the gluing arrows? We see in Figure 15 that on the original
5. TIC-TAC-TOE ON A TORUS AND KLEIN BOTTLE 95
Figure 14. A rotation of torus tic-tac-toe by 90◦ does not change

which squares are adjacent.
board square 1 is adjacent to squares 2, 4, 3, and 9, but after the rotation square
1 is adjacent to squares 2, 4, 7, and 9. Thus a strategy starting with square 1
cannot be rotated to get a strategy starting at square 7. But this doesn’t mean
that the four corners of Klein bottle tic-tac-toe are necessarily NOT equivalent. It
just means that if we want to show that the corners are equivalent, we need to find
another way to go from a strategy starting at one corner to a strategy starting at
a different corner.
Figure 15. A rotation of Klein bottle tic-tac-toe by 90◦ does

change what squares are adjacent.
Another operation we can do with the numbers on torus tic-tac-toe is to scroll

down one row (again without changing the gluing arrows). By scrolling down one
row, we mean moving the numbers in the bottom row to the top while pushing the
other rows of numbers down. We can see in Figure 16 that again which squares are
adjacent to which hasn’t changed. So whatever strategy we have for making the
first move in one row can be scrolled down to get a strategy for making the first
move in a different row. Thus a first move in any row is equivalent to a first move
in another row, assuming the move is in the same column.
Figure 16. Scrolling the torus board down one row does not
change what squares are adjacent.
In Figure 17, we scroll the Klein bottle board down one row. Notice that the
bottom row is flipped over when it moves to the top. You can see in Figure 17 that
scrolling down does not change which squares are adjacent. Thus, as we saw on
96 4. ORIENTABILITY
Figure 17. Scrolling the Klein bottle board down one row does
not change what squares are adjacent.
the torus tic-tac-toe board, a first move in any row on the Klein bottle tic-tac-toe
board is equivalent to a first move in another row, assuming the move is in the
same column. However, you should notice that the columns on the left and the
right together form one long column with six entries.
In the Exercises, you will further analyze which first moves are equivalent in
Klein bottle tic-tac-toe.
6. The Projective Plane
The Klein bottle is not the only non-orientable 2-dimensional surface. There
are actually infinitely many non-orientable surfaces that we will learn about in
Chapter 6. Here we focus on a particularly important non-orientable surface known
as the projective plane, denoted by P 2 . The gluing diagram for the projective plane
is a disk with every pair of diametrically opposite points on the boundary glued
together. For example, in Figure 18 each pair of points with the same letter are
glued together.
Figure 18. In a projective plane every pair of opposite points on

the boundary are glued together.
In order to understand what happens along the gluing circle, let’s watch as
A. Square takes his pet Starry for a walk in a projective plane (see Figure 19). The
middle image of the figure shows A. Square after he has crossed the gluing circle
but Starry has not. At this moment, they look at each other across the inside of
the disk and each thinks that the other is mirror reversed. However, if they look
at each other back across the gluing circle, they will seem normal. Once Starry
has followed A. Square across the gluing circle in the third image, when he and
A. Square see each other across the inside of the disk, they will again seem normal.
B. Triangle has been watching A. Square and Starry as they take their walk.
From her perspective, in the third image they are both mirror reversed. We can see
from our 3-dimensional vantage point that B. Triangle is in fact correct, A. Square
6. THE PROJECTIVE PLANE 97
Figure 19. A. Square takes his pet Starry for a walk in a projec-
tive plane.
and Starry have taken an orientation reversing path and are now both mirror re-
versed. This tells us that the projective plane is non-orientable. What happens if
A. Square and Starry go for a second walk across the gluing circle?
Of course, we’d like to be able to visualize the projective plane after we’ve glued
up all of the pairs of opposite points on the boundary of the disk. But it turns out
that it’s even more difficult to visualize the glued up projective plane than it was
to visualize the glued up Klein bottle. We start with the gluing diagram on the
left in Figure 20. The disk only has two arrows because the entire left arc is glued
to the entire right arc with a twist as indicated by the direction of the arrows. In
order to see how the boundary is glued up, we pretend the inside of the disk isn’t
there. First we bring the top and bottom dots together to get a figure eight. We
want to glue the circles together so that the arrows match. To do this, we have to
first twist one of the circles at the vertex so that the arrows are lined up (as we’ve
done to the right circle in the third image of the figure). At this point it’s easy to
glue the circles together.
Figure 20. We can glue up the boundary of P 2 if we pretend the

inside of the disk isn’t there.
Now we want to see what happens to the inside of the disk during this process.
On the left side of Figure 21 we have deformed the disk into a hemisphere to give us
a 3-dimensional view of what’s going on. The first two pictures in the figure follow
the first two pictures of Figure 20. But in the third picture we see that the inside
of the disk gets in the way of twisting the right circle so that the arrows are lined
up. Thus we won’t be able to glue the two circles together as we did in Figure 20.
Instead, in the fourth image of Figure 21 we pinch each circle into a line segment so
that we can glue the two line segments together. Finally, we obtain the fifth image.
Keep in mind that the fifth image in Figure 21 is not actually the projective
plane since it contains segments where there should be circles. However, you may
recall that Figure 10 was also not actually a Klein bottle since it intersected itself
along a circle.
98 4. ORIENTABILITY
Figure 21. We pinch the two gluing circles into line segments so
that we can glue them together.
7. The Projective Plane with a Disk Removed
We saw in the last section that we can either think of a projective plane in
terms of its gluing diagram or as the object we would get by gluing together the
two circles in the third image of Figure 21. We now introduce another way to think
about the projective plane.
In Figure 22, we’ve cut out a piece of a disk from each side of the projective
plane, like taking bites out of opposite sides of an apple. We can stretch each of
these pieces into a half-disk and then glue the two half-disks together to get the
disk on the right of Figure 22. Note that we had to flip one of the half-disks over
in order to line up the black and white arrows. The double headed arrow and the
triangular arrow on the boundary of the disk tell us how to glue it back to the apple
core.
Figure 22. We cut two half-disks out of a projective plane and

then glue them together.
In Figure 23 we see that the apple core can be deformed to a gluing diagram for
a Möbius strip by straightening the edges so that it looks like a rectangle with the
top and bottom glued together with a twist. In this way, we see that a projective
plane can be thought of as a Möbius strip with a disk glued along its boundary.
You could try to do this process in reverse to create a projective plane. Start
with a Möbius strip together with a large floppy disk, made of cloth or rubber. Use
tape or staples to start attaching the boundary of your disk to the boundary of
your Möbius strip. Continue attaching the boundaries for as long as you can until
you get stuck and can’t continue. This experiment will give you a better sense of
what goes wrong when you try to glue up a projective plane in R3 .
7. THE PROJECTIVE PLANE WITH A DISK REMOVED 99
Figure 23. A projective plane can be cut up into a Möbius strip

and a disk.
In Figure 22, we chose a particular disk to cut out of a projective plane so that
we could easily recognize that what remained was a Möbius strip. In fact, it doesn’t
matter what disk we remove from a projective plane, what remains will still be a
Möbius strip. It just might be harder to recognize.
In Figure 24, we cut a disk out of the center of the gluing diagram of the
projective plane. We know that the washer that remains is supposed to be a gluing
diagram for a Möbius strip. But we can’t glue up this Möbius strip in R3 .
Figure 24. A projective plane with a disk cut out in the center
of the gluing diagram.
In order to better understand the Möbius strip in Figure 24, we redraw the
projective plane as we drew the hemisphere in Figure 21. Then the disk that we’ve
removed is the southern polar cap of this hemisphere. In Figure 25, we try to glue
up the Möbius strip as we did the projective plane in Figure 21. The picture on
the right of Figure 25 is not really a Möbius strip because we have squeezed the
two gluing circles into line segments just as we did in Figure 21.
Figure 25. A cross-cap represents a projective plane with a disk removed.

100 4. ORIENTABILITY
We call this picture a cross-cap. We can think of a cross-cap as representing

either a Möbius strip or a projective plane with a disk removed. The one advantage
of drawing a cross-cap over the usual drawing of a Möbius strip is that the boundary
of a cross-cap is a flat circle rather than a twisted circle. This will be useful in
Chapter 6 when we want to attach a projective plane with a disk removed to other
surfaces.
8. Projective 3-Dimensional Space
Now that we understand the projective plane, we would like to create an anal-
ogous 3-dimensional space. First, recall that the 3-dimensional analog of a disk
is a solid ball (which we denoted by B 3 in Chapter 1). To obtain the projective
3-dimensional space P 3 , we glue together every pair of opposite points along the
boundary of a ball, just as we glued together opposite points along the boundary
of a disk to obtain P 2 . Figure 26 illustrates some of the pairs of opposite points
that are glued together along the boundary of the ball.
Figure 26. To obtain P 3 , we glue opposite points on the bound-

ary of a ball.
In Figure 27, we illustrate a Möbius strip that is contained in a horizontal cross

section of P 3 . We saw in Figure 2 that if A. Square walks along the center line
of a Möbius strip, he will come back as his mirror image. But when A. 3D-Girl
walked along a Möbius strip in Figure 4 she did not come back as her mirror image.
Let’s consider what happens to A. 3D-Girl when she walks along the Möbius strip
in Figure 27.
In the third picture of the figure, she comes back to the spot on the strip where
she started, but she is now upside down and underneath the strip (as indicated by
the dotted lines). Because of the gluing, her left hand has gone from the right side
of the picture to the left side of the picture. However, if you compare the third
picture with the first picture, you can see that though she is under the strip, her
left hand is still a left hand. Thus just as in Figure 4, she has not become her
mirror image. In the Exercises you will show that no matter what path A. 3D-Girl
takes in P 3 , she will not be reversed. Thus, in fact, P 3 is orientable. This seems
surprising since P 2 is non-orientable and P 3 was constructed analogously.
9. 1-SIDED AND 2-SIDED SURFACES 101
Figure 27. A. 3D-Girl walks along a Möbius strip inside P 3 .
9. 1-Sided and 2-Sided Surfaces
We saw in Figures 4 and 27 that when A. 3D-Girl walked along a Möbius strip,
she ended up standing on the opposite “side” of the strip from where she started.
Since she can get from one “side” of the Möbius strip to the other by walking along
it in this way, we would like to say that the Möbius strip has only one side. On
the other hand, as we see in Figure 28, no matter what path A. 3D-Girl follows on
the surface of a disk she remains on the same side. In particular, she cannot climb
over the edge of the disk. For this reason, we would like to say that a disk in R3
has two sides.
Figure 28. No matter where A. 3D-Girl walks on a disk she stays

on the same side.
Another way to think about the distinction between a Möbius strip and a disk
is to observe that we could seamlessly paint a disk with two colors, blue on top and
red on the bottom. But we cannot seamlessly paint a Möbius strip with two colors.
Before we define 1-sided and 2-sided, let’s recall from Chapter 3 that a surface
has the same local topology as R2 and a 3-manifold has the same local topology
as R3 . Suppose that a surface is in a 3-manifold and p is a point in the surface.
Then there is a small 3-dimensional region around p which looks like a ball whose
intersection with the surface looks like a disk. We can see this ball and disk for a
point in a Möbius strip in R3 in Figure 29.
Figure 29. A ball around the point p which intersects the Möbius
strip in a disk.
So when we say that a 3-dimensional person walks along a Möbius strip and
comes back on the other side, we mean that the person starts on one side of this
disk and comes back on the other side of the disk. In particular, we now have the
following definition.
Definition. A surface contained in a 3-manifold is said to be 1-sided if a

3-dimensional creature standing on a small disk in the surface can walk along a
path in the surface and come back on the other side of that disk. If there is no such
path, then the surface is said to be 2-sided.
Using this definition, we can see that the Möbius strip in Figure 4 is 1-sided,
and the disk in Figure 28 is 2-sided. In fact, any Möbius strip in R3 will be 1-sided,
and any disk, sphere, or curved torus in R3 will be 2-sided (we’ll see why in the
next section).
It is important to note that sidedness is an extrinsic property of a surface in a
3-manifold, since it has to do with the experience of a 3-dimensional creature and
hence cannot be detected by a 2-dimensional creature in the surface. On the other
hand, orientability of a surface is an intrinsic property because it can be detected
from within the surface.
People often think that saying a surface is 1-sided is equivalent to saying it’s
non-orientable, but is this always true? We’ll learn the answer to this question in
the next section.
10. NON-ORIENTABILITY AND 1-SIDEDNESS 103
10. Non-orientability and 1-Sidedness
Figure 30 illustrates a 3-dimensional Klein bottle K 3 obtained by gluing the

top and bottom of a cube with a left-right flip and gluing the other pairs of opposite
sides straight.
Figure 30. The gluing diagram for a 3-dimensional Klein bottle.
Because of the gluing arrows on the cube, the grey square cutting the cube in
half in Figure 31 is the gluing diagram for a torus, which we know is orientable. To
figure out if this torus is 1-sided or 2-sided, we illustrate A. 3D-Girl walking upward
along the right side of the torus. Before passing through the ceiling, we see that
her left hand is still swollen from her bee sting. Because the ceiling is glued to the
floor with a left-right flip, as she passes through the ceiling, she emerges from the
floor on the left side of the torus. Thus, this torus is a 1-sided orientable surface in
K 3 . Observe that both before and after she passes through the ceiling, the hand on
the tail side of the black and white arrow is the one that’s swollen. Has A. 3D-Girl
been reversed?
Figure 31. A. 3D-Girl walks along a 1-sided torus in K 3 .
In Figure 32, we illustrate three perpendicular surfaces in the K 3 whose gluing

diagram is in Figure 30. For each surface, we have labeled it according to whether
it’s 1-sided or 2-sided and whether it’s orientable or non-orientable. You should
check that the surfaces each have the orientability and sideness properties listed.
This example shows that a single 3-manifold can contain surfaces that are orientable
and 1-sided, orientable and 2-sided, and non-orientable and 2-sided.
In spite of the above surfaces in K 3 , the following theorem shows that for sur-
faces in orientable 3-manifolds, 1-sidedness and non-orientability are in fact equiv-
alent.
Figure 32. Orientability is independent of sidedness in K 3 .
Theorem 4.1. Let S be a surface in an orientable 3-manifold M . Then S is

1-sided if and only if S is non-orientable.
The statement of our theorem contains the phrase “if and only if”. This means
that to prove it we need to show both “if S is 1-sided, then S is non-orientable”
and “if S is non-orientable, then S is 1-sided”. We think of the first statement as
the forwards direction of “if and only if” because we are assuming the beginning
of the sentence and proving the end of the sentence. In the proof, we will use the
symbol (=⇒) to indicate that we are about to prove the forwards direction. We
think of the statement “if S is non-orientable, then S is 1-sided” as the backwards
direction of “if and only if” because we are assuming the end of the sentence and
proving the beginning of the sentence. We indicate that this is what we are about
to prove by using the symbol (⇐=).
Proof. (=⇒) Suppose that S is a 1-sided surface in an orientable 3-manifold

M . We prove that S is non-orientable with the help of A. 3D-Girl.
Since S is 1-sided, A. 3D-Girl can find path on S to walk along which will take
her to the other side of the surface. Before beginning her journey, she dips her left
foot in black paint so that it leaves footprints as she walks. In fact, there’s so much
black paint on her foot that her footprints are significantly bigger than her foot.
Since M is orientable, no matter what path she walks on, she will not be reversed.
So her left foot will remain a left foot throughout her journey. We don’t know what
the surface is or what the 3-manifold looks like, but we do know that A. 3D-Girl’s
journey has taken her from one side of the surface to the other side of the surface,
and hence at the start and finish of her trip her left footprints must look like those
in Figure 33.
In the meantime, A. Square who lives in the surface, thinks her footprint is a
2-dimensional creature whose name is A. Footprint. He observes that A. Footprint
has gone for a walk and has become reversed. From this he concludes that the
surface he lives in must be non-orientable. Hence indeed S is non-orientable.
(⇐=) Now suppose that S is a non-orientable surface in M . We prove that S is
1-sided in M , again with the help of A. 3D-Girl.
Since S is non-orientable, it contains an orientation reversing path that reverses
any 2-dimensional creature who walks along it. A. 3D-Girl follows this path, leav-
ing footprints as she goes. Since the footprints are 2-dimensional creatures, they
will become reversed as the result of following an orientation reversing path. In
10. NON-ORIENTABILITY AND 1-SIDEDNESS 105
Figure 33. A. 3D-Girl comes back on the other side of the surface
but her left foot remains a left foot.
particular, her left footprint will have become a right footprint by the end of her
trek.
If A. 3D-Girl stayed on the same side of the surface throughout her journey and
her left footprint became a right footprint, her left foot would have to have become
a right foot as illustrated in Figure 34. But this could only happen if A. 3D-Girl had
become reversed, which is impossible since the 3-manifold she lives in is orientable.
Thus, in fact, her journey must have taken her to the other side of the surface so
that her footprint could become reversed without her becoming reversed. It now
follows that the surface is indeed 1-sided.
Figure 34. A. 3D-Girl stays on the same side of the surface, but
her left footprint becomes a right footprint.
Notice that at the end of a proof we use the symbol to tell the reader that
the proof is done. You will see this at the end of the proofs in this book as well as
in many other math books.
11. Exercises
1. A tube connecting two spaces is sometimes called an Einstein-Rosen bridge (see

Figure 35).
(a) What evidence could A. Square look for that might indicate that Flatland
was two planes with an Einstein-Rosen bridge connecting them?
(b) Suppose two parallel planes in R3 are connected by an Einstein-Rosen
bridge such that all Squares in the two planes are oriented in a parallel
way from the perspective of a 3-dimensional person looking down on the
two planes. Do the Squares in one plane appear to the Squares in the
other plane to be reversed?
(c) Is the surface consisting of two parallel planes connected by an Einstein-
Rosen bridge an orientable surface?
(d) If two planes are connected by two Einstein-Rosen bridges, is the surface
necessarily orientable?
(e) By analogy with our description of a 2-dimensional Einstein-Rosen bridge,
describe a 3-dimensional Einstein-Rosen bridge. Suppose there is a 3-
dimensional Einstein-Rosen bridge attaching one R3 to another copy of
R3 . Is the space orientable?
2. We saw in the chapter that left and right can morph into one another in a
non-orientable surface. Is this also true for clockwise and counterclockwise?
3. Suppose A. Square lives on a Möbius strip.
(a) How does he see himself in the distance? Does he appear to be reversed?
Draw an extended diagram.
(b) Suppose A. Square stands still and watches B. Triangle walk around and
around the strip. Just as B. Triangle is passing him after going once
around the strip, A. Square looks ahead at all the copies of himself and
her. Which, if any, of them will appear to be reversed? Draw an extended
diagram of a Möbius strip containing A. Square and B. Triangle at this
moment in time.
4. Suppose A. Square lives alone in a small Klein bottle and each side of A. Square
is colored a different color. Draw an extended diagram of A. Square’s space.
5. Suppose you live in a 3-dimensional Klein bottle with the front and back wall
glued with a left-right flip and the other pairs of opposite walls glued straight.
11. EXERCISES 107
(a) As you look around, you will observe many images of yourself. Which of
these images are reversed?
(b) Suppose that A. 3D-Girl walks once through the front wall and returns
(from the back wall) to stand next to you. Which images of her appear to
be reversed?
6. Does a 3-dimensional Klein bottle necessarily contain a 2-dimensional Klein
bottle? If so, where? How many Klein bottles does it contain? Does it contain
a pair of perpendicular Klein bottles?
7. List the numbered squares in every diagonal in the Klein bottle tic-tac-toe
board illustrated in Figure 13. Do some diagonals contain more numbers than
others?
8. Recall that two moves in Klein bottle tic-tac-toe are said to be equivalent if the
strategy for the rest of the game is precisely analogous.
(a) Explain why two board positions that are related by a vertical or horizontal
reflection of the board are equivalent.
(b) Start with the Klein bottle tic-tac-toe board illustrated in Figure 13. Is
the board we obtain by scrolling up one row equivalent to the original
board? Is the board we get by scrolling to the right one column equivalent
to the original board? Explain.
(c) List all first moves that are equivalent to placing an X in the upper left
corner of the board.
(d) List all first moves that are equivalent to placing an X in the center of the
board.
(e) How many non-equivalent first moves are there in Klein bottle tic-tac-toe?
9. For each of the non-equivalent first moves that you found in Exercise 8, list
the optimal moves for each player. Determine whether the first player will
necessarily win Klein bottle tic-tac-toe if both players play optimally.
10. Draw a picture illustrating a P 2 contained in P 3 . Is your P 2 1-sided or 2-sided?
Explain.
11. Explain why A. 3D-Girl cannot be reversed by any path she takes in P 3 .
12. How would you define a projective 1-dimensional space P 1 that is analogous to
P 2 ? Is P 1 orientable or non-orientable?
13. For each of the surfaces listed below, state one or more intrinsic topological
and/or geometric properties that enables you to distinguish that surface from
ALL of the other surfaces in the list. State whether each property you refer to
is topological or geometric.
• a plane
• a flat torus
• a flat Klein bottle
• a sphere
• the surface bounding a cube
14. Discuss the sidedness and orientability of the grey surfaces in the 3-manifolds
in Figure 36.
15. For each of the gluing diagrams in Figure 36, draw two intersecting surfaces
perpendicular to the grey surface and determine their sidedness and orientabil-
ity.
16. Determine whether the surface in Figure 37 is orientable. Is it 1-sided or 2-sided
in R3 ?

CHAPTER 5
Flat Manifolds
Topics:
• Flat surfaces
• Polygonal gluing diagrams
• Cone points and anti-cone points
• Extended diagrams of cones
• Flat manifolds with cubical gluing diagrams
• Turn manifolds
• Extended diagrams of flat 3-manifolds
1. Flat Surfaces
By now you’re probably comfortable with the flat torus and flat Klein bottle
obtained by abstractly gluing opposite sides of a square. One reason that we like
these gluing diagrams is that their local geometry is the same as that of a plane.
By contrast, the local geometries of the curved torus in R3 and the curved Klein
bottle in R4 are not nearly so nice. For this reason, when we refer to a torus or
Klein bottle, you should always think of the gluing diagram illustration of it, rather
than imagining it as a curved surface in 3 or 4 dimensions.
In general, we’ll say a surface is flat if its local intrinsic geometry is the same as
that of a plane. This makes sense since a plane corresponds to our intuitive notion
of the word “flat”. In Chapter 3, we saw that the sum of the angles of any triangle
on a sphere is greater than 180◦ , and the sum of the angles of a triangle on a curved
torus depends on where the triangle is located on the torus. Since the sum of the
angles of a triangle is a local intrinsic geometric property, this tells us that neither
a sphere nor a curved torus has the same local intrinsic geometry as R2 , and hence
neither is flat. This again corresponds to our intuition since we wouldn’t want to
call a sphere or a curved torus “flat”. Although our intuition about flatness only
applies to surfaces, we can extend the concept to spaces of any dimension.
Definition. An n-dimensional manifold is said to be flat if its local intrinsic
geometry is the same as that of Rn .
In particular, this means that the geometry of any flat manifold is homogeneous
and the sum of the angles of any triangle in a flat manifold is 180◦ . So to show
that a surface is not flat, all we need to do is find one triangle in the surface whose
angle sum is not 180◦ . The following theorem tells us that we can also prove that
a surface is flat just by looking at angle sums of triangles.
109
110 5. FLAT MANIFOLDS
Theorem 5.1. A surface is flat if and only if the sum of the angles of any
triangle in the surface is equal to 180◦ .
This theorem is quite useful, but it’s not easy to prove. In particular, why does
simply knowing about the sums of angles of triangles tell us that all of the local
intrinsic geometric properties of a surface are the same as they are in R2 ? You
might see the proof of this result if you take a course in differential geometry.
We saw in Chapter 3 that the sum of the angles of every triangle in the flat
torus is 180◦ . It now follows from this theorem that the flat torus is indeed flat,
which is what our intuition told us anyway. To see how we can apply Theorem 5.1
to a new surface, let’s consider the hexagonal gluing diagram in Figure 1. Any
triangle which is in the interior of the hexagon is planar and hence the sum of its
angles is 180◦ . If a triangle goes out one edge and in another but doesn’t contain a
corner, we can cut out the two pieces of the triangle and tape them together to get
a planar triangle as illustrated in Figure 1. Thus the sum of the angles would again
be 180◦ . Can you think of how you would you check that the sum of the angles
of a triangle containing a corner is yet again 180◦ ? After checking that this is the
case for all three types of triangles, we can apply Theorem 5.1 to conclude that the
surface in Figure 1 is indeed flat.
Figure 1. The sum of the angles of this triangle in the glued

surface is equal to 180◦ .
In addition to triangles on a flat torus, in Chapter 3 we saw that the sum of

the angles of any triangle on an infinite cylinder is 180◦ . Thus we can also use
Theorem 5.1 to conclude that the infinite cylinder is also flat. Since the glued
up cylinder appears to be curved, our intuition about what flat means has led us
somewhat astray.
2. Flat Gluing Diagrams
We would like to find an easier method to determine whether a surface is flat

than by checking the angle sum of every triangle. In Chapter 3, we saw that
every triangle in the flat torus corresponds to infinitely many identical triangles in
the extended diagram (see Figure 2). Since the extended diagram is a plane, we
concluded that every triangle in the flat torus has an angle of sum 180◦ . But how
did we know that the extended diagram of a surface is a plane?
To answer this question, observe that when we exit through one side of the
square gluing diagram, we re-enter from the opposite side. This means that we
get the extended diagram by gluing together one square after another as we go up,
down, back, and forth forever. These squares fit neatly together to form a plane
2. FLAT GLUING DIAGRAMS 111
Figure 2. A triangle in the flat torus and in the extended diagram.
because the edges of the squares are glued together in pairs and four squares come
together at each corner. We illustrate the way the four corners of the square fit
together in Figure 3.
Figure 3. Copies of the square gluing diagram for a torus fit

neatly together to form a plane.
The theorem below shows that we can use this idea more generally to check if
surfaces are flat without having to draw an extended diagram or consider the angle
sum of every triangle in the surface.
Theorem 5.2. Let S be a surface with a polygonal gluing diagram. Then S is
flat if and only if the edges of the polygon are glued together in pairs and the sum
of the angles around each glued up vertex is 360◦ .
Recall from Chapter 4 that if we want to prove a theorem containing the phrase
“if and only if”, we have to prove two statements. In this case, the statements that
we have to prove are the following.
(1) If S is flat, then the edges of the polygon are glued together in pairs and
the sum of the angles around each glued up vertex is 360◦ .
(2) If the edges of a polygon are glued together in pairs and the sum of the
angles around each glued up vertex is 360◦ , then S is flat.
We think of the first statement as the forwards direction of the proof because
we are assuming the beginning of the “if and only if” statement and proving the
end. We use the symbol (=⇒) to indicate that we are about to prove the forwards
direction. We think of the second statement as the backwards direction of the proof
because we are assuming the end of “if and only if” and proving the beginning. In
this case, we indicate what we are about to prove by using the symbol (⇐=).
Proof. (=⇒) Suppose that the surface S defined by the polygonal gluing
diagram is flat. We prove as follows that the edges must be glued together in pairs
and that the sum of the angles around each glued up vertex is 360◦ .
Recall that if a surface is flat, then its local intrinsic geometry at every point is
the same as that of R2 . In particular, every point must have exactly 360◦ around
it. To find the angles around a vertex v in the glued up surface, we have to add
together the angles around all of the vertices that are glued to v. This means that
in order to have 360◦ around each vertex in the glued up surface, the sum of the
angles around any collection of vertices that are glued together must be 360◦ .
Also, each point on an edge which is not a vertex has precisely 180◦ around it
in the polygon before the gluing. Thus we must glue the edges together in pairs in
order to have 2 × 180◦ = 360◦ around each such point.
(⇐=) Suppose that in the gluing diagram for S the edges are glued together in
pairs and the sum of the angles around each glued up vertex is 360◦ . We prove as
follows that S is flat.
Observe that since the polygonal gluing diagram is drawn on a plane, any point
in the inside of the polygon already has the same local intrinsic geometry as R2 . So
let’s consider a point p which is not a vertex but is on an edge of the polygon. Since
edges are glued together in pairs, there are two edges in the glued up surface that
contain p. To evaluate the local intrinsic geometry around p, we can tape together
a little semicircle around p on each of these edges. Because the two semicircles are
flat and we are gluing them together along straight edges, we get a flat disk around
p in the glued up surface. Thus the local intrinsic geometry around p is the same
as it is around a point in R.
Finally, to evaluate the local intrinsic geometry around a vertex v in the glued
up polygon, we cut out a wedge around each vertex glued to v and align the wedges
one after the other, as we would if we had served too many plates of pizza and we
were putting the untouched slices back together into the pizza box. Since the sum
of the angles around the vertices glued to p is 360◦ , the circular wedges around
these vertices will fit nicely together so that the region around v has the same local
intrinsic geometry as a point in R. We illustrate this in Figure 4 for a hexagonal
gluing diagram where a group of three vertices are glued together.
Figure 4. When these wedges are glued together we get 360◦

around the black vertex.
2. FLAT GLUING DIAGRAMS 113
Since every point in the glued up surface has the same local intrinsic geometry
as a point in R2 , we can conclude that the surface is indeed flat.
As an example, let’s use Theorem 5.2 to show that the surface illustrated in
Figure 5 is flat. Since the edges are glued together in pairs, we only need to check
that the sum of the angles around any collection of vertices that are glued together
is 360◦ . Before we compute these angle sums, let’s determine which vertices are
glued together.
Figure 5. We use Theorem 5.2 to show that this surface is flat.
Since the triple arrows are glued together, vertices 2 and 6 are glued together
and vertices 3 and 5 are glued together. Now since the solid black arrows are glued
together, vertices 2 and 4 are glued together and vertices 1 and 5 are glued together.
Finally, since the white triangular arrows are glued together, vertices 4 and 6 are
glued together and vertices 1 and 3 are glued together. Altogether, this means that
vertices 2, 4, and 6 are all glued together, and vertices 1, 3, and 5 are all glued
together. We indicate this in Figure 5, by coloring the even numbered vertices grey
and the odd numbered vertices black.
Now since all of the angles of a regular hexagon are 120◦ , the sum of the angles
around every glued up vertex is 360◦ . So even though we don’t know what our
surface would look like if we tried to physically glue it up, we can conclude from
Theorem 5.2 that the surface is indeed flat. Note that we will refer to this method
of determining which vertices are glued together as “matching up arrows”.
For the special case when our gluing diagram is a square, we can restate The-
orem 5.2 more simply. We call the following result a corollary because it follows
from Theorem 5.2.
Corollary 2. A surface whose gluing diagram is a square is flat if and only
if its edges are glued in pairs and all four vertices are glued together.
As an application of Corollary 2, we consider the gluing diagram of the Klein
bottle in Figure 6. Observe that opposite edges are glued together. Hence edges
are indeed glued together in pairs. Also, by matching up arrows as we did for the
hexagon in Figure 5, we can check that all four vertices are glued together. Thus
by Corollary 2 we can conclude that the Klein bottle is indeed flat, without having
to construct the extended diagram.
Next, let’s consider the surface we get from the gluing diagram in Figure 7.
The edges are again glued together in pairs, though the edges that are glued are
adjacent rather than opposite one another. By matching the arrows, we see that
the vertices are not all glued together. Rather the pair of white vertices are glued
Figure 6. We can apply Corollary 2 to show that this gluing

diagram for the Klein bottle is flat.
together, and the remaining two vertices are not glued to any other vertex. Thus by
Corollary 2, the surface in Figure 7 is not flat. Actually, the surface is topologically
the same as a surface that we’ve seen before. Can you identify it?
Figure 7. We can apply Corollary 2 to show that the surface with

this gluing diagram is not flat.
3. The Point of a Cone
We now consider what happens if we have a polygonal gluing diagram where

edges are glued in pairs, but the sum of the angles around some vertex is not 360◦ .
In this case, the surface would be flat everywhere except at these glued up vertices.
We introduce the following definition in order to be able to describe such “bad”
points.
Definition. If the sum of the angles around a point is less than 360◦ , then the
point is said to be a cone point. If the sum of the angles around a point is more
than 360◦ , then the point is said to be an anti-cone point.
A cone point looks like the point of a cone. In Figure 8, we illustrate a cone
point with 90◦ around it. The surface is flat except at the cone point. In Figure 8,
we removed 270◦ from the plane in order to get a cone angle of 90◦ .
In general, we can create a cone point with angle θ by removing an angle of
360◦ − θ from a region of the plane and then gluing together the edges of the wedge
that remains. Don’t panic if the symbol θ looks foreign to you—it should! It’s a
lower-case Greek letter, which is often used for angles in trigonometry. The letter
θ is spelled “theta”, which is pronounced to rhyme with the cheese feta and the
word “laytah” as in “catch you laytah”.
A. Square has heard that there is a cone point in Flatland. But it’s in a region
of Flatland that is supposed to be a bad neighborhood, so none of his friends or
family have dared to go there. He has no idea what a cone point is, but decides to
risk going to a bad neighborhood to see one for himself. On the left side of Figure 9,
3. THE POINT OF A CONE 115
Figure 8. A cone whose cone point has 90◦ around it.
Figure 9. A. Square tries unsuccessfully to explore a cone point.
we can see A. Square as he is starting his trek up the back of the cone towards the
cone point.
In the left-hand picture in Figure 9, A. Square’s sides are illustrated with dashed
lines because he’s on the back side of the cone. In the middle picture, he thinks he
sees the back of another square who’s also walking towards the cone point. As he
moves closer to the cone point, the other square seems to be deliberately bumping
into him. A. Square goes to the left to move out of the other guy’s way, but the
guy moves to the left too so that their top corners are still touching. With this
guy blocking him, it’s impossible for A. Square to get any closer to the cone point.
A. Square thinks that this other square is a real jerk because even when A. Square
tries to take a completely different path to the cone point, that other square gets
in his way. No wonder people say this is a bad neighborhood if the squares who
live here are all jerks like this guy. Eventually, A. Square gives up and goes home,
murmuring under his breath about never wanting to come to this neighborhood
again.
A cone is not the only surface with a cone point. For example, consider the
gluing diagram for the surface in Figure 10. Although the edges of the gluing
diagram are glued in pairs, we saw in Figure 7 there is less than 360◦ around each
of the glued up vertices. As a result, this surface is flat except at these three
Figure 10. This gluing diagram has three cone points.
cone points. If you glue the surface up, it will resemble the surface of an apple
turnover. Topologically this surface is the same as a sphere, but geometrically it’s
quite different.
4. Using Extended Diagrams to Understand Cone Points
An extended diagram can help us visualize what it would be like to live in a

universe with a cone point. As a first example, let’s consider an infinite cone with
a cone angle of 90◦ . We can think of the gluing diagram for the cone as an infinite
wedge. To obtain the extended diagram, we glue together copies of the wedge along
their edges, as we did to obtain the extended diagram of the flat torus. However,
since the cone angle is 90◦ , once we’ve glued together four of these wedges, we can
just glue the remaining edge of the fourth wedge to the remaining edge of the first
edge to get a plane, as illustrated in Figure 11.
Figure 11. The extended diagram for a cone with a cone angle of 90◦ .
Since the extended diagram of a cone with cone angle 90◦ is a plane, we might
be tempted to think that the cone is flat. In particular, we know that every triangle
in the extended diagram has an angle sum of 180◦ . However, in the Exercises in
Chapter 3 we saw that a triangle containing the cone point has an angle sum of
more than 180◦ . To understand this apparent contradiction, below we consider
triangles containing the cone point in the cone and in the extended diagram.
In the gluing diagram on the left side of Figure 12, we see a path of three
segments with right angles between them. The endpoints of the path come together
as a result of the gluing, creating a triangle in the cone whose angle sum is more
than 180◦ . However, because the extended diagram consists of four wedges which
are each identical to the gluing diagram, the three segments making up the triangle
in the cone are repeated in each of these wedges. This creates a cross shape rather
than a triangle in the extended diagram, as illustrated on the right in Figure 12.
4. USING EXTENDED DIAGRAMS TO UNDERSTAND CONE POINTS 117
Figure 12. A triangle in the gluing diagram containing the cone

point is not a triangle in the extended diagram.
Thus, the fact that the extended diagram is a plane does not imply that the angle
sum of every triangle in the cone is 180◦ .
Now let’s consider the triangle in the extended diagram illustrated on the left
in Figure 13. Since the extended diagram is a plane, this triangle has an angle
sum of 180◦ . But all four wedges in the extended diagram are superimposed in the
gluing diagram. Hence the sides of the triangle in the extended diagram intersect
in the gluing diagram as illustrated on the right in Figure 13. (You should trace
the image on the left, then cut out the four wedges, and put them one on top of
the other to check that you agree with the image on the right.) Thus the triangle
in the extended diagram does not represent a triangle in the gluing diagram.
Figure 13. A triangle in the extended diagram containing the

cone point is not a triangle in the gluing diagram.
These examples should convince you that triangles that contain the cone point
do not correspond to triangles in the extended diagram, and triangles that contain
the cone point in the extended diagram do not correspond to triangles in the cone.
In particular, the cone with a cone angle of 90◦ is not flat, even though its extended
diagram is a plane.
We can use the extended diagram of a cone to determine how many images
A. Square sees of himself. For example, Figure 14 illustrates A. Square in the
extended diagram of the cone with a cone angle of 90◦ . In this case, there are four
copies of A. Square in the extended diagram, each rotated 90◦ from the previous
one. A. Square can see three images of himself and believes that he is one of four
squares in Flatland. He thinks that the other three squares are so shy that they
always run away whenever he tries to approach them.
Figure 14. In a cone with a cone angle of 90◦ , A. Square sees

three other squares.
Next we consider what A. Square sees in a cone with a cone angle of 73◦ . In
this case, it is not as easy to create an extended diagram because 73 does not divide
evenly into 360. As a result, the extended diagram spirals around and around the
cone point before the final wedge is attached to the original wedge. But we don’t
need to draw the whole extended diagram in order to determine what A. Square
sees. Since light travels in a straight line, his vision cannot wrap around the cone
point, just like your vision cannot wrap around a building to see who’s approaching
on a perpendicular street. In fact, as we can see in Figure 15, A. Square can only
see four images of himself, two on each side of the cone point. The squares adjacent
to him on either side are rotated by 73◦ relative to him, and the next squares after
those are rotated by 73◦ relative to the previous ones.
Figure 15. In a cone with a cone angle of 73◦ , A. Square sees

four other squares.
We have numbered the wedges to make it easier to talk about them. In the
diagram it looks like the “real” A. Square is standing in wedge 3. But in fact, it’s
not that there is one real A. Square and the others are fake. We have just picked
the image of A. Square in wedge 3 to use to trace his line of sight. Observe that
since each wedge has an angle of 73◦ , wedge 1 and wedge 5 overlap in a region with
angle (5 × 73◦ ) − 360◦ = 5◦ . This gives A. Square the impression that the squares
in regions 1 and 5 are closer together than the squares in any other two adjacent
regions. But keep in mind that A. Square does not see the numbers or the lines
separating one wedge from the next. He just sees the other squares who are always
5. ANTI-CONE POINTS 119
running away from him; and wherever he goes, the whole picture moves with him.
In particular, the overlap region will always be exactly opposite him.
For our next example, let’s imagine that A. Square lives in a cone with a cone
angle of 300◦ . This means each wedge has an angle of 300◦ . Hence the next images
of him are in wedges that are 300◦ away in either direction. However, as we can
see on the left side of Figure 16, when A. Square is looking to the left of the cone
point, he can’t see the square which is at an angle of 300◦ clockwise around the cone
point from him. On the other hand, as we can see on right side of the figure, when
A. Square is looking to the right of the cone point, he can’t see the square which is
at an angle of 300◦ counterclockwise around the cone point from him. Thus in this
space, A. Square doesn’t see any copies of himself.
Figure 16. In a cone with a cone angle of 300◦ , A. Square doesn’t

see any other squares.
5. Anti-cone Points
In contrast to a cone point which has an angle with less than 360◦ around it,
an anti-cone point has an angle with more than 360◦ around it. One way we can
construct an anti-cone point is by cutting a slit in a region of the plane and inserting
a wedge. If we add a wedge with angle θ, we will obtain an anti-cone point with an
angle of 360◦ + θ around it. For example, in Figure 17 we inserted a wedge with an
angle of 45◦ giving us an anti-cone point with angle 360◦ + 45◦ = 405◦ around it.
Figure 17. We add a wedge of 45◦ to create an anti-cone point

with an angle of 405◦ around it.
In Figure 18, A. Square and B. Triangle go on an expedition in search of an

anti-cone point. They never do find an anti-cone point, but as they walk side-
by-side each following a straight line, they seem to be getting farther apart. This
Figure 18. A. Square and B. Triangle search in vain for an anti-

cone point.
makes A. Square worry about whether B. Triangle really likes him as much as he
likes her.
The hexagon in Figure 19 is the gluing diagram for a torus. (Can you see
why?) However, it has a cone point of 120◦ at vertex p and an anti-cone point of
5 × 120◦ = 600◦ , made up of the vertices labeled q (you should convince yourself
that indeed all of the vertices labeled q are in fact glued together). This hexagonal
gluing diagram yields a torus whose local geometry is different from both the curved
torus and the flat torus.
Figure 19. A gluing diagram for a torus with a cone point and
an anti-cone point.
6. We Show that the 3-Torus Is Flat
We can construct flat 3-manifolds using a method similar to the one we used
to construct flat surfaces. However, instead of starting with a polygonal gluing
diagram in the plane, we start with a 3-dimensional polyhedron in R3 in which
faces are glued together in pairs. In order for the glued up space to be flat, every
point must have a region around it whose local intrinsic geometry is the same as
R3 . This means that each point must have a region around it that looks like a ball.
We’ll see how this works with the gluing diagram for the 3-torus T 3 , illustrated in
Figure 20.
Since the top and bottom faces of the cube in Figure 20 are glued together, the
top front edge is glued to the bottom front edge. Since the back and front faces are
glued together, the bottom front edge is glued to the bottom back edge. Finally,
since the top and bottom faces are glued together, the bottom back edge is glued
to the top back edge. In this way, we can conclude that the four dotted edges are
glued together. You should use a similar argument to convince yourself that the
6. WE SHOW THAT THE 3-TORUS IS FLAT 121
Figure 20. A gluing diagram for T 3 .
four thick grey edges are glued together, and the four thin black edges are glued
together. You should also check that all eight vertices are glued together.
To see if this glued up cube has the same local intrinsic geometry as R3 , we
consider four types of points in the gluing diagram: interior points of the cube,
points on a face of the cube, points on an edge of the cube, and points around a
vertex of the cube. In Figure 21, we illustrate a region around a point of each type.
We see as follows that each of these glued up regions is a ball.
Figure 21. There is a ball around each type of point in T 3 .
• A point of Type 1: This type of point is in the interior of the cube, so

a region around it is a ball.
• A point of Type 2: Because faces are glued in pairs, two points at the
same spot on opposite faces of the cube get glued together. Around each
of these points, there’s half a ball inside the cube. These two halves are
glued together to produce a whole ball around the glued up point.
• A point of Type 3: A region around an edge point of the cube looks
like a quarter of a solid ball. As we saw in Figure 20, edges are glued in
groups of four. Thus four quarters of a ball are glued together to produce
a complete ball around the glued up point.
• A point of Type 4: A region around each vertex of the cube looks
like an eighth of a solid ball. Since all eight vertices are glued together,
these eight eighths of a ball are glued together to produce a complete ball
around the glued up point.
Since every point in T 3 has a region around it that looks like a ball, T 3 has the
same local intrinsic geometry as R3 . Thus T 3 is indeed a flat 3-manifold.
7. A Method to Determine if a Glued Up Cube Is Flat
We would like to have a result like Theorem 5.2 to check whether a 3-dimen-
sional gluing diagram is flat. It’s a little more complicated in 3 dimensions than
it was for surfaces, so we focus on gluings of a cube. Since we want our glued up
space to be a manifold, it can’t have a boundary. Thus we must have pairs of faces
glued together. But there are additional requirements needed to make sure the
manifold is flat. In particular, as we saw in Figure 21, the region around an edge of
a cube looks like a quarter of a solid ball. So we need precisely four such edges to
be glued together so that the local intrinsic geometry of the glued up manifold will
be the same as that of R3 . Also, a region around a vertex of a cube looks like an
eighth of a solid ball. So we need precisely eight such regions to be glued together
to obtain the local intrinsic geometry of R3 . If all of these requirements are met,
then the local intrinsic geometry of our glued up manifold should be exactly the
same as that of R3 . These arguments should give you an idea of why the following
theorem is true. You should look back at Corollary 2 to see that Theorem 5.3 is
the 3-dimensional analogue.
Theorem 5.3. A 3-manifold whose gluing diagram is a cube is flat if and only
if its faces are glued in pairs, its edges are glued together in groups of four, and all
eight vertices are glued together.
As an example of how to use Theorem 5.3, let’s consider the 3-dimensional
Klein bottle K 3 , illustrated in Figure 22. From the arrows we see that the top and
bottom faces of the cube are glued together with a back-front flip, and the other
two pairs of opposite faces are glued together straight.
Figure 22. A gluing diagram for K 3 .
In order to use Theorem 5.3, we need to check that the edges of the cube are
glued together in groups of four and all eight vertices are glued together. As we
consider each edge, we “color” it so that the ones that are glued together are colored
in the same way. Similarly, as we consider each vertex, we color those which are
glued together with the same color. As we see in Figure 23, all eight vertices are
colored black and the edges are colored in groups of four. This shows us that we
can apply Theorem 5.3 to conclude that the 3-dimensional Klein bottle is flat.
Next we consider the gluing of the cube in Figure 24. In this case, the top
and the bottom faces of the cube are glued together with a front-back flip and the
front and back faces are glued together with a top-bottom flip. Such a 3-manifold
is not a 3-dimensional Klein bottle because two pairs of opposite faces are glued
with a flip. We have colored the edges in the figure to indicate which ones are glued
8. EXTENDED DIAGRAMS OF GLUED UP CUBES 123
Figure 23. Edges are glued together in groups of four, and all
Figure 24. This 3-manifold is not flat because its edges are not
glued together in groups of four.
together. Observe that there are only two thick black edges and only two dotted
edges. Since Theorem 5.3 is an “if and only if” statement, we can conclude that
the 3-manifold obtained from this gluing diagram is not flat.
8. Extended Diagrams of Glued Up Cubes
Before we construct the extended diagram of a glued up cube, let’s recall the
extended diagrams of a flat torus and flat Klein bottle. In particular, we saw in
Chapter 1 that the extended diagram of T 2 is a plane made up of infinitely many
copies of the gluing diagram. If A. Square lives in a flat T 2 , he will see copies of
himself going off forever upwards, downwards, leftwards, and rightwards. Figure 25
illustrates a 3 × 3 grid of the extended diagram with A. Square in it.
In Chapter 4 we saw that the extended diagram of the flat Klein bottle is also
a plane made up of infinitely many copies of the gluing diagram. If we glue the top
of the square to the bottom of the square with a flip and glue the left and right
sides straight, then the extended diagram for K 2 with A. Square in it is illustrated
in Figure 26.
We are now ready to consider the extended diagram of a flat 3-manifold ob-
tained by gluing up a cube. We know by Theorem 5.3 that such a 3-manifold is flat
if and only if the faces of the cube are glued in pairs, the edges are glued in groups
of four, and all eight vertices are glued together. This corresponds exactly to how
infinitely many cubes could be glued together to create R3 . That is, if we imagine
R3 as made out of infinitely many cubes, each face would be shared by two cubes,
each edge would be shared by four cubes, and each vertex would be shared by eight
Figure 25. A 3 × 3 grid of the extended diagram for T 2 with

A. Square in it.
Figure 26. A 3 × 3 grid of the extended diagram for K 2 with

A. Square in it.
cubes. Thus if a glued up cube is flat, then its extended diagram is R3 , made up
of infinitely many copies of the gluing diagram.
We can get a sense of what it’s like to live in such a 3-manifold by looking at
copies of A. Cube in an extended diagram. To make it easier to see what’s going
on, we color the top face of A. Cube white, her bottom face orange, her front face
black, her back face purple, her right face grey, and her left face yellow. Because
this book is in black and white, we denote the faces that are orange, yellow, and
purple, by the letters O, Y, and P. If a white face has no letter on it, then it’s white.
We illustrate the colored faces of A. Cube in Figure 27.
Now let’s suppose that A. Cube lives in T 3 . If she looks towards the front wall
of the gluing diagram, her line of vision goes through the front wall and into the
back wall, hitting her in the back. So imagines she is seeing the back of another
cube. Similarly, if she looks towards the back, the top, the bottom, the right, and
the left, she will see her opposite side. In fact, A. Cube sees identical images of
herself going off forever upwards, downwards, backwards, forwards, leftwards, and
rightwards. In Figure 28, we illustrate a 3 × 3 × 3 block of the extended diagram of
T 3 with A. Cube in it. Since you can’t see all sides of a cube at once, it’s helpful
8. EXTENDED DIAGRAMS OF GLUED UP CUBES 125
Figure 27. The colored faces of A. Cube.
P P P
P P P
P P P
P P P
P P P
P P P
Figure 28. The extended diagram of T 3 with A. Cube in it.
to remember which colors are opposite which. In particular, we can’t see her back
side, but we remember that purple is opposite black. So her back side must be
purple. Also, we can’t see her right side, but we remember that grey is opposite
yellow. So we know that her right side is grey. You can refer back to Figure 27 if
you forget how her faces are colored.
All of the cubes in Figure 28 are identical, because each wall of the gluing
diagram for T 3 is glued to the opposite wall straight. By contrast, Figure 29
illustrates an extended diagram of K 3 with A. Cube in it. We know that K 3 is
obtained from a cube by gluing two pairs of opposite faces straight and one pair
with a flip. We can determine which pair is glued with a flip by analyzing the
colored cubes in the extended diagram.
Let’s start with the cube in the front lower left corner of the figure with an
arrow pointing at it. We know that the back of the cube is purple since the front is
black, the right is grey because the left is yellow, and the bottom is orange because
the top is white.
Figure 29. The extended diagram of K 3 with A. Cube in it.
Now let’s consider the cube just to the right of the cube with the arrow pointing
at it. Since the top of this cube is white, we know that there couldn’t have been
a top-bottom flip of the right and left walls of the gluing diagram. Also, since the
front of this cube is black, we know there couldn’t have been a back-front flip of
the right and left walls of the gluing diagram. From this we can conclude that the
two side walls are glued straight.
Next we consider the cube just above the one with the arrow pointing at it.
Since the left face of this cube is grey, but the left face of the original cube was
yellow, there must have been a left-right flip of the top and bottom faces in the
gluing diagram. On the other hand, since the upper cube and lower cube both have
black in front, there could not have been a front-back flip of the top and bottom
walls of the gluing diagram. Finally, when we look out the back of the cube in the
lower left, we see an identical cube just behind it. Thus it follows that the front
and back walls must be glued straight. This tells us everything about how the walls
of the gluing diagram were glued.
Using this type of reasoning, if you’re given an extended diagram of a glued
up cube, you should be able to determine what the gluing is, and if you’re given
a gluing diagram for a cube, you should be able to draw an extended diagram. In
the Exercises, you will have a chance to practice going back and forth between a
gluing diagram and an extended diagram for a glued up cube.
9. OTHER TYPES OF GLUINGS OF A CUBE 127
9. Other Types of Gluings of a Cube
So far, we have only considered gluing the faces of a cube straight or with a
flip. Now we will see what happens when we glue a pair of faces of a cube with a
rotation. In particular, we define a 14 -turn manifold as the manifold obtained by
gluing one pair of opposite faces with a 90◦ rotation, while gluing the other pairs
of opposite faces straight. In Figure 30, the top and bottom faces are glued with a
90◦ rotation. This means that when A. 3D-Girl goes out the top of the cube, she
comes in the bottom of the cube rotated by 90◦ . By checking that the edges are
glued in groups of four and all eight vertices are glued together, we see that the
resulting manifold is indeed flat. Is this manifold orientable?
Figure 30. Since the top and bottom of this cube are glued with
a 90◦ rotation, when A. 3D-Girl goes out the top she comes in the
bottom rotated by 90◦ .
In a similar way, we can create 12 -turn and 34 -turn manifolds by gluing opposite
faces straight except for one pair which is glued with a 180◦ or a 270◦ rotation.
Observe that we can’t create turn manifolds in two dimensions since if you start
with a polygon, the edges can only be glued straight or with a flip. In particular,
gluing one edge to another with a 90◦ rotation wouldn’t make sense, and a 180◦
rotation of an edge is the same as a flip.
All of the glued up cubes we’ve considered so far have pairs of opposite faces
glued. We can also have a gluing diagram where some number of adjacent faces of
the cube are glued together. For example in Figure 31, the top face is glued to the
front face, the back face is glued to the bottom face, and the right and left faces
are glued together. In the figure, we have colored the edges and vertices to indicate
which ones are glued together. You can imagine gluing adjacent faces together by
folding along the edge between them. Thus the gluing in Figure 31 is obtained by
folding along the dashed line and the dotted line before gluing up the left and right
faces. Since there are only two thick black edges and only one dashed edge and one
dotted edge, this gluing diagram is definitely not flat. Notice that the endpoints of
the thick black edge are glued together, as are the endpoints of the dashed edge,
as well as those of the dotted edge. This means that the thick black edge becomes
a circle of cone points with 180◦ around each point on it. The dashed edge and
Figure 31. The top and front faces are glued, the back and bot-
tom faces are glued, and the right and left faces are glued.
the dotted edge each form another circle of cone points with only 90◦ around each
point.
A cube is not the only 3-dimensional gluing diagram. We can create 3-manifolds
by gluing up pairs of faces on any polyhedron. In Exercises 18 and 19 we consider
3-manifolds obtained by gluing together opposite faces of a dodecahedron. As you
glue up pairs of faces of a polyhedron, remember that you can check if the glued up
3-manifold is flat by determining whether every point on each edge and vertex has
a 3-dimensional ball around it. In particular, you might want to look at Figure 21
to help you.
10. Exercises
1. Draw a gluing diagram for a 2-manifold that is homogeneous at all but five
points. Draw a 2-manifold that is homogeneous at all but n points.
2. For each of the conditions listed below, draw an octagonal gluing diagram with
edges glued in pairs and indicate whether your glued up surface is orientable
or non-orientable.
(a) The surface has least one cone point.
(b) The surface has at least one anti-cone point.
(c) The surface has both a cone point and an anti-cone point.
3. Explain how you know that the surface whose gluing diagram is illustrated in
Figure 19 is a torus, and how you know it has one cone point and one anti-cone
point.
4. What surface is obtained by gluing up the hexagon in Figure 32? Does this
surface have any cone points or anti-cone points?

10. EXERCISES 129
5. Suppose A. Hexagon lives in the surface illustrated in Figure 32 and has a

different color on each side. Draw an extended diagram of A. Hexagon in her
universe.
6. Compare the global topology and geometry of a flat torus and the hexagonal
gluing diagram in Figure 32.
7. What surface do you get by gluing each side of a square to itself by folding the
side in half. Does the surface have any cone points or anti-cone points? If so,
how many?
8. Imagine a large square with a smaller square cut out of it. Glue together
opposite sides of each square. What surface do you get?
9. Suppose A. Square lives alone in the space illustrated in Figure 33.
(a) Draw an extended picture of his space.
(b) Is his space flat?
10. Suppose that A. Square lives alone in his space, and imagine that he is translu-
cent so he can see through himself. For each of the spaces listed below, deter-
mine how many squares A. Square can see in addition to himself. In each case,
draw a picture to illustrate your conclusion.
(a) An infinitely long cylinder.
(b) An infinite cone with a cone angle of 45◦ .
(c) An infinite cone with a cone angle of 50◦ .
(d) An infinite cone with a cone angle of 200◦ .
11. Draw a polyhedral gluing diagram which is homogeneous everywhere except
along one circle where every point on that circle has less than 360◦ around
it and along another circle where every point is has more than 360◦ around
it. Note this means that the space has a circle of cone points and a circle of
anti-cone points.
12. Suppose that you live in a 14 -turn manifold and you talk on the phone to some-
one who lives in a 34 -turn manifold. Could you determine that your manifolds
were different? What if you talk to someone in a 12 -turn manifold?
13. Draw three gluing diagrams of a cube whose extended diagrams are different, in
which each have two pairs of opposite faces glued with flips and the remaining
pair of opposite faces glued straight. Explain how you know that the extended
diagram of each of your spaces is distinct from the others.
14. How can you tell the difference between a glued up cube in which one pair of
opposite faces is glued with a flip and the rest are glued straight, from a glued
up cube in which one pair of opposite faces is glued with a 180◦ rotation and
the rest are glued straight?
15. Suppose that we glue opposite faces of a cube in each of the ways described
below. Is it possible to get a flat manifold? If so explain how.
(a) One pair of sides is glued with a flip, and the other two pairs are each
glued with a 180◦ rotation.
(b) One pair of faces is glued with a flip, and the other two pairs are each
glued with a 90◦ rotation.
(c) Each pair of faces is glued with a flip.
16. Start with a cube and glue each of the vertical sides to itself by folding the side
in half along a floor to ceiling line. Then glue the bottom of the cube to the top
of the cube with a 14 -turn. How many circles does this manifold have where it
is not homogeneous? What is the angle around each of these non-homogeneous
circles?
17. Consider the 3-manifold whose gluing diagram is illustrated in Figure 24. If
we take a cross section of the gluing diagram parallel to the left and right
faces, we obtain a square gluing diagram. Draw a picture of the gluing diagram
with arrows to indicate the gluings along the edges. Determine whether the
corresponding surface is flat, and identify the surface.
18. A dodecahedron is a polyhedron with 12 faces that are pentagons (see Fig-
ure 34). If the pentagons are transparent, you can look through one and see
that the opposite pentagon is rotated by 10 1
× 360◦ = 36◦ . In order to glue
opposite faces, the vertices of the pentagons must line up. So we must rotate
a face by an odd number times 36◦ before gluing it to the opposite face.

10. EXERCISES 131
Consider the space obtained from a dodecahedron by gluing each pair of

1
opposite faces with a 10 -turn. In other words we rotate the faces by 36◦ before
we glue them to their opposite faces. How many vertices are glued together?
How many edges are glued together? Is the 3-manifold flat?
19. Suppose you live in a dodecahedron with opposite pairs of faces glued with
1
a 10 -turn and you are suddenly transported to a dodecahedron with opposite
3
pairs of faces glued with a 10 -turn. Can you tell the difference between your
new space and your original one?
20. Figures 35, 36, and 37 are extended diagrams for spaces created by gluing
opposite faces of a cube. A. Cube is alone in each space. Her coloring is
illustrated in Figure 27. Explain how the faces of the cube were glued together
to create the extended diagram for each figure.


10. EXERCISES 133
21. Trace Figure 38 onto a piece of paper and then color it to illustrate the view
inside of a 14 -turn manifold containing A. Cube.
22. Create a gluing of a cube where pairs of opposite faces are glued, but not all
CHAPTER 6
Connected Sums of Spaces
Topics:
• Connected sums of surfaces
• Classification of surfaces versions 1 and 2
• Euler characteristic
• Genus
• Connected sums of 3-manifolds
1. Einstein-Rosen Bridges
A first step in trying to understand the shape of our universe is to make a list
of all the possibilities. If we had such a list, we might be able to compare evidence
about the shape of our universe to topological and geometric properties of spaces
on the list in order to make a short list of serious contenders. However, so far,
the list of 3-dimensional universes that we have learned about is not very long. In
particular, the only 3-manifolds we have discussed are R3 , S 3 , T 3 , K 3 , together
with other manifolds obtained by gluing up a cube. But these are not the only
possibilities. Just as we can put together bricks to make complicated buildings,
we can put together familiar spaces to make new more complicated spaces. As
usual, before we describe how to do this in 3 dimensions, we consider the analogous
problem in two dimensions.
Perhaps the simplest way to combine two surfaces is to put them next to each
other, and then find a way to connect them to form a single surface. In one of
the Exercises in Chapter 4, we joined two planes together with a cylindrical tube
known as an Einstein-Rosen bridge (see Figure 1 below). This tube provides a way
for A. Square and his friends to get from one of the planes to the other, just as
an ordinary bridge enables us to get from one island to another. We can describe
the construction of an Einstein-Rosen bridge between two planes as removing the
interior of a disk from each plane and then adding a tube that goes from the
boundary of the hole in one plane to the boundary of the hole in the other plane.
Notice that after we attach the tube, the circles at the top and the bottom along
which it’s glued are different from normal circles in the plane because they bound
infinite regions on both sides.
In this chapter, we will not pay any attention to the geometry of our spaces.
This means we are free to shorten an Einstein-Rosen bridge between two planes
until it has essentially no length at all. The result is two planes which are joined
135
136 6. CONNECTED SUMS OF SPACES
Figure 1. An Einstein-Rosen bridge joining two planes.
together along a single circle. We can visualize this by removing the tube in Figure 1
and gluing the circle at the top of the tube to the circle at the bottom of the tube.
Figure 2 illustrates a gluing diagram for this space, where the grey regions
have been removed from the two planes and the arrows indicate how to connect the
boundary circles together. When A. Square goes through one of these circles, he is
instantly transported to the other circle, as illustrated in the figure.
Figure 2. These two planes with holes are joined together along
the circle.
The same idea enables us to connect any two surfaces along a circle. In particu-
lar, we remove the interior of a disk in each surface, and then we attach a tube that
goes from the boundary of the hole in one surface to the boundary of the hole in the
other surface as illustrated on the left side of Figure 3. Because we are concerned
with the topology of the new surface and not the geometry, we can then shorten
the tube between the surfaces until it has no length at all. In this way, the surfaces
are glued together along a circle rather than joined by a tube, as illustrated on the
right side of Figure 3.
Figure 3. We can shrink the tube joining two surfaces until it’s
just a circle.
2. Connected Sums of Surfaces
We formalize the ideas from the last section with the following definition.
Definition. Let A and B be surfaces. The connected sum of A and B,
written as A#B, is the surface obtained by removing the interior of a disk from
each of A and B and then gluing the remaining pieces together along the boundaries
of the holes.
3. ARITHMETIC PROPERTIES OF THE CONNECTED SUM 137
Figure 4 illustrates the connected sum of two tori. In the first step, we put
the two tori side by side with the connected sum symbol between them. When we
remove a disk from each one, it looks like we have two hollow doughnuts with a bite
taken out of each. Finally, we stick the two partially eaten donuts together along
their boundaries to get the double doughnut. The resulting surface is a 2-holed
torus denoted by T 2 #T 2 , or more succinctly by 2T 2 .
Figure 4. The connected sum of two tori.
To take the connected sum of three tori, we can first add two tori to get the 2-
holed torus 2T 2 , and then add another torus T 2 to get a 3-holed torus. Symbolically
we can express this as:
T 2 #T 2 #T 2 = (T 2 #T 2 )#T 2 = 2T 2 #T 2 = 3T 2 .
In a similar way we can write the connected sum of two copies of any surface
A as A#A = 2A, and we can write the connected sum of n copies of A as:
nA = A#A# · · · #A .

n terms
You might wonder what would happen if the two tori in Figure 4 were glued
together along a different circle. Since we are focusing on topology here, the disks
that we remove from A and B can slide around on each of the surfaces so that
A#B does not depend on where the gluing takes place. We say that the connected
sum operation is well defined to mean that it does not depend on anything other
than A and B.
3. Arithmetic Properties of the Connected Sum

The word “sum” gives us the idea that the connected sum is, at least in some
ways, similar to normal addition of numbers. In the Exercises, you will show
that the connected sum operation obeys the commutative property (expressed as
a + b = b + a) and the associative property (expressed as a + (b + c) = (a + b) + c)).
We’ll explore some of the other properties of the connected sum below, and we’ll
see that connected sums do share many (but not all) of the properties of sums of
numbers.
In ordinary addition, the number 0 is said to be an identity element because
for any number a we have 0 + a = a and a + 0 = a. We see as follows that
the sphere S 2 serves as an identity element under the connected sum operation.
Figure 5 illustrates the connected sum of a torus T 2 and a sphere S 2 . Since we are
only considering topology here, we can deform the resulting surface to our heart’s
delight. In particular, we can shrink the blister on the torus to a small lump, which
eventually becomes absorbed into the surface of the torus itself. Thus, we see that
T 2 #S 2 = T 2 . In fact, S 2 #T 2 = T 2 is also true, and the torus in Figure 5 can be
replaced with any surface X. Thus we see that S 2 is indeed an identity element for
the connected sum.
Figure 5. The connected sum of a torus and a sphere is a torus.
In ordinary addition, any number a has an additive inverse −a such that

a + (−a) = 0. Since S 2 plays the role of 0 for the connected sum, we might
wonder whether every surface has an inverse under the connected sum operation.
In other words, if we are given a surface A, can we find a surface B such that
A#B = S 2 ? If such an inverse surface exists, we might want to call it −A rather
than B. We can ask the related question of whether we are allowed to “cancel”
under the connected sum operation. That is, if we know that A#B = A#C, can
we conclude that B = C? For ordinary addition, we know that if 2 + x = 2 + 3,
then we can cancel the 2’s to get x = 3. Before we answer these questions, let’s
consider how the connected sum works with gluing diagrams.
4. Gluing Diagrams for nT 2 and nP 2
We start with the gluing diagram of a torus. Figure 6 shows some pictures of
how we might remove a disk from the gluing diagram. Note that in the diagram on
the right, the disk that we have removed is shaped like an eye rather than a disk.
But that’s OK because an eye shape can be deformed to a disk, and we are only
interested in topology right now rather than geometry.
Figure 6. Different types of disks on the gluing diagram for a torus.
Since we saw above that the connected sum is well defined, it doesn’t matter
where we remove the disk from our surface. Thus we can choose to remove whichever
disk is most convenient. You might guess that the best disk to remove is the one
right in the center of the picture. But, in fact, we will use the picture on the right
because after removing the disk, we can cut open the gluing diagram so that the
circle around the disk we’ve removed becomes an edge in a pentagon (see Figure 7).
This is allowed because all five vertices in the pentagon are glued together (you
should check this), so the two endpoints of the new edge represent the same point.
In particular, the new edge still represents a circle in the pentagonal gluing diagram.
The connected sum of the two tori is then formed by attaching two copies of
this pentagonal gluing diagram together along the two new edges (corresponding
4. GLUING DIAGRAMS FOR nT 2 AND nP 2 139
Figure 7. Changing the gluing diagram from a square with a hole

in it to a pentagon.
∪ ∪
Figure 8. A gluing diagram for the connected sum of two tori.
to the boundaries of the two disks that were removed). In Figure 8 we see that the
gluing diagram for 2T 2 is an octagon.
Notice that in Figure 8, we use the union symbol ∪ to mean that we want the
two pieces glued together. This symbol is different from the connected sum symbol
#, which we put between two surfaces before we have removed the disks from each
surface. For example, look at the way the symbol # is used on the left side of
Figure 5.
Next we use a similar method to obtain the gluing diagram for the connected
sum of two projective planes. We start with two copies of the gluing diagram for
P 2 . Then we remove the inside of a small disk at the corner of each gluing diagram
as illustrated on the left side of Figure 9. This enables us to open up the diagrams
into triangles. Finally, these triangles are glued together along their new edges to
form the quadrilateral gluing diagram on the right of Figure 9.
Figure 9. A gluing diagram for the connected sum of two projec-

tive planes.
We can repeat this process to get a hexagonal gluing diagram for 3P 2 =

P #P 2 #P 2 . In fact, for any number n, we can similarly obtain a gluing dia-
2
gram with 2n sides for nP 2 and one with 4n sides for nT 2 . We can also find a
gluing diagram for the connected sum of some number of tori together with some
number of projective planes.
5. The Classification of Surfaces

The following theorem gives us the surprising result that all finite surfaces
can be obtained as connected sums of the three surfaces S 2 , T 2 , and P 2 . You
might see the proof of this theorem in a more advanced topology class. The word
“classification” in the title of the theorem is often used in mathematics to mean
that the result describes all possibilities.
Theorem 6.1 (The Classification of Surfaces, version 1). Every surface with
finite area is either S 2 or the connected sum nT 2 #mP 2 for a pair of non-negative
integers m and n, at least one of which is positive.
The Classification of Surfaces together with the method we learned in the last
section enables us to create a gluing diagram for any finite surface. However, the
gluing diagrams that we create in this way are not the only ones for a given surface.
For example, in Figure 9 we illustrated a gluing diagram for 2P 2 = P 2 #P 2 . We
obtain a different gluing diagram for 2P 2 as follows.
On the left side of Figure 10 we’ve redrawn the gluing diagram from Figure 9 as
a square. Then in the second step of Figure 10, we’ve cut the gluing diagram along
a diagonal to obtain two triangles. We then rotate the top triangle and flip over
the bottom triangle so that the double arrows line up in the third step. Finally, we
glue the double arrows together and then reshape the diagram into a new square.
We can now recognize this gluing diagram as that of a Klein bottle. This wasn’t
apparent from Figure 9. The technical term for this kind of manipulation of a
gluing diagram is cut and paste. In particular, we say that Figure 10 uses cut and
paste to show that P 2 #P 2 = K 2 .
Figure 10. We use cut and paste to show that P 2 #P 2 = K 2 .
In the Exercises, you will use cut and paste to show that
3P 2 = T 2 #P 2 .
From this equation together with the equation P 2 #P 2 = K 2 , we obtain the fol-
lowing lemma.
Lemma 1. K 2 #P 2 = (P 2 #P 2 )#P 2 = 3P 2 = T 2 #P 2 .
This result can be used to answer our earlier question about whether cancelation
works for the connected sum operation. In particular, observe that since K 2 is non-
5. THE CLASSIFICATION OF SURFACES 141
orientable and T 2 is orientable, we know that K 2 = T 2 . Thus we cannot cancel the

P 2 from both sides of the equation
K 2 #P 2 = T 2 #P 2 .
It is important to keep this in mind as you work with the connected sum operation.
Earlier we had also asked whether the connected sum operation had an inverse.
Starting again with the result that K 2 #P 2 = T 2 #P 2 , we show below that P 2
cannot have an inverse. Suppose that P 2 has an inverse Q. Then we could add Q
to both sides of the equation K 2 #P 2 = T 2 #P 2 and use the associative property
(which you will prove in the Exercises) to get
K 2 #(P 2 #Q) = (K 2 #P 2 )#Q = (T 2 #P 2 )#Q = T 2 #(P 2 #Q).
Since Q is the inverse of P 2 , we obtain the equation

K 2 #S 2 = T 2 #S 2 .
From this we would again get the false equation K 2 = T 2 . Thus P 2 cannot have
an inverse.
Finally, notice that the equation 3P 2 = T 2 #P 2 (which you will prove in the
Exercises) allows us to avoid mixing tori and projective planes in connected sums.
For example, the connected sum 2T 2 #P 2 can be written as follows:
2T 2 #P 2 = T 2 #(T 2 #P 2 ) = T 2 #3P 2 = (T 2 #P 2 )#2P 2 = 3P 2 #2P 2 = 5P 2 .
In the Exercises, you are asked to generalize this computation to show that we
can rewrite any mixed sum of tori and projective planes as a sum of just projective
planes. This result together with the Classification of Surfaces, version 1, gives us
the following.
Theorem 6.2 (The Classification of Surfaces, version 2). Every surface with
finite area is either S 2 , nT 2 , or nP 2 for some positive integer n.
It now follows that every finite orientable surface is either S 2 or nT 2 , and
every finite non-orientable surface is nP 2 . Since S 2 #T 2 = T 2 and S 2 #P 2 = P 2 ,
we can get any finite surface by starting with S 2 and adding on some number of
tori or some number of projective planes. We can visualize these two possibilities
as follows.
Starting with S 2 #T 2 in Figure 11, we absorb the base of T 2 into S 2 until the
intersection of T 2 and S 2 splits into two circles. Then we separate the circles so
that the torus becomes a tube that goes from the sphere back to itself. Thus instead
of visualizing S 2 #T 2 as a sphere with a torus attached, we can visualize it as a
sphere with a handle attached. By attaching n handles to S 2 , we obtain nT 2 .
Figure 11. S 2 #T 2 is topologically the same as S 2 with a handle.

Figure 12. We can represent a projective plane with a disk re-

moved as a cross-cap.
We would like to find an operation analogous to attaching handles that will

enable us to visualize adding projective planes to S 2 . In Chapter 4, we learned
that we can represent a projective plane with a disk removed as a cross-cap (see
Figure 12). Keep in mind that a cross-cap is not really a projective plane with
a disk removed because in the fourth image of Figure 12 we squeezed the two
gluing circles into line segments before gluing them together. However, drawing a
projective plane with a disk removed in this way allows us to represent S 2 #P 2 as a
sphere with a cross-cap attached. Thus we can visualize S 2 #nP 2 as a sphere with
n cross-caps attached.
We can now visualize the Classification of Surfaces, version 2, as saying that
every surface can be obtained from a sphere by attaching some number of handles
or some number of cross-caps. For example, in Figure 13 we illustrate 3T 2 as a
sphere with three handles and 2P 2 as a sphere with two cross-caps. Note that the
number of handles or cross-caps could be zero in the case where the surface is just
a sphere.
Figure 13. We represent 3T 2 by the image on the left and 2P 2

by the image on the right.
6. Dividing a Surface into Vertices, Edges, and Faces
Suppose that A. Square knows that his space is finite and he knows the Clas-
sification of Surfaces, version 2. He would like to determine whether he is living
6. DIVIDING A SURFACE INTO VERTICES, EDGES, AND FACES 143
Figure 14. An atlas of a cube with one page for each face.
in S 2 , nT 2 , or nP 2 . So he decides to systematically map out every region of his

universe and collect the maps together into an atlas. Like an atlas of the earth, each
page represents some region. The gluings would only occur between one page and
another. For example, suppose that his space is the surface of a cube. Each face
of the cube could correspond to one page of his atlas. Altogether, the atlas would
have six pages corresponding to the six faces of a cube. In Figure 14, the center
of each page has two words to indicate which face of the cube it represents. These
labels describing the position of the page in three dimensions (top, bottom, front,
back, etc.) would be meaningless to A. Square, but is useful for us to visualize how
the pages form a cube.
If you’ve ever used a road atlas to navigate, you might recall that a country
is divided into square regions and there is a detailed road map of each region on
a separate page. If you want to go off the top of one page, there is a number
indicating the page you should turn to in order to continue going in that direction.
A similar thing holds for A. Square’s atlas. If he wants to know where he would be
if he went off the top of a particular page, he would look for the letter on another
page that matched the letter on the top of his page. For example, if he went off
the top of the page labeled “right face”, he would see the letter j which would tell
him that to continue going in that direction he should enter the right side of the
page labeled “back face”. The pages also contain arrows on the edges to indicate
how the pair of pages with the same letter are glued together.
As A. Square is making an atlas of his space, he decides to keep track of the
number of pages, the number of edges, and the number of vertices in his atlas. In
order to avoid having to keep track of the topology of the regions represented by
Figure 15. A region in the shape of a washer must be split into

two pages so that each page can be deformed to a disk.
Figure 16. A tetrahedron and a dodecahedron.
different pages, he makes the rule that the region on each page must be topologically
the same as a disk. This means for example that if a region was shaped like a washer
(which mathematicians call an annulus), he would have to split it into two pages,
as illustrated in Figure 15.
We will refer to the pages of A. Square’s atlas as faces, keeping in mind the rule
that every face can be deformed to a disk. For the atlas of the cube in Figure 14,
there are 8 vertices, 12 edges, and 6 faces. In Figure 16 we illustrate a tetrahedron
and a dodecahedron. You should check that the tetrahedron has 4 vertices, 6 edges,
and 4 faces, and the dodecahedron has 20 vertices, 30 edges, and 12 faces.
Since A. Square is trying to determine the topology of his universe, what’s
important to him is how the faces are glued together, not the specific size or shape
of each face. In fact, the surfaces of a cube, a tetrahedron, and a dodecahedron
are all topologically the same as a sphere. Yet the information he is collecting on
the number of faces, edges, and vertices of each are different. In the next section
we’ll learn about a wonderful formula that A. Square can put these numbers into to
obtain a single number that will help him determine the topology of his universe.
7. The Euler Characteristic of a Surface
In the middle of the eighteenth century the mathematician Leonhard Euler

(whose name is pronounced “Oiler”) introduced the following formula for surfaces
with finite area.
Definition. Let S be a surface which is divided into a finite collection of v
vertices, e edges, and f faces, where a face is a region that can be deformed to a
disk. We define the Euler characteristic of S to be the number
χ(S) = v − e + f.
Note that the symbol χ is a Greek letter, which is pronounced “kai” to rhyme
with “sky”. The Euler characteristic is represented by χ because both characertistic
and χ start with a hard k sound.
Let’s compute the Euler characteristics of a cube, tetrahedron, and dodecahe-
dron by using the numbers of faces, edges, and vertices that we determined in the
last section.
χ(cube) = 8 − 12 + 6 = 2.
χ(tetrahedron) = 4 − 6 + 4 = 2.
χ(dodecahedron) = 20 − 30 + 12 = 2.
It’s not a coincidence that all three of these values are equal to 2. It turns
out that the Euler characteristic is a topological property of a surface. Since the
7. THE EULER CHARACTERISTIC OF A SURFACE 145
cube, the tetrahedron, and the dodecahedron are all topologically equivalent, any
topological property that one of these three surfaces has, the other two surfaces
must also have. More generally, we have the following powerful theorem, whose
proof you might see in a more advanced topology class.
Theorem 6.3 (Euler Characteristic Theorem). The Euler characteristic is a
topological property of a surface with finite area. In particular, the value of the
Euler characteristic is the same no matter how the surface is divided into faces,
edges, and vertices.
If it really doesn’t matter how we divide a surface into faces, edges, and vertices,
we might as well pick an easy way of dividing it up. In fact, the easiest way is to
draw a gluing diagram and consider the whole region of the gluing diagram as a
single face. Then we just have to count how many distinct edges and vertices there
are and plug these numbers into the Euler characteristic formula.
For instance, let’s compute Euler characteristics using the gluing diagrams for
the surfaces S 2 , T 2 , P 2 , and K 2 , as illustrated in Figure 17. We aren’t used to
seeing S 2 represented by a gluing diagram. To see that the first gluing diagram
represents a sphere, we just fold over the top to attach the arrows. This gives us
the surface of an apple turnover, which is topologically the same as S 2 .
Figure 17. We use these gluing diagrams to compute Euler characteristics.
Using the gluing diagram in Figure 17, we see that the sphere has 1 face, 1
edge (since the two arrows are glued together), and 2 vertices (one at each end of
the arrows), so we have
χ(S 2 ) = 2 − 1 + 1 = 2.
Note that this agrees with our computations for the cube, the tetrahedron, and the
dodecahedron, which are each topologically the same as the sphere.
Next, we see that the torus consists of 1 square face, 2 edges (one horizontal
and the other vertical), and just 1 vertex (since all four corners of the face are glued
together), so we have
χ(T 2 ) = 1 − 2 + 1 = 0.
In the case of the projective plane, we have 1 face, 1 edge, and 1 vertex, so
χ(P 2 ) = 1 − 1 + 1 = 1.
Finally, for the Klein bottle there is 1 square face, 2 edges, and 1 vertex (since
all four corners of the face are glued together), so
χ(K 2 ) = 1 − 2 + 1 = 0.
Since P 2 and K 2 have different Euler characteristics, by the Euler Characteris-
tic Theorem (Theorem 6.3) they must be topologically distinct. On the other hand,
the Euler characteristics of T 2 and K 2 are the same even though we know that T 2
and K 2 are different surfaces, since T 2 is orientable and K 2 is not.
8. The Euler Characteristic of Connected Sums
The following theorem gives us an easy way to compute the Euler characteristic
of a connected sum.
Theorem 6.4. Let M1 and M2 be finite area surfaces. Then
χ(M1 #M2 ) = χ(M1 ) + χ(M2 ) − 2.
Proof. Since M1 and M2 are finite area surfaces, they can each be divided
into some number of faces, edges, and vertices. Suppose that M1 has f1 faces, e1
edges, and v1 vertices, and M2 has f2 faces, e2 edges, and v2 vertices. For each of
M1 and M2 , we add a circle attached to a vertex such that the circle is contained
entirely in a single face. Since the face containing the new circle is topologically
the same as a disk, the new circle will bound a region that is also topologically the
same as a disk. In Figure 18, the new circle bounds the grey region.
Figure 18. In both M1 and M2 , we add a circle attached to a

vertex such that the circle is contained in a single face.
In both M1 and M2 , this new circle counts as an additional edge and bounds an
additional face. Thus M1 will now have f1 + 1 faces, e1 + 1 edges, and v1 vertices,
and M2 will have f2 + 1 faces, e2 + 1 edges, and v2 vertices. Notice that the Euler
characteristics of M1 and M2 do not change.
In order to create the connected sum M1 #M2 , we remove a disk from each one
and then glue the surfaces together along the circular boundaries. Since it doesn’t
matter what disk we remove, we might as well remove the new faces bounded by
the circles we’ve added to M1 and M2 .
Now M1 #M2 has f1 + f2 faces, since we have removed the new faces from each
of M1 and M2 . When we count the edges of M1 #M2 , we have to account for the
original edges of M1 , the original edges of M2 , and the new circle in each. However,
the new circles are glued together in M1 #M2 . Thus we should only count this new
edge once. This means that the total number of edges in M1 #M2 is e1 + e2 + 1.
Similarly the vertex on the boundary of the face that we’ve removed from M1 is
glued to the vertex on the boundary of the face that we’ve removed from M2 .
Hence, the number of vertices of M1 #M2 is v1 + v2 − 1. We can now compute the
Euler characteristic of M1 #M2 as follows:
χ(M1 #M2 ) = (v1 + v2 − 1) − (e1 + e2 + 1) + (f1 + f2 )
= (v1 − e1 + f1 ) + (v2 − e2 + f2 ) − 2 = χ(M1 ) + χ(M2 ) − 2,
as desired.
9. THE GENUS OF A SURFACE 147
Recall from the previous section that χ(T 2 ) = 0 and χ(P 2 ) = 1. Thus the
theorem tells us that if we add T 2 to a surface M , then we have
χ(M #T 2 ) = χ(M ) + 0 − 2 = χ(M ) − 2,
and if we add P 2 to M , we have
χ(M #P 2 ) = χ(M ) + 1 − 2 = χ(M ) − 1.
Using this result together with induction, you will prove the following corollary in
the Exercises.
Corollary. The Euler characteristic of the orientable surface nT 2 is given

by the formula
χ(nT 2 ) = 2 − 2n,
and the Euler characteristic of the non-orientable surface nP 2 is given by the for-
mula
χ(nP 2 ) = 2 − n.
Observe that the corollary together with the Euler Characteristic Theorem
(Theorem 6.3) gives us the following theorem.
Theorem 6.5. Two finite area surfaces are topologically the same if and only
if they have the same Euler characteristic and either both are orientable or both are
non-orientable.
This result is very convenient for A. Square. If he can explore all of his universe
and create an atlas, he can count up the faces, edges, and vertices in his atlas and
compute the Euler characteristic. Then he can compare the Euler characteristic
that he found with the formulas in the corollary. If he gets an odd number, he knows
right away that his universe is nP 2 , and he can determine the value of n from his
computation of the Euler characteristic. For example, suppose he determines that
his universe has an Euler characteristic of −47. Then he can use the second formula
from the corollary to get 2 − n = −47, and from this he can conclude that his
universe is 49P 2 . But if the number that he gets is even, then he has to do further
explorations to determine whether or not his universe has an orientation reversing
path. Once he finds this out, he can use the correct formula from the corollary to
determine exactly what surface he lives in. For example, if he determines that his
universe is orientable and has Euler characteristic −10, then he uses the formula
2 − 2n = −10 from which he concludes that his universe is 6T 2 .
9. The Genus of a Surface
After discovering Theorem 6.5, A. Square is so excited that he rushes to find

B. Triangle to tell her about it. However, to his surprise, B. Triangle has been
busy developing her own approach to distinguishing finite area surfaces. Instead
of dividing surfaces into faces, edges, and vertices as A. Square did, B. Triangle
has been considering the topological properties of different types of circles that can
occur on a surface. In particular, she’s interested in non-separating circles, which
are those you could cut along without dividing the surface into two regions (see
Figure 19). In Chapter 3, we saw that there are infinitely many non-separating
circles on a torus. On the other hand, if we cut along any circle on a sphere, we will
separate it into two pieces. Note that we are using the word “circle” here to mean
a loop that is topologically the same as a circle. In particular, we are not allowing
these circles to intersect themselves.
Figure 19. This circle does not separate the torus into two regions.
B. Triangle has discovered that the more complicated a surface is, the more
non-separating circles it contains. She makes the following definition.
Definition. The genus of a surface is the maximum number of disjoint cir-
cular cuts that can be made without disconnecting it.
The condition that the circular cuts be disjoint is important. For example,
we saw that there are infinitely many non-separating loops on a torus, but any
two of them intersect. In fact, as we can see in Figure 20, even the two simplest
non-separating circles on a torus (a meridian and a longitude) will intersect.
Figure 20. After cutting a torus open along a meridian, any ad-
ditional circular cut will disconnect it.
If we cut a torus along a meridian, it will become a cylinder. Now any circular
cut that we make on the cylinder will disconnect it. In fact, no matter what non-
separating circle we use to cut open a torus, we’ll get a cylinder; and once we get
a cylinder, we can’t make another circular cut without disconnecting it. According
to B. Triangle’s definition this means that the genus of a torus is 1.
Next let’s consider the 2-holed torus 2T 2 . Figure 21 shows two disjoint non-
separating circles in 2T 2 , each of which wraps meridionally around one of the tori.
The existence of these two circles shows that the genus of 2T 2 is at least 2.
When we cut along the circles in Figure 21 it doesn’t disconnect the surface.
But we’d like to know if we can make another cut in Figure 21 without disconnecting
the surface. To figure this out, let’s deform the cut open 2-holed torus into a sphere
with four holes. We’ve illustrated this deformation in Figure 22. We can then check
that cutting the sphere with four holes along any circle will disconnect it. So it
seems reasonable that the genus of 2T 2 is actually equal to 2. But we don’t know
9. THE GENUS OF A SURFACE 149
Figure 21. A pair of disjoint non-separating circles in 2T 2 .
whether we could have made three circular cuts in 2T 2 without disconnecting it if

we had made our first two cuts differently. We’ll come back to this later, so let’s
not worry about it for the moment.
Figure 22. This cut open 2-holed torus can be deformed to a

sphere with four holes.
Recall that K 2 = 2P 2 , and as we saw in Section 1, we can visualize the Klein

bottle as a sphere with two cross-caps attached. Recall that a cross-cap is made
by squeezing two gluing circles into line segments and gluing them together, as in
Figure 12.
On the left side of Figure 23 we illustrate K 2 = S 2 #P 2 #P 2 , where we use
cross-caps in place of projective planes with holes in them. Now we cut each cross-
cap along the line segment representing the gluing circle. On the right side of
Figure 23, we see the result of cutting open K 2 = S 2 #P 2 #P 2 along these two
circles. Since this does not separate the Klein bottle, we know that the genus of
the Klein bottle is at least 2.
Figure 23. Cutting along the two gluing circles in K 2 =

S 2 #P 2 #P 2 does not separate the surface.
In fact, we can see from Figure 24 that once we cut our Klein bottle open along
these two circles, we can deform it to get a cylinder, which can be further deformed
to get a sphere with two holes. Now if we cut along any circle in the sphere with
two holes, we’ll disconnect it. So it seems likely that the genus of the Klein bottle
equals 2. But again we don’t know if we could have made more circular cuts without
disconnecting it if we had chosen our first two cuts differently.
Figure 24. We can deform this cut open Klein bottle to a sphere
with two holes.
Figure 25. A non-separating circle in the Klein bottle.
10. The Genus of nT 2 and nP 2
We can generalize the strategies illustrated in Figures 21 and 23 to cut along

n circles in any nT 2 or nP 2 . For nT 2 we can cut along a meridional circle on each
of the n tori without disconnecting it. For nP 2 , we can cut along each of the n
gluing circles represented by line segments running down the middle of each of the
n cross-caps. In both cases, the collection of circles we cut along are disjoint and
don’t disconnect the surface. These constructions show that the genus of nT 2 and
of nP 2 is at least n. Thus we have the inequalities

genus nT 2 ≥ n and genus nP 2 ≥ n.
Furthermore, we know that after cutting along these circles, each of our sur-
faces becomes a sphere with some number of holes. So any additional circles will
necessarily separate it into pieces. However, we don’t know whether there is a bet-
ter choice of circles which would allow us to cut along more than n circles without
disconnecting the surface. For example, we might find more circles if we allowed
the circles to wrap multiple times around several handles or cross-caps, like the
(p, q)-curves on a torus that we saw in Chapter 3. After all, if we had looked
at the projection of the Klein bottle shown in Figure 25, we may have imagined
that we could not do better than a single circle wrapping meridionally around the
main cylinder forming the bottle; nevertheless, Figure 23 shows that two circles are
possible.
The following theorem, whose proof you might see in a more advanced topology
class, tells us that in fact there is no way to make more than n circular cuts on nT 2
or nP 2 without disconnecting it.
Theorem 6.6. The genus of all finite area surfaces is given by the formulas

genus S 2 = 0, genus nT 2 = n, and genus nP 2 = n.
11. CONNECTED SUMS OF 3-MANIFOLDS 151
When B. Triangle explains the idea of genus to A. Square, the two of them get
to work to see if there is a relationship between the Euler characteristic and the
genus of a surface. They know that a sphere has Euler characteristic 2 and genus
0, and nT 2 has Euler characteristic 2 − 2n and genus n. On the other hand, for
nP 2 , the Euler characteristic is 2 − n and the genus is n. They summarize this
relationship in the following theorem.
Theorem 6.7. The Euler characteristic of an orientable surface S of genus n
is given by the formula
χ(S) = 2 − 2n.
The Euler characteristic of a non-orientable surface of genus n is given by the
formula
χ(S) = 2 − n.
Notice that the first formula works even if the surface is a sphere which has
genus 0.
11. Connected Sums of 3-Manifolds
We would like to define a 3-dimensional connected sum operation which is

analogous to the connected sum of surfaces. Such a definition would enable us to
combine 3-manifolds into more complicated ones, just as we used the connected
sum to combine surfaces into more complicated surfaces. Recall that we defined
the connected sum of two surfaces as the surface we get by removing the interior
of a disk from each and then gluing what remains together along the boundaries of
the holes. Since the analogue of a disk is a ball, if we are working with 3-manifolds,
we have to remove the interior of a ball rather than a disk from each. In particular,
we make the following definition.
Definition. Let A and B be 3-manifolds. The connected sum of A and B,
written A#B, is the 3-manifold obtained by removing the interior of a ball from
each of A and B and then gluing the remaining parts together along the boundaries
of the holes.
As with surfaces, the connected sum of two 3-manifolds is a topological notion,
so it does not depend on where the balls are removed. Just as we did with surfaces,
we can use connected sums to construct an infinite number of 3-manifolds. For
instance, we are already familiar with the 3-torus T 3 , the projective 3-dimensional
space P 3 , and the 3-dimensional Klein bottle K 3 . Combining T 3 with each of these
manifolds, we obtain T 3 #T 3 , T 3 #P 3 , and T 3 #K 3 . But visualizing connected sums
of 3-manifolds is not as easy as it was for surfaces. In fact, we have the same problem
in visualizing connected sums of 3-manifolds that Flatlanders have in visualizing
connected sums of surfaces.
A. Square has been reading this chapter along with you. So he understands
connected sums of surfaces, but his friends don’t. Figure 26 illustrates a good way
for A. Square to explain to his friends what it would be like to live in the connected
sum of two tori. A. Square tells his friends that you start with two gluing diagrams
for a torus. Then you remove a grey disk from each, and glue the surfaces together
along their boundaries. This means that if you live in one of the tori and you
Figure 26. When A. Square enters one grey circle, he will be

instantly transported to the other grey circle.
enter the grey circle you will be instantly transported to the other torus. Once his
friends understand T 2 #T 2 , he can use similar drawings with other gluing diagrams
to explain the connected sum of any pair of surfaces.
We can use a similar picture and explanation to help us visualize the connected
sum of 3-manifolds. In Figure 27, we obtain 2T 3 by removing a grey ball from
each T 3 and then gluing the two spaces together along their spherical boundaries.
This means that if A 3D-Girl lives in one T 3 and enters the grey sphere she will be
instantly transported to the other T 3 .
Recall from earlier in the chapter that S 2 is an identity element for the con-
nected sum of surfaces. In other words, for any surface X, we have the connected
sum X#S 2 = X. We see as follows that S 3 plays the same role for connected sums
of 3-manifolds. Suppose that X is a 3-manifold, and we want to evaluate X#S 3 .
This means we have to remove a ball from each of X and S 3 and then glue what
remains together. However, we learned in Chapter 1 that just as S 2 is topologically
the same as two disks glued together along their boundaries, S 3 is topologically the
same as two balls glued together along their boundaries. Thus when we remove a
ball from S 3 , we are left with another ball, which we will call B. Now to get X#S 3 ,
we remove a ball from X and then glue in B. Since B is just replacing the ball we
removed from X, we end up with X#S 3 = X.
While we have been comparing the connected sum operation to ordinary ad-
dition, we can also think of it terms of multiplication. In this case, S 2 plays the
role of the number 1, which is the identity in multiplication. Since any surface with
Figure 27. When A. 3D-Girl puts her arm through one grey
sphere, it comes out of the other grey sphere.
12. EXERCISES 153
finite area is either S 2 , nT 2 , or nP 2 , and T 2 and P 2 cannot be broken down any

further, we can think of T 2 and P 2 as prime numbers. However, unlike the integers
which contain infinitely many primes, T 2 and P 2 are the only prime surfaces.
On the other hand, like the integers, there are infinitely many prime 3-mani-
folds. While there is no purely topological classification of 3-manifolds like the
Classification of Surfaces, there is a geometric classification of 3-manifolds. This
result, known as the Geometrization Conjecture, was proved by the Russian math-
ematician Grigori Perelman, earning him the two most prestigious prizes in math-
ematics, the Fields Medal (in 2006) and the Clay Millennium Prize (in 2010), the
latter is worth one million dollars. Surprisingly, Perelman declined both awards,
stating that he was “not interested in money or fame”.
12. Exercises
1. For each n > 1, construct nT 2 as a glued up polygon. Then determine how

many cone points and anti-cone points your glued up surface has.
2. For each n > 1, construct nP 2 as a glued up polygon. Then determine how
many cone points and anti-cone points your glued up surface has.
3. In this exercise we consider whether the connected sum operation obeys the
same rules as ordinary addition of numbers.
(a) Is the connected sum operation commutative? In other words, is it always
true that X#Y = Y #X for any surfaces X and Y ?
(b) Is the connected sum operation associative? In other words, is it always
true that X#(Y #Z) = (X#Y )#Z for any surfaces X, Y , and Z?
(c) Explain why no surface other than S 2 has an inverse with respect to the
connected sum operation?
4. (Hard.) Use cut and paste operations to show that the two surfaces illustrated
in Figure 28 have the same topology. Explain how this proves that T 2 #P 2 =
3P 2 .

5. We saw in the chapter that the connected sum of n tori and m projective planes
can be written as a connected sum of some number of projective planes, as long
as m is at least 1. Write a formula for the number of projective planes that
must be used. Use induction together with the equation T 2 #P 2 = 3P 2 , to
prove that your formula is correct. (Hint: Try to induct on only one variable.)
Why does this imply that every surface other than S 2 is either nT 2 or nP 2 ?
6. Explain how T 2 can be obtained by gluing pairs of sides of any polygon with
an even number of sides.
7. Suppose that Flatland is a non-trivial connected sum. How could A. Square
detect it topologically? Suppose our universe is a non-trivial connected sum.
How could we detect it topologically?
8. Using a strip of paper, construct a Möbius strip. Cut the strip along the center
circle; this is the dotted line in the gluing diagram on the left in Figure 29.
Now cut along the other curve; this is the dotted line in the gluing diagram on
the right Figure 29. What do you notice in each case?
9. Use the theorem about the Euler characteristic of connected sums together with
induction to prove that the Euler characteristic of the orientable surface nT 2 ,
is given by the formula
χ(nT 2 ) = 2 − 2n,
and the Euler characteristic of the non-orientable surface nP 2 is given by the
formula
χ(nP 2 ) = 2 − n.
10. Can you find two finite surfaces which have the same Euler characteristic but
are topologically distinct? How could you tell the surfaces apart?
11. Use the Euler characteristic and the Classification of Surfaces, version 2, to
determine what surface is represented by the hexagonal gluing diagram in Fig-
ure 30.

12. EXERCISES 155
12. We have seen that K 2 #P 2 , T 2 #P 2 , and 3P 2 are different representations of the

same surface. We did this using cut and paste operations. Find the number of
vertices, edges, and faces, and the Euler characteristic for the gluing diagrams
associated with each of these three representations.
13. Suppose that X is a surface and M is obtained by removing n disks from X.
What is the relationship between the Euler characteristics of M and X?
14. B. Triangle is trying to identify a surface by making disjoint circular cuts. Her
third cut disconnects the surface. What surface might she have?
15. Explain why cutting a sphere with four holes along any circle will disconnect
it.
16. Show that for any n, cutting a sphere with n holes along a circle will disconnect
it.
17. Find the Euler characteristic of a genus 5 non-orientable surface. Characterize
this surface as one of the surfaces listed in the Classification of Surfaces, version
2.
18. Let M be a surface with genus m, and let N be a surface with genus n. Find
a formula for the genus of the connected sum M #N in terms of m and n.
CHAPTER 7
Products of Spaces
Topics:
• Products of sets
• Products of spaces
• Checklist to show a space is a product
• Topological uniqueness of products
• Geometric products
• Geometric products of flat spaces
• Visualizing geometric products
• Isotropic and non-isotropic spaces
• A 3-manifold which is homogeneous but not isotropic
1. Products of Sets
While the connected sum operation enabled us to combine two surfaces or

two 3-manifolds to create new ones, we cannot use it to combine two manifolds of
different dimensions. In this chapter we will see how to use the product operation
to combine spaces and then we’ll construct some new 3-manifolds from 2-manifolds
and 1-manifolds. We begin by defining products of sets, then move on to study the
topology of products of spaces, and finally consider the geometry of products later
in the chapter.
Before we give a formal definition of the product of sets, let’s consider the
illustration of R2 in Figure 1. The black point in the figure has coordinates (1, 2)
because a vertical line through the point hits the X-axis at 1 and a horizontal line
through the point hits the Y -axis at 2. The idea of a product of two sets is motivated
by the idea that every point in R2 can be expressed by a pair of coordinates in this
way.
Definition. Let X and Y be sets. The product X × Y is defined to be the set
of ordered pairs (x, y) where x is in X and y is in Y . We refer to the sets X and
Y as the factors of the product X × Y .
According to this definition, we see that R × R is the set of ordered pairs (x, y)
where both x and y are real numbers. This corresponds exactly to the definition of
R2 . Thus we can write R × R = R2 .
Observe that since the product X × Y is defined as the set of ordered pairs,
if the pairs (x, y) and (a, b) differ in either of their coordinates, then they must be
157
158 7. PRODUCTS OF SPACES
Figure 1. We use vertical and horizontal lines through a point in

R2 to find its coordinates.
different elements of the set X × Y . For example, the point (1, 2) in R2 can’t be
written in any other way as an ordered pair.
As our next example, let’s consider the sets A = {1, 2, 3, 4} and B = {1, 2, 3}.
The product A × B is obtained by listing all ordered pairs of points whose first
coordinate is in the set A and whose second coordinate is in the set B. We can
write this product as the set
{(1, 1), (2, 1), (3, 1), (4, 1), (1, 2), (2, 2), (3, 2), (4, 2), (1, 3), (2, 3), (3, 3), (4, 3)}.
Since A and B are each a finite set of numbers, it is easy to check that the above
set of ordered pairs satisfies the definition of A × B. However, it is easier to think
about the points in A × B if we organize the set into the rectangular formation
illustrated in Figure 2. We can then describe the product both as three rows of
four points and as four columns of three points.
Figure 2. The product of A = {1, 2, 3, 4} and B = {1, 2, 3} writ-

ten in a rectangular formation.
While the product of A = {1, 2, 3, 4} and B = {1, 2, 3} is just a product of sets,

the product R×R can also be thought of as a product of spaces. That is, if we think
of R as a 1-dimensional space rather than as an infinite set of numbers, then the
product R × R = R2 is a 2-dimensional space whose topological form comes from
the topological forms of the factors. However, the above definition of the product
of sets does not tell us how to visualize a product of spaces. This is what we will
explore in the next section.
2. PRODUCTS OF SPACES 159
2. Products of Spaces
We already know how to visualize the product R × R, since we know that

this product equals R2 and we can visualize R2 as a plane. We will now use our
understanding of R2 as a product of two lines as the basis for a more general method
of visualizing products of spaces. We begin by analyzing the plane in more detail
below.
The first step is to choose two lines in the plane to be our axes. We pick a
horizontal line which we call the X-axis, and a vertical line which we call the Y -
axis. Note that we will use capital letters for the axes to avoid confusing the axes
with points, whose coordinates will be written with lower case letters. Observe that
the X-axis and the Y -axis intersect in precisely one point. Also, for each point x
on the X-axis, there is a vertical line disjoint from the Y -axis which intersects the
X-axis in the point x. We denote this line by {x} × R. Similarly, for each point y
on the Y -axis, there is a horizontal line disjoint from the X-axis which intersects
the Y -axis in the point y. We denote this line by R × {y}. Furthermore, each
horizontal line intersects each vertical line in precisely one point, and this point has
coordinates (x, y). We illustrate this in Figure 3.
Figure 3. The lines {x} × R and R × {y} meet at the point (x, y).
For example, {3} × R is a vertical line which intersects the X-axis at the point
3, and R × {8} is a horizontal line which intersects the Y -axis in the point 8. Also,
the vertical line {3} × R intersects the horizontal line R × {8} at the point (3, 8).
Now let p be a point in the plane. Then there is exactly one vertical line {x}×R
which goes through p and exactly one horizontal line R × {y} which goes through
p. This gives us a way to write every point p in the plane as a coordinate pair (x, y)
in precisely in one way, which we know is required by the definition of R × R as a
product of sets.
Generalizing the above analysis of the plane as the product of two lines, we
now have the following Checklist, which will be referenced frequently below.
Checklist to show that a space S is the product X × Y
(1) Choose copies of X and Y in S intersecting in precisely one point, which

will be the X-axis and Y -axis, respectively.
(2) Check that for every point x on the X-axis, S contains a copy of Y disjoint
from the Y -axis whose intersection with the X-axis is the point x. We
denote this copy of Y by {x} × Y .
(3) Check that for every point y on the Y -axis, S contains a copy of X disjoint
from the X-axis whose intersection with the Y -axis is the point y. We
denote this copy of X by X × {y}.
(4) Check that each {x} × Y intersects each X × {y} in precisely one point.
(5) Check that each point p in S is contained in precisely one of the {x} × Y
and precisely one of the X × {y}. We denote p by the coordinate pair
(x, y).
Let’s verify that if we follow these steps, then S will indeed be the product of
X and Y . Let x be in X and y be in Y . By Step (2) we know that S contains
{x} × Y , and by Step (3) we know that S contains X × {y}. By Step (4) {x} × Y
and X × {y} intersect in precisely one point p. And finally by Step (5) the point p
is (x, y). Thus every coordinate pair (x, y) is a point in S. Furthermore, by Step
(5) we know that every point in S can be written as a coordinate pair (x, y) in
precisely in one way. Thus it follows from the definition of the product of sets in
Section 1 that S = X × Y .
Before doing some examples, let’s recall that in Section 1 we wrote the product
of the sets A = {1, 2, 3, 4} and B = {1, 2, 3} in the form of a rectangle, and we then
described the set A × B as both three rows of four points and four columns of three
points.
We can use similar language to describe Steps 2–5 of the Checklist. In particu-
lar, we can rephrase Step (2) as saying that S is an “X of Y ’s” where each copy of
Y meets the X-axis in one point, and we can rephrase Step (3) as saying that S is a
“Y of X’s” where each copy of X meets the Y -axis in one point. Then we rephrase
Step (4) as saying that every copy of X meets every copy of Y in precisely one
point. Finally, we rephrase Step (5) as saying that every point in S is on precisely
one of the copies of X and one of the copies of Y .
Using this terminology and some pictures, we can now more easily check that
a plane is the product of two lines. In particular, after picking our axes, we observe
in Figure 4(a) that there is a vertical line through each point on the X-axis meeting
the X-axis in a single point. Note that we cannot really draw a line through every
point on the X-axis. So we just draw a finite collection of lines and leave the rest
to your imagination. Then we observe in Figure 4(b) that there is a horizontal line
through each point on the Y -axis meeting the Y -axis in a single point. Finally, we
check that each horizontal line meets each vertical line in precisely one point and
each point of the plane is on precisely one horizontal and one vertical line.
In the next two sections, we use a similar method to show that several 2-
dimensional spaces can be expressed as products of 1-dimensional spaces.
3. A. Square and A. Pentagon Are Products
We begin by using the Checklist to show that a square is the product of two
line segments. We denote a line segment by the letter I to stand for interval, since
we can think of a line segment as an interval in R. From now on, we will use the
3. A. SQUARE AND A. PENTAGON ARE PRODUCTS 161
Figure 4. R2 = R×R is both a union of vertical lines and a union

of horizontal lines.
words interval, segment, and line segment interchangeably. Thus we show that a
square is I × I.
First we choose the axes to be the line segments on the bottom and left side
of the square. Then in Figure 5 we do Steps (2) and (3) by illustrating the square
is both a union of horizontal segments which each intersect the vertical axis in
precisely one point, and a union of vertical segments which each intersect the hor-
izontal axis in precisely one point. Furthermore, since each vertical segment meets
each horizontal segment in one point, we have done Step (4).
Figure 5. I × I written as a union of horizontal intervals and a

union of vertical intervals.
Finally, we illustrate Step (5) in Figure 6 by observing that every point in the
square is contained in precisely one horizontal segment and one vertical segment.
This shows that our representation of I × I as a square is correct.
Figure 6. A square is the product I × I.
After reading the beginning of this section, A. Square starts bragging to his
friends that squares and rectangles are superior to other shapes because they’re
the only shapes that are products. His rectangular and square friends are feeling
Figure 7. A pentagon is also the product I × I.
pretty good, since they always secretly believed they were superior. And now they
knew why. Several polygonal shapes have been arguing with A. Square, demanding
that he explain what’s so great about being a product. A. Square responds that
products are more beautiful, and ends with, “If you won’t admit that, then you’re
just being defensive.” Finally A. Pentagon pushes her way to the front of the crowd
and says, “Well, you’re just ignorant, because all polygons are products.” At which
point she draws Figure 7, which even A. Square has to admit is beautiful.
In fact, A. Pentagon is right. As we see in the image on the left side of Figure 7,
a pentagon is a black interval of dark grey intervals. Note that we call the dark grey
arcs “intervals” because they are topologically the same as intervals even though
they aren’t straight. We’ll discuss the geometry of products later in the chapter.
Also, we see in the center image of Figure 7 that the pentagon is also a black interval
of light grey intervals. Furthermore, as we see on the right side of the figure, each
dark grey interval intersects each light grey interval in exactly one point, and every
point in the pentagon is contained in precisely one dark grey interval and one light
grey interval. This includes all of the points on the edges of the pentagon. For
example, a point on the top left edge of the pentagon is an endpoint of one of
the dark grey intervals and is contained in one of the light grey intervals. In the
Exercises you will show that a triangle and a hexagon are also products of two
intervals.
4. Some Products where One Factor Is a Circle
For our next example, we show that a cylinder can be expressed as the product
of a circle S 1 and a line segment I. In Figure 8, we illustrate a circle of intervals
going along the horizontal S 1 -axis and an interval of circles going along the vertical
I-axis. We can see that each of the intervals intersects the S 1 -axis in one point and
each of the circles intersects the I-axis in one point
In Figure 9, we do Step (4) by checking that each circle meets each segment in
a single point. Finally, we do Step (5) by observing that every point in the cylinder
is contained in precisely one horizontal circle and one vertical segment. Thus we
can conclude that the cylinder is in fact the product S 1 × I.
We now show that an annulus (which, as we saw in Chapter 6, is the mathemat-
ical word for a washer) can also be expressed as the product S 1 × I. In Figure 10,
we illustrate the S 1 -axis as a big circle and the I-axis as a horizontal line segment.
Here we have a circle of intervals which are radial spokes going along the S 1 -axis
4. SOME PRODUCTS WHERE ONE FACTOR IS A CIRCLE 163
Figure 8. The cylinder is circle of intervals and an interval of circles.
Figure 9. A cylinder is the product S 1 × I.
and an interval of concentric circles going along the I-axis. Each of the intervals
intersects the S 1 -axis in one point and each of the circles intersects the I-axis in
one point. Also each circle meets each segment in a single point, and every point in
the annulus is contained in an interval. Thus an annulus is also the product S 1 × I.
Figure 10. An annulus is also the product S 1 × I.
As our last example, we see as follows that a torus can be expressed as the
product S 1 × S 1 . In Figure 11, we illustrate the product for both the curved torus
and the flat torus. On the curved torus, one S 1 -axis is a longitudinal circle and the
other S 1 -axis is a meridional circle. We can see that each meridional circle intersects
each longitudinal circle in precisely one point. So the curved torus is indeed a
longitude of meridional circles and a meridian of longitudinal circles. Furthermore,
every point of T 2 is on precisely one longitudinal circle and one meridional circle.
For the flat torus in Figure 11, the two vertical edges of the gluing diagram
represent a single S 1 -axis, and the two horizontal edges of the gluing diagram
represent the other S 1 -axis. Because of the gluings, the horizontal and vertical
segments in the square are actually circles. Thus again we have a circle of circles
in two different ways, and every point is on precisely one vertical segment and one
horizontal segment.
Figure 11. The curved and flat torus T 2 are each the product
S 1 × S1.
5. Examples of Spaces Incorrectly Expressed as Products
In order to better understand the Checklist, in this section we will explore some
2-dimensional spaces which are incorrectly expressed as products. In each case we
identify which steps of the Checklist are satisfied and which ones are violated.
We begin by considering the shape in Figure 12 which looks like a rectangle
with a semicircle on each end.
Figure 12. Can this shape be expressed as S 1 × I?
Figure 13. This attempt at expressing the shape in Figure 12 as

S 1 × I violates Steps (2), (4), and (5) of the Checklist.
Let’s try to express this shape as the product S 1 × I with a vertical S 1 -axis
and a horizontal I-axis as illustrated on the left in Figure 13.
Since the axes meet in just one point, Step (1) is satisfied. Now observe that
the space is a circle of horizontal intervals. However, with the exception of the
top and bottom intervals, all of these horizontal intervals meet the S 1 -axis in two
points rather than just one, violating Step (2). On the other hand, the space is an
interval of circles which each meet the I-axis in precisely one point. Thus Step (3)
is satisfied. But again with the exception of the top and bottom intervals, all of the
horizontal intervals meet each of the circles in two points. Thus Step 4 is violated.
Finally, all of the points in the interior of the shape in Figure 12 are on more
than one of the circles. For example, we see two circles containing the point p.
Thus Step (5) is also violated. From this we can conclude that Figure 13 is not a
representation of S 1 × I. Note that we actually only need to know that one of the
steps of the Checklist is violated to reach this conclusion.
5. EXAMPLES OF SPACES INCORRECTLY EXPRESSED AS PRODUCTS 165
Figure 14. This attempt at expressing a Möbius strip as S 1 × I

violates Steps (3) and (4) of the Checklist.
Next, let’s consider whether a Möbius strip could be represented as the product
S 1 × I as illustrated in Figure 14. Here, the S 1 -axis is the center circle in black in
the left image of the figure, and the I-axis is the vertical interval in black in the
right image of the figure. Since the two axes intersect in a single point, Step (1) is
satisfied. Also on the left side of the figure we see that the Möbius strip is a circle
of intervals where each interval meets the S 1 -axis in one point. Hence Step (2) is
satisfied. On the right side we see that it is also an interval of circles, but every
circle except the center circle intersects the I-axis in two points rather than just
one. For example, we see that the grey circle wraps twice around the Möbius strip,
intersecting the I-axis in the points p and q. Thus Step (3) is violated. Similarly,
apart from the center circle, all of the circles intersect all of the intervals in two
points. Thus Step (4) is violated. On the other hand, every point on the Möbius
strip is on precisely one interval and one circle. Thus Step (5) is satisfied. We
conclude that Figure 14 is not a representation of the product S 1 × I.
Our final example is based on the Earth, where locations are often described
by their longitude and latitude. A longitude is a great circle going north-south and
a latitude is a horizontal circle going east-west. We will now consider whether S 2
is a product of two circles where one axis is the equator (which is a latitude) and
the other axis is a longitude. We illustrate our axes in Figure 15. In addition to
the equator, we’ve drawn four other latitudes, and in addition to our longitude axis
we’ve drawn two other longitudes. However, you should imagine longitudes and
latitudes through every point.
Figure 15. S 2 is not the product of an equator and a longitude.
Observe that since the axes meet in two points, Step (1) is violated. In addition,
for every point on the equator, there is a longitude, but these longitudes are not
disjoint from the longitude axis, and each longitude meets the equator in two points
rather than one. So Step (2) is (doubly) violated. Also, for every point on the
longitude axis there is a latitude, but again each latitude meets the longitude axis
in two points. So Step (3) is also violated. In fact, all of the longitudes meet all of
the latitudes in two points. Thus Step (4) is also violated. Finally observe that the
North and South Poles are on every longitude and on no circle of latitude. Thus
Step (5) is (doubly) violated. Hence this attempt at writing S 2 as S 1 × S 1 violates
every Step of the Checklist.
In the Exercises you will have the chance to consider some other spaces which
are incorrectly expressed as products, and determine which steps of the Checklist
are and are not violated.
6. The Topological Uniqueness of Products
In Section 3, we saw that both a cylinder and an annulus can be expressed as

the product S 1 × I. It may seem odd that two different spaces can be expressed
by the same product. If we applied this to products of numbers, it would mean
that you could get more than one result by multiplying the same two numbers.
This seems ridiculous. We know that 2 times 3 equals 6, no matter whether you
do the multiplication on the North Pole, in a Halloween costume, or standing on
your head. So what’s the catch? How can we get two different spaces by taking the
product of the same two factors?
The answer is that your intuition is right. It’s impossible to get different re-
sults. Rather, we had mistakenly assumed that a cylinder and an annulus were
different; when they are actually topologically the same. In Figure 16, we can see
a deformation taking a cylinder to an annulus by pulling open the top circle while
keeping the bottom circle fixed. So a cylinder and an annulus have the same in-
trinsic topology, and there is only one space which describes S 1 × I after all. Phew.
These spaces are geometrically different, but we’ll get to that later in the chapter.
Figure 16. A cylinder has the same intrinsic topology as an annulus.
In fact, this result is true for all products as we see in the following theorem,
whose proof you could see if you take a rigorous course on topology.
Theorem 7.1. Any pair of spaces which can be expressed as the product of the
same two factors has the same intrinsic topology.
Theorem 7.1 is useful for showing that a given space is not a particular product.
For example, in the last section we showed that S 2 is not the product of an equator
and a longitude. However, this proof required finding at least one step of the
Checklist which is violated, and when we were done we could only conclude that
7. THE DIMENSION OF PRODUCT SPACES 167
S 2 was not a product of these particular circles. We didn’t know that S 2 couldn’t
be expressed as S 1 × S 1 in some other way. Now since we know that a sphere is
topologically distinct from a torus, we know immediately from Theorem 7.1 that
there is no way that the sphere can be expressed as S 1 × S 1 .
7. The Dimension of Product Spaces
So far, we have only been considering products of two 1-dimensional spaces.

You may have noticed that all of spaces that we obtained in this way are 2-
dimensional. Also, we know that 1 + 1 = 2 (even if you’re on the North Pole,
in a Halloween costume, or standing on your head). It turns out that the same rule
of addition of dimensions holds for products of any dimension. We state this in the
next theorem, whose proof you might see if you take an advanced course on point
set topology.
Theorem 7.2. If X is a space of dimension n and Y is a space of dimension
m, then the product X × Y is a space of dimension n + m.
If it seems strange to you that when we take the product of spaces we add the
dimensions rather than multiplying them, think about what we do to exponents
when we multiply numbers raised to powers. For example, 102 × 103 is 105 and not
106 . This is consistent with the product R × R = R2 .
According to Theorem 7.2, the product of a 1-dimensional space X and a 0-
dimensional space should be a space of dimension 1 + 0 = 1. You might have
thought that no space could have fewer than one dimension. In fact, the only 0-
dimensional space is a single point y. So we can visualize the product X × {y}, as
the copy of X that we described in Step 2 of the Checklist (illustrated in Figure 3
for R2 ). So it’s not surprising that X × {y} has the same dimension as X. This is
true regardless of the dimension of X. Since such a product is just a copy of X, we
say that it is trivial, meaning it’s not interesting. This is similar to the fact that
we don’t factor a number n into n × 1. Every number could be written this way,
but it wouldn’t be interesting. So we don’t bother doing it.
Thus if we want to write a 2-dimensional space as a non-trivial product, we
know that each factor must be a 1-dimensional space. However, the only 1-
dimensional spaces are S 1 , I, R, and R+ (which represents a ray that starts at
0 and includes all the the non-negative real numbers). Note that we don’t include
the ray R− (of non-positive numbers) in our list, because it is topologically the
same as R+ . Hence the only 2-dimensional spaces which are non-trivial products
are obtained as a product of two of these 1-dimensional spaces. So far, we have
seen the products R × R, I × I, S 1 × I, and S 1 × S 1 . In the Exercises, you will
construct the products I × R+ , S 1 × R, S 1 × R+ , and R+ × R+ .
If we want to construct a 3-dimensional space as a non-trivial product, then
one factor must be 2-dimensional and the other factor must be 1-dimensional. In
order for our 3-dimensional space to be a manifold, it can’t have any boundary.
This means that neither factor can have any boundary. If in addition we want the
3-manifold to have finite volume, then both of the factors have to be finite. In
particular, this means the 1-dimensional factor must be S 1 and the 2-dimensional
space must be a surface with finite area. Recall from Chapter 6 that by the Clas-
sification of Surfaces, version 2, every surface with finite area is either S 2 , nT 2 ,
or nP 2 for some positive integer n. Thus the only 3-manifolds with finite volume
that can be obtained as non-trivial products have one of the three forms S 2 × S 1 ,
nT 2 × S 1 , or nP 2 × S 1 . We explain how to visualize S 2 × S 1 and nT 2 × S 1 in the
next section. You will explore the product nP 2 × S 1 in the Exercises.
8. Visualizing S 2 × S 1 and nT 2 × S 1
As we often do, we begin by tackling an easier problem. In this case, the easier
problem is to visualize S 2 ×I. Imagine an intact hard-boiled egg where the yolk has
completely disappeared without a trace, or possibly never existed. Now suppose
that you pierce the egg white all over by toothpicks that go from the outside where
the shell would be to the inside where the yolk would begin. In Figure 17, we
illustrate some of the horizontal toothpicks, but you should imagine that there is
a toothpick piercing through every point from the outer layer towards the center
where the yolk would be. This yolkless hard-boiled egg is S 2 × I, where the outer
layer is the S 2 -axis, and one of the toothpicks is the I-axis. You should check that
this satisfies the five steps of the Checklist.
Figure 17. S 2 × I can be visualized as the egg white of a hard-

boiled egg that has been pierced everywhere by toothpicks.
Now that we can visualize S 2 × I, we can use it to create S 2 × S 1 . Recall that

if we glue together the endpoints of an interval, it becomes a circle. So in order to
turn S 2 × I (which is an S 2 of intervals and an interval of S 2 ’s) into S 2 × S 1 (which
is an S 2 of circles and a circle of S 2 ’s), we want to glue together the endpoints of
each toothpick so that it becomes a circle. Thus we abstractly glue the outer layer
of the egg white to the inner layer. We use the word “abstractly” here to emphasize
that this gluing cannot be done in R3 .
In Figure 18, we see A. 3D-Girl who is on vacation in S 2 × S 1 . She trips over a
stone feels like she’s losing her balance, so she grabs hold of a nearby stick to steady
herself. Unfortunately, at just that moment, someone starts pulling her down by
the leg. When she returns from her vacation, she tells all of her friends never to go
to S 2 × S 1 because the people there are so aggressive and rude.
Having visualized S 2 × S 1 , we are now ready to imagine nT 2 × S 1 for all n ≥ 1.
We begin again with an easier visualization, which in this case is to picture a
chocolate bunny. Such Easter treats are usually made of a thick layer of chocolate
which is hollow inside. The mathematical name for shapes of this form is thickened
surfaces. In fact, our yolkless hard-boiled egg is an example of a thickened surface,
8. VISUALIZING S 2 × S 1 AND nT 2 × S 1 169
Figure 18. Someone is pulling A. 3D-Girl’s leg.
where the surface is a sphere. Now imagine a thick layer of chocolate in the shape of
an n-holed torus, which is empty inside just like a chocolate bunny. Note that most
chocolate bunnies have an opening at the bottom where you can stick your finger
to make the chocolate bunny into a finger puppet (though if you’re not careful the
chocolate bunny could break into pieces or melt while you’re doing your puppet
show). Our thickened n-holed torus has no such opening.
A thickened n-holed torus is in fact nT 2 × I. The nT 2 -axis is the outer layer
of chocolate and the I-axis is a segment (or toothpick) perpendicular to the nT 2 -
axis that could be used to measure the thickness of the chocolate. For example, in
Figure 19 we illustrate half of the thickened torus T 2 × I. This half looks like a
piece of macaroni. Hence you can imagine T 2 × I either as two pieces of macaroni
glued together on their ends or as a bagel made of chocolate which is hollow inside
(like a chocolate bunny, but in the shape of a bagel). This description of nT 2 × I is
both an nT 2 of intervals (or toothpicks) and an interval of nT 2 (chocolate layers).
Again you should check that this description satisfies the steps of our Checklist.
Figure 19. Half of a thickened torus.
In order to get nT 2 × S 1 , we again want to glue together the endpoints of each

interval so that it becomes a circle just as we did to get S 2 × S 1 from S 2 × I. We do
this by abstractly gluing the inside copy of nT 2 to the outside copy of nT 2 . Thus
at each point of the surface nT 2 there is now a circle where there had been a line
segment.
It’s as if we glued the inside surface of a chocolate bunny to the outside surface
of the chocolate bunny, again assuming the bunny has no finger hole in the bottom.
The problem with such a glued up bunny is that, because it has no boundary, there
would be no way to take the first bite. Wouldn’t that be frustrating if someone
Figure 20. A. Square cannot give B. Triangle a chocolate sphere,

because a sphere can’t be contained in Flatland.
gave you a (chocolate bunny) × S 1 as a present and you couldn’t even take a bite?
But for that matter, nobody could give you such a present because one 3-manifold
can’t fit inside of another.
A. Square would have the same problem if he wanted to give B. Triangle a
chocolate sphere as a present (see Figure 20). Even though a sphere is 2 dimensional,
there would be no way that a sphere could fit inside of Flatland unless Flatland
itself were the sphere. But then A. Square and B. Triangle would be part of the
sphere, and if it were made of chocolate, then a 3-dimensional person might eat
Flatland with A. Square and B. Triangle in it. But we have to stop this line of
reasoning because it’s becoming entirely too gruesome for a math book.
9. Geometric Products
So far, we’ve learned what a product is and how to correctly and incorrectly
construct products. We’ve also learned how to construct and visualize various
2-dimensional and 3-dimensional product spaces. We now want to consider the
geometry of products as well as their topology. Once we understand the definition
of a geometric product, we’ll go back through some of the examples of 2-dimensional
and 3-dimensional products to see if we can construct them in a geometric way.
In order to be able to say anything about the geometry of a product, we need
to construct the product more carefully than we’ve been doing so far. In particular,
we need to add restrictions on the size of the copies of the two factors as well as on
the angles between them.
Definition. A product X × Y is said to be geometric if all of the following
conditions hold:
(1) All subsets of the form X × {y} have the same size.
(2) All subsets of the form {x} × Y have the same size.
(3) Every subset of the form X × {y} is perpendicular to every subset of the
form {x} × Y .
As a first example, let’s consider the product S 1 × I. We saw earlier, that
S × I could be viewed either as a cylinder or as an annulus . In the cylinder,
1
all of the circles S 1 × {y} are the same size, and all of the intervals {x} × I are
the same length (see the left side of Figure 21). Thus conditions (1) and (2) of
the definition of a geometric product are satisfied. Moreover, each interval {x} × I
is perpendicular to each circle S 1 × {y}, as required by condition (3). Hence, the
cylinder is indeed a geometric representation of the product S 1 × I.
Let’s see what happens when we view an annulus as the product S 1 × I. In this
case, the intervals {x} × I are all the same size and intersect the circles S 1 × {y}
10. GEOMETRIC PRODUCTS OF FLAT SPACES 171
Figure 21. A cylinder is a geometric S 1 × I, but an annulus is not.
at right angles. Thus, conditions (2) and (3) are satisfied. However, as can be seen
on the right side of Figure 21, the concentric circles S 1 × {y} are not all the same
size, failing condition (1) of the Definition. Thus topologically an annulus is S 1 × I
but the product is not geometric.
10. Geometric Products of Flat Spaces
We’d like to see how we can use the geometry of the factors in a geometric
product to understand the geometry of the product. Since we understand flat
geometry best, let’s look at geometric products of flat spaces first. In Chapter 5,
we defined an n-dimensional manifold to be flat if its local intrinsic geometry is
the same as that of Rn . In other words, any measurement you can make in an
arbitrarily small region of a flat n-manifold would have the same value if you made
the same measurement in an arbitrarily small region of Rn . This includes lengths,
areas, angles, volumes, etc.. Since we want to be able to consider spaces which have
boundaries, we need to extend the definition of flat to n-dimensional spaces that
are not necessarily manifolds.
Definition. An n-dimensional space is flat if it has the same local intrinsic
geometry as a subset of Rn .
It follows from this definition that any 2-dimensional space that you can draw
on a piece of paper is flat. For example, a square, an octagon, and a rectangle are
all flat. However, be careful. Just because these shapes are all flat, doesn’t mean
they all have the same local intrinsic geometry. They each have the same local
intrinsic geometry as a subset of the plane, but these subsets are not the same.
Suppose we want to compare the local intrinsic geometry of a square, an oc-
tagon, and a rectangle. If we consider a point in the interior of any of these shapes,
a small region around it will look exactly like a comparable small region of R2 . So
the local intrinsic geometry around interior points of all three shapes is the same.
But now let p be one of the corners of the square. Then any small region around p
is a 90◦ wedge that looks like a quarter of a pizza (which you could eat as a single
slice if either the pizza is very small or you are very hungry). However, there is
no point on an octagon which has a region around it that looks like a 90◦ wedge.
Rather, a region around one of the corners of an octagon looks like a 135◦ wedge.
On the other hand, on a rectangle a small region around a corner is a 90◦ wedge,
just like on a square. Thus the local intrinsic geometry of a square and a rectangle
are the same, and are both different from that of an octagon.
We saw in Section 6 that there are only four 1-dimensional spaces, and they
are R, R+ , I, and S 1 . The question is which of these are flat? Certainly R itself is
flat, and R+ and I are also flat since they are both subsets of R. But what about
S 1 ? Since S 1 is a 1-manifold, it is flat if its local intrinsic geometry is the same as
R. This doesn’t mean that S 1 has to be contained in R (which it isn’t). Rather, it
means that every point of S 1 has to have a small region around it whose intrinsic
geometry is the same as that of a comparable small region of R.
At first this may seem unlikely. Since a round circle is curved, how could it also
be flat? One way to see that S 1 is indeed flat is to represent it with a gluing diagram
which is an interval whose endpoints are glued together. Any point other than an
endpoint certainly has the same local intrinsic geometry as R. In Figure 22, we
consider a small region around the endpoints of the interval. Since the endpoints
p and q are glued together, a region around this glued up point consists of the
two grey segments illustrated in the figure. However, the grey segments are glued
together at the points p and q. Thus they become a single grey segment, which is
identical to the grey segment around the point r on R. Thus every point on the
glued up segment has the same local intrinsic geometry as R. Hence this gluing
diagram for S 1 is flat.
Figure 22. The local intrinsic geometry at the endpoints of the

interval is the same as at the point r.
Right now you may be saying “OK sure, a gluing diagram for S 1 is flat just
like a gluing diagram for a torus is flat, but what about a round circle?” In fact,
unlike a torus, we can take the gluing diagram for S 1 and glue it up into a round
circle without shrinking it or stretching it in any way. Keep in mind that the gluing
diagram for S 1 is 1-dimensional—it’s not like gluing up the ends of a piece of string
cheese. It may help to recall from Chapter 3 that we can also glue up the diagram
for a cylinder without shrinking or stretching it in any way. In fact, we can see this
for ourselves by gluing together one pair of opposite sides of a piece of paper. Thus
the local intrinsic geometry of a flat gluing diagram for a cylinder is exactly the
same as that of a curved cylinder. Now, we can think of our round circle as one of
the boundary circles of a piece of paper that we’ve glued up into a cylinder. Thus
the local intrinsic geometry of the round circle is exactly the same as that of the
gluing diagram. Hence, whether we think of the round circle as a glued up a line
segment or as one of the boundary circles of a cylinder, we reach the conclusion
that the round circle is flat. This means that in fact every 1-dimensional space is
flat.
The following theorem (whose proof you might see in a differential geometry
course) tells us that taking geometric products of flat spaces is as nice as it could
be.
Theorem 7.3. A geometric product of flat spaces is flat.
For example, in addition to the cylinder in Figure 21, the four spaces illustrated
in Figure 23 are geometric products because the horizontal and vertical factors all
have the same length and meet at right angles. Hence by Theorem 7.3 we can
immediately conclude that each of these spaces is flat.
11. A FLATLAND-FRIENDLY GEOMETRIC S 1 × I 173
Figure 23. By Theorem 7.3, these geometric products of flat

spaces are flat.
You may recall that we know from Theorem 5.1 in Chapter 5 that a surface
is flat if and only if the sum of the angles of any triangle in the surface is equal
to 180◦ . However, using Theorem 7.3 is easier than checking the sums of angles
of every triangle, and applies to spaces of any dimension with boundary as well
as to surfaces (remember that we use the word “surface” interchangeably with 2-
manifold). In fact, since every 1-dimensional space is flat, the geometric product of
any two 1-dimensional spaces is flat. Thus any geometric product which has factors
chosen from I, S 1 , R or R+ is flat. How many flat 2-dimensional spaces does this
give us?
11. A Flatland-Friendly Geometric S 1 × I
A. Square is a little frustrated because he heard us say that S 1 × I is flat, but

when he pictures it as an annulus, it isn’t flat. Sure he can picture a flat S 1 × I
using a gluing diagram like the second picture in Figure 23. But he wants to know if
there’s a way to interpret his normal picture of an annulus as a geometric product.
In fact, there is a way for A. Square to visualize the geometric product S 1 × I
without a gluing diagram that is reminiscent of how we explained a 3-dimensional
cube to him back in Chapter 2. Recall that we drew the “Flatland-friendly” picture
of a cube in Figure 24 and told him that the inner square is the same size as the
outer square, but farther away in the third dimension. While the third dimension
doesn’t really mean anything to him, he was able to suspend his disbelief long
enough that he could accept this explanation. As 3-dimensional people we can see
Figure 24 as a tunnel with a square cross section, where the smaller square is just
deeper inside the tunnel.
Figure 24. A Flatland-friendly drawing of a 3-dimensional cube.

We now show A. Square the annulus in Figure 25 and explain that he can see
this as a geometric product of S 1 and I by imagining that all of the concentric
circles are really the same size, the inner ones are just farther away in the third
dimension. In fact, this picture of the annulus is not much different from that
of the cube. We can see it as a tunnel with a circular cross section, where the
smaller circles are deeper inside the tunnel. Furthermore, since parallel circles on a
cylinder are geodesics, we tell A. Square that the concentric circles on the annulus
are actually geodesics. This means that if two points are on the same circle, then
the shortest path between them follows the circle. For example in Figure 25 the
grey arc is the shortest path between its endpoints.
Figure 25. The grey arc is a geodesic in this Flatland-friendly

illustration of a geometric S 1 × I.
It would be hard for A. Square to check that the sums of angles of any triangle in
this geometric S 1 × I are all 180◦ , since the sides of a triangle have to be geodesics.
For example, if he considers the grey shape in Figure 26, only one of the three
arcs in the boundary looks like a geodesic. But if we draw the same shape on the
cylinder, we see that all three arcs are geodesics and hence the shape is indeed a
triangle. This will seem quite strange to A. Square. However, he remembers that
in Figure 24 the trapezoidal shapes were actually squares, so he accepts that this
bulging shape is actually a triangle.
Figure 26. A triangle in the geometric S 1 × I.
12. Flat 3-Dimensional Spaces as Geometric Products
In this section, we will learn to visualize flat 3-dimensional spaces which are
geometric products of a flat 2-dimensional space and a flat 1-dimensional space. Any
1-dimensional space will do, since all of them are flat. For the flat 2-dimensional
12. FLAT 3-DIMENSIONAL SPACES AS GEOMETRIC PRODUCTS 175
space, we can take any of the geometric products of two 1-dimensional spaces or
one of the spaces T 2 or K 2 which we already know are flat.
As a first example, let’s consider a geometric product of a square X = I × I
and an interval Y = I. On the left side of Figure 27, we draw a copy of the square
as the horizontal X-axis on the bottom and copies of the interval vertically at every
point on the square. On the right side of Figure 27, the vertical interval in the left
corner is the Y -axis, and you should imagine that at every point on it there is a
horizontal square. As we can see from the figure, the product (I × I) × I is a square
of intervals and an interval of squares, satisfying all of the steps of our Checklist.
Since the product is geometric, by Theorem 7.3 the cube is flat. In fact, we already
know a cube is flat because it has the same local intrinsic geometry as a subset of
R3 .
Figure 27. The geometric product of a square and an interval is a cube.
Observe that we will get the same product if we put the square factor on any
one of the faces of the cube, as long as the interval factor is perpendicular to the
square. Also, notice that we read the product (I × I) × I as “the product of a
square and an interval”. Whereas, we read the product I × (I × I) as “the product
of an interval and a square”. In either case, the geometric product is a cube.
For our next example, let’s take the flat torus T 2 as our 2-dimensional space
and then use any one of the 1-dimensional spaces as the other factor. For example,
in Figure 28, we illustrate the geometric product of a flat torus T 2 and an interval
I. In both pictures, the torus factor is on the bottom and the interval factor is
vertical. At every point in the torus we can visualize a perpendicular interval, and
at every point on one of these intervals we can visualize a perpendicular torus. Thus
each of the horizontal cross sections of this space is a flat torus that is identical to
the one on the bottom.
Figure 28. The geometric product of a flat torus and an interval.

In Chapter 5 we visualized flat 3-manifolds using a cubical gluing diagram with

2-dimensional arrows on the faces. We can do the same thing for any geometric
product of a 2-dimensional space whose factors are a square gluing diagram and a 1-
dimensional space. For example, we can turn the illustration of T 2 × I in Figure 28
into a cubical gluing diagram with 2-dimensional arrows as follows. Since each
horizontal cross section of T 2 × I is a flat torus with the same gluing arrows as the
square at the bottom, the front and back faces of the cube must be glued together
straight, and the left and right faces of the cube must be glued together straight.
Also, since there is an interval perpendicular to each horizontal cross section, none
of the endpoints of the intervals should be glued together or to an endpoint of
another interval. This means that the top of the cube cannot be glued to the
bottom of the cube. Thus the gluing arrows on the cube must be as illustrated in
Figure 29.
Figure 29. A cubical gluing diagram for T 2 × I.
Now we can use the gluing diagram for T 2 × I to create gluing diagrams for
geometric products of T 2 and other 1-dimensional spaces. For example, the gluing
diagram for T 2 × R is like that of T 2 × I, except instead of being a cube with a top
and a bottom, the gluing diagram would have the form of an infinite column with
a square cross section. This is analogous to the gluing diagram for S 1 × R that we
saw in Figure 23 which was an infinite strip with an interval as the cross section.
Similarly, the gluing diagram for T 2 × R+ would be a column which went on forever
upward but had a square face on the bottom. In Figure 30, we use dotted lines to
indicate that the gluing diagram goes on forever in that direction.
Figure 30. Gluing diagrams for S 1 × R and S 1 × R+ with dotted

lines indicating the gluing diagram goes on forever.
The one remaining geometric product which has the flat torus as one factor
and a 1-dimensional space as the other factor is T 2 × S 1 . In this case, we start with
the gluing diagram for T 2 × I, and then we want each interval to become a circle.
13. A GEOMETRIC nT 2 × S 1 177
Figure 31. The product T 2 × S 1 is T 3 .
This means that we want to glue together the endpoints of each of the intervals in
Figure 28. As a result, the top face of the cube in Figure 29 must be glued to the
bottom face. We can see from Figure 31 that the resulting 3-manifold is the 3-torus
T 3.
Note that just as we could write the product of a square and an interval as
either (I × I) × I or I × (I × I), we can write the product of a torus and a circle as
either (S 1 ×S 1 )×S 1 or S 1 ×(S 1 ×S 1 ). Thus we don’t actually need any parenthesis
in these products. We can simply say that a cube is I ×I ×I and T 3 is S 1 ×S 1 ×S 1 .
In the Exercises, you will explore geometric products where one factor is a flat
K 2 and the other factor is a 1-dimensional space.
13. A Geometric nT 2 × S 1
Observe that when we take a geometric product of two homogeneous spaces,

the local intrinsic geometry of the product will also be homogeneous. By contrast,
in Section 8, we learned to visualize nT 2 ×S 1 for n > 1 as a thickened n-holed torus
made of chocolate where the inside layer is glued to the outside layer. While such
a visualization is appealing (as chocolate always is), the geometry of this nT 2 × S 1
is hard to analyze. The one thing we can say is that it’s not homogeneous, since
the curved n-holed torus itself is non-homogeneous.
However, this is not the only way to visualize nT 2 × S 1 . In Chapter 6, we
learned that for n > 1 we can obtain nT 2 abstractly as a glued up 4n-gon where all
of the vertices are glued together into a single anti-cone point. While this glued up
surface still isn’t homogeneous, it’s only “bad” at the anti-cone point and otherwise
it’s flat. So let’s see if we can understand the geometry of a geometric product of
this nT 2 and an interval. We illustrate the geometric product 2T 2 × I in Figure 32.
You should imagine that there is an identical octagonal gluing diagram at every
horizontal level of the product as there was an identical square gluing diagram at
every horizontal level of Figure 28.
Let v denote the anti-cone point of the glued up 4n-gon. We saw in Section 7
that the product of any space X with a single point is just a copy of X. Thus the
product of the anti-cone point v with an interval is an interval. It follows that the
product nT 2 × I contains the interval {v} × I of anti-cone points and otherwise it’s
flat. What we mean by an interval of anti-cone points is that for any point x on
I, the cross section of nT 2 × I of the form nT 2 × {x} contains the anti-cone point
(v, x) which has 180◦ (4n − 2) degrees around it (you may want to see if you can
prove that this is indeed the sum of the angles of a 4n-gon). So we can think of
{v} × I as an “anti-cone interval” inside of the geometric product nT 2 × I. It’s the
only “bad interval” in nT 2 × I and otherwise nT 2 × I is flat.
Figure 32. The product 2T 2 × I.
As we saw in the last section, to go from the product of a surface and an

interval to the product of a surface and a circle, we simply glue together the top
and bottom face of our 3-dimensional gluing diagram. Thus we can obtain the
geometric product nT 2 × S 1 by gluing together the top and bottom 4n-gon in
nT 2 × I (illustrated for the octagon in Figure 33). This will give us nT 2 × S 1 which
is flat everywhere except along an “anti-cone circle”, which we obtain by gluing
together the endpoints of the “anti-cone interval” in nT 2 × I.
Figure 33. The product 2T 2 × S 1 is flat except for an anti-cone

circle along the glued up vertices.
We saw in Chapter 6 that we can obtain every finite surface as a glued up

polygon which is flat except for at some number of cone points or anti-cone points.
Thus we can take the geometric product of the gluing diagram for any finite surface
and S 1 to get a 3-manifold which has flat geometry everywhere except at a finite
number of cone circles or anti-cone circles.
On one hand, such 3-manifolds are nice because they are flat almost every-
where. But on the other hand, such 3-manifolds are not nice because they’re not
homogeneous. If we started with a homogeneous surface and took a geometric
product, we would end up with a homogeneous 3-manifold. In the next chapter
we’ll learn how to create homogeneous geometries for nT 2 and nP 2 .
14. A GEOMETRIC S 2 × S 1 179
14. A Geometric S 2 × S 1
We would now like to create S 2 × I and S 2 × S 1 as geometric products. We

begin by recalling from Section 8 that we visualized S 2 × I as a yolkless hard-boiled
egg (illustrated in Figure 17). Since we want all of the layers of our S 2 × I to have
homogeneous geometry, let’s imagine that the hard-boiled egg is perfectly round
so that each of the layers from the outside to the inside is geometrically a sphere.
This is nice, but it’s still not a geometric product, since the concentric spheres get
smaller as we go towards where the yolk would be.
Before we try to visualize S 2 × I as a geometric product, let’s remember that
when we taught A. Square to picture S 1 × I as a geometric product, we started by
reminding him of the Flatland-friendly picture of a cube. So we begin our discussion
of a geometric S 2 ×I by recalling our drawing of a 4-dimensional cube from Chapter
2 (see Figure 34). In particular, since we have no 4-dimensional perspective, we
had to accept that the inner cube is the same size as the outer cube but farther
away in the fourth dimension. In fact, the 3-dimensional cross sections which are
parallel to the outer cube are all cubes of exactly the same size.
Figure 34. A 3-dimensional drawing of a 4-dimensional cube.
Now we want to think of the geometric product S 2 × I in the same way. In

particular, we imagine that, in spite of the way it looks, our yolkless hard-boiled
egg is spherical and all of the concentric spheres from the outside to the inside are
exactly the same size. When we think of S 2 × I in this way, we have to accept that
if two points are on the same spherical layer, then the geodesic between them will
be part of a great circle contained in that sphere, rather than an arc which cuts
across many spherical layers.
Thus the edges of a triangle whose vertices are all on the same spherical layer
must all be arcs of great circles contained in that sphere (see Figure 35). This
means that the sum of the angles of such a triangle will necessarily be greater than
180◦ . In particular, this shows that the geometry of S 2 × I is not flat. But this is
not a surprise, since we knew that S 2 was not flat.
By contrast with our spherical cross sections, we now consider a horizontal
cross section of the geometric S 2 × I. In the drawing on the left of Figure 36, this
cross section looks like an annulus. However, since all of the concentric spheres in
S 2 × I are actually the same size, all of the circles in this cross section should also
be the same size. Thus we can either visualize this annulus the way that A. Square
imagines a geometric S 1 × I, or we can visualize it as a cylinder rather than as an
Figure 35. The sum of the angles of this triangle in a spherical

layer of S 2 × I is greater than 180◦ .
Figure 36. A horizontal cross section of a geometric S 2 × I is a

geometric S 1 × I.
annulus. Either way, this means that the sum of of the angles of every triangle in
this cross section must be 180◦ .
Thus at any point in S 2 ×I, the local geometry is flat if you look in the direction
of a cross section of the form S 1 × I (as in Figure 36). Whereas, the local geometry
is spherical if you look in the direction of a spherical cross section (as in Figure 35).
In other words, the local geometry at a point depends on the direction in which
you look. We’ll discuss this strange phenomenon in the next section.
But in the meantime, let’s create the geometric product S 2 × S 1 by abstractly
gluing the inside and outside spheres of our geometric S 2 × I. This means that each
line segment (which could be a toothpick measuring the thickness of the egg white)
becomes a circle of S 2 × S 1 . As we saw earlier, the geometry of S 2 is homogeneous
and the geometry of S 1 is homogeneous, and their product is geometric. Thus
we end up with a nice homogeneous, geometric representation of the 3-manifold
S 2 × S 1 , even if there is something a little strange going on when you look in
different directions.
15. Isotropic and Non-isotropic Spaces
In order to better understand how the geometry of a space can change depend-
ing on the direction in which you look, let’s suppose that we are talking about
the amount of illumination in a region rather than the local geometry of a region.
Suppose that you are in the dessert at noon with the sun overhead, there is not
a cloud in the sky, and there are no houses or trees to create shadows. No mat-
ter where you walk, the amount of light is the same. In this sense you could say
16. EXERCISES 181
that the illumination is “homogeneous”. However, if you look directly up at the

sun, the light will be more intense than if you look straight ahead; and if you look
straight ahead there will be more light than if you look down at your shoes. In this
situation, you could say that the amount of light that hits your eye depends on the
direction in which you look.
While this seems normal for illumination, it might seem strange to live in
a universe where the geometry changed when you looked in different directions.
It order to talk about such strange spaces more easily, we make the following
definition.
Definition. A space is said to be isotropic if the local geometry at every point
is the same in all directions.
The word isotropic comes from the Greek words iso (meaning equal) and tropos
(meaning direction). Next time you are out with a friend on a cloudless day, you
could remark that the amount of light is “homogeneous but not isotropic”. Your
friend could either be impressed with your vocabulary or think you’re crazy. But
either way, it would give you a chance to talk about the math you’re learning.
We saw in the last section that the geometric product S 2 × I is not isotropic
because if you look in the direction of a spherical layer the geometry is different
than if you look towards the inside of the egg white. Also, the local geometry at
a point on the inside sphere or outside sphere of S 2 × I will be different from that
of a point in the interior of S 2 × I because such points are in the boundary of the
space. Thus S 2 × I is neither isotropic nor homogeneous.
On the other hand, we saw in the last section that the geometric S 2 × S 1 is
homogeneous, and since the geometric S 2 × I was not isotropic, it now follows that
the geometric S 2 ×S 1 is not isotropic. This means that at any point in S 2 ×S 1 if you
look in different directions, the geometry is different, but this strange directional
geometry will be exactly the same at every point in the manifold.
The idea of a manifold which is homogeneous but not isotropic is indeed strange.
This is quite different from T 3 , K 3 , and the other flat 3-manifolds that we studied in
Chapter 5. In particular, every flat 3-manifold is isotropic as well as homogeneous,
since its local geometry is the same as R3 . Even the 3-dimensional sphere S 3 turns
out to be isotropic as well as homogeneous.
16. Exercises
1. Show that an infinite cylinder can be expressed as the product of S 1 and R.

2. Recall that we use the symbol R+ to represent the ray consisting of all real num-
bers greater than or equal to 0. Describe the following 2-dimensional products:
I × R+ , S 1 × R+ , R + × R+ .
3. How many topologically distinct 2-dimensional spaces are there which are prod-
ucts of two 1-dimensional spaces? How many topologically distinct 2-manifolds
are there which are products of two 1-dimensional manifolds?
4. List all 2-manifolds with finite area which can be written as a non-trivial prod-
uct. Explain how you know you have them all.
5. Explain what would go wrong if you tried to express a Möbius strip as S 1 × I

where the S 1 -axis is a circle other than the center circle of the Möbius strip and
the I-axis is the vertical interval illustrated in black on the right in Figure 14.
6. Explain why neither S 2 nor a Möbius strip can be a product of two 1-dimen-
sional spaces.
7. Could a pentagon be expressed as I × I with all of the segments straight as
illustrated in Figure 37. If not, name a step of the Checklist which is violated.
8. Illustrate a triangle and a hexagon as the product of two intervals. Then explain
why neither of these products is geometric.
9. Are the descriptions of nT 2 × I and nT 2 × S 1 in Section 8 geometric? Explain
why or why not.
10. In this problem we want to determine what goes wrong if we try to express a
plane as a product of a circle and a line as illustrated in Figure 38. In particular,
list all of the steps from the Checklist which are violated and explain how.
Note that the axes in Figure 38 are in bold black. Besides the S 1 -axis,
which has radius 1, we’ve drawn parallel circles with radii 2 and 3. Similarly,
we’ve drawn a few lines parallel to the R-axis. Even though the figure includes
only a few circles and lines, you should imagine that there are circles of every
radius and a vertical line through every point.
11. Suppose S = X × Y and a and b are distinct points of X. Prove that {a} × Y
and {b} × Y are disjoint.
16. EXERCISES 183
12. Create a sphere as a glued up square. Then describe the geometry of a geometric
product of your glued up square and S 1 . In particular, discuss where it’s flat
and where it’s not flat.
13. Draw a picture of a geometric nP 2 × S 1 which is flat everywhere except along
some number of circles. How many circles of cone points or anti-cone points
does the space have?
14. Describe the geometric products (S 1 × I) × S 1 and (S 1 × S 1 ) × I, and illustrate
each with a cubical gluing diagram. Are the two products the same or different?
15. Draw a picture of what you would get if you physically glued up the faces of
the gluing diagram in Figure 29.
16. Draw cubical gluing diagrams for K 2 × I, K 2 × R, K 2 × R+ , and K 2 × S 1 . Is
one of the spaces a familiar flat manifold?
17. Describe the following 3-dimensional products: S 2 × R+ , S 2 × R,
R + × R+ × R+ .
18. Imagine that the picture on the left in Figure 36 is supposed to be a geometric
S 2 × S 1 rather than a geometric S 2 × I. Describe the cross section illustrated
in the picture. Is the cross section flat?
19. Can you find a manifold other than S 2 × S 1 which is homogeneous but not
isotropic?
20. Can you find a manifold which is isotropic but not homogeneous?
21. Let A be a surface which is not flat, and suppose that A × S 1 is a geometric
product. Prove that A × S 1 is not flat.
22. Suppose that we live in a universe that we knew was one of S 2 × S 1 , P 2 × S 1 ,
or T 3 . How could we tell the difference between these three spaces?
23. Is the product operation distributive over the connected sum operation? In
other words, for any spaces X, Y , and Z, is it always true that X × (Y #Z) =
(X × Y )#(X × Z)? Give a proof or counterexample.
CHAPTER 8
Geometries of Surfaces
Topics:
• Euclid’s axioms
• Spaces that don’t satisfy one of the axioms
• Alternative axioms
• Spherical geometry
• Areas of disks in different surfaces
• Hyperbolic geometry
• Helping A. Square visualize a hyperbolic plane
• Homogeneous geometries for all finite surfaces
1. Euclid’s Axioms
Recall that a surface is flat if it has exactly the same local intrinsic geometric
properties as a plane. This means that all flat surfaces are homogeneous and
isotropic because the plane is homogeneous and isotropic. Also, all triangles in
a flat surface have angle sum equal to 180◦ because all triangles in the plane have
angle sum equal to 180◦ .
With all of our attention on the local geometry of flat surfaces, we might assume
that the only global geometric property which distinguishes a plane from other flat
surfaces is that it has infinite area. But we saw in Chapter 3 that an infinite cylinder
is flat, yet it also has infinite area. In order to distinguish the global geometry of an
infinite cylinder from that of a plane, we could observe that some of the geodesics
on a cylinder are finite whereas no geodesics on the plane are finite. Perhaps what
distinguishes the global geometry of a plane from all other flat surfaces is that its
area is infinite and all of its geodesics are infinite.
Let’s consider the geometry of the surface we get by removing a single point
from the plane. We can’t analyze the local geometry at the missing point, because
the missing point isn’t part of the space. If we look at the local intrinsic geometry
at any other point, it’s still exactly the same as it is at any point in the plane. In
particular, no matter how close a point is to the hole, there is a region around it
which is small enough that it doesn’t contain the hole. This means that the local
intrinsic geometry of a region around any point in our surface is identical to that of
a region in the plane. Hence the plane with a point missing is also flat. Also, all of
its geodesics are infinite, just like the geodesics in the plane. So what distinguishes
the global geometry of a plane with a hole from that of a plane without a hole?
185
186 8. GEOMETRIES OF SURFACES
We could keep coming up with global geometric properties that enable us to

distinguish one flat surface after another from a plane, but why do all this work
when it’s been done for us more than two thousand years ago? Euclid, who was a
philosopher around 300 BC, wrote a list of five axioms that completely describe the
local and global geometry of a plane. All other statements about the plane, such as
the fact that the sum of the angles of a triangle is 180◦ or the fact that a plane has
infinite area, can be proven from just these five axioms. We now consider Euclid’s
axioms one by one to understand what they mean and to get a sense of what sorts
of spaces don’t satisfy them.
Axiom 1. For any two distinct points, there is a geodesic that passes through
both points.
Even though Axiom 1 seems quite innocuous, it is easy to find spaces which
do not satisfy it. In fact, A. Square found such a space. Enticed by an ad that
promised a “once-in-a-lifetime trip to a far away and exotic place”, A. Square took
a vacation to the plane with the point at the origin missing. When he asked at the
hotel desk if they could tell him the shortest path to the beach, the person replied,
“No. There is no shortest path to the beach”. At first he thought the person was
making a joke. But as he found out in Figure 1, every path from the hotel to the
beach has to go around the origin, and there is no single path which is shorter than
all others. Thus the plane with the origin removed is an example of a space where
Axiom 1 fails.
Figure 1. There is no geodesic that A. Square can take to get to

the beach.
Axiom 2. Any geodesic can be extended to a geodesic line.

Note that in this chapter we use the term geodesic line for a geodesic which
doesn’t terminate in either direction. On a plane the geodesic lines are straight
lines, whereas on a sphere the geodesic lines are great circles. We’ll use the term
geodesic segment for a geodesic that terminates in both directions, and geodesic ray
for a geodesic which terminates in one direction but not in the other. On a plane
the geodesic segments are line segments and the geodesic rays are rays that go off
arbitrarily far in one direction but not the other. On a sphere the geodesic segments
1. EUCLID’S AXIOMS 187
are arcs of great circles, and there are no geodesic rays. The term geodesic will be
used for any geodesic line, geodesic segment, or geodesic ray.
Certainly in a plane, any geodesic segment or geodesic ray can be extended to
get a geodesic line. On the other hand, as A. Square discovers on his next vacation,
not all 2-dimensional spaces satisfy Axiom 2.
A. Square had not intended to take a vacation over Thanksgiving because
the holiday is too short and it’s considered the heaviest travel period in the year.
However, he saw an ad that caught his eye:
Tired of pointless vacations? Want a vacation that includes all
of the positives and none of the negatives? Love chocolate? If
you answered yes to all three of these questions, then Chocotopia
is your ideal vacation destination.
This ad seemed like it was targeted directly at him. He was willing to fight the
crowds to spend four full days at Chocotopia. Chocotopia was located on a half-
plane, which seemed fine. At least it didn’t have any points missing! He was so
excited that he booked a package deal which included all of the chocolate you could
eat.
Almost immediately after he arrived, he began helping himself to small samples
of chocolate bars, chocolate mousse, brownies, chocolate truffles, chocolate sundaes,
and chocolate chip cookies. But he saved his favorite—chocolate cupcakes—for last.
Unfortunately, the chocolate cupcake that he ate on top of all the samples caused
a bit of indigestion. Based on past experience, he decided that taking a long walk
would be the perfect remedy. Unfortunately, as we see in Figure 2, after walking
only a short way, he was unable to continue walking in the same direction. Even
veering off slightly, was no better. Resigned, he went back to the hotel to take a
nap and rest up for his next chocolate opportunity.
Figure 2. A. Square can’t continue walking in this direction.
Axiom 3. Given any point a and any positive number r, there is a circle of
radius r with center a.
In order to understand this axiom, we need to be clear about what we mean
by a circle. Euclid defined a circle as follows.
Definition. A circle with center a and radius r is defined to be a single curve
which bounds a region containing the point a such that the distance from a to every
point on the curve is equal to r.
We will see as follows that an infinite cylinder violates Axiom 3. In particular,
we show that it does not have circles of every possible radius.
Suppose that c is the circumference of the infinite cylinder, let a be a point

in the cylinder, and choose r > 2c . On the left side of Figure 3, we illustrate
a gluing diagram for the infinite cylinder in which we have attempted to draw a
circle centered at a with radius r. However, since the circumference of the infinite
cylinder is the width of the gluing diagram and r > 2c , the diameter of the circle
we are trying to draw is larger than the width of the gluing diagram. Thus our
“circle” will cross over the gluing line and intersect itself as we see on the left side
of the figure.
Figure 3. The set of points whose distance from a is r consists

of two curves which together bound the grey region.
Now observe in Figure 3 that the shortest paths from a to x and from a to y
are along the dotted segments, though the segment going from a to x crosses over
the gluing line so it looks like it’s in two pieces. The lengths of these segments are
shorter than r, and hence the distance from a to the points x and y is less than r.
Similarly, we see that of all the points on the vertical eye shape except for the top
and bottom points are less than r away from a.
Hence the points which are precisely r units from a are those lying on the
two grey curves which together bound the grey region illustrated on the right in
Figure 3. Since neither of these curves by itself bounds a region, neither curve
satisfies Euclid’s definition of a circle. Thus the infinite cylinder does not have a
circle with radius r and center a, and hence violates Axiom 3.
Axiom 4. All right angles are equal.

In order to understand this axiom, we need to be clear about what we mean by
a right angle and what we mean by two angles being equal. We define the second
one first. In particular, we say that two angles in a surface are equal if there is a
rigid motion of the surface taking one to the other. This means that even if the
two angles are made of steel rods which are completely rigid, you could still move
one to the other.
It may seem like it’s not necessary to define a right angle, since we all know
that right angles are those that have 90◦ . This definition of a right angle works
just fine for the plane, but we want to consider right angles on other surfaces where
it might not be so easy to measure 90◦ . So instead we define a right angle as the
angle formed when two geodesics intersect in a T-shape so that the two resulting
1. EUCLID’S AXIOMS 189
angles are equal. This definition makes sense in any surface since we know what it
means for two angles to be equal. In Figure 4, we see two T-shapes. In each case,
the angles on either side of the T are equal, and hence they are right angles.
Figure 4. Two right angles in the plane.
Thus Axiom 4 is saying that we can move any right angle to any other right
angle by a rigid motion. You may recall from Chapter 3 that two objects in a space
have the same extrinsic geometry if and only if there is a rigid motion of the space
taking one object to the other. If we think of a right angle is an object in our space,
then we can restate Axiom 4 as saying that any pair of right angles have the same
extrinsic geometry.
After his stressful vacations on the plane with a point missing and the half-
plane, A. Square decides that he doesn’t want to waste his next vacation going to
some other part of a plane. The following year he sees an ad that says, “Build your
own windsurf board, and learn how to sail on the beautiful blue seas of a curved
torus”. The idea of building a windsurf board and sailing it in the ocean sounds
great, and a curved torus is a complete surface without missing parts or boundaries,
so what could possibly go wrong?
A. Square’s first day on the torus is great. Listening to the surf crashing in the
background, he eagerly builds his board and attaches the mast. He finds it a bit
odd that the materials for the mast and board seem quite pliable. When he asks
about this, the instructor responds by saying “the materials need some flexibility so
they can communicate with different regions of the ocean”. This doesn’t make any
sense to A. Square, since a windsurf board can’t talk like a square or a triangle can.
But his instructor has a thick toroidal accent, and A. Square is too embarrassed
to ask him to repeat. The instructor does manage to convey to A. Square that he
should keep the mast at a right angle to the board at all times, which A. Square
promises he will do.
After his instructor checks that his windsurf board is sturdy and ocean worthy,
A. Square rushes to the beach to give it a whirl. For a while, he is whipping through
the waves having a great time, but just as the sea seems to get calmer, he finds that
he is having a harder and harder time staying on the board. Suddenly he realizes
that the problem is that the board has changed from being curved towards his body
to being curved away (see Figure 5). No wonder he keeps sliding away from the
mast. He can’t figure out what he did wrong, since the mast and board are still at
right angles.
Scared that he might fall off and get eaten by sharks, A. Square quickly sails
back to shore. He is about to call the instructor over to show him what happened
to the board, when he notices that in fact the board is exactly the way it was when
he built it. He had so much fun sailing, that he shrugs it off as an illusion created
Figure 5. A. Square’s windsurf board has changed from being

curved towards his body to being curved away.
by too much sun. A. Square didn’t realize it at the time, but the “different regions
of the ocean” that his instructor referred to were what caused his boat to bend,
and what causes the curved torus to violate Axiom 4.
Axiom 5. Given any point a and geodesic line L not containing a, there is
exactly one geodesic line containing a and that does not intersect L.
Figure 6. In a plane, there is always a line through a parallel to L.
Axiom 5 is illustrated for the plane in Figure 6. The above statement of Axiom 5
is not in the original form stated by Euclid. Rather it is a simpler version of the
axiom given by John Playfair in the fifth century AD.
To see that Axiom 5 doesn’t hold for a sphere, we need look no further than
A. Square’s next vacation. He enjoyed windsurfing so much on the torus that
this year he’s eager to take another windsurfing vacation. After hearing that the
windsurfing on a sphere is out of this world, he buys a ticket and packs a bag.
When A. Square rents his windsurf board on the sphere, he has to sign a
statement that he won’t cross into the roped off swimming area. Anyone caught
sailing over the rope has to pay a hefty fine. He paddles with his board to a few
feet outside of the swimming area and plans to sail parallel to the rope. Once he
begins sailing, he notices that there’s a good tailwind which is pushing the board
straight along a geodesic. Mistakenly, he figures that he doesn’t have to worry
about crossing into the swimming area because the tailwind will keep him going
along the black circle in Figure 7. He’s having a great time riding the waves, happy
that he’s having no trouble staying on his board.
2. FLAT SURFACES AND EUCLIDEAN GEOMETRY 191
Figure 7. A. Square assumes the tailwind wind is pushing him

along the black circle.
Figure 8. The tailwind actually pushes him along a geodesic

which crosses into the swimming area.
But he suddenly notices that the geodesic he’s following is taking him straight
into the rope rather than parallel to it, as illustrated in Figure 8. In a panic,
he quickly capsizes his windsurf board just as it hits the rope. Fortunately, the
lifeguard was so busy flirting with a cute pentagon that she didn’t notice A. Square’s
close call.
2. Flat Surfaces and Euclidean Geometry
If a surface satisfies Axioms 1–5, then we say that it has Euclidean geometry
or simply that it’s Euclidean. For example, as you probably learned when you
took geometry, the plane satisfies Axioms 1–5 and hence is Euclidean. This is no
surprise since Euclid’s axioms were intended to describe the geometry of the plane.
On the other hand, a space violating one or more of the Axioms is not Euclidean.
So none of A. Square’s unfortunate vacation destinations is Euclidean, which was
why they each had such strange geometric properties. Since some of these non-
Euclidean surfaces are flat and others aren’t, we might think that being flat and
being Euclidean are completely independent characteristics of surfaces. However,

we see as follows that this is not the case.
In Theorem 5.1 from Chapter 5, we saw that a surface is flat if and only if
the sum of the angles of any triangle in the surface is equal to 180◦ . Now we have
the following theorem which relates the sum of the angles in a triangle to Euclid’s
Axioms.
Theorem 8.1. Let S be a surface which satisfies Axioms 1–4. Then S satisfies
Axiom 5 if and only if the sum of the angles of any triangle in S is equal to 180◦ .
This result together with Theorem 5.1 shows that every Euclidean surface is
flat and every flat surface which satisfies Axioms 1–4 is Euclidean. Keep in mind
that there are examples like the plane with a point removed, the half-plane, and
an infinite cylinder which are flat but violate at least one of Axioms 1–4. In the
Exercises, you will explore some other examples of flat surfaces which violate one
or more of the axioms.
In addition to the property of Euclidean surfaces that the sum of the angles of
a triangle is 180◦ , you may recognize some other properties of Euclidean surfaces
which have names like the Pythagorean Theorem or the Side-Angle-Side Theorem,
and you may even have proved them in a geometry class. We won’t present the
proofs of these results or of Theorem 8.1, but you can read about them and other
properties of Euclidean surfaces in a textbook on Euclidean geometry.
3. Some Alternative Axioms
In many languages the word for the article “a” and the word for the number
“one” are the same. If we replace the word “a” by the word “one” in Axiom 1,
then Axiom 1 would become the following.
Alternative Axiom 1: For any two distinct points, there is one geodesic that
passes through both points.
Both Axiom 1 and Alternative Axiom 1 are true for the plane. But not all
spaces which satisfy Axiom 1 satisfy this stronger alternative version of the axiom,
as A. Square discovered when he and his twin brother B. Square went to the flat
torus for their winter vacation.
A. Square and B. Square are mirror images of one another, which has left some
people wondering if they were identical at birth and then somehow one took an
orientation reversing path without the other. The Square brothers spent most of
their vacation on the flat torus building elaborate snow forts and having snowball
fights. In Figure 9, A. Square realized that if he positioned himself exactly halfway
around the flat torus from his brother, he could throw two snowballs at once to hit
the front and back of B. Square’s head at exactly the same time. He thought this
was a great discovery, but B. Square didn’t.
In fact many books on Euclidean geometry use Alternative Axiom 1 instead
of Axiom 1. Euclid himself assumes in some of his proofs that there is a unique
line through any pair of points. Thus according to many historians it is likely that
Euclid intended to include uniqueness in the statement of Axiom 1.
Axiom 5 is the axiom which has gotten the most attention from mathematicians
and historians because it is regarded as the least self-evident of the axioms. For
4. SPHERICAL TRIGONOMETRY 193
Figure 9. A. Square throws snowballs to hit B. Square from the

back and front simultaneously.
hundreds of years, mathematicians tried unsuccessfully to derive Axiom 5 from the

other four axioms. We now know that this is impossible.
In fact, there are two ways that Axiom 5 can be violated. To understand these
two ways, let’s see what happens when we negate Axiom 5. In particular, consider
the statement:
It is not the case that, given any point a and geodesic line L not
containing a, there is exactly one geodesic line containing a and
that does not intersect L.
We could say this more simply as follows.
There is a point a and geodesic line L not containing a such that
it is not the case that there is exactly one geodesic line containing
a and that does not intersect L.
This statement implies that either there is no geodesic line containing a that
is disjoint from L or there is more than one such line. In the first case, every
line through a intersects L. In the second case, there are multiple lines through a
that do not intersect L. We refer to these two possibilities as “First Alternative
Axiom 5” and “Second Alternative Axiom 5”. For the rest of this chapter, we focus
on spaces in which one of these two alternatives to Axiom 5 is satisfied.
4. Spherical Trigonometry
As A. Square saw in Figure 8, a sphere does not satisfy Euclid’s fifth axiom. In
fact, every pair of great circles on a sphere intersects in two points. In Figure 10,
we illustrate a great circle L and a point a disjoint from L. As we can see in the
figure, any great circle containing a will intersect L. Thus a sphere satisfies the
following alternative axiom.
First Alternative to Axiom 5. Given any point a and any geodesic line L not
containing a, every geodesic line through a intersects L.
Because a sphere satisfies this alternative to Axiom 5, the geometry of a sphere
is quite different from that of a plane. For example, recall from Chapter 3 that
the sum of the angles of any triangle in a sphere is more than 180◦ ; and the bigger
Figure 10. There is no great circle containing a which doesn’t

intersect L.
the triangle is, the greater the angle sum. Consider a spherical triangle whose
angles are θ1 , θ2 , and θ3 , and whose complementary angles are ϕ1 , ϕ2 , and ϕ3 , as
illustrated in Figure 11. Recall from Chapter 5 that θ is a Greek letter pronounced
“theta”, which rhymes with the cheese feta. The Greek letter ϕ is “phi”, which is
pronounced “fee” and sometimes “fie”, but never “fo” or even “fum”.
Figure 11. Two complementary triangles in a sphere.
Notice that the three geodesic segments in Figure 11 actually bound two tri-
angles: one on the inside with angles θ1 , θ2 , and θ3 , and one on the outside with
angles ϕ1 , ϕ2 , and ϕ3 . We say that this pair of triangles are “complementary”.
But be careful never to confuse the complementary triangles in Figure 11 with the
complimentary triangles in Figure 12.
Figure 12. Two complimentary triangles.
What happens to the angle sum ϕ1 + ϕ2 + ϕ3 of the outer triangle if the angles
θ1 , θ2 , and θ3 of the inner triangle get smaller? Since there are 360◦ around every
4. SPHERICAL TRIGONOMETRY 195
point, we see that the angles in the complementary triangle are ϕ1 = 360◦ − θ1 ,
ϕ2 = 360◦ − θ2 , and ϕ3 = 360◦ − θ3 . We know that θ1 + θ2 + θ3 will always be bigger
than 180◦ , but if we make the triangle smaller and smaller, this sum will get closer
and closer to 180◦ . If we do this, the sum ϕ1 + ϕ2 + ϕ3 = 3 × 360◦ − (θ1 + θ2 + θ2 ) =
1080 − (θ1 + θ2 + θ2 ) gets very close to 900◦ . So the angle sum of a spherical triangle
is always between 180◦ and 900◦ .
A. Square loves having snowball fights with his brother B. Square. Even though
they don’t realize it, they are unconsciously using trigonometry all the time. When
A. Square aims a snowball at B. Square, A. Square’s brain must first determine
the distance between them. A. Square looks at B. Square, and his eye registers the
angle θ formed by the light rays coming from the top and bottom of B. Square. In
Figure 13, this angle θ is 45◦ .
Figure 13. A. Square sees B. Square in a flat universe.
The light rays from the top and bottom of B. Square together with the edge of
B. Square’s face form a triangle. A. Square’s brain uses the angle θ together with
trigonometry to estimate how far to throw a snowball. Your brain does this all the
time without you being aware of it. This happens even if you hated trigonometry,
swore you’d never use it, and couldn’t tell a sohcahtoa from a cohsahtoa if you met
one walking down the street.
If A. and B. Square are in a flat surface as they do in Figure 13, then the sum
of the angles of any triangle is 180◦ . Hence the angle between the bottom light ray
and B. Square’s face must be 45◦ , since the triangle is a right triangle and θ = 45◦ .
Using Euclidean geometry, we can now deduce that the triangle in Figure 13 is
isosceles. This means that the length of the top light ray is exactly equal to the
height of B. Square. Since A. Square is the mirror image of B. Square, A. Square
knows B. Square’s height like the back of his hand (or better yet, like the front of
his face), and hence also the distance between them. A. Square’s brain cranks out
computations nearly instantaneously and A. Square launches his snowball before
B. Square even notices that A. Square has a snowball in his hand. Our brains make
similar computations when we throw a ball, a snowball, or even a frisbee.
However, the same calculations wouldn’t apply if A. Square and B. Square were
on a sphere; which is exactly where they go for their next vacation. After listening
to A. Square talk endlessly about his great windsurfing vacation on the sphere
(without mentioning his close call with the swimming area), B. Square wants to
see the sphere for himself. Though B. Square pretends to be disappointed that the
only time they can go to the sphere is over winter break, he’s actually eager to go
somewhere snowy so that he can get back at A. Square with his own well-placed
snow balls.
As soon as they check in at their hotel on the sphere, they run outside to have
a snowball fight. We see in Figure 14 that the light rays from the top and bottom
of B. Square together with B. Square’s face form a triangle as they did before.
Also, again the angle between the top light ray and B. Square’s face is a right angle
and θ = 45◦ . But, in this case, the sides of the triangle are arcs of great circles
rather than line segments, and the triangle has two right angles in addition to the
45◦ angle. Furthermore, the geodesic segment between A. Square and the top of
B. Square’s head is almost half of a great circle, whereas B. Square’s height is only
a quarter of a great circle. This means that the distance between A. Square and
B. Square is almost twice the height of B. Square.
Figure 14. A. Square sees B. Square in a sphere.
In A. Square’s eagerness to throw the first snowball, his brain uses Euclidean
trigonometry to aim. As a result, A. Square thinks that the distance between
them is the same as B. Square’s height, and his snowball lands only halfway to its
intended target. B. Square teases him about this for the rest of their vacation on
the sphere.
5. The Area of a Disk in a Sphere
After their snowball fight, the Square brothers see an ice-skating rink in the
distance. Thinking that maybe ice-skating will be less frustrating than a snowball
fight, A. Square rushes up to the entrance of the rink. They don’t see where to
go to rent ice-skates, and after some arguing about which way it should be, they
decide to go left. The rental counter was actually to the right of the entrance, and
hence they have to go all the way around what seems to be a fairly small ice-skating
rink to get there. With skates and tickets in hand, they are about to enter the rink
and start skating when A. Square sees a sign that says “8 skaters in the rink”.
Always an ace at math, A. Square is eager to figure out the area of the skating
rink in order to know if it will be crowded or not. Since A. Square’s edges are
straight and he is a bit self-centered, he declares that his height and his width are
each 1 unit. B. Square is perfectly happy to use A. Square’s edges as the unit
of measurement because he has exactly the same dimensions as his brother. This
means that the area of each of the brothers is 1 square unit, which sounds perfect
to the Square brothers.
5. THE AREA OF A DISK IN A SPHERE 197
Remembering the geometry he learned in high school, A. Square tells his

brother that since the rink is circular with radius r, its circumference is given
by C = 2πr and its area is given by A = πr 2 . He then solves the circumference
equation for r to get
C
r= .
2π
From their walk around the rink, A. Square guesses that the circumference of the
rink is 10 units. He then plugs C = 10 into the equation for the radius, and
computes the radius of the rink to be
10 5
r= = .
2π π
Next, he plugs this value for r into the equation A = πr 2 , and finds that the area
of the rink is
2
5 25
A = πr 2 = π = .
π π
Using his calculator he finds that the area of the skating rink is roughly equal to 8
square units (you should check A. Square’s computation with your own calculator).
If there are 8 skaters in the rink, each of which has area equal to 1 square unit like
the Square brothers, then there would be no room to even enter the rink. If the 8
skaters are somewhat thinner than the Square brothers, then they might be able to
enter the rink, but skating with them would still be like sardines skating together
in a can. Proud of his geometric prowess, A. Square declares that they should go
do something else a bit roomier, like watch the snow fall.
Just as they are about to give up on skating, B. Square thumps his forehead
and then thumps A. Square’s forehead for good measure. A. Square has made
a mistake in his calculations! (Can you see where?) A. Square mistakenly used
Euclidean rather than spherical geometry to find the area of the rink. A circle of
circumference 2πr encloses an area of πr 2 in a plane, but the same circumference
encloses more area in a sphere. For example, suppose that the skating rink takes
up exactly one hemisphere. Looking at Figure 15, we see that the area of the top
hemisphere is larger than the area of the planar disk with the same circumference.
Figure 15. The area of a hemisphere is larger than the area of a

flat disk with the same circumference.
We can find the exact area of the hemisphere as follows. Suppose that r is the
radius of the sphere. Then the length of the equator is C = 2πr. Hence the radius
of the rink is again
C
r= .
2π
Since the surface area of a sphere is

4πr 2 ,
the area of one hemisphere must be
A = 2πr 2 .
Using A. Square’s observation that the circumference of the rink is roughly 10 units,
we plug in C = 10 to again get that the value of the radius is
10 5
r= = .
2π π
But now we plug this value into the formula for the area of a hemisphere to find
that the area of the rink is
2
2 5 50
A = 2πr = 2π = ,
π π
which we find with a calculator is approximately 16 square units.
So the ice-skating rink could be as much as twice as spacious as A. Square
had previously thought. Having twice your body size to move in a skating rink is
better than skating in a can of sardines, but it’s still not great. Since the other
skaters all appear to be quite skinny, A. Square agrees to enter the skating rink (see
Figure 16).
Figure 16. Because the skating rink is on a sphere, it is more

spacious than A. Square had thought.
In his calculations for both the snowball fight and the ice skating rink, A. Square
mistakenly assumed that the geometry of a sphere would be the same as that of a
plane, forgetting that triangles and circles in a sphere have more area because they
bulge outwards. In fact, the same is true for any geometric shape in a sphere. As a
result, angle sums of polygons will be bigger in a sphere than in a plane. Also, just
as for spherical triangles, the bigger the spherical n-gon is compared to the size of
the sphere, the bigger its angle sum will be. For instance, if all of the vertices of an
n-gon lie on a great circle, then the n-gon will have angle sum n × 180◦ . Figure 17
6. MAPS OF THE EARTH 199
Figure 17. This hemisphere has six vertices and hence is a hexagon.
shows a hexagon whose angle sum is 6 × 180◦ = 1080◦ . Furthermore, such an n-gon
will be a hemisphere, and hence will have area 2πr 2 (where r is the radius of the
sphere).
6. Maps of the Earth
After returning to their flat universe, A. Square and B. Square decide to make
a scrapbook with drawings of the skating rink and the other things they saw during
their visit to the sphere. But they keep arguing about how to draw the skating
rink. The problem is that since spherical and flat geometry are so different, there
is actually no way to draw spherical distances and angles accurately on a piece of
paper.
Mapmakers encounter the same problem as the Square brothers when they
make flat maps of the Earth like the one shown in Figure 18. To explain this map
to a Flatlander, we would need to first explain that if we cross the left edge of the
map, we reappear on the right edge. This is not hard for Flatlanders to understand
because they are used to drawings of the flat torus. But we also tell them that the
entire top edge is shrunk to a single point which is called the North Pole, and the
entire bottom edge is shrunk to another point which we call the South Pole.
Each vertical line on the flat map (known as a longitude) corresponds to half
a great circle through the North and South Poles. The horizontal lines (known as
latitudes) correspond to parallel circles on the Earth whose circumferences decrease
as you move away from the equator towards either of the poles. Apart from the
line corresponding to the equator, none of the latitudes are great circles, and hence
do not represent geodesics.
Looking at Figure 18, let’s compare the distance between points A and B to
the distance between points C and D. Although these two distances look like they
are equal in length on the map, in reality the distance between C and D is shorter
than the distance between A and B. This is because the shortest path between
points A and B is an arc of a great circle on the sphere which follows the curved
grey arc illustrated in Figure 18. On the other hand, since points C and D are on
the equator, the line segment between them on our map corresponds exactly to the
arc of a great circle on the sphere. If you have ever looked at flight paths on airline
maps, you have probably noticed that they are represented by curved arcs rather
than straight lines. This is because airlines take the shortest route, which means
flight paths follow great circles on the Earth.
Figure 18. A planar map of the Earth cannot be accurate.
The map we have drawn in Figure 18 is known as a Mercator map. The ad-
vantage of this map is that it preserves angles and the shapes of small regions. For
instance, longitudes and latitudes form right angles on the Mercator map, as they
do on a globe. Also, if you compare a small patch of the Mercator map with the
corresponding small region on a globe, then you’ll notice that shapes within that
small region are nearly the same. But this is not true for large regions. Can you
find a large area in Figure 18 whose shape is different from what it is on a globe?
The Mercator map has been used as the primary map for steering ships since the
1500s because it preserves angles and is relatively accurate on a small scale. Even
today, Google Maps uses a variant of the Mercator map to make its maps.
Maps other than the Mercator map might have less distortion on a large scale,
but some distort angles while others divide up the Earth in strange ways. There is
always a trade-off. A flat map of the Earth will never be completely free of distor-
tion, because as we have seen spherical geometry is quite different from Euclidean
geometry.
7. Hyperbolic Geometry
In this section, we consider an axiom system consisting of Euclid’s first four

axioms but the fifth axiom is now replaced by:
Second Alternative to Axiom 5. For any given line L and any point a not on
L, there are multiple lines through a that do not intersect L.
7. HYPERBOLIC GEOMETRY 201
The hyperbolic plane illustrated in Figure 19 satisfies such an axiom system.

Note that for the rest of this chapter we will use the term “flat plane” to refer to
the usual plane in order to distinguish it from the hyperbolic plane. Topologically
the hyperbolic plane is the same as the flat plane, but geometrically it is like a
saddle at every point. The best word to describe the hyperbolic plane is “floppy”.
In Figure 19, we illustrate two geodesics through the point a neither of which
intersects the geodesic L. In fact, there are infinitely many geodesics through a
that don’t intersect L.
Figure 19. There are infinitely many geodesics through the point
a which do not intersect the geodesic line L.
Furthermore, all triangles on a hyperbolic plane have angle sums which are less
that 180◦ , and the bigger the triangle is, the smaller its angle sum will be. This
makes hyperbolic triangles bulge inwards and look thin like the white triangle on
the right side of Figure 19.
Recall that when A. and B. Square were having a snowball fight in the sphere,
A. Square’s Euclidean brain thought that B. Square was closer than he really was.
In Figure 20, we see that in the hyperbolic plane the light rays going from the top
and bottom of B. Square to A. Square’s eye together with B. Square’s face make
a triangle. However, the triangle is thinner than it would be in a flat plane. As a
result, A. Square’s Euclidean brain would deduce that B. Square is further away
than he really was, and he would throw a snowball flying past B. Square’s head.
If you want to make your own hyperbolic plane so that you can get a feel for
how floppy it is, there are many fun ways to do so. You can learn to crochet a
hyperbolic plane and to make a hyperbolic blanket by following the directions on
these websites:
• http://www.math.cornell.edu/∼dwh/papers/crochet/crochet.html
• http://geometrygames.org/HyperbolicBlanket/index.html
Just as a spherical circle encloses more area than a circle of the same circum-
ference in the flat plane, a hyperbolic circle encloses less area than a circle of the
Figure 20. A. Square thinks B. Square is farther away than he

really is.
same circumference in the flat plane. The circular ice-skating rink in Figure 21 has
a circumference of roughly 10 times the height of A. Square. Even though there
are only 8 other skaters in the rink, all of whom are much skinnier than A. Square
and B. Square, it is too crowded for them to skate. In this case, A. Square and
B. Square really would be better off watching snow fall than trying to ice skate in
this crowded rink.
Figure 21. This hyperbolic skating rink encloses less area than a
rink with the same radius in the flat plane.
8. A. Square Learns to Draw a Hyperbolic Plane
Although we can visualize the hyperbolic plane as a floppy plane in space, it is

much more difficult for A. Square to visualize a floppy plane. If you try to visualize
a “floppy” 3-dimensional space, you’ll see the problem.
8. A. SQUARE LEARNS TO DRAW A HYPERBOLIC PLANE 203
To help A. Square visualize the hyperbolic plane, let’s first recall how we ex-
plained the geometric product S 1 × I to him in Chapter 7. We showed him the
picture of an annulus in Figure 22, and told him to imagine that all of the concentric
circles are actually geodesics of the same size. In addition, we told him that the
shortest path between two points on one of these circles goes along the circle, as
illustrated by the grey arc in the figure. This seemed counterintuitive to him, since
he expected the shortest path to be along a straight segment. However, he was
willing to ignore his intuition in order to visualize S 1 × I as a geometric product.
We used a similar technique ourselves to visualize the 3-dimensional space S 2 × I
as a geometric product.
Figure 22. In the geometric product S 1 × I, the shortest path

between these points is along the grey arc.
Since there’s not enough space in a flat plane to draw a floppy plane, we’ll
have to use a similar type of distorted drawing to enable A. Square to visualize a
hyperbolic plane. We begin by drawing a circle in the plane and telling A. Square
that we’re going to draw the hyperbolic plane inside of this circle. Since a hyperbolic
plane has no boundary, the hyperbolic plane won’t include the circle bounding the
region. At first A. Square objects to the idea that a region bounded by a circle
could have infinite area. But then he remembers that in Chapter 6 when we defined
the connected sum of two planes, the circle where the planes were glued together
bounded an infinite area on both sides (see Figure 23).
Figure 23. A gluing diagram for the connected sum of two planes.
Next, we tell him that for the hyperbolic plane there is not only an infinite
amount of area inside of the circle, but some of the geodesics there are pretty
strange as well. He accepts the idea of strange geodesics, since he accepted the
strange geodesics for the annulus in Figure 22. We tell him that now there are two
different types of geodesics: diameters of the disk and pieces of circles meeting the
boundary of the disk at right angles. Since we’re not including the boundary of the
region, these geodesics have no endpoints. Figure 24 shows some geodesics in the
hyperbolic plane.
Figure 24 includes two geodesics through a point a together with a disjoint
geodesic L. Once he has accepted the strange geodesics, A. Square can see that the
Second Alternative to Axiom 5 is satisfied by the hyperbolic plane.
Figure 24. Geodesics are diameters of the disk and circular arcs
meeting the boundary circle at right angles.
Figure 25. The grey segments between points a and b and points
c and d are longer than the circular arcs between them.
In order to help A. Square remember that straight line segments which are not
part of a diameter are not geodesics, we show him Figure 25, and tell him that the
grey line segments between points a and b and points c and d are actually longer
than the black circular arcs between these pairs of points.
This is reminiscent of the geometric S 1 × I in Figure 22, where the shortest
path was along an arc of a circle rather than along a straight segment. However,
because we’ve had to squish the infinite hyperbolic plane into the space of a disk,
the closer to the boundary circle we get, the more distances are distorted. For
example, we can see in Figure 25 that the hyperbolic distance between a and b is
actually greater than the hyperbolic distance between c and d, even though their
Euclidean distances are identical.
Observe that the shaded region in the lower right of Figure 25 is bounded by
three circular arcs which all meet the boundary circle in right angles. Such arcs
are geodesics, and hence the shaded region is a hyperbolic triangle. Because two of
the sides of the triangle curve inward, the angles are much smaller than they would
be in a Euclidean triangle. In fact, if all three of the vertices of a triangle are very
close to the boundary circle, the angle sum of the triangle will be very close to 0◦ .
The same is true for n-gons in the hyperbolic plane. In particular, an n-gon
can have an angle sum as close to 0◦ as we like. Figure 26 shows an octagon in
the hyperbolic plane whose angle sum is 8 × 45◦ = 360◦ . This will be important
to us in the next section when we try to find homogeneous geometries for all finite
surfaces.
9. HOMOGENEOUS GEOMETRIES FOR ALL nT 2 WITH n ≥ 2 205
Figure 26. A hyperbolic octagon whose angles are 45◦ .
9. Homogeneous Geometries for all nT 2 with n ≥ 2
It turns out that if a finite surface has homogeneous geometry, then its local
intrinsic geometry must be the same as that of either a flat plane, a sphere, or a
hyperbolic plane. We would like to figure out how to construct each finite surface
so that it has one of these three homogeneous geometries.
Recall from the Classification of Surfaces, version 2 (in Chapter 6), that every
finite surface is topologically the same as either S 2 , nT 2 , or nP 2 for some n > 0.
Since we know that K 2 = 2P 2 and T 2 can both be constructed to have flat geometry
and S 2 has spherical geometry, we are left with the task of finding homogeneous
geometries for the surfaces nT 2 for all n > 1 and nP 2 for all n = 2.
Recall that we saw the following theorem in Chapter 5.
Theorem 8.2. Let S be a surface with a polygonal gluing diagram. Then S is
flat if and only if the edges of the polygon are glued together in pairs and the sum
of the angles around each glued up vertex is 360◦ .
This theorem is useful for showing that a surface obtained from a polygonal
gluing diagram is flat. But if the sum of the angles around a glued up vertex is not
360◦ , the theorem only tells us that the glued up surface isn’t flat. It doesn’t tell
us if there is another way to construct the surface to make it homogeneous.
For example, we saw in Chapter 6 that 2T 2 is obtained from the octagonal
gluing diagram illustrated in Figure 27. However, each vertex of an octagon has
135◦ around it, and all of the vertices in the diagram are glued together (you should
remind yourself of how we know this). This means that the glued up octagon has
a single anti-cone point with 135◦ × 8 = 1080◦ around it. Thus we know by
Theorem 8.2 that this gluing diagram is not flat. However, there could be another
gluing diagram for 2T 2 which is homogeneous, though not necessarily flat.
In order to allow us to create homogeneous gluing diagrams which are not flat,
we use the following generalization of Theorem 8.2 to polygonal gluing diagrams
in other homogeneous surfaces. The proof of this theorem requires techniques that
you might learn in a differential geometry class.
Theorem 8.3. Let S be a surface with a polygonal gluing diagram in a homo-
geneous surface. Then S has the same geometry as the homogeneous surface if and
only if its edges are glued together in pairs and the sum of the angles around every
glued up vertex is 360◦ .
Figure 27. All of the vertices in this gluing diagram for 2T 2 are
glued together.
We’ll see how to use Theorem 8.3 by applying it to an octagonal gluing diagram
for T 2 with the same gluing arrows as the one illustrated in Figure 27 but in a
hyperbolic plane rather than a flat plane. Recall from the last section that polygons
in the hyperbolic plane have smaller angles than corresponding polygons in a flat
plane, and the closer the vertices are to the boundary of the disk, the smaller the
angles will be. Thus if we draw the octagonal gluing diagram for 2T 2 on a hyperbolic
plane rather than on a flat plane, its angles will be smaller. Moreover, as we saw
in Figure 26, we can choose the vertices of the octagon on the hyperbolic plane in
such a way that each of the angles is equal to precisely 45◦ . While A. Square has
to accept the drawing of the octagonal gluing diagram on the left side of Figure 28,
we can also draw the gluing diagram on the floppy hyperbolic plane illustrated on
the right.
Now since all eight vertices of the hyperbolic octagon are glued together and
all eight angles are 45◦ , the glued up vertex will have a total of 45◦ × 8 = 360◦
around it. Thus we can apply Theorem 8.3 to conclude that the 2T 2 obtained from
this gluing diagram has homogeneous hyperbolic geometry.
Recall from Chapter 6 that for any n ≥ 2, the n-holed torus nT 2 can be
obtained from a gluing diagram which is a 4n-gon. However just as we saw with
Figure 28. In the hyperbolic plane we can find an octagonal glu-

ing diagram for 2T 2 whose angles are 45◦ .
10. A HOMOGENEOUS GEOMETRY FOR P 2 207
the octagon, all of the vertices in the 4n-gon are glued together. Thus we can
create an nT 2 which has homogeneous hyperbolic geometry by drawing the 4n-gon
on a hyperbolic plane and by moving the vertices apart until each vertex angle is
◦
◦
exactly 360
4n . In this way, the glued up vertex will have exactly 360 around it, and
by Theorem 8.3 the nT 2 obtained from this gluing diagram will have homogeneous
hyperbolic geometry.
10. A Homogeneous Geometry for P 2
Next we would like to construct a projective plane P 2 which has homogeneous

geometry. The usual gluing diagram for P 2 is on the left in Figure 29. We can’t
apply Theorem 8.3 to this diagram, since the diagram is not a polygon. However,
by splitting each of the edges in half and then straightening the four edges out, we
obtain the square gluing diagram on the right. At this point, we still can’t apply
Theorem 8.3, since the vertices of the gluing diagram are glued together in pairs
and hence there is only 90◦ × 2 = 180◦ around each glued up vertex.
Figure 29. This give us a polygonal gluing diagram for the pro-
jective plane.
Thus we need to increase the angle around each vertex from 90◦ to 180◦ so that
when we glue the vertices together in pairs, there will be 360◦ around each glued
up vertex. Since any polygon on a sphere has larger angles than the corresponding
polygon in a plane, in this case we want to draw our a gluing diagram on a sphere
Figure 30. A gluing diagram for P 2 on a sphere.
Now the bigger the polygon is on the sphere, the bigger its angles will be. Since
we want the angle around each vertex to be 180◦ , we want to move the vertices of
the square apart until they all lie on a single great circle. This stretches the square
out to cover an entire hemisphere as illustrated in Figure 31. The hemisphere may
not look like a “square”, but it has four vertices and four geodesic edges, and hence
it is in fact a “spherical square” (not to be confused with a “squarical sphere”).
Figure 31. A square gluing diagram for P 2 in which the angle

around each vertex is 180◦ .
The angle at each vertex is now 180◦ . So after gluing the vertices together in pairs,
there will be exactly 360◦ around each vertex, just as we wanted. Thus we can
apply Theorem 8.3 to conclude that the P 2 obtained from this gluing diagram has
spherical homogeneous geometry.
Recall that 2P 2 = K 2 has a square gluing diagram with flat geometry. In the
Exercises, you will construct an nP 2 with homogeneous geometry for each n > 2.
Thus in fact every finite surface can be constructed so that it has homogeneous
geometry.
11. Uniqueness of Homogeneous Geometries for Surfaces
We have seen that it is possible to construct every finite surface so that it has
one of the three homogeneous geometries. Also, we know that different construc-
tions of the same surface can have different geometry. For example, the flat torus
has flat geometry while the curved torus has non-homogeneous geometry. Perhaps
a surface could be constructed in two different ways, so that one has one homoge-
neous geometry and the other has a different homogeneous geometry. Maybe we
could even find a surface which can have each of the three homogeneous geometries
depending on how it was constructed.
As an example, let’s consider two different gluing diagrams for a torus. In
the Exercises in Chapter 5, we showed that the hexagonal gluing diagram on the
right in Figure 32 is a torus. Perhaps the two tori in Figure 32 have two different
homogeneous geometries. The square gluing diagram has all four vertices glued
together, and the hexagonal gluing diagram has two groups of three vertices glued
together. Since 4 × 90◦ = 360◦ and 3 × 120◦ = 360◦ , it follows from Theorem 8.2
that in fact both of these diagrams have flat geometry. Drawing either of these
gluing diagrams on a hyperbolic plane will cause the glued up vertices to have less
Figure 32. These are two different gluing diagrams for a torus.
12. EXERCISES 209
than 360◦ , and hence the surface will have one or two cone points. Drawing either
of these gluing diagrams on a sphere will cause the glued up vertices to have more
than 360◦ , and hence the surface will have one or two anti-cone points.
Perhaps with more work we could find a completely different gluing diagram for
a torus that could be drawn on a hyperbolic plane or a sphere with no cone points
or anti-cone points. In fact, the following theorem shows that this is impossible.
Theorem 8.4 (The Uniformization of Surfaces). Every finite surface can be
constructed with exactly one homogeneous geometry.
This theorem follows from a beautiful result known as the Gauss-Bonnet For-
mula, which you might learn about if you take differential geometry.
12. Exercises
1. Prove that a rectangular flat torus whose sides have different lengths does not
satisfy Euclid’s Axiom 3.
2. Prove that a square flat torus does not satisfy Axiom 3.
3. Prove that a flat Klein bottle does not satisfy Axiom 3.
4. Prove that a sphere does not satisfy Axiom 3.
5. Consider a sphere with radius r. Let a be a point on the sphere. Is there a
circle on the sphere with center at a and radius πr? Are there circles with
radius less than πr?
6. Prove that the sphere and the infinite cylinder don’t satisfy Alternative Axiom
1.
7. Prove that the surface in Figure 33 violates Axiom 5.
8. Prove that an infinite cone violates Axiom 5.

9. Prove that a flat torus satisfies Axioms 1, 2, 4, and 5.
10. Given any two points in a flat torus, show that there are infinitely many geodesic
segments between them.
11. Do all of the geodesic lines in a flat torus have finite length?
12. Explain why the flat plane with the origin removed does not satisfy Axiom 5.
13. Describe a surface together with a pair of points p and q, such that there are
four geodesic segments between p and q with the same length.
14. Show that the flat Klein bottle satisfies Axioms 1, 2, and 4.
15. Explain why the flat Klein bottle does not satisfy Axiom 5.
16. Show that a sphere satisfies Axioms 1, 2, and 4.
17. For each of Euclid’s five axioms, determine whether it is a statement about the
local geometry of a surface or the global geometry of a surface.
18. In what ways is the map of Earth in Figure 34 accurate or inaccurate?
19. Prove that the sum of the angles of a flat n-gon is (n − 2) × 180◦ .
20. Suppose A. Square and B. Square are in a small sphere but have had a bit of
brotherly disagreement. So A. Square is camped out at the North Pole and
B. Square is at the South Pole. If we rotate the sphere to compare the North
Pole and the South Pole, to us A. Square and B. Square look identical, except
that B. Square is the mirror image of A. Square. What does B. Square look
like to A. Square?
21. Suppose B. Square couldn’t stand being in the sphere with A. Square anymore
and has somehow transported himself home to the flat torus, leaving A. Square
in the sphere by himself. Would A. Square be able to see himself? And if so,
how would he look? In particular, in what ways would the view be the same
or different from the view of B. Square when he was at the opposite pole?
22. Suppose A. Square and his brother B. Square are again both in a small sphere.
A. Square watches as his brother walks away from him. Does the image of
B. Square get bigger or smaller as he walks away? At what point is the image
the smallest and at what point is it the largest?
23. We say that two triangles are congruent if they have the same intrinsic geometry.
This means that each of the sides and each of the angles in one have to equal
each of the corresponding sides and angles of the other. In Euclidean geometry,
there are four congruence theorems:
• Side-Side-Side (SSS) says that if two triangles have all three side lengths
equal, then the two triangles are congruent.
12. EXERCISES 211
• Angle-Angle-Side (AAS) says that two triangles are congruent if two pairs
of angles are equal and the lengths of a pair of corresponding sides not in
between the two angles are also equal.
• Side-Angle-Side (SAS) says that if two pairs of sides have equal length and
the corresponding angles between the two sides have equal length, then the
triangles are congruent.
• Angle-Side-Angle (ASA) says that two triangles are congruent if two pairs
of angles are equal and the corresponding sides between the two angles
have equal length.
Which, if any, of the SSS, AAS, SAS, and ASA theorems hold for triangles
in a sphere?
24. Explain how to create an nP 2 with hyperbolic geometry for any n > 2. Explain
why this method won’t work for n = 1 or n = 2.
Part 2
Knots
CHAPTER 9
Introduction to Knot Theory
Topics:
• Knots and links
• Projections
• Crossing number
• Prime and composite knots
• Knot tables
• A brief history of knot theory
• Reidemeister moves
• Tricolorability
• Invertibility
1. 1-Dimensional Universes
Our 1-dimensional friend A. Dash is feeling sad. Just yesterday, A. Square told
her about all the different 2-dimensional universes he had learned about: the plane,
the sphere, the torus, the Klein bottle, and infinitely many more. A. Dash began
pondering the different 1-dimensional universes, and could only think of two: an
infinite line and a circle. How boring!
“Don’t worry, I’m sure there must be others”, exclaims A. Square, and he
begins sketching the squiggly shapes in Figure 1. But after a few minutes, A. Square
realizes that all of the shapes he has drawn could actually be deformed to ordinary
lines or circles. Poor A. Dash really did have only two possible universes. A. Square
began to feel very lucky that he did not have to live in a boring 1-dimensional
universe.
Figure 1. Some 1-dimensional universes in the plane.

215
216 9. INTRODUCTION TO KNOT THEORY
While the intrinsic topology of 1-dimensional universes might be a bit boring,

the extrinsic topology of 1-dimensional spaces in R3 is anything but boring. We
tried to explain to A. Dash and A. Square that there are infinitely many different
knotted universes that she could live in, like the ones illustrated in Figure 2. She
was very excited to hear this, and wanted to visualize a knotted universe. But
unfortunately, neither A. Dash nor A. Square can understand what a knot is, since
they can’t imagine how one arc can wrap around another arc. This is because,
by its very nature, knottedness is an extrinsic property of circles in 3-dimensional
space.
Figure 2. A. Dash in some knotted universes.
Even though it might sound like it’s not a theory, knot theory is an important
area of topology that has some surprising applications, including to the study of
DNA and proteins, as we’ll see in Chapters 14 and 15.
The broad goal of knot theory is to develop methods to determine whether a
given knot can be deformed to another knot, or even to an ordinary circle. This is
not as easy as it might seem. Think about how hard it can be to untangle fishing
line or thread when it’s tangled up, and such strings have loose ends so you should
be able to unthread the tangles. When we study knots, we require that the ends of
our string be glued together so that if there is a knot, it will be “trapped” in the
loop. Otherwise, every knotted string could be deformed to an unknotted arc, as
we see in Figure 3.
Figure 3. A knotted arc is topologically the same as an unknotted arc.
Mathematically speaking, we define a knot as a circular loop in R3 , and we

define a link as one or more disjoint loops in R3 . Note that this means that a knot
can also be considered to be a link with only one component. Each loop of a link is
2. WHEN ARE TWO KNOTS EQUIVALENT? 217
called a component. For example, in Figure 4, A. Dash is in one component of a link,

and it would be impossible for her to get to the other component, or even to know
that it existed. In fact, since you can’t get from one to the other, each component
of a link is its own universe, regardless of how they may be linked together in
3-dimensional space. Imagine if our universe were somehow linked together with
another universe in a higher dimension but we had no way of knowing that the
other universe even existed.
Figure 4. A. Dash in one component of a link.
2. When Are Two Knots Equivalent?
We say that two knots or links are equivalent or even the same if one can be
deformed to the other in R3 . Figure 5 shows two knots that are equivalent. Can you
visualize a deformation that transforms the knot on the left into the knot on the
right? At this point it would be a good idea to beg, borrow, or steal an extension
cord (the shorter the better). Such a cord will come in handy to do deformations
in this and the next two chapters. Tie your extension cord into the knot on the left
in Figure 5, then stick the plug end of it into the socket end to make a circle. Now
see if you can deform your cord into the knot on the right in Figure 5.
Figure 5. These two knots are equivalent.
Rephrasing this notion of sameness in the language of Chapter 3, we say that

two knots or links are equivalent if they have the same extrinsic topology in R3 . In
contrast to earlier chapters, where we focused on the intrinsic properties of a space,
the study of knots and links is focused on extrinsic topology. In particular, every
knot has the same intrinsic topology as an unknotted circle—they are all the same
to A. Dash—and every link with n components has the same intrinsic topology as
n unknotted and unlinked circles.
It is because they exist in 3-dimensional space that knots and links are interest-
ing. Unfortunately, since the pages of this book are only 2 dimensional, we need to
represent knots and links in a 2-dimensional way. A projection is a 2-dimensional
drawing of a knot or link where we indicate undercrossings (i.e., places where part
of the knot or link passes under another part in R3 ) with gaps in the arcs. All the
pictures of knots and links you’ve seen so far in this chapter have been projections.
You can think of a projection as the shadow cast by shining a bright light on the
knot or link from above, with gaps to indicate crossings. As we deform a knot or
link in R3 , its shadow will change, giving us many different projections of the same
knot or link. For example, Figure 5 shows two projections of the same knot.
Any knot that can be deformed to the projections in Figure 5 is called the
trefoil knot. Notice that both of the projections in Figure 5 have three crossings.
Do all projections of the trefoil have three crossings? Well, no; it’s pretty easy to
deform the trefoil so it will have additional crossings, as illustrated in Figure 6.
Figure 6. Projections of the trefoil with different numbers of crossings.
However, it’s often more difficult to remove crossings from a projection than to
add crossings. Is there a projection of the trefoil with fewer than three crossings?
We use c(K) to denote the fewest number of crossings of any projection of a knot
or link K, and refer to c(K) as the crossing number of K. We might guess that the
crossing number of the trefoil is three. But looks can be deceiving. In Figure 7, we
see a projection of a knot with six crossings. But the crossing number of this knot
is less than six. Can you see how you might deform it to remove some crossings?
Again, your extension cord may come in handy here.
Figure 7. The crossing number of this knot is less than six.
The simplest knot is an ordinary unknotted circle, which is called the unknot.
It is the only knot that has crossing number zero. Similarly, any link with crossing
number zero is called an unlink . We say that the unknot and unlink are trivial,
while all other knots and links are said to be non-trivial. Figure 8 shows a projection
of a trivial knot. See if you can deform it so that it has no crossings.
Figure 8. A complicated projection of the unknot.
One goal of knot theory is to be able to tell if a knot is trivial or not. You
might think that a computer program could do this for us: just feed the computer
a knot projection, and after a few seconds, out pops an answer! Unfortunately, no
3. THE MIRROR IMAGE OF A KNOT OR LINK 219
such computer program exists. It is an unsolved problem of knot theory to find

and implement an algorithm that distinguishes trivial knots from non-trivial knots
in a reasonable amount of time.
3. The Mirror Image of a Knot or Link
Even if you don’t know whether a knot is trivial, you can still do some inter-
esting things with it. For example, imagine holding a knot up to a mirror. Now
you have two knots: the one in your hand, and the one in the mirror. It’s natural
to wonder whether you can deform the knot in your hand so that it looks like its
mirror image. In Figure 9, A. Square is standing in front of a projection of the
trefoil looking in a mirror.
Figure 9. A. Square standing in front of a knot projection looking

in a mirror.
Let’s say our original knot is called K. Then we call its mirror image K ∗ .
Starting with a projection of K, we can draw a projection of K ∗ without using a
mirror. We just switch all of the crossings in K from over to under, and under to
over. To see why this works, imagine the mirror is in the plane of the paper. Every
place the knot goes above the page, it’s mirror image goes below the page, and
every place the knot goes below the page, it’s mirror image goes above the page.
Figure 10 illustrates a link and its mirror image.
Figure 10. To get the mirror image, we switch all of the crossings.
So, is a knot or link really different from its mirror image? It’s not hard to show
that the links in Figure 10, known as Hopf links, are actually the same. Just turn
over one component of K without moving the other, and you’ll get K ∗ . However,
not all knots and links are equivalent to their mirror images; in fact, most are not.
The trefoil knot, and its mirror image (in Figure 9) are not equivalent, though we
won’t be able to show this until Chapter 10.
4. The Connected Sum of Two Knots
In Chapter 6, we used connected sums to build complicated manifolds from

simpler ones. We will now use a similar idea to take familiar knots and build more
complicated ones. To take the connected sum of knots K1 and K2 , we cut each
knot into a knotted arc and glue the resulting endpoints together as illustrated in
Figure 11. This will yield a new knot, denoted by K1 #K2 .
Figure 11. The connected sum of two knots.
The knot on the left side of Figure 11 is the trefoil knot that we saw previously,
and the knot on the right is the figure eight knot. These knots come up often. So
it’s good to know their names. In particular, if you ran into one of them at a party,
it would be embarrassing not to know its name.
The connected sum of knots can be thought of as a 1-dimensional version of
the connected sum of two surfaces or two 3-manifolds that we studied in Chapter 6.
Just as we can create infinitely many different surfaces by taking connected sums
of tori and projective planes, we can use the connected sum operation to create
infinitely many different knots. For example, for any number n, we can create a
knot which is the connected sum of n trefoil or figure eight knots. In Figure 12, we
illustrate a knot which is the connected sum of three figure eight knots.
Figure 12. This knot is the connected sum of three figure eight knots.
In Chapter 6, we compared the properties of the connected sum operation for

surfaces with ordinary addition for numbers. In particular, recall that a sphere is
an identity element for connected sums of surfaces because the connected sum of
5. A BRIEF HISTORY OF KNOT THEORY 221
any surface with a sphere will be topologically the same as the original surface.
Similarly, the unknot is an identity element for connected sums of knots.
We can also compare the connected sum operation to multiplication of integers.
In particular, a positive integer is said to be prime if it cannot be expressed as a
product of two integers neither of which is equal to the number 1. Every positive
integer which is not 1 and which is not prime is said to be composite. For example,
the number 3 is prime, but the number 6 is composite. If we applied this concept to
connected sums of finite surfaces, we could say that a surface is prime if it cannot
be expressed as the connected sum of two surfaces, unless one of the two surfaces
is topologically equivalent to a sphere. By the Classification of Surfaces we know
that every finite surface has the form S 2 , nT 2 , or nP 2 for some n. Hence the only
prime surfaces are T 2 and P 2 .
Now we apply this concept to connected sums of knots. In particular, we say a
knot is prime if it can’t be expressed as a connected sum of two knots, unless one
of them is the trivial knot (the unknot). Every knot which is not the trivial knot
and which is not prime is said to be composite. For example, the trefoil knot and
the figure eight knot are both prime, whereas the knot in Figure 12 is composite.
However, while the study of finite surfaces which are prime is not very interesting
because there are only two prime surfaces, there are infinitely many prime knots.
You can think of prime numbers, together with the operation of multiplication,
as forming the building blocks of all positive integers. This is why prime numbers
play a big role in number theory. Prime knots play a similar role in knot theory.
Since every non-trivial knot is either prime or a connected sum of prime knots,
we study prime knots as a way to understand all knots. In fact, just as every
number can be expressed uniquely as product of prime numbers, every knot can be
expressed uniquely as the connected sum of prime knots.
Figure 13 is a table listing every prime knot with fewer than nine crossings.
Each knot in the table is denoted with two numbers: the first number is the crossing
number of the knot, and the subscript is just a way to keep track of different knots
with the same crossing number. To save space, mirror images are not listed in the
table. For example, the trefoil is denoted by 31 . There is no 32 , because the only
other knot with crossing number 3 is the mirror image of 31 , which is denoted as
3∗1 . Similarly, the figure eight knot is the only knot whose crossing number is 4.
On the other hand, there are two different knots with crossing number 5: one is
denoted by 51 and the other is denoted by 52 .
As you might guess, there are significantly more 9 crossing knots than 8 crossing
knots, and still more 10 crossing knots than 9 crossing knots. Knot theorists have
compiled tables of all prime knots with up to 16 crossings – a total of 1, 701, 935
distinct knots. Not surprisingly we don’t include the table here. In fact, such a
table is only available online.
5. A Brief History of Knot Theory
Today’s knot tables trace their origins back to 19th-century Scotland. While
the United States was embroiled in civil war, Scottish physicists were trying to
develop a model for the atom. We know today that atoms are made of a dense,
positively charged nucleus surrounded by a cloud of electrons. But in the 19th
century, physicists had very different theories about atoms. In particular, their
Figure 13. A table of prime knots with up to 8 crossings.
ideas grew out of a theory, introduced by Descartes in 1644, that all of space and
matter were filled with an invisible substance called ether. This theory took off in
the 19th century when physicists believed space needed to be filled with some kind
of medium in order to explain how light could behave like a wave and yet travel
outside of the atmosphere. They referred to this medium as luminiferous ether to
indicate its role in light propagation.
At the time, scientists knew that elements, such as carbon, oxygen, etc., existed
and also that matter was composed of atoms. However, the actual structure of
atoms was still a mystery. One prominent theory came from the work of physicists
Peter Guthrie Tait and Sir William Thomson (a.k.a. Lord Kelvin, of the Kelvin
temperature scale). Inspired by the behavior of smoke rings that Tait had observed
in his laboratory, they theorized that atoms were swirling tubes, or “vortices”, in the
6. REIDEMEISTER MOVES 223
ether. But these tiny tubular whirlpools were not simple geometric circles. Rather,
they were knotted or linked, and the type of knot or link in an ether whirlpool
determined the element.
Thomson thought the spectroscopic properties of each element were reflected in
the crossings of its projection. For example, he hypothesized that a sodium atom,
which emits two prominent spectral lines when energized, should correspond to a
vortex with crossing number 2. Perhaps a trefoil in the ether was really a helium
atom, and a figure eight knot could be a lithium atom (see Figure 14).
Figure 14. Atoms were believed to be knotted or linked vortices

in the ether.
In the mid-19th century, physicists were beginning to create tables of elements

that much later would be replaced by the modern periodic table. Tait, Thomson,
and their colleague James Maxwell (famous for his equations of electromagnetism)
realized that if this knotted vortex theory was correct, a table of distinct knots was
equivalent to a table of the elements. Creating such a table would be an important
scientific contribution, and Tait thought an analysis of crossing number would be
a good place to start.
With the help of a few other early knot theorists, Tait worked for many years
to create an accurate knot table based on crossing number. While their work was
imperfect and non-rigorous, they noticed many important patterns. In particular,
Tait made several conjectures about the properties of knots and their crossings, two
of which we’ll explore in the next chapter.
Unfortunately, all of this scientific interest in knots would soon come to an
abrupt end. The ether vortex model for the atom was falling out of favor among
physicists, and by 1887 the famous Michelson-Morley experiment shook the sci-
entific world by showing conclusively that luminiferous ether did not exist! Sadly,
since the primary motivation for studying knots was to create a periodic table based
on the vortex model, knot theory was abandoned for many years. It wasn’t until the
1920s that interest in knots was re-ignited by topologists, and Tait’s conjectures—
still unproven—were rediscovered.
6. Reidemeister Moves
A. Square has been listening quietly on the sidelines while we’ve been discussing
knot tables and the history of knot theory. He’s interested in the table of knots
in Figure 13 because it’s filled with 2-dimensional pictures that he can appreciate.
The problem is that when A. Square looks at a projection, he sees it as a collection
of disjoint arcs in Flatland. For example, A. Square sees the projection of a trefoil
in Figure 15 as three disconnected arcs that are free to move away from each other.
In order to help him understand that the arcs are actually joined together,
without losing the important 3-dimensional information of which strand goes under
Figure 15. A. Square sees a projection of a knot as a collection

of disconnected arcs in Flatland.
and which strand goes over, we add a grey arc where the gaps are. We’ll call the
projection with the grey arcs a Flatland-friendly projection of a knot rather than
just a projection. Now we explain to A. Square that if he can deform one Flatland-
friendly projection of a knot to another, then the knots are equivalent. For example,
he can deform the Flatland-friendly projection on the left in Figure 16 to the one
on the right. Keep in mind that this is a planar deformation, meaning that the
deformation takes place in the plane. He is not sliding a grey arc under one of the
arcs of the projection. Rather, he sees the black and grey together as forming a little
cross that moves along as he deforms and rotates the Flatland-friendly projection
in the plane.
Figure 16. A. Square can deform the Flatland-friendly projection

on the left to the one on the right.
A. Square is disappointed to learn that you can’t get from any knot to any
equivalent knot by just using those planar deformations that he can do. However,
A. Square is in luck. The German mathematician Kurt Reidemeister showed in the
1920s that two knots are equivalent to one another if and only if you can go from a
projection of one knot to a projection of the other by using combinations of three
small “moves” in the plane together with planar deformations. The reason that we
say these moves are small is because, when we do one of the moves, we don’t change
anything else about the projection. By applying these moves one after the other to
different parts of the projection, we can make big changes in the projection. But
each individual move is just a small change in a very specific spot.
We will illustrate the three moves on knot projections and explain what we’re
doing in terms of 3-dimensional deformations. But we tell A. Square to fill in the
gaps in the projections with grey arcs and to think about any moves as allowable
changes in order to make a knot into a Flatland-friendly projection in the plane,
even if he can’t see with his 2-dimensional vision why these moves are valid.
Reidemeister I move: A kink can be added or removed as in Figure 17. The

double headed arrows in the figure indicate that you can do the move going from
6. REIDEMEISTER MOVES 225
either of the kinks to no kink, or from no kink to either kink. With this move, a
crossing is removed or added to the projection. We denote this move by (R1).
Figure 17. (R1) A kink can be added or removed.
Reidemeister II move: An arc can be slid over or under another arc as in Figure
18. With this move, two crossings are removed or added to the projection. We
denote this move by (R2).
Figure 18. (R2) An arc can be slid over or under another arc.
Reidemeister III move: An arc can be slid under or over a crossing as illustrated
by the thick black arc in Figure 19. With this move, the number of crossings remain
the same, but the arc is positioned differently in the projection. We denote this
move by (R3).
Figure 19. (R3) An arc can be slid under or over a crossing.
It’s easy to check that any two projections related by a Reidemeister move (or
a sequence of finitely many such moves) must represent equivalent knots. However,
it’s not easy to see why repeatedly using just these three moves together with planar
deformations would give you all possible deformations of a knot. The proof of the
following important theorem is quite technical, so it is unlikely that you would see
it in a course. However, if you are interested, you can find several different proofs
of it on the web.
Theorem 9.1 (Reidemeister’s Theorem). Any deformation of a knot or link in
3-dimensional space can be achieved by planar deformations together with a finite
number of the three Reidemester moves.
Reidemeister’s Theorem tells us that if we know that two projections represent
equivalent knots, then we can go from one to the other with a finite sequence of
Reidemeister moves, denoted by (R1), (R2), and (R3), along with planar deforma-
tions denoted by (P). For example, we observed that the two projections in Figure 5
represent equivalent knots. It follows that there must be a sequence of Reidemeister
moves and planar deformations taking one to the other. Figure 20 shows one such
sequence, but there are many other possibilities.
Figure 20. A sequence of Reidemeister moves and planar deformations.
While Figure 20 illustrates how you could deform a projection using Reidemeis-
ter moves, the Reidemeister moves should not actually be used for this purpose. In
particular, deforming a projection with Reidemeister moves is much slower, harder,
more confusing, and more tedious than deforming the knot or link in 3-dimensional
space with the help of an extension cord. So when you’re doing the Exercises, re-
member that you should NEVER use Reidemeister moves to prove that two knots
are equivalent unless you are specifically told to do so.
If we don’t use Reidmeister moves to deform one knot to another, than why
do we even care about Reidemester’s Theorem? The point of the theorem is that
it gives us a way to prove that certain properties of knots and links are preserved
under any deformation. In particular, if we can show that a property does not
change under any of the three Reidemeister moves or planar deformation, then we’ll
know that the property doesn’t change under any of the infinitely many possible
deformations of a knot or link in 3-dimensional space. Hence if one knot has such
a property and another knot doesn’t, then we can conclude that the knots are not
equivalent. We’ll see how this works in the next section.
7. Coloring Knots with Three Colors
You’re so excited about all the cool math you’ve been learning lately that you
decide to take your knot table with you to Thanksgiving dinner at your grand-
parents’ house. While you’re busy stuffing yourself with your second helping of
stuffing, your eight-year-old cousin, Felicity, finds your knot table and starts col-
oring on it with her crayons. She traces the arcs of the knot projections with her
three favorite colors (black, grey, and striped). After noticing that at each crossing
there are three arcs which come together, she makes the following list of rules for
herself.
Felicity’s Very Special Coloring Rules:
(1) More than one color must be used for every knot.
(2) At each crossing either the three arcs that come together must all be the
same color or each arc must be a distinct color.
7. COLORING KNOTS WITH THREE COLORS 227
She skips the unknot because it’s boring and starts coloring the other knot
projections. She’s pretty happy with her coloring of the projection of the trefoil
knot (illustrated in Figure 21) because she succeeded in following her rules.
Figure 21. Felicity’s coloring of a projection of the trefoil.
The next knot in the table is the figure eight knot. She starts with the arc on
the right, and colors it black (as in Figure 22). Then she looks at the long arc on
the left and decides to color it grey.
Figure 22. Felicity colors the right arc black.
Since black and grey come together at the bottom left crossing, by Felicity’s
Rule (2), the third arc at this crossing must be striped (see Figure 23).
Figure 23. Because black and grey come together at this crossing,
by Rule (2) the third arc must be striped.
Now there’s only one arc left to color—the short arc in the middle. At the top
crossing, Felicity has already colored one arc grey and one black, so by Rule (2)
she has to color the middle arc striped. But this means that the middle crossing
has two arcs that are striped and one that is grey, which violates Rule (2) (see
Figure 24).
Felicity gets frustrated and wads up the paper. You go over to console her and
ask why she’s upset. She smooths out the crumpled paper and explains her game
and what didn’t work. You try to be helpful and observe that she hasn’t tried all
the options yet. You suggest that she color the left arc black instead of grey. But
then you see that this implies by Rule (2) that she must also color the bottom arc
black, as illustrated in Figure 25.
Again she is left with one arc to color. Felicity doesn’t want a monochromatic
picture because that would violate Rule (1). But because of Rule (2) she is forced
Figure 24. The crossing in the center violates Rule (2).
Figure 25. Since two arcs at the bottom left crossing are black,
the third arc must be as well.
to color the middle arc black. Before she has a complete meltdown, you teach her to
play torus tic-toe-toe and let her go first. After winning torus tic-tac-toe, Felicity
quickly forgets about coloring knots.
The coloring game that Felicity was playing turns out to be useful for distin-
guishing knots. Let’s take a moment to formalize Felicity’s rules.
Definition. A projection of a knot or link is said to be tricolorable if there

is a way to color the arcs with one of three colors such that:
(1) More than one color is used.
(2) At each crossing, either only one color is used or all three colors are used
A knot or link is said to be tricolorable if it has a tricolorable projection.
Figure 26. Each crossing has either only one or all three colors.
Notice that we have to color a projection of a knot rather than a knot itself
because a knot itself is a curve in space and doesn’t have crossings. It is projec-
tions of the knot that require one arc to pass in front of another and hence have
crossings. However, this definition is telling us that if a single projection of a knot
is tricolorable, then we will say that the knot itself is tricolorable even if some other
projection of the knot is not tricolorable.
8. TRICOLORABILITY AND KNOT EQUIVALENCE 229
8. Tricolorability and Knot Equivalence
Our experience with Felicity shows us that the trefoil knot is tricolorable. But
we haven’t shown that the figure eight knot is not tricolorable. We’ve only shown
that the particular projection of the figure eight in Figure 22 is not tricolorable.
Perhaps we can deform the figure eight to a position that has a tricolorable projec-
tion. After all, there are infinitely many different ways to deform a knot.
This is where Reidemeister’s Theorem is useful, because it tells us that any
two projections of a given knot can be connected by deformations in the plane
together with a finite sequence of Reidemeister moves. Deforming the arcs in the
plane certainly doesn’t change whether a projection is tricolorable. If, in addition,
we could show that each of the three Reidemeister moves doesn’t change whether
a projection is tricolorable, then we would know that no deformation, no matter
how complicated, could cause a tricolorable projection to change into one which
was not tricolorable or vice versa. This would tell us that for a given knot, either
all projections are tricolorable or none of them are.
Theorem 9.2. The tricolorability of a knot or link projection does not change
when a single Reidemeister move is applied.
Proof. Let’s suppose that we start with a knot or link projection which is
tricolored and we do a single Reidemeister move. We want to show that the new
projection is also tricolorable. Since a Reidemeister move only changes a projection
in a small disk (which we will refer to as “the disk”), we want to show that it’s
possible to color the new projection within this disk so that the arcs exiting the disk
are the same color as they were before and at least two colors are still used in the
overall projection. If we can do this, we’ll have a tricoloring of our new projection,
and hence we’ll be able to conclude that doing this one Reidemeister move did not
change the tricolorability of the projection.
We’ll prove this for Reidemeister moves (R1) and (R2) and give some hints
about how to prove it for (R3). You can fill in the details for (R3) in the Exercises.
Reidemeister I move: First suppose that we want to apply an (R1) move to get
rid of a kink (this means going from left to right in Figure 27). Observe in Figure 27
that just two arcs come together at the crossing of a kink. By Felicity’s Rule (2),
the projection within the disk must be a single color, say black, and hence the arcs
leaving the disk are both colored black. Now coloring the single arc that’s left black
will be consistent with the coloring of the arcs leaving the disk. Also, we have not
changed the total number of colors used in the overall projection. Thus, this gives
us a tricoloring of the overall projection after the (R1) move.
Next suppose we want to apply an (R1) move to add a kink (this means going
from right to left in Figure 27). Then the original projection in the disk is a single
Figure 27. Tricoloring within a small region before and after an

(R1) move.
arc, and hence must be colored with a single color, say black. When we add the
kink we color its two arcs black so that again the coloring in the region is consistent
with the coloring of the arcs leaving the disk. Again we have not changed the total
number of colors used in the overall projection. Thus, we still have a tricoloring of
the overall projection after the (R1) move.
So doing a single (R1) move will not change a tricolorable projection into a not
tricolorable projection.
Reidemeister II move: First suppose that the arcs in the disk where we’re going
to do an (R2) move are originally all the same color, say black. Then, as we saw
in the proof for (R1) moves, whether we are going from the projection on one of
the sides of Figure 28 to the projection in the middle or vice versa, by coloring the
new arcs in the disk black the coloring will be consistent with the coloring of the
arcs leaving the disk and we won’t have changed the number of colors used in the
overall projection. Thus we will still have a tricoloring of the new projection.
Figure 28. If the arcs in the disk are all black, then we can color
them all black after doing an (R2) move.
Next suppose that the arcs in the disk are not all a single color, and we are
doing an (R2) move to remove two crossings. Since one arc goes over both crossings,
it must have only one color (say black as illustrated on the far left or far right of
Figure 29). Say the top of the other arc is grey. Then the middle arc is forced to be
striped, and hence the remaining bottom arc is forced to be grey. Thus prior to the
(R2) move, the arcs in the disk must be colored as illustrated on the far left or far
right in Figure 29. After doing the (R2) move to remove the two crossings, we color
one of the new arcs black and the other new arc grey so that they match up with
the colors of the arcs leaving the disk. Also, since the coloring in the disk uses more
than one color, the overall coloring satisfies Felicity’s Rule (1) of tricolorability.
Thus we have a tricoloring of our new projection.
Figure 29. The arcs in the disk are not all the same color, and
we do an (R2) move.
Finally, suppose that the arcs in the disk are not all a single color, and we are
doing an (R2) move to add two crossings. Then originally there are no crossings
in the disk, and we can assume one arc is black and the other is grey as illustrated
in the middle two drawings in Figure 29. Now after we do the (R2) move, we can
8. TRICOLORABILITY AND KNOT EQUIVALENCE 231
color the new arc that is created striped. This will give us a tricoloring within the
disk which matches up with the coloring of the arcs exiting the disk. Also since the
coloring in the disk uses all three colors, the overall coloring satisfies Felicity’s Rule
(1) of tricolorability. This again gives us a tricoloring of the new projection.
Reidemeister III move: This proof is one of the Exercises. It’s similar to the
proof for Reidemeister II moves. However, for (R2) moves we saw that there are
two cases depending on whether or not all of the arcs in the disk are the same color.
Since there are more crossings involved in an (R3) move, there are more cases that
you have to consider. For example, in Figure 30 we illustrate three different types
of colorings inside of the disk where an (R3) move will take place.
Figure 30. Three different colorings inside of the disk where an

(R3) move will take place.
Try to list all possible colorings of the arcs in the disk. Then see if one of your
diagrams can be turned around or the colors can be interchanged to get another
one of your diagrams. If so, then these two cases are analogous and, hence, don’t
need separate proofs. Once you eliminate the analogous cases, there should be five
cases remaining. In each case, you need to show that it’s possible to color the
new projection within the disk so that the arcs exiting the disk are the same color
as they were before the (R3) move and at least two colors are still used in the
projection.
In the arguments above for moves (R1), (R2), and (R3) we showed that if a
projection of a knot or link is tricolorable and we do a single Reidemeister move,
then the resulting projection will still be tricolorable. But if we start with a pro-
jection which is NOT tricolorable and do a single Reidemeister move, could we end
up with a projection which IS tricolorable? In fact, this follows from what we have
already shown. In particular, if we ended up with a tricolorable projection, we
would just start with that projection and do the Reidemeister move in reverse to
show that our original projection was also tricolorable.
Thus once you have completed the argument for (R3) moves, we will have shown
that doing a single Reidemeister move will neither change a tricolorable projection
to one which is not tricolorable nor will it change a not tricolorable projection to
one which is tricolorable.
Now we can use Theorem 9.2 together with Reidemeister’s Theorem to prove
the following corollary.
Corollary. One projection of a knot or link is tricolorable if and only if every
projection of the link is tricolorable.
Since our corollary contains the phrase “if and only if” we need to prove both “if
some projection is tricolorable, then every projection is tricolorable” and “if every
projection is tricolorable then a particular projection is tricolorable”. Recall that

we use the symbol (=⇒) to indicate that we are about to prove the first statement,
and (⇐=) to indicate that we’re about to prove the second statement.
Proof. (=⇒) Suppose that some projection of a knot or link is tricolorable.

By Reidemeister’s Theorem, any other projection of the link can be obtained by
a finite sequence of Reidemeister moves together with planar deformations. As we
remarked earlier, a planar deformation will not have any effect on tricolorability.
Also by Theorem 9.2, after each subsequent Reidemeister move, the new projection
is still tricolorable. Hence, after a finite sequence of Reidemeister moves and planar
deformations, the projection is still tricolorable. Thus every projection of the knot
or link is tricolorable.
(⇐=) If every projection of a knot or link is tricolorable, then certainly any
particular projection is tricolorable.
We could restate this corollary as saying that if a pair of knots or links is

equivalent and a projection of one of the knots or links is tricolorable, then any
projection of the other one must also be tricolorable. Also, it follows from the
corollary that if one projection of a knot or link is not tricolorable, then no pro-
jection is tricolorable. Thus we can now conclude that the figure eight knot is not
tricolorable, since Felicity showed us that one projection of it is not tricolorable.
Since the trefoil is tricolorable and the figure eight is not, we’ve thus shown that
the trefoil and the figure eight knot are not equivalent.
Observe that the unknot is not tricolorable because it has a projection with no
crossings which would have to be colored with a single color, violating Rule (1) of
tricolorability. On the other hand a trivial link with two components is tricolorable,
since even though the link has a projection with no crossings, each component can
be colored with a different color. Since the figure eight and the unknot are both
not tricolorable, we don’t yet know how to prove that the figure eight knot is not
equivalent to the unknot. Similarly, if two knots or links are both tricolorable, then
we cannot distinguish them from each other in this way.
9. Oriented Knots and Invertibility
A. Dash is fond of taking walks around her universe, stopping occasionally for a
picnic. Sometimes, though, after some wine and cheese, she forgets which direction
she is going. She’d prefer not to retrace her steps, so she decides to mark the
direction she’s walking with an arrow every time she stops.
A knot with a chosen direction like this is said to be oriented. The direction,
or orientation, is shown by one or more arrows on the knot, drawn consistently like
signs on a one-way street (see Figure 31). In order to orient a link, we have to put
an orientation on each component of the link.
If K is an oriented knot or link, then −K denotes K with the opposite orienta-
tion on every component. Figure 32 shows that when the trefoil knot is turned over,
we get the same projection that we started with, but with the arrow reversed. Since
the trefoil knot is denoted by 31 , we say that 31 is equivalent to −31 . However, not
every oriented knot is equivalent to itself with its orientation reversed.
10. CONNECTED SUMS OF NON-INVERTIBLE KNOTS 233
Figure 31. A. Dash gives her knotted universe an orientation.
Figure 32. When we turn over the trefoil knot its projection is
the same but the direction of the arrow is reversed.
Definition. If an oriented knot K can be deformed so that it has an identical

projection but with its orientation reversed, then we say the knot is invertible. If
no such deformation exists, then we say the knot is non-invertible.
We saw in Figure 32 that the trefoil knot is invertible. In fact, most knots in the
table in Figure 13 are invertible. It was not until 1963 that the first non-invertible
knot was even discovered. The knot with the fewest crossings that is non-invertible
is 817 (illustrated in Figure 33), though this was not the first knot which was shown
to be non-invertible.
Figure 33. 817 is the non-invertible knot with the fewest number
of crossings.
10. Connected Sums of Non-invertible Knots

When working with non-invertible knots, the definition of the connected sum
operation is slightly more complicated than what we gave in Section 4. In particular,
if K1 and K2 are non-invertible knots, then we have to orient the knots before we can
take their connected sum so that we know which end of one knotted arc to attach
to which end of the other knotted arc. This is not a problem for the connected sum
of invertible knots, since an invertible knot can be reversed by a deformation.
Definition. Let K1 and K2 be oriented knots. The connected sum K1 #K2
is formed by cutting each knot and gluing the endpoints of K1 to the endpoints of
K2 in such a way that the arrow going out of K1 is attached to the arrow going
into K2 .
The following examples involving the non-invertible knot 817 show why it’s im-
portant to specify an orientation when forming the connected sum of non-invertible
knots. We begin by giving 817 an orientation, and then use the above definition
to create the knot 817 #817 in Figure 34. The original direction of the orientation
doesn’t matter, but once we assign an orientation, we have to stick with it. In par-
ticular, both copies of 817 have to have the same orientation before we take their
connected sum. If we want to use the opposite orientation on one of the knots, then
we write −817 rather than 817 . Observe in Figure 34 that since we have glued the
arrow going out of one 817 to the arrow going into the other 817 , the arrows on the
connected sum match up so that 817 #817 has a single orientation.
Figure 34. 817 #817 .
If we had glued the arrow going out of one 817 to the arrow going out of the
other 817 , we would end up with conflicting orientations on the connected sum.
You wouldn’t want to be driving down a narrow one-way street and encounter two
one-way signs pointing head to head, just as you see another car speeding towards
you.
In Figure 35 we illustrate the connected sum 817 # − 817 . Notice that since the
knots have opposite orientations, we need to turn one knot over in order to line
up the endpoints so that we can glue them correctly. Since 817 and −817 are not
equivalent, the connected sums 817 #817 and 817 # − 817 are not equivalent.
Figure 35. 817 # − 817 .
One way to illustrate connected sums of oriented knots without having to keep
track of all of the crossings is as beads with pictures on them. If a knot is invertible,
then its bead has a picture that is the same regardless of which hole is up and which
hole is down. If the knot is non-invertible, then its bead looks different if we turn
it upside down. Figure 36 illustrates a bead representing an invertible knot and
another bead representing a non-invertible knot.
We can make a necklace out of two non-invertible beads in two different ways,
as illustrated in Figure 37. If your beads represent the knot 817 , the necklace on
the left represents the connected sum 817 #817 because the two beads are both
oriented stem to flower. The necklace on the right represents the connected sum
817 # − 817 because the left bead is oriented stem to flower and the right bead is
oriented flower to stem. This example shows that orientation really does matter for
connected sums of non-invertible knots.
11. EXERCISES 235
Figure 36. An invertible bead and a non-invertible bead.
Figure 37. We can make a necklace out of two non-invertible

beads in two different ways.
Let’s say Felicity asked you and A. Square to each show her the connected sum
of two unoriented 817 knots. You might come up with the knot in Figure 34, without
any arrows, while A. Square might draw the knot in Figure 35 (again, without any
arrows). These knots are non-equivalent, but both you and A. Square would have
done everything correctly. This shows us that the connected sum operation without
orientations is ambiguous for non-invertible knots. Thus, while it still makes sense
to talk about the connected sums of unoriented knots in general, we have to be
careful to always use orientations if one of our knots might be non-invertible.
11. Exercises
1. Use Reidemeister moves to show that the links in Figure 38 are equivalent. In
other words, show a sequence of moves I, II, and III taking one projection to
the other.
2. Use Reidemeister moves to show that the knot in Figure 39 is really the unknot.
3. Prove that every knot has a projection with over 1000 crossings.
4. Show that the third Reidemeister move preserves tricolorability. This proof is
similar to the proof in the text for Reidemeister II moves. However, there are
more cases because you have to consider all the ways that the arcs exiting a
small disk around the (R3) move might be colored. By turning the diagram
around or interchanging colors, you can see that some of the cases are analogous
and hence don’t need separate proofs. Once you eliminate the analogous cases,
there will only be five cases remaining. In each case, you need to show that
it’s possible to color the new projection within the disk so that the arcs exiting
the disk are the same color as they were before the (R3) move and at least two
colors are still used in the projection.
5. Consider all of the seven crossing knots in the tables and determine which ones
are tricolorable.
6. Draw a picture of 31 #41 , and show that it is tricolorable. If you used some
other knot in place of 41 in this connected sum, would the connected sum still
be tricolorable?
7. Let K1 be any tricolorable knot, and let K2 be any knot (which may or may
not be tricolorable). Prove that K1 #K2 is tricolorable.
8. There are many different ways to color a tricolorable knot. For example all of
the black arcs could be colored grey, while all of the grey arcs are changed to
striped, and all of the striped arcs are colored black.
(a) Show that the projection of the trefoil knot on the left in Figure 40 has
exactly six tricolorings.
(b) Find two different tricolorings of the trefoil knot where one can be obtained
from the other by simply permuting the three colors. Such tricolorings are
called equivalent. How many non-equivalent tricolorings does the trefoil
knot have?
(c) How many tricolorings does the square knot (shown on the right in Fig-
ure 40 right) have?
(d) How many non-equivalent tricolorings does the square knot have?
9. Show that the two projections of the link in Figure 41 are equivalent.
10. Prove that the link in Figure 41 is not tricolorable and therefore is not the
trivial link of two components.
11. EXERCISES 237
Figure 41. Illustration for Exercises 9 and 10.
11. Draw two crossings in the plane, and consider all possible ways to connect the
different endpoints. Then use all of the ways of connecting the endpoints as
cases in a proof that there can be no non-trivial knots with only two crossings.
How does this imply that the crossing number of a trefoil knot is 3?
12. Give an upper bound for the number of links with crossing number n (don’t
try to find the least upper bound).
13. Prove that for all unoriented knots K1 and K2 , we have the inequality
c(K1 #K2 ) ≤ c(K1 ) + c(K2 ).
14. Here we will show that 41 is equivalent to 4∗1 . To do this, draw a deformation
of the figure eight knot (shown in Figure 42) to its mirror image. Don’t use
Reidemeister moves! Instead, try to make your deformation short and easy to
follow. You may want to color the arcs of your knot with different colors to
make it clear which part of the knot is being deformed at each step.
15. Prove that the two links illustrated in Figure 41 are equivalent.
16. Determine how many ways there are to orient a link with n components.
17. Recall that the knot 817 (shown in Figure 43) is non-invertible. Are the knots
817 #817 or (−817 )#817 invertible? You may want to draw beads to represent
the knots in this problem.
18. Is it true that −(31 #817 ) = −31 # − 817 ? Justify your answer.
19. Is 31 #817 equivalent to (−31 )#817 ? Is 31 #817 invertible? You may want to
draw beads to represent the knots in this problem.
20. If K1 is an invertible prime knot and K2 is a non-invertible prime knot, are
K1 #K2 and (−K1 )#K2 equivalent? Is K1 #K2 necessarily invertible? You
may want to draw invertible and non-invertible beads to represent the knots in
this problem.
21. Prove that any knot can be unknotted by changing some number of crossings.
Hint: Think about how a water slide always goes down, and use that as a model
for your projection.
CHAPTER 10
Invariants of Knots and Links
Topics:
• Linking number
• Writhe
• Tait conjectures
• Unknotting number
• Bridge number
• Stick number
• Lattice number
• Seifert surfaces and genus
1. What’s an Invariant?
After a long day of studying, you settle into a comfy chair and flip on the
television. Surfing through the channels, you stop at a new reality show. One of
the participants is named Buttercup. “Funny”, you say to yourself. “The only
person I’ve ever known named Buttercup is my cousin.” You haven’t seen your
cousin since she was a kid, so who knows what she looks like now. You pause and
look more closely at the TV screen trying to figure out if it could really be her.
Unfortunately, as much as you would like to be related to a reality TV star,
having the first name of your cousin is no proof that this woman is your cousin.
You continue watching the show, making a mental note of everything you see or
hear about Buttercup. She has brown eyes and black hair just like your cousin.
She mentions she was born in New York, and she has a fear of heights. You search
your brain trying to remember if the same things were also true about your cousin.
“Fiddlesticks”, you mutter and flip off the TV. It has been too long. You won’t
be able to figure this out without asking your mother. Even so, the time spent
watching the reality show has (for perhaps the first time ever!) not been a waste.
It reminded you of your other favorite pastime—comparing knots.
The primary goal of knot theory is to develop methods to determine whether
two knots are equivalent or not. This goal is similar to what you were just trying
to do with the two Buttercups. Consider the two knot projections in Figure 1.
Do these projections represent the same knot or different knots? The Buttercup
problem might take some time to solve (and a little help from your mother), but
it’s no match in difficulty to this knot puzzle. This particular puzzle took over 50
years to solve.
239
240 10. INVARIANTS OF KNOTS AND LINKS
Figure 1. Perko showed that these two projections represent the

same knot.
If we can find a deformation from one knot in Figure 1 to the other, then
they are equivalent. However, looking for a deformation won’t help us show that
the knots are different. Suppose you sit at a table, trying to deform one knotted
extension cord into the other, for years, without ever succeeding. This might provide
strong evidence that the knots are not equivalent, however you can’t be 100%
sure. Maybe just one more day of playing with the cord would enable you to show
that the knots are equivalent. In fact, the pair of knots in Figure 1, were listed
separately in the original knot tables. But in 1974, the two projections were shown
to be equivalent by Kenneth Perko, a lawyer who had studied knot theory as an
undergraduate. Since then, these two projections have been known as the Perko
pair.
While the Perko pair turned out to be the same knot, we’d like a method that
would enable us to prove that knots that seem different really are different. For
example, so far we haven’t even been able to prove that the figure eight knot cannot
be deformed to the unknot.
Recall from Chapter 3 that we said that a topological property of a space is a
characteristic of the space that is unchanged when the space is deformed. Such
properties enable us to distinguish spaces. For example, we know that the torus
and the Klein bottle are topologically distinct because the torus has the property
of being orientable whereas the Klein bottle does not. Looking for distinguishing
characteristics was also what you were trying to do with the two Buttercups. If
the TV Buttercup was born in New York and your cousin Buttercup was born in
California, then you would know they must be different people. We would like to
find topological properties that enable us to distinguish different knots in a similar
way.
Definition. An invariant of a knot or link is a property that does not change
when the knot or link is deformed in space.
This means that in order to show that something is an invariant, we have to
check that it won’t change when we deform the knot or link. If a quantity is defined
over all possible projections of a knot or link, then it’s an invariant by definition.
Before we introduce some specific knot invariants, let’s think about how we
use characteristics to distinguish people that we don’t know well. We often recog-
nize people based on physical traits like eye color, hair color, facial hair, glasses,
or height. We can imagine that each of these characteristics sorts people into dif-
ferent buckets. Of course, these buckets are figurative and not literal, but that’s
immaterial—the idea is the same. For instance, if we consider eye color, we might
have different buckets filled with blue-eyed people, green-eyed people, brown-eyed
people, people with bright purple contact lenses that they never remove, and so on.
2. CROSSING NUMBER, TRICOLORABILITY, AND NUMBER OF COMPONENTS 241
The buckets might be labeled according to the eye color for easy reference. If you
have two acquaintances who belong to different buckets (maybe one has green eyes
and the other has brown eyes), then you can tell them apart immediately using this
“bucket system”. However, if two people belong to the same bucket because they
both have bright purple contact lenses that they never remove, then you will need
to move on to a different bucket system—maybe hair color—to distinguish between
them.
Invariants work the same way for knots and links. We can think of an invari-
ant as a bucket system in which knots and links are sorted into different buckets
according to a particular invariant. If we have two equivalent knots or links, then
they will always be placed in the same bucket no matter how many bucket systems
we use because the value of any invariant will be the same. On the other hand, if
we find an invariant that places a pair of knots or links in different buckets (because
they have different values of the invariant), then we can be certain that the knots
or links are not equivalent to one another. But be careful. Just because a pair of
knots or links are in the same bucket or even several of the same buckets doesn’t
mean they are necessarily equivalent. This is just like the fact that two different
people can be placed in the same two buckets because they both have brown eyes
and black hair.
2. Crossing Number, Tricolorability, and Number of Components
Our first example of an invariant of knots and links is the crossing number.
Recall from Chapter 9 that the crossing number c(K) of a knot or link K is the
fewest number of crossings of any projection of K. This means no matter how
we deform K, a projection of it will have at least c(K) crossings. To compute the
crossing number, we start with any projection K and count the number of crossings.
We then try to deform K in such a way that the number of crossings is reduced.
If we find a projection with fewer crossings, then we repeat the process with this
new projection. If there is no projection of K with fewer than n crossings, then
c(K) = n. Since the crossing number is defined in over all possible projections of
the knot or link, by definition the crossing number is an invariant.
The problem with this method of computing the crossing number is that we
might not know whether in fact there is a projection of K with fewer than n
crossings. For example, Figure 2 illustrates a knot with 15 crossings. Perhaps if we
deformed this knot, we would find a projection with fewer than 15 crossings. Maybe
if we worked at it long enough, we could even show that this knot could be deformed
to the unknot. But just because we can’t figure out how to reduce the number of
crossings doesn’t mean there isn’t some way to do it. For this reason the crossing
number is difficult to compute, and hence it’s not very usable for distinguishing
knots.
Tricolorability which was introduced in Chapter 9, is another example of a
knot and link invariant. Recall that in Chapter 9 we showed that tricolorability is
preserved by all three Reidemeister moves as well as by planar deformation. Since
Reidemeister’s Theorem tells us that we can get from one projection of a knot or
link to any other projection via a finite sequence of Reidemeister moves and planar
deformations, we know that tricolorability is unchanged by any deformation. It
follows that tricolorability is indeed an invariant.
Figure 2. Can we find a projection of this knot with fewer than 15 crossings?
Also, in contrast with crossing number, tricolorability is not hard to deter-

mine. There are only so many ways to color a projection following the rules of
tricolorability. So if we have a little more patience than Felicity had, we can try
each possibility and reach a definite conclusion about whether or not a projection
is tricolorable. However, because tricolorability only divides knots and links into
those that are tricolorable and those that aren’t, it too has limited usefulness.
For example, since the trefoil knot is tricolorable and the figure eight knot is
not, we know that they are topologically distinct. But since both the figure eight
knot and the unknot are not tricolorable, tricolorability won’t help us determine if
they are topologically the same or different. In particular, just because two knots
are both tricolorable or both are not tricolorable doesn’t mean they are the same.
Similarly, just because two people are both called Buttercup or both have brown
eyes doesn’t mean they’re the same person.
Another example of an invariant is the number of components of a link. Each
component of a link is a disjoint loop and a deformation of a link cannot add or
remove a component. Thus, every projection of a link will have the same number
of components. Also, like tricolorability, the number of components of a link is easy
to compute. But also like tricolorability, the number of components doesn’t help
us distinguish many knots or links. In fact, all knots are links with one component.
So the number of components won’t help us distinguish knots at all. In addition,
the number of components of a link is so easy to see that you wouldn’t even be
wondering if two links with different numbers of components are equivalent. It’s
like if we had a bucket system for distinguishing people, and there was one bucket
for mannequins and another for humans. This wouldn’t help us much since the
difference between a person and a mannequin is pretty obvious.
So far we have three invariants—crossing number, tricolorability, and the num-
ber of components—none of which is particularly helpful. Each one is either hard
to compute and hence not very usable, or it doesn’t distinguish many knots or links
and hence is not very useful. In fact, there is no known invariant that is computable
and distinguishes every pair of inequivalent knots or links. We may have to try sev-
eral invariants before we are able to distinguish a particular pair of knots or links.
Thus, we want to have lots of invariants at our disposal—the more, the better!
3. Positive and Negative Crossings
You are deep in thought trying to imagine an invariant that would be both
easy to compute and useful, when your phone rings. It’s our 1-dimensional friend
3. POSITIVE AND NEGATIVE CROSSINGS 243
A. Dash who loves talking to higher-dimensional beings because they seem so com-
plicated and mysterious. “Remember how I like to use arrows to keep myself from
backtracking on my afternoon walks?” she asks. “I have a feeling arrows are useful
for other things, too. Maybe you can use them to come up with a knot invariant?”
A. Dash’s enthusiasm is infectious, but she doesn’t have any concrete plans for what
to do. Rather, she asserts that with all of your complicated 3-dimensional thoughts
you should be able to figure that out yourself.
It’s understandable that A. Dash is not sure how to proceed. After all, she
can’t even visualize a knot or link. How could she possibly distinguish a pair of
knots or links? But we might have more luck. First let’s notice in Figure 3 that
if there are no arrows on a projection, then every crossing looks like every other
crossing if you just rotate it properly. However, we will see below that if we orient
our knot or link, all the crossings in a projection can be divided into two different
types which we can use our hands to distinguish.
Figure 3. If we rotate the crossing on the left by 90◦ , we obtain

the crossing on the right.
A crossing is said to be positive or right-handed if when you put your right

thumb on the over strand so that it points in the direction of the arrow, then the
rest of the fingers on your right hand point under the strand in the direction of the
arrow on the undercrossing (see Figure 4).
Figure 4. A positive (right-handed) crossing.
A crossing is said to be negative or left-handed if when you put your left thumb
on the over strand so that it points in the direction of the arrow, then the rest of
the fingers on your left hand point under the strand in the direction of the arrow
on the undercrossing (see Figure 5).
In contrast with Figure 3, in Figure 6 we see that no matter how we rotate a
positive crossing, we won’t get a negative crossing. A similar thing is true if we
rotate a negative crossing. So even though you may have to rotate a picture (or
your head) to figure out if a crossing is positive or negative, you don’t have to worry
that you might get a different result depending on how you rotate it.
Figure 7 illustrates an oriented projection of the trefoil knot. We can now go
through the crossings one by one, using our hands to determine which ones are
Figure 5. A negative (left-handed) crossing.
Figure 6. A positive crossing is still positive, no matter how it’s rotated.
positive and which ones are negative. In fact, this particular projection has three
negative crossings and no positive crossings (you should check this). What if you
oriented the knot the other way? Would the crossings still be negative or would
they become positive?
Figure 7. This oriented projection has three negative crossings.
When we assign a positive or negative sign to a crossing of an oriented knot or

link, we call it a signed crossing. Signed crossings are going to play an important
role in helping us distinguish knots and links, but it will take us a couple of sections
to get there.
4. Writhe
Now instead of simply counting the total number of crossings of a projection

of a knot or link, we can consider an oriented projection and count the number
of positive crossings and the number of negative crossings separately. We see as
follows that we can combine these two numbers into a single number, which tells
us something about the way the knot is twisting in space.
Definition. The writhe w(L) of an oriented link projection L is given by the
formula

total number of total number of
w(L) = − .
positive crossings negative crossings
5. LINKING NUMBER 245
Since the knot projection in Figure 7 has no positive crossings and three nega-
tive crossings, its writhe is −3. If we reverse the orientation on the knot (changing
the direction of all the arrows) as in Figure 8, the negative crossings remain nega-
tive. Thus the writhe does not change.
Figure 8. The orientation of this projection is opposite that of

Figure 7, but the writhe is the same.
At this point, you might be tempted to rush off and phone A. Dash to tell her
we’ve found the invariant she was hoping for. But we need to slow down. We’ve
made progress on her challenge, but we haven’t figured it out yet. The writhe is
not actually an invariant, even though it is independent of how a knot is oriented.
Consider the two projections of the same knot in Figure 9. One has writhe −3 and
the other has writhe 0. In fact, by adding twists to a projection, we can create a
projection of an equivalent knot with any writhe we want. Thus the writhe depends
on the particular projection and hence is not an invariant. Despite this, we will see
in Section 6 that the writhe can help us distinguish knots or links which are in a
special family.
Figure 9. These two projections of the same knot have different writhes.
5. Linking Number
While the writhe is not an invariant, it turns out that a closely related value
is an invariant of two component links. Now, instead of counting all positive and
negative crossings, we will only count the crossings between the two components of
the link.
Definition. Consider an oriented projection of a link with components K1
and K2 . The linking number of K1 and K2 is defined by the formula
⎡⎛ ⎞ ⎛ ⎞⎤
1
lk(K1 , K2 ) = ⎣⎝ positive crossings ⎠ − ⎝ negative crossings ⎠⎦ .
2 between K1 and K2 between K1 and K2
Figure 10 illustrates two oriented links together with their linking numbers.
Note that we don’t consider all of the crossings of the projection on the right, only
the four crossings between the different components. You might be surprised that
the link on the right has linking number 0, since it’s not the trivial link. However,
the linking number is measuring the number of times that one component wraps
around the other component, not whether or not the link is trivial. In particular,
for the link on the right, the long component goes up and back through the small
component, so it’s not really wrapping around at all.
Figure 10. Some links with their linking numbers.
In order to prove that the linking number is an invariant of oriented links, we

use Reidemeister’s Theorem as we did in Chapter 9 to prove that tricolorability
is an invariant. In particular, we have to show that the linking number will not
change as a result of a planar deformation or doing any of the three Reidemeister
moves. Proving this is one of the Exercises.
While the linking number is an invariant of an oriented link, changing the
orientation of just one of the components might change the linking number. For
example, reversing the orientation of one of the components of the link on the left
in Figure 10 gives us the link in Figure 11, which has linking number 1 rather than
−1.
Figure 11. This link has linking number 1.
Note that reversing the orientation on one of components of a link will change
the signs of all of the crossings between the two components, but will not otherwise
affect the linking number. This means that the absolute value of the linking number
will be the same regardless of the orientation on the link. Thus we can define the
following invariant of unoriented links.
Definition. Consider an unoriented link with components K1 and K2 . The
unsigned linking number of K1 and K2 , is |lk(K1 , K2 )| for any orientation on
K1 and K2 .
In some sense, the unsigned linking number tells us how much the two compo-
nents are intertwined with each other; the larger the number, the more intertwined.
6. NUGATORY CROSSINGS AND ALTERNATING KNOTS AND LINKS 247
And since the linking number is unchanged by Reidemeister moves and the unsigned
linking number is independent of orientation, the unsigned linking number is an in-
variant of unoriented two component links. For example, since the links in Figure 10
have different unsigned linking numbers, we know they must be distinct.
The unsigned linking number is an invariant for two component links that is
more powerful than tricolorability. Now you can call A. Dash and impress her with
how complicated and mysterious you are to have figured out this new invariant.
6. Nugatory Crossings and Alternating Knots and Links
It has been a few weeks since that evening when you thought you saw your
cousin Buttercup on TV. By a strange coincidence your Mom tells you that your
cousin Buttercup will be in town for a few days. So you invite her over to see for
yourself if she was the woman that you saw on the TV show. As soon as she walks
in, it’s obvious that she was not that woman. But you laugh about it together and
then tell her what you’ve been learning about knot theory. You explain the idea of
invariants, giving her the example of the crossing number of a knot. Buttercup sees
the 15 crossing projection from Figure 2 lying on your desk and asks, “So, what’s
the crossing number of this knot?”
Truth be told, all you know about this knot is that its crossing number is no
more than 15. In order to avoid having to admit this to her, you quickly try to
flip through this chapter to see if you can learn more about the crossing number of
this knot. In fact, you’re in luck. While the crossing number is difficult to compute
for most knots and links, there is a special family of knots and links for which it’s
easy to compute. And the knot from Figure 2 is a member of this family. Before
we describe the family, we need a few more definitions.
Definition. A crossing in a knot or link projection is said to be nugatory if
it can be removed by twisting part of the knot (see Figure 12).
Figure 12. We can twist the parts of the knots in the dotted
circles to remove these crossings.
According to an English dictionary, the word “nugatory” means “worthless,

unimportant, or invalid”. For example, you could use it in a sentence such as “All
of your excuses for not wanting to study math with me are nugatory”. The knot
theory definition of nugatory is not all that different in the sense that a nugatory
crossing contributes nothing to a link. For example, each projection in Figure 12
has a nugatory crossing (indicated by an arrow) that can easily be eliminated by
twisting the part of the knot in the dotted circle. None of the other crossings in
these projections are nugatory.
Definition. A projection of a link is said to be reduced if it contains no

nugatory crossings.
Just because a projection is reduced doesn’t mean you can’t remove more cross-
ings by deforming it in some other way. For example, in Figure 13 we illustrate a
reduced projection of the unknot. See if you can deform it to a planar circle.
Figure 13. An unknot with no nugatory crossings.
Definition. A projection of a knot or link is said to be alternating if consec-

utive crossings on each component alternate between going over and going under.
A knot or link is said to be alternating if it has an alternating projection.
For example, in Figure 14 we illustrate an alternating and a non-alternating
knot.
Figure 14. An alternating and a non-alternating knot.
7. Tait’s Conjectures about Alternating Knots
It turns out that link projections which are both alternating and reduced are
really great to work with. In 1870, P. G. Tait, (who constructed the first knot
tables) made several important assertions, including two about reduced alternat-
ing projections. Since Tait gave intuitive arguments rather than proofs of these
assertions, they came to be known as Tait’s Conjectures. His first conjecture about
alternating knots was finally proven in 1986, more than 100 years after he first
asserted it. While it is still known as “Tait’s Conjecture”, since it is now proven,
we refer to it as a theorem. You might see the proof of this theorem if you take a
more advanced knot theory course.
7. TAIT’S CONJECTURES ABOUT ALTERNATING KNOTS 249
Theorem 10.1 (Tait Conjecture #1). A reduced alternating projection of a

knot or link has the fewest number of crossings of any projection of that knot or
link.
This theorem should certainly help you impress your cousin Buttercup with
your vast knowledge of knot theory. In particular, since the knot projection in
Figure 2 is reduced and alternating, we know by Theorem 10.1 that the number of
crossings is as small as possible. Thus you can confidently tell Buttercup that the
knot on your desk indeed has crossing number 15. Similarly, since the projection
of the figure eight knot on the left side of Figure 14 is reduced and alternating, we
know immediately that it has crossing number 4. This enables us to distinguish
the figure eight knot from the unknot, which of course has crossing number 0.
Note that the crossing number is still hard to compute for knots and links that
don’t have reduced alternating projections, and remains a mysterious invariant in
other ways. In fact, the following remains an unproven conjecture.
Conjecture: For any knots K1 and K2 , c(K1 #K2 ) = c(K1 ) + c(K2 ).
For example, the crossing number of a trefoil knot is 3 and the crossing number
of a figure eight knot is 4. In Figure 15 we illustrate a projection of the connected
sum of a trefoil and a figure eight knot which has seven crossings. Is it possible to
draw it in some other way with only six or even fewer crossings? Can we we apply
Theorem 10.1 in this example?
Figure 15. Is there a projection of this connected sum with fewer

than seven crossings?
Among P. G. Tait’s other ideas was a conjecture about the writhe. We saw
in Section 4 that the writhe is not an invariant of knot or links. In particular,
Figure 9 illustrated how we can use Reidemeister I moves to make a given oriented
knot or link have whatever writhe we want. However, we see in the following
theorem that if we restrict ourselves to reduced alternating oriented projections,
then writhe actually is an invariant. You might see the proof of this theorem in a
more advanced course on knot theory.
Theorem 10.2 (Tait Conjecture #2). All reduced alternating projections of a
given oriented knot or link have the same writhe.
Since the writhe of a knot projection does not depend on its orientation, it
follows from this theorem that if a pair of knots have reduced alternating projections
with different writhes, then the knots are not equivalent.1 For example, both of the
projections in Figure 16 have six crossings and are reduced and alternating. You
can check that the knot on the left has writhe equal to −2, while the knot on the
right has writhe equal to 0. Thus we know that the knots are not equivalent.
1 The situation is slightly more complicated for links, since then the writhe depends on the
orientation of each component.

Figure 16. These knots are not equivalent since they have re-
duced alternating projections with different writhes.
Theorem 10.2 can also help us distinguish links and their mirror images. In the
Exercises for Chapter 9, you showed that the figure eight knot and its mirror image
are equivalent. However, many knots are not equivalent to their mirror images. For
example, we can use Theorem 10.2 to show that the trefoil knot and its mirror image
are distinct. Exercises 22, 23, and 24 will help you further explore the relationship
between alternating knots and links and their mirror images.
8. What Proportion of Knots Are Alternating?
Remember that we originally started discussing alternating knots because your

cousin Buttercup asked you how you knew that a particular projection of a knot
with 15 crossings couldn’t be deformed to have fewer crossings. It’s a good thing
the projection was reduced and alternating. That made things easy since we had
Theorem 10.1 in our back pocket. But what are the chances that if Buttercup
chooses another knot projection at random it would also be alternating?
Let’s consider the knot table from Chapter 9. Working our way from the
unknot up through the eight-crossing knots, we see that every knot on the table is
alternating except for 819 , 820 , and 821 . From this information we might conclude
that almost all knots are alternating.
As nice as this conclusion seems, it is not actually true. Table 1 shows what
happens as we look at prime knots with more and more crossings (question marks
indicate quantities that are not currently known). Look at the column labeled
“Proportion alternating”, and observe that this number falls markedly after the
first nine lines. Though the actual proportion of alternating knots is not known for
crossing numbers greater than 16, it was proven in 2009 that this proportion gets
closer and closer to zero as crossing numbers get bigger and bigger.
So you were a bit lucky with Buttercup’s question. If Buttercup randomly chose
another knot with a high crossing number, the likelihood of it being an alternating
knot would be low.
8. WHAT PROPORTION OF KNOTS ARE ALTERNATING? 251
Table 1. The number of prime knots of a given crossing number

and the proportion of them that are alternating.
Crossing Known to be Known to be Total Proportion

number alternating non-alternating number alternating
0 1 0 1 1
3 1 0 1 1
4 1 0 1 1
5 2 0 2 1
6 3 0 3 1
7 7 0 7 1
8 18 3 21 .86
9 41 8 49 .84
10 123 42 165 .75
11 367 185 552 .66
12 1288 888 2176 .59
13 4878 5110 9988 .49
14 19536 27436 46972 .42
15 85263 168030 253293 .34
16 379799 1008906 1388705 .27
17 1769979 ??? ??? ???
18 8400285 ??? ??? ???
19 40619385 ??? ??? ???
20 199631939 ??? ??? ???
21 990623857 ??? ??? ???
22 4976016485 ??? ??? ???
9. Seifert Surfaces
Our next invariant will connect knots and surfaces—two worlds which we have
studied separately until now. First, we need to specify what we will mean by the
word “surface” in this context. Previously, we used the word “surface” interchange-
ably with “2-manifold”. In particular, a surface was not allowed to have a boundary,
because we were studying surfaces as possible 2-dimensional universes. Now that
we will be using surfaces to analyze knots, all of our surfaces will have boundaries.
Definition. We define a Seifert surface for a knot to be an orientable surface

whose boundary is the knot.
For example, a disk is a Seifert surface for the unknot. (In fact, the unknot is
the only knot which has a disk as a Seifert surface.) Figure 17 shows another Seifert
surface for the unknot. This Seifert surface was obtained by attaching a handle to
a disk. Similarly, we can attach two, three, or any number of handles to a disk to
get distinct Seifert surfaces for the unknot. This creates an infinite collection of
Seifert surfaces for the unknot.
Figure 17. A Seifert surface for the unknot.
While it was easy to list infinitely many Seifert surfaces for the unknot, finding
Seifert surfaces for non-trivial knots requires a little more work. For example,
suppose we want to find a Seifert surface for the trefoil knot. The first surface we
think of is the one illustrated in Figure 18. This is a surface whose boundary is the
knot, but it won’t be a Seifert surface unless it’s orientable. Unfortunately, as we
can see in the figure, when A. Square goes for a walk around the surface, he comes
back as his mirror image. A. Square remembers this phenomenon—it happened
when he walked around the Möbius strip in Chapter 4. It means that the surface
is non-orientable, and hence it’s not a Seifert surface.
Let’s try again to find a Seifert surface for a trefoil knot. Recall from Theo-
rem 4.1 in Chapter 4 that since R3 is an orientable 3-manifold, a surface in R3 is
2-sided if and only if it’s orientable. To check whether a surface in R3 is 2-sided, we
can shade it with two colors using the rule that whenever we come to a crossing,
we have to switch colors. We call such a coloring a checkerboard shading because
it reminds us of a checkerboard where squares alternate between white and black.
Notice that we can’t color the projection in Figure 19 with a checkerboard shading
because we’ll be forced to have two adjacent regions that are the same color.
9. SEIFERT SURFACES 253
Figure 18. This surface is non-orientable, and hence it’s not a

Seifert surface.
Figure 19. We can’t color this surface with a checkerboard shading.
In our quest to find an orientable surface with the trefoil knot as its boundary,
let’s start with a different projection of the knot. On the left side of Figure 20, we
illustrate a checkerboard shading of a different projection of the trefoil knot. Notice
that at each crossing we switch colors, and hence this surface has two sides. You
can also check that the surface is orientable by sending A. Square for walks around
it, noticing that no matter what path A. Square follows, when he returns to his
starting point he is not mirror reversed. Thus, the surface on the left in Figure 20
is a Seifert surface for the trefoil knot.
Once we have one Seifert surface for a knot, we can create infinitely many
Seifert surfaces by adding handles to our surface, just as we did for the unknot.
For example, on the right in Figure 20, we illustrate a new Seifert surface that we
get by attaching a handle to the surface on the left. Observe that adding handles
in this way does not change the orientability of the surface.
Remember how A. Dash was convinced that orienting her knot could lead to
a useful invariant? We are in the same situation here with surfaces. The Seifert
surfaces we’ve constructed ought to give us some helpful information about knots.
Figure 20. Two Seifert surfaces for a trefoil knot.
Just like A. Dash, we may not see how we should proceed right away, but what we
learned about surfaces in Chapter 8 will come in handy now.
10. The Genus of a Knot
In Chapter 6, we defined the genus of a finite surface without boundary to

be the maximum number of disjoint circular cuts that can be made in a surface
without disconnecting it. Then Theorem 6.6 told us that the genus of all finite area
surfaces is given by:

genus S 2 = 0, genus nT 2 = n, genus nP 2 = n.
Since for orientable surfaces the genus tells us exactly what the surface is, the
genus seems like it might help us analyze Seifert surfaces for knots. However, in
order to apply the definition of genus to a Seifert surface, we first need to get rid
of the boundary.
Definition. Let S be a Seifert surface for a knot K. We define the genus of
S to be the genus of the surface that we get by abstractly gluing a disk along K.
Keep in mind that this is an abstract gluing, since a non-trivial knot cannot
bound a disk in R3 . Instead we can think of the gluing as instant transport between
the boundary of the Seifert surface and the disk. With this definition of the genus
of a Seifert surface, we are ready to define our invariant.
Definition. The genus of a knot is the smallest genus of any Seifert surface
for the knot.
Notice that by its definition, the genus of a knot is a non-negative integer. Also,
since the genus is defined over all possible Seifert surfaces for any projection of the
knot, then by definition it’s an invariant. However, because each knot has infinitely
many Seifert surfaces, finding the one with the smallest genus seems like it could
be a daunting task. However, once you have one Seifert surface for a knot, you at
least know that the genus of the knot can’t be any bigger than the genus of that
surface. Below, we tackle two examples which are not so daunting.
For our first example, let’s determine the genus of (you guessed it!) the unknot.
As we remarked above, the unknot has a disk as a Seifert surface. In order to find
the genus of this Seifert surface, we have to glue another disk along its boundary.
Recall from Chapter 1 that gluing
two disks together along their boundary results
in a sphere. Now, since genus S 2 = 0 and 0 is the smallest possible value for the
10. THE GENUS OF A KNOT 255
Figure 21. A deformation of our surface to two light grey disks

with three twisted bands joining them.
genus of any surface, we can conclude that the unknot has genus equal to 0. In
fact, the unknot is the only knot which has genus equal to 0, since the unknot is
the only knot which bounds a disk in R3 . Though you might have to wait until you
take an advanced topology course before you see a proof of this seemingly simple
fact.
Next, let’s see if we can determine the genus of a trefoil knot. We begin by
finding the genus of the Seifert surface illustrated on the left in Figure 21. This
Seifert surface looks complicated, but we know that once we glue a disk to the
boundary we will get a finite orientable surface without boundary. Hence, according
to the Classification of Surfaces that we saw in Chapter 6, it is a connected sum of
some number of tori.
Before we glue a disk onto the boundary, we will do some manipulations of
the Seifert surface to make it easier to recognize. We will describe our steps as we
go along and illustrate each with a picture. As you read this, make sure that you
understand how to get from each picture to the next one.
Our first step is to deform the surface on the left in Figure 21 so that it looks
like two disks joined together by three twisted bands as illustrated in the center of
the figure. Next we flip the right disk over, turning it 180◦ along the dotted circle
in the direction of the arrow. This gives us the image on the right. Notice that
this step removes the twist in the bottom band, but adds a twist to each of the
other bands and causes the top and middle bands to switch places on the disk on
the right side.
Since we are trying to determine the genus of the surface rather than its position
in R3 , we only care about its intrinsic topology. Thus we are free to cut the surface
apart, as long as we carefully glue the exact same points back together. In Figure 22
we cut open the two doubly twisted bands from Figure 21, undo the twists, and
then glue them back together. This makes the surface on the right a lot simpler
than the one on the left. Notice that when we reattach the bands in the drawing
on the right, we have to be careful to glue together exactly the same points that
we cut apart. This means that the top two bands still have to criss-cross even after
we untwist them individually.
Next starting with the image on the right in Figure 22, we fatten the bottom
band to get the image at the top left in Figure 23. Then we stretch out the bottom
and curve it up in the back to create a bowl shape with a dark inside. Next we
smooth out the picture so that it looks like a basket with two handles. Finally, we
fatten the lower handle, and slide the top handle so that it goes all the way from
the left side of the basket to the right side of the basket, lifting it so that we only
see its dark side.
Figure 22. We cut the two doubly twisted bands, untwist them,
and then reattach them.
Figure 23. We deform the surface to get a basket with two handles.
Next in Figure 24, we make the lower handle shorter and shorter while making
the top handle wider and wider, to get the picture on the right.
Figure 24. We make the lower handle shorter and the top handle wider.
Next, on the left side of Figure 25, we have pulled up the light grey underside
around the hole. Then in the center of Figure 25, we enlarge the light grey while
decreasing the size of the dark grey, until we obtain a torus with a long rectangular
strip removed from the top. Finally, we shrink the rectangular hole until it becomes
a small round hole. Viola! We have a torus with a disk removed. If we glued the
disk back in, we would obtain a torus, which we know has genus 1. Thus, the genus
of the Seifert surface that we started with in Figure 21 is 1.
It follows that the genus of the trefoil knot is at most one. While you might
think that we still need to consider the remaining infinite number of other Seifert
surfaces for the trefoil, we can actually stop here. Since the unknot is the only
11. USING EULER CHARACTERISTIC TO COMPUTE GENUS 257
Figure 25. We enlarge the light grey while shrinking the dark
grey to get a torus with a disk removed.
knot with genus equal to 0, we know there is no Seifert surface for the trefoil whose
genus is less than 1. So the trefoil knot must have genus exactly equal to 1.
11. Using Euler Characteristic to Compute Genus
If all of the pictures in Figures 21 through 25 overwhelmed you, then you’ll

probably prefer the following technique to compute the genus of a Seifert surface.
We begin by recalling the definition of the Euler characteristic of a surface from
Chapter 6.
Definition. Let S be a surface without boundary which is made up of v ver-
tices, e edges, and f faces, where a face is defined as a shrinkable region. We define
the Euler characteristic of S to be the number χ(S) = v − e + f.
In Chapter 6, A. Square and B. Triangle determined that an orientable surface
with genus n has Euler characteristic χ = 2 − 2n. We will now go backwards and
use the Euler characteristic of a Seifert surface to determine its genus.
To see how this works, let’s start with the Seifert surface on the left in Figure 21.
We divide it into vertices, edges, and faces as shown on the left in Figure 26.
Counting each, we find that there are 12 vertices, 18 edges, and 5 faces. However,
to determine the genus, we need to glue a disk to the boundary of the Seifert
surface. We divide the boundary of this disk into the same number of vertices
and edges that are on the boundary of the Seifert surface. Then we abstractly
glue the new disk onto the boundary of the Seifert surface so that vertices are
glued to vertices and edges are glued to edges. This does not change the total
number of vertices or edges. but it adds one to the total number of faces. This
gives us a total of 6 faces. Since the vertices and edges have already been counted
when we counted those of the Seifert surface, we don’t need to add them into our
totals. Thus, the Euler characteristic of the Seifert surface together with the disk
Figure 26. We divide the Seifert surface into vertices, edges, and
faces, and then add a disk along the boundary.
is χ = v − e + f = 12 − 18 + 6 = 0. Since an orientable surface with genus n has

Euler characteristic χ = 2 − 2n, we set 2 − 2n = 0 and solve to find that the genus
of the Seifert surface is n = 1.
This method is a lot easier than following all of the pictures to see that the
Seifert surface is a torus with a disk removed. On the other hand, manipulating a
Seifert surface to get a surface that we recognize gives us more intuition about the
surface than adding and subtracting numbers. So both methods are useful.
12. A Potpourri of Knot Invariants
So far, we have learned about several knot invariants in some detail. We are
like spectators at a fireworks show where we watch with attention as each of the
fireworks ascends, bursts, and spreads across the sky. Now we close this chapter
with a quick discussion of some additional knot invariants. This is not meant to
be a complete list of invariants or even a detailed explanation of the invariants
described. Rather, the goal of this section is to simply to give you a sense of the
variety of invariants out there, which we hope will motivate you to learn more
knot theory. It’s like the finale of a firework show, where fireworks burst one upon
another in contrasting heights, shapes, and colors. You might even find yourself
spontaneously saying a few oooh’s and aaah’s. Enjoy!
12.1. Unknotting Number. We kick off our “fireworks finale” with an in-

variant which reminds us of the winter holidays. Think beyond your warm fuzzy
memories and recall the yearly fight you have with the strings of holiday lights
stuffed in a box in a closet. Each year you struggle to untangle the mess and ask
yourself, “How many times do I have to pull the whole string of lights through a
tiny loop to untangle this stupid thing?” You wish you could just magically pass
one piece of wire through another to instantly untangle the whole thing.
This age-old holiday untangling problem can be recast in the setting of knot
theory. Suppose you start with a non-trivial knot (similar to your tangled up string
of lights if one end was plugged into the other). You want to magically convert
this knot into an unknot. In one of the Exercises from Chapter 9, you showed that
any knot projection can be turned into a projection of the unknot by changing
some number of crossings from undercrossings to overcrossings and vice versa as
illustrated in Figure 27. Keep in mind that changing crossings is not a deformation.
So being able to change a knot to the unknot in this way does not mean your original
knot was equivalent to the unknot.
Figure 27. This is how we change a crossing.
Definition. The unknotting number of a knot K, denoted u(K), is defined

as the smallest number of crossings that must be changed to obtain the unknot among
all possible projections of K.
12. A POTPOURRI OF KNOT INVARIANTS 259
Since the unknotting number is defined over all possible projections of a knot,
it’s an invariant. We compute it for the trefoil knot as follows. First observe that
because we know that the trefoil knot is truly knotted, every one of its projections
will require at least one crossing change to produce a projection of the unknot.
Recall that we denote the trefoil knot by 31 . Hence, we write u(31 ) ≥ 1. We can
prove that the unknotting number of the trefoil is exactly 1 by finding a projection
in which one crossing change produces a projection of an unknot. In Figure 28, we
change the crossing inside of the small grey circle to go from the left picture (which
is the trefoil knot) to the right picture (which is the unknot). In fact, changing any
crossing in this projection of the trefoil will give us the unknot.
Figure 28. Switching this crossing will change the trefoil into an unknot.
It has been shown that any knot with unknotting number 1 is necessarily prime.
However, there are prime knots whose unknotting numbers are more than 1. For
example, in Figure 29 we make a crossing change on the knot 71 , and then deform
it to get the knot 51 . Then we make a crossing change on 51 , and then deform it
to get the trefoil knot 31 . As we saw above, we can then make a crossing change
on the trefoil to get the unknot. Thus, we can state with certainty that u(71 ) ≤ 3
and u(51 ) ≤ 2. However, perhaps some smaller number of crossing changes will
also do the trick. Perhaps if we start with a different projection of 71 or 51 and do
a single crossing change, we could get the unknot. In fact, the knot 51 really does
have unknotting number 2, and the knot 71 really does have unknotting number 3.
But these facts are not easy to prove.
According to the definition, in order to determine the unknotting number of
a knot, we need to find a projection which requires the fewest possible crossing
changes to get the unknot. Surprisingly, such a projection is not necessarily one
with the minimal number of crossings. For example, the illustration of the knot 108
on the left in Figure 30 has ten crossings, which is the minimal crossing number of
the knot. For this projection, we would need to change three crossings to get the
unknot. However, if we deform the knot to the 14-crossing projection on the right,
then we only need to change two crossings to get the unknot.
Examples like this tell us that just seeing what happens when you change differ-
ent crossings in a minimal crossing projection might not be sufficient to determine
the unknotting number of a knot. Since it’s not possible to consider infinitely many
different projections of a knot to check which one requires the fewest number of
crossing changes, the unknotting number is generally difficult to compute.
Figure 29. We see that u(71 ) ≤ 3 and u(51 ) ≤ 2.
Figure 30. The projection on the left requires more crossing

changes to get the unknot than the one on the right.
12.2. Bridge Number. Your friends love going to amusement parks, but you
do not. Each roller coaster seems to be hurtling you either to an early death or to
losing your lunch. To hide your fright, you take up the habit of counting the number
of peaks on each ride. Other riders wonder why you’re shouting out numbers on
the ride, but you don’t care. Any diversion from these near-death experiences is
welcome! Moreover, you are getting good practice for our next knot invariant, the
so-called bridge number of a knot.
The bridge number is defined as follows. Starting with a knot projection, we
pick a direction which we will call up. As we trace along the projection, we will
sometimes be moving upwards and sometimes downwards, just like when you’re
riding on a roller coaster. We pay no attention to when we’re going forwards and
backwards or side to side. Points at which we change from going upwards to going
downwards—similar to the peaks of the roller coaster—are called bridges. These
look like local maxima of a function. Note that valleys where we change direction
from going downwards to going upwards, are not bridges.
Definition. The bridge number of a knot is defined as the smallest number

of bridges among all possible projections of the knot and all possible choices for the
up direction of a projection.
12. A POTPOURRI OF KNOT INVARIANTS 261
Figure 31. This designation of up makes the figure eight knot

have three bridges.
Since the bridge number is defined over all possible projections, by definition
it’s an invariant. However, like the crossing number and the unknotting number,
the bridge number is difficult to compute.For instance, suppose that we consider the
projection of the figure eight knot shown in Figure 31 using the indicated direction
as up. We highlight the bridges with thick grey arcs. Since this projection and
direction has three bridges, we know that the bridge number of the figure eight
knot is at most 3.
On the other hand, in Figure 32 we rotate our projection in order to designate
a different direction as up. In this new projection, we find that there are only two
bridges. It can be shown that the only knot which has a projection with just one
bridge is the unknot. Since we know that the figure eight knot is not equivalent to
the unknot (because the projection in Figure 32 is reduced and alternating), this
means that the figure eight knot has no projection with fewer than two bridges. In
particular, the projection in Figure 32 has the minimum number of bridges. Thus
the bridge number of the figure eight knot is indeed 2.
Figure 32. This projection of the figure eight knot has two bridges.
12.3. Stick Number. Until now, we have drawn our knots as curves in space.
This certainly makes knot theory visually appealing. But nothing is preventing us
from considering polygonal constructions of knots. In other words, we can construct
knots from straight line segments (called sticks) that are glued at their endpoints
to form a single knotted polygon in 3-dimensional space. Every knot can be con-
structed in this way by using lots of very short sticks.
Definition. We define the stick number of a knot as the smallest number
of sticks needed to construct that particular knot.
Since the stick number is again defined over all possible stick constructions of
a knot, the stick number is another invariant. Note that there is no requirement
that all of the sticks be the same length, and you should not assume that the sticks
have any particular thickness.
The unknot has stick number 3 since it is impossible to make a circle with fewer
than three sticks. In Figure 33, we illustrate a stick construction of the trefoil knot
with six sticks and a stick construction of the figure eight knot with seven sticks.
Figure 33. Stick constructions of the trefoil and the figure eight knot.
In order to prove that the trefoil has stick number 6 and the figure eight has
stick number 7, we would have to show that each of the following statements holds:
(1) Every 3-stick, 4-stick, and 5-stick configuration represents an unknot.
(2) The illustration on the left side of Figure 33 is a 6-stick configuration of
the trefoil knot.
(3) The illustration on the right side of Figure 33 is a 7-stick configuration
of the figure eight knot, but there is no 6-stick configuration of the figure
eight knot.
The difficulty involved in proving each of these steps is exacerbated by the fact
that we need to consider polygonal configurations in 3-dimensional space and cannot
rely entirely on 2-dimensional projections. For instance, even though Figure 34
looks like a 5-stick projection of a trefoil knot and an 8-stick projection of the knot
818 , this is an optical illusion. In fact, it’s impossible to make these constructions
with real sticks in 3-dimensional space. In both of these examples, the choice of
over- and under-crossings in the projections will force one or more of the sticks
to bend. You should try making these configurations yourself using matchsticks,
toothpicks, or popsicle sticks to see what happens.
Figure 34. Some configurations that are impossible to construct

with real sticks.
12.4. Lattice Number. The lattice number is similar to the stick number,
but now in addition to using sticks to build our knots, the sticks are not allowed to
go any which way. Rather, we require the sticks to be of equal length and parallel
to the x, y, or z-axes in R3 . Another way of thinking about this, is to imagine
that all of R3 is filled with a grid like you would see on graph paper but in three
dimensions instead of two dimensions. We can only create knots out of edges that
go along the 3-dimensional grid, as illustrated in Figure 35. We call this a lattice
configuration of a knot.
13. EXERCISES 263
Figure 35. A lattice configuration of a trefoil knot with 24 vertices.
Definition. The lattice number of a knot is the smallest number of vertices

in any lattice configuration of that knot.
Again we know this is an invariant because it’s defined over all possible lattice
configurations. Perhaps it is surprising to realize just how large lattice numbers
tend to be. In particular, Figure 35 shows a lattice configuration of the trefoil knot
which has a minimum number of vertices. If you count the number of vertices in
this configuration, you will find that there are in fact 24.
Looking back over this chapter, we’ve done all kinds of things to knots and
links to help us distinguish them. We colored them, we built them out of sticks,
we put arrows on them, we counted their peaks, and we related them to surfaces.
Each of these ideas led us to an invariant. Nevertheless, a common theme in this
journey has been that the most useful invariants are often hard to compute. We
will spend the next chapter developing a single invariant which is both useful and
not too difficult to compute.
13. Exercises
1. Prove that the writhe is preserved under Reidemeister moves II and III, but
not Reidemeister move I.
2. Prove that the writhe of a knot projection is independent of the choice of
orientation.
3. Draw three different projections of the knot 52 : one with writhe −5, one with
writhe −4, and one with writhe −6.
4. Prove that the linking number of an oriented link projection will not change as
a result of a planar deformation or doing any of the three Reidemeister moves.
Then use Reidemeister’s Theorem to conclude that the linking number is an
invariant. Hint: You may want to look back in Chapter 9 to see how we proved
that tricolorability would not change as a result of the first two Reidemeister
moves.
5. (Hard.) Prove that the linking number is always an integer. In other words,
prove that
⎛ ⎞ ⎛ ⎞
⎝ positive crossings ⎠ − ⎝ negative crossings ⎠
between K1 and K2 between K1 and K2
is always an even number.

6. For any n, prove that there are infinitely many non-equivalent oriented two-
component links with linking number n, each of whose components is unknot-
ted.
7. Let L denote an oriented link with components J and K. Let −J and −K
denote J and K with their orientations reversed, and let L∗ denote the mirror
image of L with oriented components J ∗ and K ∗ . Suppose that lk(J, K) = n.
Find lk(−J, K), lk(J, −K), lk(−J, −K), and lk(J ∗ , K ∗ ). Justify your conclu-
sions.
8. Suppose that an oriented link with components J and K can be deformed to
its mirror image while preserving the orientation on each component. Prove
that lk(J, K) = 0.
9. Find an unoriented link with components J and K which can be deformed to
its mirror image, but regardless of how we orient the link, lk(J, K) = 0.
10. Find the unsigned linking number of each of the links in Figure 36. Can you
use the linking number to determine if some of the links are non-equivalent?
11. The three links in Figure 37 are all different. Can you prove it? Try to use two
different invariants in your analysis.
12. The unsigned linking number is an invariant of two component links. In Fig-
ure 38 you’ll find two links, each with three components. How might the un-
signed linking number help you show that these links are non-equivalent? Hint:
Every three component link contains several two component links.
13. EXERCISES 265
13. The left link in Figure 38 is special because if you delete any component, the link
falls apart. In particular, the linking number between any pair of components
is 0. This link is considered to be very beautiful and has even appeared in
various works of art and architecture. More generally, we have the following
definition.
Definition. A link with n components, L1 , . . . , Ln , is said to be Brunnian
if the link is non-trivial, but removing any one component gives the trivial link .
For every n > 2, find a Brunnian link with n components.

14. How many non-equivalent two component links are there having a projection
with precisely two crossings? Justify your conclusion.
15. The projection of 63 in the knot table is alternating. Find another projection
of 63 that is non-alternating. Can you find a non-alternating projection of 63
which has no nugatory crossings?
16. The projection in Figure 39 is not reduced. Find another projection of this
knot that is reduced.
17. Find the crossing number of the knot in Figure 40. Carefully explain how you
arrived at your answer and include pictures.
18. Find the crossing number of the knot in Figure 41. Carefully explain how you
arrived at your answer and include pictures.
19. Let K1 and K2 be knots. Prove that if K1 and K2 are alternating, then their
crossing numbers satisfy the equation c(K1 #K2 ) = c(K1 ) + c(K2 ). Hint: First
show that since K1 and K2 are alternating, the connected sum K1 #K2 can be
deformed to an alternating projection.
20. Find the writhes of all the prime 7-crossing knot projections given in the knot
table. Using this information, which knots can you conclude must be distinct
from one another?
21. Compute the writhes of the projections of 820 and 821 given in the table. Why
can’t we use the writhe to conclude that 820 and 821 must be different knots?
22. Let L denote an oriented knot or link projection, and let L∗ denote the mirror
image of L. How is w(L∗ ) related to w(L)?
23. Let L be a reduced alternating oriented knot or link projection, and let L∗ be
its mirror image.
(a) If L and L∗ are equivalent, what does Tait’s Second Conjecture say about
their writhes?
(b) Combining your results from part (a) and Exercise 22, what must be true
of w(L)?
(c) Use your observation to prove that the knot 51 cannot be deformed into
its mirror image. List all prime knots with up to six crossings that cannot
be equivalent to their mirror images.
24. Let L be an alternating link with odd crossing number. Show that L cannot
be deformed into its mirror image.
25. Show that any knot projection can be changed into an alternating projection
(of a possibly different knot) by switching some number of crossings. If the
projection you started with had n crossings, what is the largest number of
crossings you would have to switch to make it alternating? Explain.
26. In Section 9 we saw that once you have one Seifert surface for a knot, you can
create infinitely many Seifert surfaces for the knot by adding handles. Is there
a way to add a handle to an orientable surface to make it non-orinetable?
27. Determine the unknotting number of the figure eight knot.
28. Determine the bridge number of the trefoil knot.
29. Show that there are infinitely many distinct knots with bridge number 2.
30. Show that if a knot projection has only one bridge, then it must be a projection
of the unknot. (Hint: Draw the one bridge, and consider how the two strands
that come down from it interact with one another.)
31. Show that every 4-stick knot is the unknot.
13. EXERCISES 267
32. Create three non-trivial links with different linking numbers such that in each
case one component has three sticks and the other component has four sticks.
33. Show that it is impossible to construct the 5-stick trefoil knot shown in Fig-
ure 34(a).
34. For each prime 7-crossing knot projection in the knot table, use shading to find
a surface whose boundary is the knot. Which of your surfaces are orientable
and which are non-orientable? Which surfaces are Seifert surfaces?
35. Calculate the genus of the knot 52 using the projection given in the knot table.
36. Calculate the genus of the knot 72 using the projection given in the knot table.
37. Complete the outline below to prove that every knot has a lattice configuration.
(a) Suppose that we start with a knot projection with n crossings. Show that,
after an appropriate planar deformation, we can find a collection of hori-
zontal lines H0 , H1 , . . . , Hn and a collection of vertical lines V0 , V1 , . . . , Vn
such that:
(i) H0 lies below the knot projection and does not intersect the projec-
tion;
(ii) Hn lies above the knot projection and does not intersect the projec-
tion;
(iii) Hi+1 lies above Hi , and there is exactly one crossing of the projection
lying between them;
(iv) V0 lies to the left of the knot projection and does not intersect the
projection;
(v) Vn lies to the right of the knot projection and does not intersect the
projection; and
(vi) Vj+1 lies the the right of Vj , and there is exactly one crossing of the
projection lying between them.
(b) Each pair of consecutive horizontal lines Hi and Hi+1 together with each
pair of consecutive vertical lines Vj and Vj+1 defines a rectangle contain-
ing at most one crossing of the projection. Show that the arcs of the
projection inside this rectangle can be deformed so that they are made up
of only vertical and horizontal line segments. Note that the entire defor-
mation should be accomplished without moving the endpoints of the arcs
(where they meet Hi , Hi+1 , Vj , or Vj+1 ) and without introducing any new
crossings.
(c) Show that the projection can be deformed so that every horizontal line
segment lies along the line y = k for a different integer value of k. Also,
show that the projection can be deformed so that every vertical line seg-
ment lies along the line x = m for a different integer value of m. Show
why this means that after these deformations every line segment in the
projection consists of some union of unit-length edges parallel to either
the x- or y-axis.
(d) Complete the lattice configuration of the knot by adding unit-length edges
parallel to the z-axis.
CHAPTER 11
Knot Polynomials
Topics:
• The bracket polynomial
• The X-polynomial
• The Jones polynomial
• The span of a Jones polynomial
• Reading the Jones polynomial from a knot table
• The state model of the Jones polynomial
1. An Introduction to Polynomial Invariants
Let’s say you cross paths with a shady character under a bridge and he holds
you at knifepoint, demanding that you compute the crossing number of a non-
alternating knot. (More likely, this is a nightmare after a late night of knot study-
ing.) Under the circumstances, you might be better off begging him to take your
money instead. Even if, instead of a crossing number, the shady character de-
manded that you compute another invariant from Chapter 10, you’d likely fare no
better. Most of the knot invariants we’ve learned about so far can be pretty tough
to calculate. That is, except for tricolorability, which doesn’t distinguish many
knots.
In this chapter, we will finally meet a useful invariant that you can compute for
any knot simply by following a list of rules. Your success is guaranteed (though your
speed may not be). We hope you never find yourself computing knot invariants for
shady characters under bridges, but if you do, we advise that you offer to compute
the invariant we’ll discuss here.
This new invariant is not a numerical value, like most of the invariants we saw in
Chapter 10. It is a special type of polynomial. By arranging the terms in order from
lowest exponent to highest exponent and listing the coefficients, a polynomial can
be represented by a finite list of numbers. For example, the polynomial 1 − 4x + 3x3
could be expressed as the list (1, −4, 0, 3). Note that the third entry is 0 because
there is no x2 term in the polynomial. Since a list of numbers can encode more
information than a single number, a polynomial invariant seems like it might be
more useful than an invariant consisting of a single number. In fact, this new
invariant is more useful in distinguishing knots than any of our previous invariants.
One last note before we get started. The invariant that we will introduce in
this chapter is not technically a polynomial because it can have negative as well as
269
270 11. KNOT POLYNOMIALS
positive integer exponents. The technical name for such an expression is a Laurent
polynomial. But knot theorists simply call it a polynomial, so we’ll do the same.
2. The Rules for the Bracket Polynomial
Our goal in this chapter is to define a polynomial invariant of knots and links.
However, as the first step we define the bracket polynomial of a link projection,
which is not actually an invariant. Then in Section 6, we will use the bracket
polynomial to create the X-polynomial, which will be an invariant. Defining the
bracket polynomial requires introducing three rules which we will explain as we go
along.
2.1. Rule 1. The bracket polynomial

L will be defined for any link projec-
tion L, but we’ll start out slowly and just define it on a projection of a circle with
no crossings at all.
It seems natural that the simplest possible projection should be assigned a very
simple polynomial. Indeed, we begin by defining:

Rule 1 : = 1.
2.2. Rule 2. For a link projection L which has one or more crossings, we want
to remove the crossings one at a time until we end up with a link with no crossings
so that we can apply Rule 1. We start by choosing one crossing and cutting the
overcrossing into two arcs as shown in Figure 1. This gives us four endpoints of
arcs where the crossing had been.
Figure 1. Cutting a crossing.
We then want to reconnect the dangling arcs in the center of the picture without
reintroducing a crossing. We do this by attaching each endpoint of the former
overcrossing (the thick arc in the figure) to an endpoint of the former undercrossing
(the thin arc in the figure). We can do this in two ways, depending on whether we go
in the clockwise or counterclockwise direction from the endpoint of an overcrossing
(illustrated in Figure 2). The dotted lines are just there to help you see the clockwise
or counterclockwise direction.
Let Lclock be the link projection produced from a projection of L by cutting
an overcrossing and reconnecting the ends clockwise, and let Lcounter be the link
projection produced from L by cutting the same overcrossing and reconnecting the
ends counterclockwise. Notice that the projections Lclock and Lcounter each contain
one fewer crossing than L.
2. THE RULES FOR THE BRACKET POLYNOMIAL 271
Figure 2. Re-attaching the ends after cutting the overcrossing.
We now introduce variables A and B and an equation that uses these variables
to relate the bracket polynomial of L to the bracket polynomials of the simpler link
projections Lclock and Lcounter :

L = A
Lclock + B
Lcounter .
This equation is saying that the bracket polynomial of L is the sum of A times
the bracket polynomial of Lclock and B times the bracket polynomial of Lcounter . It
may seem strange that the variables are A and B rather than x and y. The capital
letters A and B have always been used for knot and link polynomials, so we stick
with them despite their strangeness. The above equation is Rule 2, but we prefer
to express it symbolically as

Rule 2 : =A +B .
The little dotted circles inside of the bracket symbols mean that the projections
of the associated links (L, Lclock , and Lcounter ) are all identical outside of these
circles.
Figure 3 shows how to apply Rule 2 to one of the crossings in a trefoil knot.
Keep in mind that when we connect the ends of the former overcrossing clockwise,
we multiply by A, and when we connect the ends counterclockwise, we multiply by
B. This is true no matter how the crossing is turned in the projection.

=A +B =A +B
Figure 3. Resolving one crossing of the trefoil knot.
Applying Rule 2 to get rid of a crossing is called resolving the crossing. By

repeatedly resolving crossings, we will break our projection into a collection of link
projections with fewer and fewer crossings, which will eventually become a collection
of circles with no crossings.
2.3. Rule 3. Our third (and final) rule says that we can remove circles with
no crossings, if we multiply by yet another variable, which we will call C.

Rule 3 : L = C
L .
For example, suppose that we start with a link projection L consisting of two
circles in a plane with no crossings at all. By applying Rule 3 followed by Rule 1,
we see below that the bracket polynomial of L is just the variable C.

=C = C(1) = C.
3. The Bracket Polynomial and Reidemeister II Moves
At this point, we have rules that can be applied repeatedly to a knot or link
projection to get rid of crossings and extra circles one after the other until we
eventually get a polynomial without any bracket symbols in it. But a big question
remains—is this polynomial a knot or link invariant? If it’s not an invariant, why
would we want to spend the time applying the rules over and over to get some com-
plicated expression with variables A, B, and C which wouldn’t help us distinguish
one knot from another?
To be an invariant, the bracket polynomial needs to be unchanged under the
three Reidemeister moves. In this section, we will see what happens when we apply
Reidemister II moves. We will consider Reidemeister I moves later, and you will
deal with Reidemeister III moves in the Exercises.
Below we illustrate what happens when we resolve the crossings that occur in
a Reidemeister II move. Make sure you can follow the steps as we describe them
line by line.

=A +B

=A A +B +B A +B

=A A + BC

+B A +B

2 2
=A + ABC + AB +B

= A2 + ABC + B 2 + AB .
3. THE BRACKET POLYNOMIAL AND REIDEMEISTER II MOVES 273
• Line 1: The left picture illustrates what the projection looks like near
the place where we will do a Reidemeister II move. We apply Rule 2 to
get rid of the top crossing. This means we multiply by A when we connect
clockwise and we multiply by B when we connect counterclockwise.
• Line 2: We apply Rule 2 to get rid of the bottom crossing. This means
that each expression on Line 1 is broken into two expressions, giving us a
total of four terms.
• Line 3: We deform the drawings so they look nicer and we use Rule 3 to
eliminate the little circle in the second term by multiplying by the variable
C.
• Line 4: We multiply the variables and write the products of variables in
alphabetical order.
• Line 5: Wecollecttogether the three terms that are being multiplied by
the symbol and put them inside of parentheses.
In order for the bracket polynomial to be invariant under Reidemeister II moves,

the polynomial needs to be the same before and after we do a Reidemeister II move.
Thus we need to know that

= .
By the above computation, we know that

2
= A + ABC + B 2 + AB .

We set this equal to to get the equation

2 2

A + ABC + B + AB =

Since cannot turn into , we can split this into two separate equa-
tions:
2 2

A + ABC + B =0
and
AB = .
This gives us the equations:

A2 + ABC + B 2 = 0 and AB = 1.
−1
The second equation implies that B = A . Hence we can substitute A−1 for
B in the first equation to get the equation
A2 + AA−1 C + A−2 = 0.
Now we can solve this equation for the variable C to get

C = −A2 − A−2 .
These computations show that if the bracket polynomial is going to be invariant
under Reidemeister II moves (which we want it to be), then the equations B = A−1
and C = −A2 − A−2 must both be true. To check that, with the substitutions
B = A−1 and C = −A2 − A−2 , the bracket polynomial will be invariant under
Reidemeister II moves, we plug these values for B and C into our computation
of the bracket polynomial for the crossings in a Reidemeister II move, and then
simplify to get the following:

=A + A−1

=A A + A−1

−1 −1
+A A +A

=A A + A−1 (−A2 − A−2 )

−1 −1
+A A +A

−1 −2
=A 2
+ AA (−A − A2
)

−1 −1 2
+ AA + (A )

= A2 + AA−1 (−A2 − A−2 ) + (A−1 )2 + AA −1

−2 −2
= (A − A − A
2 2
+A ) +

= .
This shows that our variable substitutions are exactly what we need in order
to guarantee that the bracket polynomial will be invariant under Reidemeister II
moves.
4. The Bracket Polynomial with Only One Variable

We now plug the expressions B = A−1 and C = −A2 − A−2 into Rules 2 and 3
of the bracket polynomial to get the definition of the bracket polynomial with only
one variable.
5. THE BRACKET POLYNOMIAL AND REIDEMEISTER I MOVES 275
Definition. The bracket polynomial

L of a link projection L is a Laurent
polynomial with variable A, defined by three rules:

(1) = 1.

−1
(2) =A +A .

(3) L = (−A2 − A−2 )
L.
We know from Section 3 that the bracket polynomial is invariant under Reide-
meister II moves, and you will prove in the Exercises that it’s also invariant under
Reidemeister III moves. Unfortunately, as we will soon see, this polynomial is not
invariant under Reidemeister I moves. Don’t despair, however. This problem will
eventually be fixed. But in the meantime, let’s practice using the rules to compute
bracket polynomials of a couple of link projections.
We previously calculated the bracket polynomial of a projection L consisting
of two planar circles and found that
L = C. Now that we only have the variable
A in our polynomial, we recalculate this polynomial using Rules 1 and 3 as follows:

= (−A2 − A−2 ) = (−A2 − A−2 )(1) = −A2 − A−2 .
Next, we compute the bracket polynomial of a projection of a Hopf link. Make

sure you know how to justify each step in the computation below.

=A + A−1

=A A + A−1

+ A−1 A + A−1
= A2 (−A2 − A−2 ) + 1 + 1 + A−2 (−A2 − A−2 )
= −A4 − 1 + 2 − 1 − A−4
= −A4 − A−4 .
5. The Bracket Polynomial and Reidemeister I Moves
In order to see what goes wrong with the bracket polynomial when we do
Reidemeister I moves, we compute the bracket polynomial of a projection of an
unknot with a single crossing. Keep in mind that the bracket polynomial is defined
on a projection of a link, not on the link itself. So we can’t just assume that every
projection of the unknot has bracket polynomial equal to 1.
Using the rules we have

= A + A−1
= A(−A2 − A−2 ) + A−1
= −A3 .
The projection differs from a circle without crossings by a single
Reidemeister I move. Thus we see that doing a Reidemeister I move to a circle
without any crossings changes the bracket polynomial from 1 to −A3 . This means
that the bracket polynomial is not invariant under Reidemeister I moves, and hence
equivalent knots do not necessarily have the same bracket polynomials. Thus we
need to be very careful to only talk about the bracket polynomial of a knot or link
projection rather than of a knot or link itself.
Recall from Chapter 10 that we defined an oriented crossing to be positive if
when you point the thumb of your right hand in the direction of the overcrossing,
you fingers curl in the direction of the under crossing. We illustrate a negative and
positive crossing in Figure 4.
Figure 4. A negative and positive crossing.
Using the designation of positive and negative crossings, we can designate a

twist as positive or negative. In particular, we refer to the twist in Figure 5 as a
“positive” twist because regardless of how we orient it, the crossing is positive. You
should use your right hand to check for yourself that both crossings illustrated in
Figure 5 are indeed positive.
Figure 5. A positive twist.
By contrast, the two twists illustrated in Figure 6 are negative.
Figure 6. A negative twist.

6. THE X-POLYNOMIAL 277
We saw in the beginning of this section that adding a positive twist to a circle
with no crossings changes the bracket polynomial from 1 to −A3 . We now see as
follows that starting with any knot or link projection L and adding a positive twist
to get a new link projection L multiplies the bracket polynomial by a factor of
−A3 :

L = = A + A−1
= A
L (−A2 − A−2 ) + A−1
L
=
L (−A3 − A−1 + A−1 ) =
L (−A3 ).
You will show in the Exercises that starting with a knot or link projection L
and adding a negative twist to get a new link projection L multiplies the bracket
polynomial by a factor of −A−3 .
6. The X-Polynomial
We saw in the last section that adding a twist to a link projection multiplies the
bracket polynomial by −A3 or −A−3 . Sadly, this means that the bracket polynomial
is not invariant under Reidemeister I moves. We will solve this problem by making
use of the writhe that we introduced in Chapter 10. Recall the following definition.
Definition. The writhe w(L) of an oriented link projection L is given by the
formula
w(L) = − .
In Chapter 10, we saw that the writhe isn’t an invariant. In fact, we could
make a knot have any writhe that we want by adding some number of positive or
negative twists. What’s amazing is that we will now use the writhe (which isn’t
an invariant) and the bracket polynomial (which isn’t an invariant) to define a new
polynomial which will be a link invariant. Sometimes two wrongs actually do make
a right! But don’t tell your parents.
Definition. Let L be an oriented link projection with writhe w(L). We define
the X-polynomial of L to be X(L) = (−A3 )−w(L)
L.
We did not need a knot or link projection to be oriented in order to compute
the bracket polynomial. However, since the writhe is only defined for oriented
projections, X(L) is only defined if L is an oriented projection. Recall that for a
knot projection, both orientations give the same writhe, so which orientation we
assign won’t matter for the X-polynomial. But this is not true for a link projection.
As an example, we compute the X-polynomial for the oriented Hopf link pro-
jection L illustrated in Figure 7.
Figure 7. An oriented Hopf link projection.

Recall that in Section 4 we computed the bracket polynomial of this (unori-

ented) link projection to be
L = −A4 − A−4 . Also you should check Figure 7
to see that w(L) = −2. Thus we can compute the X-polynomial of this oriented
projection to be
X(L) = (−A3 )−(−2) (−A4 − A−4 )
= A6 (−A4 − A−4 )
= −A10 − A2 .
We are now ready to prove the marvelous, magnificent, miraculous result that
the X-polynomial is indeed an invariant of oriented links. This is so awesome that
we want to state it as a theorem so we can sit and admire it before we prove it.
Theorem 11.1. The X-polynomial is an invariant of oriented links.

Proof. Let L be an oriented link projection. If we can show that X(L) is
preserved by planar deformations and the three Reidemeister moves, then it will
follow from Reidemeister’s Theorem that X(L) is indeed an invariant. Since the
writhe and the bracket polynomial are both defined in terms of crossings, planar
deformations will have no effect on them. Thus the X-polynomial will also be
invariant under planar deformations. So we can move on to the Reidemeister moves.
We saw that the bracket polynomial
L is preserved by Reidemeister II moves,
and you will verify in the Exercises that
L is also preserved by Reidemeister III
moves. In the Exercises for Chapter 10, you proved that the writhe is also preserved
by Reidemeister II and III moves. Since X(L) is the product of two quantities that
are each invariant under Reidemeister II and III moves, it follows that X(L) is also
invariant under Reidemeister II and III moves. At this point, all that is left to do
is to prove that X(L) is invariant under Reidemeister I moves.
Let L denote L with a positive twist added. This means that L has exactly
one more positive crossing than L. Hence w(L ) = w(L) + 1. Also, we saw in
Section 5 that
L =
L(−A3 ). So we have

X(L ) = (−A3 )−w(L )
L
= (−A3 )−w(L)−1
L(−A3 )
= (−A3 )−w(L) (−A3 )−1 (−A3 )
L
= (−A3 )−w(L)
L
= X(L).
Thus, the X-polynomial is invariant under adding or removing a positive twist.
As you will check in the Exercises, by an analogous argument, the X-polynomial is
invariant under adding or removing a negative twist.
Therefore, the X-polynomial is invariant under planar deformation and all three
Reidemeister moves, and hence it is an invariant of oriented links.
Time for a celebration. At long last, we have defined the X-polynomial and
proven that it is indeed an invariant. You are now armed with a knot invariant
which you can compute by yourself or with a friend, for any oriented knot or link
(even in dark alleys or under bridges if necessary). Since the X polynomial is an
invariant, we can now talk about the X-polynomial of an oriented link instead of
always having to say the X-polynomial of an oriented link projection. Nonetheless,
7. THE JONES POLYNOMIAL 279
we have to fix a projection in order to compute the bracket polynomial as a step

in the computation of the X-polynomial. However, the particular projection we
choose will not affect the value of the X-polynomial.
In order to see what might happen to the X-polynomial when the orientation
of a single component of a link is reversed, let’s consider the oriented Hopf link
projection L in Figure 7, and let L denote the oriented Hopf link projection in
Figure 8. These two oriented projections differ only in the orientation of one of
their components. Since L and L are the same if we ignore the orientations,
L =
L . However, w(L) = −2 and w(L ) = 2. Thus
X(L) = (−A3 )−2
L
= A−6
L
and
X(L ) = (−A3 )2
L
= (−A3 )2
L
= A6
L.
Thus the X-polynomial can be affected by changing the orientation of one
component. In the Exercises, you will explore what happens when the orientations
on all components of a link are reversed.
Figure 8. A Hopf link whose orientation is different from that of Figure 7.
7. The Jones Polynomial
The X-polynomial was originally discovered by Vaughan Jones in a different

form known as the Jones polynomial, which is denoted by V (L) for an oriented
knot or link L. In spite of being called a “polynomial”, it’s not actually a poly-
nomial or even a Laurent polynomial like the X-polynomial, because its exponents
can be fractions as well as integers. The Jones polynomial and the X-polynomial
represent the same information about a link though the rules for computing them
are somewhat different. In fact, you can get from one to the other by an easy sub-
stitution. To change the X-polynomial X(L) (whose variable is the letter A) into
the Jones polynomial V (L) (whose variable is the letter q), we use the substitution
A = q −1/4 . We can also go from the Jones polynomial to the X-polynomial by
using the substitution q = A−4 . For example, since the we know the X-polynomial
of the oriented Hopf link in Figure 7 is
X(L) = −A10 − A2 ,
we find that the Jones polynomial of this oriented Hopf link is
V (L) = −q −5/2 − q −1/2 .
In your previous math classes, the variable of a function was probably written
inside of the parentheses after the name of the function. For example, you’ve
probably worked with the function f (x) = x2 . By contrast, even though knot and
link polynomials have variables like A and q, we write the knot or link itself inside
of the parentheses next to the name of the function rather than the variable. In
other words, if L is an oriented knot or link, we write X(L) for the X-polynomial
of L and V (L) for the Jones polynomial of L, rather than writing X(A) and V (q).
The reason for this distinction is that a function like f (x) = x2 takes in a real
number x and spits out a new real number x2 , whereas a knot and link polynomial
takes in an oriented knot or link and spits out a polynomial rather than a single
real number. Of course once you have found the polynomial of a particular knot,
you could plug in numbers for the variable and get numbers out. But we’re not
interested in the specific numbers you would get out—we’re interested in comparing
the polynomials of different oriented knots or links in order to know that one can’t
be deformed to the other.
The Jones polynomial is a powerful tool in knot theory because it’s not hard to
compute and it distinguishes a lot of knots. For example, we know that the knots
51 and 52 are distinct because they have different polynomials. The same is true
for the non-alternating knots 819 and 820 , illustrated in Figure 9. Note that, since
the Jones polynomial of a knot is independent of the orientation of the knot, we
are free to omit the orientation, as we’ve done in Figure 9.
Figure 9. We know that the knots 819 and 820 are distinct be-
cause they have different polynomials.
You may wonder whether the X-polynomial (or, equivalently, the Jones poly-
nomial) distinguishes every pair of inequivalent links. Unfortunately, this is not the
case. For example, the two knots illustrated in Figure 10 have the same Jones poly-
nomial but are distinct. If you look carefully, you’ll see that they look quite similar
but are not identical. Of course this doesn’t prove that they aren’t equivalent. In
fact, knot theorists used the genus to distinguish them by showing that the one on
the left has genus 3, while the one on the right has genus 2.
We might at least hope that the Jones polynomial could distinguish all non-
trivial knots from the unknot. This may be possible. As of 2015, no non-trivial
knot with Jones polynomial equal to 1 has been found. But who knows? Maybe
there is some complicated knot with Jones polynomial equal to 1 somewhere out
there just waiting to be discovered.
In spite of the fact that Jones polynomials can in theory have fractional powers,
Jones polynomials of knots actually only have integer exponents. By contrast, links
with at least two components have Jones polynomials whose exponents are multiples
of 1/2.
7. THE JONES POLYNOMIAL 281
Figure 10. These knots have the same Jones polynomial but are distinct.
Many knot tables include Jones polynomials compactly by displaying the low-
est exponent in the polynomial together with a list of numbers representing the
coefficients of all of the terms. For example, next to the trefoil knot in Table 1
is the code {−4} − 1 1 0 1. The {−4} indicates that −4 is the lowest exponent
in the Jones polynomial of the trefoil, and the list of numbers that follow are the
coefficients of the variables of each subsequent exponent. So −1 is the coefficient
of q −4 , 1 is the coefficient of q −3 , 0 is the coefficient of q −2 , 1 is the coefficient of
q −1 , and that’s it. This means that the Jones polynomial of the trefoil knot is
V (31 ) = −q −4 + q −3 + q −1 .
Note that since the coefficient of q −2 is 0, we omit the q −2 term from the polynomial.
Now, by substituting q = A−4 into the Jones polynomial, we find that the
X-polynomial of the trefoil knot is
X(31 ) = −A16 + A12 + A4 .
Since Jones polynomials have been calculated (with the help of computers) for knots
with up to 11 crossings, it is now easy to find the X-polynomials of these knots by
simply doing a substitution of q = A−4 , as we did for the trefoil knot above.
If you look at Table 1, you will see that the lists of coefficients in the table get
longer as the crossing number increases—until we reach the non-alternating knots
819 , 820 , and 821 at the end of our table. This gives us the impression that the
more crossings a knot has, the more terms are in its Jones polynomial.
To help keep track of the number of terms in the Jones polynomial (including
terms with coefficient 0), we define the span of a Jones polynomial V (L), written as
span(V (L)), to be the difference between the highest power and the lowest power of
the variable q in V (L). For example, for the trefoil knot, since the Jones polynomial
is V (31 ) = −q −4 + q −3 + q −1 , we have span(V (31 )) = −1 − (−4) = 3.
It turns out that for any alternating knot K, the span of its Jones polynomial
is equal to its crossing number. We could express this symbolically by saying that
for alternating knots, we have span(V (K)) = c(K). In Chapter 10, we learned
about two of Tait’s Conjectures which took more than 100 years to be proved. The
first of these conjectures said that a reduced alternating projection of a knot or link
has the smallest number of crossings of any projection. This important result was
proved using the span of the Jones polynomial.
Table 1. Jones polynomials for knots with up to 8 crossings.
Knot Jones coefficients Knot Jones coefficients

31 {-4} -1 1 0 1 85 {0} 1 -1 3 -3 3 -4 3 -2 1
41 {-2} 1 -1 1 -1 1 86 {-7} 1 -2 3 -4 4 -4 3 -1 1
51 {-7} -1 1 -1 1 0 1 87 {-2} -1 2 -2 4 -4 4 -3 2 -1
52 {-6} -1 1 -1 2 -1 1 88 {-3} -1 2 -3 5 -4 4 -3 2 -1
61 {-4} 1 -1 1 -2 2 -1 1 89 {-4} 1 -2 3 -4 5 -4 3 -2 1
62 {-5} 1 -2 2 -2 2 -1 1 810 {-2} -1 2 -3 5 -4 5 -4 2 -1
63 {-3} -1 2 -2 3 -2 2 -1 811 {-7} 1 -2 3 -5 5 -4 4 -2 1
71 {-10} -1 1 -1 1 -1 1 0 1 812 {-4} 1 -2 4 -5 5 -5 4 -2 1
72 {-8} -1 1 -1 2 -2 2 -1 1 813 {-3} -1 3 -4 5 -5 5 -3 2 -1
73 {2} 1 -1 2 -2 3 -2 1 -1 814 {-7} 1 -3 4 -5 6 -5 4 -2 1
74 {1} 1 -2 3 -2 3 -2 1 -1 815 {-10} 1 -3 4 -6 6 -5 5 -2 1
75 {-9} -1 2 -3 3 -3 3 -1 1 816 {-6} -1 3 -5 6 -6 6 -4 3 -1
76 {-6} -1 2 -3 4 -3 3 -2 1 817 {-4} 1 -3 5 -6 7 -6 5 -3 1
77 {-3} -1 3 -3 4 -4 3 -2 1 818 {-4} 1 -4 6 -7 9 -7 6 -4 1
81 {-6} 1 -1 1 -2 2 -2 2 -1 1 819 {3} 1 0 1 0 0 -1
82 {-8} 1 -2 2 -3 3 -2 2 -1 1 820 {-5} -1 1 -1 2 -1 2 -1
83 {-4} 1 -1 2 -3 3 -3 2 -1 1 821 {-7} 1 -2 2 -3 3 -2 2
84 {-5} 1 -2 3 -3 3 -3 2 -1 1
8. The State Model for Computing the Bracket Polynomial
Hopefully you are feeling confident with your newfound understanding of the
X-polynomial and the Jones polynomial. But before you get too comfortable, ask
yourself the following question: Is there anything better than having a step-by-
step procedure to compute the X-polynomial? Tough question, right? But there is
something better: knowing not just one but two step-by-step procedures to compute
the X-polynomial. Yes, there is another way to compute it, and we would be remiss
if we did not end this chapter with a description of this alternate approach.
Recall that we previously determined the bracket polynomial of a knot or link
projection by resolving crossings one at a time and each time applying Rule 2. We
did this repeatedly until we obtained a collection of circles with no crossings, whose
bracket polynomial we could find using Rules 1 and 3.
Our new method for computing the bracket polynomial of a knot or link pro-
jection will work much faster. Instead of eliminating the crossings one at a time, we
simultaneously eliminate every crossing in the projection. Then we find all possible
ways to reconnect the four dangling ends of each crossing. We call this process
splitting the crossings. For example, in Figure 11, we illustrate a Hopf link projec-
tion at the top, and all the different ways that we can split the crossings under it.
Since the Hopf link projection has two crossings and each crossing can be split in
two different ways, we end up with four diagrams, each of which has zero crossings.
8. THE STATE MODEL FOR COMPUTING THE BRACKET POLYNOMIAL 283
Figure 11. A Hopf link projection and its four states.
Figure 12. Rotating the upper arc counterclockwise sweeps out

the A regions. The other regions are B regions.
Each of the diagrams with zero crossings will be called a state of our original
projection. In Figure 11, we obtain four states. In general, a link projection with
n crossings has 2n states.
Next we go on a labeling spree. We want to label each split crossing in each
state with either an A or a B. To do this we use Figure 12. In particular, starting
with the arc going over a crossing, we rotate it counterclockwise until it lines up
with the arc going under the crossing. This will sweep out two regions, which are
both labeled with the letter A. We then label the other two regions with the letter
B.
If we split a crossing so that the two A-regions are joined together, we call the
split an A-split. If we split a crossing so that the two B-regions are joined together,
we call the split a B-split. We illustrate these two types of splits in Figure 13.
Figure 13. A crossing can be split in these two ways.
After labeling each split in this fashion, we then label each state by indicating
which splits came from A-splits and which splits came from B-splits. In Figure 14,
we have labeled all of the states for the Hopf link projection.
Next, for each state S, let a(S) denote the number of A-splits in S, let b(S)
denote the number of B-splits in S, and let |S| denote the total number of unknotted
circles in S. In Figure 15, we determine these values for each of the states of our
Hopf link projection.
We are now ready to put all of these quantities together to compute the bracket
polynomial.
Figure 14. The A-splits and B-splits in the states of this Hopf
link projection.
Figure 15. The values of a(S), b(S), and |S| for the states of the
Hopf link projection.
Definition. Let L be a knot or link projection. For each state S in L, we

define a term of the form
Aa(S) A−b(S) (−A2 − A−2 )|S|−1 .
The sum of all of these terms is said to be the state model of the bracket polynomial
of L.
To see how this works, in Figure 16 we compute the term that each state of
the Hopf link projection contributes to the state model.
Adding up the terms in Figure 16 we get
1 + (−1 − A4 ) + 1 + (−A4 − 1) = −A4 − A−4.
Observe that this polynomial agrees with the bracket polynomial that we com-
puted for this projection in Section 4. In fact, this is true in general. For any knot
or link projection, we get the same polynomial whether we use the state model or
the rules for the bracket polynomial described in Section 4. You might see a proof
of this if you take a more advanced knot theory class. In any case, now you have
9. EXERCISES 285
Figure 16. The terms that each state of the Hopf link projection
contributes to the state model.
two methods to compute the bracket polynomial of a knot or link projection, and
either way you can use the result to get the X-polynomial for the knot or link, and
once you have that by a simple substitution you can find the Jones polynomial.
9. Exercises
1. Show that starting with a knot or link projection L and adding a negative twist
to get a new link projection L multiplies the bracket polynomial by a factor of
−A−3 .
2. (a) Find the bracket polynomial of each of the unknot projections in Figure 17.
Figure 17. Illustration for Exercise 2(a).
(b) Find the bracket polynomial of each of the unknot projections in Figure 18.
Figure 18. Illustration for Exercise 2(b).
3. Find the bracket polynomial of the Hopf link projection in Figure 19.
4. Using the definition of the bracket polynomial and your answers to Exercise 2,
find the bracket polynomial of the link projection given in Figure 20.
5. Use Rules 1 and 3 together with induction to show that the bracket polynomial
of a planar link projection L (that is, a projection with no crossings) which has
n components is
L = (−A2 − A−2 )n−1 .
6. The goal of this problem is to compute the Jones polynomial of the trefoil knot
illustrated in Figure 21 by following the outline below.
(a) First, compute the bracket polynomial for this projection.
(i) Start by applying Rule 2 to any crossing. This gives us two new
projections.
(ii) Next, use planar deformations to deform the two new projections to
look like projections we’ve seen in the previous exercises. Since you
already know the bracket polynomials for those projections, you can
plug them in.
(iii) Once you have plugged in the bracket polynomials for both pro-
jections, simplify the bracket polynomial and write it so that the
exponents are in descending order.
(b) Now, find the writhe of the projection in Figure 21 and use it to compute
X(31 ).
(c) Finally, use the substitution A = q −1/4 to compute V (31 ), the Jones poly-
nomial of the trefoil. The exponents of your polynomial should now be
in ascending order so that you can compare your answer to the Jones
polynomial for 31 given in Table 1.
7. Prove that the bracket polynomial is invariant under a Reidemeister III move.
8. Prove that the X-polynomial is invariant under adding or removing a negative

twist.
9. If L is an oriented knot or link, then −L denotes the knot or link with the
orientation of every component reversed and L∗ denotes the mirror image of L.
How is X(L) related to X(−L)? How is X(L∗ ) related to X(L)? Justify your
conclusions.
9. EXERCISES 287
10. Use Exercise 9 to show that the trefoil knot cannot be deformed to its mirror
image.
11. Compute the Jones polynomial of the figure eight knot, then check your answer
using Table 1. Hint: Start by applying bracket polynomial Rule 2 to the
crossing indicated in Figure 22, and remember that you can plug in bracket
polynomials that you’ve already computed.
12. Use the Jones polynomials given in Table 1 to find a knot with eight crossings
that cannot be deformed to its mirror image. (Remember that illustrations of
these knots are given in the knot table in Chapter 9.) Explain your reasoning.
13. Find an oriented link that cannot be deformed to its mirror image, yet has
linking number zero. Explain how you know that your link cannot be deformed
to its mirror image.
14. Let L+ , L− , and L0 be three oriented link projections that are identical except
at a single crossing where the three links differ as indicated in Figure 23.
The goal of this problem is to follow the steps below to prove the equation
q −1 V (L+ ) − qV (L− ) + (q − 2 − q 2 )V (L0 ) = 0.
1 1
(a) First write equations for the bracket polynomials

L+ and
L− in terms
of the variable A and the bracket polynomial of the link with the crossing
removed.
(b) Use the equations from step (a) to get an equation relating
L+ ,
L− ,
and
L0 .
(c) Use the equation from step (b) to get an equation relating X(L− ), X(L+ ),
and X(L0 ).
(d) Then use the substitution q = A−4 in the equation from step (c) to get
the required equation.
Part 3
Molecules
CHAPTER 12
Mirror Image Symmetry

from Different Viewpoints
Topics:
• Mirror image symmetry
• Geometric chirality
• Chemical chirality
• Understanding diagrams of molecules
• Euclidean rubber gloves
• Topological rubber gloves
• Topological chirality
1. Mirror Image Symmetry
In Chapter 3, we investigated spaces from both a topological and a geometric

viewpoint, and we explored the consequences of each. In Chapters 9, 10, and 11
we considered the topology of knots and links, and we saw that knot theory is
historically linked to chemistry. In Chapters 12–15, we apply techniques we’ve
learned in our study of topology, geometry, and knot theory to the study of non-
rigid molecules. Don’t worry, you don’t need to know any chemistry or biology to
understand these chapters.
If you are left handed, you have probably had the frustrating experience of
trying to take notes in a classroom that has only right-handed desks (see Figure 1).
Sitting at a right-handed desk makes writing with your left hand more challenging.
You could stand on your head on the seat of a right-handed desk so that your
left-hand would be next to the table, but if you’ve done this, you know that taking
notes while standing on your head presents other challenges. If you were designing
your own classroom, it might make sense to select desks which have mirror image
symmetry, so that both right- and left-handed students could use them.
An object that does not have mirror symmetry has two mirror forms, which
we can think of as the left-handed form and the right-handed form. For example,
the mirror image of a right-handed desk is a left-handed desk. Generalizing from
our experience with left-handed students having to use right-handed desks, we see
that when two objects without mirror image symmetry come together, the way
they interact depends on their handedness. For example a right foot and a left
foot interact differently with a left shoe. On the other hand, a right foot and a left
291
292 12. MIRROR IMAGE SYMMETRY FROM DIFFERENT VIEWPOINTS
Figure 1. A right-handed and left-handed desk.
foot interact the same way with a sock, because a sock has mirror image symmetry.
That is, unless it’s a toe sock whose bottom and top are different.
Our observations about asymmetric objects can also be applied to the molecules
in our bodies. In particular, most of the molecules in living things don’t have mir-
ror image symmetry. For example, DNA spirals in a left-handed helix, and DNA,
RNA, and proteins are all made of left-handed building blocks. In fact, scientists
who study the origins of life believe that life can’t exist without such molecular
asymmetry. As a result of this asymmetry, we react differently to the right-handed
and left-handed forms of an asymmetric molecule, just as a right foot reacts differ-
ently to a right shoe and a left shoe. For example, one form of the molecule carvone
smells like spearmint and the mirror form smells like caraway. The difference in
smells is the result of our olfactory system (which is asymmetric) reacting differ-
ently to the mirror forms of carvone. Similarly, most naturally occurring amino
acids taste bitter, while their mirror images taste sweet.
Pharmaceutical companies pay particular attention to mirror image symmetry,
because both the effectiveness of a drug and the severity of its side effects can depend
on which of the two forms is given. The pain reliever ibuprofen has two forms only
one of which is effective, while the mirror form has no significant effect on your
body. Ibuprofen, like almost all medications, is synthesized in a 50:50 mix of the
two mirror forms. Since it is expensive to separate the two forms, ibuprofen is sold
as a mixture. The fact that half of the molecules making up each pill are the useless
mirror forms is not important. On the other hand, the mirror image of the pain
reliever naproxen is actually toxic to your liver. So pharmaceutical companies are
obliged to go through the expense of separating the two forms so that they only sell
the harmless form. Sometimes when the two forms of a medication are separated, it
turns out that they each have a different use. For example, the medication Darvon
is a painkiller, while its mirror form Novrad is used as a cough medicine.
From these examples we see that knowing whether a molecule has mirror image
symmetry is important. But what exactly is meant by “mirror image symmetry”
is not quite as a simple as it sounds. For example, consider a stretchy right-handed
glove. If we turn it inside out, we will get a left-handed glove. Suppose that the
inside and outside of the glove are identical. Would we say that the glove has mirror
image symmetry? In the next few sections, we introduce three different approaches
to the study of mirror image symmetry: geometric, chemical, and topological.
2. GEOMETRIC SYMMETRY 293
2. Geometric Symmetry
We begin with geometric symmetry because this corresponds most closely to

our intuition. Recall from Chapter 3 that two objects in R3 have the same extrinsic
geometry if there is a rigid motion taking one to the other. In the same chapter,
we defined a rigid motion to be any motion of an object in space that doesn’t
involve stretching, shrinking, or bending. Such motions can also be thought of as
a combination of rotations and translations of the object in space.
We are now interested in whether an object and its mirror image are geomet-
rically the same. That is, we want to know whether we can go from an object to
its mirror image by a motion that doesn’t involve stretching, shrinking, or bending.
For example, could such a motion change the right-handed desk in Figure 1 to a
left-handed desk? If it could, left-handed students could just move the chair in the
right way, and they wouldn’t have to stand on their heads.
Consider the knot projection illustrated on the left side of Figure 2. The mirror
image (illustrated on the right) is obtained by switching all of the undercrossings
and overcrossings. At first glance you might not say that this projection has mirror
symmetry. Yet, if we rotate the knot on the left by 90◦ we get the knot on the
right. So geometrically this knot has mirror image symmetry.
Figure 2. A knot and its mirror image.
When we ask whether or not an object is geometrically the same as its mirror
image, it may seem like it could depend on where we place the mirror. Actually,
at this very moment, A. Square is wondering the same thing about mirror images
in Flatland. In Figure 3, we see A. Square experimenting with mirrors in different
positions to determine whether all of his mirror images are the same. This experi-
ment also provides A. Square with the opportunity to admire himself in the mirror
from all different angles, which is something he always enjoys.
Figure 3. A. Square and his image in different mirrors.
Note that the mirrors in the figure are illustrated as dotted line segments and
A. Square’s mirror images are illustrated in grey. From this figure, we can see
that A. Square can get from any one of his mirror images to any other by a rigid
motion of the mirror image in Flatland. Thus all of A. Square’s mirror images are
geometrically the same, regardless of the position of the mirror. This is also true
in 3-dimensional space. No matter where you are relative to a mirror, your mirror
image will be the same. But you might want to verify this for yourself.
3. Geometric Chirality and Achirality
In order to determine whether a molecule is the same or different than its

mirror image, we need to physically represent the molecule in space. We do this
with a graph where each vertex represents an atom or a group of atoms, and each
edge represents a molecular bond. Chemists have developed a nice way to draw
pictures of these graphs which includes information about what is in front and what
is behind the plane of the paper.
For example, consider the amino acid L-alanine represented by the graph in
Figure 4. The line segments in the figure indicate edges which lie in the plane of
the paper. The dark triangular segment indicates an edge which comes out of the
plane of the paper towards you with the widest part of the triangle closest to you.
The dashed triangular segment indicates an edge which goes back behind the plane
of the paper with the widest part of the triangle farthest from you. Thus the C,
CH3 , and H are in the plane of the paper, while the NH2 is in front of the paper and
the CO2 H is behind the plane of the paper. In 3-dimensional space, the vertices
labelled CH3 , H, NH2 , and CO2 H lie at the corners of a regular tetrahedron. Don’t
worry if you don’t know what each of these letters stands for. Just realize that
different letters represent different atoms.
Figure 4. This graph represents the 3-dimensional structure of

the amino acid L-alanine.
The molecule L-alanine together with its mirror image are illustrated in Figure
5. If we imagine that the molecule is rigid, no matter how we rotate or translate it,
we will never obtain its mirror image. Thus geometrically this molecule is different
from its mirror image.
Now let’s consider the molecule illustrated in Figure 6. This molecule is differ-
ent from L-alanine because it has two H’s (representing hydrogen) rather than just
one. Is this molecule geometrically different from its mirror image?
In order to be able to refer to molecules and other objects which are geometri-
cally different from their mirror image we introduce the following definition.
Definition. An object is said to be geometrically chiral if there is no rigid
motion taking it to its mirror image. Otherwise, it’s said to be geometrically
achiral.
4. CHEMICAL CHIRALITY AND ACHIRALITY 295
Figure 5. The molecule L-alanine together with its mirror image.
Figure 6. Is this molecule geometrically different from its mirror image?
For example, the molecule in Figure 5 is geometrically chiral, while the molecule
in Figure 6 is geometrically achiral.
At first this definition may seem confusing because an object is chiral if it is
NOT the same as its mirror image and it is achiral if it IS the same as its mirror
image. Normally, we add the prefix “a” to a word to indicate that something is not
the case. For example, a person is said to be apolitical if he/she is NOT political.
So our definitions of chiral and achiral seem backwards.
However, like many definitions, the definition of the word chiral makes more
sense if you understand its etymology. The word chiral (which rhymes with spiral)
comes from the ancient Greek word χειρ which means hand. There is no rigid
motion taking a left hand to a right hand. Thus a hand is a prototypical example
of a chiral object. So an object which is chiral has the trait of being “like a hand”,
while an object which is achiral has the trait of being “not like a hand”. In this
sense, the prefix “a” added to the word chiral has the same meaning as it does
when it is added to the word political. If you ever get confused between chiral and
achiral, just remember that the word without the “a” means hand.
4. Chemical Chirality and Achirality
As we mentioned in Section 1, because of the asymmetry of the molecules in

our bodies, we may react differently to the right-handed and left-handed forms of
a medication. Chemists want to know whether a given molecule is the same as its
mirror image in order to predict how it will react with other molecules. This can be
determined experimentally because molecules which are different from their mirror
image rotate polarized light at room temperature, while molecules which are the
same as their mirror image do not. Here by “the same” we mean chemically the
same. We formalize this with the following definition.
Definition. A molecule is said to be chemically chiral if it cannot trans-

form itself to its mirror image at room temperature. Otherwise, it is said to be
chemically achiral.
By transform itself, we mean the chemical structure of the molecule enables it

to change into its mirror image. Such a transformation is not a chemical reaction,
but rather a conversion of the molecule back and forth between the two mirror
forms of the structure. Saying that a molecule is chemically achiral says something
about the behavior of the molecule, not necessarily its geometry.
Before we look at a molecular example, let’s consider the behavior of a right-
handed glove. A right-handed glove is geometrically chiral because it cannot be
rotated or translated to get a left-handed glove, which is its mirror image. This
means that as a rigid object, it is distinct from a left-handed glove. But suppose
we are interested in how the glove interacts with a hand rather than its geometry.
We might observe that our right-handed glove can be turned inside out and then
it becomes a perfectly wearable left-handed glove (assuming there aren’t a lot of
threads and what-not on the inside). This means that the glove is achiral in a
practical sense even though it’s geometrically chiral. Similarly, our definition of
being chemically achiral is a practical definition about molecular behavior rather
than a geometric definition.
If a molecule is geometrically achiral, then as a rigid object its structure is
identical to that of its mirror image. It follows that it must be chemically achiral
as well, since no chemical transformations are necessary to change it into its mirror
image. Equivalently, we could argue that if a molecule is chemically chiral, then
no chemical transformations allow it to transform itself into its mirror image, and
hence it is geometrically chiral as well.
We illustrate the relationship between the collection of all chemically achiral
molecules and the collection of all geometrically achiral molecules in Figure 7. This
picture tells us that the set of all geometrically achiral molecules is contained in
the set of all chemically achiral molecules. The question is, are there molecules
which are chemically achiral, but not geometrically achiral? Or are the ovals in
Figure 7 actually equal? We have seen that a glove can be geometrically chiral yet
practically achiral. But can there be a molecule that has this property? We will
answer this question in the next section.
Chemists normally use the word “chiral” for both geometrically chiral molecules
and chemically chiral molecules. Lord Kelvin, who created the first knot tables,
coined the term chiral to refer to an object that is geometrically different from its
mirror image. In particular, in 1884, Kelvin wrote the following:
I call any geometrical figure, or group of points chiral, and say

it has chirality, if its image in a plane mirror, ideally realized,
cannot be brought to coincide with itself.
4. CHEMICAL CHIRALITY AND ACHIRALITY 297
Figure 7. The set of all geometrically achiral molecules is con-

tained in the set of all chemically achiral molecules.
Modern organic chemistry textbooks often paraphrase Lord Kelvin’s definition

as follows:
Definition. A chiral molecule is one that is not superimposable on its mirror

image.
This definition means that a molecule is chiral if as a rigid object it is not
identical to its mirror image. This is precisely the same as our definition of being
geometrically chiral. The problem with applying this definition to molecules is that
it assumes that all molecules are rigid. But some molecules are large enough that
they can be floppy. For example, DNA is floppy and can even be knotted (as we’ll
see in Chapter 14). There are other synthetic and biological polymers that are not
as long as DNA, but can still be somewhat flexible. Furthermore, even some small
molecules have certain bonds around which a piece of the molecule can rotate. For
example, let’s consider the molecule in Figure 8.
Figure 8. In the next section, we’ll show that this molecule is

geometrically chiral but chemically achiral.
In order to understand the 3-dimensional structure of Figure 8, recall that a

dark triangular segment indicates an edge which comes out of the plane of the paper
towards you with the widest part of the triangle closest to you. A dashed triangular
segment indicates an edge which goes behind the plane of the paper with the widest
part of the triangle farthest away from you. These rules help us understand the
positions of the four hexagons in the figure. In particular, we see that the left-most
and right-most hexagons are behind the plane of the paper, because of the way the
dashed segments connect them to the central part of the molecule. The hexagon
which is the second from the left in the figure is in the plane of the paper, and the
hexagon which is second from the right is perpendicular to the plane of the paper
with half of it in front and half of it behind the page. You should carefully check
that you agree with all of the notes on the figure saying what’s behind and what’s
in front of the page.
The molecule in Figure 8 is rigid except for the two pieces on the sides which
rotate simultaneously as indicated by the arrows. We call these rotating pieces
propellers. In contrast with molecules we’ve seen earlier in the chapter, the molecule
in Figure 8 has corners with no letters on them. Such corners should be seen as
vertices of the graph that represent carbon atoms. The letter C is often omitted
from molecular graphs because carbons are so common. Note that the NO2 and
the O2 N represent the same group of atoms. The order in which the letters are
written is meant to indicate that the N is attached to the adjacent hexagon.
5. Chemical Achirality and Geometric Chirality of Figure 8
The twins, Leftie and Rightie, have been begging to appear in this book. They
think it’s not fair that A. 3D-Girl is practically the only 3-dimensional person who
is in the book. They have even accused us of not wanting them because they’re not
ambidextrous. So we’ve agreed to include them in this section to help us explain
why the molecule in Figure 8 is chemically achiral but geometrically chiral.
We begin by drawing the molecule in Figure 9 with the pieces on the left
and right sides replaced by Leftie and Rightie, respectively. Because of the striped
triangles in Figure 8, attaching the left-most and right-most hexagons to the central
part of the molecule, Leftie and Rightie are standing behind the plane of the paper.
Thus Leftie has her left arm forward to hold on to her oxygen and Rightie has her
right arm forward to hold on to her oxygen. Since the molecule is rigid everywhere
except for the rotating propellers, Leftie must always have her left arm forward,
and Rightie must always have her right arm forward.
Figure 9. Leftie and Rightie have replaced the left and right
pieces of the molecule.
Now let’s imagine a vertical mirror to the right of the molecule in Figure 10.
We’ve seen that it doesn’t matter where the mirror is, so we’ve chosen this one
because it makes the mirror image easy to draw. The molecule in the mirror
(illustrated in Figure 10) looks the same as the original except that the positions
5. CHEMICAL ACHIRALITY AND GEOMETRIC CHIRALITY OF FIGURE 8 299
Figure 10. The mirror image of the molecule from Figure 9.
of the vertical and horizontal hexagons are switched. Also, in the mirror image,
Leftie is wearing a dress with a backwards R and Rightie is wearing a dress with
a backwards L. This may seem confusing, but since Leftie only has a left arm
and Rightie only has a right arm, even when they wear each other’s clothes turned
inside out, there is no ambiguity about which twin is which.
We show as follows that the molecule in Figure 9 can transform itself to its
mirror image in Figure 10. First we rotate the molecule in Figure 9 by 90◦ around
a horizontal axis as illustrated in Figure 11. Since the molecule is in solution, it
can rotate freely.
Figure 11. We rotate the molecule by 90◦ around a horizontal axis.
Figure 12. This is the result of rotating the molecule by 90◦ .

As we see in Figure 12, this causes the vertical hexagon to become horizontal
and the horizontal hexagon to become vertical. We’re happy about this because
this is the way the hexagons are in the mirror image of the molecule. However, the
oxygens are now in back and in front of the page, and Leftie and Rightie are now
horizontal.
To fix the oxygens and the twins, we simply let the propellers spin until they are
as they should be. In this way, we end up with the mirror image that was illustrated
in Figure 10. Since rotating the molecule by 90◦ and letting the propellers spin are
both chemically possible, this shows that the molecule is chemically achiral.
Next we want to show that the molecule is geometrically chiral. This means
we have to show that no rigid motion can take the molecule to its mirror image.
We begin by freezing the molecule in its original position so that the propellers
can’t spin. In Figure 13, we see that in the original molecule, Leftie is parallel to
the adjacent hexagon. Hence, as a rigid object, she will always have her left arm
forward and be parallel to the adjacent hexagon. In fact, whether or not she’s rigid,
she doesn’t have a right arm so she can’t put her right arm forward.
Figure 13. Leftie is parallel to the adjacent hexagon in the orig-

inal form.
In the mirror image (illustrated in Figure 14), Leftie is perpendicular to the

adjacent hexagon, whereas Rightie is parallel to the adjacent hexagon. Since no
rigid motion can take a pair of parallel hexagons to a pair of perpendicular hexagons,
and Leftie cannot become Rightie, as a rigid object the original molecule is different
from its mirror image. Thus the molecule is geometrically chiral.
Figure 14. Leftie is perpendicular to the adjacent hexagon in the

mirror image.
6. EUCLIDEAN RUBBER GLOVES 301
Figure 15. Not every chemically achiral molecule is geometrically achiral.
This molecule illustrates that the concepts of chemical achirality and geometric
achirality are not equivalent. In particular, the question that was in Figure 7 can
now be replaced by the molecule in Figure 8. The new diagram is illustrated in
Figure 15.
6. Euclidean Rubber Gloves
There is a name for molecules like the one illustrated in Figure 8, which we
define below.
Definition. A molecule is said to be a Euclidean rubber glove if it is

chemically achiral, but it is chemically impossible for it to attain a position which
is geometrically achiral.
In the last section, we demonstrated that the molecule in Figure 8 is a Euclidean
rubber glove. In order to see why these types of molecules are called “rubber
gloves” remember that a right-handed rubber glove becomes a left-handed rubber
glove when it is turned inside out. However, despite it being made of rubber or
even latex, a rubber glove does not have the flexibility to get into a position which
is geometrically achiral. It’s the physical characteristics of a rubber glove which
prevent it from attaining a position which would be rigidly the same as its mirror
image. The word “Euclidean” in this definition refers to physical, chemical, or
geometric constraints that prevent the molecule from getting into a geometrically
achiral position.
For example, if we ignored all such constraints for the molecule in Figure 8,
we could flatten it into the page of the paper as we’ve done in the image on the
bottom of Figure 16. Notice that the drawing at the bottom of Figure 16 has no
Figure 16. Ignoring chemical constraints, we could flatten this

molecule into a plane.
black triangles or dashed triangles because no part of it is in front of or behind

the plane of the paper. This includes the horizontal hexagon in the molecule which
has now become vertical in the flattened picture. If we imagine that the mirror
is in the plane of the paper, we can see that the flattened out molecule would be
its own mirror image. In fact, any object that is contained in a plane is its own
mirror image in 3-dimensional space. Thus a flattened out drawing of the molecule
in Figure 8 would be geometrically achiral. However, flattening out the molecule is
not chemically possible.
7. Geometrically Chiral and Achiral Knots
It is very tempting to think that every object which can change into its mirror
image can be deformed into a geometrically achiral position if it’s sufficiently flex-
ible. In particular, we might wonder whether every molecule which is a Euclidean
rubber glove can be flattened out so that it’s contained in a plane like the molecule
in Figure 16. To explore whether this is true, let’s first consider some knots.
We saw in Chapter 9 that the figure eight knot can be deformed to its mirror
image. However, there is no rigid motion taking the usual projection of a figure
eight knot to its mirror image (see Figure 17). For example, if you turn over the
projection of the figure eight knot on the left, you will obtain the same projection
again rather than its mirror image. In particular, this projection of the figure eight
knot is not geometrically achiral.
With some effort you can deform the figure eight knot to the projection illus-
trated in Figure 18, which we saw previously can be rotated by 90◦ to obtain its
mirror image. Thus Figure 18 is a geometrically achiral projection of the figure
eight knot. What makes this projection difficult to find is that it has twice as many
crossings as the usual projection of the figure eight knot.
7. GEOMETRICALLY CHIRAL AND ACHIRAL KNOTS 303
Figure 17. Turning over the figure eight knot gives us the same
projection rather than the mirror image.
Figure 18. A geometrically achiral projection of the figure eight

knot together with its mirror image.
Figure 19. 817 can be deformed to its mirror image but cannot
be deformed to a geometrically achiral position.
It turns out that the eight crossing knot 817 (illustrated in Figure 19) is the
first example in the knot tables of a knot which can be deformed to its mirror image
but cannot be deformed to a geometrically achiral position. However, the proof of
this is not easy.
In fact, if an infinitely long string contains 817 , then it can be deformed to a
projection which is geometrically achiral. We illustrate such a projection in Figure
20, where we see that the knotted line can be rotated by 180◦ to obtain its mirror
image. However, as a knotted circle, 817 cannot be deformed to a projection which
can be transformed to its mirror image by a rigid motion. For example, imitating
Figure 20, we can deform 817 to the projection in Figure 21. However, if we rotate
this projection by 180◦ , then even though the knotted arc goes to its mirror image,
the semi-circular arc will go to the top when it should remain at the bottom. This
Figure 20. This knotted line can be rotated by 180◦ to obtain its
mirror image.
Figure 21. If we rotate this projection by 180◦ , the arc at the

bottom goes to the top.
knot demonstrates that not every object that can change into its mirror image
necessarily can be deformed to a position which is geometrically achiral.
Let’s compare the properties of the knot 817 to those of the molecule illustrated
in Figure 8. Since the molecule is a Euclidean rubber glove, it is chemically achiral
but geometrically chiral. However, we saw that if we were to ignore all chemical
constraints, the molecule could be flattened out into a geometrically achiral position.
By contrast, the knot 817 can be deformed to its mirror image, but cannot be
deformed to a geometrically achiral position.
8. A Topological Rubber Glove
We might wonder if there could be a molecule that behaves like the knot 817 .
That is, is there a molecule which is chemically achiral but can’t be deformed
to a geometrically achiral position? This question was asked in 1983, but the first
example of such a molecule wasn’t synthesized until 1997. The molecule (illustrated
in Figure 22) is a link with a pair of adjacent hexagons at the top that can rotate
around the bonds connecting them to the rest of the molecule (as indicated by the
arrows). Note that the H3 C and the CH3 represent the same group of atoms. The
order in which the letters are written is meant to indicate that the C is attached
to the adjacent hexagon in both molecules.
We see as follows that this molecule is chemically achiral. Since the two com-
ponents of the link are not connected to one another, the lower ring can turn over
without affecting the top ring. When the lower ring does turn over, the crossings
between the two rings switch and the H3 C goes from sticking out on the left to
sticking out on the right. In this way, the turned over lower ring looks exactly like
the lower ring in the mirror image. Since the pair of adjacent hexagons on the top
ring rotates, the pair can change from being slanted up to being slanted down. This
means the pair of hexagons can go from their position in the original molecule to
their position in the mirror image. Thus the molecule can chemically transform
itself into its mirror image, and hence it is indeed chemically achiral.
Next we want to convince ourselves that, even if the molecule in Figure 22 were
completely flexible, it could not be deformed to a position which is geometrically
achiral. Suppose it could be deformed into a position which is geometrically achiral.
Once it’s in such a position, it’s frozen. This means the slanted hexagons can no
longer rotate. Since the components of the link are not identical, the rigid motion
taking the deformed molecule to its mirror image must take each component to
itself. But the bottom component doesn’t have mirror image symmetry because
the H3 C is only on one side, and the top component doesn’t have mirror image
8. A TOPOLOGICAL RUBBER GLOVE 305
Figure 22. A molecular link and its mirror image.
symmetry because the pair of slanted hexagons are not symmetric. So the only
way for the molecule to be geometrically achiral would be if both components were
actually lying in the plane of the mirror. But because the components are linked,
they cannot simultaneously lie in a plane. Hence the molecule cannot be deformed
to a position which is geometrically achiral.
By analogy with our definition of a Euclidean rubber glove, we now make the
Definition. A molecule is said to be a topological rubber glove if it is

chemically achiral, but even if it were completely flexible it could not be deformed
to a position that is geometrically achiral.
Recall that a molecule is a Euclidean rubber glove if it is chemically achiral,

but it is chemically impossible to attain a position which is geometrically achiral.
In other words, it is chemically impossible for a Euclidean rubber glove to get into a
geometrically achiral position. A topological rubber glove has the stronger property
that even if we ignore all chemical constraints and imagine that the molecule is a
completely flexible graph, then the molecule still couldn’t get into a position that
is geometrically achiral. The molecule illustrated in Figure 22 is an example of a
topological rubber glove.
Figure 23 provides an illustration of the relationship between chemical achirality
and geometric achirality. rubber glove. Note that the molecules from both Figure
8 and 22 are in the larger oval but not the smaller oval because they are chemically
achiral but not geometrically achiral. The difference between these two molecules
is that chemical constraints prevent the molecule in Figure 8 from attaining a
geometrically achiral position, while even ignoring chemical constraints the molecule
in Figure 22 cannot be deformed to a geometrically achiral position.
Figure 23. The relationship between chemically achiral and geo-

metrically achiral molecules.
9. Topological Chirality
When we defined geometric chirality, we were taking a geometric viewpoint

by treating all molecules as if they were completely rigid. As we have seen, some
molecules are rigid, while others have pieces that can rotate around certain bonds,
and large molecules (like DNA) can even be pretty flexible. So a geometric view-
point does not always correspond to chemical reality. In fact, a mathematical char-
acterization of chirality for all molecules is impossible because the level of rigidity
of a molecule depends on its chemistry not just its geometry. In this section, we will
take a topological viewpoint by treating all molecules as if they were completely
flexible. The chemical reality is wedged between the geometric viewpoint and the
topological viewpoint.
Definition. A molecule is said to be topologically chiral if it cannot be
deformed to its mirror image assuming complete flexibility. Otherwise, it is said to
be topologically achiral.
Any chemical motion of a molecule (whether rigid or not) can be thought of as
a deformation of the molecule. Thus if a molecule can chemically transform itself
into its mirror image, then it can be deformed into its mirror image. This means
that chemical achirality implies topological achirality. It follows that if a molecule
is topologically chiral, it is also chemically chiral. Furthermore, a molecule whose
chirality comes from its rigidity can be forced to change into its mirror image by
heating it enough. However, if a molecule is topologically chiral, then bonds would
have to break for it to change into its mirror image. Thus it will not change into
its mirror image even when heated. So topological chirality is a more enduring
type of chirality than geometric chirality. For this reason, topological chirality is
an interesting concept for chemists.
On the other hand, let’s consider the L-alanine molecule in Figure 24. This
molecule is small and hence quite rigid. Thus it cannot chemically transform itself
9. TOPOLOGICAL CHIRALITY 307
Figure 24. This molecule is chemically chiral but topologically achiral.
into its mirror image. However, if we pretend that the molecule is completely
flexible, we could grab the CO2 H and push it to the left, while pulling the H to the
right. In this way, we could deform the molecule to its mirror image. Hence the
L-alanine molecule is chemically chiral but topologically achiral.
The diagram in Figure 25 illustrates the relationship between topologically achi-
ral molecules, chemically achiral molecules, and geometrically achiral molecules. In
particular, we have seen that if a molecule is geometrically achiral, then it is chem-
ically achiral. But if it’s geometrically chiral, it may or may not be chemically
chiral, depending on the rigidity of the molecule. On the other hand if a molecule
is topologically chiral, then it is chemically chiral, independent of how rigid the
Figure 25. The relationship between topologically achiral, chem-

ically achiral, and geometrically achiral molecules.
molecule is. But if a molecule is topologically achiral, it may or may not be chem-
ically achiral, depending on the rigidity of the molecule. So geometry is useful to
prove chemical achirality, and topology is useful to prove chemical chirality.
The molecular Möbius ladder illustrated in Figure 26 is an example of a topo-
logically chiral molecule. This molecule is called a Möbius ladder because it looks
like a Möbius strip except it is made from a ladder with three rungs rather than from
a strip of paper. But how can we show that a molecule like the molecular Möbius
ladder is topologically chiral? In order to do this, we would need to demonstrate
that no deformation could take the molecule to its mirror image. This reminds
us of the task of showing two knots are not equivalent. In Chapters 9, 10, and
11 we addressed this problem by developing invariants. However, there aren’t as
many invariants for molecules as there are for knots and links. In the next chapter,
we will learn several different methods for showing that a molecule is topologically
chiral.
Figure 26. A molecular Möbius ladder is an example of a topo-

logically chiral molecule.
10. Exercises
1. Explain how we know that geometric chirality does not imply chemical chirality,
and chemical chirality does not imply topological chirality.
2. Prove that if the molecule illustrated in Figure 27 were completely flexible, it
could be deformed into a plane. Is the molecule geometrically chiral? Is it
topologically chiral? Given that this molecule is rigid, is it chemically chiral?

10. EXERCISES 309
3. (Hard.) Prove that if we don’t distinguish between the rungs and sides of
the molecular Möbius ladder illustrated in Figure 26, then the graph could be
deformed to its mirror image.
4. Draw a step-by-step deformation taking a Möbius ladder with four rungs (il-
lustrated in Figure 28) to a position where the circle representing the sides of
the ladder lies in a plane.

5. Consider a closed ladder like the molecular Möbius ladder in Figure 26, but
with three half-twists instead of just one. Prove that there is no deformation
that takes the closed ladder to its mirror image taking sides to sides and rungs
to rungs.
6. We can define an object (not just a molecule) to be a Euclidean rubber glove if
it can be deformed to its mirror image, but for physical, chemical, or geometric
reasons it cannot be deformed to a geometrically achiral position. Explain why
a real rubber glove is a Euclidean rubber glove. Give another example of a
familiar object which is a Euclidean rubber glove, and explain why.
7. We can define an object to be a topological rubber glove if it can be deformed
to its mirror image, but even if it’s completely flexible, it cannot be deformed
to a geometrically achiral position. Is a trefoil knot a topological rubber glove?
What about a figure eight knot?
8. Prove that if a molecule is a topological rubber glove, then it is a Euclidean
rubber glove, but not the converse.
9. Is the molecule illustrated in Figure 8 a topological rubber glove? Explain your
reasoning.
10. (Hard.) Prove that the Kuratowski cyclophane molecule (illustrated in Fig-
ure 29) can be deformed to a geometrically achiral position. Note that the
dotted edge goes behind the solid edges.

CHAPTER 13
Techniques to Prove Topological Chirality
Topics:
• A molecular knot
• Using knot polynomials to prove topological chirality
• Using 2-fold branched covers to prove topological chirality
• Using chiral subgraphs to prove topological chirality
• Graph automorphisms and their properties
• A combinatorial method to prove topological chirality
1. Topological Chirality
In Chapter 12, we learned about three different types of chirality: geometric,

chemical, and topological. In order to determine if a given molecule is geometrically
chiral, we treat it as a rigid object and look at it from all different angles to see if
there is a rigid motion from it to its mirror image. Chemical chirality is determined
experimentally. In order to prove that a molecule is topologically chiral, we have to
prove that it is impossible to deform to its mirror image assuming complete flexibil-
ity. This is most difficult type of chirality to prove, because there are infinitely many
possible deformations of a given molecular graph. However, by treating molecules
as topological objects, we can use techniques which build on material we’ve learned
about knot theory to show particular molecules are topologically chiral.
In this chapter, we will present four different techniques that have been used to
prove specific molecules are topologically chiral. None of the techniques presented
here (or any technique currently known) will work for all possible molecules. The
best we can do is to develop a collection of techniques that we can choose among
when confronted with a molecular graph. The more techniques we know, the better
the odds are that one of them will work for the given molecule.
This is analogous to what we learned about knot invariants in Chapters 9,
10, and 11. We would like to be able to prove that a particular knot is different
from another knot. If we can show that a particular invariant of the two knots is
different, then we know the knots are different. But there is no single invariant that
we can use to distinguish every possible pair of knots. So the more invariants we
know, the more likely it is that one of them will work to distinguish a given pair of
knots.
311
312 13. TECHNIQUES TO PROVE TOPOLOGICAL CHIRALITY
2. Molecular Knots
Knots have been found to occur naturally in both DNA and proteins (which we
will learn about in Chapters 14 and 15). However, it took over 50 years of attempts
by organic chemists before the first molecular knot was successfully synthesized in
1989. Not surprisingly, the first molecular knot to be synthesized was a trefoil knot
(illustrated in Figure 1). If we want to be precise, the molecular trefoil knot is
not actually a knot according to the definition we gave in Chapter 9 because it
has clumps of hexagons (known by chemists as aromatic rings) at various places
along what should just be a circle. We call it a molecular knot because the overall
structure of the molecule is that of a knot. That is, if you look at the picture from
far away without your glasses or contact lenses, you’ll see the knot rather than all
the hexagons and the many copies of the letter O.
Figure 1. A molecular trefoil knot.
In fact, this molecular graph contains many trefoil knots. Each hexagon has
two different paths going from one end to the other. Depending on which of these
paths we choose, we will trace a out a different trefoil. For example, in Figure 2, we
use thick grey lines to illustrate two of the trefoils contained in the graph. Can you
see where the two knots differ? You should also see if you can find other trefoils in
the graph.
Figure 2. Two trefoil knots contained in this molecular graph.

3. TECHNIQUE 1: KNOT POLYNOMIALS 313
Notice that all of the non-trivial knots in the molecular graph are identical
trefoil knots. In particular, the graph contains many left-handed trefoil knots (il-
lustrated on the left in Figure 3), but no right-handed trefoil knots (illustrated on
the right in Figure 3). In addition to all of the trefoil knots, the graph contains
some small planar circles made of one, two, or three hexagons. But planar circles
are identical to their mirror images, so they aren’t important to us here.
Figure 3. A left-handed trefoil and a right-handed trefoil.
3. Technique 1: Knot Polynomials
In order to prove that the molecular trefoil knot is topologically chiral we need
to review what we learned about knot polynomials in Chapter 11. In particular,
recall that the X-polynomial is defined for an oriented knot or link projection L
by the formula X(L) = (−A3 )−w(L)
L, where w(L) denotes the writhe of L, and
L is the bracket polynomial of L. The writhe of an oriented link is the difference

between the number of positive crossings and the number of negative crossings.
Also, recall that the bracket polynomial is defined by the rules below:

Rule 1. = 1.

Rule 2. =A + A−1 .

Rule 3. L = (−A2 − A−2 )
L.
In one of the Exercises in Chapter 11 you were asked to show that the trefoil
knot cannot be deformed to its mirror image. If you did this exercise, you might
have proved the following more general theorem.
Theorem 13.1. Let L be a knot with mirror image L∗ . Then the X-polynomial
X(L∗ ) is obtained from X(L) by reversing the signs of all of the exponents.
For a projection of a knot, the value of the writhe is the same regardless of
the orientation on the knot, which is why we don’t need to orient our knot in the
statement of Theorem 13.1.
Proof. We begin by comparing the writhe of a knot projection L to that of
its mirror image L∗ . When we take the mirror image of the projection, the positive
and negative crossings switch, causing the writhe to change sign. Thus we have
∗
w(L∗ ) = −w(L). It follows that (−A3 )−w(L ) = (−A3 )w(L) = (−A−3 )−w(L) . Thus
the X-polynomial of the mirror image is given by
∗
X(L∗ ) = (−A3 )−w(L )
L∗ = (−A−3 )−w(L)
L∗ .
Next let’s consider how the bracket polynomial of L∗ compares with that of
L. Suppose that Rule 2 of the bracket polynomial is used to remove a particular
crossing. If we apply Rule 2 to the same crossing in the mirror image, we get the
same result except that A and A−1 are switched. Furthermore, replacing A by A−1
has no effect on the formula in Rule 3, since −(A−1 )2 − (A−1 )−2 = −A2 − A−2 .
Also, switching A and A−1 has no effect on the outcome of Rule 1. It follows that
L∗ can be obtained from

L by switching A and A−1 .
∗
The formula (−A3 )−w(L ) = (−A−3 )−w(L) that we derived above can be re-
∗
stated by saying that we obtain (−A3 )−w(L ) from (−A3 )−w(L) by switching A and
A−1 . Putting all of this together, we conclude that X(L∗ ) is the same as the X(L)
except that the signs of the exponents have been reversed. This is exactly what the
theorem says.
Returning now to the task of showing that the molecular trefoil knot is topo-
logically chiral, let T be a left-handed trefoil knot. From Chapter 11, we know that
the X-polynomial of T is X(T ) = −A16 + A12 + A4 . Theorem 13.1 tells us that the
X-polynomial of the mirror image is X(T ∗ ) = −A−16 + A−12 + A−4 . This could
be also shown by direct computation, but it’s easier to use the theorem. Since
X(T ) = X(T ∗ ), we conclude that the left-handed trefoil knot cannot be deformed
to the right-handed trefoil knot. Hence the trefoil knot is topologically chiral. Fi-
nally, since the molecular trefoil knot in Figure 1 contains left-handed trefoils but
no right-handed trefoils, and a trefoil knot is topologically chiral, our molecular
trefoil knot must also be topologically chiral.
We say that a knot polynomial is symmetric with respect to the signs of its
exponents if switching the sign of every exponent does not change the polynomial.
For example, the polynomial −A16 + A12 + A4 is not symmetric with respect to
the signs of its exponents, whereas the polynomial −A2 − A−2 is symmetric with
respect to the signs of its exponents. If we compute the X-polynomial of a knot and
observe that it is not symmetric with respect to the signs of its exponents, then we
immediately know that the knot is topologically chiral. The same statement is also
true for the Jones polynomial, since the X-polynomial and the Jones polynomial
are related by the change of variable q = A−4 . This means that we can determine
that a knot is topologically chiral just by looking at a knot table containing Jones
polynomials. For instance, a knot table shows that the Jones polynomial for the
trefoil knot is q + q 3 − q 4 , which has only positive exponents, and hence does not
have exponents which are symmetric with respect to sign. Hence we immediately
know from the knot table that the trefoil is topologically chiral. These arguments
should convince you of the truth of the following theorem.
Theorem 13.2. If the Jones polynomial (or X-polynomial ) of a knot is not
symmetric with respect to the signs of the exponents, then the knot is topologically
chiral.
Though this is a useful technique to detect topological chirality, it does have
some limitations. First of all, the technique can only be used on a molecule which
contains a knot or a link. Secondly, even if the molecule does contain a knot or
4. TECHNIQUE 2: 2-FOLD BRANCHED COVERS 315
link, if its Jones polynomial happens to be symmetric with respect to the signs
of its exponents, we can’t conclude anything about whether or not the knot is
topologically achiral. Finally, in the case of the molecular trefoil knot, all of the
knots in the molecular graph were left-handed trefoils. Hence, proving that a trefoil
knot is topologically chiral is sufficient to know that the molecule is topologically
chiral. But if a molecule contains both a knot and its mirror image, it may or
may not be topologically chiral. For example, Figure 4 illustrates a molecular knot
which contains both a left-handed trefoil and a right-handed trefoil. This molecule
is chemically achiral, and hence topologically achiral.
Figure 4. A chemically achiral molecular knot.
Now imagine a vertical mirror dividing the knot in half. The mirror would
simply interchange the two sides of the molecule, taking the molecule to itself.
Thus the molecule is actually geometrically achiral.
4. Technique 2: 2-Fold Branched Covers
In 1983, the chemist David Walba synthesized a molecular Möbius ladder by

forcing the ends of a 3-rung molecular ladder to join with a half-twist (see Figure 5).
Remember from Chapter 12 that corners in molecular drawings should be treated
as vertices of the graph representing carbon atoms. The three vertical pairs of
edges that we see in the illustration represent double bonds joining a pair of carbon
atoms.
Figure 5. The molecular Möbius ladder.

Walba wanted to assert that some of his molecules did indeed have the form of
a Möbius ladder, but the molecules were too small to see in a microscope. Walba
showed experimentally that some of his molecules were chemically chiral, and he
wanted to argue that this was consistent with their having the form of Figure 5. He
could prove that a Möbius ladder was geometrically chiral, but the molecules that
he had synthesized were long enough that they had some flexibility. So knowing
that the molecules were geometrically chiral wasn’t sufficient to conclude that they
were chemically chiral.
Walba suspected that the molecular Möbius ladder was topologically chiral,
but he couldn’t prove it. He knew that a Möbius strip with the twist going one
way could not be deformed to a Möbius strip with the twist going the other way.
However, there are more ways to deform a Möbius ladder than a Möbius strip.
In 1986, the topologist Jon Simon solved Walba’s problem by finding a link in
the 2-fold branched cover of the Möbius ladder whose topological chirality he could
use to prove the topological chirality of the molecular Möbius ladder. This was the
first time deep topological techniques were used to solve a chemically motivated
question, and as such marked the beginning of the interdisciplinary field of chemical
topology.
We won’t define 2-fold branched covers because the definition is too technical.
You could see the definition if you took a geometric topology course. Here we give
a sketch of the main idea’s of Simon’s proof, without going into the technicalities.
The first step of the proof is to simplify the graph of the Möbius ladder by
omitting all of the atoms and the double bonds, and distinguishing the rungs from
the sides by coloring the rungs black and the sides grey (see Figure 6).
Figure 6. A colored Möbius ladder.
In fact (as you may have shown in the Exercises in Chapter 12), if we don’t
distinguish between rungs and sides, then there is actually a deformation from
the graph of the Möbius ladder to its mirror image. However, such a deformation
interchanges a pair of rungs of the ladder (which represent carbon-carbon double
bonds) with a pair of chains of carbons and oxygens contained in the sides of the
ladder. Interchanging carbon-carbon bonds with chains of carbons and oxygens
makes no sense from a chemical point of view.
Hence we will only allow deformations of the graph in Figure 6 which take
grey edges to grey edges and black edges to black edges. We draw the different
rungs with different types of segments, not because the rungs are different from
one another, just to make it easier to keep track of what’s going on later in our
argument.
Since we’re trying to prove that the Möbius ladder graph is topologically chiral,
we can assume the graph is completely flexible. With this flexibility, we can deform
the circle making up the sides of the ladder so that it lies in a plane, as illustrated
4. TECHNIQUE 2: 2-FOLD BRANCHED COVERS 317
Figure 7. We deform the Möbius ladder so that the grey circle

lies in a plane.
on the right in Figure 7. You should check for yourself that the graph can indeed
be deformed to look like this.
We have numbered the rungs in Figure 7 to help us keep track of their vertical
order. In particular, observe that in Figure 7, rung 1 is at the bottom, rung 2 is in
the middle, and rung 3 is at the top.
In Figure 8, we illustrate two copies of the Möbius ladder, one above the other.
To get the 2-fold branched cover of the Möbius ladder, we need to glue the top
and bottom copies together along the grey circle. In order to be able to fit the two
copies together, in the bottom graph we have pushed all of the rungs down below
the level of the plane containing the grey circle. Nonetheless, in the bottom graph,
rung 3 is still in the top position, rung 2 is still in the middle position, and rung 1
is still in the lowest position.
Figure 8. Both of these Möbius ladders have rung 3 on top, rung

2 in the middle, and rung 1 on the bottom.
Now we glue the top and bottom copies of the Möbius ladder together along
the grey circle in such a way that the endpoints of identical rungs are lined up as
illustrated on the left side of Figure 9. This is the 2-fold branched cover of the
Möbius ladder.
Figure 9. If we erase the grey circle in the 2-fold branched cover,

then we get a link.
Observe in the left image in Figure 9 that the 2-fold branched cover consists
of one grey circle and three black circles. Each black circle is the result of gluing
together a pair of identical rungs along their endpoints. By erasing the grey circle,
we obtain the link with three components illustrated on the right.
This link has three components but they are not the Borromean rings. Rather it
is a link with the property that each component is linked with each of the other two
components. Jon Simon proved that this link is topologically chiral by analyzing
the linking numbers of each pair of components. His proof required considering all
possible ways of orienting each of the components and computing all of the linking
numbers in each case. An outline of the proof is given in the Exercises. By proving
that this link is topologically chiral, Simon was able to conclude that the molecular
Möbius ladder is topologically chiral as well.
The topological chirality of the Möbius ladder is important not only because it’s
an interesting molecule, but because it can be used to prove that more complicated
molecules are topologically chiral. We see an example of this in the next section.
5. Technique 3: Chiral Subgraphs
This method was developed by the chemist Kurt Mislow to prove that the mol-
ecule triple layered naphthalenophane (illustrated in Figure 10) was topologically
chiral. A naphthalene molecule consists of a pair of adjacent hexagons and is the
primary ingredient in mothballs. The name “triple layered naphthalenophane” is
easy to remember because the molecule does indeed have three layers, each consist-
ing of a naphthalene.
The idea of this technique is to search for a subgraph contained in our molecule
which is known to be topologically chiral and use it to prove that the entire graph
is topologically chiral. For this method to work, the topologically chiral subgraph
must have the property that any deformation of the molecule to its mirror image
would necessarily take this subgraph to a mirror image of itself. In the case of triple
layered naphthalenophane, we will show that there is Möbius ladder hiding in the
molecule which has the required property.
We define the length of a path or a circle in a graph as the number of vertices
contained in it. Remember that each corner in a molecular graph represents a
carbon atom, and hence counts as a vertex. A key observation that we will use in
5. TECHNIQUE 3: CHIRAL SUBGRAPHS 319
Figure 10. Triple layered naphthalenophane.
our proof that triple layered naphthalenophane is topologically chiral is that the
graph contains a special circle C that is longer than any other circle in the graph.
This special circle is illustrated in grey in Figure 11. The length of the grey circle
is 35 (you should check this yourself). In the Exercises, you are asked to show
that this circle is indeed longer than any other circle in the graph. Observe that in
Figure 11 the three black edges in the center of the graph are the only edges which
have both vertices on the circle C. So both C and these three edges play a special
role in our graph.
Figure 11. C is the longest circle in triple layered naphthalenophane.
We will prove that triple layered naphthalenophane is topologically chiral by

contradiction. This means that we begin by assuming that the graph of triple
layered naphthalenophane can be deformed to its mirror image, assuming complete
flexibility. Observe that no matter how we deform a graph, we do not change the
number of vertices in a circle or a path. So our deformation of the graph to its
mirror image must take the circle C to a circle of the same length in the mirror
image, and hence must take the edges with both vertices on C to edges with both
vertices on the corresponding circle in the mirror image. This means that our
deformation taking triple layered naphthalenophane to its mirror image must also
take the colored graph on the left in Figure 12 to its mirror image.
However, the graph in Figure 12 can be deformed to a Möbius ladder by smooth-
ing out the circle C and spreading out the three grey edges around C. In the illus-
tration on the right in Figure 12 we erase all of the corners in order to make the
Möbius ladder look like the ones we saw in the previous section. Notice that the
rungs of our Möbius ladder are black and the sides are grey as they were in our proof
Figure 12. C together with these three edges can be deformed to

a Möbius ladder.
that the molecular Möbius ladder is topologically chiral. Since we know that there
is no deformation of this colored Möbius ladder to its mirror image, the graph of
triple layered naphthalenophane also could not be deformed to its mirror image. In
particular, this means that triple layered naphthalenophane is topologically chiral.
6. Some Graph Theory
The next technique that we introduce uses the graphical structure of a molecule
to prove that no matter how the structure is situated in 3-dimensional space, it will
be topologically chiral. In order to explain this technique, first we need to learn
some terminology from graph theory.
An abstract graph is a collection of vertices, together with a collection of edges
joining some of the vertices. In particular, an abstract graph is independent of how
the graph is situated in space, just like an abstract gluing of a surface is independent
of how the surface is situated in space. If we are interested in talking about how
a graph is sitting in space, we refer to its position in space as an embedding of the
graph. In the language of Chapter 3, if we are talking about an abstract graph,
then we are considering the graph intrinsically. Whereas if we are talking about an
embedding of a graph in space, then we are considering the graph extrinsically.
In Chapters 1–8, we saw that when we study 3-manifolds we always think of
them intrinsically, because as 3-dimensional beings we are incapable of knowing
anything about how our 3-dimensional space could be sitting inside of a higher-
dimensional space. Similarly, A. Square always studies 2-manifolds intrinsically,
and A. Dash always studies 1-manifolds intrinsically. We can consider 2-dimensional
spaces extrinsically and A. Square can consider 1-dimensional spaces extrinsically.
But poor A. Dash can only consider 0-manifolds extrinsically, which is pretty boring
since every 0-manifold is just a point.
In Chapters 9–11, we saw that knots have to be studied extrinsically, because
intrinsically every knot is just a circle. Molecular graphs are noteworthy because
they can either be studied extrinsically as objects in space as well as intrinsically
just in terms of what is connected to what. So far, in this chapter and the previous
one, we have only been considering molecular graph extrinsically. The fourth tech-
nique that we will present for proving that a molecule is topologically chiral treats
molecular graphs intrinsically.
6. SOME GRAPH THEORY 321
The valence of a vertex in a graph is the number of edges that contain that
vertex. The distance between two vertices in a graph is the fewest number of edges
contained in a path from one of the vertices to the other. Two vertices are adjacent
if there is an edge in the graph joining them.
Before we introduce more terminology, let’s practice using these definitions on
the Möbius ladder graph illustrated in Figure 13. If we want to talk about the
abstract graph of the Möbius ladder, then we ignore the twist in the front of the
graph. On the other hand, if we are considering the embedding of the graph, then
this twist is a crucial piece of information. To determine the valence of a vertex, we
only need to look at an abstract graph. For the Möbius ladder graph, the valence
of each vertex is 3. We can further observe that vertex 1 in the Möbius ladder is
adjacent to vertices 2, 4, and 6; whereas, vertices 1, 3, and 5 are each a distance 2
apart.
Figure 13. An embedding of the Möbius ladder graph.
If you’re confident about the terminology we’ve introduced so far, then it’s time
to introduce a little more.
Definition. An automorphism of an abstract graph is a permutation of the
vertices which takes each pair of adjacent vertices to a pair of adjacent vertices. An
automorphism of a molecular graph is also required to take atoms of a given type
to atoms of the same type.
When we look for automorphisms of a graph, the embedding of the graph
is not important. What is important is which vertices are connected to which.
As an example, we again consider the Möbius ladder graph in Figure 13. The
permutation which interchanges vertices 2 and 6 and interchanges vertices 3 and 5
is an automorphism. The fact that this automorphism can’t be achieved just by
turning the graph upside down in space is irrelevant, since we are only considering
the abstract graph. Observe that the permutation which just interchanges vertices
2 and 5 and doesn’t move any other vertices is not an automorphism of Figure 13
because vertex 2 is adjacent to vertex 3, while vertex 5 is not adjacent to vertex 3.
The following properties of automorphisms may seem obvious. But proving
them from the above definitions is worth doing in order to check your understanding.
You’ll have the opportunity to prove them in the Exercises.
Properties of Automorphisms:
(1) Any automorphism of an abstract graph takes vertices of a given valence

to vertices of the same valence.
(2) Any automorphism of an abstract graph takes a pair of vertices which are
a certain distance apart to a pair of vertices which are the same distance
apart.
All right, now we’re ready to learn one final definition.

Definition. The order of an automorphism is the smallest number n such
that doing the automorphism n times takes each vertex back to its original position.
For example, the automorphism of the Möbius ladder in Figure 13 which in-
terchanges vertices 2 and 6 and vertices 3 and 5 has order 2, because if we do the
automorphism twice, then every vertex returns to its original position; and if we
only did the automorphsim once, every vertex would not stay in its original posi-
tion. The identity automorphism (also known as the trivial automorphism) which
doesn’t move any vertex, is the only automorphism that has order 1.
In Figure 14 we illustrate two very important graphs: the complete graph on
five vertices K5 , and the complete bipartite graph on two sets of three vertices K3,3 .
Note that all of the vertices in the figure are represented by dots. In particular, the
places where edges appear to intersect in Figure 14 are neither vertices nor actual
intersections. We draw the graphs this way because we are considering them as
abstract graphs, and hence we don’t want to specify a particular embedding of the
graphs in space.
Figure 14. The graphs K5 and K3,3 .
The graphs K5 and K3,3 are special because they are the “smallest graphs”
which cannot be embedded in a plane without any edges intersecting. We are in-
terested in embeddings of graphs in a plane because such an embedding is its own
mirror image, where the mirror is the plane itself. This means that any embed-
ding of a graph in a plane is topologically achiral. In fact, such an embedding is
geometrically achiral.
7. Technique 4: A Combinatorial Approach
The following theorem, whose proof uses advanced techniques in topology, al-
lows us to prove that many abstract graphs have no topologically achiral embedding
in 3-dimensional space.
Theorem 13.3. If a graph has no order 2 automorphisms and contains either

of the graphs K5 or K3,3 , then any embedding of the graph in R3 is topologically
chiral.
7. TECHNIQUE 4: A COMBINATORIAL APPROACH 323
Figure 15. The molecule ferrocenophane.
We will now use Theorem 13.3 to show that the molecule ferrocenophane (il-
lustrated in Figure 15) is topologically chiral. To help you understand this figure,
you should note that the Fe represents an iron atom, the O represents an oxygen
atom, and the rest of the atoms are carbons.
While it is certainly doesn’t look like it at first glance, ferrocenophane contains
the complete graph K5 . To see this, we first observe in Figure 14 that as an abstract
graph, K5 is just a set of five vertices together with edges joining every pair of
vertices. We illustrate a K5 which is contained in ferrocenophane in Figure 16. In
the K5 on the right in Figure 16, only the dots with numbers on them represent
vertices. Though the corners in the graph of ferrocenophane represent carbon
atoms, in this K5 the corners have no significance other than to help us see that
K5 is contained in the molecule.
Figure 16. Ferrocenophane contains the graph K5 .
In order to use Theorem 13.3, we must also show that ferrocenophane has no
order 2 automorphisms. We will actually show that the only automorphism that
ferrocenophane has is the identity automorphism. We begin by letting f be an au-
tomorphism of the abstract graph of ferrocenophane. Recall that an automorphism
of a molecular graph is required to take atoms of a given type to atoms of the same
type. This means that f must take the single oxygen atom to itself as illustrated
in Figure 17. If an automorphism takes a vertex to itself, we will say it fixes the
vertex.
Figure 17. f must fix the single oxygen atom.
Figure 18. f must fix the carbon atom adjacent to the oxygen.
Now it follows from the Properties of Automorphisms that f must take a pair
of adjacent vertices to a pair of adjacent vertices. In particular, f must fix the only
carbon which is adjacent to the oxygen atom, as shown in Figure 18.
Next observe that there are two vertices in addition to the oxygen atom which
are also adjacent to this carbon. However, one vertex has valence 2 and the other
vertex has valence 4. Recall that we know from the Properties of Automorphisms
Figure 19. f must fix each of these vertices.

8. EXERCISES 325
that an automorphism must preserve the valence of a vertex. Hence f must take
each of these vertices to itself, as illustrated in Figure 19.
We can continue arguing in this way, going through the vertices one after the
other, to conclude that in fact every vertex of ferrocenophane must be fixed by the
automorphism f . But this means that f is actually the identity automorphism.
Thus we have shown that the identity is the only automorphism of ferrocenophane.
In particular, ferrocenophane cannot have an automorphism of order 2.
Since we have now shown that ferrocenophane satisfies both of the hypotheses
of Theorem 13.3, we can conclude that not only is ferrocenophane topologically
chiral, but any molecule which has the same abstract graph as ferrocenophane
must be topologically chiral as well.
In fact, Theorem 13.3 has nothing to do with a particular embedding of a graph
in R3 . Once we check that the conditions of Theorem 13.3 are satisfied for a graph,
as we did for ferrocenophane, then even if we change the embedding of the graph
in space it will still be topologically chiral. This seems surprising since topological
chirality has to do with comparing an embedding of a graph with the mirror image
embedding of the same graph. This observation motivates us to make the following
definition.
Definition. A graph is said to be intrinsically chiral if every embedding of
it in R3 is topologically chiral.
The idea of this definition is that a graph is intrinsically chiral if its topological
chirality is an intrinsic property of the graph, rather than depending on the extrinsic
topology of the graph in R3 . Using this terminology, we now restate Theorem 13.3
as follows.
Theorem 13.4. If a graph has no order 2 automorphisms and contains either
of the graphs K5 or K3,3 , then it is intrinsically chiral.
In the Exercises, you will use Theorem 13.4 to show that some other molecules
are intrinsically chiral.
8. Exercises
1. In this exercise you will follow the steps below to complete the proof that
the molecular Möbius ladder is topologically chiral by proving that the link in
Figure 17 is topologically chiral.

(a) Assign an orientation to each component of the link and compute the
linking numbers lk(J, K), lk(J, L), and lk(K, L).
(b) Compute the linking numbers of the mirror image of the oriented link.
(c) Show that changing the orientation of a single component in the mirror
image will change the number of negative linking numbers by an even
number.
(d) Use the above steps to prove that the link in Figure 20 is topologically
chiral.
2. Prove that the circle C in Figure 11 is longer than any other circle in the
graph. Hint: Any longest circle must travel through all three pairs of adjacent
hexagons.
3. Illustrate a deformation from the graph on the left of Figure 12 to the colored
Möbius ladder.
4. Consider a graph like triple layered naphthalenophane, but with a row of four
hexagons on each of the three layers instead of just two. Prove that this graph
is topologically chiral.
5. Prove that the graph K3,3 is not intrinsically chiral by finding a topologically
achiral embedding of K3,3 in R3 .
6. Figure 21 illustrates the Simmons-Paquette molecule. All of the atoms in this
molecule are carbons except for the three oxygens labeled with O’s. Prove that
this molecule is intrinsically chiral.
7. The molecular sphere in Figure 22 is called a fullerene. The figure illustrates a

fullerene together with a cap which has 3-fold symmetry. That is, rotating the
molecule around a vertical axis by 120◦ will not change the way it looks. In
particular, the way that the cap is attached in the back of the molecule is the
same as on the two sides. Prove that this molecule contains either K5 or K3,3 ,
and contains a “special circle” such that any automorphism of the molecule
takes that circle to itself.
8. Use Exercise 7 to prove that the molecule illustrated in Figure 22 is intrinsically
chiral.
9. Prove that the protein illustrated in Figure 23 is topologically chiral. You may
use the fact that any deformation of the protein to its mirror image must send
the N to the N in the mirror image and the C to the C in the mirror image.
8. EXERCISES 327
10. Prove that any automorphism of a graph takes vertices of a given valence to
vertices of the same valence.
11. Prove that any automorphism of a graph takes a pair of vertices which are a
given distance apart to a pair of vertices which are the same distance apart.
12. Prove that a 3-rung molecular Möbius ladder has an automorphism of order 6.
13. Let C denote a circle with six vertices. Prove that the order of any automor-
phism of C divides 6.
14. Find topogically achiral embeddings of the abstract graphs K5 and K6 illus-
trated in Figure 24. Note there are no intersections between the edges. We draw
them with intersecting segments because we don’t want to specify a particular
embedding of the graphs in space.
15. Figure 25 illustrates a Möbius ladder with four rungs both as an embedded
graph in the form of a Möbius strip (on the left) and as an abstract graph
(on the right). In the picture of the abstract graph, note that the edges do not
intersect and there is no vertex in the center of the picture. Draw an embedding
of this graph which is topologically achiral.
16. Show that the molecule illustrated in Figure 26 contains the abstract graph of
a Möbius ladder with four rungs.
17. Prove that the graph illustrated in Figure 26 is topologically achiral.

CHAPTER 14
The Topology and Geometry of DNA
Topics:
• The biological structure of DNA
• Supercoiling
• The linking of the two backbones
• The average writhe of the axis
• The twist of one backbone around the axis
• The equation Lk = T w + Wr
• How to measure supercoiling
• Replication
• Site specific recombination
• Rational tangles
• Tangle operations
• The tangle model of site specific recombination
• Solving the mystery of Tn3 resolvase
1. Synthetic versus Biological Molecules
In Chapters 12 and 13 we learned that organic chemists have created molec-

ular versions of knots, links, Möbius ladders, and other interesting 3-dimensional
graphs. Synthesizing such topologically complex molecules is no mean feat. In fact,
despite many years of trying to synthesize a variety of different molecular knots, the
only knots which have been synthesized as of 2015 (without using DNA or other
biologically based polymers) are the five illustrated in Figure 1.
Figure 1. These are the only knots that have been synthesized by chemists.
In fact, most molecules that are created in a laboratory are small and relatively
rigid, so that their geometry is important, but their topology is not. If you have
ever looked at the leaflets that come with medications, you might have noticed that
the chemical diagrams on them are pretty simple. Perhaps at some point, there
329
330 14. THE TOPOLOGY AND GEOMETRY OF DNA
might be a revolutionary new medicine synthesized in the form of a knot. But in

view of the difficulty involved, it’s unlikely to happen any time in the near future.
In contrast with molecules designed and created from scratch by chemists,
biological molecules like DNA1 and proteins are sufficiently long and floppy that
they can become tangled up all by themselves, just like your headphones get tangled
when you stuff them into your pocket. In fact, while organic chemists are trying
to synthesize molecules which are more knotted, molecular biologists are trying
to better understand how DNA keeps from becoming too knotted. This means
that topology and knot theory are not only important tools for understanding the
chirality of synthetic molecules (as we saw in Chapter 13), they are also important
tools for understanding the structure and behavior of biological molecules that
occur naturally.
2. The Biology of DNA
Imagine if you had the most important job on the planet. No, not being the
President of the United States. That’s way too minor. The job is making sure that
life continues. The job description involves designing and operating a system of
reproduction that passes on complex genetic information while introducing genetic
diversity and allows cells to duplicate and grow during reproduction and throughout
the life cycle.
While this job seems much too difficult and important for you or me to take on,
DNA is happy to do it, and its success at the job is entirely due to its knowledge of
topology and geometry. There is no doubt that the job is stressful, but DNA even
has its own system to relieve its stress without having to join a gym. The rest of
this chapter will be devoted to learning a few of the secrets of its success, though
we won’t include enough information that DNA might worry that we’re thinking
of trying to take over its job. Rather, we’ll learn just enough for us to appreciate
how incredibly smart and competent DNA is.
In this section and the next we’ll learn a little bit about the biological aspects
of the job of DNA. Then, starting in Section 4, we’ll learn about the topology and
geometry that DNA uses to do its job.
The most important information that DNA has is coded in a sequence of letters,
known as the genetic code. The more complicated the organism, the longer the
sequence needs to be to keep track of all the information. For humans, this sequence
contains roughly 3 billion letters. The letters making up the sequence are chosen
from among A, T, C, and G. These letters stand for four bases known as adenine,
thymine, cytosine, and guanine, which are attached to a long string of alternating
sugar and phosphate molecules making up a strand of DNA.
However, DNA is so clever and good at its job that right from the beginning
it instituted a system of having a backup copy of everything. In particular, there
are actually two strands of DNA wound together, which each keep track of all of
the important information. But the sequence of bases on the two strands aren’t
identical. Such clever DNA would think that’s much too simple a scheme. Rather,
each base A on one strand bonds with the base T on the other strand, and each
base C on one strand bonds with the base G on the other. So if the sequence
ATGGA occurs on one strand, then it must be bonded with the sequence TACCT
1 In case you were wondering, DNA is short for DeoxyriboNucleic Acid.
2. THE BIOLOGY OF DNA 331
on the other strand. Since the sequence of A’s, T’s, C’s, and G’s that occurs on one
strand determines the sequence of A’s, T’s, C’s, and G’s that occurs on the other
strand, we only need to list one of these two sequences in order to specify the entire
genetic code. But if somehow there are bases on one strand that don’t bond with
the corresponding bases on the other string, then the DNA knows right away that
some funny business has happened on one of the strands, and it can try to figure
out who made a mistake and fix it.
Figure 2 illustrates the sugars, phosphates, and bases on two strands of a small
segment of DNA. Each sugar is denoted by the letter S and each phosphate is
denoted by the letter P . The bases that are bonded together on a DNA molecule
are called base pairs. For example the segment of DNA illustrated in Figure 2 has
five base pairs. Looking at the figure, we can think of the base pairs as rungs on a
molecular ladder.
Figure 2. The sequence ATGGA on one strand binds with the

sequence TACCT on the other strand.
Occasionally, one can find a cell in which there is a DNA molecule that only
has one strand instead of two. In order to distinguish this single stranded form of
DNA from the usual DNA molecules, we refer to the two stranded form as duplex
DNA. Figure 3 illustrates how the two strands of duplex DNA are twisted together
in the form of a right-handed double helix.
Figure 3. The two strands of DNA are twisted together in a

right-handed double helix.
In Figure 3, the bonds between the bases are illustrated as vertical bars, so
that it looks more like a ladder than in the close up picture in Figure 2. If you
count the number of vertical bars in one full turn of the helix, you will find that
there are about 10.5. In general, duplex DNA is said to be in a relaxed state if the
double helix makes one complete right-hand turn for every 10.4–10.5 base pairs.
This means that on average the strands of the duplex DNA will make two complete
turns for every 21 = 2 × 10.5 base pairs. If there are more than 10.5 base pairs per
turn or fewer than 10.4 base pairs per turn, then the DNA molecule is said to be
stressed.
We call the two strands the backbones of the DNA molecule. This terminology
may be confusing because you would naturally expect that DNA should have only
one backbone, just like a person or an animal or even a fish. Imagine you go out
to dinner with some friends, and you order a whole fish to share. When it arrives
steaming, smelling delicious, and looking at you with one of its eyes, you confidently
offer to cut the flesh off the bones and serve it to your friends in nice neat portions.
However, when you cut the fish open, you discover that there are actually two
backbones that are twisted together in a double helix with the flesh stuck between
them. Getting the flesh out without causing bits of fish to start flying around seems
virtually impossible. So you say “just kidding” and pass the platter to someone
else to cut up.
There are two possible forms for a DNA molecule: closed up in a loop or open
like a wiggly arc in space. DNA of the first type is said to be circular, while DNA
of the second type is said to be linear. Note that circular DNA is not necessarily
planar. In fact, it can be quite wiggly like some of the complicated unknots that
we saw in Chapter 9. For circular DNA, the twisting of the double helix is trapped
in the molecule unless one or both strands of the molecule are cut. While the DNA
of organisms such as bacteria and viruses is normally circular, this is not the case
for human DNA. You might think this means that duplex DNA easily becomes
untwisted. However, even when DNA isn’t circular, the ends are usually fixed to a
membrane so that the double helix cannot untwist. Thus, to make it easier to study
the topology of DNA, we imagine that all DNA is circular even though technically
this is not always the case.
3. The Problem of Packing DNA into a Cell
If the above job description for DNA did not sound challenging enough, there
are further difficulties which you only realize when you look at the details of the job
more closely. The problem is that the length of each DNA molecule ranges from
1,000 to 500,000 times the diameter of the cell which contains it. In order to help
us visualize the problem of packing this much DNA into a cell, let’s imagine scaling
the cell up so that it’s the size of a basketball. Then human duplex DNA would be
a 120 mile long thread made from two super thin threads (the backbones) twisted
together. If this much thread were randomly stuffed into a basketball, it would
likely result in a tangled mess that would be impossible to undo. Think about how
tangled up your headphones get when you stuff them into your pocket, and then
imagine how hard it would be to untangle them if they were 120 miles long and
were inside of a basketball.
One of the most important aspects of the job of DNA is to enable each cell
to split into two identical cells. This is necessary for any organism to grow and
create offspring. When cells divide, the DNA molecules within the cells split into
two identical DNA molecules through a process known as replication. This involves
the two strands of the double helix separating and each creating a new strand of
DNA which binds to it. Since the sequence of bases in one backbone determines the
sequence of bases in the other backbone, one strand is all that’s needed to create a
duplex DNA molecule which is a copy of the original. It’s impressive that DNA has
the capability to duplicate itself in this way, since no other molecules are known
4. SUPERCOILING 333
to have this capability. But in order for DNA to be able to replicate, it can’t be
tangled up randomly in the cell like 120 miles of thread stuffed into a basketball.
Another important aspect of the job of DNA is creating genetic diversity. This
means it’s not enough for the DNA to just copy the genetic code of one of the
parents to create the genetic code of an offspring. Rather each time an offspring is
created, the genetic codes of the parents are combined in a new way. Have you ever
wondered how you can be so different from your siblings when you have the same
parents? How is it possible for you to have light brown hair when your mother
has black hair and your father has strawberry blond hair? Even if you imagine the
genes for hair color are like paint, if you mix black and yellow paint, you are likely
to get an ugly shade of green rather than a normal light brown hair color. Why
can’t a mother with black hair and a father with orange hair produce a child with
black and orange striped hair? Cats seem to be able to do this, so why can’t we?
A key component to answering all of these questions is that during reproduc-
tion, segments of DNA are rearranged to create new genetic combinations in a
process known as recombination. One way this works is that distant sections of
DNA with identical codes are brought together and the endpoints are cut and in-
terchanged. But how can two identical sections of DNA end up near one another
if the DNA is randomly packed into a cell like a 120 mile long thread stuffed into
a basketball? Furthermore, this cutting and reattaching has to be done without
causing the DNA to become knotted or tangled so badly that it makes it impossible
for it to do the other aspects of its job.
You may recall from our discussion of the unknotting number in Chapter 10
that we wanted to change a crossing in order to create a new knot whose minimal
crossing number was less than that of the original. But there was no good way
to figure out what projection to use and which crossing to change. In the case of
recombination, somehow the DNA is able to know how to cut and reattach to make
the tangling better rather than worse. This means DNA has to be able to think
globally but act locally. But how is it able to do this when it’s packed so tightly
into a cell?
4. Supercoiling
As we saw in the last section, DNA needs a good way to organize itself so
that it can be packed into a cell without uncontrolled tangling occurring. To un-
derstand how DNA solves this problem, let’s consider a similar problem that yarn
manufacturing companies face. If yarn were sold as a tangled mess stuffed inside
of a basketball, nobody would buy it. Instead, yarn is wound up into forms known
as cakes, skeins, and hanks. Though must people call all three forms a skein. In
fact, a cake is what you would get if you wound yarn around a stick, a skein is an
oblong shape where the yarn is pulled out from the center, and a hank is illustrated
in Figure 4. To make a hank, you wind the yarn around two pegs or two arms
to create a loose coil. Then you twist the two sides of the coil tightly together.
Finally, you fold the twisted up yarn in half and again twist the two sides together.
This gives the yarn a nice compact form which holds its shape and doesn’t tangle.
In order to fit neatly into a cell, DNA needs to have a compact shape which
holds together and doesn’t tangle, just like yarn. Duplex DNA is naturally wound
up in a double helix, as illustrated for circular DNA on the left side of Figure 5. If
Figure 4. These are the steps to winding yarn into a hank.
two parts of the double helix come together and twist, as illustrated on the right
side of Figure 4, the DNA will become more compact, which will allow it to fit more
easily into a cell without tangling. We refer to twists like those on the right side of
Figure 5 as supercoils.
Figure 5. When duplex DNA twists up like this it becomes com-

pact and doesn’t tangle, just like a hank of yarn.
In order to see for yourself how supercoiling occurs, you can do the following
experiment with two pieces of rope and a friend. Grab the end of one rope in
each hand and tell your friend to hold the other two ends together in one hand.
Then begin twisting up your ends so that what you get looks like a double helix
as illustrated on the left of Figure 6. Now keep twisting the two ropes together
tighter and tighter. Eventually, it will be harder to keep twisting because you are
creating stress on the ropes. If you keep twisting anyway, two pieces of the double
helix rope will come together and start to supercoil, as illustrated on the right side
of Figure 6. This supercoiling relieves some of the stress on the twisted ropes. The
more you twist up the two ropes, the more supercoiled it will become. You and
your friend can forcibly remove the supercoils by moving away from each other to
pull the ropes tight. But as soon as you come back together so that the ropes are
no longer taut, the supercoils return. For this reason, we say that the supercoiling
of the two ropes is stable.
Next, imagine making a loop out of a pair of ropes which are loosely twisted up
in a double helix but do not supercoil even when the ropes are not taut. You could
then turn part of the double helix over to add a supercoil as illustrated on the right
of Figure 7. But this supercoil will go away if you lift the ropes up. Hence, in this
case, we say that the supercoiling of the two ropes is unstable.
Supercoiling is certainly one of DNA’s most brilliant ideas. Maybe it should
even win a Nobel Prize for this idea. Supercoiling keeps the strands neat and
5. VISUALIZING DNA AS A CIRCULAR RIBBON 335
Figure 6. If you keep twisting ropes together, they will eventually

become supercoiled.
Figure 7. We can create unstable supercoiling by turning over

part of the twisted rope.
organized, and packed efficiently in a cell. It allows distant parts of the DNA to be
close together to facilitate recombination. It also facilitates the unwinding of the
two strands to allow replication to take place.
In the next six sections, we’ll learn about the geometry and topology of duplex
DNA as a result of supercoiling.
5. Visualizing DNA as a Circular Ribbon
In order to best describe the geometry and topology of DNA, we visualize a

duplex DNA molecule as a twisted ribbon whose edges are the two backbones. We
illustrate a short piece of such a ribbon in Figure 8. If you imagine duplex DNA
this way, then the center of the ribbon is what is known as the axis of the DNA.
Remember from Section 2 that we define duplex DNA to be in a relaxed state if
the double helix makes one complete right-hand turn for every 10.4–10.5 base pairs.
When an unknotted DNA molecule is in such a state, its axis will be roughly planar,
just like the rope we saw on the left side of Figure 7. However, if a knotted DNA
molecule is in a relaxed state, its axis still cannot be planar since a knot cannot lie
in a plane.
For circular DNA, you may wonder if the two ends of the ribbon can be joined
so that we get a Möbius strip or some other non-orientable surface. To answer
this question, let’s remember that the backbones of a DNA molecule consist of
alternating sugar and phosphate molecules, with one base attached to each sugar.
When sugars and phosphates bond together, each sugar has a special site on one
end denoted by a 3 and a special site on the other end denoted by a 5 . To
visualize this, imagine that each sugar molecule is a white extension cord, with a 3
designating the socket on one end and a 5 designating the plug on the other end,
Figure 8. We visualize DNA as a ribbon that twists around its

central axis.
Figure 9. We think of a sugar molecule as an extension cord with

a 3 designating the socket and a 5 designating the plug.
Next imagine that each phosphate molecule is a dark grey extension cord with
a 3 designating its socket and a 5 designating its plug. The 5 site on a phosphate
molecule bonds to the 3 site on a sugar molecule, and the 3 site on a phosphate
molecule bonds to the 5 site on a sugar molecule. Thus the backbone of a DNA
molecule can be pictured as a long chain of alternating white and grey extension
cords, where the plug of one cord fits neatly into the socket of the next cord, and
so on as illustrated in Figure 10.
Figure 10. We can think of a backbone of a DNA molecule as a

chain of alternating white and grey extension cords.
Observe that no matter how many extension cords we put together, one end
of the chain will always be a socket and the other end of the chain will always be
a plug. For DNA, this means that one end of each backbone will be a 3 , and the
other end will be a 5 . Furthermore, for a duplex DNA molecule, one backbone
ends in 3 while the adjacent backbone ends in a 5 as illustrated in Figure 11.
Now let’s return to our visualization of DNA as a twisted ribbon, but this time
we label the ends of each edge of the ribbon with a 3 and a 5 as in Figure 12. As
you can see in the figure, the 3 and the 5 on the black edge, and the 3 and the 5
on the grey edge are on opposite ends of the ribbon.
6. THE LINKING NUMBER OF THE BACKBONES 337
Figure 11. At each end of duplex DNA, one backbone ends in 3

while the other backbone ends 5 .
Figure 12. The 3 and the 5 on the black edge, and the 3 and
the 5 on the grey edge are on opposite ends of the ribbon.
In order to close the ribbon to make a loop, the 3 site at the end of one of
the backbones has to bond with a 5 site at the other end of the ribbon. As we
can see in Figure 12, the only way this can happen is if the ends of the grey strand
join together and the ends of the black strand join together. So each backbone of
a closed up duplex DNA molecule forms its own loop. This means that the DNA
ribbon cannot be a Möbius strip.
6. The Linking Number of the Backbones
We saw above that if we think of a duplex DNA molecule as a circular ribbon

twisting around its axis, then each backbone forms its own loop. Furthermore,
because of the twisting of the strands in the double helix, the two backbones will
be the components of a non-trivial link. Hence we can think of a circular DNA
molecule either in terms of the knot (or unknot) in the axis of the ribbon or in
terms of the link between the two backbones. In Figure 13, the knotted axis of the
DNA molecule is illustrated with a thick grey line and the linked backbones are
illustrated with thin black lines.
We arbitrarily give the axis of the DNA an orientation, and then orient each
of the backbones so that they go in the same direction as the axis (see Figure 14).
Now that the backbones are oriented, we can keep track of their linking with the
linking number that was introduced in Chapter 10. In particular, we make the
Figure 13. We can think of this DNA molecule in terms of either

the knotted axis or the link of the two backbones.
Figure 14. We put an orientation on the axis, and then orient

each of the backbones in the same direction as the axis.
Definition. Consider an oriented projection of the backbones K1 and K2 of

a DNA molecule. The linking number Lk is defined by the formula
⎡⎛ ⎞ ⎛ ⎞⎤
1
Lk = ⎣⎝ positive crossings ⎠ − ⎝ negative crossings ⎠⎦ .
2 between K and K between K and K
1 2 1 2
The most important thing that we learned about the linking number in Chapter
10 is that it’s a topological invariant, which means that deforming the link will not
change its linking number. For example, suppose that the axis of our DNA is an
unknot like the one on the left in Figure 5. Putting some supercoils in the axis,
as illustrated on the right, does not change the linking number between the two
backbones. Changing the orientation of the axis also does not change the linking
number, since it would change the orientation of both backbones.
The linking number Lk keeps track of important topological information about
a DNA molecule, which we will return to later after we introduce two other quan-
tities which give us some important geometric information about the molecule.
7. THE AVERAGE WRITHE OF THE AXIS 339
7. The Average Writhe of the Axis
Recall the following definition from Chapter 10.

Definition. The writhe w(L) of an oriented knot projection K is given by
the formula

w(K) = − .
This number does not depend on the orientation of the knot. However, as we
saw in Chapter 10, the writhe is only an invariant for knot projections which are
alternating and reduced. In particular, the writhe may change when we deform a
knot, and it may even change when we consider a projection from a different angle.
Hence the writhe is not a topological property of a knot. But we will see how to use
the writhe to get a geometric property of DNA.
Recall from Chapter 3 that an extrinsic property of an object in space is geo-
metric if it doesn’t change when the object undergoes a rigid motion. We see as
follows that the writhe is not a geometric property because it can change when
a knot is rotated. Consider the unknot obtained by wrapping a rigid steel band
around two perpendicular soda cans as illustrated in Figure 15. The projection that
we see on the left side has no crossings and thus has writhe equal to 0. However,
if we remove the cans and rotate the the steel band by 45◦ , we’ll get the projec-
tion illustrated on the right side of Figure 15. This second projection has a single
negative crossing regardless of how it’s oriented (you should check this), and thus
has a writhe of −1. In order to convince yourself that the second projection can be
obtained by rotating the first projection by 45◦ , you could try letting your index
fingers play the role of the cans and a rubber band wrapped around your fingers
play the role of the steel band.
Figure 15. These two projections of a rigid loop have different writhe.
Thus sadly the writhe is neither a topological nor a geometric property of

an arbitrary knot, and hence might seem pretty useless. However, the following
definition turns the writhe of a rigid loop into a useful geometric property.
Definition. The average writhe of a rigid loop, denoted by Wr, is the av-
erage of the writhes over all possible projections of the loop.
We want to emphasize that the average writhe is only defined for rigid loops.
That is, we’re not considering the writhe over all projections of any deformation
of a loop. Rather, we imagine that the loop is made of steel. We can turn it in
different ways to get different projections, but we can’t deform it at all.
The notion of taking the average of infinitely many different projections might
seem a bit strange. However, if you have taken calculus, you might have seen this
kind of average as an application of integration. But don’t worry if you haven’t
taken calculus. Since Wr is an average over all possible projections, the quantity
Wr is unchanged when a loop is rigidly rotated. This means that Wr is indeed a
geometric property of rigid loops.
Unfortunately, for most loops, Wr can only be computed with the aid of a
computer. However, to get a feel for it, we can compute the value of Wr for some
very simple examples by hand. To begin with let’s imagine a planar circle with no
crossings. To compute its average writhe we are supposed to consider the writhes
of all possible projections. But if we consider a plane perpendicular to the plane
containing the circle cutting the circle in half, one semi-circle will be on top of the
plane and the other semi-circle will be below the plane. Thus the projection of the
two semi-circles will be on top of each other so that the projection looks like a line
segment. In Figure 16, both semi-circles project to the grey line segment in the
plane. This type of projection is not allowed because it doesn’t have crossings that
we can count. You can check that any other projection of our planar circle will
have zero crossings. Thus we can conclude that the average writhe is Wr = 0.
Figure 16. Both semi-circles project to the grey line segment.
Next let’s determine the average writhe for the steel band illustrated in Fig-
ure 15 (again a rubber band around your index fingers may be helpful). Observe
that for some projections, the loop will have no crossings at all, while for others
it will have a single negative crossing. In fact, these are the only possibilities for
legitimate projections, that is where one arc does not lie directly on top of another.
Thus, for every projection of the steel band, the writhe is either 0 or −1. This tells
us that the average writhe for the loop is somewhere between −1 and 0. Though
we can’t assign it a precise value, we can see from this example that in general, the
average writhe of a loop is not an integer, even though for any given projection the
writhe is an integer. This makes sense since the average of a collection of integers
is not always an integer. For example, the average of the numbers 1 and 2 is 1.5.
8. The Twist of a Backbone around the Axis
We have now seen that for any closed loop of duplex DNA, we can obtain
information about the geometry of the molecule by computing the average writhe
Wr of its central axis (though we may need the help of a computer to do so). Also,
we can obtain topological information about the molecule by computing the linking
number of the two backbones oriented in the direction of the axis. But we need to
introduce one more quantity before we can completely describe the geometry and
topology of circular duplex DNA.
Definition. The twist of a circular duplex DNA molecule, denoted by T w,
is the total amount that one of the backbones of the DNA winds about the central
axis.
8. THE TWIST OF A BACKBONE AROUND THE AXIS 341
At first glance, the amount that one backbone winds around the axis seems like
it should equal the linking number between the two backbones. To see why this is
not always the case, let’s consider the ribbon illustrated in Figure 17. Observe that
for this ribbon, neither backbone twists around the central axis, and so T w = 0.
However, you should check that after orienting the two backbones in the same
direction, they are indeed linked with Lk = −1.
Figure 17. For this ribbon, Lk = −1 and T w = 0.
The following visualization may help you understand the twist better. We start
by drilling a hole in the center of a penny, and threading the oriented axis of the
DNA through the hole from the tail side of the penny to the head side of the penny.
You can picture the penny as a bead on a necklace where the chain is the DNA
axis.
Now imagine that the top of Lincoln’s head is touching one of the backbones
of the DNA, as illustrated in Figure 18. We slide the penny up along the axis in
the direction of the arrow, always keeping the top of Lincoln’s head against the
backbone. In Figure 18 we see that as the penny moves up the axis, Lincoln will
have to rotate counterclockwise so that the top of his head remains on the backbone.
Figure 18. The twist measures the total rotation of Lincoln’s

head as the penny moves along the axis.
The twist T w will measure the total rotation of Lincoln’s head as the penny
goes all the way around the axis until it returns to the place on the axis where
it started. Each full (360◦ ) counterclockwise rotation contributes +1 to the twist,
and each full clockwise rotation contributes −1 to the twist. Note that the penny
may return to where it started on the axis without Lincoln’s head having the same
orientation it had at the beginning. If this is the case, the twist will not be an
integer.
We compute the twist for the simple ribbons illustrated in Figure 19. In the two
illustrations on the left, the axis of the ribbon is in a plane and the backbones don’t
wind around the axis at all. So in these cases, T w = 0. In the third illustration
from the left, the axis is again in a plane, but each backbone winds negatively
around the axis. Thus in this case, T w = −1. Finally, in the example on the right,
the axis is not in a plane, but even so the backbones don’t wind around it. So again
we have T w = 0.
Figure 19. Some ribbons and their twists.
What is interesting to note about these examples is that the two ribbons il-
lustrated on the left of Figure 19 are topologically equivalent, and the two ribbons
illustrated on the right are topologically equivalent. However, it is easier to find a
deformation between the left two ribbons than between the right two ribbons. You
can do the following experiment with a belt to convince yourself that in fact the
right two ribbons are topologically equivalent. When your belt is not around your
pants (or your waist), put a negative twist in it and then buckle it so that it looks
like the picture that is third from the left in Figure 19. Putting this twist in the
belt creates stress. When you let go of the belt, this stress will cause the belt to
deform itself so that it looks like the picture on the far right side of Figure 19.
In spite of the simple examples illustrated in Figure 19, in general, the twist
is not an integer, and (like the average writhe) it is not easy to calculate by hand.
For example, consider the ribbon illustrated in Figure 20. The axis is represented
by the dotted black loop and one of the backbones is represented by the solid black
loop. The line segments between the axis and black backbone are there to help you
visualize how the black backbone is twisting around the axis. If you look at the
figure carefully, you will see that the twist of the ribbon is somewhere between 0
and 1, though we can’t determine the precise number just by looking at it.
Since the two pictures on the right in Figure 19 are topologically equivalent but
have different values of T w, we know that the twist of a ribbon is not a topological
Figure 20. The twist of this ribbon is between 0 and 1.

9. THE RELATIONSHIP BETWEEN LINKING, TWISTING, AND WRITHING 343
property. It turns out that, like the average writhe, the twist is actually a geometric
property of a ribbon.
9. The Relationship between Linking, Twisting, and Writhing
Looking back at what we’ve done in the last three sections, we have defined
three quantities that each tell us something about the geometry or the topology of
circular duplex DNA. The linking number Lk of the two backbones is a topological
property, whereas both the average writhe of the axis Wr and the twist T w between
a backbone and the axis are geometric properties. As a result, any deformation of
the molecule will leave Lk fixed, but may change T w and Wr. Surprisingly, it turns
out that these three quantities are related by the equation:
Lk = T w + Wr.
We won’t prove this equation is true. After taking a course on differential

geometry, you could read a proof that this equation holds. Rather, we will focus
on computing some examples. In particular, in Figure 21 we find the values of Lk
and Wr for the ribbons whose T w we found in Figure 19. In the first and second
illustrations on the left, the axis is a planar circle, and so the average writhe is
Wr = 0. Also, since the backbones don’t twist around each other or around the
axis, we have Lk = T w = 0. In the third picture, the axis is again a planar circle,
but in this case the backbones twist around each other and around the axis. Thus
we have Wr = 0 and Lk = T w = −1. Finally, we observe that the ribbon on the
right is nearly flat. It comes out of the page ever so slightly so that one part of the
ribbon can pass over the other. This means that in almost every direction the writhe
is −1. We might imagine that the writhe is zero when we project in a direction
perpendicular to the plane of the page. However, in that direction, pieces of the
projection lie on top of one another as they did for the planar circle in Figure 16.
Hence this projection is not allowed. In fact, there are almost no projections where
the writhe is zero. So averaging over all projections, we see that Wr = −1. Also,
we already know from Figure 17 that Lk = −1 and T w = 0. Thus for each of the
ribbons in Figure 21, the given values satisfy the equation Lk = T w + Wr.
Figure 21. Lk, T w, and Wr for the ribbons in Figure 19.
For our next example, let’s consider a circular unknotted duplex DNA molecule
with 315 base pairs which is in a relaxed state. Recall, that because it’s unknotted
and in a relaxed state, the backbones make one complete right-hand turn for every
10.4–10.5 base pairs and the axis is roughly planar. Thus we can divide the total
number of base pairs by 10.5 to determine that the linking number between the
two backbones is Lk = 315/10.5 = 30. Also, since the axis is roughly planar, we
can assume that Wr = 0. Hence it follows from the equation Lk = T w + Wr that
T w = 30.
Now suppose that we want to twist up the axis of our molecule so that Wr =
−3, as illustrated on the right side of Figure 22, and we want our configuration
to be stable. If a supercoil is added to the axis of our duplex DNA molecule
without increasing the linking between the two backbones, then the supercoil will
be unstable, like the supercoil that we added to our circular ropes in Figure 7.
Figure 22. We want to twist up the axis so that Wr = −3.
In order to create stable supercoiling as we did in Figure 6, we have to increase

the linking between the two backbones to put enough stress on the DNA that it
will be forced to supercoil. To do this, we cut one backbone, twist it around the
other backbone, and then re-attach it as in Figure 23.
Figure 23. To increase Lk we cut one backbone, twist it around

the other backbone, and then re-attach the ends.
Let’s say that we perform the operation illustrated in Figure 23 to add +1 to

Lk. Then one or both of T w and Wr must increase in order to add +1 to the right
side of the equation Lk = T w + Wr. Remember that even though Lk is always an
integer, the values of T w and Wr are not necessarily integers. It turns out that if
Lk is increased by +1, the most stable configuration of DNA occurs when the value
of Wr increases by 0.75 and the value of T w increases by 0.25. Since 3 × .25 = .75,
this means that for each full turn added between the two backbones, Wr increases
by 3 times as much as T w increases. Similarly, for each full turn removed from the
two backbones, Wr decreases by 3 times as much as T w decreases.
Now let’s return to our example of a circular unknotted duplex DNA molecule
with 315 base pairs, which we want to have a stable configuration with Wr = −3.
Since we’ve decreased Wr by 3, we must also decrease T w by 1. However, by the
equation T w + Wr = Lk, this means that the linking number between the two
backbones must be decreased by (−1) + (−3) = −4. Hence, the linking number
between the backbones in the new configuration will be Lk = 30 − 4 = 26.
10. REPRESENTING THE SUPERCOILING WITH A SINGLE NUMBER 345
10. Representing the Supercoiling with a Single Number
We would like to use the linking number of the two backbones as a way to
measure how supercoiled a DNA molecule is. However, just knowing what the
linking number is doesn’t give us enough information. Analogously, if all you know
about a dog is its weight, you couldn’t reach any conclusions about its shape. For
example, suppose you know that a full grown dog weighs 20 pounds. This would be
an enormous amount for a Yorkshire Terrier but very little for a German Shepherd.
If you knew the size of the dog in addition to its weight, you would be able to have
a sense of whether it was underweight or overweight. Similarly, the linking number
only gives us meaningful information about supercoiling if we put it in the context
of how long the molecule is, which we measure by its number of base pairs.
Thus we want to compare the linking number of an arbitrary duplex DNA
molecule to the linking number of a relaxed DNA molecule with the same number
of base pairs, which we denote by Lkrelax . Since relaxed DNA has the form of a
right-handed double helix and the two backbones have parallel orientations, all of
the twists between the two backbones are positive. This means that Lkrelax > 0.
Now the linking difference, Lk − Lkrelax is the number of full turns between the two
backbones that need to be added or removed in order to change an arbitrary DNA
molecule with linking number Lk into a relaxed form. If we divide this difference by
Lkrelax , we get the number of turns that need to be added or removed per full turn
between the two backbones to change it into a relaxed form. Thus, the amount of
supercoiling of an arbitrary DNA molecule can be represented by the value
Lk − Lkrelax
.
Lkrelax
We say a molecule is negatively or positively supercoiled according to whether
this number is negative or positive. Since this value is per full turn, we are able to
compare the supercoiling of DNA molecules with different numbers of base pairs.
The bigger the absolute value

Lk − Lkrelax

Lkrelax
is, the more tightly supercoiled the molecule.
As an example, let’s return to the unknotted DNA molecule that we considered
at the end of the last section which had 315 base pairs and Lk = 26. Recall that in
a relaxed state, the linking number of this molecule was 30. Thus, the supercoiling
is represented by
26 − 30
≈ −0.133.
30
Now let’s compare the supercoiling of this molecule to that of an unknotted
DNA molecule that has 630 base pairs and has Lk = 56. To find the supercoiling of
our second DNA molecule, we first compute the linking number of a relaxed DNA
molecule which has 630 base pairs as Lkrelax = 630/10.5 = 60. Thus we find that
the supercoiling of our second molecule is given by
56 − 60
≈ −0.067.
60
We know both molecules are negatively supercoiled since −0.133 and −0.067
are both negative numbers. However, since | − 0.133| > | − 0.067|, we can conclude
that the molecule with 315 base pairs is more supercoiled than the molecule with
630 base pairs.
To explain this, let’s observe that the difference in the linking number of the
given molecule and a relaxed molecule of the same length is Lk − Lkrelax = −4
in both cases. However, the molecule with only 315 base pairs has to absorb this
linking difference over a shorter length of DNA than the molecule with 630 base
pairs. Thus the shorter molecule will encounter more torsional stress and, as a
result, will have more supercoiling.
We finish this section by introducing some terms which capture what we’ve
learned. A relaxed DNA molecule has Lk − Lkrelax = 0; and if the molecule is
unknotted, then we also have Wr = 0. If a DNA molecule has Lk < Lkrelax ,
then we say that the DNA is underwound because the winding between the two
backbones is less than it would be if the molecule were in a relaxed state. This
means that the molecule has more than 10.5 base pairs per full turn between the
two backbones. To relieve the stress that this creates, the value of Wr decreases.
If the molecule is unknotted, this means that Wr becomes negative, in which case
we say the molecule is negatively supercoiled.
On the other hand, if Lk > Lkrelax , then we say that the DNA molecule is
overwound because now there are fewer than 10.5 base pairs per full turn between
the two backbones. To relieve the stress that this creates, the value of Wr increases.
If the molecule is unknotted, this means that Wr becomes positive, and we say the
molecule is positively supercoiled.
In fact, in the cells of almost all organisms the DNA is negatively supercoiled,
and it is this supercoiling that allows it to fit compactly and efficiently into cells.
Nonetheless, in general, the values of Lk and T w are positive, as they are for relaxed
DNA molecules.
11. Replication
Recall that in order to duplicate itself for reproduction and cell growth, a
duplex DNA molecule has to split into two identical DNA molecules through a
process known as replication. The first step in this process is that one after another
the bonds between the pairs of bases on the two backbones break apart. If we think
of these bonds as the teeth of a zipper, breaking these bonds is like unzipping the
zipper. Since the linking number between the backbones doesn’t change while they
are being pulled apart, the twists between them are all being pushed into the part
of the DNA that is still zipped up. This means there are fewer and fewer base
pairs for the same number of twists that there were before replication began (see
the middle picture in Figure 24).
Observe that the fewer base pairs that a DNA molecule has, the smaller Lkrelax
will be. We remarked at the end of the last section that DNA in a cell is negatively
supercoiled as a result of having more than 10.5 base pairs between each full turn of
the backbone. This means that the twisting between the two backbones of DNA in
a cell is more spread out than it would be for DNA of the same length in a relaxed
state. It follows that before replication, we have Lk < Lkrelax .
Once replication begins and the backbones start to unzip, the twists between
them are forced into the zipped up part, which has fewer and fewer base pairs, as
illustrated in Figure 24. Let Lkrelax represent the linking number of a relaxed DNA
11. REPLICATION 347
Figure 24. Unzipping the backbones pushes the twisting into a

shorter segment, which leads to positive supercoiling.
molecule which has the same number of base pairs as the zipped up part. Then
Lkrelax will get smaller and smaller as the DNA unzips. Since Lk is unchanged
throughout the unzipping process, Lkrelax eventually becomes smaller than Lk. As
we saw in the last section, when Lk > Lkrelax , DNA is positively supercoiled in
order to relieve the stress created by having fewer than 10.5 base pairs per full
twist of the backbones. Thus the shorter and shorter the zipped up part becomes,
the more it becomes positively supercoiled, as illustrated on the right of Figure 24.
Eventually, when there are too many positive supercoils to fit in the remaining
zipped up part, the unzipping process is forced to stop. This would put an end to
replication and the future of life as we know it, if it weren’t for the work of a special
enzyme that comes to the rescue in the nick of time. But first we pause to bring
you a word from our sponsors—the topoisomerase enzymes.
In general, enzymes are large protein molecules that facilitate important molec-
ular changes in living organisms. There are many types of enzymes, all of whose
names end with the suffix ase. The topoisomerases are a family of enzymes which
maintain the intrinsic structure of DNA molecules while changing their extrinsic
topology, which is why they have the prefix topo. There are two types of topoiso-
merases: Type I which cuts one backbone of the DNA to remove twists between
the two backbones and then reconnects the backbone to itself; and Type II which
cuts both backbones to change the topology of the axis and then reconnects each
backbone to itself.
Topoisomerase I comes to the rescue, solving the problem of excessive super-
coiling during replication by cutting one backbone, allowing it to untwist around
the other backbone, and then regluing the two ends as illustrated in Figure 25. In
this way, positive supercoiling is reduced, allowing the unzipping process and thus
replication to continue.
By regulating supercoiling during replication, Type I topoisomerases enable
cells to grow and reproduce. However, not all cell growth and reproduction is a
good thing. The growth of cancerous cells in our bodies is one of the scariest things
Figure 25. Topoisomerase I cuts one backbone, untwists it

around the other backbone, then reglues the ends.
that can happen to us. Topoisomerase I inhibitors have been used as cancer fight-
ing drugs because they prevent cancer cells from replicating. Unfortunately, such
medications can also prevent certain healthy cells from replicating. In particular,
topoisomerase I inhibitors interfere with the growth of hair cells, which is why
chemotherapy often causes hair loss.
12. Site Specific Recombination
While replication plays a crucial role in reproduction, it is not unique in its

importance. DNA is also involved in the equally important process of recombina-
tion. In fact, recombination is what enables genetic diversity and allows species to
evolve over time.
There are several types of recombination which you might learn about in a
course on genetics. In this section, we discuss a particular type of recombination
known as site specific recombination in which two small pieces of duplex DNA are
brought together, cut, and then reattached in a different way. We focus on this type
of recombination because it can change the extrinsic topology of DNA, and can even
create knotted or linked molecules. In our study of site specific recombination, we
will focus exclusively on topology and ignore the geometry of the molecules.
Note that in contrast with the earlier part of this chapter, we are now thinking
of a duplex DNA molecule as a single string, ignoring the fact that it actually
consists of two backbones which twist together to form a double helix. If you find
it hard to forget about the the two strands, keep in mind that you are used to
thinking about a rope as a single string, even though if you look closely you can
see that it too consists of multiple strands wound together.
We begin with a little vocabulary to help us talk about the steps of site specific
recombination. Prior to the start of recombination, a duplex DNA molecule is
called the substrate. The first step of recombination is that two identical short
pieces of genetic code known as sites come together, most often as a result of
supercoiling together with random motion inside of the cell. When they’re close
enough together, a recombinase enzyme binds to both of the sites at once. At this
point we can visualize the enzyme as a small ball with the two sites inside it. The
DNA molecule together with this ball is called the synaptic complex . Next the
recombinase cuts open the two sites, moves the endpoints of the cut open sites
13. KNOTTED AND LINKED PRODUCTS OF RECOMBINATION 349
Figure 26. The steps of site specific recombination.
around within the ball, and then reattaches them, possibly in a different way. The
resulting DNA molecule is said to be the product of the recombination.
We illustrate these steps in Figure 26, where we’ve drawn the substrate as an
unknot, and the two sites as small black rectangles. In the first picture, the sites
are far apart. But as a result of supercoiling and random motion, the sites are
brought together so that the enzyme can bind to both of them at once to obtain
the synaptic complex in the third picture. In the fourth picture the recombinase
has cut the ends of the two sites. Then in the final picture the ends are interchanged
and glued to the new strands to obtain the product. Observe that in regluing the
ends of the sites, a new crossing has been introduced. In this particular example,
the product of recombination is knotted though the substrate was not. Thus the
recombinase has indeed changed the topology of the molecule.
13. Knotted and Linked Products of Recombination
What makes site specific recombination tricky to study is that molecular bi-
ologists can observe the substrate and the products, but they can’t observe what
is happening during the actual process of recombination. In particular, they don’t
know what the synaptic complex looks like, and they can’t tell whether there are
twists that are caught inside or outside of the enzyme ball. For example, in Fig-
ure 26 we can see that there are three twists that are outside of the enzyme ball
that end up as crossings in the knotted product. Imagine that you’re a scientist
and all you can see is the first and last picture of Figure 26. It would be hard to
figure out exactly how the enzyme managed to change the unknotted substrate into
the particular knotted product.
Sometimes scientists get lucky and can find clues that help them piece together
what the process might have been. For example, suppose a recombinase acts on an
unknotted substrate to obtain a knotted product that scientists can identify, then
the recombinase acts on that knotted product to obtain a link that scientists can
again identify. Now scientists have two pieces of information rather than just one.
If the recombinase acts a third time, the scientist can get even more information.
It’s like a detective story—the more times the bad guy commits his crime, the easier
it is for the detective to figure out the pattern.
However, in the case of knotted and linked products, scientists want to know
the order in which the knots and links are produced so that they can guess how one
could lead to another. But they can’t watch the process or stop it and restart it.
All their information has to be obtained from the collection of knotted and linked
products they have at the end of the process. The trick to figuring out the order in
which the knots and links were produced is by determining the quantities of each.
The largest number of knotted or linked products are the result of recombination on
the original unknotted substrate. The second largest number of knotted or linked
products are the result of recombination on the first products, and so on. Knowing
what knots and links are produced and in what order gives the scientists a major
clue as to what the enzyme did to create them. But the mystery is still not solved.
In order to understand the difficulty of this problem, let’s consider an exper-
iment conducted by the molecular biologists Wasserman, Dungan, and Cozzerelli
in 1985. They observed that the enzyme Tn3 resolvase acted on an unknotted
circular substrate, and the product of one recombination was the Hopf link. The
second recombination produced the figure eight knot, and the third resulted in the
Whitehead link. There were also some 62 knots that resulted from the fourth re-
combination. They could not detect any products beyond these. These knotted
and linked products are illustrated in the order they were produced in Figure 27.
Figure 27. Consecutive products of the enzyme Tn3 resolvase.
You can see from Figure 27 that knowing the knotted and linked products
of recombination and the order in which they were produced does not reveal the
process. You might want to pause here to see if you can figure out how the enzyme
did it. (Hint: It was not in the library with the candlestick.) Before we can present
the solution to this mystery, we need to take a detour in the next three sections to
learn about tangles. As you will see, the study of tangles is closely related to the
study of knots and links. So the next few sections should bring back fond memories
of Chapters 9, 10, and 11.
14. An Introduction to Tangles
While most people use the word “tangle” to refer to a string in a messy jumbled
heap, mathematicians use the term in a more precise way. In particular, we define
a tangle to be a ball containing two strings with endpoints along the equator at the
compass points NE, NW, SW, and SE. In Figure 28, we illustrate a tangle where
the endpoints of one string are at the NE and SE points, and the endpoints of the
other string are at the NW and SW points. We say that two tangles are equivalent
if we can deform the strings of one to the strings of the other while keeping the
four endpoints of the strings fixed at the compass points.
The simplest tangles are those with two vertical strings or two horizontal strings
with no crossings (see Figure 29). The tangle with two horizontal strings is known as
the 0-tangle because horizontal lines have slope 0, and the tangle with two vertical
14. AN INTRODUCTION TO TANGLES 351
Figure 28. An example of a tangle.
Figure 29. The 0-tangle and the ∞-tangle.
strings is known as the ∞-tangle because vertical lines have slope ∞. Note that
the strings in these tangles do not have to be straight lines.
We can create new inequivalent tangles from the 0-tangle and the ∞-tangle by
moving the endpoints of the strings around within the boundary of the ball. For
example, let’s start with the 0-tangle and twist the NE and SE endpoints around
each other within the boundary of the ball to get the new tangle illustrated on the
right in Figure 30. Note that the tangles we create in this way will not in general
be equivalent to the 0- or ∞-tangle, because a pair of tangles is only equivalent if
we can get from one to the other by a deformation keeping the endpoints of the
strings fixed.
We can get infinitely many different tangles by twisting the endpoints of the
strings of the 0- and ∞-tangles in the boundary of the ball in different ways. The
exact tangle that we get will depend on how we twist. To keep track of our twist-
ing, we designate a twist as positive if the overcrossing has positive slope, and we
designate a twist as negative if the overcrossing has negative slope. If we twist the
NE and SE endpoints n times around each other so that we get n crossings and the
slope of each overcrossing is positive we label our tangle (+n) or simply (n). If we
twist the NE and SE endpoints n times around each other and the slope of each
overcrossing is negative, we label our tangle (−n). We refer to tangles obtained in
this way as integer tangles. We illustrate some examples in Figure 31.
Figure 30. If we twist the NE and SE endpoints around each

other, we get a new tangle.
Figure 31. Some integer tangles.
15. Rational Tangles
We can form more complicated tangles by starting with a 0-tangle or an ∞-

tangle and alternating between twisting the NE and SE endpoints of the strings
around each other some number of times and twisting the SW and SE endpoints
around each other some number of times. When we twist the NE and SE endpoints
around each other we say that we are adding horizontal twists, and when we twist
the SW and SE endpoints around each other we say that we are adding vertical
twists. If we start with the 0-tangle, we must begin by twisting the NE and SE
endpoints; and if we start with the ∞-tangle we must begin by twisting the SW
and SE endpoints. Notice that we never twist the NE and NW endpoints or the
NW and SW endpoints around each other.
For example, the tangle on the far right in Figure 32 was obtained by starting
with the 0-tangle and twisting the NE and SE endpoints horizontally around each
other three times in the positive direction, followed by twisting the SW and SE
endpoints vertically around each other twice in the negative direction, and then
twisting the NE and SE endpoints horizontally around each other once in the pos-
itive direction. In order to keep track of the twisting and the order in which we
did it, we denote this tangle by the tangle vector (3, −2, 1). If you haven’t studied
vectors before, don’t worry. A vector is just a list of numbers in a particular order.
Figure 32. The steps to creating a (3, −2, 1)-tangle from a 0-tangle.
In order to make it clear from a tangle vector which twists are horizontal and
which twists are vertical, we make a rule that the last entry in any tangle vector
must correspond to some number of horizontal twists, even if that number is 0.
For example, in the third step of Figure 32, there is a 0 in the last entry of the
tangle vector because the last thing we did was add vertical twists. Because of this
rule, whenever we start with the 0-tangle, the tangle vector we get will have an odd
number of entries. You can see in Figure 32 that all of the vectors have an odd
number of entries.
15. RATIONAL TANGLES 353
If we start with the ∞-tangle, we begin by twisting the SW and SE endpoints

vertically around each other. Since the last entry of the tangle vector still has to be
some number of horizontal twists, in this case the vector will always have an even
number of entries. Thus looking at whether a tangle vector has an even number or
an odd number of entries tells us which of the two tangles we started with.
For example, in Figure 33 we construct a tangle with the same number of
twists as we did in Figure 32, but this time we start with the ∞-tangle instead
of the 0-tangle. Now we begin by twisting the SW and SE endpoints vertically
around each other three times in the positive direction, then we twist the NE and
SE endpoints horizontally around each other twice in the negative direction, and
finally we twist the SW and SE endpoints horizontally around each other once in
the positive direction. We illustrate our construction of this tangle in Figure 33.
We can see that the vectors representing the steps in this construction are different
from those in Figure 32. In particular, now they all have an even number of entries.
Figure 33. The steps to creating a (3, −2, 1, 0)-tangle from the
∞-tangle.
Any tangle that we can obtain in this manner starting with the 0 or ∞-tangle
is called a rational tangle and is represented by a vector of the form (a1 , a2 , . . . , an ),
where each entry corresponds to some number of twists of the NE and SE or SW
and SE endpoints and the last entry represents some number of horizontal twists.
Note that an is the only term in the tangle vector that is allowed to be zero.
We call tangles constructed in this way “rational” because to each such tangle
β
we can associate a rational number α by computing the so-called continued fraction
of the vector (a1 , a2 , . . . , an ) using the following rule.
β 1
= an −
α 1
an−1 −
1
··· −
1
a2 −
a1
To get a feel for what is going on, we compute the continued fraction corre-
sponding to the (3, −2, 1)-tangle in Figure 32 as follows.
10 1
=1−
7 1
−2 −
3
John H. Conway showed that two rational tangles are equivalent if and only if
their continued fractions are equal. This means that to determine if two rational
tangles are the same or different, we only need to compute these continued fractions
to see if they are equal. This is a very powerful result that makes it easy to
distinguish rational tangles.
Rational tangles are incredibly nice. They can be coded up by vectors, and
distinguished easily by computing their continued fractions. But not all tangles are
rational. For example, in Figure 34 we illustrate a non-rational tangle. You can
tell that this tangle isn’t rational because no matter how you move the endpoints
around on the boundary of the tangle ball, you can’t undo the tangle.
Figure 34. A non-rational tangle.
16. Operations on Tangles
Just as we have the connected sum operation that enables us to add two knots
together to get a more complicated knot, we have an addition operation for tangles
that we can use to combine two tangles to get a more complicated tangle. In
particular, we add tangles A and B together by connecting the NE and SE endpoints
of A to the NW and SW endpoints of B and extending the remaining endpoints to
the boundary of a larger ball, as illustrated in Figure 35. The result is called the
sum of A and B and is denoted by A + B. Observe that even though tangles A
and B in Figure 35 are rational, the tangle A + B is not rational.
Figure 35. In this example, the sum of rational tangles is not rational.
We can create a simpler example by adding together two ∞-tangles. In this

case, as we see in Figure 36, the sum has an extra component in the center, and
hence is not rational.
We can also start with a tangle A, and then form a knot or link by adding an
arc to the top and the bottom of the tangle as illustrated in Figure 37. The result
is called the numerator closure of A and is denoted by N (A). As we see in Figure
37, we can also combine these two operations to get N (A + B).
With this background in the theory of rational tangles and tangle operations,
we are now ready to use tangles to analyze how site specific recombination produces
particular knots and links.
17. THE TANGLE MODEL OF SITE SPECIFIC RECOMBINATION 355
Figure 36. The sum of two ∞-tangles has an extra component,

and hence is not rational.
Figure 37. The numerator closure of the tangles A and A + B.
17. The Tangle Model of Site Specific Recombination
The idea of a mathematical model is that you state a number of reasonable

assumptions about a real situation, and then you use mathematics to derive con-
clusions. We would like to use tangles to model site specific recombination. The
first assumption of our model is that a recombinase enzyme has the form of a ball,
and that this ball together with the two sites inside of it (which we refer to as the
enzyme ball) looks like a tangle.
If you look back at the drawing of the synaptic complex in Figure 26, you can
see that we drew the enzyme ball with the sites inside of it as a 0-tangle. In reality,
we don’t know that the enzyme together with the sites looks the way we drew it.
Maybe there are additional pieces of the substrate that go in and out of the enzyme
and are tangled together with the two sites, or maybe the two sites themselves are
twisted together. If we knew exactly what was going on inside the enzyme, we
wouldn’t need a model. But thinking of the enzyme as a ball and the pieces of the
substrate inside of the enzyme as the two strings of a tangle seems like a reasonable
assumption.
In our model, we will refer to the tangle inside of the enzyme ball prior to
recombination as the site tangle and denote it by T . Next we model the twisted
part of the substrate outside of the enzyme as another tangle that we refer to as
the substrate tangle and denote it by S. Finally, we model the substrate as the
numerator closure N (S + T ). Note that this means all of the crossings of the
projection must be in one of the two tangles. We illustrate a substrate expressed
as the numerator closure N (S + T ) on the left side of Figure 38.
The next assumption of our model is that the action of the enzyme takes place
entirely inside of the enzyme ball and only affects the pieces of DNA which are in
the site tangle T . In particular, the part of the molecule outside of the enzyme
ball remains fixed while the enzyme is acting on T . This means that the enzyme
Figure 38. According to our model, the enzyme changes the sub-
strate N (S + T ) into the product N (S + R).
causes the tangle T to be replaced by a new tangle R that we will refer to as the
recombination tangle. Hence the enzyme changes the numerator closure N (S + T )
representing the substrate into the numerator closure N (S + R) representing the
product as illustrated in Figure 38.
For a given recombinase, if we know what knot or link the substrate and the
product are, then we can set N (S + T ) equal to the knot or link of the substrate
and N (S + R) equal to the knot or link of the product. This gives us two tangle
equations with the tangles S, R, and T as the unknowns. Recall from when you
took algebra that if you have two equations and three unknowns you won’t be
able to solve to get specific values for the unknowns. The same is true for tangle
equations. In particular, with two tangle equations and three unknown tangles we
cannot solve the equations for the tangles S, R, and T .
However, if there is a second or even a third knotted or linked product, then
we can obtain more equations. But first we have to add one more assumption
to our model, which is that if the enzyme acts multiple times, it always binds at
the same place and always acts the same way. This means that each recombi-
nation event adds the same tangle R next to the previous one. Hence the
first product of recombination is N (S + R), the second product of recombination
is N (S + R + R), the third product is N (S + R + R + R) and so on. Using this
assumption, Figure 39 illustrates the products we would get from Figure 38 after
one and two recombination events.
Figure 39. The first product is N (S +R) and the second product
is N (S + R + R).
18. APPLYING THE MODEL TO THE ENZYME TN3 RESOLVASE 357
18. Applying the Model to the Enzyme Tn3 Resolvase
We now apply the tangle model to analyze the products of the enzyme Tn3
resolvase. In fact, the tangle model was originally developed by Claus Ernst and De
Witt Sumners precisely to explain how Tn3 resolvase acted on an unknotted sub-
strate to produce first the Hopf link, then the figure eight knot, then the Whitehead
link, and finally the 62 knot, as illustrated in Figure 27.
Based on this consecutive list of knotted and linked products, our model gives
us the tangle equations:
N (S + T ) = unknot,
N (S + R) = Hopf link,
N (S + R + R) = figure eight knot,
N (S + R + R + R) = Whitehead link,
N (S + R + R + R + R) = 62 knot.
Ernst and Sumners then used advanced techniques from topology to solve these
tangle equations for the tangles R, S, and T . Their solution, illustrated in Figure 40,
shows the steps that the enzyme took to get from one knot or link to the next, and
thus solves the mystery of Figure 27.
Figure 40. How the products of Tn3 resolvase occurred.
The following theorem expresses the same result in terms of tangle vectors.
Theorem 14.1. Suppose that S, T , and R are rational tangles that satisfy the
following equations:
(1) N (S + T ) = unknot,
(2) N (S + R) = Hopf link,
(3) N (S + R + R) = figure eight knot,
(4) N (S + R + R + R) = Whitehead link.
Then S = (3, 0), R = (−1), and N (S + R + R + R + R) is the 62 knot.
You can check in Figure 40 that S is indeed a vertical tangle with three positive
crossings and R is a single negative crossing. The tangle model together with
this theorem shows that knowing the knots and links in the substrate together
with the first three products of Tn3 resolvase is enough to not only determine
what the tangles S and R are, but to correctly predict that the fourth product of
recombination will be the 62 knot.
19. Exercises
1. Explain why a closed duplex DNA molecule always contains an even number
of half twists.
2. Find Lk, Wr, and T w for the ribbon in Figure 41, then, check that the equation
Lk = T w + Wr holds.
3. Draw a picture of a ribbon that has Lk = 0 and Wr = 2.
4. Suppose that a duplex DNA molecule is in a stable configuration. If you add

3 to the value of Wr, how would you need to change Lk so that the new
configuration is also stable?
5. Consider two unknotted DNA molecules with stable configurations. The first
one has 630 base pairs and Wr = 6, and the second one has 210 base pairs and
Wr = −3.
(a) For each molecule find the value of

Lk − Lkrelax
.
Lkrelax
(b) Determine which molecule is more tightly supercoiled.
6. Consider the ribbon illustrated in Figure 42 which is like the one illustrated
on the right of Figure 21 but with two twists instead of one. The ribbon is
almost entirely in the plane of the page, but comes out of the plane ever so
slightly in two places so that one piece can cross over the other. The enzyme
Topoisomerase II works by cutting both backbones of duplex DNA and reat-
taching them so as to make a crossing change in the central axis as illustrated
in Figure 42. What effect will this enzyme action have on the values of Lk, Wr,
and T w?
19. EXERCISES 359
7. Recall that in site specific recombination two identical short segments of DNA
are brought together by the recombinase. These segments are strings of letters
which must be lined up precisely in the enzyme ball. Rather than writing out
the letters in each segment, we can represent them as arrows as illustrated in
the middle pictures of Figure 43. Prior to recombination, the two arrows could
either represent the same or opposite orientations on the circular substrate. If
the orientation of the two arrows represents the same orientation on the circular
substrate, then we say that the sites are direct repeats. If the arrows on the
sites represent opposite orientations on the substrate, then we say that the sites
are inverted repeats. These are illustrated in Figure 43. Explain how we know
that if a recombinase acts once on a circular substrate with inverted repeats,
then the product is a knot.
8. If a recombinase acts on a circular substrate an even number of times with

direct repeats then the product must be a knot, and if it acts an odd number
of times the product must be a link. Explain why this must be the case.
9. Draw pictures of the rational tangles (−2, 3, 4) and (−2, 3, 4, 0).

10. Find the continued fractions for the tangles (−2, 3, 4) and (−2, 3, 4, 0).
11. Give examples of tangles A and B other than (0) and (∞) such that N (A) is
the unknot and N (B) is the unlink.
12. Give examples of tangles A and B other than (0) and (∞) such that N (A + B)
is the unknot. Note A and B do not have to be rational.
13. Figures 40 and 44 illustrate that the products of recombination with Tn3 re-
solvase can be written as numerator closures of rational tangles in two different
ways. Consider the continued fractions of the two different tangles whose clo-
sure is the 62 knot, and determine whether the tangles are equivalent.
14. Illustrate deformations taking the knots and links in Figure 44 to those in
Figure 40.
CHAPTER 15
The Topology of Proteins
Topics:
• The structure of proteins
• Which knots occur in proteins
• Knots and links in proteins
• Non-planar graphs in metalloproteins
• Non-planar graphs in the protein nitrogenase
• Möbius ladders with three and four rungs in metalloproteins
• Möbius ladders in small proteins
• Recommendations for further reading
1. An Introduction to Proteins
In Chapter 14, we learned that DNA is a long floppy molecule whose topology is
just as important as its geometry. But DNA isn’t the only floppy biological molecule
whose topology is important. Proteins are made up of long chains of amino acids,
just as DNA is made up of long chains of sugars and phosphates. The chain of
amino acids, known as the backbone of a protein, is not as long as the backbone of
a DNA molecule, but it is long enough that it lacks a rigid structure and behaves
like a floppy coil before folding up into a rigid functional shape. Scientists want to
understand how protein folding occurs because many illnesses including Alzheimer’s
disease, Parkinson’s disease, type 2 diabetes, and even many allergies are the result
of proteins folding incorrectly.
It is believed that the topology of a protein structure can affect its ability to fold
properly. To shed light on the role topology might play, scientists are interested in
studying proteins containing knots, links, and non-planar graphs. Such structures
are not very common among all proteins, but they are by no means unheard of.
In fact, knotted proteins have been found in a wide range of organisms from yeast
to humans. Furthermore, when a protein loses or gains a knot or link through
the process of evolution, it seems to cause a change in the behavior or function of
the protein. For example, proteins containing knots and links have been observed
to have increased stability relative to unknotted and unlinked proteins. Yet the
precise biological function of knotting and linking in proteins is still unknown.
In this chapter, we will learn about protein structures which cannot be deformed
into a plane. In order to help us focus on the topology of such structures, we will
treat them as if they are completely flexible, even though in reality proteins are
only partially flexible. Nonetheless, if we can show that a completely flexible model
361
362 15. THE TOPOLOGY OF PROTEINS
cannot be deformed to lie in a plane, then the actual protein, with somewhat less
flexibility than the model, also cannot be deformed to lie in a plane. Thus, for
the purpose of identifying non-planarity within protein structures, we can assume
complete flexibility, even though this is not physically accurate.
In order to study the topology of a particular protein, we first need to draw
a picture of it. On left side of Figure 1, we see an illustration of a protein that
you might find in a biochemistry paper or textbook. Unfortunately, this type of
illustration, while relatively accurate, is so complicated that it’s a challenge to
see which pieces of the protein pass behind and which pieces pass in front. Such
information is crucial if we are looking for knots or links in the drawing. Thus we
will deform our proteins to create much simpler drawings that omit many of the
twists and turns, but make it easy to identify all of the crossings. We will refer
to these simplified drawings as the underlying topological structure of the protein.
For the protein in Figure 1, we can remove all of the crossings so that we obtain a
squiggly segment in the plane as its underlying topological structure. The two ends
of this segment are known as the termini of the protein. In order to distinguish one
endpoint from the other, one is labeled as the C-terminus and the other is labeled
as the N-terminus.
Figure 1. We simplify the biochemical diagram on the left to get

the underlying topological structure on the right.
2. Trapping Knots in Proteins
Recall that in Chapter 9 we defined a knot to be a circular path in R3 . The

requirement that the path be circular is so that the knot will be trapped within the
circle, as the knot on the left side of Figure 2 is. If we don’t require this, a knot
like the one on the right side of Figure 2 can be easily undone. You probably have
ample experience with pulling your earbuds out of your pocket only to find them
completely tangled up. In spite of how annoying this can be, you know that with
enough patience you can remove all the knots by systematically unthreading one of
the ends back through any loops.
While the backbones of protein structures are normally open segments like
shoelaces rather than circular paths, if a knot is far away from the endpoints of the
segment, then from a biochemical viewpoint, it is trapped within the structure. In
2. TRAPPING KNOTS IN PROTEINS 363
Figure 2. The knot on the left is trapped, whereas the knot on

the right can be removed.
particular, the energy required for the termini to unthread the knot is so large that
this is unlikely to occur. So even though most proteins are open segments like the
one on the right side of Figure 2, we want to create a model of them which prevents
any knots from wriggling off the ends.
Many approaches have been developed to creating such a model. On the left
side of Figure 3 we start with a protein in the form of an open segment. Then we
replace the termini by arrows indicating a straight ray that goes on indefinitely.
These infinite rays prevent the knot in the middle figure from slipping off the
structure. The problem with this approach is that if one or both of the termini are
near the tangling of the knot, then different knots can be trapped in the structure
depending on the direction in which the rays are extended. For example, in the
right image of Figure 3, if we extend the bottom endpoint by a ray that goes under
the arc, we will get a trefoil knot, but if we extend at a slightly different angle so
that it goes over the arc, we will get the unknot.
Figure 3. The knot on the left is trapped in the structure, while

the knot on the right may or may not be.
Another approach to modeling a knotted linear protein structure is to place it

in the center of a large ball. The termini are then extended by straight segments
going to the spherical boundary of the ball and are finally joined by an arc in the
boundary to create a closed loop. In Figure 4, the extensions to the boundary are
drawn as grey dashed lines, and the arc in the boundary is drawn as a grey curve.
However, the same problem can occur that we saw on the right side of Figure 3
when we extended the termini indefinitely. In particular, different knots may be
obtained depending on how the arcs are extended to the boundary of the ball.
One way to resolve the problems with both approaches is to list all the knots
that can be obtained by extending the ends in every possible direction, and then the
Figure 4. We put the knotted arc in a ball, extend the endpoints

to the boundary, and then join them by an arc.
knot that occurs the most frequently is considered to be the one in the structure.
To do this more formally, a probability is assigned to each knot and the knot with
the highest probability is the one considered to be in the structure.
When the termini of a protein are quite close to each other, adding a small arc
between them gives the knot with the highest probability. For example, in Figure 5,
we connect the termini together with a short arc to obtain the knot on the right.
In this chapter, when we are looking at knots in open linear proteins, we require
the termini to be close to each other so there is no ambiguity about which knot is
formed.
Figure 5. Because the termini are close together, we can add a

short arc joining them to obtain a knot.
3. Which Knots Occur in Proteins?
Whatever method is taken to trapping knots in proteins, amongst the tens of

thousands of proteins identified as of 2015, only a few hundred have been recognized
to contain knots. Furthermore, among the hundreds of knotted proteins, all contain
one of the four knots 31 , 41 , 52 , and 61 , illustrated in their circular forms in Figure
6.
Note that Figure 6 includes both mirror forms of the trefoil knot, since both
have been identified in proteins. Since the figure eight knot 41 can be deformed to
its mirror image, only one form of it is listed in Figure 6. By contrast, the 52 and
61 knots are distinct from their mirror images, yet surprisingly these knots have
only been found in proteins in the forms illustrated in Figure 6. Perhaps even more
surprising is the fact that the relatively simple 51 knot (illustrated in Figure 7) has
yet to be observed in any protein structure.
3. WHICH KNOTS OCCUR IN PROTEINS? 365
Figure 6. As of 2015, these are the only knots that have been
found in proteins.
Figure 7. This five crossing knot has never been identified in a

protein structure.
Various theories have been proposed as to why certain knots have been found
and other knots have not. One such theory is based on how proteins might become
knotted in the first place. According to this theory, the protein’s underlying topo-
logical structure looks like a twisted bobby pin with a loop at the top. If one of the
ends of the protein threads through this loop and the other end comes close to the
threaded end, then adding a short arc between the two termini traps the knot, as
illustrated for the 52 knot in Figure 8.
Figure 8. The 52 knot could be the result of threading one ter-

minus through the loop of a twisted bobby pin.
You may recall that among the various knot invariants that we introduced
in Chapter 10, we defined the unknotting number to be the smallest number of
crossings that must be changed to obtain the unknot among all possible projections
of a given knot. If a knot is obtained by threading one terminus through the loop
of a twisted bobby pin, then changing one of the crossings of the arc joining the
termini will result in an unknot. This means that any non-trivial knot that occurs
in a protein in this way, has unknotting number 1. In Figure 9 we illustrate how to
unknot the knot from Figure 8 by changing a single crossing of the threading arc.
Figure 9. Changing the crossing created by the threading gives

us the unknot.
It turns out that the knot 51 has unknotting number 2 (though this is not easy
to prove). It follows that 51 can’t be formed in this way. So if the only way a
knot could form is through threading one end of a protein through the loop of a
twisted bobby pin, then this would explain why the knot 51 has never been found.
However, it’s not known whether there are in fact other ways that proteins can
become knotted.
The other big mystery that remains is why the trefoil knot occurs in proteins
in both mirror forms, but the 52 and 61 knots only occur in one of their mirror
forms.
4. Examples of Knotted and Linked Proteins
Roughly half of all proteins are known as metalloproteins because they contain
metal atoms. Because of the extra structure provided by the metal atoms, the
graphs of these molecules are more complex than other proteins and hence more
likely to contain knots and links.
The trefoil knot 31 was the first knot to be discovered in a protein structure,
and it was found in the metalloprotein illustrated in Figure 10. In contrast with
the protein knots we saw in the last section, which were entirely contained in
the backbones of the proteins, the knot in Figure 10 includes two disulfide bonds
between different pieces of the backbone (illustrated as thick grey segments). In
general, disulfide bonds are relatively strong bonds that join different parts of the
protein backbone with each other or with metal atoms which are attached to the
protein. Disulfide bonds are important biologically because their formation guides
how the protein folds and, once the folding is completed, these bonds hold the
protein in its folded form.
The trefoil knot is by far the most common knot identified in proteins and
has been found in some forms of carbonic anhydrase, a metalloprotein present in
virtually all living organisms from humans to animals to plants. In particular,
trefoil knots appear in one type of human carbonic anhydrase found in red blood
cells.
If we only consider knots in the backbones of proteins rather than allowing
disulfide bonds to be included as part of the knot, then there are only about forty
proteins which contain knots and almost all of these knots are trefoils. However,
there are a few such knotted proteins which contain the 41 knot, and only single
proteins have been found to contain the 52 and 61 knots.
Interestingly, the protein with the 52 knot, known as ubiquitin hydrolase, is
contained in human brains. So if someone tells you that your brain is tied in knots,
4. EXAMPLES OF KNOTTED AND LINKED PROTEINS 367
Figure 10. This knotted protein includes the thick grey segments
representing disulfide bonds.
they’re right! This protein is quite important in preventing neurodegenerative

disorders. In particular, a mutation in the gene encoding it seems to be correlated
with Parkinson’s and Alzheimer’s diseases. Some researchers believe that the knot
in this structure is what helps prevent it from degrading. This theory is based on
the fact that protein degradation can occur when proteins are threaded through a
narrow hole. If there’s a knot in the protein, it could make the protein too fat to
fit through such a hole. If you sew, you’ve probably encountered a similar situation
when trying to thread a needle with a piece of thread containing a knot (see Figure
11). However, the existence of such knots in proteins is good because it prevents
disease, while such knots in thread are just annoying.
Figure 11. An unknotted thread can pass through the hole of a

needle, but a knotted thread cannot.
Topological links in proteins are believed to provide stability to help organisms

survive in harsh environments. For example, protein links have been found in
organisms that live in the extreme heat found in volcanos. The Hopf link is by far
the most common of the protein links. It occurs in several distinct ways, a couple
of which are illustrated in Figure 12.
The image on the left side of Figure 12 shows two protein backbones (one in
thin grey and the other in thin black) that form a link if you include the disulfide
bonds illustrated as thick grey segments. The image on the right shows a linked
protein structure in which a heme porphyrin ring makes up one ring (illustrated in
black) and the protein backbone together with a central iron atom make up another
Figure 12. Some Hopf links in protein structures.
(illustrated in grey). Note the Fe represents an iron atom and the N represents a
nitrogen atom.
Other types of links can occur as a result of multiple heme porphyrin rings
binding to a protein. In Figure 13, each heme porphyrin ring is illustrated as a
black squares and the protein backbone is contained in the dark grey together with
dotted grey curve. Note that the black squares represent the same structure as the
black ring with the nitrogens in the link on the right of Figure 12. The dotted lines
are not part of the link. We illustrate the underlying topological structure of the
link on the right. This link is called a key ring link because it looks like a big grey
key chain with four black keys hanging from it.
Figure 13. This protein structure is a key ring link with four keys.
Perhaps the most interesting example of a protein link is the structure that
protects the virus bacteriophage HK97, illustrated in Figure 14. While you might
think that viruses are always bad, bacteriophages are good viruses because they
fight bacterial infections. So we should be happy that this virus has a link that
looks like chainmail protecting it against the evil bacteria. The chainmail around
the bacteriophage is actually a 72 component link known as a capsid whose stability
comes from the linking. It is made up of 12 pentagonal rings and 60 hexagonal rings,
linked into a spherical shape. In Figure 14, we highlight a pentagon in grey to make
it easier to see it among all the hexagons.
5. NON-PLANAR GRAPHS IN METALLOPROTEINS 369
Figure 14. The front hemisphere of a protein link made up of 12

pentagonal rings and 60 hexagonal rings.
Whether a protein link contains 2, 5, or 72 components, linking seems to greatly

increase the stability of the resulting structure.
5. Non-planar Graphs in Metalloproteins
As we saw above, protein structures can be non-planar because they contain

knots or links. However, these are not the only types of non-planarity that can
occur in proteins. In particular, some proteins are non-planar because they are
attached to a metal cluster, like the iron-sulfur cluster illustrated in Figure 15. So
if you see an iron-sulfur box with four legs walking down the street looking like a
four-legged tarantula, you’ll recognize it as a metal cluster whose nitrogen atoms
might be attached to a protein.
Figure 15. An iron-sulfur cluster resembles a four-legged bug.
Notice that in spite of its cube-like form, if the iron-sulfur cluster in Figure 15
isn’t attached to anything, it could be deformed into a plane. It’s the combination
of the metal cluster together with the protein backbone that makes a structure
non-planar. For example, let’s consider the protein structure in Figure 16 which
contains the iron-sulfur cluster from Figure 15. We’ve drawn the backbone of the
protein as a straight line to make it easier to see, not because it’s actually straight.
If we imagine that the protein in Figure 16 is completely flexible, maybe we
could deform it into the plane. But what if we try unsuccessfully to deform it into
a plane? We might suspect that it can’t be done, but we’d like to be sure. With
Figure 16. Is this protein structure non-planar?
a bit of work we can check that Figure 16 doesn’t contain any non-trivial knots or
links. So if it can’t be deformed to lie in a plane, there must be some other reason.
Before showing that the protein in Figure 16 is indeed non-planar, we need to
learn a little bit more about non-planar graphs. First let’s recall from Chapter 13
that when we talk about an abstract graph, we are ignoring how the graph is situated
in space. If we want to consider the specific way that a graph is sitting in R3 we
refer to it as an embedding of the graph in space. Thus a knot is a particular
embedding of a circle in space. Using the language we learned in Chapter 3, we
would say that all knots have the same intrinsic topology, but whether or not a
knot is non-trivial depends on its extrinsic topology.
For graphs, the situation is a little more complicated. A graph in 3-dimensional
space can be non-planar because of either its extrinsic or its intrinsic topology. In
Figure 17 we illustrate three graphs whose non-planarity is entirely due to their
extrinsic topology. This means that none of these graphs can be deformed to lie
in a plane, but each has a planar embedding. In other words, each could be taken
apart and reassembled in the plane. You should see if you can draw each graph in
the plane without any crossings. Note that while it’s easy to see that the first graph
contains a knot and the second graph contains a link, showing that the graph on
the right can’t be deformed to lie in a plane is not easy.
Figure 17. These graphs are non-planar because of their extrinsic topology.
5. NON-PLANAR GRAPHS IN METALLOPROTEINS 371
In the Exercises in Chapter 10, we defined a link to be Brunnian if the link is

non-trivial but removing any single component leaves a trivial link. Graphs, such
as the one on the right in Figure 17, are said to be Brunnian because they are non-
planar, but if we remove any single edge, the graph that remains can be deformed
to lie in a plane. You should test this out by removing each edge in turn.
In contrast with the graphs in Figure 17 whose extrinsic topology makes them
non-planar, there are some graphs whose non-planarity comes from their intrinsic
topology. This means that if you took them apart and reassembled them, you still
wouldn’t be able to put them in a plane. For example, you may recall from Chapter
13 that the graphs K5 and K3,3 are special because they are the “smallest graphs”
which cannot be embedded in a plane without any edges intersecting. For K5 this
means that no matter how you put five vertices in a plane and start drawing edges
between pairs of vertices, you would not be able to draw non-intersecting edges
between every pair of vertices. You might want to try this to see how many edges
you can draw before two of them are forced to cross. Because the non-planarity of
K5 and K3,3 is independent of how they are embedded in space, we say that these
graphs are intrinsically non-planar.
Figure 18. The graphs K5 and K3,3 cannot be embedded in a plane.
Note that in Figure 18, the vertices of the graphs are represented by dots.
This is to distinguish them from the places where edges appear to intersect, which
are neither vertices nor actual intersections. We draw the graphs as if they had
intersecting edges because we are thinking of them as abstract graphs, and hence
we don’t want to specify a particular embedding of the graphs in 3-dimensional
space.
Now we return to the task of showing that the protein structure in Figure 16
is non-planar. If we can prove that it contains one of the intrinsically non-planar
graphs K5 or K3,3 , then the protein itself would be intrinsically non-planar. This
is because if there were a planar embedding of Figure 16, then it would contain a
planar embedding of K5 or K3,3 , which we know is impossible.
In Figure 19, we show that the protein structure in Figure 16 contains the graph
K3,3 . To do this, we label two sets of three vertices such that every vertex in one
set is connected by a disjoint path to every vertex in the other set. In the center
image, we have deleted all of the edges that won’t be part of our K3,3 . One of the
two sets of three vertices (shown as black dots in the center image) consists of three
of the four iron atoms that occur on the corners of the cube-like structure. The
other set of three vertices (shown as dark grey dots) consists of two sulfur atoms on
the corners of the cube-like structure together with one atom on the backbone. We
delete the light grey vertices from the center image and then observe that the dotted
edges, grey edges, and black edges form disjoint paths connecting every black dot
Figure 19. This protein contains a K3,3 graph and hence is non-planar.
to every dark grey dot. On the right in Figure 19, we illustrate K3,3 as an abstract
graph with the same color coding of vertices and edges. It now follows that this
protein structure is indeed intrinsically non-planar. So it’s not surprising that we
weren’t able to deform it to lie a plane.
As we saw in the above example, if a subgraph of a protein structure is non-
planar, then the protein structure itself is non-planar. However, this is not the only
way of showing that a structure is non-planar. If a smaller graph can be obtained
from a larger graph by collapsing edges and/or omitting edges or vertices, we say
the smaller graph is a minor of the larger graph. For example, in Figure 20, the
graph on the right is obtained from the graph on the left by collapsing the dotted
edges and deleting the grey vertices and the thick grey edge. Thus the graph on
the right is a minor of the graph on the left.
Figure 20. The graph on the right is obtained by collapsing the

dotted edges and deleting the grey vertices and grey edge.
We can use the following theorem, whose proof you might see in a graph theory
course, to show that certain protein structures are intrinsically non-planar.
Theorem 15.1 (Kuratowski’s Theorem). A graph is intrinsically non-planar
if and only if it contains K5 or K3,3 as a minor.
We saw above that the protein structure in Figure 19 is intrinsically non-planar
because it contains K3,3 . In Figure 21, we see that this protein also contains the
complete graph K5 as a minor. In particular, starting with the graph on the left, we
collapse the dashed edges and delete the grey vertices and grey edges to obtain the
6. THE PROTEIN STRUCTURE NITROGENASE 373
Figure 21. We collapse the dashed edges and delete the grey
vertices and edges to obtain K5 .
graph in the middle which has vertices numbered 1 through 5 and disjoint paths
between every pair of vertices. We can compare the graph in the middle to the
one on the right to see that indeed we have K5 . Thus by Kuratowski’s Theorem,
this gives us another way of showing that this protein structure is intrinsically
non-planar.
6. The Protein Structure Nitrogenase
In this section, we analyze the protein structure nitrogenase. This structure

is important because it’s responsible for turning nitrogen in the atmosphere into
nitrogen that can be consumed by bacteria, which are then consumed by other
organisms. However, nitrogenase is so complicated that biochemists studied it for
decades before it was finally completely mapped out in 2011.
Figure 22. A schematic drawing of the proteins in nitrogenase

and the metal clusters in the MoFe protein.
Figure 23. An M-cluster of the MoFe protein in nitrogenase.
Nitrogenase actually consists of two proteins, known as the Fe protein and then
MoFe protein, which work together. We focus here on the MoFe protein, which itself
contains four metal clusters: two of which are known as M-clusters and two of which
are known as P-clusters. To help you keep track of what is contained in what, you
should refer to the schematic drawing in Figure 22. Note that the clusters that
we’re interested in are the ones represented by grey disks in the figure. We will
show below that both the M-clusters and the P-clusters of the MoFe protein are
intrinsically non-planar.
Figure 24. An M-cluster contains K3,3 and hence is non-planar.

6. THE PROTEIN STRUCTURE NITROGENASE 375
Figure 25. A P-cluster attached to the backbones of nitrogenase.
We start with Figure 23, which is an M-cluster. Recall from Chapters 12 and
13 that thick wedges in a molecular diagram indicate bonds that are coming out of
the page. Note that the dashed lines in the figure indicate that the carbon atom in
the center is bound to the iron atoms around it.
In Figure 24, we show that the M-cluster is intrinsically non-planar by speci-
fying two sets of three vertices in a K3,3 graph together with edges between them.
In particular, in the second image, we label the top left sulfur atom, the central
carbon, and the top right sulfur atom by the letters a, b and c, and we label the
top left iron, bottom left iron, and central right iron by the numbers 1, 2, and 3.
We use dotted, grey, and black lines to highlight disjoint paths from the vertices a,
b, and c to the vertices 1, 2, and 3. In the third image we have deleted the edges
that are not part of the K3,3 . The last image shows the same color scheme for the
Figure 26. The P-cluster together with part of the backbone con-
tains K3,3 .
usual illustration of K3,3 . This shows that the M-cluster contains a K3,3 graph,
and hence by Kuratowski’s Theorem it is indeed intrinsically non-planar.
Now we turn our attention to a P-cluster of the MoFe protein. While the M-
cluster is intrinsically non-planar by itself, the P-cluster is only non-planar if we
consider it together with the backbone of the protein. Figure 25 illustrates one of
these clusters and its attachment to the two protein backbones, illustrated with
dotted curves.
In order to prove that this structure is non-planar, we show that it contains a
K3,3 graph. Starting with the structure from Figure 25, we remove all of the edges
that are not part of the K3,3 to get the middle illustration of Figure 26. Notice
that the arc joining vertex 2 to vertex a uses the grey part of the backbone. The
final illustration of K3,3 is color coded to look like the middle illustration. In the
Exercises you will show that the P-cluster not only contains an additional K3,3 ,
but also contains K5 as a minor.
7. Möbius Ladders in Metalloproteins
In Chapters 12 and 13 we learned that a Möbius ladder is a graph that can be

embedded in space in the form of a Möbius strip. Now we will see that such graphs
can actually be found in some proteins.
In Figure 27 we label the vertices of a Möbius ladder with three rungs to show
that, as abstract graph, it is identical to K3,3 . While the figure on the left looks
quite different from the one on the right, you should check that exactly the same
vertices are connected by edges in both graphs. In particular, in both graphs, every
vertex with a letter is connected to every vertex with a number, and no two letters
or two numbers are connected to each other.
Figure 27. As an abstract graph, a Möbius ladder with three

rungs is the same as K3,3 .
We saw in Figure 24 that the M-cluster of the MoFe protein in nitrogenase

contains the abstract graph K3,3 . Figure 28 illustrates a step-by-step deformation
of the K3,3 in the M-cluster to the Möbius ladder in Figure 27 that has the form of a
Möbius strip. In the Exercises you are asked to explain the steps of the deformation
shown in Figure 28.
Since a Möbius ladder with three rungs is an embedding of the non-planar
graph K3,3 , any protein structure containing a Möbius ladder with at least three
rungs is necessarily non-planar. However, just as knowing which particular knot
is contained in a protein structure gives us more information about the structure
than knowing just that it contains some unspecified knot, knowing that a protein
structure contains a K3,3 in the form of an embedded Möbius ladder gives us more
7. MÖBIUS LADDERS IN METALLOPROTEINS 377
Figure 28. A deformation of the K3,3 in the M-cluster of nitro-

genase to a Möbius ladder with three rungs.
information than just knowing that it contains some unspecified embedding of the
abstract graph K3,3 .
Even more surprising than the fact that nitrogenase contains a Möbius ladder
with three rungs in the form of a Möbius strip is the fact that protein structures can
contain Möbius ladders with four rungs in the form of Möbius strips. In particular,
Figure 29 depicts a protein structure containing an iron-sulfur cluster which we
will show contains such a Möbius ladder. We begin by illustrating the underlying
Figure 29. The underlying topological form of a protein structure

with an iron-sulfur cluster.
Figure 30. A deformation taking the iron-sulfur cluster from Fig-

ure 29 to a Möbius ladder with four rungs.
topological form of the protein structure on the right with numbers to help us see
the Möbius ladder.
In Figure 30 we omit the four thin grey edges from the cube-like form in the
center of the molecule. We can then shorten the edge connecting vertices 1 and 8
and the edge connecting vertices 4 and 5 to obtain the black infinity symbol with
four dotted curves connecting the two lobes in the second illustration in Figure 30.
We can then push the two lobes together in the back and rotate the graph to get
a Möbius ladder with four rungs in the form of a Möbius strip illustrated on the
right in Figure 30.
8. Möbius Ladders in Small Proteins
Möbius ladders can even be found in small proteins that do not contain metal
clusters. This includes some proteins in the cyclotide family. These small proteins
were originally discovered in the Democratic Republic of Congo where they are a
chemical component of a medicinal plant that was traditionally used to stimulate
labor and facilitate childbirth. Because cyclotides are much smaller than normal
proteins, some biochemists do not even consider them to be proteins. Nonetheless,
cyclotides are important because they have tremendous potential in the develop-
ment of medications to fight HIV, cancer, and antibiotic-resistant bacterial infec-
tions. Because of their stability, they have also been proposed as a scaffolding to
carry medications to a specified site in the body. For example, a cyclotide might be
used to deliver a chemotherapy drug directly to the site of a tumor and only release
it locally so it doesn’t get into the bloodstream. This might enable chemotherapy
drugs to have fewer side effects.
Figure 31. A cyclotide with three disulfide bonds.

8. MÖBIUS LADDERS IN SMALL PROTEINS 379
Figure 32. A deformation taking nerve growth factor to a Möbius

ladder with three rungs.
In the Exercises, you will show that the cyclotide illustrated in Figure 31 can
be deformed to a Möbius ladder with three rungs that has the form of a Möbius
strip. Notice that the backbone of this protein is actually circular, in contrast with
the backbones of most proteins. This cyclotide is often referred to as a “cysteine
knot motif” though it does not actually contain a knot. In order to make sense of
this terminology, we could interpret “knot motif” to simply mean that the cyclotide
is topologically non-planar.
Recall from Section 1 that when the termini of a protein molecule are close
to each other, adding a small arc between them gives the knot with the highest
probability. We use a similar method in Figure 32 to join the termini of the protein
known as nerve growth factor, after which we can deform it to a Möbius ladder. In
the Exercises you are asked to explain the steps of this deformation.
Nerve growth factor is a small protein that is important for growth and func-
tioning of nerve cells. It was discovered in the 1950s by the scientists Rita Levi-
Montalcini and Stanley Cohen, who were later awarded a Nobel Prize for their
discovery. Like the cyclotide in Figure 31, nerve growth factor is said to have a
Figure 33. This graph cannot be deformed into a Möbius ladder.

“cysteine knot motif”, which we can again interpret as referring to the non-planarity
of the structure.
The middle steps of the deformations illustrated in Figures 28, 30, and 32 each
include a loop with three or four edges joining different parts of the loop. It might
be tempting to assume that any such embedded graph can be deformed to a Möbius
ladder. But this is not always the case. In particular, you will show in the Exercises
that the graph illustrated in Figure 33 cannot be deformed into a Möbius ladder.
9. Goodbye for Now
We hope that reading this book has given you the background necessary to
explain to your friends and family what topology is and why it’s important. In
particular, if scientists want to understand the universe, they need to begin with
an understanding of the ideas of geometry and topology that were introduced in
Chapters 1–8. Knots are so ubiquitous in everyday life (think about your shoelaces
or computer cords, not to mention the 52 knots in your brain) that it seems like
everyone should want to learn the theory of them. But even beyond untangling
our shoelaces and cords, as we saw in Chapters 12–15, knots, links, and non-planar
graphs can be found in surprising places, such as molecular structures ranging from
small synthetic molecules to DNA and proteins. Perhaps the most important thing
you have gotten out of this book is the realization that what seems like abstract
esoteric mathematics at first glance may turn out to be fun to study and useful in
the real world. As we say “goodbye”, we hope this book has whet your appetite
to learn more mathematics, with the knowledge that no matter how theoretical an
area of mathematics seems, there is probably an application of it just waiting to be
discovered.
If you are interested in reading other books about topology and geometry and
knot theory and their applications, then you might enjoy the following:
• Weeks, J.R. (2002). The Shape of Space. Boca Raton, FL: CRC Press,
Taylor & Francis Group.
• Armstrong M.A. (1983). Basic Topology. New York: Springer Science+
Business Media, Inc.
• Messer, R., & Staffin, P. (2006) Topology Now! Washington DC: Mathe-
matical Association of America.
• Goodman, S.E. (2005). Beginning Topology. Providence, RI: American
Mathematical Society.
• Adams, C. (2004). The Knot Book. Providence, RI: American Mathe-
matical Society.
• Cromwell, C. (2004). Knots and Links. Cambridge: Cambridge Univer-
sity Press.
• Livingston, C. (1993). Knot Theory. Washington DC: Mathematical
Association of America.
• Murasugi, K. (1996). Knot theory and Its Applications. New York:
Springer Science+Business Media, Inc.
• Flapan, E. (2000). When Topology Meets Chemisty: A topological look
at molecular chirality. Cambridge: Cambridge University Press and the
Mathematical Association of America.
10. EXERCISES 381
10. Exercises
1. The structure in Figure 34 represents a protein backbone. Show that this may
or may not represent a knotted structure depending on how the endpoints of
the structure are extended.
2. Illustrate how a trefoil knot could be the result of threading one end of a protein
backbone through a twisted hairpin.
3. Show that the 51 knot could be obtained from a twisted hairpin by doing one
threading followed by another.
4. Draw a single threading of the unknot that creates a 61 knot. Draw two thread-
ings that together go from the unknot to a 62 knot.
5. Recall that an abstract graph of a Möbius ladder with three rungs is topolog-
ically equivalent to K3,3 . Is the abstract graph of a Möbius ladder with four
rungs topologically equivalent to K4,4 ?
6. Draw embeddings of Möbius ladders with three and four rungs which cannot
be deformed to the usual embedding that looks like a Möbius strip.
7. (a) Show that the iron-sulfur cluster in Figure 15 can be deformed to lie in a
plane.
(b) Show that without the backbones, the P-cluster in Figure 25 can be de-
formed to lie in a plane.
8. The P-cluster in Figure 25 contains multiple K3,3 graphs. Find one that is
distinct from the one presented in Figure 26.
9. Show that the P-cluster of nitrogenase contains K5 as a minor.
10. Explain the steps of the deformation illustrated in Figure 28.
11. Explain the steps of the deformation illustrated in Figure 32.
12. The structure in Figure 31 represents the “cyclotide knot motif”. Show that
this structure can be deformed to a Möbius ladder with three rungs in the form
of a Möbius strip.
13. In Figure 31, we say that the disulfide bonds have connectivity (1, 4), (2, 5),
(3, 6). This notation indicates that there are disulfide bonds between vertices
1 and 6, between vertices 2 and 4, and between vertices 3 and 5. Draw a pro-
tein structure with disulfide connectivity (1, 3), (2, 5), (4, 6). Is your drawing
intrinsically non-planar, or does whether or not it can be deformed into a plane
depend on the embedding of the structure in space?
14. Using the same labeling system as in the above problem, list the connectivity
of a protein that contains a Möbius ladder with four rungs.
15. Consider the graph of the protein in Figure 33 with the N and C termini
connected. Does this protein contain a Möbius ladder? If not, is it planar?
(a) In the exercises in Chapter 10, we defined a link to be Brunnian if the link
is non-trivial, but removing any single component gives a trivial link. Is
the chain mail protein link shown in Figure 14 Brunnian? Why or why not?
(b) How many rings have non-zero linking number with a given pentagonal
ring?
(c) How many rings have non-zero linking number with a given hexagonal
ring?
16. Illustrate a planar embedding of each of the graphs in Figure 17.
17. Show that if you remove any single edge from the graph on the right side of
Figure 17, then graph can be deformed to lie in a plane.
18. Show that if you remove or collapse a single edge of the graphs K5 and K3,3 ,
then you obtain a graph that can be embedded in the plane.
Index
2-fold branched cover, 316 crossing

negative (left-handed), 243, 313
Abbott, Edwin A., 5 nugatory, 247
achiral overcrossing, 270, 293, 351
chemically, 296, 315 positive (right-handed), 243, 313
geometrically, 294 resolving, 271
topologically, 306, 322 split (A-split or B-split), 283
Allis, Victor, 16 splitting, 282
angle, 39, 52, 170 undercrossing, 217, 243, 270, 293
right, 171, 188 crossing number, 218, 241, 269, 281
sum, 59, 109, 185 cube, 19
anti-cone point, 114, 205, 209 0-dimensional, 36
area 1-dimensional, 36
finite, 84, 140 2-dimensional, 36
infinite, 6 3-dimensional, 36, 173
associative, 137, 153 4-dimensional, 40, 179
atom, 221, 294, 315 n-dimensional, 47
automorphism, 321 solid, 41
identity, 322 cut and paste, 140
order, 322 cycloid, 378
cylinder, 15
B 3 , 13, see also ball, 3-dimensional Connect Four, 15
ball, 21 curved, 62
3-dimensional, 13, 128 finite, 31
boundary, 7 flat, 62
bridge number, 260 infinite, 6, 62, 110
slanted, 46
chiral
chemically, 296, 311, 316 D 2 , 13, see also disk
geometrically, 294, 311 dimension, 6, 13
intrinsically, 325 disk, 7, 21, 59, 96
topologically, 306, 311 open, 74
circle, 6 distance, 31, 50
great, 9, 58 distributive, 183
circumference, 197 DNA, 216, 292, 297, 312, 330
commutative, 137, 153 axis, 335
cone point, 114, 209 backbone, 332
cone, infinite, 26 base pairs, 331
connected sum duplex, 331
of knots, 220, 233, 249 overwound, 346
of surfaces, 203, 255 relaxed, 331, 335
continued fraction, 353 site, 335, 348
Conway, John H., 353 stressed, 331
cross-cap, 100, 142 substrate, 348
383
384 INDEX
twist, 340 alternating, 248, 281

underwound, 346 composite, 221
equivalent, 217
Einstein-Rosen bridge, 106, 135 figure eight, 220, 227, 240, 249, 261, 302,
enzyme, 347 350, 357
ball, 355 invariant, 240, 269, 278
recombinase, 348 left or right handed, 313
Euler characteristic, 144, 257 mirror image, 219
Euler, Leonhard, 144 non-alternating, 248, 269, 280
extended diagram, 17, 61, 93 non-trivial, 218, 365
oriented, 232, 243, 286
Flatland, 5
prime, 221
genus, 148 projection, 217
of a knot, 254 square, 236
of a Seifert surface, 254 trefoil, 218, 242, 253, 271, 281, 286, 312,
geodesic, 53, 174 363
Geometrization Conjecture, 153 trivial, 218
geometry
lattice
Euclidean, 191
configuration, 262
extrinsic, 52, 293
number, 263
intrinsic, 52, 91, 109
line, 6
local, 70, 180
geodesic, 186, 192
spherical, 197
straight, 8
gluing, 13
link, 216
abstractly glued, 15, 62
alternating, 248
gluing diagram, 13, 61, 88, 109, 138, 145
Brunnian, 265
hexagonal, 110, 154, 208
component, 217
octagonal, 128, 206
Hopf, 219, 275, 350, 357, 367
polygonal, 111, 205
invariant, 272
polyhedral, 122
oriented, 244, 277
graph, 294, 298, 311
reduced, 248
complete, 322
trivial, 218, 265
complete bipartite, 322
Whitehead, 350, 357
embedding, 320, 370, 376
linking number, 287
non-planar, 361, 376
longitude, 78, 148, 199
handedness (left and right handed), 291, loop, 67
301 non-separating, 147
handle, 6, 141 unshrinkable, 68
adding, 252 Lord Kelvin, 222, 296
hemisphere, 13, 84, 97
Möbius
homogeneous, 71, 92, 109, 181, 185, 205
ladder, 308, 315, 329, 376
I, see also interval, 160 strip, 88, 335, 376
identity element, 137, 220 manifold, 71
induction, 147, 154 1-manifold, 85
inverse, 138, 153 2-manifold, 71, 91, 128
isotropic, 181 3-manifold, 72, 120, 151
n-manifold, 71
Jones, Vaughan, 279 flat, 109, 171
turn, 127
K 2 , see also Klein bottle, 91 mathematical model, 355
K 3 , see also Klein bottle, 3-dimensional, 92 Maxwell, James, 223
Klein bottle, 90, 113, 145 meridian, 78, 148
3-dimensional, 92, 122, 151 mirror reversed, 87
flat, 93, 109 Mislow, Kurt, 318
tic-tac-toe, 93 molecule, 291, 311, 329, 361
Klein, Felix, 90
knot, 216 non-orientable, 89, 141
INDEX 385
orientable, 89, 127, 141 S 1 , 13, see also circle

S 2 , 13, see also sphere, 2-sphere
(p, q)-curve, 79 S 3 , 20, see also sphere, 3-sphere
P 1 , see also projective space, saddle surface, 60
1-dimensional, 107 Seifert surface, 252
P 2 , see also projective plane, sidedness
2-dimensional, 96 1-sided, 102
P 3 , see also projective space, 2-sided, 102, 252
3-dimensional, 100 Simon, Jon, 316
parallel, 190 space, 8
path sphere, 5
orientation reversing, 89, 147 2-sphere, 22
shortest, 8, 58, 186
3-sphere, 21
straightest, 8
state model, 283
Perko pair, 240
stereographic projection, 75
Perko, Kenneth, 240
stick number, 261
plane, 5
supercoil, 334
flat, 4, 50
negatively supercoiled, 345
hyperbolic, 201
positively supercoiled, 345
point
stable, 334
at infinity, 75
unstable, 334
polynomial
symmetry
X-, 270, 277, 313
3-fold, 326
bracket, 270, 313
mirror image, 291
Jones, 279, 314
synaptic complex, 348
product, 157
geometric, 170, 203
T 2 , 13, see also torus, 2-torus
projective plane, 96, 139, 207
T 3 , 19, see also torus, 3-torus
projective space
Tait’s Conjectures, 248
1-dimensional, 107
Tait, Peter Guthrie, 222, 248
3-dimensional, 100, 151
tangle, 350
property
0-tangle, 350
extrinsic, 50, 102, 216
∞-tangle, 351
geometric, 52, 109, 339, 343
equation, 356
global, 69, 70
equivalent, 350
local, 69
integer, 351
topological, 67, 144, 240, 339
numerator closure, 354
protein, 216, 361
rational, 353
backbone, 361
recombination, 356
Pythagorean Theorem, 192
site, 355
R2 , 13, see also plane substrate, 355
R3 , 13 sum, 354
R4 , 21, 34 vector, 352
Rn , 71 tetrahedron, 42
recombination, 333 topology
product, 349 extrinsic, 64, 217, 347
Reidemeister moves, 224, 272 intrinsic, 66, 91, 217
Reidemeister’s Theorem, 225, 241 torus, 5
Reidemeister, Kurt, 224 2-holed, 5, 68, 137, 148
repeats 2-torus, 22
direct, 359 3-holed, 5, 137
inverted, 359 3-torus, 19, 62, 120
replication, 332, 346 curved, 15, 50, 102
right angle, 41 extended diagram, 18
rigid motion, 52, 293, 300, 339 flat, 15, 61, 92, 107, 109
rubber glove knotted, 65
Euclidean, 301 unknotted, 65
topological, 305 triangle, 42
386 INDEX
complementary, 194
congruent, 210
equilateral, 42
hyperbolic, 201
isosceles right, 195
planar, 61, 110
spherical, 194
tricolorable, 228, 242
twist, 351
horizontal, 352
negative, 351
positive, 276, 351
vertical, 352
Uniformization of Surfaces, 209

universe, 3, 7
1-dimensional, 31
4-dimensional, 31
infinite-dimensional, 31
unknot, 218
unknotting number, 258, 365
unlink, 218
vertex, 315
distance, 321
valence, 321
Walba, David, 315

writhe, 244, 277, 313, 339
average, 339
This book is an elementary introduction to geometric topology and its applica-
tions to chemistry, molecular biology, and cosmology. It does not assume any
PDWKHPDWLFDO RU VFLHQWLÀF EDFNJURXQG VRSKLVWLFDWLRQ RU HYHQ PRWLYDWLRQ WR
study mathematics. It is meant to be fun and engaging while drawing students
in to learn about fundamental topological and geometric ideas. Though the
ERRNFDQEHUHDGDQGHQMR\HGE\QRQPDWKHPDWLFLDQVFROOHJHVWXGHQWVRUHYHQ
eager high school students, it is intended to be used as an undergraduate text-
book.
7KHERRNLVGLYLGHGLQWRWKUHHSDUWVFRUUHVSRQGLQJWRWKHWKUHHDUHDVUHIHUUHGWR
LQWKHWLWOH3DUWGHYHORSVWHFKQLTXHVWKDWHQDEOHWZRDQGWKUHHGLPHQVLRQDO
FUHDWXUHVWRYLVXDOL]HSRVVLEOHVKDSHVIRUWKHLUXQLYHUVHDQGWRXVHWRSRORJLFDO
and geometric properties to distinguish one such space from another. Part 2 is
DQLQWURGXFWLRQWRNQRWWKHRU\ZLWKDQHPSKDVLVRQLQYDULDQWV3DUWSUHVHQWV
applications of topology and geometry to molecular symmetries, DNA, and
proteins. Each chapter ends with exercises that allow for better understanding
of the material.
7KHVW\OHRIWKHERRNLVLQIRUPDODQGOLYHO\7KRXJKDOORIWKHGHÀQLWLRQVDQG
WKHRUHPV DUH H[SOLFLWO\ VWDWHG WKH\ DUH JLYHQ LQ DQ LQWXLWLYH UDWKHU WKDQ D
ULJRURXVIRUPZLWKVHYHUDOKXQGUHGVRIÀJXUHVLOOXVWUDWLQJWKHH[SRVLWLRQ7KLV
DOORZV VWXGHQWV WR GHYHORS LQWXLWLRQ DERXW WRSRORJ\ DQG JHRPHWU\ ZLWKRXW
getting bogged down in technical details.
For additional information

and updates on this book, visit
www.ams.org/bookpages/mbk-96
AMS on the Web

www.ams.org
MBK/96

Knots, Molecules, and The Universe - An Introduction To Topology (Flapan)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Knots, Molecules, and The Universe - An Introduction To Topology (Flapan)

Uploaded by

Copyright:

Available Formats

KNOTS, MOLECULES,

AND THE UNIVERSE:

Maia Averett David Clark Lew Ludwig

AMERICAN MATHEMATICAL SOCIETY

For additional information and updates on this book, visit

Library of Congress Cataloging-in-Publication Data

Chapter 1. An Introduction to the Shape of the Universe 3

Chapter 2. Visualizing Four Dimensions 29

Chapter 3. Geometry and Topology of Diﬀerent Universes 49

Chapter 5. Flat Manifolds 109

Chapter 6. Connected Sums of Spaces 135

Chapter 7. Products of Spaces 157

2. Products of Spaces 159

Chapter 8. Geometries of Surfaces 185

Chapter 9. Introduction to Knot Theory 215

Chapter 10. Invariants of Knots and Links 239

Chapter 11. Knot Polynomials 269

Chapter 12. Mirror Image Symmetry from Diﬀerent Viewpoints 291

Chapter 13. Techniques to Prove Topological Chirality 311

Chapter 14. The Topology and Geometry of DNA 329

Chapter 15. The Topology of Proteins 361

Emille Davie Lawrence co-wrote Chapters 1 and 8, edited the pref-

solutions for exercises in Chapters 1 and 8.

the entire manuscript and the solutions.

provided technical support during PCMI, TEXed solutions and created

and edited ﬁgures throughout the manuscript. In particular, the pictures

edited solutions to exercises.

some solutions to exercises in Chapters 8 and 15.

An Introduction to the Shape of the Universe

1. To Inﬁnity and Beyond

Figure 1. A. 3D-Girl waving to the crowd as her spaceship blasts oﬀ.

2. A. Square and His Universe

Figure 2. A portrait of A. Square.

A. Square lives in a 2-dimensional universe. He and his space have no thickness,

Figure 3. A. Square on the surfaces of a sphere, a torus, and a

We say that A. Square’s universe is 2-dimensional because he can only move in

Figure 4. A. Dot lives in a circle and A. Dash lives in a knotted circle.

Figure 5. Some 2-dimensional universes with inﬁnite area.

Figure 6. If A. Square lived in a disk, when he reached the edge

As we think about possible universes for A. Square and for ourselves, we do

3. Straightest Paths in Flatland

Figure 7. Straightest paths on a plane and a sphere.

In order to see if a similar method would enable A. Square to diﬀerentiate

Figure 8. A straightest path on a torus.

4. Exploring the Shape of a Cave

Having watched A. Square use straightest paths to compare diﬀerent universes,

Figure 11. A cave which looks like the inside of a doughnut.

Figure 12. A cave with one four-way intersection.

Figure 13. These caves each have two three-way intersections.

Figure 14. On the left there’s a path from a three-way intersec-

5. Creating Universes by Gluing

Figure 15. A sphere consists of two hemispheres that are glued

Figure 16. When A. Square arrives at the boundary of one disk,

Figure 17. Gluing up opposite sides of a square.

As 3-dimensional people, we can physically glue up opposite sides of the square

If we want to physically glue up opposite sides of a square, the square needs to

6. Games on a Gluing Diagram for a Cylinder and a Torus

One way to check your understanding of abstract gluings is to play games on a

Figure 20. The game of Connect Four on a cylinder.

Figure 21. X has won this game of torus tic-tac-toe.

Figure 22. A. Square sees a copy of himself up ahead.

Figure 23. A portion of an extended diagram of a ﬂat torus with

Figure 24. The three black X’s in an extended diagram represent