Geometri PDF
Uploaded by
Taufiq Abdul GaniGeometri PDF
Uploaded by
Taufiq Abdul GaniGeometry
with an Introduction to
Cosmic Topology
2018 Edition
Michael P. Hitchman
Michael P. Hitchman
Department of Mathematics
Lin eld College
McMinnville, OR 97128
http://mphitchman.com
© 2017-2018 by Michael P. Hitchman
This work is licensed under the Creative Commons Attribution-ShareAlike
4.0 International License. To view a copy of this license, visit http://
creativecommons.org/licenses/by-sa/4.0/.
2018 Edition
ISBN-13: 978-1717134813
ISBN-10: 1717134815
A current version can always be found for free at
http://mphitchman.com
Cover image: Five geodesic paths between two points in a two-holed torus with
constant curvature.
Preface
Geometry with an Introduction to Cosmic Topology approaches geometry
through the lens of questions that have ignited the imagination of stargazers
since antiquity. What is the shape of the universe? Does the universe have an
edge? Is it in nitely big?
This text develops non-Euclidean geometry and geometry on surfaces
at a level appropriate for undergraduate students who have completed a
multivariable calculus course and are ready for a course in which to practice
the habits of thought needed in advanced courses of the undergraduate
mathematics curriculum. The text is also suited to independent study, with
essays and discussions throughout.
Mathematicians and cosmologists have expended considerable amounts of
eort investigating the shape of the universe, and this eld of research is called
cosmic topology. Geometry plays a fundamental role in this research. Under
basic assumptions about the nature of space, there is a simple relationship
between the geometry of the universe and its shape, and there are just three
possibilities for the type of geometry: hyperbolic geometry, elliptic geometry,
and Euclidean geometry. These are the geometries we study in this text.
Chapters 2 through 7 contain the core mathematical content. The text
follows the Erlangen Program, which develops geometry in terms of a space and
a group of transformations of that space. Chapter 2 focuses on the complex
plane, the space on which we build two-dimensional geometry. Chapter 3
details transformations of the plane, including Möbius transformations. This
chapter marks the heart of the text, and the inversions in Section 3.2 mark
the heart of the chapter. All non-Euclidean transformations in the text
are built from inversions. We formally de ne geometry in Chapter 4, and
pursue hyperbolic and elliptc geometry in Chapters 5 and 6, respectively.
Chapter 7 begins by extending these geometries to dierent curvature scales.
Section 7.4 presents a uni ed family of geometries on all curvature scales,
emphasizing key results common to them all. Section 7.5 provides an informal
development of the topology of surfaces, and Section 7.6 relates the topology of
surfaces to geometry, culminating with the Gauss-Bonnet formula. Section 7.7
discusses quotient spaces, and presents an important tool of cosmic topology,
the Dirichlet domain.
Two longer essays bookend the core content. Chapter 1 introduces the
geometric perspective taken in this text. In my experience it is very helpful
to spend time discussing this content in class. The Coneland and Saddleland
activities (Example 1.3.5 and Example 1.3.7) have proven particularly helpful
for motivating the content of the text. In Chapter 8, after having developed
two-dimensional non-Euclidean geometry and the topology of surfaces, we
glance meaningfully at the present state of research in cosmic topology. Section
8.1 oers a brief survey of three-dimensional geometry and 3-manifolds, which
iii
iv
provide possible shapes of the universe. Sections 8.2 and 8.3 present two
research programs in cosmic topology: cosmic crystallography and circles-in-
the-sky. Measurements taken and analyzed over the last twenty years have
greatly altered the way many cosmologists view the universe, and the text ends
with a discussion of our present understanding of the state of the universe.
Compass and ruler constructions play a visible role in the text, primarily
because inversions are emphasized as the basic building blocks of transfor-
mations. Constructions are used in some proofs (such as the Fundamental
Theorem of Möbius Transformations) and as a guide to de nitions (such as
the arc-length dierential in the hyperbolic plane). We encourage readers to
practice constructions as they read along, either with compass and ruler on
paper, or with software such as The Geometer's Sketchpad or Geogebra. Some
Geometer's Sketchpad templates and activites related to the text can be found
at the text's website.
Reading the text online An online text is fabulous at linking content,
but we emphasize that this text is meant to be read. It was written to tell
a mathematical story. It is not meant to be a collection of theorems and
examples to be consulted as a reference. As such, online readers of this text
are encouraged to turn the pages using the arrow buttons on the page as
opposed to clicking on section links. Read the content slowly, participate in
the examples, and work on the exercises. Grapple with the ideas, and ask
questions. Feel free to email the author with questions or comments about the
material.
Changes from the previously published version For those familiar with
the oringial version of the text published by Jones & Bartlett, we note a few
changes in the current edition. First, the numbering scheme has changed, so
Example and Theorem and Figure numbers will not match the old hard copy.
Of course the numbering schemes on the website and the new print options
of the text do agree. Second, several exercises have been added. In sections
with additional exercises, the new ones typically appear at the end of the
section. Finally, Chapter 7 has been reorganized in an eort to place more
emphasis on the family (Xk , Gk ), and the key theorems common to all these
geometries. This family now receives its own section, Section 7.4. The previous
Section 7.4 (Observing Curvature in a Universe) has been folded into Section
7.3. Finally, the essays in Chapter 8 on cosmic topology and our understanding
of the universe have been updated to include research done since the original
publication of this text, some of which is due to sharper measurements of the
temperature of the cosmic microwave background radiation obtained with the
launch of the Planck satellite in 2009.
Acknowledgements
Many people helped with the development of this text. Je Weeks' text The
Shape of Space inspired me to rst teach a geometry course motivated by
cosmic topology. Colleagues at The College of Idaho encouraged me to teach
the course repeatedly as the manuscript developed, and students there provided
valuable feedback on how this story can be better told.
I would like to thank Rob Beezer, David Farmer, and all my colleagues
at the UTMOST 2017 Textbook Workshop and in the PreTeXt Community
for helping me convert my dusty tex code to a workable PreTeXt document
in order to make the text freely available online, both as a webpage and as a
printable document. I also thank Jennifer Nordstrom for encouraging me to
use PreTeXt in the rst place.
I owe my colleagues at The College of Idaho and the University of Oregon
a debt of gratitude for helping to facilitate a sabbatical during which an earlier
incarnation of this text developed and was subsequently published in 2009 with
Jones & Bartlett. Richard Koch discussed some of its content with me, and Jim
Dull patiently discussed astrophysics and cosmology with me at a moment's
notice.
The pursuit of detecting the shape of the universe has rapidly evolved over
the last twenty years, and Marcelo Rebouças kindly answered my questions
regarding the content of Chapter 8 back in 2008, and pointed me to the
latest papers. I am also grateful to the authors of the many cosmology papers
accessible to the amateur enthusiast and written with the intent to inform.
Finally, I would like to thank my family for their support during this
endeavor, for letting this project permeate our home, and for encouraging
its completion when we might have done something else, again.
v
vi
To my son, Jasper
Contents
Preface iii
Acknowledgements v
1 An Invitation to Geometry 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 A Brief History of Geometry . . . . . . . . . . . . . . . . . . . 3
1.3 Geometry on Surfaces: A First Look . . . . . . . . . . . . . . . 7
2 The Complex Plane 13
2.1 Basic Notions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Polar Form of a Complex Number . . . . . . . . . . . . . . . . 15
2.3 Division and Angle Measure . . . . . . . . . . . . . . . . . . . . 17
2.4 Complex Expressions . . . . . . . . . . . . . . . . . . . . . . . . 19
3 Transformations 25
3.1 Basic Transformations of C . . . . . . . . . . . . . . . . . . . . 25
3.2 Inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 The Extended Plane . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4 Möbius Transformations . . . . . . . . . . . . . . . . . . . . . . 47
3.5 Möbius Transformations: A Closer Look . . . . . . . . . . . . . 54
4 Geometry 63
4.1 The Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2 Möbius Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5 Hyperbolic Geometry 72
5.1 The Poincaré Disk Model . . . . . . . . . . . . . . . . . . . . . 72
5.2 Figures of Hyperbolic Geometry . . . . . . . . . . . . . . . . . 77
5.3 Measurement in Hyperbolic Geometry . . . . . . . . . . . . . . 81
5.4 Area and Triangle Trigonometry . . . . . . . . . . . . . . . . . 90
5.5 The Upper Half-Plane Model . . . . . . . . . . . . . . . . . . . 103
6 Elliptic Geometry 109
6.1 Antipodal Points . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.2 Elliptic Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.3 Measurement in Elliptic Geometry . . . . . . . . . . . . . . . . 120
6.4 Revisiting Euclid's Postulates . . . . . . . . . . . . . . . . . . . 128
ix
x Contents
7 Geometry on Surfaces 130
7.1 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
7.2 Elliptic Geometry with Curvature k>0 . . . . . . . . . . . . . 134
7.3 Hyperbolic Geometry with Curvature k < 0 . . . . . . . . . . . 137
7.4 The Family of Geometries (Xk , Gk ) . . . . . . . . . . . . . . . . 142
7.5 Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
7.6 Geometry of Surfaces . . . . . . . . . . . . . . . . . . . . . . . . 162
7.7 Quotient Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
8 Cosmic Topology 178
8.1 Three-Dimensional Geometry and 3-Manifolds . . . . . . . . . 178
8.2 Cosmic Crystallography . . . . . . . . . . . . . . . . . . . . . . 188
8.3 Circles in the Sky . . . . . . . . . . . . . . . . . . . . . . . . . . 195
8.4 Our Universe . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
A List of Symbols 202
Index 205
1
An Invitation to Geometry
How can it be that mathematics, being after all a product of
human thought which is independent of experience, is so admirably
appropriate to the objects of reality?
Albert Einstein
Out of nothing I have created a strange new universe.
János Bolyai
1.1 Introduction
Imagine you are a two-dimensional being living in a two-dimensional universe.
Mathematicians in this universe often represent its shape as an in nite plane,
exactly like the xy -plane you've used as the canvas in your calculus courses.
Your two-dimensional self has been taught in geometry that the angles of
◦
any triangle sum to 180 . You may have even constructed some triangles to
check. Builders use the Pythagorean theorem to check whether two walls meet
at right angles, and houses are sturdy.
a c
Figure 1.1.1: Measure a, b, and c and check whether a2 + b2 = c2 . If equality
holds, the corner is square!
The in nite plane model of the two-dimensional universe works well enough
for most purposes, but cosmologists and mathematicians, who notice that
everything within the universe is nite, consider the possibility that the
universe itself is nite. Would a nite universe have a boundary? Can it
have an edge, a point beyond which one cannot travel? This possibility is
1
2 1. An Invitation to Geometry
unappealing because a boundary point would be physically dierent from the
rest of space. But how can a nite universe have no boundary?
In a stroke as bold as it is simple, a two-dimensional mathematician suggests
that the universe looks like a rectangular region with opposite edges identi ed.
Consider a at, two-dimensional rectangle. In fact, visualize a tablet screen.
Now imagine that you are playing a video game called Asteroids. As you shoot
the asteroids and move your ship around the screen, you nd that if you go o
the top of the screen your ship reappears on the bottom; and if you go o the
screen to the left you reappear on the right. In Figure 1.1.2 there are just ve
asteroids. One has partially moved o the top of the screen and reappeared
below, while a second is half way o the right hand edge and is reappearing
to the left.
Figure 1.1.2: A nite two-dimensional world with no boundary.
Thus, the top edge of the rectangle has been identi ed, point by point,
with the bottom edge. In three dimensions one can physically achieve this
identi cation, or gluing, of the edges. In particular, one can bend the rectangle
to produce a cylinder, being careful to join only the top and bottom edges
together, and not any other points. The left and right edges of the rectangle
have now become the left and right circles of the cylinder, which themselves
get identi ed, point by point. Bend the cylinder to achieve this second gluing,
and one obtains a donut, also called a torus.
Figure 1.1.3: The video screen in Figure 1.1.2 is equivalent to a torus.
Of course your two-dimensional self would not be able to see this torus
surface in 3-space, but you could understand the space perfectly well in its
rectangle-with-edges-identi ed form. It is clearly a nite area universe without
any edge.
A sphere, like the surface of a beach ball, is another nite area two-
dimensional surface without any edge. A bug cruising around on the surface
of a sphere will observe that locally the world looks like a at plane, and that
the surface has no edges.
Consideration of a nite-area universe leads to questions about the type
of geometry that applies to the universe. Let's look at a sphere. On small
scales Euclidean geometry works well enough: small triangles have angle sum
◦
essentially equal to 180 , which is a de ning feature of Euclidean geometry.
But on a larger scale, things go awry. A very large triangle drawn on the
◦
surface of the sphere has an angle sum far exceeding 180 . (By triangle, we
1.2. A Brief History of Geometry 3
mean three points on the surface, together with three paths of shortest distance
between the points. We'll discuss this more carefully later.)
Consider the triangle formed by the north pole and two points on the
◦
equator in Figure 1.1.4. The angle at each point on the equator is 90 , so the
◦
total angle sum of the triangle exceeds 180 by the amount of the angle at the
north pole. We conclude that a non-Euclidean geometry applies to the sphere
on a global scale.
Figure 1.1.4: A triangle on a sphere.
In fact, there is a wonderful relationship between the topology (shape) of
a surface, and the type of geometry that it inherits, and a primary goal of
this book is to arrive at this relationship, given by the pristine Gauss-Bonnet
equation
kA = 2πχ.
We won't explain this equation here, but we will point out that geometry
is on the left side of the equation, and topology is on the right. So, if a two-
dimensional being can deduce what sort of global geometry holds in her world,
she can greatly reduce the possible shapes for her universe. Our immediate
task in the text is to study the other, non-Euclidean types of geometry that
may apply on surfaces.
1.2 A Brief History of Geometry
Geometry is one of the oldest branches of mathematics, and most important
among texts is Euclid's Elements. His text begins with 23 de nitions, 5
postulates, and 5 common notions. From there Euclid starts proving results
about geometry using a rigorous logical method, and many of us have been
asked to do the same in high school.
Euclid's Elements served as the text on geometry for over 2000 years, and it
has been admired as a brilliant work in logical reasoning. But one of Euclid's
ve postulates was also the center of a hot debate. It was this debate that
ultimately led to the non-Euclidean geometries that can be applied to dierent
surfaces.
Here are Euclid's ve postulates:
1. One can draw a straight line from any point to any point.
2. One can produce a nite straight line continuously in a straight line.
3. One can describe a circle with any center and radius.
4. All right angles equal one another.
5. If a straight line falling on two straight lines makes the interior angles
on the same side less than two right angles, the two straight lines, if
produced inde nitely, meet on that side on which the angles are less
than two right angles.
4 1. An Invitation to Geometry
Does one postulate not look like the others? The rst four postulates are
short, simple, and intuitive. Well, the second might seem a bit odd, but all
Euclid is saying here is that you can produce a line segment to any length you
want. However, the 5th one, called the parallel postulate , is not short or
simple; it sounds more like something you would try to prove than something
you would take as given.
Indeed, the parallel postulate immediately gave philosophers and other
thinkers ts, and many tried to prove that the fth postulate followed from
the rst four, to no avail. Euclid himself may have been bothered at some level
by the parallel postulate since he avoids using it until the proof of the 29th
proposition in his text.
In trying to make sense of the parallel postulate, many equivalent
statements emerged. The two equivalent statements most relevant to our study
are these:
0
5 . Given a line and a point not on the line, there is exactly one
line through the point that does not intersect the given line.
500 . The sum of the angles of any triangle is 180 .
◦
Reformulation 50 of the parallel postulate is called Playfair's Axiom after the
Scottish mathematician John Playfair (1748-1819). This version of the fth
postulate will be the one we alter in order to produce non-Euclidean geometry.
The parallel postulate debate came to a head in the early 19th century.
Farkas Bolyai (1775-1856) of Hungary spent much of his life on the problem
of trying to prove the parallel postulate from the other four. He failed, and
he fretted when his son János (1802-1860) started following down the same
tormented path. In an oft-quoted letter, the father begged the son to end the
obsession:
For God's sake, I beseech you, give it up. Fear it no less than the
sensual passions because it too may take all your time and deprive
you of your health, peace of mind and happiness in life.
1
But János continued to work on the problem, as did the Russian mathematician
Nikolai Lobachevsky (1792-1856). They independently discovered that a well-
de ned geometry is possible in which the rst four postulates hold, but the
fth doesn't. In particular, they demonstrated that the fth postulate is not
a necessary consequence of the rst four.
In this text we will study two types of non-Euclidean geometry. The
rst type is called hyperbolic geometry, and is the geometry that Bolyai and
Lobachevsky discovered. (The great Carl Friedrich Gauss (1777-1855) had
also discovered this geometry; however, he did not publish his work because
he feared it would be too controversial for the establishment.) In hyperbolic
geometry, Euclid's fth postulate is replaced by this:
5H. Given a line and a point not on the line, there are at least two
lines through the point that do not intersect the given line.
◦
In hyperbolic geometry, the sum of the angles of any triangle is less than 180 ,
a fact we prove in Chapter 5.
The second type of non-Euclidean geometry in this text is called elliptic
geometry, which models geometry on the sphere. In this geometry, Euclid's
fth postulate is replaced by this:
1 See for instance, Martin Gardner's book The Colossal Book of Mathematics , W.W.
Norton & Company (2001), page 176.
1.2. A Brief History of Geometry 5
5E. Given a line and a point not on the line, there are zero lines
through the point that do not intersect the given line.
◦
In elliptic geometry, the sum of the angles of any triangle is greater than 180 ,
a fact we prove in Chapter 6.
The Pythagorean Theorem The celebrated Pythagorean theorem depends
upon the parallel postulate, so it is a theorem of Euclidean geometry. However,
we will encounter non-Euclidean variations of this theorem in Chapters 5 and 6,
and present a uni ed Pythagorean theorem in Chapter 7, with Theorem 7.4.7,
a result that appeared recently in [20].
The Pythagorean theorem appears as Proposition 47 at the end of Book I
of Euclid's Elements, and we present Euclid's proof below. The Pythagorean
theorem is fundamental to the systems of measurement we utilize in this text,
in both Euclidean and non-Eucidean geometries. We also remark that the nal
proposition of Book I, Proposition 48, gives the converse that builders use: If
we measure the legs of a triangle and nd that c2 = a2 + b2 then the angle
opposite c is right. The interested reader can nd an online version of Euclid's
Elements here [29].
Theorem 1.2.1 The Pythagorean Theorem. In right-angled triangles the
square on the side opposite the right angle equals the sum of the squares on the
sides containing the right angle.
Proof. Suppose we have right triangle ABC as in Figure 1.2.2 with
right angle at C, and side lengths a, b, and c, opposite corners A, B and C ,
respectively. In the gure we have extended squares from each leg of the
triangle, and labeled various corners. We have also constructed the line through
C parallel toAD, and let L and M denote the points of intersection of this line
with AB and DE , respectively. One can check that ∆KAB is congruent to
∆CAD. Moreover, the area of ∆KAB is one half the area of the square AH .
This is the case because they have equal base (segment KA) and equal altitude
(segment AC ). By a similar argument, the area of ∆DAC is one half the area
of the parallelogram AM . This means that square AH and parallelogram AM
2
have equal areas, the value of which is b .
One may proceed as above to argue that the areas of square BG and
2
parallelogram BM are also equal, with value a . Since the area of square
BD, which equals c2 , is the sum of the two parallelogram areas, it follows that
a2 + b2 = c2 .
The arrival of non-Euclidean geometry soon caused a stir in circles outside
the mathematics community. Fyodor Dostoevsky thought non-Euclidean
geometry was interesting enough to include in The Brothers Karamazov, rst
published in 1880. Early in the novel two of the brothers, Ivan and Alyosha, get
reacquainted at a tavern. Ivan discourages his younger brother from thinking
about whether God exists, arguing that if one cannot fathom non-Euclidean
geometry, then one has no hope of understanding questions about God.
2
One of the rst challenges of non-Euclidean geometry was to determine
its logical consistency. By changing Euclid's parallel postulate, was a system
created that led to contradictory theorems? In 1868, the Italian mathematician
Enrico Beltrami (1835-1900) showed that the new non-Euclidean geometry
could be constructed within the Euclidean plane so that, as long as Euclidean
2 See, for instance, The Brothers Karamazov, Fyodor Dostoevsky (a new translation by
Richard Pevear and Larissa Volokhonsky), North Point Press (1990), page 235.
6 1. An Invitation to Geometry
K C
A L B
D M E
Figure 1.2.2: Proving the Pythagorean theorem
geometry was consistent, non-Euclidean geometry would be consistent as well.
Non-Euclidean geometry was thus placed on solid ground.
This text does not develop geometry as Euclid, Lobachevsky, and Bolyai
did. Instead, we will approach the subject as the German mathematician Felix
Klein (1849-1925) did.
Whereas Euclid's approach to geometry was additive (he started with
basic de nitions and axioms and proceeded to build a sequence of results
depending on previous ones), Klein's approach was subtractive. He started
with a space and a group of allowable transformations of that space. He
then threw out all concepts that did not remain unchanged under these
transformations. Geometry, to Klein, is the study of objects and functions
that remain unchanged under allowable transformations.
Klein's approach to geometry, called the Erlangen Program after the
university at which he worked at the time, has the bene t that all three
geometries (Euclidean, hyperbolic and elliptic) emerge as special cases from
a general space and a general set of transformations.
The next three chapters will be devoted to making sense of and working
through the preceding two paragraphs.
Like so much of mathematics, the development of non-Euclidean geom-
etry anticipated applications. Albert Einstein's theory of special relativity
illustrates the power of Klein's approach to geometry. Special relativity, says
Einstein, is derived from the notion that the laws of nature are invariant with
respect to Lorentz transformations.
3
Even with non-Euclidean geometry in hand, Euclidean geometry remains
central to modern mathematics because it is an excellent model for our local
geometry. The angles of a triangle drawn on this paper do add up to 180 .
◦
Even galactic triangles determined by the positions of three nearby stars have
◦
angle sum indistinguishable from 180 .
However, on a larger scale, things might be dierent.
Maybe we live in a universe that looks at (i.e., Euclidean) on smallish
scales but is curved globally. This is not so hard to believe. A bug living in
a eld on the surface of the Earth might reasonably conclude he is living on
an in nite plane. The bug cannot sense the fact that his at, visible world
is just a small patch of a curved surface (Earth) living in three-dimensional
space. Likewise, our apparently Euclidean three-dimensional universe might
3 Relativity: The Special and General Theory , Crown Publications Inc (1961), p. 148.
1.3. Geometry on Surfaces: A First Look 7
be curving in some unseen fourth dimension so that the global geometry of the
universe might be non-Euclidean.
Under reasonable assumptions about space, hyperbolic, elliptic, and
Euclidean geometry are the only three possibilities for the global geometry of
our universe. Researchers have spent signi cant time poring over cosmological
data in hopes of deciding which geometry is ours. Deducing the geometry of
the universe can tell us much about the shape of the universe and perhaps
whether it is nite. If the universe is elliptic, then it must be nite in volume.
If it is Euclidean or hyperbolic, then it can be either nite or in nite. Moreover,
each geometry type corresponds to a class of possible shapes. And, if that isn't
exciting enough, the overall geometry of the universe may be fundamentally
connected to the fate of the universe. Clearly there is no more grand application
of geometry than to the fate of the universe!
Exercises
1. Use Euclid's parallel postulate to prove the alternate interior angles
theorem. That is, in Figure 1.2.3(a), assume the line BD is parallel to the line
AC . Prove that ∠BAC = ∠ABD.
2. Use Euclid's parallel postulate and the previous problem to prove that
◦
the sum of the angles of any triangle is 180 . You may nd Figure 1.2.3(b)
helpful, where segment CD is parallel to segment AB .
D B B D
A C A C
(a) (b)
Figure 1.2.3: Two consequences of the parallel postulate.
1.3 Geometry on Surfaces: A First Look
Think for a minute about the space we live in. Think about objects that live
in our space. Do the features of objects change when they move around in our
space? If I pick up this paper and move it across the room, will it shrink? Will
it become a broom?
If you draw a triangle on this page, the angles of the triangle will add
◦
to 180 . In fact, any triangle drawn anywhere on the page has this property.
Euclidean geometry on this at page (a portion of the plane) is homogeneous :
the local geometry of the plane is the same at all points. Our three-dimensional
space appears to be homogeneous as well. This is nice, for it means that if we
3 3
buy a 5 ft freezer at the appliance store, it doesn't shrink to 0.5 ft when we
get it home. A sphere is another example of a homogenous surface. A two-
dimensional bug living on the surface of a sphere could not tell the dierence
(geometrically) between any two points on the sphere.
The surface of a donut in three-dimensional space (see Figure 1.3.1) is not
homogeneous, and a two-dimensional bug living on this surface could tell the
dierence between various points. One approach to discovering dierences in
geometry involves triangles.
8 1. An Invitation to Geometry
Figure 1.3.1: This torus surface is not homogeneous.
It is an important matter to decide what we mean, exactly, by a triangle on
a surface. A triangle consists of three points and three edges connecting these
points. An edge connecting point A to point B is drawn to represent the path
of shortest distance between A and B. Such a path is called a geodesic . For
the two-dimensional bug, a straight line from A to B is simply the shortest
path from A to B.
On a sphere, geodesics follow great circles. A great circle is a circle
drawn on the surface of the sphere whose center (in three-dimensional space)
corresponds to the center of the sphere. Put another way, a great circle is
a circle of maximum diameter drawn on the sphere. The circles a and b in
Figure 1.3.2 are great circles, but circle c is not.
Figure 1.3.2: Geodesics on the sphere are great circles.
In the Euclidean plane, geodesics are Euclidean lines. One way to determine
a geodesic on a surface physically is to pin some string at A and draw the string
tight on the surface to a point B. The taut string will follow the geodesic from
A to B. In Figure 1.3.3 we have drawn geodesic triangles on three dierent
surfaces.
Figure 1.3.3: Depending on the shape of the surface, geodesic triangles can
◦
have angle sum greater than, less than, or equal to 180 .
Getting back to the donut, a two-dimensional bug could use triangles to
tell the dierence between a convex point on an outer wall, and a saddle-
shaped point on an inner wall (see Figure 1.3.1). A bug could draw a triangle
about the convex point, determine the angle sum, and then move around the
surface to a saddle-shaped point, and determine the angle sum of a new triangle
(whose legs are the same length as before). The bug would scratch her head at
the dierent angle sums before realizing she'd stumbled upon something big.
1.3. Geometry on Surfaces: A First Look 9
She'd go home, write up the result, emphasizing the fact that a triangle in the
◦
rst convex region will have angle sum greater than 180 , while a triangle
◦
in the saddle-shaped region will have angle sum less than 180 . This happy
bug will conclude her donut surface is not homogeneous. She will then sit back
and watch the accolades pour in. Perhaps even a Nobel prize. Thus, small
triangles and their angles can help a two-dimensional bug distinguish points
on a surface.
The donut surface is not homogeneous, so let's build one that is.
Example 1.3.4: The Flat Torus.
Consider again the world of Figure 1.1.2. This world is called a at
torus. At every spot in this world, the pilot of the ship would report at
◦
surroundings (triangle angles add to 180 ). Unlike the donut surface
living in three dimensions, the at torus is homogeneous. Locally,
geometry is the same at every point, and thanks to a triangle check,
this geometry is Euclidean. But the world as a whole is much dierent
than the Euclidean plane. For instance, if the pilot of the ship has a
powerful enough telescope, he'd be able to see the back of his ship. Of
course, if the ship had windows just so, he'd be able to see the back
of his head. The at torus is a nite, Euclidean two-dimensional world
without any boundary.
Example 1.3.6: Coneland.
Here we build cones from at wedges, and measure angles of some
triangles.
a. Begin with a circular disk with a wedge removed, like a pizza
missing a slice or two. Joining the two radial edges produces a cone.
Try it with a cone of your own to make sure it works. Now, with
the cone at again, pick three points, labeled A, B , and C, such that
C is on the radial edge. This means that in this attened version of
the cone, point C actually appears twice: once on each radial edge, as
in Figure 1.3.6. These two representatives for C should get identi ed
when you join the radial edges.
B
A θ
C C
Figure 1.3.6: A triangle on a cone.
b. Draw the segments connecting the three points. You should
get a triangle with the tip of the cone in its interior. (This triangle
should actually look like a triangle if you re-form the cone.) If you
10 1. An Invitation to Geometry
don't get the tip of the cone on the inside of the triangle, adjust the
points accordingly.
c. With your protractor, carefully measure the angle θ subtended
by the circular sector. To emphasize θ's role in the shape of the cone,
we let S(θ) denote the cone surface determined by θ.
d. With your protractor, carefully measure the three angles of your
triangle. The angle at point C is the sum of the angles formed by the
triangle legs and the radial segments. Let ∆ denote the sum of these
three angles.
e. State a conjecture about the relationship between the angle θ
and ∆, the sum of the angles of the triangle. Your conjecture can be
in the form of an equation. Then prove your conjecture. Hint: if you
draw a segment connecting the 2 copies of point C, what is the angle
sum of the quadrilateral ABCC ?
Example 1.3.7: Saddleland.
Repeat the previous exercise but with circle wedges having θ > 2π .
Identifying the radial edges in this case produces a saddle-shaped
surface. [To create such a circle wedge we can tape together two wedges
of equal radius. One idea: Start with a disk with one radial cut, and a
wedge of equal radius. Tape one radial edge of the wedge to one of the
slit radial edges of the disk. Then, identifying the other radial edges
should produce Saddleland.]
Remember, a homogeneous surface is a space that has the same local
geometry at every point. Our at torus is homogeneous, having Euclidean
geometry at every point. However, our cones S(θ) in the previous exercises
are not homogeneous (unless θ happens to be 2π ). If a triangle in S(θ) does
not contain the tip of the cone in its interior, then the angles of the triangle
will add to π radians, but if the triangle does contain the tip of the cone in
its interior, then the angle sum will not be π radians. A two-dimensional bug,
then, could conclude that S(θ) is not homogeneous.
b H c
c b
Figure 1.3.8: A hexagonal video screen
Example 1.3.9: A non-Euclidean surface.
Consider the surface obtained by identifying the edges of the hexagon
as indicated in Figure 1.3.8. In particular, the edges are matched
according to their labels and arrow orientation. So, if a ship ies o the
1.3. Geometry on Surfaces: A First Look 11
hexagonal screen at a spot on the edge marked a, say, then it reappears
at the matching spot on the other edge marked a.
Suppose the pilot of a ship wants to y around one of the corners of
the hexagon. If she begins at point H, say, and ies counterclockwise
around the upper right corner as indicated in the diagram, she would
y o the screen at the top near the start of an a edge. So, as she
made her journey, she would reappear in the lower left corner near the
start of the other a edge. Continuing around she would complete her
journey after circling this second corner.
◦
However, the angle of each corner is 120 , and gluing them together
will create a cone point, as pictured below. Similarly, she would nd
that the other corners of the hexagon meet in groups of two, creating
two additional cone points. As with the Coneland Example 1.3.5, the
pilot can distinguish a corner point from an interior point here. She
can look at triangles: a triangle containing one of the cone points will
◦
have angle sum greater than 180 ; any other triangle will have angle
◦
sum equal to 180 .
a c
c
So the surface is not homogeneous, if it is drawn in the plane.
However, the surface does admit a homogeneous geometry. We can
get rid of the cone points if we can increase each corner angle of the
◦
hexagon to 180 . Then, two corners would come together to form a
◦
perfect 360 patch about the point.
But how can we increase the corner angles? Put the hexagon on the
sphere! Imagine stretching the hexagon onto the northern hemisphere
of a sphere (see Figure 1.3.10). In this case we can think of the 6
points of our hexagon as lying on the equator. Then each corner
angle is ◦
180 , each edge is still a line (geodesic), and when we glue
◦
the edges, each pair of corner angles adds up to exactly 360 , so the
surface is homogeneous. The homogeneous geometry of this surface is
the geometry of the sphere (elliptic geometry), not the geometry of the
plane (Euclidean geometry).
a c b
b c a
Figure 1.3.10: A surface with homogeneous elliptic geometry.
It turns out every surface can be given one of three types of homogeneous
geometry: Euclidean, hyperbolic, or elliptic. We will return to the geometry of
surfaces (and of our universe) after we develop hyperbolic and elliptic geometry.
12 1. An Invitation to Geometry
If it doesn't make a whole lot of sense right now, don't sweat it, but please use
these facts as motivation for learning about these non-Euclidean geometries.
Exercises
1. Work through the Coneland Example 1.3.5.
2. Work through the Saddleland Example 1.3.7.
3. Circumference vs Radius in Coneland and Saddleland. In addition
to triangles, a two-dimensional bug can use circles to screen for dierent
geometries. In particular, a bug can study the relationship between the radius
and the circumference of a circle. To make sure we think like the bug, here's
how we de ne a circle on a surface: Given a point P on the surface, and a real
number r > 0, the circle centered at P with radius r is the set of all points r
units away from P, where the distance between two points is the length of the
shortest path connecting them (the geodesic).
a. Pick your favorite circle in the plane. What is the relationship between
the circle's radius and circumference? Is your answer true for any circle in the
plane?
b. Consider the Coneland surface of Example 1.3.5. Construct a
circle centered at the tip of the cone and derive a relationship between its
circumference and its radius. Is C = 2πr here? If not, which is true: C > 2πr
or C < 2πr?
c. Consider the Saddleland surface of Example 1.3.7. Construct a
circle centered at the tip of the saddle and derive a relationship between its
circumference and its radius. Is C = 2πr here? If not, which is true: C > 2πr
or C < 2πr?
2
The Complex Plane
To study geometry using Klein's Erlangen Program, we need to de ne a space
and a group of transformations of the space. Our space will be the complex
plane.
2.1 Basic Notions
The set of complex numbers is obtained algebraically by adjoining the number
i to the set R of real numbers, where i is de ned by the property that i2 = −1.
We will take a geometric approach and de ne a complex number to be an
ordered pair (x, y) of real numbers. We let C denote the set of all complex
numbers,
C = {(x, y) | x, y ∈ R}.
Given the complex number z = (x, y), x is called the real part of z , denoted
Re(z); and y is called the imaginary part of z , denoted Im(z). The set of
real numbers is a subset of C under the identi cation x ↔ (x, 0), for any real
number x.
Addition in C is componentwise,
(x, y) + (s, t) = (x + s, y + t),
and if k is a real number, we de ne scalar multiplication by
k · (x, y) = (kx, ky).
Within this framework, i = (0, 1), meaning that any complex number (x, y)
can be expressed as x + yi as suggested here:
(x, y) = (x, 0) + (0, y)
= x(1, 0) + y(0, 1)
= x + yi.
The expression x+yi is called the Cartesian form of the complex number.
This form can be helpful when doing arithmetic of complex numbers, but it
can also be a bit gangly. We often let a single letter such as z or w represent a
complex number. So, z = x + yi means that the complex number we're calling
z corresponds to the point (x, y) in the plane.
It is sometimes helpful to view a complex number as a vector, and complex
addition corresponds to vector addition in the plane. The same holds for scalar
multiplication. For instance, in Figure 2.1.1 we have represented z = 2 + i,
w = −1 + 1.5i, as well as z + w = 1 + 2.5i, as vectors from the origin to these
13
14 2. The Complex Plane
points in C. The complex number z −w can be represented by the vector from
w to z in the plane.
z+w
w w z−w
i z z
Figure 2.1.1: Complex numbers as vectors in the plane.
We de ne complex multiplication using the fact that i2 = −1.
(x + yi) · (s + ti) = xs + ysi + xti + yti2
= (xs − yt) + (ys + xt)i.
The modulus of z = x + yi, denoted |z|, is given by
p
|z| = x2 + y 2 .
Note that |z| gives the Euclidean distance of z to the point (0,0).
The conjugate of z = x + yi, denoted z, is
z = x − yi.
In the exercises the reader is asked to prove various useful properties of the
modulus and conjugate.
Example 2.1.2: Arithmetic of complex numbers.
Suppose z = 3 − 4i and w = 2 + 7i.
Then z + w = 5 + 3i, and
z · w = (3 − 4i)(2 + 7i)
= 6 + 28 − 8i + 21i
= 34 + 13i.
A few other computations:
4z = 12 − 16i
p
|z| = 32 + (−4)2 = 5
zw = 34 − 13i.
Exercises
1. z + w, sz , |z|, and z · w.
In each case, determine
a. z = 5 + 2i, s = −4, w = −1 + 2i.
b. z = 3i, s = 1/2, w = −3 + 2i.
c. z = 1 + i, s = 0.6, w = 1 − i.
2. 2
Show that z · z = |z| , where z is the conjugate of z .
2.2. Polar Form of a Complex Number 15
3. Suppose z = x + yi and w = s + ti are two complex numbers. Prove the
following properties of the conjugate and the modulus.
a. |w · z| = |w| · |z|.
b. zw = z · w.
c. z + w = z + w .
d. z + z = 2Re(z). (Hence, z + z is a real number.)
e. z − z = 2Im(z)i.
f. |z| = |z|.
4. A Pythagorean triple consists of three integers (a, b, c) such that
a2 + b2 = c2 . We can use complex numbers to generate Pythagorean triples.
Suppose z = x + yi where x and y are positive integers. Let
a = Re(z 2 ) b = Im(z 2 ) c = zz.
a. Prove that a2 + b2 = c2 .
b. Find the complex number z = x + yi that generates the famous triple
(3,4,5).
c. Find the complex number that generates the triple (5,12,13).
d. Find ve other Pythagorean triples, generated using complex numbers
of the form z = x + yi, where x and y are positive integers with no common
divisors.
2.2 Polar Form of a Complex Number
A point (x, y) in the plane can be represented in polar form (r, θ) according to
the relationships in Figure 2.2.1.
(x, y)
r x = r cos(θ)
θ y = r sin(θ)
Figure 2.2.1: Polar coordinates of a point in the plane
Using these relationships, we can rewrite
x + yi = r cos(θ) + r sin(θ)i
= r(cos(θ) + i sin(θ)).
This leads us to make the following de nition. For any real number θ, we
de ne
eiθ = cos(θ) + i sin(θ).
For instance, eiπ/2 = cos(π/2) + i sin(π/2) = 0 + i · 1 = i.
i0
Similarly, e = cos(0) + i sin(0) = 1, and it's a quick check to see that
eiπ = −1, which leads to a simple equation involving the most famous numbers
in mathematics (except 8), truly an all-star equation:
eiπ + 1 = 0.
z = x + yi and (x, y) has polar form (r, θ) then z = reiθ is called
If the
polar form of z. The non-negative scalar |r| is the modulus of z, and the
angle θ is called the argument of z , denoted arg(z ).
16 2. The Complex Plane
Example 2.2.2: Exploring the polar form.
On the left side of the following diagram, we plot the points z =
2eiπ/4 , w = 3eiπ/2 , v = −2eiπ/6 , u = 3e−iπ/3 .
z −3 + 4i
4 r
v
α θ
3
u
To convert z = −3 + 4i to polar form, refer to the right side of
√
the diagram. We note that r = 9 + 16 = 5, and tan(α) = 4/3, so
θ = π − tan−1 (4/3) ≈ 2.21 radians. Thus,
−1
−3 + 4i = 5ei(π−tan (4/3))
≈ 5e2.21i .
Theorem 2.2.3. The product of two complex numbers in polar form is given
by
reiθ · seiβ = (rs)ei(θ+β) .
Proof. We use the de nition of the complex exponential and some
trigonometric identities.
reiθ · seiβ = r(cos θ + i sin θ) · s(cos β + i sin β)
= (rs)(cos θ + i sin θ) · (cos β + i sin β)
= rs[cos θ cos β − sin θ sin β + (cos θ sin β + sin θ cos β)i]
= rs[cos(θ + β) + sin(θ + β)i]
= rs[ei(θ+β) ].
Thus, the product of two complex numbers is obtained by multiplying their
magnitudes and adding their arguments, and
arg(zw) = arg(z) + arg(w),
where the equation is taken modulo 2π . That is, depending on our choices
for the arguments, we have arg(vw) = arg(v) + arg(w) + 2πk for some integer
k.
Example 2.2.4: Polar form with r ≥ 0.
When representing a complex number z in polar form as z = reiθ , we
may assume that r is non-negative. If r < 0, then
reiθ = −|r|eiθ
= (eiπ ) · |r|eiθ since − 1 = eiπ
= |r|ei(θ+π) , by Theorem 2.2.3.
2.3. Division and Angle Measure 17
Thus, by adding π to the angle if necessary, we may always assume
that z = reiθ where r is non-negative.
Exercises
1. Convert the following points to polar form and plot them: 3 + i, −1 − 2i,
3 − 4i, 7, 002, 001, and −4i.
2. Express the following points in Cartesian form and plot them: z = 2eiπ/3 ,
w = −2eiπ/4 , u = 4ei5π/3 , and z · u.
3. Modify the all-star equation to involve 8. In particular, write an
expression involving e, i, π, 1, and 8, that equals 0. You may use no other
numbers, and certainly not 3.
4. If z = reiθ , prove that z = re−iθ .
2.3 Division and Angle Measure
The division of the complex number z by w 6= 0, denoted z
w , is the complex
number u z = w · u.
that satis es the equation
1
For instance,
i = −i because 1 = i · (−i).
In practice, division of complex numbers is not a guessing game, but can
be done by multiplying the top and bottom of the quotient by the conjugate
of the bottom expression.
Example 2.3.1: Division in Cartesian form.
We convert the following quotient to Cartesian form:
2+i 2 + i 3 − 2i
= ·
3 + 2i 3 + 2i 3 − 2i
(6 + 2) + (−4 + 3)i
=
9+4
8−i
=
13
8 1
= − i.
13 13
Example 2.3.2: Division in polar form.
Suppose we wish to nd z/w where z = reiθ and w = seiβ 6= 0. The
reader can check that
1 1
= e−iβ .
w s
Then we may apply Theorem 2.2.3 to obtain the following result:
z 1
=z·
w w
1
= reiθ · e−iβ
s
r i(θ−β)
= e .
s
18 2. The Complex Plane
So,
z
arg = arg(z) − arg(w)
w
where equality is taken modulo 2π .
Thus, when dividing by complex numbers, we can rst convert to
polar form if it is convenient. For instance,
√ iπ/4
1+i 2e 1 1
=√ = e−iπ/2 = − i.
−3 + 3i 18e i3π/4 3 3
Angle Measure Given two rays L1 and L2 having common initial point, we
let∠(L1 , L2 ) denote the angle between rays L1 and L2 , measured from L1
to L2 . We may rotate ray L1 onto ray L2 in either a counterclockwise direction
or a clockwise direction. We adopt the convention that angles measured
counterclockwise are positive, and angles measured clockwise are negative, and
admit that angles are only well-de ned up to multiples of 2π . Notice that
∠(L1 , L2 ) = −∠(L2 , L1 ).
To compute ∠(L1 , L2 ) where z0 is the common initial point of the rays, let
z1 be any point on L1 , and z2 any point on L2 . Then
z2 − z0
∠(L1 , L2 ) = arg
z1 − z0
= arg(z2 − z0 ) − arg(z1 − z0 ).
Example 2.3.3: The angle between two rays.
Suppose L1 and L2 are rays emanating from 2 + 2i. Ray L1 proceeds
along the line y = x and L2 proceeds along y = 3 − x/2 as pictured.
L1
3 + 3i
2 + 2i θ
4+i L2
To compute the angle θ in the diagram, we choose z1 = 3 + 3i and
z2 = 4 + i. Then
∠(L1 , L2 ) = arg(2 − i) − arg(1 + i) = − tan−1 (1/2) − π/4 ≈ −71.6◦ .
◦
That is, the angle from L1 to L2 is 71.6 in the clockwise direction.
The angle determined by three points.
If u, v, and w are three complex numbers, let ∠uvw denote the angle θ
from ray
−
→
vu to
−→.
vw In particular,
w−v
∠uvw = θ = arg .
u−v
2.4. Complex Expressions 19
θ
u
v
For instance, if u = 1 on the positive real axis, v = 0 is the origin in C, and
z is any point in C, then ∠uvz = arg(z).
Exercises
1. Express
1
x+yi in the form a + bi.
2. Express these fractions in Cartesian form or polar form, whichever seems
more convenient.
1 1 4+i 2
, , , .
2i 1 + i 1 − 2i 3 + i
3. Prove that|z/w| = |z|/|w|, and that z/w = z/w.
4. Suppose z = reiθ and w = seiα are as shown below. Let u = z · w. Prove
that ∆01z and ∆0wu are similar triangles.
u
w
0 1
5. Determine the angle ∠uvw where u = 2 + i, v = 1 + 2i, and w = −1 + i.
6. Suppose z is a point with positive imaginary component on the unit circle
shown below, a = 1 and b = −1. Use the angle formula to prove that angle
∠bza = π/2.
b 0 a
2.4 Complex Expressions
In this section we look at some equations and inequalities that will come up
throughout the text.
20 2. The Complex Plane
Example 2.4.1: Line equations.
The standard form for the equation of a line in the xy -plane is
ax + by + d = 0. This line may be expressed via the complex variable
z = x + yi. For an arbitrary complex number β = s + ti, note that
βz + βz = (sx − ty) + (sy + tx)i + (sx − ty) − (sy + tx)i
= 2sx − 2ty.
It follows that the line ax + by + d = 0 can be represented by the
equation
αz + αz + d = 0 (equation of a line)
where α = 12 (a − bi) is a complex constant and d is a real number.
Conversely, for any complex number α and real number d, the
equation
αz + αz + d = 0
determines a line in C.
We may also view any line in C as the collection of points equidistant from
two given points.
Theorem 2.4.2. Any line in C can be expressed by the equation |z−γ| = |z−β|
for suitably chosen points γ and β in C, and the set of all points (Euclidean)
equidistant from distinct points γ and β forms a line.
Proof. Given two points γ and β in C, z is equidistant from both if and
only if |z − γ|2 = |z − β|2 . Expanding this equation, we obtain
(z − γ)(z − γ) = (z − β)(z − β)
|z| − γz − γz + |γ|2 = |z|2 − βz − βz + |β|2
2
(β − γ)z + (β − γ)z + (|γ|2 − |β|2 ) = 0.
This last equation has the form of a line, letting α = (β − γ) and
d = |γ|2 − |β|2 .
Conversely, starting with a line we can nd complex numbers γ and β that
do the trick. In particular, if the given line is the perpendicular bisector of the
segment γβ , then |z − γ| = |z − β| describes the line. We leave the details to
the reader.
Example 2.4.3: Quadratic equations.
Suppose z0 is a complex constant and consider the equation z 2 = z0 .
A complex number z that satis es this equation will be called a square
√
root of z0 , and will be written as z0 .
iθ0
If we view z0 = r0 e in polar form with r0 ≥ 0, then a complex
number z = reiθ satis es the equation z 2 = z0 if and only if
reiθ · reiθ = r0 eiθ0 .
2.4. Complex Expressions 21
In other words, z satis es the equation if and only if r2 = r0 and
2θ = θ0 (modulo 2π ).
As long as r0 is greater than zero, we have two solutions to the
equation, so that z0 has two square roots:
√
± r0 eiθ0 /2 .
√
For instance, z2 = i has two solutions. Since i = 1eiπ/2 , i =
√ √ √
±e iπ/4
. In Cartesian form, i = ±( 22 + 22 i).
More generally, the complex quadratic equation αz 2 + βz + γ = 0
where α, β, γ are complex constants, will have one or two solutions.
This marks an important dierence from the real case, where a
quadratic equation might not have any real solutions. In both cases we
may use the quadratic formula to hunt for roots, and in the complex
case we have solutions
p
−β ± β 2 − 4αγ
z= .
2α
For instance, z 2 + 2z + 4 = 0 has two solutions:
√
−2 ± −12 √
z= = −1 ± 3i
2
√
since −1 = i.
Example 2.4.4: Solving a quadratic equation.
Consider the equation z 2 − (3 + 3i)z = 2 − 3i. To solve this equation
for z we rst rewrite it as
z 2 − (3 + 3i)z − (2 − 3i) = 0.
We use the quadratic formula with α = 1, β = −(3 + 3i), and
γ = −(2 − 3i), to obtain the solution(s)
p
(3 + 3i)2 + 4(2 − 3i)
3 + 3i ±
z=
√ 2
3 + 3i ± 8 + 6i
z= .
2
To determine the solutions in Cartesian form, we need to evaluate
√
8 + 6i. We oer two approaches.
√ The rst approach considers the
following task: Set x + yi = 8 + 6i and solve for x and y directly by
squaring both sides to obtain a system of equations.
√
x + yi = 8 + 6i
2
(x + yi) = 8 + 6i
x − y 2 + 2xyi = 8 + 6i.
2
Thus, we have two equations and two unknowns:
x2 − y 2 = 8 (1)
2xy = 6. (2)
22 2. The Complex Plane
In fact, we also know that x2 + y 2 = |x + yi|2 = |(x + yi)2 | =
|8 + 6i| = 10, giving us a third equation
x2 + y 2 = 10. (3)
Adding equations (1) and (3) yields x2 = 9 so x = ±3. Substituting
x = 3 into equation (2) yields y = 1; substituting x = −3 into (2) yields
y = −1. Thus we have two solutions:
√
8 + 6i = ±(3 + i).
√
We may also use the polar form to determine 8 + 6i. Consider
iθ
the right triangle determined by the point 8 + 6i = 10e pictured in
the following diagram.
8 + 6i
10
√ √
We know 8 + 6i = ± 10eiθ/2 , so we want to nd θ/2. Well, we
can determine tan(θ/2) easily enough using the half-angle formula
sin(θ)
tan(θ/2) = .
1 + cos(θ)
The right triangle in the diagram shows us that sin(θ) = 3/5 and
cos(θ) = 4/5, so tan(θ/2) = 1/3. This means that any point reiθ/2 lives
on the line through the origin having slope 1/3, and can be described
√
by k(3
√ + i) for some scalar k . Since 8 + 6i √ has this form,
√ it follows
that
√ + i) for some k . Since | 8 + 6i|
8 + 6i = k(3 √ = 10 , it follows
that |k(3 + i)| = 10, so k = ±1. In other words, 8 + 6i = ±(3 + i).
Now let's return to the solution of the original quadratic equation
in this example:
√
3 + 3i ± 8 + 6i
z=
2
3 + 3i ± (3 + i)
z= .
2
Thus, z = 3 + 2i or z = i.
Example 2.4.5: Circle equations.
If we let z = x + yi and z0 = h + ki, then the complex equation
|z − z0 | = r (equation of a circle)
describes the circle in the plane centered at z0 with radius r > 0.
To see this, note that
|z − z0 | = |(x − h) + (y − k)i|
p
= (x − h)2 + (y − k)2 .
2.4. Complex Expressions 23
So |z − z0 | = r is equivalent to the equation (x − h)2 + (y − k)2 = r2
of the circle centered atz0 with radius r.
|z − 3 − 2i| = 3 describes the
For instance, set of all points that are
3 units away from 3 + 2i. All such z form a circle of radius 3 in the
plane, centered at the point (3, 2).
Example 2.4.6: Complex expressions as regions.
Describe each complex expression below as a region in the plane.
1. |1/z| > 2.
Taking the reciprocal of both sides, we have |z| < 1/2, which is
the interior of the circle centered at 0 with radius 1/2.
2. Im(z) < Re(z).
Set z = x + yi in which case the inequality becomes y < x. This
inequality describes all points in the plane under the line y = x,
as pictured below.
3. Im(z) = |z|.
p
Setting z = x + yi, this equation is equivalent to y = y 2 + x2 .
Squaring both sides
p we obtain 0 = x2 , so that x = 0. It
follows that y = y 2 = |y| so the equation describes the points
(0, y) with y ≥ 0. These points determine a ray on the positive
imaginary axis.
Im(z) = |z|
1
2
Im(z) <Re(z)
Moving forward, lines and circles will be especially important objects for
us, so we end the section with a summary of their descriptions in the complex
plane.
Lines and circles in C.
Lines and circles in the plane can be expressed with a complex variable
z = x + yi.
The line ax + by + d = 0 in the plane can be represented by the
equation
αz + αz + d = 0
1
where α= 2 (a − bi) is a complex constant and d is a real number.
The circle in the plane centered at z0 with radius r > 0 can be
represented by the equation
|z − z0 | = r.
24 2. The Complex Plane
Exercises
1. Use a complex variable to describe the equation of the line y = mx + b.
Assume m 6= 0. In particular, show that this line is described by the equation
(m + i)z + (m − i)z + 2b = 0.
2. In each case, sketch the set of complex numbers z satisfying the given
condition.
a. |z + i| = 3.
b. |z + i| = |z − i|.
c. Re(z) = 1.
d. |z/10 + 1 − i| < 5.
e. Im(z) > Re(z).
f. Re(z) = |z − 2|.
3. Suppose u, v, w are three complex numbers not all on the same line. Prove
that any point z in C is uniquely determined by its distances from these three
points. Hint: Suppose β and γ are complex numbers such that |u−β| = |u−γ|,
|v − β| = |v − γ| and |w − β| = |w − γ|. Argue that β and γ must in fact be
equal complex numbers.
4. Find all solutions to the quadratic equation z 2 + iz − (2 + 6i) = 0.
3
Transformations
Transformations will be the focus of this chapter. They are functions rst
and foremost, often used to push objects from one place in a space to a more
convenient place, but transformations do much more. They will be used to
de ne dierent geometries, and we will think of a transformation in terms of
the sorts of objects (and functions) that are unaected by it.
3.1 Basic Transformations of C
We begin with a de nition.
De nition 3.1.1. Given two sets A and B, a function f :A→B is called
one-to-one (or 1-1) if whenever a1 6= a2 in A, then f (a1 ) 6= f (a2 ) in B . The
function f is called onto if for any b in B there exists an element a in A such
that f (a) = b. A transformation on a set A is a function T : A → A that is
one-to-one and onto.
Following are two schematics of functions. In the rst case, f :A→B is
onto, but not one-to-one. In the second case, g:A→B is one-to-one, but not
onto.
f g
A B A B
A transformation A has an inverse function, T −1 , characterized by the
T of
−1
property that the compositions T ◦ T (a) = a and T ◦ T −1 (a) = a for all a in
−1
A. The inverse function T is itself a transformation of A and it undoes T
−1
in this sense: For elements z and w in A, T (w) = z if and only if T (z) = w.
In this section we develop the following basic transformations of the plane,
as well as some of their important features.
Basic Transformations of C.
General linear transformation: T (z) = az + b, where a, b are in C
with a 6= 0.
Special cases of general linear transformations:
◦ Translation by b: Tb (z) = z + b
25
26 3. Transformations
◦ Rotation by θ about 0: Rθ (z) = eiθ z
◦ Rotation by θ about z0 : R(z) = eiθ (z − z0 ) + z0
◦ Dilation by factor k > 0: T (z) = kz
Re ection across a line L: rL (z) = eiθ z + b, where b is in C, and θ
is in R.
Example 3.1.2: Translation.
Consider the xed complex number b, and de ne the function Tb : C →
C by
Tb (z) = z + b.
The notation helps us remember that z is the variable, and b is a
complex constant. We will prove that Tb is a transformation, but this
fact can also be understood by visualizing the function. Each point
in the plane gets moved by the vector b, as suggested in the following
diagram.
Tb (v)
b
Tb (w)
v
For instance, the origin gets moved to the point b, (i.e., Tb (0) = b),
and every other point in the plane gets moved the same amount and in
the same direction. It follows that two dierent points, such as v and
w in the diagram, cannot get moved to the same image point (thus,
the function is one-to-one). Also, any point in the plane is the image
of some other point (just follow the vector −b to nd this pre-image
point), so the function is onto as well.
We now oer a formal argument that the translation Tb is a
transformation. Recall, b is a xed complex number.
That Tb is onto :
To show that Tb w denote an arbitrary element of C.
is onto, let
We must nd a complex number z such that Tb (z) = w. Let z = w − b.
Then Tb (z) = z + b = (w − b) + b = w . Thus, T is onto.
That Tb is one-to-one :
To show that Tb is 1-1 we must show that if z1 6= z2 then
Tb (z1 ) 6= Tb (z2 ). We do so by proving the contrapositive. Recall,
the contrapositive of a statement of the form If P is true then Q is
true is If Q is false then P is false. These statements are logically
equivalent, which means we may prove one by proving the other. So, in
the present case, the contrapositive of If z1 6= z2 then Tb (z1 ) 6= Tb (z2 )
is If Tb (z1 ) = Tb (z2 ), then z1 = z2 . We now prove this statement.
Suppose z1 and z2 are two complex numbers such that Tb (z1 ) =
Tb (z2 ). Then z1 + b = z2 + b. Subtracting b from both sides we see that
z1 = z2 , and this completes the proof.
3.1. Basic Transformations of C 27
Example 3.1.3: Rotation about the origin.
Let θ be an angle, and de ne Rθ : C → C by Rθ (z) = eiθ z.
Rθ (z)
θ z
This transformation causes points in the plane to rotate about the
origin by the angle θ. (If θ > 0 the rotation is counterclockwise, and
ifθ < 0 the rotation is clockwise.) To see this is the case, suppose
z = reiβ , and notice that
Rθ (z) = eiθ reiβ = rei(θ+β) .
Example 3.1.4: Rotation about any point.
To achieve a rotation by angle θ about a general point z0 , send points in
the plane on a three-leg journey: First, translate the plane so that the
center of rotation, z0 , goes to the origin. The translation that does the
trick is T−z0 . Then rotate each point by θ about the origin (Rθ ). Then
translate every point back (Tz0 ). This sequence of transformations has
the desired eect and can be tracked as follows:
T−z R Tz
0
z 7−−−→ θ
z − z0 7−−→ eiθ (z − z0 ) 7−−→
0
eiθ (z − z0 ) + z0 .
In other words, the desired rotation R is the composition Tz0 ◦Rθ ◦T−z0
and
R(z) = eiθ (z − z0 ) + z0 .
That the composition of these three transformations is itself a
transformation follows from the next theorem.
Theorem 3.1.5. If T and S are two transformations of the set A, then the
composition S ◦ T is also a transformation of the set A.
Proof. We must prove that S◦T :A→A is 1-1 and onto.
That S ◦ T is onto :
Suppose c is in A. We must nd an element a in A such that S ◦ T (a) = c.
Since S is onto, there exists some element b in A such that S(b) = c.
Since T is onto, there exists some element a in A such that T (a) = b.
Then S ◦ T (a) = S(b) = c, and we have demonstrated that S ◦ T is onto.
That S ◦ T is 1-1 :
Again, we prove the contrapositive. In particular, we show that if
S ◦ T (a1 ) = S ◦ T (a2 ) then a1 = a2 .
If S(T (a1 )) = S(T (a2 )) then T (a1 ) = T (a2 ) since S is 1-1.
And T (a1 ) = T (a2 ) implies that a1 = a2 since T is 1-1.
Therefore, S ◦ T is 1-1.
28 3. Transformations
Example 3.1.6: Dilation.
Suppose k > 0 is a real number. The transformation T (z) = kz is
called a dilation ; such a map either stretches or shrinks points in the
plane along rays emanating from the origin, depending on the value of
k.
Indeed, if z = x + yi, thenT (z) = kx + kyi, and z and T (z) are
on the same line through the origin. If k > 1 then T stretches points
away from the origin. If 0 < k < 1, then T shrinks points toward the
origin. In either case, such a map is called a dilation.
Given complex constants a, b with a 6= 0 the map T (z) = az + b is called a
general linear tranformation . We show in the following example that such
a map is indeed a transformation of C.
Example 3.1.7: General Linear Transformations.
Consider the general linear transformation T (z) = az + b, where a, b
are in C and a 6= 0. We show T is a transformation of C.
That T is onto :
Let w denote an arbitrary element of C. We must nd a complex
number z such that T (z) = w. To nd this z , we solve w = az + b for
z . So, z = a1 (w − b) should work (since
a 6= 0, z is a complex number).
1 1
Indeed, T ( (w − b)) = a ·
a a (w − b) + b = w. Thus, T is onto.
That T is one-to-one :
To show that T is 1-1 we show that if T (z1 ) = T (z2 ), then z1 = z2 .
If z1 and z2 are two complex numbers such that T (z1 ) = T (z2 ),
then az1 + b = az2 + b. By subtracting b from both sides we see that
az1 = az2 , and then dividing both sides by a (which we can do since
a 6= 0), we see that z1 = z2 . Thus, T is 1-1 as well as onto, and we
have proved T is a transformation.
Note that dilations, rotations, and translations are all special types
of general linear transformations.
We will often need to gure out how a transformation moves a collection
of points such as a triangle or a disk. As such, it is useful to introduce the
following notation, which uses the standard convention in set theory that a∈A
means the element a is a member of the set A.
De nition 3.1.8. Suppose T : A → A is a transformation and D is a subset
of A. The image of D, denoted T (D), consists of all points T (x) such that
x ∈ D. In other words,
T (D) = {a ∈ A | a = T (x) for some x ∈ D}.
For instance, if L is a line and Tb is translation by b, then it is reasonable
to expect that T (L) is also a line. If one translates a line in the plane, it ought
to keep its linear shape. In fact, lines are preserved under any general linear
transformation, as are circles.
Theorem 3.1.9. Suppose T is a general linear transformation.
a. T maps lines to lines.
b. T maps circles to circles.
Proof. a. We prove that if L is a line in C then so is T (L). A line L is
described by the line equation
αz + αz + d = 0
3.1. Basic Transformations of C 29
for some complex constant α and real number d. Suppose T (z) = az + b is
a general linear transformation (so a 6= 0). All the points in T (L) have the
form w = az + b where z satis es the preceeding line equation. It follows that
z = a1 (w − b) and when we plug this into the line equation we see that
w−b w−b
α +α +d=0
a a
which can be rewritten
α α αb αb
w+ w+d− − = 0.
a a a a
Now, for any complex number β the sum β + β is a real number, so in the
αb αb
above expression, d − (
a + a ) is a real number. Therefore, all w in T (L)
satisfy a line equation. That is, T (L) is a line.
b. The proof of this part is left as an exercise.
Example 3.1.10: The image of a disk.
The image of the disk D = {z ∈ C | |z − 2i| ≤ 1} under the
transformation T : C → C given by T (z) = 2z + (4 − i) is the disk
T (D) centered at 4 + 3i with radius 2 as pictured below.
4i
4 + 3i
2i
D T (D)
2 4
We will be interested in working with transformations that preserve angles
between smooth curves. A planar curve is a function r : [a, b] → C mapping
an interval of real numbers into the plane. A curve is smooth if its derivative
exists and is nonzero at every point. Suppose r1 and r2 are two smooth curves
in C that intersect at a point. The angle between the curves measured from
r1 to r2 , which we denote by ∠(r 1 , r 2 ), is de ned to be the angle between the
tangent lines at the point of intersection.
De nition 3.1.11. A transformation T of C preserves angles at point z0
if ∠(r 1 , r 2 ) = ∠(T (r 1 ), T (r 2 )) for all smooth curves r 1 and r 2 that intersect
at z0 . A transformation T of C preserves angles if it preserves angles at all
points in C. A transformation T of C preserves angle magnitudes if, at
any point in C, |∠(r 1 , r 2 )| = |∠(T (r 1 ), T (r 2 ))| for all smooth curves r 1 and
r 2 intersecting at the point.
Theorem 3.1.12. General linear transformations preserve angles.
Proof. Suppose T (z) = az + b where a 6= 0. Since the angle between curves
is de ned to be the angle between their tangent lines, it is su cient to check
that the angle between two lines is preserved. Suppose L1 and L2 intersect at
z0 , and zi is on Li for i = 1, 2, as in the following diagram.
30 3. Transformations
L2
z2
L1
z0 z1
Then,
z2 − z0
∠(L1 , L2 ) = arg .
z1 − z0
Since general linear transformations preserve lines, T (Li ) is the line through
T (z0 ) and T (zi ) for i = 1, 2 and it follows that
T (z2 ) − T (z0 )
∠(T (L1 ), T (L2 )) = arg
T (z1 ) − T (z0 )
az2 + b − az0 − b
= arg
az1 + b − az0 − b
z2 − z0
= arg
z1 − z0
= ∠(L1 , L2 ).
Thus T preserves angles.
De nition 3.1.13. A xed point of a transformation T : A → A is an
element a in the set A such that T (a) = a.
b 6= 0, the translation Tb of C has no xed points. Rotations of C and
If
dilations of C have a single xed point, and the general linear transformation
T (z) = az + b has one xed point as long as a 6= 1. To nd this xed point,
solve
z = az + b
for z. T (z) = 2z + (4 − i) of
For instance, the xed point of the transformation
Example 3.1.10 is found by solving z = 2z +4−i, for z , which yields z = −4+i.
So, while the map T (z) = 2z + (4 − i) moves the disk D in the example to the
disk T (D), the point −4 + i happily stays where it is.
De nition 3.1.14. A Euclidean isometry is a transformation T of C with
the feature that |T (z) − T (w)| = |z − w| for any points z and w in C. That is, a
Euclidean isometry preserves the Euclidean distance between any two points.
Example 3.1.15: Some Euclidean isometries of C.
It is perhaps clear that translations, which move each point in the
plane by the same amount in the same direction, ought to be isometries.
Rotations are also isometries. In fact, the general linear transformation
T (z) = az + b will be a Euclidean isometry so long as |a| = 1:
|T (z) − T (w)| = |az + b − (aw + b)|
= |a(z − w)|
= |a||z − w|.
3.1. Basic Transformations of C 31
So, |T (z) − T (w)| = |z − w| ⇐⇒ |a| = 1. Translations and rotations
about a point in C are general linear transformations of this type, so
they are also Euclidean isometries.
Example 3.1.16: Re ection about a line.
Re ection about a line L is the transformation of C de ned as
follows: Each point on L gets sent to itself, and if z is not on L, it gets
sent to the point z∗ such that line L is the perpendicular bisector of
segment zz ∗ .
z∗
Re ection about L is de ned algebraically as follows. If L happens
to be the real axis then
rL (z) = z.
For any other line L we may arrive at a formula for re ection by
rotating and/or translating the line to the real axis, then taking the
conjugate, and then reversing the rotation and/or translation.
For instance, to describe re ection about the line y = x + 5, we may
translate vertically by −5i, rotate by − π4 , re ect about the real axis,
π
rotate by
4 , and nally translate by 5i to get the composition
z 7→ z − 5i
π
7→ e− 4 i (z − 5i)
π π
7→ e− 4 i (z − 5i) = e 4 i (z + 5i)
π π π
7→ e 4 i · e 4 i (z + 5i) = e 2 i (z + 5i)
π
7→ e 2 i (z + 5i) + 5i.
π
Simplifying (and noting that e 2 i = i), the re ection about the line
L:y =x+5 has formula
rL (z) = iz − 5 + 5i.
In general, re ection across any line L in C will have the form
rL (z) = eiθ z + b
for some angle θ and some complex constant b.
Re ections are more basic transformations than rotations and translations
in that the latter are simply careful compositions of re ections.
Theorem 3.1.17. A translation of C is the composition of re ections about
two parallel lines. A rotation of C about a point z0 is the composition of
re ections about two lines that intersect at z0 .
Proof. Given the translation Tb (z) = z + b let L1 be the line through the
origin that is perpendicular to segment 0b as pictured in Figure 3.1.18(a). Let
32 3. Transformations
L2 be the line parallel to L1 through the midpoint of segment 0b. Also let ri
denote re ection about line Li for i = 1, 2.
Now, given any z in C, let L be the line through z that is parallel to vector b
(and hence perpendicular to L1 and L2 ). The image of z under the composition
r2 ◦ r1 will be on this line. To nd the exact location, let z1 be the intersection
of L1 and L, and z2 the intersection of L2 and L, (see the gure). To re ect
z about L1 we need to translate it along L twice by the vector z1 − z . Thus
r1 (z) = z + 2(z1 − z) = 2z1 − z .
Next, to re ect r1 (z) about L2 , we need to translate it along L twice by
the vector z2 − r1 (z). Thus,
r2 (r1 (z)) = r1 (z) + 2(z2 − r1 (z)) = 2z2 − r1 (z) = 2z2 − 2z1 + z.
Notice from Figure 3.1.18(a) that z2 − z1 is equal to b/2. Thus r2 (r1 (z)) =
z+b is translation by b.
Rotation about the point z0 by angle θ can be achieved by two re ections.
The rst re ection is about the line L1 through z0 parallel to the real axis,
and the second re ection is about the line L2 that intersects L1 at z0 at an
angle of θ/2, as in Figure 3.1.18(b). In the exercises you will prove that this
composition of re ections does indeed give the desired rotation.
L1 L2 z
r2 ◦ r1 (z) L2
θ
r1 (z) r2 ◦ r1 (z) L θ
z1 z2 2 L1
z z0
b
0 b
2 r1 (z)
(a) (b)
Figure 3.1.18: (a) Translations and (b) rotations are compositions of
re ections.
We list some elementary features of re ections in the following theorem.
We do not prove them here but encourage you to work through the details.
We will focus our eorts in the following section on proving analogous features
for inversion transformations, which are re ections about circles.
Theorem 3.1.19. Re ection across a line is a Euclidean isometry. Moreover,
any re ection sends lines to lines, sends circles to circles, and preserves angle
magnitudes.
In fact, one can show that any Euclidean isometry can be expressed as the
composition of at most three re ections. See, for instance, Stillwell [10] for a
proof of this fact.
Theorem 3.1.20. Any Euclidean isometry is the composition of, at most,
three re ections.
Exercises
1. Is T (z) = −z a translation, dilation, rotation, or none of the above?
3.2. Inversion 33
2. Show that the general linear transformation T (z) = az + b, where a and b
are complex constants, is the composition of a rotation, followed by a dilation,
followed by a translation. Hint: View the complex constant a in polar form.
3. Prove that a general linear transformation maps circles to circles.
4. Suppose T is a rotation by 30
◦
about the point 2, and S is a rotation by
◦
45 about the point 4. What is T ◦ S? Can you describe this transformation
geometrically?
5. Suppose T (z) = iz + 3 and S(z) = −iz + 2. Find T ◦ S. What type of
transformation is this?
6. Find a formula for a transformation of C that maps the open disk
D = {z | |z| < 2} to the open disk D0 = {z | |z −i| < 5}. Is this transformation
unique, or can you think of two dierent ones that work?
7. x = k.
Find a formula for re ection about the vertical line
8. Find a formula for re ection about the horizontal line y = k .
9. Find a formula for re ection in the plane about the line y = mx + b,
where m 6= 0. Hint: Think about what angle this line makes with the positive
x-axis.
10. Prove that the construction in Figure 3.1.18(b) determines the desired
rotation.
11. S(z) = kz is a dilation about the origin. Find an equation for a dilation
of C by factor k about an arbitrary point z0 in C.
3.2 Inversion
Inversion oers a way to re ect points across a circle. This transformation plays
a central role in visualizing the transformations of non-Euclidean geometry, and
this section is the foundation of much of what follows.
Suppose C r and center z0 . Inversion in the circle
is a circle with radius
C sends a point z 6= z0 to the point z ∗ de ned as follows: First, construct the
ray from z0 through z . Then, let z ∗ be the unique point on this ray that
satis es the equation
|z − z0 | · |z ∗ − z0 | = r2 .
The point z∗ is called the symmetric point to z with respect to C .
r z∗
z
z0
Inversion in a circle centered at z0 is a transformation on the set C − {z0 }
consisting of all complex numbers except z0 . We usually denote inversion in
the circle C by by iC (z) = z∗. In the next section we will discuss how to extend
this transformation in a way to include the center z0 .
You will work through several features of circle inversions in the exercises,
including how to construct symmetry points with compass and ruler (see
Figure 3.2.18). We note here that iC xes all the points on the circle C,
and points inside the circle get mapped to points outside the circle and vice
34 3. Transformations
versa. The closer z gets to the center of the circle, the further iC (z) gets from
the circle.
Example 3.2.1: Inversion in the unit circle.
The unit circle in C, denoted S1 , is the circle with center z0 = 0 and
radius r = 1. The equation for the point z ∗ symmetric to a point z 6= 0
with repsect to S1
thus reduces from |z − z0 | · |z ∗ − z0 | = r2 to
|z| · |z ∗ | = 1.
Moreover, z∗ is just a scaled version of z since they are on the same
ray through the origin. That is, z ∗ = kz for some positive real number
k . Plug this description of z ∗ into the symmetry point equation to see
|z|·|kz| = 1, which implies k = 1/|z|2 . Thus, z ∗ = (1/|z|2 )z . Moreover,
|z|2 = z · z , so inversion in the unit circle S1 may be written as
iS1 (z) = 1/z.
The following formula for inversion about an arbitrary circle can be
obtained by composition of inversion in the unit circle with some general linear
transformations. The details are left to Exercise 3.2.1.
Inversion in the circle C centered at z0 with radius r.
Inversion in the circle C centered at z0 with radius r is given by
r2
iC (z) = + z0 .
(z − z0 )
Example 3.2.2: Inverting some gures in a circle.
Below we have inverted a circle, the letter `M,' and a small grid across
the circle C centered at z0 . It looks as if the image of the circle is
another circle, which we will soon prove is the case. We will also prove
that lines not intersecting the ccenter of C get inverted into circles. It
follows that the line segments in the `M' get mapped to arcs of circles.
iC (z)
iC (w)
w
z iC (v)
iC (p) v
p
z0
u iC (u)
As Example 3.2.2 suggests, the distinction between lines and circles gets
muddied a bit by inversion. A line can get mapped to a circle and vice versa.
In what follows, it will be helpful to view re ection in a line and inversion in a
circle as special cases of the same general map. To arrive at this view we rst
make lines and circles special cases of the same general type of gure.
3.2. Inversion 35
De nition 3.2.3. A cline is a Euclidean circle or line. Any cline can be
described algebraically by an equation of the form
czz + αz + αz + d = 0
where z = x + yi is a complex variable, α is a complex constant, and c, d are
real numbers. If c = 0 the equation describes a line, and if c 6= 0 and |α|2 > cd
the equation describes a circle.
The word cline (pronounced `Klein') might seem a bit forced, but it
represents the shift in thinking we aim to achieve. We need to start thinking of
lines and circles as dierent manifestations of the same general class of objects.
What class? The class of clines.
Letting α = a + bi and z = x + yi, the cline equation czz + αz + αz + d = 0
can be written as
c(x2 + y 2 ) + [ax − by + (ay + bx)i] + [ax − by − (ay + bx)i] + d = 0
which simpli es to
c(x2 + y 2 ) + 2(ax − by) + d = 0.
If c = 0 then we have the equation of a line, and if c 6= 0 we have the
equation of a circle, so long as a2 + b2 > cd. In this case, the equation can be
put into standard form by completing the square. Let's run through this.
If c 6= 0,
c(x2 + y 2 ) + 2ax − 2by + d = 0
2a 2b d
x2 + x + y 2 − y = −
c c c
2 2 2 2
2 2a a 2 2b b d a b
x + x+ +y − y+ =− + +
c c c c c c c
2 2
a2 + b2 − cd
a b
x+ + y− =
c c c2
and we have the equation of a circle so long as the right-hand side (the radius
term) is positive. In other words, we have the equation of a circle so long as
a2 + b2 > cd. We summarize this information below.
The cline equation.
Given c, d ∈ R, α ∈ C, if c 6= 0, the cline equation
czz + αz + αz + d = 0
gives a circle with center z0 and radius r, where
r
|α|2 − cd
Re(α) Im(α)
z0 = − , and r= ,
c c c2
so long as |α|2 > cd. If c = 0, the cline equation gives a line.
From now on, if you read the phrase inversion in a cline, know that this
means inversion in a circle or re ection about a line, and if someone hands you
a cline C, you might say, Thanks! By the way, is this a line or a circle?
We note here the construction of a cline through three points in C. This
construction is used often in later chapters to generate gures in non-Euclidean
geometry.
36 3. Transformations
Theorem 3.2.4. There exists a unique cline through any three distinct points
in C.
Proof. Suppose u, v , and w are distinct complex numbers. If v is on the line
through u and w then this line is the unique cline through the three points.
Otherwise, the three points do not lie on a single line, and we may build a
circle through these three points as demonstrated in Figure 3.2.5. Construct
the perpendicular bisector to segment uv , and the perpendicular bisector to
segment vw. These bisectors will intersect because the three points are not
collinear. If we call the point of intersection z0 , then the circle centered at z0
through w is the unique cline through the three points.
z0 w
Figure 3.2.5: Constructing the unique circle through three points not on a
single line.
Theorem 3.2.6. Inversion in a circle maps clines to clines. In particular, if
a cline goes through the center of the circle of inversion, its image will be a
line; otherwise the image of a cline will be a circle.
Proof. We prove the result in the case of inversion in the unit circle.
The general proof will then follow, since any inversion is the composition of
this particular inversion together with translations and dilations, which also
preserve clines by Theorem 3.1.9.
Suppose the cline C is described by the cline equation
czz + αz + αz + d = 0,
where c, d ∈ R, α ∈ C.
We want to show that the image of this cline under inversion in the unit
circle, iS1 (C), is also a cline. Well, iS1 (C) consists of all points w = 1/z , where
z satis es the cline equation for C. We show that all such w live on a cline.
If z 6= 0 then we may multiply each side of the cline equation by 1/(z · z)
to obtain
1 1 11
c+α +α +d = 0.
z z zz
But since w = 1/z and w = 1/z , this equation reduces to
c + α · w + α · w + dww = 0,
or
dww + α · w + α · w + c = 0.
Thus, the image points w form a cline equation. If d=0 then the original
cline C passed through the origin, and the image cline is a line. If d 6= 0 then
C did not pass through the origin, and the image cline is a circle. (In fact,
we must also check that |α|2 > dc. This is the case because the original cline
2
equation ensures |α| > cd.)
3.2. Inversion 37
We will call two clines orthogonal if they intersect at right angles. For
instance, a line is orthogonal to a circle if and only if it goes through the
center of the circle. One very important feature of inversion in C is that
clines orthogonal to C get inverted to themselves. To prove this fact, we rst
prove the following result, which can be found in Euclid's Elements (Book III,
Proposition 36).
Lemma 3.2.7. Suppose C is the circle with radius r centered at o, and p is a
point outside C . Let s = |p − o|. If a line through p intersects C at points m
and n, then
|p − m| · |p − n| = s2 − r2 .
Proof. Suppose the line throughp does not pass through the center of C , as
in the diagram below. Let q mn, and let d = |q −o|
be the midpoint of segment
as in the diagram. Note also that the line through q and o is the perpendicular
bisector of segment mn. In particular, |m − q| = |q − n|.
n
q
m r
d
p s o
The Pythagorean theorem applied to ∆pqo gives
|p − q|2 + d2 = s2 , (1)
and the Pythagorean theorem applied to ∆nqo gives
|q − n|2 + d2 = r2 . (2)
By subtracting equation (2) from (1), we have
|p − q|2 − |q − n|2 = s2 − r2 ,
which factors as
(|p − q| − |q − n|)(|p − q| + |q − n|) = s2 − r2 .
Since |p − q| − |q − n| = |p − m| and |p − q| + |q − n| = |p − n|, the result follows.
The case that the line through p goes through the center of C is left as an
exercise.
We note that the quantity s2 − r2 in the previous lemma is often called the
power of the point p with respect to the circle C . That is, if circle C has radius
r and a point p is a distance s from the center of C then the quantity s2 − r2
is called the power of the point p.
Theorem 3.2.8. Suppose C is a circle in C centered at z0 , and z 6= z0 is not
on C . A cline through z is orthogonal to C if and only if it goes through z ∗ ,
the point symmetric to z with respect to C .
38 3. Transformations
Proof. Assume C is the circle of radius r centered at z0 , and D is a cline
through a point z 6= z0 not on C . Let z ∗ denote the point symmetric to z with
respect to C.
First, suppose D is a line through z. A line through z passes through z∗ if
and only if it passes through the center of C, which is true if and only if the
line is orthogonal to C. Thus, the line D through z contains z∗ if and only if
it is orthogonal to C, and the theorem is proved in this case.
Now assume D is a circle through z . Let o and k denote the center and
radius of D, respectively. Set s = |zo −o|, and let t denote a point of intersection
of C and D as pictured below.
z
C z0 o
s
r k
D
t
We must argue that C and D is on D . Now,
are orthogonal if and only if z∗
C andD are orthogonal if and only if ∠otz0 is right, which is the case if and
2 2 2
only if r = s − k by the Pythagorean theorem. Applying Lemma 3.2.7 to
the point z0 (which is outside D ) and the line through z0 and z , we see that
|z0 − z| · |z0 − w| = s2 − k 2 , (1)
where w is the second point of intersection of the line with circle D.
Note also that as symmetric points, z and z∗ satisfy the equation
|z0 − z| · |z0 − z ∗ | = r2 . (2)
Thus, if we assume z∗ is on D, w, in which
then it must be equal to the point
case equations (1) and (2) above tell us s2 − k 2 = r2 . It follows that D is
2 2 2
orthogonal to C . Conversely, if D is orthogonal to C , then s − k = r , so
∗ ∗ −→
|z0 − w| = |z0 − z |. Since z and w are both on the ray z0 z it must be that
z ∗ = w. In other words, z ∗ is on D.
Corollary 3.2.9. Inversion in C takes clines orthogonal to C to themselves.
Theorem 3.2.10. Inversion in a cline preserves angle magnitudes.
Proof. The result was stated for lines in Theorem 3.1.19. Here we assume
C is a circle of inversion. Consider two curves r1 and r2 that intersect at a
point z that is not on C or at the center of C . Recall, ∠(r1 , r2 ) = ∠(L1 , L2 )
where Li is the line tangent to curve ri at z , for i = 1, 2. We may describe
this angle with two circles C1 and C2 tangent to the tangent lines L1 and
L2 , respectively, with the additional feature that the circles meet the circle of
inversion C at right angles, as in Figure 3.2.11. Indeed, C1 is the circle through
z and z ∗ whose center is at the intersection of lines m1 and k , where m1 is the
line through z that is perpendicular to L1 , and k is the perpendicular bisector
∗ ∗
of segment zz . Circle C2 also goes through z and z , and its center is on the
intersection of k and the line m2 through z that is perpendicular to L2 .
3.2. Inversion 39
k
C L1
m1
C2
z
z∗
m2
C1
L2
Figure 3.2.11: Inversion in a circle preserves angle magnitudes.
The advantage to describing ∠(L1 , L2 ) with these circles is that the image
of the angle, ∠(iC (L1 ), iC (L2 )), is also described by these two circles, at
∗
their other intersection point z . Notice that these angles will have opposite
signs. For instance, in Figure 3.2.11, our initial angle is negative, described
by sweeping arc C1 clockwise onto C2 , but in the image, we sweep iC (C1 )
counterclockwise onto iC (C2 ). We leave it as an exercise for the reader to
check that the angle of intersection of C1 and C2 at z∗ is the same magnitude
as the angle between C1 and C2 at z.
Now we show that inversion preserves angle magnitudes for angles that
occur on the circle C (i.e., z is on C ). C0
be a concentric circle to C . Then
Let
iC (z) = S◦iC 0 where S is a dilation of C whose xed point is the common center
0 0
of circles C and C (see Exercise 3.2.12). Since our angle is not on circle C , iC 0
preserves the magnitude of the angle by reason of the preceding argument. The
dilation S preserves angles according to Theorem 3.1.12. Thus iC preserves
angle magnitudes as well. We leave the case of the angle occurring at the
origin to the next section. Bearing that exception in mind, this completes the
proof.
Another important feature of inversion in a cline is that it preserves
symmetry points.
Theorem 3.2.12 Inversion preserves symmetry points. Let iC denote
inversion in a cline C . If p and q are symmetric with respect to a cline D,
then iC (p) and iC (q) are symmetric with respect to the cline iC (D).
Proof. Assume C is the cline of inversion, and assume p and q are symmetric
with respect to a cline D as in Figure 3.2.13 (where C and D are represented
as circles).
40 3. Transformations
F E
C
p
p∗
E∗ q
∗
q
F ∗ D∗
D
Figure 3.2.13: Inversion preserves symmetry points: If p and q are symmetric
with respect to D and we invert about cline C then the image points are
symmetric with respect to the image of D.
We may construct two clines E and F that go through p and q . In the
gure, cline E is a circle and cline F is a line. These clines intersect D at right
angles (Theorem 3.2.8). Since inversion preserves clines and angle magnitudes,
we know that E ∗ = iC (E)
F ∗ = iC (F ) are clines intersecting the cline
and
D = iC (D) at right angles. Both E ∗ and F ∗ contain p∗ = iC (p), so they both
∗
∗ ∗
contain the point symmetric to p with respect to D (Theorem 3.2.8), but
∗ ∗ ∗ ∗
the only other point common to both E and F is q = iC (q). Thus, p and
∗ ∗
q are symmetric with respect to D .
We close the section with two applications of inversion.
Theorem 3.2.14 Apollonian Circles Theorem. Let p, q be distinct points
in C, and k > 0 a positive real number. Let D consist of all points z in C such
that |z − p| = k|z − q|. Then D is a cline.
Proof. k = 1, the set D is a Euclidean line, according to Theorem 2.4.2,
If
so we assume k 6= 1. Let C be the circle centered at p with radius 1. Suppose
z is an arbitrary point in the set D. Inverting about C , let z ∗ = iC (z) and
q ∗ = iC (q) as in the following diagram.
z∗
1
q
p q∗
Observe rst that ∆pz ∗ q ∗ and ∆pqz are similar.
Indeed, |p − z| · |p − z ∗ | = 1 = |p − q| · |p − q ∗ | by the de nition of the
inversion transformation, so we have equal side-length ratios
|p − z ∗ | |p − q ∗ |
= ,
|p − q| |p − z|
3.2. Inversion 41
and the included angles are equal, ∠q ∗ pz ∗ = ∠qpz.
It follows that
|z − q| |z ∗ − q ∗ |
= ∗ ,
|p − q| |z − p|
from which we derive
|z − q|
|z ∗ − q ∗ | = |z ∗ − p| ·
|p − q|
|z − q| 1
= [|z ∗ − p| · |z − p|] · ·
|z − p| |p − q|
1 1
=1· · .
k |p − q|
Thus, the set D of all points z satisfying |z − p| = k|z − q| has image iC (D)
under this inversion consisting of all points z ∗ on a circle centered at q ∗ with
radius (k|p − q|)−1 . Since inversion preserves clines and p is not on iC (D), it
follows that D itself is a circle.
As we let k run through all positive real numbers, we obtain a family
of clines, called the circles of Apollonius of the points p and q. We
note that p and q are symmetric with respect to each cline in this family (see
Exercise 3.3.2).
C1
p z1 q zo L1
C2
t
C
Figure 3.2.15: Finding two points symmetric with respect to a line and circle.
Theorem 3.2.16. Suppose we have two clines that do not intersect, and at
least one of them is a circle. Then there exist two points, p and q , that are
symmetric with respect to both clines.
Proof. L, and the other is a circle C centered
First, assume one cline is a line
at the point z0 L1 be the line through z0 that
as pictured in Figure 3.2.15. Let
is perpendicular to L, and let z1 be the point of intersection of L and L1 . Next,
construct the circle C1 having the diameter z0 z1 . Circle C1 intersects circle C
at some point, which we call t. Notice that ∠z0 tz1 is right, and so the circle
C2 centered at z1 through t is orthogonal to C . Furthermore, the center of C2 ,
z1 , lies on line L, so C2 is orthogonal to L. Let p and q be the two points at
which C2 intersects L1 . By construction, and by using Theorem 3.2.8, p and q
are symmetric to both C and L.
Now, suppose C1 and C2 are circles that do not intersect. We may rst
∗
perform an inversion in a circle C that maps C1 to a line C1 , and C2 to another
∗
circle, C2 , as suggested in Figure 3.2.17 (any circle C centered on a point of C1
works). Then by reason of the preceeding argument, there exist two points p
∗ ∗
and q that are symmetric with respect to C1 and C2 . Since inversion preserves
42 3. Transformations
symmetry points, iC (p) and iC (q) are symmetric with respect to both iC (C1∗ )
∗ ∗ ∗
and iC (C2 ). But iC (C1 ) = C1 and iC (C2 ) = C2 so we're found two points
symmetric to both C1 and C2 . (In fact, we have one exception. If C1 and C2
are concentric circles, this strategy will produce points iC (p) and iC (q), one of
which is the center of C, and we have not yet extended the notion of inversion
to include the center. We do so in the next section in such a way that the
theorem applies to this exceptional case as well.)
C1∗
C1 C2
C2∗
Figure 3.2.17: By inversion we may transform two circles to a circle and a
line.
Exercises
1. Prove the general formula for inversion in a circle C centered at z0 with
radius r. In particular, show in this case that
r2
iC (z) = + z0 .
(z − z0 )
2. Constructing the symmetric point to z when z is inside the circle of
inversion.
Prove that for a point z inside the circle C with center z0 (Figure 3.2.18(a)),
the following construction nds the symmetry point of z. (1) Draw the ray
from z0 through z. (2) Construct the perpendicular to this ray at z. Let t be
a point of intersection of this perpendicular and C. (3) Construct the radius
z0 t. (4) Construct the perpendicular to this radius at t. The symmetric point
z ∗ is the point of intersection of this perpendicular and ray −
z→
oz.
3.2. Inversion 43
(1)
(4)
z∗ (1)
t t z
(3) z z∗
z0
z0 (2)
C C (2)
(a) (b)
Figure 3.2.18: Constructing the symmetric point (a) if z is inside the circle
of inversion; (b) if z is outside the circle of inversion.
3. Constructing the symmetric point to z when z is outside the circle of
inversion.
z outside the circle C with center z0 (Figure 3.2.18(b)),
Prove that for a point
the following construction nds the symmetry point of z . (1) Construct the
circle having diameter z0 z . Let t be a point of intersection of the two circles.
∗
(2) Construct the perpendicular to z0 z through t. Let z be the intersection
of this perpendicular with segment z0 z .
4. Suppose T1 is inversion in the circle |z| = r1 , and T2 is inversion in the
circle |z| = r2 , where r1 , r2 > 0. Prove that T2 ◦ T1 is a dilation. Conversely,
show any dilation is the composition of two inversions.
5. Determine the image of the line y = mx + b (when b 6= 0) under inversion
in the unit circle. In particular, show that the image is a circle with center
p
(−m/2b, 1/2b) and radius (m2 + 1)/4b2 . Hint: Refer to Exercise 2.4.1.
6. Determine the image of the line L given by y = 3x + 4 under inversion in
the unit circle. Give a careful plot of the unit circle, the line L, and the image
of L under the inversion.
7. Prove that inversion in the unit circle maps the circle (x − a)2 + (y − b)2 =
2
r to the circle
2 2 2
a b r
x− + y− =
d d d
where d = a2 + b2 − r2 , provided that d 6= 0.
8. Determine in standard form the image of the circle C given by
(x − 1)2 + y 2 = 4 under inversion in the unit circle. Give a careful plot of
the unit circle, the circle C, and the image of C under the inversion.
9. True or False? If a circle C gets mapped to another circle under inversion
in the unit circle, then the center of C gets mapped to the center of the
image circle, iS1 (C). If the statement is true, prove it; if it is false, provide a
counterexample.
10. Suppose C and D are orthogonal circles. Corollary 3.2.9 tells us that
inversion in C maps D to itself. Prove that this inversion also takes the interior
of D to itself.
11. Finish the proof of Theorem 3.2.10 by showing that the angle of
intersection at z∗ equals the angle of intersection at z in Figure 3.2.11.
12. Suppose C is the circle |z −z0 | = r and C is the circle |z −z0 | = r0 . Find
0
the stretch factor k in the dilation S(z) = k(z − z0 ) + z0 so that iC = S ◦ iC 0 .
44 3. Transformations
13. Complete the proof of Lemma 3.2.7 by proving the case in which the
line through p passes through the center of C.
3.3 The Extended Plane
Consider again inversion about the circle C given by |z − z0 | = r, and observe
that points close to z0 get mapped to points in the plane far away from from
z0 . In fact, a sequence of points in z0 will be inverted to a
C whose limit is
sequence of points whose magnitudes go to ∞. Conversely, any sequence of
points in C having magnitudes marching o to ∞ will be inverted to a sequence
of points whose limit is z0 .
With this in mind, we de ne a new point called the point at in nity ,
denoted ∞. Adjoin this new point to the plane to get the extended plane ,
+
denoted as C . Then, one may extend inversion in the circle C to include the
+
points z0 and ∞. In particular, inversion of C in the circle C centered at z0
+ +
with radius r , iC : C → C , is given by
2
r
(z−z ) + z0 if z 6= z0 , ∞;
0
iC (z) = ∞ if z = z0 ;
z
0 if z = ∞.
Viewing inversion as a transformation of the extended plane, we de ne z0
and ∞ to be symmetric points with respect to the circle of inversion.
The space C+ will be the canvas on which we do all of our geometry, and
it is important to begin to think of ∞ as one of the gang, just another point
to consider. All of our translations, dilations, and rotations can be rede ned
to include the point ∞.
So where is ∞ in C+ ? You approach ∞ as you proceed in either direction
along any line in the complex plane. More generally, if {zn } is a sequence of
complex numbers such that |zn | → ∞ as n → ∞, then we say lim zn = ∞. By
n→∞
convention, we assume ∞ is on every line in the extended plane, and re ection
across any line xes ∞.
Theorem 3.3.1. Any general linear transformation extended to the domain
C+ xes ∞.
Proof. If T (z) = az + b where a and b are complex constants with a 6= 0,
then by limit methods from calculus, as |zn | → ∞, |azn + b| → ∞ as well.
Thus, T (∞) = ∞.
So, with new domain C+ , we modify our xed point count for the basic
transformations:
The translation Tb of C+ xes one point (∞).
The rotation about the origin Rθ of C+ xes 2 points (0 and ∞).
The dilation T (z) = kz of C+ xes 2 points, (0 and ∞).
The re ection rL (z) of C+ about line L xes all points on L (which now
includes ∞).
3.3. The Extended Plane 45
Example 3.3.2: Some transformations not xing ∞.
The following function is a transformation of C+
i+1
T (z) = ,
z + 2i
a fact we prove in the next section. For now, we ask where T sends ∞,
and which point gets sent to ∞.
We tackle the second question rst. The input that gets sent
to ∞ is the complex number that makes the denominator 0. Thus,
T (−2i) = ∞.
To answer the rst question, take your favorite sequence that
marches o to ∞, for example, 1, 2, 3, . . .. The image of this
sequence, T (1), T (2), T (3), . . . consists of complex fractions in which
the numerator is constant, but the denominator grows unbounded in
magnitude along the horizontal line Im(z) = 2. Thus, the quotient
tends to 0, and T (∞) = 0.
As a second example, you can check that if
iz + (3i + 1)
T (z) = ,
2iz + 1
then T (i/2) = ∞ and T (∞) = 1/2.
We emphasize that the following key results of the previous section extend to
C+ as well:
There exists a unique cline through any three distinct points in C+ . (If
one of the given points in Theorem 3.2.4 is ∞, the unique cline is the
line through the other two points.)
Theorem 3.2.8 applies to all points z not on C, including z = z0 or ∞.
Inversion about a cline preserves angle magnitudes at all points in C+
(we discuss this below).
Inversion preserves symmetry points for all points in C+ (Theorem 3.2.12
holds if p or q is ∞).
Theorem 3.2.16 now holds for all clines that do not intersect, including
concentric circles. If the circles are concentric, the points symmetric to
both of them are ∞ and the common center.
Stereographic Projection We close this section with a look at stereo-
graphic projection. By identifying the extended plane with a sphere, this map
oers a very useful way for us to think about the point ∞.
De nition 3.3.3. The unit 2-sphere, denoted S2 , consists of all the points
in 3-space that are one unit from the origin. That is,
S2 = {(a, b, c) ∈ R3 | a2 + b2 + c2 = 1}.
We will usually refer to the unit 2-sphere as simply the sphere. Stereo-
graphic projection of the sphere onto the extended plane is de ned as follows.
Let N = (0, 0, 1) denote the north pole on the sphere. For any point P 6= N
−−→
on the sphere, φ(P ) is the point on the ray N P that lives in the xy -plane. See
Figure 3.3.4 for the image of a typical point P of the sphere.
46 3. Transformations
z
N
P = (a, b, c)
x
φ(P )
y
Figure 3.3.4: Stereographic projection.
The stereographic projection map φ can be described algebraically. The line
−−→
through N = (0, 0, 1) and P = (a, b, c) has directional vector N P = ha, b, c−1i,
so the line equation can be expressed as
~r(t) = h0, 0, 1i + tha, b, c − 1i.
This line intersects the xy -plane when its z coordinate is zero. This occurs
1 a b
when t=1−c , which corresponds to the point (
1−c , 1−c , 0).
Thus, for a point (a, b, c) on the sphere with c 6= 1, stereographic projection
φ: S2 → C+ is given by
a b
φ((a, b, c)) = + i.
1−c 1−c
Where does φ send the north pole? To ∞, of course. A sequence of points
on S2 that approaches N will have image points in C with magnitudes that
approach ∞.
Angles at ∞ If we think of ∞ as just another point in C+ , it makes sense
to ask about angles at this point. For instance, any two lines intersect at ∞,
and it makes sense to ask about the angle of intersection at ∞. We can be
guided in answering this question by stereographic projection, thanks to the
following theorem.
Theorem 3.3.5. Stereographic projection preserves angles. That is, if two
curves on the surface of the sphere intersect at angle θ, then their image curves
in C+ also intersect at angle θ.
Thus, if two curves in C+ intersect at ∞ we may de ne the angle at
which they intersect to equal the angle at which their pre-image curves under
stereographic projection intersect. The angle at which two parallel lines
intersect at ∞ is 0. Furthermore, if two lines intersect at a nite point p
as well as at ∞, the angle at which they intersect at ∞ equals the negative
of the angle at which they intersect at p. As a consequence, we may say that
inversion about a circle preserves angle magnitudes at all points in C+ .
Exercises
1. In each case nd T (∞) and the input z0 such that T (z0 ) = ∞.
a. T (z) = (3 − z)/(2z + i).
b. T (z) = (z + 1)/eiπ/4 .
c. T (z) = (az + b)/(cz + d).
3.4. Möbius Transformations 47
2. Suppose D is a circle of Apollonius of p and q. Prove that p and q
are symmetric with respect to D. Hint: Recall the circle C in the proof of
Theorem 3.2.14. Show that p and q get sent to points that are symmetric with
respect to iC (D).
3. Determine the inverse stereographic projection function φ−1 : C+ → S2 .
In particular, show that for z = x + yi 6= ∞,
x2 + y 2 − 1
−1 2x 2y
φ (x, y) = , , .
x2 + y 2 + 1 x2 + y 2 + 1 x2 + y 2 + 1
3.4 Möbius Transformations
Consider the function de ned on C+ by T (z) = (az+b)/(cz+d) where a, b, c and
d are complex constants. Such a function is called a Möbius transformation
if ad − bc 6= 0. Transformations of this form are also called fractional linear
transformations. The complex number ad − bc is called the determinant of
T (z) = (az + b)/(cz + d), and is denoted as Det(T ).
Theorem 3.4.1. The function
az + b
T (z) =
cz + d
is a transformation of C+ if and only if ad − bc 6= 0.
Proof. First, suppose T (z) = (az+b)/(cz+d) and ad−bc 6= 0. We must show
that T is a transformation. To show T is one-to-one, assume T (z1 ) = T (z2 ).
Then
az1 + b az2 + b
= .
cz1 + d cz2 + d
Cross multiply this expression and simplify to obtain
(ad − bc)z1 = (ad − bc)z2 .
Sincead − bc 6= 0 we may divide this term out of the expression to see z1 = z2 ,
so T T is onto.
is 1-1 and it remains to show that
+ +
Suppose w in C is given. We must nd z ∈ C such that T (z) = w . If
w = ∞, then z = −d/c (which is ∞ if c = 0) does the trick, so assume w 6= ∞.
To nd z such that T (z) = w we solve the equation
az + b
=w
cz + d
for z, which is possible so long as a and c are not both 0 (causing the z terms
to vanish). Since ad − bc 6= 0, we can be assured that this is the case, and
solving for z we obtain
−dw + b
z= .
cw − a
Thus T is onto, and T is a transformation.
To prove the converse we show the contrapositive. We suppose ad − bc = 0
and show T (z) = (az + b)/(cz + d) is not a transformation by tackling two
cases.
Case 1 : ad = 0. In this case, bc = 0 as well, so a or d is zero, and b
or c is zero. In all four scenarios, one can check immediately that T is not a
48 3. Transformations
transformation of C+ .
For instance, if a = c = 0 then T (z) = b/d is neither
+
1-1 nor onto C .
Case 2 : ad 6= 0. In this case, all four constants are non-zero, and a/c = b/d.
Since T (0) = b/d and T (∞) = a/c, T is not 1-1, and hence not a transformation
+
of C .
Note that in the preceeding proof we found the inverse transformation
of a Möbius transformation. This inverse transformation is itself a Möbius
transformation since its determinant is not 0. In fact, its determinant equals
the determinant of the original Möbius transformation. We summarize this
fact as follows.
Theorem 3.4.2. The Möbius transformation
az + b
T (z) =
cz + d
has the inverse transformation
−dz + b
T −1 (z) = .
cz − a
In particular, the inverse of a Möbius transformation is itself a Möbius
transformation.
If we compose two Möbius transformations, the result is another Möbius
transformation. Proof of this fact is left as an exercise.
Theorem 3.4.3. The composition of two Möbius transformations is again a
Möbius transformation.
Just as translations and rotations of the plane can be constructed from
re ections across lines, the general Möbius transformation can be constructed
from inversions about clines.
Theorem 3.4.4. A transformation of C+ is a Möbius transformation if and
only if it is the composition of an even number of inversions.
Proof. We rst observe that any general linear transformation T (z) = az +b
is the composition of an even number of inversions. Indeed, such a map is a
dilation and rotation followed by a translation. Rotations and translations are
each compositions of two re ections (Theorem 3.1.17), and a dilation is the
composition of two inversions about concentric circles (Exericise 3.2.4). So, in
total, we have that T (z) = az + b is the composition of an even number of
inversions.
Now suppose T is the Möbius transformation T (z) = (az + b)/(cz + d). If
c = 0 then T is a general linear transformation of the form T (z) = ad z + db , and
we have nothing to show.
So we assume c 6= 0. By doing some long division, the Möbius
transformation can be rewritten as
az + b a (bc − ad)/c
T (z) = = + ,
cz + d c cz + d
which can be viewed as the composition T3 ◦ T2 ◦ T1 (z), where T1 (z) = cz + d,
bc−ad a
T2 (z) = 1/z and T3 (z) = c z + c . Note that T1 and T3 are general linear
transformations, and
1 1
T2 (z) = =
z z
3.4. Möbius Transformations 49
is inversion in the unit circle followed by re ection about the real axis. Thus,
each Ti is the composition of an even number of inversions, and the general
Möbius transformation T is as well.
To prove the other direction, we show that if T is the composition of two
inversions then it is a Möbius transformation. Then, if T is the composition of
any even number of inversions, it is the composition of half as many Möbius
transformations and is itself a Möbius transformation by Theorem 3.4.3.
Case 1 : T is the composition of two circle inversions. Suppose T = iC1 ◦iC2
where C1 is the circle |z −z1 | = r1 and C2 is the circle |z −z2 | = r2 . For i = 1, 2
the inversion may be described by
ri2
iCi = + zi ,
z − zi
and if we compose these two inversions we do in fact obtain a Möbius
transformation. We leave the details of this computation to the reader but
note that the determinant of the resulting Möbius transformation is r12 r22 .
Case 2 : T is the composition of one circle inversion and one line re ection.
Re ection in the line can be given by rL (z) = eiθ z + b and inversion
in the
2
r
circle C is given by iC = + zo where z0 and r are the center and
radius
z−zo
of the circle, as usual. Work out the composition and you'll see that we have a
iθ 2
Möbius transformation with determinant e r (which is non-zero). Its inverse,
the composition iC ◦ rL , is also a Möbius transformation.
Case 3 : T is the composition of two re ections. Either the two lines of
re ection are parallel, in which case the composition gives a translation, or
they intersect, in which case we have a rotation about the point of intersection
(Theorem 3.1.17). In either case we have a Möbius transformation.
It follows that the composition of any even number of inversions yields a
Möbius transformation.
Since Möbius transformations are composed of inversions, they will embrace
the ner qualities of inversions. For instance, since inversion preserves clines,
so do Möbius transformations, and since inversion preserves angle magnitudes,
even number of inversions).
Möbius transformations preserve angles (as an
Theorem 3.4.5. Möbius transformations take clines to clines and preserves
angles.
The following xed point theorem is useful for understanding Möbius
transformations.
Theorem 3.4.6. Any Möbius transformation T : C+ → C+ xes 1, 2, or all
points of C . +
Proof. To nd xed points of T (z) = (az + b)/(cz + d) we want to solve
az + b
= z,
cz + d
for z, which gives the quadratic equation
cz 2 + (d − a)z − b = 0. (1)
If c 6= 0 then, as discussed in Example 2.4.3, equation (1) must have 1 or 2
solutions, and there are 1 or 2 xed points in this case.
If c = 0 and a 6= d, then the transformation has the form T (z) = (az + b)/d,
which xes ∞. From equation (1), z = b/(d − a) 6= ∞ is a xed point as well.
So we have 2 xed points in this case.
If c=0 and a = d, then equation (1) reduces to 0 = −b, so b = 0 too, and
the transformation is the identity transformation T (z) = (az + 0)/(0z + a) = z .
This transformation xes every point.
50 3. Transformations
With this xed point theorem in hand, we can now prove the Fundamental
Theorem of Möbius Transformations, which says that if we want to induce a
one-to-one and onto motion of the entire extended plane that sends my favorite
three points (z1 , z2 , z3 ) to your favorite three points (w1 , w2 , w3 ), as dramatized
below, then there is a Möbius transformation that will do the trick, and there's
only one.
z2
z1
w3
w1 w2
z3
Figure 3.4.7: We can build a Möbius transformation that sends z1 7→ w1 ,
z2 7→ w2 , and z3 7→ w3 .
Theorem 3.4.8 Fundamental Theorem of Möbius Transformations.
There is a unique Möbius transformation taking any three distinct points of
C+ to any three distinct points of C+ .
Proof. Supposez1 , z2 , and z3 are distinct points in C+ , and w1 , w2 , and w3
+
are distinct points in C . We show there exists a unique Möbius transformation
that maps zi 7→ wi for i = 1, 2, 3. To start, we show there exists a map, built
from inversions, that maps z1 7→ 1, z2 7→ 0 and z3 7→ ∞. We do so in the case
that z3 6= ∞. This special case is left to the exercises.
First, invert about any circle centered at z3 . This takes z3 to ∞ as desired.
0 0
Points z1 and z2 no doubt get moved, say to z1 and z2 , respectively, neither
0
of which is ∞. Second, do a translation that takes z2 to 0. Such a translation
0 00
will keep ∞ xed, and take z1 to some new spot z1 in C. Third, rotate
00
and dilate about the origin, (which keeps 0 and ∞ xed) so that z1 moves
to 1. This process yields a composition of inversions that maps z1 7→ 1,
z2 7→ 0, and z3 7→ ∞. However, this composition actually involves an odd
number of inversions, so it's not a Möbius transformation. To make a Möbius
transformation out of this composition, we do one last inversion: re ect across
the real axis. This keeps 1, 0, and ∞ xed. Thus, there is a Möbius
transformation taking any three distinct points to the points 1, 0, ∞. For
now, we let T denote the Möbius transformation that maps z1 7→ 1, z2 7→ 0
and z3 7→ ∞.
One can similarly construct a Möbius transformation, call it S , that takes
w1 7→ 1, w2 7→ 0, and w3 7→ ∞.
−1
If we let S denote the inverse transformation of S , then the composition
−1
S ◦ T is a Möbius transformation, and this transformation does what we set
out to accomplish as suggested by Figure 3.4.9. In particular,
S −1 ◦ T (z1 ) = S −1 (1) = w1
S −1 ◦ T (z2 ) = S −1 (0) = w2
S −1 ◦ T (z3 ) = S −1 (∞) = w3 .
Finally, to prove that this Möbius transformation is unique, assume that
there are two Möbius transformations U and V that map z1 7→ w1 , z2 7→ w2
3.4. Möbius Transformations 51
1
0
∞
T S
z1 w1
z2 w2
z3 S −1 ◦ T w3
Figure 3.4.9: A schematic for building a Möbius transformation that sends
zi 7→ wi for i = 1, 2, 3. Go through the points 1, 0, ∞.
and z3 7→ w3 . Then V −1 ◦ U is a Möbius transformation that xes z1 , z2 ,
and z3 . According to Theorem 3.4.6 there is only one Möbius transformation
that xes more than two points, and this is the identity transformation. Thus
V −1 ◦ U (z) = z for all z ∈ C+ . Similarly, U ◦ V −1 (z) = z , and it follows that
U (z) = V (z) for all z ∈ C+ . That is, U and V are the same map.
There is an algebraic description of the very useful Möbius transformation
mapping z1 7→ 1, z2 7→ 0 and z3 7→ ∞ that arose in the proof of Theorem 3.4.8:
(z − z2 ) (z1 − z3 )
T (z) = · .
(z − z3 ) (z1 − z2 )
The reader can check that the map works as advertised and that it is indeed
a Möbius transformation. (While it is clear that the transformation has the
form (az + b)/(cz + d), it might not be clear that the determinant is nonzero.
It is, since the zi are distinct.) We also note that if one of the zi is ∞, the
form of the map reduces by cancellation of the terms with ∞ in them. For
instance, if z2 = ∞, the map that sends z1 7→ 1, ∞ 7→ 0 and z3 7→ ∞ is
T (z) = (z1 − z3 )/(z − z3 ).
The Möbius transformation that sends any three distinct points to 1, 0,
and ∞ is so useful that it gets its own name and special notation.
De nition 3.4.10. The cross ratio of 4 complex numbers z, w, u, and v,
where w, u, and v are distinct, is denoted (z, w; u, v), and
z−u w−v
(z, w; u, v) = · .
z−v w−u
If z is a variable, and w, u, and v are distinct complex constants, then
T (z) = (z, w; u, v) is the (unique!) Möbius transformation that sends w →
7 1,
u 7→ 0, and v 7→ ∞.
Example 3.4.11: Building a Möbius transformation.
Find the unique Möbius transformation that sends 1 7→ 3, i 7→ 0, and
2 7→ −1.
One approach: Find T (z) = (z, 1; i, 2) and S(w) = (w, 3; 0, −1). In
this case, the transformation we want is S −1 ◦ T .
To nd this transformation, we set the cross ratios equal:
(z, 1; i, 2) = (w, 3; 0, −1)
z−i 1−2 w−0 3+1
· = ·
z−2 1−i w+1 3−0
52 3. Transformations
−z + i 4w
= .
(1 − i)z − 2 + 2i 3w + 3
Then solve for w:
−3zw + 3iw + 3i − 3z = 4[(1 − i)z − 2 + 2i]w
−3z + 3i = [3z − 3i + 4[(1 − i)z − 2 + 2i]]w
−3z + 3i
w= .
(7 − 4i)z + (−8 + 5i)
Thus, our Möbius transformation is
−3z + 3i
V (z) = .
(7 − 4i)z + (−8 + 5i)
It's quite easy to check our answer here. Since there is exactly one
Möbius transformation that does the trick, all we need to do is check
whether V (1) = 3, V (i) = 0 and V (2) = −1. Ok... yes... yes... yep!
We've got our map!
Theorem 3.4.12 Invariance of Cross Ratio. Suppose z0 , z1 , z2 , and z3 are
four distinct points in C+ , and T is any Möbius transformation. Then
(z0 , z1 ; z2 , z3 ) = (T (z0 ), T (z1 ); T (z2 ), T (z3 )).
Proof. Let T be an arbitrary Möbius transformation, and de ne S(z) =
(z, z1 ; z2 , z3 ), which sends z1 7→ 1, z2 7→ 0, and z3 7→ ∞. Notice, the
composition S ◦ T −1 is a Möbius transformation that sends T (z1 ) 7→ 1,
T (z2 ) 7→ 0, and T (z3 ) 7→ ∞. So this map can be expressed as a cross ratio:
S ◦ T −1 (z) = (z, T (z1 ); T (z2 ), T (z3 )).
Plugging T (z0 ) into this transformation, we see
S ◦ T −1 (T (z0 )) = (T (z0 ), T (z1 ); T (z2 ), T (z3 )).
On the other hand, S ◦ T −1 (T (z0 )) = S(z0 ), which equals (z0 , z1 ; z2 , z3 ).
So we have proved that
(z0 , z1 ; z2 , z3 ) = (T (z0 ), T (z1 ); T (z2 ), T (z3 )).
Example 3.4.13: Do four points lie on a single cline?
In addition to de ning maps that send points to 1, 0, and ∞, the
cross ratio can proclaim whether four points lie on the same cline: If
(z, w; u, v) is a real number then the points are all on the same cline;
if (z, w; u, v) is complex, then they aren't. The proof of this fact is left
as an exercise.
Take the points 1, i, −1, −i. We know these four points lie on the
circle |z| = 1, so according to the statement above, (1, i; −1, −i) is a
real number. Let's check:
1+1 i+i
(1, i; −1, −i) = ·
1+i i+1
2 2i
=
1+i1+i
4i
=
(1 − 1) + 2i
3.4. Möbius Transformations 53
4i
=
2i
= 2. (Yep!)
Another important feature of inversion that gets passed on to Möbius
transformations is the preservation of symmetry points. The following result
is a corollary to Theorem 3.2.12.
Corollary 3.4.14. If z and z ∗ are symmetric with respect to the cline C ,
and T is any Möbius transformation, then T (z) and T (z ∗ ) are symmetric with
respect to the cline T (C).
We close the section with one more theorem about Möbius transformations.
Theorem 3.4.15. Given any two clines C1 and C2 , there exists a Möbius
transformation T that maps C1 onto C2 . That is, T (C1 ) = C2 .
Proof. Let p1 C1 and q1 and q1∗ symmetric with respect to
be a point on
C1 . Similarly, let p2 be a point on C2 and q2 and q2∗ be symmetric with respect
to C2 . Build the Möbius transformation that sends p1 7→ p2 , q1 7→ q2 and
q1∗ 7→ q2∗ . Then T (C1 ) = C2 .
Exercises
1. Find a transformation of C+ that rotates points about 2i by an angle
π/4. Show that this transformation has the form of a Möbius transformation.
2. Find the inverse transformation of T (z) = 3z+i
2z+1 .
3. Prove Theorem 3.4.3. That is, suppose T and S are two Möbius
transformations and prove that the composition T ◦ S is again a Möbius
transformation.
4. Prove that any Möbius transformation can be written in a form with
determinant 1, and that this form is unique up to sign. Hint: How does the
determinant of T (z) = (az + b)/(cz + d) change if we multiply top and bottom
of the map by some constant k?
5. Find the unique Möbius transformation that sends 1 7→ i, i 7→ −1, and
−1 7→ −i. What are the xed points of this transformation? What is T (0)?
What is T (∞)?
6. Repeat the previous exercise, but send 2 → 0, 1 → 3 and 4 → 4.
7. Prove this feature of the cross ratio: (z, z1 ; z2 , z3 ) = (z, z1 ; z2 , z3 ).
8. Prove that the cross ratio of four distinct real numbers is a real number.
9. Prove that the cross ratio of four distinct complex numbers is a real
number if and only if the four points lie on the same cline. Hint: Use the
previous exercise and the invariance of the cross ratio.
10. Do the points 2 + i, 3, 5, and 6+i lie on a single cline?
11. More on Möbius transformations.
a. Give an example of a Möbius transformation T such that T (z) 6= T (z)
for some z C+ .
in
b. Suppose T is a Möbius transformation that sends the real axis onto
itself. Prove that in this case, T (z) = T (z) for all z in C.
12. Is there a Möbius transformation that sends 1 to 3, i to 4, -1 to 2+i
and −i to 4 + i? Hint: It may help to observe that the input points are on a
single cline.
54 3. Transformations
13. Find the xed points of these transformations on C+ . Remember that
∞ can be a xed point of such a transformation.
2z
a. T (z) = 3z−1
b. T (z) = iz
−iz
c. T (z) = (1−i)z−1
14. Find a Möbius transformation that takes the circle |z| = 4 to the straight
line 3x + y = 4. Hint: Track the progress of three points, and the rest will
follow.
15. Find a non-trivial Möbius transformation that xes the points -1 and
1, and call this transformation T. Then, let C be the imaginary axis. What is
the image of C under this map. That is, what cline is T (C)?
16. Suppose z1 , z2 , z3 are distinct points in
+
C . Show that by an even
number of inversions we can map z1 7→ 1, z2 7→ 0, and z3 7→ ∞ in the case
that z3 = ∞.
3.5 Möbius Transformations: A Closer Look
To visualize Möbius transformations it is helpful to focus on xed points and,
in the case of two xed points, on two families of clines with respect to these
points.
Given two points p and q in C+ ,a type I cline of p and q is a cline
that goes through p and q, and a type II cline of p and q is a cline with
respect to which p and q are symmetric. Type II clines are also called circles
of Apollonius (see Exercise 3.3.2). Figure 3.5.1 shows some type I and type II
clines of p and q. The type II clines of p and q are dashed.
p q
Figure 3.5.1: Type I clines (solid) and Type II clines (dashed) of p and q.
By Theorem 3.2.8, any type II cline of p and q intersects any type I cline of
p and q at right angles. Furthermore, because Möbius transformations preserve
clines and symmetry points, we can be assured that Möbius transformations
preserve type I clines as well as type II clines. In particular, if C is a type I
cline of p and q, then T (C) is a type I cline of T (p) and T (q). Similarly, if C
is a type II cline of p and q , then T (C) is a type II cline of T (p) and T (q). We
can use this to our advantage.
For instance, the type I clines of the points 0 and ∞ are, precisely, lines
through the origin, while the type II clines of 0 and ∞ are circles centered
at the origin. (Remember, inversion in a circle takes the center of the circle
to ∞.) The type I clines in this case are clearly perpendicular to the type
II clines, and they combine to create a coordinate system of the plane (polar
coordinates), as pictured in Figure 3.5.2(a). We can move this system of clines
3.5. Möbius Transformations: A Closer Look 55
p q
(a) (b)
Figure 3.5.2: (a) Type I clines (solid) and type II clines (dashed) of points 0
and ∞ (solid); (b) A Möbius transformation sending 0 7→ p and ∞ 7→ q sends
type I and II clines of 0 and ∞ to type I and II clines of p and q , respectively.
by considering a Möbius transformation that maps 0 to p and ∞ q (where
to
p, q 6= ∞). Lines through the origin get mapped to type I clines of p and q ,
and circles through the origin get mapped to type II clines of p and q . The
result is a system of clines that serves as a general coordinate system for the
plane. Each point z in the plane is at the intersection of a single type I cline
of p and q and a single type II cline of p and q, and these two clines intersect
at right angles.
Let's get back to xed points and how they can help us describe Möbius
transformations. We consider the case that 0 and ∞ are xed before proceeding
to the general case.
Example 3.5.3: Fixing 0 and ∞.
Suppose T (z) = (az + b)/(cz + d) is a Möbius transformation that xes
0 and ∞. In this case, the form of the Möbius transformation can be
simpli ed. In particular, since T (0) = 0, it follows that b = 0. And
a
since T (∞) = ∞, it follows that c = 0. Thus, T (z) =
d z which may
be written as
T (z) = reiθ z.
With T in this form, it is clear that if T xes 0 and ∞, then T is a
combination of a dilation (by factor r) and a rotation about the origin
(by a factor θ). We may assume that r > 0 in the above equation,
because if it is negative, we can turn it into a positive constant by
adding π to the angle of rotation.
A dilation by r will push points along lines through the origin.
These lines are precisely the type I clines of 0 and ∞. All points in
the plane either head toward ∞ (if r > 1) or they all head toward 0 (if
0 < r < 1). Of course, if r = 1 there is no dilation.
Meanwhile, rotation about 0 by θ pushes points along circles
centered at the origin. These circles are precisely the type II clines
of 0 and ∞.
Now suppose T is a Möbius transformation that xes two nite points p
and q (neither is ∞). Let
z−p
S(z) =
z−q
56 3. Transformations
be a Möbius transformation that takes p to 0 and q to ∞. Let U be the Möbius
transformation determined by the composition equation
U = S ◦ T ◦ S −1 (1)
Notice
U (0) = S ◦ T ◦ S −1 (0) = S ◦ T (p) = S(p) = 0,
and
U (∞) = S ◦ T ◦ S −1 (∞) = S ◦ T (q) = S(q) = ∞.
That is, is a Möbius transformation that xes 0 and ∞.
U So, by
Example 3.5.3, U is a rotation, a dilation, or some combination of those, and
U looks like U (z) = reiθ z .
In any event, focusing on T again and using equation (1), which can be
rewritten as S ◦ T = U ◦ S , we arrive at the following equation, called the
normal form of the Möbius transformation in this case.
Normal form, two xed points.
The normal form of a Möbius transformation T xing distinct points p
and q (neither of which is ∞):
T (z) − p z−p
= reiθ ·
T (z) − q z−q
This normal form is much more illuminating than the standard a, b, c, d
form because, although the map is still described in terms of four constants
(p, q, r, θ ), each constant now has a simple geometric interpretation: p and q
are xed points, r is a dilation factor along type I clines of p and q, and θ is a
rotation factor around type II clines of p and q.
In particular, thanks to composition equation (1) we can view T as the
composition T = S −1 ◦ U ◦ S. T moves points according to
With this view,
a three-leg journey. Think of a general point z clinging to the intersection
of a single type II cline of p and q and a single type I cline of p and q (see
Figure 3.5.4).
S U (S(z))
T (z)
z
p q S(z)
S −1
Figure 3.5.4: Tracking the image of z if T xes p and q.
First, z gets sent via S to S(z), which is at the intersection of a line through
the origin and a circle centered at the origin. Second, U (which has the form
U (z) = reiθ z ), sends S(z) along this line through the origin (by dilation factor
r), and then around a new circle centered at the origin (by rotation factor θ)
−1
to the point U (S(z)). Third, S sends U (S(z)) back to the intersection of a
3.5. Möbius Transformations: A Closer Look 57
type I cline of p and q and a type II cline of p and q . This point of intersection
is S −1 (U (S(z))) and is equivalent to T (z).
Though fatigued, our well-traveled point realizes there's a shortcut. Why
go through this complicated wash? We can understand T as follows: T will
push points along type I clines of p and q (according to the dilation factor r)
and along type II clines of p and q (according to the rotation factor θ ).
iθ
We emphasize two special cases of this normal form. If |re | = 1 there is
no dilation, and points simply get rotated about type II clines of p and q as
in Figure 3.5.5. Such a Möbius transformation is called an elliptic Möbius
transformation .
q
p
Figure 3.5.5: An elliptic Möbius transformation xing p and q swirls points
around type II clines of p and q.
The second special case occurs when θ = 0. Here we have a dilation factor
r, but no rotation. All points move along type I clines of p and q, as in
Figure 3.5.6. A Möbius transformation of this variety is called a hyperbolic
Möbius transformation . A hyperbolic Möbius transformation xing p and
q either sends all points away from p and toward q or vice versa, depending on
the value of r.
q
p
Figure 3.5.6: A hyperbolic Möbius transformation xing p and q pushes
points away from one xed point and toward the other along type I clines of p
and q.
If we are not in one of these special cases, then T is simply a combination
of these two, and a Möbius transformation of this type is often called
loxodromic .
If a Möbius transformation xes two nite points, say p and q , and it is not
the identity transformation, then some nite point gets sent to ∞. Moreover,
∞ gets sent to some nite point. The point sent to in nity is called the pole
of the transformation and is often denoted z∞ . That is, T (z∞ ) = ∞. The
inverse pole of T is the image of ∞ under the map, which is often denoted
as w∞ . That is, T (∞) = w∞ . There is a simple relationship between the four
points p, q, z∞ , and w∞ .
58 3. Transformations
Lemma 3.5.7. Suppose T is a Möbius transformation that xes distinct nite
points p and q , sends z∞ to ∞, and sends ∞ to w∞ . Then p + q = z∞ + w∞ .
Proof. Suppose T satis es the conditions of the lemma. Then T has normal
form
z−p T (z) − p
=λ ,
z−q T (z) − q
where λ = reiθ . Plug z = z∞ into the normal form to see
z∞ − p
= λ · 1.
z∞ − q
Plug z=∞ into the normal form to see
w∞ − p
1=λ .
w∞ − q
Next solve each equation for λ, set them equal, cross multiply, and simplify
as follows to get the result:
z∞ − p w∞ − q
=
z∞ − q w∞ − p
(z∞ − p)(w∞ − p) = (w∞ − q)(z∞ − q)
p2 − pw∞ − pz∞ = q 2 − qw∞ − qz∞
p2 − q 2 = p(z∞ + w∞ ) − q(z∞ + w∞ )
(p − q)(p + q) = (p − q)(z∞ + w∞ )
p + q = z∞ + w∞ . (since p 6= q )
This completes the proof.
Theorem 3.5.8. If T is a Möbius transformation that xes two distinct, nite
points p and q , sends z∞ to ∞, and sends ∞ to w∞ , then
w∞ z − pq
T (z) = .
z − z∞
Proof. In the proof of Lemma 3.5.7, we found that the constant λ in the
normal form of T is
z∞ − p
λ= .
z∞ − q
It follows that T has the normal form
z−p z∞ − p T (z) − p
= · .
z−q z∞ − q T (z) − q
Solve this expression for T (z) and reduce it using the fact that p+q =
z∞ + w∞ to get the expression for T that appears in the statement of the
theorem. The details are left to the reader.
Example 3.5.9: Fix -1 and 1 and send i 7→ ∞.
By Lemma 3.5.7, the Möbius transformation sends ∞ to −i, so by
Theorem 3.5.8,
−iz − (1)(−1) −iz + 1
T (z) = = .
z−i z−i
3.5. Möbius Transformations: A Closer Look 59
Example 3.5.10: Analyze a Möbius transformation.
Consider the Möbius transformation
(6 + 3i)z + (2 − 3i)
T (z) = .
z+3
First we nd the xed points and the normal form of T. To nd
the xed points we solve T (z) = z for z.
(6 + 3i)z + (2 − 3i)
=z
z+3
(6 + 3i)z + (2 − 3i) = z 2 + 3z
z 2 − (3 + 3i)z − (2 − 3i) = 0.
Hey! Wait a moment! This looks familiar. Let's see ... yes! We
showed in Example 2.4.4 that this quadratic equation has solutions
z=i and z = 3 + 2i.
So the map has these two xed points, and the normal form of T is
T (z) − i z−i
=λ .
T (z) − (3 + 2i) z − (3 + 2i)
To nd the value of λ, plug into the normal form a convenient value
of z. For instance, T (−3) = ∞, so
−3 − i
1=λ .
−3 − (3 + 2i)
It follows thatλ = 2, so T is a hyperbolic map that pushes points
along clines through i and 3 + 2i. Below is a schematic for how the
+ 2
map pushes points around in C . Notice T (0) =
3 − i, T (1) = 2, and
T (4i) = 2.16 + 4.12i. Points are moving along type I clines of i and
3 + 2i away from i and toward 3 + 2i.
From the original description of T we observe that the pole of the
map is z∞ = −3, and the inverse pole of the map is w∞ = 6+3i. Notice
that z∞ , w∞ , and the two xed points all lie on the same Euclidean line.
This will always be the case for a hyperbolic Möbius transformation
(Exercise 3.5.9).
4i T (4i)
w∞
3 + 2i
z∞ 0 1 T (1)
T (0)
Example 3.5.11: Fix i and 0 and send 1 7→ 2.
The normal form of this map is
T (z) − i z−i
=λ .
T (z) − 0 z−0
60 3. Transformations
Since T (1) = 2 we know that
2−i
= λ(1 − i).
2
Solving for λ we have
3 1
λ= + i,
4 4
and the map is loxodromic.
√
Expressing λ in polar form, λ = reiθ , gives r = 410 and θ =
arctan(1/3). So T pushes points along type I clines of i and 0 according
to the scale factor r and along type II clines of i and 0 according to
the angle θ .
Now we consider Möbius transformations that x just one point. One such
Möbius transformation comes to mind immediately. For any complex number
d, the translation T (z) = z + d xes just ∞. In the exercises, you prove
that translations are the only Möbius transformations that x ∞ and no other
point.
1
Now suppose T xes p 6= ∞ S(z) = z−p
(and no other point). Let be
−1
a Möbius transformation taking p to ∞, and let U = S ◦ T ◦ S . Then
U (∞) = S(T (S −1 (∞))) = S(T (p)) = S(p) = ∞, and U xes no other point.
Thus, U (z) = z + d for some complex constant d.
The composition equation S ◦ T = U ◦ S gives the following equation called
the normal form of a Möbius transformation T xing p 6= ∞ (and no
other point):
Normal form, one xed point p 6= ∞.
1 1
= +d
T (z) − p z−p
Observe that U (z) = z + d pushes points along lines parallel to one another
in the direction of d (as in the right of Figure 3.5.12). All of these parallel
−1
lines meet at ∞ and are mutually tangent at this point. The map S takes
this system of clines to a system of clines that meet just at p, and are tangent
to one another at p, as pictured. The slope of the single line in this system
depends on the value of the constant d. In fact, the single line in the system
of clines is the line through p and T (∞) (see Exercise 3.5.12 for details).
S
d
p
0
S −1
Figure 3.5.12: A parabolic map xing p pushes points along clines that are
mutually tangent at p.
3.5. Möbius Transformations: A Closer Look 61
A map that xes just p will push points along such a system of clines that
are mutually tangent at p. Such a map is called parabolic . In a sense, a
parabolic map sends points both toward and away from p along these clines,
just as any translation pushes points along a line toward ∞ and also away from
∞.
Example 3.5.13: Normal form, one xed point..
Consider T (z) = (7z − 12)/(3z − 5). To nd its normal form we start
by nding its xed points.
z = T (z)
z(3z − 5) = 7z − 12
2
3z − 12z + 12 = 0
z 2 − 4z + 4 = 0
(z − 2)2 = 0
z = 2.
So T is parabolic and has normal form
1 1
= + d.
T (z) − 2 z−2
To nd d plug in the image of another point. Using the original
description of the map, we know T (0) = 2.4 so
1 1
= +d
0.4 −2
so that d = 3. The normal form is then
1 1
= + 3.
T (z) − 2 z−2
Exercises
1. Complete the proof of Theorem 3.5.8.
2. Analyze each of the Möbius transformations below by nding the xed
points, nding the normal form, and sketching the appropriate coordinate
system of clines, being sure to indicate the motion of the transformation.
z
a. T (z) = 2z−1
−z
b. T (z) = (1+i)z−i
c. T (z) = 3iz−5
z−i
3. A transformation T is called an involution if it is its own inverse. If this
is the case, then T ◦T is the identity transformation. Prove that if a Möbius
transformation T is an involution and not the identity transformation, it must
be elliptic.
4. Suppose a Möbius transformation T has the following property: There
are distinct points a, b, c in the complex plane C such that T (a) = b, T (b) =
c, T (c) = a.
a. What is the image of the unique cline through a, b, and c under T?
b. Explain why the triple composition T ◦T ◦T is the identity transforma-
tion.
c. Prove that T is elliptic.
62 3. Transformations
5. Prove that if the Möbius transformation T xes just ∞, then T (z) = z +d
for some complex constant d.
6. Find the Möbius transformation that xes 2 and 4 and sends 2+i to ∞.
7. Use the normal form to build and classify a Möbius transformation that
xes 4 and 8 and sends i to 0.
8. Suppose T is an elliptic Möbius transformation that xes the distinct,
nite points p and q.
a. Prove that the points z∞ and w∞ as de ned in Lemma 3.5.7 lie on the
perpendicular bisector of segment pq .
b. Show that T is the composition of two inversions about clines that
contain p and q . Hint: Think about which inversion xes p and q and takes
z∞ to ∞?
9. Suppose the Möbius transformation T xes the distinct, nite points p and
q and sends z∞ to ∞ and ∞ to w∞ . By Lemma 3.5.7 we know p+q = z∞ +w∞ .
Use the normal form of T to prove the following facts:
a. If T is elliptic then the four points p, q , z∞ , and w∞ form a rhombus.
Under what conditions is this rhombus actually a square?
b. If T is hyperbolic then these 4 points all lie on the same Euclidean line.
c. If T is loxodromic, then under what conditions do these four points
determine a rectangle?
10. Prove that any pair of nonintersecting clines in C may be mapped by
a Möbius transformation to concentric circles. Hint: By Theorem 3.2.16 there
are two points p and q in C that are symmetric with respect to both clines.
What happens if we apply a Möbius transformation that takes one of these
points to ∞?
11. Suppose T = iC1 ◦ iC2 where C1 and C2 are clines that do not
intersect. Prove that T has two xed points and these points are on all clines
perpendicular to both C1 and C2 .
12. Suppose T (z) is parabolic with normal form
1 1
= + d.
T (z) − p z−p
1
Prove that the line through p and p+ d gets sent to itself by T.
13. Analyze T (z) = [(1 + 3i)z − 9i]/[iz + (1 − 3i)] by nding the xed
points, nding the normal form, and sketching the appropriate system of clines
indicating the motion of the transformation.
14. Find a parabolic transformation with xed point 2+i for which
T (∞) = 8.
15. Given distinct points p, q , and z in C, prove there exists a type II cline
of p and q that goes through z.
4
Geometry
Recall the two paragraphs from Section 1.2 that we intended to spend time
making sense of and working through:
Whereas Euclid's approach to geometry was additive (he started
with basic de nitions and axioms and proceeded to build a sequence
of results depending on previous ones), Klein's approach was
subtractive. He started with a space and a group of allowable
transformations of that space. He then threw out all concepts that
did not remain unchanged under these transformations. Geometry,
to Klein, is the study of objects and functions that remain
unchanged under allowable transformations.
Klein's approach to geometry, called the Erlangen Program after
the university at which he worked at the time, has the bene t
that all three geometries (Euclidean, hyperbolic and elliptic)
emerge as special cases from a general space and a general set of
transformations.
+
We now have both the space (C ) and the transformations (Möbius transfor-
mations), and are just about ready to embark on non-Euclidean adventures.
Before doing so, however, one more phrase needs de ning: group of transfor-
mations. This phrase has a precise meaning. Not every collection of transfor-
mations is lucky enough to form a group.
4.1 The Basics
De nition 4.1.1. A collection G of transformations of a set A is called a
group of transformations if G has the following three properties:
1. Identity : G contains the identity transformation T :A→A de ned by
T (a) = a for all a ∈ A.
2. Closure : If T and S are two transformations in G, then the composition
T ◦S is in G.
3. Inverses : If T is in G, then the inverse T −1 is in G.
The reader who has seen group theory will know that in addition to the
three properties listed in our de nition, the group operation must satisfy
a property called associativity. In the context of transformations, the
group operation is composition of transformations, and this operation is
always associative: if R, S , and T are transformations of a set A, then the
transformation (R ◦ S) ◦ T equals the transformation R ◦ (S ◦ T ). So, in the
present context of transformations, we omit associativity as a property that
needs checking.
63
64 4. Geometry
Example 4.1.2: Group of translations.
Let T denote the collection of all translations of the plane C. In
particular, for each b ∈ C, let Tb : C → C denote the translation
Tb (z) = z + b. The set T consists of all Tb , for all b ∈ C. That is,
T = {Tb | b ∈ C}.
Show that T is a group of transformations.
Solution. To verify T forms a group, we must check the three
properties.
1. T contains the identity: Since 0 ∈ C, T contains T0 (z) = z + 0 =
z, which is the identity transformation of C.
2. T has closure: Suppose Tb and Tc are in T . Then Tb ◦ Tc (z) =
Tb (z + c) = (z + c) + b = z + (b + c). But this map is
exactly the translation Tb+c , which is in T since b + c ∈ C.
Thus, the composition of two translations is again a translation.
Notationally, we have shown that Tb ◦ Tc (z) = Tb+c (z).
3. T contains inverses: SupposeTb is in T , and consider T−b , which
is inT since −b ∈ C. Note that Tb ◦ T−b (z) = Tb (z − b) =
(z−b)+b = z , and T−b ◦Tb (z) = T−b (z+b) = (z+b)−b = z . Thus,
T−b is the inverse of Tb , and this inverse is in T . Notationally,
Tb−1 = T−b : the inverse of translation by b is translation back by
−b.
De nition 4.1.3. Let S be any set, and G a group of transformations on S.
The pair (S, G) geometry . A gure in the geometry is any subset
is called a
A of S . An element of S is called a point in the geometry. Two gures A and
B are called congruent , denoted A ∼ = B , if there exists a transformation T in
G such that T (A) = B .
Although a gure in a geometry (S, G) is de ned to be a subset of S, we
make one abuse of notation and sometimes treat points as gures. For instance,
we might write a ∼
= b for two points a and b in S when, formally, we mean
{a} ∼
= {b}. Incidentally, a ∼
= b in the geometry (S, G) means there exists a
transformation T ∈G such that T (a) = b.
Let's look at some examples now to help sort through these de nitions.
Example 4.1.4: Finite group of rotations.
Consider the set H = {R0 , Rπ/2 , Rπ , R3π/2 } consisting of four rotations
of C about the origin (by 0, π/2, π , and 3π/2 radians). We observe rst
i0
that H forms a group. Since R0 (z) = e z = z , H contains the identity
transformation on C. The set also satis es closure, and the reader
can check all possible compositions. For instance, R3π/2 ◦ Rπ = Rπ/2 .
Finally, the inverse of each transformation in H is again in H . Check
−1 −1 −1
that R0 = R0 , Rπ/2 = R3π/2 , Rπ−1 = Rπ , and R3π/2 = Rπ/2 . Thus,
H is a group and we may study the geometry (C, H). For instance, is
the circle C given by |z − i| = .5 congruent to the circle D given by
|z| = .5? Well, is there a transformation in H that maps C onto D?
No! The only four circles congruent to C are pictured below. These
are found by rotating C about the origin by 0, π/2, π , or 3π/2 radians,
the only allowable transformations in this geometry.
4.1. The Basics 65
C
i
Notice also that any point z 6= 0 is congruent to four points: z,
eiπ/2 z , eiπ z , and ei3π/2 z . How many points are congruent to z = 0?
Are all lines congruent in this geometry? Nope. We are only allowed
these few rotations, so we have no way to map the line y = x, say, to
the line y = x + 1.
Example 4.1.5: A two-element group.
Consider re ection of C across the real axis, given by r(z) = z . Since
r◦r is the identity map, the set G = {1, r} is a group of transformations
on C, and we may de ne the geometry (C, G). Notice that while 3 + i
is congruent to 3 − i in this geometry, it is not congruent to −3 + i.
Also, the circle |z − 2i| = 1 is congruent to the circle |z + 2i| = 1 but
not the circle |z − 3i| = 1.
Example 4.1.6: Translational geometry.
Let T denote the group of translations in Example 4.1.2, and consider
the geometry (C, T ). We call this geometry translational geometry .
Which gures in Figure 4.1.7 are congruent in this geometry?
Remember, two gures are congruent if we can nd a transformation
that moves one gure on top of the other. Since our allowable moves
here are translations, we cannot change the radius of a circle (that's a
dilation), and we cannot rotate objects. So, in translational geometry
the only gures congruent in Figure 4.1.7 are H and L.
B E F
C
G H
L
A
Figure 4.1.7: Figures in the plane.
De nition 4.1.8. A collection D of gures in a geometry (S, G) is called an
invariant set if, for any gure A in D and any transformation T in G, T (A)
is also in D . A function f de ned on D is called an invariant function if
f (B) = f (T (B)) for any gure B in D and any transformation T in G.
66 4. Geometry
For instance, suppose D is the set of all lines in C. Let f be the function
that takes a line to its slope. In translational geometry, (C, T ), the set D of
all lines is an invariant set because if A is any line, then so is its image, T (A),
under any translation T in T. Furthermore, f is an invariant function because
any translation of any line preserves the slope of that line.
Of course, two gures in an invariant set need not be congruent. For
instance, in translational geometry the set D of all lines is an invariant set,
although if lines A and B in D have dierent slopes then they are not congruent.
This feature of the set D makes it seem too big, in some sense. Can invariant
sets be more exclusive, containing only members that are congruent to one
another? You bet they can.
De nition 4.1.9. A set of gures D in a geometry is called minimally
invariant if no proper subset of it is also an invariant set.
For instance, the set of all lines is not a minimally invariant set in
translational geometry because it has proper subsets that are also invariant
sets. One such subset consists of all lines with slope 8.
Theorem 4.1.10. An invariant set D of gures in a geometry (S, G) is
minimally invariant if and only if any two gures in D are congruent.
Proof. First assume D is a minimally invariant set in the geometry (S, G),
and suppose A and B are arbitrary gures in D. We must show that A∼= B.
We begin by constructing a new set of gures, the one consisting of A and
all transformations of A. In particular, de ne
A = {T (A) | T ∈ G}.
Notice that for any T ∈ G, T (A) is in the set D since D is invariant. This
means that A is a subset of D.
Furthermore, A itself is an invariant set, thanks to the group nature of G.
In particular, if C is any member of A, then C = T0 (A) for some particular T0
in G. Thus, applying any transformation T to C ,
T (C) = T (T0 (A)) = T ◦ T0 (A)
and since T ◦ T0 is again a transformation in G, T ◦ T0 (A) lives in A.
So we've established two facts: (1) A is a subset of D, and (2) A is an
invariant set. Since D is a minimally invariant set it follows by de nition that
A = D. This means that the given set B, which is in D, is also in A. That is,
A∼= B.
The proof of the other direction is left as an exercise for the reader.
The proof of Theorem 4.1.10 illustrates a convenient way to nd minimally
invariant sets: If A is a gure in (S, G), then A = {T (A) | T ∈ G} is a
minimally invariant set.
Example 4.1.11: Euclidean geometry.
Euclidean geometry (C, E), where E consists of
is the geometry
all transformations of the form T (z) = eiθ z + b, where θ is a real
number and b is in C. Note that E consists of precisely those general
linear transformations of the form T (z) = az + b in which |a| = 1.
4.1. The Basics 67
In the exercises, you check that this collection is indeed a group of
transformations.
The group E includes rotations and translations, but not dilations.
Let's take a look at some familiar properties of objects that should be
invariant in Euclidean geometry.
The Euclidean distance between two points z1 and z2 is de ned
to be |z1 − z2 |. To show that this is an invariant function of (C, E), we
need to show that for any T in the group E, the distance between z1
and z2 equals the distance between T (z1 ) and T (z2 ):
|T (z1 ) − T (z2 )| = |(eiθ z1 + b) − (eiθ z2 + b)|
= |eiθ (z1 − z2 )|
= |eiθ ||z1 − z2 |
= |z1 − z2 |. (since |eiθ | = 1)
Thus, Euclidean distance is preserved in (C, E).
Angles are preserved as well. We have already proved that
general linear transformations preserve angles (Theorem 3.1.12), and
Euclidean transformations are general linear transformations, so angles
are preserved in (C, E).
De nition 4.1.12. A geometry (S, G) is called homogeneous if any two
points in S are congruent, and isotropic if the transformation group contains
rotations about each point in S.
Example 4.1.13: Homogeneous geometry.
Translational geometry (C, T ) is homogeneous because there is a
translation that will map any point of C to any other point of C. That
is, any two points of C are congruent. Of course, without rotations,
translational geometry is not isotropic. Here's a formal argument that
(C, T ) is homogeneous:
Suppose p and q are arbitrary points in C. We must nd a
translation T in T such that T (p) = q . Let w = q − p, and consider
the translation Tw in T . Then Tw (p) = p + w = p + (q − p) = q . Thus
Tw (p) = q and p ∼= q . Since p and q are arbitrary points in C it follows
that (C, T ) is homogeneous.
Euclidean geometry (C, E) is homogeneous since it contains all
translations, but the geometries of Example 4.1.4 and Example 4.1.5
are not.
De nition 4.1.14. A metric for a geometry (S, G) is an invariant function
d : S×S → R mapping each ordered pair (x, y) of elements from S to a real
number such that
1. d(x, y) ≥ 0 for all x, y ∈ S and d(x, y) = 0 if and only if x = y.
2. d(x, y) = d(y, x) for all x, y ∈ S .
3. (Triangle inequality) d(x, z) ≤ d(x, y) + d(y, z) for all x, y, z ∈ S .
Example 4.1.15: Euclidean metric.
The Euclidean metric is de ned by d(z, w) = |z − w|. We have already
shown that d is preserved under Euclidean transformations, and the
rst two conditions of being a metric follow directly from the de nition
68 4. Geometry
of modulus. We establish the triangle inequality by direct computation
in the following lemma.
Lemma 4.1.16. For any points z, w, v in C,
|z − w| ≤ |z − v| + |v − w|.
Proof. If v =w then the the result holds, so we assume v 6= w. Since d
is invariant under Euclidean transformations, we may assume that v = 0 and
w = r > 0 is a point on the positive real axis. (Translate the plane by −v to
send v to 0, and then rotate about 0 until the image of w under the translation
lands on the positive real axis.) Thus, it's enough to show that for any complex
number z and any positive real number r,
|z − r| ≤ |z| + r.
Notice that
|z − r| ≤ |z| + r ⇐⇒ |z − r|2 ≤ (|z| + r)2
⇐⇒ (z − r)(z − r) ≤ |z|2 + 2r|z| + r2
⇐⇒ |z|2 − r(z + z) + r2 ≤ |z|2 + 2r|z| + r2
⇐⇒ −(z + z) ≤ 2|z|
⇐⇒ −2Re(z) ≤ 2|z|
⇐⇒ −Re(z) ≤ |z|.
By letting z = a + bi, we may restate the last inequality as
p
−a ≤ a2 + b2 ,
√ √
which is true since −a ≤ |a| = a2 ≤ a2 + b2 .
Exercises
1. Find a particular translation to prove that in Figure 4.1.7 H ∼
= L in
translational geometry.
2. Let A be the set of all circles in C centered at the origin, and let G be
the set of all inversions about circles in A. That is,
G = {iC | C ∈ A}
Is G a group of transformations of C+ ? Explain.
3. Prove that the group E of Euclidean transformations of C is indeed a
group.
4. Let G be the set of all dilations of C+ . That is
G = {T (z) = kz | k ∈ R, k > 0}.
Is G a group of transformations of C+ ? Explain.
5. True or False? Determine whether the statement is true or false, and
support your answer with an argument.
a. Any two lines are congruent in Euclidean geometry (C, E).
b. Any two circles are congruent in Euclidean geometry (C, E).
6. Prove that if a set of gures D is invariant in a geometry (S, G), and any
two gures in D are congruent, then D is minimally invariant.
4.2. Möbius Geometry 69
7. Describe a minimally invariant set of translational geometry that contains
the gure D from Figure 4.1.7.
8. Rotational geometry is the geometry (C, R) where R is the group of
rotations about the origin. That is
R = {Rθ (z) = eiθ z | θ ∈ R}.
a. Prove that R is a group of transformations.
b. Is D = {all lines in C} an invariant set in rotational geometry? Is it a
minimally invariant set?
c. Find a minimally invariant set of rotational geometry that contains the
circle |z − (2 + i)| = 4.
d. Is (C, R) homogeneous? Isotropic?
9. Prove that the function v(z1 , z2 ) = z1 − z2 is invariant in translational
geometry (C, T ) but not rotational geometry (C, R).
10. Prove that the following function is a metric for any geometry (S, G).
(
0 if x = y;
d(x, y) =
1 if x 6= y .
11. Prove that (C, E) is isotropic. That is, show the group E contains all
rotations about all points in C.
12. Which gures from Figure 4.1.7 are congruent in (C, E)?
13. Let's create a brand new geometry, using the set of integers Z =
{. . . , −2, −1, 0, 1, 2, . . .}. For each integer n, we de ne the transformation
Tn : C → C by Tn (z) = z + ni. Let G denote the set of all transformations Tn
for all integers n. That is, G = {Tn | n ∈ Z}.
a. Prove that (C, G) is a geometry.
b. Consider the set of gures D consisting of all lines in the plane with
slope 4. Is D an invariant set of (C, G)? Is it minimally invariant? Explain.
c. My favorite line, for clear and personal reasons, is y = x + 8. Please
describe a minimally invariant set of gures containing this line.
d. Determine the set of points in C congruent to i in this geometry. Is C
homogeneous?
4.2 Möbius Geometry
We spent a fair amount of time studying Möbius transformations in Chapter
3, and this will pay dividends now.
De nition 4.2.1. Möbius Geometry is the geometry (C+ , M), where M
denotes the group of all Möbius transformations.
Without actually stating it, we essentially proved that M is a group
of transformations. Namely, we proved that the inverse of a Möbius
transformation is again a Möbius transformation, and that the composition
of two Möbius transformations is a Möbius transformation. We remark that
the identity map on C+ , T (z) = z , is a Möbius transformation (of the form
T (z) = (az + b)/(cz + d), where a = d = 1, and b = c = 0), so M is a group.
Below we recast the key results from Chapter 3 in geometric terms:
Any two clines are congruent in Möbius Geometry (Theorem 3.4.15).
70 4. Geometry
The set of all clines is a minimally invariant set of Möbius Geometry
(Theorems 3.4.5 and Theorem 3.4.15).
The cross ratio is an invariant of Möbius Geometry (Theorem 3.4.12).
Angle measure is an invariant of Möbius Geometry (Theorem 3.4.5).
While we're at it, let's restate three other facts about Möbius transforma-
tions:
Any transformation in M is uniquely determined by the image of three
points.
If T in M is not the identity map, then T xes exactly 1 or 2 points.
Möbius transformations preserve symmetry points.
What else? Euclidean distance is not an invariant function of Möbius
Geometry. To see this, one need look no further than the map T (z) = 1/z .
If p=2 and q =3 (two points on the real axis) then d(p, q) = |p − q| = 1.
However, their image points T (p) = 1/2 and T (q) = 1/3 have a Euclidean
distance between them of 1/6. So our old-fashioned notion of distance goes
out the window in Möbius geometry.
We emphasize that angles are preserved in Möbius geometry, which is a
good thing. Why is this a good thing? Remember that in the distant past,
humanity set out looking for a geometry in which Euclid's rst 4 postulates
hold true, but the 5th one fails. The 4th postulate states that all right angles
equal one another. This means that if Ralph is holding a right angle over in
the corner, and Randy is holding one down the block somewhere, we ought to
be able to transform one onto the other and see that the angles are the same.
Transformations do not change angles.
Rather than pursue the very general Möbius geometry, we take the preceed-
ing facts and apply them straight away to two of its special subgeometries,
hyperbolic geometry and elliptic geometry.
Exercises
1. Which gures in Figure 4.2.2 are congruent in (C+ , M).
2. Describe a minimally invariant set in
+
(C , M) containing the triangle
comprised of the three vertices 0, 1, and i and the three Euclidean line segments
connecting them. Be as speci c as possible about the members of this set.
3. Suppose p and q are distinct, nite points in C+ . Let G consist of all
elliptic Möbius transformations that x p and q. We consider the geometry
(C+ , G).
a. Show that G is a group of transformations.
b. Determine a minimally invariant set in (C+ , G) that contains the
Euclidean line through p and q.
c. Determine a minimally invariant set in (C+ , G) that contains the
perpendicular bisector of segment pq .
d. For any point z 6= p, q in C+ , characterize all points in C+ congruent to
z.
e. Is (C+ , G) homogeneous?
4. Repeat the previous exercise for the set G consisting of all hyperbolic
Möbius transformations that x p and q.
4.2. Möbius Geometry 71
A B C D
E F G H
Figure 4.2.2: Which of these gures are congruent in (C+ , M)?
5
Hyperbolic Geometry
Hyperbolic geometry can be modelled in many dierent ways. We will focus
here on the Poincaré disk model, developed by Henri Poincaré (1854-1912)
in around 1880. Poincaré did remarkable work in mathematics, though he
was never actually a professor of mathematics. He was particularly interested
in the relationship between mathematics, physics, and psychology. He began
studying non-Euclidean geometry in detail after it appeared in his study of two
apparently unrelated disciplines: dierential equations and number theory.
1
Poincaré took Klein's view that geometries are generated by sets and groups of
transformations on them. We consider a second model of hyperbolic geometry,
the upper half-plane model, in Section 5.5.
5.1 The Poincaré Disk Model
De nition 5.1.1. The Poincaré disk model for hyperbolic geometry is
the pair (D, H) where D consists of all points z in C such that |z| < 1, and H
consists of all Möbius transformations T for which T (D) = D. The set D is
called the hyperbolic plane , and H is called the transformation group in
hyperbolic geometry.
We note that H does indeed form a group of transformations, a fact that
is worked out in the exercises. Throughout this chapter the unit circle will be
called the circle at in nity , denoted by S1∞ . Of course, the circle at in nity
is not included in the hyperbolic plane D but bounds it. The circle at in nity
will be used extensively in our investigations. We note here that any Möbius
transformation that sends D to itself also sends S1∞ to itself.
0
C
S1∞
Figure 5.1.2: Inversion about a cline orthogonal to the unit circle takes D to
D.
1 See Arthur Miller's chapter in [13] for a discussion of Poincaré's diverse interests.
72
5.1. The Poincaré Disk Model 73
Consider a cline C S1∞ , as in
that is orthogonal to the circle at in nity
1
Figure 5.1.2. If we invert about C , S∞ is inverted to itself. Moreover, this
1
inversion takes the interior of S∞ , namely the hyperbolic plane D, to itself as
well, (Exercise 3.2.10). It follows that compositions of two such inversions is
a Möbius transformation that sends D to itself, and is thus in the group H.
These inversions play an important role in hyperbolic geometry, and we give
them a name.
An inversion in a cline C that is orthogonal to S1∞ is called a re ection
of the hyperbolic plane , or, a hyperbolic re ection .
It turns out that these re ections generate all the maps in H. For instance,
rotation about the origin is a Möbius transformation that sends D to itself, so it
is in H. But rotation about the origin is also the composition of two re ections
about lines that intersect at the origin. Since any line through the origin meets
the unit circle at right angles, re ection about such a line is a re ection of the
hyperbolic plane, so rotation about the origin is the composition of two such
re ections. We now prove the following general result.
Theorem 5.1.3. Any Möbius transformation in H is the composition of two
re ections of the hyperbolic plane.
Proof. Suppose T is a Möbius transformation that sends D to itself. This
means some point in D, say z0 , gets sent to the origin, 0. Let z0∗ be the point
symmetric to z0 with respect to S1∞ . Since T sends the unit circle to itself, and
Möbius transformations preserve symmetry points, it follows that T sends z0∗
1
to ∞. Furthermore, some point z1 on S∞ gets sent to the point 1.
∗
If z0 = 0, then z0 = ∞, and T xes 0 and ∞. Then by Example 3.5.3,
iθ
T (z) = re z is a dilation followed by a rotation. However, since T also sends D
onto D, the dilation factor must be r = 1. So T is simply a rotation about the
origin, which is the composition of two hyperbolic re ections about Euclidean
lines through the origin.
Now assume z0 6= 0. In this case, by using z0 , z0∗ , and z1 as anchors, we
may achieve T via two hyperbolic re ections, as follows:
First, invert about a circle C orthogonal to S1∞ that sends z0 to the origin.
Such a circle does indeed exist, and we will construct it now. As in
Figure 5.1.4, draw circle C1 with diameter 0z0∗ . Let p be a point of intersection
1 ∗
of C1 and the unit circle S∞ . Construct the circle C through p centered at z0 .
∗ 1 1
Since ∠0pz0 is right, S∞ is orthogonal to C , so inversion about C sends S∞
∗
to itself. Furthermore, since z0 gets sent to ∞ and symmetry points must be
preserved, inversion in C sends z0 to 0.
∗
Thus, the rst inversion takes z0 to 0 and z0 to ∞. To build T we must
also send z1 to 1. Note that inversion in circle C will have sent z1 to some
0 1
point z1 on the unit circle (since S∞ is sent to itself ). Now re ect across the
0 0
line through the origin that bisects angle ∠10z1 . This sends z1 to 1, sends
D to D, and leaves 0 and ∞ xed. Composing these two inversions yields
∗
a Möbius transformation that sends z0 to 0, z0 to ∞ and z1 to 1. Since a
Möbius transformation is uniquely determined by the image of three points,
this Möbius transformation is T.
Notice that in proving Theorem 5.1.3 we proved the following useful fact.
Lemma 5.1.5. Given z0 in D and z1 on S1∞ there exists a transformation in
H that sends z0 to 0 and z1 to 1.
Therefore, one may view any transformation in H as the composition of
two inversions about clines orthogonal to S1∞ . Moreover, these maps may
be categorized according to whether the two clines of inversion intersect zero
74 5. Hyperbolic Geometry
C1
z0∗
0 z0
C
S1∞
Figure 5.1.4: Constructing a circle C orthogonal to S1∞ about which z0 is
inverted to 0.
times, once, or twice. In Figure 5.1.6 we illustrate these three cases. In each
case, we build a transformation T in H by inverting about the solid clines in
the gure ( rst about L1 , then about L2 ). The gure also tracks the journey
of a point z under these inversions, rst to z0 by inverting about L1 , then
0
onto T (z) by inverting z about L2 . The dashed clines in the gure represent
some of the clines of motion, the clines along which points are moved by the
transformation. Notice these clines of motion are orthogonal to both clines of
inversion. Let's work through the three cases in some detail. If the two clines
L1 L1
0
z z
p z
0
z
p
T (z)
T (z)
L2
L2 (b)
(a) p (b)
L1
z
L2
z0
T (z)
q
(c)
Figure 5.1.6: The three types of transformations in H: (a) hyperbolic
rotation (b) parallel displacement (c) hyperbolic translation.
of inversion, L1
L2 , intersect inside D, say at the point p, then they also
and
intersect outside D at the point p∗ symmetric to p with respect to the unit circle
1
since both clines are orthogonal to S∞ . This scene is shown in Figure 5.1.6(a).
5.1. The Poincaré Disk Model 75
The resulting Möbius transformation will x p (and p∗ ), causing a rotation of
∗
points in D around p along type II clines of p and p . Not surprisingly, we will
call this type of map in H a rotation of the hyperbolic plane about the
point p; or, if the context is clear, we call such a map a rotation about p.
If the clines of inversion intersect just once, then it must be at a point p
on the unit circle. Otherwise the symmetric point p∗ , which would be distinct
from p, would also belong to both clines, giving us two points of intersection.
The resulting map moves points along circles in D that are tangent to the unit
circle at this point p. Circles in D that are tangent to the unit circle are called
horocycles , and this type of map is called a parallel displacement. See
Figure 5.1.6(b).
If the clines of inversion do not intersect, then at least one of the clines
must be a circle; and according to Theorem 3.2.16, there are two points p and
q symmetric with respect to both clines. These two points will be xed by
the resulting Möbius transformation, since each inversion sends p to q and q
to p. Furthermore, these two xed points must live on the unit circle by a
symmetric points argument (the details of which are left as an exercise). In
L1 and L2 do not intersect, then they are
other words, if the clines of inversion
type II clines of the xed points q of the resulting transformation in H.
p and
Moreover, this transformation will push points along type I clines of p and q .
We call such a Möbius transformation in H a translation of the hyperbolic
plane or, simply, a translation . See Figure 5.1.6(c).
q
T 2 (z)
T (z)
L1
T 2 (z) p
T (z)
L2
L2 z z L1
p
(a) (b)
Figure 5.1.7: Moving an `M' about the hyperbolic plane, by (a) rotation
about p; and (b) translation xing ideal points p and q.
Example 5.1.8: Moving an `M' around in (D, H).
Figure 5.1.7(a) depicts the image of the gure M (resembling the letter
`M') under two applications of a rotation T of the hyperbolic plane
about point p. This rotation is generated by two inversions about
clines L1 L2 that intersect at p (and meet S1∞ at right angles).
and
The gure M is pictured, as is T (M ) and T (T (M )). The gure also
tracks successive images of a point z on M : z gets mapped to T (z)
2
which gets mapped to T (z) = T (T (z)). By this map T , any point z
∗
in D rotates around p along the unique type II cline of p and p that
contains z .
Figure 5.1.7(b) depicts the image of M under two applications of
a translation of the hyperbolic plane. The translation is generated by
76 5. Hyperbolic Geometry
two inversions about non-intersecting clines L1 and L2 (that meet S1∞
at right angles). The xed points of the map are the points p and q
on the circle at in nity. Under this translation, any point z in D heads
away from p and toward q on the unique cline through the three points
p, q, and z.
We now derive the following algebraic description of transformations in
H:
Transformations in H.
Any transformation T in the group H has the form
z − z0
T (z) = eiθ
1 − z0z
where θ is some angle, and z0 is the point inside D that gets sent to 0.
Assume z0 is in D, z1 is on the unit circle S1∞ , and that the map T in H
sends z0 7→ 0, z0∗ 7→ ∞ and z1 7→ 1. Using the cross ratio,
z1 − z0∗ z − z0
T (z) = (z, z1 ; z0 , z0∗ ) = · .
z1 − z0 z − z0∗
But z0∗ = 1
z0 , so
z1 − 1/z0 z − z0 z0 z1 − 1 z − z0
T (z) = · = · .
z1 − z0 z − 1/z0 z1 − z0 z0 z − 1
Now, the quantity
z0 z1 − 1
z1 − z0
is a complex constant, and it has modulus equal to 1. To see this, observe that
since 1 = |z1 |2 = z1 z1 ,
z0 z1 − 1 z0 z1 − z1 z1
=
z1 − z0 z1 − z0
−z1 (z1 − z0 )
= .
z1 − z0
Since |z1 | = 1 and, in general |β| = |β| we see that this expression
has modulus 1, and can be expressed as eiθ for some θ. Thus, if T is a
transformation in H it may be expressed as
z − z0
T (z) = eiθ
1 − z0z
where θ is some angle, and z0 is the point inside D that gets sent to 0.
Is the converse true? Is every transformation in the form above actually a
member of H? The answer is yes, and the reader is asked to work through the
details in the exercises.
Exercises
1. Prove that (D, H) is homogeneous.
5.2. Figures of Hyperbolic Geometry 77
2. Suppose T is a Möbius transformation of the form
z − z0
T (z) = eiθ ,
1 − z0z
where z0 is in D. Prove that T maps D to D by showing that if |z| < 1 then
|T (z)| < 1. Hint: It is easier to prove that |z|2 < 1 implies |T (z)|2 < 1.
3. Construct the xed points of the hyperbolic translation de ned by the
inversion of two nonintersecting clines that intersect S1∞ at right angles, as
shown in the following diagram.
S1∞
4. Suppose p and q are two points in D. Construct two clines orthogonal
to S1∞ that, when inverted about in composition, send p to 0 and q to the
positive real axis.
5. Prove that any two horocycles in (D, H) are congruent.
6. Prove that H is a group of transformations.
5.2 Figures of Hyperbolic Geometry
The Euclidean transformation group, E, consisting of all (Euclidean) rotations
and translations, is generated by re ections about Euclidean lines. Similarly,
the transformations in H are generated by hyperbolic re ections, which are
inversions about clines that intersect the unit circle at right angles. This
suggests that these clines ought to be the lines of hyperbolic geometry.
De nition 5.2.1. A hyperbolic line in (D, H) is the portion of a cline inside
D that is orthogonal to the circle at in nity S1∞ . A point on S1∞ is called an
ideal point . Two hyperbolic lines are parallel if they share one ideal point.
Figure 5.2.2: A few hyperbolic lines in the Poincaré disk model.
Theorem 5.2.3. There exists a unique hyperbolic line through any two distinct
points in the hyperbolic plane.
78 5. Hyperbolic Geometry
Proof. Let p and q D. Construct the point p∗
be arbitrary points in
1
symmetric to p with respect to the unit circle, S∞ . Then there exists a cline
∗ 1
through p, q and p , and this cline will be orthogonal to S∞ , so it gives a
hyperbolic line through p and q. Since there is just one cline through p, q and
p∗ , this hyperbolic line is unique.
Which hyperbolic lines happen to be portions of Euclidean lines (instead
of Euclidean circles)? A Euclidean line intersects a circle at right angles if and
only if it goes through the center of the circle. Thus, the only hyperbolic lines
that also happen to be Euclidean lines are those that go through the origin.
One may also use a symmetric points argument to arrive at this last fact.
Any Euclidean line goes through ∞. To be a hyperbolic line (i.e., to be
orthogonal to S1∞ ), the line must also pass through the point symmetric to
∞ with respect to the unit circle. This point is 0. Thus, to be a hyperbolic
line in (D, H), a Euclidean line must go through the origin.
Theorem 5.2.4. Any two hyperbolic lines are congruent in hyperbolic geom-
etry.
Proof. We rst show that any given hyperbolic line L is congruent to the
hyperbolic line on the real axis. Suppose p
L, and v is one of its
is a point on
ideal points. By Lemma 5.1.5 there is a transformation T in H that maps p to
∗
0, v to 1, and p to ∞. Thus T (L) is the portion of the real axis inside D, and
L is congruent to the hyperbolic line on the real axis. Since any hyperbolic
line is congruent to the hyperbolic line on the real axis, the group nature of H
ensures that any two hyperbolic lines are congruent.
L
p
1
0 T (L)
Figure 5.2.5: Any hyperbolic line is congruent to the hyperbolic line on the
real axis.
Theorem 5.2.6. Given a point z0 and a hyperbolic line L not through z0 ,
there exist two distinct hyperbolic lines through z0 that are parallel to L.
Proof. Consider the case where z0 is at the origin. The line L has two ideal
points, call them u and v, as in Figure 5.2.7. Moreover, since L does not go
through the origin, Euclidean segment uv is not a diameter of the unit circle.
Construct one Euclidean line through 0 and u, and a second Euclidean line
through 0 and v. (These lines will be distinct because uv is not a diameter of
the unit circle.) Each of these lines is a hyperbolic line through 0, and each
shares exactly one ideal point with L. Thus, each is parallel to L. The fact
that the result holds for general z0 is left as an exercise.
5.2. Figures of Hyperbolic Geometry 79
0
v
Figure 5.2.7: Through a point not on a given hyperbolic line L there exist
two hyperbolic lines parallel to L.
Figure 5.2.7 illustrates an unusual feature of parallel lines in hyperbolic
geometry: there is no notion of transitivity. In Euclidean geometry we
know that if line L is parallel to line M, and line M is parallel to line
N, then line L is parallel to line N. This is not the case in hyperbolic
geometry.
Example 5.2.8: Hyperbolic triangles.
Three points in the hyperbolic plane D that are not all on a single
hyperbolic line determine a hyperbolic triangle. The hyperbolic
triangle ∆pqr is pictured below. The sides of the triangle are portions
of hyperbolic lines.
Are all hyperbolic triangles congruent? No. Since transformations
in H are Möbius transformations they preserve angles, so triangles with
dierent angles are not congruent.
q
r
The next section develops a distance function for the hyperbolic plane. As
in Euclidean geometry, we want to be able to compute the distance between
two points, the length of a path, the area of a region, and so on. Moreover, the
distance function should be an invariant; the distance between points should
not change under a transformation in H. With this in mind, consider again a
hyperbolic rotation about a point p, as in Figure 5.1.6(a). It xes the point p
and moves points around type II clines of p and p∗ . If the distance between
points is unchanged under transformations in H, then all points on a given
type II cline of p and p∗ will be the same distance away from p. This leads us
to de ne a hyperbolic circle as follows.
De nition 5.2.9. Suppose p is any point in D, and p∗ is the point symmetric
to p with respect to the unit circle. A hyperbolic circle centered at p is a
Euclidean circle C inside D that is a type II circle of p and p∗ .
80 5. Hyperbolic Geometry
Figure 5.2.10 shows a typical hyperbolic circle. This circle is centered at
point p and contains point q. Construction of such a circle may be achieved
with compass and ruler as in Exercise 5.2.5.
p p∗
Figure 5.2.10: A hyperbolic circle centered at p through q.
Theorem 5.2.11. Given any points p and q in D, there exists a hyperbolic
circle centered at p through q .
∗
Proof. Given p, q ∈ D, construct p , the point symmetric to p with respect
1 ∗
to S∞ . Then by Exercise 3.5.15 there exists a type II cline of p and p that
1
goes through q . This type II cline lives within D because S∞ is also a type
∗
II cline of p and p , and distinct type II clines cannot intersect. This type II
cline is the hyperbolic circle centered at p through q.
Exercises
1. Suppose C is a hyperbolic circle centered at z0 through point p.
Show that there exists a hyperbolic line L tangent to C at p, and that L
is perpendicular to hyperbolic segment z0 p.
2. Constructing a hyperbolic line through two given points.
a. Given a point p in D, construct the point p∗ symmetric to p with respect
to the unit circle (see Figure 3.2.18).
b. Suppose q D. Construct the cline through p, q , and
is a second point in
p∗ . Call this cline C. C intersects the unit circle at right angles.
Explain why
c. Consider the portion of cline C you constructed in part (b) that lies in
D. This is the unique hyperbolic line through p and q . Mark the ideal points
of this hyperbolic line.
3. Can two distinct hyperbolic lines be tangent at some point in D? Explain.
4. Suppose L is a hyperbolic line that is part of a circle C. Can the origin
of the complex plane be in the interior of C? Explain.
5. Constructing a hyperbolic circle centered at a point p through a point q .
Suppose p and q are two points in D, and that q is not on the line through
p and p∗ - the point symmetric to p with respect to the unit circle.
a. Find the center of the Euclidean circle through p, p∗ , and q. Call the
center point o.
b. Construct the segment oq .
c. Construct the perpendicular to oq at q. This perpendicular intersects
the Euclidean line through p and p∗ . Call the intersection point o0 .
0
d. Construct the Euclidean circle centered at o through q.
e. Prove that this circle is the hyperbolic circle through q centered at p.
5.3. Measurement in Hyperbolic Geometry 81
6. Explain why Theorem 5.2.6 applies in the general case, when z0 is not at
the origin.
7. Given a point and a hyperbolic line not passing through it, prove that
there is a hyperbolic line through the point that is perpendicular to the given
line. Is this perpendicular unique?
5.3 Measurement in Hyperbolic Geometry
In this section we develop a notion of distance in the hyperbolic plane. If
someone is standing at point p and wants to get to point q, he or she should
be able to say how far it is to get there, whatever the route taken.
The distance formula is derived following the approach given in Section 30
of Boas' text [2]. We rst list features our distance function ought to have,
and use the notation that dH (p, q) represents the hyperbolic distance from p
to q in the hyperbolic plane D.
1. The distance between 2 distinct points should be positive.
2. The shortest path between 2 points should be on the hyperbolic line
connecting them.
3. If p, q, and r are three points on a hyperbolic line with q between the
other two then dH (p, q) + dH (q, r) = dH (p, r).
4. Distance should be preserved by transformations in H. (A lunch pail
shouldn't shrink if it is moved to another table.) In other words, the
distance formula should satisfy
dH (p, q) = dH (T (p), T (q))
for any points p and q in D, and any transformation T in H.
5. In the limit for small distances, hyperbolic distance should be propor-
tional to Euclidean distance.
Perhaps the least obvious of the features listed is the last one. One theme
of this text is that locally, on small scales, non-Euclidean geometry behaves
much like Euclidean geometry. A small segment in the hyperbolic plane is
approximated to the rst order by a Euclidean segment. Small hyperbolic
triangles look like Euclidean triangles and hyperbolic angles correspond to
Euclidean angles; the hyperbolic distance formula will t with this theme.
To nd the distance function, start with a point's distance from the origin.
Given a point z in D, rotate about 0 so that z gets sent to the point x = |z|
on the positive real axis.
We may nd a hyperbolic line L about which x gets re ected to the origin.
Such a hyperbolic line is constructed in the proof of Theorem 5.1.3. Recall, the
line L is on the circle centered at x∗ (the point symmetric to x with respect
1 1
to to S∞ ) that goes through the points at which S∞ intersects the circle with
∗
diameter 0x . Let x+h be a point near x on the positive real axis, and suppose
x + h gets inverted to the point w, as depicted in Figure 5.3.1. One can show
(in Exercise 5.3.1) that
−h
w= .
1 − x2 − hx
82 5. Hyperbolic Geometry
w 0 x x+h
Figure 5.3.1: Re ection of the hyperbolic plane sending x to 0 and x+h to
w.
If distance is to be preserved by transformations in H,
dH (x, x + h) = dH (0, w). (1)
Also, 0, x, and x + h are all on the same hyperbolic line (the real axis), so
assuming h>0
dH (0, x) + dH (x, x + h) = dH (0, x + h). (2)
For x ∈ R de ne the function d(x) = dH (0, x) which is the hyperbolic
distance of x to the origin. Then (2) and (1) may be combined to give
d(x + h) − d(x) = d(w).
Divide both sides by h to get
d(x + h) − d(x) d(w)
= .
h h
As h→0 we obtain
d(w)
d0 (x) = lim .
h→0 h
We now interrupt this derivation with an important point. In the limit
for small w, the hyperbolic distance of w from 0, d(w), is proportional to the
Euclidean distance |w − 0| = |w|. Since w is the image of x + h under the
inversion and x gets inverted to 0, it follows that w → 0 as h → 0. So, we
assume that
d(w)
lim =k
h→0 |w|
for some constant k. Following convention, we set the constant of proportion-
ality to k = 2, as this makes length and area formulas look very nice later on.
Now, back to the derivation.
d(w)
d0 (x) = lim
h→0 h
d(w) |w|
= lim
h→0 |w| h
h
= lim 2 ·
h→0 (1 − x2 − hx)h
2
= .
1 − x2
5.3. Measurement in Hyperbolic Geometry 83
d(x) we integrate:
To get back to the distance function
Z Z
2 1 1
dx = + dx (partial fractions)
1 − x2 1−x 1+x
= − ln(1 − x) + ln(1 + x)
1+x
d(x) = ln ,
1−x
so we arrive at the following distance formula.
The hyperbolic distance from 0 to z .
The hyperbolic distance from 0 to a point D z in is
1 + |z|
dH (0, z) = ln .
1 − |z|
Notice that if z inches its way in D out toward the circle at in nity (i.e.,
|z| → 1), the hyperbolic distance from 0 to z approaches ∞. This is a good
thing. Thinking of Euclid's postulates, this notion of distance satis es one of
our fundamental requirements: One can produce a hyperbolic segment to any
nite length.
To arrive at a general distance formula dH (p, q), observe something curious.
The hyperbolic line through 0 and x has ideal points -1 and 1. Furthermore,
the expression (1 + x)/(1 − x) corresponds to the cross ratio of the points 0,
x, 1, and -1. In particular,
0−1 x+1 1+x
(0, x; 1, −1) = · = .
0+1 x−1 1−x
Thus,
dH (0, x) = ln((0, x; 1, −1)).
We can now derive a general distance formula, assuming the invariance of
distance under transformations in H. There is a transformation T in H that
takes p to the origin and q to some spot on the positive real-axis, call this spot
x (see Figure 5.3.2). Thus,
dH (p, q) = dH (T (p), T (q)) (invariance of distance)
= dH (0, x)
= ln((0, x; 1, −1))
= ln((p, q; u, v)), (invariance of cross ratio)
where u and v are the ideal points of the hyperbolic line through p and q.
To be precise, u is the ideal point you would head toward as you went from p
to q, and v is the ideal point you would head toward as you went from q to p.
84 5. Hyperbolic Geometry
v q
p T
−1 0 x 1
Figure 5.3.2: To nd the distance between p and q, we may rst transform
p to 0 and q to the positive real axis.
A working formula for dH (p, q)
One may compute the hyperbolic distance between p and q by rst nding
the ideal points u and v of the hyperbolic line through p and q and then
using the formula dH (p, q) = ln((p, q; u, v)). In practice, nding coordinates
for these ideal points can be a di cult task, and it is often simpler to compute
the distance between points by rst moving one of them to the origin. (This
simpler approach uses the fact that hyperbolic distance is preserved under
transformations in H. This fact will be proved shortly.)
One transformation in H that sends p to 0 has the form
z−p
T (z) = .
1 − pz
The map T sends q to some other point, T (q), in D. Assuming again that
T preserves distance, it follows that dH (p, q) = dH (0, T (q)), and
1 + |T (q)|
dH (p, q) = ln .
1 − |T (q)|
q−p
Making the substitution T (q) = provides us with the following
1 − pq
working formula for the hyperbolic distance between two points.
Theorem 5.3.3. Given two points p and q in D, the hyperbolic distance
between them is
|1 − pq| + |q − p|
dH (p, q) = ln .
|1 − pq| − |q − p|
Example 5.3.4: The distance between two points.
For instance, suppose p = 12 i, q =
1 1
2 + 2 i, z = .95e
i5π/6
and w = −.95.
Then dH (p, q) ≈ 1.49 units, while dH (z, w) ≈ 4.64 units.
5.3. Measurement in Hyperbolic Geometry 85
p q
z
1.49
4.64
w
The arc-length dierential
Now that we can compute the distance between two points in the hyperbolic
plane, we turn our attention to measuring the length of any path that takes us
from p to q.
De nition 5.3.5. A smooth curve is a dierentiable map from an interval
of real numbers to the plane
r : [a, b] → C
such that r 0 (t) exists for all t and never equals 0.
In the spirit of this text, we write r(t) = x(t) + iy(t), in which case
r 0 (t) = x0 (t) + iy 0 (t).
Recall that in calculus we rst approximate the Euclidean length of a given
smooth curve r(t) = x(t)+iy(t) by summing the contributions of small straight
line segments having Euclidean length
∆s = |r(t + ∆t) − r(t)|
p
= [x(t + ∆t) − x(t)]2 + [y(t + ∆t) − y(t)]2
s 2 2
x(t + ∆t) − x(t) y(t + ∆t) − y(t)
= + |∆t|.
∆t ∆t
r(t + ∆t)
r(t) ∆s
r(a)
r(b)
As ∆t → 0 we obtain the Euclidean arc-length dierential
p
ds = (dx/dt)2 + (dy/dt)2 dt,
which may be expressed as
ds = |r 0 (t)| dt.
For instance, we may compute the (Euclidean) circumference of a circle
with radius a as follows. Consider r : [0, 2π] → C by r(t) = a cos(t) + ia sin(t).
This map traces a circle of radius a centered at the origin. To nd the length
of this curve, which we denote as L(r), compute the integral
Z 2π
L(r) = |r 0 (t)| dt
0
86 5. Hyperbolic Geometry
Z 2π
= | − a sin(t) + ia cos(t)| dt
0
Z 2π q
= a2 sin2 (t) + a2 cos2 (t) dt
0
Z 2π
= a dt
0
= 2πa.
In the hyperbolic plane, we may deduce the arc-length dierential by a
similar argument. Suppose r is a smooth curve in D given by r(t) = x(t)+iy(t),
for a ≤ t ≤ b. One may approximate the length of a tiny portion of the curve,
say from r(t) to r(t+∆t), by the hyperbolic distance between these two points,
dH (r(t), r(t + ∆t)). To compute this distance, we rst send the point r(t) to
0 by the transformation
z − r(t)
T (z) = ,
1 − r(t)z
so that
dH (r(t), r(t + ∆t)) = ln[1 + |T (r(t + ∆t))|] − ln[1 − |T (r(t + ∆t))|].
To arrive at an arc-length dierential, we want to let ∆t approach 0. As
this happens, T (r(t + ∆t)) approaches T (r(t)), which is 0. From calculus we
also know that ln(1 + x) ≈ x for x very close to 0. Thus, for small ∆t, we have
dH (r(t), r(t + ∆t)) ≈ |T (r(t + ∆t))| + |T (r(t + ∆t))|
r(t + ∆t) − r(t)
= 2 ·
1 − r(t)r(t + ∆t)
r(t+∆t)−r(t)
∆t
=2· · |∆t|.
|1 − r(t)r(t + ∆t)|
Now, as ∆t → 0, the numerator in the above quotient goes to |r 0 (t)| and
2
the denominator goes to 1 − |r(t)| , and we arrive at the following hyperbolic
arc-length dierential.
De nition 5.3.6. If r : [a, b] → D is a smooth curve in the hyperbolic plane,
de ne the length of r, denoted L(r), to be
Z b
2
L(r) = |r 0 (t)|dt.
a 1 − |r(t)|2
One can immediately check that the hyperbolic distance between two points
in D corresponds to the length of the hyperbolic line segment connecting them.
Theorem 5.3.7. The arc-length de ned above is an invariant of hyperbolic
geometry. That is, if r is a smooth curve in D, and T is any transformation
in H, then L(r) = L(T (r)).
The proof of this theorem is left as an exercise. One can prove that
hyperbolic re ections preserve arc-length as well. This should come as no
surprise, given the construction of the distance formula at the start of this
section. Still, one can prove this fact from our de nition of arc-length
(Exercise 5.3.6). Thus, all hyperbolic re ections and all transformations in
H are hyperbolic isometries : they preserve the hyperbolic distance between
points in D.
Another consequence of the invariance of distance, when applied to
hyperbolic rotations, is the following:
5.3. Measurement in Hyperbolic Geometry 87
Corollary 5.3.8. All points on a hyperbolic circle centered at p are equidistant
from p.
Proof. Suppose u and v are on the same hyperbolic circle centered at p.
That is, these points are on the same type II cline with respect to p and p∗ ,
so there exists a hyperbolic rotation that xes p and maps u to v . Thus,
dH (p, u) = dH (T (p), T (u)) = dH (p, v). It follows that any two points on the
hyperbolic circle centered at p are equidistant from p.
We are now in a position to argue that in the hyperbolic plane, the shortest
path (geodesic) connecting two points p and q is along the hyperbolic line
through them.
Theorem 5.3.9. Hyperbolic lines are geodesics; that is, the shortest path
between two points in (D, H) is along the hyperbolic segment between them.
Proof Sketch : We rst argue that the geodesic from 0 to a point c on the
positive real axis is the real axis itself.
Suppose r(t) = x(t) + iy(t) for a ≤ t ≤ b, is an arbitrary smooth curve
from 0 to c (so r(a) = 0 and r(b) = c).
Suppose further that x(t) is nondecreasing (if our path backtracks in the x
direction, we claim the path cannot possibly be a geodesic). Then
Z b
2 p
L(r) = (x0 (t))2 + (y 0 (t))2 dt.
a 1 − [(x(t))2 + (y(t))2 ]
The hyperbolic line segment from 0 to c can be parameterized by r 0 (t) =
x(t) + 0i for a ≤ t ≤ b, which has length
Z b
2 p
L(r 0 ) = 2
(x0 (t))2 dt.
a 1 − [x(t)]
The curve r0 is essentially the shadow of r on the real axis.
One can compare the integrands directly to see that L(r) ≥ L(r 0 ).
Since transformations in H preserve arc-length and hyperbolic lines, it
follows that the shortest path between any two points in D is along the
hyperbolic line through them.
Corollary 5.3.10. The hyperbolic distance function is a metric on the
hyperbolic plane. In particular, for any points p, q, u in D
1. dH (p, q) ≥ 0, and dH (p, q) = 0 if and only if p = q ;
2. dH (p, q) = dH (q, p); and
3. dH (p, q) + dH (q, u) ≥ dH (p, u).
Proof. Recall our formula for the hyperbolic distance between two points
in Theorem 5.3.3:
|1 − pq| + |q − p|
dH (p, q) = ln .
|1 − pq| − |q − p|
This expression is always non-negative because the quotient inside the
natural log is always greater than or equal to 1. In fact, the expression equals
1 (so that the distance equals 0) if and only if p = q.
Note further that this formula is symmetric. Interchanging p and q leaves
the distance unchanged.
Finally, the hyperbolic distance formula satis es the triangle inequality
because hyperbolic lines are geodesics.
88 5. Hyperbolic Geometry
Example 5.3.11: Two paths from p to q .
Two paths from p = .5i to q = .5 + .5i are shown below: the (solid)
hyperbolic segment from p to q , and the (dashed) path r that looks
like a Euclidean segment. Which path is shorter?
We may compute the length of the hyperbolic segment connecting
p and q with the distance formula from Theorem 5.3.3. This distance
is approximately 1.49 units.
By contrast, consider the path in D corresponding to the Euclidean
line segment from p to q . This path may be described by r(t) = t + 21 i
for 0≤t≤ 1
2 . Then r 0 (t) = 1 and
Z 1
2 2
L(r) = dt
0 1 − (t2 + 14 )
≈ 1.52.
It is no surprise that the hyperbolic segment connecting p to q is
a shorter path in (D, H) than the Euclidean line segment connecting
them.
r
p q
Example 5.3.12: Perpendicular bisectors in (D, H).
For any two points p and q in D, we may construct the perpendicular
bisector to hyperbolic segment pq by following the construction in
Euclidean geometry. Construct both the hyperbolic circle centered at
p that goes through q and the hyperbolic circle centered at q that goes
through p. The hyperbolic line through the two points of intersection
of these circles is the perpendicular bisector to segment pq , labeled L
in the following diagram.
Hyperbolic re ection about L maps p to q and q to p. Since
hyperbolic re ections preserve hyperbolic distances, each point on L
is hyperbolic equidistant from p and q. That is, for each z on L,
dH (z, p) = dH (z, q).
5.3. Measurement in Hyperbolic Geometry 89
L q
In Euclidean geometry one uses perpendicular bisectors to construct the
circle through three noncollinear points. This construction can break down in
hyperbolic geometry. Consider the three points p, q, and r in Figure 5.3.13.
The corresponding perpendicular bisectors do not intersect. There is no point
in D hyperbolic equidistant from all three of these points. In particular,
in hyperbolic geometry, there need not be a hyperbolic circle through three
noncollinear points.
Lpq
q
Lqr p
Lpr r
Figure 5.3.13: Three noncollinear points need not determine a circle in
hyperbolic geometry.
Exercises
1. Suppose 0 < x < 1 and L is a hyperbolic line about which x gets inverted
to the origin. (Such an inversion was constructed in Theorem 5.1.3.) For a
real number h, let w be the image of x+h under this inversion. Prove that
−h
w= 1−x2 −hx .
2. Determine a point in D whose hyperbolic distance from the origin is
2,003,007.4 units.
3. Suppose L C is any cline through the ideal
is any hyperbolic line, and
points of L. For any point z
L, its perpendicular distance to C is the length
on
of the hyperbolic segment from z to C that meets C at right angles. Prove
that the perpendicular distance from C to L is the same at every point of L.
Hint: Use the fact that distance is an invariant of hyperbolic geometry.
4. Determine the hyperbolic distance from the point p = 0.5 to the point
q = 0.25 + 0.5i.
5. Prove Theorem 5.3.7.
90 5. Hyperbolic Geometry
6. Hyperbolic re ections preserve distance in (D, H)
a. Use the de nition of arc-length to prove that hyperbolic re ection about
the real axis preserves arc-length.
b. Use part (a) and Theorem 5.3.7 to argue that hyperbolic re ection about
any hyperbolic line preserves arc-length in (D, H).
7. Suppose z0 is in the hyperbolic plane and r > 0. Prove that the set C
consisting of all points z in D such that dH (z, z0 ) = r is a Euclidean circle.
5.4 Area and Triangle Trigonometry
The arc-length dierential determines an area dierential and the area of a
region will also be an invariant of hyperbolic geometry. The area of a region
will not change as it moves about the hyperbolic plane. We express the area
formula in terms of polar coordinates.
De nition 5.4.1. Suppose a region R in D is described in polar coordinates.
The area of R in (D, H), denoted A(R), is given by
ZZ
4r
A(R) = drdθ.
R (1 − r2 )2
The integral in this formula is di cult to evaluate directly in all but the
simplest cases. Following is one such case.
Example 5.4.2: The area of a circle in (D, H).
Suppose our region is given by a circle whose hyperbolic radius is a.
Since area is an invariant, we may as well assume the circle is centered
at the origin. Let x be the point at which the circle intersects the
positive real axis (so 0 < x < 1), as pictured below. Then, by the
distance formula
1+x
a = ln .
1−x
0 a x
Solving for x, we have
ea − 1
x= .
ea + 1
This circular region may be described in polar coordinates by
0 ≤ θ ≤ 2π and 0 ≤ r ≤ x. The area of the region is then given
by the following integral, which we compute with the u-substitution
u = 1 − r2 :
Z 2π Z x
4r
drdθ
0 0 (1 − r2 )2
5.4. Area and Triangle Trigonometry 91
Z 2π Z 1−x2
−2
= dθ du
0 1 u2
1−x2
2
= 2π
u
1
2
= 2π − 2
1 − x2
x2
= 4π .
1 − x2
Replace x in terms of a to obtain
(ea − 1)2 (ea + 1)2 (ea − 1)2
4π · = 4π
(ea + 1)2 (ea + 1)2 − (ea − 1)2 4ea
a 2
e −1
= 4π
2ea/2
2
− e−a/2
a/2
e
= 4π .
2
This last expression can be rewritten using the hyperbolic sine function,
evaluated at a/2. We investigate the hyperbolic sine and cosine functions in
the exercises but note their de nitions here.
De nition 5.4.3. The hyperbolic sine function , denoted sinh(x), and the
hyperbolic cosine function , denoted cosh(x), are functions of real numbers
de ned by
ex − e−x ex + e−x
sinh(x) = and cosh(x) = .
2 2
The area derivation in Example 5.4.2 may then be summarized as follows.
Theorem 5.4.4. The area of a hyperbolic circle with hyperbolic radius r is
4π sinh2 (r/2).
Other regions are not as simple to describe in polar coordinates. An
important area for us will be the area of a
2
3 -ideal triangle
, the gure that
results if two of the three vertices of a hyperbolic triangle are moved to ideal
points. See Figure 5.4.6.
Theorem 5.4.5. The area of a 32 -ideal triangle having interior angle α is equal
to π − α.
The proof of this theorem is given in the following section. The proof there
makes use of a dierent model for hyperbolic geometry, the so-called upper
half-plane model.
92 5. Hyperbolic Geometry
α
p
Figure 5.4.6: A
2
3 -ideal triangle having interior angle α has area equal to
π − α.
An ideal triangle consists of three ideal points and the three hyperbolic
lines connecting them. It turns out that all ideal triangles are congruent (a
fact proved in the exercises); the set of all ideal triangles is minimally invariant
in (D, H).
−1 1
Figure 5.4.7: All ideal triangles are congruent to this one.
Theorem 5.4.8. Any ideal triangle has area equal to π.
Proof. Since all ideal triangles are congruent, assume our triangle ∆ is the
ideal triangle shown in Figure 5.4.7.
2
But then ∆ can be partitioned into two
3 -ideal triangles by drawing the
vertical hyperbolic line from 0 along the imaginary axis to ideal point i. Each
2
3 -ideal triangle has interior angle π/2, so ∆ has area π/2 + π/2 = π .
It is a remarkable fact that π is an upper bound for the area of any triangle
in (D, H). No triangle in (D, H) can have area as large as π , even though side
lengths can be arbitrarily large!
Theorem 5.4.9. The area of a hyperbolic triangle in (D, H) having interior
angles α, β , and γ is
A = π − (α + β + γ).
5.4. Area and Triangle Trigonometry 93
q
R1
β
p α R3
γ
R2
r
t
Figure 5.4.10: Determining the area of a hyperbolic triangle.
Proof. Consider Figure 5.4.10 containing triangle ∆pqr. We have extended
segment qp to the ideal point t, u is an ideal point of line rq , and v is an ideal
point of line pr. The area of the ideal triangle ∆tuv is π . Notice that regions
2
R1 , R2 , and R3 are all
3 -ideal triangles contained within the ideal triangle.
Consider R1 , whose ideal points are u and t, and whose interior angle is ∠uqt.
Since the line through q and r has ideal point u, the interior angle of R1 is
∠uqt = π − β . Similarly, R2 has interior angle π − α and R3 has interior angle
π − γ.
Let R denote the triangle region ∆pqr whose area A(R) we want to
compute. We then have the following relationships among areas:
π = A(R) + A(R1 ) + A(R2 ) + A(R3 )
= A(R) + [π − (π − α)] + [π − (π − β)] + [π − (π − γ)].
Solving for A(R),
A(R) = π − (α + β + γ),
and this completes the proof.
In Euclidean geometry trigonometric formulas relate the angles of a triangle
to its side lengths. There are hyperbolic trigonometric formulas as well.
b
γ α
a
c
β
Figure 5.4.11: Relating angles and lengths in a hyperbolic triangle.
94 5. Hyperbolic Geometry
Theorem 5.4.12. Suppose a hyperbolic triangle in D has angles α, β and γ
and opposite hyperbolic side lengths a, b, c, as pictured in Figure 5.4.11. Then
the following laws hold.
a. First hyperbolic law of cosines.
cosh(c) = cosh(a) cosh(b) − sinh(a) sinh(b) cos(γ).
b. Second hyperbolic law of cosines.
cos(α) cos(β) + cos(γ)
cosh(c) = .
sin(α) sin(β)
c. Hyperbolic law of sines.
sinh(a) sinh(b) sinh(c)
= = .
sin(α) sin(β) sin(γ)
The proofs of these laws are left as exercises for the interested reader. The
following result follows from the rst hyperbolic law of cosines.
Corollary 5.4.13 Hyperbolic hypotenuse theorem. In a right hyperbolic
triangle with hyperbolic side lengths a and b, and hypotenuse c,
cosh(c) = cosh(a) cosh(b).
The second hyperbolic law of cosines also leads to an interesting result. In
the hyperbolic plane, if we nd ourselves at point z , we may infer our distance
c to a point w by estimating a certain angle, called the angle of parallelism of
z to the line L through w that is perpendicular to segment zw. The following
is a picture of this scene:
α w
z c
In this setting, ∆zwu is a 31 -ideal triangle, and the second hyperbolic law
of cosines applies with γ = 0 and β = π/2 to yield the following result.
Corollary 5.4.14. Suppose z and w are points in D, and L is a hyperbolic
line through w that is perpendicular to hyperbolic segment zw. Suppose further
that u is an ideal point of L. Let α = ∠wzu and c = dH (z, w). Then
1
cosh(c) = .
sin(α)
The angle of parallelism is pursued further in Section 7.4.
Example 5.4.15: Flying around in (D, H).
Suppose a two-dimensional ship is plopped down in D. What would the
pilot see? How would the ship move? How would the pilot describe the
world? Are all points equivalent in this world? Could the pilot gure
5.4. Area and Triangle Trigonometry 95
out whether the universe adheres to hyperbolic geometry as opposed
to, say, Euclidean geometry?
Recall what we know about hyperbolic geometry. First of all, any
two points in the hyperbolic plane are congruent, so the geometry is
homogeneous. The pilot could not distinguish between any two points.
Second, the shortest path between two points is the hyperbolic
line between them, so light would travel along these hyperbolic lines,
assuming light follows geodesics. The pilot's line of sight would follow
along these lines, and the ship would move along these lines to y as
quickly as possible from p to q , assuming no pesky asteroid elds block
the path. To observe a galaxy at point q from the point p (as in the
diagram below), the pilot would point a telescope in the direction of
line L, the line along which the light from the galaxy travels to reach
the telescope.
With a well-de ned metric, we can say more. The pilot will view
the hyperbolic plane as in nite and without boundary. In theory, the
pilot can make an orbit of arbitrary radius about an asteroid located
anywhere in the space.
To test for hyperbolic geometry, perhaps the pilot can turn to
triangles. The angles of a triangle in the hyperbolic plane sum to less
◦
than 180 , but only noticeably so for large enough triangles. In our
disk model of hyperbolic geometry, we can easily observe this angle
de ciency. In the gure below, triangle ∆zuw has angle sum about
130◦ , and ∆pqr has angle sum of about 22◦ ! Whether an intrepid 2-D
explorer could map out such a large triangle depends on how much
ground she could cover relative to the size of her universe. We will
have more to say about such things in Chapters 7 and 8.
p
u
z w
L
q
a q
xi
a p
a
0 a x
Figure 5.4.16: A journey that would trace a square in the Euclidean plane
does not get you home in (D, H).
96 5. Hyperbolic Geometry
Example 5.4.17: Hyperbolic squares?
Simply put, hyperbolic squares don't exist. In fact, no four-sided
gures with four right angles exist, if we assume the sides are hyperbolic
segments. If such a gure existed, its angle sum would be 2π . But such
a gure could be divided along a diagonal into two triangles whose
total angle sum must then be 2π as well. This means that one of the
triangles would have angle sum at least π, which cannot happen.
On the other hand, there is no physical obstruction to a two-
dimensional explorer making the following journey in the hyperbolic
plane: Starting at a point such as p in Figure 5.4.16, head along a line
◦
in a certain direction for a units, turn right (90 ) and proceed in a line
for a more units, then turn right again and proceed in a line a units,
and then turn right one more time and proceed in a line for a units.
Let q denote the point at which the explorer arrives at the end of this
journey. In the Euclidean plane, q will equal p, because the journey
traces out a square built from line segments. However, this is not the
case in (D, H) (though if we connect p and q with a hyperbolic line we
obtain a geodesic pentagon with (at least) three right angles!). In the
exercises we investigate the distance between p and q as a function of
the length a.
However, we can build a four-sided gure closely resembling a
rectangle, if we drop the requirement that the legs be hyperbolic line
segments.
Through any point 0 < a < 1 on the positive real axis, we may
construct a hyperbolic line L1 through a that is perpendicular to the
real axis. Also construct a hyperbolic line L2 through −a that is
perpendicular to the real axis. Now pick a point z on L1 and construct
the cline arc C1 through z, 1 and −1. Also construct the cline arc C2
through z, 1 and −1. This creates a four sided gure, which we call
a block . We claim that each angle in the gure is right, and that
opposite sides have equal length. Moreover, z and a can be chosen so
that all four sides have the same length. This gure isn't a rectangle,
however, in the sense that not all four sides are hyperbolic segments.
The arcs C1 and C2 are not hyperbolic lines.
L2 L1
z
C1
−1 1
−a a
C2
z
Figure 5.4.18: Blocks in D: Four-sided, right-angled gures whose opposite
sides have equal length.
While squares don't exist in the hyperbolic plane, we may build right-angled
regular polygons with more than four sides using hyperbolic line segments. In
5.4. Area and Triangle Trigonometry 97
fact, for each triple of positive real numbers (a, b, c) we may build a right-angled
hexagon in the hyperbolic plane with alternate side lengths a, b, and c. We
encourage the reader to work carefully through the construction of this hexagon
in the proof of Theorem 5.4.19. We use all our hyperbolic constructions to get
there.
Theorem 5.4.19. For any triple (a, b, c) of positive real numbers there exists a
right-angled hexagon in (D, H) with alternate side lengths a, b, and c. Moreover,
all right-angled hexagons with alternate side lengths a, b, and c are congruent.
Proof. We prove the existence of a right-angled hexagon with vertices
v0 , v1 , · · · , v5 such that dH (v0 , v1 ) = a, dH (v2 , v3 ) = b, and dH (v4 , v5 ) = c.
C
v4
v5
D c v3
b
v2
B
a
v0 v1
Figure 5.4.20: Building a right hexagon in the hyperbolic plane that has
alternate side lengths (a, b, c).
First, let v0 be the origin in the hyperbolic plane, and place v1 on the
positive real axis so that dH (v0 , v1 ) = a. Note that
ea − 1
v1 = .
ea + 1
Next, construct the hyperbolic line A perpendicular to the real axis at the
point v1 . This line is part of the cline that has diameter v1 v1∗ .
Pick any point v2 on the line A. For the sake of argument, assume that v2
lies above the real axis, as in Figure 5.4.20.
Next, construct the hyperbolic line B perpendicular to A at the point v2 .
This line is part of the cline through v2 and v2∗ with center on the line tangent
to A at v2 .
Next, construct the point v3 on line B that is a distance b away from v2 .
This point is found by intersecting B with the hyperbolic circle centered at v2
with radius b. (To construct this circle, we rst nd the scalar k so that the
hyperbolic distance between kv2 and v2 is b.)
Next, draw the perpendicular C to line B at v3 .
Then construct the common perpendicular of C and the imaginary axis,
call this perpendicular D. We construct this common perpendicular as follows.
First nd the two points p and q symmetric to both C and the imaginary axis.
∗ 1
Then nd p , the point symmetric to p with respect to S∞ . The cline through
∗
p, q , and p is perpendicular to C , the imaginary axis and S1∞ , so this gives us
our line D .
If C and the imaginary axis intersect, no such perpendicular exists (think
triangle angles), so drag v2 toward v1 until these lines do not intersect. Then
98 5. Hyperbolic Geometry
construct D as in the preceeding paragraph. Let v4 and v5 be the points of
intersection of D with C and the imaginary axis, respectively.
This construction gives us a right angled hexagon such that dH (v0 , v1 ) = a
and dH (v2 , v3 ) = b. dH (v4 , v5 ) = c. Notice that this last distance
We also want
is a function of the position of vertex v2 on line A. As v2 goes along A to v1
there is a point beyond which D no longer exists, and as v2 goes along A to the
circle at in nity, the length of segment v4 v5 takes on all positive real values. So,
by the intermediate value theorem, there is some point at which the segment
has length c. Finally, all right-angled hexagons with alternate side lengths a,
b, and c are congruent to the one just constructed because angles, hyperbolic
lines, and distances are preserved under transformations in H.
Example 5.4.21: Inscribe a circle in an ideal triangle.
We show that if one inscribes a circle in any ideal triangle, its points of
tangency form an equilateral triangle with side lengths equal to
√ 2 ln(ϕ)
where ϕ is the golden ratio (1 + 5)/2.
Since all ideal triangles are congruent, we choose one that is
convenient to work with. Consider the ideal triangle with ideal points
-1, 1, and i.
The hyperbolic lineL1 joining -1 and i is part of the circle C1 with
radius 1 centered at −1 + i. The hyperbolic line L2 joining i and 1 is
part of the circle C2 with radius 1 centered at 1 + i. Let C denote the
circle with radius 2 centered at −1 + 2i, as pictured below.
−1 + 2i
C
i
−1 + i 1+i
p q
L1 c L2
C1 −1 C2
1
0
Inversion about C gives a hyperbolic re ection of D that maps L1
onto the real axis. Indeed, the circle C1 , since it passes through the
center of C, gets mapped to a line - the real axis, in fact. Moreover,
hyperbolic re ection across the imaginary axis maps L1 onto L2 . Let c
be the point of intersection of these two hyperbolic lines of re ection,
as pictured. The hyperbolic circle with hyperbolic center c that passes
through the origin will be tangent to the real axis, and will thus inscribe
the ideal triangle. Let the points of tangency on L1 and L2 be p and
q, respectively.
The point q can be found analytically as the point of intersection
1
of circles C and C2 . Working it out, one nds q= 5 + 52 i. Thus,
1 + |q|
dH (0, q) = ln
1 − |q|
5.4. Area and Triangle Trigonometry 99
√1
1 +
5
= ln
1− √1
5
√
5+1
= ln √
5−1
√
( 5 + 1)2
= ln √ √
( 5 − 1)( 5 + 1)
√
( 5 + 1)2
= ln
4
= ln(ϕ2 )
= 2 ln(ϕ).
Exercises
1. Properties of sinh(x) = (ex − e−x )/2 and cosh(x) = (ex + e−x )/2.
a. Verify that sinh(0) = 0 and cosh(0) = 1.
d d
b. Verify that
dx [sinh(x)] = cosh(x) and dx [cosh(x)] = sinh(x).
2 2
c. Verify that cosh (x) − sinh (x) = 1.
d. Verify that the power series expansions for cosh(x) and sinh(x) are
x2 x4
cosh(x) = 1 + + + ···
2 4!
x3 x5
sinh(x) = x + + + ··· .
3! 5!
2. Prove that the circumference of a hyperbolic circle having hyperbolic
radius r is C = 2π sinh(r).
3. The hyperbolic plane looks Euclidean on small scales.
a. Prove
4π sinh2 (r/2)
lim+ = 1.
r→0 πr2
Thus, for small r, the Euclidean formula for the area of a circle is a good
approximation to the true area of a circle in the hyperbolic plane.
b. Prove
2π sinh(r)
lim+ = 1.
r→0 2πr
Thus, for small r, the Euclidean formula for the circumference of a circle
is a good approximation to the true circumference of a circle in the hyperbolic
plane.
4. Prove that all ideal triangles are congruent in hyperbolic geometry. Hint:
Prove any ideal triangle is congruent to the one whose ideal points are 1, i,
and -1 (see Figure 5.4.7).
5. An intrepid tax collector lives in a country in the hyperbolic plane. For
collection purposes, the country is divided into triangular grids. The collector
is responsible for collection in a triangle having angles 12◦ , 32◦ , and 17◦ . What
is the area of the collector's triangle? Can the entire space D be subdivided
into a nite number of triangles?
6. 1 1 1
2 , and 2 + 2 i.
Consider the hyperbolic triangle with vertices at 0,
Calculate the area of this triangle by determining the angle at each vertex.
Hint: To determine the angle at a vertex it may be convenient to move it to
the origin via an appropriate transformation in H.
100 5. Hyperbolic Geometry
7. Recall the block constructed in Example 5.4.17. Prove that all four angles
are 90◦ , and that for any choice of z, opposite sides have equal hyperbolic
length.
8. Find a formula for the area of an n-gon, comprised of n hyperbolic line
segments in terms of its n interior angles α1 , α2 , · · · , αn . Hint: Decompose the
n-gon into triangles.
9. Building a hyperbolic octagon with interior angles 45◦ .
π
Letr be a real
number such that 0 < r < 1. The eight points vk = rei 4 k for k = 0, 1, · · · , 7
determine a regular octagon in the hyperbolic plane, as shown in Figure 5.4.22.
Note that v0 is the real number r. The interior angle of each corner is a function
of r. We nd the value of r for which the interior angle is 45◦ .
a. Prove that the center of the circle containing the hyperbolic line through
v0 and v1 is
1 + r2 1 + r2
z0 = + tan(π/8)i.
2r 2r
Hint: The center z0 is on the perpendicular bisector of the Euclidean segment
v0 v0∗ v0 v1 .
and also on the perpendicular bisector of the Euclidean segment
2
∗
b. b = 1+r
Let
2r be the midpoint of segment v0 v0 and show that
∠v0 z0 b = π/8 precisely when the interior angles of the octagon equal π/4.
c. Using ∆v0 bz0 and part (b), show that the interior angles of the octagon
equal π/4 precisely when
1+r 2
|b − v0 | 2r −r
tan(π/8) = = 1+r 2
.
|z0 − b| 2r · tan(π/8)
d. Solve the equation in (c) for r to obtain r = (1/2)(1/4) .
v2
v1
v3 z0
v4
v0 b vo∗
v5 v7
v6
Figure 5.4.22: Building an octagon with interior angles equal to 45◦ .
10. Suppose we construct a regular
1 2
n-gon in the hyperbolic plane from
n−1
the corner points r, re n 2πi , re n 2πi , · · · , re n 2πi where 0 < r < 1. Calculate
the hyperbolic length of any of its sides.
11. Prove the rst hyperbolic law of cosines by completing the following
steps.
a. Show that for any positive real numbers x and y,
2 2
x +y x2 − y 2
cosh(ln(x/y)) = and sinh(ln(x/y)) = .
2xy 2xy
b. Given two points p and q in D, let c = dH (p, q). Use the hyperbolic
distance formula from Theorem 5.3.3 and part (a) to show
(1 + |p|2 )(1 − |q|2 ) − 4Re(pq)
cosh(c) = .
(1 − |p|2 )(1 − |q|2 )
5.4. Area and Triangle Trigonometry 101
c. Now suppose our triangle has one vertex at the origin, and one point on
the positive real axis. In particular, suppose p = r (0 < r < 1) and q = keiγ
(0 < k < 1), with angles α, β, γ and hyperbolic side lengths a, b, and c as in
Figure 5.4.23.
α c
b
γ
β
0 a p 1
Figure 5.4.23: A hyperbolic triangle with one corner at the origin and one
leg on the positive real axis.
Show
1 + r2 2r
cosh(a) = ; sinh(a) = ;
1 − r2 1 − r2
1 + k2 2k
cosh(b) = ; sinh(b) = .
1 − k2 1 − k2
d. Show that for the triangle in part (c),
cosh(c) = cosh(a) cosh(b) − sinh(a) sinh(b) cos(γ).
e. Explain why this formula works for any triangle in D.
12. In this exercise we prove the hyperbolic law of sines. We assume our
triangle is as in Figure 5.4.24. Thus, q = keiγ for some 0<k<1 and p=r
for some real number 0 < r < 1. Suppose further that the circle containing
side c has center z0 and Euclidean radius R, shown in the gure, and that mq
is the midpoint of segment qq ∗ and mp is the midpoint of segment pp∗ , so that
∆z0 mq q and ∆pmp z0 are right triangles.
q∗
mq
α z0
q
R
β
α
b R
c
γ β
0 a p mp p∗
Figure 5.4.24: Deriving the hyperbolic law of sines.
102 5. Hyperbolic Geometry
a. Verify that the triangle angles α and β correspond to angles ∠mq z0 q
and ∠pz0 mp , respectively.
b. Notice that sin(α) = |mq − q|/R and sin(β) = |mp − p|/R. Verify that
2
1−r 2
|mq − q| = 1/k−k
2 = 1−k
2k and that |mp − p| = 2r .
c. Verify that
sinh(a) sinh(b)
= .
sin(α) sin(β)
d. Explain why we may conclude that for any hyperbolic triangle in D,
sinh(a) sinh(b) sinh(c)
= = .
sin(α) sin(β) sin(γ)
13. Prove the second law of hyperbolic cosines. Hint: This result follows
from repeated applications of the rst law and judicial use of the two identities
cos2 (x) + sin2 (x) = 1 and cosh2 (x) − sinh2 (x) = 1.
14. Recall the journey in Example 5.4.17 in which a bug travels a path that
would trace a square in the Euclidean plane. For convenience, we assume the
starting point p is such that the rst right turn of 90◦ occurs at the point x on
the positive real axis and the second turn occurs at the origin. (This means
that x = (ea − 1)/(ea + 1).) The third corner must then occur at xi. We
have reproduced the journey with some more detail in Figure 5.4.25. In this
exercise we make use of hyperbolic triangle trig to measure some features of
this journey from p to q in terms of the length a of each leg.
a. Determine the hyperbolic distance between p and the origin (corner two
of the journey). In particular, show that dH (0, p) = cosh2 (a). Note that if
this journey had been done in the Euclidean plane, the corresponding distance
√ √
would be 2a. Is cosh2 (a) close to 2a for small positive values of a?
1
b. Let θ = ∠x0p. Show that tan(θ) =
cosh(a) . What is the corresponding
angle if this journey is done in the Euclidean plane? What does θ approach as
a → 0+ ?
c. Show that ∠x0p = ∠0px.
4
d. Show that dH (p, q) = cosh (a)[1 − sin(2θ)] + sin(2θ).
e. Let α = ∠qp0, b = dH (0, p) and c = dH (p, q) Show that
sinh(b)
sin(α) = cos(2θ).
sinh(c)
f. Determine the area of the pentagon enclosed by the journey if, after
reaching q we return to p
along the geodesic. In particular, show that the area
3π
of this pentagon equals
2 − 2(θ + α). What is the corresponding area if the
journey had been done in the Euclidean plane?
g. Would any of these measurements change if we began at a dierent point
in the hyperbolic plane and/or headed o in a dierent direction initially than
the ones in Figure 5.4.25?
5.5. The Upper Half-Plane Model 103
a q
xi
c
a α p
b
a
0
θ
a x
Figure 5.4.25: A journey that would trace a square in the Euclidean plane
does not get you home in (D, H). But how close will the nish point q be to
the starting point p?
5.5 The Upper Half-Plane Model
The Poincaré disk model is one way to represent hyperbolic geometry, and for
most purposes it serves us very well. However, another model, called the upper
half-plane model, makes some computations easier, including the calculation
of the area of a triangle.
De nition 5.5.1. The upper half-plane model of hyperbolic geometry
has space U consisting of all complex numbers z such that Im(z) > 0, and
transformation group U consisting of all Möbius transformations that send U
to itself. The space U is called the upper half-plane of C.
The Poincaré disk model of hyperbolic geometry may be transferred to the
upper half-plane model via a Möbius transformation built from two inversions
as follows:
1. Invert about the circle C centered at i passing through -1 and 1 as in
Figure 5.5.2.
2. Re ect about the real axis.
C
i
−1 1
Figure 5.5.2: Inversion in C maps the unit disk to the upper-half plane.
Notice that inversion about the circle C xes -1 and 1, and it takes i to
∞. Since re ection across the real axis leaves these image points xed, the
composition of the two inversions is a Möbius transformation that takes the
unit circle to the real axis. The map also sends the interior of the disk into
104 5. Hyperbolic Geometry
the upper half plane. Notice further that the Möbius transformation takes ∞
to −i; therefore, by Theorem 3.5.8, the map can be written as
−iz + 1
V (z) = .
z−i
This Möbius transformation is the key to transferring the disk model of
the hyperbolic plane to the upper half-plane model. In fact, when treading
back and forth between these models it is convenient to adopt the following
convention for this section: Let z denote a point in D, and w denote a point
in the upper half-plane U, as in Figure 5.5.3. We record the transformations
linking the spaces below.
Going between (D, H) and (U, U).
The Möbius transformation V mapping D to U, and its inverse V −1 , are
given by:
−iz + 1 iw + 1
w = V (z) = and z = V −1 (w) = .
z−i w+i
Some features of the upper half-plane model immediately come to light.
Since V is a Möbius transformation, it preserves clines and angles. This means
that the ideal points in the disk model, namely the points on the circle at
in nity, S1∞ , have moved to the real axis and that hyperbolic lines in the disk
model have become clines that intersect the real axis at right angles.
V
w
z
real axis
S1∞
V −1
Figure 5.5.3: Mapping the disk to the upper half-plane.
De ne the hyperbolic distance between two points w1 , w2 in the upper half-
plane model, denoted dU (w1 , w2 ), to be the hyperbolic distance between their
pre-images in the disk model.
Suppose w1 and w2 are two points in V whose pre-images in the unit disk
are z1 and z2 , respectively. Then,
dU (w1 , w2 ) = dH (z1 , z2 ) = ln((z1 , z2 ; u, v)),
where u and v are the ideal points of the hyperbolic line through z1 and z2 .
But, since the cross ratio is preserved under Möbius transformations,
dU (w1 , w2 ) = ln((w1 , w2 ; p, q)),
where p, q are the ideal points of the hyperbolic line in the upper half-plane
through w1 and w2 . In particular, going from w1 to w2 we're heading toward
ideal point p.
5.5. The Upper Half-Plane Model 105
Example 5.5.4: The distance between ri and si.
For r>s>0 we compute the distance between ri and si in the upper
half-plane model.
The hyperbolic line through ri and si is the positive imaginary axis,
having ideal points 0 and ∞. Thus,
dU (ri, si) = ln((ri, si; 0, ∞))
ri − 0 si − ∞
= ·
ri − ∞ si − 0
r
= ln .
s
Example 5.5.5: The distance between any two points.
To nd the distance between any two points w1 and w2 in U, we rst
build a map in the upper half-plane model that moves these two points
to the positive imaginary axis. To build this map, we work through
the Poincaré disk model.
By the transformationV −1 we send w1 and w2 back to D. We let
z−z1
z1 = V −1 (w1 ) z2 = V −1 (w2 ). Then, let S(z) = eiθ 1−z
and
1z
be the
transformation in (D, H) that sends z1 to 0 with θ chosen carefully so
that z2 gets sent to the positive imaginary axis. In fact, z2 gets sent
−1
to the point ki where k = |S(z2 )| = |S(V (w2 ))| (and 0 < k < 1).
Then, applying V to the situation, 0 gets sent to i and ki gets sent to
1+k −1 1+k
1−k i. Thus, V ◦ S ◦ V sends w1 to i and w2 to
1−k i, where by the
previous example the distance between the points is known:
dU (w1 , w2 ) = ln(1 + k) − ln(1 − k).
Describing k in terms of w1 and w2 is left for the adventurous reader.
We do not need to pursue that here.
We now derive the hyperbolic arc-length dierential for the upper half-
plane model working once again through the disk model. Recall the arc-length
dierential in the disk model is
2|dz|
ds = .
1 − |z|2
Since z = V −1 (w) = iw+1
w+i we may work out the arc-length dierential in
terms of dw. We will need to take the derivative of a complex expression,
which can be done just as if it were a real valued expression. Here we go:
2|dz|
ds =
1 − |z|2
iw+1
2|d w+i |
iw+1
= (z = w+i )
iw+1 2
1 − w+i
|iw + 1|2
2|i(w + i)dw − (iw + 1)dw|
= 1− (chain rule)
|w + i|2 |w + i|2
4|dw|
=
|w + i| − |iw + 1|2
2
106 5. Hyperbolic Geometry
4|dw|
=
(w + i)(w − i) − (iw + 1)(−iw + 1)
4|dw|
=
2i(w − w)
|dw|
= .
Im(w)
This leads us to the following de nition:
De nition 5.5.6. The length of a smooth curve r(t) for a ≤ t ≤ b in the
upper half-plane model (U, U), denoted L(r), is given by
b
|r 0 (t)|
Z
L(r) = dt
a Im(r(t))
Example 5.5.7: The length of a curve.
To nd the length of the horizontal curve r(t) = t + ki for a ≤ t ≤ b,
note that r 0 (t) = 1 and Im(r(t)) = k. Thus,
b
b−a
Z
1
L(r) = dt = .
a k k
dw
From the arc-length dierential ds = Im(w) comes the area dierential:
De nition 5.5.8. In the upper half-plane model (U, U) of hyperbolic geome-
try, the area of a region R described in cartesian coordinates, denoted A(R),
is given by ZZ
1
A(R) = dxdy.
R y2
Example 5.5.9: The area of a 23 -ideal triangle.
2
Suppose w∈U is on the unit circle, and consider the
3 -ideal triangle
1w∞ as pictured.
α
w
π−α
In particular, suppose the interior angle at w is α, so that w =
ei(π−α) where 0 < α < π.
2
The area of this
3 -ideal triangle is thus
Z 1 Z ∞
1
A= √ dydx
cos(π−α) 1−x2 y2
5.5. The Upper Half-Plane Model 107
Z 1
1
= √ dx.
cos(π−α) 1 − x2
√
With the trig substituion cos(θ) = x, so that 1 − x2 = sin(θ) and
− sin(θ)dθ = dx, the integral becomes
0
− sin(θ)
Z
= dθ
π−α sin(θ)
= π − α.
2
It turns out that any
3 -ideal triangle is congruent to one of the form 1w∞
where w is on the upper half of the unit circle (Exercise 5.5.3), and since our
transformations preserve angles and area, we have proved the area formula for
2
a
3 -ideal triangle.
Theorem 5.5.10. The area of a 23 -ideal triangle having interior angle α is
equal to π − α.
Exercises
1. What becomes of horocycles when we transfer the disk model of
hyperbolic geometry to the upper half-plane model?
2. What do hyperbolic rotations in the disk model look like over in the
upper half-plane model? What about hyperbolic translations?
3. Give an explicit description of a transformation that takes an arbitrary
2
-ideal triangle in the upper half-plane to one with ideal points 1 and ∞ and
3
an interior vertex on the upper half of the unit circle.
4. Determine the area of the triangular region pictured below. What is the
image of this triangle under V −1 in the disk model of hyperbolic geometry?
Why doesn't this result contradict Theorem 5.4.9?
1 + 2i
i 1+i
5. Another type of block. Consider the four-sided gure pqst in (D, H) shown
in the following diagram. This gure is determined by two horocycles C1 and
C2 , and two hyperbolic lines L1 and L2 all sharing the same ideal point. Note
that the lines are orthogonal to the horocycles, so that each angle in the four-
◦
sided gure is 90 .
a. By rotation about the origin if necessary, assume the common ideal point
is i and use the map V to transfer the gure to the upper half-plane. What
does the transferred gure look like in U? Answer parts (b)-(d) by using this
transferred version of the gure.
pq and st are equal.
b. Prove that the hyperbolic lengths of sides
c equal the hyperbolic length of the leg pt along the larger radius
c. Let
horocycle C1 , and let d equal the hyperbolic length of the leg sq on C2 . Show
x
that c = e d where x is the common length found in part (b).
d. Prove that the area of the four-sided gure is c − d.
108 5. Hyperbolic Geometry
6
Elliptic Geometry
Elliptic geometry is the second type of non-Euclidean geometry that might
describe the geometry of the universe. In this chapter we focus our attention
on two-dimensional elliptic geometry, and the sphere will be our guide. The
chapter begins with a review of stereographic projection, and how this map is
used to transfer information about the sphere onto the extended plane. We
develop elliptic geometry in Sections 6.2 and 6.3, and then pause our story
in Section 6.4 to re ect on what we have established, geometry-wise, before
moving on to geometry on surfaces in Chapter 7.
6.1 Antipodal Points
Recall, the unit 2-sphere S2 consists of all points (a, b, c) in R3 for which
2 2 2 2
a + b + c = 1, and S may be mapped onto the extended plane by the
stereographic projection map φ : S2 → C+ de ned by
(
a
+ b i if c 6= 1;
φ(a, b, c) = 1−c 1−c
∞ if c = 1.
Two distinct points on a sphere are called diametrically opposed points
if they are on the same line through the center of the sphere. Diametrically
opposed points on the sphere are also called antipodal points. If P = (a, b, c)
is on S2 then the point diametrically opposed to it is −P = (−a, −b, −c). It
turns out that φ maps diametrically opposed points of the sphere to points in
the extended plane that satisfy a particular equation.
De nition 6.1.1. Two points z and w in C+ are called antipodal points if
they satisfy the equation
z · w = −1.
Furthermore, we set 0 and ∞ to be antipodal points in C+ . If z and w are
antipodal points, we say w is antipodal to z, and vice versa.
Each z in C+ has a unique antipodal point, za , given as follows:
−1/z
if z 6= 0, ∞;
za = ∞ if z = 0;
z = ∞.
0 if
Since − z1 = − |z|1 2 z , we note that za is a scaled version of z, so that z and
za live on the same Euclidean line through the origin. Again, since 0 and ∞
are antipodal points, this notion extends to all points z ∈ C+ .
109
110 6. Elliptic Geometry
Lemma 6.1.2. Given two diametrically opposed points on the unit sphere,
their image points under stereographic projection are antipodal points in C+ .
Proof. First note that the north pole N = (0, 0, 1) and the south pole
S = (0, 0, −1) φ to ∞
are diametrically opposed points and they get sent by
C+ ; so the lemma holds in this case.
and 0, respectively, in
Now suppose P = (a, b, c) is a point on the sphere with |c| 6= 1, and
Q = (−a, −b, −c) is diametrically opposed to P . The images of these two
points under stereographic projection are
a b −a b
φ(P ) = + i and φ(Q) = − i.
1−c 1−c 1+c 1+c
If we expand the following product
a b −a b
φ(P ) · φ(Q) = + i · + i ,
1−c 1−c 1+c 1+c
we obtain
a2 + b2
φ(P ) · φ(Q) = −
1 − c2
which reduces to −1 since a2 + b2 + c2 = 1.
Thus, diametrically opposed points on the sphere get mapped via stereo-
graphic projection to antipodal points in the extended plane.
A cline C in C+ is called a great circle if whenever z is on C, so is its
antipodal point za . Some great circles are drawn in Figure 6.1.3. We note that
the unit circle is a great circle in C+ as is any line through the origin.
za
z 0 1
wa
Figure 6.1.3: Five great circles in C+ .
To construct a great circle in the extended plane, it is enough to ensure
that it passes through one particular pair of antipodal points. This can be
proved with the aid of stereographic projection, but an alternative proof is
given below, one that does not leave the plane, but rather uses a proposition
from Book III of Euclid's Elements.
Lemma 6.1.4. If a cline in C+ contains two antipodal points then it is a great
circle.
6.1. Antipodal Points 111
Proof. Suppose C C+ containing antipodal points p and pa ,
is a cline in
and suppose q is any other point on C . We show qa is also on C .
If C is a line, it must go through the origin since p and pa are on the same
line through the origin. Since q and qa are also on the same line through the
origin, if q is on C then qa is too.
If C is a circle, then the origin of the plane is in the interior of C and the
chord ppa contains the origin, as pictured in Figure 6.1.5. The line through q
and the origin intersects C at another point, say w . We show w = qa .
pa
0
p
q
Figure 6.1.5: A cline C containing a pair of antipodal points p and pa must
be a great cirlce. In the gure, w = qa .
The intersecting chords theorem (Book III, Proposition 35 of Euclid's
Elements ) applied to this gure tells us that |p| · |pa | = |q| · |w|. (We leave
the proof to Exercise 6.1.7. The proof essentially follows that of Lemma 3.2.7,
except we consider a point inside the given circle.) As antipodal points,
|p| · |pa | = 1, and it follows that |w| = 1/|q|. Since the segment wq contains
the origin, it follows that w is the point antipodal to q.
The following theorem tells us that inversions will play a central role in
elliptic geometry, just as they do in hyperbolic geometry.
Theorem 6.1.6. Re ection of S2 about a great circle corresponds via stereo-
graphic projection to inversion about a great circle in C+ .
We work through the relationship in one case, and refer the interested
reader to [10] for the general proof.
Example 6.1.7: Re ection of S2 about the equator.
We argue that re ection of S2 about the equator corresponds to
inversion about the unit circle.
First of all, stereographic projection sends the equator of S2 to the
+ 2
unit circle in C . Now, re ection of S across the equator sends the
∗
point P = (a, b, c) to the point P = (a, b, −c). We must argue that
∗
φ(P ) and φ(P ) are symmetric with respect to the unit circle.
∗
If |c| = 1 then P and P correspond to the north and south poles,
and their image points are 0 and ∞, and these points are symmetric
with respect to the unit circle. So, assume |c| =
6 1. Notice that
a b a b
φ(P ) = + i and φ(P ∗ ) = + i
1−c 1−c 1+c 1+c
112 6. Elliptic Geometry
are on the same ray beginning at the origin. Indeed, one is the positive
scalar multiple of the other:
1−c
φ(P ∗ ) = φ(P ).
1+c
Moreover,
1−c
|φ(P )| · |φ(P ∗ )| = · |φ(P )|2
1+c
1 − c a2 + b2
= ·
1 + c (1 − c)2
a2 + b2
=
1 − c2
= 1.
Again, the last equality holds because a2 + b2 + c2 = 1.
Thus, φ(P )
∗
and φ(P ) are symmetric with respect to the unit circle. It follows that
2
inversion in the unit circle, iS1 , corresponds to re ection of S across
the equator, call this map R, by the equation
iS1 ◦ φ = φ ◦ R.
We end the section with one more feature of the stereographic projection
map. The proof can be found in [10].
Theorem 6.1.8. The image of a circle on S2 via stereographic projection is a
cline in C+ . Moreover, the pre-image of a circle in C+ is a circle on S2 . The
pre-image of a line in C+ is a circle on S2 that goes through N = (0, 0, 1).
S2 can
In fact, one can oer a constructive proof of this theorem. A circle on
2
be represented as the intersection of S with a plane Ax+By +Cz +D = 0 in 3-
2
dimensional space. One can show that the circle in S de ned by Ax+By+Cz+
2 2
D = 0 gets mapped by φ to the cline (C +D)(u +v )+2Au+2Bv+(D−C) = 0
in the plane (described via u, v cartesian coordinates); conversely the circle
|w − w0 | = r in C gets mapped by the inverse function φ−1 to the circle in S2
2 2 2 2
given by the plane −2x0 x − 2yo y + (1 − |w0 | + r )z + (1 + |w0 | − r ) = 0
where w0 = x0 + y0 i.
Example 6.1.9: The image of a circle under φ.
Consider the circle on S2 x = − 12 . In
de ned by the vertical plane
standard form, this plane has constants A = 2, B = C = 0, and D = 1,
so the image under φ is the circle
(u2 + v 2 ) + 4u + 1 = 0.
Completing the square we obtain the circle
(u + 2)2 + v 2 = 3
√
having center (−2, 0) and radius 3.
Example 6.1.10: The pre-image of a circle under φ.
6.1. Antipodal Points 113
The pre-image of the circle |w − (3 + 2i)| = 4 in C is the circle on S2
de ned by the plane
−2 · 3x − 2 · 2y + (1 − 13 + 4)z + (1 + 13 − 4) = 0
or
−3x − 2y − 4z + 5 = 0.
Exercises
1. Constructing an antipodal point. Suppose z is a point inside the unit
circle. Prove that the following construction, which is depicted in Figure 6.1.11,
gives za , the point antipodal to z: (1) Draw the line through z and the origin;
(2) draw the line through the origin perpendicular to line (1), and let T be on
line (2) and the unit circle; (3) construct the segment zT ; (4) construct the
perpendicular to segment (3) at point T. Line (4) intersects line (1) at the
point za . Use similar triangles to prove that za = − |z|1 2 z.
(2)
(1)
0
(3)
(4)
za
T
Figure 6.1.11: Constructing the antipodal point to z.
2. Explain why any great circle in C+ either contains 0 or has 0 in its
interior.
3. Characterize those great circles in C+ that are actually Euclidean lines.
4. Prove that re ection of S2
across the great circle through (0,0,1), (0,0,-1)
and (1,0,0) corresponds via stereographic projection to re ection of C+ across
the real axis.
5. Determine the image under φ of the circle z = 1/2 on the unit sphere.
6. Explain why re ection of S2
across any longitudinal great circle (i.e. a
great circle through the north and south poles) corresponds to re ection of C+
across a line through the origin.
7. Prove the intersecting chords theorem in two parts.
a. Suppose C is a circle with radius r o. Suppose p is a point
centered at
inside C and a line through p intersects C at points m and n, as pictured
below. If we let s = |p − o| prove that |p − m| · |p − n| = r2 − s2 .
114 6. Elliptic Geometry
n
p
s r
m
o
b. Prove the intersecting chords theorem: If mn and ab are any two chords
of C passing through a given interior point p, then
|m − p| · |a − p| = |m − p| · |b − p|.
6.2 Elliptic Geometry
As was the case in hyperbolic geometry, the space in elliptic geometry is
derived from C+ , and the group of transformations consists of certain Möbius
transformations. We rst consider the transformations.
De nition 6.2.1. Let S consist of all Möbius transformations T that preserve
antipodal points. In other words, S consists of all Möbius transformations T
with the property that if z and za are antipodal points in C+ then T (z) and
T (za ) are antipodal points in C+ .
We leave it to the reader to verify that S actually forms a group.
Considering the antipodal point construction in Figure 6.1.11, it seems clear
that rotation of C+ about the origin should preserve antipodal points. So,
these rotations should be in S. Let's run through this.
Example 6.2.2: Rotations about 0 are in S .
Suppose R(z) = eiθ z is a rotation about the origin and z and za are
antipodal points. If z = ∞ or 0, then rotation about the origin xes z ,
and it also xes its antipodal point, so the map R preserves antipodal
1
points in these cases. Now suppose z 6= 0, ∞. Then za = − . Since
z
−iθ
R(za ) = − e z , it follows that
e−iθ
R(z) · R(za ) = −eiθ z · = −1.
z
Thus, the image points R(z) and R(za ) are still antipodal points.
Rotations about the origin belong to the group S.
Now we consider what a typical transformation in S looks like.
Suppose the Möbius transformation T is in S, and that z and w are
antipodal points. Then T (z) · T (w) = −1. Since z and w are antipodal points,
w = − z1 , so T (z) · T (−1/z̄) = −1, or
−1
T (z) = . (1)
T (−1/z̄)
6.2. Elliptic Geometry 115
Assume T (z) = (az + b)/(cz + d), and that it has determinant ad − bc = 1.
(Recall, from Exercise 3.4.4 that we can always write a Möbius transformation
in a form with determinant 1, and this form is unique up to sign.) Now,
−a/z̄ + b
T (−1/z̄) =
−c/z̄ + d
−ā/z + b̄
=
−c̄/z + d¯
b̄z − ā
= ¯ .
dz − c̄
Substituting into equation (1) of this derivation yields
az + b ¯ + c̄
−dz
= .
cz + d b̄z − ā
The transformation on the right also has determinant one, so these two
transformations are identical up to sign. We can assume that d = −ā and that
c = b̄, so that T may be expressed as follows:
az + b
T (z) = .
b̄z − ā
We can make this general form look a lot like the general form for a
transformation in H. To do so, rst multiply each term by −1/ā, assuming
a 6= 0. (If a = 0, what would the transformation look like?)
− āa z − b
ā
T (z) =
− āb̄ z + 1
− āa (z + ab )
=
− āb̄ z + 1
z − z0
= eiθ · ,
z0z + 1
where eiθ = −a/ā and z0 = −b/a. Thus, we have derived the following
algebraic description of transformations in S:
Transformations in S .
Any transformation T in the group S has the form
z − z0
T (z) = eiθ
1 + z0z
for some angle θ and some z0 in C+ .
One can show that the following converse holds: Any transformation having
the form above preserves antipodal points. It follows that for any point
z0 ∈ C+ , there exists a transformation T in S such that T (z0 ) = 0, and
since rotations about the origin live in S we can prove the following useful
result (see Exercise 6.2.1).
Lemma 6.2.3. Given distinct points z0 and z1 in C+ , there exists a
transformation in S that sends z0 to 0 and z1 to the real axis.
We now prove a theorem that helps us visualize the maps in S.
116 6. Elliptic Geometry
Theorem 6.2.4. If a Möbius transformation preserves antipodal points, then
it is an elliptic Möbius transformation.
Proof. Suppose T is a Möbius transformation that preserves antipodal
points. If T is not the identity map then it must x one or two points. However,
if T preserves antipodal points and xes a point p, then it must x its antipodal
point pa . Thus, T must have two xed points and T has normal form
T (z) − p z−p
=λ .
T (z) − pa z − pa
To show T is an elliptic Möbius transformation, we must show that |λ| = 1.
Setting z = 0 the normal form reduces to
T (0) − p p
=λ .
T (0) − pa pa
Solve for T (0) to obtain
ppa − λppa
T (0) = .
pa − λp
On the other hand, setting z=∞ and solving for T (∞), one checks that
p − λpa
T (∞) = .
1−λ
Since T preserves antipodal points, T (0) · T (∞) = −1; therefore, we have
ppa − λppa p − λpa
· = −1.
pa − λp 1−λ
If we expand this expression and solve it for λλ (using the fact that
p · pa = −1, and p 6= pa ), we obtain λλ = 1, from which it follows that
|λ| = 1. Thus T is an elliptic Möbius transformation.
In Exercise 3.5.8 we showed that any elliptic Möbius transformation is the
composition of two inversions about clines that intersect at the two xed points.
In the case of a Möbius transformation that preserves antipodal points, these
two xed points must be antipodal to each other. It follows by Lemma 6.1.4
that the two clines of inversion are great circles in C+ . Thus, we may view each
transformation of S as the composition of two inversions about great circles.
This is reassuring. By Theorem 6.1.6, these inversions correspond to re ections
of the sphere about great circles, and composing two of these re ections of
the sphere yields a rotation of the sphere. The transformation group S,
then, consists of Möbius transformations that correspond via stereographic
projection to rotations of the sphere. We summarize these facts below.
Theorem 6.2.5. Any transformation in S is the composition of two inversions
about great circles in C+ , and it corresponds via stereographic projection to a
rotation of the unit 2-sphere.
Before turning to the space in elliptic geometry, we make one more
comment about the group S. The reader may have noticed a strong similarity,
algebraically, between the maps in S and the maps in H, the transformation
group in hyperbolic geometry. Recall, transformations in H have the form
z − z0
T (z) = eiθ . (transformation in H)
1 − z0z
6.2. Elliptic Geometry 117
The single sign dierence between the algebraic desciption of maps in
S from the maps in H is not a coincidence. We could have de ned the
transformations in hyperbolic geometry as those Möbius transformations that
preserve symmetric points with respect to the unit circle. That is, we could
have de ned the group H to consist of all Möbius transformations satisfying
this property: If z · w̄ = 1 then T (z)·T (w) = 1. If we had then asked what such
a T would look like, we would have gone through the argument as we did in
this section, with a +1 initially instead of a -1. This sign dierence propagates
to the sign dierence in the nal form of the description of the map.
The space for elliptic geometry
One could let the space be all of C+ . If we take this tack, then we are
reproducing the geometry of the sphere in C+ . We call the geometry (C+ , S)
spherical geometry . Distance can be de ned to match distance on the unit
sphere, and great cirlces in C+ will be geodesics. Rather than develop these
details with this choice of spaces, we will focus instead on a dierent space.
We do so because we want to build a geometry in which there is a unique line
between any two points. This is not quite true on the sphere, and so not quite
true in (C+ , S). If two points in C+ are antipodal points, such as 0 and ∞,
then there are in nitely many lines (great circles) through these points.
So, we choose to work primarily with a space in which this feature of many
distinct lines through two points vanishes. The trick is to identify antipodal
points. That is, the space we will consider is actually the space C+ with
antipodal points identi ed. What does this space look like?
Remember the at torus from Chapter 1? Each point on the boundary of
the rectangle is identi ed with the corresponding point on the opposite edge.
The two points are fused together into a single point.
In elliptic space, every point gets fused together with another point, its
antipodal point. So, for instance, the point 2 + i gets identi ed with its
antipodal point
5
i
− 52 −
. In elliptic space, these points are one and the same.
With the at torus, we could visualize the space after identifying points by
wrapping it up in three-dimensional space. This was possible since we were
only identifying the edges of the rectangle. In the present case every point in
C+ gets paired up. To help visualize the space here we look at a region in the
plane that contains one representative from each pair.
Consider the closed unit disk, consisting of all complex numbers z such that
|z| ≤ 1. For each point w outside this disk, its antipodal point wa is inside
the disk. Thus, the closed unit disk contains a representative from each pair
of antipodal points. However, there is some redundancy: for a point w on the
boundary of the disk (|w| = 1), its antipodal point wa is also on the boundary.
To account for this redundancy, we identify each point on the boundary of the
closed unit disk with its antipodal point.
This will be our model for the space in elliptic geometry, and this space is
called the projective plane.
De nition 6.2.6. The projective plane , denoted P2 , consists of all complex
numbers z such that |z| ≤ 1 with the additional feature that antipodal points
on the unit circle are identi ed.
We can think of this space as the closed unit disk with its two edges (top-half
circle and bottom-half circle) identi ed according to the arrows in Figure 6.2.7.
Notice the pleasant journey a bug has taken from p to q in this gure. From
p she heads o toward point c, which appears on the boundary of our model.
When she arrives there she simply keeps walking, though in our model we see
118 6. Elliptic Geometry
her leave the screen and reappear at the antipodal point ca . She has her
sights set on point d and saunters down there, continues on (reappearing at
da ), and heads on to q, hungry but content.
da
ca
q
p
c
d
Figure 6.2.7: A leisurely stroll from p to q in the projective plane P2 .
De nition 6.2.8. The disk model for elliptic geometry , (P2 , S), is the
geometry whose space is P2 and whose group of transformations S consists of
all Möbius transformations that preserve antipodal points.
Because the transformations of S are generated by inversions about great
circles, these circles ought to determine the lines in elliptic geometry.
De nition 6.2.9. An elliptic line in (P2 , S) is the portion of a great circle
+
in C that lives in the closed unit disk.
Two elliptic lines have been constructed in Figure 6.2.10.
a
c
b d
Figure 6.2.10: Elliptic lines go through antipodal points.
Theorem 6.2.11. There is a unique elliptic line connecting two points p and
q in P2 .
Proof. Suppose p and q are distinct points in P2 . This means q 6= pa
+
as points in C . Construct the antipodal point pa , which gives us three
distinct points in C+ : p, q and pa . There exists a unique cline through these
three points. Since this cline goes through p and pa , it is an elliptic line by
Lemma 6.1.4.
Note that elliptic lines through the origin are Euclidean lines, just as was the
case in the Poincaré model of hyperbolic geometry. As a result, to prove facts
about elliptic geometry, it can be convenient to transform a general picture to
the special case where the origin is involved.
6.2. Elliptic Geometry 119
Theorem 6.2.12. The set of elliptic lines is a minimally invariant set of
elliptic geometry.
Proof. By de nition, any transformation T in S preserves antipodal points.
Thus, if L is an elliptic line, then T (L) is as well, and the set of elliptic lines
is an invariant set of elliptic geometry.
To show the set is minimally invariant, we appeal to Theorem 4.1.10, and
prove that any two elliptic lines are congruent. To see this, notice that any
elliptic line L is congruent to the elliptic line on the real axis. Indeed, for any
points z0 and z1 on L, Lemma 6.2.3 ensures the existence of a transformation
T in S that sends z0 to the origin, and z1 to the real axis. It follows that T (L)
is the real axis. Since all elliptic lines are congruent to the real axis, any two
elliptic lines are congruent.
Theorem 6.2.13. Any two elliptic lines intersect in P2 .
Proof. Given any two elliptic lines, apply a transformation T in S that
sends one of them to the real axis. It is enough to prove that any elliptic line
in P2 must intersect the real axis. Suppose M is an arbitrary elliptic line in
2
P and z is a point on M. If Im(z) =0 then z is on the real axis and we are
done.
If Im(z) > 0, then z lies above the real axis. It follows that Im(za ) <0
by the de nition of the antipodal point za . Since M contains both z and za it
intersects the real axis at some point.
If Im(z) < 0, then Im(za ) > 0 and M must intersect the real axis as before.
In either case, M must intersect the real axis, and it follows that any two
elliptic lines must intersect.
As an immediate consequence of this theorem, there is no notion of parallel
lines in elliptic geometry.
Corollary 6.2.14. If p in P2 is not on the elliptic line L, then every elliptic
line through p intersects L.
Exercises
1. The transformation group in elliptic geometry.
a. Prove that S is a group of transformations.
b. For each θ ∈ R and z0 ∈ C, prove that the following Möbius
transformation is in S :
z − z0
T (z) = eiθ .
1 + z0 z
c. For each θ ∈ R, prove that T (z) = eiθ z1 is in S.
d. Use (b) and (c) to prove that for any distinct points p, q ∈ C+ there
exists a transformation in S that sends p to 0 and q to a point on the positive
real axis, thus proving Lemma 6.2.3.
2. Prove that the disk model for elliptic geometry is homogeneous.
3. Given a point z not on an elliptic line L, prove there exists an elliptic
line through z that intersects L at right angles. Hint: First transform z to a
convenient location in P2 .
4. Is there a nonidentity transformation in S that xes two distinct points
in P2 ? If so, nd one, otherwise explain why no such transformation exists.
5. Find a transformation in S that sends the point
1 1
2 to the point 2 + 12 i.
120 6. Elliptic Geometry
6.3 Measurement in Elliptic Geometry
Rather than derive the arc-length formula here as we did for hyperbolic
geometry, we state the following de nition and note the single sign dierence
from the hyperbolic case. This sign dierence is consistent with the sign
dierence in the algebraic descriptions of the transformations in the respective
geometries.
De nition 6.3.1. If r : [a, b] → P2 is a smooth curve in (P2 , S), the length
of r, denoted L(r), is given by
b
2|r 0 (t)|
Z
L(r) = dt.
a 1 + |r(t)|2
The area of a gure R described in polar coordinates in the elliptic space
P2 , denoted A(R), is given by
ZZ
4r
A(R) = drdθ.
R (1 + r2 )2
Theorem 6.3.2. Arc-length is an invariant of elliptic geometry.
The proof of this theorem is left as an exercise, and is essentially the same
as the proof that hyperbolic arc-length is an invariant of hyperbolic geometry,
from which it follows that area is invariant. One can also prove that the
shortest path between two points is along the elliptic line between them. That
is, geodesics follows elliptic lines.
However, one must be careful when measuring the shortest path between
points in the projective plane. If p and q are two points in P2 then there
is exactly one elliptic line through them both, but we may view this line as
consisting of two elliptic segments, both of which connect p to q. That is, if a
bug nds herself at point p and wants to walk along an elliptic line to point
q, she can do so by proceeding in either direction along the line, as shown in
Figure 6.3.3. The elliptic distance between p and q, which we denote by
dS (p, q), is then the minimum value of the two segment lengths.
(1)
p
(2)
Figure 6.3.3: There is a single elliptic line joining points p and q, but two
elliptic line segments. The distance from p to q is the shorter of these two
segments.
Theorem 6.3.4. The distance between two points p and q in (P2 , S) is
q−p 1 + pq
dS (p, q) = min 2 arctan
, 2 arctan
.
1 + pq q−p
6.3. Measurement in Elliptic Geometry 121
Proof. We rst determine the elliptic distance between the origin and a
point x (with 0 < x ≤ 1) on the positive real axis.
The elliptic line through 0 and x lives on the real axis, and we may
parameterize the eastbound segment connecting 0 to x by r(t) = t for
0 ≤ t ≤ x. (The westbound segment from 0 to x is clearly not shorter
than the eastbound segment.) The length of this segment is
x x
2|r 0 (t)|
Z Z
2|1|
dt = dt
0 1 + |r(t)|2 0 1 + |t|
2
Z x
1
=2 dt
0 1 + t2
= 2 arctan(x).
Thus,dS (0, x) = 2 arctan(x), for 0 < x ≤ 1.
To determine the distance dS (p, q) between arbitrary points in P2 , we may
rst apply a transformation in S that sends p to the origin, and q to a point
on the positive real axis. By an appropriate choice of θ , the transformation
z−p
T (z) = eiθ
1 + pz
will do the trick. Now, T is a Möbius transformation of the entire extended
plane, and it may take q outside the unit disk, in which case qa will be
mapped to a point on the real axis inside the unit disk. So, either |T (q)|
or |T (qa )| will be a real number between 0 and 1. If we call this number x,
then dS (p, q) = dS (0, x), since transformations in S preserve distance between
points.
Now,
q−p
|T (q)| = ,
1 + pq
and the reader can check, using the fact that qa = −1/q , that
1 + pq
|T (qa )| =
.
q−p
It follows that
q−p 1 + pq
dS (p, q) = min 2 arctan
, 2 arctan
.
1 + pq q−p
This completes the proof.
With the sphere as our model, we can check our formulas against
measurements on the sphere. For instance, dS (0, 1) = 2 arctan(1) = π/2. The
elliptic segment from 0 to 1 corresponds via stereographic projection to one-
quarter of a great circle on the unit sphere. Any great circle on the unit sphere
has circumference 2π , π/2 on the
so one-quarter of a great circle has length
sphere. We also note that the distance formula dS (0, x) = 2 arctan(x) applies
+
to spherical geometry (C , S) for all positive real numbers x, and this distance
matches the corresponding distances of the points on the unit 2-sphere, see
Exercise 6.3.12.
We emphasize that π/2 is an upper bound for the distance between two
points in (P2 , S). However, there is no upper bound on how long a journey
along an elliptic line can be. If Bormit the bug wants to head out from point
p and travel r units along any line, Bormit can do it, without obstruction, for
any r > 0. Of course, if r is large enough, Bormit will do laps on this journey.
We say a path is a geodesic path if it follows along an elliptic line.
122 6. Elliptic Geometry
De nition 6.3.5. In(P2 , S), the elliptic circle centered at z0 with radius r
consists of all points z ∈ P2 such that there exists a geodesic path of length r
from z0 to z.
Each transformation T in S is an elliptic Möbius transformation by
Theorem 6.2.4 that xes two antipodal points, say p and pa . So T pushes
points along type II clines of p and pa , and since transformations preserve
distance between points, these type II clines of p and pa determine elliptic
circles; all points on these type II clines are equidistant from p.
Now, suppose p and q are any distinct points in P2 . There exists a type II
cline of p and pa that goes through q. If this cline lives entirely inside the closed
unit disk it represents the elliptic circle centered at p through q. Of course,
this cline may not live entirely inside the disk, as is the case in Figure 6.3.6.
But each point on the type II cline of p and pa through q has antipodal point
on the type II cline of p and pa through qa . So, in P2 we may represent the
elliptic circle centered at p through q by the portions of these two type II clines
of p and pa that live in the closed unit disk.
q
p
pa
qa
Figure 6.3.6: The elliptic circle centered at p through q in P2 may consist of
portions of two distinct clines.
We note also that elliptic circles centered at the origin correpond to
Euclidean circles centered at the origin. In particular, a Euclidean circle
centered at the origin with Euclidean radius a (with 0 < a < 1) corresponds
to an elliptic circle centered at the origin with elliptic radius 2 arctan(a).
Theorem 6.3.7. An elliptic circle in P2 with elliptic radius r < π/2 has
circumference C = 2π sin(r).
The proof of this theorem is left as an exercise. Circles with elliptic radius
greater than or equal to π/2 are also investigated in the exercises. They may
not look like circles!
The area of a triangle We now turn our attention to nding a formula for
the area of a triangle in elliptic geometry. We begin with lunes. A lune is
the region in P2 bounded between two elliptic lines. How do two elliptic lines
bound a region? Two lines trap a region because we are identifying antipodal
points on the unit circle. A bug living in the shaded region of P2 pictured
in Figure 6.3.8 would be able to visit all shaded points without crossing the
boundary walls determined by the two elliptic lines. The shaded region is a
single, connected region bounded by two lines. So what is the area of this
region?
6.3. Measurement in Elliptic Geometry 123
Figure 6.3.8: A lune in P2 with angle α.
Lemma 6.3.9 The area of a lune. In (P2 , S), the area of a lune with angle
α is 2α.
Proof. To compute the area of a lune, rst move the vertex of the lune
to the origin in such a way that one leg of the lune lies on the real axis,
as in Figure 6.3.10. Then half of the lunar region can be described in polar
coordinates by 0≤r≤1 and 0 ≤ θ ≤ α.
Figure 6.3.10: A lune whose elliptic lines intersect at the origin.
So the area of a lune having angle α is given by
Z α Z 1
4r
A=2 drdθ (Let u = 1 + r2 )
0 0 (1 + r2 )2
Z α 1
−2
=2 dθ
0 1 + r 2 0
Z α
=2 1 dθ
0
= 2α.
Example 6.3.11: Triangle area in (P2 , S).
Triangle ∆pqr below is formed from 3 elliptic lines. Notice that each
corner of the triangle determines a lune, and that the three lunes cover
the entire projective plane, with some overlap. In particular, the three
lunes in sum cover the triangle three times, so the sum of the three
lune areas equals the area of the entire projective plane plus two times
the area of the triangle.
124 6. Elliptic Geometry
p
α
β q
r γ
The area of the entire projective plane is 2π (see Exercise 6.3.9), so
we have the following relation:
2π + 2 · A(∆pqr) = A(Lune p) + A(Lune q) + A(Lune r)
2π + 2 · A(∆pqr) = (2α + 2β + 2γ)
A(∆pqr) = (α + β + γ) − π.
We summarize the result of this example with the following theorem.
Theorem 6.3.12. In elliptic geometry (P2 , S), the area of a triangle with
angles α, β, γ is
A = (α + β + γ) − π.
From this theorem it follows that the angles of any triangle in elliptic
◦
geometry sum to more than 180 .
We close this section with a discussion of trigonometry in elliptic geometry.
We derive formulas analogous to those in Theorem 5.4.12 for hyperbolic
triangles. We assume here that the triangle determined by distinct points p, q
and z in (P2 , S) is formed by considering the shortest paths connecting these
three points. So triangle side lenghths will not exceed π/2 in what follows.
q
α c
b
γ β
r
0 a
Figure 6.3.13: An elliptic triangle with side lengths and angles marked.
Theorem 6.3.14. Suppose we have an elliptic triangle with elliptic side
lengths a, b, and c, and angles α, β, and γ as in Figure 6.3.13. Then the
following laws hold:
a. Elliptic law of cosines
cos(c) = cos(a) cos(b) + sin(a) sin(b) cos(γ).
b. Elliptic law of sines
sin(a) sin(b) sin(c)
= = .
sin(α) sin(β) sin(γ)
6.3. Measurement in Elliptic Geometry 125
Proof. a. Position our triangle conveniently, with one corner at the origin,
one on the positive real axis at the point r (0 < r ≤ 1), and one at the point
q = keiγ (with 0 < k ≤ 1) as in Figure 6.3.13. Then a = dS (0, r) = 2 arctan(r),
so that
cos(a) = cos(2 arctan(r))
= cos2 (arctan(r)) − sin2 (arctan(r))
by the cosine double angle formula. If we set θ = arctan(r) we may use the
following right triangle to rewrite the above description of cos(a) as follows:
1 r2 2
r
cos(a) = − √ 1+
1 + r2 1 + r2 r
2
1−r
= .
1 + r2 θ
1
On the other hand,
sin(a) = sin(2 arctan(r))
= 2 sin(arctan(r)) cos(arctan(r)),
from which it follows that
2r
sin(a) = .
1 + r2
Turning our attention to the side with length b, b = dS (0, q) = 2 arctan(|q|) =
2 arctan(k), since q = keiγ with k > 0. It follows that
1 − k2 2k
cos(b) = and sin(b) = .
1 + k2 1 + k2
And the third side? Theorem 6.3.4 tells us
iγ
iγ
ke − r
c = dS (r, ke ) = 2 arctan
.
1 + rkeiγ
Since in general cos(2 arctan(x)) = (1 − x2 )/(1 + x2 ), it follows that
|1 + rkeiγ |2 − |keiγ − r|2
cos(c) = ,
|1 + rkeiγ |2 + |keiγ − r|2
which can be expanded using |z|2 = z · z , and then reduced to obtain
1 + r2 k 2 − k 2 − r2 + 2rk(eiγ + e−iγ )
cos(c) = .
1 + r2 + k2 + r2 k2
Now, eiγ + e−iγ = 2 cos(γ), so we have
(1 − r2 )(1 − k 2 ) + 2rk2 cos(γ)
cos(c) =
(1 + r2 )(1 + k 2 )
1 − r2 1 − k2 2r 2k
= · + · cos(γ)
1 + r2 1 + k2 1 + r2 1 + k2
= cos(a) cos(b) + sin(a) sin(b) cos(γ).
126 6. Elliptic Geometry
Though we worked out the formula for a conveniently located triangle,
it holds for any triangle in elliptic geometry because angles and distances are
preserved by transformations in elliptic geometry, and there is a transformation
that takes any triangle to this convenient location.
b. To prove the elliptic law of sines, rst construct the circle containing
side c. This circle goes through r and q and their antipodal points −1/r and
qa as pictured in Figure 6.3.15. We let p denote the center of the circle, and R
its Euclidean radius.
q
α
R
1
mr 0
γ β
-
r
α r
β R
p mq
qa
Figure 6.3.15: Deriving the elliptic law of sines.
Let mr be the midpoint of the segment connecting r and −1/r, and let mq
be the midpoint of the segment connecting q and qa . So,
1 + r2
1 1
|mr − r| = r+ = ,
2 r 2r
and
1 iγ 1 + k 2
1
|mq − q| = k +
e = .
2 k 2k
∆pmr r is right, and ∠mr rp = π/2−β so that ∠rpmr = β .
Note further that
From this right triangle we seesin(β) = (1 + r2 )/(2rR).
Similarly, ∆pmq q is right, and we have ∠mq pq = α, and sin(α) =
(1 + k 2 )/(2kR).
Comparing ratios,
sin(a) 2r 2kR
= ·
sin(α) 1 + r2 1 + k2
2k 2rR
= ·
1 + k 1 + r2
2
sin(b)
= .
sin(β)
To see that the ratio sin(c)/ sin(γ) must match the preceding common ratio,
note that transformations in elliptic geometry preserve distances and angles,
so we may transform our triangle above to one in which the length c is now
on the real axis with one end at the origin. The argument above ensures that
the ratio sin(c)/ sin(γ) then matches one of the other ratios, and so all three
agree.
6.3. Measurement in Elliptic Geometry 127
Corollary 6.3.16 Elliptic hypotenuse theorem. In a right triangle in
(P2 , S) with elliptic side lengths a and b, and hypotenuse c,
cos(c) = cos(a) cos(b).
Exercises
1. Prove Theorem 6.3.7. Namely, prove that the circumference of a circle in
elliptic geometry is C = 2π sin(r), where r < π/2 is the elliptic radius. Hint:
Assume your circle is centered at the origin.
2. Prove
2π sin(r)
lim = 1.
r→0+ 2πr
Thus, for small r, the Euclidean formula for the circumference of a circle is
a good approximation to the true circumference of a circle in elliptic geometry.
3. Circles with large radius.
a. Give a careful sketch of the circle centered at 0 with elliptic radius π/2.
b. Give a careful sketch of the circle centered at 0 with elliptic radius 2π/3.
c. Give a careful sketch of the circle centered at 0 with elliptic radius π .
d. Give a careful sketch of the circle centered at 0 with elliptic radius 2π .
4. Let C be an elliptic circle with center z0 and elliptic radius r > 0. For
what value(s) of r is C an elliptic line? For what value(s) of r is C a single
point?
5. Determine the elliptic distance between
1 1
2 and 2 i.
6. Prove that the area of a circle in elliptic geometry with radius r < π/2 is
A = 4π sin2 (r/2).
7. Prove
4π sin2 (r/2)
lim = 1.
r→0+ πr2
Thus, for small r, the Euclidean formula for the area of a circle is a good
approximation to the true area of a circle in elliptic geometry.
8. Prove that arc-length is an invariant of elliptic geometry.
9. Prove that the area of P2 is 2π . Thus, unlike the hyperbolic case, the
space in elliptic geometry has nite area.
10. An intrepid tax collector lives in a country in the elliptic space P2 . For
collection purposes, the country is divided into triangular grids. The collector
observes that the angles of the triangle she collects in are 92◦ , 62◦ , and 27◦ .
What is the area of her triangle? (Be sure to convert the angles to radians.)
Can the entire space P2 be subdivided into a nite number of triangles?
11. Find a formula for the area of an n-gon in elliptic geometry (P2 , S),
given that its n angles are α1 , α2 , · · · , αn .
12. In this exercise we show the distance between any two points p and q in
spherical geometry (C+ , S) is
q−p
dS (p, q) = 2 arctan ,
1 + pq
128 6. Elliptic Geometry
and that dS (p, q) corresponds to the distance on the sphere between φ−1 (p)
−1
and φ (q).
a. The de nition of arc-length in spherical geometry (C+ , S) is the same
2
as the one for (P , S). Using this de nition, follow the proof of Theorem 6.3.4
to show that for any positive real number C+ , dS (0, x) = 2 arctan(x).
x in
+
b. Use invariance of arc-length to explain why, for arbitrary p and q in C ,
q−p
dS (p, q) = 2 arctan .
1 + pq
c. Suppose x>0 is a real number. Determine φ−1 (x), the point on the
unit sphere corresponding to x via stereographic projection. What is φ−1 (0)?
−1 −1
d. Determine the distance between φ (0) and φ (x) on the sphere. In
particular, show that this distance equals arccos((1 − x2 )/(1 + x2 )). Hint: the
distance between these points will equal the angle between the vectors to these
points, and this angle can be found using the formula cos(θ) = ~v · w
~ for two
unit vectors.
e. Show that for x > 0, arccos((1 − x2 )/(1 + x2 )) = 2 arctan(x). Hint: You
may nd the following half-angle formula useful: tan(θ/2) = tan(θ)/(sec(θ) +
1).
13. Prove that (P2 , S) is isotropic.
6.4 Revisiting Euclid's Postulates
Without much fanfare, we have shown that the geometry (P2 , S) satis es the
rst four of Euclid's postulates, but fails to satisfy the fth. This is also the
case with hyperbolic geometry (D, H). Moreover, the elliptic version of the
fth postulate diers from the hyperbolic version. It is the purpose of this
section to provide the proper fanfare for these facts.
Recall Euclid's ve postulates:
1. One can draw a straight line from any point to any point.
2. One can produce a nite straight line continuously in a straight line.
3. One can describe a circle with any center and radius
4. All right angles equal one another.
5. Given a line and a point not on the line, there is exactly one line through
the point that does not intersect the given line.
That the rst postulate is satis ed in (P2 , S) is Theorem 6.2.11.
The second postulate holds here even though our elliptic space is nite. We
can extend line segments inde nitely because the space has no boundary. If we
are at a point in the space, and decide to head o in a certain direction along
an elliptic straight line, we can walk for as long and far as we want (though we
would eventually return to our starting point and continue making laps along
the elliptic line).
The third postulate follows by a similar argument. In the previous section
we de ned a circle about any point in P2 . And since we can walk an arbitrarily
6.4. Revisiting Euclid's Postulates 129
long distance from any point, we can describe a circle of any radius about the
point.
The fourth postulate follows since Möbius transformations preserve angles
and the maps in S are special Möbius transformations.
The fth postulate fails because any two elliptic lines intersect (Theo-
rem 6.2.13). Thus, given a line and a point not on the line, there is not a
single line through the point that does not intersect the given line.
Recall that in our model of hyperbolic geometry, (D, H), we proved that
given a line and a point not on the line, there are two lines through the point
that do not intersect the given line.
So we have three dierent, equally valid geometries that share Euclid's
rst four postulates, but each has its own parallel postulate. Furthermore,
on a small scale, the three geometries all behave similarly. A tiny bug living
on the surface of a sphere might reasonably suspect Euclid's fth postulate
holds, given his limited perspective. A tiny bug in the hyperbolic plane would
reasonably conclude the same. Small triangles have angles adding up nearly
◦
to 180 , and small circles have areas and circumferences that are accurately
described by the Euclidean formulas πr2 and 2πr. We explore geometry on
surfaces in more detail in the next chapter.
7
Geometry on Surfaces
In hyperbolic geometry (D, H) and elliptic geometry (P2 , S), the area of a
triangle is determined by the sum of its angles. This is a signi cant dierence
from Euclidean geometry, in which a triangle with three given angles can be
built to have any desired area. Does this mean that if a bug lives in a world
adhering to elliptic geometry, it can never stumble upon a triangle with three
right angles having area 3π ? Yes and no. In the elliptic geometry as de ned in
Chapter 6, no such triangle exists because a triangle with 3 right angles must
have area
π
( π2 + π π
2 + 2 ) − π = 2 . So the answer appears to be yes. However,
2
the elliptic geometry (P , S) models the geometry of the unit
sphere, and this
choice of sphere radius is somewhat arbitrary. What if the radius of the sphere
changes? Imagine a triangle with three right angles having one vertex at the
north pole and two vertices on the equator. If the sphere uniformly expands,
the angles of the triangle will stay the same, but the area of the triangle will
increase. So, if a bug is convinced she lives in a world with elliptic geometry,
but is also convinced she has found a triangle with three right angles and area
3π , the bug might be drawn to conclude she lives in a world modeled on a
larger sphere than the unit 2-sphere.
The key geometrical property of a space dictating the relationship between
the angles of a triangle and its area is called curvature. Curvature also dictates
the relationship between the circumference of a circle and its radius.
7.1 Curvature
Consider the smooth curve in Figure 7.1.1. The curvature of the curve at a
point is a measure of how drastically the curve bends away from its tangent
line, and this curvature is often studied in a multivariable calculus course.
The radius of curvature at a point corresponds to the radius of the circle that
best approximates the curve at this point. The radius r of this circle is the
reciprocal of the curvature k of the curve at the point: k = 1/r.
Figure 7.1.1: The curvature of a curve.
The curvature of a surface (such as the graph of a function z = f (x, y)) at
a particular point is a measure of how drastically the surface bends away from
130
7.1. Curvature 131
its tangent plane at the point. There are three fundamental types of curvature.
A surface has positive curvature at a point if the surface lives entirely on one
side of the tangent plane, at least near the point of interest. The surface
has negative curvature at a point if it is saddle-shaped, in the sense that the
tangent plane cuts through the surface. Between these two cases is the case
of zero curvature. In this case the surface has a line along which the surface
agrees with the tangent plane. For instance, a cylinder has zero curvature, as
suggested in Figure 7.1.2(c).
(a) (b) (c)
Figure 7.1.2: The curvature of a surface at a point can be (a) positive; (b)
negative; or (c) zero.
This informal description of curvature makes use of how the surface is
embedded in space. One of Gauss' great theorems, one he called his Theorem
Egregium , states that the curvature of a surface is an intrinsic property of the
surface. The curvature doesn't change if the surface is bent without stretching,
and our tireless two-dimensional inhabitant living in the space can determine
the curvature by taking measurements.
A two-dimensional bug living in the hyperbolic plane, the projective plane,
or the Euclidean plane would notice that a small circle's circumference is
related to its radius by the Euclidean formula c ≈ 2πr. In Euclidean
geometry this formula applies to all circles, but in the non-Euclidean cases,
the observant bug might notice in large circles a signi cant dierence between
the actual circumference of a circle and the circumference predicted by c = 2πr.
Large circles about a cup-shaped point with positive curvature will have
circumference less than that predicted by Euclidean geometry. This fact
explains why a large chunk of orange peel fractures if pressed at onto a table.
Large circles drawn around a saddle-shaped point with negative curvature will
have circumference greater than that predicted by the Euclidean formula.
Calculus may be used to precisely capture this deviation between the
Euclidean-predicted circumference 2πr and the actual circumference c for
circles of radius r in the dierent geometries.
Recall that in the hyperbolic plane, c = 2π sinh(r); in the Euclidean plane
c = 2πr; and in the elliptic plane c = 2π sin(r). In Figure 7.1.3 we have
c
graphed the ratio
2πr where c is the circumference of a circle with radius r in
(a) the hyperbolic plane; (b) the Euclidean plane; and (c) the elliptic plane.
132 7. Geometry on Surfaces
1.2
(a)
c/(2πr)
1
(b)
ratio
(c)
0.8
0.2 0.4 0.6 0.8 1
radius r
Figure 7.1.3: Plotting the ratio c/2πr against r in (a) hyperbolic geometry,
(b) Euclidean geometry, and (c) elliptic geometry.
In all three cases, the ratio c/2πr approaches 1 as r shrinks to 0.
Furthermore, in all three cases the derivative of the ratio approaches 0 as
r → 0+ . But with the second derivative of the ratio we may distinguish these
geometries. It can be shown that the curvature at a point is proportional to
this second derivative evaluated in the limit as r → 0+ . We will not derive
this formula for curvature, but will use this working de nition as it appears in
Thurston's book [11].
De nition 7.1.4. Suppose a circle of radius r about a point p is drawn in a
1
space , and its circumference is c. The curvature of the space at p is
d2
c
k = −3 lim+ 2 .
r→0 dr 2πr
Since we are interested in worlds that are homogeneous and isotropic, we
will focus our attention on worlds in which the curvature is the same at all
points. That is, we investigate surfaces of constant curvature.
Example 7.1.5: The curvature of a sphere.
Consider the sphere with radius s in the following diagram, and note
the circle centered at the north pole N having surface radius r. The
circle is parallel to the plane z = 0, has Euclidean radius x, and hence
circumference 2πx.
N
r
x
θ s
1 the term `space' is intentionally vague here. Our space needs to have a well-de ned
metric, so that it makes sense to talk about radius and circumference. The space might
be the Euclidean plane, the hyperbolic plane or the sphere. Other spaces are discussed in
Section 7.5.
7.1. Curvature 133
But x = s sin(θ) and r = θ · s, from which we deduce x = s sin( rs ),
and in terms of the surface radius r of the circle, its circumference is
r
c = 2πs sin .
s
The curvature of the sphere at N is thus
d2 2πs sin(r/s)
k = −3 lim .
r→0+ dr 2 2πr
Cancelling the 2π terms and replacing sin(r/s) with its power series
expansion, we have
r3 r5
d2 s( rs − 6s 3 + 120s5 − · · · )
k = −3 lim+ 2
r→0 dr r
2 2
r4
d r
= −3 lim+ 2 1 − 2 + − ···
r→0 dr 6s 120s4
12r2
−1
= −3 lim+ + − ··· .
r→0 3s2 120s4
Note that all the terms of the second derivative after the rst have
powers of r in the numerator, so these terms go to 0 as r → 0+ , and
the curvature of the sphere at the north pole is 1/s2 . In fact because
the sphere is homogeneous, the curvature at any point is
1
k= .
s2
Example 7.1.6: Curvature of the hyperbolic plane.
Because hyperbolic geometry is homogeneous and its transformations
preserve circles and lengths, the curvature is the same at all points
in the hyperbolic plane. We choose to compute the curvature at the
origin.
Recall, the circumference of a circle in (D, H) is c = 2π sinh(r). To
compute the curvature, use the power series expansion
r3 r5
sinh(r) = r + + + ··· .
3! 5!
d2 2π sinh(r)
k = −3 lim+ 2
r→0 dr 2πr
2
r2 r4
d
= −3 lim+ 2 1 + + + ···
r→0 dr 3! 5!
1 12r2
= −3 lim + + ··· .
r→0+ 3 5!
Again, each term of the second derivative after the rst has a power
of r in its numerator, so in the limit as r → 0+ , each of these terms
vanishes. Thus, the curvature of the hyperbolic plane in (D, H) is
k = −1.
134 7. Geometry on Surfaces
Exercises
1. Use our working de nition to show that the curvature of the projective
plane in elliptic geometry is 1. Recall, c = 2π sin(r) in this geometry.
2. Use our working de nition to explain why the curvature of the Euclidean
plane is k = 0.
7.2 Elliptic Geometry with Curvature k>0
One may model elliptic geometry on spheres of varying radii, and a change in
radius will cause a change in the curvature of the space as well as a change in
the relationship between the area of a triangle and its angle sum.
For any real number k > 0, we may construct a sphere with constant
curvature k. According to Example 7.1.5, the sphere centered at the origin
√
with radius 1/ k works. Geometry on this sphere can be modeled down in
the extended plane via stereographic projection. This geometry will be called
elliptic geometry with curvature k > 0. √
Consider the sphere S2k centered at the origin of R3 with radius 1/ k .
De ne stereographic projection φk : S2k → C+ , just as we did in Section 3.3, to
obtain the formula
(
a√ b√ √1 ;
1−c k
+ 1−c k
i if c 6= k
φk (a, b, c) =
∞ if c= √1 .
k
Diametrically opposed points on S2k get mapped via φk to points z and za
−1
that satisfy the equation za =
kz , by analogy with Lemma 6.1.2. We call two
+ 2
such points in C antipodal with respect to Sk , or just antipodal points if the
value of k is understood.
Our model for elliptic geometry with curvature
√ k has space P2k equal to
the closed disk in C of radius 1/ k , with antipodal points of the boundary
identi ed. This space is simply a scaled version of the projective plane from
Chapter 6.
The group of transformations, denoted Sk , consists of those Möbius
transformations that preserve antipodal points with respect to S2k . That is,
T ∈ Sk if and only if the following holds:
1 1
if za = − then T (za ) = − .
kz kT (z)
The geometry (P2k , Sk ) with k > 0 is called elliptic geometry with
curvature k. Note that (P21 , S1 ) is precisely the geometry we studied in
Chapter 6.
The transformations of C+ in the group Sk correspond precisely with
rotations of the sphere S2k . One can show that transformations in Sk have
the form
z − z0
T (z) = eiθ .
1 + kz0 z
7.2. Elliptic Geometry with Curvature k>0 135
We de ne lines in elliptic geometry with curvature k to be clines with the
1
property that if they go through z then they go through za = − kz . These lines
correspond precisely to great circles on the sphere S2k .
The arc-length and area formulas also slip gently over from Chapter 6 to
this more general setting.
The arc-length of a smooth curve r in P2k is
b
2|r 0 (t)|
Z
L(r) = dt.
a 1 + k|r(t)|2
As before, arc-length is an invariant, and the shortest path between two
points is along the elliptic line through them. In the exercises we derive a
formula for the distance between points in this geometry. The greatest possible
√
distance between two points in (P2k , Sk ) turns out to be π/(2 k).
The area of a region R given in polar form is computed by the formula
ZZ
4r
A(R) = drdθ.
R (1 + kr2 )2
To compute the area of a triangle, proceed as in Chapter 6. First, tackle
the area of a lune , a 2-gon whose sides are elliptic lines in (P2k , Sk ).
Lemma 7.2.1. Assume k > 0. A lune in (P2k , Sk ) with interior angle α has
area 2α/k.
Proof. Without loss of generality, we may consider the vertex of our lune
to be the origin. As before, elliptic lines through the origin must also pass
through ∞, so our two lines forming the lune are Euclidean lines. After a
convenient rotation, we may further assume one of these lines is the real axis,
so that the lune resembles the one in Figure 6.3.10. To compute the area of
the lune, compute the integral
√
Z α Z 1/ k
4r
A=2 drdθ.
0 0 (1 + kr2 )2
2
Letting
√u = 1 + kr so that du = 2krdr, the bounds of integration change
from [0, 1/ k] to [1, 2]. Then,
Z α Z 2 Z α
2 du 4 1 2α
A=2 2
dθ = dθ = .
0 k 1 u k 0 2 k
Thus, the angle of a lune with interior angle α is 2α/k .
We remark that the lune with angle
√ π actually covers the entire disk of
radius 1/ k . Thus, the area of the entire space
√ P2k is 2π/k , which matches
√
half the surface area of a sphere of radius 1/ k . We often call s = 1/ k the
radius of curvature for the geometry; it is the radius of the disk on which
we model the geometry.
Also, the integral computation in the proof of Lemma 7.2.1 reveals the
following useful antiderivative:
−2
Z
4r
2 2
dr = + C.
(1 + kr ) k(1 + kr2 )
This fact may speed up future integral computations.
Lemma 7.2.2. In elliptic geometry with curvature k, the area of a triangle
with angles α, β, and γ is
1
A= (α + β + γ − π).
k
136 7. Geometry on Surfaces
Proof. As in the case k = 1, the area of any triangle may be determined
from the area of three lunes and the total area of P2k , as depicted in
Example 6.3.11.
Example 7.2.3: Triangles on the Earth.
The surface of the Earth is approximately spherical with radius about
6375 km. Therefore, the geometry on the surface of the Earth can be
reasonably modeled by (P2k , Sk ) where k = 1/63752 km−2 . The area of
a triangle on the Earth's surface having angles α, β, and γ is
1
A= (α + β + γ − π).
k
Can you nd the area of the triangle formed by Paris, New York, and
Rio? Use a globe, a protractor, and some string. The string follows a
geodesic between two points when it is pulled taut.
Exercises
1. Prove that for k > 0, any transformation in Sk has the form
z − z0
T (z) = eiθ ,
1 + kz0 z
where θ is any real number and z0 is a point in P2k . Hint: Follow the derivation
of the transformations in S found in Chapter 6.
2. Verify the formula for the stereographic projection map φk .
√
3. Assume k > 0 and let s = 1/ k . Derive the following measurement
formulas in (P2k , Sk ).
a. The length of a line segment from 0 to x, where 0<x≤s is
dk (0, x) = 2s arctan(x/s).
b. The circumference of the circle centered at the origin with elliptic radius
√
r < π/(2 k) is C = 2πs sin(r/s).
c. The area of the circle centered at the origin with elliptic radius
√
2 2 r
r < π/(2 k) is A = 4πs sin .
2s
4. In this exercise we investigate the idea that the elliptic formulas
in Exercise 7.2.3 for distance, circumference, and area approach Euclidean
formulas when k → 0+ .
a. Show that the elliptic distance dk (0, x) from 0 to x, where 0 < x ≤ s,
approaches 2x as k → 0+ (twice the usual notion of Euclidean distance).
b. Show that the elliptic circumference of a circle with elliptic radius r
approaches 2πr as k → 0+ .
c. Show that the elliptic area of this circle approaches πr2 as k → 0+ .
5. Triangle trigonometry in (P2k , Sk ).
Suppose we have a triangle in (P2k , Sk ) with side lengths a, b, c and angles
α, β, γ as pictured in Figure 6.3.13.
a. Prove the elliptic law of cosines in (P2k , Sk ):
√ √ √ √ √
cos( kc) = cos( ka) cos( kb) + sin( ka) sin( kb) cos(γ).
7.3. Hyperbolic Geometry with Curvature k<0 137
b. Prove the elliptic law of sines in (P2k , Sk ):
√ √ √
sin( ka) sin( kb) sin( kc)
= = .
sin(α) sin(β) sin(γ)
7.3 Hyperbolic Geometry with Curvature k<0
We may do the same gentle scaling of the Poincaré model of hyperbolic
geometry as we did in the previous section to the disk model of elliptic
geometry. In particular, for each negative number k<0 we construct a model
for hyperbolic geometry with curvature k.p
We de ne the space Dk to be the open disk of radius 1/ |k| centered
p at
the origin in C. That is, Dk consists of all z in C such that |z| < 1/ |k| . In
p
this setting, the circle at in nity is the boundary circle |z| = 1/ |k|.
The group Hk consists of all Möbius transformations that send Dk to itself.
The geometry (Dk , Hk ) with k < 0 is called hyperbolic geometry with
curvature k. Pushing analogy with the elliptic case, we may de ne the group
of transformations to consist of all Möbius transformations with this property:
if z and z ∗ are symmetric with respect to the circle at in nity then T (z) and
T (z ∗ ) are also symmetric with respect to the circle at in nity. Noting that the
∗ 1 1
point symmetric to z with respect to this circle is z = = − , we draw
|k|z kz
the satisfying conclusion that T ∈ Hk if and only if the following holds:
1 1
if z∗ = − then T (z ∗ ) = − .
kz kT (z)
Thus, the group Hk in the hyperbolic case has been de ned precisely as the
group Sk in the elliptic case. Furthermore, one can show that transformations
in Hk have the form
z − z0
T (z) = eiθ ,
1 + kz0 z
where z0 is a point in Dk .
Straight lines in this geometry are the clines in C+ orthogonal to the circle
at in nity. By Theorem 3.2.8, a straight line in (Dk , Hk ) is precisely a cline
with the property that if it goes through z then it goes through its symmetric
−1
point .
kz
The arc-length and area formulas also get tweaked by the scale factor, and
now look identical to the formulas for elliptic geometry with curvature k.
The arc-length of a smooth curve r in Dk is
b
2|r 0 (t)|
Z
L(r) = dt.
a 1 + k|r(t)|2
The area of a region R given in polar coordinates is computed by the formula
ZZ
4r
A(R) = drdθ.
R (1 + kr2 )2
As in Chapter 5 when k was xed at -1, the area formula is a bear to use,
and one may convert to an upper half-plane model to determine the area of
2
a
3 -ideal triangle in Dk . The ambitious reader might follow the methods of
138 7. Geometry on Surfaces
2
3 -ideal triangle in (Dk , Hk ) (k < 0) with
Section 5.5 to show that the area of a
1
interior angle α is − (π − α).
k
With this formula in hand, we can derive the area of any triangle in Dk in
terms of its angles, exactly as we did in Chapter 5.
Lemma 7.3.1. In hyperbolic geometry with curvature k, the area of a triangle
with angles α, β , and γ is
1
A= (α + β + γ − π).
k
Observing negative curvature
Suppose we are located at a point z in a hyperbolic universe with curvature k.
We see in the distance a hyperbolic line L that seems to extend inde nitely. We
might intuitively see the point w on the line that is closest to us, as suggested
in Figure 7.3.2. Now suppose we look down the road a bit to a point v . If v is
close to w the angle ∠wzv will be close to 0. As v gets further and further away
from w, the angle will grow, getting closer and closer to the angle θ = ∠wzu,
where u is an ideal point of the line L.
v
θ w
z d
Figure 7.3.2: The angle of parallelism θ of a point z to a line L.
A curious fact about hyperbolic geometry is that this angle θ, which is
called the angle of parallelism of z to L, is a function of z 's distance d
to L. In Section 5.4 we saw that cosh(d) = 1/ sin(θ) in (D, H). In particular,
one may deduce the distance d to the line L by computing θ . No such analogy
exists in Euclidean geometry. In a Euclidean world, if one looks farther and
◦
farther down the line L, the angle will approach 90 , no matter one's distance
d from the line. The following theorem provides another formula relating the
angle of parallelism to a point's distance to a line.
Theorem 7.3.3 Lobatchevsky's formula. In hyperbolic geometry with
curvature k, the hyperbolic distance d of a point z to a hyperbolic line L is
related to the angle of parallelism θ by the formula
√
tan(θ/2) = e− |k|d
.
Proof. For this proof, let s = √1 . Note that s is the Euclidean radius of
|k|
the circle at in nity in the disk model for hyperbolic geometry with curvature
k. Since angles and lines and distances are preserved, assume z is the origin
and L is orthogonal to the positive real axis, intersecting it at the point x (with
0 < x < s).
7.3. Hyperbolic Geometry with Curvature k<0 139
s
r
θ
x x∗
L
Figure 7.3.4: Deriving Lobatchevsky's formula.
Recall the half-angle formula
tan(θ)
tan(θ/2) = .
sec(θ) + 1
According to Figure 7.3.4, tan(θ) = r/s, where r is the Euclidean radius of
the circle containing the hyperbolic line L. Furthermore, sec(θ) = (x + r)/s,
so
r
tan(θ/2) = . (1)
x+r+s
We may express x and r in terms of the hyperbolic distance d from 0 to x.
In Exercise 7.3.2 we prove the hyperbolic distance from 0 to x in (Dk , Hk ) is
s+x
d = s ln
s−x
so that
ed/s − 1
x=s· .
ed/s + 1
Expressr in terms of d by rst expressing it in terms of x. Note that
segment xx∗ is a diameter of the circle containing L, where x∗ = −1 kx is the
point symmetric to x with respect to the circle at in nity. Thus, r is half the
∗
distance from x to x :
1 + kx2
r=− .
2kx
Replacing k with −1/s2 , we have
s2 − x2
r= .
2x
One checks that after writing x in terms of d, r is given by
2sed/s
r= .
e2d/s − 1
Substitute this expression for r into the equation labeled (1) in this proof,
and after a dose of satisfactory simplifying one obtains the desired result:
tan(θ/2) = e−d/s .
p
Since s = 1/ |k| this completes the proof.
140 7. Geometry on Surfaces
Parallax If a star is relatively close to the Earth, then as the Earth moves
in its annual orbit around the Sun, the star will appear to move relative to
the backdrop of the more distant stars. In the idealized picture that follows,
e1 and e2 denote the Earth's position at opposite points of its orbit, and the
star s is orthogonal to the plane of the Earth's orbit. The angle p is called the
parallax , and in a Euclidean universe, p determines the star's distance from
the Sun, D, by the equation D = d/ tan(p), where d is the Earth's distance
from the Sun.
e1
d
p
Sun s
D
e2
We may determine p by observation, and d is the radius of Earth's orbit
around the Sun (d is about 8.3 light-minutes.) In practice, p is quite small,
d
so a working formula is
p D =
. The rst accurate measurement of parallax
was recorded in 1837 by Friedrich Bessel (1784-1846) . He found the stellar
◦
parallax of 0.3 arc seconds (1 arc second = 1/3600 ) for star 61 Cygni, which
put the star at about 10.5 light-years away.
If we live in a hyperbolic universe with curvature k, a detected parallax
puts a bound on how curved the universe can be. Consider Figure 7.3.5. As
before, e1 and e2 represent the position of the Earth at opposite points of its
orbit, so that the distance between them is 2d, or about 16.6 light-minutes.
Assume star s is on the positive real axis and we have detected a parallax p,
so that angle ∠e2 se1 = 2p.
e2
α
2d
2p u
e1 a s
Figure 7.3.5: A detected parallax in a hyperbolic universe puts a bound on
its curvature
The angle α = ∠e1 e2 s in Figure 7.3.5 is less than the angle of parallelism
θ = ∠e1 e2 u. Noting that tan(x) is an increasing function and applying
Lobatchevsky's formula it follows that
√
tan(α/2) < tan(θ/2) = e− |k|2d
.
We may solve this inequality for |k|:
√
tan(α/2) < e− |k|2d
p
ln(tan(α/2)) < − |k|2d ln(x) is increasing
7.3. Hyperbolic Geometry with Curvature k<0 141
2
ln(tan(α/2))
> |k|. x2 is decreasing for x<0
2d
To get a bound for |k| in terms of p, note that α ≈ π/2 − 2p (the triangles
◦
used in stellar parallax have no detectable angular deviation from 180 ), so
2
ln(tan(π/4 − p))
|k| < .
2d
We remark that for values of p near 0, the expression ln(tan(π/4 − p)) has
linear approximation equal to −2p, so a working bound for k , which appeared
2
in Schwarzschild's 1900 paper [27], is |k| < (p/d) .
Exercises
1. Prove that for k < 0, any transformation in Hk has the form
z − z0
T (z) = eiθ ,
1 + kz0 z
where θ is any real number and z0 is a point in Dk . Hint: Follow the derivation
of the transformations in H found in Chapter 5.
2.
p
Assume k < 0 and let s = 1/ |k|. Derive the following measurement
formulas in (Dk , Hk ).
a. The length of a line segment from 0 to x, where 0 < x < s is
s+x
dk (0, x) = s ln .
s−x
Hint: Evaluate the integral by partial fractions.
b. The circumference of the circle centered at the origin with hyperbolic
radius r is c = 2πs sinh(r/s).
c. The area of the circle centered at the origin with hyperbolic radius r is
A = 4πs2 sinh2 ( 2s
r
).
3. Let's investigate the idea that the hyperbolic formulas in Exercise 7.3.2 for
distance, circumference, and area approach Euclidean formulas when k → 0− .
a. Show that the hyperbolic distance dk (0, x) from 0 to x, where 0 < x < s,
−
approaches 2x as k → 0 .
b. Show that the hyperbolic circumference of a circle with hyperbolic radius
r approaches 2πr as k → 0− .
c. Show that the hyperbolic area of this circle approaches πr2 as k → 0− .
4. Triangle trigonometry in (Dk , Hk ).
Suppose we have a triangle in (Dk , Hk ) with side lengths a, b, c and
p angles
α, β, and γ as pictured in Figure 5.4.11. Suppose further that s = 1/ |k|.
a. Prove the hyperbolic law of cosines in (Dk , Hk ):
cosh(c/s) = cosh(a/s) cosh(b/s) − sinh(a/s) sinh(b/s) cos(γ).
b. Prove the hyperbolic law of sines in (Dk , Hk ):
sinh(a/s) sinh(b/s) sinh(c/s)
= = .
sin(α) sin(β) sin(γ)
5. As k < 0 approaches 0, the formulas of hyperbolic geometry (Dk , Hk )
approach those of Euclidean geometry. What happens to Lobatchevsky's
142 7. Geometry on Surfaces
formula as k approaches 0? What must be the angle of parallelism θ in the
limiting case? Is this value of θ independent of d?
6. Bessel determined a parallax of p = .3 arcseconds for the star 61 Cygni.
Convert this angle to radians and use it to estimate a bound for the curvature
constant kif the universe is hyperbolic. The units for this bound should be
−2
light years (convert the units for the Earth-Sun distance to light-years).
7. The smallest detectable parallax is determined by the resolving power of
our best telescopes. Search the web to nd the smallest detected parallax to
date, and use it to estimate a bound on k if the universe is hyperbolic.
7.4 The Family of Geometries (Xk , Gk )
A strong connection exists between the family (P2k , Sk ) of elliptic geometries
with curvature k > 0 and the family (Dk , Hk ) of hyperbolic geometries
with curvature k < 0. The two families sport identical descriptions of the
transformation group, identical descriptions of straight lines, identical arc-
length and area formulas, as well as identical formulas for the area of a triangle.
We may symbolize this connection with the following description of an
in nite family of geometries, one for each real number k. This general
description will allow us to elegantly express some important features common
to hyperbolic, Euclidean, and elliptic geometry.
De nition 7.4.1. For each real number k the geometry (Xk , Gk ) has space
Dk
if k < 0;
Xk = C if k = 0;
2
Pk if k > 0,
and group of transformations Gk consisting of all Möbius transformations of
the form
z − z0
T (z) = eiθ ,
1 + kz0 z
whereθ ∈ R and z0 is a point in Xk . Moreover, the unique line through two
pointsp and q in (Xk , Gk ) is the unique cline through the points p, q , and
−1/(kp). A smooth curve r : [a, b] → Xk has arc-length given by
b
2|r 0 (t)|
Z
L(r) = dt.
a 1 + k|r(t)|2
The area of a polar region R in (Xk , Gk ) is given by
ZZ
4r
A(R) = drdθ.
R (1 + kr2 )2
As we have seen, these geometries manifest themselves in strikingly dierent
ways. If k > 0, the sum of the angles of a triangle must be greater than π ;
and ifk < 0 the sum of the angles of a triangle must be less than π . If k > 0
2
the space Xk = Pk has nite area; if k < 0, Xk = Dk has in nite area. If
k > 0 the circumference of a circle with radius r is less than 2πr; if k < 0 the
circumference is greater than 2πr .
What about when k = 0? We may check that (X0 , G0 ) corresponds to
Euclidean geometry, though with a scaled metric. In particular, lines in
(X0 , G0 ) correspond to Euclidean lines (since k = 0, −1/(kp) = ∞, so the
7.4. The Family of Geometries (Xk , Gk ) 143
unique line through p and q is the Euclidean line), and when k=0 the arc-
length is simply twice the usual Euclidean arc-length. So while we are scaling
distances in (X0 , G0 ), Euclidean geometry applies: triangles are Euclidean
◦
triangles and have angle sum equal to 180 . Triangles with a right angle are
Euclidean right triangles and satisfy the Pythagorean theorem.
Thus, we treat (Xk , Gk ) as one big family of geometries. The sign of k
dictates the type of geometry we have, and the magnitude of k dictates the
radius of the disk in which we model the geometry (unless k = 0 in which case
the space is C). Morevoer, Euclidean geometry (X0 , G0 ) marks the edge of the
knife from which we move into a hyperbolic world is k drops below 0, and into
an elliptic world if k rises above 0.
We now summarize some results established in the previous sections and
emphasize key features common to all (Xk , Gk ).
First and foremost, we note that arc-length is an invariant function of
(Xk , Gk ) and that the arc-length ensures that the shortest path from p to q
in (Xk , Gk ) is along the line between them. We have discussed these facts in
the cases k = −1, 0, 1, and the result holds for arbitrary k . So, the arc-length
formula provides a metric on (Xk , Gk ): Given p, q ∈ Xk , we de ne dk (p, q) to
be the length of the shortest path from p to q . The circle in (Xk , Gk ) centered
at p through q consists of all points in Xk whose distance from p equals dk (p, q).
Theorem 7.4.2. For all real numbers k, (Xk , Gk ) is homogeneous and
isotropic.
z−p
Proof. Given any point p in Xk ,T (z) = 1+kpz in Gk
the transformation
maps p to the origin. So all points in Xk are congruent to 0. By the group
structure of Gk it follows that any two points in Xk are congruent, so the
geometry is homogeneous.
To show(Xk , Gk ) is isotropic we consider three cases.
k < 0 then
If
p (Xk , Gk ) models hyperbolic geometry on the open disk with
radius s = 1/ |k|. As such, Gk contains the sorts of Möbius transformations
discussed in Chapter 5 and pictured in Figure 5.1.6. In particular, for any
point p ∈ Dk , Gk contains all elliptic Möbius transformations that swirl points
around type II clines of p and −1/(kp), the point symmetric to p with respect
to the circle at in nity. These maps are preciley the rotations about the point
p in this geometry: they rotate points in Xk around cicles centered at p.
If k = 0, then transformations in G0 have the form T (z) = eiθ (z − z0 ).
Now, rotation by angle θ about the point p in the Euclidean plane is given by
T (z) = eiθ (z − p) + p. Setting z0 = p − pe−iθ we see that this rotation indeed
lives in G0 .
If k > 0 then (Xk , Gk ) models elliptic geometry on the projective plane with
√
radius s = 1/ k . As such, for each p ∈ Xk , Gk contains all elliptic Möbius
transformations that x p and pa , the point antipodal to p with respect to
the circle with radius s. Such a map rotates points around type II clines with
respect to p and pa . Since these type II clines correspond to circles in Xk
centered at the xed point, it follows that Gk contains all rotations.
Thus, for all k ∈ R, (Xk , Gk ) is isotropic.
Theorem 7.4.3. Suppose k is any real number, and we have a triangle in
(Xk , Gk ) whose angles are α, β, and γ and whose area is A. Then
kA = (α + β + γ − π).
Proof of this tidy result has already appeared in pieces (see Exercise 1.2.2,
Lemma 7.2.2, and Lemma 7.3.1); we emphasize that this triangle area formula
reveals the locally Euclidean nature of all the geometries (Xk , Gk ): a small
144 7. Geometry on Surfaces
◦
triangle (one with area close to 0) will have an angle sum close to 180 . Observe
also that the closer |k| is to 0, the larger a triangle needs to be in order to detect
an angle sum dierent from 180◦ . Of course, if k = 0 the theorem tells us that
the angles of a Euclidean triangle sum to π radians.
Theorem 7.4.4. Suppose a convex n-sided polygon (n ≥ 3) in (Xk , Gk ) has
interior angles αi for i = 1, 2, . . . , n. The area A of the n-gon is related to its
interior angles by n
X
kA = αi − (n − 2)π.
i=1
Proof. A convex n-gon can be divided into n − 2 triangles as in Figure 7.4.5.
Observe that the area of the n-gon equals the sum of the areas of these triangles.
∆1 ∆2
∆3
∆4
∆6 ∆5
Figure 7.4.5: Splitting an n-gon into n−2 triangles in the case n = 8.
By Theorem 7.4.3, the area Ai of the ith triangle ∆i is related to its angle
sum by
X
kAi = ( angles in ∆i ) − π.
Thus,
n−2
X
kA = kAi
i=1
n−2
X X
= ( angles in ∆i − π).
i=1
Now, the total angle sum of the n−2 triangles equals the interior angle
sum of the n-gon, so it follows that
n
X
kA = αi − (n − 2)π.
i=1
This completes the proof.
Lemma 7.4.6. Suppose k ∈ R, s = √1
|k|
and 0 < x < s is a real number (if
k = 0, we just assume 0 < x). In (Xk , Gk ), the circle centered at 0 through x
has area
4πx2
.
1 + kx2
Proof. Consider the circle centered at the origin that goes through the point
x on the positive real axis, where 0 < x < s. The circular region matches the
polar rectangle 0 < θ < 2π and 0 < r < x, so the area is given by
Z 2π Z x
4r
2 2
drdθ.
0 0 (1 + kr )
7.4. The Family of Geometries (Xk , Gk ) 145
Evaluating this integral gives the result, and the details are left as an
exercise.
Theorem 7.4.7 Uni ed Pythagorean Theorem. Suppose k ∈ R, and we
have a geodesic right triangle in (Xk , Gk ) whose legs have length a and b and
whose hypotenuse has length c. Then
k
A(c) = A(a) + A(b) − A(a)A(b),
2π
where A(r) denotes the area of a circle with radius r as measured in (Xk , Gk ).
Proof. Suppose k ∈ R. If k = 0 the equation reduces to c2 = a2 + b2 ,
which is true by the Pythagorean Theorem 1.2.1! Otherwise, assume k 6= 0
and let s = √1 , as usual. Without loss of generality we may assume our right
|k|
triangle is de ned by the points z = 0, p = x, and q = yi, where 0 < x, y < s.
By construction, the legs zp and zq are Euclidean segments, and ∠pzq is right.
Let a = dk (z, p), b = dk (z, q), and c = dk (p, q). By Lemma 7.4.6,
4πx2 4πy 2
A(a) = , and A(b) = .
1 + kx2 1 + ky 2
To nd A(c), we rst nd dk (p, q). By invariance of distance,
dk (p, q) = dk (0, |T (q)|),
z−p |yi−x|
where T (z) = 1+kpz . Let t = |T (q)| =
|1+kxyi| , which is a real number.
Now, A(c) is the area of a circle with radius c, and the circle centered at 0
through t has this radius, so
4πt2
A(c) = .
1 + kt2
x2 +y 2
Using the fact that t2 = 1+k2 x2 y 2 , one can now check by direct substitution
that
k
A(c) = A(a) + A(b) − A(a)A(b).
2π
While we have proved the theorem, it feels a bit like we have missed the
best part - discovery of the relationship. For more on this, we encourage the
reader to consult [20].
Example 7.4.8: Detecting curvature with triangles.
Suppose a two-dimensional bug in (Xk , Gk ) walks along a line for a > 0
◦
units, turns left 90 , and walks on a line for another a units, thus
creating a right triangle with legs of equal length. Let c denote the
hypotenuse of this triangle. The diagram below depicts such a route,
in each of the three geometry types. For convenience, we assume the
journey begins at the origin and proceeds rst along the positive real
axis.
p
p
c c
c p a a
a θ θ
0 θ 0 0
a x a x a x
(a) (b) (c)
146 7. Geometry on Surfaces
It turns out the value of c as a function of a can reveal the curvature
k of the geometry. If k = 0 the Pythagorean theorem tells us that
c2 = 2a2 . For k < p 0, the hyperbolic
p law of cosines (Exercise 7.3.4)
tells us that cosh( |k|c) = cosh2 ( |k|a). For √ k > 0, the2 elliptic
√ law
of cosines (Exercise 7.2.5) tells us that cos( kc) = cos ( ka). We
may solve each of these equations for c, using the fact that a and c
are positive, to give us c as a function of a and the curvature k . We
have plotted these functions for k = −1, 0, 1 below. When k < 0 the
length of the hypotenuse of such a triangle is slightly longer than that
predicted by the Euclidean formula (the Pythagorean theorem); when
k>0 the length is smaller than the length predicted by the Euclidean
formula.
k = −1
2
k=0
c
1.5
hypotenuse
k=1
0.5
0.2 0.4 0.6 0.8 1 1.2 1.4
lega
We close by summarizing measurement formulas for these geometries. Ex-
cept for the case k=0 these formulas were proved in the exercises of the pre-
vioius two sections. The case k = 0 is tackled in Exercise 7.4.1.
Measurement Formulas in (Xk , Gk ).
Suppose k ∈ R, s = √1 and 0 < x < s is a real number (with the
|k|
convention that if k=0 we simply require 0 < x).
In (Xk , Gk ), the distance from 0 to x is given by
s+x
s ln if k < 0;
s−x
dk (0, x) =
2x if k = 0;
2s arctan(x/s) if k > 0.
A circle with radius r as measured in (Xk , Gk ) has area A(r) given
by
2 r
2
4πs sinh ( 2s )
if k < 0;
A(r) = πr2 if k = 0;
4πs2 sin2 ( 2s
r
) if k > 0,
7.4. The Family of Geometries (Xk , Gk ) 147
A circle with radius r as measured in (Xk , Gk ) has circumference
C(r) given by
2πs sinh(r/s)
if k < 0;
C(r) = 2πr if k = 0;
2πs sin(r/s) if k > 0.
Exercises
1. Check that the measurement formulas in (Xk , Gk ) are correct when
k = 0. In particular, show that d0 (0, x) = 2x for any x > 0 on the real axis,
and that a circle with radius r as measured in (X0 , G0 ) has area A(r) = πr2
and circumference C(r) = 2πr.
2. Complete the proof of Lemma 7.4.6.
3. Use De nition 7.1.4 to prove that for all real numbers k , the curvature
of(Xk , Gk ) is indeed equal to k. Hint: Tackle three cases: k < 0, k = 0, and
k > 0.
4. Suppose an intrepid team of two-dimensional explorers sets out to
determine which 2-dimensional geometry is theirs. Their cosmologists have told
them there world is homogeneous, isotropic, and metric, so they believe that
the geometry of their universe is modeled by (Xk , Gk ) for some real number
k. They carefully measure the angles and area of a triangle. They nd the
◦ ◦ ◦ 2
angles to be 29.2438 , 73.4526 , and 77.2886 , and the area is 8.81 km . Which
geometry is theirs? What is the curvature of their universe?
5. Prove that in (Xk , Gk ) the derivative of area with respect to r is
d
circumference:
dr [A(r)] = C(r).
6. Suppose a two-dimensional bug in (Xk , Gk ) traces the right triangle route
from 0 to p as depicted in Example 7.4.8. Argue that for a given value of a, the
hyperbolic hypotenuse length exceeds the Euclidean hypotenuse length, which
exceeds the elliptic hypotenuse length. Hint: One might prove that for all
√ k,
dc d2 c
when a = 0, c = 0 da |
and
a=0 = 2 ; and then show that for a > 0,
da is
2
positive for negative values of k , and it is negative for positive values of k .
7. Suppose a team of two-dimensional explorers living in (Xk , Gk ) travels 8
◦
units along a line. Then they turn right (90 ) and travel 8 units along a line.
At this point they nd they are 12 units from their starting point. Which type
of geometry applies to their universe? Can they determine the value of k from
this measurement? If so, what is it?
8. Repeat the previous exercise using the measurements a = 8 units and
c = 11.2 units.
9. Suppose a team of two-dimensional explorers living in (Xk , Gk ) nds
themselves at point p. They travel 8 units along a line to a point z , turn right
◦
(90 ) and travel another 8 units along a line to the point q. At this point
they measure ∠pqz = .789 radians. Which type of geometry applies to their
universe? Can they determine the value of k from this measurement? If so,
what is it?
10. Suppose a team of two-dimensional explorers living in (Xk , Gk ) tethers
one of their team to a line 18 scrambles long (a scramble is the standard unit for
measuring length in this world - and 24 scrambles equals one tubablast). They
swing the volunteer around in a circle, and though he laughs maniacally, he is
148 7. Geometry on Surfaces
able to record with con dence that he traveled 113.4 scrambles. Assuming
these measurements are correct, which type of geometry applies to their
universe? Can they determine the value of k from these measurements? If
so, what is it?
7.5 Surfaces
Back in Chapter 1 we motivated the study of non-Euclidean geometry with a
question in cosmology: What is the shape of the universe? We discussed the
idea that dierent shapes inherit dierent types of geometry. Indeed, the rules
of geometry are dierent on a sphere than on a at plane. In this section we
take a break from geometry in order to discuss the shapes themselves.
In topology one studies those features of a space that remain unchanged if
the space is stretched or otherwise continuously deformed. Such features of a
space are called topological features.
The features of a space that do change under such deformations are
geometric features. For instance, as a ball is in ated its volume, curvature,
and surface area change; these are geometric properties. On the other hand,
no matter how big or lumpy the ball gets, or how it is stretched (unless it
pops!), a loop drawn on the surface of the ball separates the surface into two
disjoint pieces. This is a topological property of the ball. A second topological
property of a space is whether any loop drawn on it can be continuously
contracted (while staying in the space) to a point. The sphere also has this
feature. Nothing about the sphere's shape prevents a loop from shrinking on
the surface to a point.
Figure 7.5.1: The sphere and torus are topologically distinct.
On the surface of a donut there are loops one can draw that do not separate
the surface into disjoint pieces. The loop that goes around the donut like an
armband in Figure 7.5.1 is one such loop. Furthermore, this loop cannot be
continuously contracted to a point while staying on the surface. This suggests
the surface of a ball and the surface of a donut are topologically dierent
shapes.
The sphere is an example of a simply connected space because any loop
drawn on the surface can be contracted to a point. The torus is an example of
a multiconnected space because there exist loops in the torus that cannot
be contracted to a point.
Roughly speaking, two spaces are topologically equivalent, or homeomor-
phic , if one can be continuously deformed to look like the other. For instance,
a circle is topologically equivalent to a square. One can be mapped onto the
other via a homeomorphism : a continuous bijection between the objects that
has a continuous inverse. One homeomorphism is suggested in the following
example.
7.5. Surfaces 149
Example 7.5.2: A circle is homeomorphic to a square.
Construct a homeomorphism as follows. Assume the square and circle
are concentric as shown below, and that z0 is the center of the circle.
Then, for each point z on the square, de ne T (z) to be the point on
−
z→
the circle through which the ray 0z passes. One can show that T is a
homeomorphism: it is a 1-1 and onto function that is continuous, and
its inverse is continuous as well.
T (z)
z
zo
We are quickly heading into the realm of topology, and must resist the
temptation to dive headlong and formally into this rich subject. In this text
we restrict our focus to a whirlwind tour of topological tools that are used
to investigate possible shapes for two- and three-dimensional universes. We
encourage the reader interested in a more formal approach to the subject to
consult any number of good texts, including [9].
Our candidate universes are examples of manifolds. A topological n-
manifold is a space with the feature that each point in the space has a
neighborhood that looks like a patch of Rn . In cosmology, the spatial section of
space-time is assumed to be a 3-manifold, and when we ask about the shape of
the universe, we are asking about the shape of this 3-manifold. In this section
we focus on 2-manifolds.
n
A bit more formally, R consists of all n-tuples of real numbers p =
(x1 , x2 , . . . , xn ). The open n-ball centered at a point p in Rn with radius
r > 0, denoted Bn (p, r), is the set
Bn (p, r) = {x ∈ Rn | |x − p| < r},
where |x − p| is the Euclidean distance between the points x and p.
For instance, an open 1-ball is an open interval in R. The open 2-ball
B2 (z0 , r) consists of all points z in C such that |z − z0 | < r. The open 3-ball
B3 (p, r) consists of all points in R3 inside the sphere centered at p with radius
r.
r r
p
p−r p p+r z0
Figure 7.5.3: Open n-balls for n = 1, 2, 3.
A topological n-manifold is a space X with the feature that each point
in X has a neighborhood that is homeomorphic to an open n-ball.
A circle is an example of a 1-manifold: each point on a circle has a
neighborhood homeomorphic to an open interval. The surface of a sphere is a
150 7. Geometry on Surfaces
2-manifold: each point on the sphere has a neighborhood that is homeomorphic
to an open 2-ball.
Example 7.5.4: Some 2-manifolds.
a. LetS consist of parallel planes in R3 . In particular, let S =
{(x, y, z) ∈ R3 | z = 0 or z = 1} as in part (a) of the following diagram.
Each point in S has a neighborhood of points that is an open 2-ball.
s
z
(a) (b)
b. The open unit disk D = {z ∈ C | |z| < 1} is a 2-manifold. Each
point in D has a neighborhood that is an open 2-ball in the plane. In
particular, if |z| = r < 1, let s = 1 − r. Then B2 (z, s) does the trick.
c. The at torus. Consider the at torus of chapter 1, which has
been redrawn in Figure 7.5.5. It consists of all points in a rectangle
drawn in the Euclidean plane, with the additional feature that each
point on the top edge is identi ed with the point on the bottom
edge having the same x-coordinate; and each point on the left edge
is identi ed with the point on the right edge having the same y-
coordinate. In Figure 7.5.5 we indicate this edge identi cation with
oriented labels: the a edges get identi ed by matching the arrow
orientations, as do the b edges.
Each point in the at torus has a neighborhood homeomorphic to
an open 2-ball.
If a point p is not on an edge, as in Figure 7.5.5, then by choosing
a radius small enough, there is an open 2-ball about the point that
misses the edges.
If a point q is on an edge but not at a corner, then by choosing a
radius small enough, there is an open 2-ball about the point that misses
the corners. This open 2-ball consists of parts of two open 2-balls from
R2 . Because of the edge identi cation, these open 2-ball halves come
together to form a perfect open 2-ball about the point.
If a point u is at a corner of the rectangle, there is an open 2-ball
about the point that consists of parts of four dierent open 2-balls from
R2 . Because of the edge identi cation, these four quarter balls come
together to form a single open 2-ball of points around the corner point.
7.5. Surfaces 151
q a u
a
b p b
q
p b u
a
Figure 7.5.5: The at torus is a 2-manifold as well as a surface
De nition 7.5.6. A surface is a closed, bounded, and connected topological
2-manifold.
Of the 2-manifolds in Example 7.5.4, only the third is a surface according
to this de nition. A set in Rn is bounded if it lives entirely within some
open n-ball of nite radius. The parallel planes manifold extends in nitely far
away from the origin in R3 so this manifold is not bounded. This manifold is
not connected either. Informally, a space is connected if it has just one piece.
The parallel planes manifold has two distinct pieces (the two planes) that are
separated in space, so it is not connected. The open unit disk D is bounded
and connected, but not closed as a set in R2 . A set S in Rn is closed if it has
the following feature: if {zk } is a sequence of points in S that converges to a
point z, then z is also in S.
To see that D is not closed, consider the sequence { 12 , 23 , 34 , 45 , . . .}. Each
point in this sequence lives in D, but the limit of the sequence, which is 1, does
not. The torus, on the other hand, is closed, boundede, and connected, so we
call it a surface. We note that another topological term, compactness, is often
used when discussing surfaces. A space living in Rn is compact if and only
if it is closed and bounded. Thus, for us, a surface is a compact, connected
2-manifold.
Rather than concern ourselves with a formal development of these topolog-
ical concepts, we take with us this informal introduction to help us to pursue
the construction of surfaces. It turns out that all surfaces can be built from
the three venerable surfaces in Figure 7.5.7 by a process called the connected
sum.
Figure 7.5.7: The sphere S2 , the torus T2 , and the projective plane P2 .
Connected Sum If X1 and X2 are surfaces, the connected sum surface,
denoted X1 #X2 , is obtained as follows:
1. Remove an open 2-ball from X1 and an open 2-ball from X2 ;
2. Connect the boundaries of these open 2-ball with a cylinder.
Note that since Xi is a surface, each point in each space has a neighborhood
homeomorphic to an open 2-ball, so we may always achieve the connected sum
of two surfaces. The result is a new surface: it is still closed, bounded, and
152 7. Geometry on Surfaces
connected, and each point still has an open neighborhood homeomorphic to an
open 2-ball.
Figure 7.5.8: Some connected sums.
Example 7.5.9: Some connected sums.
Let T2 denote the torus surface in R3 and S2 denote the sphere, as
usual. Figure 7.5.8 depicts two connected sums: T2 #T2 , and S2 #T2 .
The surface T2 #T2 is called the two-holed torus, and it is topologically
equivalent to the surface labeled H2 in Figure 7.5.10. One can shrink
the length dimension of the connecting cylinder to make the one look
like the other. The surface S2 #T2 is homeomorphic to the torus T2 .
To see this, observe that if one removes an open-2 ball from a sphere
and attaches one end of a cylinder to the sphere along the boundary
of the removed disk, the result is homeomorphic to a closed disk, as
suggested in the following diagram. So, if one removes an open 2-ball
from a surface X and then caps the hole with this sphere-with-cylinder
shape, the net eect is patching the hole. In the arithmetic of connected
sums, the sphere plays the role of 0: S2 #X = X for any surface X.
For each integer g ≥ 1, the handlebody surface of genus g, denoted Hg ,
is de ned as
g copies
z }| {
Hg = T2 #T2 # · · · #T2 .
The surface Hg gets its name from the fact that it is topologically equivalent
to a sphere with g handles attached to it. For this reason, we set H0 = S2 . A
few handlebody surfaces are pictured in Figure 7.5.10.
7.5. Surfaces 153
H0 H2
H4
Figure 7.5.10: Some handlebody surfaces.
For each integer g ≥ 1, de ne the cross-cap surface of genus g by
g copies
z }| {
Cg = P2 #P2 # · · · #P2 .
A cross-cap is the space obtained by removing an open 2-ball from the
projective plane P2 . The cross-cap surface Cg is topologically equivalent to a
sphere with g open 2-balls removed and replaced with cross-caps.
The surface C1 is the projective plane. This space does not embed in R3 as
the handlebody surfaces do, but it can be represented as a disk with antipodal
points identi ed, as we did in Chapter 6. Figure 7.5.11 depicts the connected
sum C1 #C1 . In Figure 7.5.11(a), we have removed an open ball from each
copy of C1 . The boundary circles of these open 2-balls must be joined, and we
indicate this by orienting the boundary circles and giving each one the same
label of c. In Figure 7.5.11(b), we have sliced open each copy of C1 at the
vertex to which the c loops are joined. The c loops are now edges that we
identify together in Figure 7.5.11(c). Thus, we may view C2 as the square
with edge identi cations as indicated in part (c) of the gure. It turns out this
space C2 is homeomorphic to the Klein bottle, a famous surface we consider
shortly.
a2 a1
c c a2
a2 a1 a2 a1
(a) c
a1
a2 a1 (c)
c c
a2 a1
(b)
Figure 7.5.11: Constructing C2 .
154 7. Geometry on Surfaces
Polygonal Surfaces
Our representation of C2 is an example of a polygonal surface . To build
a polygonal surface, start with a nite number of polygons having an even
number of edges and then identify edges in pairs.
If the surface is built from a single polygon, the edge identi cations of
the polygon can be encoded in a boundary label . Each edge of the polygon
gets assigned a letter and an orientation. Edges that are identi ed have the
same letter, and we obtain a boundary label by traversing the boundary of our
polygon in the counter-clockwise direction (by convention) and recording the
letters we encounter, with an exponent of +1 if our walk is in the direction
of the oriented edge, and an exponent of -1 if our walk is in the opposite
direction of the oriented edge. For instance, a boundary label for the surface
C2 found in Figure 7.5.11 is a1 a1 a2 a2 . Furthermore, we may inductively show
that by repeating the connected sum operation of Figure 7.5.11 to Cg−1 #C1 ,
the surface Cg can be represented as a 2g -gon having boundary label
a1 a1 a2 a2 · · · ag ag .
It turns out that all surfaces can be constructed this way, an important and
useful result. We have seen that the torus can be constructed as a polygonal
surface: take a rectangle and identify the edges according to the boundary
label aba−1 b−1 . In Figure 7.5.13 we demonstrate that the 2-holed torus can be
represented as a regular octagon with boundary label
(a1 b1 a−1 −1 −1 −1
1 b1 )(a2 b2 a2 b2 ).
In general, for g ≥1 the surface Hg can be represented by a regular 4g -gon
having boundary label
(a1 b1 a−1 −1 −1 −1 −1 −1
1 b1 )(a2 b2 a2 b2 ) · · · (ag bg ag bg ).
The remarkable theorem here is that the handlebody surfaces and the cross-
cap surfaces account for all possible surfaces, without redundancy.
Theorem 7.5.12. Any surface is homeomorphic to the sphere S2 , a handle-
body surface Hg with g ≥ 1, or a crosscap surface Cg with g ≥ 1. Moreover,
no two surfaces in this list are homeomoprhic to each other.
Two proofs of this theorem are oating around the literature now. The
classic proof, which makes use of cell divisions, can be found, for instance,
in [9]. The new proof, due to John Conway, bypasses the arti cial constructs
in the classic proof, and can be found in [12].
To summarize, we have two ways to think about surfaces. First, the
classi cation theorem above can be restated as follows: any surface is
homeomorphic to the sphere, the sphere with some number of handles attached,
or the sphere with some number of cross-caps attached. Second, any surface
can be constructed from a 2m-gon with its edges identi ed in pairs, for some
m ≥ 1.
7.5. Surfaces 155
b2 a1
a2 b1
a2 a2 b1 b1
c c b2 a1
c
b2 a1
(a) a2 b1
b2 a2 b1 a1
b2 a1
(c)
a2 c c b1
b2 a1
(b)
Figure 7.5.13: H2 constructed from an octagon.
Characterizing a surface
If somebody throws a surface at us, how do we know which one we're catching?
One way to characterize a surface is to determine two particular topological
invariants: its orientability status and its Euler characteristic. Let's discuss
what these features of a surface are.
Orientability status
The Möbius strip is obtained from a rectangle by identifying the left and
right edges with a twist:
Figure 7.5.14: A Möbius strip.
Notice that a Möbius strip has an orientation-reversing path. A clock
rotating clockwise, if it heads along the Möbius strip, will eventually return to
its starting place to nd that it is now rotating counterclockwise.
finis
h
start
A surface is called non-orientable if it contains a Möbius strip. If a
surface does not contain a Möbius strip then it is orientable . The Möbius
strip itself is not a surface as we've de ned it because it has an edge. Points
on this edge don't have any neighborhoods that look like open 2-balls (they
156 7. Geometry on Surfaces
look more like half balls). But the Klein bottle in the following example is a
non-orientable surface.
Example 7.5.15: The Klein bottle.
The Klein bottle looks a lot like the torus, but there's a twist. The
top and bottom edges are identi ed as they were for the torus, but the
left and right edges are oriented oppositely. More formally, it is the
polygonal surface obtained from a square with boundary label aba−1 b,
and the Klein bottle is denoted by K2 .
Why is the Klein bottle non-orientable? A bug leaving the screen
on the right near the top would reappear on the left near the bottom.
But take a closer look, the bug has become mirror-reversed.
b b
This orientation-reversing path exists because of a Möbius strip
lurking in the Klein bottle. (Conisder, for instance, the thin horizontal
strip formed by the dashed segments in the gure.)
The same bug living in the torus would never nd itself mirror
reversed as a result of traveling in its surface. The torus is orientable.
A surface's orientability status is a topological invariant. This means that if
S and T are homeomorphic surfaces then they must have the same orientability
status. Notice that there exists a Möbius strip in each Cg (take a thin strip
from the middle of one a1 edge to the middle of the other), so that all cross-cap
surfaces are non-orientable. On the other hand, all the handlebody surfaces
Hg (g ≥ 0) are orientable.
Euler Characteristic
In addition to an orientability status, each surface has attached to it an
integer called the Euler characteristic. The Euler characteristic is a topological
invariant, and it can be calculated from a cell division of a surface, which is a
kind of tiling of the surface by planar faces. Let's make this more precise.
An n-dimensional cell , or n-cell , is a subset of a space whose interior is
homeomorphic to an open n-ball in Rn . For instance, a 1-cell, also called an
edge , is a set whose interior is homeomorphic to an open interval; a 2-cell, or
face , is a set whose interior is homeomorphic to an open 2-ball in the plane. We
call points in a space 0-cells. A 0-cell is also called a vertex (plural vertices ).
De nition 7.5.16. A cell complex C is a collection of cells in some space
subject to these two conditions:
1. The interiors of any two cells in the complex are disjoint
2. The boundary of each cell is the union of lower-dimensional cells in C.
A cell complex C is called an n-dimensional cell complex , or n-complex,
if it contains an n-cell, but no higher-dimensional cells.
7.5. Surfaces 157
Example 7.5.17: Some cell complexes.
A 1-complex is commonly called a graph : it consists of vertices and
edges. At each end of an edge is a vertex (possibly the same vertex
at each end), and no two edges intersect in their interiors. The left
side of the following diagram shows a 1-complex with seven 0-cells and
ve 1-cells. To the right is a 2-complex with one 0-cell and one 2-cell.
The entire boundary of the 2-cell is attached to the single 0-cell, thus
creating a well-known surface, the 2-sphere.
Example 7.5.18: The Platonic solids..
The ve Platonic solids can be viewed as 2-complexes if we ignore the
space bounded by their faces. We've pictured all ve, and given vertex,
edge, and face counts for each one.
Of course, the region bounded by the faces of each Platonic solid
is homeomorphic to an open 3-ball, so the Platonic solids can also be
viewed as 3-complexes; each solid having exactly one 3-cell.
v=8,e=12,f=6
v=4, e=6, f=4 v=6, e=12,f=8
v=20,e=30,f=12
v=12,e=30,f=20
De nition 7.5.19. A cell division of a space X is a cell complex C that is
homeomorphic to X.
For instance, each Platonic solid (viewed as a 2-complex) is a cell division
of S2 . To construct a homeomorphism, map each point of the Platonic solid
to a point on a sphere by projection as suggested in Figure 7.5.20.
158 7. Geometry on Surfaces
Figure 7.5.20: A dodecahedron and sphere are homeomorphic.
Example 7.5.21: Attempted cell divisions of H1 .
Three cell divisions of the torus are pictured below, along with one
failed cell division. In each valid cell division, we count the number
of faces, edges, and vertices of the cell division. To make an accurate
count, one must take the edge identi cation into account.
(a) (b)
(c) (d)
Cell division (a) has two vertices, six edges, and four faces. One
vertex is in the center of the rectangle, and the other vertex is in the
corner (remember, all four corners get identi ed to a single point).
As for the edges, four emanate from the center vertex, and we have
two others: the horizontal edge along the boundary of the rectangle
(appearing twice), and the vertical edge along the boundary (also
appearing twice). Thus, the edges of the rectangle are also edges of
the cell division, so the faces in this cell division are triangles, and
there are four of them.
Cell division (b) has six vertices, eight edges, and two faces. There
are four vertices in the interior of the rectangle, one vertex on the
horizontal boundary of the rectangle, and one vertex on the vertical
boundary. Notice that the corner point of the rectangle is not a vertex
of the cell division. To count the edges, observe that four edges form
the inner diamond, and one edge leaves each vertex of the diamond,
for a total of eight edges. The boundary edges of the rectangle do not
form edges in this cell division. Counting faces, we have one inside
7.5. Surfaces 159
the diamond and one outside the diamond. Convince yourself that the
region outside the diamond makes just one face.
You can check that cell division (c) has one vertex in the center of
the rectangle, two edges (one is horizontal, the other is vertical), and
one face.
The attempt (d) fails to be a cell division of the torus. Why is this?
At rst glance, we have four vertices, four edges, and two faces. The
trouble here is the face outside the inner square. It is not a 2-cell -
that is, its interior is not homeomorphic to an open 2-ball. To see this,
note that this region contains a loop that does not separate the face
into two pieces. Can you nd such a loop? Since no open 2-ball has
this feature, the region in question is not homeomorphic to an open
2-ball.
De nition 7.5.22. The Euler characteristic of a surface S , denoted χ(S),
is
χ(S) = v − e + f
where v , e, and f denote the number of 0-cells (vertices), 1-cells (edges), and
2-cells (faces), respectively, of a cell division of the surface.
The Euler characteristic is well-de ned. This means that dierent cell
divisions of the same surface will determine the same value of χ. Furthermore,
this simple number is a powerful topological invariant: If two surfaces have
dierent Euler characteristics then they are not homeomorphic. However, the
Euler characteristic alone doesn't completely characterize a surface: if two
surfaces have the same Euler characteristic, they need not be homeomorphic.
The Euler characteristic of the sphere is 2. Each Platonic solid in
Example 7.5.18 is a cell division of S2 , and a count reveals χ(S2 ) = 2.
2
The Euler characteristic of the torus T is 0. Each valid cell division in
Example 7.5.21 yields v − e + f = 0.
Each polygonal surface induces a cell division of the surface. This cell
division has a single face, and after identifying the edges in pairs, the corners
and edges of the underlying polygon correspond to vertices and edges in
the cell division of the surface. To determine the Euler characteristic of a
polygonal surface, be careful to make edge and vertex counts after the edge
identi cations.
Theorem 7.5.23. The handlebody surface Hg has Euler characteristic χ(Hg ) =
2 − 2g , for all g ≥ 0. The cross-cap surface Cg has Euler characteristic
χ(Cg ) = 2 − g , for all g ≥ 1.
Proof. We have already seen that the Euler characteristic of the sphere is 2,
so the result holds for H0 . For g ≥ 1 consider the standard polygonal represen-
tation of Hg as a 4g -gon with boundary label (a1 b1 a−1 −1 −1 −1
1 b1 ) · · · (ag bg ag bg ).
One checks that all the corners come together at a single point, so our cell
division of Hg has a single vertex. For instance, consider the 2-holed torus
in Figure 7.5.24. Starting at the lower right-hand corner labeled (1), begin
traversing the corner point in a clockwise direction. After hitting the b1 edge
near its initial point, one reappears on the other b1 edge near its initial point.
Keep circling the corner (according to the sequence indicated) until you re-
turn to the starting point. Notice that all eight corners are traversed before
returning to the starting point. So the cell division determined by the polygon
will have a single vertex. Since the 4g edges are identi ed in pairs, the cell
division has 2g edges, and there is one face, the interior of the polygon. Thus,
χ(Hg ) = 1 − 2g + 1 = 2 − 2g .
160 7. Geometry on Surfaces
a2
b2 b1
5 2
4 7
a2 a1
3 8
6 1
b2 b1
a1
Figure 7.5.24: This cell division of H2 has a single vertex.
The cross-cap surface Cg can be represented by the polygonal surface
obtained by identifying the edges of a 2g -gon according to the boundary label
(a1 a1 ) · · · (ag ag ). Once again, all the corners come together at a single point in
the edge identi cation, and the number of edges in the cell division is half the
number of edges in the 2g -gon. So the cell division determined by the polygon
has one vertex, g edges, and one face. Thus, χ(Cg ) = 2 − g .
In light of the previous theorem and the surface classi cation theorem, a
surface is uniquely determined by its orientability status and Euler character-
istic. We summarize the classi cation in the table below.
χ orientable non-orientable
2 H0
1 C1
0 H1 C2
-1 C3
-2 H2 C4
-3 C5
-4 H3 C6
. . .
. . .
. . .
Table 7.5.25: Classi cation of surfaces
This completes our brief, somewhat informal foray into the topology of
surfaces. Again, several good sources provide a rigorous development of these
ideas, including [9]. In the next section we turn to the task of attaching
geometry to a surface.
7.5. Surfaces 161
a a
b b
c b
d c
c a
c d
b a
(a) (b)
a d
e c
d b
d b
c a
c a
c b
e b
d a
(c) (d)
Figure 7.5.26: Four polygonal surfaces.
Exercises
1. Find the Euler characteristic of each surface in Figure 7.5.26. Then
determine whether each surface is orientable or non-orientable. Then classify
the surface.
2. Classify the polygonal surface built from a hexagon having boundary
label abca−1 b−1 c−1 .
3. With the aid of Figure 7.5.27, convince yourself that the connected sum
of two projective planes is homeomorphic to the Klein bottle as de ned in
Example 7.5.15. At the top of Figure 7.5.27 there are two projective planes
(with boundary labels a1 a1 and a2 a2 ), each with a disk removed. In the
connected sum, join the boundaries of the removed disks, which can be achieved
by joining the s1 arcs together and the s2 arcs together. The b and c edges in
the rst projective plane (and the d and e edges in the second projective plane)
indicate cuts we will make to the space, portrayed in the subsequent pictures.
Thus, by removing a disk from each projective plane, and cutting as indicated
by the b, c, d, and e edges, we obtain four rectangles (topologically), all of
whose edges get identi ed as indicated. Convince yourself that by moving
these rectangles around (either by rotation, or re ection about a vertical or
horizontal axis) we produce the Klein bottle.
4. What surface in our catalog do we obtain from the connected sum of a
torus and a projective plane? That is, what is H1 #C1 ? Explain your answer.
5. Show that in the polygonal representation of Cg , all corner points come
together at a single point when the edges are identi ed.
162 7. Geometry on Surfaces
a1 a2
1 3 s1
s1
b c d e
s2 s2
2 4
a1 a2
c c e e
a1 1 2 a1 a2 3 s1 4 a2
s1 s2 s2
b b d d
b e d c
a1 2 s2 4 a2 3 s1 1 a1
c d e b
Figure 7.5.27: C2 is homeomorphic to the Klein bottle K2 .
7.6 Geometry of Surfaces
If you've got a surface in your hand, you can nd a homeomorphic version
of the surface on which to construct hyperbolic geometry, elliptic geometry,
or Euclidean geometry. And the choice of geometry is unique: No surface
admits more than one of these geometries. As we shall see, of the in nitely
many surfaces, all but four admit hyperbolic geometry (two admit Euclidean
geometry and two admit elliptic geometry). Thus, if you randomly generate a
constant curvature surface for a two-dimensional bug named Bormit, Bormit
will no doubt live in a world with hyperbolic geometry.
Chances are, too, that a constant curvature surface cannot be embedded in
three-dimensional space. Only the sphere has this nice feature. In fact, if X
is any surface that lives in R3 , like the handlebody surfaces in Figure 7.5.10,
then it must have at least one point with positive curvature. Why is this?
The surface in R3 is bounded, so there must be some sphere in R3 centered
at the origin that contains the entire surface. Shrink this sphere until it just
bumps into the surface at some point. The curvature of the surface at this
point matches the curvature of the sphere, which is positive.
We have seen that any surface is homeomorphic to a polygonal surface
representing Hg or Cg , for some g , and we now show that each of these may be
given a homogeneous, isotropic and metric geometry (so that it has constant
curvature). A polygonal surface can only run into trouble homogeneity-wise
where the corners come together. Any point in the interior of the polygon has
◦
a nice 360 patch of space about it, as does any point in the interior of an
7.6. Geometry of Surfaces 163
edge. However, if the angles of the corners that come together at a point do
◦
not add up to 360 , then the surface has either a cone point or a saddle-point,
◦
depending on whether the angle sum is less than or more than 360 . As we
saw in Chapter 1 (Exercises 1.3.5 and Checkpoint 1.3.7), in either case, we
will not have a homogeneous geometry; a two-dimensional bug would be able
to distinguish (with triangles or circles) a cone point from a at point from a
saddle point.
To smooth out such cone points or saddle points we change the angles at
the corners so that the angles do ◦
add up to 360 . We now have the means for
doing this. If we need to shrink the corner angles, we can put the polygon in
the hyperbolic plane. If we need to expand the angles, we can put the polygon
in the projective plane.
Example 7.6.1: C3 admits hyperbolic geometry.
The standard polygonal representation of C3 = P2 #P2 #P2 , is a
hexagon having boundary label a1 a1 a2 a2 a3 a3 , as in Figure 7.6.2. All
six corners of the hexagon come together at a single point. In the
◦
Euclidean plane, a regular hexagon has corner angles equal to 120 . To
avoid a saddle point when joining the six corners together, shrink the
◦
corner angles to 60 . A tiny copy of a regular hexagon in the hyperbolic
◦
plane will have corner angles just under 120 . If the hexagon grows so
that its vertices approach ideal points, its corner angles will approach
◦ ◦
0 . At some point, then, the interior angles will be 60 on the nose.
(We may construct this precise hexagon as well. See Exercise 7.6.4.) If
C3 is built from this hexagon living in the hyperbolic plane, the surface
inherits the geometry of the space in which it nds itself; that is, the
surface C3 admits hyperbolic geometry, a nice homogeneous, isotropic
and metric geometry.
a2
a2
a3 a2 a3 a2
a3 a1 a3 a1
a1
a1
Figure 7.6.2: C3 admits hyperbolic geometry.
Example 7.6.3: An elliptic polygonal surface.
Revisiting Example 1.3.9, consider the hexagon with boundary label
abcabc. The six corners of this polygonal surface come together in
groups of two. These corners create cone points because the angle sum
◦
of the two corners coming together is less than 360 in the Euclidean
plane. We can avoid cone points by putting the hexagon in P2 . A
small regular hexagon in the projective plane will have corner angles
◦
just slightly greater than 120 , but we need each corner angle to expand
◦
to 180 . We may achieve these angles by expanding the hexagon until it
164 7. Geometry on Surfaces
covers the entire projective plane. In fact, the surface of Example 1.3.9
is the projective plane, and it admits elliptic geometry.
Theorem 7.6.4. Each surface admits one homogeneous, isotropic, and metric
geometry. In particular, the sphere (H0 ) and projective plane (C1 ) admit
elliptic geometry. The torus (H1 ) and Klein bottle (C2 ) admit Euclidean
geometry. All handlebody surfaces Hg with g ≥ 2 and all cross-cap surfaces Cg
with g ≥ 3 admit hyperbolic geometry.
The following section oers a more formal discussion of how any surface
admits one of our three geometries but we present an intuitive argument here.
The sphere and projective plane admit elliptic geometry by construction:
The space in elliptic geometry is the projective plane, and via stereographic
projection, this is the geometry on S2 .
The torus and Klein bottle are built from regular 4-gons (squares) whose
edges are identi ed in such a way that all 4 corners come together at a point.
In each case, if we place the square in the Euclidean plane all corner angles are
π/2, so the sum of the angles is 2π , and our surfaces admit Euclidean geometry.
Each handlebody surfaces Hg for g ≥ 2 and each cross-cap surfaces Cg for
g ≥ 3 can be built from a regular n-gon where n ≥ 6. Again, all n corners
come together at a single point. A regular n-gon in the Euclidean plane has
interior angle (n − 2)π/n radians, so the corner angles sum to (n − 2)π radians.
This angle sum exceeds 2π radians since n ≥ 6. Placing a small version of this
n-gon in the hyperbolic plane, the corner angle sum will be very nearly equal
to (n − 2)π radians and will exceed 2π radians, but as we expand the n-gon
the corners approach ideal points and the corner angles sum will approach
0 radians. Thus, at some point the angle sum will equal 2π radians on the
nose, and the polygonal surface built from this precise n-gon admits hyperbolic
geometry. Exercise 7.6.4 works through how to construct this precise n-gon.
Of course, one need not build a surface from a regular polygon. For instance,
the torus can be built from any rectangle in the Euclidean plane and it will
inherit Euclidean geometry. So while the type of geometry our surface admits
is determined, we have some exibility where certain geometric measurements
are concerned. For instance, there is no restriction on the total area of the
torus, and the rectangle on which it is formed can have arbitrary length and
width dimensions. These dimensions would have a simple, tangible meaning
to a two-dimensional bug living in the surface (and might be experimentally
determined). Each dimension corresponds to the length of a geodesic path
that would return the bug to its starting point. A closed geodesic path in a
surface is a path that follows along a straight line (in the underlying geometry)
that starts and ends at the same point. Figure 7.6.5 shows three closed geodesic
paths, all starting and ending at a point near a bug's house. The length of one
path equals the width of the rectangle, the length of another equals the length
of the rectangle, and the third follows a path that is longer than the rst two.
7.6. Geometry of Surfaces 165
Figure 7.6.5: A at torus has many closed geodesic paths. The length of the
shortest closed geodesic path equals the length of the short side of the rectangle
on which the torus is modeled.
Even in hyperbolic surfaces, where the area of the surface is xed (for a
given curvature) by the Gauss-Bonnet formula, which we prove shortly, there is
freedom in determining the length of closed geodesic paths.
Example 7.6.6: Building hyperbolic surfaces from pants.
If we make three slices in the two-holed torus we obtain two pairs of
pants, as indicated in the following gure.
d1 c2 e1
d2 c1 e2
c2 c2
c1 c1
d2 d2 e2 e2
d1 d1 e1 e1
We label our cuts so that we can stitch up our surface later. Match
the ci , di , and ei edges to recover the two-holed torus. Any pair of pants
can be cut into two hexagons by cutting along the three vertical seams
in the pants. It follows that the two-holed torus can be constructed
from four hexagons, with edges identi ed in pairs as indicated below.
166 7. Geometry on Surfaces
c1 c2
c2
c1 a1 a3 a1 a3
a1 a3
a2
d2 d2 d1 d1 d2 d2
d1 d1 a2 a2
c1 c2
c2
c1 b1 b3 b1 b3
b1 b3
b2
e2 e2 e1 e1 e2 e2
e1 e1
b2 b2
The four hexagons represent a cell division of H2 having 6 vertices,
12 edges, and 4 faces. In the edge identi cation, corners come together
◦
in groups of 4, so we need each corner angle to equal 90 in order to
endow it with a homogeneous geometry. We know we can do this in
the hyperbolic plane. Moreover, according to Theorem 5.4.19, there is
freedom in choosing the dimensions of the hexagons. That is, for each
triple of real numbers (a, b, c) there exists a right-angled hexagon in D
with alternate lengths a, b, and c. So, there exists a two-holed torus for
each combination of six seam lengths (a1 , a2 , a3 , b1 , b2 , and b3 ).
A surface that admits one of our three geometries will have constant
curvature. The reader might have already noticed that the sign of the curvature
will equal the sign of the surface's Euler characteristic. Of course the magnitude
of the curvature (if k 6= 0) can vary if we place a polygonal surface in a scaled
version of P2 or D. That is, while the type of homogeneous geometry a surface
admits is determined by its Euler characteristic (which is determined by its
shape), the curvature scale can vary if k 6= 0. By changing the radius of a
sphere, we change its curvature (though it always remains positive). Similarly,
the surface C3 in Figure 7.6.2 has constant curvature -1 if it is placed in the
hyperbolic plane of Chapter 5. However, it can just as easily nd itself in the
hyperbolic plane with curvature k = −8. Recall, the hyperbolic plane with
curvature k < 0 is modeled
p on the open disk in C centered at the origin with
radius 1/ |k|. Placing the hexagonal representation of C3 into this space so
that its corner angle sum is still 2π produces a surface with constant curvature
k.
We have nally arrived at the elegant relationship between a surface's
curvature k, its area, and its Euler characteristic. This relationship crystalizes
the interaction between the topology and geometry of surfaces.
Theorem 7.6.7 Gauss-Bonnet. The area of a surface with constant curva-
ture k and Euler characteristic χ is given by the formula kA = 2πχ.
√
Proof. The sphere with constant curvature k has radius equal to 1/ k , and
area equal to 4π/k . Since the sphere has Euler characteristic 2, the Gauss-
Bonnet formula holds in this case. The projective plane P2 with curvature k
has area equal to 2π/k , and Euler characteristic equal to 1, so the result holds
in this case as well. The torus and the Klein bottle each have k = 0 and χ = 0,
7.7. Quotient Spaces 167
so in this case the Gauss-Bonnet formula reduces to the true statement that
0 = 0.
Any surface of constant negative curvature cam be represented by a regular
n-sided polygon with n ≥ 6. Furthermore, this polygon can be placed in the
hyperbolic plane with curvature k < 0, so that its interior angles sum to 2π
radians. According to Theorem 7.4.4,
n
kA = 2π − (n − 2)π = 2π 2 − ,
2
where A is the area of the n-gon.
Now, for g ≥ 2 , Hg is represented by a 4g -gon so that
kA = 2π(2 − 2g) = 2πχ(Hg ).
For g ≥ 3, Cg is represented by a 2g -gon so that
kA = 2π(2 − g) = 2πχ(Cg ).
This completes the proof.
Suppose a two-dimensional cosmologist believes she lives in a surface of
constant curvature (because her universe looks homogeneous and isotropic). If
she can deduce the curvature of the universe and its total area, she will know its
Euler characteristic. That is, by measuring these geometric properties she can
deduce the shape of her universe, or, at worst, narrow down the possibilities
to two. If χ is 2 or an odd integer then the cosmologist will know the shape
of the universe. If χ<2 is even, her universe will have one of two possible
shapes - one shape orientable, the other non-orientable.
Exercises
1. Suppose our intrepid team of two-dimensional explorers from Exer-
cise 7.4.4, after an extensive survey, estimates with high con dence that the
2 2
area of their universe is between 800,000 km and 900,000 km . Assuming
their universe is homogeneous and isotropic, what shapes are possible for their
universe? Can they deduce from this information the orientability status of
their universe?
2. Suppose a certain constant curvature 3-holed torus H3 has area 5.2 km .
2
What must be the area of a constant curvature C3 surface so that they have
the same curvature?
3. Building a hyperbolic surface from pairs of pants. To construct Hg for
g ≥ 2 from pairs of pants, how many do we need? Express your answer in
terms of g.
4. Suppose we want to build a regular n-gon in (D, H) from the vertices
k
in 2π
vk = re for k = 0, 1, . . . , n − 1 so that the n interior angles sum to 2π .
p
Prove that this is the case when r = cos(2π/n). Hint: It may be helpful to
refer to Exercise 5.4.9 where we proved this result in the case n = 8.
7.7 Quotient Spaces
We may bend a sheet of paper and join its left and right edges together to
obtain a cylinder. If we let I2 = {(x, y) ∈ R2 | 0 ≤ x ≤ 1, 0 ≤ y ≤ 1} represent
168 7. Geometry on Surfaces
our square piece of paper, and C = {(x, y, z) ∈ R3 | x2 + y 2 = 1, 0 ≤ z ≤ 1}
represent a cylinder, then the map
p : I2 → C by p((x, y)) = (cos(2πx), sin(2πx), y)
models this gluing process.
Figure 7.7.1: Bending a sheet of paper into a cylinder.
This map tries very hard to be a homeomorphism. The map is continuous,
onto, and it is almost one-to-one with a continuous inverse. It fails in this
endeavor only where we join the left and right edges: the points (0, y) and
(1, y) in I2 both get sent by p to the point (1, 0, y). But p is nice enough
to induce a homeomorphism between the cylinder and a modi ed version of
the domain I2 , obtained by dividing out of I2 the mapping redundancies so
that the result is one-to-one. The new version of I2 is called a quotient space.
We develop quotient spaces in this section because all surfaces and candidate
three-dimensional universes can be viewed as quotient spaces. We need the
notion of an equivalence relation on a set. To get this, we need the notion of
a relation.
A relation on a set S is a subset R of S × S . In other words, a relation
R consists of a set of ordered pairs of the form (a, b) a and b are in
where
S . If (a, b) is an element in the relation R, we may write aRb. It is common
to describe equivalence relations, which we de ne shortly, with the symbol ∼
instead of R. So, when you see a ∼ b this means the ordered pair (a, b) is in
the relation ∼, which is a subset of S × S.
De nition 7.7.2. An equivalence relation on a set A is a relation ∼ that
satis es these three conditions:
1. Re exivity: x∼x for all x∈A
2. Symmetry: If x∼y then y∼x
3. Transitivity: If x∼y and y∼z then x ∼ z.
For any element a ∈ A, the equivalence class of a, denoted [a], is the
subset of all elements in A that are related to a by ∼. That is,
[a] = {x ∈ A | x ∼ a}.
Example 7.7.3: An equivalence relation.
De ne z ∼ w in C if and only if Re(z) − Re(w) is an integer and
Im(z) = Im(w). For instance, (−1.6 + 4i) ∼ (2.4 + 4i) since the
dierence of the real parts (-1.6 - 2.4 = -4) is an integer and the
imaginary parts are equal. To show ∼ is an equivalence relation, we
check the three requirements.
7.7. Quotient Spaces 169
1. Re exivity: Given z = a + bi, it follows that z ∼ z because
a−a=0 is an integer and b = b.
2. Symmetry: Suppose z ∼ w. Then Re(z) − Re(w) = k for some
integer k and Im(z) = Im(w). It follows that Re(w) − Re(z) =
−k is an integer and Im(w) = Im(z). In other words, w ∼ z .
3. Transitivity: Suppose z ∼ w and w ∼ v . We must show z ∼ v .
Since z ∼ w, Re(z) − Re(w) = k for some integer k , and since
w ∼ v , Re(w) − Re(v) = l for some integer l. Notice that
k + l = [Re(z) − Re(w)] + [Re(w) − Re(v)]
= Re(z) − Re(v).
So, Re(z) − Re(v) is an integer. Furthermore, we have Im(z) =
Im(w) = Im(v). Thus, z ∼ v.
The equivalence class of a point z = a + bi consists of all points
w = c + bi where a − c is an integer. In other words, c = a + n for some
integer n, so w = z + n and we may express the equivalence class as
[z] = {z + n | n ∈ Z}.
A partition of a set A consists of a collection of non-empty subsets of A
that are mutually disjoint and have union equal to A. An equivalence relation
on a set A serves to partition A by the equivalence classes. Indeed, each
equivalence class is non-empty since each element is related to itself, and the
union of all equivalence classes is all of A by the same reason. That equivalence
classes are mutually disjoint follows from the following lemma.
Lemma 7.7.4. Suppose ∼ is an equivalence relation on A, and a and b are
any two elements of A. Then either [a] and [b] have no elements in common,
or they are equal sets.
Proof. Suppose there is some element c that is in both [a] and [b]. We show
[a] = [b] by arguing that each set is a subset of the other.
That [a] is a subset of [b]: Suppose x is in [a]. We must show that x is in
[b] as well. Since x is in [a], x ∼ a. Since c is in [a] and in [b], c ∼ a and c ∼ b.
We may use these facts, along with transitivity and symmetry of the relation,
to see that x ∼ a ∼ c ∼ b. That is, x is in [b]. Therefore, everything in [a] is
also in [b].
We may repeat the argument above to show that [b] is a subset [a]. Thus,
if [a] and [b] have any element in common, then they are entirely equal sets,
and this completes the proof.
In light of Lemma 7.7.4, an equivalence relation on a set provides a natural
way to divide its elements into subsets that have no points in common. An
equivalence relation on A, then, determines a new set whose elements are the
distinct equivalence classes.
De nition 7.7.5. If ∼ is an equivalence relation on a set A, the quotient
set of A by ∼ is
A/∼ = {[a] | a ∈ A}.
We will be interested in quotients of three spaces: the Euclidean plane C,
the hyperbolic plane D, and the sphere S2 . If we build a quotient set from one
of these spaces, we will call a region of the space a fundamental domain of
the quotient set if it contains a representative of each equivalence class of the
quotient and at most one representative in its interior.
170 7. Geometry on Surfaces
Orbit Spaces
We may construct a natural quotient set from a geometry (X, G).
The group structure of G de nes an equivalence relation ∼G on X as follows:
For x, y ∈ X , let
x ∼G y if and only if T (x) = y for some T ∈ G.
Indeed, for each x ∈ X , x ∼G x because the group G must contain the identity
transformation, so the relation is re exive. Next, if x ∼G y then T (x) = y for
−1 −1
some T in G. But the group contains inverses, so T is in G and T (y) = x.
Thus y ∼G x, and so ∼G is symmetric. Third, transitivity of the relation
follows from the fact that the composition of two maps in G is again in G.
Given geometry (X, G) we let X/G denote the quotient set determined by
the equivalence relation ∼G . In this setting we call the equivalence class of a
point x in X , the orbit of x. So, the orbit of x consists of all points in the
space X to which x can be mapped under transformations of the group G:
[x] = {y ∈ X | T (x) = y for some T ∈ G}.
Put another way, the orbit of x is the set of points in X congruent to x in the
geometry (X, G).
Note that if the geometry G is homogeneous, then any two points in X
are congruent and, for any x ∈ X , the orbit of x is all of X . In this case the
quotient set X/G consists of a single point, which is not so interesting. We
typically want to consider orbit spaces X/G in which G is a small group of
transformations.
We say that a group of transformations G of X is a group of homeo-
morphisms of X if each transformation in G is continuous. In this case, we
call X/G an orbit space . If X has a metric, we say that a group of transfor-
mations of X is a group of isometries if each transformation of the group
preserves distance between points.
Example 7.7.6: Building a topological cylinder.
Consider the horizontal translation T1 (z) = z + 1 of C. This
transformation is a (Euclidean) isometry of C and it generates a group
of isometries of C as follows. Put T1 and T1−1 in the group, along with
any number of compositions of these transformations. Fortunately, any
number of compositions of these two maps results in an isometry that
is easy to write down. Any nite composition of copies of T1 and T1−1
indicates a series of instructions for a point z: at each step in the long
composition z moves either one unit to the left if we apply T1−1 or one
unit to the right if we apply T1 . In the end, z has moved horizontally
by some integer amount. That is, any such composition can be written
as Tn (z) = z + n for some integer n. We let hT1 i denote the group
generated by T1 , and we have
hT1 i = {Tn (z) = z + n | n ∈ Z}.
The orbit of a point p under this group of isometries is
[p] = {p + n | n ∈ Z}.
A fundamental domain for the orbit space C/hT1 i is the vertical
strip consisting of all points z with 0≤ Re(z) ≤ 1, as in the following
gure. Every point in C is related to a point in this shaded vertical
7.7. Quotient Spaces 171
strip. Furthermore, no two points in the interior of the strip are related.
By passing to the quotient, we are essentially rolling up the plane in
to an in nitely tall cylinder. The rolling up process is described by the
map p : C → C/hT1 i given by p(z) = [z].
Example 7.7.7: A quotient space from rotations.
The rotation R π2 of C by π/2 about the origin generates a group of
isometries of C consisting of four transformations. We generate the
group as before, by considering all possible compositions of R π and
2
R−1
π . This group turns out to be nite: Any combination of these
2
rotations produces a rotation by 0, π/2, π , or 3π/2 radians, giving us
hR π2 i = {1, R π2 , Rπ , R 3π
2
}.
The orbit of the point 0 is simply {0} because each transformation
in the group xes 0, but the orbit of any other point in C is a four-
element set. For instance, the orbit of 1 is [1] = {i, −1, −i, 1}.
It turns out that every surface can be viewed as a quotient space of the
form M/G, where M is either the Euclidean plane C, the hyperbolic plane D,
or the sphere S2 , and G is a subgroup of the transformation group in Euclidean
geometry, hyperbolic geometry, or elliptic geometry, respectively. In topology
terminology, the space M is called a universal covering space of the orbit
space M/G.
Example 7.7.8: H1 as quotient of C.
Suppose a and b are positive real numbers. Let hTa , Tbi i be the group of
homeomorphisms generated by the horizontal translation Ta (z) = z + a
and the vertical translation Tbi (z) = z + bi.
This group contains all possible compositions of these two transfor-
mations and their inverses. Thus, the orbit of a point z consists of all
complex numbers to which z can be sent by moving z horizontally by
some integer multiple of a units, and vertically by some integer mul-
tiple of b units. An arbitrary transformation in Γ = hTa , Tbi i has the
form
T (z) = z + (ma + nbi)
where m and n are integers. A fundamental domain for the orbit
space consists of the rectangle with corners 0, a, a + bi, bi. The
resulting quotient space is homeomorphic to the torus. Notice that
points on the boundary of this rectangle are identi ed in pairs. In
fact, the fundamental domain, with its boundary point redundancies,
172 7. Geometry on Surfaces
corresponds precisely to our polygonal surface representation of the
torus.
If the space M has a metric and our group of homeomorphisms is su ciently
nice, then the resulting orbit space inherits a metric from the universal covering
space M. To be su ciently nice, we rst need our homeomorphisms to
be isometries. The group of isometries must also be xed-point free and
properly discontinuous. The group G is xed-point free if each isometry
in G (other than the identity map) has no xed points. The group G is
properly discontinuous if every x in X has an open 2-ball Ux about it
whose images under all isometries in G are pairwise disjoint. The interested
reader is encouraged to see [10] or [9] for more detail. If our group G is a
xed-point free, properly discontinuous group of isometries, then the resulting
orbit space inherits a metric from M.
Consider the quotient space in Example 7.7.7. The group here is a group of
isometries, since rotations preserve Euclidean distance, but it is not xed-point
free. All maps in G have xed points (rotation about the origin xes 0). This
prevents the quotient space from inheriting the geometry of its mother space.
2πr
Indeed, a circle centered at [0] with radius r would have circumference
4 ,
which doesn't correspond to Euclidean geometry.
The group of isometries in the torus example is xed-point free and properly
discontinuous, so the following formula for the distance between two points [u]
and [v] in the orbit space C/hTa , Tbi i is well-de ned:
d([u], [v]) = min{|z − w| | z ∈ [u], w ∈ [v]}.
Figure 7.7.9 depicts two points in the shaded fundamental domain, [u]
and [v]. The distance between them equals the Euclidean distance in C of
the shortest path between any point in equivalence class [u] and any point in
equivalence class [v]. There are many such nearest pairs, and one such pair
is labeled in Figure 7.7.9 where z is in [u] and w is in [v]. Also drawn in the
gure is a solid line (in two parts) that corresponds to the shortest path one
would take within the fundamental domain to proceed from [u] to [v]. This
path marks the shortest route a ship in the video game from Chapter 1 could
take to get from [u] to [v].
7.7. Quotient Spaces 173
[u]
w [v]
Figure 7.7.9: The distance between two points in the torus viewed as a
quotient of C.
Example 7.7.10: P2 as quotient of S2 .
Let Ta : S2 → S2 be the antipodal map Ta (P ) = −P . This map is an
isometry that sends each point on S2 to the point diametrically opposed
−1
to it, so it is xed-point free. Since Ta = Ta , the group generated
by this map consists of just 2 elements: Ta and the identity map. The
2
quotient space S /hTa i is the projective plane.
Example 7.7.11: H2 as quotient of D2 .
We may build a regular octagon in the hyperbolic plane whose interior
angles equal π/4 radians. We may also nd a hyperbolic transformation
that takes an edge of this octagon to another edge. Labelling the edges
as in the following diagram, let Ta be the hyperbolic isometry taking
one a edge to the other, being careful to respect the edge orientations.
We construct such a map by composing two hyperbolic re ections about
hyperbolic lines: the hyperbolic line containing the rst a edge, and the
hyperbolic line m that bisects the b edge between the a edges. Since
the two lines of re ection do not intersect, the resulting map in H is
a translation in H and has no xed points in D (the xed points are
ideal points).
c b
d
a
c m
b
d a
De ne Tb , Tc , and Td similarly and consider the group of isometries
of D generated by these four maps. This group is a xed-point
174 7. Geometry on Surfaces
free, properly discontinuous group of isometries of D, so the resulting
quotient space inherits hyperbolic geometry.
The distance between two points [u] and [v] in the quotient space
is given by
dH ([u], [v]) = min{dH (z, w) | z ∈ [u], w ∈ [v]}.
Geodesics in the quotient space are determined by geodesics in the
hyperbolic plane D.
Topologically, the quotient space is homeomorphic to H2 , and the
octagon pictured above serves as a fundamental domain of the quotient
space. Moving copies of this octagon by isometries in the group
produces a tiling of D by this octagon. Each copy of the octagon would
serve equally well as a fundamental domain for the quotient space.
Figure 7.7.12 displays a portion of this tiling, including a geodesic
triangle in the fundamental domain, and images of it in neighboring
octagons.
c b
d a
c b
d a
Figure 7.7.12: A triangle in H2 , with various images of it.
All surfaces Hg for g≥2 and Cg for g≥3 can be viewed as quotients of D
by following the procedure in the previous example.
Start with a perfectly sized polygon in D. The polygon must have corner
angle sum equal to 2π radians, and the edges that get identi ed must have equal
length so that an isometry can take one to the other. (In every example so
far, we have used regular polygons in which all sides have the same length, but
asymmetric polygons will also work.) Next, for each pair of oriented edges to be
identi ed, nd a hyperbolic isometry that maps one onto the other (respecting
the orientation of the edges). The group generated by these isometries creates
a quotient space homeomorphic to the space represented by the polygon, and it
inherits hyperbolic geometry. Note also that the initial polygon can be moved
by the isometries in the group to tile all of D without gaps or overlaps.
7.7. Quotient Spaces 175
Dirichlet Domain
We end this section with a discussion of the Dirichlet domain, which is an
important tool in the investigation of the shape of the universe. Suppose we
live in a surface described as a quotient M/G where M is either C , D, or S2 ,
and G is a xed-point free and properly discontinuous group of isometries of the
space. For each point x in M de ne the Dirichlet domain with basepoint
x to consist of all points y in M such that
d(x, y) ≤ d(x, T (y))
for all T in G, where it is understood that d(x, y) is Euclidean distance,
hyperbolic distance, or elliptic distance, depending on whether M is C , D,
or S2 , respectively.
At each basepoint x in M , the Dirichlet domain is itself a fundamental
domain for the surface M/G, and it represents the fundamental domain that
a two-dimensional inhabitant might build from his or her local perspective. It
is a polygon in M (whose edges are lines in the local geometry) consisting of
all points y that are as close to x or closer to x than any of its image points
T (y) under transformations in G.
We may visualize a Dirichlet domain with basepoint x as follows. Consider
a small circle in M centered at x. Construct a circle of equal radius about all
points in the orbit of x. Then, begin in ating the circle (and all of its images).
Eventually the circles will touch one another, and as the circles continue to
expand let them press into each other so that they form a geodesic boundary
edge. When the circle has lled the entire surface, it will have formed a polygon
with edges identi ed in pairs. This polygon is the Dirichlet domain.
At any basepoint in the torus of Example 7.7.8 the Dirichlet domain will
be a rectangle identical in proportions to the fundamental domain. In general,
however, the shape of a Dirichlet domain may be dierent than the polygon on
which the surface was built, and the shape of the Dirichlet domain may vary
from point to point, which is rather cool.
b b
i
c c
a 1 a 2
Figure 7.7.13: Building a Klein bottle from a hexagon.
Example 7.7.14: Dirichlet domains in a Klein bottle.
Consider the surface constructed from the hexagon in Figure 7.7.13,
which appeared in Levin's paper on cosmic topology [23]. Assume the
hexagon is placed in C with its six corners at the points 0, 1, 2, 2 + i,
1+i and i.
This polygonal surface represents a cell division of a surface with
three edges, two vertices, and one face. The Euler characteristic is
thus 0, so the surface is either the torus or Klein bottle. In fact, it
is a Klein bottle because it contains a Möbius strip. We may tile the
Euclidean plane with copies of this hexagon using the transformations
T (z) = z + 2i (vertical translation) and r(z) = z + (1 + 2i) (a
176 7. Geometry on Surfaces
transformation that re ects a point about the horizontal axis y = 1
and then translates to the right by one unit). The following gure
shows the shaded fundamental domain A and its images under various
combinations of T and r.
b b b b
c T (A) c T ◦ r2 (A) c
a a a a a a
c r−1 (A) c r(A) c r3 (A) c
b b b b b b b
c A c r2 (A) c r4 (A) c
a
a a a a a a
c T −1 ◦ r−1 (A) c T −1 ◦ r(A) c T −1 ◦ r3 (A) c
b b b b b b
If Γ is the group of transformations generated by T and r, the
quotient space C/Γ is the Klein bottle, and its geometry is Euclidean,
inherited from the Euclidean plane C.
It turns out that the Dirichlet domain at a basepoint in this space
can vary in shape from point to point. Exercise 7.7.4 investigates the
shape of the Dirichlet domain at dierent points.
Exercises
1. Show that the Dirichlet domain at any point of the torus in Example 7.7.8
is an a by b rectangle by completing the following parts.
a. Construct an a by b rectangle to be the fundamental domain, and place
eight copies of this rectangle around the fundamental domain as in Figure 7.7.9.
Then plot a point x in the fundamental domain, and plot its image in each of
the copies.
b. For each image x0 of x, construct the perpendicular bisector of the
segment xx0 . The eight perpendicular bisectors enclose the Dirichlet domain
based at x. Prove that the Dirichlet domain is also an a by b rectangle.
2. Construct C3 as a quotient of D by a group of isometries of D. Be as
explicit as possible when de ning the group of isometries.
3. Explain why the g -holed torus Hg can be viewed as a quotient of D by
hyperbolic isometries for any g ≥ 2.
4. Consulting Example 7.7.14, show that the Dirichlet domain at any point
z on the line Im(z) = 1/2, such as the one in Figure 7.7.15, is a square. Show
that the Dirichlet domain at any point z on the line Im(z) =0 is a rectangle.
7.7. Quotient Spaces 177
[w] [z]
a [w] [w]
[z] [z]
b
[z] [w] z [z]
a [w] w
[w] [w]
[z] [z]
b
[w] [z]
a
Figure 7.7.15: The Dirichlet domain at a point is obtained by considering
perpendicular bisectors with its nearest images, and its shape can vary from
point to point, as at points z and w. In the gure all vertically oriented edges
are c edges. All horizontally oriented edges in a given row have the same label,
either a or b as indicated.
8
Cosmic Topology
Cosmic topology can be described as the eort to determine the shape of
our universe through observational techniques. In this chapter we discuss two
programs of research in cosmic topology: the cosmic crystallography method
and the circles-in-the-sky method. Both programs search for topology by
assuming the universe is nite in volume without boundary. The chapter
begins with a discussion of three-dimensional geometry and some 3-manifolds
that have been given consideration as models for the shape of our universe.
8.1 Three-Dimensional Geometry and 3-Manifolds
Recall that R3 is the set of ordered triples of real numbers (x, y, z), and a
3-manifold is a space with the feature that every point has a neighborhood
that is homeomorphic to an open 3-ball. We assume that the shape of the
universe at any xed time is a 3-manifold. Evidence points to a universe that
is isotropic and homogeneous on the largest scales. If this is the case, then just
as in the two-dimensional case, the universe admits one of three geometries:
hyperbolic, elliptic, or Euclidean. This section oers a brief introduction to the
three-dimensional versions of these geometries before turning to 3-manifolds.
The reader is encouraged to see [12] for a broader intuitive discussion of these
ideas, or [11] for a more rigorous approach.
Euclidean Geometry in Three Dimensions Euclidean geometry is the
geometry of our experience in three dimensions. Planes look like in nite
tabletops, lines in space are Euclidean straight lines. Any planar slice of 3-
space inherits two-dimensional Euclidean geometry.
The space for three-dimensional Euclidean geometry is R3 , and we may use
3
boldface notation v to represent a point in R . The group of transformations in
this geometry consists of all rotations of R3 about lines and all translations by
vectors in R3 . The distance between points v = (x1 , y1 , z1 ) and w = (x2 , y2 , z2 )
is given by the Euclidean distance formula
p
|v − w| = (x1 − x2 )2 + (y1 − y2 )2 + (z1 − z2 )2 .
Any transformation in the group can be expressed as a screw motion. A
screw motion is a transformation of R3 consisting of a translation in the
direction of a line followed by a rotation about that same line. Of course,
rotations about lines and translations are special cases of this more general
map.
178
8.1. Three-Dimensional Geometry and 3-Manifolds 179
Hyperbolic Geometry in Three Dimensions The Poincaré disk model
of hyperbolic geometry may be extended to three dimensions as follows. Let
the space H3 consist of all points inside the unit ball in R3 . That is, let
H3 = {(x, y, z) ∈ R3 |x2 + y 2 + z 2 < 1}.
The unit 2-sphere S2 bounds H3 , and is called the sphere at in nity. Points
3
on the sphere at in nity are called ideal points and are not points in H .
The group of transformations for three-dimensional hyperbolic geometry
is generated by inversions about spheres that are orthogonal to the sphere at
in nity. Inversion about a sphere is de ned analogously to inversion in a circle.
Suppose S is a sphere in R3 centered at v0 with radius r, and v is any point
in R 3
. De ne the point symmetric to v with respect to S to be the point v∗
−
v−
→
on the ray 0v such that
|v − v0 ||v ∗ − v0 | = r2 .
One may prove that inversion about a sphere S will send spheres orthogonal
to S to themselves. So, composing two inversions about spheres orthogonal to
S2∞ generates an orientation preserving transformation of hyperbolic three-
space H3 . The transformation group consists of all such compositions. Lines
in this geometry correspond to arcs of clines in H3 that are orthogonal to the
sphere at in nity (see line L in Figure 8.1.1). These lines are geodesics in H3 .
3
As in the two-dimensional case, Euclidean lines through the origin of H are
also hyperbolic lines. A plane in this geometry corresponds to the portion of a
sphere or Euclidean plane inside H3 that meets the sphere at in nity at right
angles, such as planes P1 and P2 in Figure 8.1.1. If we restrict our attention
to any plane in H3 , we recover the two-dimensional hyperbolic geometry of
Chapter 5.
P1
P2
L S2∞
Figure 8.1.1: One model of hyperbolic space H3 , the open unit 2-ball.
Elliptic Geometry in three dimensions Three-dimensional elliptic geom-
etry is derived from the geometry that the 3-sphere S3 inherits as a subspace
4
of the Euclidean space R . The 3-sphere consists of all points in 4-dimensional
space one unit from the origin:
S3 = {(x, y, z, w) ∈ R4 | x2 + y 2 + z 2 + w2 = 1}.
Great circles in S3 are circles of maximum diameter drawn in the space.
Great circles correspond to geodesics in the space. Similarly, a great 2-sphere
in S3 is a 2-sphere of maximum diameter drawn in the space. As a subspace of
180 8. Cosmic Topology
S3 , a great 2-sphere inherits the elliptic geometry of Chapter 6. The 3-sphere
is discussed in more detail in Example 8.1.3.
The transformation group for elliptic geometry in three dimensions is
conveniently described by viewing R4 as the set of quaternions. A quaternion
has the form a + bi + cj + dk where a, b, c, d i, j, and k
are real numbers and
are imaginary numbers with the feature that i2
= −1 and the
= j2 = k2
additional property that the product ijk = −1. The constant a is called the
scalar term
√ of the quaternion q = a + bi + cj + dk. The modulus of q
is |q| = a2 + b2 + c2 + d2 . A quaternion q is called a unit quaternion , if
|q| = 1. The conjugate of q , denoted q ∗ , is the quaternion q ∗ = a−bi −cj −dk.
∗
One can check that q · q = |q|2 . The 3-sphere then consists of all unit
quaternions.
The transformation group for the geometry on S3 is generated by multipli-
cation on the left and/or right by unit quaternions. For xed unit quaternions
u and v, the map T : S3 → S3 by T (q) = uqv is a typical transformation in
three-dimensional elliptic geometry. These transformations preserve antipodal
points on the 3-sphere - distinct points on S3 that are on the same Euclidean
4
line through the origin in R .
As in the two-dimensional case, three-dimensional Euclidean, hyperbolic,
and elliptic geometries are homogenous, isotropic and metric. Furthermore,
triangles distinguish the geometries: in Euclidean space R3 , a triangle has
angle sum equal to π radians; in S3 any triangle angle sum exceeds π radians;
3
in H any triangle angle sum is less than π radians. As a result, if we place
a solid like a dodecahedron in H3 (so that the faces are portions of planes
3
in H ), the angles at which the corners come together are smaller than the
corner angles of the Euclidean dodecahedron. These angles shrink further as
the corner points of the solid approach ideal points on the sphere at in nity.
On the other hand, placing the dodecahedron in S3 will increase the angles at
the corners.
This exibility is very useful. Whereas surfaces can be constructed
from polygons by identifying their edges in pairs, many 3-manifolds can be
constructed from a 3-complex (such as the Platonic solids) by identifying its
faces in pairs. A 3-manifold built in this way will admit either Euclidean
geometry, hyperbolic geometry, or elliptic geometry depending on how, if at
all, the corner angles need to change so that the corners come together to
form a perfect patch of three-dimensional space. Now we investigate a few
3-manifolds and the geometry they admit.
Example 8.1.2: The 3-torus.
We may think of the 3-torus T3 as a cubical room with the following
face (or wall) identi cations: if we propel ourselves (with a jet pack)
up through the ceiling, we reappear through the oor directly below
where we had been. For instance, the elliptical points in the following
gure are identi ed. Each point on the top face is identi ed with the
corresponding point on the bottom face. If we y through the left wall
we reappear through the right wall at the corresponding spot (see the
triangles in the gure). If we y through the front wall we reappear
through the back wall at the corresponding spot (see the squares in the
gure).
8.1. Three-Dimensional Geometry and 3-Manifolds 181
The 3-torus is the three-dimensional analogue of the torus video
screen we saw on the rst pages of this text. In fact, a slice of the
3-torus in a plane parallel to one of the faces is a torus. One checks
that under the face identi cation, all eight corners come together in
a single point. The angles at these corners are such that they form a
perfect patch of three-dimensional space. We do not need to in ate
the corner angles (in elliptic space) or shrink the angles (in hyperbolic
space) in order for the corners to come together perfectly.
We can also view the 3-torus as an orbit space as discussed in
Section 7.7. In particular, let Tx , Ty , Tz represent translations of R3 by
one unit in the x-direction, y -direction, and z -direction, respectively.
That is, Tx (x, y, z) = (x + 1, y, z), Ty (x, y, z) = (x, y + 1, z) and
Tz (x, y, z) = (x, y, z +1). These three transformations generate a group
3
of transformations Γ of R , and any transformation in the group has
the form T (x, y, z) = (x + a, y + b, z + c) where a, b, c are integers. A
3
fundamental domain of the quotient space R /Γ is the unit cube
I3 = {(x, y, z) | 0 ≤ x, y, z ≤ 1},
and all the images of this cube under the transformations in Γ tile R3 .
3
The 3-torus is a quotient of R generated by Euclidean isometries, and
it inherits Euclidean geometry.
The 3-torus is one of just ten compact and connected 3-manifolds admiting
Euclidean geometry. Of these ten, six are orientable, and four are non-
orientable. The 3-torus is orientable, and the other ve orientable Euclidean
manifolds are presented below in Example 8.1.7 and Example 8.1.9.
The elliptic 3-manifolds have also been classi ed. There are in nitely many
dierent types, and it turns out they are all orientable. The 3-sphere is the
simplest elliptic 3-manifold, and Albert Einstein assumed the universe had this
shape when he rst solved his equations for general relativity. He found a static,
nite, simply connected universe without boundary appealing for aesthetic
reasons, and it cleared up some paradoxes in physics that arise in an in nite
universe. However, the equations of general relativity only tell us about the
local nature of space, and do not x the global shape of the universe. In 1917,
the Dutch astronomer Willem De Sitter (1872-1934) noticed that Einstein's
solutions admitted a another, dierent global shape: namely three-dimensional
elliptic space obtained from the 3-sphere by identifying antipodal points, just
as we did in Chapter 6 in the two-dimensional case.
Example 8.1.3: The 3-sphere.
Recall, the 3-sphere consists of all points in 4-dimensional space 1 unit
from the origin:
S3 = {(x, y, z, w) | x2 + y 2 + z 2 + w2 = 1}.
One model of S3 consists of two solid balls in R3 whose boundary
2-spheres are identi ed point for point, as in the following gure. The
182 8. Cosmic Topology
gure shows a path from point a to point b in S3 that goes through the
point p on the boundary of the left-hand solid ball before entering the
right-hand solid ball through the point on its boundary with which p
has been identi ed (also called p).
p p
The 3-sphere has interesting features, often accessible by analogy
with the 2-sphere. If you slice the 2-sphere with a plane in R3 , the
intersection is a circle (or a single point if the plane is tangent to the
2-sphere). Moreover, the circle of intersection is a geodesic on the 2-
sphere if it is a great circle: a circle of maximum radius drawn on the
2-sphere. Bumped up a dimension, if you slice the 3-sphere with R3 ,
the intersection is a 2-sphere (or a single point if the slice is tangent to
S3 ). Moreover, the 2-sphere of intersection is a great 2-sphere if it is
a 2-sphere of maximum diameter drawn in S3 . The boundary 2-sphere
common to both solid balls in the gure is a great 2-sphere of S3 .
3
One may also view S by stereographically projecting it into the
space obtained from R3 by adding a point at ∞, much as we identi ed
S2 with the extended plane via stereographic projection in Section 3.3.
Let c3
R denote real 3-space with a point at in nity attached to it. To
construct the stereographic projection map, it is convenient to rst
express R3 in terms of pure quaternions. A pure quaternion is a
quaternion q whose scalar term is 0. That is, a pure quaternion has
form q = bi + cj + dk. The point (x, y, z) in R is identi ed with
3
the
the pure quaternion q = xi + y j + z k.
Though we are working in four-dimensional space, Figure 3.3.4
serves as a guide in the construction of the stereographic projection
map. Let N = (0, 0, 0, 1) be the north pole on S3 . If P = (a, b, c, d) is
3
any point on S other than N , construct the Euclidean line through N
and P. This line has parametric form
L(t) = h0, 0, 0, 1i + tha, b, c, d − 1i.
We de ne the image of P under stereographic projection to be the
intersection of this line with the three-dimensional subspace consisting
of all points (x, y, z, 0) in R4 . This intersection occurs when 1 + t(d −
1) = 0, or when
1
t= .
1−d
Then the stereographic projection map φ : S3 → R
c3 is given by
i + 1−d j + 1−d k
(
a b c
1−d if d 6= 1;
φ((a, b, c, d)) =
∞ if d = 1.
8.1. Three-Dimensional Geometry and 3-Manifolds 183
In this way, we can transfer the geometry of the 3-sphere into
real 3-space with a point at in nity attached to it. For instance,
if P = (a, b, c, d) is on the 3-sphere, its antipodal point is −P =
(−a, −b, −c, −d) and one can show (in the exercises) that these points
3
get mapped by φ to points in R (pure quaternions) q = φ(P ) and
u = φ(−P ) with the property that q · u∗ = −1. As a result, two points
c3 are called antipodal points if they satisfy the equation
q and u in R
∗
q · u = −1.
Example 8.1.4: The Poincaré dodecahedral space.
The Poincaré dodecahedral space has been given consideration
as a model for the shape of our universe (see [25]). Start with a
dodecahedron living in R3 (we're thinking of it as a solid now, not
a surface) and identify opposite faces with a one-tenth clockwise twist
(rotation by 2π/10 radians). It is cumbersome to indicate all face
identi cations, so the gure below shows the identi cation of the front
face with the back face. For instance, the two corners labeled with a 1
get matched.Notice: no points in the interior of the dodecahedron get
identi ed.
With this face identi cation the twenty corners of the dodecahedron
come together in ve groups of four (one group of 4 is marked with
vertices in the gure), and the angles are too small for the corners
to create a full open ball of three-dimensional space when they come
together. Placing the dodecahedron in elliptic space in ates the corner
angles, and we may make these angles fat enough to determine a perfect
three-dimensional patch of space.
Extending the ideas of Section 7.7 to this case, one can show that the
Poincaré dodecahedral space space is a quotient of the 3-sphere S3 by a
group of isometries. The dodecahedron pictured here is a fundamental
domain of the space. Just as the polygons in Section 7.7 tiled the space
in which they lived, and the cubical fundamental domain of the 3-torus
tiles R3 , the dodecahedron tiles S3 . Although far from obvious, it turns
3
out that S is tiled by 120 copies of this dodecahedron!
Hyperbolic 3-manifolds have not been completely classi ed, but a remark-
able relationship exists between geometry and shape in this case: a connected
orientable 3-manifold supports at most one hyperbolic structure. This marks
an important dierence from the two-dimensional case. If a two-dimensional
being knows that she lives in a two-holed torus, say, then the area of her
universe is determined by its curvature, thanks to the Gauss-Bonnet formula,
but other geometric properties may vary, such as the length of minimal-length
184 8. Cosmic Topology
closed geodesic paths in the universe, as we saw in Example 7.6.6. No such
freedom exists in the three-dimensional case.
Here is one hyperbolic 3-manifold, also built as a quotient of the dodeca-
hedron.
Example 8.1.5: The Seifert-Weber space.
Identify opposite faces of the dodecahedron with a three-tenths clock-
wise turn. With this identi cation, all twenty corners of the dodeca-
hedron come together to a single point, and no such identi cation is
possible unless we drastically shrink the corner angles. We may do
this by placing the dodecahedron in hyperbolic space. The resulting
hyperbolic 3-manifold is called the Seifert-Weber space .
1 1
Example 8.1.6: Lens spaces.
The lens spaces L(p, q) form an in nite family of elliptic 3-manifolds.
Assume p>q are positive integers whose largest common factor is 1.
The lens spaces may be de ned as follows. Consider the unit solid ball
living in R3 whose boundary is the unit 2-sphere. The ball consists
of all points (x, y, z) such that x2 + y 2 + z 2 ≤ 1. Any point on the
boundary 2-sphere can be expressed in coordinates (z, t) where z is a
complex number, t is real, and |z|2 + t2 = 1. The lens space L(p, q)
is obtained from the solid ball by identifying each point (z, t) on the
boundary 2-sphere with the point(e(2πq/p)i z, −t). Note that the north
pole of the boundary 2-sphere is (0, 1) and it is identi ed with the
south pole (0, −1). Any other point u in the northern hemisphere is
identi ed with a single point in the southern hemisphere. This point
is found by re ecting u across the xy -plane, then rotating around the
z -axis by 2πq/p radians. Each point on the equator is identi ed with
p − 1 other points.
The lens space L(p, q) can be obtained from a cell division of the
unit solid ball. The cell division has p + 2 vertices: the north pole n,
the south pole s, and p equally spaced vertices along the equator, label
them v0 , v1 , · · · , vp−1 . The following gure depicts this scene for p = 5.
There is an edge connecting vi to each of its neighbors on the equator,
creating p edges around the equator. There is an edge connecting each
vi with n and an edge connecting each vi with s, as pictured, for a
total of 3p edges. The vertices and edges create 2p triangular faces on
the boundary sphere, half of them in the northern hemisphere. The
3-complex has a single 3-cell, corresponding to the interior of the solid
ball.
8.1. Three-Dimensional Geometry and 3-Manifolds 185
v0 v4 v3
v1 v2
s
L(5, 2)
Let Ni denote the face in the northern hemisphere whose vertices are
vi , vi+1 , n. Let Si be the face in the southern hemisphere whose vertices
are vi , vi+1 , and s. The lens space L(p, q) is created by identifying face
Ni with face Si+q where the sum i + q is taken modulo p. For instance,
in L(5, 2), the shaded face N0 in the gure is identi ed with the shaded
face S2 (and the circle of points pictured in N0 gets sent to the circle
of points in S2 ). The faces N1 and S3 are identi ed, as are N2 with S4 ,
N3 with S0 and N4 with S1 .
Example 8.1.7: The Hantschze-Wendt manifold.
This interesting manifold can be constructed from two identically sized
cubes that share a face, as in the following gure. The resulting box
has 10 faces to be identi ed in pairs. A three-dimensional bug with
a jet pack living in the manifold would experience the following face
identi cations. The top and bottom faces are identi ed just as they
were in the 3-torus. Each point on the top back face (see the circle
in the gure) is identi ed with a single point on the bottom back
face. This point is the re ection of the given point across the vertical
segment bisecting the face. The top front face and the bottom front
face are identi ed the same way. The gure shows two square points
on these faces that are identi ed. The top left face and bottom right
◦
face are identi ed with a 180 rotation (see the elliptical point and
its image point), as are the bottom left face and the top right face
(the two triangular points are identi ed). This 3-manifold also inherits
Euclidean geometry.
A more elegant description of this manifold makes use of the orbit space
construction of Section 7.7. Consider a unit cube sitting in R3 . Let T1 be
186 8. Cosmic Topology
◦
the following Euclidean transformation: Rotate by 180 about segment 1 in
Figure 8.1.8(a) and then translate along the length of that segment. The
original cube and its image are pictured in Figure 8.1.8(b). Let T2 be de ned
similarly using segment 2 of Figure 8.1.8(a). The original cube and its image
under T2 are pictured in Figure 8.1.8(c). The transformation T3 is de ned
the same way using segment 3, and Figure 8.1.8(d) depicts its eect on the
original cube. Each of these transformations is a screw motion. In particular,
◦
each screw motion consists of rotation by 180 and translation by unit length.
The Hantzsche-Wendt manifold is then the quotient of R3 by the group of
Euclidean isometries generated by the three screw motions, and it inherits
Euclidean geometry.
(3)
(2) (1)
(a) (b)
(c) (d)
Figure 8.1.8: Describing the Hantzsche-Wendt Manifold via a group of
transformations.
By repeated applications of the three transformations and their inverses one
may produce image cubes covering half of R3 in a checkerboard pattern. The
3
other points in R are themselves grouped into cubes and form the other half of
the checkerboard pattern. It follows that the ten-faced solid in Example 8.1.7
is a fundamental domain of the manifold, and the face identi cations are
determined by the transformation group.
Example 8.1.9: Four more Euclidean 3-manifolds.
Two orientable 3-manifolds arise from slight changes to the 3-torus
construction of Example 8.1.2. The quarter turn manifold results
◦
when the front and back faces of a cube are identi ed with a 90
rotation, while the remaining 2 pairs of opposite faces are identi ed
directly, as they were for the 3-torus. Figure 8.1.10(a) depicts this
8.1. Three-Dimensional Geometry and 3-Manifolds 187
identi cation. The F on the front face is identi ed with an F on the
back face by a one-quarter turn.
The half turn manifold results when the front and back faces of
◦
a cube are identi ed with a 180 rotation, while the remaining 2 pairs
of opposite faces are identi ed directly, as they were for the 3-torus.
See Figure Figure 8.1.10(b).
The remaining two connected, compact, orientable Euclidean 3-
manifolds have a hexagonal prism as the fundamental domain. A
hexagon can be used to tile the plane, and copies of a hexagonal prism
as in Figure 8.1.10(c) can tile R3 . Note that the prism has two hexag-
onal faces and the other six faces are parallelograms. The one-third
turn manifold is obtained as follows: directly identify parallelograms
that are opposite one another, and identify the two hexagonal faces
◦
with a rotation of 120 , as in Figure 8.1.10(c). Theone-sixth turn
manifold is obtained by identifying the parallelograms as before, but
◦
the two hexagonal faces by a rotation of just 60 , as in Figure 8.1.10(d).
More detailed discussions of these and the non-orientable Euclidean 3-
manifolds can be found in [12] and [16].
F
F
F F
(a) (b)
F
F
120◦ 60◦
F
(c) (d)
Figure 8.1.10: Four Euclidean 3-manifolds: (a) the quarter turn manifold;
(b) the half turn manifold; (c) the one-third turn manifold; and (d) the one-
sixth turn manifold.
Exercises
188 8. Cosmic Topology
1. Investigating quaternions.
a. Show that ij = k, jk = i, and ki = j.
b. Show that ji = −k, kj = −i, and ik = −j.
The conjugate of a quaternion q = a + bi + cj + dk is q = a − bi − cj − dk.
∗
c.
√
The modulus of q is |q| = a2 + b2 + c2 + d2 . Prove that q · q ∗ = |q|2 for any
quaternion.
d. Prove that |uv| = |u| · |v| for any quaternions. Thus, the transformations
in three-dimensional elliptic geometry, which have the form T (q) = uqv where
u, q, v are unit quaternions, do send a point of S3 to a point on S3 .
2. Suppose P = (a, b, c, d) and −P = (−a, −b, −c, −d) are diametrically
opposed points on S3 . Prove that
φ(P ) · φ(−P )∗ = −1,
where φ : S3 → R
c3 is stereographic projection.
3. Verify that in the 3-torus of Example 8.1.2, all eight corners come together
at a single point.
4. Verify that in the Poincaré dodecahedral space of Example 8.1.4, the
corners come together in ve groups of four. Specify the ve groups.
5. Verify that in the Seifert-Weber space of Example 8.1.5, all 20 corners
come together in a single point.
6. What happens if you identify each pair of opposite faces in a cube with
a one-quarter twist? Convince yourself that not all 8 corners come together at
a single point, so that the result is not a Euclidean 3-manifold.
8.2 Cosmic Crystallography
Imagine once again that we are two-dimensional beings living in a two-
dimensional universe. In fact, suppose we are living in the torus in Figure 8.2.1
at point E (for Earth). Our world is homogeneous and isotropic, and adheres
to Euclidean geometry. Our lines of sight follow Euclidean lines. If we can
see far enough, we ought to be able to see an object, say G (for galaxy), in
dierent directions. Three dierent lines of sight are given in the gure.
(2)
G
(3)
(3) (1)
E
(2)
Figure 8.2.1: Seeing multiple images of the same object in the torus.
In fact, if we suppose for a minute that we can see as far as we wish, then
we would be able to see G by looking in any direction that produces a line of
sight with rational slope. In reality, we can't see forever, and this limitation
produces a visual boundary. We will let robs denote the distance to which
we can see, which is the radius of our observable universe. To have any hope
8.2. Cosmic Crystallography 189
of seeing multiple images of the same object, the diameter of our observable
universe, 2robs , must exceed some length dimension of the universe.
Getting back to the torus, the easiest way to nd the directions in which
one can view G is to tile the plane with identical copies of the torus. Place the
Earth at the same point of each copy of the rectangle, and the same goes for
other objects such as G. Figure 8.2.2 displays a portion of the tiling, and our
visual boundary. According to Figure 8.2.2, in addition to the instance of G
in the fundamental domain, 5 of its images would be visible.
G G G
E E E
(2)
G (3) (1)
G G
E E E
robs
G G G
E E E
visual boundary
Figure 8.2.2: Detecting multiple images with a tiling of the universal
covering.
Practically speaking, detecting multiple images of the same object is
complicated by the nite speed of light. Since the lines of sight in Figure 8.2.2
have dierent lengths, we see the object G at dierent times in its evolution.
In reality, galaxies evolve dramatically over time. So even if we found an image
of G looking in some longer direction, it might look so dierent we wouldn't
recognize it.
Still, humanity has pondered the tantalizing possibility that one of the
many distant galaxies we've detected with our telescopes is actually the Milky
Way galaxy. Based on recent lower-bound estimates of the size of our universe,
however, it is now clear that we will not be treated to such a sight.
Rather than spotting dierent images of a particular object, perhaps we
can detect multiple images of the same object indirectly. Consider a catalog
of similar objects that don't evolve too rapidly (such as galaxy superclusters).
Assume we have observed N such objects and they appear to be sprinkled
randomly about the universe (the universe is homogeneous and isotropic after
all). In Figure 8.2.3 we have generated a two-dimensional random distribution
of point sources.
190 8. Cosmic Topology
10
10 5 0 5 10
10
Figure 8.2.3: The positions of N objects in a catalog.
Rather than hunt for two copies of the same source, we compute the distance
between each pair of sources in the catalog (we must make an assumption about
the geometry of space to compute these distances), giving us N (N − 1)/2
distances. If the catalog contains no repeat images, then the distances ought
to follow a Poisson probability distribution. The histogram of the N (N − 1)/2
distances is called a pair separation histogram (PSH) and is given in
Figure 8.2.4.
Pair Separation Histogram
relative
frequency
0
2R
distance
Figure 8.2.4: A pair separation histogram in a simply connected universe.
Now consider the catalog in Figure 8.2.5(a). As in Figure 8.2.3, the
observable radius has been scaled to 10 units. The objects we can observe
may look evenly distributed, but in fact there are multiple images of the same
object. In this simulation the universe is a torus and the observable radius
exceeds the dimensions of our universe. Placing ourselves at the origin in this
catalog, our Dirichlet domain in this torus universe has been superimposed on
the catalog in Figure 8.2.5(b).
10 10
5 5
10 5 0 5 10 10 5 0 5 10
5 5
10 10
(a) (b)
Figure 8.2.5: The positions of N objects in a catalog.
8.2. Cosmic Crystallography 191
The pair separation histogram for this simulated catalog appears in
Figure 8.2.6. Notice the spikes in the histogram. Some distances are occurring
with higher frequency than one would expect by chance alone. At rst glance,
it looks like the spikes occur at distances of about 10, 14, and 17 units.
Pair Separation Histogram for a flat torus universe
0 5 10 15 20
Figure 8.2.6: A pair separation histogram in a torus universe.
What causes these spikes? Look at the catalog plot again in Figure 8.2.5(b),
and nd an object near the top edge of the fundamental domain. We've
highlighted a group of objects that look a bit like a sled. There is a copy
of this object just below the bottom edge of the fundamental domain. The
distance between a point in this sled and its image below equals the length of
the width dimension of our torus, which is 14 in this simulation. Indeed, there
are lots of points for which we see an image displaced vertically in this manner,
so this distance will occur lots of time in the pair separation histogram.
Another copy of our big sled appears just to the right of the right edge of
the fundamental domain. So, the distance between a point and its horizontally
displaced image will be equal to the length dimension of the torus, which is
10 in this simulation. There are also lots of these pairs in the catalog - in fact
more than before since this dimension of the fundamental rectangle is smaller.
This accounts for the larger spike in the histogram above the distance 10.
Finally, there are some objects in the fundamental domain for which
a diagonally displaced copy is visible. The length of this diagonal is
√
102 + 142 ≈ 17.2, and this accounts for the third, smallest spike in the
histogram.
Spikes appear in the PSH precisely because the torus universe C/Γ is
constructed from isometries that move each point in the space by the same
distance. A transformation T of a metric space with the property that
d(p, T (p)) = d(q, T (q))
for all points p and q in the space is called a Cliord translation . Any
translation in C is a Cliord translation; every point gets moved the same
distance. However, a rotation about the origin is not: the further a point is
from the origin, the further it moves. In the exercises you prove that non-trivial
isometries in the hyperbolic plane are not Cliord translations.
Recall that to create the PSH of a catalog in the cosmic crystallography
method we must rst make an assumption about the geometry of the universe.
The PSH in Figure 8.2.6 was generated by computing the Euclidean distances
between points in the catalog. If we assumed a hyperbolic universe and used
the hyperbolic metric to produce the PSH, the spikes would vanish.
In essence, the method of cosmic crystallography pours over catalogs of
astronomical objects, computing distances between all pairs of objects in the
catalog, and then looking for spikes in the pair separation histogram. The
192 8. Cosmic Topology
detection of a spike in the histogram that cannot be reasonably explained by
chance in the distribution of objects indicates a nite universe.
The method of cosmic crystallography has limitations beyond the obvious
challenge of accurately measuring astronomical distances. As you will see in
the exercises, dierent shapes might produce identical spikes, so nding a spike
in a PSH doesn't precisely determine the shape of our universe (though it would
certainly narrow down the list of candidates). All ten Euclidean 3-manifolds
will reveal a spike (or spikes) in the PSH if we can see far enough to detect the
nite dimensions. Some of the elliptic 3-manifolds will produce spikes in the
PSH, but no hyperbolic 3-manifold would reveal itself by this method, since
hyperbolic isometries are not Cliord translations.
To date, no statistically signi cant spikes have been found in the pair
separation histograms computed from real catalogs. Two good surveys of this
method, including information about generating simulated catalogs, can be
found in [26] and [24].
More sensitive methods have been proposed that might detect any 3-manifold,
regardless of geometry, and we look at one such method below.
Example 8.2.7: Collecting correlated pairs.
In a catalog of images, there are two types of pairs that might generate
recurring distances. The cosmic crystallography method outlined
previously detects what have been called Type II pairs in the literature:
A Type II pair is a pair of points of the form {p, T (p)}, where T is
a transformation from the group of isometries used to generate the
manifold. If T transforms each point the same distance (i.e., if T is
a Cliord translation), then this common distance will appear in the
PSH as a spike.
The other type of recurring distance can arise from what's been
called a Type I pair of points in the catalog. A Type I pair consists of
any pair {p, q} of points. If we can see images of these points in a copy
of the fundamental domain, say T (p), T (q), then since transformations
preserve distance, d(p, q) = d(T (p), T (q)), and this common distance
will have occurred at least twice in the PSH. The gure below shows
a portion of the torus tiling of the plane depicted in Figure 8.2.2. The
two types of pairs of points are visible: Type I pairs are joined by
dashed segments, and Type II pairs are joined by solid segments.
I T (q)
T (p)
II
II
q
p I
Type I pairs will not produce discernible spikes in the PSH. Even
in simulations for which several images of a pair of points are present,
8.2. Cosmic Crystallography 193
the spike generated by this set of pairs having the same distance is not
statistically signi cant.
The collecting correlated pairs (CCP) method, outlined below,
attempts to detect the Type I pairs in a catalog.
Suppose a catalog has N objects, and let P = N (N − 1)/2 denote
the number of pairs generated from this set. Compute all P distances
between pairs of objects, and order them from smallest to largest. Let
∆i denote the dierence between the (i + 1)st distance and the ith
distance. Notice ∆i ≥ 0 for all i, and i runs from 1 up to P − 1.
Now, ∆i = 0 for some i if two dierent pairs of objects in the catalog
have the same separation. It could be that unrelated pairs happen to
have the same distance, or that the two pairs responsible for ∆i = 0 are
of the form {p, q} and {T (p), T (q)} (Type I pairs). (We're assuming
there are no type II pairs in the catalog.)
Let Z equal the number of the ∆i 's that equal zero. Then
Z
R=
P −1
denotes the proportion of the dierentials that equal zero. This single
number is a measure, in some sense, of the likelihood of living in a
multiconnected universe.
In a real catalog involving estimations of distances, one wouldn't
expect Type I pairs to produce identical distances, so instead of using
Z as de ned above, one might let Z equal the number of the ∆i 's that
are less than , where is some small positive number. For more details
on this method, see [28].
Exercises
1. The Klein Bottle. We may view the Klein bottle as a quotient of C by
the group of isometries generated by T1 (z) = z + i and T2 (z) = z + 1 + i. A
fundamental domain for the quotient is the unit square in C. The edges of the
square are identi ed as pictured. The a edges are identi ed as they would be
for a torus, but the b edges get identi ed with a twist.
a
i
b b
a 1
Figure 8.2.8: Building the Klein bottle as a quotient of C.
a. Verify that T1 maps the bottom edge of the unit square to the top edge
of the unit square, and that T2 maps the left edge of the unit square to the
right edge with a twist.
b. Determine the inverse transformations T1−1 and T2−1 .
c. We may compose any number of these four transformations T1 , T2 , T1−1 ,
−1
and T2 to tile all of C with copies of the unit square. In the following gure
we have indicated in certain squares the transformation (built from the four
above) that moves the unit square into the indicated square. Complete the
194 8. Cosmic Topology
gure below by indicating a composition of the four transformations that maps
the unit square to the indicated square in C.
T12
T1 T1 ◦ T2
i
T2−1 T2 T22
0 1
T1−1 T1−1 ◦ T2
d.Verify that T2 ◦ T1 = T1−1 ◦ T2 .
e. Show that T1 is a Cliord translation of C but T2 isn't. It follows that
in a Klein bottle universe, cosmic crystallography would only be able to detect
one dimension of the Klein bottle.
2. Figure 8.2.9 shows a simulated catalog from a two-dimensional Euclidean
universe with the corresponding pair separation histogram. There are two
Euclidean surfaces: the Klein bottle and the torus. Based on the cosmic
crystallography analysis, what do you think is the shape of the universe?
Explain your answer. Also, make an estimate at the total area of the universe.
Catalog of Standard Candles
10
Pair Separation Histogram
10 5 0 5 10
0 5 10 15 20
10
Figure 8.2.9: A catalog of images for a two-dimensional Euclidean universe,
with corresponding PSH.
3. Prove that the transformation Ta in Example 7.7.11 used to generate
the two-holed torus as a quotient of D is not a Cliord translation. Can you
generalize the argument to show that any isometry in D that takes one edge
of a regular n-gon to another edge is not a Cliord translation?
4. A rough estimate for the number of images of an object one might see
in a catalog can be made by dividing the volume of space occupied by the
catalog by the volume of the Dirichlet domain at our position in the universe.
Suppose we live in an orientable two-dimensional universe. In fact, suppose we
8.3. Circles in the Sky 195
live in Hg for some g ≥ 2, and our Dirichlet domain is the standard 4g -gon as
in Figure 7.5.24. Set the curvature of the universe to k = −1.
a. According to Gauss-Bonnet, what is the area of the Dirichlet domain?
b. What is the area of the observable universe, as a function of robs ? That
is, what is the area of a circle in (D, H) with radius robs ?
c. Determine the ratio of the area of the observable universe (A(O.U.)) to
the area of the fundamental domain (A(F.D.)). Your ratio A(O.U.)/A(F.D.)
will depend on robs and g.
d. Complete the following table in the case g = 2. Assume robs has units
in light-years.
robs A(O.U.) A(F.D.) Ratio
2
4
6
8
Table 8.2.10: Estimating the number of images of an object one might see
in a catalog
e. Repeat part (d) in two other cases: g=4 and g = 6.
5. Repeat the previous exercise in the case of the non-orientable two-
dimensional universe, Cg .
8.3 Circles in the Sky
Immediately after the big bang, the universe was so hot that the usual
constituents of matter could not form. Photons could not move freely in
space, as they were constantly bumping into free electrons. Eventually, about
350,000 years after the big bang, the universe had expanded and cooled to
the point that light could travel unimpeded. This free radiation is called
the cosmic microwave background (CMB) radiation, and much of it is still
travelling today. The universe has cooled and expanded to the point that this
radiation has stretched to the microwave end of the electromagnetic spectrum,
having a wavelength of about 1 or 2 millimeters.
The CMB radiation is coming to us from every direction, and it has all
been travelling for the same amount of time - and at the same speed. This
means that it has all traveled the same distance to reach us at this moment.
Thus, we may think of the CMB radiation that we can detect at this instant
as having come from the surface of a giant 2-sphere with us at the sphere's
center. This giant 2-sphere is called the last scattering surface (LSS).
It is perhaps comforting to think that everyone in the universe has their
own last scattering surface, that everyone's LSS has the same radius, and that
this radius is growing in time.
The CMB radiation coming to us has a temperature that is remarkably
uniform: it is constant to a few parts in 100,000, which makes the temperature
of the radiation in the LSS very nearly perfectly uniform. As Craig Hogan
points out in [14], this is much smoother than a billiard ball. Nonetheless,
there are slight variations in the temperature. These variations, due to slight
imbalances in the distribution of matter in the early universe, were predicted
well before they were nally found (when our instruments became sensitive
196 8. Cosmic Topology
enough to detect them). These very slight temperature dierences might reveal
the shape of the universe.
Imagine our universe is a giant 3-torus. Assume a fundamental domain for
the universe is a rectangular box as shown in Example 8.1.2, and that this box
is our Dirichlet domain (we're at the center of this box). We may tile R3 with
copies of this fundamental domain, placing ourselves in the same position of
each copy of the fundamental domain. Now, imagine our last scattering surface
in the fundamental domain. In fact, there will be a copy of our last scattering
surface surrounding each copy of us in each copy of the fundamental domain.
If our last scattering surface is small relative to the size of the fundamental
domain, as in Figure 8.3.1(a), then it will not intersect any of its copies.
However, if the last scattering surface is large relative to the size of the
fundamental domain, as in Figure 8.3.1(b), then it will intersect one or more
of its copies. Moreover, adjacent copies of the LSS will intersect in a circle. In
this happy case, our last scattering surface will contain circles with matching
temperature distributions. Look again at Figure 8.3.1(b). We have three copies
of our fundamental domain pictured as well as three copies of the LSS (only one
of which is shaded to make the situation less cluttered). Two vertical circles
of intersection appear in the gure. From our point of view at the center of
the LSS, the two images of the circle will be directly opposite one another in
the sky. Since these circles are one and the same, the temperature distribution
around the two circles will agree. Therein lies the hope. Scan the temperature
distribution in the last scattering surface for matching circles.
(a) (b)
Figure 8.3.1: The LSS compared to the size of a 3-torus universe. In (a)
the LSS is small, so it won't reveal the shape of the universe; in (b) the LSS
intersects itself and will have circles of matching temperature distributions.
This strategy for detecting a nite universe is called the circles-in-the-
sky method, which, in cosmic topology, has advantages over the cosmic
crystallography method. In theory, the circles-in-the-sky method can be used
to detect any compact manifold, regardless of the geometry it admits. Also,
the search for matching circles is independent of a metric. One doesn't need
to make a claim about the geometry of the universe to detect a nite universe.
This method is computationally very intensive. The search for matching
circles on this giant 2-sphere involves the analysis of a six parameter space: the
center (θ1 , φ1 ) of one circle on the LSS, the center of the second circle (θ2 , φ2 ),
the common angular radius α of the two circles (since these circles are copies
of the same circle they will have the same radius), and the relative phase of the
two circles, say β. (See the diagram that follows.) In general, β 6= 0 if the face
identi cations in the 3-manifold involve rotations. It remains for us to analyze
whether a statistically signi cant correlation exists between the temperatures
as we proceed around the circles.
8.3. Circles in the Sky 197
(θ1 , φ1 )
(θ2 , φ2 )
When comparing the size of the LSS relative to the size of space, it is
convenient to de ne the following length dimension. The injectivity radius
at a point in a manifold, denoted rinj , is half the distance of the shortest closed
geodesic path that starts and ends at that point. A necessary condition, then,
for detecting matching circles in the LSS at our location is that our observable
radius robs exceeds our injectivity radius rinj .
Example 8.3.2: Detecting the 3-torus from the LSS.
If the universe is a 3-torus, and our LSS has diameter larger than some
dimension of the 3-torus, then the LSS will intersect copies of itself, and
the matching circles would be diametrically opposed to one another on
the LSS. Suppose our Dirichlet domain in a 3-torus universe is an a
by b by c box in R3 , where a < b < c. Our LSS is then centered at
the center of the box. The injectivity radius of the universe is a/2.
In the following gure, we assume that robs , the radius of the LSS, is
greater than a/2 but less than b/2. In this case, the circles-in-the-sky
method would detect one pair of matching circles in the temperature
distribution of the LSS. From the Earth E , we would observe that circle
C1 , when traced in the counterclockwise direction, matches circle C2
when traced in the clockwise direction, with no relative phase shift.
b
C1
E
a
C2
c
If the size is right, all six compact orientable Euclidean 3-manifolds would
have matching circles that are diametrically opposed to one another on the
LSS. The phase shift on these matching circles will be non-zero if the faces are
identi ed with a rotation.
Example 8.3.3: LSS in a Poincaré dodecahedral space.
If we live in a Poincaré dodecahedral space and our LSS is large enough,
we might see six pairs of matching circles, each pair consisting of
diametrically opposed circles in the sky with matching temperature
distributions after a relative phase shift of 36◦ . The following gure
indicates the matching circles that would arise from the identi cation
of the front face and rear face of the dodecahedron. From the Earth E,
198 8. Cosmic Topology
we would observe that circle C1 when traced in the counterclockwise
direction matches circle C2 when traced in the clockwise direction, with
◦
a phase shift of 36 .
5
1 2
c2 2
c1
4 3
5
3
LSS
In general, the matching circles we (might) see can depend not only on the
shape of the universe, but also on where we happen to be in the universe.
This follows because the Dirichlet domain can vary from point to point
(see Example 7.7.14 for the two-dimensional case). Of the 10 Euclidean 3-
manifolds, only the 3-torus has the feature that the Dirichlet domain is location
independent. Some (but not all) elliptic 3-manifolds have this feature, and in
any hyperbolic 3-manifold, the Dirichlet domain depends on your location.
Thus, if we do observe matching circles, it can not only reveal topology but
also the Earth's location in the universe.
Searches to date have focused on circles that are diametrically opposed to
one another on the LSS (or nearly so). This restriction reduces the search
space from six parameters to four. Happily, most detectable universe shapes
would have matching circles diametrically opposite one another, or nearly so.
At the time of this writing, no matching circles have been found, and this
negative result places bounds on the size of our universe. For instance, an
article written by the people who rst realized a small nite universe would
imprint itself on the LSS [19], concludes from the absence of matching circles
that the universe has topology scale (i.e., injectivity radius rinj ) bigger than
24 gigaparsecs, which works out to 24 × 3.26 × 109 ≈ 78 billion light-years. So,
a geodesic closed path trip in the universe would be at least 156 billion light
years long.
This stupendous distance is hard to fathom, but it appears safe to say
that we might abandon the possibility of gazing into the heavens and seeing a
distant image of our beloved Milky Way Galaxy.
In addition to [19], other accessible papers have been written on the circles-
in-the-sky method, as well as the cosmic crystallography method. (See [23],
[26], and [24].) Je Weeks also discusses both programs of research in The
Shape of Space, [12].
8.4 Our Universe
Our universe appears to be homogeneous and isotropic. The presence of cosmic
microwave background radiation is evidence of this: it is coming to us from
every direction with more or less constant temperature. This uniformity can
be explained by the in ationary universe theory. The theory, pioneered in the
1980s by Alan Guth and others, states that during the rst 10−30 seconds (or
8.4. Our Universe 199
so) after the big bang, the universe expanded at a stupendous rate, causing
the universe to appear homogeneous, isotropic, and also at.
The assumptions of isotropy and homogeneity are remarkably fruitful when
one approaches the geometry and topology of the universe from a mathematical
point of view. Under these assumptions, three possibilities exist for the
geometry of the universe - the three geometries that have been the focus of
this text. Each geometry type has possible universe shapes attached to it, and
Section 8.1 showcases some of the leading compact candidates.
The mathematical point of view gives us our candidate geometries, but
attempts at detecting the geometry from the mathematical theory have proved
unsuccessful. For instance, no enormous, cosmic triangle involving parallax
has produced an angle sum su ciently dierent from π radians to rule out
Euclidean geometry.
Adopting a physical point of view, we have another way to approach the
geometry of the universe. Einstein's theory of general relativity ties the
geometry of the universe to its mass-energy content. If the universe has a
high mass-energy content, then the universe will have elliptic geometry. If the
universe has a low mass-energy content, then it will be hyperbolic. If it has
just a precise amount, called the critical mass density, the universe will be
Euclidean. From a naive point of view, this makes it seem highly unlikely that
our universe is Euclidean. If our mass-energy content deviates by the mass
of just one hydrogen atom from this critical amount, our universe fails to be
Euclidean.
A little notation might be helpful. It turns out that from Einstein's eld
equations, the mass-energy density of the universe, ρ, is related to its curvature
k by the following equation, called the Friedmann Equation:
8πG k
H2 = ρ − 2.
3 a
Here G is Newton's gravitational constant; H is the Hubble constant
measuring the expansion rate of the universe; k = −1, 0, or 1 is the curvature
constant; and a is a scale factor. In fact, a and H are both changing in time,
but may be viewed as constant during the present period. Current estimates
of H (see [17] or [22]) are in the range of 68 to 70 kilometers per second per
megaparsec, where 1 megaparsec is 3,260,000 light-years.
In a Euclidean universe, k=0 and solving the Friedmann equation for ρ
gives us the critical density
3H 2
ρc = .
8πG
This critical density is about 1.7 × 10−29 grams per cubic centimeter, and
is the precise density required in a Euclidean universe.
We let Ω equal the ratio of the actual mass-energy density ρ of the universe
to the critical one ρc . That is,
ρ
Ω= .
ρc
Then, if Ω < 1 the universe is hyperbolic; if Ω > 1 the universe is elliptic;
and if Ω = 1 on the nose then it is Euclidean.
Until the late 1990s all estimates of the mass-energy content of the universe
put the value of Ω much less than 1, suggesting a hyperbolic universe. In fact,
dierent observational techniques for estimating the total mass-energy content
200 8. Cosmic Topology
of the universe put the value of Ω at about 1/3, contrary to the value of 1
predicted by the in ationary universe model.
But at the dawn of the 21st century, this all changed with detailed mea-
surements of the cosmic microwave background radiation, and the remarkable
discovery that the universe is expanding at an accelerated rate.
Careful analysis of the WMAP data on the cosmic background radiation
1
suggests that the universe is at, or nearly so, in agreement with the theory of
in ation. (This analysis is dierent than the circles in the sky method, which
searches for shape.) The ve-year estimate from the WMAP data (see [21])
put the value of Ω at
Ω = 1.0045 ± .013.
So if the mass-energy density is about 1/3 of what is required to get us
to a Euclidean universe, but it appears from the CMB that the universe is
Euclidean, or nearly so, some other form of energy must exist. Evidence for
this was presented in 1999 (see, for instance, [18]) in the form of observations
of distant exploding stars called Type Ia supernovae. These distant supernovea
are fainter than expected for a universe whose expansion rate is slowing down,
suggesting that the universe is accelerating its expansion. In 2011, Saul
Perlmutter, Brian Schmidt, and Adam Riess won the Nobel Prize in Physics
for their work on this discovery.
When Einstein rst proposed the 3-sphere as the shape of the universe,
his theory predicted that the 3-sphere should be expanding or collapsing. The
idea of a static universe appealed to him, and he added a constant into his
eld equations, called the cosmological constant, whose role was to counteract
gravity and prevent the universe from collapsing in on itself. But at roughly the
same time, Edwin Hubble, Vesto Slipher and others discovered that galaxies
in every direction were receding from us. Moreover, galaxies farther away were
receding at a faster rate, implying that the universe is expanding. Einstein
withdrew his constant.
But now the constant has new life, as it can represent the repulsive dark
energy that seems to be counteracting gravity and driving the accelerated
expansion of the universe. So the density parameter in the Friedmann equation
can have two components: ρM , which is the mass-energy density associated
with ordinary and dark matter (the mass-energy density that cosmologists
have been estimating by observation); and ρΛ , which is the dark energy, due
to the cosmological constant. In this case, Friedmann's equation becomes
8πG k
H2 = (ρM + ρΛ ) − 2
3 a
and dividing by H2 we have
8πG 8πG k
1= ρM + ρΛ − .
3H 2 3H 2 (aH)2
We let
ρM ρΛ −k
ΩM = , ΩΛ = , Ωk = .
ρc ρc (aH)2
So the simple equation
1 = ΩM + ΩΛ + Ωk
1 The Wilkinson Microwave Anisotropy Probe was launched in 2001 to carefully plot the
temperature of the cosmic microwave background radiation.
8.4. Our Universe 201
fundamentally describes the state of the universe. The in ationary universe
model suggests that Ωk ≈ 0 , which is supported by recent reports. Nine-
year analysis of the WMAP data combined with measurements of the Type Ia
Supernaovae (see [22]) suggest
−0.0066 < Ωk < 0.0011,
with ΩΛ ≈ .72 and ΩM ≈ 0.28. As the universe evolves, the values of the
density parameters may change, though the sum will always equal one.
Now is probably as good a time as any to tell you that the fate of the
universe is tied to its mass-energy content and the nature of the dark energy,
which is tied to the the geometry of the universe, which is tied to the topology
of the universe.
If the cosmological constant is zero then the relationship is simple: if Ω>1
so that we live in an elliptic 3-manifolds, the universe will eventually begin to
fall back on itself, ultimately experiencing a big crunch. If Ω = 1, a nite
Euclidean universe will have one of 10 possible shapes (6 if we insist on an
orientable universe) and its expansion rate will asymptotically approach 0, but
it will never begin collapsing. If Ω < 1, the universe will continue to expand
until everything is so spread out we will experience a big chill.
With a nontrivial vacuum energy the situation changes. While the
gravitational force due to the usual mass-energy content of the universe tends
to slow the expansion of the universe, it seems the dark energy causes it to
accelerate. Whether the universe is hyperbolic, elliptic, or Euclidean, if the
dark energy wins the tug of war with gravity, the curvature of the universe
would approach 0 as it continued to accelerate its expansion, and the density
of matter in the universe would approach 0.
The European Space Agency launched the Planck satellite in 2009. In its
orbit at a distance of 1.5 million kilometers from the Earth, the Planck satellite
has given us improved measurements of the CMB temperature, enabling
sharper estimates of cosmological parameters, as well as more re ned data
on which to run circles-in-the-sky tests. Alas, no circles have been detected.
Regarding the curvature of the universe, the Planck team concludes in [17], in
agreement with the WMAP team, that our universe appears to be at to a one
standard deviation accuracy of 0.25%.
In short, the universe appears to be homogeneous, isotropic, nearly at, and
dominated by dark energy. The current estimates for Ωk leave the question of
the geometry of the universe open, though just barely. It is still possible that
our universe is a hyperbolic or an elliptic manifold, but the curvature would
have to be very close to 0. If the universe is a compact, orientable Euclidean
manifold, we have six dierent possibilities for its shape. Since Euclidean
manifold volumes aren't xed by curvature, there is no reason to expect the
dimensions of a Euclidean manifold to be close to the radius of the observable
universe. But if the size is right, the circles-in-the-sky method would reveal
the shape of the universe through matching circles.
Perhaps we will be treated to matching circles some day. Perhaps not.
Perhaps the universe is just too big. In any event, pursuing the question of the
shape of the universe is a remarkable feat of the human intellect. It is inspiring
to think, especially looking up at a clear, star lled night sky, that we might
be able to determine the shape of our universe, all without leaving our tiny
planet.
A
List of Symbols
Symbol Description Page
C the complex plane 13
S1 the unit circle 34
iC (z) inversion in the circle C 34
∞ the point at ∞ 44
C+ the etended complex plane 44
(C, T ) translational geometry 65
(C, E) Euclidean geometry 66
(C+ , M) Möbius geometry 69
(D, H) the Poincaré disk model for hyperbolic geome- 72
try
D the hyperbolic plane 72
S1∞ the circle at ∞ in (D, H) 72
P2 the projective plane 117
(P2 , S) the disk model for elliptic geometry 118
(P2k , Sk ) the disk model for elliptic geometry with cur- 134
vature k
(Dk , Hk ) the disk model for hyperbolic geometry with 137
curvature k
(Xk , Gk ) 2-dimensional geometry with constant curva- 142
ture k
Rn real n-dimensional space 149
X1 #X2 the connected sum of two surfaces 151
T2 the torus 152
Hg the handlebody surface of genus g 152
Cg the cross-cap surface of genus g 153
K2 the Klein bottle 156
χ(S) the Euler characteristic of a surface 159
X/G the quotient set built from geometry (X, G) 170
H3 hyperbolic 3-space 179
S3 the 3-sphere 179
T3 the 3-torus 180
202
References
Textbooks
[1] Anderson, J.W. Hyperbolic Geometry, 2nd Ed. Springer, London, 2005.
[2] Boas, R.P. Invitation to Complex Analysis. Random House, New York,
1987.
[3] Coxeter, H.S.M. Introduction to Geometry, 2nd Ed. Wiley, New York,
1961.
[4] Heath, T.L. The Thirteen Books of Euclid's Elements, 2nd Ed. Dover,
New York, 1956.
[5] Henle, M. Modern Geometries: The Analytic Approach. Prentice Hall,
New Jersey, 1997.
[6] Hilbert, D., and H. Cohn-Vossen, Geometry and the Imagination. AMS
Chelsea Publishing, Rhode Island, 1952.
[7] Jennings, G.A. Modern Geometry with Applications. Springer, 1998.
[8] Schwerdtfeger, H. Geometry of Complex Numbers. Dover, New York,
1979.
[9] Sieradski, A.J. An Introduction to Topology and Homotopy. PWS-KENT
Publishing Company, Boston, 1992.
[10] Stillwell, J. Geometry on Surfaces. Springer, New York, 1992.
[11] Thurston, W.P. Three-Dimensional Geometry and Topology, Vol. I.
Princeton, New Jersey, 1997.
[12] Weeks, J.R. The Shape of Space, 2nd Ed.. Dekker, New York, 2002.
Other Books
[13] Gray J., ed. The Symbolic Universe: Geometry and Physics 1890-1930.
Ofxord University Press, New York, 1999.
[14] Hogan, C.J. The Little Book of the Big Bang. Copernicus Springer-
Verlag, New York, 1998.
[15] Rucker, R. Geometry, Relativity and the 4th Dimension. Dover, New
York, 1977.
Articles
[16] Adams, C. and J. Shapiro. The Shape of the Universe: Ten Possibilities.
American Scientist, Sept - Oct 2001, 89, 443453.
[17] Ade, P. A. R. and others. Planck 2015 results. XIII. Cosmological
parameters. e-Print: arXiv:1502.01589v3 [astro-ph.CO], 2016.
203
204 A. List of Symbols
[18] Bahcall, N., J. Ostriker, S. Perlmutter, and P. Steinhardt. The Cosmic
Triangle: Revealing the State of the Universe. Science, 1999, 284
no. 5419, 14811488.
[19] Cornish, N.J., D.N. Spergel, G.D. Starkman, and E. Komatsu. Con-
straining the Topology of the Universe. Physical Review Letters, (2004),
92 no. 20, 1302.
[20] Foote, R. L. A Uni ed Pythagorean Theorem in Euclidean, Spherical,
and Hyperbolic Geometries. Math. Magazine, February 2017 90 no. 1,
5969.
[21] Hinshaw, G. and others. Five-Year Wilkinson Microwave Anisotropy
Probe (WMAP) Observations: Data Processing, Sky Maps, and Basic
Results. e-Print: arXiv:0803.0732, 2008.
[22] Hinshaw, G. and others. Nine-Year Wilkinson Microwave Anisotropy
Probe (WMAP) Observations: Cosmological Parameter Results. eThe
Astrophysical Journal: Supplement Series, 2013.
[23] Levin, J. Topology and the Cosmic Microwave Background. Phys. Rep.,
2002, 365, 251333.
[24] Luminet, J.-P. and M. Lachieze-Rey. Cosmic Topology. Phys. Rep.,
(1995), 254, 135214.
[25] Luminet, J.-P., J.R. Weeks, A. Riazuelo, R. Lehoucq, R., and J.-P.
Uzan. Dodecahedral space topology as an explanation for weak wide-angle
temperature correlations in the cosmic microwave background. Nature,
(2003), 425, 59395.
[26] Rebouças, M.J., and G.I. Gomero. Cosmic Topology: A Brief Overview.
Phys. Rep., December, 2004, 34 no. 4A, 13581366.
[27] Schwarzschild, K. On the Permissible Curvature of Space. Vierteljahrschrift
d. Astronom. Gesellschaft, (1900), 35, 33747.
[28] Uzan, J.-P., R. Lehoucq, and J.-P. Luminet. New developments in the
search for the topology of the Universe. Nuclear Physics B Proceedings
Supplements, January 2000, 80, C425+.
Web Resources
[29] Joyce, D.E. Euclid's Elements, http://aleph0.clarku.edu/~djoyce/java/
elements/elements.html (1994)
Index
(Xk , Gk ), 142 Bolyai, Farkas, 4
(Dk , Hk ) with k < 0, 137 bounded set in Rn , 151
(P2k , Sk ) with k > 0, 134
Hg , 152
cell complex, 156
∞, 44
n-dimensional, 156
P2 , 117
edge, 156
S1 , 34
face, 156
i, 13
vertex, 156
2-sphere, 45
cell division, 157
3-manifold
circles of Apollonius, 41, 54
3-sphere, 181
Cliord translation, 191
3-torus, 180
half turn manifold, 187 cline, 35
Hantschze-Wendt, 185 closed geodesic path, 164
one-sixth turn manifold, 187 closed set in Rn , 151
one-third turn manifold, 187 compact space, 151
Poincaré dodecahedral space, complex number, 13
183 addition of, 13
quarter turn manifold, 186 argument, 15
Seifert-Weber space, 184 Cartesian form, 13
3-sphere, 181 conjugate, 14
3-torus, 180 division of, 17
8, the greatness of, 15 imaginary part, 13
modulus, 14
all-star equation, 15
multiplication, 14
angle
polar form, 15
between curves, 29
real part, 13
between rays, 18
scalar multiplication, 13
determined by three points,
cone point, 11, 163
18
connected sum, 151
transformation preserves, 29
construction
transformation preserves
antipodal point, 113
magnitudes, 29
cline through three distinct
angle of parallelism, 138
+ points, 36
antipodal points of C , 109
elliptic line through two
Apollonian Circles Theorem, 40
points, 118
Beltrami, Enrico, 5 hyperbolic circle centered at p
Bessel, Friedrich, 140 through q, 80
block, 96 hyperbolic line through two
Bolya, János, 4 points, 80
205
206 Index
hyperbolic re ection sending general linear transformation, 28
a point to the origin, 73 general theory of relativity, 199
octagon in (D, H) with 45◦ geodesic, 8
interior angles, 100 geometry, 64
right-angled hexagon in (Xk , Gk ), 142
(D, H), 97 congruent gures in, 64
symmetric point iC (z) Euclidean, 66
when z is inside C , 42 gure in, 64
when z is outside C , 43 homogeneous, 67
cosmic background radiation hyperbolic geometry, Poincaré
(CMB), 195 disk model, 72
cosmic crystallography, 191 invariant function, 65
collecting correlated pairs, 193 invariant set, 65
pair separation histogram, 190 isotropic, 67
cosmic topology, 178 metric, 67
circles-in-the-sky method, 196 minimal invariant set, 66
cosmic crystallography point, 64
method, 191 rotational, 69
cross ratio, 51 translational, 65
invariance of, 52 geometry with curvature k = 0,
cross-cap, 153 142
curvature of a space at a point, geometry with curvature k > 0,
132 134
curve, 29 geometry with curvature k < 0,
smooth, 29, 85 137
golden ratio, 98
De Sitter, Willem, 181
great circle, 8
Dirichlet domain, 175
great circle in C+ , 110
Dostoevsky, Fyodor, 5
group of homeomorphisms, 170
group of isometries, 170
Einstein, Albert, 6, 181, 200
group of transformations, 63
general theory of relativity,
Guth, Alan, 198
199
special theory of relativity, 6
half turn manifold, 187
elliptic geometry, 118
handlebody surface of genus g, 152
arc-length, 120
Hantschze-Wendt manifold, 185
area, 120
Hogan, Craig, 195
circle, 122
homeomorphic spaces, 148
distance formula, 120
homeomorphism, 148
line, 118
homogeneous geometry, 67
lune, 122
horocycle, 75
elliptic geometry with curvature k,
Hubble constant, 199
134
Hubble, Edwin, 200
equivalence relation, 168
hyperbolic cosine function, 91
Erlangen Program, 6, 63
hyperbolic geometry
Euclidean distance between
points, 67
H, the transformation group
in hyperbolic geometry,
Euler characteristic, 159
72
extended plane, 44
2
-ideal triangle, 91
3
Friedmann Equation, 199 D, the hyperbolic plane, 72
fundamental domain, 169 arc-length, 86
area, 90
Gauss, Carl Friedrich, 4, 131 circle, 79
Gauss-Bonnet Formula, 166 circle at in nity, 72
Index 207
circle centered at p through q, loxodromic, 57
construction of, 80 normal form
distance between points, 83 one xed point, 60
ideal point, 77 parabolic, 61
ideal triangle, 92 pole of a, 57
line, 77 manifold, 149
line through p and q, metric, 67
construction of, 80 Euclidean, 67
parallel lines, 77 hyperbolic, 87
Poincaré disk model, 72 multiconnected space, 148
upper half-plane model, 103
one-sixth turn manifold, 187
hyperbolic geometry with
one-third turn manifold, 187
curvature k, 137
one-to-one, 25
hyperbolic plane, 72
onto, 25
re ection of, 73
rotation of, 75
open n-ball, 149
orbit of a point, 170
translation of, 75
orbit space, 170
hyperbolic refelction, 73
orthogonal clines, 37
hyperbolic sine function, 91
hyperbolic transformation group, parallax, 140
72 of star 61 Cygni, 140
parallel displacement, 75 Parallel Postulate, 4
rotation, 75 partition, 169
translation, 75 Perlmutter, Saul, 200
planar curve, 29
image of a set, 28
Planck Satellite, 201
in ationary universe theory, 198
Platonic solids, 157
inverse pole of a Möbius
Playfair's Axiom, 4
transformation, 57
Playfair, John, 4
inversion in a circle, 33
Poincaré, Henri, 72
involution transformation, 61
point at ∞, 44
isometry
pole of a Möbius transformation,
Euclidean, 30
57
xed-point free, 172
polygonal surface
hyperbolic, 86
boundary label, 154
properly discontinuous, 172
power of a point, 37
isotropic geometry, 67
projective plane, 117
Klein Bottle, 193 quarter turn manifold, 186
Klein bottle, 156 quaternion, 180
Klein, Felix, 6, 72 conjugate, 180
modulus, 180
last scattering surface (LSS), 195 pure, 182
lens space, L(p, q), 184 scalar term, 180
Lobachevsky, Nikolai, 4 unit, 180
Lobatchevsky's formula, 138 quotient set, 169
X/G built from geometry
Möbius geometry, 69
(X, G), 170
Möbius strip, 155
Möbius transformation, 47 radius of curvature, 135
determinant of, 47 real n-dimensional space, 149
elliptic, 57 re ection, 31
fundamental theorem of, 50 relation on a set, 168
hyperbolic, 57 Riess, Adam, 200
inverse pole of a, 57 rotation, 27
208 Index
saddle point, 163 Euclidean, 178
Schmidt, Brian, 200 hyperbolic, 179
screw motion, 178 Thurston, William, 132
Seifert-Weber space, 184 transformation, 25
simply connected space, 148 dilation, 28
Slipher, Vesto, 200 xed point, 30
smooth curve, 29 fractional linear, 47
sphere general linear, 28
diametrically opposed points, inverse, 25
109 involution, 61
spherical geometry, 117 Möbius, 47
square root, 20 preserves angle magnitudes,
stereographic projection, 45 29
surface, 151 preserves angles, 29
cross-cap surface of genus g re ection, 31
(Cg ), 153 rotation, 27
handlebody surface a genus g screw motion, 178
(Hg ), 152
type I cline of p and q , 54
non-orientable, 155
type II cline of p and q , 54
orientable, 155
polygonal, 154
unit 2-sphere, 45
symmetric points
unit circle, 34
with respect to a circle, 33
universal covering space, 171
with respect to a sphere, 179
Theorem Egregium, 131 Weeks, Je, 198
three-dimensional geometry Wilkinson Microwave Anisotropy
elliptic, 180 Probe (WMAP), 200
You might also like
- Fundamentals of Cosmology Notes by Professor Komissarov Leeds UK 2016100% (1)Fundamentals of Cosmology Notes by Professor Komissarov Leeds UK 201690 pages
- Designazoointegratedgeometryareaperimeterproject89% (9)Designazoointegratedgeometryareaperimeterproject37 pages
- Geometry of Lengths, Areas, and Volumes Two-Dimensional Spaces by Cannon, James W.100% (3)Geometry of Lengths, Areas, and Volumes Two-Dimensional Spaces by Cannon, James W.133 pages
- Math Charmers Tantalizing Tidbits For The Mind - Posamentier and HauptmanNo ratings yetMath Charmers Tantalizing Tidbits For The Mind - Posamentier and Hauptman300 pages
- Geometry With An Introduction To Cosmic Topology 2018 Edition Michael P. HitchmanNo ratings yetGeometry With An Introduction To Cosmic Topology 2018 Edition Michael P. Hitchman238 pages
- (SS) Transitions and Trees, Hans HüttelNo ratings yet(SS) Transitions and Trees, Hans Hüttel291 pages
- (Ebook) Your Cosmic Context: An Introduction To Modern Cosmology by Todd Duncan, Craig Tyler ISBN 9780132400107, 0132400103 PDF DownloadNo ratings yet(Ebook) Your Cosmic Context: An Introduction To Modern Cosmology by Todd Duncan, Craig Tyler ISBN 9780132400107, 0132400103 PDF Download61 pages
- LAB211 Assignment: Title Background ContextNo ratings yetLAB211 Assignment: Title Background Context4 pages
- Design Review of Helmet For Aesthetic and Ergonomic ConsiderationsNo ratings yetDesign Review of Helmet For Aesthetic and Ergonomic Considerations10 pages
- A First YeaR CoursE in GeometrY Evan Chen 2019No ratings yetA First YeaR CoursE in GeometrY Evan Chen 201936 pages
- Personalized Meal Plans Instruction GuideNo ratings yetPersonalized Meal Plans Instruction Guide3 pages
- Lectures On Convex Sets: Niels LauritzenNo ratings yetLectures On Convex Sets: Niels Lauritzen93 pages
- Geometry of Curves and Surfaces: Weiyi Zhang Mathematics Institute, University of WarwickNo ratings yetGeometry of Curves and Surfaces: Weiyi Zhang Mathematics Institute, University of Warwick78 pages
- Tian, Canonical Metrics in Kahler Geometry PDFNo ratings yetTian, Canonical Metrics in Kahler Geometry PDF107 pages
- The Handbook of Computer-Aided Geometric Design: January 2002No ratings yetThe Handbook of Computer-Aided Geometric Design: January 200217 pages
- Rewording The Brain How Cryptic Crosswords Can Improve Your Memory and Boost The Power and Agility of Your Brain Research PDF Download100% (14)Rewording The Brain How Cryptic Crosswords Can Improve Your Memory and Boost The Power and Agility of Your Brain Research PDF Download15 pages
- Wayne's Fruit Loops Music Making LessonsNo ratings yetWayne's Fruit Loops Music Making Lessons16 pages
- Trainng report-BSNL: Monday, July 2, 2007No ratings yetTrainng report-BSNL: Monday, July 2, 200740 pages
- Layers of The Atmosphere Power Point PDFNo ratings yetLayers of The Atmosphere Power Point PDF16 pages
- Family Worship, Family Worship, J. H. Merle D'aubignéNo ratings yetFamily Worship, Family Worship, J. H. Merle D'aubigné18 pages
- Dept. of Chemistry, Rajabazar Science College, 92-Acharya Prafulla Chandra Road, University of Calcutta, Kolkata - 700009, West Bengal, IndiaNo ratings yetDept. of Chemistry, Rajabazar Science College, 92-Acharya Prafulla Chandra Road, University of Calcutta, Kolkata - 700009, West Bengal, India6 pages
- Practice Exam For Final Exam Acct301 With AnswersNo ratings yetPractice Exam For Final Exam Acct301 With Answers9 pages
- Cico Plast-N: Normal Water Reducing Admixture / Plasticiser For ConcreteNo ratings yetCico Plast-N: Normal Water Reducing Admixture / Plasticiser For Concrete2 pages
- Playdor School Bus Schedule - 15 April 2024 To 26 July 2024-1No ratings yetPlaydor School Bus Schedule - 15 April 2024 To 26 July 2024-11 page
