Lecture Notes in Modern Geometry

RUI WANG

The content of this note mainly follows John Stillwell’s book geometry of surfaces.

1 The euclidean plane

1.1 Approaches to euclidean geometry

Our ancestors invented the geometry over euclidean plane. Euclid [300 BC] understood euclidean plane via points,
lines and circles. A motivation of Euclid’s method was to answer the question that what can be done with ruler and
compass only. Euclid’s geometry is based on logic deductions from axiom system. (The rigorous axiom system
was given by Hilbert [1899].) The proofs are usually tricky and simple but quite isolated from other branches of
mathematics.
The viewpoint of modern geometry is to study euclidean plane (and more general, euclidean geometry) using sets
and numbers. This idea dates back to Descartes (1596-1650) and is referred as analytic geometry. On one side,
this brings an effective way in understanding geometry; on the other side, the intuition from geometry stimulates
solutions of problems purely from algebras. (A famous example might be Fermat’s last theorem which was solved
by Andrew Wiles in 1995 using the most advanced algebraic geometry. ) From this point of view, modern geometry
successfully makes mathematics as a whole, which is the spirit of the math from 20 century’s.
In fact, starting from Euclid’s time, people are trying to ask whether one can remove the parallel axiom from the
axiom system and set up all results from euclidean geometry. The answer turns to be negative. People found that
there are three different types of geometry based on different assumption as replacements for parallel axiom. It
was Riemann [1840] who clarified the basic viewpoints and opened the chapter of modern geometry. Riemann’s
idea basically includes:
• consider points in a n-dimensional space as n-tuple of numbers;
• consider the distance between two points as a distance function;
• introduce the concept of curvature which reflects the geometry of the space.

Different choices of metrics correspond to different geometry. From Riemann’s point of view, the eulidean plane
corresponds to a curvature zero metric over R2 .
Though in general curvature is defined from point to point, if we add another assumption that the curvature is
a constant, we will see that the situation gets much simplified. More concretely, the geometry of spaces now is
completely reflected by its isometries. The idea of understanding geometry by studying its isometries dates back
to Klein [1872]. In particular, this builds up a bridge between classical euclidean geometry (Euclid’s method) and
Riemannian geometry of constant curvatures. Our lectures will take this point of view.
2 Rui Wang

1.2 Isometries

Consider the set of pair of real numbers

R2 := {(x, y)|x, y ∈ R}.
The euclidean distance is a function d : R2 × R2 → R defined as
p
d((x1 , y1 ), (x2 , y2 )) = (x1 − x2 )2 + (y1 − y2 )2 .
This function represents Pythagorean distance of two points in the plane as what we know from Euclid’s method.

Definition 1.1 An euclidean isometry for R2 is a map f : R2 → R2 satisfying that for any (x1 , y1 ), (x2 , y2 ) ∈ R2 ,
d(f (x1 , y1 ), f (x2 , y2 )) = d((x1 , y1 ), (x2 , y2 )).
We use Iso(R2 , d) to denote the set of all euclidean isometries for R2 .

Example 1.2 We give three important examples of euclidean isometries.

(1) Translation by (α, β).
t(α,β) : R2 → R2 , (x, y) 7→ (x + α, y + β).
(2) Reflection about x-axis.
Rfx : R2 → R2 , (x, y) 7→ (x, −y).
(3) Rotation around the origin by θ -angle counter-clockwise.
rO,θ : R2 → R2 , (x, y) 7→ (x cos θ − y sin θ, x sin θ + y cos θ).
Check: These are all isometries.

These isometries lists in previous Example 1.2 have nice representations via complex numbers.

Example 1.3 (1) Translation by z0 = α + iβ .

tz0 : C → C, z 7→ z + z0 .
(2) Reflection about x-axis.
Rfx : C → C, z 7→ z̄.
(3) Rotation around the origin by θ -angle counter-clockwise.
rO,θ : C → C, z 7→ eiθ z.

Using complex numbers, the euclidean distance can be expressed as

d(z1 , z2 ) = |z1 − z2 |.
Use it, check again these three maps are isometries.

Exercise 1.4 (1) Assume f and g are two isometries of the euclidean plane. Prove that the composition g ◦ f
is also an isometry.
(2) Prove that the following two definitions for a line in the euclidean plane are equivalent.
Lecture Notes in Modern Geometry 3

(a) A line in the euclidean plane is a set

{(x, y) ∈ R2 |ax + by + c = 0}
for some a, b, c ∈ R with a2 + b2 6= 0;
(b) A line in the euclidean plane is a set
L(a0 ,b0 ),(a1 ,b1 ) := {(x, y) ∈ R2 |d((x, y), (a0 , b0 )) = d((x, y), (a1 , b1 ))}
for some (a0 , b0 ), (a1 , b1 ) ∈ R2 with (a0 , b0 ) 6= (a1 , b1 ). Here d denotes the euclidean distance.
(3) Use the second definition in (2) to prove: an isometry maps a line to a line. (Remark: In later lectures, we
are going to introduce distance functions other than the euclidean distance. Then the first way of defining a
line turns out to be not good any more because a line defined in that way is not preserved under isometries.
However, the second definition still makes sense: notice in this definition, we only use the distance function.)
(4) Prove the three isometries given in Example 1.2 are all one-to-one and onto. Find their inverses.

Intuition tells us, not only the reflection about x-axis, a reflection about any line is an isometry; Not only the
rotation around the origin, a rotation around any point in R2 is an isometry.

Example 1.5 (1) Assume p = (α, β) ∈ R2 . Denote by rp,θ : R2 → R2 the rotation around p by θ -angle
counter-clockwise. Then
−1
rp,θ = t(α,β) ◦ rO,θ ◦ t(α,β) .

(2) Assume L is a line in R2 . Denote by RfL : R2 → R2 the reflection about L (How to define a reflection?).
Notice that if we map L to x-axis, then the reflection will be the standard one that about x-axis. For this, we
need to be a little careful for the following two cases:
• Case: L intersect x-axis at some point p = (α, 0) via θ as the angle from the positive direction of
−1
x-axis to L. Then we can rotate L to x-axis by rp,−θ = rp,θ , and hence
−1
RfL = rp,θ ◦ Rfx ◦ rp,θ .
Further, we can express rp,θ using (1).
• Case: L is parallel to x-axis. Assume L can be written as y = β . Then we can translate L to x-axis
−1
via t(0,−β) = t(0,β) . Similarly, for this case,
−1
RfL = t(0,β) ◦ Rfx ◦ t(0,β) .

We are familiar with these expressions of the form φ ◦ ψ ◦ φ−1 which is called conjugation, from linear
algebra or more general from group theory. In general, the appearance of this form indicates we are doing
some coordinate change.

Exercise 1.6 Represent RfL and rp,θ using C and check your answers via examples.

From the expressions of RfL and rp,θ , we see that they are compositions of translations, reflections about x-axis
and rotations around origin. In fact, we are going to prove any euclidean isometry can be written as compositions
of these three.
4 Rui Wang

1.3 Reflections

Take two lines L1 , L2 in the euclidean plane. They either intersect or parallel (i.e. not intersect). Let’s first see:
(1) If L1 intersects L2 at some point p, then RfL2 ◦ RfL1 is the rotation around p for the double of the oriented
angle from L1 to L2 .
p
(2) If L1 ∩ L2 = ∅, then RfL2 ◦ RfL1 is some translation t(α,β) with the amount α2 + β 2 as double of the
oriented distance from L1 to L2 .
Conversely, we prove the following result.

Theorem 1.7 (1) Any rotation rp,θ can be decomposed as

rp,θ = RfL2 ◦ RfL1
for any two lines L1 ∩ L2 = {p} with the oriented angle from L1 to L2 as 21 θ .
(2) Any translation t(α,β) can be decomposed as
t(α,β) = RfL2 ◦ RfL1
p
1
for any two lines L1 ∩ L2 = ∅ with the oriented distance from L1 to L2 as 2 α2 + β 2 .

You can definitely check the proof directly by doing some calculation. However, a more geometric way is to follow
the scheme:
(1) Prove the results for the simplest cases: rO,θ and t(0,β) .
(2) Prove that general cases can be reduced to these simplest cases using conjugation by isometries.
Let’s take t(α,β) as an example for how this works:
Step1. Show that we can find some isometry f so that
t(α,β) = f ◦ t(0,β) ◦ f −1 .
Step2. Using the results for the simplest case t(0,β) to write
t(0,β) = RfL2 ◦ RfL1 .
Step3. Then we have
t(α,β) = f ◦ t(0,β) ◦ f −1
= f ◦ RfL2 ◦ RfL1 ◦ f −1
= (f ◦ RfL2 ◦ f −1 ) ◦ (f ◦ RfL1 ◦ f −1 )
= Rff (L2 ) ◦ Rff (L1 ) .
The rotation case is exactly the same and is left to you to finish.

Exercise 1.8 Prove the set of translations and rotations is closed under composition. (Closed means for any two
maps of translations or rotations, their composition is still a translation or a rotation.)
Lecture Notes in Modern Geometry 5

1.4 The three reflections theorem

We have seen that both translation and rotation can be written as compositions of reflections. In this section,
we prove an even stronger result that any euclidean isometry can be decomposed into reflections, which makes
reflection play an essential role in understanding isometries (we are going to discuss more on this point of view
along our lectures). More concretely, let’s prove

Theorem 1.9 (The three reflections theorem) Any euclidean isometry can be written as compositions of one or
two or three reflections.

To prove this theorem, the first difficulty needs to be overcome is we need to give a way to characterize any
euclidean isometry. For this, we first show that

Lemma 1.10 Any a euclidean isometry is uniquely determined by the image of three points which are not in a
line.

Proof Let’s take A, B, C in the plane R2 and assume that they are not in a line. Assume f is an isometry. If there
is another isometry f 0 so that
f (A) = f 0 (A), f (B) = f 0 (B), f (C) = f 0 (C).
Let’s prove f = f 0 .
For this, we need to show for any p ∈ R2 , we have f (p) = f 0 (p). Assume this is not the case, i.e., we can find
some p so that f (p) 6= f 0 (p).
Notice that d(f (z), f (p)) = d(f (z), f 0 (p)) for any z = A or B or C. This requires f (A), f (B), f (C) live in the line
Lf (p),f 0 (p) determined by f (p) and f 0 (p) (see the definition of a line in Exercise 1.4). Now we show that this will
require A, B, C live in the same line which contradicts with our assumption.
To see this, WLOG we can assume d(f (A), f (B)) + d(f (B), f (C)) = d(f (A), f (C)). Then using the fact that f is an
isometry, we have
d(A, B) + d(B, C) = d(A, C).
A moment of calculation tells us, A, B, C must satisfy some linear relation as they all in a set {(x, y)|ax+by+c = 0}
for some a, b not both zero. This shows they are in a line.
Then we are done with this proof.

Now with the help of this lemma, we can write any isometry f by A, B, C as three points not in a line and their
images A0 , B0 , C0 under f . Let’s show then f can be decomposed into at most reflections.
We still start from the simplest case:
• Case A = A0 , B = B0 , C = C0 : By the proof of Lemma 1.10, this map must be the identity map. By picking
any line L, we can write
idR2 = RfL ◦ RfL .
• Case A = A0 , B = B0 , C 6= C0 : By the proof of Lemma 1.10, this map must be the reflection about the line
going through A, B. Hence we are done.
6 Rui Wang

• Case A = A0 , B 6= B0 , C 6= C0 : We have two possibilities for this case: Let’s denote by LBB0 and LCC0 the
two lines determined by B, B0 and C, C0 respectively (notation as in the definition of a line in Exercise 1.4).
They both go through A.
(1) If LBB0 = LCC0 =: L, we show this isometry must be the reflection about L.
(2) If LBB0 6= LCC0 . We show this isometry is a rotation around A. Recall that we have shown that a
rotation can be written as a composition of two reflections, we are done for this case too.
• Case A 6= A0 , B 6= B0 , C 6= C0 : For this case, we first take a translation to map A to A0 . Then we reduce
it to one of the first three cases we have done. There is only one trouble that we need to take care of in
addition. For a translation, we know it can be decomposed into two reflections. In case after translation, we
get the case (2) above, then we have four reflections to composite but we want up to three. To resolve this
issue, notice that we can take a common line, say L, for both translation and rotation. Then this map is a
composition RfL2 ◦ RfL ◦ RfL ◦ RfL1 which is the same as RfL2 ◦ RfL1 since reflection twice about the same
line is just the identity map. Then we are done with the proof of this theorem.

1.5 Orientation-preserving isometries

For any simple loop in R2 without self intersection point, it divides R2 into two connected regions. (In fact, this
is a nontrivial result called the Jordan curve theorem. But since it matches our intuition and more important in our
case, we don’t consider general loops, we just accept it without a proof.) Given an isometry f , we have shown
in Lemma 1.10 that f is uniquely characterized by A, B, C not in a line and A0 := f (A), B0 := f (B), C0 := f (C).
Consider the loop ABCA, it divides R2 into two regions, we call the bounded region the interior of the loop ABCA
and denote it by Int(ABCA). Now imagine we stand on a point of the loop ABCA, with our head towards the
positive direction of z-axis (x, y, z-axises satisfy the right-hand-rule.), we define the positive direction of the loop
ABCA as the direction so that the interior Int(ABCA) is on our left-hand side. It is either ABCA or ACBA. (Here
by the notation ABCA, we don’t consider the order of A, B, C, A but we use ABCA to denote the direction from A
to B to C and back to A. )
Then we consider the loop A0 B0 C0 A0 , and define the orientation for it in the above way. We call f is an orientation-
preserving isometry, if A0 B0 C0 A0 and ABCA are both positive or both negative directions; Otherwise, we call f
is an orientation-reversing isometry. (In fact, this definition is not complete, because we need to prove that such
defined orientation-preserving or orientation-reversing is independent of choices of the points A, B, C. This is left
to you as an exercise.)
The following result is a basic property.

Proposition 1.11 (1) A composition of two orientation-preserving isometries is orientation-preserving;

(2) A composition of two orientation-reversing isometries is orientation-preserving;
(3) A composition of an orientation-preserving isometry and an orientation-reversing isometry is orientation-
reversing.

Remark 1.12 One may denote an orientation-preserving isometry by 1 and denote an orientation-reversing
isometry by −1, then the pattern above is just like 1 · 1 = 1; (−1) · (−1) = 1; 1 · (−1) = −1 = (−1) · 1. In fact,
this is saying that we can construct a surjective group homomorphism from Iso(R2 , d) to Z2 ∼ = ({±1}, ·), which
maps orientation-preserving isometries to 1 and maps orientation-reversing isometries to −1. By this, we also see
Lecture Notes in Modern Geometry 7

the set of orientation-preserving isometries, which we denote by Iso+ (R2 , d) is a normal subgroup of Iso(R2 , d)
because it is the kernel of this group homomorphism. Using the fundamental theorem of group homomorphism,
the quotient group Iso(R2 , d)/Iso+ (R2 , d) ∼
= Z2 and hence the subgroup Iso+ (R2 , d) has index 2.

1.6 Classification of euclidean isometries

Now using the three reflections theorem, we prove the classification theorem for euclidean isometries.
For this, we first introduce a terminology called glide reflection. Assume L is a line, a glide reflection is a translation
in L direction composite with the reflection RfL . Notice that the order of the translation and the reflection doesn’t
matter, i.e., these two maps commute. It is easy to check that glide reflection is either a reflection about L or three
reflections about L, M, N with M, N both perpendicular to L. Now we prove the reverse part: if an isometry is a
reflection or a composition of three reflections, then it is a glide reflection.
Clearly, after proving this, we will conclude the following classification result for euclidean isometries for R2 .

Theorem 1.13 A euclidean isometry of R2 is either a translation, a rotation or a glide reflection. Moreover, an
orientation-preserving isometry is either a translation or a rotation; an orientation-reversing isometry must be a
glide reflection.

The case of reflection is obviously a glide reflection with trivial translation. We now prove that any isometry
RfN ◦ RfM ◦ RfL is a glide reflection.

Proof Case 1. M ∩ N = {p}. Denote by M 0 the unique line going through p and perpendicular to L. Denote
by M 0 ∩ L = {q}, which are going to use later. Next we replace N by some N 0 going through p so that
RfN ◦ RfM = RfN 0 ◦ RfM0 .
(Recalling from Theorem 1.7 that such N 0 is uniquely determined.) It follows RfN ◦ RfM ◦ RfL = RfN 0 ◦
RfM0 ◦ RfL .
Now we take L0 as the unique line going through q and perpendicular to N 0 . Take M 00 as the unique line
going through q so that
RfM0 ◦ RfL = RfM00 ◦ RfL0 .
This follows
RfN ◦ RfM ◦ RfL = RfN 0 ◦ RfM0 ◦ RfL = RfN 0 ◦ RfM00 ◦ RfL0 .
In particular, from the construction, we notice that both N 0 and M 00 are perpendicular to L0 , which makes
RfN 0 ◦ RfM00 ◦ RfL0 a glide reflection about L0 . We are done with this case.
Case 2. M ∩ N = ∅. This one is left to you as a homework problem and here is a hint: Consider two subcases:
(1) M ∩ L = ∅. For this case, these three lines are parallel and their composition is just a reflection.
(2) M ∩ L = {p}. For this case, you can work out a similar construction as Case 1 starting with replacing
M by the line M 0 going through p and perpendicular to N .
8 Rui Wang

1.7 The group structure of Iso(R2 , d)

We introduce a terminology called group. A group is a set G together with a binary operation · so that
(1) The binary operation is associative;
(2) There is an identity element;
(3) Every element has an inverse element.
For our case, the set G is taken as Iso(R2 , d) and the binary operation is taken as composition. With respect to
composition, Iso(R2 , d) is a group and called the euclidean isometry group.
In general a subgroup of a group (G, ·) is a subset H ⊂ G which is closed under the binary operation · and the H
with respect to the binary operation restricted to it is a group. A normal subgroup of a group G is a subgroup H
so that every element of the form g−1 · h · g ∈ H for each g ∈ G, h ∈ H . We don’t want to spend too much time
on the abstract definition, but you should have the following examples in your mind.

Example 1.14 (1) Assume L is a line, {idR2 , RfL } is a group with respect to composition. In fact this group is
isomorphic to Z2 ;
(2) The set of translations {t(α,β) |(α, β) ∈ R2 } forms a group with respect to composition. In fact this group is
isomorphic to (C, +).

The group Iso(R2 , d) has many subgroups:

(1) For any p ∈ R2 , all rotations around p forms a subgroup, which is isomorphic to (U(1), ·);
(2) For any line L in R2 , {idR2 , RfL } forms a subgroup which is isomorphic to Z2 ;
(3) All translations form a subgroup which is isomorphic to (C, +);
(4) All translations and rotations forms a normal subgroup, which is Iso+ (R2 , d).
In general, for a group G, any subset S ⊂ G can generate a subgroup of G by including all possible multiplications
from elements in S and their inverses. We denote this subgroup as < S >. In particular, if S contains only one
element, say g, then the subgroup generated by g is a cyclic subgroup with g as a generator. We use < g > to
denote this group.

Example 1.15 (1) Take z0 ∈ C, the translation tz0 generates a subgroup < tz0 > whose elements are of the
forms tnz0 , n ∈ Z. This subgroup is isomorphic to Z.
2π
(2) The take θ = n for some n ∈ Z+ . The rotation rp,θ generates a finite cyclic group which is isomorphic to
Zn .

The following statement is main reason we introduce the concept of group here. Assume Γ is a subgroup of
Iso(R2 , d). Define a relation over R2 as
p ∼Γ q if and only if q = f (p) for some f ∈ Γ.

Lemma 1.16 The relation ∼Γ is an equivalence relation over R2 whenever Γ is a subgroup of Iso(R2 , d). As a
result, the set of equivalence classes R2 / ∼Γ is well-defined.
Lecture Notes in Modern Geometry 9

Proof (1) Reflexive: Take any p ∈ R2 , notice that p = idR2 (p). Since Γ is a subgroup of Iso(R2 , d), the
identity element idR2 ∈ Γ, then p ∼Γ p.
(2) Symmetric: Assume p ∼Γ q. Then we can find some f ∈ Γ so that q = f (p). Since Γ is a subgroup of
Iso(R2 , d), the inverse map f −1 as the inverse of f is also in Γ. We write
p = f −1 (f (p)) = f −1 (q),
which shows that q ∼Γ p.
(3) Transitive: Assume p ∼Γ q and q ∼Γ r. Then we can find some f , g ∈ Γ so that
q = f (p), r = g(q).
By writing r = g(f (p)) = (g ◦ f )(p) and noticing that g ◦ f ∈ Γ since Γ is a subgroup of Iso(R2 , d), this
proves p ∼Γ r.

To simplify notation, we denote R2 / ∼Γ by R2 /Γ.

In next section, we are going to understand the geometry over R2 /Γ when Γ is “good”.

Example 1.17 Take Γ =< t(1,0) >. As a set, R2 /Γ can be identified with the strip [0, 1) × R. More precisely,
every point in this strip corresponds to a representative from a equivalence class.
Our next question is, how to give a natural distance function to this set R2 /Γ?

2 Euclidean surfaces

2.1 Metric spaces

Assume X is a set, a function

dX : X × X → R
is called a distance function, if dX satisfies the following three properties:
(1) dX (x, y) = dX (y, x);
(2) dX (x, y) ≥ 0 for any x, y ∈ X , and dX (x, y) = 0 if and only if x = y;
(3) dX (x, z) ≤ dX (x, y) + dX (y, z) for any x, y, z ∈ X .
A set with a distance is called a metric space and we use the pair (X, dX ) to denote.

Example 2.1 (1) The euclidean distance is a distance function over R2 .

(2) Similar as for R2 , we can introduce the euclidean distance for Rn as
p
d((x1 , x2 , · · · , xn ), (y1 , y2 , · · · , yn )) = (x1 − y1 )2 + (x2 − y2 )2 + · · · + (xn − yn )2 .
10 Rui Wang

(3) Consider the previous example R2 /Γ with Γ =< t(1,0) >. Denote by [p] the equivalence class on p. Define
a function
dΓ : R2 /Γ × R2 /Γ → R
as
dΓ ([p], [q]) = min d(p0 , q0 ).
p0 ∈[p],q0 ∈[q]

In fact, we should take the notation of inf instead of min to define the distance dΓ in general. The difference
between inf and min can be seen from the following example:
1
inf{x > 0| } = 0
x
but there is no x > 0 so that 1/x = 0. For this case, it is not proper to use notation min. In another word,
when we use min, we need to make sure that the minimal value can be taken by some element in the set we
are considering. (The same explanation goes to the difference between sup and max. ) However, for our
current case of Γ =< t(1,0) >, it is ok to use min, because we can prove that there must be some p0 ∈ [p]
and q0 ∈ [q] so that
(1) d(p0 , q0 ) = inf d(p0 , q0 ).
p0 ∈[p],q0 ∈[q]

The proof is left to you as an exercise.

Before we check dΓ is a distance function by definition, you should convince yourselves that this definition
matches our intuition for the geometry of cylinder.

Lemma 2.2 (R2 /Γ, dΓ ) is a metric space.

Proof We only need to check the three properties for a distance function.
(a) By definition, dΓ ([p], [q]) = inf p0 ∈[p],q0 ∈[q] d(p0 , q0 ) ≥ 0. Clearly, if [p] = [q], then dΓ ([p], [q]) = 0.
Now, notice that we can take p0 ∈ [p] and q0 ∈ [q] so that
d(p0 , q0 ) = dΓ ([p], [q]).
Then dΓ ([p], [q]) = 0 indicates p0 = q0 and then [p] = [p0 ] = [q0 ] = [q].
(b) dΓ ([p], [q]) = inf p0 ∈[p],q0 ∈[q] d(p0 , q0 ) = inf p0 ∈[p],q0 ∈[q] d(q0 , p0 ) = dΓ ([q], [p]).
(c) Take three equivalence classes [p], [q], [r] ∈ R2 /Γ, and pick p0 ∈ [p], r0 ∈ [r] so that
dΓ ([p], [r]) = d(p0 , r0 ).
Notice that we can further take q0 , q00 ∈ [q] so that dΓ ([p], [q]) = d(p0 , q0 ) and dΓ ([q], [r]) = d(q00 , r0 ).
(Why?)
Since both q0 , q00 are in [q], we can find some t ∈ Γ so that q00 = t(q0 ). Then
d(q00 , r0 ) = d(t(q0 ), t(t−1 )(r0 )) = d(q0 , t−1 (r0 )).
To simplify notation, let’s denote r00 = t−1 (r0 ). Now we have estimates
dΓ ([p], [r]) = d(p0 , r0 ) ≤ d(p0 , r00 ) ≤ d(p0 , q0 ) + d(q0 , r00 ) = d(p0 , q0 ) + d(q00 , r0 )
= dΓ ([p], [q]) + dΓ ([q], [r]),
and the proof is done.
Lecture Notes in Modern Geometry 11

Definition 2.3 Assume (Xi , di ), i = 1, 2, are two metric spaces. A bijective map f : X1 → X2 is called an
isometry if it satisfies that
d1 (x1 , y1 ) = d2 (f (x1 ), f (y1 )), for any x1 , y1 ∈ X1 .

Clearly, if f is an isometry, then f −1 is also an isometry.

The set of metric spaces has a relation ∼ defined by isometries in this way:
(X1 , d1 ) ∼ (X2 , d2 ) if and only if there exists some isometry from X1 to X2 .
Check that such relation is an equivalence relation over metric spaces. We say two metric spaces (X1 , d1 ) and
(X2 , d2 ) are isometric, if (X1 , d1 ) ∼ (X2 , d2 ).

Example 2.4 In calculus class, we have learned that a cylinder (of radius 1) can be defined as a surface in R3
C := {(x, y, z) ∈ R3 |x2 + y2 = 1}.
The cylinder C posses a distance function defined as the smallest arc length along all paths over C connecting the
two points, i.e.,
Z 1
dC (p, q) = inf |γ̇(t)|dt.
γ is smooth and γ(0)=p,γ(1)=q,γ⊂C 0

Exercise 2.5 (1) Check dC is a distance function on C;

(2) Try to prove (C, dC ) and (R2 /Γ, dΓ ) with Γ =< t(2π,0) > are isometric.

Proof For this, we first introduce a bijective map between C and R2 /Γ. Denote the map
y
φ : C → R2 , (x, y, z) 7→ (arctan , z)
x
and φ̄ = [φ] : C → R2 /Γ. (The natural domain of the function arctan is (−∞, ∞) with range (−π, π). We
extend it by defining φ(−1, 0) = (−π, 0). ) It is not hard to check the so-defined map φ̄ is a bijection.
For any p, q ∈ C, a smooth path connecting p, q over C can be written as a parametrized curve
r(t) = (x(t), y(t), z(t)),
with r(0) = p, r(1) = q and x(t)2 + y(t)2 = 1 for any t ∈ [0, 1]. The distance between p and q by definition is the
shortest arc length. Hence we don’t need to consider all curves connecting p, q but can focus on the ones whose
image under φ live in some fundamental region with
|φ(r(t2 )) − φ(r(t1 ))| ≤ π for any t1 , t2 ∈ [0, 1].
(You need some argument to show this. Though a rigorous proof is not required, you need to convince yourselves by
geometry intuition.) Now WLOG, let’s assume p = (−1, 0, 0) and q = (cos θ0 , sin θ0 , zq ) with θ0 ∈ [−π, 0]. We
calculate the arc length of r for this case using the cylinder coordinates: r̃(θ) = (cos θ, sin θ, z̃(θ)), θ ∈ [−π, θ0 ],
as follows
Z θ0 q
LR3 (r) = (cos θ)0 2 + (sin θ)0 2 + z̃0 (θ)2 dθ
−π
Z θ0 p
= 1 + z̃0 (θ)2 dθ
−π
12 Rui Wang

On the other hand, the image of r under φ is the parametrized curve φ(r̃(θ)) = (θ, z̃(θ)), θ ∈ [−π, θ0 ], in R2 .
Notice that the arc length of this curve is calculated as
Z θ0 p
LR2 (φ(r̃)) = 1 + z̃0 (θ)2 dθ.
−π
Hence from the calculation above,
LR3 (r) = LR2 (φ(r̃)).
Now combine with the definition
dC (p, q) := inf LR3 (r)
r
= inf LR2 (φ(r̃))
r
= d(φ(p), φ(q))
= dΓ (φ̄(p), φ̄(q)).
where the last second equality uses the fact that on the euclidean plane R2 , the shortest arc length connecting two
points is the line segment and the shortest arc length is the euclidean distance.
This concludes that dC is a distance function since dΓ is a distance function as we have shown. Moreover, (C, dC ),
(R2 /Γ, dΓ ) are isometric via the map φ̄.

Exercise 2.6 Assume (Xi , di ), i = 1, 2 are two metric spaces and φ : X1 → X2 is an isometry. Prove that φ maps
a circle in X1 to a circle in X2 with the same circumference. Here we define the circumference of a circle as the
supremum of the sum of all line segment connecting adjacent division points on the circle.

2.2 Locally euclidean surfaces

Assume (X, dX ) is a metric space. Take  > 0 and a point x ∈ X . A disk centered at p with radius  is defined as
U(X,d) (x; ) = {y ∈ X|dX (x, y) ≤ }.
If metric space is the euclidean plane (R2 , d), we denote such disk by D(x; ).
When we restrict the metric U(X,d) (x; ), we obtain a metric space (U(X,d) (x; ), dX ). For the case in euclidean plane,
we call it a euclidean disk.

Definition 2.7 A metric space (S, dS ) is called locally euclidean surface, if for any p ∈ S, there exists some  > 0,
so that the  disk centered at p is isometric to some euclidean disk.

Example 2.8 (1) The euclidean plane (R2 , d) is a locally euclidean plane: For each p ∈ R2 , we can take  = 1
and the translation t−p is the isometry from D(p; 1) to D(0; 1).
(2) An open subset in (R2 , d) is a subset U ⊂ R2 so that each x ∈ U , there exists some disk D(x; ) centered at
x inside U . Any open subset U of R2 with respect to the distance function d is locally euclidean.
(3) (R2 /Γ, dΓ ) with Γ generated by a translation is locally euclidean.
Proof WLOG, we take Γ =< t(1,0) >. Pick any point [p] ∈ R2 /Γ and p0 = (x0 , y0 ) ∈ [p]. Notice that
U(R2 /Γ,dΓ ) (p0 ; 14 ) is isometric to the euclidean disk centered at (x0 , y0 ) with radius 14 by the definition of dΓ .
This shows that R2 /Γ is locally euclidean.
Lecture Notes in Modern Geometry 13

2.3 More examples

In this section, we construct more examples in a similar way as we construct the cylinder R2 / < t(1,0) >.

Example 2.9 The twisted cylinder. Assume f is a proper glide reflection (i.e., not a reflection). R2 / < f > is
a twisted cylinder. It is also a locally euclidean surface. A portion of a twisted cylinder is called a Möbius band
whose picture is as given below:

Remark 2.10 (1) A cylinder is orientable while a twisted cylinder is not orientable.

(Over a Möbius band, when the ant comes back to the same point for the first time, its antennae pointing
to the opposite direction as it starts off. Viewing the surface in R3 (which is orientable), this description is
equivalent to state the Möbius band is one-sided. )
(2) Both a cylinder and a twisted cylinder are unbounded (not compact).

Example 2.11 (1) The torus. Take Γ =< t(1,0) , t(0,1) >. Then R2 /Γ is the torus.

(2) The Klein bottle. Take Γ =< t(1,0) , f >, where f is a glide reflection along the y-axis with translation t(0,1) .
Then R2 /Γ is the Klein bottle.
14 Rui Wang

Remark 2.12 (1) A torus is orientable while a Klein bottle is not orientable. Try to find a Möbius band in a
Klein bottle.
(2) Both torus and Klein bottle are compact locally euclidean surfaces.
So far, we have constructed four types of locally euclidean surfaces.
Now let’s see two ‘bad’ examples which are either not a metric space or not locally euclidean.
Example 2.13 (1) Consider the subgroup Γ generated by two translations as follows:
Γ =< t(0,1) , t(0,√2) > .
Over the quotient space R2 /Γ, we try to define a distance function dΓ as
dΓ ([p], [q]) = inf d(p0 , q0 ).
p0 ∈[p],q0 ∈[q]

We take two point [(0, 0)], [(0, 1)] ∈ R2 /Γ. Notice that
√ √
[(0, 0)] = {(0, m + 2n)|m, n ∈ Z}, [(0, 1)] = {(0, m + 2n + 1)|m, n ∈ Z}.
Notice that for any  > 0, we can always find m, n ∈ Z so that
√
|m + 2n − 1| < .
√
(In fact, {m + 2n|m, n ∈ Z} is dense in R.) Hence
dΓ ([(0, 0)], [(0, 1)]) = 0
but [(0, 0)] 6= [(0, 1)]. This shows that dΓ is not even a metric.
(2) Γ =< rO, π4 >. (R2 /Γ is an orbifold.) We show now (R2 /Γ, dΓ ) is not locally euclidean. Consider the
point [(0, 0)] ∈ R2 /Γ. Assume there exists some local isometry
φ : R2 /Γ → R2 .
WLOG, we can assume φ([(0, 0)]) = (0, 0) ∈ R2 . Take a circle γ in R2 /Γ centered at [(0, 0)] with radius
. Recalling from Exercise 2.6, φ(γ) should be a circle in R2 with radius  since φ is local isometry and
π
preserves the circumference. The circumference of γ is 2π
8 = 4 , while the circumference of φ(γ) is 2π.
We get contradiction and there is no such local isometry φ.
Notice in this example, there exists some point p ∈ R2 (the origin for this example), which is a fixed point
of the subgroup Γ. In general, when this happens, we can not expect locally euclidean property.
In next section, we abstract some terminology to exclude the phenomenons show up in these two bad examples.
Lecture Notes in Modern Geometry 15

2.4 Conditions on group actions

We have shown that any subgroup Γ of Iso(R2 , d) induces a set of equivalent classes, which we denote by R2 /Γ.
Moreover, there is a natural surjective map
π : R2 → R2 /Γ, p 7→ [p].
From the viewpoint of group action, each equivalence class [p] is a Γ-orbit through p. Hence the space R2 /Γ
consists of (disjoint) orbits of Γ. For this reason, we also call [p] a Γ-orbit through p when we want to emphasize
the group Γ-action on R2 .

Definition 2.14 Assume V is a subset of R2 . Call q ∈ R2 is a limiting point of V , if every euclidean disk centered
at q contains infinitely many points in V .

From the definition, a subset V has no limiting point means that every point in q ∈ R2 , we can find a disk center
at q, so that the disk contains only at most one point from V . As a result, a subset containing only finite points has
no limiting point. However, be careful that a subset without limiting point is not necessary finite.

Example 2.15 (1) Z × Z ⊂ R2 has no limiting point.

(2) Every point in R2 is a limiting point of Q × Q ⊂ R2 .

Definition 2.16 A subgroup Γ of Iso(R2 , d) acts on R2 discountinuously, if every orbit has no limiting point.

Example 2.17 (1) The subgroups that we use to construct cylinder, twisted cylinder, torus and the Klein bottle
all act on R2 discountinuously.
(2) (This is Example 2.13 (1).) Consider the subgroup Γ generated by two translations as follows:
Γ =< t(0,1) , t(0,√2) > .

Consider the orbit through (0, 0) ∈ R2 , which is

√
[(0, 0)] = {(0, m + 2n)|m, n ∈ Z}.
√
Notice that the subset {m + 2n|m, n ∈ Z} is dense in R, hence Γ action on R2 is not discontinuous.

Next, we introduce another concept from group actions. Assume Γ is a subgroup of Iso(R2 , d). For each point
p ∈ R2 , the following subset of Γ
Γp := {g ∈ Γ|g(p) = p}
is a subgroup of Γ and is called the isotropy group of Γ at p.

Definition 2.18 We say a subgroup Γ of Iso(R2 , d) acts on R2 freely, if for every point p ∈ R2 , the isotropy
group Γp contains only the identity map.

Example 2.19 (1) The subgroups that we use to construct cylinder, twisted cylinder, torus and the Klein bottle
all act on R2 discountinuously.
(2) Consider the subgroup generated by a reflection RfL (which contains only two elements). Then every point
on L is fixed by the two elements and hence this group action is not free.
16 Rui Wang

(3) (This is Example 2.13 (2).) Consider the subgroup generated by a rotation rp,θ with θ 6= 0. Then p is fixed
by the whole subgroup hence this action is not free.

Proposition 2.20 Assume Γ is a proper subgroup of Iso(R2 , d). The quotient surface (R2 /Γ, dΓ ) is a locally
euclidean metric space if and only if the subgroup Γ acts on R2 discontinuously and freely.

Proof We have seen from examples that if the group action of Γ is not free or not discontinuous, R2 /Γ is not
locally euclidean. This shows that the locally euclidean property implies the discontinuous and free properities.
Conversely, assume Γ acts on R2 discontinuously and freely, then we can show that Γ must be generated by one
translation or a glide reflection, or by two linear independent translations or glide reflections. For these four cases,
we have shown that they are locally euclidean. (In the proof here, we used the three reflections theorem which
avoids the using of assumption of properness for the group action. )

From the proof, in fact we have derived the following result.

Corollary 2.21 Assume Γ is a subgroup of Iso(R2 , d). The quotient surface (R2 /Γ, dΓ ) is a locally euclidean
metric space if and only if it is one of the following four metric spaces:
(1) a cylinder. For this case, Γ is generated by a translation;
(2) a twisted cylinder. For this case, Γ is generated by a glide reflection;
(3) a torus. For this case, Γ is generated by two linearly independent translations;
(4) a Klein bottle. For this case, Γ is generated by two linearly independent glide reflections;

In next section, we are going to consider more general locally euclidean surfaces and explain the corresponding
Killing-Hopf theorem.

2.5 The Killing-Hopf theorem

To introduce the Killing-Hopf theorem, we need two mild conditions which we introduce now.

Definition 2.22 A metric space (X, dX ) is called path connected, if for any two points x, y ∈ X , there exists a
continuous path in X joining x and y.

For example, the whole euclidean plane is path connected, but the euclidean plane excluding x-axis is not path
connected. For a non path connected metric space, we can just look at its path connected components instead. For
a metric space, if it is path connected then any both open and closed subset is either empty or the whole set. (The
latter condition that any both open and closed subset is either empty or the whole set is called connectness. In
general, a path connected space must be connected, but a connected space may not be path connected. )

Definition 2.23 A metric space (X, dX ) is called complete, if every Cauchy sequence converges to some point in
X.
 Lecture Notes in Modern Geometry 17

This definition is too analytic and not easy to use in geometry. Luckily, mathematicians prove that for a metric
space which is locally euclidean (in fact, much more general than locally euclidean), it is complete if and only if
every line segment (which is called geodesic for general cases) can be continued indefinitely. This is the property
for a complete metric space that we are going to use. (Usually, this property is referred as geodesic completeness.
)

Theorem 2.24 (Killing-Hopf theorem for euclidean case) Each complete, connected locally euclidean surface
is one of the following four metric spaces of the form R2 /Γ:

(1) a cylinder;
(2) a torus;
(3) a twisted cylinder;
(4) a Klein bottle.

Before we give a proof, let’s see the importance of this theorem: The locally euclidean condition is a local condition,
while the classification of types of four quotient surfaces is a global topology condition. A mathematical statement
becomes extremely important once it indicates that certain local information could lead to a global result. For
example, using the Killing-Hopf theorem, we can immediately conclude:

Example 2.25 (1) The topological 2-sphere can not be locally euclidean.
(2) The surfaces with more than one hole can not be locally euclidean.

We explain the key idea in the proof.

Step 1. We construct a covering π : R2 → S, i.e., we look for a surjective map π so that it is also local isometry.
First we take an arbitrary point oS ∈ S as the image of (0, 0) ∈ R2 by π . Then by the locally euclidean
property, there exists a small disk D() in R2 centered at (0, 0) and a small disk US (oS ; ) in S centered at
oS which are isometric to each other. We denote this isometry by π : D() → US (oS ; ).
Notice that the image of every ray in D() by π is a line segment in S. Every point in R2 belongs to a unique
ray. By extending the corresponding image segment, we define a map from R2 to S. The definition is based
on the completeness assumption for S. Next we need to use the interplay of connected, completeness and
locally euclidean property to show that π is a local isometry and surjective. Since the proof is a little subtle
and beyond this class, we omit here. (Interested readers can refer the textbook. ) This map is called the
pencil map and was constructed first by Hopf [1925].
Step 2. First, we look at the set of euclidean isometries which are compatible with the covering map π :

Γ := {f ∈ Iso(R2 , d)|π ◦ f = π}.

It is not hard to see Γ is a subgroup of Iso(R2 , d). Next, we can show that R2 /Γ is isometric to S and this
finishes the proof.
18 Rui Wang

3 The sphere

3.1 The sphere S2 and its isometry group

In Section 1, we have studied the metric space (R2 , d) which is called the euclidean plane. In this section, we
introduce an analogue, the sphere, and derive parallel results as we have obtained for the euclidean plane. To do
this, the first thing we need to do is to define a distance function on the sphere.
A unit sphere is a surface in R3 defined as
S2 := {(x, y, z) ∈ R3 |x2 + y2 + z2 = 1},
and a distance function for S2 is defined as the shorter arc length between two points through the great circle.
Equivalently, we can write the distance as
1
dS2 (p, q) = 2 arcsin d(p, q),
2
where d is the euclidean distance for R3 . In this section, when we mention the sphere S, we always mean the
metric space (S2 , dS2 ). (If we just want to refer the topological sphere, we use the notation S2 instead. )

Exercise 3.1 (1) Check (S2 , dS2 ) is a metric space.

(2) Show that this distance dS2 is the same as the distance defined as the shortest path connecting the two points
over the sphere. (It is easier to use spherical coordinates to calculate arc length.)
(3) Show that (S2 , dS2 ) is not locally euclidean. (Hint: Try to calculate the circumference of a circle with small
radius and compare it with the one of its image in the euclidean plane.)

As for the euclidean plane, we now study the isometry group for Iso(S2 , dS2 ). We can also define rotation and
reflection for the current situation, but since we have shown the three reflection theorem for euclidean plane, we
might focus on the key concept first: Let’s define reflections over S2 . Recalling that to describe a euclidean
reflection, we only need a line which is defined in Exercise 1.4 (4) using distance only. We define a line in S2 as
the set
Cp,q := {s ∈ S2 |dS2 (p, s) = dS2 (q, s)},
for two distinct points p, q ∈ S2 . This definition directly leads to the fact that an isometry of S2 maps a line to a
line.

Exercise 3.2 Check that a line in S2 is a great circle in the sphere. As a result, an isometry of S2 maps a great
circle to a great circle.

From this, we also see that any two distinct points on the sphere S2 , there exists a unique line (i.e., a great circle)
going through them. This is the same as the euclidean plane. However, now for S2 , the parallel axiom no longer
holds: Given a line and a point which is not on the line, there is no great circle going through the point and without
intersecting the given line. This causes the difference of geometry between the euclidean plane and the sphere.
Now we define the reflection. Assume C is a line (i.e., a great circle) in S2 . A reflection about C which we denote
by RfC maps every point p ∈ S2 to the point q ∈ S2 so that the line determined by p, q as Cp,q defined as above is
the given line C.
Lecture Notes in Modern Geometry 19

Exercise 3.3 Prove that map RfC is well-defined. For this, you need to prove that such q exists and is unique.
Moreover, check that a reflection is an isometry over S2 and RfC ◦ RfC = idS2 .
Assume ` is an axis through the origin with a positive direction being chosen. We can define the rotation around
` for points on S2 as rotations on the plane which going through the point the perpendicular to `. We use r`,θ to
denote a rotation with θ -angle.
For isometries of the euclidean plane, we also consider translations and rotations and proved that a translation or
a rotation can be written as compositions of two reflections. Whether the composition of two refections gives
a translation or a rotation is determined by whether the two lines have intersection or not. For the sphere S2 ,
every two distinct lines (great circles) must intersect at two points which are antipodal to each other, i.e., the axis
connecting these two points must go through the origin. Their composition is a rotation around this axis with the
double angle between these two great circles. Conversely, any rotation around an axis through the origin can be
decomposed into two reflections. In another word, for the sphere S2 , there are no translations but only rotations.
This will bring a different result for the corresponding spherical Killing-Hopf theorem.
We now state the three reflections theorem for the sphere whose proof is a strict analogue of the corresponding
result for the euclidean plane.
Theorem 3.4 (The three reflections theorem for the sphere) Any isometry for S2 can be written as compositions
of one or two or three reflections.
Exercise 3.5 Prove the three reflections theorem for the sphere case.
Among these isometries, the orientation-preserving ones and orientation-reversing ones are defined in the same
way as for the case of Iso(R2 , d). Similarly, we use the notation Iso+ (S2 ) to denote the set of orientation-preserving
isometries, but now it contains only rotations. The same as the euclidean case, Iso+ (S2 ) is a normal subgroup of
Iso(S2 ) with index 2. People also use SO(3) to denote Iso+ (S2 ). We are going to discuss more on it in Section ??.
For the orientation-reversing isometries, we can similarly prove that any composition of three reflections is a
glide reflection, but now, a glide reflection means a rotation composited with a reflection along the great circle
perpendicular to the rotation axis.

3.2 Locally spherical surfaces

Theorem 3.6 (The Killing-Hopf theorem for spherical surfaces) A connected complete locally spherical surface
is either a sphere or a projective plane (RP2 ).
We explain the key difference between the current Killing-Hopf theorem and the one for the euclidean plane. The
point is, a nontrivial subgroup Γ of Iso(S2 ) which discountinuously and freely acts on S2 can only be the one
generated by the glide reflection fπ whose rotation angle is π .
Exercise 3.7 Explain the above point in details. In particular, explain why the glide reflection with other angles
can not generate a discountinuous and free group action.
The quotient surface S2 / < fπ >, which is called the projective plane (because it can be understood as the space
of lines in R3 ), can be understood as gluing the antipodal points on a sphere. As the Klein bottle, it is compact,
without boundary and not orientable (one-sided surface). It is not embedded in R3 either.
Exercise 3.8 Find a fundamental region for the projective plane and try to find a Möbius band in the fundamental
region.
20 Rui Wang

3.3 Relation to the complex plane

For the euclidean plane, we know as a set, it is bijective to the complex plane C and we can represent basic
isometries by maps over it. For the sphere, we can also do it (as a set only) using the so-called stereographic
projection. Of course, the sphere and the complex plane are not isometric but they are conformal in the sense that
the angles are preserved.
We explain the construction of stereographic projection first. Let’s denote the north pole (0, 0, 1) by N . For any
point p ∈ S2 which is not N , we can find a unique line in R3 going through N and p. This line intersection with
the {z = −1}-plane at a unique point, which we define as the image of p under the stereographic projection. By
this way, every point on S2 \ {N} one-to-one corresponds to point on the {z = −1}-plane which can be2 identified
with the complex plane C. We can formally assign the infinity as the image of the north pole. Denote by C the
complex plane adding the infinity, then the stereographic projection is a bijection from S2 to C.

Explicitly, the stereographic projection pr has the expression

2x 2y
pr(x, y, z) = ( , ) ∈ C.
1−z 1−z
Its inverse map
4u 4v u2 + v2 − 4
pr−1 (u, v) = ( 2 , , ) ∈ S2 .
u + v2 + 4 u2 + v2 + 4 u2 + v2 + 4
The stereographic projection maps the great circle {(x, y, 0) ∈ S2 |x2 +y2 = 1} to the circle {(u, v, −1)|u2 +v2 = 4}.
From this, clearly it is not an isometry (Why?). Instead, it assigns C a different metric space structure from the
euclidean plane. One important property for stereographic projection is that it is conformal, i.e., it preserves angles.
Moreover it maps a circle on S2 to a circle in the {z = −1}-plane.
Lecture Notes in Modern Geometry 21

Exercise 3.9 Assume C is the great circle in S2 determined by the plane ax + by + cz = 0 with a2 + b2 + c2 .
Then its stereographic projection is the circle in the plane {(x, y, z) ∈ R3 |z = −1} with center (− 2a 2b
c , − c , −1) and
2
radius |c| .

3.4 The inversions

Consider a circle in the complex plane centered at z0 ∈ C with radius R as Cz0 ,R := {w ∈ C||w − z0 | = R}. A
inversion about Cz0 ,R is defined as a map
InvCz0 ,R : C \ {z0 } → C \ {z0 }
as
z0 w̄ + (R2 − |z0 |2 )
InvCz0 ,R (w) = .
w̄ − z¯0
(We can also include z0 by extending C to C = C ∪ {∞}. ) The geometric meaning of inversion is as shown in
the following picture that the inversion of the point P is P0 with the relation that
|SP| · |SP0 | = R2 .

In fact, the same picture also shows the relation between the inversion and the reflection about a great circle in the
sphere S2 .

Proposition 3.10 Assume C is a great circle in S2 whose image under the stereographic projection is denoted as
C0 . We know from last section that C0 is a circle in the complex plane C ∼= {(x, y, z) ∈ R3 |z = −1}. There is the
relation between the reflection about a great circle C and the inversion about the circle C0 as
pr ◦ RfC = InvC0 ◦ pr.
Here pr : S2 → C denotes the stereographic projection.

Further combine with Exercise 3.9, we get the following result.

Proposition 3.11 Assume C is the great circle in S2 determined by the plane ax+by+cz = 0 with a2 +b2 +c2 = 1.
Then the inversion about the stereographic projection of C is
2(a + bi)w̄ − 4c
InvC0 (w) = .
−cw̄ − 2(a − bi)

Recall that a rotation is a composition of two reflections, from Proposition 3.11 we obtain the following expression
for a rotation under the stereographic projection.
22 Rui Wang

Proposition 3.12 Any rotation r over the sphere S2 has the following expression
2uw + 4v
pr ◦ r ◦ pr−1 (w) = ,
−v̄w + 2ū
with some u, v ∈ C with |u|2 + |v|2 = 1.

Proof We sketch the proof and leave the details to be checked by you. Assume r is decomposed into two
reflections as RfC2 ◦ RfC1 whose inversions under stereographic projections are
2λj w̄ − 4νj
pr ◦ RfCj ◦ pr−1 = , j = 1, 2.
−νj w̄ − 2λj
Here λj ∈ C and νj ∈ R with |λj |2 + νj2 = 1.
Then we calculate their composition and get
2uw + 4v
pr ◦ r ◦ pr−1 (w) = (pr ◦ RfC2 ◦ pr−1 ) ◦ (pr ◦ RfC1 ◦ pr−1 )(w) =
−v̄w + 2ū
where
u = λ¯1 λ2 + ν1 ν2 , v = λ1 ν2 − λ2 ν1 .
Direct checking shows that |u|2 + |v|2 = 1.

We remark that the coefficients 2 and 4 will disappear if we do the stereographic projection to the xy-plane instead.
2uw+4v
This expression is interesting in the following sense: Consider the map which maps the map −v̄w+2ū to the matrix
 
u v
. Then we find that the composition of maps commutes with the matrix multiplications.
−v̄ ū
The image set of 2 × 2 complex matrices forms a subgroup which is called the group of special unitary matrices
 
u v
SU(2) := { |u, v ∈ C, |u|2 + |v|2 = 1}.
−v̄ ū

On the other hand, a rotation over S2 corresponds to an orientation-preserving linear map which preserves norms
for the linear space R3 . From linear algebra class, we know such matrices form a subgroup of 3 × 3 real matrices
which is called the group of special orthogonal matrices
SO(3) := {A ∈ GL3 (R)|AAT = I, det(A) = 1}.
By this way, the above construction in fact gives a map
 
u v 2u · (·) + 4v
Φ : SU(2) → SO(3), → r(·) = .
−v̄ ū −v̄ · (·) + 2ū

The above construction has shown that the map Φ is surjective. However, it is not injective but a 2-to-1 map.

Remark 3.13 From our interpretation for the group SO(3), it is clear to see that it has a smooth manifold structure
which is diffeomorphic to RP3 since it corresponds to equivalence classes by antipodal points in C2 ∼ = R4 . It is
connected, but not simply-connected.
The non-contractible loop corresponds to the Balinese plate trick https://en.wikipedia.org/wiki/Plate_
trick.
Lecture Notes in Modern Geometry 23

3.5 The Gauss-Bonnet formula

For three points A, B, C which are not in a line on the euclidean plane, it is well-known that the sum of interior
angles of the triangle ∆ABC is π . Now we ask the same question for three points A, B, C which are not in a great
circle on the unit sphere S2 . A spherical triangle is a region in S2 formed by three great circular arcs intersecting
pairwise in three vertices. For example, in the following picture, the ∆ABC is a spherical triangle.

An interior angle of a spherical triangle is defined as the angle between the two planes where the two adjacent great
circular arcs live in. Then we can ask, what is the sum of the three interior angles in a spherical triangle? Some
special examples show us that this sum must be great than π but is no longer a constant. Is this possible for us to
obtain an equality for it then? We now try to derive a formula for the general case.
Consider the spherical triangle ∆ABC as in the above picture and notice that
Area(ACA1 B) = Area(∆ABC) + Area(∆BCA1 ).
α
Here A1 is the antipodal point of A and thus Area(ACA1 B) = 2π · 4πR2 = 2αR2 , where R is the radius of the
sphere. Similarly, we also have
Area(∆ABC) + Area(∆ACB1 ) = 2βR2
Area(∆ABC) + Area(∆ABC1 ) = 2γR2 .
Sum them, we get
2(α + β + γ)R2 = 3Area(∆ABC) + (Area(∆BCA1 ) + Area(∆ACB1 ) + Area(∆ABC1 )).
Further notice that
Area(∆BCA1 ) + Area(∆ACB1 ) + Area(∆ABC1 ) = 2πR2 − Area(∆ABC).
Put these together, we obtain the following formula
1
α+β+γ =π+ Area(∆ABC).
R2
This formula not only shows that the sum of the three interior angles in a triangle is greater than π , but also
explicitly relate the difference with the area of the triangle.
In fact, the constant R12 also has a geometric meaning, which is nothing but the Gaussian curvature of the sphere
with radius R. In general, the Gaussian curvature of a surface at a point is defined as the multiplication of the
maximum and minimum values of the curvatures of all curves in the surface through the point. (Euler proved in
1760 that the curves in the surface which take the maximum and minimum values are perpendicular to each other
at the point. )
24 Rui Wang

A more general formula for triangles in a constant curvature surface is then given as
(2) α + β + γ = π + κ · Area(∆ABC)
where κ is the Gaussian curvature. The euclidean plane has zero Gaussian curvature and then this formula is just
the one that states the sum of interior triangles in a euclidean triangle is π . In later lectures, we are going to study
the hyperbolic plane whose Gaussian curvature is constant −1 and then the sum of angles of a hyperbolic triangle
is π minus the hyperbolic area of the triangle.
The formula (2) is referred as (a simple version of) the famous Gauss-Bonnet formula.

Exercise 3.14 (1) Prove the following formula for any spherical polygon
Area(polygon) + sum of exterior angles = 2π.
(Notice that this formula is independent of the number of vertices of the polygon.)
(2) Try to prove that two spherical triangles with the same angles must be isometric. (This result shows another
difference between the spherical geometry and the euclidean geometry.)

4 The hyperbolic plane

4.1 The pseudosphere

As we have seen that the sphere is a surface of constant positive curvature. The nonzero curvature of the sphere
make it has different geometry from the euclidean plane. Next, it is natural to look for some surface with negative
constant curvature and then the geometry on this surface is expected to be opposite to the sphere. Locally, we know
that a point on a surface with a negative curvature is a saddle point.

Let’s now introduce example which has saddle points everywhere.

u −u
Consider the curve v = cosh u = e +e2 in the uv-plane. Then we draw its involute curve, which is called tractrix,
as the red curve shown in the following picture. The involute curve is determined by the condition that
|PQ| = arc length PS
and that PQ is tangent to the hyperbolic consine curve at P.
Lecture Notes in Modern Geometry 25

Next, we obtain the revolution of the tractrix about the u-axis. The surface we obtain in this way is called a
pseudosphere. The Gaussian curvature of the pseudosphere at Q can be calculated as
1 1
κ=− · .
|PQ| |QR|
We assume P has coordinate (u, cosh u) in the uv-plane. Then
Z ur
dv eu − e−u
|PQ| = arc length PS = 1 + ( )2 du = sinh u = .
0 du 2
On the other hand, use the slope of the tangent line PR at P, we can calculate out that R has coordinate (u−coth u, 0),
where
sinh u eu − e−u
coth u := = u .
cosh u e + e−u
1
It follows |QR| = |PR| − |PQ| = sinh u . As a result, we have finished showing that the Gaussian curvature is
constant −1 over the pseudosphere.
We end this section by discussing the defect of the pseudosphere.
(1) This surface is not smooth at u = 0 as we can see that the point (0, 1) in the uv-plane is sharp which makes
the whole circle from rotation become singularities. If we remove this circle from the surface, the resulting
surface is not complete.
(2) It is not easy to describe ‘lines’ for the pseudosphere.
In fact, Hilbert in 1901 proved that any surface of constant negative curvature smoothly embedded in R3 can not be
complete. This in fact break our dreams to look for a model as S2 for the negative curvature case. For this reason,
we have to get rid of the requirement that the metric for the surface is induced from the euclidean metric R3 .

4.2 The hyperbolic plane

The hyperbolic plane is a metric space whose set of points can be taken as the upper half plane
H2 := {(x, y) ∈ R2 |y > 0}.
26 Rui Wang

We use the metric on the pseudosphere induces a metric on H2 as follows: define x as the rotation angle around
u-axis and τ the arc length from S to Q. We can use (x, τ ) to parametrize the pseudosphere. Then we calculate
the infinitesimal distance between points (x, τ ) and (x + dx, τ + dτ ) over the pseudosphere as
ds2 = (vdx)2 + (dτ )2
= v2 (dx)2 + (dτ )2 .
Here (u, v) is the uv-coordinate for the point (x, τ ). A little calculation shows that v = e−τ . Hence the infinitesimal
distance is p
ds = e−2τ (dx)2 + (dτ )2 .

We change parameter again by defining y = eτ , then dy = eτ dτ and then this infinitesimal distance can be written
as p
dx2 + dy2
ds = .
y
Notice that, though by change of variable, our parameters (x, y) has range [0, 2π) × (1, ∞), we can safely extend
the domain of (x, y) to the whole upper half plane H2 and regard it as an infinitesimal distance on H2 . From now
on, we use the notation dsH2 to denote it.
Using dsH2 , we define a metric over the upper half plane H2 as the following way: For any two points p, q ∈ H2 ,
the hyperbolic distance is defined as
Z
dH2 (p, q) = inf dsH2
γ:[0,1]→H2 ,γ(0)=p,γ(1)=q γ
Z 1p 0 2
x (t) + y0 (t)2
= inf dt
γ:[0,1]→H2 ,γ(0)=p,γ(1)=q 0 y(t)

By the construction, we have the following conclusion.

Lemma 4.1 (H2 , dH2 ) is a metric space. It is called the hyperbolic plane.

In fact, we are going to show later that the path which realizes the minimal distance is either lines parallel to y-axis
or half circles with centers at x-axis.
To give an intuition for the difference between the hyperbolic distance and the euclidean distance, we see the
following example.

Example 4.2 Consider the half circle with center C = (c, 0) and radius r on the upper half plane. Let’s calculate
the arc length from P to Q along this half circle as shown in the following picture.
Lecture Notes in Modern Geometry 27

We can parametrize the curve as

x = c + r cos θ, y = r sin θ, θ ∈ [α, β].
We calculate the hyperbolic arc length as
p β
x0 (θ)2 + y0 (θ)2
Z
dθ
α y(θ)
Z β
1
= dθ
α sin θ
1 (cos β − 1)(cos α + 1)
= log | |.
2 (cos β + 1)(cos α − 1)
We notice that, this arc length is independent of the radius r but only depends on angles.
For the points A = (x, y) and B = (x0 , y0 ) = (x, y0 ), we can calculate the hyperbolic distance as
Z y0 √ 2
0 + 12
dy = log y0 − log y.
y y

4.3 The Poincaré disk

Now we introduce another model for the hyperbolic plane. Consider the map from C → C defined as
iz + 1
φ(z) =
z+i
The map φ is a bijective, orientation-preserving map. It maps 1 to 0, maps 0 to −i, maps ∞ to i. Moreover, it
maps the upper half plane to the open unit disk
D2 := {z ∈ C||z| < 1},
with the real axis to the unit circle. We can calculate out the inverse map which is
w+i
ψ(w) = .
iw + 1
Using the complex expression of the infinitesimal distance
p
dx2 + dy2 |dz|
dsH2 = = ,
y Imz
we can formally calculate the corresponding infinitesimal distance in the unit disk D2
w+i
|d iw+1 | 2|dw|
dsD2 = w+i = .
Im( iw+1 ) 1 − |w|2

This calculation in fact shows that the metric space (D2 , dD2 ) with the distance function induced from dsD2 is
isometric to (H2 , dH2 ). In anther word, these two models are different coordinate system for the same geometric
space. We are going to interchange the use of these two models freely according to our different purposes.
There are two important properties for the D2 -model that we would like to point out:
28 Rui Wang

(1) It maps the euclidean lines parallel to y-axis and half circles centered on x-axis in H2 to arcs which are
perpendicular to the boundary circle of the disk D2 (including diameters). These are in fact D2 -lines. In
particular, it maps the y-axis in H2 to the y-axis in D2 and maps the unit circle center at origin in H2 to the
x-axis in D2 .
(2) The map φ : C → C in fact √ is the reflection about y-axis composed with the inversion about the circle
centered at −i with radius 2. From this, we can at least see that the map φ preserves angles. (In fact, all
such kind of maps which are only in z (i.e., with no z̄) preserve angles.)

4.4 Hyperbolic isometries

In this section, we first see five examples of hyperbolic isometries. Some of them have simple expressions in the
H2 -model and some have simple expressions in the D2 -model. Then we see that these are isometries are plenty
enough to move every hyperbolic triangle into some favored positions. Using these, we are able to obtain the three
reflections theorem just as for the euclidean case.
Before we start, we give the following lemma whose proof we skip.

Lemma 4.3 A smooth map f = (f1 , f2 ) : H2 → H2 is a hyperbolic isometry if and only if

p
(df1 )2 + (df2 )2 |dz|
= .
f2 Imz

Now let’s see the five examples and the check of they are hyperbolic isometries using Lemma 4.3 is left to you.
The first three examples have simple expression in the upper half plane model H2 .
(1) For any α ∈ R, define tα : H2 → H2 as
tα (z) = z + α.
This map is called a limit rotation. The reason to regard it as a rotation is it is a composition two reflections
with the two reflection axises intersecting at infinity.
(2) For any ρ ∈ R+ , define the dilatation
dρ (z) = ρz.
This is map is in fact a translation, because it is a composition of two reflections about half circles centered
at the origin.
(3) We can express the reflection about the y-axis by the following map
Rfy (z) = −z̄.

Exercise 4.4 Try to express the above three isometries in the Poincaré disk model D2 .

The next two examples of isometries have simple expression in the disk model.

(1) For any θ ∈ R, define the rotation rθ : D2 → D2 as

rθ (z) = eiθ z.
It can be decomposed into two reflections about diameters.
Lecture Notes in Modern Geometry 29

(2) We can express the reflection about the x-axis by the following map
Rfx (z) = z̄.

Exercise 4.5 Try to express the above two isometries in the upper half plane model H2 . In particular, the reflection
about the x-axis in D2 -model is in fact the inversion about the unit disk in H2 model, i.e., the map z 7→ 1z̄ .

Lemma 4.6 For any two points A, B in H2 , we can find construct a hyperbolic isometry using the above five basic
isometries to map both A, B to the y-axis.

Proof Assume A = (αA , βA ) and B = (αB , βB ). We first use the translation t−αA to map A to A0 = (0, βA ) and
at the same time, B is mapped to the point B0 = (αB − αA , βB ). Then we can dilate and make the A0 to i by
multiplying β1A . The B0 is mapped to some B00 .
Then we move to the D2 -model, in which the i in H2 -model becomes the origin. So, we can rotate the corresponding
image of B00 to some point on the y-axis. Notice that the preimage of the y-axis in the H2 -model is also the y-axis,
we are done with the construction.

From now on, when we try to prove any property which is supposed to be invariant under isometries, we can apply
this lemma to simplify the case we are dealing with.
As an application, let’s show now the result we mentioned before.

Proposition 4.7 For any two points A, B in H2 , the shortest hyperbolic distance is realized by either of the
following two cases:
(1) the euclidean line connected AB if AB is parallel to the y-axis;
(2) the arc of a circle centered on x-axis which joins A, B if AB is not parallel to the y-axis.

Proof We start with the simplest case.

(1) If A, B are both on the y-axis. WLOG, we can assume any path connecting AB is parametrized as (x(y), y)
with y ∈ [a, b], where a, b are the y-coordinates of A0 and B0 . Then we calculate
Z bp 0 2 Z b
x (y) + 1 1
dy ≥ dy,
a y a y
and the equality can be taken if and only if the latter is just the hyperbolic arc length of the line segment AB.
We are done for this case.
(2) If only A is on the y-axis.
(a) If A = i. Then we can take the rotation on the D2 -model around the center 0 to map B to the y-axis.
Then apply (1), we know the shortest distance is realized by lines over y-axis. Then we only need to
understand the image of y under rotation. Notice that the rotation of θ -angle on the H2 -model can be
expressed as
iz + 1 eiθ iz+1
z+i + i
z 7→ eiθ 7→ iθ iz+1 .
z+i ie z+i + 1
Hence when z in on the y-axis, i.e., z = yi, we can do some calculation and show that every point on
the image has equal euclidean distance to the point cot θ . Then this shows that it maps the y-axis to
the half circle centered at cot θ with radius csc θ . We are done for this case too.
30 Rui Wang

(b) If A is not i. We can do dilatation to map A to i. Then we only need to show that half circle centered
at x-axis is still some half circle centered at x-axis after dilatation, which is obvious since the image
of the circle at x0 ∈ R is x0 + reiθ and its image after dilatation ρ is ρx0 + ρreiθ , which is still a half
circle centered at x-axis. We are done with this case too.
(3) If neither A nor B on the y-axis. We can use limit rotation, i.e., real translation to move A to y-axis. Then
this reduces to one of the cases are done. We only need to argue that a half circle centered at x-axis is still
some half circle centered at x-axis after real translation, which is definitely true. Hence all cases are done.

In general, such lines which realize the shortest distance are called geodesic lines for the corresponding metric.
For example, the geodesic lines for the euclidean metric are just the lines is usual sense. The geodesic lines for the
spherical metric over S2 are great circles.
All these three types of geometry share the following important property that a geodesic line can be characterized
using the equidistance property for two points.

Proposition 4.8 A hyperbolic geodesic line can be expressed as the set of hyperbolic equidistance to two points
in H2 .

Proof Similar as previous treatment, we can deal with the simplest case with A, B share equal euclidean distance
to the y-axis and AB parallel to the x-axis in the euclidean sense. For this case, it is not hard to see that the set of
hyperbolic equidistance to A, B is just the y-axis.
For general case, we can use the basic hyperbolic isometries to transfer any A, B to the previous case. The details
are left to you.

As a consequence, we can define the reflection about a hyperbolic geodesic line just as what we did for the spherical
case. (See the definition and discussion around Exercise 3.3. ) Further using the basic isometries, we mimic the
proof for the euclidean case and are able to obtain the three reflections theorem again.

Theorem 4.9 Any hyperbolic isometry can be written as compositions of one or two or three reflections.

As a corollary, every hyperbolic isometry is surjective and we obtain

Corollary 4.10 Iso(H2 ) forms a group.

4.5 Hyperbolic isometries under complex coordinates

From the three reflections theorem for the hyperbolic case, we know that to understand hyperbolic isometries
under complex coordinates, the first step is to express reflections under complex coordinates. For this, we have the
following lemma.

Lemma 4.11 In the H2 -model, the reflection about the hyperbolic line C, i.e, a semicircle centered on x-axis, is
the inversion map. Here, we also include the case that C is a half line parallel to the y-axis and for this case, the
reflection is the same as the euclidean reflection.
Lecture Notes in Modern Geometry 31

Proof For the special case that C is a half line parallel to the y-axis, it is clear from euclidean geometry to see
that a hyperbolic reflection is the same as the euclidean reflection.
We now look at general cases.
(1) If C is the half unit circle centered at the origin. Notice that for this case, the image of C in the D2 -model
is the x-axis, and the reflection in the D2 -model is conjugate map. Hence, the reflection about C in the
H2 -model is the composition of the following maps
z 7→ φ(z) 7→ φ(z) 7→ ψ(φ(z)),
where φ : H2 → D2 is φ(z) = iz+1 w+i
z+i and ψ(w) = iw+1 is its inverse map. Direct calculation shows that it
maps z to 1z̄ , i.e., the inversion about the unit circle C.
1
(2) If C is a radius r half circle centered at the origin. We can use the dilatation by r to map C to the unit circle
C1 . The reflection then is the composition of the following maps
z z z r2
z 7→ 7→ RfC1 ( ) 7→ r · RfC1 ( ) = .
r r r z̄
This is the inversion about the circle C.
(3) For any radius r half circle C centered at some x0 ∈ R. We can use limit rotation t−x0 to map it to a circle
center at the origin. Then this is the case (2). It follows the reflection is the composition of the following
maps
r2 r2 x0 z̄ + (r2 − x02 )
z 7→ z − x0 7→ 7→ + x0 = .
z − x0 z − x0 z̄ − x0
Clearly, this is an inversion about the circle C.

Then by direct calculation, we will further obtain the following statement which was first discovered by Poincaré.

Theorem 4.12 (Poincaré) The H2 -isometries are either of the form

az + b
f (z) = ,
cz + d
where a, b, c, d ∈ R and ad − bc = 1 or of the form
−az̄ + b
f̄ (z) = ,
−cz̄ + d
where a, b, c, d ∈ R and ad − bc = 1.
The former are orientation-preserving isometries and the latter are orientation reversing isometries.

In the D2 -model, the above theorem has the following interpretation.

Theorem 4.13 The D2 -isometries are either of the form

az + b
f (z) = ,
b̄z + ā
32 Rui Wang

where a, b ∈ C and |a|2 − |b|2 = 1 or of the form

az + b
f̄ (z) = ,
b̄z + ā
where a, b ∈ C and |a|2 − |b|2 = 1.
The former are orientation-preserving isometries and the latter are orientation reversing isometries.

We end this section by the following remark.

Remark 4.14 We can construct a map from the orientation preserving hyperbolic isometries Iso+ (H2 ) to the
group of projective special linear group PSL2 (R) as
 
+ 2 az + b a b
Iso (H ) → PSL2 (R) : f (z) = 7→ .
cz + d c d
This is a group isomorphism.
On the other hand, the group PSL2 (R) is also the group Aut(H2 ) of biholomorphic maps from H2 to itself (This
can be proved by the Schwarz lemma). This is an important fact which indicates that the hyperbolic geometry is
closely related to complex analysis. (The euclidean isometry group doesn’t have such connection to Aut(C).)

4.6 Geometric properties of hyperbolic isometries

By the three reflections theorem, all hyperbolic isometries are among the three classes: reflection, two reflections
and three reflections. Moreover, compositions of two reflections are orientation preserving hyperbolic isometries.
Let’s first understand the geometry of such isometries.
When we consider compositions of two reflections in euclidean geometry, whether it is a rotation or a translation
completely determined by the relation of two reflection lines. Here we have similar results, but the relation of two
geodesic lines in hyperbolic geometry is more complicated.

Definition 4.15 Assume C1 and C2 are two hyperbolic geodesic lines. We say they
(1) intersect, if C1 ∩ C2 6= ∅;
(2) ultraparallel, if after extended to the boundary in the D2 -model, C1 and C2 have no intersection;
(3) asymptotic, if C1 and C2 have no intersection in D2 but intersect at boundary.

The three cases divide orientation preserving hyperbolic isometries into three types:
(1) If C1 ∩ C2 6= ∅, we can see from euclidean geometry that the intersection is a unique point in the hyperbolic
plane. Then the composition of two reflections about them is a rotation around this point for double angle
of their intersection angle. In particular, if both C1 and C2 are diameters in the D2 -model or equivalently,
intersect at i in the H2 -model, their composition is just the standard rotation about the origin in D2 .
(2) If C1 and C2 are ultraparallel, then the composition of two reflections about them is a translation of double
hyperbolic distance between them. In particular, if both C1 and C2 are center at 0, their composition is just
the dilatation.
Lecture Notes in Modern Geometry 33

(3) If C1 and C2 are asymptotic, then the composition of two reflections about them is a limit rotation. In
particular, if C1 intersects C2 at i in the D2 -model, this is tc where c is the double of the euclidean distance
between C1 , C2 .
This leads to the following classification of hyperbolic isometries.

Theorem 4.16 (Classification of hyperbolic isometries) Each H2 -isometry is either a

(1) rotation;
(2) limit rotation;
(3) translation; or
(4) glide reflection.

Proof We first deal with orientation preserving case. Any orientation preserving isometry can be written as a
composition of two reflections. Let’s use L, M to denote the two reflection hyperbolic lines. We need to consider
the following three cases:
(1) L ∩ M 6= ∅. From the euclidean geometry, two semicircles centered at x-axis, or a semicircle and a line
parallel to y-axis intersect at only one point. We can show that this isometry must be the rotation with
respect to this intersection point.
(2) L and M are asymptotic, which means that L and M has no intersection in H2 , but if we look at them in the
D2 -model, they intersect at the boundary of the unit disk. For this case, we can show that their composition
is a limit rotation.
(3) L and M has no intersection even with taking infinity into consideration. For this case, we can show that
under good coordinates, it is a dilatation and hence a H2 -translation.
For the orientation reversing case, we can show that it must be a glide reflection.

4.7 The Gauss–Bonnet formula for the hyperbolic case

5 Hyperbolic surfaces

No time to finish, so I only sketch some main results.