Geometry1 ws07

Boris Springborn Geometry I Lecture 1 Winter Semester 07/08
Introduction
It seems natural that a course entitled Geometry should begin with the question:
What is geometry?
Right now, I would like to answer this question in the form of a short historic overview of the
subject. Geometry is, after all, something that people have been doing for a very long time.
The following brief history of geometry will be incomplete, inaccurate (true history is much more
complicated) and biased (we will ignore what happened in India or China, for example). It is a
shortened, smoothed out version of history that is meant only as a rough explanation of how the
material that will be covered in this course came into being.
The word geometry comes from the Greek word γεωμετρία, which is a composite of the words
for earth and measure. Geometry began as the science of measuring the earth, or surveying, and
it began ∼2000 BC in Egypt and Mesopotamia (Babylon, in today’s Iraq). These were among
the first great civilizations and they depended on agriculture along the rivers Nile in Egypt, and
Tigris and Euphrates in Mesopotamia. These rivers would periodically inundate and fertilize the
surrounding land, which made periodic surveying necessary to delimit the fields. The science
of geometry developed from this, with applications also in construction and astronomy. The
Egyptians and Babylonians could compute areas and volumes of simple geometric figures, they
had some approximations for π, and they already knew Pythagoras’ theorem. Strangely though,
no records of general theorems or proofs have survived from this period. Egyptian papyri and
Babylonian clay tables with their cuneiform script contain only worked exercises. Maybe they
did not state general theorems, maybe they just did not write them down, or maybe they did but
these documents did not survive. Basically we have no idea how they conducted their research.
This changed with the period of Greek geometry (Thales ∼600 BC to Euclid ∼300 BC). They
clearly stated general theorems for which they gave proofs. That is, they deduced more compli-
cated statements from simpler ones by logical reasoning. This suggests putting all statements in
order so that each statement is proved using only statements that have previously been proved.
By necessity, one must begin with a few (as few as possible) hopefully very simple statements
that are accepted without proof. In Euclid’s Elements, geometry (most or even all of what was
known at the time) is presented in this form. It begins with a few definitions and postulates
(today we say axioms) from which all theorems are deduced one by one. These postulates were
simple statements like “there is a unique straight line through two points” and “two lines in-
tersect in a unique point or they are parallel”. But one of the postulates was considered more
complicated and less obvious than the others, the parallel postulate: “Given a line and a point
not on the line, there is a unique parallel to the line through the point.” For centuries to come,
people tried to prove this one postulate using the other, simpler ones, so that it could be elim-
inated from the unproved postulates. One way people tried to prove the parallel postulate was
to assume instead that there are many parallels and derive a contradiction. But even though
some strange theorems could be deduced from this alternative parallel postulate (like that there
is an upper bound for the area of triangles) no true contradiction would appear. This finally lead
to the realization that the alternative parallel postulate did not contradict the other postulates.
Instead it leads to a logically equally valid version of geometry which is now called hyperbolic
geometry (Lobachevsky 1829, Bólyai 1831). Later it was realized that one may also assume that
there are no parallels (the other postulates also have to be changed a little for this), and the
resulting geometry is called elliptic geometry. This is simply the geometry on the sphere, where
pairs of opposite points are considered as one point, and lines are great circles. Both hyperbolic
and elliptic geometry are called non-Euclidean geometries, because their axioms are different
from Euclid’s.
Another important development in geometry was the introduction of coordinates by Descartes
and Fermat in the first half of the 17th century. One could then describe geometric figures and
prove theorems using numbers. This way of doing geometry was called analytic geometry, as
opposed to the old way beginning with geometric axioms, which was called synthetic geometry.
By the late 19th/early 20th century, it was proved that both approaches are in fact equivalent:
One can either start with axioms for numbers and use them to define the objects of geometry,
or one can start with axioms of geometry and define numbers geometrically, one gets the same
theorems.
1
The study of the rules of perspective in painting (da Vinci & Dürer ∼1500) lead to the devel-
opment of projective geometry (Poncelet, 1822), dealing with the question: Which properties of
geometric figures do not change under projections? For example, straight lines remain straight
lines, but parallels do not remain parallel.
Another type of geometry is Möbius geometry, which deals with properties that remain unchanged
under transformations mapping circles to circles (such as inversion on a circle). Then there is
also Lie geometry (about which I will say nothing now) and there are other types of geometry.
Klein’s Erlangen Program (1871) provided a systematic treatment of all these different kinds of
geometry and their interrelationships. It also provided a comprehensive and maybe surprising
answer to the question: What is geometry? We will come back to this.
Contents of this course

• spherical geometry
• hyperbolic geometry
• projective geometry
• Möbius geometry
• Lie geometry
• some other stuff
Spherical geometry
n n+1 n+1
p
The n-dimensional unit sphere is S = {x ∈ R | kxk = 1} ⊂ R , where kxk = hx, xi
xi yi is the standard Euclidean scalar product. We will consider mainly S 2 ⊂ R3 .
P
and hx, yi =
A great circle in S 2 is the intersection of S 2 with a plane through the origin, E = {x ∈ Rn+1 |
hx, ni = 0}, knk = 1. (The intersection with a plane not through 0 is called a small circle.) The
points ±n ∈ S 2 are called the poles of the great circle. For x, y ∈ S 2 , x 6= ±y, there is a unique
great circle through x and y. (If x = ±y, there is a one parameter family.) Two great circles
intersect in two diametrically opposite points.
Theorem. The shortest continuously differentiable curve connecting two points x, y ∈ S 2 is the
shorter arc of the great circle through x and y. Its length is d(x, y) := arccoshx, yi.
Proof. The second sentence of the theorem is clear. To prove the first sentence, let γ : [t0 , t1 ] →
S 2 be a continuously differentiable curve from x to y. We have to show length(γ) ≥ d(x, y),
with equality only if γ is an arc of a great circle. We may assume that γ(t) = x only for t = t0
and γ(t) = y only for t = t1 . Let f (p) = arccoshx, pi, which is defined where |hx, pi| ≤ 1
and differentiable where strict inequality holds. For p ∈ S 2 \ {x, −x} let v(p) = grad f (p) −
hgrad f (p), pi p. A direct calculation shows that kvk = 1. Then
Z t1 Z t1 Z t1
d(x, y) = f (y)−f (x) = hgrad f (γ(t)), γ 0 (t)i dt = hv(γ(t), γ 0 (t)i dt ≤ kγ 0 (t)k dt = length(γ),
t0 t0 t0
where we have used hγ 0 , γi = 0 and the Cauchy-Schwarz inequality. Equality only holds if γ 0
is always in the direction of v(γ). This can be used to show that γ must be an arc of a great
circle.
In fact, this proof works also if γ is only assumed piecewise C 1 .
Corollaries. (i) The shortest curves connecting two opposite points are halves of great circles.
(ii) The function d(x, y) is a metric on the sphere.
2
Hemispheres and digons
A hemisphere is the intersection of S 2 with a half-space
H = {x ∈ R3 | hx, ni ≥ 0}, knk = 1.
A hemisphere is bounded by a great circle. One of its poles, n, lies inside the hemisphere, the
other, −n, outside.
A spherical digon is the intersection of two hemispheres.
The interior angle α and the exterior angle α̂ of the digon satisfy α + α̂ = π and cos α̂ = hm, ni,
so cos α = −hm, ni. The exterior angle is the spherical distance between the poles m and n.
The area of the digon is 2α.
(One could alternatively define a spherical digon as a spherical region bounded by two half great
circles. Then the interior angle α could exceed π. The formula for the area would still hold.)
Spherical triangles
There are actually several different sensible definitions for spherical triangles:
1. The intersection of three hemispheres with poles not on one great circle. Such triangles are
called Euler triangles.
2. A spherical region bounded by three great circular arcs. Such triangles are either an Euler
triangle, the outside of an Euler triangle, the union of an Euler triangle and a hemisphere, or
the difference of a hemisphere and an Euler triangle.
3. An oriented closed curve on the sphere consisting of three great circular arcs. (The arcs may
intersect each other.) Such spherical triangles were considered by Möbius. To measure the
angles of such a triangle sensibly, the sphere must be given an orientation.
4. Finally, a spherical triangle in the sense of Study consists of (i) three points A, B, C ∈ S 2 ,
(ii) the three great circles gAB , gBC , gCA through A and B, B and C, C and A, each of which
is given an arbitrary orientation, and (iii) numbers a, b, c, α̂, β̂, γ̂ ∈ R such that: (a) If one
moves from A along gAB the signed distance c, then one reaches B. Similarly for the other
points and great circles. (b) If one rotates gAB around A by the signed angle α̂, it is moved
into gCA with correct orientation. Similarly for the other points and great circles. Triangles
are considered different even if the corresponding numbers differ only by integer multiples
of π.
3
We will now consider Euler triangles, and “spherical triangle” will mean “Euler triangle”, unless
stated otherwise.
In the definition above, an Euler triangle is defined in terms of hemispheres. But an Euler
triangle is also determined by its vertices:
Suppose A, B, C ∈ S 2 do not all lie on one great circle. (That is, A, B, C are linearly independent
unit vectors in R3 ). Then the intersection of S 2 with the cone
C = {λA + µB + νC ∈ R3 | λ, µ, ν ≥ 0}
is an Euler triangle. Indeed, there are unique unit vectors A0 , B 0 , C 0 ∈ S 2 with
hA0 , Ai > 0, hA0 , Bi = 0, hA0 , Ci = 0,

hB 0 , Ai = 0, hB 0 , Bi > 0, hB 0 , Ci = 0,
hC 0 , Ai = 0, hC 0 , Bi = 0, hC 0 , Ci > 0,
and C is the intersection of the half-spaces HA0 , HB 0 , HC 0 , where HX = {x ∈ R3 | hX, xi ≥ 0}.

The side lengths a, b, c and exterior angles α̂, β̂, γ̂ (see figure above) are determined by
cos a = hB, Ci, cos α̂ = hB 0 , C 0 i, etc.
The interior angles are α = π − α̂, etc.

The triangle with vertices A0 , B 0 , C 0 is called the polar triangle of the triangle with vertices
A, B, C.
The polar triangle of the polar triangle is the original triangle.
The side lengths of the polar triangle are the exterior angles of the original triangle and vice
versa.
The side lengths a, b, c of a spherical triangle satisfy the inequalities
−a + b + c > 0, a − b + c > 0, a + b − c > 0, a + b + c < 2π.
The first three inequalities follow directly from the triangle inequality of the spherical metric d.
The third follows from
d(A, B) + d(B, C) + d(C, A) < d(A, B) + d(B, −A) + d(−A, C) + d(C, A) = 2π.
| {z } | {z }
π π
Applying the same reasoning to the polar triangle, we get for the exterior angles of a spherical
triangle
−α̂ + β̂ + γ̂ > 0, α̂ − β̂ + γ̂ > 0, α̂ + β̂ − γ̂ > 0, α̂ + β̂ + γ̂ < 2π,
and hence for the interior angles
−α + β + γ < π, α − β + γ < π, α + β − γ < π, α + β + γ > π.
Theorem. The area of a spherical triangle with interior angles α, β, γ is α + β + γ − π.

Proof. The three great circles through A and B, B and C, C and A divide the sphere into eight
triangles. Two of them are the triangle A, B, C and the triangle −A, −B, −C. They have the
same area because they are symmetric with respect to the origin. The other six triangles all have
a side in common with either triangle A, B, C or triangle −A, −B, −C and complement these
to six digons, two with angle α, two with angle β, two with angle γ. Altogether, the six digons
cover each of the triangles A, B, C and −A, −B, −C three times, and the rest of the sphere once.
So
4(α + β + γ) = sum of area of the six digons
= area(S 2 ) + 2 area(4A, B, C) + 2 area(4 − A, −B, −C)
= 4π + 4 area(4A, B, C).
4
Theorem. Let a, b, c ∈ R. Then

(i) a, b, c satisfy the inequalities
− a + b + c > 0, a − b + c > 0, a + b − c > 0, a + b + c < 2π (∗)
if and only if there is a spherical triangle with side lengths a, b, c.

(ii) If it exists, this triangle is unique up to an orthogonal transformation of R3 .
Proof. (i) The “if” part was shown in the last lecture. To see converse, assume a, b, c satisfy (∗).
Note first that this implies 0 < a, b, c < π. Now ponder the following picture:
(ii) Suppose A, B, C and Ã, B̃, C̃ are the vertices of two triangles with corresponding sides of
equal length. Because corresponding scalar products are equal (hA, Bi = hÃ, B̃i, etc.), the linear
map T : R3 → R3 with T (A) = Ã, T (B) = B̃, T (B) = B̃ is orthogonal.
Remarks. (i) There is ofn
course
an analogous theorem regarding the angles of a spherical triangle.
a
o
(ii) The region D := ∈ R3 a, b, c satisfy (∗) is the interior of the tetrahedron with
b

c
0 π π 0 αb
vertices 0 , π , 0 , π . The region of possible vectors of exterior angles βb is also D.
0 0 π π γ
b
By the previous theorem (and its analogue for angles), the side lengths determine the angles and
vice versa. This gives a bijection D → D. If the side lengths approach a face (vertex) of D, then
the corresponding angles approach a vertex (face) of D.
Theorem. Side lengths and exterior angles of a spherical triangle satisfy the equations
− cos a + cos b cos c
cos α̂ = , (side cosine theorem)
sin b sin c
− cos α̂ + cos β̂ cos γ̂
cos a = , (angle cosine theorem)
sin β̂ sin γ̂
and four more equations obtained by simultaneous permutations of a, b, c and α̂, β̂, γ̂.
The following proof using Gram matrices exemplifies a general method which will be useful
again later. The Gram matrix for a (finite) sequence of vectors v1 , . . . vk is the symmetric matrix
(hvi , vj i)ki,j=1 of pairwise scalar products.
Proof. Let V = (A B C) ∈ R3×3 be the matrix whose columns are the vertices of the spherical
triangle, considered as column vectors. Then the Gram matrix for A, B, C is
   
hA, Ai hA, Bi hA, Ci 1 cos c cos b
G = V t V = hB, Ai hB, Bi hB, Ci = cos c 1 cos a .
hC, Ai hC, Bi hC, Ci cos b cos a 1
5
(Note for later that det G = (det V )2 > 0.) Similarly, let W = (A0 B 0 C 0 ) be the matrix of poles.
Their Gram matrix is
 0 0
hA0 , B 0 i hA0 , C 0 i
  
hA , A i 1 cos γ̂ cos β̂
G0 = W t W = hB 0 , A0 i hB 0 , B 0 i hB 0 , C 0 i = cos γ̂ 1 cos α̂ .
hC 0 , A0 i hC 0 , B 0 i hC 0 , C 0 i cos β̂ cos α̂ 1
Also,
hA0 , Ai hA0 , Bi hA0 , Ci
   0 
hA , Ai 0 0
t
W V = hB 0 , Ai hB 0 , Bi hB 0 , Ci =  0 hB 0 , Bi 0  =: D
hC 0 , Ai hC 0 , Bi 0
hC , Ci 0 0 hC 0 , Ci
is a diagonal matrix with positive entries. So W t = DV −1 and W = (V t )−1 D, and
G0 = DV −1 (V t )−1 D = D(V t V )−1 D = DG−1 D. (∗∗)
The inverse of G is
sin2 a
 
− cos c + cos a cos b − cos b + cos c cos a
1 
G−1 = − cos c + cos a cos b sin2 b − cos a + cos b cos c .
det G
− cos b + cos c cos a − cos a + cos b cos c sin2 c
2 1 2
Substitute this into (∗∗) and consider diagonal elements: One finds 1 = D11 det G sin a, therefore
√ √ √
det G det G det G
D11 = sin a , and similarly D22 = sin b , D33 = sin c . Now consider for example element
1
(3, 2) in (∗∗): cos α̂ = D33 det G (− cos a + cos b cos c)D22 . This is the side cosine theorem.
The angle cosine theorem is the side cosine theorem applied to the polar triangle.
Theorem. Side lengths and interior angles of a spherical triangle satisfy
sin a sin b sin c
= = , (sine theorem)
sin α sin β sin γ
s
α sin( a−b+c
2 ) sin( a+b−c
2 )
tan = , (half-angle theorem)
2 sin( −a+b+c
2 ) sin( a+b+c
2 )
v
cos( −α+β+γ ) cos( α+β+γ
u
a u 2 2 )
tan = t− α−β+γ α+β−γ
. (half-side theorem)
2 cos( 2 ) cos( 2 )
Proof. In terms of the interior angle α, the side cosine theorem says cos α = cos a−cos b cos c
sin b sin c .
2 α 2 α
Using cos α = 2 cos 2 − 1 = 1 − 2 sin 2 and other trigonometric identities, one obtains the
equations
s
α sin( −a+b+c
2 ) sin( a+b+c
2 )
cos = ,
2 sin b sin c
s
α sin( a−b+c
2 ) sin( a+b−c
2 )
sin = ,
2 sin b sin c
which are also of independent interest. Dividing one by the other, one obtains the half-angle
theorem. The half-side theorem is the half-angle theorem for the polar triangle. To prove the
sine theorem, consider
α α P
sin α = 2 sin cos = ,
2 2 sin b sin c
where r
−a + b + c a − b + c a + b − c a + b + c
P = 2 sin sin sin sin .
2 2 2 2
So sin α sin b sin c = P . But since the expression for P is symmetric in a, b, c, one has equally
sin α sin b sin c = sin β sin c sin a = sin γ sin a sin b = P.
Divide by sin a sin b sin c to obtain the sine theorem.
6
In principle, all relations between side lengths and angles of a spherical triangle can be derived
from the inequalities for the sides and the side cosine theorem (or alternatively from the inequali-
ties for the angles and the angle cosine theorem) using only algebra and analysis, without further
recourse to geometry. We have derived the half-angle theorem and the sine theorem in this way.
On the other hand, the following proof of Napier’s rule, which is due to Napier himself, uses a
remarkable geometric construction.
Theorem (Napier’s rule). Consider a right-angled spherical triangle
with γ = π2 , and let ā = π2 − a, b̄ = π2 − b. Then
cos c = sin b̄ sin ā = cot α cot β,

cos β = sin α sin b̄ = cot c cot ā,
cos ā = sin c sin α = cot β cot b̄,
cos b̄ = sin β sin c = cot ā cot α,
cos α = sin ā sin β = cot b̄ cot c.
That is: “The cosine of any part is equal to the product of sines of
opposite parts and to the product of cotangents of adjacent parts.” (The
“parts” are ā, b̄, α, c, β, in this cyclic order.)
Proof. The first line of equations follows directly from the cosine rules cos γ = cos sin c−cos a cos b
a sin b
and cos c = cos γ+cos α cos β
sin α sin β . To prove the remaining equations, assume first that a, b, c, α, β < π2 .
Consider the following construction. Draw the two great circles that have A and B as poles.
Together with the extended sides of the original triangle, they form 4 other right angled triangles:
(The five triangles form a right-angled pentagram called the pentagramma mirificum. In its
center there is a spherical pentagon each vertex of which is the pole of the opposite side.) Note
that a1 = π2 − c, b1 = π2 − β, α1 = π2 − a, c1 = π2 − b, β1 = α, so that
ā1 = c, b̄1 = β, α1 = ā, c1 = b̄, β1 = α.
This proves the other equations of Napier’s rule under the assumption of acute angles and sides.
Now suppose that a side length a, b, c or an angle α, β is greater than π2 . (The remaining cases
where one is equal to π2 consist of doubly or triply right-angled triangles for which Napier’s rule
can easily be checked.) It can be shown that, first, for one of the neighbor triangles (into which
the sphere is separated by the same great circles) a, b, c, α, β < π2 , and, second, if Napier’s rule
holds for one of the neighbor triangles, it holds for all of them.
7
Stereographic projection
0
Project the unit sphere S 2 ⊂ R3 from the “north pole” e3 = 0 to the plane x3 = 0. This map
1
2 2
σ : S \ {e3 } → R is called stereographic projection. One easily derives the following equations
for σ and its inverse:
 
x1
1 x1
σ x2  = ,
1 − x3 x2
x3
 
2u1
u1 1
σ −1 = 2  2u2 .
u2 u1 + u22 + 1
u21 + u22 − 1
Theorem. Stereographic projection σ maps circles in S 2 which contain e3 to lines in R2 and all
other circles in S 2 to circles in R2 . All circles and lines in R2 are images of circles in S 2 .
Remark. The fact that circles through e3 are mapped to lines is geometrically clear: A circle
through e3 is the intersection of S 2 with a plane through e3 . So all projection rays lie in this
plane, and the circle is mapped to the line in which it intersects the image plane x3 = 0.
Proof. A circle in S 2 is the intersection of S 2 with a plane
E = {x ∈ R3 | hx, ni = d}, where knk = 1, 0 ≤ d < 1.
It contains e3 iff d = he3 , ni = n3 . A point u ∈ R2 in the image of the circle iff σ −1 (u) ∈ E, that
is, iff
1
2u1 n1 + 2u2 n2 + (u21 + u22 − 1)n3 ,

d= 2
u1 + u22 + 1
or equivalently,
0 = (n3 − d)(u21 + u22 ) + 2u1 n1 + 2u2 n2 − (n3 + d).
If n3 = d, this is the equation for a line. Otherwise, divide by n3 − d, complete the squares, and
use knk2 = 1 to obtain
n 1 2 n 2 2 1 − d2
u1 + + u2 + − = 0.
n3 − d n3 − d (n3 − d)2
√
n1 1−d2
This is the equation for a circle with center c = − n31−d

n2 and radius r = |n3 −d| .
To show (without further calculations) that every circle and every line in R2 is the image of a
circle in S 2 , you can make an argument using the fact that three points in R2 uniquely determine
a line or circle through them and three points in S 2 uniquely determine a circle through them.
8
Theorem. Stereographic projection is conformal, that is, it preserves angles: If two curves in
S 2 \ {e3 } intersect at some angle, then their image curves in R2 intersect at the same angle.
Proof. Let γ̂, η̂ : (−ε, ε) → S 2 \ {e3 } be two curves with γ̂(0) = η̂(0) = p̂, and let γ = σ ◦ γ̂,
η = σ ◦ η̂ be their images curves in R2 . Let v̂ = γ̂ 0 (0), ŵ = η̂ 0 (0), v = γ 0 (0), w = η 0 (0), and let
p = γ(0) = η(0) = σ(p̂). The intersection angles α̂ and α between γ̂, η̂ and γ, η, are determined
by
hv̂, ŵi hv, wi
cos α̂ = p , cos α = p .
hv̂, v̂ihŵ, ŵi hv, vihw, wi
We want to show that α̂ = α. In fact, we will show that
4 4 4
hv̂, ŵi = (hp,pi+1)2 hv, wi, hv̂, v̂i = (hp,pi+1)2 hv, vi, hŵ, ŵi = (hp,pi+1)2 hw, wi,
and this implies cos α̂ = cos α, and hence α̂ = α.

First we derive equations for v̂, ŵ in terms of v, w. By the equation for σ −1 we have

2γ1
1
γ̂ = hγ,γi+1 2γ2
hγ,γi−1,
so
2γ10

2γ1
2hγ,γ 0 i
γ̂ 0 = 1
hγ,γi+1
2γ20 − (hγ,γi+1)2
2γ2
2hγ,γ 0 i hγ,γi−1,
γ10

2γ1
2 hγ,γ 0 i
= hγ,γi+1
γ20 − hγ,γi+1
2γ2 ,
hγ,γ 0 i hγ,γi−1
and hence
v1 2p1
0 2 v2 hp,vi
v̂ = γ̂ (0) = hp,pi+1 − hp,pi+1
2p2 .
hp,vi hp,pi−1
In the same way one gets

w1 2p1
2 w2 hp,wi
ŵ = hp,pi+1 − hp,pi+1
2p2 ,
hp,wi hp,pi−1
so

4 hp,wi
hv̂, ŵi = (hp,pi+1) 2 hv, wi + hp, vihp, wi − hp,pi+1 2hp, vi + hp, vi hp, pi − 1
| {z }
hp,vi(hp,pi+1)

hp,vi hp,vihp,wi
2
− hp,pi+1 2hp, wi + hp, wi hp, pi − 1 + (hp,pi+1)2 4hp, pi + hp, pi − 1
| {z } | {z }
hp,wi(hp,pi+1) (hp,pi+1)2
4
= (hp,pi+1) 2 hv, wi,
and similarly for hv̂, v̂i and hŵ, ŵi.

All of this works in the same way also for the n-dimensional sphere S n ∈ Rn+1 . (But for n = 1,
the projection from the circle to the line, the two theorems are trivial.) Stereographic projection
from en+1 is σ : S n \ {en+1 } → Rn ,
x1
!
1 x2
.. ,
σ(x) =
1 − xn+1 x.n
2u1 !
1 ..
−1 .
σ (u) = 2u
hu, ui + 1 hu,ui−1 n
It maps (n − 1)-dimensional spheres in S n (intersections of S n with affine hyperplanes in Rn+1 )

to (n − 1)-dimensional spheres and planes in Rn , and it is conformal.
9
Bilinear and quadratic forms

We will define n-dimensional hyperbolic space as
H n = x ∈ Rn+1 x21 + x22 + . . . + x2n − x2n+1 = −1, xn+1 > 0

= x ∈ Rn+1 hx, xi = −1, xn+1 > 0 ,

where now hx, yi = x1 y1 +. . .+xn yn −xn+1 yn+1 , and lengths of curves in H n and angles between
them are measured using this scalar product instead of the normal Euclidean scalar product. We
will need to be familiar with such indefinite scalar products, and since we will deal with general
bilinear and quadratic forms later, it may be a good idea to refresh some material from linear
algebra.
Let V be an n-dimensional vector space over a field K. (We will be interested in the cases
K = R and K = C.) A bilinear form on V is a function b : V × V → K, which is linear in
each argument. If e1 , . . . , en is a basis of V , then the matrix of the bilinear form b is B ∈PK n×n
with Bij =Pb(ei , ej ). If x, y ∈ K n are the coordinate vectors for v, w ∈ V , that is v = i xi ei
t
P
and w = i y i e i , then b(v, w) = x By = ij Bij x i yj . If f1 , . . . , fn is another basis with
fj = i Tij ei , and p, q ∈ K n are the coordinate vectors of v, w in this new basis, then x = T p,
P
y = T q, and the matrix of b with respect to the new basis is B̃ = T t BT .

The bilinear form B is symmetric if b(v, w) = b(w, v) for all v, w ∈ V . A bilinear form is
symmetric if its matrix with respect to one (and hence every) basis is symmetric.
A quadratic form on V is a function q : V → K for which there exists a bilinear form b such that
q(v) = b(v, v) for all v ∈ V. (◦)
If q is a quadratic form, then there exists a unique symmetric bilinear form satisfying (◦). Hence
symmetric bilinear forms and quadratic forms are in one-to-one correspondence. Quadratic forms
are homogeneous polynomials of degree 2 in (any) coordinates.
Example. x21 + x1 x2 + x22 is a quadratic form on K 2 . For example, it comes from the bilinear
form x1 y1 + x1 y2 + x2 y2 , which is not symmetric. But it also comes from the symmetric bilinear
form x1 y1 + 21 x1 y2 + 12 x2 y1 + x2 y2 .
Symmetric bilinear forms
Suppose b is symmetric and q(v) = b(v, v). The kernel of b is
ker b = {v ∈ V | b(v, w) = 0 for all w ∈ V }.
This is a linear subspace of V . The bilinear and quadratic forms b, q are called degenerate
if ker b 6= {0}, and non-degenerate if ker b = {0}. The form b is degenerate iff its matrix
with respect to one (hence every) basis has determinant 0. Let U0 = ker b and let U be any
complementary subspace, so that V = U ⊕ U0 . Then the restrictions b|U and q|U are non-
degenerate bilinear/quadratic forms on U .
There exist bases ê1 , . . . , ên of V such that
b(êi , êj ) = 0 if i 6= j. ()
The basis vectors ei with b(êi , êi ) = 0 form a basis for U0 = ker b. Assume the basis is ordered so
that these come last. Then the matrix of b with respect to this basis is diagonal with diagonal
elements λ1 , . . . , λr , 0, . . . , 0, where λi 6= 0 and r = n − dim ker b. In the coordinates u1 , . . . , un
with respect to this basis, q is a sum of squares:
λ1 u21 + λ2 u22 + . . . + λr u2r .
A basis ê1 , . . . , ên satisfying () and the corresponding coordinates u1 , . . . , un can be found using
the generalized Gram-Schmidt orthogonalization procedure or by completing the squares.
10
Generalized Gram-Schmidt orthogonalization procedure
The ordinary Gram-Schmidt orthogonalization procedure takes as input a basis e1 , . . . , en of V

and a positive definite symmetric bilinear form b. It works like this:
for i = 1 to n − 1 do
for j = i + 1 to n do
b(e ,e )
ej ← ej − b(eii ,eji ) ei
end for
end for
This may not work if b is not positive definite because, for some i, b(ei , ei ) may be 0. If that is the
case, do the following: If there is among the ei+1 , . . . , en a basis vector, say ek , with b(ek , ek ) 6= 0,
then swap ei and ek and continue. If that that is not possible because b(ei+1 , ei+1 ), . . . , b(en , en )
are all 0, then there must be some ek (i + 1 ≤ k ≤ n) with b(ei , ek ) 6= 0, because b was assumed
to be non-degenerate. Assign ei + ek and ei − ek to ei and ek and continue.
If b is degenerate, find a basis for ker b and a basis for a complementary subspace and apply the
orthogonalization procedure to the latter.
Warning: This algorithm is not numerically stable. Its main purpose is to prove the existence of
a diagonalizing basis.
Completing the squares
Examples. Suppose in some coordinates q is

x1 2 + x2 2 + 2x3 2 + x1 x2 + x1 x3 + x2 x3
=(x1 + 12 x2 + 12 x3 )2 + 43 x2 2 + 47 x3 2 + 12 x2 x3
=(x1 + 12 x2 + 12 x3 )2 + 34 (x2 + 31 x3 )2 + 20
12 x3 2
| {z } | {z } |{z}
u1 u2 u3
= u1 2 + 3
4 u2 2 + 20
12 u3 2 .
It may happen that there is no square to complete:
x1 x2 + x2 x3 = 14 (x1 + x2 )2 − 14 (x1 − x2 )2 + x2 x3 = 41 y1 2 + 41 y2 2 + 12 (y1 − y2 )x3 ,

| {z } | {z }
y1 y2
continue as before.
Case K = C
p
We can always make all λ1 , . . . , λr equal 1 by dividing ek (k = 1, . . . , r) by b(ek , ek ) (arbitrary
choice of square roots).
Pr Thus, any symmetric bilinear form on a complex vector space is in
suitable coordinates k=1 xk yk .
Case K = R
Let
i+ = (number of êk for which b(êk , êk ) > 0),
i− = (number of êk for which b(êk , êk ) < 0),
i0 = (number of êk for which b(êk , êk ) = 0) = dim ker b.
The numbers i+ , i− , i0 do not depend on the particular basis, but only on b. (Why?) The
numbers i+ and i− are the positive and negative index of b. The signature of b is (i+p , i− , i0 ), also
written (i+ , i− ) if i0 = 0. We can normalize the êk (k = 1, . . . , r) by dividing by |b(êk , êk )|.
Thus, one obtains a basis with b(êi , êi ) = ±1 or 0, and any symmetric bilinear form on a real
vector space is in suitable coordinates
i+ i+ +i−
X X
xk yk − xk yk .
k=1 k=i+
11
Scalar products
A scalar product is a non-degenerate symmetric bilinear form. Scalar products are often written
h·, ·i or (·, ·). Vectors v, w with hv, wi = 0 are called orthogonal to each other.
Any n-dimensional complex vector space with a scalar
Pnproduct can by an appropriate choice of
basis be identified with Cn with the scalar product 1 xk yk .
Rp,q denotes the real vector space Rp+q , equipped with the scalar product
p
X p+q
X
hx, yi = xk yk − xk yk ,
k=1 k=p+1
which has signature (p, q). By an appropriate choice of basis, any real vector space with a scalar
product with signature (p, q) can be identified with Rp+q . A vector v in Rp,q (or any other real
vector space with scalar product) is called spacelike, timelike, or lightlike if hv, vi > 0, < 0, or = 0.
The set of lightlike vectors is called the light cone. A Euclidean scalar product is a scalar product
with negative index i− = 0. A Lorentz scalar product is a scalar product with negative index
i− = 1. A vector space with a Euclidean or Lorentz scalar product is a Euclidean or Lorentz
vector space, respectively. An orthonormal basis for Rp,q is a basis e1 , . . . , en with hei , ej i = 0 if
i 6= j, hei , ei i = 1 for i = 1, . . . , p and hei , ei i = −1 for i = p + 1, . . . , n.
Theorem. Let V be a real vector space with scalar product h·, ·i, let e1 , . . . , en be a basis, let
Bij = hei , ej i, and let
B11 ··· B1k
!

B11 B12
a0 = 1, a1 = B11 , a2 = det B21 B22 , . . . ak = det .. .. , . . . an = det B.
. .
Bk1 ··· Bkk
Suppose that none of the ak are 0. Then the negative index of the scalar product h·, ·i is equal to
the number of sign changes in the sequence a0 , a1 , . . . , an .
Orthogonal transformations
Let V be a vector space with scalar product h·, ·i. An orthogonal transformation on V is a linear
map T : V → V with hT v, T wi = hv, wi for all v, w ∈ V . The orthogonal transformations form
a group, the orthogonal group of V , h·, ·i. The orthogonal group of Cn with standard scalar
product is denoted by O(n, C). The orthogonal group of Rp,q is denoted by O(p, q), or if q = 0
also by O(p) or O(p, R). The corresponding matrix groups are also denoted by O(n, C), O(p, q).
The columns of a matrix in O(p, q) form an orthonormal basis of Rp,q . The determinant of an
orthogonal transformation is ±1. The orthogonal transformations with determinant +1 form
subgroups, the special orthogonal groups, denoted by SO(n, C) and SO(p, q).
Lorentz vector spaces

R1,1 is R2 with hx, yi = x1 y1 − x2 y2 . So hx, xi = x21 − x22
> >
and this is = 0 if |x1 | = |x2 |. The figure also shows another
< <
orthonormal basis.
R2,1 is R3 with hx, yi = x1 y1 + x2 y2 − x3 y3 . So hx, xi =
> p >
x21 + x22 − x23 and this is = 0 if x21 + x22 = |x3 |.
< <
In a Lorentz vector space, any non-zero vector which is

orthogonal to a timelike vector is spacelike.
An orthogonal transformation T ∈ O(n, 1) either maps
each connected component of the hyperboloid hx, xi = −1
onto itself, or it interchanges the two components. Those
that map each component onto itself form a subgroup,
O+ (n, 1). Also, SO+ (n, 1) = O+ (n, 1) ∩ SO(n, 1). In
all, O(n, 1) has 4 connected components, each consisting
of the transformations with determinant either +1 or −1
and either fixing or interchanging the components of the
hyperboloid hx, xi = −1.
12
Hyperbolic geometry
Hyperbolic space of n dimensions is
H n = {x ∈ Rn,1 | hx, xi = −1, xn+1 > 0},
where hx, yi = x1 y1 + . . . + xn yn − xn+1 yn+1 is the Lorentz scalar product of Rn,1 . Lengths of
curves and angles between them are measured using this scalar product (and not the Euclidean
scalar product of Rn+1 ): The length of a curve γ : [t1 , t2 ] → H n is
Z t2 p
length(γ) = hγ 0 (t), γ 0 (t)i dt
t1
and the angle α between two curves γ, η : (−ε, ε) → H n intersecting in p = γ(0) = η(0) with
non-zero velocities v = γ 0 (0), w = η 0 (0) is determined by
hv, wi
cos α = p .
hv, vihw.wi

hv,wi
Why is this well defined? Why is hγ 0 , γ 0 i never negative and why is √ ≤ 1? For

hv,vihw.wi
v ∈ Rn,1 let v ⊥ = {w ∈ Rn,1 | hv, wi = 0}. If hv, vi < 0 (v is timelike) then the restriction of the
scalar product h·, ·i to v ⊥ is a Euclidean scalar product. Now hγ, γi = −1 implies hγ, γ 0 i = 0,
so γ 0 (t) ∈ γ(t)⊥ . In the formulas for lengths and angles, the Lorentz scalar product is therefore
applied to vectors in a subspace of Rn,1 on which it is Euclidean.
One-dimensional hyperbolic space
H 1 is one branch of the hyperbola x1 2 − x2 2 = −1 in R1,1 .
sinh s . (Because cosh2 s − sinh2 s = 1.)

This is a curve which can be parameterized as γ(s) = cosh s
2 2
Now γ 0 (s) = cosh 0 0
s

sinh s , so hγ (s), γ (s)i = cosh s − sinh s = 1 and
Z s2 p
length(γ|[s1 ,s2 ] ) = hγ 0 (s), γ 0 (s)i = s2 − s1 .
s1
On the other hand,
hγ(s1 ), γ(s2 )i = sinh s1 sinh s2 − cosh s1 cosh s2 = − cosh(s1 − s2 ).
(Because cosh(x + y) = cosh x cosh y + sinh x sinh y.) Hence, the hyperbolic distance d(p1 , p2 ) of
two points p1 , p2 ∈ H 1 is given by
− cosh d(p1 , p2 ) = hp1 , p2 i.
Hyperbolic lines
A hyperbolic line in n-dimensional hyperbolic space H n is a non-empty intersection of H n with

a 2-dimensional linear subspace U of Rn,1 .
13
Proposition. If U is a 2-dimensional linear subspace of Rn,1 with U ∩ H n 6= ∅, then the restric-

tion h·, ·i|U has signature (1, 1).
Proof. By assumption, there is a u ∈ U with hu, ui = −1. Extend u to a basis u, v of U , and let
ṽ = v + hu, viu. Then u, ṽ is a basis of U with hu, ṽi = 0. Because any non-zero vector which is
orthogonal to a timelike vector is spacelike, hṽ, ṽi > 0. So h·, ·i|U has signature (1, 1).
By an appropriate choice of basis, any 2-dimensional U intersecting H n can therefore be identified
with R1,1 . The hyperbolic line U ∩ H n is thus identified with 1-dimensional hyperbolic space
H 1 ∈ R1,1 .
Remark. In the same way, any non-empty intersection of H n with a (k + 1)-dimensional subspace
can be identified with H k .
For two points p1 , p2 ∈ H n , there is a unique hyperbolic line containing them: span(p1 , p2 ) ∩ H n .
Theorem. The shortest piecewise continuously differentiable curve connecting two points p1 , p2 ∈
H n is the hyperbolic line segment between them. Its hyperbolic length is
d(p1 , p2 ) = arcosh(−hp1 , p2 i).
This can be proved in the same way as we proved the corresponding theorem for the sphere.
Two-dimensional hyperbolic space
The hyperbolic plane H 2 is one component of the hyper-

boloid of two sheets x1 2 + x2 2 − x3 2 = −1 in R2,1 . Any
2-dimensional subspace U of R2,1 which intersects H 2 is
U = {x ∈ R2,1 | hx, ni = 0}
for some n ∈ R2,1 with hn, ni = 1. The vector −n would

give the same subspace, but up to sign, the unit normal
n of U is unique. Thus, the spacelike unit vectors in R2,1
are in 2-to-1 correspondence with the hyperbolic lines in
H 2 . They are in 1-to-1 correspondence with the hyperbolic
half-planes
{x ∈ R2,1 | hx, ni ≥ 0} ∩ H 2 .
Proposition. Let n1 , n2 ∈ R2,1 with hn1 , n1 i = hn2 , n2 i = 1, and let l1 , l2 be the corresponding
hyperbolic lines, li = {x ∈ H 2 | hx, ni i = 0}. Assume n1 6= ±n2 , so that l1 and l2 are different
lines. Then the following statements are equivalent.
(i) The lines l1 and l2 intersect.
(ii) The restriction of h·, ·i to span(n1 , n2 ) has signature (2, 0).
(iii) |hn1 , n2 i| < 1.
Proof. (i)⇒(ii): If l1 ∩ l2 6= ∅ then there is an x ∈ H 2 with hx, n1 i = hx, n2 i = 0. So x⊥ =
span(n1 , n2 ), and (ii) follows because any non-zero vector orthogonal to a timelike vector is
spacelike.
(ii)⇒(i): If the restriction of the scalar product to span(n1 , n2 ) has signature (2, 0), then its
restriction to the orthogonal complement span(n1 , n2 )⊥ must have signature (0, 1), so the com-
plement intersects H2 .
(ii)⇔(iii): The vectors n1 , n2 form a basis of span(n1 , n2 ). In this basis, the matrix of the
restriction of h·, ·i is

hn1 , n1 i hn1 , n2 i 1 hn1 , n2 i
B= = .
hn2 , n1 i hn2 , n2 i hn2 , n1 i 1
So B11 = 1 and det B = 1 − hn1 , n2 i2 . The equivalence of (ii) and (iii) follows from last lecture’s
signature theorem (and the fact that the matrix of a non-degenerate bilinear form has non-zero
determinant).
14
Now suppose the hyperbolic lines l1 , l2 intersect in x ∈ H 2 , and let hi = {y ∈ H 2 | hy, ni i ≥ 0}

be half-planes bounded by l1 , l2 . Since the Lorentz scalar product of R2,1 is Euclidean on the
subspace x⊥ = span(n1 , n2 ), we can measure angles between vectors in x⊥ in the usual way.
The exterior angle α̂ of the half-planes h1 , h2 at x is determined by
cos α̂ = hn1 , n2 i.
The interior angle α is π − α̂, so

− cos α = hn1 , n2 i.
In particular, l1 and l2 intersect orthogonally if hn1 , n2 i = 0.
Remark. A hyperbolic rotation (of H 2 ) with center x ∈ H 2 is a map T ∈ O(2, 1) with T (x) = x
and which is a Euclidean rotation on x⊥ . The exterior angle between the two half-spaces h1 , h2
is the angle of the hyperbolic rotation mapping one to the other.
Proposition. (i) If x ∈ H 2 and l is a hyperbolic line, then there is a unique hyperbolic line
through x which intersects l orthogonally.
(ii) If l1 , l2 are two hyperbolic lines with unit normals n1 , n2 such that |hn1 , n2 i| > 1 (so the lines
do not intersect), then there is a unique line l3 intersecting both l1 and l2 orthogonally.
Proof. Exercise.
Hyperbolic triangles
Let A, B, C ∈ H 2 be three points in the hyperbolic plane. Assume that they do not all lie on one
hyperbolic line (this is equivalent to assuming A, B, C to be linearly independent). The hyperbolic
triangle with vertices A, B, C is the intersection of H 2 ⊂ R2,1 with the set of non-negative linear
combinations
{λA + µB + νC | λ, µ, ν ∈ R≥0 }.
The side lengths a = d(B, C), b = d(C, A), c = d(A, B) satisfy
− cosh a = hB, Ci, − cosh b = hC, Ai, − cosh c = hA, Bi.
Let A0 , B 0 , C 0 be the spacelike unit vectors such that the half-plane bounded by the line through
B, C and containing A is
hA0 = {x ∈ H 2 | hA0 , xi ≥ 0},
and analogously for B 0 and C 0 . Then the hyperbolic triangle with vertices A, B, C is also the
intersection hA0 ∩ hB 0 ∩ hC 0 . The interior angles α, β, γ at A, B, C satisfy
− cos α = hB 0 , C 0 i, − cos β = hC 0 , A0 i, − cos γ = hA0 , B 0 i.
Theorem. The side lengths and interior angles of a hyperbolic triangle satisfy
− cosh a + cosh b cosh c
cos α = , (hyperbolic side cosine theorem)
sinh b sinh c
cos α + cos β cos γ
cosh a = . (hyperbolic angle cosine theorem)
sin β sin γ
Proof (sketch). This can be proved in the same way as we proved the spherical cosine theorems
using the Gram matrices G, G0 of A, B, C and A0 , B 0 , C 0 . Only now the scalar product is
1 0 0
hx, yi = xt Ey with E= 01 0 .
0 0 −1
So if V = (A B C) and W = (A0 B 0 C 0 ), then

−1 − cosh c − cosh b 1 − cos γ − cos β
G = V t EV = − cosh c −1 − cosh a , G0 = W t EW = − cos γ 1 − cos α ,
− cosh b − cosh a −1 − cos β − cos α 1
and D = W t EV is a diagonal matrix with positive elements on the diagonal. Continue as in the
spherical case . . .
15
Remark. The spherical cosine theorems for a sphere of radius R (instead of 1) are
a
cos R − cos Rb cos Rc a cos α + cos β cos γ
cos α = b c
, cos = .
sin R sin R
R sin β sin γ
One gets the hyperbolic cosine theorems by setting R = i. That’s why it is sometimes said that
hyperbolic geometry is the geometry on a sphere with imaginary radius.
From the hyperbolic cosine theorems, one can derive
sinh a sinh b sinh c
= = (hyperbolic sine theorem)
sin α sin β sin γ
in the same way in which we derived the spherical sine theorem from the spherical cosine theo-
rems. One can also derive hyperbolic versions of the half-angle and half-side theorems, and other
formulas of spherical trigonometry.
Theorem. (i) A hyperbolic triangle with side lengths a, b, c ∈ R>0 exists if and only if the triangle
inequalities are satisfied. (ii) A hyperbolic triangle with angles α, β, γ ∈ (0, π) exists if and only
if
α + β + γ < π.
Remark. We will see later that the area of a hyperbolic triangle is π − (α + β + γ).
Proof (sketch). 1. Show that a symmetric (3 × 3)-matrix G is the Gram matrix of 3 linearly
independent vectors in R2,1 if and only if the bilinear form xt Gy has signature (2, 1).
−1 − cosh c − cosh b
2. Consider G = − cosh c −1 − cosh a . Because
− cosh b − cosh a −1
G11 G12
= − sinh2 c < 0,

G11 = −1 < 0, det G21 G22
the signature of G is (2, 1) if and only if det G < 0 (by the theorem from Lecture 6).
3. Show the remarkable identity
−a+b+c a−b+c a+b−c a+b+c

det G = −4 sinh 2 sinh 2 sinh 2 sinh 2 ,
and use it to prove part (i) of the theorem.

1 − cos γ − cos β
Part (ii) can be shown in the same way by considering G0 = − cos γ 1 − cos α and using the
− cos β − cos α 1
identity
det G0 = −4 cos −a+b+c a−b+c a+b−c a+b+c

2 cos 2 cos 2 cos 2 .
16
The Klein model of the hyperbolic plane

Project H 2 ⊂ R2,1 to the plane x3 = 1 with 0 ∈ R2,1
as center of projection. This is central projection,
because 0 is the center of the hyperboloid of which
x1
H 2 is one sheet. A point xx2 ∈ H 2 is mapped to
3
x
1 x12 2
x3 x3 , and all of H is mapped to the inside of the
unit circle in the image plane x3 = 1. Hyperbolic
lines are mapped to secants of this unit circle. If we
forget about the x3 -coordinate, we get an image of
the hyperbolic plane which is called the Klein model.
Thus, the Klein model of the hyperbolic plane is the
unit disk From W. P. Thurston, Three-dimensional geometry and Topology.
D2 = {u ∈ R2 | u21 + u22 < 1},

x1
corresponds to x13 xx12 ∈ D2 , and, inversely, u1

where x2
x3
∈ H2 u2 ∈ D2 corresponds to
u1
√ 1 u2 ∈ H 2 . Angles and lengths in are measured in H 2 .
1−u21 −u22 1
For two hyperbolic lines l1 , l2 with unit normals n1 , n2 ∈ R2,1 , there are three possibilities:
(1) |hn1 , n2 i| < 1. (2) |hn1 , n2 i| = 1. (3) |hn1 , n2 i| > 1.
The lines intersect. The lines do not intersect and The lines do not intersect and
their images in the Klein model their images in the Klein model
intersect on the unit circle. intersect outside the unit circle.
In cases 2 and 3 the lines do not intersect, thus they are parallel. However, to distinguish the
two cases, parallel is sometimes used to mean only lines in case 2, and lines in case 3 are then
called ultra-parallel.
Angle of parallelism
If l ⊂ H 2 is a hyperbolic line and x ∈ H 2 is a point not on
l, then there exist two parallels (in the narrow sense) to l
through x and infinitely many ultra-parallels. The angle α
between one of the parallels and the perpendicular through
x is called the angle of parallelism. It depends only on the
distance b from x to l. In fact, α = 2 arctan e−b (exercise).
More distance formulas

Let x ∈ H 2 and let l = {y ∈ H 2 | hy, ni = 0} be a hyperbolic line.
Proposition. The point on l that is closest to x is the intersection xp of l with the line lp through
x that is perpendicular to l.
Proof. The case x ∈ l is trivial, so assume x 6∈ l. Let γ : R → l ⊂ H 2 be a parameterization
of l with unit speed. So hγ, γi = −1, hγ, ni = 0, and hγ 0 , γ 0 i = 1. Let f (s) = d(x, γ(s)) =
arcosh(−hx, γ(s)i). Convince yourself that f (s) → ∞ as s → ±∞. So f must attain a minimum,
say at s = s0 . Then f 0 (s0 ) = 0, and this implies hx, γ 0 (s0 )i = 0. Let xp = γ(s0 ), np = γ 0 (s0 ),
and let lp = {y ∈ H 2 | hy, np i = 0}. Then
• xp is the point on l closest to x,
• hx, np i = 0 so x ∈ lp .
• hxp , np i = 0 because hγ, γ 0 i = 0, so xp ∈ lp ,
• hn, np i = 0 because hγ 0 , ni = 0, so l ⊥ lp .
17
Proposition. The distance d(x, l) from x to l satisfies |hx, ni| = sinh d(x, l).
Proof. Let V = (x xp n np ) and let E be the diagonal matrix with 1, 1, −1 on the diagonal.
Then
 
hx,xi hx,xp i hx,ni hx,np i −1 − cosh d(x ,x) hx,ni 0
!
p
hxp ,xi hxp ,xp i hxp ,ni hxp ,np i  − cosh d(xp ,x) −1
0 = det V t EV = det  hn,xi hn,xp i hn,ni hn,np i
= det hn,xi 0
0
1
0
0
hnp ,xi hnp ,xp i hnp ,ni hnp ,np i 0 0 0 1
= 1 + hx, ni2 − cosh2 d(xp , x) = − sinh2 d(xp , x) + hx, ni2

and d(xp , x) = d(x, l) by the previous proposition.
Of course the sign of hx, ni depends on whether or not x is contained in the half-plane hx, ni ≥ 0.
Proposition. The distance d(l1 , l2 ) between two lines li = {y ∈ H 2 | hy, ni i = 0} with |hn1 , n2 i| > 1
satisfies |hn1 , n2 i| = cosh d(l1 , l2 ).
Proof (sketch). This can be proved in the same way as the previous proposition, but letting
V = (x n1 n2 np ), where lp = {y ∈ H 2 | hy, np i = 0} is the common perpendicular to l1 , l2 , and
x = l1 ∩ lp .
The sign of hn1 , n2 i depends on whether or not one of the half-planes hy, ni i ≥ 0 contains the
other.
Summary. Let x1 , x2 ∈ H 2 , let n, n1 , n2 be unit spacelike vectors and let l, l1 , l2 be the cor-
responding lines. The two sides of each line are marked + and − according to the sign of the
scalar product of points in H 2 with the chosen normal on that side.
hx1 , ni = sinh d(x1 , l)

hx1 , x2 i = − cosh d(x1 , x2 ) hx2 , ni = − sinh d(x2 , l)
hn1 , n2 i = cosh d(l1 , l2 ) hn1 , n2 i = cos α̂ = − cos α
hn1 , n2 i = − cosh d(l1 , l2 )
Hyperbolic “trilaterals”
Let us define a hyperbolic trilateral as a non-empty intersection of three half-planes, of which
none is contained in another. If we only consider the generic cases where pairs of boundary lines
intersect or are ultra-parallel then there are four types of trilaterals according to the number of
vertices. In the figures, the common perpendiculars of ultra-parallel lines are drawn in yellow.
(1) (2) (3) (4)
The first case is the case of triangles. In the other cases, one can also derive trigonometric
formulas in the same way as we did for triangles. Of particular interest are trilaterals of type (4),
which correspond to right-angled hexagons. For those one obtains the cosine and sine theorems
cosh a + cosh b cosh c sinh a0 sinh b0 sinh c0
cosh a0 = and = = .
sinh b sinh c sinh a sinh b sinh c
18
Intersections of H 2 ⊂ R2,1 with planes

Any plane in R2,1 is determined by an equation of the form hx, vi = b with
v ∈ R2,1 \ {0} and b ∈ R. There are three types of planes according to whether v
is spacelike, timelike or lightlike.
• If hv, vi < 0, we may assume that hv, vi = −1 and v3 > 0, so v ∈ H 2 . A plane
of this type which intersects H 2 is of the form hv, xi = − cosh r. The intersection
is therefore a circle with center v and radius r.
• If hv, vi > 0, we may assume that hv, vi = 1 and b > 0. A plane of this type
always intersects H 2 and is of the form hv, xi = sinh r. The intersection with H 2
is therefore a curve of constant distance from the line hv, xi = 0. (For r = 0, it is
the line itself.)
• If hv, vi = 0, a non-empty intersection with H 2 is called a horocircle.
The Poincaré disk model of the hyperbolic plane

0
Project H 2 to the plane x3 = 0 with e3 = 0 as center
−1
of projection. This is stereographic projection of H 2 . It maps
H 2 to the unit disk of the plane x3 = 0. Analytically, it is the
map σH 2 : H 2 → D2 ,
 
x1
1 x1
σH 2 x2  = ,
x3 + 1 x2
x3
 
2u1
u1 1
σH 2 −1 =  2u2 .
u2 1 − u1 2 − u2 2
1 + u1 2 + u2 2
Theorem. Stereographic projection of σH 2 : H 2 → D2

maps intersections of H 2 ⊂ R2,1 with planes in R2,1 to
intersections of D2 with circles and lines.
In particular, hyperbolic lines are mapped to intersections
of D2 with circles and lines that intersect the unit circle
∂D2 orthogonally. 7 lines, 2 circles, 1 horocircle (left).
This can be shown by a calculation like in the case of S . 2 Curves of constant distance form a
line (right).
Theorem. Consider two curves γ̂, η̂ : (−ε, ε) → H 2 in the hyperbolic plane with γ̂(0) = η̂(0) = p̂.
Let γ = σH 2 ◦ γ̂ and η = σH 2 ◦ η̂ be their images in D2 under stereographic projection, and let
v̂ = γ̂ 0 (0), ŵ = η̂ 0 (0), v = γ 0 (0), w = η 0 (0). Then
4
hv̂, ŵiR2,1 = hv, wiR2 ,
(1 − p1 2 − p2 2 )2
where h·, ·iR2,1 is the Lorentz scalar product of R2,1 and h·, ·iR2 is the standard Euclidean scalar
product of R2 .
This, too, can be shown by a calculation like in the case of S 2 .
Hence σH 2 is conformal in the sense that curves in H 2 intersecting at
some angle are mapped to curves inside the unit circle of the Euclidean
plane intersecting at the same angle.
One can measure hyperbolic lengths and angles directly in D2 by using
the variable scalar product
4
gp (v, w) = hv, wiR2 . (∗)
(1 − p1 2 − p2 2 )2
For example, the hyperbolic length of a curve in H 2 ⊂ R2,1 which σH 2

maps to γ : [t1 , t2 ] → D2 is
Z t2 q
gγ(t) (γ 0 (t), γ 0 (t)) dt.
t1
19
In the image of H 2 under stereographic projection, lengths appear scaled down by the variable
factor 12 (1 − p1 2 − p2 2 ). The image in D2 of an object in the hyperbolic plane gets smaller and
smaller as it moves towards the boundary circle ∂D2 .
A Riemannian metric on an open set U ⊆ Rn is a variable Euclidean scalar product. More
precisely, it is a C ∞ map g : U × Rn × Rn → R, (p, v, w) 7→ gp (v, w), such that for each
p ∈ U , gp (·, ·) is a Euclidean scalar product on Rn . (Thus, gp (v, w) = v t G(p)w with a matrix
G depending on p ∈ U .) One can then measure lengths of curves and angles in U using the
Riemannian metric, and this is called Riemannian geometry.
A Riemannian metric g on U ⊆ Rn is called conformal if
gp (u, v) = λ(p)hu, viRn
for some function λ : U → R>0 . If this is the case, angles measured using g are equal to the
Euclidean angles measured using h·, ·iRn .
Thus, gp as defined by equation (∗) is a conformal Riemannian metric on D2 . The unit disk D2
with this Riemannian metric is called the Poincaré disk model of the hyperbolic plane.
From Klein model to Poincaré disk model via the hemisphere model
We have encountered two ways to map H 2 ∈ R2,1 to the unit disk. Central projection gives the
Klein model and stereographic projection gives the Poincaré disk model. The composition
central stereographic
D2 −−−−−−→ H 2 −−−−−−−−→ D2
projection projection
is a peculiar self-map of the unit disk D2 , which maps secants of D2 to circles orthogonal to the
boundary ∂D2 .
Proposition. The same map D2 → D2 is also the result of the

following construction: First, project D2 ⊂ R2 ⊂ R3 orthogo-
nally down to the lower hemisphere of S 2 ⊂ R3 . Then project
stereographically back to D2 .
From Prasolov & Tikhomirov,

Geometry .
x1
1 x1

Proof. Let x = x2
x3
∈ H 2 . Via central projection, this corresponds to x3 x2 in the Klein
model. Orthogonally projecting down to the lower hemisphere of S 2 gives
 
x1 /x3 
x1
  
x1
x2 /x3 1 1
= x =  x2 
  
√ 2
x3 x3
 q
− 1 − ( xx13 )2 − ( xx23 )2 − x3 2 − x1 2 − x2 2 −1
Projecting this point back to the unit disk (by stereographic projection of S 2 ) results in
x1
x1 1 x3 1 x1
σ x13 x2 = x = ,
1 + x1
2
−1
x3
3
x3 + 1 x2
which is the same as σH 2 (x).

x1 x1
The calculation in the proof shows that xx2 7→ x13 x2 maps H 2 to the lower hemisphere
x1 x1 3 −1
2 1
of S . Equally x 7→ x3
x 2 x 2 maps H to the upper hemisphere of S 2 . These images of H 2
2
3 1
are called the (lower and upper) hemisphere models. The hemisphere models are also conformal,
and circles, horospheres, lines, and curves of constant distance from lines are represented by
(parts of) circles in S 2 .
20
The Poincaré half-plane model
One obtains the Poincaré half-plane model of H 2 by projecting

the upper hemisphere model 1stereographically
from a point on
the equator, say from e1 = 0 to the plane x1 = 0. This maps
0
the equator (minus e1 ) to the x2 -axis and the upper hemisphere
to the upper half-plane of the the x2 , x3 -plane which we identify
with the upper half-plane
2
= uu12 u2 > 0 ⊂ R2 .

H+ From W.P. Thurston, Three-Dimensional Ge-
ometry and Topology.
Since stereographic projection is conformal and maps circles to circles and lines, hyperbolic lines
2
are represented in H+ by half circles meeting the u1 -axis orthogonally and vertical lines. You
2
can show by a direct calculation that hyperbolic lengths and angles can be measured in H+ by
using the Riemannian metric
1
gp (v, w) = 2 hv, wiR2 .
u2
2
The half-plane H+ with this Riemannian metric is called the Poincaré half-plane model of the
2
hyperbolic plane. In H+ , hyperbolic lengths appear scaled by the variable factor u2 which is the
Euclidean distance to the boundary.
Two examples for length calculations in the half-plane model (and some remarks)

2 sin t
(1) Consider the curve γ : [0, α] → H+ , γ(t) = r . Its hyperbolic length is
cos t
Z αq Z αs
1 cos t cos t
gγ(t) (γ 0 (t), γ 0 (t)) dt = r , r dt
0 0 r2 cos2 t − sin t − sin t
Z α
1 1 1 + sin α α π
= dt = log = log tan + . (Check it out!)
0 cos t 2 1 − sin α 2 4

2 0
(2) Consider the curve η : [t1 , t2 ] →H+ , η(t) = . Its hyperbolic length is
t
Z t2 q Z t2 s Z t2
0 0
1 0 0 1
gη(t) (η (t), η (t)) dt = , dt = dt
t1 t1 t2 1 1 t1 t
t2
= log t2 − log t1 = log .
t1
Note that in the first example, the length does not depend on r, and in the second example, the
length depends only on the quotient t2 /t1 . In fact, scaling transformations uu12 7→ λ uu12 (with
λ > 0) of the upper half-plane represent isometries of the hyperbolic plane. For example, scaling
21
by the factor 2 makes all objects in the upper half-plane look twice as large. At the same time,
all distances from the boundary also double. In effect, hyperbolic lengths stay the same.
Horizontal translations uu12 7→ u1u+c

2
of the upper half-plane also represent isometries of H 2 ,
and so do reflections on vertical lines.
Calculating hyperbolic areas in the half-plane model

If Euclidean lengths in the upper half-plane model have to be scaled by a variable factor of 1/u2
to get the hyperbolic length, then Euclidean area has to be scaled by the factor (1/u2 )2 . So the
2
hyperbolic area of a region R ⊂ H+ is
Z
1
area(R) = 2
du1 du2 .
R u2
Theorem. The area of a hyperbolic triangle with interior angles α, β, γ is

π − α − β − γ.
Remark. In Lecture 8, I sketched a purely analytic proof for the fact

that the angle sum in a hyperbolic triangle is always less than π. A
more visual argument can be made using the Poincaré disk model and
moving one vertex of the triangle to the center of the disk.
To prove the theorem, consider first a hyperbolic triangle

T (α, β, 0) with one vertex “at infinity”. (Strictly speak-
ing, this is not a triangle but what we called a trilateral.)
The figure on the right shows such a triangle in the Klein
model, in the Poincaré disk model, and in the Poincaré
half-plane (where the infinite vertex was used to project
from the hemisphere model).
Z
1
area(T (α, β, 0)) = 2
du1 du2
T (α,β,0) u2
Z cos β Z ∞
1
= √ 2
du2 du1
u1 =cos(π−α) u2 = 1−u1 2 u2
Z cos β ∞ !
1
= − du1
u1 =cos(π−α) u2 √1−u1 2
Z cos β
1
= √ du1
u1 =cos(π−α) 1 − u1 2
cos β
= − arccos u1 cos(π−α) = π − α − β.
For β → ∞ one obtains the area of a triangle with two vertices “at infinity”:
area(T (α, 0, 0)) = π − α.
Now we can calculate the area for a triangle T (α, β, γ) with angles α, β, γ:
area(T (α, β, γ)) = area(T (α, 0, 0)) − area(T (β1 , γ1 , 0)) − area(T (β2 , 0, 0)
= (π − α) − (π − β1 − γ1 ) − (π − β2 )
= π − α − (π − β1 − β2 ) − (π − γ1 ) = π − α − β − γ.
This proves the theorem.
Concluding remarks
(1) All the models we have discussed exist also for higher dimensional hyperbolic space.
(2) We have defined hyperbolic space as one sheet of a hyperboloid and then derived the other
models from it. Actually, any metric space isometric to our H n is called hyperbolic space and
H n . The hyperboloid is just a model like the others, called the hyperboloid model.
22
Projective Geometry
Introduction
Consider projecting a plane E to another

plane E 0 from a point P not on E or E 0 .
Every point in E has an image in E 0 except
points on the vanishing line of E, which is the
intersection of E with the plane parallel to E 0
through P . Every point in E 0 has a preimage
in E except points on the vanishing line of E 0 ,
which is the intersection of E 0 with the plane
parallel to E through P .
The projection maps lines to lines. A family of parallel lines in E is mapped to a family of lines
in E 0 which intersect in a point on the vanishing line.
Idea: Introduce, in a addition to the ordinary points of E, new points which correspond to points
on the vanishing line of E 0 . In the same way, introduce new points of E 0 which are images of the
vanishing line of E. These new points are called points at infinity, and the extended planes are
called projective planes. The projection becomes a bijection between projective planes. Parallel
lines in E intersect in a point at infinity. The points at infinity of E form a line called the line
at infinity which corresponds to the vanishing line of E 0 .
Drawing a floor tiled with square tiles
Suppose you have already drawn the first tile. (I don’t want to go
into the details of how one can construct the image of the first tile,
even though that is interesting and not difficult.) The figure shows
how the other tiles can then be constructed.
Analytic treatment
−1 x1
Suppose E is the x1 x2 -plane, E 0 is the x2 x3 -plane, and P = 0 . A point A = x2 ∈ E
0 1 0
is mapped a point A0 = y1 ∈ E 0 , and by solving A0 = P + t(A − P ) for t one finds that
y2
y1 = x1x+12
and y2 = x1x+1
1
. So in terms of the coordinates x1 , x2 of plane E and y1 , y2 of plane
E 0 , the projection is the function
y1 x1
1 x2

y2 = f x2 := x1 +1 x1 .
The vanishing line of E is the line x1 = −1, and the vanishing line of E 0 is the line y2 = 1.
x1

Introduce homogeneous u1 coordinates: Instead of using two numbers x2 to describe a point in E,
use three numbers uu2 such that x1 = uu31 and x2 = uu32 . The homogeneous coordinates for a
3
x1

point are not unique: x2 are homogeneous coordinates for the point xx12 , but for any λ 6= 0,

λx1 1
λx2 are also homogeneous coordinates for the same point. In the same way, use homogeneous
λ v1
coordinates vv2 with y1 = vv13 and y2 = vv23 to describe a point yy12 ∈ E 0 . Let us write the

3 u1
projection f in terms of homogeneous coordinates. Let uu2 be homogeneous coordinates for
v1 3
x1 y1
f xx12 . Then

x2 and let v
v
2 be homogeneous coordinates for y2 =
3
u2
v1 x2 u3 u2
= y1 = = u1 = ,
v3 x1 + 1 u3 +1 u1 + u3
v2 u1
= ... = ,
v3 u1 + u3
so we may choose
        
v1 u2 0 1 0 u1 u1
v2  =  u1  = 1 0 0 u2  =: fˆ u2  .
v3 u1 + u3 1 0 1 u3 u3
23
Using homogeneous coordinates, the projection may thus be written as a linear map fˆ : R3 → R3 .
Moreover,
u1 fˆ is bijective! Points on the vanishing line x1 = −1 of E have homogeneous coordinates
v1
u2 with u1 +u3 = 0, and fˆ maps these to vectors v2 , which are not homogeneous coordinates
u3 0
for any point in E 0 . But we can interpret them as homogeneous coordinates for a point at infinity
of the extended plane.
Thus, each non-zero vector u ∈ R3 represents a point in a projective plane (of which it is the
vector of homogeneous coordinates) and two vectors u, u0 represent the same point if and only if
u0 = λu for some λ 6= 0.
Projective geometry
Projective geometry deals with the properties of figures that remain Pappus’ theorem
unchanged under projections. An example for a theorem of projec- C0
tive geometry is Pappus’ theorem. It talks only about points, lines, 0 B0
A
and the incidence relation between points and lines.
C 00 B 00 A00
We will see that a curve being a conic section is a projective prop-
erty. But the distinction between circles, ellipses, parabolas and
hyperbolas is not. The photograph shows a circle being projected A
to a parabola. B C
Also the distinction between ordinary points and points at infinity
is not a projective property, because as we have seen, a projection
can map ordinary points to points at infinity and vice versa. So
from the point of view of projective geometry, points at infinity of
a projective plane are not distinguished from ordinary points and
the line at infinity is a line like any other.
Basic definitions
Let V be a vector space over a field F . The projective space of V is the set P(V ) of 1-dimensional
subspaces of V . If the dimension of V is n + 1, then the dimension of the projective space P(V ) is
n. A 1-dimensional projective space is called a projective line and a 2-dimensional one is called
a projective plane. An element of P(V ) (that is, a 1-dimensional subspace of V ) is called a point
of the projective space.
If v ∈ V \{0}, then we write [v] := span v. So [v] is a point in P(V ), and v is called a representative
vector for this point. If λ 6= 0 then [λv] = [v] and λv another representative vector for the same
point.
Suppose we chose a basis v1 , . . . , vn+1 of V . This gives an identification of V with F n+1 and of
P (V ) with P (F n+1 ). A vector v ∈ V has a basis representation
n+1
X
v= xj vj
j=1
and x1 , . . . , xn+1 ∈ F are the coordinates of v with respect to the basis. These coordinates of
v ∈ V are the homogeneous coordinates of the point [v] ∈ P(V ). If λ 6= 0, then λx1 , . . . , λxn+1
are also homogeneous coordinates of [v].
Let U ⊂ P(V ) be the subset of points
for which a particular homogeneous coordinate, say xn+1 ,
Pn+1
does not vanish: U = [v] ∈ P(V ) v = j=1 xj vj with xn+1 6= 0 . Then the map Un+1 → F n ,
 u1   x1 /xn+1   u1 
u2 x2 /xn+1 u2 Xn
v 7−→  ..  =   is a bijection with inverse  ..  7−→ uj vj + vn+1 ,
 
..
. . . j+1
un xn /xn+1 un
and u1 , . . . , un are called affine coordinates of v ∈ Un+1 .

We will mostly consider the case where the base field F of the vector space is the field of real
numbers R. In this case, the concepts of a point, line or curve, etc., have their intuitively
geometric meaning. But many theorems of projective geometry hold for arbitrary base fields. In
particular, when dealing with curves and surfaces defined by algebraic equations, it is natural to
use the base field C. Finite fields are used in elliptic curve cryptography.
One usually writes RPn for P(Rn+1 ) and CPn for P(Cn+1 ). More generally, if V is any real or
complex vector space, then P (V ) is called an RPn or CPn , respectively.
24
Basic examples
The points of RP1 , the one-dimensional real projective space or the

real projective line are the 1-dimensional subspaces
x1 x1 2
where xx12 6= 0.

x2 = R x2 ⊂ R ,
The points with homogeneous coordinate x2 6= 0 are described by

x1 /x2 ∈ R. On the other hand, all
one affine coordinate represen-
tative vectors x01 ∈ R2 \ {0} represent the same point 10 ∈ RP1 .

So one can think of RP1 as R plus one additional point (which is

reasonably denoted by ∞).
The points of RP2 , the two-dimensional real projective space or the
real projective plane are the 1-dimensional subspaces
h x1 i x1 x1
3
x2 = R x2
x x
⊂ R , where x2
x
6= 0.
3 3 3
The points with homogeneous coordinateh x1xi3 6= 0 are described by two affine coordinates x1 /x3 ,
x2 /x3 . On the other hand, the points x2 form the 1-dimensional real projective space P(U ),
0x1
where U ⊂ R3 is the subspace U = { x2 ∈ R3 }. So RP2 can be thought of as R2 plus a
0
projective line.
In general one can think of RPn as Rn plus an additional RPn−1 .
In the same way, the complex projective line CP1 is C plus one additional point ∞. In complex
analysis, CP1 = C ∪ {∞} is called the extended complex plane and denoted by C.b
Projective subspaces
A projective subspace of the projective space P(V ) is a projective space P(U ), where U is a vector
subspace of V . If k is the dimension of P(U ) (that is, k + 1 is the dimension of U ), then P (U )
is called a k-plane in P (V ). In particular, for k = 1 it is called a line, for k = 2 a plane and for
k = n − 1 a hyperplane in P (V ).
Exercise. How many points are there in the projective plane P (Z2 3 )? How many lines? How
many points does each line contain? How many lines pass through each point?
Proposition. Through any two distinct points in a projective space there passes a projective line.
Proposition. Two distinct lines in a projective plane intersect in a unique point.
(Proofs by linear algebra.)
In general, if P(U1 ) and P(U2 ) are two projective subspaces of P (V ), then the intersection
P(U1 ) ∩ P(U2 ) is the projective subspace P(U1 ∩ U2 ). The projective span or join of P(U1 ) and
P(U2 ) is the projective subspace P (U1 + U2 ).
Exercise. Show that a point is in the join of two projective subspaces if and only if it is on a
line joining a point in one of the subspaces with a different point in the other. (Actually, there
one a very degenerate case in which this is not true.)
From the dimension formula of linear algebra,
dim(U1 + U2 ) = dim U1 + dim U2 − dim(U1 ∩ U2 ),
one obtains
dim(P(U1 + U2 )) = dim P(U1 ) + dim P(U2 ) − dim(P(U1 ) ∩ P(U2 )).
25
Desargues’ theorem
Theorem (Desargues). Let A, A0 , B, B 0 , C, C 0 be points of a projec-

tive plane such that the lines AA0 , BB 0 and CC 0 intersect in one
point P .
Then the intersection points C 00 = AA0 ∩ BB 0 , A00 = BB 0 ∩ CC 0
and B 00 = CC 0 ∩ AA0 lie on one line.
If this theorem is considered in an affine plane instead of a projec-
tive plane, it breaks up into several special cases which have to be
considered separately. For example: Under the same conditions, if
AB k A0 B 0 and BC k B 0 C 0 then CA k C 0 A0 .
Some preparation is necessary before the following proof. Let P (V ) be an n-dimensional projec-
tive space. Then n + 2 points in P (V ) are said to be in general position if one (hence both) of
the following equivalent conditions are satisfied:
(i) No n + 1 of the points are contained in an (n − 1)-dimensional projective subspace.
(ii) Any n + 1 of the points have linearly independent representative vectors.
So three points on a line are in general position if they are distinct, four points in a plane are
in general position if no three of them lie on a line, and five points in a 3-dimensional projective
space are in general position if no four of them lie in a plane.
Lemma. Let P (V ) be an n-dimensional projective space and suppose P1 , . . . Pn+2 ∈ P (V ) are
in general position. Then representative vectors v1 , . . . , vn+1 ∈ V may be chosen so that
v1 + v2 + · · · + vn+1 = vn+2 .
This choice is unique up to a common factor. That is, if ṽ1 , . . . , ṽn+1 is another choice of
representative vectors with ṽ1 + ṽ2 + · · · + ṽn+1 = ṽn+2 , then ṽk = λvk for some λ 6= 0.
Proof of the lemma. Let w1 , . . . , wn+2 be any representative vectors for the points P1 , . . . Pn+2 .
They are linearly dependent because dim V = n + 1. So
n+2
X
aj wj = 0
j=1
for some aj which are not all zero. In fact, no ak can be zero, because that would mean that
there are n + 1 among the wj which are linearly dependent. Hence we may choose
v1 = a1 w1 , v2 = a2 w2 , ... vn+1 = an+1 wn+1 , vn+2 = −an+1 wn+1 .
To see the uniqueness
Pn+1 claim, suppose λ1 v1 , . . . , λn+2 vn+2 is another choice of representative
vectors with 1 λk vk −λn+2 vn+2 = 0. This amounts to a system of equations of rank n+ 1 for
the n+2 variables λk . So the solution space is 1-dimensional and hence λ1 = λ2 = . . . = λn+2 .
Proof of Desargues’ theorem. If A, A0 , P are not distinct the statement of the theorem is obvious.
(Check this.) So we may assume that A, A0 , P are distinct and also B, B 0 , P and C, C 0 , P . But
then A, A0 , P are three points on a line in general position. So by the lemma we may choose
representative vectors a, a0 , p ∈ V with a + a0 = p. For the same reason we may also choose
representative vectors b, b0 and c, c0 so that b + b0 = p and c + c0 = p. Then
a + a0 = b + b0 = c + c0 .
This implies a − b = b0 − a0 . Obviously, the vector a − b = b0 − a0 is in the span of a and b and
also in the span of a0 and b0 . So the point [a − b] = [b0 − a0 ] ∈ P(V ) lies on the line AB and on
the line A0 B 0 , hence it is the point of intersection, C 00 . Similarly, A00 = [b − c] = [c0 − b0 ], and
B 00 = [c − a] = [a0 − c0 ]. But
(a − b) + (b − c) + (c − a) = 0,
which means that vectors (a − b), (b − c), (c − a) are linearly dependent and so they span a
subspace of dimension at most 2. Therefore, C 00 , A00 and B 00 lie on a line.
26
Desargues’ theorem says:

If the lines joining corresponding points of two triangles meet
in one point, then the intersections of corresponding sides lie
on one line.
The converse is also true:
If the intersections of corresponding sides of two triangles lie
on one line, then the lines joining corresponding points meet in
one point.
Surprisingly, the converse statement is in fact equivalent to the
original statement after a permutation of the point labels (see
figure right).
The Desargues configuration turns out to be very symmetric

in the sense that there are many permutations of the labels
A, B, C, A0 , B 0 , C 0 , A00 , B 00 , C 00 and P preserving the relevant in-
cidences.
Desargues’ theorem also holds for triangles in two different

planes of a 3-dimensional projective space P (V ). In this case,
it can be proved without any calculations: The intersection
points of corresponding sides lie on the line in which the planes
of the two triangles intersect.
The planar version of Desargues’ theorem can also be proved without any calculations if the
third dimension is used:
Proof (a 3d proof of Desargues’ theorem). Let E be the plane of the two triangles ABC, A0 B 0 C 0
and the point P . Choose a line through P which is not in E and two points X and Y on it. The
lines XA and Y A0 lie in one plane, so they intersect in a point Ã. Similarly, let B̃ = XB ∩ Y B 0
and C̃ = XC ∩ Y C 0 . Now the intersection of the line ÃB̃ and the plane E lies on the line AB,
because the plane XÃB̃ intersects E in AB. Similarly, ÃB̃ ∩ E also lies on the line A0 B 0 , so
ÃB̃ ∩ E = C 00 . In the same way, B̃ C̃ ∩ E = A00 and C̃ Ã ∩ E = B 00 . Hence A00 , B 00 and C 00 lie on
the line where E intersects the plane ÃB̃ C̃.
27
The preceding proof also suggests the following 3-dimensional way to generate any planar De-
sargues configuration. This construction also reflects the high degree of combinatorial symmetry
of the configuration. Let P1 , P2 , P3 , P4 , P5 be five points in general position in a 3-dimensional
projective space, and let E be a plane that contains none of these points. Let lij = lji be the 10
lines joining Pi and Pj (i 6= j). The 10 points Pij where these lines intersect E form a Desargues
configuration. If (i, j, k, r, s) is any permutation of (1, 2, 3, 4, 5), then the points Pij , Pjk , Pki al-
ways lie on a line (the intersection of the plane Pi Pj Pk with E), which we denote by grs . Any
one of the points Puv lies on the line gxy if the four indices uvxy are different. So there are three
lines through each point and three points on each line. Corresponding points of the triangles
Pir , Pjr , Pkr and Pis , Pjs , Pks are joined by the lines gjk , gki , gij , which all pass trough Prs . The
intersection points of corresponding sides all lie on the line grs .
The same Desargues figure contains therefore 5·4/2 = 10 pairs of triangles satisfying the condition
of Desargues’ theorem.
Pappus’ theorem
Theorem (Pappus). Let A, B, C be points on one line in a C0

0 B0
projective plane P (V ), and let A0 , B 0 , C 0 be points on another A
line. Then the points C 00 B 00 A00
C 00 = AB 0 ∩ A0 B, A00 = BC 0 ∩ B 0 C, B 00 = CA0 ∩ C 0 A
A B
lie on a line. C
Proof. At most one of the points A0 , B 0 , C 0 may lie on the line through A, B, C, so we may
assume without loss of generality that A, B, B 0 , C 0 are in general position. Then we may choose
representative vectors a, b, b0 , c0 ∈ V with a + b + b0 = c0 .Now b, b0 form
a,0 1abasis for V in which
1 0
the homogeneous coordinate vectors for A, B, B 0 , C 0 are 0 , 1 , 0 , 1 . So we may assume
0 0 1 1
without loss of generality that
h1i h0i h0i h1i
A = 0 , B = 1 , B0 = 0 , C 0 = 1 .
0 0 1 1
Then h1i h1i

C= c and A0 = 1
0 a
for some c and a. Now

nh x1 i o nh 1 0 io h1i
AB 0 = x2 x2 = 0 and A 0
B = s 1 +t 1 so C 00
= 0 ,

x3 a 0 a
nh x1 i o nh 0 1 io h1i
BC 0 = x2 x1 − x3 = 0 and B 0 C = s 0 + t c so A00 = c ,

x3 1 0 1
nh x1 i o nh 1 1 io h a+c−1 i
0 0 00
CA= x x2 − x3 = 0 and CA = s c + t 1 so B =

2 ac
x 3 0 a ac
(where B 00 is obtained by a quick calculation on the side). Now

1 1 a+c−1
(c − 1) 0 + a c − ac = 0,
a 1 ac
so the vectors are linearly dependent and hence A00 , B 00 , C 00 lie on a line.
28
The synthetic approach to projective geometry

We have defined a projective space P (V ) of a vector space V over a field as the set of 1-dimensional
subspaces of V . Of course this definition is based on basic axioms of algebra: the field axioms and
the vector space axioms. This section provides a rough outline of how the theory of projective
spaces P (V ) can be based on geometric axioms.
The following definition of a projective space in terms of geometric axioms (due to Oswald Veblen
& John W. Young, 1908) is not equivalent to our definition of a projective space of a vector space
over a field. (One obtains an equivalent definition if Pappus’ theorem is added as an independent
axiom; see the structure theorem below.)
A projective space P = (P, L) is a set P, the elements of which are called points, together with
a set L of subsets of P , which are called lines, such that the following axioms are satisfied.
Axiom 1. For any two distinct points there exists a unique line which contains both points.
Axiom 2. If A, B, C ∈ P are three distinct points and l ∈ L is a line that intersects the lines
AB and AC in distinct points, then l intersects the line BC.
Axiom 3. Every line contains at least three points.
Axiom 1 implies that two lines intersect in at most one point. Axiom 2 is a clever way of saying
that two lines in one plane always intersect without first defining what a plane is.
A projective subspace of P is a subset U ⊆ P of points such that the line through any two points
of U is contained in U. Together with the subset of lines {l ∈ L | l ⊆ U }, the subspace is a
projective space in its own right. The intersection of two projective subspaces is a projective
subspace.
If S ⊆ P is any set of points, then the projective span of S is the smallest projective subspace
containing S, or equivalently, the intersection of all projective subspaces which contain S.
The dimension of the projective space P is the smallest number n for which there exist n + 1
points P1 , . . . , Pn+1 ∈ P such that P is the projective span of {P1 , . . . , Pn+1 }.
The Axioms 1–3 together with the assertion that the dimension of P is 2 are equivalent to the
following axioms for a projective plane. (Can you prove this equivalence? It is a little tricky.)
Axiom P1. Same as Axiom 1.
Axiom P2. Any two lines have non-empty intersection.
Axiom P3. Same as Axiom 3.
Axiom P4. There are at least two different lines.
If the dimension of P is at least 3, then Desargues’ theorem can be deduced from Axioms 1–3.
The 3D proof of the last lecture works in this setting, it uses only the incidence relations between
points, lines and planes, and does not involve any calculations. However, there are projective
planes in which Desargues’ theorem does not hold. The purely 2-dimensional proof does not
work here because it is based on calculations.
A projective plane in which Desargues’ theorem holds is called a Desarguesian plane.
Theorem (Veblen & Young). Any projective space in which Desargues’ theorem holds (that is,
any projective space of dimension ≥ 3 and any Desarguesian plane) is isomorphic to a projective
space P (V ) of a vector space V over a skew field F . If Pappus’ theorem also holds in P , then F
is a field.
(Two projective spaces are isomorphic if there is a bijection between their points that maps
lines to lines. A skew field satisfies all field axioms except that the multiplication may not be
commutative. You may check that our computational proof of Pappus’ theorem does not work
if multiplication is not commutative.)
In a projective plane, Desargues’ theorem can be deduced from Pappus’ theorem. (This was
demonstrated by Hessenberg in 1905. It is not obvious at all.)
Thus, any theorem that holds in any projective space P (V ) of a vector space V over a filed can
also be deduced from Axioms 1–3 together with Pappus’ theorem as independent axiom, and
vice versa. Further axioms of order and of continuity have to be added to single out the real
projective spaces RPn (just like further axioms have to be added to the general field axioms to
single out field of reals).
29
Projective transformations
Let V, W be (n + 1)-dimensional vector spaces, and let f : V → W be an invertible linear map.

Since f maps any 1-dimensional subspace [v] ⊆ V to a 1-dimensional subspace [f (v)] ⊆ W , it
defines an invertible map P (V ) → P (W ). A map between projective spaces which arises in this
way is called a projective transformation:
A map fˆ : P (V ) → P (W ) is a projective transformation if there is an invertible linear map
f : V → W such that fˆ([v]) = [f (v)].
A projective transformation maps lines in P (V ) to lines in P (W ) and generally k-planes to
k-planes.
In homogeneous coordinates, a projective transformation is represented by matrix multiplication:
x.1
A point in P (V ) with homogeneous coordinates x = .. is mapped to the point in P (W )
xn+1
with homogeneous coordinates y = Ax for some invertible (n + 1) × (n + 1) matrix A. In affine
coordinates ui = xi /xn+1 , wi = yi /yn+1 (i = 1, . . . , n) the map is a so-called fractional linear
transformation: Pn
j=1 aij uj + ai,n+1
wi = Pn .
j=1 an+1,j uj + an+1,n+1
Each wi is the quotient of two affine linear functions of the uj , where the denominator is the
same for all i.
Proposition. Two invertible linear maps f, g : V → W give rise to the same projective trans-
formation P (V ) → P (W ) if and only if g = λf for some scalar λ 6= 0.
Proof. “⇐”: If g = λf , then [g(v)] = [λf (v)] = [f (v)].
“⇒”: Suppose [g(v)] = [f (v)] for all v ∈ V \ {0}. This implies g(v) = λ(v)f (v) for some non-zero
scalar λ(v) which may a priori depend on v. We have to show that it does not. So suppose
v, w ∈ V \ {0}. If v, w are linearly dependent, then it is obvious from the definition of λ(v) that
λ(v) = λ(w). So assume v, w are linearly independent. Now
g(v + w) = g(v) + g(w) = λ(v)f (v) + λ(w)f (w)
but also
g(v + w) = λ(v + w)f (v + w) = λ(v + w)(f (v) + f (w)).
Since f (v) and f (w) are also linearly independent this implies λ(v) = λ(v + w) = λ(w).
The projective transformations P (V ) → P (V ) form a group called the projective linear group
P GL(V ). It is the quotient of the general linear group GL(V ) of invertible linear maps V → V
by the normal subgroup of non-zero multiples of the identity: P GL(V ) = GL(V )/{λI}λ6=0 .
Example 1. Affine transformations
Consider an affine transformation Rn → Rn , u 7→ M u + b with M ∈ GL(n, R) and b ∈ Rn . In

xi
homogeneous coordinates x1 , . . . , xn+1 with ui = xn+1 , this map can be written
    
x1 x1
 ..   M b 
 .. 
 .  
. 
  7−→  
 .
 xn  
   xn 
xn+1 0 ··· 0 1 xn+1
| {z }
A
The affine map x 7→ M x + b can be extended to the projective transformation
RPn → RPn , [x] 7→ [Ax].
It maps the plane at infinity xn+1 = 0 to the plane at infinity. Conversely, any projective
transformation which maps the plane xn+1 = 0 to itself corresponds in the affine coordinates u
to an affine map Rn → Rn .
30
Example 2. Central projection

Let E1 = P (U1 ) and E2 = P (U2 ) be two hyperplanes in a projective space P (V ) (for example,
two lines in a projective plane) and let W ∈ P (V ) be a point not in E1 or E2 . Then the central
projection from E1 to E2 with center W is the map fˆ : E1 → E2 that maps a point A ∈ E1 to
the intersection of E2 with the line through W and A.
Proposition. The central projection fˆ is a projective transformation E1 → E2 .
Proof. We have to show that fˆ comes from an invertible linear map f : U1 → U2 . Note that W ,
as point in P (V ), is a 1-dimensional subspace of V . Since it does not lie in E2 , W ∩ U2 = {0}.
This means that V is the direct sum
V = W ⊕ U2 ,
and there are two linear maps pW : V → W and pU2 : V → U2 (the projections onto W and
U2 ) such that for any v ∈ V , pW (v) and pU2 (v) are the unique vectors in W and U2 such that
v = pW (v) + pU2 (v).
Claim: The central projection fˆ comes from the linear map pU2 |U1 , the restriction of pU2 to U1 .
To see this, let a ∈ U1 be a representative vector of A ∈ E1 . Then pU2 (a) 6= 0, because pU2 (a) = 0
would mean a ∈ W , but U1 ∩ W = {0} because by assumption E1 does not contain W . This
shows that pU2 |U1 is invertible, because it an injective linear map U1 → U2 and dim U1 = dim U2 .
Now pU2 (a) ∈ U2 , so [pU2 (a)] ∈ E2 . Also a = pW (a) + pU2 (a), or
pU2 (a) = a − pW (a),
so pU2 (a) ∈ [a] + W , which means that [pU2 (a)] is in the (projective) line through A ∈ P (V ) and
W ∈ P (V ). Hence [pU2 (a)] is the intersection of E2 with the line through W and A, so it is the
image of A under the central projection.
One can consider also more general types of projections. For example let l1 and l2 be two lines
in a 3-dimensional projective space P (V ), and let l0 be a line that does not intersect l1 or l2 .
Then the projection l1 → l2 with the line l0 as center of projection is defined as follows: A point
A ∈ l1 is mapped to the intersection of l2 with the plane spanned by l0 and A. This map l1 → l2
is also a projective transformation, and the proof is the same (apart from obvious modifications).
Most generally, in an n-dimensional projective space P (V ), one can project one k-plane E1 to
another k-plane E2 from any (n − k − 1)-plane EC which does not intersect E1 or E2 as center
of projection; and this is a projective transformation E1 → E2 .
Theorem. Let P (V ) and P (W ) be two n-dimensional projective spaces and suppose

A1 , . . . , An+2 ∈ P (V ) and B1 , . . . , Bn+2 ∈ P (W )
are in general position. Then there exists a unique projective transformation fˆ : P (V ) → P (W )
with fˆ(Ai ) = Bi for i = 1, . . . , n + 2.
Proof. Existence: By the lemma of Lecture 13, we may choose representative Pn+1 vectors aP
1 , . . . an+2
n+1
for A1 , . . . , An+2 and b1 , . . . , bn+2 for B1 , . . . , Bn+2 such that 1 ai = an+2 and 1 bi =
bn+2 . Also by the general position assumption, a1 , . . . , an+1 and b1 , . . . , bn+1 are bases of V
and W , respectively. Hence there is an invertible linear map f : V → W with f (ai ) = bi for
i = 1, . . . , n + 1. But then also
Pn+1 Pn+1 Pn+1
f (an+2 ) = f 1 ai = 1 f (ai ) = 1 bi = bn+2 .
So f maps the 1-dimensional subspaces Ai = [ai ] ⊆ V to Bi = [bi ] ⊆ W for i = 1, . . . , n + 2.
Uniqueness: Let g : V → W be another invertible linear map with g(ai ) ∈ Bi for i = 1, . . . , n + 2.
Then b̃i = g(ai ) would be another set of representative vectors for the Bi with
Pn+1 Pn+1 Pn+1
b̃n+2 = g(an+2 ) = g i ai = i g(ai ) = i b̃i .
By the uniqueness part of the lemma from Lecture 13, this implies b̃i = λbi for some λ 6= 0, so
g = λf , and g and f induce the same projective transformation P (V ) → P (W ).
31
Theorem. Let l1 , l2 be two different lines in a projective plane P (V ). A projective transformation

l1 → l2 is a central projection if and only if it maps the intersection l1 ∩ l2 to itself. Otherwise
it is the composition of two central projections l1 → l → l2 .
Proof.
One can classify projective transformations fˆ : P (V ) → P (V ) according to the normal forms

of the corresponding linear maps f : V → V . Note that fixed points of fˆ correspond to 1-
dimensional eigenspaces of f . Let us consider a projective transformation fˆ : RP1 → RP1
corresponding to the linear map f : R2 → R2 . The following cases are possible:
1. f has two distinct real eigenvalues. Then it has two linearly independent eigenvectors. So the
projective transformation fˆ has two fixed points. The eigenvectors of f form a basis of R2 in
λ1 0 1
which f has the matrix 0 λ2 . In the affine coordinate u of RP corresponding to this basis,
the fixed points are 0 and ∞ and the projective map fˆ is the scaling transformation u 7→ λλ21 u.
In other affine coordinates, fˆ looks like the projected image of a scaling transformation.
2. f has two linearly independent eigenvectors to the same eigenvalue. Then f is a multiple of
the identity and fˆ is the identity.
3. f has a double eigenvalue but only one linearly independent eigenvector, so fˆ has only
one fixed point. Then there is a basis in which the matrix of f has the Jordan normal form
λ 1 = λ 1 1/λ . In the affine coordinate u corresponding to this basis, the fixed point is ∞

0 λ 0 1
and fˆ is the translation u 7→ u + λ1 . In another chice of affine coordinate, fˆ looks like a projected
image of a translation.
4. f has a pair of complex conjugate
√ eigenvalues. Then there is a basis of R2 in which the matrix
a −b 2 2 cos ϕ − sin ϕ
of f has the form b a = a + b sin ϕ cos ϕ for some ϕ, and we may forget about the
√
factor a2 + b2 . So fˆ comes from a linear map which looks in some basis of R2 like a rotation.
32
A proof of Pappus’ theorem using central projections

Let X = AC ∩ C 00 A00 . We have to show
X = B 00 . Consider the sequence of central
projections
A0 B0 C 00
AC 0 −→ l −→ BC 0 −→ AC 0 ,
where the centers of the projections are

written above the arrows. They map
A 7−→ A 7−→ R 7−→ A

Q 7−→ B 7−→ B 7−→ Q
B 00 7−→ C 7−→ A00 7−→ X
C 0 7−→ P 7−→ C 0 7−→ C 0 .
So the composition is a projective transformation AC 0 → AC 0 which fixes the three points
A, Q, C 0 , so it must be the identity. It also maps B 00 to X, so B 00 = X.
The cross ratio

The cross ratio of four distinct points Pi = xyii ∈ RP1 (i = 1, . . . , 4) is the number

det xy11 xy22 det xy33 xy44

(x y − x2 y1 )(x3 y4 − x4 y3 )
cr(P1 , P2 , P3 , P4 ) := = 1 2 .
det xy22 xy33 det xy44 xy11

(x2 y3 − x3 y4 )(x4 y1 − x1 y4 )
It is well defined because the expressions on the right hand side do not depend on the choice of
representative vectors (xi , yi ) but only on the points Pi .
To derive an expression for the cross ratio in terms of the affine coordinate u = xy , we assume at
first that no yi is 0 so that no ui is ∞:
y1 y2 xy11 − xy22 y3 y4 xy33 − xy44

(u − u2 )(u3 − u4 )
cr(P1 , P2 , P3 , P4 ) = = 1 =: cr(u1 , u2 , u3 , u4 ).
y2 y3 xy22 − xy33 y4 y1 xy44 − xy11

(u2 − u3 )(u4 − u1 )
Even if one of the ui is ∞, one gets correct results using this formula if one “cancels infinities”.
For example, if y1 = 0 so that u1 = ∞, one has
x1 y2 y3 y4 xy33 − xy44

x1 y2 (x3 y4 − x4 y3 ) u3 − u4
cr(P1 , P2 , P3 , P4 ) = = x2 x3 = − ,

(x2 y3 − x3 y3 )(−x1 y4 ) y2 y3 (−x1 y4 ) y2 − y3 u2 − u3
so the following calculation gives the correct result:

: −1

(∞−u2 )(u3 − u4 ) u3 − u4
cr(∞, u2 , u3 , u4 ) = =− .

(u2 − u3 ) 4 − ∞)
(u u2 − u3
Proposition. (i) If f : RP1 → RP1 is a projective transformation and f (Pi ) = Qi , then
cr(P1 , P2 , P3 , P4 ) = cr(Q1 , Q2 , Q3 , Q4 ).
(ii) Ifv, w is any basis of R2 and x̃i , ỹi are homogeneous coordinates for Pi in this basis (that is,
Pi = x̃i v + ỹi w ) and ũ = x̃ỹ is the corresponding affine coordinate, then one may just as well
use these coordinates to compute the cross-ratio:
(x̃1 ỹ2 − x̃2 ỹ1 )(x̃3 ỹ4 − x̃4 ỹ3 ) (ũ1 − ũ2 )(ũ3 − ũ4 )
cr(P1 , P2 , P3 , P4 ) = =
(x̃2 ỹ3 − x̃3 ỹ4 )(x̃4 ỹ1 − x̃1 ỹ4 ) (ũ2 − ũ3 )(ũ4 − ũ1 )
All this works not only for the real projective line RP1 but also for the complex projective line
CP1 and any other projective space P (V ) of a 2-dimensional vector space V over any field.
! If v, w is a basis of V , then the cross ratio of four points Pi with homogeneous coordinates !
x̃i , ỹi in this basis is defined defined by the equation above, and this is independent of the
choice of basis.
33
Proposition. The cross-ratio cr(P1 , P2 , P3 , P4 ) is the affine coordinate of the image of P1 under
the projective transformation that maps P2 , P3 , P4 to the points with affine coordinates 0, 1, ∞.
Corollary. The cross ratio of four distinct points can take all values except 0, 1, ∞.
Proposition. There exists a projective transformation that maps four distinct points P1 , P2 , P3 , P4
of a line to four distinct points Q1 , Q2 , Q3 , Q4 on the same or another line if and only if
cr(P1 , P2 , P3 , P4 ) = cr(Q1 , Q2 , Q3 , Q4 ).
The cross ratio depends on the order of the points. How does it change if the points are permuted?
• The cross ratio does not change if I simultaneously interchange two of the points and the
remaining two:
cr(u1 , u2 , u3 , u4 ) = cr(u2 , u1 , u4 , u3 ) = cr(u3 , u4 , u1 , u2 ) = cr(u4 , u3 , u2 , u1 ).
This is easy to see from the equation for the cross ratio in terms of the ui .
• Of the 24 permutations of u1 , u2 , u3 , u4 , I need therefore only consider the six which fix u1 and
permute u2 , u3 , u4 .
• If i, j, k, l is a permutation of 1, 2, 3, 4, then
cr(u1 , u2 , u3 , u4 ) = cr(v1 , v2 , v3 , v4 ) ⇐⇒ cr(ui , uj , uk , ul ) = cr(vi , vj , vk , vl ).
Indeed, there is a projective transformation that maps u1 , u2 , u3 , u4 to v1 , v2 , v3 , v4 if and only

if there is one that maps ui , uj , uk , ul to vi , vj , vk , vl . (It’s the same map in both cases.)
• Hence if cr(u1 , u2 , u3 , u4 ) = q = cr(q, 0, 1, ∞), then
(q − 1)(0 − ∞)
cr(u1 , u3 , u2 , u4 ) = cr(q, 1, 0, ∞) = = 1 − q,
(1 − 0)(∞ − q)
(q − 0)(∞ − 1) q
cr(u1 , u2 , u4 , u3 ) = cr(q, 0, ∞, 1) = = ,
(0 − ∞)(1 − q) q−1
(q − ∞)(1 − 0) 1
cr(u1 , u4 , u2 , u3 ) = cr(q, ∞, 1, 0) = = ,
(∞ − 1)(0 − q) q
(q − 1)(∞ − 0) q−1 1
cr(u1 , u3 , u4 , u2 ) = cr(q, 1, ∞, 0) = = =1− ,
(1 − ∞)(0 − q) q q
(q − ∞)(0 − 1) 1
cr(u1 , u4 , u2 , u3 ) = cr(q, ∞, 0, 1) = = .
(∞ − 0)(1 − q) 1−q
34
Projective involutions of the real projective line

For any four points A, B, C, D ∈ RP1 there is a unique projective transformation f : RP1 → RP1
with f (A) = B, f (B) = A, f (C) = D, f (D) = C, because cr(A, B, C, D) = cr(B, A, D, C). The
transformation f is an involution, that is, f 6= identity but f ◦ f = identity.
A pair of points {A, B} ⊂ RP1 separates another pair {C, D} ⊂ RP1 if C and D are in different
connected components of RP1 \ {A, B}.
{A, B} separates {C, D} ⇐⇒ cr(A, C, B, D) < 0.
The involution f has no fixed points if {A, B} separates {C, D}, otherwise it has two fixed points.
If f has two fixed points P and Q, then for all X ∈ RP1 , cr(X, P, f (X), Q) = −1.
For any two points P, Q ∈ RP1 there is a unique projective involution of RP1 that fixes P and Q.
If A, B, P, Q are four points in RP1 , then one says the pair {A, B} separates the pair {P, Q}
harmonically, if cr(A, P, B, Q) = −1.
The complete quadrilateral
Theorem (on the complete quadrilateral).

Let A, B, C, D ∈ RP2 be four points in general posi-
tion, let P = AB ∩ CD, Q = AD ∩ BC, ` = P Q,
X = ` ∩ BD, Y = ` ∩ AC. Then the pair of points
{P, Q} on ` separates the pair {X, Y } harmonically:
cr(P, X, Q, Y ) = −1.
Here are two proofs for this theorem, one computa-

tional, one using a projective involution of the plane.
Proof (by computation). Since the points are in general position, there is a projective transfor-
mation of RP2 that maps A, B, C, D to
h 1 i h −1 i h −1 i h 1 i
1 , 1 , −1 , −1 .
1 1 1 1
If the theorem holds for the projected figure, it holds also for the original one. It is thus enough
to verify the theorem for
h1i h −1 i h −1 i h 1 i
A = 1 , B = 1 , C = −1 , D = −1 .
1 1 1 1
h1i h0i h 1 i h1i

In this case, ` is the line x3 = 0, and P = 0 , Q = 1 , X = −1 , Y = 1 . Homogeneous
0 0 0 0
coordinates on ` are obtained by dropping the x3 -coordinate (which is zero), so
det 10 −1
1

det 01 11
cr(P, X, Q, Y ) = 1 0
11
= −1.
det −1 1 det 1 0
35
Proof (using a projective involution of RP2 ). Since A, B, C, D are in general position, there is
a projective transformation of the plane RP2 that maps A 7→ B, B 7→ A, C 7→ D, D 7→ C. It
is an involution of RP2 which maps the lines AB and CD onto themselves. It maps the line
AD to BC and vice versa. Hence, the points P and Q are fixed, and the line ` is mapped to
itself. Since the line AC is mapped onto BD and vice versa, X is mapped to Y and Y to X.
Thus, the restriction to ` is an involution of ` with fixed points P, Q and interchanging X, Y . So
cr(P, X, Q, Y ) = −1.
Projective involutions of the real projective plane
Suppose f : RP2 → RP2 is a projective involution of

the real projective plane. Let A ∈ RP2 be a point
which is not a fixed point, and A0 = f (A). Then also
A = f (A0 ) and hence the line AA0 is mapped to itself.
Let B ∈ RP2 be a point not on this line which is also
not a fixed point, and B 0 = f (B). Then the line BB 0
is also mapped to itself, so P = AA0 ∩ BB 0 is a fixed
point of f .
The restriction of f to the line AA0 is an involution of AA0 with a fixed point P , so it has another
fixed point Q, and this is the point such that {P, Q} separates {A, A0 } harmonically. Equally,
the restriction of f to the line BB 0 is an involution of BB 0 with fixed points P and R such that
{P, R} separates {B, B 0 } harmonically. Now f fixes every point on the line ` = QR. (Can you
see why?) Thus:
Any projective involution of RP2 has a whole line ` of fixed points and another fixed point P 6∈ `.
Conversely, if ` is a line in RP2 and P is a point not on `, then there is a unique projective
involution f that fixes P and every point on `. This is the projective reflection on ` and P .
Indeed if X, Y are any two points on `, and any representative
−1 0 0 vectors of P, X, Y are chosen as
3
basis of R , then the matrix of f in this basis must be 0 1 0 .
0 01
(What does this reflection look like in an affine chart in which ` is the line at infinity? What
does it look like if P is a point at infinity?)
Happy holidays!
artwork: Franz Pedit creative directors: Ilya & Lysander
36
The fundamental theorem of real projective geometry

We know that projective transformations map lines to lines. For real projective spaces they are
in fact all transformations that map lines to lines:
Fundamental theorem of real projective geometry. If a bijective map RPn → RPn (n > 1)
maps lines to lines, then it is a projective transformation.
Remarks. • Note that it is not necessary to assume that the map is continuous.
• A map RPn → RPn maps lines to lines if and only if it maps k-planes to k-planes. (Why?)
• The corresponding statement for CPn is false. For example, the map CPn → CPn ,
   
z1 z̄1
 ..   . 
 .  7−→  .. 
zn+1 z̄n+1
is bijective and maps lines to lines, but it is not a projective transformation. The fundamental
theorem for general projective spaces P (V ) of a vector space V over a field F says the following:
If f : P (V ) → P (V ) is bijective and maps lines to lines, then f comes from an almost linear map
ϕ : V → V . A map ϕ : V → W between vector spaces over F is called almost linear if for all
u, v ∈ V , λ ∈ F ,
ϕ(u + v) = ϕ(u) + ϕ(v) and ϕ(λv) = α(λ)ϕ(v),
where α : F → F is a field automorphism. (For example, complex conjugation is an automor-
phism of C. The field R of real numbers has no automorphism except the identity.)
Corollary. A bijective map f : Rn → Rn which maps lines to lines is an affine transformation
f (x) = Ax + b for some A ∈ GL(n, R), b ∈ Rn .
(Because any such map Rn → Rn can be extended to a bijective map RPn → RPn which maps
lines to lines and the hyperplane at infinity to itself. (How?))
For simplicity, I will present a proof of the fundamental theorem only for the case n = 2 of the
real projective plane. This already contains all the important ideas, so you can figure out for
yourself how it works for n > 2. The proof depends on the following two lemmas.
Lemma. Suppose a map f : RP2 → RP2 is bijective and maps lines to lines. Then if A, B, C, D
are four points on a line in RP2 and cr(A, B, C, D) = −1, then cr(f (A), f (B), f (C), f (C)) = −1
as well.
Proof. Use the theorem on the complete quadrilateral.
Lemma. Suppose a map f : RP1 → RP1 of the line has the property that if A, B, C, D ∈ RP1
are four points with cr(A, B, C, D) = −1, then also cr(f (A), f (B), f (C), f (D)) = −1. Then f is
a projective transformation.
Proof. We will show that if f also fixes 0, 1, and ∞, it must be the identity. This implies the
lemma: For general f let g be the projective transformation that maps f (0) 7→ 0, f (1) 7→ 1,
f (∞) 7→ ∞. Then the composition g ◦ f satisfies the assumptions of the theorem and fixes
0, 1, ∞. If it is the identity, then f = g −1 is a projective transformation.
So assume in addition that f fixes 0, 1, ∞. Then for all x, y ∈ R:
f (x)+f (y)
(1) f ( x+y
2 )= 2 , because cr(x, x+y
2 , y, ∞) = −1.
(2) f (2x) = 2f (x), because cr(0, x, 2x, ∞) = −1.
(3) f (x + y) = f (x) + f (y). This follows from (1) and (2).
(4) f (−x) = −f (x) because 0 = f (0) = f (x + (−x)) = f (x) + f (−x).
(5) f (nx) = nf (x) for n ∈ Z. This follows from (3) and (4).
(6) f (qx) = qf (x) for q ∈ Q. This follows from (5).
(7) f (q) = q for q ∈ Q because f (q) = f (q · 1) = qf (1) = q · 1.
(8) f (x2 ) = f (x)2 . This follows form (4) and cr(−x, 1, x, x2 ).
(9) x > 0 ⇒ f (x) > 0. This follows from (8) because the a real number is positive if and only if
it is the square of a real number.
(10) f is increasing on R. This follows from (3,4,9) because
0<x−y =⇒ 0 < f (x − y) = f (x) − f (y).
Finally: An increasing function on R which fixes the rationals is the identity. (Why?)
37
Proof (of the fundamental theorem, n = 2). We will show that if f also fixes the four points
h1i h0i h0i h1i
P1 = 0 , P2 = 1 , P3 = 0 , P4 = 1 ,
0 0 1 1
it must be the identity. This implies the theorem: For general f let g : RP2 → RP2 be the pro-
jective transformation that maps f (Pi ) to Pi . Then the composition g ◦ f is bijective, maps lines
to lines and fixes the points Pi . If it is the identity, then f = g −1 is a projective transformation.
So assume that f : RP2 → RP2 is bijective, maps lines to lines and fixes P1 , P2 , P3 , P4 . Let
X ∈ RP2 be any point not on the line P1 P2 (which we consider as the line at infinity). We will
show that f (X) = X.
Let `1 = P3 P1 and `2 = P3 P2 . Since f fixes these
points, it maps `1 to `1 and `2 to `2 . By the lemmas,
the restrictions f |ì : ì → ì are projective transfor-
mations. But f |`1 fixes P1 , P3 , and E1 = P2 P4 ∩ `1 ,
so it is the identity. Equally, f |`2 fixes P2 , P3 , and
E2 = P1 P4 ∩`2 , so it is the identity. Hence f fixes also
X1 = P2 X ∩ `1 and X2 = P1 X ∩ `2 . Since f maps
lines to lines, X = X1 P2 ∩ X2 P1 implies f (X) =
f (X1 )f (P2 ) ∩ f (X2 )f (P1 ) = X1 P2 ∩ X2 P1 = X.
We have shown that f (X) = X for all X not on P1 P2 .
But then it also fixes all points on P1 P2 . (Why?)
Hence, f is the identity.
Localized version of the fundamental theorem. Let U be a subset of RPn that contains
an open ball B ⊂ Rn ⊂ RPn . Suppose an injective map f : U → RPn maps lines to lines in
the following sense: If ` is a line in RPn which intersects U , then there is a line `0 such that
f (` ∩ U ) = `0 ∩ f (U ). Then f is the restriction of a projective transformation of RPn to U .
Again, for simplicity, I will present a proof for the case n = 2 only. This already contains all the
important ideas so you can figure out for yourself how it works for n > 2.
Proof. (for n = 2) Define a map fˆ : RP2 → RP2 as follows. For X ∈ B let fˆ(X) = f (X). If
X 6∈ B, let `1 , `2 be two lines through X that intersect B and let f (X) be the intersection of the
lines `01 and `02 , the images of `1 , `2 under f (in the sense explained in the theorem). This point
is well defined because it does not depend on the choice of `1 and `2 . To see this, use Desargues’
theorem to show that if `1 , `2 , `3 are three lines that intersect B and all go through one point
outside B, then their images under f intersect in one point. You have to convince yourself that
you always have enough room in the open ball to construct (the relevant part of) a Desargues
figure. (See left figure below.)
We have defined fˆ using only information about f on B. In fact, fˆ coincides with f on U .
(Why?) Further, fˆ maps lines to lines. To see this, use (the inverse) Desargues’ theorem to
show that fˆ maps three points on a line to three points on a line. Again you have to convince
yourself that you have enough room in B to construct (the relevant part of) a Desargues figure.
(See right figure below.) Finally, by the fundamental theorem (global version), fˆ is a projective
transformation.
38
Duality
In homogeneous coordinates x1 , x2 , x3 , the equation for a line in a projective plane is
a1 x1 + a2 x2 + a3 x3 = 0,
where not all coefficients ai are zero. The coefficients a1 , a2 , a3 can be seen as homogeneous
coordinates for the line, because if we replace in the equation ai by λai for some λ 6= 0 we get
an equivalent equation for the same line. Thus, the set of lines in a projective plane is itself
a projective plane, the dual plane. Points in the dual plane correspond to lines in the original
plane. Moreover, if we consider in the above equation the xi as fixed and the ai as variables, we
get an equation for a line in the dual plane. Points on this line correspond to lines in the original
plane that contain [x]. Thus, a the points on a line in the dual plane correspond to lines in the
original plane through a point.
It makes sense to look at this phenomenon in a basis independent way and for arbitrary dimen-
sion. It boils down to the duality of vector spaces.
Let V be a finite dimensional vector space over a field F .
The dual vector space V ∗ of V is the vector space of linear functions V → F (linear forms on V ).
If v1 , . . . , vn is a basis of V , the dual basis of V ∗ is ϕ1 , . . . , ϕn with ϕi (vj ) = δij . In particu-
lar dim V = dim V ∗ . But there is no natural way to identify V ∗ with V . (“Natural” means
independent of any arbitrary choices. In this case: choice of a basis.)
There is, however, a natural identification of V with V ∗∗ : A vector v ∈ V is identified with the
linear form V ∗ → F , ϕ 7→ ϕ(v). With this identification, V is also the dual vector space of V ∗ .
Let f : V → W be a linear map. The dual linear map f ∗ : W ∗ → V ∗ is defined by f ∗ (ψ)(v) =
ψ(f (v)). Note that the dual map “goes in the opposite direction”. If f is invertible, then
∗
f ∗ −1 = f −1 is a map V ∗ → W ∗ .
If U ⊆ V is a linear subspace, the annihilator of U is the linear subspace
U 0 = {ϕ ∈ V ∗ | ϕ(v) = 0 for all v ∈ U } ⊆ V ∗
of linear forms that vanish on U .

This provides a correspondence between subspaces of V with subspaces of V ∗ .
The dimensions of U and U 0 are related by
dim U + dim U 0 = dim V.
Indeed, let v1 , . . . , vk be a basis for U and extend it to a basis v1 , . . . , vn of V . Let ϕ1 , . . . , ϕn be

the dual basis of V ∗ . Then (one sees easily that) ϕk+1 , . . . , ϕn is a basis of U 0 .
(In fact, the above dimension formula is just a coordinate free way of saying that each linearly
independent homogeneous equation in the coordinates reduces the dimension of the solution
space by 1.)
If U1 and U2 are subspaces of V , then
(U1 ∩ U2 )0 = U1 0 + U2 0 and (U1 + U2 )0 = U1 0 ∩ U2 0 .
(Can you see this?)

Now let P (V ) be the n-dimensional projective space of an (n + 1)-dimensional vector space V .
The dual projective space is P (V ∗ ).
A point [v] ∈ P (V ) corresponds to the hyperplane P ([v]0 ) ∈ P (V ∗ ), and a point [ϕ] ∈ P (V ∗ )
corresponds to the hyperplane P ([ϕ]0 ) in P (V ). Note that the points of the hyperplane P ([ϕ]0 )
correspond to the hyperplanes in P (V ) that contain [v].
In general, a k-plane P (U ) ⊆ P (V ) corresponds to the plane P (U 0 ) ⊆ P (V ∗ ) of dimension
dim U 0 − 1 = dim V − dim U − 1 = (n + 1) − (k + 1) − 1 = n − k − 1.
The points in P (U 0 ) correspond to the hyperplanes in P (V ) that contain P (U ).
39
Let us take another look at duality for projective planes. (Hyperplanes in a plane are lines.) To
aid the imagination, let us focus on the real projective plane RP2 = P(R3 ) and its dual plane
∗
P(R3 ) which we denote by RP2∗ (although everything holds in general).
So each point in RP2 corresponds to a line in RP2∗ and vice versa. The points on a line in RP2
correspond to the lines through the corresponding point in RP2∗ . Lines through a point in RP2
correspond to the points on the corresponding line in RP2∗ .
Every theorem about RP2 can also be read as a theorem about RP2∗ . This leads to the following
duality principle:
From every theorem that talks only about incidence relations between points and lines in a pro-
jective plane, one obtains another valid theorem by interchanging the words “point” and “line”
(and the phrases “goes through” and “lies on”).
For example, the theorem that is obtained from the Desargues theorem in this way (the dual
Desargues theorem) turns out to be the converse of Desargues’s theorem.
We had seen that the the converse of Desargues is equivalent to Desargues, so Desargues’s
theorem turns out to be self-dual. The same is true for Pappus’s theorem. (Check it out.)
Note that four lines through a point in RP2 correspond to four points on a line in RP2∗ . But for
four points on a line we had defined the cross ratio. Via duality this gives us a definition for the
cross ratio of four lines through a point.
P4
Proposition. Let l1 , l2 , l3 , l4 be four lines `4
through a point P in RP2 . Let l be a line not
P3
containing P and let P1 , P2 , P3 , P4 be the inter- P `3
sections of the four lines li with l. Then P2
`2
P1
cr(l1 , l2 , l3 , l4 ) = cr(P1 , P2 , P3 , P4 ).
` `1
This proposition is an immediate consequence of the following one.
Proposition. Let P be a point in RP2 and let l∗ be the corresponding line in RP2∗ , so that each
point of l∗ corresponds to a line through P . Let l be a line in RP2 that does not contain P . Then
the map l∗ → l that maps a point of l∗ to the intersection of the corresponding line with l is a
projective transformation.
Proof. Let P = [v1 ], and let [v2 ], [v3 ] be two points on l. Then v1 , v2 , v3 is a basis of R3 . Let
∗
ϕ1 , ϕ2 , ϕ3 be the dual basis of R3 . The line l∗ is spanned by [ϕ2 ], [ϕ3 ]. Hence the points [ϕ] ∈ l∗
have representative vectors ϕ = sϕ2 + tϕ3 , and s, t are homogeneous coordinates on l∗ . The line
in RP2 corresponding to [ϕ] intersects l in a point [v] such that v = xv2 + yv3 and
0 = ϕ(v) = (sϕ2 + tϕ3 )(xv2 + yv3 ) = sx + ty.
This is the case for x = t, y = −s. So the map l∗ → l in question comes from the linear map
sϕ2 + tϕ3 7→ tv2 − sv3 .
40
Conic sections
The Euclidean point of view
The conic sections are ellipses (including circles), parabolas, hyperbolas, and the degenerate
cases of a pair of lines, which may degenerate further to one “double” line, and a single point.
ellipse hyperbola parabola

y y y
b
r2 b
r1 r2 r1
−a x x
−f f a −f −a a f r1
f r2
x
−b −f y = −f
x 2 y 2 x 2 y 2
+ = 1. − = 1. y = ax2
a b a b
r1 + r2 = const. r1 = r2
|r1 − r2 | = const.
√ √ f= 1
f = a2 − b2 f = a2 + b2 4a
two intersecting lines two parallel lines one “double” line

y
y
b a
a x x
−a
x 2 y 2
− =0 y2 = a y2 = 0
a b
What do these curves have in common?

1. They arise as intersection of a plane with a cone (or cylinder in the case of two parallel lines).
Hence the name.
The figure on the right illustrates Dandelin’s proof for the case of an ellipse.
2. They are all described by quadratic equations in the two Euclidean coordinates. In fact:
Theorem. The set of solutions of any quadratic equation in two variables u, v,
au2 + 2buv + cv 2 + du + ev + f = 0 (∗)
is empty or a conic section.

More precisely: There is a change of coordinates ( uv ) = A( xy ) + t with A ∈ O(2), t ∈ R2 which
reduces (∗) to one of the standard forms above. Do you still know how to prove this?
41
Optical properties of the conic sections
Here are two proofs for the case of an ellipse:
Proof #1 (more analytic). Let γ : R → R2 be a parameterization

of an ellipse, like for example γ(t) = ( ab cos t
sin t ). The direction of
0
the tangent at a point γ(t) is γ (t) and
hγ 0 , γ − F1 i hγ 0 , γ − F2 i
cos(α1 ) = , cos(α2 ) = − .
kγ 0 k kγ − F1 k kγ 0 k kγ − F2 k
Now kγ − F1 k + kγ − F2 k = const. implies
0 = (kγ − F1 k + kγ − F2 k)0
hγ 0 , γ − F1 i hγ 0 , γ − F2 i
= +
kγ 0 k kγ − F1 k kγ 0 k kγ − F2 k
= cos α1 − cos α2 .
Proof #2 (more synthetic). Let P be a point of the ellipse. Ex-

tend the line segment F2 P a distance of r1 beyond P . Call the
new endpoint of the extended segment F20 . Claim: The tangent
of the ellipse at P is the perpendicular bisector ` of F1 F20 . Indeed,
P lies on ` because it has equal distance r1 from F1 and F20 . Con-
sider any other point P̃ on ` and let r̃1 be its distance to both F1
and F20 and let r̃2 be its distance to F2 . Then r̃1 + r̃2 > r1 + r2
so P̃ does not lie on the ellipse. Hence, ` intersects the ellipse in
precisely one point, P , which proves the claim. Now the equality
of the angles follows easily.
The advantage of the second proof is that it suggests another theorem:

Theorem. Let c be a circle with center F2 and let F1 be a point inside c. The locus of the centers
of all circles that go through F1 and touch c is an ellipse with foci F1 and F2 .
42
The projective point of view
Five points determine a conic section
Pencils of conic sections
Pascal’s theorem
to be completed
43
Inside and outside
The pole-polar relationship
The dual conic and Brianchon’s theorem
to be completed
44
The rational paramteterization of conics
Steiner’s projective generation of conic sections
to be completed
45
Quadrics
Quadrics are the generalization of conic sections to arbitrary dimension: They are the sets defined
by one quadratic equation in the coordinates. Conic sections are the special case of quadrics in
the plane.
The Euclidean point of view
A quadric in the Euclidean space Rn is defined by a quadratic equation

xT Ax + bT x + c = 0, with A a symmetric n × n matrix, b ∈ Rn , c ∈ R.
This can be brought to normal form by a Euclidean motion x 7→ M x + v, with M ∈ O(n, R),
v ∈ Rn . In R3 , the following cases can occur:1
ellipsoid elliptic paraboloid 2-sheeted hyperboloid
( xa )2 + ( yb )2 + ( zc )2 = 1 z = ( xa )2 + ( yb )2 ( xa )2 + ( yb )2 − ( zc )2 = −1
1-sheeted hyperboloid hyperbolic paraboloid plus some degenerate cases:
• cones and cylinders over a conic

• two planes
• one “double” plane
• one line
• one point
• the empty set.
( xa )2 + ( yb )2 − ( zc )2 = 1 z = ( xa )2 − ( yb )2
The projective point of view
If q is quadratic form on a vector space V , q 6= 0, then

Q = {[v] ∈ P(V ) | q(v) = 0}
is called a quadric in P(V ). Since Q is defined by a quadratic polynomial in the homogeneous
coordinates, the algebraic properties of the base field of scalars of V play an important role. We
will only consider R and comment on C.
In RP3 , there are only three non-degenerate cases depending on the signature of q:
(0) (+ + ++) or (− − −−), the case of definite q leading to an empty quadric Q = ∅. We exclude
this case from now on.
(1) (+ + +−) or (+ − −−). In an affine image of RP3 , Q looks like an ellipsoid, an elliptic
paraboloid, or a 2-sheeted hyperboloid, depending on whether the plane at infinity does not
intersect, is tangent to, or intersects the quadric (without being tangent).
(2) (+ + −−). In an affine image of RP3 , Q looks like a 1-sheeted hyperboloid or a hyperbolic
paraboloid, depending on whether the plane at infinity intersects (without being tangent) or
is tangent to Q. (In this case, any plane meets Q.)
If q is degenerate, let U0 = ker q, and let U1 be a complementary subspace. Then Q is the union
of all lines joining a point in P(U0 ) with a point in the non-degenerate quadric in P(U1 ) defined
by the restriction q|U1 .
In CPn , there is up to projective transformations only one non-degenerate quadric. There are n
degenerate ones, depending on the rank of q (which can be 1, . . . , n).
1 The three images in the first row are taken from Wikipedia.
46
Proposition. If Q is a (non-empty) non-degenerate quadric in RPn , then its defining bilinear

form is uniquely determined up to a scalar factor.
Proof. Suppose q and q̃ determine the same quadric. Hence q(v, v) = 0 ⇔ q̃(v, v) = 0. We
want to show q̃ = λq. Suppose q has signature (k, n + 1 − k), where 1 ≤ k ≤ n + 1, and let
e1 , . . . , ek , f1 , . . . , fn+1−k be an orthonormal basis for q (with q(ei , ei ) = 1 and q(fm , fm ) = −1).
We are done if we have shown that q̃(ei , ej ) = λδij , q̃(fm , fl ) = −λδml , and q̃(ei , fm ) = 0. To
see this, first note that for any i = 1, . . . , k and m = 1, . . . , n + 1 − k,
q(ei ± fm , ei ± fm ) = q(ei , ei ) ± 2q(ei , fm ) + q(fm , fm ) = 0,
so
0 = q̃(ei ± fm , ei ± fm ) = q̃(ei , ei ) ± 2q̃(ei , fm ) + q̃(fm , fm )
This implies q̃(ei , fm ) = 0 and q̃(ei , ei ) = −q̃(fm , fm ) = λ for some λ ∈ R independent of i, m.
To see that q̃(ei , ej ) = 0 for i 6= j and q̃(fm , fl ) = 0 for m 6= l, make a similar argument starting
with q(v, v) = 0 for v = (ei ± ej ) + (fm ± fl ) (here the signs may be chosen independently).
Inside and outside

Any (non-empty) non-degenerate quadric divides RPn into two connected components. If the
signature is (n, 1), then inside and outside can be defined in the same way as for conics: A point
is inside iff every line through it intersects the quadric in two points.
Question: How can the two components be distinguished in general if the signature is (k, m)
with k 6= m?
If the signature is neutral (k = m), the two components can be interchanged by a projective
transformation.
Lines in a quadric
A line intersects a quadric in RPn either not at all, in two points, in one point, or it lies entirely
in the quadric. In the last two cases, the line is called a tangent. In RP3 , the only non-degenerate
quadrics that contain lines are the ones with neutral signature (+ + −−).
Proposition. Let Q be a quadric in RP3 with neutral signature. Through any point in Q there
are precisely two lines lying entirely in Q.
Proof. To see that there are no more than two lines through a point [p] ∈ Q lying entirely in
Q, show that any such line must lie in the plane q(p, · ) = 0, and note that the intersection of Q
with a plane is a conic section, so it cannot contain more that two lines.
To see that there are actually two such lines, we may assume (after a change of coordinates,
if necessary) that Q is the quadric x21 + x22 − x23 − x24 = 0. This equation is equivalent to
(x1 + x3 )(x1 − x3 ) + (x2 + x4 )(x2 − x4 ) = 0, and, after changing to new coordinates
y1 = x1 + x3 , y2 = x1 − x3 , y3 = −(x2 + x4 ), y4 = x2 − x4 ,
to
y1 y2 − y3 y4 = 0.
Now the map y1 s1 s2
s1 s2 y2 t1 t2
f: RP1 × RP1 −→ Q, t1 , t2 −
7 → y3 = s1 t2
y4 t1 s2
y1
y s s1 y1
is actually a bijection RP1 × RP1 ↔ Q. Indeed, if y23 ∈ Q, then t11 is determined by t1 = y4
y4
or by st11 = yy32 . (It can happen that one of the right hand sides is 00 , but not both. If neither
is 00 , they are equal.) Similarly, st22 is determined by st22 = yy13 or by st22 = yy42 . For any point

P = f (P1 , P2 ) ∈ Q, the images of the functions f (P1 , · ) : RP1 → Q and f ( · , P2 ) : RP1 → Q are
two lines through P lying entirely in Q.
In fact this proof also shows:
• Q contains two families of pairwise skew lines, and each line of the first family intersects each
line of the second family.
• Since RP1 is homeomorphic to the circle S 1 , Q is homeomorphic to S 1 ×S 1 , so it is topologically
a torus.
47
Polarity
A non-degenerate symmetric bilinear form q on a vector space V defines a relation between the
points and hyperplanes of P(V ): To each point [v] ∈ P(V ) corresponds the polar hyperplane
{[w] ∈ P(V ) | q(v, w) = 0},
and to each hyperplane there is a corresponding point, its pole. Note that
[x] ∈ polar hyperplane of [y] ⇐⇒ [y] ∈ polar hyperplane of [x] ⇐⇒ q(x, y) = 0.
More generally, let U ⊆ V be a (k + 1)-dimensional linear subspace of V , and let n + 1 = dim V .

The orthogonal subspace of U (with respect to q) is
U ⊥ = {w ∈ V | q(u, w) = 0 for all u ∈ U }.
The dimension of U ⊥ is dim V − dim U = n − k, and U ⊥⊥ = U . The k-plane P(U ) and the
(n − k − 1)-plane P(U )⊥ in P(V ) are called polar to each other. Polarity (with respect to q)
is therefore a one-to-one relation between k-planes and (n − k − 1)-planes in the n-dimensional
projective space P (V ). In particular, if n = 3, polarity is a relation between points and planes
and between lines and lines.
Proposition. Let Q be a (non-empty) non-degenerate quadric in RPn (CPn ) defined by the
symmetric bilinear form q, and let X ∈ Q, Y ∈ RPn (CPn ). Then
The line XY is tangent to Q ⇐⇒ X is in the polar hyperplane of Y .
Proof. Let X = [x], Y = [y]. Then q(x, x) = 0 because X ∈ Q. The line XY is tangent to Q
either if it intersects Q in no other point but X, or if it is contained entirely in Q. The points
on the line XY except X are parameterized by [tx + y] with t ∈ R (C). Such a point lies in Q if
0 = q(tx + y, tx + y) = t2 q(x, x) + 2tq(x, y) + q(y, y) = 2tq(x, y) + q(y, y).
This equation for t has one solution if q(x, y) 6= 0, it has no solution if q(x, y) = 0 and q(y, y) 6= 0,
and it is satisfied for all t if q(x, y) = q(y, y) = 0. So the line XY contains no other points of Q
except X or lies entirely in Q precisely if q(x, y) = 0.
This provides a simple geometric interpretation of the polarity relationship between points and
hyperplanes in the case when the polar hyperplane intersects Q: The tangents from a point to
the quadric touch the quadric in the points in which the quadric intersects the polar hyperplane.
If a quadric in RP3 is illuminated by a point light source
outside the quadric (or by parallel light), the borderline
between light and shadow on the quadric is a conic in
the polar plane; and the shadow that the quadric throws
on some other another plane is a projected image of this
conic.
What about the polarity between lines in RP3 ? If a
point moves on a line, the polar planes rotate about a
line, and these two lines are polar to each other.
Projective transformations that map a quadric to itself
Let Q be a (non-empty) non-degenerate quadric in RPn defined by the symmetric bilinear form q
with signature (k, n+1−k). If f : Rn+1 → Rn+1 is a linear map which is orthogonal with respect
to q (that is, q(x, y) = q(f (x), f (y)) for all x, y ∈ Rn+1 ), then the projective map [x] 7→ [f (x)]
clearly maps Q to Q. If the signature is not neutral (that is, if k 6= n + 1 − k), then these are all
projective maps that map Q to Q:
Proposition. If the signature is not neutral, then any projective transformation that maps Q to
Q comes from a linear map which is orthogonal with respect to q.
Hence, under the assumption of non-neutral signature, the group of projective transformations
mapping Q to Q is P O(k, n + 1 − k), the projective orthogonal group for signature (k, n + 1 − k).
48
Proof. Suppose [x] 7→ [f (x)] maps Q to Q. This means that the symmetric bilinear forms q and
q̃ defined by q̃(x, y) = q(f (x), f (y)) define the same quadric. In the last lecture we saw that this
means q̃ = λq for some λ ∈ R \ {0}. Hence q(f (x), f (y)) = λq(x, y) for all x, y ∈ R. We will
show that λ is positive. Then √1λ f defines the same projective transformation and is orthogonal
with respect to q. Now to see that λ is positive, let e1 , . . . , en+1 be an orthonormal basis with
respect to q. Then f (e1 ), . . . , f (en+1 ) is still an orthogonal basis. If λ were negative, it would
contain n + 1 − k spacelike and k timelike vectors. This cannot be, because every orthogonal
basis contains k spacelike and n + 1 − k timelike vectors
To see that the non-neutral signature assumption is really necessary,
x1 consider
x3 the quadric Q
defined by x1 + x2 − x3 − x4 = 0. The projective transformation x3 7→ xx41 maps Q to Q,
2 2 2 2 x2
x4 x2
but it does not come from an orthogonal map in O(2, 2). Note that it also interchanges the two
connected components of RP3 \Q. The following is true in general: If a projective transformation
maps a non-degenerate quadric to itself and each connected component of the complement of
Q to itself, then it comes from an orthogonal transformation. (In the case of neutral signature,
these form only a subgroup of index 2 within the group of all projective transformations mapping
Q to Q).
The projective model of hyperbolic space

We had defined n-dimensional hyperbolic space in the hyperboloid model as
H n = {x ∈ Rn+1 | hx, xi = −1 and xn+1 > 0} where hx, yi = x1 y1 + · · · + xn yn − xn+1 yn+1 ,
equipped with the metric d defined by cosh d(x, y) = −hx, yi.
n
In the projective model, n-dimensional hyperbolic space consists of the set Hproj of points inside
n
a quadric Q ⊂ RP defined by a symmetric bilinear form q of signature (n, 1),
n
Hproj = {[x] ∈ RPn | q(x, x) < 0},
equipped with the metric dproj defined by
|q(x, y)|
cosh dproj ([x], [y]) = p .
q(x, x)q(y, y)
This is indeed a metric space isometric to H n . First of all, if we use in RPn homogeneous
coordinates with respect to an orthonormal basis for q, then q = h · , · i, and the map
H n −→ Hproj
n
, x 7−→ [x]
is bijective and an isometry.
The projective model is actually the same as the Klein model (seen from the projective point of
view, which was how Klein saw it in the first place): The points inside the unit circle—or unit
sphere in higher dimension—are obviously the points inside a quadric of the right signature, and
from the projective point of view we might as well take any other such quadric.
n
The group of isometries of Hproj is the projective orthogonal group
P O(n, 1) of projective transformations mapping Q to Q (and hence
also the inside of Q to the inside of Q).
Via the polarity relation, points outside Q correspond to hyperplanes
n
in Hproj . Two hyperplanes intersect orthogonally if one contains the
pole of the other. (Do you see why? What is the formula for the
angle between two intersecting hyperplanes polar to two points [v]
and [w] outside Q? What is the formula for their distance if they
do not intersect?)
The following proposition is due to Arthur Cayley. (Actually, its history is usually told like this:
Cayley discovered this way of defining a metric inside a quadric in terms of a cross ratio. Then
Klein realized that this was a model for the non-Euclidean geometry discovered by Lobachevsky.)
n
Proposition. For two points A, B ∈ Hproj , let X, Y be the points
where the line AB intersects the quadric Q, labelled such that X, B
separates A, Y . Then
1
dproj (A, B) = 2 log cr(B, X, A, Y ).
49
Proof. First convince yourself that cr(B, X, A, Y ) > 1, so that the logarithm is positive (see
Lectures 17 and 18). We will prove the proposition by showing that
|q(a, b)|
cosh( 12 log cr(B, X, A, Y )) = p .
q(a, a)q(b, b)
Suppose A = [a] and B = [b], and introduce an affine parameter on the line AB by t 7→ [a + tb].
The points A and B correspond to the parameter values t = 0 and t = ∞. The parameter values
for X and Y are the roots t1,2 of the quadratic equation
0 = q(a + tb, a + tb) = q(a, a) + 2q(a, b)t + q(b, b)t2 .

(∞−t1 )(0−t2 ) t2
On the one hand, cr(B, X, A, Y ) = (t1 −0)(t2 −∞) = t1 , so
t2 t2
1 1
q q
1
log cr(B, X, A, Y ) = 21 e 2 log t1 + e− 2 log t1 = 1 t2 t

cosh 2 2 t1 + 1
t2 .
On the other hand, q(a, a) + 2q(a, b)t + q(b, b)t2 = q(b, b)(t − t1 )(t − t2 ) implies
2q(a, b) q(a, a)
t1 + t2 = − and t1 t2 = .
q(b, b) q(b, b)
Since t1 and t2 have the same sign, and so do q(a, a) and q(b, b), we get
1
q
t2
q
t |t1 + t2 | |q(a, b)|
2 t1 + 1
t2 = √ =p .
2 t1 t2 q(a, a)q(b, b)
Klein’s Erlangen program

For his inaugural lecture at the university of Erlangen in 1872, Felix Klein prepared a paper
dealing with the question “What is geometry? ”. In it, he breaks radically with the traditional
point of view that geometry is the study of the “true” space around us. Instead, geometry should
be viewed as
the study of invariants under a group of transformations.
Actually this was not meant as an abstract formal definition of the field of geometry. On the
contrary, Klein saw a “joy in the pure form or shape” as the characteristic mark of a geometer,
and he always emphasized the importance of space intuition. The point of the Erlangen program
was to provide an organizing principle for the overabundant material that had accumulated in
geometry, or rather, in the different geometries that had been discovered.
Let us see how some familiar geometries fit into this scheme. Every particular geometry deals with
properties of figures in some space that remain invariant under some group of transformations
of the space:
geometry space transformation group some invariant properties & quantities
Euclidean Rn the group of Euclidean motions distance of two points, angle of two lines
x 7→ Ax + v, A ∈ O(n), b ∈ Rn
projective RPn the group of projective transfor- the cross ratio, being a k-plane, being a
mations, P GL(n + 1) quadric, incidence relations
n
hyperbolic Hproj P O(n, 1) hyperbolic distances and angles
n
spherical S O(n + 1) spherical distances and angles
similarity Rn the group of similarity transfor- angles, ratios of distances
mations x 7→ λAx + v, λ ∈ R>0 ,
A ∈ O(n), v ∈ Rn
affine Rn the group of affine transforma- parallelism, ratios of distances between
tions x 7→ Ax + v, A ∈ GL(n), points on a line
v ∈ Rn
Actually, one does not distinguish between two geometries if there is a bijection between the
spaces (or at least between subsets of them) which induces an isomorphism of the transformation
50
groups. Instead, one then speaks of different models for the same geometry. For example,
we have considered different models for hyperbolic geometry: The points inside a quadric of
signature (n, 1) are in 1-to-1 correspondence with the points of the upper sheet of a 2-sheeted
hyperboloid, and the projective transformations that fix the quadric correspond to the orthogonal
transformations of Rn,1 that map each sheet to itself. We have also seen the conformal Poincaré
models, and we will soon discuss what the corresponding transformation groups are.
Elliptic geometry is the geometry on the sphere but pairs of opposite points are considered as
one point. So the space is S n with opposite points identified and the group is O(n + 1)/{±I}.
(Both I and −I fix every point of this space.) Another model for elliptic geometry is RPn with
group P O(n + 1), the group of projective transformations that fix a definite symmetric bilinear
form. The advantage of elliptic geometry over spherical geometry is that two lines intersect in
one point, and there is a unique line through every two points.
Klein’s Erlangen program emphasizes the transformation projective
MMM
group rather than the space on which it acts. If the
qqqqq MMM
transformation group of one geometry is a subgroup of x q &
hyperbolic elliptic affine
the transformation group of another geometry, the first
is called a subgeometry of the second. For example, the
space Rn of Euclidean geometry can be seen as a sub- similarity
set of RPn —the complement of a particular hyperplane
xn+1 = 0 which is considered “at infinity”. The group Euclidean
of Euclidean motions then corresponds
to the group of
projective transformations A0 v1 with A ∈ O(n). Euclidean geometry is thus a subgeometry
of projective geometry. The diagram illustrates the subgeometry relationship for some familiar
geometries.
The invariants of one geometry are also invariants of any of its subgeometries
(because the group is smaller). The same is true for theorems. Every theorem of
projective geometry is also a theorem of Euclidean geometry. The converse is not
true, but often one can see a Euclidean theorem as special case of a projective
theorem. For example, the theorem on the inscribed angle over a chord of a circle
is a special case of Steiner’s theorem on the projective generation of conics.
Klein’s group theoretical point of view on geometry illuminates the relationship between the
different geometries. This is not only of theoretical interest, it is a great practical help in the
day-to-day business of geometric research. When confronted with a geometric problem, it is
usually an excellent idea to ask first: In which geometry should I treat this problem?
Let me illustrate this with a real-live example. A couple of months
ago, Wolfgang Schief told me about the following striking theorem
which he had discovered. It had somehow come up in his research
on integrable systems and he had good reason to believe it was true.
Theorem (W. K. Schief). Consider three pairs a1 , a2 ; b1 , b2 ; c1 , c2
of lines in the plane. If the four intersection points ai ∩ bj lie on a
circle and the four intersection points bi ∩ cj lie on a circle, then the
four intersection points ci ∩ aj also lie on a circle.
How to prove this? It looks like a theorem of Euclidean geometry, or
more precisely, of similarity geometry. But it can also be interpreted
in terms of projective geometry and this leads to a surprisingly sim-
ple proof. Pairs of lines and circles are all conic sections. If we consider the Euclidean plane as
the complement in RP2 of the line x3 = 0, then the circles are characterized among the (non-
empty) non-degenerate quadrics a11 x21 + 2a12 x1 x2 + 2a13 x1 x3 + a22 x22 + 2a23 x2 x3 + a33 x23 = 0 by
the homogeneous linear equations in the coefficients a11 − a22 = 0 and a12 = 0. (A circle with
center (c1 , c2 ) ∈ R2 and radius r has equation x21 + x22 − 2c1 x1 x3 − 2c2 x2 x3 + (c21 + c22 − r2 )x23 = 0.)
Recall that a conic corresponds to a point in the 5-dimensional projective space of the vector
space of quadratic forms. So a non-degenerate conic is a circle if is lies in a particular 3-plane in
that projective space. Now the three pairs of lines correspond to points, and the three quadru-
ples of intersection points correspond to pencils of non-degenerate conics, that is, to three lines
connecting these points. If the 3-plane of circles intersects two of these lines, it also intersects
the third. Thus the theorem is reduced to the fact that if a line in RPn intersects two sides of a
triangle it also intersects the third.
51
Möbius geometry
Elementary model
Pn
Consider Rn with the standard Euclidean scalar product hx, yi = 1 xi yi .
Reflection in a hyperplane {x : hx − a, vi = 0} is the map
hx − a, vi
x 7−→ x0 = x − 2 v.
hv, vi
Reflection (or inversion) in a hypersphere with center c and radius r is the

map
r2
x 7−→ x0 = c + (x − c).
kx − ck2
Note that x0 lies on the same ray emanating from c and kx−ck·kx0 −ck = r2 .
Inversion in a sphere is an involution, except that the center c has no image
and no preimage. We fix this by adding one extra point, ∞, to Rn and
we declare it to be the image and preimage of c. We also declare that
reflections in hyperplanes map ∞ to ∞. Then both kinds of reflections are involutions on
Rn ∪ {∞}.
A Möbius transformation of Rn ∪ {∞} is a composition of reflections in hyperplanes and hyper-
spheres. The Möbius transformations form a group called the Möbius group denoted by Möb(n).
A Möbius transformation is orientation reversing or preserving depending on whether it is the
composition of an odd or even number of reflections. The subgroup of orientation preserving
Möbius transformations is called the special Möbius group and denoted by SMöb(n) or Möb+ (n).
The Möbius group contains all similarity transformations:
• A translation x 7→ x + v is the composition of two reflections in parallel hyperplanes.
• An orthogonal transformation x 7→ Ax with A ∈ O(n) is the composition of at most n reflec-
tions in hyperplanes through the origin. (Can you prove this?)
• A scaling transformation x 7→ λx with λ > 0 is the composition of √ a reflection in the unit
sphere followed by a reflection in a sphere with center 0 and radius λ. (Check this.)
Proposition. A Möbius transformation maps any hyperplane or hypersphere to a hyperplane or
hypersphere.
Proof. This is true for all similarity transformations. (These map hyperplanes to hyperplanes
and hyperspheres to hyperspheres.) A reflection in a sphere with center c and radius r is the same
as the similarity transformation x 7→ 1r (x − c) mapping the sphere to the unit sphere, followed
by inversion in the unit sphere, followed by the inverse similarity transformation x 7→ rx + c.
(Check this.) So it remains to show that inversion in the unit sphere maps spheres and planes
to spheres and planes. One could consider hyperspheres and hyperplanes separately but we will
treat both cases simultaneously. Any hypersphere or hyperplane is determined by an equation
of the form
pkxk2 − 2hv, xi + q = 0 with kvk2 − pq > 0.
If p = 0, the inequality implies v 6= 0 so the equation describes a hyperplane. If p 6= 0, it describes
a hypersphere. Indeed, divide through by p to obtain
q
0 = kxk2 − 2h p1 v, xi + p = kx − p1 vk2 − 1
p2 kvk
2
+ pq .
q
q
This is a sphere with center p1 v and radius 1
p2 kvk
2 − p . (The assumed inequality ensures that
the expression under the square root is positive.) Now for x0 = 1
kxk2 x one obtains
pkx0 k2 − 2hv, x0 i + q = 0 ⇐⇒ qkxk2 − 2hv, xi + p = 0.
So x0 is contained in a particular hyperplane or hypersphere if and only if x is contained in some

other hyperplane or hypersphere.
52
From now on we will consider hyperplanes as a special cases of hyperspheres that contain ∞. So
hypersphere will mean hypersphere or hyperplane.
Proposition. Any bijective map f : Rn ∪ {∞} → Rn ∪ {∞} which maps hyperspheres to hyper-
spheres is a Möbius transformation.
Proof. (i) Suppose f (∞) = ∞. Then f maps hyperplanes to hyperplanes. Then it also maps
lines to lines, because a line is the intersection of n −1 hyperplanes. By the fundamental theorem
of projective geometry (or rather the corollary of it, see Lecture 19), the restriction f |Rn is an
affine transformation. Since it also maps spheres to spheres it must be a similarity.
(ii) Suppose f (∞) = c 6= ∞. Let g be the inversion in a sphere with center c. Then g ◦ f also
maps hyperspheres to hyperspheres and also ∞ to ∞. By (i) it is a similarity transformation,
so f = g ◦ g ◦ f is a Möbius transformation.
Proposition. The Möbius transformations are conformal.
Proof. Since the similarity transformations are conformal it remains only to show that inversion
in the unit sphere is conformal. Let t 7→ γ(t), t 7→ η(t) be two parameterized curves intersecting
in γ(t0 ) = η(t0 ). The intersection angle α is determined by
hγ 0 (t0 ), η 0 (t0 )i
cos α = .
kγ 0 (t0 )kkη 0 (t0 )k
1 1
Let γ̂ = hγ,γi γ, η̂ = hη,ηi η, be the image curves after inversion in the unit sphere. One finds
that
1
γ̂ 0 = hγ, γiγ 0 − 2hγ, γ 0 iγ ,

hγ, γi2
and similarly for η̂ 0 . From this one obtains hγ̂ 0 , γ̂ 0 i = hγ,γi
1 0 0 0
2 hγ , γ i, so kγ̂ k =
1 0
kγk2 kγ k, and in
the same way kη̂ 0 k = kηk1 0
2 kη k. Using γ(t0 ) = η(t0 ) =: p one finds that
1
hγ̂ 0 (t0 ), η̂ 0 (t0 )i = hγ 0 (t0 ), η 0 (t0 )i
kpk4
and hence
hγ 0 (t0 ), η 0 (t0 )i hγ̂ 0 (t0 ), η̂ 0 (t0 )i
0 0
= 0
kγ (t0 )kkη (t0 )k kγ̂ (t0 )kkη̂ 0 (t0 )k
Two-dimensional Möbius geometry

This case is special because we can identify R2 with C, and R2 ∪ {∞} with the extended complex
plane Cb = C ∪ {∞}, which is the same as CP1 , the complex projective line (see Lecture 13). The
orientation preserving and reversing similarity transformations are z 7→ az + b and z 7→ az̄ + b
(a 6= 0), reflection in the real line is z 7→ z̄, and inversion in the unit circle |z| = 1 is the map
z 7→ |z|z 2 = z̄1 .
b = CP1 are
Proposition. The orientation preserving and reversing Möbius transformations of C
precisely the maps of the form
az + b az̄ + b a b

z 7−→ and z 7−→ with det c d = ad − bc 6= 0.
cz + d cz̄ + d
Proof. First, these transformations form a group: The transformations of the first kind are the
projective transformations of CP1 , and the transformations of the second kind are compositions
of these with complex conjugation z 7→ z̄. (Note that first performing a transformation of the
first kind and then complex conjugation also leads to a transformation of the second kind.) This
groups contains the similarity transformations and inversion in the unit sphere, so it contains the
Möbius group. On the other hand, it is not bigger than the Möbius group, because any of these
transformations is a composition of reflections and similarity transformations: If c = 0, they are
just similarity transformations. Otherwise, this follows from
az + b a bc − ad
= +
cz + d c c(cz + d)
and the equation obtained by replacing z by z̄.
53
Two-dimensional Möbius geometry (continued)

We have identified the extended real plane R2 ∪ {∞} with the complex projective line CP1 =
Cb = C2 ∪ {∞}, and we have seen that the orientation preserving Möbius transformations are
the complex projective transformations of CP1 : SMöb(2) = PGL(2, C). This has the following
immediate consequences:
• The orientation preserving Möbius transformations of the plane preserve the complex cross
1 −z2 )(z3 −z4 )
ratio cr(z1 , z2 , z3 , z4 ) = (z
(z2 −z3 )(z4 −z1 ) . If f is an orientation reversing Möbius transformation,
then cr(f (z1 ), f (z2 ), f (z3 ), f (z4 )) = cr(z1 , z2 , z3 , z4 ).
• For any three points z1 , z2 , z3 and any three points w1 , w2 , w3 , there is a unique orientation
preserving Möbius transformation f ∈ PGL(2, C) with f (zi ) = wi . There is also a unique
orientation reversing one mapping zi 7→ wi , namely f followed by an inversion in the circle
through w1 , w2 , w3 .
• Four points z1 , z2 , z3 , z4 lie on a circle if their cross ratio is real. Moreover, they are in that
cyclic order on the circle if cr(z1 , z2 , z3 , z4 ) < 0.
When we talked about the Poincaré disk and half-plane models of the hyperbolic plane, we did
not say what the corresponding hyperbolic isometries were. We can do so now:
• The isometries of the hyperbolic plane in the Poincaré disk model are the Möbius transforma-
tions that map D2 → D2 . These are the maps
az + b az̄ + b
z 7→ and z 7→ with a, b ∈ C and |a|2 − |b|2 > 0.
b̄z + ā b̄z̄ + ā
• The isometries of the hyperbolic plane in the half-plane model are the Möbius transformations
that map the upper half-plane Im z > 0 to the upper half-plane. These are the maps
az + b az̄ + b
z 7→ with ad − bc > 0 and z 7→ with ad − bc < 0,
cz + b cz̄ + b
and a, b, c, d ∈ R in both cases.
• The Poincaré disk and half-plane models are related by the map D2 → {z ∈ C | Im z > 0},
z+i
z 7→ iz+1 . Indeed, this maps 0 7→ i, 1 7→ 1, i 7→ ∞, −1 7→ −1, −i 7→ 0. So it maps the circle
through 1, i, −1, −i (the unit circle) to the circle through 1, ∞, −1, 0 (the real line); and since
it maps 0 to i, it maps the inside of the unit circle to the upper half-plane.
The projective model of Möbius geometry

Stereographic projection maps Rn ∪{∞} to the n-dimensional sphere S n ⊂ Rn+1 ⊂ RPn+1 . (The
point ∞ is mapped to the north pole.) Hyperspheres in Rn ∪ {∞} are mapped to hyperspheres
in S n , which are intersections of S n with hyperplanes of RPn+1 . The Möbius transformations
of Rn ∪ {∞} are characterized by the property that they map hyperspheres to hyperspheres.
Hence they correspond to those transformations of S n that map intersections with hyperplanes
to intersections with hyperplanes. The projective transformations of RPn+1 that map S n → S n
have this property. Hence P O(n + 1, 1) ⊂ Möb(n). In fact, these are all Möbius transformations:
P O(n + 1, 1) = Möb(n). To see this let us examine which transformations of S n correspond
to the scalings, translations and orthogonal transformation of Rn and to inversion in the unit
sphere.
First, S n ⊂ RPn+1 is
Pn+1
S n = {[x] ∈ RPn+1 | x ∈ Rn+2 , hx, xi = 0}, where hx, yi = 1 xi yi − xn+2 yn+2 .
In 0

So we can write hx, yi = xT Ey, where E = 0 −1 and In is the n × n identity matrix. So
hAx, Ayi = xT AT EAx and hence
A ∈ O(n + 1, 1) ⇐⇒ AT EA = E ⇐⇒ A−1 = EAT E.
Stereographic projection Rn → S n is
 
2u
1 2u in hom.
u 7−→ ==== (u, u) − 1
(u, u) + 1 (u, u) − 1 coords
(u, u) + 1
where ( · , · ) is the standard Euclidean scalar product of Rn .
54
• Orthogonal transformations u 7→ M u, M ∈ O(n). Stereographic projection maps

2M u M 00 2u M 00
M u 7−→ (M u,M u)−1 = 0 10 (u,u)−1 and 0 10 ∈ O(n + 1, 1).
(M u,M u)+1 0 01 (u,u)+1 0 01
• Scalings u 7→ λu. Stereographic projection maps

2λu 2u
In 0 0

1
2u
λu 7−→ λ22 (u,u)−1 = λ(u,u)− λ1 = 0 1 1
2 (λ+ λ )
1 1
2 (λ− λ ) (u,u)−1 .
λ(u,u)+ λ 1 1 1 1 (u,u)+1
λ (u,u)+1 0 2 (λ− λ) 2 (λ+ λ)
| {z }
−1 S(λ)
To see that S(λ) ∈ O(n + 1, 1), check that S (λ) = S( λ1 ) and
= ES T (λ)E. S( λ1 )
• Translations u 7→ u + v. Stereographic projection maps
2u+2v
In −v v
2u

u + v 7−→ (u,u)+2(u,v)+(v,v)−1 = v T 1− 12 (v,v) 12 (v,v) (u,u)−1 .
(u,u)+2(u,v)+(v,v)+1 v T − 21 (v,v) 1+ 12 (v,v) (u,u)+1
| {z }
T (v)
To see that T (v) ∈ O(n + 1, 1), check that T −1 (v) = T (−v) and T (−v) = ET T (v)E.
1
• Inversion in unit sphere u 7→ (u,u) u. Stereographic projection maps
2
u
1 (u,u)
1
2u In 0 0 2u In 0 0
u 7−→ (u,u)
−1 = 1−(u,u) = 0 −1 0 (u,u)−1 , and 0 −1 0 ∈ O(n + 1, 1).
(u, u) 1
+1 1+(u,u) 0 0 1 (u,u)+1 0 0 1
(u,u)
Thus we have the correspondences:

elementary model projective model
R ∪ {∞} ←→ S n ⊂ RPn+1
n
Möb(n) ←→ P O(n + 1, 1)
polarity
hypersphere ⊂ Rn ∪ {∞} ←→ hyperplane ⊂ RPn+1 intersecting S n ←→ point outside S n
Hyperspheres (including hyperplanes) in Rn ∪ {∞} correspond to points outside S n , that is to

points [s] with hs, si > 0. Indeed we had seen that hyperspheres in Rn ∪ {∞} are
determined
2u
by an equation of the form p(u, u) − 2(u, v) + q = 0 with (v, v) − pq > 0. For x = (u,u)−1 ,
(u,u)+1
−v
1
this equation is equivalent to hx, si = 0 with s = 2 (p−q) . In particular, a hypersphere in
− 12 (p+q)
−c
n 1 2
R with center c and radius r corresponds to the point [s] with s = λ 2 (1−(c,c)+r ) for some
2 1
2 (1+(c,c)−r )
λ 6= 0. This can be used to show:
Proposition. The hyperspheres corresponding to two points [s1 ], [s2 ] outside S 2 intersect at an
angle θ determined by cos θ = √ hs1 ,s2 i . (Since the right hand side is only determined up to
hs1 ,s1 ihs2 ,s2 i
sign, the angle θ is only defined up to θ ↔ π − θ.)
Proof. First check that hsi , si i = λ2i ri2 and
1
hs1 , s2 i = λ1 λ2 (r12 + r22 − (c1 − c2 , c1 − c2 )).
2
hs1 ,s2 i r12 +r22 −(c1 −c2 ,c1 −c2 )
So √ =± 2r1 r2 .
hs1 ,s1 ihs2 ,s2 i
Now use (c1 − c2 , c1 − c2 ) = r12 + r22 − 2r1 r2 cos α (see Figure).
Möbius geometric pencils of circles

Lines in RP3 correspond to 1-parameter families of
circles in R2 ∪ {∞}. The left figure shows two
such families corresponding to a line intersecting S 2
(blue) and the polar line, which does not intersect S 2
(green). The right figure shows two families corre-
sponding to two polar lines tangent to S 2 . In both
cases, the blue and green circles are orthogonal to
each other.
55
Relationship between Möbius and other geometries

Möbius geometry deals with properties of figures in S n ⊂ RPn+1 that are invariant under the
group P O(n + 1, 1) of projective transformations of RPn+1 that map S n → S n . Thus, n-
dimensional Möbius geometry is a subgeometry of (n + 1)-dimensional projective geometry. The
same group, P O(n+1, 1), also maps B n+1 (the inside of S n ) to itself. This gives the Klein model
of (n+1)-dimensional hyperbolic geometry. So n-dimensional Möbius geometry can be seen as
the geometry of the points in the ideal boundary of (n + 1)-dimensional hyperbolic space.
For a point P = [p] ∈ RPn , let GP be the subgroup of P O(n + 1, 1) consisting of all projective
transformations that map P 7→ P (in addition to mapping S n → S n ). These also map the polar
plane of P to itself.
If P is outside S n , then the polar plane intersects B n+1 , and the geometry of this intersection
with the group GP is n-dimensional hyperbolic geometry.
If P is the center of S n , then the polar plane is the plane at infinity, so GP is the group of affine
transformations mapping S n to itself. This is the group of orthogonal transformations. So the
space S n with the group GP is n-dimensional spherical geometry. If P is any other point inside
S n , one obtains a Möbius geometrically equivalent model for n-dimensional spherical geometry.
If P is the north pole of S n , then GP corresponds (via stereographic projection) to the Möbius
transformations of Rn ∪ {∞} that fix ∞. These are the similarity transformations. Thus, S n
with GP is a model for n-dimensional similarity geometry. If P is any other point in S n , one
obtains a Möbius geometrically equivalent model of similarity geometry.
If P ∈ S n , the group GP consists of all projective transformations that come from orthogonal
maps A ∈ O(n + 1, 1) with Ap = λp for some λ ∈ R \ {0}. Because p is a lightlike vector λ is
not always equal to ±1. (For example consider the orthogonal transformations that correspond
to scalings in Rn ∪ {∞}; see last lecture.) If instead of GP , one considers the (projectivized)
group of all A ∈ O(n + 1, 1) with Ap = p, then one obtains a model for n-dimensional Euclidean
geometry.
(n + 1)-dimensional projective
RPn+1 , P GL(n + 2, R)
h hh
hhh
shhhh
(n + 1)-dimensional hyperbolic o / n-dimensional Möbius
points inside S n , P O(n + 1, 1) S n ⊂ RPn+1 , P O(n + 1, 1)
fff f WWWWW
fff ffff WWWWW
WW+
sff
n-dimensional hyperbolic n-dimensional spherical n-dimensional similarity

n-dimensional Euclidean
The paraboloid model of Möbius geometry

Stereographic projection maps a point u ∈ Rn ∪ {∞} to the
point
u1 e1 + . . . + un en + (1 − kuk2 )en+1 + (1 + kuk2 )en+2

in S n ⊂ RPn+1 . Now let

e0 = 12 (en+2 − en+1 ), e∞ = 12 (en+1 + en+2 ),
so that en+1 = e∞ − e0 and en+2 = e∞ + e0 . (The subscripts
0 and ∞ are chosen because [e0 ] and [e∞ ] are the south and
north pole of S n , and these are the images of 0 and ∞ under
stereographic projection. Written in the basis e1 , . . . , en , e0 , e∞ , the image of u is
u1 e1 + . . . + un en + kuk2 e0 + e∞ .

If we divide by the e∞ -coordinate to dehomogenize, this maps u to the point (u, kuk2 ) on the
paraboloid un+1 = kuk2 . So in the new coordinates, stereographic projection becomes vertical
projection to a paraboloid.
56
For the new basis vectors, the Lorentz scalar product is he0 , e0 i = he∞ , e∞ i = 0, he0 , e∞ i = − 21 .
So the matrix for the Lorentz scalar product in the new basis is
1 
..
.
1 .
 
0 −1

2
− 21 0
Because the paraboloid model is just a different projective image of the sphere model, spheres
in Rn ∪ ∞ are mapped to the intersection of the paraboloid with a plane, which corresponds
via polarity to a point outside the the paraboloid. Let us derive an explicit formula for the
correspondence between spheres and points outside the paraboloid. Consider a sphere in Rn
with center c and radius r. A point u ∈ Rn belongs to the sphere if it satisfies the equation
kuk2 − 2(c, u) + kck2 − r2 = 0.
Let (y1 , . . . , yn , y0 , y∞ ) = λ(u1 , . . . , un , kuk2 , 1) be the homogeneous coordinates of the image
point on the paraboloid. In terms of these, the sphere equation is
−2c1 y1 − . . . − 2cn yn + y0 + (kck2 − r2 )y∞ = 0,
or
c1 y1 + . . . + cn yn − 12 y0 − 12 (kck2 − r2 )y∞ = 0,
and this can be written

Pn 2 2
Pn
1 ci ei + (kck − r )e0 + e∞ , 1 yi ei + y0 e0 + y∞ e∞ = 0.
So in homogeneous coordinates with respect to the basis e1 , . . . , en , e0 , e∞ , a sphere with center
c and radius r corresponds to the point
c1 , . . . , cn , kck2 − r2 , 1 .

Note
that this is vertically below the point in the paraboloid corresponding to the center c, which
is c1 , . . . , cn , kck2 , 1 .
In the same way, a hyperplane (v, u) − d = 0 corresponds to the point [v1 , . . . , vn , −2d, 0]. Thus,
in the new coordinates, hyperspheres and hyperplanes correspond to points [s1 , . . . , sn , s0 , s∞ ]
with s21 + . . . + s2n − s0 s∞ > 0. If s∞ = 0, the point corresponds to a hyperplane, otherwise it
corresponds to a sphere with radius r determined by
1 2
r2 = (s + . . . + s2n − s0 ).
s∞ 1
Lie geometry
Suppose we distinguish between between differently oriented
spheres. An unoriented sphere corresponds to two oriented
spheres which consist of the same points but differ in their
orientation. The different orientations can be visualized by
drawing arrows pointing inwards or outwards as shown in
the figure. Let us define a signed radius for oriented spheres
by saying that the signed radius is just the radius if the
arrows point outward, and minus the radius if the arrows
point inward. We define coordinates for oriented spheres by r>0 r<0
appending the signed radius to the Möbius geometric coordinates for spheres. Thus, in the basis
of the paraboloid model, an oriented sphere with center c and signed radius r has homogeneous
coordinates
c1 , . . . , cn , kck2 − r2 , 1, r .

These homogeneous coordinates are not independent anymore, because the first n + 2 already
determine r2 . A point [y1 , . . . , yn , y0 , y∞ , yn+3 ] corresponds to an oriented sphere only if
y12 + . . . + yn2 − y0 y∞ − yn+3
2
= 0.
This is the equation of a quadric in RPn+2 called the Lie quadric. We have thus established a
correspondence between the oriented spheres in Rn ∪ {∞} and the points in the Lie quadric.
57
Lie geometry (continued)

In the last lecture, we have established a bijection between the space of oriented spheres in
Rn ∪ {∞} and the Lie quadric in RPn+2 , which is defined by the quadratic form
y12 + . . . + yn2 − y0 y∞ − yn+3

2
= 0. (∗)
An oriented sphere in Rn ∪ {∞} with center c and signed radius r corresponds to the point
[y1 , . . . , yn , y0 , y∞ , yn+3 ] = c1 , . . . , cn , kck2 − r2 , 1, r

in the Lie quadric.

Remarks. (1) While we do distinguish between differently oriented spheres, we do not distinguish
between spheres and points. A point is just a sphere with radius r = 0, corresponding to a point
in the Lie quadric with yn+3 = 0. Consequently (and reassuringly) points do not come with two
different orientations.
(2) A oriented hyperplane (n, u) − d = 0 in Rn with unit normal vector n can be seen as the
limit of a sphere with radius r and center (d + r)n as r → ∞. For the corresponding points in
the Lie quadric one has
2 r→∞
(d + r)n, (d + r)2 − r2 , 1, r = ( dr + 1)n, dr + 2d, 1r , 1 −→ n, 2d, 0, 1 .

So the oriented hyperplanes correspond to the points in the Lie quadric with y∞ = 0.
A Lie transformation is a transformation of the space of oriented spheres that corresponds to a
projective transformation of RPn+2 which maps the Lie quadric to itself.
Examples. (1) A Möbius transformations (considered in the paraboloid model) is a projective
transformation  y1    y1  
.. ..
 y.n  7−→ A ·  y.n 
y0 y0
y∞ y∞
The transformations of the space of spheres that correspond to projective transformations of

RPn+2 mapping the Lie quadric to itself are called Lie transformations. So a Lie transformation
maps spheres to spheres, but it does in general not map points to points. Indeed, points are
spheres with radius r = 0, so they correspond to the points in the Lie quadric with yn+3 = 0,
and this condition is in general not preserved by a projective transformation that fixes the Lie
quadric. Lie transformations are not point transformations. They are transformations of the
space of oriented spheres. How can these transformations be described geometrically? Let us
examine what it means for the corresponding oriented spheres if two points in the Lie quadric
are polar to each other.
The bilinear form corresponding to the quadratic form (∗) is
qL (y, ỹ) = y1 ỹ1 + . . . + yn ỹn − 12 y0 ỹ∞ − 21 y∞ ỹ0 − yn+3 ỹn+3 .
Two circles with centers c, c̃ and radii r, r̃ correspond to the points
c, kck2 − r2 , 1, r and c̃, kc̃k2 − r̃2 , 1, r̃

in the Lie quadric. These are polar to each other with respect to the Lie quadric if
0 = (c, c̃) − 21 kck2 − r2 − 21 kc̃k2 − r̃2 − rr̃ = 21 (r − r̃)2 − kc − c̃k2 ,

that is, if |r − r̃| = kc − c̃k. This is the case if the spheres touch and
the orientations of the spheres agree in the point of contact (this means
that the arrows point in the same direction). Indeed, if the signs of the
radii are different (as in the top figure) then the condition for oriented
contact is |r| + |r̃| = kc − c̃k; if the signs are equal (as in the bottom
figure) then the condition is |r| − |r̃| = kc − c̃k. Thus:
58
Two points in the Lie quadric are polar to each other if the corresponding
oriented spheres are in oriented contact.
You may want to convince yourself that this is true also for the case
when one or both spheres are in fact hyperplanes or points. For a point
and a sphere, oriented contact means that the point is contained in the
sphere. An immediate consequence of this is:
A Lie transformation maps spheres in oriented contact to spheres in
oriented contact.
In fact, the following is true (although I will not prove this):
Any transformation of the space of spheres which maps spheres in oriented contact to spheres in
oriented contact is a Lie transformation.
So Lie geometry is the geometry in the space of oriented spheres that studies invariants under
the group of transformations that preserve oriented contact.
The Möbius transformations also map spheres to spheres and preserve oriented contact.
If [y] and [ỹ] are two points in the Lie quadric that are polar to each other, then the whole line
[sy + tỹ] is contained in the Lie quadric. It corresponds to the 1-parameter family of spheres
that are pairwise in oriented contact at one point.
What is the signature of the Lie quadric? If we use the coordinates
yn+1 = 12 (y∞ − y0 ) and yn+2 = 21 (y∞ + y0 )
instead of y∞ and y0 , then y∞ = yn+1 + yn+2 and y0 = yn+2 − yn+1 and the quadratic form (∗)
is
y12 + . . . + yn2 + yn+1
2 2
− yn+2 2
− yn+3 ,
so the signature is (n + 1, 2).
59

Geometry1 ws07

Uploaded by

Copyright:

Available Formats

Geometry1 ws07

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Geometry1 ws07

Uploaded by

Copyright:

Available Formats

Boris Springborn Geometry I Lecture 1 Winter Semester 07/08

Contents of this course

Hemispheres and digons

A hemisphere is the intersection of S 2 with a half-space

H = {x ∈ R3 | hx, ni ≥ 0}, knk = 1.

is an Euler triangle. Indeed, there are unique unit vectors A0 , B 0 , C 0 ∈ S 2 with

hA0 , Ai > 0, hA0 , Bi = 0, hA0 , Ci = 0,

and C is the intersection of the half-spaces HA0 , HB 0 , HC 0 , where HX = {x ∈ R3 | hX, xi ≥ 0}.

cos a = hB, Ci, cos α̂ = hB 0 , C 0 i, etc.

The interior angles are α = π − α̂, etc.

−a + b + c > 0, a − b + c > 0, a + b − c > 0, a + b + c < 2π.

−α̂ + β̂ + γ̂ > 0, α̂ − β̂ + γ̂ > 0, α̂ + β̂ − γ̂ > 0, α̂ + β̂ + γ̂ < 2π,

and hence for the interior angles

−α + β + γ < π, α − β + γ < π, α + β − γ < π, α + β + γ > π.

Theorem. The area of a spherical triangle with interior angles α, β, γ is α + β + γ − π.

Theorem. Let a, b, c ∈ R. Then

− a + b + c > 0, a − b + c > 0, a + b − c > 0, a + b + c < 2π (∗)

if and only if there is a spherical triangle with side lengths a, b, c.

G0 = DV −1 (V t )−1 D = D(V t V )−1 D = DG−1 D. (∗∗)

sin α sin b sin c = sin β sin c sin a = sin γ sin a sin b = P.

Divide by sin a sin b sin c to obtain the sine theorem.

cos c = sin b̄ sin ā = cot α cot β,

ā1 = c, b̄1 = β, α1 = ā, c1 = b̄, β1 = α.

E = {x ∈ R3 | hx, ni = d}, where knk = 1, 0 ≤ d < 1.

and this implies cos α̂ = cos α, and hence α̂ = α.

In the same way one gets

and similarly for hv̂, v̂i and hŵ, ŵi.

It maps (n − 1)-dimensional spheres in S n (intersections of S n with affine hyperplanes in Rn+1 )

Bilinear and quadratic forms

H n = x ∈ Rn+1 x21 + x22 + . . . + x2n − x2n+1 = −1, xn+1 > 0

= x ∈ Rn+1 hx, xi = −1, xn+1 > 0 ,

y = T q, and the matrix of b with respect to the new basis is B̃ = T t BT .

q(v) = b(v, v) for all v ∈ V. (◦)

Symmetric bilinear forms

Suppose b is symmetric and q(v) = b(v, v). The kernel of b is

ker b = {v ∈ V | b(v, w) = 0 for all w ∈ V }.

b(êi , êj ) = 0 if i 6= j. ()

λ1 u21 + λ2 u22 + . . . + λr u2r .

Generalized Gram-Schmidt orthogonalization procedure

The ordinary Gram-Schmidt orthogonalization procedure takes as input a basis e1 , . . . , en of V

Completing the squares

Examples. Suppose in some coordinates q is

x1 x2 + x2 x3 = 14 (x1 + x2 )2 − 14 (x1 − x2 )2 + x2 x3 = 41 y1 2 + 41 y2 2 + 12 (y1 − y2 )x3 ,

Lorentz vector spaces

In a Lorentz vector space, any non-zero vector which is

H n = {x ∈ Rn,1 | hx, xi = −1, xn+1 > 0},

One-dimensional hyperbolic space

H 1 is one branch of the hyperbola x1 2 − x2 2 = −1 in R1,1 .

sinh s . (Because cosh2 s − sinh2 s = 1.)

On the other hand,

hγ(s1 ), γ(s2 )i = sinh s1 sinh s2 − cosh s1 cosh s2 = − cosh(s1 − s2 ).

− cosh d(p1 , p2 ) = hp1 , p2 i.

A hyperbolic line in n-dimensional hyperbolic space H n is a non-empty intersection of H n with

Proposition. If U is a 2-dimensional linear subspace of Rn,1 with U ∩ H n 6= ∅, then the restric-

d(p1 , p2 ) = arcosh(−hp1 , p2 i).

Two-dimensional hyperbolic space

The hyperbolic plane H 2 is one component of the hyper-

for some n ∈ R2,1 with hn, ni = 1. The vector −n would

b(êi , êj ) = 0 if i 6= j. ()