1

Pappuss Theorem: Nine proofs and three

variations

Bees, then, know just this fact which is of service to themselves, that the hexagon is greater than the square and the
triangle and will hold more honey for the same expenditure of material used in constructing the different figures.
We, however, claiming as we do a greater share in wisdom than bees, will investigate a problem of still wider
extent, namely, that, of all equilateral and equiangular
plane figures having an equal perimeter, that which has
the greater number of angles is always greater, and the
greatest plane figure of all those which have a perimeter
equal to that of the polygons is the circle.
Pappus from Alexandira, ca. 340 AD

Everything in the world is strange and marvelous to wellopen eyes.

Jose Ortega y Gasset

We will start our journey trough projective geometry in a slightly uncommon way. We will have a very close look at one particular geometric theorem
namely the hexagon theorem of Pappus. Pappus of Alexandria lived around
290350 AD and was one of the last great greek geometers of antiquity. He
was the author of a series of several books (some of them are unfortunately
lost) that covered large parts of the mathematics known at this time. Among
other topics his work addressed questions in mechanics, dealt with the volume/circumference properties of circles and gave even a solution to the angle
trisection problem (with the additional help of a conic). The reader may take
this first chapter as a kind of overture to the remainder of the book. Without
any harm one can also skip this chapter on first reading and come back to it
later.

12

1 Pappuss Theorem: Nine proofs and three variations

B
A

Fig. 1.1. Three versions of Pappuss Theorem.

1.1 Pappuss Theorem and projective geometry

The theorem that we will investigate here is known as Pappuss hexagon Theorem and usually attributed to Pappus of Alexandria (though it is not clear
whether he was the first mathematician who knew about this theorem). We
will later see that this theorem is special in several respects. Perhaps the most
important property is that in a certain sense Pappuss Theorem is the smallest theorem expressible in elementary terms only. The only objects involved
in the statement of Pappuss Theorem are points and lines and the only relation needed in the formulation of the theorem is incidence. Properly stated
the theorem consists only of nine points and nine lines and there is no such
theorem with fewer items. Another remarkable fact is that the incidence configuration underlying Pappuss Theorem has beautiful symmetry properties.
Some of them obvious, some of them slightly hidden.
Theorem 1.1 (Pappuss hexagon Theorem). Let A, B, C be three points
on a straight line and let X, Y, Z be three points on another line. If the lines
AY , BZ, CX intersect the lines BX, CY , AZ, respectively then the three
points of intersection are collinear.
Here intersecting means that two lines have exactly one point in common.
The nine points of Pappuss Theorem are the two triples of points on the
initial two lines and the three points of intersection which finally turn out
to be collinear. The nine lines are the two initial lines, the six zig-zag lines
between the points and finally the line on which the three intersection points
lie. Figure 1.1 shows several instances of Pappuss Theorem. The six black
points correspond to the initial points, whereas the three white points are the
intersections that turn out to be collinear. Observe that in our examples the
positions of the nine points and lines (taken as a set) are identical. However,
the role of the initial two triples of points is played by different points in each
example. The first example shows the picture most often drawn in textbooks
with the final conclusion line between the two initial lines. The second picture
shows that the role of these three lines can be freely interchanged. The last

1.1 Pappuss Theorem and projective geometry

13

Fig. 1.2. An almost parallel bundle of lines which meets at a point far on the right.

picture shows that also one of the inner lines can play the role of the conclusion
line (by symmetry of the construction this line can be an arbitrary inner
line). In fact the automorphism group of the combinatorial structure behind
Pappuss Theorem admits that any pair of lines that do not have a point of
the configuration in common can be taken as initial lines for the theorem.
The exact formulation of the theorem already has some subtleties, which
we want to mention here. The theorem as stated above requires that the pairs
of lines (AY , BX), (BZ, CX) and (BX, AZ), actually intersect, so that we can
speak of the collinearity of the intersection points. Stated as in Theorem 1.1
Pappuss Theorem is perfectly valid in euclidean geometry. However, if we
interpret it in Euclidean geometry it does not exhaust its full generality. There
are essentially two different ways how it can happen that two lines a and b
may not intersect in Euclidean geometry. Either they are identical (then they
have infinitely many points in common) or they are parallel (then they have
no point in common). Now, projective geometry is an extension of Euclidean
geometry in which points are added that are infinitely far away. By this we can
properly speak of the intersection of parallel lines (it lies at infinity) and we
get an interpretation of Pappuss Theorem in which all instances of parallelism
are covered as well.
The essence of real projective geometry may be summarized in the following two sentences: Bundles of parallel lines meet at an infinite point. All
infinite points are incident to a line at infinity. Thus (real) projective geometry is an extension of Euclidean geometry by certain elements at infinity.
In the next two chapters we will in depth elaborate on this extension of Euclidean geometry. In this chapter we will be content with a kind of pre-formal
understanding of it.
Imagine a horizontal line a and a line b that is almost parallel to it. Both
lines meet (since they are not parallel), but the point of intersection will be
relatively far out. If the line b has a small negative slope the intersection point
will be far to the right of the picture. If the slope of b is small but positive
the intersection point will be far out left. What happens if we move line b
continuously from the situation with small negative slope via zero slope to
the situation with small positive slope. The point of intersection will first

14

1 Pappuss Theorem: Nine proofs and three variations

C
B
A

Fig. 1.3. Euclidean Version of Pappuss Theorem.

move further and further to the right (In fact it can be arbitrarily far away).
In the situation with zero slope both lines are parallel and the intersection
point vanishes. After this the point comes back from a very far position on
the left side. Projective geometry now eliminates the special case of parallel
lines by postulating an additional point at infinity on the parallels. Figure 1.2
shows a bundle of lines that meet in a point very far out on the right. If this
point would move to infinity then the lines would eventually become parallel.
It is important to notice that in the concept of projective geometry one
assumes the existence of many different points at infinity: one for each bundle of parallel lines. All these points together form the line at infinity ! .
By introducing these additional elements special cases get eliminated from
geometry. As a matter of fact, these extensions imply that in the projective
plane any two distinct points will have a unique line connecting them and any
two distinct lines will have a unique point of intersection (it just may be at
infinity). Furthermore from an intrinsic viewpoint of the projective plane the
infinite elements are not distinguishable from the finite elements. They have
exactly the same incidence properties. (For more details see the next chapter).

1.2 Euclidean versions of Pappuss Theorem

By passing to a projective framework we get two kinds of benefit. First of all
we extend the scope in which Theorem 1.1 (in exactly the same formulation)
is valid. Any point or any line may as well be located at an infinite position
the theorem stays still true (we will prove this later). On the other hand we
may get interesting Euclidean specializations of Pappuss Theorem by sending
elements to infinity. One of them is given by the theorem below:
Theorem 1.2 (An Euclidean version of Pappuss Theorem). Consider
two straight lines a and b in euclidean geometry. Let A, B, C be three points on
a and let X, Y, Z be three points on b. Then the following holds: If AY || BX
and BZ || CY then automatically AZ || CX.

1.2 Euclidean versions of Pappuss Theorem

15

Fig. 1.4. Euclidean version of Pappuss Theorem with points at infinity and line at
infinity added (left). The straight version (right).

For a drawing of this theorem see Figure 1.3. Figure 1.4 illustrates how
the parallelism of lines is translated to the projective setup. If AY || BX
then these two lines intersect (projectively) at a point at infinity. Similarly
we get an infinite intersection for BZ || CY . Pappuss Theorem (in its
projective version) states that and and the intersection of AZ with
CX are collinear. Since and span the line ! at infinity AZ with CX
must be parallel as well. In other words the conclusion line has been sent to
infinity. The drawing on the right shows a straightened version of the situation
with the conclusion line at a finite location. Observe the similarity of the
combinatorics. If we later on will have introduced the concept of projective
transformation we will see that by a suitable transformation we can send any
instance of Pappuss Theorem to the above situation. Thus our Euclidean
version is essentially equivalent to the full Pappuss Theorem and not just a
special case of it.
We will start our collection of proofs with two proofs of Theorem 1.2.
It should be remarked in advance that most of our proofs will be algebraic
and rely on translations of geometric facts to algebraic identities. There is a
general problem with algebraic proofs: one should never divide by zero! This
seemingly obvious fact leeds to many difficulties and misunderstandings if geometric theorems are concerned. Very often proofs work perfect in generic
situations in which no points or lines coincide or additional collinearities occur, but in certain degenerate cases they may break down. In fact, many
algebraic proofs given in geometry textbooks suffer from these effect and a
whole branch of the current ongoing research deals with the proper treatment
of non-degeneracy conditions. The very statement of Theorem 1.1 carries nondegeneracy conditions in stating that the three crucial pairs of lines should
actually intersect.
In our investigations we will bypass these degeneracy problems by assuming a few (rather strong) generic non-degenericity properties. All nine points

16

1 Pappuss Theorem: Nine proofs and three variations

|OR|
|OP |
=
|OQ|
|OS|

B
A

O
Z

Fig. 1.5. Euclidean Version of Pappuss Theorem (left). Relation of parallels and
segment ratios (right).

of the configuration should be mutually distinct and all nine lines of the configuration should be pairwise distinct. If additional non-degeneracy assumptions
are necessary we will state it in the context of the proof.
Our first proof is extremely simple but (in its naive version) also of limited
scope. It will be based on ratios of segment lengths. We present the proof in a
version that works only under the following two additional assumptions. The
two initial lines must intersect in a point O. The triples of points on these
lines should not be separated by O. By introducing oriented lengths the proof
can be easily extended to get rid of the second assumption. But we will not
do this here.
Proof one: Segment Ratios. By |P Q| we denote the distance of two points
P and Q. Our first proof relies on the following fact, which is a well known
from school lessons on elementary geometry (compare Figure 1.5, right). Let
a and b be two lines intersecting at O and let P and Q be two points on a
not separated by O. Similarly let R and S be two points on b not separated
by O. Then P R and QS are parallel if and only if
|OP |
|OR|
=
.
|OQ|
|OS|
Using this fact and the hypotheses of the Theorem the parallelity of AY
and BX implies that
|OA|
|OY |
=
|OB|
|OX|
Similarly, the parallelity of BZ and CY implies that
|OZ|
|OB|
=
|OC|
|OY |
Since none of the six points is allowed to coincide with O none of the denominators in the above expression is zero. Multiplying the two left sides of the
equations and the two right sides of the equations and canceling the terms
|OB| and |OY | we obtain:

1.2 Euclidean versions of Pappuss Theorem

17

|OA|
|OZ|
=
.
|OC|
|OX|
This in turn is equivalent to the fact that AZ and CX are parallel.

!
"

At first sight the above proof seems to be very simple: Multiply two equations cancel out terms and get the result. Still it has several drawbacks. One
of the main problems is that we translated parallelism into ratios of lengths
of segments. This translation only works literally if the decisive points are not
separated by the intersection of the lines. One can circumvent this problem by
considering oriented line segments. The sign of the ratios used in our proof
would be negative if the points are separated by O and positive otherwise.
However, to make this formally correct one should provide a case by case
analysis that proves that the signs really have the desired behavior. A closer
look shows that the proof is problematic since we introduced the auxiliary
point O and we made the proof dependent on its existence. The complete
proof breaks down if the lines a and b were parallel and point O would not
exist at all. In fact the Euclidean version of Pappuss Theorem does not at all
depend on these special position requirements. The following proof uses only
the six points of Theorem 1.2. However we will need three slightly less trivial
facts concerning polynomials and oriented areas of triangles and quadrangles.
Fact 1: Oriented triangle area.
For three points the A, B, C with coordinates (ax , ay ), (bx , by ) and (cx , cy ) we
can express the oriented area of the triangle (A, B, C) by a polynomials in
the coordinates. To be more specific the desired polynomial is

a x b x cx
1
det ay by cy = ax by + bx cy + cx ay ax cy bx ay cx by .
2
1 1 1

In fact, the specific formula is not important for our next proof. What is more
important is the meaning of oriented: If the sequence of points (A, B, C) are in
counterclockwise order then the area will be calculated with positive sign. If
they are in clockwise order we will get a negative sign. If the three points are
collinear then the triangle vanishes and the area will be zero. We will denote
the triangle area by area(A, B, C).
Fact 2: Oriented quadrangle area.
The oriented area of a quadrangle !(A, B, C, D) can be defined as:
area(A, B, C, D) = area(A, B, D) + area(B, C, D).
This function as again a polynomial of the coordinates of the points. If the
boundary of this triangle (the polygonal chain from A to B to C to D and back
tom A) is free of self intersections, then the usual area is calculated (with sign
depending of the orientation). However if the polygon has self intersections
then one of the triangles in the sum contributes a positive value and the other

18

1 Pappuss Theorem: Nine proofs and three variations

Fig. 1.6. Area of a quadrangle. The convex case (left) and a self-intersecting zeroarea case (right).

a negative value. The area of a self intersecting quadrangle (A, B, C, D) is

zero if and only if the two triangles involved in the sum have equal areas with
opposite signs. Since both triangles share the edge (B, D) the zero case implies
that A and C have the same altitude over this edge. In other words the line
through A and C is parallel to the line through B and D. All together we
obtain:
AC || BD if and only if area(A, B, C, D) = 0.
Fact 3: Zero polynomials.
If a polynomial on several variables is zero in a full dimensional region of
the space of parameters, then it must be the zero polynomial. In other words
if we have a polynomial that evaluates to zero at a certain point and also
for all small perturbations away from that point, then it must be the zero
polynomial.
Now, we have collected everything to formulate a proof of Pappuss theorem by area arguments. The following proof was given as a motivating example by D. Fearnly Sander in an article on the conceptual power of areas for
theorem proving.
Proof two: Area Method. Consider six points A, B, C, X, Y, Z in the euclidean
plane located at positions that roughly resemble the situation in Figure 1.7
on the left. This figure can be considered as being composed from two
triangles (A, C, B), (X, B, Z), and two quadrangles !(B, Y, X, A) and
!(C, Z, Y, B). The sum of the oriented area (with counterclockwise vertex labels) of these tiles equals the area of the surrounding quadrangle
!(C, Z, X, A). Thus we have:
+
+
+
+

area(A, C, B)
area(X, Y, Z)
area(B, Y, X, A)
area(C, Z, Y, B)
area(C, Z, X, A) = 0.

1.2 Euclidean versions of Pappuss Theorem

19

C
B

Fig. 1.7. Pappos proof by the area method.

The expression on the left is obviously a polynomial and it does not depend
on the exact position of the points (since for our argument only the fact that
all involved polygons are labeled counterclockwise and and the fact that the
inner tiles decompose the outer quadrangle was relevant). Hence by Fact 3 this
formula must hold for arbitrary positions of the six points even in degenerate
cases. Now let the six points correspond to the points in Pappuss Theorem.
The hypotheses of the Theorem 1.2 state that (A, B, C) and (X, Y, Z) are two
collinear triples of points. Furthermore we have AY || XB and BZ || Y C. In
terms of areas this means:
area(A, C, B) = area(X, Y, Z) =
area(B, Y, X, A) = area(C, Z, Y, B) = 0.
This, implies immediately that we also have
area(C, Z, X, A) = 0
since otherwise the above area-sum formula would be violated. Hence we have
AZ || XC and the theorem is proved.
!
"

This proof is conceptually by far less trivial then our first one, but as a
benefit we get several things for free. In essence the proof says that if four of
the areas in the formula above vanish then the last one has to vanish as well.
In this form the theorem holds without any restrictions. It covers even the
case of coinciding points.
As a second benefit we may observe that this proof is very useful for generalizations. We may consider the drawing in Figure 1.7 as the projection of
a three dimensional prism over a triangle. The five facets of the prism (two
triangles and three quadrangles) correspond to the five areas involved in the
proof. We can play a similar game with every three dimensional polyhedron
that has only triangles and quadrangles in its boundary. This gives an infinite
collection of incidence theorem for which Pappuss Theorem is the smallest
example. The reader is invited to explore this field on his/her own. For instance, what would be the corresponding theorem if we consider a cube as
underlying combinatorial structure?

20

1 Pappuss Theorem: Nine proofs and three variations

Fig. 1.8. Three versions of Pappuss Theorem.

Before we start to investigate proofs of Pappuss Theorem based on concepts of projective geometry we will present some more interesting instances
of Pappuss theorem. They are drawn in Figure 1.8. Lines that seem to be parallel in the drawings are really assumed to be parallel. The first picture shows
a nice instance that unfolds the order-three symmetry that is inherent to the
Pappuss Theorem. The other two pictures show Euclidean specializations in
which some of the points are sent to infinity. So the Euclidean instance in the
second drawing could be formulated as
Theorem 1.3 (Another Euclidean version of Pappuss Theorem).
Start with a triangle A, B, C. Draw a point D on the line AB. From there
draw a parallel to AC and form the intersection with BC. From this intersection draw a parallel to AB and form the intersection with AC and continue
this procedure as indicated in the picture. After six steps you will reach point
P again.
The patient reader is invited to find out how the drawings in Figure 1.8
correspond to the labeling in our original version of the theorem.

1.3 Projective Proofs of Pappuss Theorem

In this section we want to present proofs in which (in contrast to the last
section) we make no particular use of prallelism. All proofs in this section will
rely only on the collinearity properties of points only. In this respect these
proofs are projective in nature since incidence and collinearity are genuine
projective concepts, while parallels are not.
The main algebraic tool used in this section are homogeneous coordinates
which will be introduced in detail in later chapters. In contrast to usual (x, y)coordinates in the plane homogeneous coordinates present points in the plane
by three coordinates (x, y, z). Coordinate vectors that differ only by a nonzero scalar multiple are considered to be equivalent. The zero vector (0, 0, 0)

1.3 Projective Proofs of Pappuss Theorem

1
2
3
4
5
6
7
8
9

1
a
d
0
g
j
0
m
p

0
b
e
1
h
k
0
n
q

0
c
f
0
i
l
1
o
r

[1, 2, 3] = 0
[1, 5, 9] = 0
[1, 6, 8] = 0
[2, 4, 9] = 0
[2, 6, 7] = 0
[3, 4, 8] = 0
[3, 5, 7] = 0
[4, 5, 6] = 0
[7, 8, 9] = 0

=
=
=
=
=
=
=
=
=

21

ce=bf
iq=hr
ko=ln
ar=cp
bj=ak
f m=do
dh=eg
gl=ij
mq=np

Fig. 1.9. Determinant cancellation for Pappuss Theorem.

is excluded from the considerations. Thus the non-zero points in a one dimensional subspace of R3 represent the same point. A usual Euclidean plane H
can be embedded in a homogeneous framework in the following way. Embed
H as an affine subspace of R3 that does not contain the origin. Each point p
of E corresponds to the one dimensional subspace Vp spanned by p and may
be represented by any non-zero vector of Vp . Conversely, each homogeneous
vector (x, y, z) spans a subspace V(x,y,z) . In general, this subspace intersects
the embedded plane H at some point p. This is the point that corresponds
to (x, y, z). It may happen that V(x,y,z) does not intersect H (this happens
whenever the subspace is parallel to H). Then there is no Euclidean point
associated to (x, y, z). In this case this homogeneous coordinate vector represents an infinite point (see Chapter 3 for details). Thus the finite and the
infinite points can be represented by homogeneous coordinates in a completely
generalized manner.
Collinearity of points in H translates to the fact that the three points in
R3 lie in a single plane (the plane spanned by the corresponding line and the
origin of R3 .) Thus if A = (x1 , y1 , z1 ), B = (x2 , y2 , z2 ) and C = (x3 , y3 , z3 ), are
homogeneous coordinates of points, then one can test collinearity by checking
the condition:

x1 y1 z1
det x2 y2 z2 = 0.
x3 y3 z3
This condition works for finite as well as for infinite points. The following
proof is based on this observation.

Proof three: Determinant cancellations. For matters of better readability we

have exchanged the labels of the points by simple digits from 1 to 9 (see
Figure 1.9). For the proof we need the additional non-degeneracy condition
that the triple of points (1, 4, 7) is not collinear. The generic non-degeneracy
conditions (no identical points and no identical lines) should be still valid.

22

1 Pappuss Theorem: Nine proofs and three variations

Assume that (1, 4, 7) is not collinear. After a suitable affine transformation

(which does not affect the incidence relations of points and lines) we may
assume w.l.o.g. that (1, 4, 7) forms an equilateral triangle. Now we embed the
plane in which our configuration resides into three space in a way that the
points 1, 4 and 7 are at the three-dimensional unit vectors (1, 0, 0), (0, 1, 0)
and (0, 0, 1).
Since the configuration is now embedded in R3 each point is represented by
3-dimensional (homogeneous) coordinates. Three points P, Q, R in our picture
are collinear if and only if the determinant of the 3 3 matrix formed by their
coordinates is zero. We abbreviate this determinant by [P QR]. The matrix in
Figure 1.9 represents the coordinates of the configuration.
The letters in the matrix represent the coordinates of the remaining points.
The generic non-degenericity assumptions imply that none of the letters can
be 0. This can be seen as follows. The triple of points (3, 4, 7) cannot be
collinear, since otherwise two of the configuration lines would coincide. However, the determinant formed by these points equals exactly a. Thus we get

abc
0 %= det 0 1 0 = a.
001

A similar argument works for each of the other variables.

With our special choice of the coordinates, each of the eight collinearities
of the hypotheses can be expressed as the vanishing of a certain 2 2 subdeterminant of the coordinate matrix. If we write down all these equations
(compare Figure 1.9), multiply all left sides and multiply all right sides we are
left with another equation mq = np which translates back to the collinearity
of (7, 8, 9). By our non-degeneracy assumptions all variables involved in the
proof will be non-zero, therefore the cancellation process is feasible.
!
"

This proof carries remarkable symmetric structures concerning the cancellation patterns among the determinants. Structurally it reduces to the fact
that all collinearities correspond to 22 determinants and that the each letter
occurs on the left as well as on the right. The first fact is highly dependent
on the choice of our basis, since only the zeros in the unit vectors allowed to
express each of the collinearieties as a 2 2 determinant.
One can circumvent this problem by an even more abstract approach. Instead of dealing with concrete coordinates of points we may deal with general
properties of determinants. A fundamental role in this context is players by
the so called Grassmann-Pl
ucker relations. These relations state that for arbitrary five points A, B, C, D, E in the projective plane the following relation
holds among the determinants of the homogeneous coordinates.
[ABC][ADE] [ABD][ACE] + [ABE][ACD] = 0.

1.3 Projective Proofs of Pappuss Theorem

23

This remarkable identity is of fundamental importance for projective geometry

and we will dedicate a large part of Chapter 6 to them. For now we take the
identity as an algebraic fact. On it we base our next proof.
Proof four: Grassmann-Pl
ucker relations. We again assume that (1, 4, 7) is
not collinear. We consider the fact that (1, 2, 3) is collinear in our theorem.
Taking this Grassmann-Pl
ucker relation
[147][123] [142][173] + [147][172] = 0
together with the fact that [123] = 0 we obtain
[142][173] = [147][172].
For each of the eight collinearities of the hypotheses we can get one such
equation:
[147][123] [142][173] + [143][172] = 0
[147][159] [145][179] + [149][175] = 0
[147][186] [148][176] + [146][178] = 0
[471][456] [475][416] + [476][415] = 0
[471][483] [478][413] + [473][418] = 0
[471][429] [472][419] + [479][412] = 0
[714][726] [712][746] + [716][742] = 0
[714][753] [715][743] + [713][745] = 0

=
=
=
=
=
=
=
=

[142][173] = [143][172]
[145][179] = [149][175]
[148][176] = [146][178]
[475][416] = [476][415]
[478][413] = [473][418]
[472][419] = [479][412]
[712][746] = [716][742]
[715][743] = [713][745]

Multiplying again all left sides and all right sides of the equations (and taking
care of the the signs of the determinants) and canceling out terms that occur
on both sides we end up with the equation.
[718][749] + [719][748]
(The cancellation is feasible since all involved determinants will be non-zero
by our non-degeneracy conditions). By the Grassmann-Pl
ucker relation
[714][789] [718][749] + [719][748]
this implies that [714][789] = 0. Since [147] was assumed to be non-zero
this implies that [789] = 0 which in turn is equivalent to the collinearity
of (7, 8, 9).
!
"
This proof is very similar to the previous one. However, working directly
on the level of determinants makes the special choice of the basis no longer
necessary. There are amazingly many theorems in projective geometry that
can be proved by this generic determinant calculus and one can base even
methods for automatic theorem proving on them. (For details on this subject
see [....]).

24

1 Pappuss Theorem: Nine proofs and three variations

Z
2

Fig. 1.10. Cevas. Theorem

the proof (right).

|AX|
|XB|

|BY |
|Y C|

|CZ|
|ZA|

= 1 (left). The pasting scheme for

Our next proof exposes a topological structure that underlies Pappuss

Theorem. The proof can be thought of as gluing together several triangular
shapes to a closed oriented surface. The fact that the surface is closed (has no
boundary) corresponds to the conclusion of the theorem.
For this proof to work out we need a kind of basic building block: The
Theorem of Ceva. Cevas Theorem states that if in a triangle the sides are
cut by three concurrent lines that pass through the corresponding opposite
vertex, then the product of the three (oriented) length ratios along each side
equals 1.
In fact, this theorem is almost trivial if one views the length ratios as ratios
of certain triangle areas. For this observe that, if the line (A, B) is cut by the
line (C, D) at a point X, then we have
area(C, D, A)
|AX|
=
,
|XB|
area(C, D, B)

()

where area(A, B, C) denotes the oriented triangle area. In order to prove

Cevas Theorem we consider the obvious identity:
area(CDA) area(ADB) area(BDC)

= 1,
area(CDB) area(ADC) area(BDA)
(note that the oriented triangle area is an alternating function and that each
triangle in the denominator occurs as well in the numerator). Applying the
above identity () we immediately get Cevas Theorem. The converse of Cevas
Theorem holds as well: If the product of the three ratios equals 1, then the
three lines in the interior will meet.
Now consider the situation in which two Ceva triangles are glued together
along an edge in a way such that they share the point on this edge. Multiplying
the two Ceva expressions we see that the ratio on the inner edge cancels out
and we are left only with terms that live on the boundary of the figure (see
Figure 1.11 (left)). We obtain

1.3 Projective Proofs of Pappuss Theorem

b6

a1

a6

b1

25

b5
D

a2

X
a5

V
b2

b4
B

a3
b3

a4

Fig. 1.11. Pasting copies Cevas. Theorem.

|AZ| |CY | |BV | |DW |

= 1.
|ZB| |Y A| |V D| |Y W |
We can extend this process to an arbitrary collection of triangles that are
glued edge-to-edge. An edge can either be used by only one triangle (then it
is a boundary edge) or it can be used by exactly two triangles. The whole collection of patched triangles should be orientable (thus we obtain an orientable
triangulated 2-manifold with boundary). All triangles of the collection should
be equipped with Ceva configurations that have the additional property that
points on interior edges are shared by the Ceva configurations of two adjacent
triangles. We consider the product of all corresponding Ceva-expressions. After cancellation of the ratios that correspond to inner edges we are left with
an expression that only contains oriented lengths ratios from the boundary.
For instance in the situation of Figure 1.11 (right) we get:
a1 a2 a3 a4 a5 a6

b1 b2 b3 b4 b5 b6

1.

The inner part of the structure completely cancels out and does not contribute
to the product on the boundary. Now, if we have a collection of triangles that
has nothing more than a a triangular boundary (i.e. a 2-manifold with a single triangular hole), then the Ceva condition on the whole is automatically
satisfied and we could paste in a final triangle which carries a Ceva configuration. In other words: If we have an orientable triangulated 2-manifold without
boundary and we have a Ceva configuration on all triangles but one (such that
the edge points are shared) then a Ceva configuration can automatically be
put on the final triangle. This is an incidence theorem. We now will show that
using the right manifold Pappuss Theorem can be put in exactly this form.
Proof five: Pasting copies of Cevas Theorem. Consider six triangles that are
arranged as in Figure 1.10 on the right. Furthermore identify opposite edges
of the hexagon as indicated in the drawing. This can be done by placing the

26

1 Pappuss Theorem: Nine proofs and three variations

Fig. 1.12. Creating Pappuss Theorem from six copies of Cevas Theorem

six triangles one over the other (think of a the hexagon made of paper and
fold it appropriately) and glue together corresponding opposite edges. Now,
place a Ceva configuration on each of the edges in a way such that whenever
two triangles meet at an edge the corresponding two points on this edge are
identified. Our considerations above show that if the Edge points are located
such that five of the triangles carry proper Ceva Configurations, then the last
Ceva Configuration is satisfied automatically. The figure in the Middle shows
the situation after all the triangle edges have been identified. Observe that
the points on the edges of the outer triangle as well as the edges themselves
do not contribute to the incidence theorem. What is left after these elements
are deleted is exactly a drawing of Pappuss Theorem
!
"
A proof very similar to this was given by H.S.M. Coxeter and S.L. Greitzer.
Their proof was based on Menelaus Configurations instead of Ceva Configurations but essentially similar. In (...) one can find an elaborate treatment
of the question which geometric theorems can be proved by similar manifold
arguments.

1.4 Conics
This section deals with generalizations and variations of Pappuss Theorem.
In particular we will study what happens if we consider pairs of lines as degenerate cases of a degree two curve (an ellipse, hyperbola or parabola) in
the plane. Degree two curves are often also called conics and they correspond
to solutions of (homogeneous) quadratic equations in homogeneous coordinates. More specifically a conic in the plane is characterized by six parameters
a, b, c, d, e, f and consists of the set of all points with homogeneous coordinates
(x, y, z) that satisfy the equation:
a x2 + b y 2 + c xy + d xz + e yz + f z 2 = 0.
Let (x, y, z) be a solution of this equation. Since the total degree in x, y, z of
each summand is the same (namely two), every scalar multiple (x, y, z) is

1.4 Conics

27

Z
B
C
X

X
Y

Fig. 1.13. Two instances of Pascals Theorem

also a solution of this equation. Thus we may think of each solution as a point
in the real projective plane. The totality of these points forms the conic. The
geometric form of the conic depends on the special values of the parameters.
Projectively there is no difference between ellipse, hyperbola and parabola.
These three cases simply reflect different ways in which the line at infinity
! intersects the conic. If there is no intersection the conic is an ellipse, if
there are two intersections the conic is a hyperbola (it has two infinite points
which correspond to the two asymptotes) if there is just one intersection, the
conic is a parabola (which turns out to be a limit case between the two other
possibilities).
There is one interesting special case that is also important from a projective point of view: the conic may degenerate into two lines (which may even
coincide). This happens whenever the the equation ax2 + by 2 + cxy + dxz +
eyz + f z 2 factorizes into two linear components:
ax2 + by 2 + cxy + dxz + eyz + f z 2 = (1 x + 1 y + 1 z) (2 x + 2 y + 2 z).
In this case the conic consists of two lines each one described by the linear
equation in one of the factors.
In general five points in the projective plane determine a unique conic
passing through each of them. Thus it is a truly projective condition whether
six points lie on a common conic or not. In Chapter 10 we will prove that
six points A, B, C, X, Y, Z are on a common conic if and only if the following
condition among the determinants of the homogeneous coordinates holds:
[ABC][AY Z][XBZ][XY C] = [XY Z][XBC][AY C][ABZ].
We will use this nice characterization to proof the following well known variation (or better generalization) of Pappuss Theorem:
Theorem 1.4 (Variation 1: Pascals Theorem). Let A, B, C, X, Y, Z be
six points on conic. If the lines AY , BZ, CX intersect the lines BX, CY ,
AZ, respectively then the three points of intersection are collinear.

28

1 Pappuss Theorem: Nine proofs and three variations

1
2

5
4
5

Fig. 1.14. Deformations of Pascal s Theorem and labeling for the proof.

Two instances of the theorem can be found in Figure 11.6. Pascalss Theorem is named after the famous Blaise Pascal and was discovered by (the
16 years old) Pascal in 1640. This is about 1300 years after the discovery of
Pappuss Theorem. Nevertheless it is obviously a generalization of Pappuss
Theorem. If the conic in Pascals Theorem degenerates to consist of two lines,
then we obtain immediately Pappuss Theorem. We will proof Theorem 1.4
by a similar determinant cancellation argument we already used our fourth
proof. Figure 11.6 shows two instances of Pascals Theorem one with an ellipse
and one with an hyperbola. If we smoothly deform the first into the second we
will pass through the degenerate situation that resembles Pappuss Theorem.
Proof six: Pascals Theorem. Again we assume for non-degeneracy reasons that
no points and no lines of the theorem coincide. For the labeling in the proof
we refer to Figure 11.6. Consider the following determinant equations.
conic:
[159] = 0
[168] = 0
[249] = 0
[267] = 0
[348] = 0
[357] = 0
[987] = 0

[125] [136] [246] [345] =

=
[157] [259] =
=
[126] [368] =
=
[245] [297] =
=
[247] [268] =
=
[346] [358] =
=
[135] [587] =
=
[287] [597] =

+ [126] [135] [245] [346]

[125] [597]
+ [136] [268]
[247] [259]
[246] [287]
+ [345] [368]
[157] [358]
+ [297] [587]

The first line encodes that the points 1, . . . , 6 lie on a conic. The next six
lines are consequences of Grassmann-Pl
ucker relations and the six collinearity
hypotheses of our theorem. If we multiply all expressions on the left and all
expressions on the right and cancel determinants that occur on both sides we
end up with the last expression which (under the non-degeneracy assumption
that [157] %= 0) implies the desired collinearity of (7, 8, 9).
!
"
Similar to Pappuss Theorem there are a variety of reformulations and
specializations. A nice reformulation is for instance: If a hexagon is inscribed
in a conic in the projective plane, then the opposite sides of the hexagon meet

1.4 Conics

29

9
15

2
A
B

7
X

3
C

Fig. 1.15. Degenerate versions of Pascals Theorem.

in three collinear points. Or if one prefers an euclidean variant of this in which

the conclusion line is sent to infinity one could state: If a hexagon is inscribed
in a conic and two pairs of opposite edges are parallel, the so is the third
pair. There is another nice way to derive even more incidence theorems from
Pascals Theorem. Assume that the conic has a fixed position. If two of the
points in Pascals Theorem that are joined by a line continuously approach
each other until the will meet their joining line will in the limit case become a
tangent to the conic at the position where the two points are located. Thus we
obtain as limit cases situations in which also tangents are involved (observe
that tangents are proper concepts of projective geometry).
Instances of degenerate versions are given in Figure 1.15. The leftmost
picture shows a smallest degenerate situation. The label 15 symbolizes that
points 1 and 5 are identified. The labeling is consistent with the labeling in
Figure 11.6. The join of 1 and 5 becomes the tangent at the point 15. One can
also read the construction in the reverse direction. If a conic C and a point
15 on it is given then one can construct the tangent at 15 by choosing four
arbitrary points 2, 3, 4, 6 on C, construct the joins and intersections as given
by the picture to finally arrive at point 9 another point on the tangent. This
fact was even also known to Pascal and one of the main applications of his
theorem. The second picture shows in essence the same situation as the first
one. However, here the point 15 has been sent to infinity and the corresponding
tangent is located at the line at infinity. By this the conic becomes a parabola
and the two other lines through 15 become parallel to the symmetry axis of the
parabola. Now the theorem reads as follows: Start with for points A, B, C, D
on a parabola. Draw two lines through C and D parallel to the symmetry axis
of the parabola. Intersect them with AD and BC, respectively. Then the join
of the two intersections is parallel to the joins of A and B. The right figure
shows an even more degenerate situation: Inscribe a triangle into a conic.
Form the tangents at the vertices. Intersect them with the opposite sides of
the triangles. The three intersections are collinear.

30

1 Pappuss Theorem: Nine proofs and three variations

Fig. 1.16. Generalizations of Pascals Theorem

1.5 More conics

We can think of Pascals Theorem being derived from Pappuss Theorem by
considering two lines that do not have a configuration point in common as a
(degenerate) conic. Pascals Theorem says that the theorem stays valid even if
the conic is not degenerate. The same process can be applied two more times
to obtain a theorem with 3 conics and three lines. For this consider the left
part of Figure 1.16. The blue conic arises from merging the upper and the
lower line of the drawing. The red and the green conic arise from merging
two other lines. Amazingly the new configuration still forms a theorem. If all
incidences except for the blue line are satisfied as indicated in the picture,
then the three white points are automatically collinear (we will prove this
in a minute). First we observe that there are two combinatorially different
ways of merging three pairs of lines in Pappuss Theorem to three conics. The
second possibility is shown in Figure 1.16 on the right. Also in this case we
get a theorem. To see that they are combinatorially different observe that in
one picture the three lines meet in a point in the other one they dont. Both
theorems are an instance of an even more general fact that is a consequence
of Bezouts Theorem from algebraic geometry. An algebraic curve of degree
d is the zero-set of a homogeneous polynomial of degree d. Thus conics are
algebraic curves of degree 2. Bezouts Theorem can be stated in the following
way: If an algebraic curve of degree n and an algebraic curve of degree m
intersect, then the number of intersections is either finite and smaller than
n m or the curves intersect in infinitely many points and share a component.
Now we can proof the following very strong statement
Theorem 1.5 (Variation 2: Cayley-Bacharach-Chasles Theorem). Let
A and B be two curves of degree three intersecting in nine proper points. If
six of these points are on a conic the the remaining three points are collinear.
Proof seven: Algebraic curves. Let A and B be the curves and let pA (x, y, z)
and pB (x, y, z) be the corresponding homogeneous polynomials of degree

1.5 More conics

31

7
4

Fig. 1.17. If two cubics intersect in line point six of which are on a conic then the
remaining three points are collinear.

three. Bezouts theorem implies that if the two curves A and B have only
finitely many points in common, then they can have at most nine points of
intersection. Call them 1, . . . , 9. And assume that 1, . . . , 6 are on a conic C
with polynomial pC . We will prove that 1, 2, 3 are collinear. Consider a linear
combination p = pA + pB of the two polynomials for some real parameter
. The polynomial p has the following properties. First it is again a degree
three polynomial. Second, it passes through all nine points 1, . . . , 9 (each of
these points was a zero of both pA and pB , so it is also a zero of any linear
combination of them). Now consider an additional point q on the conic C
distinct from 1, . . . , 6. There is a such that p also passes through p (to find
we just have to solve a linear equation pA (q) + pB (q) = 0). Consider p
with this specific value . The curve p passes through 1, . . . , 6 and through q.
Thus it shares these seven points with the conic C. Bezouts Theorem implies
that p must have C as one component. Thus we have p = pC L with a
linear equation L (otherwise p cannot have degree three). This implies that
the points 6, 7, 8 are all contained in the line described by the linear equation
L.
!
"
The situation of the theorem is sketched in Figure 1.17. This theorem has
been independently discovered by several people. Most probably Chasles was
the first person to discover this theorem in a slightly more general version
in 1885. As so often in mathematics the theorem is usually attributed to
other people in this case namely to Cayley and to Bacharach who published
similar results later than Chasles. The theorems shown in Figure 1.16 are
immediate specializations of this theorem. There the two curves of degree
three decompose into the product of a quadratic curve (the conic) and a linear
curve (the line). So the two red components of the picture form one curve of
degree three and the two green components for the other one. The rest is a
literal application of the above theorem. One can even go one step further and
consider Pappuss original theorem as a direct consequence of Theorem 1.5.

32

1 Pappuss Theorem: Nine proofs and three variations

For this one simply has to consider three of the lines as one cubic three and
another three as the other cubic. The color coding in Figure 1.17 makes the
decomposition clear.

1.6 Complex numbers and circles

We are almost at the end of our journey around Pappuss Theorem. In this
section we want to take the considerations of the last chapter still a little
further and draw a surprising connection to the geometry of circles in the
plane. Circles are an intrinsically Euclidean concept. Thus if we do so we
have again to talk about the exact position of our line at infinity ! . As
already mentioned in Section 1.3 homogeneous coordinates can be considered
as embedding the Euclidean plane into R3 at some affine hyperplane. This
time (and this will be done quite often later in the book) we will chose the
affine hyperplane {(x, y, z)|z = 1} for this embedding. Thus a point with
euclidean coordinates (x, y) can be represented by homogeneous coordinates
(x, y, 1) or any non-zero scalar multiple of this vector. The infinite points are
those with coordinates (x, y, 0).
We now want to study circles under this special embedding. A circle is
a special conic. Thus we want to find out, which quadratic equations will
correspond to circles. A circle is usually given by its center (cx , cy ) and a
radius r. In Euclidean geometry the circle equation can be written as
(x cx )2 + (y cy )2 r2 = 0.
Expanding this term and interpreting it in homogeneous coordinates with
z = 1 we can rewrite it:
(x cx z)2 + (y cy z)2 r2 z 2 =
x 2cx xz + c2x z 2 + y 2 2cy yz + c2y z 2 r2 z 2 =
x2 + y 2 2cx xz 2cy yz + (c2x + c2y r2 )z 2 = 0
2

The last line gives the interpretation of the circle in parameters of a general
conic. The circle is a special conic where the coefficients of x2 and y 2 are equal
and where the coefficient of xy vanishes.
There is a surprising (and very deep) connection of circles and complex
numbers. Let us investigate what happens when we intersect a circle with
the line at infinity. In other words we search for solutions of the above equation with z = 0. Clearly the solution must be complex, since no circle has
real intersections with the line at infinity (this properties is only possessed
by hyperbolas and parabolas). In the circular case for z = 0 the equation
degenerates to
x2 + y 2 = 0.
Up to scalar multiples we get the two solutions

1.6 Complex numbers and circles

33

2
5

6
7

2
5
5

1
2

3
1

7
1

6
7

4
3
6

Fig. 1.18. Metamorphoses of theorems.

I = (1, i, 0)

and

J = (1, i, 0).

These solutions are complex points at the line at infinity. Moreover (and this
is an important fact!) they do not depend on the specific choice of the specific
circle. Thus we can say: All circles pass through I and J and any conic passing
through these points is a circle.
This fact is perhaps the most important connection of Euclidean and projective geometry. It allows us to express relations about circles as incidence

34

1 Pappuss Theorem: Nine proofs and three variations

H
G
A

Fig. 1.19. Miquels Theorem.

relations of conics that involve the points I and J. In a very strong sense we
could say that: Every Euclidean incidence theorem can be expressed as a projective theorem in which two points play the special role of I and J. We will
later on dedicate an entire chapter to this effect. Here we will make a small
application of it in the context of Pappuss Theorem. Consider again the two
generalizations given in Figure 1.16. These two pictures are reproduced again
in the first row of 1.18. In these pictures the points in which three conics meet
are marked by white dots. In the same way as we assumed in Section 1.2 that
a certain line is located at the line at infinity we will now assume that in
each picture two of these points are located at the points I and J. All other
points should stay at real positions. A conic that passes through I and J is
a circle. Thus the conics in our theorem become circles (this is similar to the
effect that two lines become parallel if their point of intersection is located
at an infinite position). So the two theorems can be interpreted as Euclidean
theorems about seven points, three lines and three circles. The corresponding
pictures are shown in Figure 1.18 in the second row. For instance, the first of
these two theorems can be stated as: Given three circles, that intersect mutually in two points. The three lines spanned by the intersections of each pair
of circles meet in a point. The meeting point corresponds to point 7 in the
original theorem.
We can even go one step further. We can interpret straight lines as circles
with infinite radius. There is a particular way of extending Euclidean geometry that reflects this way of thinking. For this we introduce one point
at infinity and assume that straight lines are circles that in particular contain this point. (A word of caution: one should not confuse this extension
of Euclidean geometry by one point with the projective plane we introduced
earlier. In the projective plane a line at infinity was introduced. The extension
by only one point used here has something to do with the projective complex

1.6 Complex numbers and circles

35

Fig. 1.20. Angles in a circle.

line and is called the one-point compactification of the Euclidean plane and
will be investigated later).
In this setup we no longer have to distinguish between lines and circles.
Lines are just circles of infinite radius. In this interpretation our two theorems
could be stated as theorems on six circles and eight points (we interpret the
infinite point just as an ordinary point. The last row of Figure 1.18 gives a
drawing of the situation in which is located at a finite position. For instance
the second theorem (which is a well known fact from circle geometry) could
be stated as follows.
Theorem 1.6 (Variation 3: Miquels Theorem). Consider four points
A, B, C, D on a circle. Draw four more circles C1 , C2 , C3 , C4 That pass through
the pairs of points (A, B), (B, C), (C, D) and (D, A), respectively. Now consider the other intersections of Ci and Ci+1 for i = 1, . . . , 4 (indixes modulo
4). These four intersections are again cocircular.
We will give an elementary proof of this theorem by calculations of angle
sums. The basic fact that we will need for this proof is illustrated in Figure
1.20. If we consider a secant AB of a circle and if we look at this secant from
two other different points C and D of the circle (which are on the same side
of AB) we will see the secant in the same angle. If the points C and D are
at opposite sides of the secant we will have complementary angles. Observe
that the angles in Figure 1.20 are assumed to be oriented angles. Thus the
complementary angle has to be counted with negative sign. If one takes care
of the orientation of the angles one could say that the difference of the two
angles at C and D will in both cases be a multiple of . Conversely four points
A, B, C, D lie on a common circle if the difference of the angles (under which
AB is seen) at C and D is a multiple of . Thus we get a characterization of
four points on a circle in terms of angles.
In principle, Miguels theorem can now easily be proved by considering
angle sums among the six involved circles. However we here will prefer a more

36

1 Pappuss Theorem: Nine proofs and three variations

algebraic approach that expresses the angle relations in terms of complex

numbers. For this assume that all eight points in the picture are finite and
consider the picture of Miquels theorem embedded in the complex number
plane C. We consider A, B, C, D from Figure 1.20 as complex numbers. Then
for instance A C forms a complex number that points into the direction
AC
we get a complex number whose
from C to A. Forming the quotient BC
argument (the angle w.r.t. the real axis) is exactly the angle at point C.
AD
Similarly BD
gives a complex number that describes the angle at point D.
We can compare these two angles by forming again the quotient of these two
AC AD
numbers: BC
/ BD . This umber will be real if and only if the two angle
differ by a multiple of .
Taking everything together we get the following characterization of four
points being cocircular (possibly with infinite radius): Four points A, B, C, D
in the complex plane are cocircular if and only if
(A C)(B D)
(B C)(A D)
is a real number.
The above expression is called a cross-ratio and we will later on see that
cross ratios play an fundamental and omnipresent role in projective geometry.
With the help of cross-ratios we can easily state a proof for Miquels theorem.
Proof eight: Cross-ratio cancellations. Assume that the quadruples of points
(A, B, C, D), (A, B, E, F ), (B, C, F, G), (C, D, G, H), (D, A, H, E) are cocircular. From this we obtain that the following cross-ratios are all real:
(AB)(CD)
(CB)(AD) ,

(F B)(AE)
(AB)(F E) ,

(CB)(F G)
(F B)(CG) ,

(HD)(CG)
(CD)(HG) ,

(AD)(HE)
(HD)(AE) .

Multiplying all these numbers and canceling out terms that occur in the numerator as well as in the denominator we are left with the expression:
(F G)(H E)
.
(H G)(F E)
Since this expression is the product of real numbers it must itself be real. By
our above observations this expresses exactly the cocircullarity of (E, F, G, H)
which is the conclusion of our theorem.
!
"

1.7 Finally...
We will end this section by an almost trivial proof of Pappuss theorem in its
full generality by simply expanding an algebraic term. Still we need a little
preparation for this. Again consider the original points of Pappuss Theorem
expressed in homogeneous coordinates. Thus we assume that the drawing

1.7 Finally...

F
I

37

Fig. 1.21. A construction sequence for Pappuss Theorem.

plane H is again embedded in R3 at a position that does not contain the

origin of R3 . As before each point p is represented by a three dimensional
vector (x, y, z). This time we will take all points of R3 {(0, 0, 0)} into account.
For this we identify the vector (x, y, z) with all of its non-zero scalar multiples
(x, y, z), %= 0. By this R3 {(0, 0, 0)} is divided into equivalence classes.
Each equivalence class represents a point of the projective plane. A point of the
drawing plane H can be represented by its actual (x, y, z) position or by any
non-zero scalar multiple of it. Conversely, for a point (x, y, z) of R3 {(0, 0, 0)}
we consider the line l(x,y,z) through it and the origin. The point in H that is
represented by (x, y, z) is the intersection of l(x,y,z) and H. If this intersection
does not exist, (x, y, z) represents an infinite point.
In this setup a straight line g in H may be considered as two dimensional
linear space spanned by the elements of g and the origin of R3 . Such a line
may by represented by a linear equation
{(x, y, z) R3 {(0, 0, 0)}|ax + by + cz = 0}
given by parameters (a, b, c) R3 {(0, 0, 0)}. Thus points as well as lines are
represented by non-zero vectors in R3 . A line g is incident to a point p if and
only if standard the scalar product +p, g, is zero.
This observation gives us the key for a very elegant method of calculating
the line that connects two points p and q. We simply need a vector g with
the property +p, g, = +q, g, = 0. Such a vector can simply be calculated by
the cross-product p q. Similarly the intersection of two lines g and h asks
for a vector p with the property +p, g, = +p, h, = 0. Thus the intersection
can be calculated by g h. So we can apply the cross product to calculate
intersections and joins in projective geometry. (We will learn much more of
this in Chapter 4.)
What happens if we try to form the join of two identical points p and q. If
p and q represent the same point they must be scalar multiples of each other
q = p. Performing the cross product we obtain: p q = p p = (p p) =
(0, 0, 0). Obtaining a zero vector as result is an indication of performing a

38

1 Pappuss Theorem: Nine proofs and three variations

degenerate operation. A similar effect happens when we try to intersect two

identical lines.
How can we test collinearity of three points p, q, r? The points are collinear
if and only if the representing vectors are linear dependent in R3 . Thus we
can test collinearity by the condition det(p, q, r) = 0.
Now we can express Pappuss Theorem as a sequence of nested cross products and a determinant. Expanding the final term and observing that it is zero
will proof the theorem
Proof nine: Brute force. We give a construction sequence for the Pappuss
configuration. We start with five free points A, B, C, D, E (compare Figure
1.21). The coordinates for the remaining four points in the construction can
be calculated by
F = (A D) (B C),
G = (A B) (D E),
H = (C D) (B E),
I = (A H) (C G).

Testing the final collinearity boils down to test whether det(E, F, I) = 0. The
following session of the computer algebra program Mathematica shows an
evaluation of these expressions. All output except for the final result has been
suppressed. The final 0 proves Pappuss Theorem.

What does this evaluation indeed proof? It shows that when we perform
the construction sequence independently of the initial choice of the coordinates of A, B, C, D, E the final determinant will be zero. This may happen for
two different reasons. Either during the construction sequence we run into a
degenerate situation (like the intersection of identical lines) that introduce a
zero-vector as an intermediate result. Or all operations were valid (this will
be the case for almost every instance) and the final points E, F, I are indeed
collinear.
!
"
A word of caution: The last proof is very general and seems to be straight
forward. Still the help of a compute is essential here. Performing the calculations per hand would require to perform all cross-products and to evaluate the
final determinant. The final term has all together 15456 summands of degree
15. They can be cancelled in pairs which gives the final conclusion.