On Projective Planes
On Projective Planes
On Projective Planes
On Projective Planes
2 4
7
3 5
6
By Johan K
ahrstr
om
ii
iii
Abstract
It was long an open question whether or not a projective plane of order 10 existed. It has now been
shown, using a computer search, that no projective plane of order 10 exists. This essay introduces
the concept of projective planes, and then looks at some of the theory on the existence of certain
projective planes. It concludes with a look at the computer search for a projective plane of order
10, and some of the theory behind it.
iv
Contents
Acknowledgement vii
Bibliography 41
v
vi CONTENTS
Acknowledgement
I would like to thank Pia Heidtmann for all her help and support, and for putting up with me
taking up much more of her time than I was supposed to (when her time often should have been
spent on marking exams). I am also very thankful for her suggestion on the topic for this essay. I
had only a vague idea of what I wanted to do, and I feel that projective planes is about as close to
this idea as is possible to come.
vii
viii ACKNOWLEDGEMENT
Chapter 1
This essay will give an introduction to a special kind of geometry called a projective plane. In
the first two chapters we define projective planes, and go through some of their most important
properties. The last two chapters will look at existence questions regarding the finite projective
planes. In particular, the last chapter looks at how it was shown that a certain finite plane does
not exist.
Projective planes are a special case of a more general structure called a geometry. As geometries
have more in common with our intuitive notion of geometry, we shall start by looking at these.
1.1 Geometries
Definition 1 (Geometry). A geometry S = (P, L) is a non-empty set P whose elements are
called points, together with a set L of non-empty subsets of P called lines satisfying:
G1: For any two distinct points p1 , p2 P , there exists exactly one line l L such that both
p1 l and p2 l.
G2: There exists a set of four points, such that given any set of three of these points, no line
exists that contains all three points.
Note that the sets P and L may be either finite or infinite. We say that a point p is on, or
incident with, a line l if p l. Similarly a line l is on, or incident with, a point p if p l. A set of
points is called collinear if there exists a line such that all points are on the line. If p and q are
two points, then pq denotes the unique line on both p and q. Obviously, qp = pq. If l1 and l2 are
lines that intersect in a point, l1 l2 denotes their point of intersection.
Using this notation, we can write G1 and G2 in a more straightforward way:
G2: There exists a set of four points, no three of which are collinear.
A set of four points, no three of which are collinear, is called a quadrangle. A line through two
points of a quadrangle is called a diagonal of the quadrangle.
Example 2 This first example will assure us that the Euclidean plane is indeed a geometry. Here,
P consist of the usual points, and L of the usual lines. It is a well-known fact that two points are
on a unique line. The points (0,0), (1,0), (0,1), (1,1) form a quadrangle. Thus, both G1 and G2
are satisfied, so the Euclidean plane is a geometry.
Example 3 S = (P, L), where P = {1, 2, 3, 4} and L = {{1, 2}, {1, 3}, {1, 4}, {2, 3}, {2, 4}, {3, 4}}
(see Figure 1.1 (a)) is a geometry. It is easy to see that any two points are on exactly one line,
and the four points of P make up a quadrangle.
1
2 CHAPTER 1. INTRODUCTION - PROJECTIVE PLANES
1 2 2
1
3 4
3 4
(a) (b)
Proof. Assume that there are two lines that intersect in at least two points, and let two such
points be p1 and p2 . But, by G1, there is exactly one line on both p1 and p2 , and we have a
contradiction.
There are many kinds of geometries, and this essay is mainly about a set of geometries called
projective planes. However, before we look at those, we shall look briefly at another set of geometries
called the affine planes as the two are closely related.
Definition 5 (Affine Plane). An affine plane A is a geometry, that satisfies the following
condition:
AP : For any line l, and any point p not on l, there exists a unique line l0 on p that does not
intersect l.
This is the famous parallel axiom of Euclidean geometry. Clearly, the Euclidean plane is an
affine plane. The geometry of Example 3 is also easily seen to be an affine plane. For example, if
l = {1, 4} and p = 3, the unique line on p that is not on l is {2, 3}. The obvious connection to the
parallel axiom justifies the following definition.
Definition 6 (Parallel Lines). Two lines in an affine planes are said to be parallel if they do
not intersect. Any line is also said to be parallel to itself. If l1 and l2 are two parallel lines, we
write this as l1 ||l2 .
Lemma 7. The relation is parallel on the lines of an affine plane is an equivalence relation.
Proof. Let l1 , l2 and l3 be three distinct lines of A. By definition l1 ||l1 , so this relation is reflexive.
Now, assume that l1 is parallel to l2 , that is l1 does not meet l2 . Then l2 does not meet l1 either, so
l2 is parallel to l1 and the relation is symmetric. To show transitivity, assume that l1 ||l2 and l2 ||l3 ,
but that l1 is not parallel to l3 . Then l1 and l3 intersect in a point p. But then p is on two lines
missing l2 , contradicting AP . Hence we must have that l1 ||l3 , and the relation is transitive.
The equivalence relation is parallel partitions the set of lines in an affine plane into parallel
classes. For a line l in an affine plane, we denote its parallel class by [l]. That is, [l] consists of all
lines parallel to l.
Lemma 8. Let p be a point of an affine plane. For each parallel class of lines, there is exactly one
line on p that belongs to the class.
Proof. Let [l] be any parallel class, l L. If l is not on p, by AP there exists a unique line on p
parallel to l, and we are done. If l is on p, we must show that no other line on p is also in [l]. But
any other line on p meets l in p, so the lines are not parallel.
1.2. PROJECTIVE PLANE 3
2 4
7
3 5
6
Example 11 Let A = (P, L) be the Euclidean plane. As we have seen earlier, this is an affine
plane. However, there is an easy way to construct a projective plane from the Euclidean plane, by
tying together all parallel lines in a point at infinity. It is done as follows.
Let P 0 = {P R {}}, where is just a symbol. We shall define the set L0 . Any line l of A
can be written as either the equation y = kx + m or x = m, where k, m R. If l is of the former
form, then let l0 = {l {k}} be an element of L0 . If l is of the latter form, then let l0 = {l {}}
be an element of L0 . Finally, let l = {R {}} (called the line at infinity) be an element of L0 .
A point on the line at infinity will be written as either (k) where k R or as (), and any point
in P will be written as (x, y) in the obvious way. A point in P will be called an ordinary point,
and a point on the line l will be called an extended point.
It should be obvious that two points of are on a unique line, so G1 is satisfied. The four points
(0, 0), (0, 1), (1, 0) and (1, 1) still form a quadrangle, so G2 is also satisfied. Let l and l0 be two
lines other than the line at infinity. If l and l0 have different slope, then they meet in one point of
the ordinary plane (and only there). If they have the same slope, then they meet in a point of the
line at infinity (and only there). Finally, l meets the line at infinity in precisely one point. Thus
P P is satisfied, so = (P 0 , L0 ) is a projective plane. This plane is called the projective real plane
The previous example suggests a way of turning any affine plane into a projective plane. This
is done as follows. Let A = (P, L) be an affine plane. Let P 0 = P {[l] | l L}, that is, P 0 is P
with one point added for each parallel class of A. Now, define the set L0 as follows. For each line
l L, let {l {[l]}} be an element of the set L0 . Finally, let {[l] | l L} be an element of L0 . Then
= (P 0 , L0 ) is a projective plane. If this is not easy to see, compare with Example 11.
Similarly, any projective plane can be turned into an affine plane by removing one line, and all
points on it. G1 still holds, as does G2. That AP holds can be seen as follows. Let p be any point
of the affine plane, let the line removed from the projective plane be l, and let l0 be any line in the
affine plane not on p. In the projective plane, l0 intersects l in precisely one point, call it q. Again,
4 CHAPTER 1. INTRODUCTION - PROJECTIVE PLANES
in the projective plane, there is one unique line on p and q. Thus, in the affine plane, the line pq
is the only line on p not meeting l0 , and thus AP holds.
Example 12 We shall turn the affine plane of Example 3 into a projective plane. As there
are three parallel classes, add the points 5, 6 and 7 to get the lines {{1, 2, 5}, {3, 4, 5}, {1, 4, 6},
{2, 3, 6}, {1, 3, 7}, {2, 4, 7}, {5, 6, 7}}.
Example 13 Let = (P, L) where P are the lines of R3 through the origin and L are the planes
of R3 through the origin. Two distinct lines through the origin define a unique plane through the
origin, so G1 is satisfied. Furthermore, two distinct planes through the origin intersect in a unique
line through the origin, so P P is also satisfied. The axes, together with the line through the origin
and (1, 1, 1) is an example of a quadrangle, so is indeed a projective plane. This example is
very important, and we shall use it in the next chapter to construct projective planes.
Definition 14 (Dual statement). If S is a statement on projective planes, then the dual state-
ment S 0 of S is the statement where the words points and lines are swapped.
Example 15 The dual statement of two points are on a line is two lines are on a point.
In the definition of a projective plane, there are three statements on points and lines:
G2: There exists a set of four points, no three of which are collinear.
G20 : There exists a set of four lines, no three of which are on a common point.
We see that G10 = P P , and P P 0 = G1. So, if we assume that G1, G2 and P P hold, then obviously
G10 and P P 0 hold. Also, by G2, there exist four points p1 , p2 , p3 and p4 , no three collinear. Thus
the lines p1 p2 , p2 p3 , p3 p4 and p4 p1 are four distinct lines (see Figure 1.3 (a)). Obviously, each of
the four points p1 , p2 , p3 and p4 are on exactly two of these lines. Now, suppose for a contradiction
that three of these lines intersect in a commont point p (which will be different from p1 , p2 , p3 and
p4 as each of these are on exactly two of the lines). But this contradicts P P , as then there are two
distinct points common to two of the three lines. Hence G20 hold.
l1
p p
1 2
l4 l2
p l
3
p
4
l3
p
(a) (b)
Figure 1.3: (a) If the three lines p1 p2 , p2 p3 and p4 p1 would intersect in a common point p, the lines p1 p2 and
p1 p4 would both contain both p and p1 , contradicting P P . (b) If there would exist a line l through the points l1 l2 ,
l2 l3 and l3 l4 , then there would be two distinct lines containing the two points l2 l3 and l3 l4 , contradicting P P 0 .
1.3. NUMERIC PROPERTIES OF FINITE PROJECTIVE PLANES 5
On the other hand, assume that G10 , G20 and P P 0 hold. Then obviously G1 and P P hold. Also,
by G20 , there exist four lines l1 , l2 , l3 and l4 , no three on a common point (see Figure 1.3 (b)).
Hence the four points l1 l2 , l2 l3 , l3 l4 and l4 l1 are all distinct (all these lines intersect by G10 ). As
these are all distinct, each of the four lines l1 , l2 , l3 and l4 are on exactly two of these points. Now
suppose, for a contradiction, that three of these points are collinear, say on the line l (which will be
different from the lines l1 , l2 , l3 and l4 as these are on only two of the points). But this contradicts
P P 0 , as then there are two distinct lines through two of the three points. Hence G2 hold.
From the preceding argument, we see that G1, G2 and P P hold if and only if G10 , G20 and P P 0
hold. Hence, for any statement proved using G1, G2 and P P , the dual statement can be proved
using G10 , G20 and P P 0 , and vice versa. We have proven the following theorem:
Theorem 16. For any true statement on projective planes, the dual statement is also true.
This simplifies our work a lot. If we prove a theorem saying something about the points of a
projective plane, the dual argument will prove the same thing about lines, and vice versa. This is
one of the beautiful symmetric properties of projective planes.
Example 17 Let = (P, L), where P = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13}, and
See figure 1.4. Note that point 4 is on the line {1, 2, 3, 4}, even though the figure does not show
this (this line is represented in the figure by the circle going through points 1, 2 and 3). This is a
projective plane.
6
2 3
5
4
11 8
12 9
13 10
Figure 1.4: A finite projective plane. Note that {1, 2, 3, 4} is a line even though this is not clear from the picture.
6 CHAPTER 1. INTRODUCTION - PROJECTIVE PLANES
If you look at our two examples of finite projective planes, they are both very symmetrical. In
the Fano plane, each line is on three points, and each point is on three lines. In the last example
each line is on four points, and each point is on four lines.
Lemma 18. Let l be any line of a projective plane , and p any point not on l. Then the line l
has exactly n + 1 points for some positive integer n, if and only if there are n + 1 lines on p.
Proof. Assume that the line l has n + 1 points. Then each of these points are on exactly one line
that is also on p, so there are at least n + 1 lines on p. Since any line on p intersects l in exactly
one point, there are no other lines on p, so there are exactly n + 1 lines on p. The dual argument
shows the converse.
Theorem 19. Let be a finite projective plane. Then there exists an integer n > 1, such that
each point is on n + 1 lines, and dually, each line has n + 1 points.
Proof. We will prove the first part, and the second part will follow from the dual argument. Let
p1 , p2 , p3 , p4 , be the points of a quadrangle in . Each of these points is certainly on at least three
lines (the lines on the point and the other points of the quadrangle). Let n > 1 be the number
such that p1 is on n + 1 lines.
Now, the lines p2 p3 , p3 p4 and p2 p4 are each on exactly n + 1 points, by Lemma 18. For each
point q in , at least one of the lines p2 p3 , p3 p4 , p2 p4 misses q. Again, by Lemma 18, there are
exactly n + 1 lines on q, and we have proven the theorem.
Definition 20 (Order of a finite projective plane). The order of a finite projective plane is
the integer n such that each point is on n + 1 lines, and each line is on n + 1 points.
By looking at our previous examples, we see that the Fano plane has order 2, as does the plane
of Figure 1.1 (b). The plane in Figure 1.4 has order 3. The following theorem counts the total
number of points, and lines, of a finite projective plane.
Theorem 21. The number of points in a projective plane of order n is n2 + n + 1, and dually
the number of lines is also n2 + n + 1.
Proof. Let p be any point of . Now, there are n + 1 lines on p. Each of these lines have n points
other than p. All of these points are distinct, as otherwise two different lines would have two points
in common, which violates G1. Thus, the number of points of other than p is n(n + 1) = n2 + n,
so the total number of points is n2 + n + 1.
By this theorem, the Fano plane should contain 22 + 2 + 1 = 7 points and lines, which it does.
Also, The plane of example 17 should have 32 + 3 + 1 = 13 points and lines, which it indeed does.
Theorem 22. The projective plane of order 2 is unique (up to renumbering of the points).
Proof. Let the plane be . We know that has 7 points, which we will denote by 1, 2, 3, 4, 5, 6
and 7. There should be three lines on the point 1, and we may, without loss of generality, assume
that these are l1 = {1, 2, 3}, l2 = {1, 4, 5} and l3 = {1, 6, 7}. Now, there are two more lines on the
point 2. One of these must be on point 4, and we may, again without loss of generality, assume
that this line is l4 = {2, 4, 6}. Then the other line on 2 must be l5 = {2, 5, 7}. Now, 4 is on one
more line. The only points that 4 is not already on a line with are 3 and 7, and thus l6 = {3, 4, 7}.
Now all points except 3, 5 and 6 are on three lines, and no two of these points are already on a
line, so l7 = {3, 5, 6} is a line. We have now constructed a projective plane of order 2, and this was
possible in essentially one way (up to renumbering of the points), so there is only one projective
plane of order 2.
From the previous theorem we see that the projective planes of Figures 1.1 (b) and 1.2 are
actually the same. This can be seen by renumbering the points of Figure 1.1 (b) to
11 44 76
22 53
37 6 5.
So, why do we use the word planes for these object. Well, this is because they are, in a sense,
two-dimensional. In linear algebra, we say that two vectors span a plane, or equivalently that three
points not on a line define a plane. The next theorem shows, that for finite projective planes, by
selecting three points, not on a line, we can generate the entire plane.
1.4. INCIDENCE MATRICES 7
Theorem 23. Let be a finite projective plane of order n. Let p, q and r be three points of ,
not on a line. Then the lines pp0 , where p0 is any point on the line qr contains all points of .
Proof. The line pq has n + 1 points. The n + 1 lines through r and these points are all distinct.
Counting the points of these lines, we get n(n + 1) + 1 = n2 + n + 1, which are all the points of
.
Example 24 Let p = 1, q = 3 and r = 7 of the Fano plane, figure 1.2. The line pq is {1, 2, 3}.
The lines through these points and r are the lines {1, 6, 7}, {2, 4, 7} and {3, 5, 7}. By theorem 23
any point of the Fano plane should be in at least one of these lines, which is easily seen to be true.
The following lemmas will be useful when proving the Bruck-Ryser theorem in chapter 3. Please
use the incidence matrix of the previous example to check their validity.
Lemma 27. Let A be the incidence matrix of a finite projective plane of order n. Let ui , be the
rows of A. Then: (
1 if i 6= j
ui uj =
n + 1 if i = j
The same holds for inner products of rows of A.
Proof. Let N = n2 + n + 1 be the number of lines/points of , or equivalently the number of
columns/rows of A. Let [ai1 , ai2 , . . . , aiN ] be the row ui , and [aj1 , aj2 , . . . , ajN ] be the row uj .
If i 6= j, these rows represent two different lines. These two lines have one unique point in
common, which means that there is exactly one m, 1 m N such that both aim and ajm are
one. Hence the product of aim and ajm is one. For all k, 1 k N and k 6= m, at least one of
aik or ajk is zero, which means that their product is zero. By this we see that the inner product
of these rows (which is the sum of all these products) are exactly one.
If i = j, these rows represent the same line. Then it is easy to see that their inner product is
exactly the sum of the ones in the row. Since there are n + 1 ones in any row, this sum is n + 1.
Lemma 28. Let A be the incidence matrix of a projective plane of order n. Then AAT = nI +J,
where I is the (n2 + n + 1) (n2 + n + 1) identity matrix and J is the (n2 + n + 1) (n2 + n + 1)
matrix having all entries equal to one.
Proof. Let aij be the entry of AAT on row i and column j. This is equal to the inner product
of rows i and j of A. By Lemma 27, aij = 1 if i 6= j, and aij = n + 1 if i = j. Thus the
diagonal of AAT will equal n + 1, and all other entries will equal 1. In short, this can be written
as AAT = nI + J, using the notation of the lemma.
8 CHAPTER 1. INTRODUCTION - PROJECTIVE PLANES
Lemma 29. Let A be the incidence matrix of a projective plane of order n. Then
2
det A = (n + 1)n(n +n)/2
.
Proof. Let B = AAT . Then det B = det(AAT ) = det A det AT = (det A)2 . As det B is a square,
it is non-negative. By Lemma 28 B has all entries 1, except for the diagonal, which has all entries
n + 1. Subtract the first row from all other rows, and then add all columns to the first column. As
all these operations preserve the absolute value of the determinant, this gives us
n+1 1 1
n + 1 1 1
..
1
n n 0
n + 1 .
=
det B =
.. .. .. .
.
..
. 1
. . ..
n 0 n
1 1 n+1
(n + 1)2 1 1
0 n 0
=
.. .. .
. . ..
0 0 n
As this is a triangular matrix, the determinant is just the product of the elements of the diagonal.
All the elements of the diagonal are positive, so
2
det B = (n + 1)2 nn +n
.
2
As (det A)2 = det B, it follows that det A = (n + 1)n(n +n)/2
.
Chapter 2
We will start this chapter with a section that shows how we can construct projective planes using
fields. We will then turn to the important concept of collineations (or automorphisms) of projective
planes.
If x = 0, then y and z can not both be zero (as v is a non-zero vector). Then, if y 6= 0 then
[v] = [y 1 v] = [0, 1, z 0 ], but if y = 0 then [v] = [z 1 v] = [0, 0, 1].
Any plane through the origin can be represented with an equation of the form ax + by + cz = 0,
where a point (x, y, z) is on the plane if it satisfies this equation. We will use the notation ha, b, ci
to denote the line of corresponding to this plane. As in the case for vectors, [a, b, c] = [a0 , b0 , c0 ]
if and only if a = ra0 , b = rb0 and c = rc0 for some non-zero element r F . Using this notation,
we se that the point [x, y, z] is on the line ha, b, ci if and only if
x
ha, b, ci y = 0
z
The fact that constructed in this way is a projective plane follows easily from elementary linear
algebra. Two lines with exactly one point in common are contained in exactly one plane in F 3 , and
as the point in common is the origin, the plane passes through the origin. Hence the two distinct
lines of F 3 corresponding to two distinct points of lie in exactly one plane through the origin,
which thus define a unique line through the two points of . Hence G1 is satisfied. The four points
[1, 0, 0], [0, 1, 0], [0, 0, 1] and [1, 1, 1] are an example of a quadrangle, so G2 is satisfied. In fact,
we may always assume that the four points of a quadrangle has these co-ordinates. Finally, two
9
10 CHAPTER 2. COLLINEATIONS AND CONSTRUCTIONS OF PROJECTIVE PLANES
Figure 2.1: (a) Two points are on one unique line. (b) Two lines intersect in exactly one point. (c) There exists
a set of four points, no three of which are collinear.
planes through the origin in F 3 intersect in exactly one line through the origin, so the two lines of
corresponding to the planes meet in a unique point, and thus P P is satisfied, and we have shown
that is a projective plane (see figure 2.1). The plane constructed in this way using the field F is
usually denoted by P G(2, F ) (meaning the projective geometry of dimension 2 constructed from
the field F ).
It is very important to know when three points are on the same line, which the following Lemma
tells us.
Lemma 30. Let p1 = [v1 ] = [x1 , y1 , z1 ], p2 = [v2 ] = [x2 , y2 , z2 ] and p3 = [v3 ] = [x3 , y3 , z3 ] be three
points of the projective plane P G(2, F ). Then the three points are collinear if and only if
x1 y1 z1
det x2 y2 z2 = 0.
x3 y3 z3
Proof. The three points are on a line if and only if v1 = v2 + v3 for some and in F . But it
is precisely when this holds that the above determinant is zero.
As we are mostly interested in finite projective planes here, the following theorem is very im-
portant. When F is a finite field, it is customary to write P G(2, q) where the prime power q is the
order of the field, as opposed to writing P G(2, Fq ) (where Fq denotes the finite field of order q).
Proof. We will prove this by counting the number of points of any line in P G(2, q). Any line of
P G(2, q) corresponds to a plane through the origin of F3q , that is a set of the form L = {v + w |
, Fq }, where v and w are vectors of F3q . Let this set correspond to the line l. The points
of P G(2, q) that are on l are the points of the form [v + w], and not both zero. As
before, [v + w] = [0 v + 0 w] if and only if = r0 and = r 0 for some r Fq . So, if
6= 1, we have [v + w] = [v + 0 w] where 0 = 1 , and each distinct 0 corresponds to
a distinct point of l. Thus there are q points that can be written in this way. If = 0 then
[v + w] = [0 v + 1 w] = [w], which is just one point. Since all points of l are accounted for
in this way, we conclude that l has q + 1 points and P G(2, q) is thus of order q.
Example 32 We will construct the projective plane of order 2 from F2 , i.e. the plane P G(2, 2). F32
consists of the 8 vectors (0, 0, 0), (0, 0, 1), (0, 1, 0), (0, 1, 1), (1, 0, 0), (1, 0, 1), (1, 1, 0) and (1, 1, 1).
Any line through the origin consist of just two points, of which the origin is one. We can thus
identify the lines through the origin with this point, and we can identify any point of P G(2, 2)
with the non-zero vector spanning the corresponding line.
2.2. COLLINEATIONS OF PROJECTIVE PLANES 11
(0, 1, 0)
(1, 1, 0) (0, 1, 1)
(1, 1, 1)
Figure 2.2: The projective plane of order 2 constructed from the finite field of order 2.
Note that in a given plane, the sum of two of the non-zero vectors is the third non-zero vector.
As we have identified the points of P G(2, 2) with the non-zero vectors of F32 , we see that the a line
of P G(2, 2) consists of the three non-zero elements of the corresponding plane (see figure 2.2).
Theorem 33. There exists a projective plane of order q for each prime power q.
Proof. By Theorem 31 the projective plane constructed from the finite field of order q has order
q. Furthermore, as there exists a finite field for each prime power, the result follows.
Example 37 Let be the projective plane of Example 11. Define by (x, y) = (x + k, y) for
any ordinary point (x, y), where k is a non-zero real number, and (z) = z for points on the line
at infinity. Then is a collineation of .
12 CHAPTER 2. COLLINEATIONS AND CONSTRUCTIONS OF PROJECTIVE PLANES
Theorem 38. The set of collineations of a projective plane = (P, L) form a group under
composition.
Proof. First we must show that the composition of two collineations is again a collineation. So,
let and be two collineations of , and let l L be any line of . Then, as is a collineation,
(l) L. Again, as is a collineation and (l) L we have that ((l)) L, so is indeed a
collineation. Let , and be three collineations of , and let p be any point of . Then
( ) (p) = ((p)) = (((p))) = ( (p)) = ( )(p),
so composition is associative. Now, be any collineation of , let be the identity collineation,
and let p be any point of . Then
(p) = ((p)) = (p) = ((p)) = (p),
so works as an identity. Finally, as any collineation is bijective, each collineation has an inverse
1 . We have, for all points p of ,
1 (p) = (1 (p)) = p = (p)
1 (p) = 1 ((p)) = p = (p).
But we must also show that 1 (l) L for any line l, i.e. that 1 is indeed a collineation. So, let
p1 and p2 be any two points of l, and let p01 = 1 (p1 ), p02 = 1 (p2 ) and let l0 be the line p01 p02 .
As is a collineation, we must have (l0 ) = l. But then
1 (l) = 1 ((l0 )) = l0 ,
and 1 is a collineation. Hence 1 acts as an inverse of any collineation of , and the set of
collineations of form a group under composition.
Example 39 We shall determine the order of the collineation group of the Fano plane. We will
do this by counting the ways of defining a bijection of the points of the plane such that it is a
collineation. First of all, there are 7 ways of choosing (1), and once that is done there are 6 ways
of choosing (2). Now, as should be a collineation, (3) must be the third point on the line
determined by (1) and (2). As there are 4 points left, there are 4 ways of choosing (4). (5)
must be remaining point on the line determined by (1) and (4), (6) must be the remaining
point on the line determined by (2) and (4), and (7) must be the remaining point on the line
determined by (3) and (4). Hence there are 7 6 4 = 168 ways of choosing , and the order
of the collineation group is 168.
c
p
f ( p1) f ( p2) 3
f ( p2 ) f ( p1)
Axis c' Axis
c
l'
l1 l2
p l1 l3
2
p p p
1 2 l2
1
f ( p3)
(a) (b)
Figure 2.3: (a) Every axial collineation has a center. (b) Assuming that there is a point p3 such that the line on
both p3 and (p3 ) does not contain c yields that the point p2 is fixed.
Lemma 43. Let be an axial collineation. Suppose that fixes a point p that is not on the axis.
Then is also a central collineation with p as center. Dually, if a central collineation fixes a line
not on the center, it is also an axial collineation.
Proof. We must show that each line on p is fixed by . Any line on p intersects the axis in precisely
one point, and both this point and p are fixed, so the line must also be fixed. Thus any line on p
is fixed by and is a central collineation.
Proof. The identity collineation is obviously both central and axial, so suppose that is an axial
collineation that is not the identity collineation. We must show that has a center, that is, a point
whose lines are fixed by .
Let p1 and p2 be two points not on the axis. If either p1 or p2 are fixed by , the by Lemma
43 is a central collineation. So suppose that (p1 ) 6= p1 and (p2 ) 6= p2 . Let the line on p1 and
(p1 ) be called l1 , and the line on p2 and (p2 ) be called l2 (see figure 2.3 (a)). As the point where
l1 intersects the axis is fixed by , l1 must also be fixed by . Similarly l2 is fixed by . Let c be
the point of intersection of l1 and l2 . As both l1 and l2 are fixed by , c must also be fixed by .
If c is not on the axis, by Lemma 43 is a central collineation.
So suppose that c is on the axis. If all lines on c are fixed, is a central collineation with center
c. So, for a contradiction, suppose that there is a line on c that is not fixed by , and let p3 be a
point on this line different from c. If p3 is fixed by then is a central collineation by Lemma
43, so assume that p3 is not fixed by . Let the line on p3 and (p3 ) be l3 , and the point where l3
intersects the axis be c0 (see figure 2.3 (b)). As c0 is fixed by the line l3 is fixed by , as above.
As the line l1 is also fixed by , the point of intersection of l1 and l3 (which is not on the axis) is
fixed by . Let l0 be the line on p2 and l1 l3 . As the point l1 l3 is fixed by , and the point where
l0 intersects the axis is fixed by , l0 must also be fixed by . But then, as l2 is fixed by , p2 is
fixed by , contradicting our previous assumption that p2 was not fixed. Hence all lines on c are
fixed by so is a central collineation, which completes the proof.
The axis of Example 36 is the line {3, 5, 6}, and the center of Example 37 is the point (0). In
both these cases the center is a point on the axis. As was hinted in the proof of Theorem 44 this is
not always the case. For example, the collineation of the projective plane of Example 11 defined
by (x, y) = (kx, ky) for any ordinary point, where k R, and (z) = z for any point on the line
at infinity has the line at infinity as axis, and the origin as center.
A collineation with center c and axis l is called a (c, l)-collineation. Is a (c, l)-collineation uniquely
determined by its center and axis? The answer is no. For example, the collineation of Example 37
has the same center and axis for any value of k. However, the following theorem does say when a
(c, l)-collineation is completely determined.
Theorem 45. A (c, l)-collineation is completely determined by its center, its axis and its action
on one point not on the axis or the center.
14 CHAPTER 2. COLLINEATIONS AND CONSTRUCTIONS OF PROJECTIVE PLANES
p
p f( )p f( )q
l r l
c r f( )p q
f( )q
q
c
(a) (b)
Figure 2.4: (a) The center is on the axis. (b) The center is not on the axis.
Proof. Let be a (c, l)-collineation and let p be any point other than the center not on the axis
such that (p) is known. Let q be any point other than the p, (p), the center, and not on the
axis. Let r be the point of intersection of the line pq and the axis. As r is fixed by , the line on
(p) and (q) is on r. The line cq is also fixed by , as c is the center. Hence (q) is the point of
intersection of the lines cq and r(p), which are both known, and we are done (see figure 2.4).
Corollary 46. A (c, l)-collineation that fixes any point other than c, not on the axis, is the
identity collineation.
Proof. Suppose there is a point p other than c not on the axis, that is fixed by . Let q be any
other point different from c not on the axis. Then, as there are two fixed point on the line qp (p
and the point of intersection with the axis), the line qp is fixed. Then q is mapped to the point of
intersection of qp and cq, which are both fixed. Hence q is fixed, and is the identity function.
Theorem 47. The set of all (c, l)-collineations of a projective plane form a group under composi-
tion of functions.
Proof. The set of (c, l)-collineations of a projective plane is obviously a subset of the set of all
collineations of , which by Theorem 38 is a group under composition of functions.
The identity is certainly a (c, l)-collineation. If two collineations and fix all lines on c and
all points of l, then it is obvious that and also fix all lines on c and all points of l.
Finally, we need to show that the inverse of a (c, l)-collineation is again a (c, l)-collineation. So,
let be a (c, l)-collineation. As is bijective, and fixes c and all points of l, 1 also fixes c and
all points of l. We must also show that 1 fixes all lines on c.
So suppose, for a contradiction, that there is a line l0 on c such that 1 (l0 ) 6= l0 . But then, as
0
l is fixed by ,
l0 = 1 ((l0 )) = 1 (l0 ) 6= l0 ,
a contradiction. Hence 1 fixes all lines on l and is indeed a (c, l)-collineation, which completes
the proof.
Example 48 The only (c, l)-collineation of the Fano plane where c / l is the identity. This follows
from the fact that any line on the center is also on the axis. Thus any line on the center has two
of its points fixed, which implies that the third point of this line is also fixed. As all points are on
a line on the center, all points of the plane must be fixed. Thus the group of (c, l)-collineations of
the Fano plane where c / l is the trivial group, consisting of only the identity.
The group of (c, l)-collineations of the Fano plane where c l however is not trivial. Any point
not on the axis can either be fixed by the collineation (in which case the collineation is the identity),
or it can be mapped to the third point of the line on this point and the center. Hence there are
only two elements in this group and it is thus isomorphic to Z2 .
2.3. THE DESARGUES CONFIGURATION 15
Definition 49 ((c, l)-transitive). A projective plane is said to be (c, l)-transitive if, for any to
points p, and p0 such that neither is in {c}l and p0 is on the line cp, there exists a (c, l)-collineation
mapping p to p0 .
The Fano plane is easily seen to be (c, l)-transitive.
Example 50 Let be the projective real plane, and let c = (0) and l be the line at infinity. We
shall show that is (c, l)-transitive. We must construct a (c, l)-collineation which, for any two
points p and p0 different from c and not on l such that p0 is on the line cp, maps p to p0 .
So, let p and p0 be any two points such that p0 is on the line cp. Then both p and p0 are on a
line with equation y = m, with the point (0) added. Thus, if p = (x, y), then p0 = (x + n, y) for
some n R. Hence, any point q = (x0 , y 0 ) is mapped to (x0 + n, y 0 ). As all points on the line at
infinity are fixed, the collineation is now defined for all points of , and we know from Example
37 that this is indeed a collineation. Thus is (c, l)-transitive.
y
z
x
u l w
v
x'
z'
y'
We will conclude this chapter with a theorem that shows that projective planes constructed from
fields are in fact always Desarguesian.
Theorem 54. Any projective plane P G(2, F ) constructed from a field F is Desarguesian.
Proof. As was noted previously, P G(2, 2) is Desarguesian, so we can assume that F 6= F2 and
we can choose the 10 points as required. Let = P G(2, F ). Let c, x, x0 , y, y 0 , z, z 0 , u, v and w be
chosen as in Figure 2.5. We must show that u, v and w are collinear. The four points c, x, y and z
form a quadrangle of , and as was noted in the beginning of this chapter these can be assumed
to have coordinates [1, 1, 1], [1, 0, 0], [0, 1, 0] and [0, 0, 1] respectively.
As x0 is on the line cx, we have that x0 = [1, 1, 1] + [1, 0, 0] = [ + , , ] = [r, 1, 1] for some
r F , as is not zero (since x0 6= c). Similarly, y 0 = [1, s, 1] and z 0 = [1, 1, t] for some s, t F .
As u is on the line xy, we have that u = [1, 0, 0] + [0, 1, 0] = [1, a, 0] for some a F . On the
other hand, u is also on the line x0 y 0 , so
u = 0 [r, 1, 1] + 0 [1, s, 1] = [ 0 r + 0 , 0 + 0 s, 0 + 0 ].
In the last chapter we saw that fields can be used to construct projective planes. If the field is
finite the projective plane is also finite, and the order of the plane is the same as the order of the
field. Thus by Galois Theorem, there exists a projective plane of order q for any primer power q.
The purpose of this essay is to consider for which orders projective planes exist. This would be
an easy task if all projective planes could be constructed from fields. Then there would exist a
unique projective plane for each prime power, and no other finite plane would exist. However, this
is not the case. For example, it is known that there exist four non-isomorphic projective planes of
order nine (see, for example [4]).
The strongest known theorem on the existence of projective planes is Theorem 33. The strongest
theorem on non-existence is the Bruck-Ryser theorem. This chapter will be entirely devoted to
the proof of this theorem. The proof relies on the algebraic properties of the incidence matrix of
a projective plane, and not so much on the geometric properties of projective planes. This makes
this proof a lot different from the proofs seen earlier in this essay.
For the proof, we will need some results from number theory, so we will start with those. The
proofs of the first two are basically just long calculations. They can be found in Appendix A. The
proof of Lemma 58 is similar to the proof of Lemma 57, but is a bit more complex, so to not get
stuck on all this number theory, its proof can be found in Appendix B.
Lemma 55 (The Four-Squares Identity). If b1 , b2 , b3 , b4 , x1 , x2 , x3 , x4 are integers, then
(b21 + b22 + b23 + b24 )(x21 + x22 + x23 + x44 ) = y12 + y22 + y32 + y42 (3.1)
17
18 CHAPTER 3. THE BRUCK-RYSER THEOREM
This means that if not all the bs are zero, we can solve equation (3.2) for the xs.
Lemma 57. If p is an odd prime, and there exists integers x1 and x2 such that x21 +x22 0 (mod p),
then p is the sum of two integer squares.
Proof. Since x21 + x22 0 (mod p), we have that rp = x21 + x22 for some positive integer r. Take
such an expression that makes r the smallest such number. The lemma is proven if we can show
that r = 1.
Well use contradiction, so assume that r > 1. Now choose u1 and u2 such that u1 x1 (mod r),
u2 x1 (mod r) and |ui | r/2 for i = 1, 2. This can be done, as of all integers n, r/2 n
r/2, one has to equal x1 (mod r), and one has to equal x2 (mod r).
Then
u21 + u22 x21 + x22 0 (mod r),
so rs = u21 + u22 for some positive integer s. Since
rs u2 + u22 (r/2)2 + (r/2)2
s= = 1 = r/2,
r r r
we have that s < r. Now,
r2 sp = (rp)(rs) = (x21 + x22 )(u21 + u22 ) = (x1 u1 x2 u2 )2 + (x1 u2 + x2 u1 )2 (3.6)
by the two-squares identity. By the way u1 and u2 were defined, we also have
x1 u1 x2 u2 x21 + x22 0 (mod r)
and
x1 u2 + x2 u1 x1 x2 + x2 x1 0 (mod r),
so both terms of the right side of (3.6) has a factor r2 . Hence,
sp = y12 + y22 ,
where y1 = (x1 u1 x2 u2 )/r, y2 = (x1 u2 + x2 u1 )/r. Hence, s is an integer less than r, such that sp
is the sum of two integer squares. But r was assumed to be the least such integer, hence we have
a contradiction. Our assumption that r > 1 must be false, so r = 1. This proves the lemma.
Lemma 58 (Lagrange). Any positive integer can be written as the sum of four squares.
As was pointed out before, the proof of the Theorem of Lagrange can be found in Appendix B.
Lemma 59. For any integer n, if the equation x2 + y 2 = nz 2 has an integer solution with x, y, z
not all zero, then n is the sum of two squares.
Proof. Suppose that there exists integer x, y, z and satisfying the equation. We may assume that
x, y and z has no common factors, as any common factor can be canceled from the equation. We
will first prove it in the case where n is a square free integer, as the general case then easily follows.
So, assume that n is a square free number, say n = p1 p2 pk , where each pi are distinct. No pi
divides both x and y, as then either n would be divisible by p2i (which it is not), or z would be
divisible by pi (which contradicts the fact that x, y and z has no common factors). Now, as we can
rewrite the equation as
x2 + y 2 = p1 p2 pk z 2 ,
we see that x2 + y 2 0 (mod pi ) for all pi . Then, by Lemma 57, all pi are sums of two squares.
Let pi = x2i + yi2 , then
n = p1 p2 pk = (x21 + y12 )(x22 + y22 ) (x2k + yk2 ).
By successively using the two-squares identity k 1 times, we can reduce the right hand side to a
sum of two integer squares. Hence the Lemma holds when n is square free.
Now, assume that n is not a square free integer, say n = mu2 , where m is square free. Then we
can rewrite the equation as
x2 + y 2 = m(uz)2 .
As m is square free, by what we previously proved, m can be written as the sum of two integer
squares, say m = r2 + s2 . Then n = mu2 = (r2 + s2 )u2 = (ru)2 + (su)2 , so n is the sum of two
integer squares, and the proof is complete.
19
Now we have all the number theoretic machinery needed to prove the Bruck-Ryser theorem. The
theorem itself may seem rather technical, but it has an important corollary, given after the proof.
z 2 = nx2 + (1)n(n+1)/2 y 2
Proof. Suppose a projective plane of order n exists. Let N = n2 + n + 1 be the number of points
and lines in the plane. Let the incidence matrix of this plane be A, which is a N N -matrix.
Now choose x = (x1 , . . . , xN ) where xi Q for 1 i N , and let z = (z1 , . . . , zN ) where
z = Ax, which expresses the zs as a linear combination of the xs. By Lemma 28 we have that
We note that
z1
zT z = (z1 , . . . , zn ) ... = z12 + + zN
2
,
zN
x1
nxT x = n(x1 , . . . , xN ) ... = n(x21 + x2N )
xN
and
1 ... 1 x1
xT Jx = (x1 , . . . , xN ) ... . . . ... ...
1 ... 1 xN
x1
= (x1 + + xN , . . . , x1 + + xN ) ...
xN
= (x1 + + xN )2 .
z12 + + zN
2
= n(x21 + + x2N ) + 2 (3.8)
where = x1 + + xN and the zs are integer linear combinations of the xs. By Lemma 58, n
can be written as the sum of four squares, so we can write n = b21 + b22 + b23 + b24 where the bs are
integers. Now we can use Lemma 55 on the bs and the xs, four at a time, but this may leave som
xs if N is not divisible by four. What possibilities do we have for N ? We get two cases: if n = 0, 4
(mod 4) then N = 1 (mod 4), if n = 1, 2 (mod 4) then N = 3 (mod 4). We start by looking at the
latter case.
So, suppose that n = 1, 2 (mod 4) so that N = 3 (mod 4). Choose xN +1 to be a rational number,
and add nx2N +1 to both sides of equation (3.8). We then get
z12 + + zN
2
+ nx2N +1 = (b21 + b22 + b23 + b24 )(x21 + + x2N + x2N +1 ) + 2 . (3.9)
Note that N + 1 = 0 (mod 4), so we can use Lemma 55 on the bs and xs four at a time to get
z12 + . . . + zN
2
+ nx2N +1 = y12 + . . . + yN
2 2
+1 + , (3.10)
where the ys are integer linear combinations of the xs. Let y = (y1 , . . . , yN +1 ) and let x0 =
(x1 , . . . , xN , xN +1 ). As the ys are integer linear combinations of the xs, there exists a (N + 1)
(N + 1)-matrix B with integer entries such that
y = Bx0 . (3.11)
20 CHAPTER 3. THE BRUCK-RYSER THEOREM
By Lemma 56, B is invertible. Thus, x0 = B 1 y. Let A0 be the matrix A with one more columna
at the far right with all entries zero. A0 is then a N (N + 1)-matrix such that A0 x0 = Ax. Then
z = Ax = A0 x0 = A0 B 1 y = Cy, (3.12)
where the matrix C has rational entries. As the xs were chosen arbitrarily, and there is a one-
to-one correspondence between the xs and the ys by the matrix B, we could see it instead of as
choosing the ys. Furthermore, as the zs are linear combinations of the ys, the following way of
choosing the ys enables us to cancel most of the terms in equation (3.10).
As the zs are rational linear combinations of the ys, z1 = c11 y1 + + c1(N +1) yN +1 for some
rational numbers c11 , . . . , c1(N +1) . If c11 6= 1 we can set
1
y1 = (c12 y2 + + c1(N +1) yN +1 )
1 c11
so that
c11
z1 = (c12 y2 + + c1(N +1) yN +1 ) + c12 y2 + + c1(N +1) yN +1
1c
c 11
11
= + 1 (c12 y2 + + c1(N +1) yN +1 )
1 c11
1
= (c12 y2 + + c1(N +1) yN +1 ) = y1 .
1 c11
If c11 = 1 we can set
1
z1 = (c12 y2 + + c1(N +1) yN +1 )
2
so that
1
z1 = (c12 y2 + + c1(N +1) yN +1 ) + c12 y2 + + c1(N +1) yN +1
2
1
= (c12 y2 + + c1(N +1) yN +1 ) = y1 .
2
In any case, y1 is expressed in terms of the other ys, and z12 = y12 . By continuing like this, we can
choose the ys such that yi2 = zi2 for i = 1, . . . , N . Equation (3.10) is then reduced to
nx2N +1 = yN
2 2
+1 + , (3.13)
z12 + + zN
2
= (b21 + b22 + b23 + b24 )(x21 + + x2N 1 ) + nx2N + 2 (3.14)
z12 + + zN
2
= y12 + + yN
2 2
1 + nxN +
2
(3.15)
By proceeding as above, we can choose the ys in such a way that they cancel the first N 1 zs.
We then get
2
zN = nx2N + 2 , (3.16)
where zN and are rational linear combinations of the independent xN . As before, we can choose
an integer value of xN such that both zN and are integers. Cleaning up the notations a bit, we
have found that if a projective plane of order n 1, 2 (mod 4) then
z 2 = nx2 y 2 (3.17)
z 2 = nx2 + y 2 (3.18)
21
has integer solutions not all zero. Noting that if n 1, 2 (mod 4) then n(n + 1)/2 is an odd
number, and if n 0, 3 (mod 4) then n(n + 1)/2 is an even number, we can write equations (3.17)
and (3.18) as
z 2 = nx2 + (1)n(n+1)/2 y 2 . (3.19)
Thus, we have proven that if a projective plane of order n exists, then equation (3.19) has integer
solutions, not all zero.
We now sum up all we know about the existence of projective planes in the following important
proposition.
Proposition 61. Let n be a positive integer, then
A) if n is a prime power, at least one projective plane of order n exists.
B) if n 1 or 2 (mod 4), and is not the sum of two squares, then no projective planes of order
n exists.
C) In any other case, a projective plane of order n may or may not exist.
Proof. Part (A) follows from theorem 33. Part (B) is true, since if n = 1 or 2 (mod 4), then by
the Bruck-Ryser theorem the equation
nx2 = y 2 + z 2
has integer solutions. By Theorem 59 this implies that n can be written as a sum of two squares.
Hence, if n is not the sum of two squares, no projective plane of order n exists. Part (C) is
obvious.
It should be noted that the Bruck-Ryser Theorem does not say anything about the existence of
projective planes of order n when n 0 or 3 (mod 4), since the equation z 2 = nx2 + y 2 always has
the solution x = 0, y = z.
Example 62 Let us conclude this chapter by using Proposition 61 to see what planes of order
less than or equal to 25 does/doesnt/might exist. First of all, by part A of proposition 61, a plane
exist for orders 2, 3, 4, 5, 7, 8, 9, 11, 13, 16, 17, 19, 23 and 25, as these are all prime powers.
Now, lets check which orders are excluded by part B of proposition 61. The numbers that are
equal to 1 or 2 mod 4 that are not prime powers are 6, 10, 14, 18, 21 and 22. On the other hand,
the numbers (greater than one, less than or equal to 25) that are sums of two integer squares are
12 + 12 = 2 12 + 22 = 5
12 + 32 = 10 12 + 42 = 17
22 + 02 = 4 22 + 22 = 8
22 + 32 = 13 22 + 42 = 20
32 + 02 = 9 32 + 32 = 18
32 + 42 = 25 42 + 02 = 16
52 + 02 = 25
We see that the numbers that are excluded by part B of Proposition 61 are 6, 14, 21 and 22 as
these are not the sums of two integer squares. For all remaining integers, projective planes of that
order may, or may not exist. We summarize these results in table 3. In the cases where conclusions
can be made on the existence of a projective plane of a certain order, the reason is indicated in
the Comment column. PP indicates that the order is a prime power, and BR indicates that no
projective plane exist due to the Bruck-Ryser Theorem.
22 CHAPTER 3. THE BRUCK-RYSER THEOREM
Table 3.1: What can be said on the existence of projective planes of small order by Proposition 61.
Chapter 4
The final proposition of the last chapter sums up our results so far regarding the existence/nonexistence
of finite projective planes. As we saw in Example 62 there are a lot of orders for which we have
no way of telling whether or not a plane of that order exists. The first such order is 10. It was an
open question for a long time, but in 1989 it was shown by Lam, Thiel and Swiercz [13] [16] using
an extensive computer search that such a plane does not exist.
This final chapter will look at how this search was done, and what mathematical ideas was used.
We will start by looking at some theory needed to understand this.
Example 64 The binary code of the Fano plane of Example 10 is the vector space spanned by
[1, 1, 1, 0, 0, 0, 0], [1, 0, 0, 1, 1, 0, 0], [1, 0, 0, 0, 0, 1, 1], [0, 1, 0, 1, 0, 1, 0], [0, 1, 0, 0, 1, 0, 1], [0, 0, 1, 1, 0, 0, 1]
and [0, 0, 1, 0, 1, 1, 0]. Examples of code words are
Since the binary code of the Fano plane is a vector space, it is vital to investigate its dimension.
We shall return to this example later, when we have developed some more machinery. We will
then, among other things, determine its dimension.
Note that all operations are done in F2 . As all code words are linear combinations of the rows
of the incidence matrix, and all operations are in F2 , we can always assume that each row occurs
at most once in the sum.
Another observation that can be made is the following. Let x be a code word that is a sum of
the rows r1 , . . . , rm . Let aij be the value of row i in column j, and let bj be the value of x in
column j. Then
Xm
bj = aij (mod 2).
i=1
From this we conclude that x has a 1 in column j if and only if an odd number of the rows
r1 , . . . , rm has a 1 in column j, and a 0 otherwise.
23
24 CHAPTER 4. THE SEARCH FOR A PROJECTIVE PLANE OF ORDER 10
2 4
7
3 5
6
In the same way as a point is on a line if the row corresponding to the line has a 1 in the column
corresponding to the point, we say that a point is on a code word if the code word has a 1 in the
column corresponding to the point. For example, the points on the code word [0, 1, 1, 1, 1, 0, 0] of
Example 64 are the points 2, 3, 4 and 5.
Instead of saying that a code word is the sum of a set of rows, we will often say that the code
word is the sum of the corresponding set of lines. So, for example, we may write x = l1 + l2 + l3 ,
where the ls are lines, which will mean the same as x = r1 + r2 + r3 , where the rs are the rows
corresponding to the lines. By the previous paragraphs, a point will be on x if and only if it is on
an odd number of the lines l1 , l2 and l3 .
We will sometimes talk about the dot product of two lines, by which we will mean the dot
product of the corresponding rows. In the same way, we may talk about the dot product of two
code words, or of the dot product of a code word and a line. Note that the dot product of two
code words counts the number of columns that both code words has a 1 in, mod 2.
Example 65 Let C be the code of Example 64. Let x = l1 + l3 + l4 + l7 . In figure 4.1 the points
of the Fano plane has been drawn, together with the lines in the sum making up x. We easily see
that the points with an odd number of lines on them are 4, 5, 6 and 7, so x = [0, 0, 0, 1, 1, 1, 1]. The
reader is advised to check that indeed you get the same result when actually doing the calculation.
Note that there are other ways of writing x as a sum of lines. For example, x = l2 + l3 = l6 + l7 .
Definition 66 (Weight). The weight w(x) of a code word x is the number nonzero entries in
x.
As each code word can be identified with the points on it, each code word of weight i corresponds
to a set of i points. The code word of example 65 has weight 4, and as we have seen it corresponds
to the four points 4, 5, 6 and 7.
Lemma 68. Let x be a code word, and l any line of . Then w(x + l) = w(x) + n + 1 2k, where
n is the order of the plane and k is the number of points where l intersects x.
Proof. Since l has weight n + 1, the total number of nonzero columns of x and l is w(x) + n + 1.
However, in the code word x + l, each of the columns that x and l has a 1 in common will have a
0, and these columns correspond precisely to the points of intersection of x and l. So each of these
points will remove one 1 from x and one 1 from l, and we get the formula.
Lemma 69. Let x = l1 + + lk be a code word in the binary code of a projective plane , where
the li s are lines of , and let l be any line of , different from the li s. If k is even then l intersects
x in an even number of points, and if k is odd then l intersects x in an odd number of points.
4.1. THE CONNECTION WITH CODING THEORY 25
Also, as we have argued earlier, the line l intersects x in a point pi if and only if L(pi ) is odd.
Furthermore, l does not intersect x in any other points, by the way we have defined pi .
Now suppose that k is even. Then by equation (4.1) an even number of the L(pi )s must be
odd. As l intersects x in precisely the points where L(pi ) is odd, l intersects x in an even number
of points.
Now suppose that k is odd. Then by equation (4.1) an odd number of the L(pi )s must be odd.
As above, l intersects x in an odd number of points.
Lemma 69 can quite readily be extended to the case where the line is actually one of the lines
in the sum of the code word.
Corollary 70. Let x be a code word in the binary code of a finite projective plane , and let l be
any line of . Then if x is the sum of an odd number of lines, l intersects x in an even number of
points, and if x is the sum of an odd number of lines, l intersects x in an odd number of points.
Proof. The case we need to consider here, which is not covered by Lemma 69, is the case where
l is one of the lines in the sum for x. So suppose that x = l1 + + lk , and let l = li for some
i, 0 i k. Let x0 = l1 + + li1 + li+1 + + lk . Then x = x0 + l, and the points of l that is
on x is precisely those points of l that is not on x0 . As any line of a projective plane always has an
odd number of points, l is on an odd number of points of x if and only if l is on an even number
of points of x0 , and vice versa.
As x0 is the sum of k 1 lines, and none of these are l, Lemma 69 shows that l intersects x0 in an
odd number of points if k is even, and an even number of points if k is odd. Hence l intersects x in
an odd number of points if k is odd, and an even number of points if k is even, and the corollary
is proven.
Corollary 71. Let x be a code word of the binary code of a finite projective plane . Then either
all sums of lines that equal x are sums of an even number of lines, or all sums are sums of an odd
number of lines.
Proof. If x could be written as both the sum of an even number of lines and an odd number of
lines, by Corollary 70 any line would intersect x in both an odd and an even number of points,
which is impossible.
Definition 72 (Orthogonal dual code). Let C be the binary code of a projective plane of
order n. The orthogonal dual code C is the subspace of FN
2 such that
C = {u FN
2 | u v = 0 for all v C},
where N = n2 + n + 1.
From linear algebra, we know that dim(C ) = N dim(C).
Lemma 73. Let C be the binary code of a finite projective plane . The elements of C that are
also in C are precisely the code words x that is a sum of an even number of lines.
Proof. First, let x0 be the sum of two lines. Let l be any line of . Then, by Lemma 69 l intersects
x0 in an even number of points, i.e. the number of columns that both has a 1 in is even. Hence
x0 l = 0. Now let y = l1 + + lk be any code word of C. Then
x0 y = x0 (l1 + + lk ) = x0 l1 + + x0 lk = 0 + + 0 = 0,
Now, let x be a code word that is the sum of an even number of lines. then we can write
x = x1 + xm , where each xi is the sum of two lines. Let y be any code word of C. Then
x y = (x1 + + xm ) y = x1 y + + xm y = 0 + + 0 = 0,
so x C .
Now, on the other hand, let x be the sum of an odd number of lines. We can write x = x0 + l
where x0 is a code word that is the sum of an even number of lines, and l is a line. Let l0 be any
line of other than l. Then, we get
x l0 = (x0 + l) l0 = x0 l0 + l l0 = 0 + 1 = 1,
Lemma 74. Let C be the binary code of a finite projective plane . Then C C is of codimension
1 in C, that is, dim(C C ) = dim(C) 1.
Proof. First of all we note that, for any two lines l, l0 of we have l l0 = 1 as they intersect in
one point. So for any line l of we have that l / C . Hence dim(C C ) < dim(C).
Now, suppose that dim(C) = k. As the lines of span C there exists a subset of k lines that is
a base for C. Let this base be B = {l1 , . . . , lk }. Consider the set D = {l1 + l2 , l1 + l3 , . . . , l1 + lk }.
Note that |D| = k 1. As B is linearly independent (as it is a base), and
we see that
2 (l1 + l2 ) + 3 (l1 + l3 ) + + k (l1 + lk ) = 0
if and only if i = 0 for 2 i k. Thus D is also linearly independent. Hence D is a base for a
vector space of dimension k 1, that we will call L. Obviously, L C, and as each element of L
is the sum of an even set of lines L C by Lemma 73. So, L C C .
Thus, dim(L) = k 1 dim(C C ) < dim(C) = k, and this inequality can only be satisfied
if dim(C C ) = dim(C) 1, and we are done.
Lemma 75. Let C be the binary code of a projective plane of order n, where n is divisible by 2
exactly once. Then C C and
n2 + n + 2
dim(C) =
2
and
n2 + n
dim(C ) = .
2
Proof. Let A be the incidence matrix of . Let N = n2 + n + 1 be the number of lines/points of
. By Lemma 29
2
det(A) = (n + 1)n(n +n)/2
.
The operations used to transform A into A2 has all been elementary operations that preserve the
absolute value of the determinant, so | det(A)| = | det(A2 )|. Furthermore, by cofactor expanding
A2 about the rightmost column, we see that det(A2 ) = (n + 1) det(B2 ).
4.1. THE CONNECTION WITH CODING THEORY 27
As A2 is invertible, B2 must also be invertible. We can thus act on B2 using the elementary
operations to put it in diagonal form, such that its determinant is preserved. We get
h1 0 0
h2 0
..
A3 = U AV =
. .. ,
.
0 hN 1 0
n+1
where U and V are unimodal matrices, that is, matrices with determinant 1. The absolute value
of the determinant is still preserved, so we have | det(A)| = | det(A3 )|. Furthermore, det(A3 ) =
h1 h2 hN 1 (n + 1). As N 1 = n2 + n, we have
2
h1 h2 hn2 +n = n(n +n)/2
.
As n is divisible by 2 exactly once, it follows that at most (n2 + n)/2 of the hs are divisible by 2.
Considering A3 as a matrix over F2 , precisely those rows i where hi is divisible by 2 will consist
of only zeros, and the set of all nonzero rows will be linearly independent. Hence the rows of A3
over F2 spans a vector space of dimension at least n2 + n + 1 (n2 + n)/2 = (n2 + n + 2)/2.
As U and V are unimodal matrices, A has the same rank as A3 . Thus
By Lemma 74, dim(C ) dim(C) 1 = (n2 + n)/2. Since dim(C) + dim(C ) = n2 + n + 1, these
inequalities imply that
dim(C) = (n2 + n)/2
and
dim(C ) + 1 = (n2 + n + 2)/2.
By Lemma 74, dim(C C ) = dim(C) 1, so in fact C C = C , which implies that C C,
and we are done.
Example 76 We can now use Lemma 75 to calculate the entire binary code of the Fano plane.
The Fano plane is of order 2, which is divisible by 2 exactly once, so the lemma applies. Hence,
dim(C) = (4 + 2 + 2)/2 = 4. Thus, each base has four elements, and the number of linear
combinations of these over F2 is 24 = 16, so |C| = 16 and there are 16 code words in C. Also,
dim(C = (4 + 2)/2 = 3, so |C | = 23 = 8, and these eight code words are also in C. We will now
calculate them.
A base for C can easily be found, one is {l1 , l2 , l3 , l4 }, where, as before, l1 = [1, 1, 1, 0, 0, 0, 0],
l2 = [1, 0, 0, 1, 1, 0, 0], l3 = [1, 0, 0, 0, 0, 1, 1], l4 = [0, 1, 0, 1, 0, 1, 0]. Using this base we can calculate
all code words of C. However, we will do it in a slightly different way, that will show us more of
the actual structure of C. Obviously, the zero code word 0 is in C, and as the sum of all lines
contains all points (as then each point is on three lines, which is odd) the unit vector 1 (having all
entries 1) is also in C. For any line l, the code word 1 + l will contain precisely the points that is
not on l. As there are seven lines, there are seven such complementary code words. Together with
the zero vector and the unit vector, this is 16 code words, so all code words of C is accounted for
this way. Hence, any code word of C is either the unit code word, a line, or the complement of
any of these.
The following list lists the code words, where the unit code word and the lines are in the left
column, and the zero code word and complements of the lines are in the right column.
[1, 1, 1, 1, 1, 1, 1] [0, 0, 0, 0, 0, 0, 0]
[1, 1, 1, 0, 0, 0, 0] [0, 0, 0, 1, 1, 1, 1]
[1, 0, 0, 1, 1, 0, 0] [0, 1, 1, 0, 0, 1, 1]
[1, 0, 0, 0, 0, 1, 1] [0, 1, 1, 1, 1, 0, 0]
[0, 1, 0, 1, 0, 1, 0] [1, 0, 1, 0, 1, 0, 1]
[0, 1, 0, 0, 1, 0, 1] [1, 0, 1, 1, 0, 1, 0]
[0, 0, 1, 1, 0, 0, 1] [1, 1, 0, 0, 1, 1, 0]
[0, 0, 1, 0, 1, 1, 0] [1, 1, 0, 1, 0, 0, 1]
28 CHAPTER 4. THE SEARCH FOR A PROJECTIVE PLANE OF ORDER 10
Eight of these code words are in C . Obviously, 0 C and 1 / C . Furthermore, as each line l
is the sum of an odd (1!) number of lines, none of these are in C . As there should be 8 elements
in C , the seven code words not yet accounted for must be in C . Thus C is the zero vector
together with the complements of the lines (the right column in the list above).
where N = n2 +n+1, and Ai is the number of code words of weight i in C. The weight-enumerator
of the orthogonal dual of C is defined in the same way.
Example 78 Let us calculate the weight-enumerator of the code C and C of the Fano plane.
By the previous example, we see that A0 = A7 = 1, A1 = A2 = A5 = A6 = 0 and A3 = A4 = 7.
Hence
WC (x, y) = x7 + 7x4 y 3 + 7x3 y 4 + y 7 .
For C , we have A0 = 1, A1 = A2 = A3 = A5 = A6 = A7 = 0 and A4 = 7. Thus
WC (x, y) = x7 + 7x3 y 4 .
The following important identity will be given without proof, as the proof requires more profound
results from Coding Theory. For a proof, see for example Hall [15].
1
WC (x, y) = WC (x + y, x y).
|C|
Lemma 80. Let x be a code word of C. Then w(x) 0 (mod 4) if x is the sum of an even number
of lines, and w(x) 3 (mod 4) if x is the sum of an odd number of lines.
Proof. We will use induction on the number of lines in the sum equal to x. We start by observing
that the zero code word (which is the sum of an even number of lines) has weight 0 (mod 4), and
that any line (which is the sum of an odd number of lines) has weight 11 3 (mod 4).
Now, suppose that x is the sum of an even number of lines, and that w(x) 0 (mod 4). We
must show that w(x + l) 3 (mod 4) for any line l in . By Corollary 70, l intersects x in an even
number of points, so let this number be 2k. By Lemma 68 we have that
On the other hand, suppose that x is the sum of an odd number of lines, let this number be
2k + 1, and that w(x) 3 (mod 4). Then we get
As the code words of weight 0 (mod 4) are sums of even sets of lines, each line intersects such
code words in an even number of points, and as the code words of weight 3 (mod 4) are sums of
odd sets of lines, each line intersects such code words in an odd number of points.
By Lemma 80 we see that for the coefficients of the weight enumerator A4i+1 = A4i+2 = 0 for
i = 0, . . . , 27. Also, the unit code word is in C as the sum of all lines contains all points, so for
each code word x of weight w(x) in C, there is a code word x + 1 of weight 111 w(x) in C, so
Ai = A111i . Hence the 28 numbers A4i , i = 1, . . . , 27 determines the entire weight enumerator.
So
27
X 27
X
WC (x, y) = A4i x1114i y 4i + A4i+3 x1084i y 4i+3 . (4.2)
i=0 i=0
As C is precisely the code words of C that is the sum of an even set of lines, they are precisely
the code words of even weight. Thus
27
X
WC (x, y) = A4i x1114i y 4i .
i=0
Now, suppose there exists a code word x of weight 3 or 7. Let p be a point not on x. As p is
not on x, and x has weight less than 11, it follows that there exists a line on p that is not on x.
But this contradicts the fact that each line intersects x in an odd number of points. We have a
contradiction, and thus no code word of weigh 3 or 7 exist, and A3 = A7 = 0.
Suppose that there exists a code word x of weight 4 or 8, and let p be a point on x. Then, as
there are 11 lines on p, there exists a line on p that is not on any of the other points of x. Thus
this line intersect x in exactly one point, which contradicts the fact that each line intersects x in
an even number of points. Hence no code word of weight 4 or 8 exist, and A4 = A8 = 0.
Finally, we will determine A11 . Any line is a code word of weight 11, so A11 111. Now, let
x be a code word of weight 11, and p1 and p2 be two points of x. Let l be the line on p1 and
p2 . Suppose that there is a point q on l that is not on x. As there are at most 9 more points of
x not on l, and there are 10 other lines through q, there must be a line l0 on q that contains no
point of x. But all lines intersects x in an odd number of points, which l0 does not, and we have a
contradiction. Thus the code words of weight 11 are precisely the lines, and A11 = 111.
Obviously, the only code word of weight 0 is the zero code word, so A0 = 1. Summing up, we
have found that A0 = 1, A1 = A2 = = A10 = 0 and A11 = 111.
Using the McWilliams identity, we find that
27
X 1
A4i x1114i y 4i = WC (x, y) = WC (x + y, x y) =
i=0
|C|
! (4.3)
27 27
1 X
1114i 4i
X
1084i 4i
A4i (x + y) (x y) + A4i+3 (x + y) (x y) .
|C| i=0 i=0
Now, for any i, we can equate the coefficient of x1114i y 4i on the right hand side (when expanded
and factored properly) to A4i . This yields a linear equation system in the A4i s. This equation
system can not explicitly be solved for the A4i s, but using the fact that A0 , . . . , A12 is known it
can be shown that each A4i depends on just A12 , A15 and A16 . Hence if these three are found, the
entire weight-enumerator of C is determined.
This observation was made in 1973 by MacWilliams, Sloane and Thompson [5]. They used a
computer to search for possible code words of order 15, but none was found, so A15 = 0. The
method used was to look at how lines of the plane would intersect the code word, and what
structure this would give the incidence matrix. They then tried to complete this incidence matrix
into a complete incidence matrix of a projective plane. If a code word of order 15 existed in the
projective plane, this would be possible (this will all be explained in more detail in the next section).
However, no starting incidence matrix could be completed into a plane, and hence A15 = 0.
As early as 1957, Hughes [2] [3] pointed out that the only possible prime divisors of the order
of the collineation group of a projective plane of order 10 were 3, 5 and 11. In 1976, Whitesides
showed that 11 does not divide the order of the group [6], and in 1979 he showed that the full
collineation group would have order 1, 3 or 5 [7]. In the following year, Anstee, Hall and Thompson
[8] showed that the collineation group could not have order 5, and in 1981 Janko and van Trung [9]
showed that it could not have order 3. Thus the collineation group of the plane would have order
30 CHAPTER 4. THE SEARCH FOR A PROJECTIVE PLANE OF ORDER 10
1, so it would be trivial. All projective planes we have seen in this essay has been symmetric in all
sorts of ways, but if a projective plane of order 10 existed, it would have virtually no symmetry at
all. This would indeed be a very strange projective plane.
The idea now, was to calculate A12 and A16 , using similar techniques as those used for calculating
A15 . If a code word of weight 12 or 16 was found to complete to a plane, a plane has been found
and existence shown. If on the other hand no such code words was found (and A12 = A16 = 0),
the weight-enumerator of C could be calculated. Then a code word that should exist (i.e. Ai 6= 0)
should be possible to complete into a plane. If this was possible, a plane would be found, but if
this was not possible, this would contradict the fact that such a code word should exist. Hence
our assumption that a projective plane of order 10 exist must be false, and we have proven that
no projective plane of order 10 exist.
In 1983, Lam, Thiel, Swiercz and McKay finished a computer search for code words of weight 12
[10]. None was found, so A12 = 0. The program used a total of 183 days of CPU time. Three years
later, Lam, Thiel and Swiercz finished a search for code words of weight 16 [12]. Again, none was
found, so A16 = 0. Now the entire weight-enumerator was known, so the number of code words
was known for each weight. Table 4.1 lists the number of code words of each weight. All weight
where the number of code words are zero are excluded from the list. As Ai = A111i , they are
listed together.
1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 1 1 1 1 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 0 0
0 1 0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 1 0
0 0 1 0 0 0 1 0 0 0 1 0 0 0 0 0 0 1 1
Table 4.2: One of the 19 6-incidence matrices found by Lam, Crossfield and Thiel. If a code word of weight
19 corresponding to this matrix would exist in a projective plane of order 10, the top left corner of the incidence
matrix for the projective plane could be assumed to look like this.
Finally, each unordered pair of points of x define a line through x. There are 19
2 = 171 unordered
pairs of points of x. No single line is accounted for in this way. As any pair of the three points on
x of a triple line defines the line, and there are three such unordered pairs for each triple line, each
triple line is accounted for three times. As any pair of the five points on x of a heavy line defines
the line, and there are ten such unordered pairs for each heavy line, every heavy line is accounted
for ten times. Thus
3b3 + 10b5 = 171. (4.6)
By solving these three equations, we get
b1 = 6 b3 = 37 b5 = 68. (4.7)
Thus there are 6 heavy lines, 37 triple lines and 68 single lines for any code word x of weight 19
in . By permuting the rows and columns of the incidence matrix for , we can assume that the
19 points of the code word are the 19 leftmost columns, the 6 heavy lines are the 6 topmost rows,
the 37 triple lines are the next 37 rows, and the 68 single lines are 68 bottom rows (see table 4.3).
The incidence of the 19 points of x with the 6 heavy lines induces an 196-incidence matrix (the
sub matrix B1 of table 4.3), where the rows represents the lines and the columns represents the
points, as usual. We will call such a 19 6-matrix a starting point configuration. Lam, Crossfield
and Thiel found [11], using a computer search, that there are 66 non-isomorphic starting point
configurations, where two starting point configurations are isomorphic if one can be produced from
the other using only row/column permutations. If a code word of weight 19 exists in , it would
be possible to extend at least one of the starting point configurations to a 111 111-incidence
matrix representing the projective plane . Two of these non-isomorphic 19 6-matrices are given
in tables 4.2 and 4.5. These can be seen not to be isomorphic to each other by noting that the
one in table 4.2 has one column with four 1s and the one in table 4.5 has no column with four
1s in it. Permuting the rows of the matrix preserves the number of 1s in each column, and
permuting columns only moves the columns so there is no way of turning one of the starting point
configurations in to the other by permuting the rows and columns. Hence the two starting point
configurations are non-isomorphic.
Now, let p be any point of . Let c1 , c3 and c5 be the number of single, triple and heavy lines
on p, respectively. If p is a point of x, then the following relations hold.
c1 + c3 + c5 = 11, (4.8)
32 CHAPTER 4. THE SEARCH FOR A PROJECTIVE PLANE OF ORDER 10
19 92
6 B1 B2
37 B3
B4
68 B5 B6
Table 4.3: We can assume that the 19 points of the code word of weight 19 are the leftmost columns in the
incidence matrix, that the 6 heavy lines are the topmost rows, the 37 triple lines are the next 37 rows and the 68
single lines are the 68 rows at the bottom.
c1 c3 c5 c1 c3 c5
2 9 0 7 4 0
3 7 1 8 2 1
4 5 2 9 0 2
5 3 3
6 1 4
(a) (b)
Table 4.4: The solutions to equations (4.8) and (4.9) (a), and equations (4.10) and (4.11) (b).
c1 + c3 + c5 = 11, (4.10)
1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 1 1 1 1 0 0 0 0 0 0
0 1 0 0 0 1 0 0 0 1 0 0 0 1 1 0 0 0 0
0 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 1 0 0
0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 0 0 1 1
Table 4.5: A starting point configuration that can not be completed into a projective plane of order 10 as it implies
the existence of a code word of weight 16.
other to lines. Hence there are 4 points on each line that is neither in x nor on any of the other
two lines. Summing up, we see that the code word y has 16 points, which contradicts the fact that
there are no code words of weight 16. So this starting point configuration can not be completed
into a complete plane.
Four other starting point configurations can be eliminated by the same argument. Also, four
other starting point configurations can be eliminated using ad hoc arguments. This eliminates 21 of
the 66 starting point configurations. This leaves us with 45 remaining starting point configurations
that need to be eliminated. This was done using computers as follows.
For each starting point configuration, all possible incidences of the 19 points with the triple lines
are found. That is, by looking at table 4.3, to determine all possibilities for B3 once B1 is chosen.
B2 is then determined by successively looking at the incidence of the points of the heavy lines that
is not in the code word. One would then proceed to determine B4 and so forth. This, however,
was never necessary as no starting point configuration could be extended to B2 . Checking all cases
took about 2 years of CPU time. As no starting point configuration could be completed into a
projective plane, we must conclude that no projective plane of order 10 exists.
(b21 + b22 + b23 + b24 )(x21 + x22 + x23 + x44 ) = y12 + y22 + y32 + y42 (A.1)
y1 = b1 x1 + b2 x2 + b3 x3 + b4 x4
y2 = b2 x1 + b1 x2 b4 x3 + b3 x4
(A.2)
y3 = b3 x1 + b4 x2 + b1 x3 b2 x4
y4 = b4 x1 b3 x2 + b2 x3 + b1 x4 .
y1 = b1 x1 b2 x2
(A.4)
y2 = b1 x2 + b2 x1 .
Proof. Straightforward calculation. The left hand side of equation (A.1), when expanded, is
(b21 + b22 + b23 + b24 )(x21 + x22 + x23 + x24 ) = b21 x21 + b21 x22 + b21 x23 + b21 x24 +
b22 x21 + b22 x22 + b22 x23 + b22 x24 +
b23 x21 + b23 x22 + b23 x23 + b23 x24 +
b24 x21 + b24 x22 + b24 x23 + b24 x24 .
For the right hand side, we will expand the y:s, one at a time. By the generalized quadratic rule,
we have
y12 = (b1 x1 + b2 x2 + b3 x3 + b4 x4 )2 = b21 x21 + b22 x22 + b23 x23 + b24 x24 +
2b1 b2 x1 x2 + 2b1 b3 x1 x3 + 2b1 b4 x1 x4 +
2b2 b3 x2 x3 + 2b2 b4 x2 x4 + 2b3 b4 x3 x4 ,
y22 = (b2 x1 + b1 x2 b4 x3 + b3 x4 )2 = b22 x21 + b21 x22 + b24 x23 + b23 x24
2b1 b2 x1 x2 + 2b2 b4 x1 x3 2b2 b3 x1 x4
2b1 b4 x2 x3 + 2b1 b3 x2 x4 2b3 b4 x3 x4 ,
y32 = (b3 x1 + b4 x2 + b1 x3 b2 x4 )2 = b23 x21 + b24 x22 + b21 x23 + b22 x24
2b3 b4 x1 x2 2b1 b3 x1 x3 + 2b2 b3 x1 x4 +
2b1 b4 x2 x3 2b2 b4 x2 x4 2b1 b2 x3 x4 ,
y42 = (b4 x1 b3 x2 + b2 x3 + b1 x4 )2 = b24 x21 + b23 x22 + b22 x23 + b21 x24 +
2b3 b4 x1 x2 2b2 b4 x1 x3 2b1 b4 x1 x4
2b2 b3 x2 x3 2b1 b3 x2 x4 + 2b1 b2 x3 x4 .
35
36 APPENDIX A. PROOF OF THE FOUR-SQUARES IDENTITY
This looks rather messy, but we see that y12 + y22 + y32 + y42 has summands of the form b2i x2j exactly
once for each i, j, = 1, 2, 3, 4. Also, each summand of the form 2bi bj xk xl occurs exactly once for
each i, j, k, l = 1, 2, 3, 4 as do the summands of the form 2bi bj bk bl , so these cancel each other.
What we are left with is exactly the left hand side of equation (A.1).
To prove the two-squares identity, we do a similar calculation. The left hand side of equation
(A.3) is
(b21 + b22 )(x21 + x22 ) = b21 x21 + b21 x22 + b22 x21 + b2 x22 .
(b1 x21 b2 x2 )2 + (b1 x2 + b2 x1 )2 = b21 x21 + b22 x22 2b1 b2 x1 x2 + b21 x22 + b22 x21 + 2b1 b2 x1 x2
= b21 x21 + b22 x22 + b21 x22 + b22 x21
Lemma 82. Let B be the 4 4 matrix determined by (A.5). Then det(B) = (b21 + b22 + b23 + b24 )2 ,
and hence B is invertible if and only if at least one of the bs is different from zero.
Proof.
b1 b2 b3 b4
b2 b1 b4 b3
det(B) =
b 3 b 4 b 1 b 2
b4 b3 b2 b1
b1 b4 b3 b2 b3 b4
= b1 b4 b1 b2 + b2 b4 b1 b2
b3 b2 b1 b3 b2 b1
b2 b3 b4 b2 b3 b4
b3 b1 b4 b3 + b4 b1 b4 b3
b3 b2 b 1 b4 b1 b2 .
= b1 D1 + b2 D2 b3 D3 + b4 D4 .
Here we have labeled the four 3 3 determinants by D1 , D2 , D3 and D4 . Their values are
b b2 b4 b3 b4 b3
D1 = b1 1 b4 b3
b2 b1 b2 b1 b1 b2
= b1 (b21 + b22 ) b4 (b1 b4 b2 b3 ) b3 (b2 b4 b1 b3 ),
b1 b2 b3 b4 b3 b4
D2 = b2
b4
b3
b2 b1 b2 b1 b1 b2
= b2 (b21 + b22 ) b4 (b1 b3 b2 b4 ) b3 (b2 b3 b1 b4 ),
b4 b3 b3 b4 b 3 b4
D3 = b2
b1 b3
b2 b1 b2 b1 b4 b3
= b2 (b1 b4 b2 b3 ) b1 (b1 b3 b2 b4 ) b3 (b23 + b24 ),
b4 b3 b3 b4 b3 b4
D4 = b2
b1 + b4
b1 b2 b1 b2 b4 b3
= b2 (b2 b4 b1 b3 ) b1 (b2 b3 b1 b4 ) + b4 (b23 + b24 ).
So, we get
37
det(B) = b1 D1 + b2 D2 b3 D3 + b4 D4
= b21 (b21 + b22 ) + b1 b4 (b1 b4 + b2 b3 ) b1 b3 (b2 b4 b1 b3 )+
b22 (b21 + b22 ) + b2 b4 (b2 b4 b3 b1 ) + b2 b3 (b2 b3 + b1 b4 )+
b2 b3 (b1 b4 + b2 b3 ) b1 b3 (b2 b4 b1 b3 ) + b23 (b23 + b24 )+
b2 b4 (b2 b4 b1 b3 ) + b1 b4 (b2 b3 + b1 b4 ) + b24 (b23 + b24 )
= (b21 + b22 )(b21 + b22 ) + (b23 + b24 )(b23 + b24 )
+2(b1 b4 + b2 b3 )(b1 b4 + b2 b3 ) + 2(b2 b4 b1 b3 )(b1 b4 b2 b3 )
= (b21 + b22 )2 + (b23 + b24 )2 + 2(b1 b4 + b2 b3 )2 + 2(b2 b4 b1 b3 )2
= b41 + 2b21 b22 + b42 + b43 + 2b23 b24 + b44 + 2b21 b24
+4b1 b2 b3 b4 + 2b22 b23 + 2b22 b24 4b1 b2 b3 b4 + 2b21 b23
P4 2 2 2 2 2 2 2
= i,j=1 bi bj = (b1 + b2 + b3 + b4 ) ,
In this appendix we will prove Lagranges Theorem, that states that any positive integer can be
written as the sum of four squares. The proof will be very similar to the proof of Lemma 57. For
the proof, we will need the following lemma
Lemma 83. For every odd integer m, there exists integers x, y and n such that x2 + y 2 + 1 = mn.
Proof. Let X = {x2 + 1 (mod m) | x = 0, . . . , (m 1)/2}. The elements of X are easily seen to be
(m 1)/2 distinct elements. Let Y = {(y 2 ) (mod m) | y = 0, . . . (m 1)/2}. The elements of Y
are also all distinct. Hence, |X| = |Y | = (m 1)/2, so |X| + |Y | = m + 1. As there are m distinct
values (mod m), X and Y must have at least one element in common. Hence there must exist x
and y, 0 x, y (m 1)/2 such that x2 + 1 (y 2 ) (mod m), or x2 + y 2 + 1 0 (mod m).
Hence, there exists an integer n such that x2 + y 2 + 1 = mn, and we are done.
Theorem 84 (Lagrange). Every positive integer can be written as the sum of four squares.
Proof. By the four-squares identity, i suffices to prove that every positive prime p can be written
as the sum of four squares. Since 2 = 12 + 12 + 02 + 02 , and 2 is the only even prime, we can
assume that p is odd.
Let a, b, c, d be integers and m a positive integer such that a2 + b2 + c2 + d2 = mp. Such integers
exist by Lemma 83. Take such an expression that minimizes m. If we can show that m = 1,
we have proven the lemma. So, for a contradiction, assume that m > 1. Now choose integers
A, B, C, D such that m/2 < A, B, C, D m/2, and A a, B b, C c, D d (mod m). By
the same argument as in the proof of Lemma 57 such integers exists.
We have that A2 + B 2 + C 2 + D2 a2 + b2 + c2 + d2 0 (mod m), so A2 + B 2 + C 2 + D2 = rm
for some non-negative integer r. We see that
rm A2 + B 2 + C 2 + D2 m2 /4 + m2 /4 + m2 /4 + m2 /4
r= = = m,
m m m
so r m.
Suppose that r = 0. This implies that A = B = C = D = 0, so a b c d 0 (mod m).
But this means that a, b, c and d are multiples of m, so a2 + b2 + c2 + d2 is divisible by m2 , and p
is divisible by m. This contradicts the fact that m > 1 and p is prime, so r can not be zero. Thus
r > 0.
Now, suppose that r = m. Then A = B = C = D = m/2. Since A a (mod m), this means
that a = m/2 + um for some integer u. Then a2 = m2 /4 + um2 + u2 m2 , so a2 m2 /4 (mod m2 ).
The same holds, of course, for b, c and d. Then,
m2 m2 m2 m2
a2 + b2 + c2 + d2 + + + m2 0 (mod m2 ),
4 4 4 4
so a2 + b2 + c2 + d2 is again divisible m2 , which implies that p is divisible by m, which contradicts
the fact that p is prime. Hence r < m.
Now, by the four squares identity,
39
40 APPENDIX B. PROOF OF THE THEOREM OF LAGRANGE
where
e = aA + bB + cC + dD
f = bA + aB dC + cD
g = cA + dB + aC bD
h = dA cB + bC + aD.
Furthermore,
e a2 + b2 + c2 + d2 0 (mod m)
f ba + ab dc + cd 0 (mod m)
g ca + db + ac bd 0 (mod m)
h da cb + bc + ad 0 (mod m),
so e, f, g and h all have a factor m, and thus e2 , f 2 , g 2 and h2 are all divisible by m2 . But then
so rp is a sum of four integer squares, and r < m, contradicting the fact that m is the smallest
such number. Our assumption that m > 1 does not hold, so we must have that m = 1.
Bibliography
[1] Hughes D. R. and Piper F. C. Projective Planes, second edition. Springer-Verlag, New York.
[2] Hughes D. R. Collineations and generalized incidence matrices. Trans Amer. Math. Soc. No.
86, 1957, pp. 284-296.
[3] Hughes D. R. Generalized incidence matrices over group algebras. Illinois J. Math, No 1, 1957,
pp. 545-551.
[4] T.G. Room and P.B. Kirkpatrick Miniquaternion Geometry. Cambridge Univ. Press, 1971.
[5] MacWilliams F. J., Sloane N. J. A. and Thompson, J. G. On the existence of a projective
plane of order 10. Journal of Combinatorial Theory, Series A, No. 14, 1973, pp. 66-78.
[6] Whitesides S. H. Projective planes of order 10 have no collineations of order 11. Proceedings
of the Seventh Southeastern Conference on Combinatorics, Graph Theory, and Computing.
Utilias Mathematica, 1976, pp. 515-526.
[7] Whitesides S. H. Collineations of projective planes of order 10, Parts I & II. Journal of
Combinatorial Theory, Series A, No 26, 1979, pp. 249-277.
[8] Anstee R. P., Hall, M Jr. and Thompson J. G. Planes of order 10 do not have a collineation
of order 5. Journal of Combinatorial Theory, Series A, No. 29, 1980, pp. 39-58.
[9] Janko Z. and van Trung T. Projective planes of order 10 do not have a collineation of order
3. J. Reine Angew. Math, No 325, 1981, pp. 189-209.
[10] Lam C. W. H., Thiel L., Swiercz S. and McKay J. The nonexistence of ovals in a projective
plane of order 10. Discrete Mathematics No. 45, 1983, pp. 319-321.
[11] Lam C. W. H., Crossfield S. and Thiel L. Estimates of a computer search for a projective
plane of order 10. Congressus Numerantium, No 48, 1985, pp. 253-263.
[12] Lam C. W. H., Thiel L. and Swiercz S. The nonexistence of code words of weight 16 in a
projective plane of order 10. Journal of Combinatorial Theory, Series A, No. 42, 1986, pp.
207-214.
[13] Lam C. W. H., Thiel L. and Swiercz S. The non-existence of finite projective planes of order
10. Can. J. Math, Vol. XLI, No. 6, 1989, pp. 1117-1123.
[14] Batten, L. M. Combinatorics of finite geometries. Cambridge University Press, 1997.
[15] Hall, M. Jr. Combinatorial Theory, second edition. John Wiley & Sons Inc., 1998.
[16] Lam, C. W. H. The Search for a Finite Projective Plane of Order 10.
http://www.cecm.sfu.ca/organics/papers/lam/
41