Incidence Geometry

ii
Preface
The student facing incidence geometry for the first time is likely to wonder if this
subject is some fanciful departure from the more familiar territory of Euclidean and other
metric geometry.
The geometry most commonly featured in high school curricula is that of the Euclidean
plane. This setting has the advantage and the disadvantage of familiarity. Certainly the
relative familiarity of the Euclidean allows many of the concepts to be accepted intuitively
in a short time, and to create an impression that these concepts are somehow useful as
representations, or at least idealizations, of the real world.
In fact the Euclidean plane is neither as simple nor as real as the typical high school
student (or teacher) supposes. Moreover the Euclidean plane is often used as a setting
for learning the axiomatic method, which Euclid so capably promoted. Unfortunately
the Euclidean plane carries far too many implied notions for the average student to fully
appreciate and capably use in proving propositions. Indeed many of these notions were
only vaguely understood by Euclid himself, or by his strongest successors. In addition to
incidence, there are also the notions of distance, angle, continuity, betweenness/separation,
etc. Moreover the Euclidean plane is no less complicated than the real number system: for
example the Euclidean plane contains subsets whose area cannot be meaningfully defined
(non-measurable subsets in the Lebesgue sense). Even if one avoids such subtleties, the array of mathematical tools arising in typical Euclidean plane geometry is rather formidable.
Thus it is not surprising that the independence of Euclids fifth postulate remained in question for centuries.
By stripping away all the extra baggage of distance, length, angle, continuity, betweenness, etc. and retaining only the notion of incidence, we find that what remains is
still quite fascinating and highly nontrivial. In this setting the student will have ample
opportunity to experience the richness of many examples, while seeing every step of the
proofs built firmly upon a surprisingly small set of axioms. In this respect the study of
incidence geometry stands alongside group theory, topology and graph theory as a subject
area from which a very small set of axioms yields surprising bounty.
We intend our choice of topics be as self-contained as possible, while highlighting
the use of tools from other areas of mathematics, including finite fields, linear algebra,
groups, number theory, algebraic geometry, coding theory and invariant theory. The main
definitions and tools needed from these areas are therefore summarized in a number of
appendices.
To any experts who happen to be peeking at this (how embarrassing!), I beg your
indulgence as I occasionally oversimplify certain notions, and even omit some major results
and research trends. For instance, I know it is inexcusable to omit flocks of cones, BLTsets, q-clans, Kantor families, explicit triality automorphisms, construction of the G2 (q)
iii
hexagons, the actual definition of a building, or indeed of a polar space, etc. Sorry. Maybe
in the next revision!
A list of errata will be posted at
http://www.uwyo.edu/moorhouse/courses/5700/
With each mistake/misprint that you encounter in this manuscript, please first check the
website to see if it has already been listed; if not, please email me at moorhous@uwyo.edu
with the necessary correction to add to this list. Thank you!
Eric Moorhouse
September, 2007
Note on First Corrected Edition

Thanks to Colin Garnett, Dan May, Reshmi Nair, Stan Payne and Ryan Price for listing
errata in the first edition of these notes. The current version takes into account corrections
based on their comments.
Eric Moorhouse
January, 2008
iv
Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
I. Incidence Structures
1. Denitions and Examples . . . . . . . . . . . . . . . . . . . . . . . . . 3
II. Ane Planes

2.
3.
4.
5.
Denitions and Examples . . . . . . . . . . . . . . . . . . . . . . . . . 9

Translation Planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Latin Squares and Nets . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Nets and Webs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
III. Projective Planes

6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
Denitions and Examples . . . . . . . . . . . . . . . . . . . . . . . . 35

Projective Completion of Ane Planes . . . . . . . . . . . 40
Advantages of the Projective Viewpoint . . . . . . . . . . 44
Closed Congurations . . . . . . . . . . . . . . . . . . . . . . . . . . . .50
Collineations and Correlations . . . . . . . . . . . . . . . . . . . 53
Classical Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
Conics and Ovals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Codes of Planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .76
The Bruck-Ryser Theorem . . . . . . . . . . . . . . . . . . . . . . . 85
Dierence Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Generalized Incidence Matrices . . . . . . . . . . . . . . . . . . .99
Blocking Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
IV. Projective and Polar Spaces

19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
Classical Ane and Projective Spaces . . . . . . . . . . . 121

Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
The Pl
ucker Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Quadratic Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Quadrics and Polar Spaces . . . . . . . . . . . . . . . . . . . . . . 148
The Klein Correspondence . . . . . . . . . . . . . . . . . . . . . . 161
Ovoids and Spreads of Projective Space . . . . . . . . . 166
Ovoids and Spreads of Polar Spaces . . . . . . . . . . . . . 172
Generalized Quadrangles . . . . . . . . . . . . . . . . . . . . . . . .178
Generalized Polygons and Buildings . . . . . . . . . . . . . 185
v
Appendices
A1.
A2.
A3.
A4.
A5.
A6.
Finite Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
Algebras and Representations . . . . . . . . . . . . . . . . . . . 211
Exterior Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .218
Coding Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
Invariant Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
vi
PART I
Incidence Structures
1. Definitions and Examples
An incidence structure, or incidence system, consists of certain objects (usually called
points, lines, planes, etc.) together with certain incidence relations between these objects.
We begin our study with just two types of objects: points and lines. This course begins
simply with point-line incidence structures (P, L, I) (often simply abbreviated (P, L)) in
which P and L are sets of points and lines respectively, and incidence relation I PL
is a binary relation indicating which point-line pairs are incident. For example consider
the incidence system with point set P0 = {P, Q, R, S, T }, line set L0 = {, m, n, r} and
incidence relation
I = {(P, ), (P, m), (Q, ), (Q, n), (T, ), (S, ), (S, r), (R, m), (R, n)}.
We informally say that P lies on and on m, but not on n or r; also passes through
P, Q, S, T but not R; etc. The incidence structure (P0 , L0 , I), or simply (P0 , L0 ), is
informally represented by the picture:
m n r
1 1 0 0 P
R...
..
.
.
.
.
.
.
r....
. ..
1 0 1 0 Q
... .....
..
...
.
.
.
.
.
.
.
m.... .....n
A = 0 1 1 0
..
..
R
...
...
...
...
.
.
.
.
.
.
...
.
.
.
.
1 0 0 1 S
P ...................................................................................................................................
Q
T
S
1 0 0 0 T
To the right we have also shown an incidence matrix for this structure: this is a matrix
with rows and columns indexed by the points and lines respectively, and with entries 0
and 1 corresponding to non-incident and incident point-line pairs, respectively. Neither the
picture nor the incidence matrix are unique; alternative choices for our example (P0 , L0 )
are given by
Q
..
.......
.. .......
....
...
....
.
.
....
.
...
....................................................
...
....
.
....
.
.
.
.
.
....
....
...
....
..
....
...
....
..
.
....
.
.
....
..
....
.
.
....
.
.
.
.
..............................................................................................
T
m
0
0
1
0
0
0
0
0
1
1
1
1
0
1
1 Q
0
T
0
P
1 R
0 S
P
m
Observe that the incidence matrix depends on the order in which points and lines are listed,
and so is not strictly unique. (We do not consider incidence matrices in the case of innite
structures, i.e. having innitely many points or lines.) Regarding the choice of picture, it is
important to note that the structure is determined by its incidence information only: there
is no relevant notion of distance, angle, betweenness, inside/outside, continuity, etc. Our
I. INCIDENCE STRUCTURES
pictures represent the same incidence structure (P0 , L0 ) because they represent exactly
the same sets of points and lines and incidences.
Consider two point-block incidence structures (P, L) and (P , L ). An isomorphism
from (P, L) to (P , L ) is a pair of bijections P P and L L , which preserves
incidence. It is natural to assume that P and L are disjoint (similarly P and L ) so that
any such pair of bijections can be considered as a single bijection P L P L . Let
us say precisely what we mean by an isomorphism, in this setting: An isomorphism from
(P, L) to (P , L ) is a bijection : P L P L , such that
(i) P = P and L = L ; and
(ii) for all P P and L, we have P i P .
An isomorphism from (P, L) to itself is called an automorphism. The set of all automorphisms of (P, L), denoted Aut(P, L), is a group under composition. Our example
(P0 , L0 ) above has just one nontrivial automorphism = (P, Q)(m, n) interchanging the
points P and Q; and simultaneously interchanging the lines m and n. In this case we
have Aut(P0 , L0 ) = = {1, }. In terms of the incidence matrix A given above, this
means that if the rst two rows are interchanged, and the 2nd and 3rd columns are also
interchanged, then the matrix A is preserved.
An automorphism group of (P, L) is a subgroup of Aut(P, L), the latter being
the (full) automorphism group. (Compare terminology: a permutation group is
a subgroup of Sn , the latter being the symmetric group.) Consider the triangle =
({P, Q, R}, {, m, n}) embedded in (P0 , L0 ); then has exactly 6 automorphisms:
R....
..
... ....
.. .....
...
...
...
.
.
.
...
.
.
.
.
.
............................................
m.........
P
Aut = (P Q)(mn), (P R)(n)

= {1, (P Q)(mn), (P R)(n), (QR)(m), (P QR)(nm), (P RQ)(mn)}.
The dual of a point-line incidence structure (P, L) is the structure (L, P) with points
and lines exchanged, and with the same (or rather the reversed) incidence relation. The
dual of our example above, along with an incidence matrix for this dual structure, is
n ..
....
.
... ...
...
...
......
.. ....
...
.......
...
...
...
.......
.
.
.
.
.
.
.
.
.
...
.
.....
...
...
...
......
...
...
...
.......
... ..... ............
...
.
.
.
.
.
.
... .. ........
.
.. . ....
...
............................................................
P
1
1
0
0
1
0
1
0
1
0
0
0
1
0
0
1
0
1
m
1n
0 r
P

Observe that an incidence matrix for the dual structure (L, P) is simply the transpose of
an incidence matrix of the original structure (P, L). A self-dual incidence structure is
one which is isomorphic to its dual. For example a triangle is self-dual. Considering the
triangle presented above, we present an explicit isomorphism from to its dual:
m..
R...
..
.. ....
... ...
P
Q
.
.
. ..
... .....
... .....
...
.
.
.
.
m..... .....n P..... ........R
Q n
m P
...
...
..
...
...
...
...
..
..
.........................................
.. n
..............................................
.

R m
n R
Q
P
Q

m
We seldom have use for point-block incidence structures in their full generality; rather
we consider those structures that satisfy certain well-chosen properties. For example a
partial linear space is a point-line incidence structure satisfying the axioms
(PLS1) Any two distinct points lie on at most one common line.
(PLS2) Every line has at least two points.
Thus for example is a partial linear space; but neither are the example (P0 , L0 ) nor its
dual, which fail the second axiom (PLS2). Note that axiom (PLS1) says that the situation
......... .................................................................... .........
................................................................................................................
1.1 Figure
(called a digon) never occurs in our structure. Although such structures are sometimes
of interest, we shall reserve the term line for subsets meeting in at most one point. If
Figure 1.1 occurs in a given structure we may use the term block in place of line, referring
to such structures rather as point-block incidence structures.
A linear space is a point-line incidence structure satisfying the stronger conditions
(LS1) Any two distinct points lie on exactly one common line.
(LS2) Every line has at least two points.
Our example (P0 , L0 ) fails both axioms, but is an example of a linear space.
When it becomes necessary to distinguish between a line and the set of its points,
we shall denote by [] the set of points on the line . For example in the structure
...
...
...
...
...
...
...
...
...
...
....................................................................
...
...
we have [] = [m] = {P } although = m. Also in Figure 1.1 we see two lines (or rather,
blocks) with the same point sets. This issue never arises in a partial linear space, where it
is easy to see that [] = [m] whenever = m.
Likewise for any point P we denote by [P ] the set of lines through P . Thus for example
in (P0 , L0 ) we have
[P ] = {, m},
[Q] = {, n},
[] = {P, Q, S, T },
etc.;
[m] = {P, R},
etc.
In (P0 , L0 ) we write m = P to say that P is the unique point on both lines and
m. We also write P Q = or simply P Q = to say that is the unique line through
both P and Q. These operations are not always dened; for example R S is undened
in (P0 , L0 ). But in any linear space we may safely write X Y for the unique line joining
two distinct points X and Y (and extend the denition to the case X = Y by writing
X X = X). Similar remarks apply for the binary operation .
I. INCIDENCE STRUCTURES
Exercises 1.
1. Give an example of a partial linear space which is not a linear space.
2. Find the automorphism group of each of the following point-line incidence structures:
(a)
(b)
3. Draw (and label) the dual of each of the point-line incidence structures shown in Exercise #2.
4. (a) Show that the dual of any partial linear space satisfies (PLS1).
(b) Give an example of a partial linear space whose dual is not a partial linear space.
PART II
Ane Planes
2. Denitions and Examples
An ane plane is an incidence system of points and lines such that
(AP1) For any two distinct points, there is exactly one line through both.
(AP2) Given any line and any point P not on , there is exactly one line through P
that does not meet .
(AP3) There exist four points such that no three are collinear.
The most familiar model for these axioms is the Euclidean plane. (A model for a set of
axioms is an example which satises the axioms. The notions of points, lines and incidence
are not dened by our axioms; rather, any particular model provides an interpretation of
these notions.) More generally, if F is any eld, one constructs an ane plane over F by
taking ordered pairs (x, y) F 2 as points; and subsets of the form y = mx + b or x = a
(for xed a, m, b F ) as lines; that is, point sets of the form
{(x, mx+b) : x F }
for m, b F ;
{(a, y) : y F } for a F.
The incidence is the natural one: a point is on a given line i it satises the required linear
equation. This plane is denoted A2 (F ), or often AG2 (F ), and is known as the classical
ane geometry of dimension 2, i.e. ane plane, over the eld F . Note that A2 (R)
is simply the Euclidean plane, but with attention given to just the incidence information,
disregarding the additional structure of distance, angle, topology, etc.
For every nite eld Fq (where q is a prime power) there is a corresponding classical
ane plane A2 (Fq ). The smallest two such planes are as shown:
...
...
.....
.
....
....
..
..
...
....
..
..
....
....
.
.
.
.
.
.
.... ...
.... .......
.... .
.. .....
.... ..
..........................................................................................................................
... ........
..... ..
.... ..
... ....
....
.... ...
..
....
....
.
.
..
...
.
.... ....
...
...
........
..
..
.........
.
.
..
..
.
.
.
.
.
.
.
...
.
.
....
.
..... ....
.. ........
.... ..
.. ....
.... ..
... .....
..
...........................................................................................................................
... .
.... ........
.... ....
.
.
.
.. .....
...
....
...
..
....
....
.
.
.
...
.
.
.
.....
.
.
.
.
..
.
..
..
.
.....
(0, 1)
(0, 0)
x=0
y=1
y=0
y=x
(1, 1)
(1, 0)
x=1
2.1a Figure
A2 (F2 ) :
4 points, 6 lines
y=x+1
..............................
............
........
........
......
......
......
.
.
.
.
.....
....
.
.
....
.
.
....
.
.
.
.
.
.
.........................
.
.
.
.
.
.... ..... ....... ..................
.
.
.
.
.
.
.
.
....
......
.
..
.
.
.
.
.
....
.
......
.
..
...
....
...
.....
...
...
...........................................................................................................................................
....
.
.
.
.
.
.... ....... .... .......
.
.
.
.. ... .....
.. ...
...
.
.
.
.
.. ..... .. .....
.
.
....
...
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
....
.
.
.
..
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.... ...
.... ...
...
.
.
...
.. ..
.
.
.
.
.
.
.
.
.
.
.
.
.
.......
.......
..
...
.. ..
.
.
.
.
.
.
.
.
.
.
.
.
....
....
..
...
.
.
...
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
. ....
. ....
.
.
.
.
..
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... .
....
....
.
.
...
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
... . ....
.... ..
.... .. .....
.
.
..
.
.
.
.
.... ...
.... .. ....
... .. ....
.
..
.
.
.
.
.
.
.
.
.
.........
. .
..
.........................................................................................................................................
.
....
.
.
........
......
.....
..
.... . ...
.... .. ....
...
.. .....
...
... ... ..
... .. ....
...
.. .....
..
.... ... .....
.... ... .......
.
.
.
.
.
....
..
.
...
.
.
.
....
... ...
.
...
....
..
....
..
.... .......
...
...
.... .......
..
....
........
..
..
... ......
......
...
..
...
..
...
... ......
..........
..........
...
.
.
.
.
.
.
.
.
.
...
........
....
....
...
..
...
.....
.....
....
....
...
... .......
.... .... .......
.... .... ...........
...
.... . ....
.... .. .... ..
.. .....
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
...
............................................................................................................................................. ....
.....
....
......
.....
....
...
......
......
....
.
.
..
.
........
.
.
.
..
.............. ..... ..................
.
.
................
...
....
...
....
...
.....
.....
.
......
.
.
.
.......
....
.........
......
............................................
2.1b Figure
A2 (F3 ) :
9 points, 12 lines
10
II. AFFINE PLANES
Observe that A2 (Fq ) has exactly q 2 points and q 2 + q lines. We now proceed to show
that every nite ane plane has n2 points and n2 + n lines for some positive integer n. Let
and m be two lines in an ane plane. We say that is parallel to m (denoted m)
if either = m or the two lines have no points in common. By (AP1), any two lines are
either parallel or they meet in a unique point.
2.2 Proposition. Parallelism is an equivalence relation on the lines of an ane
plane.
Proof. Parallelism is clearly a reexive symmetric relation on the class of lines. To prove
that this relation is transitive, let i be lines of an ane plane (i = 1, 2, 3) such that 1 2
and 2 3 . We may assume that 1 , 2 , 3 are distinct. If 1 and 3 are not parallel then
they meet in a point P ; but then 1 and 3 are distinct lines through P parallel to 2 ,
contrary to (AP2). Thus 1 3 and we are done.
2.3 Theorem. In an ane plane, any two lines have the same number of points,
nite or innite. More precisely, given any two lines and , there is a bijection
between the points on and the points on .
Proof. Let and be distinct lines. Thus there exists a point P on but not on , and
a point P on but not on . Let p = P P (the unique line joining P and P ). For an
arbitrary point R on , let r be the unique line through R parallel to p. Since r p ,
the lines r and must meet, say in a point R . The map R R is a well-dened map
from points of to points of . Interchanging and (also P and P ) gives a similar map
from the points of to the points of , which maps R R; but this is clearly the inverse
of the previous map.
The order of an ane plane is the number of points on any given line of the plane.
It is clear from the axioms that the order is at least two. Observe that the classical nite
ane plane A2 (Fq ) has the same order as the nite eld Fq used to construct the plane,
namely q. Moreover the Euclidean plane has innite order (more precisely, its order is
the cardinality of the real numbers, namely |R| = 20 , the cardinality of the continuum).
Similarly A2 (Q) has countably innite order |Q| = 0 .
2.4 Theorem. Let A be an ane plane of nite order n. Then every point of A lies
on exactly n + 1 lines, representing the distinct parallel classes of lines. Moreover every
parallel class of lines consists of n lines which partition the points of A. In particular
A has exactly n2 points and n2 + n lines.
11
Proof. Let be any line of A. There exists a line m not parallel to . (Observe that
has n 2 points, and by (AP3), not all points lie on . If P is any point on and Q
is any point not on , we may take m = P Q.) Let M1 , . . . , Mn be the points of m, and
for each i {1, 2, . . . , n}, let i be the unique line through Mi parallel to . (Note that
{1 , . . . , n }; we may assume that = 1 .) We see that every line parallel to is one of
1 , . . . , n . Indeed if is a line parallel to , then m so meets m in a point Mi
which forces = i . Moreover the point sets of 1 , . . . , n partition the points of A; for
if P is any point and is the unique line through P parallel to , then as we have seen,
{1 , . . . , n }.
By (AP3) and (AP1) there is at least one line . Let {1 , . . . , n } be its parallel class.
Since the lines 1 , . . . , n partition the points of A, and each i has exactly n points, it
follows that A has exactly n2 points. Let P be any point of A and let r be the number
of lines through P . Count in two dierent ways the number of points Q distinct from
P . Since every point Q = P determines a unique line P Q with n points, and every line
through P has n 1 points distinct from P , we have n2 1 = (n 1)r so that r = n + 1.
We now give an example of an ane plane that is not classical. The easiest example
is perhaps the following innite ane plane, formed by perturbing the Euclidean plane.
The Moulton plane has point set R2 , and lines are of the form x = a (for a R constant,
the vertical lines); and the lines y = m x + b for m, b R; here the binary operation
is dened by

mx,
if m 0 or x 0;
mx=
2mx, if m > 0 and x > 0.
Note that if m 0 then the line y = m x + b is simply a Euclidean line of non-positive
slope. If m > 0 then the line y = m x + b bends as it crosses the y-axis; the slope of the
portion to the right of the y-axis is double that of the portion to the left of the y-axis.
2.5 Theorem. The incidence system dened above is in fact an ane plane.
The proof, which we omit, is a straightforward if slightly technical matter of checking
cases.
The following table lists the number of isomorphism types of ane planes of order
n for some small values of n. Complete enumeration of planes has been accomplished to
date only for n 10; thus for n 11 only lower bounds are currently available.
n
no. of ane
planes of order n
n
no. of ane
planes of order n
10
11
13
16
17
19
23
25
27
29
31
88
2959
31
12
II. AFFINE PLANES
The most basic open problems in nite geometry may be stated as follows:
If there exists an ane plane of order n, must n be a prime power?
Is every ane plane of prime order p isomorphic to the classical plane A2 (Fp )?
These problems will be reformulated in terms of projective planes in Part III, where we
will describe the very modest progress to date towards answering these questions.
Exercises 2.
1. A t-(v, k, ) design is a point-block incidence structure with v points such that each block contains
exactly k points, and such that every t-set of points is contained in exactly blocks. Show that an
ane plane of order n 2 is the same thing as a 2-(n2 , n, 1) design.
2. Find complex coordinates for the four unlabelled points in the following incidence structure so that
the indicated incidences hold in the complex ane
plane:
......
......... ...............
......
.....
.....
.....
....
..........
.
.
. (0,2) ............... .......................
.
.
....
....
....................................
....................................
.
.... ....
...
...
...... ... ...... ...... ... ...... ...... ..... (2,2)
. .
.
. .
...
........ .... ............. .... ............. ....
.
...
.
. .. .. .... .... .. .... .... ..
.
.
..
.... ..
.... .. ....
... .. .....
..
.
.
..........
.........
........
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
.
.
.
.
.
.
....
.
..........(1,1)
....
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
. .
.
.... ...... ....... .... ....... ........ .... .... .....
...
.. ......... .. ......... ... ... .....
..
...
.. ..... ..... .. ..... ..... .. ......
.... ... ......
...
.... .. ....
... .....
....
............................................................................ ..
(0,0)
......
................. .............. (2,0) ....
................
...
....
......
.....
.........
......
....................
Conclude that the ane plane of order three embeds in the complex ane plane A2 (C). Similarly
show that A2 (F3 ) embeds in A2 (Fp ) for innitely many primes p (which ones?).
3. An inversive plane, or a M
obius plane, is an incidence structure (P, C) with a point set P and a
.....
set C of blocks called circles, such that
...... .. .......
.
....
.....Q
(IP1) Every set of 3 distinct points lies on a unique circle.
..
..
..
... ..........................C
.....
...
(IP2) If P, Q are points and C is a circle containing P but not .....
P...........
..
..
.
...
.
.
..

.
.
...
.. ...
Q, then there is a unique circle C containing Q such that
..
.......
.. ...
.
.
.
..
.
.......
...
.

.
.

....
.
.
C C = {P }.
.
.
......
C
.......................
(IP3) There exist 4 points not on a common circle.
(a) If S is a sphere in Euclidean 3-space, show that the points and circles lying on S form an inversive
plane. (This example is known as the real inversive plane.)
(b) If (P, C) is any inversive plane with a point P P, show that (P .......{P }, L) is an ane plane where
L is the collection of all point sets C .......{P } where P C C.
(c) Show that a nite inversive plane is the same thing as a 3-(n2 +1, n+1, 1) design for some n 2.
(Equivalently, a nite incidence point-block structure is an inversive plane i it is a 3-(n2 +1, n+1, 1)
design for some n 2.) We call n the order of the inversive plane.
(d) How many circles does an inversive plane of order n have?
(e) Explicitly construct an inversive plane with as few points as possible.
4. Consider the more general Moulton plane constructed using

mx,
if m 0 or x 0;
mx =
cmx, if m > 0 and x > 0
where c R is xed. Which values of c give ane planes? Which values of c give the Euclidean plane
(up to isomorphism)? Can two distinct values of c give isomorphic planes? Explain.
3. Translation Planes
13
Let V be a 2n-dimensional vector space over a eld F . A spread of V is a set consisting
of n-dimensional subspaces which partition the nonzero vectors. If F = Fq then V has
q 2n 1 nonzero vectors and so the number of spread members is
|| =
q 2n 1
= q n + 1.
qn 1
We typically write = {V0 , V1 , V2 , . . . , Vqn }. The members Vi are called the components of the spread, and they satisfy Vi Vj = V for all i = j. Given a spread in V we
consider the point-line incidence structure A() dened by
(i) Points are vectors in V .
(ii) Lines are components of , or more generally cosets of the form v + U where
v V and U .
3.1 Theorem. The incidence structure A() is an ane plane. The group of translations x x+w for w V is an automorphism group acting regularly on the points
of this plane.
Proof. The last assertion is clearly true since the translation x x+w maps a typical
line w + U to the line (w+x) + U. Let x, x V be two distinct points; we must show that
there is a unique line passing through x and x . There is no loss of generality in assuming
x = 0; otherwise translate by the vector x to shift x and x to some pair of distinct
vectors, one of which is zero. Now there is a unique member U containing x, and U
is the unique line containing x and 0. Thus (AP1) holds.
Consider a non-incident point-line pair, say x V and w + U where x
/ w + U. Again
we may assume x = 0; otherwise translate by x. Now U is a line through x = 0 which is
disjoint from w + U. To show that this line is unique with these properties, consider any
line U through 0 disjoint from w + U. If U = U then V = U U so w = u + u
for some u U, u U ; but then u = wu U (w + U ) = , a contradiction. This
proves (AP2).
To verify (AP3), let U, U be distinct spread members, and consider any nonzero vectors u U, u U . It is straightforward to check that no three of 0, u, u , u+u are
collinear.
3.2 Example.
Let V be a 2-dimensional vector space over a eld F , and let be the
collection of all 1-dimensional subspaces. Then A() is simply the classical ane plane
A2 (F ).
14
II. AFFINE PLANES
3.3 Example: Regular Spreads.

Let V = C2 considered as a 4-dimensional vector
space over F = R. Let be the collection of all subspaces U < V which are 1-dimensional
over C, considered as 2-dimensional subspaces over R. It is easy to verify that A()
=
2
A (C), so again this construction gives nothing new. More generally if E F is any
eld extension of degree n, and V = E 2 considered as a 2n-dimensional vector space over
F , we may take to be the collection of all U < V which are 1-dimensional subspaces
over E, hence n-dimensional over F ; but then A()
= A2 (E). Such spreads are called
regular for reasons that we will explain in Section 26 (not just because someone thought
the word regular wasnt used enough in mathematical contexts). The ane planes dened
by regular spreads, are simply the classical ane planes.
To obtain examples that are genuinely new (not just disguised versions of classical
planes) consider a nite eld F = Fq . Consider the 2n-dimensional vector space
V = {(x, y) : x, y F n }.
Suppose M0 , M1 , . . . , Mqn 1 are n n matrices over F such that Mi Mj is nonsingular
whenever i = j. (There are many examples to show this is possible; the smallest is given
in Example 3.5 below.) For i = 0, 1, 2, . . . , q n 1 dene Vi to be the subspace y = xMi ,
i.e.
Vi = {(x, xMi ) : x F n }.
Also dene Vqn to be the subspace x = 0, i.e.
Vqn = {(0, y) : y F n }.
Let = {V0 , V1 , . . . , Vqn }.
3.4 Theorem. The set dened as above is a spread in V .
The matrices M0 , M1 , . . . , Mqn 1 used in the construction of are called spread matrices, or slope matrices for the spread . There is a veritable industry in constructing
examples of such sets of slope matrices, consuming entire careers of certain mathematicians.
Proof of Theorem 3.4.
Consider distinct i, j {0, 1, 2, . . . , q n 1} and suppose

(x, xMi ) = (x, xMj ) Vi Vj ;
then x(Mi Mj ) = 0. Since Mi Mj is nonsingular, this forces x = 0 and so Vi Vj = {0}.

Also if (x, xMi ) = (0, y) Vi Vqn then x = 0 = y so Vi Vqn = 0. Since has q n +1
members, by the remarks at the beginning of this Section we are done.
15
The eld Fqn can be represented as a subring of the n n matrices over Fq (see
Appendix A1.2). The resulting set of n n matrices forms a spread set as above; however,
not surprisingly, the resulting translation plane is isomorphic to A2 (Fqn ) since such spreads
are regular as observed in Example 3.3. For example the following nine matrices over F3
form a matrix representation of F9 :

0 0
1
,
0 0
0

0
2
,
1
0

0
0
,
2
2

1
1
,
0
2

1
2
,
1
2

1
0
,
2
1

2
1
,
0
1

2
2
,
1
2

2
.
1
Accordingly they dene a regular spread, which (boringly) yields the classical translation
plane A2 (F9 ). The following example is dierent, and therefore deemed more interesting.
3.5 Example: The Hall Plane of Order 9.
0
0

0
1
,
0
0

0
2
,
1
0

0
1
,
2
1

1
2
,
2
2
Let F = F3 . The nine matrices

2
0
,
1
2

1
0
,
0
1

2
1
,
0
2

2
2 1
,
2
1 1
satisfy the required condition that the dierence of any two is nonsingular. This gives rise
to the Hall plane of order 9. This plane is one of seven nonisomorphic ane planes of
order 9, the smallest order for which nonclassical planes exist. One way to verify that this
plane is non-classical, would be to verify that it does not satisfy Desargues Theorem or
Pappus Theorem (Section 11). In principle this can be done directly from an incidence
matrix, but we omit the details. In Section 13 we will give another verication that the
Hall plane of order 9 is nonclassical, using 3-ranks of incidence matrices.
This example generalizes to an innite family of translation planes, one of order q 2
for every prime power q, called the Hall planes; see [31] for details of this construction.
3.6 Example: An Innite Family of Nonclassical Planes.
Let F = Fq where q
is an odd prime power, and let E = Fq2 F be its quadratic extension. Recall that of the
q 2 1 nonzero elements of E, exactly half are squares. The map
E E,
x xq
is an automorphism of E xing every element of F , and mapping squares to squares. Dene

a new binary operation on E by

xy =
xy,
q
x y,
if y is a square (possibly zero);

if y is a nonsquare.
Consider the incidence structure with point set V = E 2 = {(x, y) : x, y E} and with
lines of the form
{(x, x m + b) : x E} for m, b E (lines of the form y = x m + b); and
{(a, y) : y E} for a E (lines of the form x = a).
16
II. AFFINE PLANES
This gives a translation plane whose components are the lines through the origin:
(i) {(x, x m) : x E} for m E (lines of the form y = x m); and
(ii) {(0, y) : y E} (the line of the form x = 0).
We check that two lines of the form (i) intersect only in {0} as required: Suppose (x, xm) =
(x, x m ) with x = 0. Since x is a square in E i xq E is a square in E, clearly m is a
square i m is a square. So either xm = xm or xq m = xq m , and in any case m = m as
required. The other requirements of a spread follow easily.
The case q = 3 gives the Hall plane of order 9. The quasields we have constructed
above are special cases of Andre quasields (see Exercise #6) and the resulting planes are
Andr
e planes.
3.7 Quasields. A loop is a set L with a binary operation such that
(L1) There exists a two-sided identity 1 L: we have 1 x = x 1 = x for all x L.
(L2) For all a L, the left-multiplication map L L, x a x is bijective; also the
right-multiplication map L L, x x a is bijective.
An example of a loop is the binary operation with Cayley table for x y given by
12345
21453
34512
45231
53124
The Cayley table of a nite loop with elements 1, 2, . . . , n and identity 1, is simply an
n n table with rst row and column (1, 2, . . . , n) (in that order); and where every row
and column contains every element exactly once. (This is a special type of Latin square;
see Section 4 for more general Latin squares.) Note that an associative loop is the same
thing as a group.
A (right) quasield is an algebraic structure Q with binary operations called addition + and multiplication such that
(Q1) Q is an abelian group with identity 0 under the addition +.
(Q2) We have 0 x = x 0 = 0 for all x Q. The nonidentity elements of Q form a
loop under multiplication .
(Q3) Q is right-distributive: (x + y) z = x z + y z for all x, y, z Q.
(Q4) If a, b, c Q with a = b, then the equation x a = x b + c has a unique solution
x Q.
If Q is nite then property (Q4) need not be stated explicitly since it follows from the
other three properties. Every eld is a quasield. A nonclassical example of a quasield is
dened in Example 3.6 above. Every quasield denes a vector space
V = Q Q = {(x, y) : x, y Q}
17
such that the subspaces of the form

{(x, x m) : x Q}
for m Q,
together with the subspace {(0, y) : y Q} form a spread of V .

Consequently every quasield Q gives rise to a translation plane of order |Q|. Conversely, every translation plane can be coordinatized by a quasield in this way. Given
a collection {M0 , M1 , . . . , Mqn 1 } of slope matrices for a spread , where M0 = 0 and
M1 = I without loss of generality (Exercise #2(b)), one obtains a quasield Q of order q n
as follows: Elements of Q are vectors in Fnq with the usual vector addition. Multiplication
is dened by
x y = xMi where y = (1, 0, 0, . . . , 0)Mi .
To see that this is well-dened, note that the matrices Mi have distinct rst rows since the
dierence of any two is nonsingular. Since there are q n slope matrices, every vector y F n
is the rst row of Mi for a unique i {0, 1, 2, . . . , q n 1}. One checks that the resulting
structure Q is a quasield, and that the plane it coordinatizes is just the translation plane
coordinatized by the original spread.
Let (P, L) be a translation plane, and let G = Aut(P, L). The group T consisting
of all translations is transitive on the points; so by Theorem A2.4, we have G = G0 T
where G0 is the stabilizer of the origin of the underlying vector space F 2n . The group G0
must in fact be the set of all semilinear transformations of F 2n preserving the spread, i.e.
mapping components to components. So if F = Fp then G0 actually consists of all linear
transformations preserving . Moreover T is a normal subgroup of G (see Appendix A2.7
where we see that the translation group is normalized by AL2n (F )) and G0 T = 1,
so G = T G0 is a semidirect product (see Appendix A2.7). The group G0 is called the
translation complement. In the case of classical planes A2 (F ), we have G0 = L2 (F ).
Exercises 3.
1. Write down an explicit set of slope matrices for the translation plane of order 25 constructed in
Example 3.6.
2. Let M0 , M1 , . . . , Mqn 1 be a set of n n slope matrices over Fq .
(a) Assuming A, B, C are n n matrices over Fq with A, B invertible, show that
{AMi B + C : 0 i < qn }
is also a set of slope matrices (The translation plane dened by this new set of slope matrices is
however isomorphic to that arising from the original set.)
(b) Show that any set of slope matrices is equivalent to a set containing 0 and I, the zero and identity
matrices of size n n over Fq .
3. Let M0 , M1 , . . . , Mqn 1 be a set of n n matrices over Fq with M0 = 0. Show that these matrices
form a set of slope matrices i M1 , . . . , Mqn 1 are invertible, and whenever 1 i < j < qn , the matrix
1
Mi Mj does not have eigenvalue 1.
18
II. AFFINE PLANES
4. Let G = M1 , M2 GL2 (F5 ) be the subgroup generated by

2 0
1 1
,
M2 =
.
M1 =
0 3
2 3
Show that the elements of G, together with the 2 2 zero matrix, form a set of slope matrices for a
spread.
Hint.
Use Exercise 3.
5. Let M0 , M1 , . . . , Mqn 1 be a set of n n slope matrices over Fq with M0 = 0, M1 = I. Let K be the

such that
set of all n n matrices A Fnn
q
AMi = Mi A
for all i.
Show that
(a) K is a eld with respect to the usual matrix addition and multiplication. (You may use Wedderburns Theorem in the nal step to conclude that K is commutative; see Appendix A3.)
(b) K has a subeld isomorphic to Fq ; thus |K| = qr for some integer r 1.
(c) We have r | n where r 1 is as in (b).
We call K the kernel of the associated translation plane.
6. Consider a nite eld F =Fq and its extension E =Fqr of degree r. Let : E E be the automorphism
x xq , so that G = {1, , 2 , . . . , r1 } is the group of all automorphisms of E xing every element
2
r1
of F ; see Appendix A1. Let N = NE/F : E F be the norm map x x1++ ++
. Also let
: F G be any map satisfying (1) = 1. Dene a new multiplication on E by

0,
if y = 0;
xy =
x(y) y, if y
= 0.
(a) Show that E is a quasield with respect to its usual addition and the new multiplication .
(b) Show that the quasields of Example 3.6 are a special case of this construction.
(c) Show that if is a homomorphism, then the quasield in (a) is associative.
This construction gives the Andr
e quasields. Those described in (c) are the Andr
e nearelds.
4. Latin Squares and Nets

Let S be a set of n 1 distinct symbols, such as {1, 2, 3, . . . , n}. An n n matrix with
entries from S and having no repeated entries in any row or column, is called a Latin
square of order n. Here are three examples of Latin squares of order 5:
01234
12340
L1 = 2 3 4 0 1 ,
34012
40123
12345
21453
L2 = 3 4 5 1 2 ,
45231
53124
bedca
caebd
L3 = a d c e b .
e badc
dcbae
Note that L1 is the Cayley table (addition table) for the additive group of integers mod 5.
More generally, a Cayley table for a group of order n is a Latin square. The example
L2 cannot be a Cayley table for any group; if it were, then {1, 2} would be a subgroup,
violating Lagranges Theorem.

Consider two Latin squares L = aij i,j and L = aij i,j with entries in the same
set S. We say L and L are orthogonal if for every pair of symbols (, ) S S occurs
19
as (aij , aij ) for some (necessarily unique) pair of indices (i, j). For example every pair of
the Latin squares
01234
01234
01234
01234
12340
23401
34012
40123
23401 , 40123 , 12340 , 34012
34012
12340
40123
23401
40123
34012
23401
12340
is orthogonal; therefore these constitute a set of four mutually1 orthogonal Latin squares. Every ane plane of order n gives rise to a set
of n1 mutually orthogonal Latin squares of order n, as follows: The
n2 points are the pairs (i, j) where i, j {1, 2, . . . , n}, thought of as
the n2 positions in an n n matrix. One parallel class of lines is given
by rows (positions with constant i-value); another is given by columns
(positions with constant j-value). Each Latin square denes a parallel
class of lines by taking the positions of each symbol S as the points
of a line in that parallel class. For example the ane plane of order 3
..........................
as shown:
........
.....
.....
Marcus Tullius
Cicero (10643 BC)
did not attend the
1969 Woodstock
Music Festival
....
....
..
...
............................
... ................................................................... ...........
.
.
.
....
.
.
.
.
.... ...............
.
.
......
. .
...
.. . .
.... ..
.... .. ....
...
...... .. .... .... .. ..... .... ..
..
.... .... ... ............... ... ............. ....
.
.
. .. .. ... ... .. .... .... ..
..
.
.
.... .. ....
... ... .....
.
.... ...
..
.
.
.
.
.
....................................................................
..
.
....
.......
.
.
.
......
.........
.
.
.
.
.
.
..
... ....
.
.
.
.
.
. ..
. .
..
.. .............. .... ............... ..... .... .....
...
.. ....... .. ........ .. ... ....
... ..... ..... ... ..... .... ... ......
...
.... .. ......
.... . ....
...
.......
. ..
.. . .
....
...... ............................................................................. ...
.......... .... ..........
...
.............
...
.....
....
......
............. ..................
...
is dened by the pair of orthogonal Latin squares

0 1 2
0 1 2
1 2 0 , 2 0 1
2 0 1
1 2 0
which specify the parallel classes
.........................
.......
......
......
....
....
....
....
...
.
.
.
.
.
.
....
..
..
...
.
.
.
.
.
.
.
.
.
.
.
..
..
.
...
.
.
.
.
.
.
.
.
..
..
..
.
...
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
..
..
..
.
.
.
.
....
.
.
.
...
..
..
.
.
.
.
.
...
.
.
.
.
..
..
.
.
.
.
.
.
..
.
.
..
....
....
..
....
.....
....
...
....
....
....
...
....
....
....
.
.
.
.
.
....
.
.
.
.
...
..... ....
....
.......
.....
.............................
and
...............................
....
......
....
....
...
....
.
.
....
..
.
....
....
.... .....
....
....
...
....
.
.
.
....
....
..
.
....
....
...
....
....
...
....
....
...
....
....
....
....
....
....
....
...
....
.
....
....
...
.
....
....
...
....
.
....
.
....
.
.... ....
.....
.... .
..
....
.
.
....
.
.
....
...
....
...
.....
......
.......
...........................
respectively. The remaining two parallel classes are specied by the rows and columns,
thus:
1
Some would say pairwise orthogonal Latin squares. Others, notably Coxeter, would say that pairwise is a
non-word made up by those who are unaware of the pre-existing word mutually.
20
II. AFFINE PLANES
.......................................................
.......................................................
......
......
......
....
..
..
..
...
...
..
..
...
..
...
and
.......................................................
....
..
..
..
...
...
..
..
...
..
...
....
..
..
..
...
...
..
..
...
..
...
Conversely, any set of k 2 mutually orthogonal Latin squares of order n species an

incidence structure consisting of n2 points and nk lines such that
(N1) Parallelism is an equivalence relation on the set of lines. (We say two lines are
parallel if they are either equal or disjoint.)
(N2) Every line has exactly n points, and every parallel class has n lines. Thus each
parallel class of lines partitions the point set.
(N3) There are k parallel classes of lines. Each point lies on exactly k lines, one from
each parallel class.
Such an incidence structure is called a k-net of order n. Note that an (n+1)-net of order
n is the same thing as an ane plane of order n. If P is a point in a k-net of order n and
1 , 2 , . . . , k are the lines of the net through P , then the point sets i ........ {P } are mutually
disjoint for i = 1, 2, . . . , k and so the number of points of the net collinear with P is
k(n 1) n2 1
from which it follows that k n+1 (and equality holds i the net is an ane plane).
Thus every set of mutually orthogonal Latin squares of order n has at most k1 members.
Given n 2, whether or not there exists an ane plane of order n, one can always ask for
the largest k for which there exists a k-net of order n. For small values of n this maximum
value of k is listed in the following table.
n
k
2
3
n
k
18
4
3
4
4
5
19
20
5
6
20
6
6
3
21
3
7
8
22
8
8
9
9
10
23
24
10
4
24
10
11
12
25
26
12
5
26
12
13
14
14
4
15
4
16
17
17
18
27
5
28
14
29
4
30
4
31
32
Note that it is still unknown whether there exist three mutually orthogonal Latin squares
of order 10. The cover of this book shows two orthogonal Latin squares of order 10.
Our construction of a k-net from a collection of k2 mutually orthogonal Latin
squares, appears to distinguish two parallel classes (the horizontal and vertical lines)
as somehow dierent from the other lines of the net. This is an accident of the presentation, which we hope to overcome by presenting nets in another way. A k-net of order n may
be specied by a collection N consisting of n2 vectors of length k (i.e. k-tuples), where
each of the k coordinates comes from a set of n symbols, such that every vector in N is
21
uniquely determined by any two of its coordinates. For example the pair of orthogonal
Latin squares1
1
1
2
3
1
2
3
a b c
c a b
b c a
corresponds to the 4-net

N = (, a, 1, 1), (, b, 1, 2), (, c, 1, 3),
(, c, 2, 1), (, a, 2, 2), (, b, 2, 3),

(, b, 3, 1), (, c, 3, 2), (, a, 3, 3) ;
the rule is that
N = {(ij , ij , i, j) : 1 i, j 3}
where ij is the (i, j)-entry of the rst Latin square, and ij is the (i, j)-entry of the second
Latin square. The lines of the net are the subsets with specied value of any one coordinate;
for example specifying the second coordinate to be c gives the line

(, c, 1, 3), (, c, 2, 1), (, c, 3, 2) .
This procedure extends to arbitrary k, and the last two coordinates behave no dierently
than the rst k2 coordinates. In particular any of the four coordinates may be used to
index the rows, and any of the three remaining coordinates may then be used to specify
the columns. For example if we index the rows by the 4th coordinate and columns by the
2nd coordinate, then the values of the 1st and 3rd coordinates become the entries of the
following two orthogonal Latin squares:
a
1
2
3
a
1
2
3
1 3 2
2 1 3
3 2 1
which give an alternative description of the same 4-net of order 3. This alternative description of nets will be examined further in Section 5.
4.1 Example: Translation Nets. Let F be an arbitrary eld, and let be a collection
of n-dimensional subspaces of F 2n , any two of which intersect only in {0}. We call
a partial spread. The members of , and their cosets in F 2n , form the lines of a
translation net on the point set F 2n . If || = k then this is in fact a k-net of order
1
Here we illustrate the historical origin of the term Latin square. We have a pair of orthogonal Latin
squares, the first with Greek letter entries, and the second with Roman letter entries; hence a pair of GraecoLatin Squares.
22
II. AFFINE PLANES
|F n |. Starting with an ane translation plane, any subset of the parallel classes will form
such a translation net. Given a translation net, one can therefore ask whether additional
parallel classes can be added to extend the net to an ane plane. This is not always
possible. However, if F is an innite eld, then given any partial spread with fewer than
|F | members, there always exists an extension of to a spread; that is, the translation
net may be completed to an ane translation plane. This may be shown by transnite
induction, an argument which relies on the Axiom of Choice (or Zorns Lemma).
Exercises 4.
1. Let L be the Latin square of order 4 arising from the Cayley table of a cyclic group of order 4. Show
that there is no Latin square of order 4 orthogonal to L.
2. Generalize Exercise 1 to the case of cyclic groups of even order.
3. Let Nk be a k-net of order 3 for k = 2, 3, 4. (Note: Each of the point-line incidence systems N2 , N3 ,
N4 is unique up to isomorphism, and is formed by choosing k parallel classes from A2 (F3 ).) Find
|Aut(Nk )| for k = 2, 3, 4.
4. A strongly regular graph with parameters (v, r, , ) is a graph with v vertices such that
(SR1) Every vertex has exactly r neighbours (i.e. is r-regular).
(SR2) Every pair of adjacent vertices has exactly common neighbours.
(SR3) Every pair of nonadjacent vertices has exactly common neighbours.
Given a point-line incidence structure (P, L), the collinearity graph of the structure is the graph
with point set P, in which two vertices are adjacent i the corresponding points are collinear. Show
that the collinearity graph of a k-net of order n, is strongly regular, and nd its parameters (v, r, , )
in terms of the parameters (n, k) of the net.
5. Nets and Webs

The most basic open problems concerning ane planes, listed at the end of Section 2, may
be reduced to certain open problems regarding codes of nets. We prefer to introduce this
approach by describing the analogous situation of webs over R and C.
Let F = R or C; you are encouraged to think rst of the more familiar case F = R.
Consider two parameterized curves
i : F F n ,
t i (t)
for i = 1, 2. We will assume that i (0) = 0; otherwise simply translate i appropriately

without changing any of the essential geometry we are about to describe. In the case F = R,
we think of i (t) as the position of particle i at time t. We require that i : F F n be
suciently smooth: in fact we assume that the n coordinates of i (t) are real or complex
analytic functions of t, according as F = R or C. Actually we do not require i (t) to be
dened for all t F , but rather just for all t in some open neighbourhood of 0 (say, in an
5. Nets and Webs
23
open interval (, ) in the real case, or in an open disk {z C : |z| < } in the complex
case). The Minkowski sum of the two curves is the surface
S = {1 (s) + 2 (t) : s, t F }.
(Here we may restrict s, t to the appropriate neighbourhood of 0 F as necessary.) The
surface S will be a smooth surface, in which every point 1 (s) + 2 (t) is uniquely specied
by the pair of coordinates (s, t), assuming that at every point of the surface S the curves
dened by s = constant and t = constant intersect transversely. This condition may
simply be stated as the requirement that the tangent vectors {1 (s), 2 (t)} be linearly
independent for all s, t in the appropriate domain.
We are especially interested in the case when there exist two additional curves 3 and
4 in S, whose Minkowski sum is also S. When this very strong condition holds, we call
such a surface S a double translation surface, following Sophus Lie who rst studied
this situation. Double translation surfaces were a prominent theme in Lies early papers, as
they arose as solutions of many interesting systems of dierential equations; in particular
many examples of minimal surfaces are examples of double translation surfaces. Note that
the surface S is ruled by four families of curves, these being translates of the four curves
i . These four families of curves form the structure of a 4-web on the points of S.
5.1 Example. Fix c F with c

/ {0, 1} and let
1 (s) = (s, 0, cs2 );
2 (t) = (0, t, t2 );
3 (u) = (u, cu, c(1 c)u2 );
4 (v) = (v, v, (c 1)v 2 ).
The Minkowski sum of the curves 1 and 2 , or of the curves 3 and 4 , is easily found to
be the quadric surface
S = {(x, y, z) F 3 : z = cx2 y 2 } F 3 .
24
II. AFFINE PLANES
5.2 Example. Fix c F with c

/ {0, 1} and let

1 (s) = s, 0, 2c s2 +(c+1)s ;

1
2 (t) = 1c 1ect , t, 2c
1e2ct ;

3 (u) = 0, u, ecu 1 ;

4 (v) = (v, 1c ln(1+v), 2c v 2 +cv .
It is straightforward to check that the Minkowski sum of the curves 1 and 2 , or of the
curves 3 and 4 , is the surface
S = {(x, y, z) F 3 : z = (x+1)ecy 1+ 2c x2 +cx} F 3 .
Here we must restrict the parameters to an appropriate neighbourhood of zero; for example
in the real case we require v > 1.
In the preceding examples n = 3. Do there exist double translation surfaces in F n
with n 4? Surprisingly, the answer is: No. Of course any surface in F 3 may be embedded
in F n where n can be as large as we like. But any double translation surface in F n with
n 3 must actually lie in a 3-dimensional subspace. This result was rst shown by Lie [40],
who actually showed much more1 than we are able to present here:
5.3 Theorem (Lie [40]). Every double translation surface S in F n for n 3, must
lie in a 3-dimensional subspace.
Before outlining a proof of this result, or describing its relevance to the study of nite
planes, we describe a generalization of Lies result. A k-web2 is a connected open neighbourhood of the origin which we denote W F 2 (or more generally a connected open
neighbourhood of a point in any F -surface, i.e. 2-dimensional manifold) together with a
k-tuple of smooth functions
xi : W F,
i = 1, 2, . . . , k,
such that at each point w W, every pair of the gradient vectors x1 (w), x2 (w), . . . ,
xk (w) F 2 are linearly independent over F . By the Inverse Function Theorem it follows
that
1
The more is that the tangent lines to the four curves i meet the plane at infinity at points whose locus is
an algebraic curve of degree 4 and genus 3. This curve need not be irreducible. If this conclusion does not make
sense to you now, take another look after reading Section 18 and maybe Section 19.
2
Actually what we define here is the special case of a 2-dimensional web. For the more general case, see
e.g. [16].
5. Nets and Webs
(5.4)
25
a general point w W is uniquely determined by any two of its coordinates

x1 (w), x2 (w), . . . , xk (w); for example given s = x1 (w) and t = x2 (w), we
may write the remaining coordinates as smooth functions of s and t.
We regard W as the point set of an incidence structure whose lines are the level curves
1
xi (a) W for every a F suciently close to zero. The condition on the derivatives
simply ensures that these curves intersect transversely, i.e. never tangentially; and assuming
W is suciently small, every point w W is uniquely determined by any two of its
coordinates x1 (w), x2 (w), . . . , xk (w). We may assume that xi (0) = 0 for all i. Consider
the vector space V = V(W, x1 , . . . , xk ) consisting of all k-tuples (1 , 2 , . . . , k ) of smooth
functions F F such that
1 (x1 (w)) + 2 (x2 (w)) + + k (xk (w)) = 0
for all w W. Also consider the subspace V0 V consisting of those k-tuples of functions
satisfying the additional condition that i (0) = 0 for all i. The quotient space V/V0 has
dimension k 1 simply because
V = V0 C
where C is the space of all k-tuples of constant functions (c1 , c2 , . . . , ck ) with c1 +c2 +
+ck = 0. The rank of the web W is by denition the dimension of V0 . We have
5.5 Theorem. The rank of a k-web W is at most (k1)(k2)/2.
For a proof in the complex case, see [16, p.49]. Webs of maximal rank (k1)(k2)/2
exist and are constructible from ane algebraic curves of degree k and maximal genus
g = (k1)(k2)/2 using Abels Theorem (or rather the converse thereof). We proceed to
show that
5.6 Proposition. Every double translation surface may be viewed as a 4-web. Moreover, Theorem 5.3 is the special case k = 4 of Theorem 5.5.
Proof. Consider a double translation surface S F n formed by curves i (i = 1, 2, 3, 4) as
above. For every point 1 (s) + 2 (t) S there exists a unique pair (u, v) = (u(s, t), v(s, t))
such that
(5.7)
Write
3 (u(s, t)) + 4 (v(s, t)) = 1 (s) + 2 (t) S.

1 (s) = 11 (s), 12 (s), . . . , 1n (s) F n ,
..
.

4 (v) = 41 (v), 42 (v), . . . , 4n (v) F n .
26
II. AFFINE PLANES
Dene the coordinate functions xi : F 2 F by

(x1 (s, t), x2 (s, t), x3 (s, t), x4 (s, t)) = (s, t, u(s, t), v(s, t)).
One checks from the assumptions on the curves i that these coordinate functions are as
required for a 4-web. For each i = 1, 2, . . . , n we have (1i , 2i , 3i , 4i ) V0 since the
i-th coordinate of (5.7) yields
1i (s) + 2i (t) 3i (u(s, t)) 4i (v(s, t)) = 0
for all s, t.
By Theorem 5.5 we may choose a basis for V0 of the form

{(1i , 2i , 3i , 4i ) : i = 1, 2, . . . , m}

where m 3 is the rank of W; also there exists an mn constant matrix C = cij F mn
such that
= C
where
11 (s)
12 (s)
1n (s)
21 (t)
22 (t)
2n(t)
=
;
31 (u(s, t)) 32 (u(s, t)) 3n (u(s, t))
41 (v(s, t)) 42(v(s, t)) 4n(v(s, t))
11(s)
12 (s)
1m (s)
(t)
22 (t)
2m (t)
21
=
.
31 (u(s, t)) 32 (u(s, t)) 3m (u(s, t))
41 (v(s, t)) 42 (v(s, t)) 4m (v(s, t))

The assertion = C simply says that the columns of are F -linear combinations of the
columns of , as required. Interpreted another way, however, it says that the rows of
are R-linear combinations of the rows of the constant matrix C, where R is the F -algebra
of smooth functions of s, t, u, v. This means that for all s, t, u, v, the rows of , namely the
vectors 1 (s), 2 (t), 3 (u), 4 (v), all lie in the row space of C, which has dimension at
most m 3. This proves Proposition 5.6.
Proof of Theorem 5.3. We restate the hypotheses in the language of 4-webs using Proposition 5.6, as follows. We have a neighbourhood W F 2 of the origin, and four coordinate
functions
xi : W F
such that at every point w W, every pair of the four gradient vectors x1 (w), . . . , x4 (w)
are linearly independent. Let
27
5. Nets and Webs
(5.8)
11
21
=
31
41
1m
12
22
32
2m
3m
42
4m
be a matrix of smooth functions ij : F F such that the columns of form a basis for
V0 ; in particular ij (0) = 0 and
(5.9)
1j (x1 (w)) + 2j (x2 (w)) + 3j (x3 (w)) + 4j (x4 (w)) = 0 for all w W.
Consider the four parameterized smooth curves

i (t) = i1 (t), i2 (t), . . . , im (t) F m ,
i = 1, 2, 3, 4.
By (5.4) there exist smooth functions u(s, t) and v(s, t) such that every point w W has
coordinates
(x1 (w), x2 (w), x3 (w), x4 (w)) = (s, t, u(s, t), v(s, t))
for some s, t; thus (5.9) becomes
1j (s) + 2j (t) + 3j (u(s, t)) + 4j (v(s, t)) = 0.
Partial dierentiation with respect to s (holding t constant) yields

(s) +
1j
u(s, t)
v(s, t)
3j (u(s, t)) +
4j (v(s, t)) = 0.
s
s
This gives the j-th coordinate in the vector relation

(5.10)
1 (s) + ()3 (u(s, t)) + ()4 (v(s, t)) = 0
in F m where () denotes smooth F -valued functions of s and t. Now we invoke (5.4) again,
this time solving for all coordinates in terms of s and u:
(x1 (w), x2 (w), x3 (w), x4 (w)) = (s, y(s, u), u,
v (s, u)).
Substituting into (5.10) gives
(5.11)
1 (s) + ()3 (u) + ()4 (

v (s, u)) = 0
where the expressions () are rewritten as smooth F -valued functions of s and u. Partial
dierentiation of (5.11) with respect to s (holding u constant) yields
(5.12)
1 (s) + ()3 (u) + ()4 (

v (s, u)) + ()4 (
v (s, u)) = 0.
28
II. AFFINE PLANES
Now let U1 F m be the subspace spanned by the tangent vectors i (0) to the curves i
at the origin, for i = 1, 2, 3, 4. Evaluating (5.11) at (s, t, u, v) = (0, 0, 0, 0), we see that
1 (0) 3 (0), 4 (0)F F m . By similar arguments with dierent subscripts, we see that
the subspace U1 F m is spanned by any two of the vectors 1 (0), 2 (0), 3 (0), 4 (0).
Let U2 F m be the subspace spanned by the tangent vectors i (0) and the bitangent
vectors i (0) for i = 1, 2, 3, 4. Evaluating (5.12) at (s, t, u, v) = (0, 0, 0, 0), we see that
1 (0) 3 (0), 4 (0), 4 (0)F . It follows that U2 = 3 (0), 4 (0), 4 (0)F .
Next we check that
(5.13)
(k)
all higher derivatives i (0) U2 for all k 1 and all i {1, 2, 3, 4}.
To see this, solve for all coordinates in terms of x1 (w) = s and x4 (w) = v as
(s, v), v).
(x1 (w), x2 (w), x3 (w), x4 (w)) = (s, y(s, v), u
Rewrite (5.12) in the form
y (s, v)) + ()4 (v) + ()4 (v) = 0
1 (s) + ()3 (
where () denotes smooth F -valued functions of s and v. Take the partial derivative of
(5.13) with respect to s (holding v constant) to obtain
y (s, v)) + ()3 (
y (s, v)) + ()4 (v) + ()4 (v) = 0.
1 (s) + ()3 (
Again evaluating at (s, t, u, v) = (0, 0, 0, 0), we obtain 1 (0) U2 , and similarly i (0)
U2 for all i. Repeated dierentiation yields a similar conclusion for higher derivatives.
Since i (0) = 0, the Taylor series for i (t) takes the form
i (t) =
tk (k)
(0)
k! i
k1
which yields i (t) U2 = 3 (0), 4 (0), 4 (0)F for all i {1, 2, 3, 4}. Thus
= 0
where is given by (5.8), is a 4 3 matrix whose entries are smooth F -valued functions
of the coordinates s, t, u, v, and

(0) 3m
(0)
31 (0) 32

(0) 42
(0) 4m
(0) F 3m
0 = 41

41
(0) 42
(0)

4m
(0)
is a constant matrix. The assertion = 0 simply expresses the requirement that the
rows of are R-linear combinations of the rows of 0 , where R is the F -algebra of smooth
5. Nets and Webs
29
functions in the coordinates s, t, u, v. Interpreted another way, however, it says that the
columns of are F -linear combinations of the three columns of matrix . Therefore V0
lies in the F -span of the three columns of .
We proceed to show how nite nets can be presented in a manner quite analogous to
our discussion of webs. We will see that nets are in some ways easier, and in other ways
more dicult, than webs. The nite case is a little easier in the sense that no smoothness
conditions are required of functions, and their identities and relations will hold globally
rather than in neighbourhoods of the origin. One diculty that we face is that many of
the arguments based on dierentiation as used in the case of webs, do not directly apply
in the nite case. This is because there is no appropriate notion of dierentiation for
functions dened over a nite eld. Although we can easily dierentiate polynomials, the
representation of a function as a polynomial is not unique, and this becomes an obstacle
if we try to mimic the proof of Theorem 5.3 in the nite case. Nor is the problem solved
by simply replacing derivatives by nite dierences.
For ease of presentation, we consider here only nets of prime order. We may view a
k-net of prime order p (where k 2) as a collection of k-tuples N Fkp such that every
w N is uniquely determined by any two of its coordinates. The points of the net are the
elements of N . The lines are the point sets with i-th coordinate equal to a for some xed
i {1, 2, . . . , k} and a Fp . The analogue of the coordinate functions of web theory, are
the maps
xi : N Fp , (a1 , a2 , . . . , ak ) ai .
For example the ane plane of order 3 is the 4-net of order 3 dened by
N = {0000, 0112, 0221, 1011, 1120, 1202, 2022, 2101, 2210}.
More generally the classical ane plane A2 (Fp ) is given by
N = {(a, b, a+b, a+2b, . . . , a+(p 1)b) : a, b Fp }
and by deleting p+1k of these coordinates we obtain a k-net of order p. In particular
the case k = 3 gives the cyclic 3-net
{(a, b, a+b) : a, b Fp },
so-called because it arises from a cyclic Latin square of order p (formed by the addition
table of Fp , a cyclic group of order p).
We dene the vector spaces V = V(N ) and V0 = V0 (N ) as before. Thus V is the
vector space over Fp consisting of all k-tuples (1 , 2 , . . . , k ) of functions Fp Fp such
that
1 (a1 ) + 2 (a2 ) + k (ak ) = 0
30
II. AFFINE PLANES
for all (a1 , a2 , . . . , ak ) N . We have

V = V0 C
where V0 V is the subspace consisting of all k-tuples satisfying the additional condition
that fi (0) = 0 for all i; and C V is the subspace consisting of k-tuples of constant
functions (c1 , c2 , . . . , ck ) Fkp with c1 +c2 + +ck = 0. Thus dim V = k 1 + dim V0.
5.14 Conjecture. dim V0 (k 1)(k 2)/2 and equality holds i N is isomorphic
to a subnet of A2 (Fp ), i.e. a net formed by k of the p+1 parallel classes of the classical
plane of order p.
To date only very limited progress has been made toward establishing this conjecture.
For example it is known to hold in the rst nontrivial case k = 3; see Theorem 5.15
below. Even a proof of the conjectured upper bound dim V0 3 for k = 4 (the nite
analogue of Theorem 5.3) would be a major breakthrough in this area! since the validity
of Conjecture 5.14 would imply that every ane plane of prime order is isomorphic to
A2 (Fp ).
5.15 Theorem ([42], [45]). Conjecture 5.14 holds for k = 3. That is, if N is a 3-net
of prime order p, then dim V0 (N ) 1, and equality holds i N is cyclic.
Proof. Let C be a complex primitive p-th root of unity, for example = e2i/p . For
any function f : Fp Fp dene the exponential sum
Sf =
f (a) Z[].
aFp
Note that |Sf | p, and equality holds i f is constant. Moreover Sf = 0 i f is a

permutation of Fp . Now given (f, g, h) V0 we have
(5.16)
f(a) + g(b) + h(c) = 0 for all (a, b, c) N ,
so
f (a)+g(b) = h(c).
Summing over all (a, b, c) N gives
(5.17)
Sf Sg = pSh .
Multiplying by Sh gives
Sf Sg Sh = p|Sh |2 .
31
By symmetry we must in fact have
Sf Sg Sh = p|Sf |2 = p|Sg |2 = p|Sh |2 .
If the latter expression does not vanish then |Sf | = |Sg | = |Sh | = 0 and so from (5.17)
we obtain |Sf | = |Sg | = |Sh | = p, so that the functions f, g, h are constant; but since
f(0) = g(0) = h(0) = 0 this means that f = g = h = 0. We may assume that this is not
the case, so that Sf = Sg = Sg = 0; that is, f, g and h are permutations of Fp . Changing
these permutations only amounts to permuting the lines in each parallel class, which is an
isomorphism of nets; therefore there is no loss of generality in assuming that
f(t) = t,
g(t) = t,
h(t) = t
for all t Fp . Now (5.16) becomes a + b c = 0 for all (a, b, c) N , so that

N = {(a, b, a+b) : a, b Fp } as required.
Exercises 5.
1. Find an explicit basis for V0 in the case of the 4-web dened in Example #5.1. What is the rank of
this web? Compare with the upper bound of Theorem 5.5.
2. Consider the two curves in R4 dened by
1 (s) = (s(s+2), s, (s+1)4 1);
2 (t) = (2t, 0, 2t2 4t).
The Minkowski sum of the two curves 1 and 2 is the surface
S : 2z = y4 + 4y3 + 4y2 + 2xy2 + 4xy + 4x x2

Compete the missing entries in the expressions
3 (u) = (u2 , u, ?);

4 (v) = (v(v+2), v, ?)
so that the curves 3 and 4 also have Minkowski sum equal to S, thereby forming a double translation
surface.
Erics white whale: Find optimal bounds for ranks of nets, and thereby
solve some of the long-standing open questions on possible orders of planes.
Or die trying!
32
34
PART III
Projective Planes
A projective plane is an incidence system of points and lines such that
(P1) For any two distinct points, there is exactly one line through both.
(P2) Any two distinct lines meet in exactly one point.
(P3) There exist four points such that no three are collinear.
Given any two distinct points P and Q, we denote the unique line joining P and Q by
P Q (pronounced P join Q) or simply P Q. Given any two lines and m, we denote
the unique point of intersection of and m by m (pronounced meet m) or simply
m. It is sometimes convenient to write P P = P and = .
The classical projective planes are constructed as follows: Let V be a 3-dimensional
vector space over an arbitrary eld F . Take as points and lines the subspaces of V of
dimension one and two, respectively. In this case P Q = P Q is the subspace spanned by
P and Q; and m = m is the intersection of the subspaces and m. Incidence is
containment: a point P lies on a line i P as subspaces of V . The resulting plane is
denoted P2 (F ) or P G2 (F ). In the case of a nite eld F = Fq we also denote this plane
P2 (q) = P2 (Fq ). The smallest projective planes, constructed from the eld F2 , is shown:
(0,1,0)
....
.........
... .... .....
.
.
. . .
... .. ...
... .... .....
...
...
.
.
....
...
.
..
...
...
1
..
...
...
.
.
.
...
.
.
.
.
... 0
.
.
.
.
.
.
.
.
... 0
.
....
...
...
..
...
...
.
.
...
.
.
.
..
...
....................................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
......... .....
. ..........
.
.
.
.
....... ...
. ........
.
.
.
.
.
.
...... ....
. ....
....
.......
.........
.......
..
.........
.
......
.
.
.
..
.
.....
.
.
.
.
.
... (0,1,1)
(1,1,0) ....
.
.
..
.......
0
.
.
.
.
..............
.
.
.. .........
.
.
.
.
.
.
.
.
.
.
1
.....
.......
...
.
.
......
.
.
.
.
.
.
......
.......
.
..
...
.
.
1
.
.
.
.
.
.
.
.
.
.......
......
..
...
...
.......
.. ..
.......
.......
.
.
.
.
.
.
.
.
.
.
.
....... .. ......
.. ....
. ..
.
.
.
.
.
.
.
.. ..
....... . ......
. ...
1
.
.
.
.. ...
.
.
.
.........
. ...
.
(1,1,1)
.
.
.
1
.
.. ...
.
.
.
...
.
... ... .........
.
0
.
.
.
...
.
.
.
.......
...
..
1
.
.
.
.
.
.
.
.
...
.
.
.
.
.
.
.
.
0 ..
.......
...
...
.
.
.
.
...
.
.
.
.
.
.
.
.......
...
.
.
...
1 ...
.
...
.
.
.
.
.
.
.
.
.
.......
.
...
..
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
1
.......
.
...
...
...
.
.
.
.
.
.
.
.
.
.
.
.
....... ..
.
.
... .........
...
1
.
.
.
.
.
.
........
.
.....
...
.
1
.
.
.
.
.
.
.
0
.
.
.
.......
...
.
... ....
.
.
.
.
.
.
.
.
.
.
.
.
.
0 ..
.......
...
.
....
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.......
...
....
..
.
.
.
1
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.......
.....
..
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....... .....
......
.
. ........
..
.
.
.
.
.
.
.
.
.
.
.
....... ...
.......
. ........
..
.
.
.
.
.
.
.
....... ...
.
.
.
.
.
.........
..
.........
.
.
.
.
.
.
.
......
.
.
.
.
.
.
.
.
..
..
(1,0,0) ................................................................................................................................................................................................................................................................ (0,0,1)
0
(1,0,1)
1
0

6.1 Figure
The smallest projective
plane P2 (F2 ):
7 points, 7 lines

We have chosen to label each point as (x, y, z), the 1-dimensional subspace spanned by
a nonzero vector (x, y, z). We refer to x, y, z as homogeneous coordinates for this
point, since these coordinates are dened only up to multiplication by a nonzero scalar
F : here (x, y, z) = (x, y, z). Each line is a 2-dimensional subspace of the form
36
III. PROJECTIVE PLANES
aX + bY + cZ = 0, where a, b, c F are not all zero. Again, scaling the coecients a, b, c by

any nonzero scalar gives the same line; so the line itself may be coordinatized as (a, b, c)T
where the column vector

a
(a, b, c)T = b
c
is dened only up to nonzero scalar multiple. In summary, points and lines may be viewed
as row and column vectors of length 3 respectively, using homogeneous coordinates in each
case; and the point (x, y, z) lies on the line (a, b, c)T i

a
0 = (x, y, z) b = ax + by + cz.
c
By choosing row vectors for points and column vectors for lines, the incidence relation is
thus expressed very simply in terms of matrix multiplication.
In the eld F2 the only choice of nonzero scalar is = 1. So lets consider the
next-smallest projective plane, coordinatized by F3 :
(0,1,0)
...................................
..................... .......... ............... ..................................
......
.
..........
.............
.
.
.
.
.
.
.
.
.
.
.
.
.
....
.
.
..........
.....
.....
.........
......
..
...................................
.....
.......................................
....
..
.........................
....... ........
..............
......... .......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...... .......
..........
..........
.
... ........
.
.
.
.
.
....... .....
.
.
.
.
.
......
.....
...... .....
..... ......
...
..... ......
.... .........
.
...... .....
...............
.
.
.
.
.
.
.
.
.
.....
.
...
.
...... ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
. .....
.
....
..... ...
.
.
.
.
.
.
.
.
.
.
.
.
.
...
.....
.
.
.........................................
.
.
.
.
.
.
.
.
.
.
.
.
.
..............................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.... . ....
..........
.
...
....
.............
.
.
..........
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.......
.... . ...
.
..
.
..........
.
......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..........
...... ..
... ......
.... ..............
..... .. ...
...... ..
.........
..... ....................
..... .... .......
............
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
...
.....
.
. .. .....
.. .... .....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
.
.
.. .... ....
..
... .
... .. ..
....
... .......
...
....
.. .... ...
.... ...
...
.... ... ........
....
....
...
..
.....
..
...
....
....
.
... ...
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.....
....
.
... ..
....
.
..
.
.
.
.
.
.
.
.
..
...
..
.
.
.
.
.
.
.
.
.
.
... ...
.
.
.
.
.
.
.... ...
....
.... ....
.
.
.
.
.
.
...
...
.
.
.
... ..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
.
....
.
.....
.
.
.
.
.
.
.
.
..
.
......
.
.
.
.
...
.
.
.
.
.
.
....
.
. .....
.
.
.. .....
.
.
..
.
.
.
.
.
.
.
......
.
.
..
.
.
.
.
.
.
.
.
....
....
.
....
..
.
.
...
.
..
.
.
.
.
.
.
.
.
.
.....
.
.
.
..
.
.
.
.
.
.
.
.
....
.
....
.....
.
..
.
.
.
.
.
.
.
.
.
.
.
....
.
.
.
.
.
...
.
.
.
.
.... . ....
.... ..
...
.
.... .. ....
.
.
.
....
.
.
.
.
.
.
.
.
.
.
.... .. ....
..... .. ...
...
.... . ....
.
.
.
.... .. ....
...
.
.
.
.....
.. ..
.
.
.
.... . ...
.... . ....
.
.
.
.
.. ....
.
.
.
.
.
.
.
.
.
.
.
.
....
...
...................................................................................................................................................................................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. ....
....
............
(0,2,1) .. ....
...
. ..
. ... .....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
....
..
...
.
.
0
..
.. ....
....
..... ............
... .......
.... ...
...
..
.... .... ........
... 0
....
......
....
...
...
.....
.....
...
....
...
.....
....
...
....
......
.....
....
..
..
....
.....
...
....
...
.
....
.
.
1
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..... ..
.... ....
...
.... ....
....
..
...
.. .
....
....
....
... ...
........
....
....
....
...
....
............
....
..
..
..
....
.........
.........
.
.
.
.
.
.
.
.
....
.
.
.
.
.
....
.
.
.
.
.
.
.
.
.
....
....
....
..
....
...
....
...
....
....
....
....
....
....
....
....
....
....
....
.
....
....
....
.....
...
....
.
.
....
....
....
....
....
....
.... ...
.... ... .......
.... .... .......
....
....
.... ..... ........
.
.... .. ....
.
.
.
.... . ....
.
.
.
....
.... ....
.... . ....
.
.... ... .....
.
.... .. ....
.
....
.
.
.
.
.
.
.......
.......
.....
......
....
...
.
....
(0,1,1) ..............................................................................................................................................................................................................................................................................................
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
.
.. (1,0,0)
.... ...
..... ... .......
... .......
....
.....
....
....
....
...
....
....
....
...
...
1 ....
....
....
....
....
....
....
....
..
..
....
....
.... .......
....
.
.... .......
.
.
.
.
0
.
.
.
.
.
.
.... ....
....
.... ....
.
....
....
....
....
....
0 ....
.....
.....
....
.......
....
..
..
.... ....
.....
....
.... .......
....
...
....
...
.... ........
..... 0
....
....
....
.
.
.
.
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
....
.
.
....
1
....
..
....
..
....
.....
......
.... .... ........
.... ... .......
.... ... ......
......
0
.... .. ....
......
.... .. ....
.... .. .....
.... . ....
.......
.
..........
.........
.
.
.
..........
.
.
.
.
.
(0,0,1) ........................................................................................................................................................................
6.2 Figure
The second-smallest
projective plane
P2 (F3 ):
13 points, 13 lines

(1,0,1)
(2,0,1)
Here we have shown homogeneous coordinates for only seven of the points and three of
the lines. In Exercise #1 you are asked to ll in the missing coordinates.
Note that if V is 3-dimensional over Fq then V has q 3 1 nonzero vectors, whereas
every point (1-dimensional subspace) contains q 1 nonzero vectors; therefore the number
of 1-dimensional subspaces (i.e. points) is (q 3 1)/(q 1) = q 2 + q + 1. The number of
lines must also be q 2 + q + 1, by applying the same argument to column vectors instead
of row vectors. Every line has q + 1 points, which we check as follows: Each line is
37
a 2-dimensional subspace, with q 2 1 nonzero vectors, partitioned into subsets of size

q 1 by the 1-dimensional subspaces, so the number of 1-dimensional subspaces (points)
is (q 2 1)/(q 1) = q + 1. Similarly (interchanging row vectors and column vectors)
every point lies on q + 1 lines. Alternatively, consider a point (1-dimensional subspace) P .
Then lines (2-dimensional subspaces) containing P correspond bijectively to 1-dimensional
subspaces of the quotient space V /P which is 2-dimensional. As we have seen, the number
of such subspaces is exactly q + 1.
We now consider more general projective planes than the classical planes P2 (F ).
6.3 Theorem. Let (P, L) be a projective plane. Then there are equally many points
and lines, i.e. the sets P and L have the same cardinality, nite or innite. Moreover
any two lines contain the same number of points, and this number equals the number
of lines through any point. If n + 1 is the number of points on every line (hence also
the number of lines through every point) then |P| = |L| = n2 + n + 1.
The number n is called the order of the projective plane. This means that the order of
the classical plane P2 (F ) is exactly |F |, the order of the eld F (nite or innite) although
the number of points on every line is |F | + 1.
Proof of Theorem 6.3. Denote by [P ] the set of lines through a point P , and by [] the
set of points on a line . If P
/ then an obvious bijection [] [P ] consists of mapping
each point Q to the line P Q.
P
...
......................
......... .. ..................
............ ... ..... ...........................
.
.
.
.
..... .... ... .... .......... ...............
..
...
......
...... ... ...
...... ................
...
...... ...
..
........
......
..... ......
...
.
.
.
.
.......................
......
...
...
...
..
.
.
.
.
.
.
.
.
.
.
..
................................
.... ..................................................................
......
.
.
.
.
.
.
.
.
.
.
.
...............................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
Let and m be distinct lines. By axiom (P3) there exists a point P

/ [] [m], and the
previous argument gives |[]| = |[P ]| = |[m]|. Thus any two lines have the same number of
points; denote this cardinality by n + 1 (nite or innite). Let P be any point. By axiom
(P3) there exists a line not passing through P ; then |[P ]| = |[]| = n + 1.
We say that the theory of projective planes admits duality in the following sense:
6.4 Theorem. If (P, L) is a projective plane, then the dual structure (L, P) is also
a projective plane.
38
Proof. Let (P, L) be a projective plane. Since (P, L) satises (P1) and (P2), the dual
structure (L, P) satises (P2) and (P1). Since (P, L) has four lines with no three concurrent by Exercise #2, (L, P) satises (P3).
By contrast, note that the dual of an ane plane is not even an ane plane. The dual
of an ane plane contains pairs of points which are not joined by any line; such a pair of
points arises as the dual of a pair of parallel lines.
We show here the projective plane of order two, along with its dual, which is also a
projective plane of order 2:
E
b
.......
........
.
..
.. .. ...
.. .. ...
... .. ....
... .. ...
... .... .....
.
.
...
..
....
...
...
..
..
...
.. .................................... .....
.
.
.........
.........
.
.
.
.
........
.....
..
.
.
.....
.
.
..
....
...
.......
.
.
.
.
. ........
.............
.
.
.
.
.
.
.
.
.
.
.......
..
......
.
....
.
.
.
.
.
.
.
.
.
.
....... .. ......
..
......
................
.....
.. ...
... ....
...................
.
.
.
.
.
... .....
.
.
.. ... .......
. ...
.
.
.
.
.
.
.
.
...
.
.
.......
.
...
...
.
.
.
.
.
.
.
.
.
...
.
.
.
.
....... ..
.
... ........
..
........
....
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. ........ .....
. ........ .....
.
.
.
.
.
.
.
....... ...
.....
..
. ........
.
.
.
.
.
.
.
.
.
.
.
.
..........
.......
..
.. .......
.......
..
.
............................................................................................................................................................
F c
C
g
b
.
... .. ....
... .. ...
... .... .....
.
.
.
...
....
...
...
..
.
...
... .............................. .....
.
.
....... ...
..........
.
.
.......
.
.
......
.
.
.......
.
.
.
....
.....
....
........
.
.
.
. .......
..
.........
.
.
.
.
.
.......
..
... . ........
.
.
.
.
.
.
.
.
.
.
....... . .......
......
...
.
.
.
.
.
.
...............
..
... .....
... ...
..........
. .
...... .... ............
.
... ....
.
.
.
.
.
.
... .....
.
.......
...
...
.
.
.
.
..
.
.
...
.
.
.
.
.
....... ..
.
... .........
...
........
....
...
...
..
............... .....
... ..................
.
.
.
.
.
.
.....
....... ...
.
.
. .........
.
.
.
.
.
.
.
.
....... ..
......
..
... ..
...
............
...
.......
...........................................................................................................................................................
f G
g
A
C
Here it happens that the dual plane is isomorphic to the original plane; we say therefore
that this plane is self-dual. Not every projective plane is self-dual, but every classical
plane P2 (F ) is:
6.5 Theorem. Every classical plane (P, L) is isomorphic to its dual.
Proof. Consider the map : (P, L) (L, P) which transposes each vector of length 3.
Writing
P = v = (x, y, z)
and
= wT = wT,
we have
P

in (P, L)
vwT = 0 i
wv T = 0

P
.
in (L, P)
There are exactly four projective planes of order 9 up to isomorphism: the classical
plane P2 (F9 ) and the Hughes plane of order 9, both of which are self-dual; the Hall plane
(the projective completion of the plane constructed in Example 3.5) and its dual.
A list of orders of the smallest known planes is given in Table 6.6. To show that
planes of order less than 9 are classical, is rather straightforward and has been known
for a long time. The classication of planes of order 9 is also relatively recent [37], and
39
relies heavily on case-checking by computer. The proof of

nonexistence of a projective plane of order 10 due to Lam
et. al. is almost as recent and also largely based on computer [36], [38]; some of the work leading up to this result
will be described in Section 13. Prior to 1989 the question of
existence of a projective plane of order 10 was probably the
most prominent open question in nite geometry. Today
the more general question of existence of a nite projective
plane whose order is not a prime power, is the most signicant open problem in nite geometry; and the second may
well be the question of whether projective planes of prime
order are necessarily classical.
6.6 Table. Projective Planes of Small Order
n
2
3
4
5
7
8
9
10
11
13
16
17
19
23
25
27
29
31
no. of planes
no. of planes
Remarks
of order n
of order n
up to isomorphism up to iso./duality
1
1
P2 (F2 )
1
1
P2 (F3 )
1
1
P2 (F4 )
1
1
P2 (F5 )
1
1
P2 (F7 )
1
1
P2 (F8 )
4
3
Lam et. al. [37]
0
0
Lam et. al. [36]; [38]
1
1
P2 (F11 )
1
1
P2 (F13 )
22
13
Royle [56]
1
1
P2 (F17 )
1
1
P2 (F19 )
1
1
P2 (F23 )
193
99
[22], [46]
13
8
[26], [46]
1
1
P2 (F29 )
1
1
P2 (F31 )
Exercises 6.
1. Finish labelling Figure 6.2 by nding homogeneous coordinates for the six remaining points and the
ten remaining lines.
2. Show that in every projective plane, there exist four lines such that no three are concurrent. (This is
the dual of axiom (P3).)
40
3. A t-(v, k, ) design is a point-block incidence structure with v points such that each block contains
exactly k points, and such that every t-set of points is contained in exactly blocks. Show that a
projective plane of order n 2 is the same thing as a 2-(n2 +n+1, n+1, 1) design.
Remark. The case n = 1 gives a 2-(3, 2, 1) design, which is just a triangle (3 points, 3 lines). This
geometry is therefore often called the projective plane of order 1, or the thin projective plane
(those of order n 2 being thick). More about this in Section 19.
4. Let S 2 be the unit sphere in R3 . For each point x S 2 denote by x S 2 the antipodal point
opposite to x. Consider the incidence system (P, L) in which points P P are dened as pairs
P = {x, x} of antipodal points of S 2 ; and lines L are great circles of S 2 , i.e. intersections of S 2
by planes passing through the centre of S 2 .
(a) Show that (P, L) is a projective plane.
(b) By considering the intersections of subspaces of R3 with the unit sphere S 2 , show that (P, L) is
isomorphic to the real projective plane P2 (R).
7. Projective Completion of Ane Planes

Let (P, L) be a projective plane, and let 0 L be any line. Deleting 0 and all its points
from (P, L) gives an incidence structure (P0 , L0 ) where
P0 = {P P : P
/ 0 },
L0 = L ........ {0 }.
It is not hard to verify that (P0 , L0 ) is an ane plane. Clearly any two points of P0 are
/ 0 , then the unique line of
joined by a unique line of L0 . If P P0 and L0 with P
L0 through P not meeting 0 , is the line P R where R = 0 P. (Note that R
/ P0 so
the lines and P R meet only at a point of P; they have no point of P0 in common.)
The map (P, L) (P0 , L0 ) described above is uniquely reversible: given an ane
plane (P0 , L0 ), one may uniquely reconstruct the missing line along with its points, thereby
forming the projective completion (P, L) of the given ane plane. This process is
described as follows:
(1) For each parallel class of lines in (P0 , L0 ), add one new ideal point to all lines in
the class.
(2) Add one new line whose points are the ideal points added in Step (1).
The new line added in Step (2) is sometimes called the line at innity and may be denoted
. From the viewpoint of the original ane plane, this new line looks rather dierent
from the other lines; but from the viewpoint of the projective plane (P, L), there is not
typically any feature that distinguishes one line from another.
7.1 Example: The Projective Plane of Order 2. Let (P, L) be the projective
plane of order 2, as shown on the left below. By deleting the line d and its three points
41
D, E, G we obtain the ane plane of order 2 with point set P0 = {A, B, C, F } and line
set L0 = {a, b, c, e, f, g}.
E
...
........
... .. ...
... .... .....
.
.
.. .. ....
...
... ...
...
...
....
..
.
.
.
. ...................... .....
.
.
. ......... ... .......... ...
.
.
.........
.
........
.
.
.
........
.
...
...
....
....
...
...........
......
.
.
.
.. ........
.
....... ......
.
.
.
.
.
.
.
.
.
.
.
.......
..
...
......
.
.
.
.
.
.
.
.
.
.
.
....... . ......
.. ..
. ..
...........
.
... ..
... ....
... .....
......................
.
.
.
.
.
.
.
... .....
. ...
... .. .........
.
.
.
.
.
.
.
.
.
...
....... ..
.
.
... .........
.
.
.
.
.
...
.........
.
.......
...
..
...
.......
........
...
... ............ .....
... ............ ......
.
.
.
.
.
.
....... ..
......
.
. ........
.
.
.
.
.
.
.
.........
.
.
.
........
..
........
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.......................................................................................................................................
F c
C
g
b
F ...........
.
... ............
.......
......
.......
.......
.
.
. ..
......
...... ....
... ...
.
.
.
.
.
.
.
.
...
.
...
.
.
.
.
.
....
.
.
.
... ........
..
...
....
..
... ............ .....
.
.
...
.....
. ........
.
.
...
.
.......
.........
.
.
.
.
..............................................................................
C
b
From the resulting ane plane we may uniquely reconstruct the original projective plane,
as follows:
(1) Add the ideal point D to both lines of the parallel class {a, c};
add the ideal point E to both lines of the parallel class {b, e};
add the ideal point G to both lines of the parallel class {f, g}.
(2) Add the ideal line d whose points are just the ideal points D, E, G added in
Step (1).
The choice of the line d P was arbitrary; by deleting any of the seven lines, along with
its points, we obtain an ane plane of order 2; and any of the resulting ane planes are
in this case isomorphic.
7.2 Example: Classical Planes. Now consider a general eld F . Let (P, L) be the
classical plane P2 (F ), and let 0 be a line. We may suppose, after a linear change of
coordinates if necessary, that 0 is the line (0, 0, 1)T . Thus P0 consists of all points
(x, y, z) not lying on 0 . All such points have z = 0, so we may multiply coordinates by
1
z to write every point of P0 uniquely in the form (x, y, 1) where x, y F are arbitrary.
Every line L0 arises from a line (a, b, c)T where (a, b) = (0, 0) since = 0 . If b = 0
then has the form y = (a/b)x (c/b); otherwise b = 0 and a = 0 so has the form
x = c/a. Clearly (P0 , L0 ) is isomorphic to the classical ane plane A2 (F ) under the
map (x, y, 1) (x, y).
The reverse process of reconstructing the projective completion of a given classical
ane plane A2 (F ) is also clear:
(1) For each m F {}, lines of slope m form a parallel class of lines in the ane
plane (P0 , L0 ). Add a new ideal point to every line of such a parallel class. (Lines
of slope are simply vertical lines of the form x = constant.)
(2) Add an new line, whose points are just the ideal points named in Step (1).
It is worth emphasizing that when completing a given ane plane to a projective plane,
each pair of parallel lines in the ane plane picks up just one ideal point of intersection
in the projective completion. If one stares down a railroad track and imagines the two
42
parallel rails meeting at innity, one might be tempted to associate two ideal points of
intersection, one at each end, + and . This viewpoint misses the fact that the
two endpoints of the projective line are eectively identied as one. In particular the real
projective line is homeomorphic to a circle S 1 . And for more general elds, which are not
generally ordered, there is no distinction between positive and negative anyway; so there
is no way to distinguish more than one sign of innity.
We also emphasize that our process of completing the ane plane to a projective
plane (whereby many new ideal points are added at innity) is distinct from forming the
one-point compactication of a plane (in which just one new ideal point is introduced).
Note that the one-point compactication of R2 is homeomorphic to the sphere S 2 , often
identied with the Riemann sphere P1 (C) = C {}, e.g. via stereographic projection.
Consider the embedding of the classical ane plane A2 (F ) in its projective completion
P2 (F ) as the complement of a line 0 , which we may take to be the line (0, 0, 1)T ; this
embedding, as described above, maps points, lines, and more general algebraic curves, via
A2 (F ) ................................................................................................................................................ P2 (F ) ........ 0
.
point (x, y)

Y
point X
Z, Z

.......................................................................................
.......................................................................................
point (x, y, 1)

point (X, Y, Z), Z = 0

Y
curve Z k f X
Z , Z =0

curve g(X, Y, Z)=0,
..................dehomogenize
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
curve g(x, y, 1)=0
g(X, Y, Z) homogeneous
curve f(x, y)=0,
deg f = k
................homogenize
.......................................................................

Y
For example consider the ane parabola y = x2 . Substituting (x, y) = X
,
Z Z , and then
2
multiplying both sides by Z to clear the denominators, the equation becomes Y Z = X 2 .
The latter curve in P2 (F ) has one more point than the original curve, namely the point
(0, 1, 0) 0 . Likewise the ane hyperbola xy = 1, when homogenized, becomes XY =
Z 2 ; here the curve picks up two points at innity, namely the points (1, 0, 0, (0, 1, 0)
0 . Of course the two conics Y Z = X 2 and XY = Z 2 are equivalent under a simple change
of variable (interchanging X and Z). Thus the original two ane conics, one a parabola
and the other a hyperbola, although not anely equivalent, nevertheless are projectively
equivalent regardless of the choice of eld F .
Put another way, one may start with the conic XY = Z 2 in the projective plane, one
may delete the line 0 : Z = 0, a secant line meeting the conic in two points, (1, 0,0) and

Y
,
(0, 1, 0), to obtain the ane hyperbola xy = 1 in the (x, y)-plane, where (x, y) = X
Z Z .
Or one may delete the line 1 : X = 0, a tangent line meeting the conic in just one
Y point

Z
2
.
(0, 1, 0), to obtain the ane parabola y = z in the (y, z)-plane, where (y, z) = X , X
We return our consideration to the case of general (not necessarily classical) ane
and projective planes. Every ane plane (P0 , L0 ) has a unique completion to a projective
43
plane (P, L) as described above. But given a projective plane (P, L), distinct choices
of line 0 , 1 L yield ane parts (by the deletion of 0 or 1 ) which in general need
not be isomorphic. If there exists Aut(P, L) such that 0 = 1 , then induces an
automorphism from the ane plane (P........[0 ], L........{0 }) to the ane plane (P........[1 ], L........{1 }).
Because the projective plane (P, L) is uniquely reconstructible from either (P........[0 ], L........{0 })
or (P ......... [1 ], L ........ {1 }), the condition that these two ane planes are isomorphic forces 0
and 1 to be in the same orbit of Aut(P, L). Thus
7.3 Theorem. Let 0 , 1 L be lines of a projective plane (P, L). Then the ane
planes (P ......... [0 ], L ........ {0 }) and (P ......... [1 ], L ........ {1 }) are isomorphic i 0 = 1 for some
Aut(P, L).
7.4 Example: Planes of Order 9. It is known [37] that there are exactly four isomorphism classes of projective planes of order 9: the classical plane P2 (F9 ), the (projective)
Hall plane of order 9 (the projective completion of the ane translation plane described
in Section 3.2), the dual of the Hall plane, and the Hughes plane of order 9. The number of line orbits in these planes is 1, 2, 2, 2 respectively. This gives a total of exactly
1 + 2 + 2 + 2 = 7 isomorphism classes of ane planes of order 9.
Exercises 7.
1. The known projective planes of order 27 are listed as follows, along with the number of line orbits of
the full automorphism group in each case:
No.
(a)
Plane
P2 (F27 )
Line Orbit Sizes
(classical)
757
(b)
Generalized twisted eld plane
1, 27, 729
(c)
(c )
Hering plane
dual of (c)
1, 756
28,729
(d)
(d )
Flag-transitive ane plane (completed)

dual of (d)
1, 756
28,729
(e) Another ag-transitive ane plane (completed)

(e )
dual of (e)
1, 756
28,729
(f)
(f )
Sherk plane
dual of (f)
1, 27, 729
1, 27,729
(g)
(g )
Andr
e plane
dual of (g)
1, 54, 702
2, 26,729
(h)
Figueroa plane
13, 312, 432
Planes (a), (b), (h) are self-dual. How many isomorphism types of ane planes of order 27 are known?
2. Show that the cubic curves y = x3 and y2 = x3 are projectively equivalent.
3. Show that the curves y = x2 and x2 + y2 = 1 are projectively equivalent over F3 and over F5 , but
not over F2 .
44
8. Advantages of the Projective Viewpoint

The axioms for projective planes are clearly simpler than those for ane planes. This
suggests that for many purposes, the projective point of view is simpler than the ane
point of view. We proceed to provide instances of this.
First of all, as we have seen, the theory of projective planes exhibits duality: the dual
of every projective plane is also a projective plane. And because the dual of a classical
plane is again classical, every theorem valid in the classical plane has a dual which is also
valid in this plane. To illustrate the utility of this principle, consider the following result,
known as the Theorem of Pappus. This basic result of classical plane geometry says that
if
and m are distinct lines in P2 (F ), meeting in a point O;

P0 , P1 , P2 are distinct points of other than O;
Q0 , Q1 , Q2 are distinct points of m other than O;
ij = Pi Qj for all i = j in {0, 1, 2};
Ri = i+1,i+2 i+2,i+1 (with subscripts mod 3)
then the three points R0 , R1 , R2 must be collinear.
8.1 Figure.
Pappus Conguration
This property does not hold in more general planes; a projective plane satises this condition i it is isomorphic to P2 (F ) for some eld F . Later (Theorem 11.2) we will prove
one direction (the only if part) of this result. The dual of Pappus Theorem, which also
must hold in classical planes, states that if
and m are distinct points in P2 (F ), joined by a line O;

P0 , P1 , P2 are distinct lines through other than O;
Q0 , Q1 , Q2 are distinct lines through m other than O;
ij = Pi Qj for all i = j in {0, 1, 2};
Ri = i+1,i+2 i+2,i+1 (with subscripts mod 3)
45
then the three lines R0 , R1 , R2 must be concurrent.
8.2 Figure.
Dual Pappus Conguration
The Theorem of Pappus yields many theorem valid in the classical ane plane A2 (F ), in
particular the Euclidean plane A2 (R). One such theorem is illustrated by the gure
where we have chosen to be a line through O other than or m; the picture shows the
ane plane formed by the complement of . In the following gure we have designated
pairs of parallel lines as usual with matching arrow markings. Here we choose itself to
be the line at innity:
And in the following example, the line 12 is taken as the line at innity:
46
Of course the original version of Figure 8.1 (with no parallel lines) also holds in the classical
ane plane. Further ane versions of this result may be listed; and then many ane
versions of the dual of Pappus Theorem may also be listed. But all of these theorems valid
in the classical ane plane are summarized a single result of classical projective geometry,
namely the original Theorem of Pappus.
A generalization of Pappus Theorem, known as Pascals Theorem, holds in the
classical projective plane P2 (F ): here the pair of lines {, m} is replaced by a conic. The
8.3 Figure.
Pascals Conguration
sense in which this generalizes Pappus Theorem, is that a pair of lines may be seen as
a degenerate conic. This theorem (whose proof we omit) has also a dual form, valid in
the classical projective plane; to dualize, points of the conic are replaced by tangent lines,
thus:
8.4 Figure.
Dual of Pascals Conguration
Both Figures 8.3 and 8.4 have multiple ane versions, as we saw with Pappus Theorem;
and all these versions are economically summarized in the single Theorem of Pascal. (In
47
fact it is possible to give one proof that covers both Pascals Theorem and Pappus Theorem.) One of the reasons for the multiple ane versions of Pascals Theorem is that in
the ane plane there are three types of nondegenerate conics: ellipses (including circles),
hyperbolas, and parabolas. If one were to prove Pascals theorem in the ane setting,
separate arguments would be required for all three cases. In the projective setting, all
three cases are the same! This is because what appears to the ane observer as three
separate types of conic (ellipses, parabolas, or hyperbolas) are simply the ane portions
of a projective conic, formed by deleting some line , which meets the conic in 0, 1 or 2
points respectively (i.e. is a passant line, a tangent line, or a secant line to the conic,
respectively):
.
.
.
...
...
...
.......................................
.
.
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.........
.....
.
...
.
.
.
.
.
.
.
.
.......
...
.
...
.
.
.
.
.
.....
..
.
.
.
.
.
.... ....
...
.
.. ...
...
.. ..
...
.. ....
...
...
...
....
...
...
.
.....
.
.
.
...
..
......
.......
........
...
.
.
.
.
.
.
.
.
.............
...
...................................................
...
...
..
Ane Ellipse
...
...
...
.......................................
.
.
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.........
.....
.
.
.
.
.
.
.
.
.
....... ....
...
.
.
.
.
.
.
........
...
.
.
.......
.
.
...
...
...
.....
...
...
........
...
.
.
.
....
. ...
.
.
.
......
.
...
.....
.......
...
.......
..........
...
..........
..................
....................................
...
...
..
Ane Parabola
...
...
.
.........................................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
..........
.....
.
.
.
.
.
.
.......
.
...
...
.
.
.
......
.
.
...
....
.....
...
...
....
.
...
...
...
...
...
..
...
...
...
.
...
....
.
.
.
.....
.
.
.
.
...
..
......
......
........
..
........
............
.........................................................
...
...
.

Ane Hyperbola
In comparing the projective and ane viewpoints to plane geometry, we recognize

several aspects of the projective viewpoint which recommend this as the easier, or more
natural approach in many situations:
(1) The axioms for projective planes are much more concise than those for ane
planes.
(2) The dual of a projective plane is simply a projective plane. The dual of an ane
plane is not an ane plane; it is just . . . well, its a dual ane plane, OK?
(3) If ones goal is to classify planes, the list of projective planes of a given order is
usually much simpler than the list of ane planes of that order. Consider for
example the 7 known ane planes of order 9, which are equivalent to the list of
4 projective planes of order 9 up to isomorphism, or simply 3 projective planes
up to isomorphism/duality (Example 7.4).
(4) It is possible for one object (such as a curve or point-line conguration) in the
projective plane to appear as several distinct objects in its various ane sections.
This means that one theorem in projective plane geometry can replace several
theorems in the ane setting; or that one theorem can require a single proof in
the projective case but many dierent proofs in the ane case.
(5) Some natural properties of the projective plane have no readily described ane
counterpart.
As an example of (5) we consider the problem of counting intersection points of plane
curves. The curves we consider are algebraic curves, i.e. zero sets of polynomials. The
degree of such a curve is the degree of its dening polynomial. As we have seen, we require
homogeneous polynomials in order to obtain well-dened curves in the projective plane.
48
The following examples of curves are dened over the real eld R, although the same
principles apply to arbitrary elds. We begin with the observation that a line (a curve of
degree 1) generally meets a conic (a curve of degree 2) in two points; the case illustrated in
Example 8.5 below is typical. In Example 8.6 the line is tangent to the circle and the point
of intersection is a double point; so if points are counted with appropriate multiplicity, the
number of intersection points is again 2. While it is possible to give a precise denition
of intersection multiplicity, as is done in any course on algebraic curves, we will not do so
here and now; we trust that a few examples will convey the general idea.
8.5 Example.
...
... (1, 3)
..........
..........................
........
. ..
... ......
..
......
... ......
....
... ....
....
.
.
...
.
.
.
... .....
.
.
.
.
...
.
... ....
....
... ..
... ..
.....
... ...
...
... ..
...
......
...
...
...
.....
...
... ....
...
.
. ...
....
.
.
...
.....
...
...
......
....
.....
.......
...
.............. ....................
...
.......
.
(3, 1)
x2 +y 2 =10
2x+y=5
8.6 Example.
8.7 Example.
...
...
..........................
...
...........
.......
.
.
.
.
.
.
..... .....
...
.
.
.
.
.
.
.
.... ...
.
.
.
.... ....
..
... ...
...
... ...
...
.. ..
...
.
.. ...
.. ..
....
...
...
..
..
...
..
...
..
...
...
.
.
..
...
..
...
.
...
.
.
...
.
...
.
.
...
.
...
.
....
.
.
...
.
....
.
.
.
...
......
..
.
.
.
.
...
.......
.
............ ...................
...
...........
..
...
.
.......................... ....
...........
....... ...
.
.
.
.
.
.
..... ..
...
.
.
.
........
.
...
......
....
....
...
.
...
.
..
.
....
.
.
.....
....
.....
...
. ..
...
.. ...
.. ....
...
.
...
.
...
...
...
...
...
...
...
.
.
...
.
...
.
...
.
....
.
..
...
.
.....
.
.
..
.
......
...
.
.
.
..
........
.
.
...
.
.
.
.
................................
...
.

(2 2, 2)
x2 +y 2 =10
2x+y=5 2
x2 +y 2 =10
2x+y= 15
2
In Example 8.7 no intersection points of the indicated line and circle are visible; this is

because the points of intersection (found algebraically) are 3+ 12 i, 32 i and 3 12 i, 32 +i .
Although the curves themselves are dened over R (meaning that the dening polynomials
have real coecients) the actual points of the curve may have coordinates in any algebraic
extension. In particular the set of all points of a curve is considered to be those points
with coordinates in the algebraic closure of the eld over which the curves are dened. All
our examples will be dened over R and so their points will have complex coordinates.
Similarly when intersecting two conics we nd typically 4 points of intersection, as
observed below.
...
...
...
... 1
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...............2
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... .............................
............
..........
............
.
...
.
.
.
.
.
.
.
........
....
...
.
.
.
.
.......
.
.
...
....
......
......
....
.....
.
.
.
.
.....
.....
...
.
.
......
...
..
.
.
.
.
...
..........
.
.
.
.
..
.
.
....................
.
.
.
.
.
.
.
.
.
.
.
.
.
.................................................................
1
...
.
...
1 .....
4
................................................................
.
.
..........................
...
.
4 ....
............
...
.........
.
....
......
...
....
.....
....
.....
.
.
.
.
.
.
.
......
....
..
.
.
.
.......
.
.
..
.....
........
...
........
.........
.
............
..........
................... ....
............ 2
2 65
..............................................................
..
4
1
..
...
..
2
..
.

,2
8.8 Example.
Four points of intersection.
Compare: deg(f) deg(g) = 22 = 4

4,
, 2
4,
xy=1
x +4y =
Why four points? Simply multiply the degrees of the two curves: 2 2 = 4. Again we must
count points with appropriate multiplicity, and include all points with complex coordinates
which simultaneously satisfy the two dening polynomial equations. Here for example we
49
intersect two parabolas and once again nd 4 points of intersection:

...
..
...
..
...
...
...
... ...........
...
.
.
. ....
...
...
... .......
...
...........
...
.........
.
.
.
.
...
....
....
....
.....
.......
......
..................................
...
...
.....
..
...
...
...
...
....
....
.....
2
......
......
.......
.......
.......
....
2y=x2 +1
8.11 Example.
(1, 1)
Four points of intersection:

(1, 1) (double point),
(1+2i, 12i), (12i, 1+2i).
2x=y + 1
In the following example we intersect a conic with a cubic to obtain 2 3 = 6 points

of intersection:
(1, 3)
8.9 Example.
Six points of intersection:
(1,
(1, 3)
(double
points),
3),
( 3, 3), ( 3, 3).
y=2x3 5x
....
...
............................
...
........
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.....
...
.....
...
...... ...
.... ...
...... .....
.......
...
... ...
.
.
... ...
.
.
.
.
...
......
.. ..
...
....
... ..
...
....
... .....
...
.
.. ..
.
.
...
.
.. ...
.
.
.... ....
...
.
.
.
.
.
...
. ..
.. .
...
... ..
... ....
... ...
...
.
.... .....
.
...
.. ..
.. ...
...
... ..
... ..
...
.. ...
... ....
...
.... ....
.
... ...
...
. .
.. ..
...
... ...
.....
..
... ..
....
...
.... .....
.....
.
...
. .
..
...
.. ...
... ....
...
... ....
... ....
... ..........
.... ........
.... .......
.
......
.....
........
...
.................................
...
.....
..

( 3, 3)
3x2 +y 2 =12
( 3, 3)
(1, 3)
Similarly, intersecting a line with a quartic (a curve of degree 4) we expect 1 4 = 4

points of intersection. The following example realizes this total with one single point and
one triple point:
8.10 Example.
Four points of intersection:
(0, 0) (triple point), (1, 0).
.
...
3......
y=x +x ....
...
...
..
..
..
..
.
.
...
.
...
...
...
...
...
...
.
.
...
..
...
..
...
...
...
....
.
.
.
.
.
...
.
.
........
...........................................................................................................................................................................................................................
....
...........
.
.
.....
.
.
.
.
.
.
..........................................
..
(1, 0)
(0, 0)
y=0
We have included this example so the student will see that what we mean (and have not
dened) by intersection multiplicity, generalizes the usual notion of multiplicity of zeroes
of a polynomial in one indeterminate: in this case f(x) = x4 + x3 = x3 (x + 1) has four
zeroes counting multiplicity, with a triple zero at 0.
Now consider the following example, which at rst seems to be a counterexample to
the principle observed up to this point. We intersect a hyperbola with a line, and we
expect to nd 2 1 = 2 points of intersection. We see only one point of intersection (with
50
real or complex coordinates); where is the missing point? It is out there, at innity!
...
....
....
....
.
.
.
.....
.....
...
.....
.....
....
.....
.....
.....
.....
.
.....
.
.
.
.....
......
.
.
.
.
.
.
.
.
......
.....
....
......
.....
.....
......
......
...... .......
......
.
.
.
.
.
.
.
.
.
.
.
.
3 5 ..............
....
.
.......
2
4 4 ........... .................................... 2
.
.
.
...
....
.
.
.
....
....
....
....
.
.
.
.
....
....
....
....
.
.
.
...
8.12 Example.
Just one ane point

of intersection: 34 , 54 .
y x =1
y=x+2
............................
.........
........
.......
......
......
......
.
.
.
.
.
......
...
.
.
.
.
......
..
.
.
.
.
......
..
.
.
.
......
.
..
.
.
.....
.
.
...
.....
.
.
.
.
.....
..
.
.
.
.
.....
.....
The two curves have homogeneous equations Y 2 X 2 = Z 2 and Y = X + 2Z. Now the
two projective points of intersection are the points
(3, 5, 4) =
34 , 54 , 1
and (1, 1, 0).
The second point lies on the line , which is dened by Z = 0.

All these examples illustrate Bezouts Theorem, which says that given two plane
curves of degree m and n respectively, assuming the two curves do not coincide or share any
components, then the number of points of intersection equals mn. This elegant statement
is valid if we abide by the following three principles:
Points whose coordinates lie in the algebraic closure of the original eld of denition,
should be included.
Points of intersection should be counted with appropriate multiplicity.
The curves should be considered as curves in the projective plane, not just in the ane
plane.
Exercises 8.
1. Consider the two Euclidean plane curves y = x2 and y = x3 . Find all points of intersection in the
complex projective plane. What are the corresponding intersection multiplicities?
9. Closed Congurations
We pause to consider a class of point-line incidence structures which slightly generalizes
the notion of a projective plane, by allowing degenerate cases which may fail the axiom
(P3). Thus a closed conguration is an incidence system of points and lines such that
(P1) For any two distinct points, there is exactly one line through both.
9. Closed Congurations
51
(P2) Any two distinct lines meet in exactly one point.

A closed conguration satisfying also (P3) (i.e. containing at least four points of which
no three are collinear) are of course projective planes; those which fail (P3) are said to
be degenerate. Note that the point-line dual of a closed conguration, is again a closed
conguration.
Closed congurations arise naturally as substructures of projective planes which are
closed under the meet and join operations, as any such substructure will inherit
the properties (P1) and (P2) from the plane in which it is embedded. To appreciate the
motivation for studying substructures which are closed under the operations of meet and
join, the student is encouraged to recall that in the study of groups, one encounters very
early the notion of a subgroup: a subset closed under the group operation, also satisfying
the usual axioms of group theory. Same for subgraphs of graphs, subelds of elds, etc. In
the case of projective planes, such closed substructures include not only subplanes (which
are projective planes in their own right) but also degenerate closed congurations.
It is not hard to see that the following lists all possible closed congurations.
9.1 Proposition. A point-line incidence structure is a closed conguration i it is
one of the following:
(i)
(ii)
(iii)
(iv)
(, );
a single point;
a single line;
a line and a point P , together with m 0 lines through P other than
, and n 0 points on other than P ;
(v) a line and a point P
/ , together with m 0 lines through P and their m
points of intersection with ; or
(vi) a projective plane.
Examples typical of cases (iv) and (v) are shown:

..
..
..
..
.........
......
..
...
.........
.....
..
...
.........
.....
.
.
.
.
.
.
.. ....
.
.
.
.
.
...
... ... ..........
.........
... ... ....... .................
.. ... ...... ............
......................
...........................................................................................................................................
9.1a Figure:
Case (iv)
P
or
P
...
............
........ .. ..........
9.1b Figure:
.......... ... ..................
.
.
.
.
.
.... .. .. ...... ...........
.... ......
..... ..
..
.......
...... ....
....
..
.
.
.
.
.
.......
Case (v)
....
..
..
...
....
....
.....
...
.
.
.
.
.
.
.
.
.
.................................................
.............
.............................
............................
............................
Two examples of closed congurations, which are the smallest cases of (iv) and (v) respectively, are the ag (an incident point-line pair) and antiag (a non-incident point-line
pair), as shown:
....
....
.......
a ag
an antiag ...........................
.......
.......
.
.
.
.
.
.
........
........
.......
.......
........
........
.......
.......
.
.
.
.
.
.
.
.
.
.
.
.
..
..
..
.......
..
.......
A typical projective plane contains many more degenerate closed congurations than
subplanes. In nite classical planes P2 (F ), every subeld K F gives rise to a subplane
52
isomorphic to P2 (K), given by those points and lines coordinatized by the subeld. In the
classical case such subplanes, and their images under automorphisms of P2 (F ), account for
all subplanes. Thus for example in a nite classical projective plane P2 (Fq ), a subplane of
order q0 exists i q = q0r for some r 1. For example we present an incidence matrix for
the projective plane P2 (F4 ) of order 4, having 21 points and 21 lines. We have partitioned
the incidence matrix so that three of its subplanes of order 2 are evident:
1101000
0110100
0011010
0001101
1000110
0100011
1010001
0000100
0000010
0000001
1000000
0100000
0010000
0001000
0010000
0001000
0000100
0000010
0000001
1000000
0100000
0100000
0010000
0001000
0000100
0000010
0000001
1000000
1101000
0110100
0011010
0001101
1000110
0100011
1010001
0000100
0000010
0000001
1000000
0100000
0010000
0001000
0001000
0000100
0000010
0000001
1000000
0100000
0010000
0100000
0010000
0001000
0000100
0000010
0000001
1000000
1101000
0110100
0011010
0001101
1000110
0100011
1010001
No such Lagrange-like condition holds for orders of subplanes in general; for example the
Hughes plane of order 9 has a subplane of order 2. However the following bound holds.
9.2 Theorem. Let (P, L) be a plane of order n with a proper subplane (P0 , L0 ) of
order n0 . Then n20 n, and equality holds i every line L meets some point of
P0 , i every point P P lies on some line of L0 .
Proof. Every L0 has (n+1) (n0 +1) = nn0 points of P........P0 . Since no two lines of
L0 share any point of P ........ P0 , the total number of points covered by lines of L0 is
(n20 + n0 + 1) + (n20 + n0 + 1)(n n0 ) n2 + n + 1 ,

points of P
points of P0
points of P ........ P0
covered by lines of L0
which simplies to
(n n0 )(n n20 ) 0.
53
10. Collineations and Correlations
Since (P0 , L0 ) is a proper subplane we have n0 < n and so n20 n. If equality holds here
then equality must hold throughout, which means that every point of P lies on some line
of L0 . A dual argument (interchanging points and lines) shows that this condition is also
equivalent to every line of L meeting some point of P0 .
A proper subplane (P0 , L0 ) of (P, L) is a Baer subplane if every line of the plane
meets some point of the subplane (P0 , L0 ). The conclusion of Theorem 9.2 shows that in a
plane of nite order n, a Baer subplane is simply a subplane of order n ; and this cannot
exist unless n is a perfect square. By the remarks above, every nite classical projective
plane of square order has Baer subplanes.
Exercises 9.
1. Let (P, L) be a projective plane of order n having a proper subplane (P0 , L0 ) of order n0 , so that
n20 n by Theorem 9.2. If n20 < n, show that n20 + n0 n. Say as much as you can about the case
of equality: n20 + n0 = n (giving examples if possible).
Hint. Suppose n20 < n, so there exists a line L not containing any point of P0 . Consider
intersections of with lines of L0 .
2. We have shown that the points and lines of P2 (F4 ) can be partitioned into three subplanes of order 2.
Can they also be partitioned into
(a) 21 ags?
(b) 21 antiags?
(c) 7 triangles (subplanes of order 1)?
Do similar partitions exist if P2 (F4 ) is replaced by a more general projective plane? Find such generalizations if possible.

Automorphisms of a projective plane (P, L) are often called collineations. (One reason
for introducing the new term collineation is to distinguish this type of automorphism
from a correlation, which we dene later in this Section.) Thus a collineation of (P, L)
may be viewed as a permutation of P L mapping points to points, and lines to lines,
such that P i P . The (full) automorphism group Aut(P, L) is the (full)
collineation group, and subgroups of Aut(P, L) are collineation groups. For example
consider the projective plane of order two, shown in Figure 10.1; here we have labelled the
points P = {1, 2, . . . , 7} and the lines L = {a, b, . . . , g}.
3
10.1 Figure
An Incidence Matrix A
of the Plane P2 (F2 )
.
.......
... .. ...
... ..... .....
.
.
.
.
... .... .....
...
...
....
...
...
.
.
.
..
.
.
... ................................... .....
.
.
.........
.........
.
.
.
.
........
.......
....
......
.....
...
..
.....
.
.
.
.
............
.
..
.. ..........
.
.
.
.
.
.
.
...
.......
......
...
.
.
.
.
.
.
.
.
.
.
....... . .......
.....
.
.................
.
......
... .....
... ....
................
.
.
.
.
.
.
.
. ...
... .. .........
.
.
... .....
.
.
.
.
.
.
...
.......
.
.
..
.
...
.
.
.
.
.
.
.
.
.
.
.
....... ..
... ........
.
...
........
....
...
.
....
..
... ............ ......
... ............ .....
.
.
.
.
.
.
.....
....... ...
.
. ........
.
.
.
.
.
.
.
.
....... ..
.
.......
.
.
.
..........
.
.
.
.
.
.. .
.
.
.
.
.
.
.
........................................................................................................................................................
7 d
6
5 e
f c
b
4
2
a
1
1
0
A=
1
0
0
0
b
0
1
1
0
1
0
0
c
0
0
1
1
0
1
0
d
0
0
0
1
1
0
1
e
1
0
0
0
1
1
0
f
0
1
0
0
0
1
1
g
1
0
1
0
0
1
1
2
3
4
5
6
7
54
The full collineation group of this plane has order 168, and is generated by the two
collineations (1234567)(abcdefg) and (14)(67)(cg)(de).
Assume that the plane (P, L) has order n. Enumerate the points and lines as P =
{P1 , P2 , . . . , PN }, L = {1 , 2 , . . . , N } where N = n2 + n + 1. Every collineation

acAut(P, L) gives rise to a pair of permutation matrices M(),
representing the
M ()

tion of on points and lines, respectively; thus M() = mij () 1i,jN and M () =

mij () 1i,jN where

mij () =
if Pi = Pj ,
0,
mij ()
if Pi = Pj ;
1,

=
0, if i = j ,
1, if i = j .

Let A = aij be the incidence matrix of A, so that

aij =
1,
0,
if Pi j ;
if Pi
/ j .
Given two collineations , Aut(P, L), the (i, j)-entry of M()M( ) is

mik ()mkj ( ) =
1,
0,
if there exists k such that Pi Pk Pj ;

otherwise
1, if Pi Pj ;
0, otherwise
= mij ( )
which shows that M()M( ) = M( ), i.e. M : Aut(P, L) GLN (Q) is a homomorphism. Note also that
1
1
M()T = M() = M( )
since M() is a permutation matrix, hence orthogonal. Similar properties hold for M :
Aut(P, L) GLN (Q) in place of M. Now if Aut(P, L) then the (i, j)-entry of M()A
is

mik ()akj = 1, if there exists k such that Pi Pk j ;
0, otherwise
k

1, if Pi j ;
=
0, otherwise
whereas the (i, j)-entry of AM () is

aik mkj () =

=
1,
0,
if there exists k such that Pi k j ;

otherwise
1,
0,
if Pi j ;
otherwise
55
and so M()A = AM (). Conversely if M1 , M2 are N N permutation matrices satisfying

1
M1 AM2 = A, then the corresponding pair of permutations of P and L is a collineation
of the plane (P, L). This proves
10.2 Theorem. Aut(P, L)
= {pairs (M1 , M2 ) of N N permutation matrices :
M1 A = AM2 }.
For example in the case of the projective plane of order two listed above, the generators
for the collineation group listed above are represented in matrix form as

1 0 0 0 1 0 1
0 1 0 0 0 0 0
0 1 0 0 0 0 0
1 0 0 0 1 0 1
0 0 1 0 0 0 01 1 0 0 0 1 0 1 1 0 0 0 1 00 0 1 0 0 0 0
0 0 0 1 0 0 00 1 1 0 0 0 1 0 1 1 0 0 0 10 0 0 1 0 0 0

0 0 0 0 1 0 0 1 0 1 1 0 0 0 = 1 0 1 1 0 0 0 0 0 0 0 1 0 0 ;
0 0 0 0 0 1 00 1 0 1 1 0 0 0 1 0 1 1 0 00 0 0 0 0 1 0

0 0 1 0 1 1 0
0 0 0 0 0 0 1
0 0 0 0 0 0 1
0 0 1 0 1 1 0
0 0 0 1 0 1 1
1 0 0 0 0 0 0
1 0 0 0 0 0 0
0 0 0 1 0 1 1

(1234567)
0
0
0
1
0
0
0

0
1
0
0
0
0
0
0
0
1
0
0
0
0
1
0
0
0
0
0
0

0
0
0
0
1
0
0
(14)(67)
0
0
0
0
0
0
1
0
1
01
0
0
01
0
0
1
0
0
0
0
1
1
0
1
0
0
0
0
1
1
0
1
0
0
0
0
1
1
0
1

A
1
0
0
0
1
1
0
0
1
0
0
0
1
1

1
1
0 1
1
0
0 = 1
0
0
0
0
1
0
0
1
1
0
1
0
0
0
0
1
1
0
1
0
0
0
0
1
1
0
1

A
(abcdef g)
1
0
0
0
1
1
0
0
1
0
0
0
1
1
1
1
00
1
0
00
0
0
0
0
0
1
0
1
0
0
0
0
0
0
0
0
0
0
0
1
0
0
0
0
1
0
0

0
0
0
1
0
0
0
(cg)(de)
0
0
0
0
0
1
0
0
0
1
0.
0
0
0
A consequence of Theorem 10.2 is

10.3 Corollary. Every collineation of a plane (P, L) has equally many xed points
and xed lines.
Proof.
Let Aut(P, L). Note that the trace of M1 () is

mii () = {i {1, 2, . . . , N} : Pi = Pi }
i
is simply the number of points P P xed by . Similarly the trace of M2 () is the

number of xed lines. However the incidence matrix A is invertible:

1
1
J = n1 (nI + J ) n(n+1)
(n+1)J = I.
A n1 AT n(n+1)
1
Thus the matrices M1 () and M2 () = A M1 ()A are similar. Since similar matrices
have the same trace, the result follows.
56
The argument used in our proof of Corollary 10.3 is based on concepts from representation theory; see Appendix A3. Here we have a pair of homomorphisms i : Aut(P, L)
GLN (Q), Mi () for i = 1, 2. Since the matrix A is invertible, the representations 1
and 2 are equivalent . The same technique applies to any invertible matrix A, where we
consider an automorphism of A as a pair of permutations of the rows and the columns;
the argument shows that any such automorphism xes equally many rows and columns.
A more general formulation of this fact is given by
10.4 Theorem. Every group G of automorphisms of a nite projective plane (P, L)
has equally many orbits on points and on lines.
Before proving Theorem 10.4, consider the following example: Let (P, L) be the projective
plane of order two shown in Figure 10.1, and let G = g, h be the group of order four
generated by g = (12)(57)(bg)(ef) and h = (15)(27)(ad)(bg). Then G has four point
orbits {1, 2, 5, 7}, {3}, {4}, {6} and four line orbits {a, d}, {b, g}, {c}, {e, f} in accordance
with Theorem 10.4. Note that the point orbits and line orbits do not have the same size!
Moreover there is no canonical bijection between point and line orbits. Nevertheless every
nonidentity element of G has the same number of xed points and xed lines, in this case
three.
For every point P P, denote the corresponding stabilizer
g
GP = {g G : P = P } and orbit P G = {P g : g G}. Recall that |P G | = [G : GP ];
see Theorem A2.2. Let P1 , . . . , Pr P be representatives of the distinct point orbits, and
1 , . . . , s L representatives of the distinct line orbits. We count

{P : P g = P }
trM1 (g) =
g
gG

= {(P, g) P G : P g = P }

|PiG ||GPi |
=
1ir
|G|
1ir
= r|G|.
Since trM1 (g) = trM2 (g) we obtain

r|G| =
trM1 (g) =
trM2 (g) = s|G|.
gG
gG
Let be a collineation of a nite projective plane (P, L), and let (P , L ) be the
substructure formed by the xed points and lines; thus
P = {P P : P = P } and L = { L : = }.
57
It is easy to see that this is a closed substructure of (P, L): if P, Q P then

(P Q) = P Q = P Q
and if , m L then
( m) = m = m.
The possibilities for a closed conguration were listed in Proposition 9.1; of these, Corollary 10.3 requires that we select only those with equally many points and lines as possible
xed substructures of collineations. This yields
10.5 Theorem. Let be a collineation of a nite projective plane (P, L). Then
the xed substructure of is a closed conguration of one of the types listed in the
following table:
case
description of
(P , L )
(i)
xed-point free
(, )
(ii)
generalized
(P, )-elation
a line and a point P , together with ...... ...........................................................

.. .................
. . ..
m 0 lines through P other than ,
.......
..........................
.................
................
................
........

and m 0 points on other than P
P
..
generalized
(iii) (P, )-homology
(iv)
planar
..
...
.....
. ..
a line and a point P
/ , together with
............P
....
.......
..... .. . ..........
..... ..... .... ...................
.
.
m 0 lines through P and their
.
.
.
... ... ... ..... .........
.
.
.
.
.
.
m points of intersection with
.......................................................................................
subplane
Technically, the identity is planar, since its xed substructure is the entire plane. In cases
(ii) and (iii), is called a generalized (P, )-perspectivity. As special cases, we call
a (P, )-perspectivity if m = n, the order of the plane. This means that xes
every point of and every line through P ; then we call a (P, )-elation if P , or a
(P, )-homology if P
/ . Also in such cases we call P the centre and the axis of the
perspectivity . For a xed group of automorphisms G Aut(P, L) and point-line pair
(P, ), the set of all g G xing every point of and every line through P forms a subgroup
denoted G(P, ) G; this group consists of the identity and all (P, )-perspectivities in G
(if any).
Another special case of (ii) and (iii) is where m = 0: in this case (P , L ) = ({P }, {}).
Here we say is a ag collineation or an antiag collineation, according as P or
P
/ . Also case (iii) includes the case m =2 where is triangular: its xed substructure
is a triangle. In this case any of the three vertices of the triangle can be considered the
centre of . Perhaps this should be considered a special case of (iv) really (a subplane of
order 1, i.e. a thin subplane; see Exercise #6.3).
More generally, every collineation group G has a xed substructure (PG , LG ) which is a
closed conguration in the plane (P, L). However, such closed congurations need not have
58
equally many points and lines (see the example given after the statement of Theorem 10.4
above, with three xed points and only one xed line). More general collineation groups
can have xed substructures of any of the types listed in Proposition 9.1.
As rst examples illustrating Theorem 10.5, we classify some collineations of the
projective plane of order 2 illustrated in Figure 10.1:
(37)(56)(bf)(cd) is an (1, a)-elation;
(24)(3576)(bdfc)(eg) is a ag collineation with centre 1 and axis a;
(235)(476)(age)(cdf) is an antiag collineation with centre 1 and axis b;
(1234567)(abcdefg) is a xed-point-free collineation.
This plane has no nontrivial homologies or planar or triangular collineations.
Consider now the classical plane P2 (F ) over a eld F . Here the full collineation group
is G = P L3 (F ), the group generated by all semilinear transformations F 3 F 3 , i.e.
transformations of the form v v A where A GL3 (F ) and Aut F ; here we view
F 3 as consisting of row vectors. (See Appendix A2.) In the group P L3 (F ), all scalar
transformations v v where 0 = F x every point and every line of P2 (F ) and
so are the identity collineation. The elation group G((1, 0, 0), (0, 1, 0)T ) consists of the
transformations
1 c 0
0 1 0 , i.e. (x, y, z) (x, y+cx, z).
0 0 1
The homology group G((1, 0, 0), (0, 1, 0)T ) consists of the transformations
c 0 0
0 1 0 , i.e. (x, y, z) (cx, y, z)
0 0 1
where c = 0. Note that the elation and homology groups with xed axis and centre are
abelian, in fact isomorphic to the additive group of F and the multiplicative group F
respectively. Nonclassical planes can however have nonabelian groups of perspectivities.
Generalizing the previous two examples, it is the case that invertible matrices of the
form I + B where B has rank one, represent perspectivities. The centre and axis are the
row and column spaces of B, respectively. (Reverse these, if the underlying vector space
F 3 is considered as column vectors.) Every nontrivial perspectivity of P2 (F ) has this form.
Diagonalizable matrices with three distinct eigenvalues, represent triangular collineations (and conversely). For example the diagonal matrix
a 0 0
0 b 0
0 0 c
has xed substructure consisting of the triangle with vertices
(1, 0, 0),
(0, 1, 0),
(0, 0, 1).
59
More generally a diagonalizable matrix with three distinct eigenvalues, has three 1-dimensional eigenspaces, and these are the vertices of its xed triangle.
A matrix of the form
1 a b
A = 0 1 c
0 0 1
with ab = 0 has a unique eigenspace (0, 0, 1) which is the centre of the corresponding
collineation; its axis is the line x = 0, i.e. (1, 0, 0)T so A represents a ag collineation.
A matrix of the form
1 0 0
A = 0 a b ,
0 c d

assuming the 22 matrix ac db has an irreducible characteristic polynomial, has (1, 0, 0)
as its unique eigenspace. In this case A represents an antiag collineation with centre
(1, 0, 0) and axis (1, 0, 0)T .
A eld automorphism Aut F acts on P2 (F ) as (x, y, z) (x , y , z ). The
corresponding collineation is planar with xed substructure P2 (K) where K = {x F :
x = x} is the xed subeld of .
10.6 Proposition. Let (P, L) be a projective plane of nite order n, and consider
any group of automorphisms G Aut(P, L). Then |G(P, )| divides n or n1 according as P or P
/ .
Proof. Let m be a line through P other than . Let S be the set of points of m other
than P , and not on . Then G(P, ) permutes S. If G(P, ) xes a point of S, as well
as xing every point of m and every line through P , then by Theorem 10.5 we must have
= 1. Thus every orbit of G(P, ) on S has size |G(P, )| by Theorem A2.2. Since |S| = n
or n1 according as P or P
/ , the result follows.
From the explicit description of the groups G(P, ) above in the classical case, we see
that the upper bound of Proposition 10.6 is attained in this case. Thus in the classical
plane P2 (F ) with full collineation group G = P L3 (F ):
If P and m is any line through P other than , then the group G(P, ) of
(P, )-elations permutes the points of m ........ {P } regularly. The group G(P, ) is
isomorphic to the additive group of F . In particular if F = Fq then G(P, ) is
elementary abelian of order q.
If P
/ and m is any line through P other than , then the group G(P, ) of
(P, )-homologies permutes the points of m ........ {P, m} regularly. The group
60
G(P, ) is isomorphic to the multiplicative group F . In particular if F = Fq

then G(P, ) is cyclic of order q 1.
An element Aut(P, L) of order two is called an involutory collineation, or
simply an involution.
10.7 Theorem. Let be an involutory collineation of a projective plane (P, L).
Then is one of the following:
(a) an elation;
(b) a homology; or
(c) Baer planar.
If the plane (P, L) is of nite order n then cases (a), (b), (c) require that n is even,
odd, or a perfect square, respectively.
Proof. It is not hard to see that every point P lies on some xed line. If L is not xed
by , then ( ) = = . In particular every non-xed line passes through
a xed point. From the list of possible xed substructures (i)(v) above, we see that the
xed points consist either of all points of some line, or a subplane (which by Theorem 9.2
is necessarily Baer). Thus every involution is either an elation, a homology or Baer planar.
If is a perspectivity then G(P, ) contains an element of order 2 and so the parity of n
follows from Proposition 10.6.
A correlation of a projective plane (P, L) is a permutation of P L which maps
points to lines, and lines to points (thus interchanging the sets P and L) while preserving
(or, more correctly, reversing) incidence. So a correlation has the property that P
i P . Such a correlation can be seen as an isomorphism between the original plane
(P, L) and the dual plane (L, P), so only a self-dual plane can have any correlations at all.
For example we illustrate here a correlation (AaDdEbCg)(BcGe)(F f) of the projective
plane of order two:
E
..
...
.........
.........
... .. ....
... .. ...
... .... ....
... .... .....
.
.
.
.
..
(AaDdEbCg)
.
. . .
... ... ...
... .. ...
(BcGe)(F f )
... ............... ...
... ........ ...
................... ... .....................
................ .... ..................
.
.
.
.
.....
....
....
....
...
....
....
....
...
....
..
..
...........
..........
...... ...
.........
...... ............ .... ............ ........
...... ............ .... ............ .......
.
.
.
.
...........
..........
... ...
. ..
... ...
... ....
. ..
... ...
............
..............
... ...
... ...
...... .. .......
... .. ........ .... ............ ..... .....
... .................. .... ................. .....
.......
.
... ...............
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
. .... ..
.. ..... ...
....
... ........ ........
....
... ......... .......
.... ..................
.... .................
....
....
.
.
...........
...........
.. .
......
.
.....
....................................................................................................................
..................................................................................................................
f G g e
F c C G
F E
g f d
A C
b
D
a
c
d
A
D
B a
a b c d e f g
.
.
.
.
..
..
..
10.8 Figure .................................. .................................. .................................. ................................................................................................................................................................
.... ... ........... .... ........... .... ................. ... ................................ ............ ...
....
....... ...
....... ......... ...... .......................... ......... . ...
.. ....
..
Incidence
.. ........
.. ............. .. ........
.. ..
.. ......
.. ..
.
... ... .... ........... ......................................................... .......................... .... .... ....
.
.
.
.
... .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
... ..
..........
......... ......................... ........ .......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... . ...... ..................... ... ........... ... .....
. .
. .
Graph of
....
.. ....... ........... ............ .. ........... .. ... ..... .... ............ .... ....
..
........................... .................. .......................... .......... .... ... .......... .... ... .......... ..... ...
.
2
.
.
.
.
.
........
........
........
. ............ ... ..... ... ..... ........
P (F2 )
............ ............ ............ .......... .......... .......... .........
A B C D E F G
We may view both collineations and correlations as automorphisms of the incidence

graph of the projective plane (P, L); this is the bipartite graph = (P, L) having
vertex set P L, and with an edge from vertex P P to vertex L i the point P lies
on the line . In fact the automorphisms of the graph are of two types:
61
(i) Those Aut preserving both P and L. These correspond to collineations of

the plane (P, L); and
(ii) Those Aut interchanging P and L, if they exist. These correspond to
correlations of the plane (P, L).
So either the plane (P, L) is self-dual, in which case Aut(P, L) (the collineation group) is a
subgroup of index 2 in Aut (identied with the group of all collineations and correlations
of (P, L)), or the plane (P, L) is not self-dual, in which case Aut may be identied with
the collineation group Aut(P, L).
Figure 10.8 shows the projective plane of order two, along with its incidence graph,
which evidently has automorphism group of order 2 168 = 336.
Note that every correlation of (P, L) must have even order. A polarity of (P, L)
is a correlation of order 2. Given a polarity and an enumeration of points as P =
{P1 , P2 , . . . , PN }, it is natural to list lines in the corresponding order L = {P1 , P2 , . . . ,
PN }; in this case the incidence matrix is symmetric since Pi Pj i Pj Pi . (Here
we have used the fact that 2 = 1 in order to simplify (Pj ) = Pj .) Conversely, if a
projective plane has a symmetric incidence matrix for some enumeration of its points Pi
and lines j , then the projective plane has a polarity Pi i .
For example the projective plane of order two illustrated in Figure 10.8 has a polarity
(Aa)(Be)(Cc)(Dg)(Ef )(F b)(Gd), giving rise to the following symmetric incidence matrix.
10.9 Figure
The Polarity
(Aa)(Be)(Cc)(Dg)(Ef )(F b)(Gd)
A
B
C
D
E
F
G
a
1
1
0
0
0
e
1
0
0
0
1
1
0
c
0
0
1
1
0
1
0
g
1
0
1
0
0
0
1
f
0
1
0
0
0
1
1
b
0
1
1
0
1
0
0
0
0
0
0
1
g
f
a
e
c
b
......................... .............................. ......................... .................................. ............................. .............................
.......
........ ..
............... ..
..........
............. ... ... .......... ...
.. ... ...... ...
............ ....
.
.
.
.
.
.
.
.
.............. ....... ........
.
.
... ... ........
.
.
.
.
.
.
.
.
...................... ........ ......... ..................................... .... ...............
....
.. ... ...........
..... ..... .. .....
..
.. ... ... ......
..... ..... .. ... .... ..
...... ....................................................................... ......................... .................... ......
...
... ... ...
.......... ... ............ ............. ...........
............ ........
....
..
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.. .......
.
.
.
.
.
.
. .. . . . . .. . .
. ..
...
... ... ... .......... .................................................. .............................. ............... .................... .......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.. ... ... .......
.
.
............... .. ..... ..... ...
................... ......... .......
.
.
.
.
.
.
.
.
.
.
.
.
.
.. ... .........
.
.
.
..... .... ....
..... ..... ................. ..
............. ......... .........
.
.
.
.
.
.
.
.
... ... ...... ...
.
.
.
.
.
.. . .
..
.......... ...
..
.. .
.......... ..
..... ....... ... ...................... ................................. ................... ....................................
..........
... ........
..... ......
........ .....
.........
....
...........
...
...
....
..
...
........
..
Let be a polarity of a projective plane (P, L). A point P is absolute if P P ;

otherwise P is nonabsolute. Similarly a line is absolute if ; otherwise is
nonabsolute. The absolute points and lines are identied by 1s on the diagonal of the
symmetric incidence matrix corresponding to . For the example illustrated in Figure 10.9,
the absolute points are A, C, G and the absolute lines are a, c, d. Since the accompanying
incidence graph lists points and lines in consistent order (corresponding via ), the graph is
symmetric about a horizontal axis of symmetry and the vertical edges identify the absolute
points and lines. Note that the absolute points are collinear, and the absolute lines are
concurrent. In general we have
10.10 Theorem. Suppose is a polarity of a projective plane (P, L) of order n,
where n is not a square. Then has n + 1 absolute points. These points form a line
if n is even, or an oval if n is odd.
62
For a proof, see [31, Thm 12.7]. For a nite classical plane P2 (Fq ), one possible polarity is
given by the transpose map which interchanges row and column vectors. Clearly, a point
(x, y, z) is absolute i
0 = (x, y, z)(x, y, z)T = x2 + y 2 + z 2 .
Such points form a nondegenerate conic if q is odd, or a line x + y + z = 0 if q is even. For
classical planes of non-square order, every polarity has this form (called an orthogonal
polarity) for some choice of coordinates. But suppose q = q02 , so that the quadratic
extension Fq Fq0 has an automorphism x x = xq0 of order two. In this case the
conjugate-transpose map v v T interchanges row and column vectors of length 3 and
gives a polarity of P2 (Fq ), called a unitary polarity. The absolute points of this polarity
are those satisfying the condition
0 = (x, y, z)(x, y, z)T = xq0 +1 + y q0 +1 + z q0 +1 .
One checks in this case that there are q03 +1 absolute points (and lines), and q02 (q02 q0 +1)
nonabsolute points (and lines). Moreover, the absolute points and nonabsolute lines form
a 2-(q03 +1, q0 +1, 1) design embedded in (P, L), called a Hermitian unital of order q0 .
We will encounter this example again in Example 18.7. Every polarity of P2 (Fq ) is either
an orthogonal polarity (for arbitrary q) or a unitary polarity (for q a square only). Such
polarities are as described above, after a change of coordinates of necessary.
Exercises 10.
1. For each of the planes indicated, nd the number of collineations of each of the types listed in Theorem 10.5.
(a) The projective plane of order 2.
(b) The projective plane of order 3.
2. Let 1 and 2 be involutory collineations of a nite projective plane. Can 1 2 be xed-point-free?
Justify your answer.
3. Let be an automorphism of a projective plane (P, L). Show that xes all lines through some point
P , i xes all points on some line . Thus has a centre i has an axis.
Note: In the nite case this follows from our classication of xed substructures of collineations, which
relies on the fact that has equally many xed points and xed lines (Corollary 10.3). However one
veries the required statement directly without any niteness assumption. Suppose xes every point
of . We may assume that every xed point of lies on (why?). Let R be a point not on , and let
P = (RR ) . Then R lies on the xed line RP ; and so every point lies on at least one xed line.
Let S be a point not on RP and consider how a xed line through S may intersect RP .
4. How many polarities does the projective plane of order two have? Explain.
5. The Hermitian unital of order 2 consists of all absolute points and nonabsolute lines with respect to
a unitary polarity of P2 (F4 ) as described above. List all points and lines of this unital, and draw it
(showing all incidences). Where have you seen a design like this before in this course? and should you
expect a similar connection between unitals and other designs we have studied, for larger values of q?
Explain.
63
11. Classical Theorems

We present here several results that may be called classical , not only because of when these
results were rst proved; but because they describe properties of the classical projective
planes P2 (F ) not shared by more general projective planes.
The group L3 (F ) of all semilinear transformations F 3 F 3 acts on the projective
plane P2 (F ). (See Appendix A2 for the notation and results of group theory used here.)
This is because every semilinear transformation maps subspaces of F 3 to subspaces of
the same dimension, preserving the incidence relation of inclusion. The normal subgroup
Z L3 (F ) consisting of all scalar transformations F 3 F 3 , v v for 0 = F , acts
trivially on P2 (F ) (i.e. it xes every point and line) and so the quotient group
P L3 (F ) = L3 (F )/Z
acts on P2 (F ). The subgroup P GL3 (F ) consists of linear transformations F 3 F 3 , but
where two linear transformations are identied if one is a scalar multiple of the other (in
which case their action on P2 (F ) is the same).
An ordered quadrangle is an ordered 4-tuple of points (A, B, C, D), no three of
which are collinear.
11.1 Fundamental Theorem of Projective Plane Geometry. The full collineation group of P2 (F ) is P L3 (F ). Its normal subgroup P GL3 (F ) is regular (i.e.
sharply transitive) on ordered quadrangles.
Proof. We prove only the second statement. Recall that the points and lines of P2 (F )
are the subspaces of F 3 of dimension 1 and 2 respectively. Let (A1 , A2 , A3 , A4 ) be an
ordered quadrangle, so that Ai = vi for i = 1, 2, 3, 4; and no three of v1 , v2 , v3 , v4 V
are linearly dependent. There exists a unique linear transformation T : F 3 F 3 such
that T (vi ) = ei for i = 1, 2, 3 where we denote the standard basis of F 3 by
e1 = (1, 0, 0);
e2 = (0, 1, 0);
e3 = (0, 0, 1).
Since T (v4 ) is not a linear combination of any two of T (vi ) = ei (i = 1, 2, 3), we must have
T (v4 ) = (a, b, c) where abc = 0. Dene an invertible linear transformation D : F 3 F 3
1
1
1
by (x, y, z) (a x, b y, c z). Then the invertible linear transformation S T : F 3 F 3
maps
1
A1 (a , 0, 0) = (1, 0, 0);
1
A2 (0, b , 0) = (0, 1, 0);

1
A3 (0, 0, c ) = (0, 0, 1);

A4 (1, 1, 1).
64
Thus P GL3 (F ) is transitive on ordered quadrangles. Now suppose a linear transformation

R : F 3 F 3 xes the ordered quadrangle (e1 , e2 , e3 , (1, 1, 1)). Since R xes ei
for i = 1, 2, 3, the matrix of R is diagonal:
a 0 0
R = 0 b 0
0 0 c
for some nonzero a, b, c F . Since R also xes (1, 1, 1), we have
(1, 1, 1) = R(1, 1, 1) = (a, b, c)
so that a = b = c and R Z, which represents the identity in P GL3 (F ). Thus P GL3 (F )
permutes ordered quadrangles regularly (sharply transitively).
The Fundamental Theorem is useful in making simplifying assumptions in any computations involving coordinates, as in the following proof.
11.2 Pappus Theorem. In Figure 8.1 the points R0 , R1 , R2 are collinear.
Proof.
By the Fundamental Theorem, we may assume that

P0 =(1, 0, 0),
P1 =(0, 1, 0),
Q0 =(0, 0, 1),
Q1 =(1, 1, 1).
We immediately obtain
=(0, 0, 1)T ,
m=(1, 1, 0)T,
O=(1, 1, 0),
P2 =(1, , 0),
Q2 =(1, 1, )
for some , F ........ {0, 1}. Next we determine

01 =(0, 1, 1)T ,
20 =(, 0, 1)T ,
10 =(1, 0, 0)T ,

12 =(, 0, 1)T ,
02 =(0, , 1)T ,

21 =(, 1, 1)T
and consequently
R0 =(1, +(1), ),
R1 =(1, , ),
R2 =(0, 1, 1).
The fact that the latter three points are collinear follows from
1 +(1)
det 1
= 0.
0
1
1
65
The following result may be proved using coordinates in a spirit similar to the previous
proof. Instead our proof makes use of a plane collineation.
11.3 Desargues Theorem. In a classical plane P2 (F ) consider two triangles ABC
and A B C sharing no sides or vertices, and suppose that the lines AA , BB and
CC are concurrent at a point O. Let P, Q, R be the intersection points AB A B ,
AC A C , BC B C respectively, as in Figure 11.4. Then P, Q, R are collinear.
11.4 Figure
The Desargues Conguration
Proof. Let = P Q. By comments following Proposition 10.6, there exists an (O, )homology mapping A A . Since xes both P and Q, while also xing every line
through O, it maps
B = P A OB P A OB = B ;
C = QA OC QA OC = C ;
BC B C .
Since xes every point of , this means that BC = B C , i.e. the three lines BC,
B C , are concurrent; thus R = BC B C lies on as required.
For each of the three preceding theorems, a converse is available. These converses,
which are usually included in the statements of the theorems, state that the only planes
with the given properties are the classical planes. A converse of the Fundamental Theorem of Projective Plane Geometry, is the fact that every projective plane admitting an
automorphism group which is transitive on ordered quadrangles, is necessarily classical.
More is true: for example the Ostrom-Dembowski-Wagner Theorem states that a nite projective plane admitting an automorphism group which permutes the points 2-transitively,
is necessarily classical.
66
Every projective plane which satises the conclusion of Desargues Theorem, is isomorphic to P2 (K) for some skeweld K. (Recall that a skeweld is roughly a not-necessarilycommutative eld; see Appendix A3 for a more complete description.) Every projective
plane which satises the conclusion of Pappus Theorem, is isomorphic to P2 (F ) for some
eld F . Thus the condition of Pappus Theorem is stronger than that of Desargues Theorem: it implies commutativity of the underlying coordinate system. Planes of the form
P2 (F ) for some eld F are called Pappian, while planes of the form P2 (K) for some
skeweld K are called Desarguesian. Now by a classical plane we may intend either
a Pappian plane or a more general Desarguesian plane; but for nite planes, these two
concepts coincide since every nite skeweld is commutative (see Theorem A3.4). The alternative name Galois plane is also commonly applied to the nite classical planes P2 (Fq )
since they are coordinatized by the nite elds Fq , also known as the Galois elds (and
sometimes therefore denoted GF (q)).
For proofs of the converses of Desargues Theorem and Pappus Theorem, see e.g. [31].
These proofs rely on the coordinatization of a general projective plane, which is accomplished using a planar ternary ring (PTR). Such a PTR consists of a set R with a
ternary operation T : R R R R, rather than the two binary operations of addition and multiplication arising in traditional ring theory. In the classical case R is just a
eld, and T (x, m, b) = xm + b; but in the general case the ternary operation cannot be
represented as a composite of two binary operations in such a convenient manner. The
greater the degree of homogeneity, or symmetry, that is assumed for the projective plane,
the nicer will be the algebraic properties of its PTR. For example translation planes, while
not as symmetric as classical planes, are coordinatized by quasields. A quasield (dened
in Section 3) is nicer than a general PTR, although not as nice as a eld (at least it is
typically neither commutative nor associative, and fails one of the distributive laws).
Exercises 11.
1. The Desargues conguration is the partial linear space shown in Figure 11.4 (with 10 points and 10
lines, including the line ). Each point of the conguration lies on three lines of the conguration; and
dually, each line passes through three points.
(a) Show that the conguration is self-dual. (As a consequence, the dual of Desargues Theorem can
be phrased as a converse of the original theorem; we omit the details.)
(b) Identify the group of automorphisms of the conguration. (Partial credit will be given for providing
generators and the order of the group. A complete answer will recognize the group as a known
group.)
2. Prove that a classical projective plane P2 (F ) has a subplane of order 2 i the characteristic of F is 2.
Hint. Use the Fundamental Theorem to choose nice coordinates for 4 of the 7 points of the subplane,
without loss of generality.
3. Prove that the ane plane of order 3 embeds in P2 (Fq ), i q 2 mod 3.
Hint. Again use the Fundamental Theorem to choose nice coordinates for 4 of the 9 points of the
embedded ane plane, without loss of generality.
67
12. Conics and Ovals

A conic in a classical plane P2 (F ) is the set C of points (x, y, z) satisfying a nonzero
homogeneous quadratic polynomial condition of the form Q(x, y, z) = 0 where
Q(X, Y, Z) = aX 2 + bY 2 + cZ 2 + dXY + eXZ + fY Z.
Note that this condition is well-dened for points (x, y, z) since if 0 = Fq then
Q(x, y, z) = 2 Q(x, y, z). We say that C is nondegenerate if it does not contain an
entire line. We show that for F = Fq a nite eld, any two nondegenerate conics are
equivalent by an invertible linear change of coordinates. Indeed we show that C can be
transformed to the conic y 2 = xz using a linear change of coordinates.
12.1 Theorem. Let C be a nondegenerate conic in a nite classical plane P2 (Fq ).
Then C has q + 1 points, of which no three are collinear. Moreover C is equivalent, by
a linear change of coordinates, to the conic y 2 = xz.
Proof. By Theorem A1.7, C has at least one point P . By a linear change of coordinates,
we may assume that P = (1, 0, 0); so without loss of generality C has the equation
bY 2 + cZ 2 + dXY + eXZ + fY Z = 0.
A typical line through P has the form
, = (0, , )T = P, (0, , )
where (, ) = (0, 0).
Here we may take

(, ) {(0, 1)} {(1, ) : Fq }
in order to list the q+1 lines through P . We claim that
(12.2)
exactly one of these lines (the tangent line at P ) meets C at a unique point
P ; and the remaining q lines through P are secant lines, each meeting C
at just one point other than P .
From (12.2) it will follow that C has exactly q + 1 points; and that no line through P meets
C in more than two points. But since P was an arbitrarily chosen point of C, in fact no
three points of C are collinear. To prove (12.2), consider an arbitrary point of other
than P ; this has the form (t, , ) for some t Fq . The condition for this point to lie
on C is that
(12.3)
b 2 f + c 2 + (d e)t = 0.
68
If (, )
/ (d, e) then the coecient of t is nonzero so there is a unique solution for t. This
gives a unique point of C on each line through P other than the line d,e . If (, ) (d, e)
then the coecient of t in (12.3) is zero and the equation becomes be2 + fde + cd2 = 0. If
this equation holds then all points of the line d,e lie on C, contrary to the nondegeneracy
of C; therefore be2 + fde + cd2 = 0 and so C contains no points of d,e other than P itself.
This proves (12.2).
It follows from (12.2) that C has at least three noncollinear points P , Q, R. Let
and m be the tangent lines at P and Q respectively, and let S = m; then no three of
P, Q, R, S are collinear. By Theorem 11.1 we may choose coordinates so that P = (1, 0, 0);
S = (0, 1, 0); Q = (0, 0, 1); R = (1, 1, 1). Now the equation of C takes the form
bY 2 + dXY + eXZ + fY Z = 0
where b + d + e + f = 0. The tangents to P and Q are given by dY + eZ = 0 and
eX + fY = 0 respectively; since these tangents pass through S = (0, 1, 0) we must have
d = f = 0 = b and so the equation for C reduces further to bY 2 + eXZ = 0 where
e = b = 0. This gives the required form Y 2 XZ = 0 for the equation of C.
A natural combinatorial generalization of a conic is provided by the following denition: A k-arc in a projective plane is a set O of k points of which no three are collinear. A
line is called a passant, a tangent or a secant of O according as | O| = 0, 1 or 2. An
oval in a projective plane of order n is an (n + 1)-arc. It is natural to ask what examples
of ovals exist other than conics; and whether larger arcs exist than ovals.
12.4 Theorem. Let O be a k-arc in a projective plane of order n. If n is odd then
k n+1. If n is even then k n+2.
Proof. Let P O. Each of the n+1 lines through P has at most one point of O other
than P itself, so that k = |O| n + 2. Suppose there exists an (n+2)-arc; we must show
that n is even. Let P O. Since every line through P must be a secant line, O has no
tangents. Let Q be a point not in O; then every line through Q is either a passant line or
a secant line. This means that |O| = n+2 is even, so n is even.
An (n+2)-arc in a plane of even order n is called a hyperoval.
Consider again the conic XZ Y 2 = 0 in P2 (Fq ), whose points are given by
C = {(1, t, t2 ) : t Fq } {(0, 0, 1)}.
One checks that the tangents at these points are given by
(t2 , 2t, 1)T for t Fq ;
(1, 0, 0)T .
69
If q is odd then these tangents (, , )T satisfy 4 2 = 0, i.e. they form a dual
conic (a conic in the dual plane); in particular no three tangents are concurrent. However
if q is even then the tangents appear more simply as
(t2 , 0, 1)T for t Fq ;
(1, 0, 0)T
and so all tangents pass through N = (0, 1, 0). Thus C {N } is a hyperoval. This
distinction in behaviour between conics in even and odd order generalizes to ovals in
arbitrary nite planes for purely combinatorial reasons:
12.5 Theorem. Let O be an oval in a projective plane of order n. Then every point
of O lies on a unique tangent; thus O has exactly n+1 tangents.
(a) If n is odd then no three tangents are concurrent.
(b) If n is even then all n+1 tangents meet in a point N . Now O {N } is a hyperoval,
the unique hyperoval containing O.
Proof. Let P O. The n points Q O ........ {P } determine n distinct secants through P so
the remaining line through P is a tangent.
Suppose n is odd. Every point T ........{P } lies on an even number of tangents including
since each tangent through T meets O in an odd number (i.e. one) of points, while the
total number of points |O| = n + 1 is even. Since each of the n points T ........ {P } lies
on at least one tangent other than , the tangents other than must meet in n distinct
points. This proves (a).
Suppose n is even and let Q
/ O. Since |O| = n+1 is odd, Q must lie on an odd
number of tangents. Since every point of O lies on a unique tangent, it follows that every
point lies on at least one tangent. Let 0 , 1 , . . . , n be the tangents and let N = 0 1 .
Let S be the set of points not covered by 0 1 , so that |S| = n(n 1). Each of the
remaining tangents j (for j = 2, 3, . . . , n) covers either n or n1 points of S, according
as j does or does not pass through N . In order that the remaining n1 tangents cover
all n(n1) points of S, we therefore require that they all pass through N .
For an oval O in a plane of even order, the common point N of intersection of all
tangents is called the nucleus or the knot of O. For an oval O in a plane of odd order n,
Theorem 12.5 shows that every point P is on 0, 1 or 2 tangents of O; the point P is called
interior, absolute or exterior with respect to an oval O accordingly. (An absolute point
is simply a point of O itself.)
70
12.6 Figure
P is exterior;
Q is absolute;
R is interior
12.7 Theorem. Let O be an oval in a plane of odd order n. Then O has

n+1 absolute points, 12 n(n1) interior points, and 12 n(n+1) exterior points;
n+1 tangents, 12 n(n1) passants, and 12 n(n+1) secants.
Moreover
(a)
(b)
(c)
(d)
(e)
(f)
Every absolute point lies on 1 tangent

and n secants.

1
Every interior point lies on 2 n+1 secants and 12 n+1 passants.

Every exterior point lies on 2 tangents, 12 n1 secants and 12 n1 passants.
Every tangent line contains 1 absolute
point and n exterior
points.

1
1
and
n+1
interior points.
Every passant line contains 2 n+1 exterior points
2

1
Every secant line contains 2 absolute points, 2 n1 exterior points and 12 n1
interior points.
In the dual plane, the tangents of O become the points of a conic O . Duality interchanges the interior, absolute and exterior points of O (respectively, O ) with the
passant, tangent and secant lines of O (respectively, O).
Proof. The number of points and lines
type follows by simple counting arguments;
each
n+1of
1
for example the n+1 tangents give 2 = 2 n(n+1) distinct points of intersection, these
being the exterior points. The number of interior points is therefore
(n2 +n+1) (n+1) 12 n(n+1) = 12 n(n1) .
The nal assertions concerning duality follow from Theorem 12.5(a). Conclusions (a), (b)
and (c) follow from the denitions by straightforward counting. Statements (d), (e) and
(f) follow by duality.
Examples of ovals are known in most of the known nite projective planes. Segres
Theorem 12.14 shows that ovals in classical planes of odd order are necessarily conics. Not
all nite projective planes have ovals; among the twenty-two known projective planes of
order 16, four have no hyperovals and hence no ovals [54].
In planes of even order, every oval extends to a hyperoval; removing any point of a
hyperoval yields an oval. So one naturally regards an oval as a hyperoval with one of its
71
points distinguished. Consider a conic C in a classical plane P2 (Fq ) where q = 2e , and let
N be its nucleus. The resulting hyperoval O = C {N } is called a regular hyperoval.
Let G be the group of all collineations of P2 (Fq ) leaving invariant the hyperoval. For q 4,
the q +2 points of the hyperoval are permuted transitively by G and all ovals are equivalent
under G. However for q 8, the group G has two orbits on the points of the hyperoval,
namely C and {N }. In this case two types of oval are available from O, namely a conic C
and a pointed conic of the form O........{P } where P C. It is sucient, and clearly easier,
to focus our attention on hyperovals rather than ovals when studying planes of even order.
For q 8, every hyperoval in P2 (Fq ) is regular. Irregular hyperovals exist in P2 (Fq )
for even q 16. The unique irregular hyperoval in P2 (F16 ) is due to Lunelli and Sce [41].
We provide here a very brief introduction to hyperovals; for a more complete survey see,
for example, [18].
Let O be a hyperoval in P2 (Fq ) where q = 2e 4. By an appropriate choice of
coordinates we may assume that O contains the points P = (1, 0, 0) and Q = (0, 1, 0).
None of the remaining q points of O may lie on the line P Q = (0, 1, 0)T , so they all have
the form (s, t, 1) with distinct ss (since no two of them are collinear with P ) and distinct
ts (since no two of them are collinear with Q). Thus
O = {(f(t), t, 1) : t Fq } {(1, 0, 0), (0, 1, 0)}
for some permutation f : Fq Fq . It is convenient to regard f(X) as represented by a
polynomial in Fq [X], which must therefore be a permutation polynomial; see Appendix
A2. We regard P Q as the line at innity, and the points of O outside this line as the graph
of f in the ane plane A2 (Fq ). Since no two ane points of O are collinear, we have
f(r)
det f(s)
f(t)
r
s
t
1
1 = 0
1
whenever r, s, t Fq are distinct; equivalently,

(12.8)
f(s) f(t)
f(r) f(t)
=
whenever r, s, t Fq are distinct.
rt
st
Of course all signs here are just + signs in characteristic two; but we write them as
signs so that the quotients in (12.8) will be clearly recognized as slopes of secants in the
graph of f. We have reduced the search for hyperovals, to the search for permutations
satisfying the condition (12.8). Again it is convenient to regard f(X) as a polynomial of
degree less than q. Accordingly, any permutation polynomial satisfying (12.8) is called an
o-polynomial. Often it is convenient to add the conditions that f(0) = 0 and f(1) = 1.
These conditions we may assume, since with appropriate choice of coordinates we may also
assume that O contains (0, 0, 1) and (1, 1, 1).
72
12.9 Example: Regular hyperovals. Let q = 2r , f(t) = t2 . The resulting hyperoval

consists of the conic XZ Y 2 = 0 together with the nucleus (0, 1, 0). Note that f is
simply the generator of Aut Fq = = {1, , 2 , . . . , r1 }, : x x2 . One checks
directly that (12.8) reduces to the condition that r + t = s + t whenever r, s, t are distinct.
This condition is clearly satised, as well as the condition that f be bijective.
r1
1
One may also let f(t) = t1/2 = t2 , i.e. f = = r1 . In this case O is clearly the
conic X 2 Y Z = 0 together with its nucleus (1, 0, 0). Here the condition (12.8) becomes
r1/2 + t1/2
s1/2 + t1/2
1
1
=
=

=
r1/2 + t1/2
r+t
s+t
s1/2 + t1/2
whenever r, s, t are distinct. Raising both sides to the power 2 gives r + t = s + t as
before.
The latter example illustrates the more general observation that if f is an o-poly1
nomial, then the inverse function f is also an o-polynomial whose hyperoval is equivalent
to that dened by f, by an interchange of the x and y coordinates.
k
12.10 Example: Segre hyperovals. Let q=2r, f(t)=t2 for some k {0, 1, 2, . . . , r1}
so that f = k is a eld automorphism. Thus f is automatically bijective. Since f is a
eld automorphism, (12.8) becomes that
k
(s t)2
(r t)2
=
rt
st
whenever r, s, t are distinct; equivalently, the map x x2 1 is bijective. This means that
gcd(2k 1, 2r 1) = 2gcd(k,r) 1 = 1, i.e. gcd(k, r) = 1. The resulting hyperovals are
irregular if k
/ {1, r1}. The smallest example arises for r = 5 and k = 2, which gives an
irregular Segre hyperoval
{(t4 , t, 1) : t F32 } {(1, 0, 0), (0, 1, 0)}
in P2 (F32 ). By the preceding remarks, the hyperoval arising from the o-polynomial t8 =
1
f (t) is equivalent to the one arising from f.
12.11 Example: The Lunelli-Sce Hyperoval. Let q = 16 and write F16 = F2 []
where 4 = + 1 (see Appendix A2.3). The unique irregular hyperoval in P2 (F16 ) may
be dened using the o-polynomial
f(t) = t12 + t10 + 11 t8 + t6 + 2 t4 + 9 t2 .
Dont even think about trying to verify by hand that this is an o-polynomial! Fortunately
this is an easy job using appropriate computer software. The remarkable thing is that
Lunelli and Sce discovered this hyperoval in 1958 by computer!
73
The subject of hyperovals (and connections to ocks and q-clans) is too big a subject
to cover here! But I cant resist giving a couple further examples.
12.12 Example: Trinomial Hyperovals. We describe two innite families of hyperovals for q = 22e+1, the Payne hyperovals and the Cherowitzo hyperovals. Both are
described using o-polynomials with only three terms. Here F
q is cyclic of order q 1 1
mod 6, so the map x x6 is bijective, which means that every x Fq has a unique
sixth root x1/6 . This is a monomial in x if the exponent is interpreted mod (q1). Dene
f(x) = x1/6 + x3/6 + x5/6 . Payne [52] showed that this is an o-polynomial.
Also, Aut(Fq ) is cyclic of odd order 2e + 1. Therefore every automorphism of Fq has
a unique square root. Let Aut(Fq ) such that 2 is the usual generator x x2 of
Aut(Fq ). Then f(x) = x + x+2 + x3+4 is an o-polynomial [17]. Cherowitzo conjectured
this result in 1985 based on computational evidence for small values of e. His proof of this
result, a decade later, was quite a tour de force.
Finally we show that ovals in classical planes of odd order are conics.
12.13 Lemma. Let O be an oval in P2 (Fq ) where q
is odd. Let A, B, C be three distinct points of O as
shown, and let A , B , C be the vertices of the triangle
formed by the tangents at A, B, C, with A opposite
A, etc. Then the three lines AA , BB , CC are concurrent.
Proof.
We may choose coordinates such that

A = (1, 0, 0),
B = (0, 1, 0),
C = (0, 0, 1),
A = (1, 1, 1).
The tangents to O at A, B, C are

(0, 1, )T ,
(1, 0, 1)T ,
(1, 1, 0)T
respectively, for some = 0; this means that B = (, , 1), C = (1, , 1).
74
Consider an arbitrary point P O........{A, B, C}, so that P has coordinates (x, y, z) where
xyz = 0. The secants AP , BP , CP have coordinates
1
(0, 1, yz )T,
(zx , 0, 1)T,
(1, xy , 0)T
respectively, where the product of the three nontrivial entries satises

1
(yz )(zx )(xy ) = 1.

Now consider a triple of lines of the form
A,a = (0, 1, a)T ,
B,b = (b, 0, 1)T ,
C,c = (1, c, 0)T
such that abc = 0. This gives an arbitrary triple of lines distinct from the sides of the
triangle ABC, but passing through A, B, C respectively. These lines must correspond (in
some order) to the secants and tangents at A, B, C considered above. Therefore

a
b
c = ()(1)(1) (1)q2 .

aFq
bFq
cFq
tangents
secants
Here the rst three factors arise from the fact that the tangents at A, B, C are A, , B,1 ,
C,1 respectively. The latter expression (1)q2 arises from the q2 triples of secants of
the form AP , BP , CP , each of which contributes a factor 1 to the product as we have
seen. By Exercise #2 each of the three products on the left is 1, so we obtain = 1.
Finally we obtain
AA = (0, 1, 1)T ,
BB = (1, 0, 1)T ,
CC = (1, 1, 0)T
and these three lines all meet at a point (1, 1, 1) as required.
12.14 Segres Theorem. If O is an oval in a nite classical plane P2 (Fq ) where q

is odd, then O is a conic.
Proof. We use the same notation as in the proof of Lemma 12.13. Denote the tangent
to O at P = (x, y, z) by (a, b, c) and consider the triangle formed by the tangents at
B, C, P . This triangle has vertices
(c, c, ba),
(b, ca, b),
(1, 1, 1)
opposite B, C, P respectively, as shown. The secants joining B, C, P to the opposite vertices

of the triangle are
(ab, 0, c)T,
(ac, b, 0)T ,
(yz, xz, x+y)T
respectively. By Lemma 12.13 these three secants are concurrent, so that
ab ac yz
0 = det 0
b
xz = (a b c)(bx + by cx cz).
c
0
x+y
Since A = (1, 1, 1) does not lie on the secant (ac, b, 0)T through C, we have
b(x+y) = c(x+z).
75
Applying the same argument to the triples ACP and ABP in place of BCP , and exploiting
the symmetry in the three coordinates, we conclude that (y+z, x+z, x+y) = (a, b, c) for
some F
q ; thus
2(xy + xz + yz) = (ax + by + cz) = 0.
Thus every point P O ........ {A, B, C}, as well as the three points A, B, C, lie on the conic
C dened by xy+xz+yz = 0. It follows that O = C.
Exercises 12.
1. How many ovals and hyperovals does a projective plane of order 2 have? Show that all ovals in this
plane are conics.
2. Show that the product of all nonzero elements in any nite eld is 1. (This fact is used in the proof
of Lemma 12.13.)
Hint:
Most terms in this product occur in pairs a, a1.
3. Consider the oval O = {0, 1, 2, 6} in the projective plane of order 3 with point set P = {0, 1, 2, . . . , 12}
and line set L = {A, B, . . . , M } as labelled below.
9
.............................
........................
.............. .......... ... ........... ......................
.........
.
.....
.
....
.... ...............................................
..................................... ........
.
.
.
.
.
.
.
.
.
..
.. ....
..............
.........
.... ....
...... ......
......
..
..........
.................
.
.
.
.
.
.
.
.
.
..........
.
.
.
.
.
...
....
.
... ....
..........
...
...
.... ... ........
.............................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.... . ....
. ..
...........
.... ...............................12
...
.
.
5.....
.
.
.
.
.
...........
...................
....... .....
.
... ........ L
......
.
.
.
.
.
.
.
.
.
.
.
.
.
.......
..................
. ...
.... . ....
......
..... ..................
....
.......
.... .. .......
.... .. .....
.... ...
.....
... .... ....
... .......
... ....
.... .... .......
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.... ....
....
.... ..... ..
...
.... ...
..
....
...
... ..........
...
.... ...
....
..I
....
.... A...
..
..
...
..
......
........
....
.......... G....
.
D
.
.
.
....
..
.
.
..
.
...
.
.
.
.
.
.
....
.
.
.
. ..... M
.. ..... ... ..... .....
.... ..
.
..
.
.
C
....
.
...
.
.
.
.... . ...
.... . ...
.
.... .. .....
.
.
.
....
.
..
.
.
.
.
.
.... . ....
.... . ....
.... . ....
.
E .... F .......
.
.........
..........
.
.
.
.
...................................
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.. .....
.....
............
......
....
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
...
....
3 .... ........
.... ...........
.. ..
.... .. 8
.... .. ....
....
...
..... .... ...J
....
.... ....... 7... ....... ...... ...
....
..
...
..... ..
....
..
.... .....
.... ....
....
...
..
...
...
.......
H
....
.......
....
......
...
.
....
.
.
.
.
.
.
.
.
.
.
....
.
. .
..
....
..... ........ ...
.... ........ .......
....
.... ... ....... .......
....
.
.
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.... . ...
.... . ...
.... . ....
.
.... ...
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
....
.
.... .. ....
.
....
......................................................................................................................................................................................................................
....
......
..
.
..........
..
..
...
...
.
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1 ... ......
.
. .
....
B ......... 10
.....2.... .......
.....
.... ... 4
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
.... ....
.... ....
...
....
......
....
......
....
....
....
....
.....
..
....
..
..
..
...........
.....
.....
.... .....
....
.....
.... .... ........ ....... .... ........ ....... .... ........
.
.
.
.
.
.... . ....
.... . ....
.... . ....
.. K
.... ......
......
. . .
. . .
.....................................................................................................................................................
11
76
(a) List the interior points; absolute points; exterior points; passant lines; tangent lines; and secant
lines for O.
(b) Complete the following sentences correctly, using your sketch of the plane:
Every interior point lies on
every absolute point lies on
every exterior point lies on
Every passant contains
every tangent contains
every secant contains
passants,
passants,
passants,
interior points,
interior points,
interior points,
tangents, and
tangents, and
tangents, and
secants;
secants;
secants.
absolute points, and

exterior points;
exterior points;
exterior points.
4. Let 1 , 2 , . . . , k be k 1 distinct lines in a projective plane (P, L) of order n. Suppose that for every
point P P, the number of i {1, 2, . . . , k} such that P i is even.
(a) Show that k n+2.
(b) Can k = n+2? Does your answer depend on n or the choice of plane (P, L)? Explain.
Hint.
(a) Consider the intersections of 2 , . . . , k with 1 . (b) Think: hyperovals.
13. Codes of Planes

Later in this section, we will require basic terminology and results of coding theory and
invariant theory, as briey summarized in Appendices A5 and A6. We begin, however, by
introducing the Smith normal form of an arbitrary integer matrix.
13.1 Theorem. Let A be an
m n matrix with integer entries. Then there is a
unique m n matrix D = dij (the Smith normal form of A) satisfying MAN = D
where M and N are integer matrices of size m m and n n respectively, each having
determinant 1, such that the entries of
integers satisfying dij = 0
D are non-negative

for i = j (i.e. D is diagonal) and d11 d22 d where = min{m, n}.
Note that if any of the diagonal entries dii is zero, then all remaining diagonal entries djj
for j i must also vanish. The diagonal entries dii , which uniquely determine the Smith
normal form of A, are called the elementary divisors of A. The uniqueness of the Smith
normal form is shown in Exercise #5. The existence follows from Exercise #6, which in
fact gives an algorithm for computing the Smith normal form. This algorithm starts with
the observation that d11 equals the greatest common divisor of all entries of A. As a rst
example, the randomly generated matrix
20
33
A0 =
14
25
16
43
40
13
22 12
40 8
22 12
26 20
44
56
2
62
77
13. Codes of Planes
may be factored as
6
7
11 11
A0 =
6
4
7
9
11
20
11
13
1
8
15 0
0
9
0
9
0
2
0
0
0
0
2
0
0
0
0
0
1 1
0
1 4
0
0 3
0
1
0
0
1 0
0 0
0 4
1 2
0 1
0 0
2
7
3
2
where the square matrices shown have determinant 1 and 1 respectively; thus the elementary divisors of A0 are 1, 2, 2, 0, which we abbreviate as 11 22 01 .
Once again let A be an arbitrary m n matrix with integer entries. For any prime p,
the p-rank of A is the rank of A over any eld of characteristic p. (The rank of the matrix
A over a eld F , depends only on the matrix A and the characteristic of F .) For example
it is clear from the Smith normal form given above, that the matrix A0 has p-rank equal
to 3 (the same as the rank over elds of characteristic zero) for p = 2, whereas its 2-rank
equals 1. More generally the p-rank of A is the number of elementary divisors of A which
are not divisible by p.
If A is an incidence matrix of an incidence structure with m points and n blocks, then
we may refer to the p-rank of A as the p-rank of the incidence structure itself. Although
there are in general exponentially many choices of incidence matrix for a given structure,
arising from permutations of the rows and columns of A, these permutations will not aect
the p-rank of A, and so the p-rank of the incidence structure is well-dened. The p-rank
serves as a useful isomorphism invariant, since it is easily computed by Gaussian elimination. This is because the only matrix entries which arise during intermediate computations
are 0, 1, 2, . . . , p1. By comparison, computing the rank of a matrix over Q is typically
very expensive, since there is no bound on the complexity of the rational entries which arise
during intermediate computations. Similarly, the algorithm suggested in Exercise #6 for
computing Smith normal forms typically leads to excessively large matrix entries during
intermediate stages of the computation, even if the resulting elementary divisors turn out
to be rather small. Fortunately there are other algorithms which help to overcome this
obstacle when computing Smith normal forms. The elementary divisors of the incidence
structure are also an isomorphism invariant, which in general give stronger information
than the p-ranks themselves, since given the elementary divisors, one can easily read o
the p-ranks for all primes p.
For example two of the four projective planes of order 9 have p-ranks as listed for
small primes p:
plane
P G2 (9)
Hall plane of order 9
2-rank 3-rank 5-rank 7-rank elementary divisors

90
90
37
41
90
90
91
91
137 318 935 901

141 310 939 901
78
The Smith normal form (or the 3-rank, which is much easier to compute) suces to
prove that these two planes are non-isomorphic. The p-ranks for primes p not dividing
the order of the plane, however, carry no information useful in distinguishing the plane;
they depend only on the order of the plane. See Exercise #4 for an explanation of this
phenomenon. Unfortunately it is not always possible to distinguish non-isomorphic objects
using p-ranks alone; for example the Hughes plane of order 9 has the same p-ranks as the
Hall plane of order 9. A good rule of thumb is that more classical (or symmetric, or nicer)
objects tend to have lower p-ranks for the relevant primes, as well as larger automorphism
groups.
Let A be an incidence matrix of a projective plane of order two. Any such incidence
matrix will do; we will take
0
0
A = 0
1
1
0
0
1
1
0
0
1
1
1
0
0
0
0
1
0
1
0
1
0
1
0
1
0
1
0
0
1
0
1
0
0
1
1
0
0
0
1
0.
0
0
We check that the elementary divisors of A (i.e. of the plane of order 2) are 14 22 61 , and
in particular its 2-rank is 4. Let C F72 be the row space of A over F2 , so that dim C = 4
and |C| = 24 = 16. The sixteen vectors of C are listed in the table in Appendix A5.8;
that is, C is the [7, 4, 3] binary Hamming code as presented there. Its codewords of weight
three and four are simply the rows of A and their complements (obtained by switching
0 1); together with the zero vector and the unique vector of weight 7, these account
for all 16 codewords. Dierent choices of incidence matrix for the projective plane of
order 2 give Hamming codes with the same parameters, diering from our code C only by
a permutation of the seven coordinates. Conversely, every binary [7, 4, 3]-code has exactly
seven codewords of weight three, and these form the rows of an incidence matrix for the
projective plane of order 2. This connection between the smallest projective plane and the
(arguably) smallest interesting code, a perfect code. But there is more. . .
Let C F82 be the linear code obtained from C by adding one extra coordinate as a
parity check. Comparing with the list in Appendix A5.8 we see that
C = {00000000, 00011110, 00101101, 00110011, . . . , 11111111}.
Note that C has 14 words of weight 4, together with the zero vector and the vector 11111111
of weight 8; thus the weight enumerator of of C is
8
4 4
8
A
C (x, y) = x + 14x y + y .
13. Codes of Planes
79
This code C is the [8, 4, 4] extended binary Hamming code. A generator matrix for C
is given by the 7 8 augmented matrix

G = A 1T
where 1 = (1, 1, 1, 1, 1, 1, 1). Since
GGT = AAT + 1T 1 = 7I + J + J = 0
(the 7 7 matrix of zeroes over F2 ) it follows that C C ; but then comparing dimensions,
that C is self-dual; that is, C = C. In particular from the MacWilliams relations A5.4 we
have
1
A
C (x, y) = A
C (x + y, x y) ,
C (x, y) = 16 A
i.e.
x8 + 14x4 y 4 + y 8 =
1
16

(x+y)8 + 14(x+y)4 (xy)4 + (xy)8 ,
which can be easily veried by routine expansion.

Several generalizations of this construction of the [7, 4, 3] binary Hamming code are
available:
If q = pr then the code (over any eld of characteristic p) of the projective plane P2 (Fq )
r
+1 . This is a special case of Theorem 22.1. However, the weight
has dimension p+1
2
enumerators of these codes are not known in general. Indeed, no weight enumerator
has currently been computed for any plane of order exceeding 8. The problem of
computing weight enumerators of general codes, is recognized as exceedingly hard: no
polynomial-time algorithm for this is known, or is likely to exist.
The [7, 4, 3] binary Hamming code is the rst member of an innite family of perfect
1-error correcting binary Hamming codes which may be constructed from higherdimensional projective spaces over F2 ; see Exercise #19.5.
If a prime p divides the order n of a projective plane, then the extended code of the
plane over any eld of characteristic p is self-orthogonal (i.e. C C ) with respect
to an appropriately
bilinear form. Now suppose
that p sharply
chosen2 symmetric

divides n, i.e. p n but p n; this condition is denoted p n. In this case we will
show that equality holds, i.e. C = C , so that C is self-dual with respect to B. These
facts we proceed to explain below.
Let A be an incidence matrix for a projective plane of order n, and let p be any prime
dividing n. The extended code C of the plane, over the eld Fp , has generator matrix

G = A 1T
80
of size N (N +1) where N = n2 +n+1 and 1 = (1, 1, . . . , 1) FN

p . Using the symmetric
+1
dened
by
bilinear form on FN
p
1 0 0 0
0 1 0 0
. . .
T
. . . . ... ...
B(x, y) = xDy = x1 y1 + x2 y2 + + xN yN xN +1yN +1 , D =
. .
0 0 1 0
0 0 0 1
Any two codewords have the form xG, yG C for some x, y FN

p , so that
B(xG, yG) = xGDGT y T = x(AAT 1T 1)y T = x(nI + J J )y T = 0

since p n. This shows that C is self-orthogonal with respect to B. Equivalently,
CD C where is the perp with
standard
dot product

respect
1to the
1
N +1
2

on Fp . In particular, dim C 2 N +1 = 2 n +n+2 .

We will show that equality in (13.2), i.e. C is self-dual, whenever p n:
(13.2)
13.3 Theorem. If a prime

p sharply
divides n, then every projective plane of order

n has p-rank equal to 12 n2 +n+2 . Moreover the extended code C is self-dual (with
respect to the form B as indicated above).

Proof. It remains only to be shown that dim C = 12 N + 1 . Since 1T lies in the column
space of A, the matrix G has the same p-rank as A, i.e. dim C = dim C. Since
(det A)2 = det(AAT ) = det(nI + J ) = (n + 1)2 nn(n+1)
by Exercise #2, we have
| det A| = (n+1)nn(n+1)/2 = d1 d2 dN

where d1 , d2 , . . . , dN are the elementary divisors of A. Sincep n, at most 12 n2 +n
of the
di s are divisible by p, so that the p-rank of A is at least 12 n2 +n+2 = 12 N +1 . This is
the reverse of the inequality (13.2), so that in fact equality must hold.
13.4 Corollary. Let p be a prime sharply dividing n, and suppose C is the extended
code of length N +1 = n2 +n+2 (over Fp ) of a projective plane of order n. Then the
weight enumerator A
C (z) satises
x + (p1)y x y

1
,
.
(x,
y)
=
A
x
+
(p1)y,
x
y
=
A
A
C
C
C

p
p
p(N +1)/2
13. Codes of Planes
81
Proof. By Theorem 13.3, equality holds in (13.2), i.e. C = CD. But D simply multiplies the last coordinate of every codeword by 1, and this does not change the weight
of any word. Thus C has the same weight enumerator as C. The result follows from the
MacWilliams relations A5.4.
Henceforth we assume for simplicity that p = 2; in this case the result of Corollary 13.4
reads
(13.5)
A
C (x, y) = A
C
x + y x y
,
.
2
2
In this case we also have

(13.6)
A
C (x, y) = A
C (y, x),
which simply expresses the fact that And = Ad : the map w 1 + w gives a bijection
between codewords of weight d and codewords of weight nd. Thus the homogeneous
polynomial A
C (x, y) C[x, y] of degree N + 1 is invariant under the group
G=
1 1 1
2 1 1
0 1
, 1 0 .
We check that this is a group of order 16:

0

, 01 1
G = 10 01 , 01 10 , 10 1
0 ,
1

1 1
1 1 1
1
, 12 11 1
.
12 11 1
1 , 2 1 1 , 2 1 1
By Moliens Theorem A6.4 the ring of invariants C[x, y]G has Hilbert series
!
1
1
8
2
2
2
1
+
+
+
+
+
16 (1t)2
(1+t)2
1t2
1+t2
1 2t+t2
1+ 2t+t2
1
.
=
2
(1 t )(1 t8 )
HC[x,y]G (t) =
By Theorem A6.5 we will be done if we can nd two algebraically independent invariants

1 and 2 of degree 2 and 8 respectively. Since all matrices in G are real orthogonal, it
is clear (cf. Example A6.2) that x2 + y 2 is invariant; let us take 1 (x, y) = x2 + y 2 . Also
the weight enumerator x8 + 14x4 y 4 + y 8 of the [8, 4, 4] extended binary Hamming code
must be invariant since it arises from a plane of order 2; we consider the slightly simpler
polynomial

2 (x, y) = 14 x8 + 14x4 y 4 + y 8 14 = x2 y 2 (x2 y 2 )2
82
which is evidently also invariant. In order to show that 1 and 2 generate the full ring
of invariants C[x, y]G , it suces to show that 1 and 2 are algebraically independent.
Suppose that there exists a nonzero polynomial
h(T1 , T2 ) = cT1i1 T2i2 + C[T1 , T2 ]
such that h(1 , 2 ) = 0. Here c = 0 and T1i1 T2i2 is the lex-highest monomial appearing in
h(T1 , T2 ): that is, for every monomial T1j1 T2j2 appearing in h, either
i1 > j1 , or
i1 = j1 and i2 j2 .
Then the coecient of x2i1 +6i2 y 2i2 in h(1 , 2 ) is c = 0; in particular, h(1 , 2 ) cannot
vanish. This shows that 1 and 2 are algebraically independent. It follows (cf. Example
A6.8) that {1 , 2 } is a fundamental set of invariants, i.e.
C[x, y]G = C[1 , 2 ].
In particular if there exists a projective plane of order n 2 mod 4, then its extended
binary code of length N +1 = n2 +n+2 must be a polynomial combination of x2 +y 2 and
x2 y 2 (x2 y 2 )2 . A stronger result may be shown: It is not hard to show (Exercise #8) that
all codewords in C have weight divisible by 4, so that A
C (x, y) is invariant under the group
1 0
G, 0 i of order 192. The ring of invariants for this larger group has Hilbert series
1
(1
t8 )(1
t24 )
and a set of fundamental invariants is

x8 + 14x4 y 4 + y 8
and
x4 y 4 (x4 y 4 )4 .
Every plane of order n 2 mod 4 has extended binary code whose weight enumerator is
a polynomial combination of the latter two polynomials.
For example this line of reasoning played a role in the proof of the nonexistence of
a projective plane of order 10. For a history of this work, see Lam [36]. Early work
on a possible plane of order 10 showed that the weight enumerator AC (x, y) would be
uniquely determined if the number of codewords of weight 12, 15 and 16 were known.
Some geometric arguments on the part of several researchers (everything from graduate
student slave labour to the intensive eorts of Fields Medalist John G. Thompson), and
extensive case-checking by computer, showed that no codewords of weight 12, 15 or 16
83
13. Codes of Planes
could exist in such a code, and that the weight enumerators of C and the extended code C
must have the form
AC (x, y) =
x111
+ 111 x100 y11
+ 24,675 x92 y19
+ 386,010 x91 y20
+ 18,864,495 x88 y23
+ 78,227,415 x
87 24
+ 2,698,398,790 x84 y27

+ 8,148,873,195 x83 y28
+ 166,383,964,620 x80 y31
+ 415,533,405,150 x79 y32
+ 5,023,148,053,500 x76 y35
+ 10,604,483,511,375 x75 y36
+ 78,347,862,432,300 x
72 39
+ 141,031,595,676,060 x71 y40

+ 653,162,390,747,370 x68 y43
+ 1,009,413,831,402,540 x67 y44
+ 2,982,186,455,878,665 x64 y47
+ 3,976,279,652,851,020 x63 y48
+ 7,582,305,834,092,682 x60 y51
+ 8,748,789,607,170,360 x
59 52
+ 10,841,059,295,003,634 x56 y55

+ 10,841,059,295,003,634 x55 y56
+ 8,748,789,607,170,360 x
52 59
..
.
A
C (x, y) =
x112
+ 111 x100 y12
+ 410,685 x92 y20
+ 97,091,910 x88 y24
+ 10,847,271,985 x84 y28

+ 581,917,369,770 x80 y32
+ 15,627,631,564,875 x76 y36
+ 219,379,458,108,360 x72 y40
+ 1,662,576,222,149,910 x68 y44
+ 6,958,466,108,729,685 x64 y48
+ 16,331,095,441,263,042 x60 y52
+ 21,682,118,590,007,268 x56 y56
+ 16,331,095,441,263,042 x52 y60
+ 6,958,466,108,729,685 x48 y64
+ 1,662,576,222,149,910 x44 y68
+ 219,379,458,108,360 x40 y72
+ 15,627,631,564,875 x36 y76
+ 581,917,369,770 x32 y80
+ 10,847,271,985 x28 y84
+ 97,091,910 x24 y88
+ 410,685 x20 y92
+ 111 x12 y100
+ y112
+ y111 ;
The fact that the extended code C = C has just 111 codewords of weight 12, arising from
the lines of the plane, means (by Exercise #7) that no hyperovals can exist in such a plane;
this result itself was a huge computational feat. The nal task was to consider all possible
congurations for the 24,675 codewords of weight 19 in C (achieved by Thompson) and
nally to show that none of these could lead to a projective plane of order 10. This required
thousands of hours of supercomputer time during the late 1980s. The nal announcement
of the nonexistence of a projective plane of order 10, spread like wildre, not only through
mathematical circles, but also through the popular media. To this day, Lam refers to this
as a computer result rather than as a proof. His work gives fuel to the debate over the
role of computers in proving mathematical results.
If we wish to determine whether or not there exist planes of order 15, 18, 20, etc.
then evidently another method will be required; if we follow the same method used to rule
out the plane of order 10, the computer resources required would vastly exceed all the
resources available in the world. Clearly, a better idea is needed!
Exercises 13.
1. Find the Smith normal form of the projective plane of order 1 with incidence matrix
84
1
1
1.
2. Let B = nI + J be an N N matrix with 1s o the main diagonal and n+1s on the main diagonal.
Let 1 be the N 1 vector of 1s, and let 1 be its orthogonal complement. Show that 1 and its
orthogonal complement 1 are eigenspaces for B and determine the corresponding eigenvalues. Hence
evaluate the determinant of B.
3. Denote by (A) the nullity of a matrix A, i.e. the dimension of the right null space of A. Show that
(AB) (A) + (B) assuming the matrix product AB is dened.
4. Let A be the incidence matrix of a projective plane of order n, so that AAT = nI + J. Using
Exercises #2,3, show that
(a) The p-rank of A is n2 +n+1 for any prime p not dividing n(n+1).
(b) The p-rank of A is n(n+1) for any prime p dividing n+1.
(c) The p-rank of A is at most n(n+1)/2 for any prime p dividing n.
5. Let A be an m n matrix, and let R = AZn , which is the additive group (i.e. Z-module) generated
by the columns of A.
(a) Show that the quotient group Zm /R is unchanged if A is replaced by M AN where M and N are
integer matrices with determinant 1, having size m m and n n respectively. That is, show
that the quotient group Zm /M AN Zn is isomorphic to Zm /R.
(b) If d11 , . . . , d are the elementary divisors of A where = min{m, n}, show that the quotient
group
Zm /R
= (Z/d11 Z) (Z/p Z).
(c) Conclude that the Smith normal form of A is unique, assuming the Fundamental Theorem of
Finitely Generated Abelian Groups: Every nitely generated abelian group G is uniquely expressible as
G
= (Z/d1 Z) (Z/dr Z)
where the non-negative integers satisfy d1 | d2 | | dr and r is the minimum number of generators
of G.
6. Let A be an m n integer matrix. An integral elementary row operation on A is any of the
following: a permutation of the rows of A, or the addition of an integer multiple of any row to any
dierent row, or the multiplication of any row by 1. We dene an integral elementary column
operation similarly.
(a) Using the extended Euclidean algorithm, show that there exists a sequence of elementary row and
column operations which when applied to A, yields an m n matrix whose (1, 1) entry equals the
gcd of all entries of A, and the remaining entries in the rst row and column are all zero.
(b) Show that by repeated application of (a) one obtains the Smith normal form of A.
(c) Use the algorithm suggested above to obtain the Smith normal form for the projective plane of
order 2.
7. Let C be the binary code of a projective plane (P, L) of even order n, and let C be the extended code
of length n2 +n+2.
(a) Show that C has minimum weight n+1, and that C has exactly n2 +n+1 codewords with this
minimum weight, whose supports are precisely the lines of (P, L).
of the extended code, has minimum weight n+2; and that minimum weight
(b) Show that the dual C

codewords in C are of just two types: extended lines of (P, L) (see (a)); and hyperovals of (P, L).
14. The Bruck-Ryser Theorem
85
(Thus for n = 2 the 14 codewords of weight 4 in the extended [8, 4, 4] binary Hamming code arise
from the 7 lines and the 7 quadrangles of the plane.)
Hint.
Use Exercise #12.4.
be the extended binary code of a projective plane (P, L) of order n 2 mod 4, so that C
is
8. Let C
has weight divisible by 4.
self-dual of length n2 +n+2. Show that every codeword in C
Hint. Every codeword is a sum (mod 2) of rows of the augmented matrix G = [A 1T ]. Use induction
on the number of rows in such a sum.

The celebrated main theorem of this section provides a general criterion for nonexistence
of projective planes of certain orders. Apart from the computer proof of the nonexistence
of a plane of order 10, this is the only such nonexistence result currently known! It states
14.1 Theorem (Bruck and Ryser [9]). Suppose there exists a projective plane of
order n 1 or 2 mod 4. Then n = a2 + b2 for some a, b Z.
In practice the fastest way to check whether n is a sum of two squares uses the prime
factorization of n; see Lemma 14.8(iii) below. For example none of the integers 6, 14, 21,
22, 30, 33 is a sum of two squares, so none of these can be the order of a projective plane.
With the exception of order 6, which was previously settled by the nonexistence of two
orthogonal Latin squares of order 6, these results were not previously known. The original
proof of Bruck and Ryser (1949) relies on a deep theorem of Hasse and Minkowski. Since
its publication, more elementary (but somewhat ad hoc) proofs have become available;
see [31] or [13]. We have chosen to present here a proof in the spirit of the original 1949
publication. Although our proof is not self-contained, we feel it is more important for
the student to place this result in a larger context (namely, questions of congruence of
rational quadratic forms) and to learn a tool that can be applied in a broader range of
settings than just the current one (i.e. questions about possible orders of projective planes).
As with many algorithmic approaches, this tool can be eectively applied without a full
understanding of why it works.
Consider a symmetric N N matrix M with rational entries. Then M represents a
rational quadratic form QM : QN Q dened by QM (v) = vMv T . Consider a change
of basis for QN of the form v = v"A where A is an invertible N N matrix over Q. Relative
to the new coordinate vector "
v , the quadratic form becomes
v T = QAM AT ("
v ).
QM (v) = vMv T = v"AMAT "
Thus the new symmetric N N matrix AMAT represents the same quadratic form as the
# is rationally
original matrix M, relative to a new basis of QN . We say that a matrix M
# = AMAT for some invertible N N matrix A with rational entries.
congruent to M, if M
86
Now a projective plane of order n 2 (with a prescribed ordering of its N points and
N lines, where N = n2 + n + 1) is equivalent to an N N matrix A of 0s and 1s such
that
(i) AJ = J A = (n+1)J ; and
(ii) AAT = nI + J
where J is the N N matrix with every entry equal to 1. Rewriting (ii) as AIAT = nI+J ,
we see that the matrices I and nI+J are rationally congruent. This property imposes
strong necessary conditions on the matrices I and nI+J , and therefore on the integer n
itself, as we describe below.
Congruence transformations are reminiscent of similarity transformations: A matrix
#
# = AMA1 for some invertible matrix A. (Equivalently,
M is similar to a matrix M, if M
# are similar if they represent the same linear transformation T : QN QN
M and M
relative to dierent bases. Indeed if w = vM then changing bases via the change-of-basis
1
matrix A for both domain and range gives w = wA
" and v = v"A, so that w
" = v"AMA .)
There is in principle a simple criterion to test whether two symmetric rational matrices
are similar: the necessary and sucient condition is that they have the same spectra, i.e.
list of eigenvalues with multiplicity. (The condition for similarity is simplied in the case
of symmetric matrices since real symmetric matrices are diagonalizable.)
We thus require a similar criterion to determine whether or not two symmetric N N
rational matrices are congruent. Fortunately such a criterion exists, and it is presented
below without proof; the student is referred to [58] for proofs and more explanation. At
least one necessary condition is clear: taking determinants we see that congruent matrices
# must satisfy det M
# = c2 det M for some nonzero c Q. Thus we dene the disM, M
criminant of a quadratic form Q, denoted disc Q, to be the determinant of the associated
matrix M. We regard the discriminant as dened only to within multiplication by the
square of some nonzero rational number, so that disc Q is a well-dened element of the
quotient group Q /(Q )2 , independent of the choice of basis; here (Q )2 < Q is the
subgroup consisting of rational squares.
We rst recall the Legendre symbol dened for an odd prime p, and an integer
a 0 mod p:
a 1,
if a is a square mod p,
=
p
1, otherwise;
This symbol has a unique
multiplicative extension to all rationals a = r/s in lowest terms

with r, s Z and p rs: one simply denes
r/s
p
rs
=
.
p
Note that this symbol is multiplicative in its upper argument: if a, b Q have no factor
of p in either the numerator or denominator when written in lowest terms, then
ab a b
=
.
p
p p
87
We now introduce the Hilbert symbol (a, b)p for an odd prime p and arbitrary nonzero
a, b Q: write
a = a0 ps , b = b0 pt
where s, t Z and the rational numbers a0 , b0 in lowest terms have no factor of p in either
the numerator or the denominator, and dene
(a, b)p =
(1)st at bs
p
0 0
= (1)st(p1)/2
a t b s
0

= (1)(p1)/2 ; see Proposition A1.3(i).
Here we have used the elementary fact that 1
p
Some explanation of the meaning of the Hilbert symbol is given in remarks found at the
end of this section; for now the reader is asked to simply work with this denition. The
Hilbert symbol is easily evaluated by the following rules, where a, b, c Q and p is an
odd prime:
(14.2) (b, a)p = (a, b)p = (a, c2 b)p ;
(14.3) (a, b)p = 1 whenever a, b Z with ab 0 mod p;
(14.4) (ac, b)p = (a, b)p (c, b)p ;

(14.5) (a, p)p = ap whenever a Z with a 0 mod p.
Auseful identity,easily proved by considering separately the three mutually exclusive cases
p n, p n+1, p n(n+1), is
(14.6) (n+1, n)p = (n+1, 1)p ; in particular, (n+1, n)p = 1.
Let M be a nonsingular symmetric N N matrix with rational entries, and for
k = 1, 2, . . . , N let Mk denote the upper-left k k submatrix of M. In particular MN = M
"M x
"T = xMk xT = 0 where
and det Mk = 0 for every k. (If xMk = 0 for some x Rk then x
" = 0, and this forces
x
" RN consists of the vector x followed by nk zeroes, from which x
x = 0.) For every odd prime p we dene

cp (M) = 1, det M p

det Mk , det Mk+1
1kN 1

p
# be nonsingular symmetric
14.7 Theorem (Hasse and Minkowski). Let M and M
# are rationally congruent i they have the
N N matrices over Q. Then M and M
#) = cp (M) for
same discriminant (i.e. to within a rational square factor) and cp(M
every odd prime p.
Before using this result to prove Theorem 14.1, we warm up with an application to 2 2
matrices. In the following, the squarefree part of n is the largest divisor of n which is
itself not divisible by any square other than 1.
88
14.8 Lemma.
alent.
Consider an integer n 1. Then the following conditions are equiv-
(i) n = a2 + b2 for some a, b Z.

(ii) n = a2 + b2 for some a, b Q.
(iii) The squarefree part of n is divisible by no primes p 3 mod 4.
(iv) (n, 1)p = 1 for every odd prime p.

(v) The matrices n0 n0 and 10 01 are rationally congruent.

s
Proof. Let p be an odd prime. If n = n0 ps where p n0 then (n, 1)p = 1
from
p
which the equivalence of (iii) and (iv) follows immediately. Both matrices in (v) have
square determinant, whereas

cp 10 01 = (1, 1)p (1, 1)p = 1;

cp n0 n0 = (1, n2 )p (n, n2 )p = (1, 1)p (n, 1)p = (n, 1)p

so the equivalence of (iv) and (v) follows from Theorem 14.7. Writing I2 = 10 01 , we see
that (v) is equivalent to
nI2 = AI2 AT = AAT
a b
a b
for some A GL2 (Q); but any such matrix has the form A = b
or
for some
a
b a
2
2
a, b Q satisfying a + b = n, so clearly (v) is equivalent to (ii).
Obviously (i) implies (ii) and the converse is well-known; in fact we show that (iii)
implies (i). Write n = m2 p1 p2 pr where m 1 and the primes p1 , p2 , . . . , pr 1 or 2
mod 4. It is well-known that pj = a2j +b2j for some aj , bj Z; now n = a2 +b2 where the integers a, b are the real and imaginary parts, respectively, of m(a1 +b1 i)(a2 +b2 i) (ar +br i)
Z[i] where i2 = 1.
Proof of Theorem 14.1. If there exists a projective plane of order n then by remarks
above, the N N identity matrix IN is rationally congruent to the N N matrix
n+1
1
1
M =
.
..
1
1
n+1
1
..
.
1
n+1
..
..
.
.
1
1
1
1
..
.
n+1
where N = n2 +n+1. Since M acts as (N + n)I = (n + 1)2 I on (1, 1, . . . , 1) and as

nI on (1, 1, . . . , 1) , we have det M = nN 1(n + 1)2 ; see Exercise #13.2. Since det M is
a rational square, the necessary condition on the discriminants is automatically satised.
89
Exactly the same reasoning gives det Mk = (n + k)nk1 for k {1, 2, . . . , N1}. From
this it follows that

(n+k, n)p (n+k+1, 1)p ,
k odd,
(det Mk , det Mk+1 )p =
(n+k+1, n)p (n+k, 1)p (n, 1)p , k even;
for example if k is odd then

(det Mk , det Mk+1 )p = (n+k)nk1 , (n+k+1)nk p = n+k, (n+k+1)n p
= (n+k, n)p (n+k, n+k+1)p = (n+k, n)p (n+k+1, 1)p
by (14.6). The case k even is similar. After cancelling duplicate factors we obtain

1)/2
(det Mk , det Mk+1 )p = (n+1, n)p (n+N, n)p (n, 1)(N
p
1kN 1
1)/2
= (n, 1)(N
p
using (14.6) and the fact that n+N = (n+1)2 is a square. Considering the remaining
factor
(1, det M)p = (1, (n+1)2 )p = 1,

we obtain
cp(M) =
1)/2
(n, 1)(N
p
(n, 1)p ,
1,
if n 1, 2 mod 4;
otherwise.
The result follows from Theorem 14.7 and Lemma 14.8.

While it is possible to simplify somewhat the latter proof by rst applying a well-chosen
congruence transformation to M, we have chosen not to avail ourselves of such tricks in
order to demonstrate the success of the straightforward approach; this we believe will best
serve the student who may in the future nd application for these techniques.
Finally, the critical Theorem 14.7 deserves some explanation, in lieu of an actual proof.
We observe that the quadratic forms x2 +y 2 and x2 y 2 are not rationally congruent because
the rst is positive denite while the second is not. This argument uses a necessary and
sucient condition for two matrices to be congruent by a real change of variable: the
number of positive eigenvalues of the associated matrix must agree, as well as the number
of negative eigenvalues. The quadratic forms x2 + y 2 and x2 + 2y 2 are congruent by a
real change of variable, but not by a rational change of variable, since the second form
represents the number 3, e.g. with (x, y) = (1, 1), whereas the rst form does not; this
was shown in our proof of Lemma 14.8 using a mod 4 argument. Similarly the quadratic
forms x2 + 2y 2 and x2 + 10y 2 are not equivalent: the rst represents the number 3 for
(x, y) = (1, 1), but the second does not represent the number 3 for any choice of x, y Q,
as can be shown by arguing either mod 3 or mod 5. More generally, some necessary
90
conditions for two rational quadratic forms to be equivalent under a rational change of
variable are that
(i) they are equivalent under a real change of variable, and
(ii) for every prime power ps , a certain necessary condition mod ps is satised.
Condition (ii) is most readily formulated as congruence over the extension Qp Q consisting of p-adic numbers, using the discriminant and the invariants cp(M) dened above.
This is analogous to the condition (i) formulated as congruence over the extension eld
Q := R Q, which is readily checked by counting positive and negative eigenvalues.
Thus the necessary conditions for equivalence of two quadratic forms over Q, are neatly
expressed as equivalence over Qp for all p {, 2, 3, 5, 7, 11, . . .}. Amazingly, as the
Hasse-Minkowski Theorem shows, these necessary conditions are also sucient! Moreover
the same conclusion holds for two rational quadratic forms that are known to be equivalent over Qp for all but possibly one p {, 2, 3, 5, 7, 11, . . . , }; this is why the prime 2
can be omitted from our statement of Theorem 14.7. And fortunately so, since congruence of rational quadratic forms over Q2 requires a slightly more delicate test than for
p {, 3, 5, 7, 11, . . .}. Incidentally, these happy facts do not hold for forms of degree 3,
where the necessary condition of equivalence over Qp for every p is not sucient to guarantee equivalence over Q; see [58, p.44].
The symbol (a, b)p as dened above for a, b Q and p {3, 5, 7, 11, . . .} is a special
case of the Hilbert symbol as dened in [58]. The reader interested in completing the
connection will note that in our case, (a, b)p takes the value 1 or 1, according as the
equation
ax2 + by 2 = z 2
does, or does not, have any nonzero solution (x, y, z) Q3p .
Exercises 14.
1. A Hadamard matrix is an n n matrix with entries 1 such that HH T = nI, i.e. the rows of H
are orthogonal. Interpret this statement as a congruence of rational quadratic forms. What are the
resulting restrictions on n? Are these necessary conditions also sucient?
2. For every m 1, let Sm be the set of all integers in {1, 2, . . . , m} which are expressible as a sum of
two squares. It is known [39] (see also [2, p.674]) that Sm has asymptotic density zero in the sense
that |Sm |/m 0 as m . Use this fact to show that asymptotically, the Bruck-Ryser Theorem
excludes about half of all positive integers as possible orders of projective planes, in the following
sense: If Em denotes the set of all integers n {1, 2, . . . , m} which fail the necessary conditions of
Theorem 14.1 (i.e. n 1 or 2 mod 4, but n is not a sum of two squares) then the ratio |Em |/m 12
as m .
15. Dierence Sets

Let G be a nite group of order N . A subset D G is a (planar) dierence set
1
if every nonidentity element g G is uniquely expressible in the form g = d1 d2 for
91
15. Dierence Sets
some d1 , d2 D. An easy counting argument shows that the parameters must satisfy
N = n2 + n + 1 where n = |D| 1. The right cosets of D in G are the subsets
Dg = {dg : d D}
for g G.
The simplest nontrivial example is given by

G = {1, x, x2 , . . . , x6 } where x7 = 1;
D = {1, x, x3 }.
In this case the right cosets of D are the subsets

D = {1, x, x3 },
Dx = {x, x2 , x4 },
..
.
Dx6 = {1, x2 , x6 }
Observe that LD = {D, Dx, Dx2 , . . . , Dx6 } is the line set of a projective plane with point
set P = G. Moreover right-multiplication by x induces a cyclic collineation which regularly
permutes both the points and the lines of this plane.
4
.......
.. ......
... .... .....
.
.
. . .
... . ...
... ..... .....
...
...
.
.
...
...
.
..
...
...........
..................... .... ......................
.
.
........
.......
.
.
.
.
......
6
...
5 ..........
....
.
..
.............
.......
.
.
.
.
.
. 2 ........ .. ........
..
.......
.
.
.
.
.
....... . .......
..
......
....... .. .......
......
.... ....
...........
... ..
. ...
... ....
.......................
.
.
.
.
.
.
.
.
.......
... .....
.
...
.
.
.. ....
.
.
.
.
.
.
.
.
.
....... ..
...
.
... .........
.
.
.
.
.
.
.
...
........
.
.......
....
..
...
.. ......
...... ...
..
... ............ ........
.... ............. .....
.
.
.
.
.
.
.
......
....... ...
. .......
..
.
.
.
.
.
.
.
.
.
.
.
..
..
.
.............................................................................................................................................................
x3
This phenomenon generalizes as follows.

15.1 Theorem. If D is a planar dierence set in G, then the right cosets Dg for
g G form the lines of a projective plane of order n = |D|1 with point set G.
Moreover the group G, acting by right-multiplication, induces a collineation group of
this plane, which regularly permutes the N = n2 +n+1 points and lines of this plane.
Proof. It is clear that the right cosets of D form the blocks of an incidence structure
(G, LD ) on which G acts by right multiplication, and that G regularly permutes the points
of this structure. Moreover there is no loss of generality in assuming that 1 D; otherwise
1
replace D by Dd for an arbitrary choice of d D (see Exercise #4). To show that this
structure is a projective plane, consider any two distinct elements of G. By transitivity of
92
the action of G, we may assume that these two elements are {1, x} where x = 1. A block
Dg contains both 1 and x i
1 = dg
and x = d g
for some d, d D
i
x = d d
and g = d
for some d, d D.
By denition, there is a unique such ordered pair d, d in D and hence any two distinct
points lie in a unique block. It follows from Exercise #6.3 that (G, LD ) is a projective
plane of order n. Since G permutes the points transitively, it must permute the lines transitively by Theorem 10.4. But also |G| = N = n2 +n+1, so in fact G must permute the
lines regularly.
The following converse of Theorem 15.1 holds:
15.2 Theorem. Let (P, L) be a projective plane of order n admitting a group G of
automorphisms permuting the points regularly. Let (P, ) be an arbitrary point-line
pair of (P, L), and let D be the set of all g G such that P g lies on . Then D is a
planar dierence set in G, and (P, L) is isomorphic to the plane (G, LD ) constructed
as above from the dierence set D.
Proof. By Theorem 10.4, G also permutes the lines of (P, L) regularly. Given any nonidentity element g G, there exists a unique line through both P and P g , and this line
1
1
can be written as h for some h G. Since both the points P h and P gh lie on , we
1
1
1
have h = d and gh = d for some d, d D; we solve to obtain g = d d .
1
Conversely suppose g = d d for some d, d D; then both P and P g lie on the line
1
d . The latter line is unique, and since the action of G on lines is regular, the element
1
d is uniquely determined; this in turn means that d = gd is uniquely determined.
Finally let (G, LD ) be the plane constructed from the planar dierence set D, and
consider the mapping : (P, L) (G, LD ) dened by acting on points as P g g and on
lines as h Dh. Since
P g h
P gh
i
gh
g Dh,
the map preserves incidence. Thus is an isomorphism of projective planes.

Let (P, L) be a nite projective plane with a collineation group acting regularly on
P, and so also on L. Fix P P and L. Write
D = {d G : P d } = {d G : d
P }.
93
15. Dierence Sets

1
Consider the map [1] : G G, g g ; then

D[1] = {d G : d P }.
This shows that the while the subset D G is a dierence set dening the original plane
(G, LD )
= (P, L), the subset D[1] G is also a dierence set dening the dual plane
(G, LD[1] )
= (L, P). Since every planar dierence set arises in this way, we have
15.3 Theorem. Let G be a nite group and let D G be any subset. Then D is a
planar dierence set in G, i the subset D[1] G is a planar dierence set. In this
case the two resulting planes (G, LD ) and (G, LD[1] ) are dual to one another.
Of course if G is abelian then the inverse map [1] is an automorphism of G mapping
D to D[1] , so necessarily the resulting planes (G, LD ) and (G, LD[1] ) are isomorphic in
this case.
While a planar dierence set D in a group G represents a line in (G, LD ), and D[1]
represents a line in the dual plane (G, LD[1] ), we may ask whether the point set D[1] G
represents anything reasonable in the original plane (G, LD ). This question has an interesting answer in the abelian case (see Exercise #5 for a counterexample in the nonabelian
case):
15.4 Theorem. Let D be a planar dierence set in a nite abelian group G. Then
the point set D[1] G is an oval.
Proof. We have |D[1] | = |D| = n+1 where the plane (G, LD ) has order n, so it suces to
show that no three points of D[1] are collinear. Suppose x, y, z D[1] Dg are distinct
1
1
1
1
1
1
and observe that x , y , z , xg , yg , zg D. Since
xyg
1 1
= (xg )(y )
1 1
= (yg )(x ) ,
1
it follows from the denition of a dierence set that xyg = 1. But a similar argument
1
shows that xzg = 1, which forces y = z, a contradiction.
We now present the standard construction of planar dierence sets in cyclic groups
of order q 2 +q+1 where q is a prime power. These are equivalent to regular collineation
groups of nite classical planes. We rst illustrate the method by constructing once again a
planar dierence set in the cyclic group of order 7, which arises from a regular collineation
94
group on the plane of order 2. Take E = F8 = F2 [] where 3 = + 1. The nonzero

elements of E are
1,
,
2 ,
3 = +1,
4 = 2 +,
5 = 3 + 2 = 2 ++1,
6 = 3 + 2 + = 2 +1;
note that 7 = 3 + = 1. The seven 2-dimensional subspaces of E over F2 are the lines
of the plane
4
.....
........
... .. ...
... .. ...
.. .... .....
.
.
. ..
...
...
...
....
...
...
..
..
...
.
.
.
. ..................................... .....
.
.
...... ..
........
.
.
.......
.
.
.....
.
.
......
.
.
.
6
5 .....
...
....
......
.
.
.
............
. .......
.
.
.
.
.
.
.
.
.
.
.
.
.......
.....
..
.
...
.
.
.
.
.
.
.
.
.
.
.
.
....... . ......
..
......
................ 2
... ..
... ....
...
... ...
...........
. ..
...... .... .............
... ....
.
.
... .....
.
.
.
.
.
.
.
.......
..
...
.
.
.
.
.
.
.
...
.
.
.
.
.
.
.
....... ..
.
............
.
.
.
...
.
.
.
.
.
.....
.
.....
....
... ............. .....
... ............ ......
.
.
.
.
.
.
....... ...
.....
.
. ........
.
.
.
.
.
.
.
.
....... ...
.
.
.......
..
.........
.
.......
.
...........................................................................................................................................................

1

3
The map E E, x x is F2 -linear; it regularly permutes the seven points, as well as

the seven lines of the plane shown above. All of this generalizes:
15.5 Theorem. Every nite classical plane admits a regular cyclic collineation
group.
Proof. Let q be a prime power, and regard E = Fq3 as a 3-dimensional vector space over
F = Fq . The points and lines of P2 (F ) may be regarded as the 1- and 2-dimensional
subspaces of this vector space. The group E is cyclic; let be a generator. The map
E E, x x is F -linear and it cyclically permutes the nonzero vectors of E. Let
T P L3 (F ) be the corresponding plane collineation, so that T permutes the points of
2
P2 (F ) cyclically and transitively. Since q +q+1 is a primitive (q 1)-st root of unity
2
in E , it is a generator of F and so T q +q+1 is the identity collineation, mapping
2
points as v q +q+1 v = x. Now the points of P2 (F ), which are the 1-spaces
i x for i = 0, 1, 2, . . . , q 2 +q, are cyclically and regularly permuted by the subgroup
T P L2 (F ). The lines must also be permuted regularly, by Theorem 10.4.
This means that every nite classical projective plane has an incidence matrix which
is circulant, i.e. the rst row of the matrix can be cycled (either left or right) to give
the remaining rows of the matrix. In the case of G = {1, x, x2 , . . . , x6 } the dierence set
15. Dierence Sets
95
D = {1, x, x3 } gives the vector (1, 1, 0, 1, 0, 0, 0) as the rst row of an incidence matrix (the
characteristic vector of the subset D G) and the remaining rows of the incidence matrix
are obtained by cycling the rst row, thus: (0, 1, 1, 0, 1, 0, 0), (0, 0, 1, 1, 0, 1, 0), etc.
There also exist nonabelian groups admitting planar dierence sets; see Exercises #2,3.
However, no known dierence sets give rise to non-Desarguesian planes. Very little is known
about the possibilities for other planar dierence sets in non-abelian groups.
The dening properties of a dierence set are neatly encoded as identities in the group
algebra. Let G be a nite group and consider the algebra of G over Q:
QG =
$
xG

ax x : ax Q .
(See Appendix A3 for a brief summary of the relevant notions used here.) For each integer
d, dene a map [d] : G G, x xd . This extends to a well-dened map QG QG, also
denoted by [d]:
[d]

ax x
=
ax xd .
xG
xG
The latter map is an algebra homomorphism whenever G is abelian; it is an isomorphism

whenever G is abelian of order not divisible by d.
15.6 Proposition. Let G be a nite group of order n2 +n+1 where n 2, and let
$
$
D G. Dene , QG by = xG x and = dD d. Then D is a planar
dierence set in G i
[1] = n + .
In this case the relation = = (n+1) also holds.
Proof. This is an immediate consequence of the denitions and the fact that = a
for all QG where a Q is the sum of the coecients in .
The next result assumes that D is a dierence set in an abelian group G of order
n +n+1.
2
15.7 Hall Multiplier Theorem. Let D be a planar dierence set in an abelian

group G of order n2 +n+1, and consider any prime p dividing n. Then the map
[p] : G G, x xp permutes the cosets Dg for g G and so is a collineation of the
projective plane (G, LD ) where LD is the set of cosets of D in G. Moreover one of
these cosets is xed by [p].
96
$
Proof. Let = xD x, so that [1] = n + . We see from the Multinomial Theorem
that p = [p] + p for some ZG, and so
[p] [1] = p [1] p [1] = p1 [1] + p1 = (n + ) p1 + p1 = + p2
for some 1 , 2 ZG. Since all coecients on the left side are non-negative, we see that
all integer coecients appearing in 2 must be non-negative. Multiplying both sides by
yields (n + 1)2 = (n2 + n + 1) + p2 , so that 2 = (n/p). Applying [1] yields also
[1]
2 = (n/p). Now

[1]

[p]
= [p] [1] [p] = [1] [1]
[p] [1] [p] [1]
= (n + )(n + )[p] = (n + )2 = n2 + 2n + 2 .
On the other hand

[1]
[1]
[p] [1] [p] [1]
= ( + p2 )( + p2 )
[1]
+ p(2 + 2
[1]
+ 2n + 2 .
= p2 2 2
= p2 2 2
[1]
) + 2
Equating these two expressions yields

[1]
p2 2 2
= n2 .
$
The expansion 2 =
xG ax x clearly has exactly one nonzero coecient ag since the
coecients are non-negative integers and the right side has a single term n2 . It follows
that 2 = (n/p)g for some g G, and so
(15.8)
[p] [1] = + ng = [1] g.
We claim that
(15.9)
[p] = g.
This will follow from (15.8) if [1] is invertible in QG. Let

=
then

[1] = (n + ) n1
1
n
1
n(n+1)2
n(n+1)2
= 1
QG;
1
(n+1)2
Thus [1] QG is invertible and so (15.9) follows.
+ n1
2
1
n(n+1)2
= 1.
97
15. Dierence Sets
The cosets of D are permuted by [p] via Dh Dghp since

(h)[p] = [p] hp = ghp;
thus [p] is a collineation of (G, LD ) where LD is the set of cosets of D in G. Note that [p]
xes the point 1 G, so by Corollary 10.3 it xes some line Dh LD .
15.10 Example: Dierence Sets in Some Small Groups. Corresponding to the
classical plane of order 3 there is a planar dierence set in the cyclic group G = {1, x, x2 , . . . ,
x12 } of order 13. In principle one may nd a primitive element (an element of order
133 1 = 2196) among the nonzero elements of F2197 and proceed as in the example
preceding Theorem 15.5. However, the Hall Multiplier Theorem aords us the following
shortcut. The orbits of = [3] Aut G on G are given by
{1},
{x, x3 , x9 },
{x2 , x5 , x6 },
{x4 , x10 , x12 },
{x7 , x8 , x11 }.
By Theorem 15.7 there exists a planar dierence set D which is a union of orbits of .
Since |D| = 4 we must include {1} and one of the orbits of size 3. Any of these will do;
take for example D = {1, x, x3 , x9 } which we easily verify to be a planar dierence set
in G.
The planar dierence set in the cyclic group of order 7 may be found by similarly
enumerating the orbits of [2] on {1, x, x2 , . . . , x6 } where x7 = 1: these orbits are
{1},
{x, x2 , x4 },
{x3 , x5 , x6 }
and both of the orbits of size 3 give planar dierence sets.

Given the diculty of proving the main questions in the theory of nite projective
planes, particularly the question of existence of planes of non-prime power order, one
might reasonably hope to make at least some progress in the special case where a regular
collineation group exists. Unfortunately to date, such progress has been quite limited.
The following result, which is representative of such progress, is included because its proof
is short and elegant. It shows that there do not exist projective planes of certain orders
n {6, 10, 12, 14, 15, 18, 21, . . .} having regular collineation groups.
15.11 Theorem (Wilbrink [66]). Let D be a planar dierence set in an abelian
group G of order n2 +n+1, and suppose the prime p divides n exactly once. We may
assume D is chosen such that D[p] = D by the Hall Multiplier Theorem.
$
$
d
and
=
(a) In the group algebra Fp G, the elements =
dD
xG x
satisfy
p1 + ( [1] )p1 = 1 + .
(b) If p {2, 3} then n = p.
98
Proof. Recall that every Fp G satises = a where a Fp is the sum of the

coecients in . The following identities clearly hold in characteristic p:
[1] = ;
2 = ;
= ;
[1] = .
From the multinomial expansion of p we also have

(15.12)
p = [p] = .

Since p |G| the algebra Fp G is semisimple by Maschkes Theorem A3.10, and so by
Theorem A3.16,
(15.13)
Every ideal I Fp G is a principal ideal generated by a unique idempotent

e Fp G; we write I = (e).
(15.14)
Given ideals I1 , I2 Fp G let Ii = (ei ) where ei is the unique idempotent

generating Ii ; then
I1 + I2 = (e1 +e2 e1 e2 );
I1 I2 = (e1 e2 )
where e1 +e2 e1 e2 and e1 e2 are the unique idempotents generating the

ideals I1 +I2 and I1 I2 respectively.
In particular consider the ideals I1 = () and I2 = ( [1] ). Multiplying (15.12) by
p2 gives ( p1 )2 = p1 so p1 I1 is its unique idempotent generator. Similarly
( [1] )p1 I2 is its unique idempotent generator. Therefore the unique idempotent
generating I1 I2 is
p1 ( [1] )p1 = p1 = .
Thus I1 I2 is 1-dimensional over Fp ; also I1 + I2 has as its unique idempotent generator
the element
p1 + ( [1] )p1 .

By Theorem 13.2 the dimension ofI1 is 12 n2 +n+2 . Since D[1] is also a planar dierence
set, the dimension of I2 is also 12 n2 +n+2 and so
dim(I1 + I2 ) = dim I1 + dim I2 dim(I1 I2 ) = n2 +n+1,
i.e. I1 + I2 = Fp G. Since 1 Fp G is the unique idempotent generating Fp G we have
p1 + ( [1] )p1 = 1. This proves (a).
If p = 2 then
$
+ [1] = 1 + =
x
x=1
where the right side has n2 +n nonzero coecients, whereas the left side has at most
2(n+1) nonzero coecients. This gives n(n+1) 2(n+1) and so n 2.
If p = 3 then
16. Generalized Incidence Matrices
(15.15)
2 + ( [1] )2 = 1 + = 2 +
$
x=1
99
x.
Writing D = {d0 , d1 , . . . , dn } and expanding

(15.16)
2 =
d2i + 2
0in
di dj ,
0i<jn

1
1
terms di dj are distinct since di dj = dk d implies di dk = d dj . Similarly
the n+1
2
distinct
from the d2i terms, so (15.16) has exactly n+1 terms with
the di dj terms are

n+1
terms with coecient 2. The same holds for the expansion of
coecient 1 and
2

[1] 2
terms with coecient 1.
) , so the left side of (15.15) has at most 2(n+1) + n+1
(
2
Thus

n2 +n 2(n+1) + n+1
2
which yields n 4.
Exercises 15.
1. Construct planar dierence sets in the cyclic groups of orders 21 and 31 by the method of Example 15.10.
2. Let G be the nonabelian group of order 21 generated by elements , subject to the dening relations
3 = 7 = 1; = 2 . Show that D = {, 2 , 4 , , 2 } is a planar dierence set in G. Explain why
the resulting projective plane (G, LD ) of order 4 must be classical (and therefore isomorphic to the
plane constructed from a planar dierence set in the cyclic group of order 21 as in Exercise #1).
3. The following construction generalizes that of Exercise #2. Consider the eld F = Fq where q 1
mod 3 and let E = Fq3 . Let E be a primitive element, i.e. the multiplicative order of is q3 1.
Dene F -linear transformations , : E E by : x xq ; : x 3(q1) x. Show that , is a
nonabelian group of order q2 +q+1 which regularly permutes the 1-dimensional F -subspaces of E.
4. Let D be a planar dierence set in a nite group G, and let a, b G. Show that the subset aDb =
{adb : d D} is also a planar dierence set. Show also that the plane (G, LaDb ) is isomorphic to
(G, LD ).
5. Show by example that the conclusion of Theorem 15.4 does not hold in the nonabelian case.
Hint. Consider the example D G of Exercise #2, and replace D by an appropriately chosen right
coset Dg.

We present a generalization of the concept of a planar dierence set, which at the same
time generalizes the notion of the incidence matrix for a projective plane. This notion was
rst introduced by Hughes [29], [30].
Let G be an automorphism group of a nite projective plane (P, L) of order n. Recall
that the number of point orbits (call this number w) also equals the number of line orbits.
Let P1 , P2 , . . . , Pw be representatives of the point orbits; and 1 , 2 , . . . , w representatives
100
of the line orbits. Denote the point stabilizers i = GPi and the line stabilizers j = Gj .
Set Aij = {g G : Pig j }. The desired relations are most easily stated in the group
algebra of G over the eld of rational numbers. We consider the elements , i , j , ij
QG dened by

=
G ; i =
i ; j =
j ; ij =
Aij
$
for i, j {1, 2, . . . , w}. For an arbitrary element = gG ag g QG dene the conjugate

1
=
ag g QG
gG
so that () = for all , QG. (In Section 15 we denoted [1] in place of .)

Also denote

|| =
ag Q
gG
so that
|| = |||| ;
| | = || ||
for all , QG. Thus the map QG Q, || is an algebra homomorphism (usually

called the augmentation map). Moreover
|| = |G|,
|i | = |i |,
|j | = |j |,
|ij | = |Aij |.
Consider the w w matrices

D = diag(1 , 2 , . . . , w );

A = ij 1i,jw ;
D = diag(1 , 2 , . . . , w );

A = ji 1i,jw
with entries in QG (note that A is the conjugate transpose of A) and the matrices
|D | = diag(|1 |, |2 |, . . . , |w |); |D | = diag(|1 |, |2 |, . . . , |w |);

|A| = |ij | 1i,jw ; J = 1 1i,jw
with entries in Z. In particular J is the w w matrix of 1s.
16.1 Theorem (Hughes [29], [30]). Given G Aut(P, L) where (P, L) is a projective plane of order n, dene the matrices D , D , A, |D |, |D |, |A| as above. Then
n2 +n+1
;
|G|
1
1
(ii) J |D | |A| = |A||D | J = (n+1)J ;
1
(i) tr(|D | ) = tr(|D | ) =

1
(iii) A |D | A = J + nD and |A ||D | |A| = |G|J + n|D | ; and

1
(iv) A|D | A = J + nD and |A||D | |A | = |G|J + n|D | .
101
Before proving these relations let us look at some examples.

16.2 Example: The Trivial Group. If G is the trivial subgroup of Aut(P, L) then
D = D = Iw is the identity matrix of order w = n2 +n+1. Moreover A is the usual
identity matrix of (P, L). In this case (i) asserts that n2 +n+1 = n2 +n+1; (ii) says that
AJ = J A = (n+1)J ; and (iii),(iv) assert that ATA = AAT = nI + J .
16.3 Example: Regular Collineation Group. If G is a regular collineation group
of a plane (P, L) then w = 1; D = D = [1] and A = [] where is arises from a planar
dierence set, as in Section 15. In this case Theorem 16.1 gives 1 = 1; || = n+1; and
= = n + .
16.4 Example. A more typical situation is illustrated below; here (P, L) is the projective
plane of order 2, and the collineation group G = , : 2 = 2 = 1, = 3 is dihedral
of order 8. The nonidentity elements of G are explicitly listed:
3
.......
.. ......
... .... .....
.
.
. . .
... .. .....
...
... ....
...
...
.
...
.
..
.
... ............................ .....
.
.
....... ..
. .........
.
.
.........
.
.
.......
.
.
......
.
.
...
......
....
....
.
.
.
.
. ........
........
.
.
.
.
.
..... ........
.
... .........
.
.
.
.
.
.
.
.
.
.
.
....... . .......
...
.....
.
.
.
.
....... . ......
..
.. ....
.............
... ...
.
... ...
... ....
....... .... .............
.
.
.
.
.
.
.
.......
... .....
.
..
. ..
.
.
.
.
.
.
.
.
.
.
.
.
....... ..
... ......
...
.
.
.
.
.
.
.
.
.
.
.
........
...
.
.......
....
..
...
... .....
...... ...
..
... ............. .....
... ............ ........
.
.
.
.
.
.
.
....... ...
.
......
.
.
.
...........
.
.
.
.
.
.
... .
...
..
............................................................................................................................................................
7 d
= (1275)(46)(afde)(bg),
= (12)(57)(bg)(ef),
= 3 = (25)(46)(ae)(df),
= (17)(25)(ad)(ef),
6
3 = (1572)(46)(aedf)(bg),
2 = 2 = (15)(27)(ad)(bg),
5 e
f c
3 = = (17)(46)(af)(de).
b
4
2
The orbits of G on points and lines are given by

1G = {1, 2, 4},
3G = {3},
4G = {4, 6},
Choosing 1, 3, 4 and
1+
D =
0
0
bG = {b, g},
cG = {c}.
a, b, c as orbit representatives, we obtain
0
0
1+
0
2
,
D =
0
0
1+ + + 3
0
0
0 1+ 2 + + 2
aG = {a, d, e, f},
2
|D | = 0
0
1++ +
0
A=
2
1+ + + 2
0
0,
4
2 0 0
|D | = 0 4 0 ;
0 0 8
4
+ + 3 + 2 0
,
|A| = 0
0
4
0
8
0
4 0
8 8.
0 8
0
0,
102
We verify the conclusions of Theorem 16.1 in this example as follows:

1
0 0
0 0
2
2
tr 0 18 0 = tr 0 14 0 = 78 ;
0 0 14
0 0 18
1
0 0
4 4 0
2 2 0
2
0 1 0 0 8 8 = 0 1 1 has column sums 3;
8
0 0 14
4 0 8
1 0 2

1
0
0
2
1
0
4 4 0
2
0 8 8 0 1 0 = 0 2 1 has row sums 3;
4
0 0 18
2 0 1
4 0 8
1+3 +
3
+ +3
+2
0
1++
1+2 + 1
+2
2
4
0
4
+ +3
+2
1+2 +
+2
4
0
4
4
8
0
0
1
8
1++
+
0 0
0
1
4
1
0
2
8 0
8
0
1
2
0
4
8
0
0
1
4
1
0
2
8 0
8
0
0
1
4
+2(1+ )
1+2 +
+2
+2(1+2
+ +3 )
0
4 0
0
4 8
1
0 8
8

4
12
0 = 8
8
8
0
1
4
1+3 +
+3
+ +3
0
+2
0
1
8
+ +3
+2
0
4 0
0
4 8
1
0 8
8
+2(1+2
+ +2 )
+2(1+ )

4
12
0 = 8
8
8
8
24
8
;
3
8
8 ;
24
1+2 +
+2
8
16
8
8
8 .
16
Proof of Theorem 16.1. Let Tj be a set of right coset representatives for j in G; then
|Tj | = [G : j ] and every line can be uniquely written in the form tj for some j
{1, 2, . . . , w} and t Tj . Counting lines gives
n2 +n+1 =

1jw
|Tj | =
|j | |G|.
1jw
Dividing both sides by |G| gives one of the relations (i), and the other follows by a dual
argument.
Let us count in two dierent ways the number of elements in the set
Mij = {(g, h) G G : Pig hj }.
103

1
Since Pig hj i Pigh j i gh Aij , and every choice of gh Aij gives rise to
|G| pairs (g, h) Mij , we have |Mij | = |G||Aij |. On the other hand the number of h G
1
such that Pig hj cannot depend on the choice of g G, since substituting h = hg for
xed g yields

|{h G : Pig hj } = |{h G : Pi hj }|
1
= |j ||{t Tj : Pi tj }|
= |j ||{ G
j : Pi }|
since each h G lies in a unique coset j t with t Tj . Thus

|Mij | = |G||j ||{ G
j : Pi }|.
Equating our two expressions for |Mij | yields
|{ G
j : Pi }| =
|Aij |
|ij |
=
.
|j |
|j |
Since the line set L is partitioned into orbits G

j for j = 1, 2, . . . , w, we have
n + 1 = |{ L : Pi }| =
|ij |
.
|j |
1jw
1
Thus the matrix |A||D | has row sums equal to n + 1, whence |A||D | J = (n+1)J .
The other relation in (ii) follows by a dual argument.
Now x i, k {1, 2, . . . , w} and consider the set
Nik = {(g, h, ) G G L : Pig , Pkh }.
$ 1
We evaluate in two dierent ways
gh QG where the summation extends over all
triples (g, h, ) Nik . Each line L is uniquely representable in the form = tj
for some j {1, 2, . . . , w} and t Tj . For this choice of line the corresponding triples
1
(g, h, tj ) arise from pairs (g, h) where g Aij t and h Akj t. The sum of gh over such
pairs (g, h) is ij kj ; then varying t Tj and j {1, 2, . . . , w} gives

gh
(g,h,)Nik
|Tj |ij kj = |G|
1jw
|j | ij kj .
1jw
If i = k then every pair (g, h) G G gives distinct points Pig , Pjh which determine
a unique line L; in this case

gh
(g,h,)Nik

g,hG
gh
= |G|.
104
If i = k then whenever the pair (g, h) G G satises Pig = Pih , there are an additional n
1
lines (i.e. n+1 instead of 1) through both Pig and Pih . Since Pig = Pih i (g, hg ) Gi ,
there are |G||i | such pairs (g, h), which together contribute a term n|G|i to our sum.
Combining with the previous case gives

gh

= |G| + nik |G|i =
(g,h,)Nik
Equating our two expressions for
gh
|G| + n|G|i , if i = k;
|G|,
otherwise.
yields
|j | ij kj = + nik i .
1jw
This veries the rst relation in (iii), and the second follows by applying the augmentation
map QG Q, ||. The proof of (iv) is similar.
An early application of Hughes Theorem 16.1 is the following:
16.5 Theorem (Hughes [29], [30]). If (P, L) is a projective plane of order n 2
mod 4 with n > 2, then its automorphism group has odd order.
Thus for example, long before it was determined that a projective plane of order 10 does
not exist, it was known that no such plane could have a collineation of order 2. We explain
the general idea of the proof of Theorem 16.5: Suppose (P, L) is a plane of order n 2
mod 4 with n > 2, and suppose there exists Aut(P, L) of order 2. By Theorem 10.7,
must be an elation. This determines the structure of the matrices A, D , etc. for the
group G = of order 2. The second relation of Theorem 16.1 shows that the integer
1
matrices |D | and 2J +n|D | are rationally congruent. But these matrices are explicitly
known in terms of n, and the method of Section 14 yields the required contradiction.
General results comparable to Theorem 16.5 are found only rarely. But Theorem 16.1
is typically applied in many concrete situations as follows. Given a positive integer n 2
which satises the necessary Bruck-Ryser condition for the possible existence of a projective
plane of order n, it is typically nevertheless a very dicult task to decide whether in fact
such a plane exists. One may therefore choose ones favourite smallish abstract group,
such as S3 or A4 or S4 , and then ask whether there exists a projective plane of order n
admitting G as a collineation group. (If n 2 mod 4 with n > 2, then Theorem 16.5 tells
us that |G| had better be odd.) Geometric arguments along the lines of Section 10 give
some information on the possible types of collineations in G (i.e. their xed substructures
as classied by Theorem 10.5). One thereby enumerates possible rational matrices |A|,
|D |, |D | satisfying the necessary conditions of Theorem 16.1. One then tries to lift
these integer matrices A, D , D in the group algebra QG still satisfying the relations
17. Blocking Sets
105
required by Theorem 16.1. In some cases one obtains a contradiction; in other cases one
may nd new planes. We have in fact found some new planes of order 25 in this way, the
Wyoming planes [48].
Exercises 16.
1. Consider the projective plane of order 2, with points and lines labelled as above, and consider the
collineation group G = , where = (346)(257)(aeg)(bdf), = (24)(56)(bc)(df ). Enumerate the
point and line orbits, choosing as orbit representatives the least (i.e. alphabetically rst) member in
each case, as we did in Example 16.4.
2. Let G be a collineation group of a nite projective plane of order n with w point orbits PiG for
w1 with all
i = 1, 2, . . . , w; and so also w line orbits G
j for j = 1, 2, . . . , w. Consider the product of n
the orbit sizes:

|PiG |
|G
nw1
j | .
i
Show that this integer is a perfect square.
Examples. Immediately following the statement of Theorem 10.4 we gave an example of a group of
order 4 acting on the projective plane of order 2, with 4 point orbits (size 1,1,1,4) and 4 line orbits
(size 1,2,2,2). In this case the product becomes 23 (1114)(1222) = 28 , a perfect square. We leave
it to the student to check that the conclusion is satised in a couple ready cases: when G is transitive
(w = 1) and when G = 1.
Hint.
Use matrix relations obtained from Theorem 16.1 and take determinants.
17. Blocking Sets

A blocking set in a projective plane (P, L) is a set B P of points such that every line
meets some point of B. How small can |B| be and still block all the lines in this way?
Since every superset of a blocking set is also a blocking set, we are primarily interested
in minimal blocking sets, i.e. blocking sets which do not properly contain any other
blocking set.
If (P, L) is a projective plane of order n, it is not hard to see that every blocking set
has size |B| n+1; for if P P is a point outside B then each of the n+1 lines through
P must contain some point of B. We can meet this trivial lower bound by choosing B to
be the point set of some line. We are more interested in a nontrivial blocking set, i.e.
one which does not contain [] for any line L.
17.1 Proposition (Bruen [10]). If (P, L) is a plane of order n then every nontrivial
blocking set has size |B| n + n+1. Equality holds i B is the point set of a Baer
subplane, in which case n is a square.
Proof. Although the result is due to Bruen, we give a simplied proof due to Blokhuis. If
n is a square and (P, L) has a Baer subplane with point set B, then clearly B is a nontrivial
blocking set of size |B| = n + n + 1.
106
Suppose B is a nontrivial blocking set of size |B| = n + k. We must show that
First observe that

k n + 1, with equality i B is the point set of a Baer subplane.

every line L meets B in at most k points. (For if [] B > k then choose P with
P
/ B. One line through P meets B in at least k+1 points, and each of the remaining n
lines through P meets B in at least 1 point, so that |B| k+1+n, a contradiction.)
For each i {1, 2, . . . , k}, let mi be the number of lines L meeting B in exactly i
points. The total number of lines in the plane is
(17.2)
mi = |L| = n2 +n+1.
1ik
Counting in two dierent ways, the number of incident point-line pairs (P, ) B L is
(17.3)
imi = (n + k)(n + 1).
1ik
Counting in two dierent ways, the number of pairs of points (P, Q) with P B and Q
/B
is
(17.4)
i(n + 1 i)mi = (n + k)(n2 + 1 k);
1ik
this is because there are mi lines having i choices of P B and n+1i choices of
Q ........B. Taking an appropriate linear combination of these relations (namely (17.4), plus
kn times (17.3), minus k times (17.2)) gives
(17.5)

(i 1)(k i)mi = n (k1)2 n .
1ik
Since the left side of (17.5) is non-negative, we obtain

k
n + 1.
If equality holds then from (17.5) we see that every line L meets B in either 1 or
k = n + 1 points. Let L0 be the set of lines L meeting B in n+1 points. Let P B,

and suppose P lies on r lines of L0 and n + 1 r lines of L ........ L0 . Counting intersections
of lines through P with points of B gives
1+r n=n+k =n+ n+1
which yields r = n + 1. Thus (B, L0 ) is a 2-(n+ n+1, n+1, 1) design embedded in
(P, L), i.e. a subplane of order n; see Exercise #6.3. The result follows.
107
17. Blocking Sets
If n is not a square, how small can a nontrivial blocking set be? It is not hard to
construct a nontrivial blocking set with as few as 3n points: take three lines a, b, c with
no point in common, and remove the three vertices A, B, C of the resulting triangle, as
...
.
... ....
shown:
... ...
....
....A
..
... ...
.. ....
...
...
...
...
.
.
...
.
...
...
.
.
...
.
.
...
.
.
.
...
.
.
.
...
.
.
.
...
.
.
.
...
.
.
.
...
.
.
..
.
.
...................................................................................................
.
...
.
.
...
.
.
.
..
This idea can be improved in the case of classical planes of odd order n = q, as follows.
Suppose a, b, c are the lines X = 0, Y = 0 and Z = 0 respectively. If a line does not
meet any of the three vertices A, B, C of the triangle (here A=(1, 0, 0), B=(0, 1, 0),
C=(0, 0, 1)) then meets a, b, c in points
(0, 1, ),
(, 0, 1),
(1, , 0)
respectively, as shown, where = 0. Since these three points are collinear we must have
0 = det
1
1
0
1 = 1 .
0
......
..
..
......
......
...... ....
......
........
.....
.
.
.
......
.
.
..
... .........
......
... .......
......
.....
......
..
.................
..
..
...
..... ............
..
......
......
.. ..........
......
......
..........
......
.......
......
.
.
.
.
.
......
... ..
.
.
.
.
......
.
..
.
.
.
.
......
.
.
.
...
.
......
.
.
.
.
.
..
......
.
.
.
.
.
.
.
......
.
..
.
.
.
.
.
......
.
.
..
.
.
.
.
......
.
.
.
.
..
.
.
.
.
.
.
.
.
..........................................................................................................................................................................................................................
......
..
.
.
.
.
.
.
.
......
.
...
.
.
.
.
......
.
.
..
......
..

A
b
(1, , 0)
(, 0, 1)
c
(0, 1, )
In particular either all three or just one of the values , , must be squares in Fq . If we
let

B = {A, B, C} (0, 1, ), (, 0, 1), (1, , 0) : Fq is a nonzero square

then B is a blocking set with |B| = 3 + 3 q1
= 32 (q + 1). This construction is due to
2
di Paola [50], who guessed that this is best possible in the case of P2 (Fp ), p prime. This
guess was veried by Blokhuis [5], using an ingenious polynomial argument:
17.6 Theorem (Blokhuis [5]).
3
2 (p + 1).
A nontrivial blocking set in P2 (Fp ) has size at least
108

Proof. Let B be a blocking set in P2 (Fp ) of size |B| = p + k + 1 where k 12 p + 1);

we must verify the equality k = 12 p + 1). There exists a line meeting B in just one point
(since |B| < 2(p + 1)). Choose coordinates (X, Y, Z) for our plane, we may assume that
the line at innity Z = 0 contains a unique point (1, 0, 0) of B, so that
B = {(1, 0, 0)} {(ai , bi , 1) : i = 1, 2, . . . , p+k}.
Every line not passing through (1, 0, 0) has the form X +uY +tZ = 0 for some t, u Fp ;
and each of these lines must pass through (ai , bi , 1) for some i, so the polynomial

F (t, u) =
(t + ai + bi u) Fp [t, u]
1ip+k
vanishes for all values of t, u in F2p . Thus

F (t, u) = (tp t)G(t, u) + (up u)H(t, u)
for some G(t, u), H(t, u) Fp [t, u], each of which has degree k. Taking the (p + k)homogeneous part of each side, we obtain

F0 (t, u) =
(t + bi u) = tp G0 (t, u) + up H0 (t, u)
1ip+k
where G0 (t, u) and H0 (t, u) are the k-homogeneous parts of G(t, u) and H(t, u) respectively.
Evaluating at u = 1 gives

f=
(t + bi ) = tp g + h
1ip+k

where f(t) = F0 (t, 1), g(t) = G0 (t, 1), h(t) = H0 (t, 1). For each i we have (t + bi ) tp t,
so that t + bi divides tp g + h (tp t)g = tg + h. Write f(t) = s(t)r(t) where s(t) is
the product of the
distinct linear factors of f(t), and r(t) contains all the repeated linear
factors. Thus s tg + h and

r f (t) = tp g + h .
Therefore f = sr divides
(tg + h)(tp g + h )
and so f = tp g + h also divides

(tg + h)(tp g + h )g (tg + h)(tp g + h)g = (tg + h)(gh g h).
But deg f(t) = p+k; deg(tg + h) = k+1; and deg(gh g h) 2k2 due to cancellation
of the t2k1 terms in gh and in g h. One possible conclusion from this is that p + k
109
18. Curves

k + 1 + 2k 2 which yields k 12 p + 1 as required. The only other possibility is
gh = g h, which yields
d h(t)
gh g h
=
= 0.
dt g(t)
g2
Since both g(t) and h(t) have degree less than p, this implies that h(t)/g(t) = c Fp , a
constant. Thus h = cg and
f(t) = (tp + c)g(t) = (t + c)p g(t).
So p of the bi s are equal to c. This means that B contains all points of the line Y = c, so
that B is trivial.
Exercises 17.
1. Show that the projective plane of order two has no nontrivial blocking set.
18. Curves
Let F be a eld and let F F be the algebraic closure of F . Let f(X, Y, Z) F [X, Y, Z]
be a nonzero homogeneous polynomial of degree d 1, so that
f(tX, tY, tZ) = td f(X, Y, Z)
for all t F . The set of points (x, y, z) in P2 (F ) such that f(x, y, z) = 0 is a (projective
algebraic) curve C of degree d. Those points of the form (x, y, z) with x, y, z F
are called the F -rational points of the curve. Curves of degree 1 or 2 are of course
lines and conics. We say that C is irreducible if f(X, Y, Z) is irreducible over F . We
will assume throughout this section that C is in fact absolutely irreducible, this being
the stronger condition that f(X, Y, Z) is irreducible over F . In particular, f(X, Y, Z) is
uniquely determined by the point set C up to nonzero scalar multiple.
A point (x, y, z) C is a singular point of C if the three partial derivatives fx , fy , fz
all vanish at (x, y, z). The curve C is smooth if none of its points are singular. Many of
the basic properties of C (algebraic, geometric and combinatorial) depend on the genus of
C. This parameter of the curve C, denoted g, is a non-negative integer whose algebraic
denition is given later in this section; but for now we observe that in familiar cases (say
F {Q, R, C} and F = C) the point set C is a complex curve and therefore a real surface.
In the smooth case it is homeomorphic to a sphere with g handles, i.e. one of the surfaces
in the list:
etc.
g=0
(sphere)
g=1
(torus)
g=2
g=3
...
110
Here we make use of the fact that a smooth complex projective algebraic curve forms a
compact connected orientable surface (2-manifold). Each such surface is determined up to
homeomorphism by the parameter g {0, 1, 2, . . .} and lies in the list above. Note that
these are surfaces, not solids; for example when g = 0 we have not a solid ball but rather
just the sphere which bounds it. Also note that non-orientable surfaces, such as the real
projective plane or the Klein bottle, do not arise; the complex analytic structure forces
the surface to be orientable. The Euler characteristic of a surface is dened by
() = V E + F
where V , E and F are the number of vertices, edges and faces in a decomposition of
into polygonal cells (triangles, squares, etc.). Although the values of V , E and F depend
on the decomposition chosen, the value of () is well-dened. Here we illustrate the fact
that the 2-sphere S 2 and the torus T 2 have Euler characteristic 2 and 0 respectively:
(S 2 ) = V E + F
= 8 12 + 6
=2
(S 2 ) = V E + F
=46+4
=2
(T 2 ) = V E + F
= 16 32 + 16
=0
The genus g may be determined from the Euler characteristic in the case of a sphere with
g handles, by the relation
() = 2 2g.
The genus of a curve C of degree d 1 satises

(18.1)
d1
2

=
1
2

d1 d2 .
and equality holds whenever C is smooth. Thus for example, smooth curves of degree 1
and 2 (that is, lines and conics) have genus 0. Note that a curve of degree zero is just a
line
P1 (C) = C {}
where C is the usual so-called complex plane (which is actually just a complex ane line,
or a real ane plane, since it has complex dimension 1 and real dimension 2) together with
a single additional point glued to the entire horizon of C. We recognize P1 (C) S 2 ,
the Riemann sphere, identied with C {} via stereographic projection.
The fact that a nondegenerate conic C in P2 (C) is also homeomorphic to S 2 may at rst

come as a surprise. This fact is a consequence of the birational equivalence f : P1 (C) C
1
which the student may regard simply as a homeomorphism such that both f and f are
18. Curves
111
expressible using rational functions (or polynomials) of the coordinates. This denition
requires some delicate interpretation: in our example we may take
C = {(x, y, z) : xz y 2 = 0, (0, 0, 0) = (x, y, z) C3 }
= {(s2 , st, t2 ) : (0, 0) = (s, t) C2 }
so that f : P1 (C) C is simply
(s, t) (s2 , st, t2 ).
The inverse map f
: C P1 (C) is dened piecewise by

(x, y), if x = 0;
(x, y, z)
(y, z), if z = 0.
Here it should be noted that x and z cannot vanish simultaneously at any point of the conic
C; and in the overlapping region where both of the conditions x = 0 and z = 0 hold, the
1
two expressions given for f ((x, y, z)) agree. Moreover this overlapping region xz = 0
includes most of the points of P1 (C) (technically, it is a Zariski dense subset; but such
technicalities we are omitting). A deep result of Hasse and Weil relates the most basic
combinatorial properties of C, namely the number of points of the curve with coordinates
in a given nite extension Fqr Fq , with the genus of the curve, as follows.
18.2 Hasse-Weil Theorem. Consider a (projective algebraic) curve C dened over
a nite eld Fq by a homogeneous polynomial f(X, Y, Z). Let Nr be the number of
Fqr -rational points of C. Then
|Nr (q r + 1)| 2gq r/2
where g is the genus of C.
Note that the projective line P1 (Fqr ) over Fqr has exactly q r + 1 points. By Theorem 18.2
this is the average or typical number of points on a given curve. For lines (curves of
genus 0) this average value is attained exactly. For typical or randomly chosen curves of
higher genus (which, by virtue of (29.1) requires higher degree) we typically nd that the
number of points is distributed rather randomly with a mean value of q r + 1 and standard
deviation proportional to g q r .
Although we do not provide here a proof of the Hasse-Weil Theorem, we mention that
it follows from much more general properties of the Zeta-function ZC (t) of the curve C.
Thats a Z (upper case Greek letter Zeta), not a Z (upper case Roman z). This function
of a complex variable t is given by
ZC (t) = exp
Nr
tr .
r
r1
112
The Riemann hypothesis for curves states that ZC (t) is in fact a rational function of t, i.e.
ZC (t) C(t), of the form
(18.3)
ZC (t) =
(1 + 1 t)(1 + 2 t) (1 + 2g t)
(1 t)(1 qt)
where g is the genus of C and 1 , . . . , 2g are algebraic integers of modulus q which occur
in complex conjugate pairs. Statement (18.3) is known as the Riemann hypothesis for C.
It was rst proved in 1973 by Deligne [23], earning him a Fields Medal; for an elementary
albeit arduous proof, see [34]. The original Riemann hypothesis concerns the location of
the zeroes of the Riemann zeta-function, this being the zeta function of the line over the
rational eld Q. By the substitution
C (s) = ZC (q s )
we see that (18.3) implies that the complex zeroes of C lie on the line Re(s) = 12 ; hence
the connection with the original Riemann Hypothesis. The assertion (18.3) is one of the
original Weil conjectures, which more generally give the structure of Zeta functions of
varieties dened over nite elds. We show that the Hasse-Weil Theorem is a consequence
of (18.3): assuming (18.3),

ln(1 + i t) ln(1 t) ln(1 qt)
ln ZC (t) =
1i2g

2i t2
3 t3
t2
t3
+ i + t + + +
2
3
2
3
1i2g

q 3 t3
q 2 t2
+
+
+ qt +
2
3
r
and by comparing coecients of t on both sides we obtain
=
(18.4)
Since |i | =
(18.5)
i t
Nr = q r + 1 + r1 + r2 + + r2g .
q by assumption, we obtain
2gq r/2 Nr (q r + 1) 2gq r/2
which is the conclusion of Theorem 18.2.

18.6 Example: The Projective Line. Consider a curve of genus 0, i.e. a projective
line or conic, say P1 = P1 (Fq ), so that Nr = q r + 1. We obtain

tr
ln ZP1 (t) =
qr + 1
r
r1
tr
r1
q r tr
r
r1
= ln(1 t) + ln(1 qt)
113
18. Curves
using the familiar Taylor expansion for ln(1 t). Thus

ZP1 (t) =
1
(1 t)(1 qt)
which is a rational function of the required form.

18.7 Example: The Hermitian Curve.
f(X, Y, Z) = X p
+1
+Yp
+1
Let F = Fq where q = p2e and let
+ Zp
+1
= X q0 +1 + Y q0 +1 + Z q0 +1
where q0 = q = pe . (This curve appeared in Section 10 as the set of absolute points with
respect to a unitary polarity.) Note that fX (X, Y, Z) = (q0 + 1)X q0 = X q0 Fq [X, Y, Z]
and similarly for the other partial derivatives. Since the gradient vector

fX (X, Y, Z), fY (X, Y, Z), fZ (X, Y, Z) = (X q0 , Y q0 , Z q0 )
does not vanish at any point of P2 (F q ), the Hermitian curve H consisting of the zeroes
of f is smooth; so by (18.1) its genus is

q0
g=
= 12 q0 q0 1 = 12 q q0 .
2
The Hasse-Weil upper bound for the number of Fq -rational points of H is
N1 q02 + 1 + 2gq0 = q02 + 1 + q02 (q0 1) = q03 + 1.
In fact this upper bound is attained, as we now show. The eld F is a quadratic extension
of K = Fq0 . The norm map
NF/K : x xq0 +1
gives a (q0 +1)-to-1 map from F to K , cyclic groups of order q02 1 and q0 1 respectively.
The F -rational points of H have the form
(0, 1, a)
and (1, a, 0)
where N (a) = 1 (q0 +1 such values of a F ); also

(1, y, z)
where y F such that 1 + y q0 +1 = 0 (there are q02 (q0 + 1) such values of y F ) and for
each such y there are q0 +1 values of z F having the required norm (1 + y q0 +1 ), for a
total of
N1 = 2(q0 + 1) + (q02 q0 1)(q0 + 1) = q03 + 1 = q 3/2 + 1.
114
Since the upper bound of (18.5) is attained, for r = 1 we obtain

1 = 2 = = 2g =
and so
(1 + qt)2g
(1 + qt)q q
ZH (t) =
=
.
(1 t)(1 qt)
(1 t)(1 qt)
18.8 Example: An Elliptic Curve. An elliptic curve is simply a curve of genus 1.

Take for example the curve E dened by
f(X, Y, Z) = Y 2 Z X 3 + XZ 2 Fq [X, Y, Z]
where q 3 mod 4. The gradient

fX (X, Y, Z), fY (X, Y, Z), fZ (X, Y, Z) = Z 2 3X 2 , 2Y Z, Y 2 +2XZ
cannot vanish at any point of E, so E is smooth. By (18.1), the genus of E is 1, i.e. E is an
elliptic curve. It may be shown (see for example [35, pp.5961]) that

Nr =
q r + 1,
if r is odd;
r/2
q + 1 2(q)
, if r is even.
This yields
ln ZE (t) = ln(1 t) ln(1 qt) 2
(q)r/2
tr
r
r1
r even
= ln[(1 t)(1 qt)]
(qt2 )s
s1
so that
ZE (t) =
= ln[(1 t)(1 qt)] + ln(1 + qt2 )
1 + qt2
.
(1 t)(1 qt)
Again, (18.3) holds.

A (nonzero) dierential 1-form on C is an expression of the form
=
g1 (X, Y, Z) dX + g2 (X, Y, Z) dY + g3 (X, Y, Z) dZ

h(X, Y, Z)
where g1 , g2 , g3 F [X, Y, Z] are homogeneous of the same degree d for some d 0,

and h F [X, Y, Z] is homogeneous of degree d+1. The degree restrictions ensure that
18. Curves
115
does not change under substitutions (X, Y, Z) (X, Y, Z) where = 0, since

d(X) = dX and similarly for Y and Z. Now consider another form
"=
g1 (X, Y, Z) dX + "
"
g2 (X, Y, Z) dY + g"3 (X, Y, Z) dZ
.
"
h(X, Y, Z)
We identify
" with whenever
" diers from only by a multiple of f or of
df =
f
f
f
dX +
dY +
dZ .
X
Y
Z
To paraphrase this: We can rewrite without changing its value, by freely using the
relations f = 0 and df = 0. Since both f and df vanish on C, the expression is therefore
well-dened at points (x, y, z) of C. We say is regular at a point (x, y, z) of C, if
the denominator h(x, y, z) = 0 (or if may be rewritten in some equivalent form, say
" as
above, with "
h(x, y, z) = 0). We say is regular (on C) if it is regular at every point of C.
In this case it may be necessary to represent using dierent (and of course equivalent)
expressions at dierent points of the curve C, as our examples will illustrate. Denote by
1C the vector space (over F ) of all regular dierential 1-forms on C. Then
(18.9)
1C has nite dimension g 0.
This gives an algebraic denition of the genus g of the curve C.

We verify that this algebraic denition of genus gives the expected answer for those
examples we have considered thus far. Consider rst the case of a projective line in P2 (F )
dened by f(X, Y, Z) = X = 0; see Example 18.6. In this case both X and dX vanish on
the line , so every nonzero dierential 1-form may be uniquely expressed in the form
=
g1 (Y, Z) dY + g2 (Y, Z) dZ
h(Y, Z)
for some homogeneous gi (Y, Z) F [Y, Z] of degree d 0 and h(Y, Z) F [Y, Z] of degree
d+1 1. We have

(ai Y + bi Z)
h(Y, Z) =
0id
for some ai , bi F with (ai , bi ) = 0. We may assume this expression for is reduced, i.e.
none of the linear factors ai Y + bi Z divides the numerator; then fails to be regular at the
point (0, bi , ai ) of C. And no amount of rewriting (by including extra X or dX terms)
with change this fact. Thus 1P1 = 0 and g = 0, in agreement with our previous nding.
Let us turn to the Hermitian curve H of Example 18.7 dened by f(X, Y, Z) = X q0 +1 +
Y q0 +1 + Z q0 +1 . Then both f and
df = X q0 dX + Y q0 dY + Z q0 dZ
116
vanish on H. Since
X df f dX = Y q0 (X dY Y dX) + Z q0 (X dZ Z dX) = 0
on H, we obtain the rst equality in the expression
(18.10)
X dY Y dX
Z dX X dZ
Z dX X dZ
=
=
Z q0
Y q0
Y q0
and the second equality in (18.10) follows by cyclically permuting the variables X, Y, Z.
Multiplying the expression (18.10) by X i Y j Z k where i, j, k 0 and i+j+k = q0 2 gives
q0
linearly independent dierential 1-forms
2
ijk =

XiY j
XiZ k
Y jZk
X
dY
Y
dX
=
Z
dX
X
dZ
=
Z
dX
X
dZ
.
Z q0 k
Y q0 j
X q0 i
At every point (x, y, z) of H, at least one of the coordinates x, y, z is nonzero, and so at
least one of the expressions for ijk listed above has nonvanishing denominator. It may
be shown that the ijk span 1H , so that
g=
dim 1H

q0
=
= 12 q0 q0 1 = 12 q q0
2
as previously indicated.
Consider now the elliptic curve E dened by f(X, Y, Z) = Y 2 ZX 3 +XZ 2 , introduced
in Example 18.8. The dening equation of E may be rewritten as
y 2 = x3 x
where x =
X
Z,
y=
Y
Z
and so
2y dy = (3x2 1) dx.
Consider the dierential

=
i.e.
=
dx
2 dy
= 2
,
y
3x 1
Z dY Y dZ
Z dX X dZ
=2
.
YZ
3X 2 Z 2
The only point of E where both Y Z and 3X 2 Z 2 vanish is (0, 1, 0), so is regular at
all points of E except possibly this one. Now rewrite the dening equation of E in the form
w2 x = x2 1
117
where x =
X
Z,
w=
Y
X,
so that
(2x w2 ) dx = 2xw dw;
=
dx
2 dw
=
xw
2x w2
=
2X 2 Z d(Y /X)
2X 3 Y 2 Z
2Z(X dY Y dX)
2(Y 2 Z + XZ 2 ) Y 2 Z
=2
X dY Y dX
Y 2 + 2XZ
where the denominator does not vanish at (0, 1, 0). Thus in fact is regular everywhere.
It may be shown that spans 1E , so that the genus of E is 1 as previously indicated.
Exercises 18.
1. Consider the projective curve C having the ane description y2 +y = x3 dened over F2 . The number
of F2r -rational points of C is
r
2 + 1,
for r = 1, 3, 5, . . .;
Nr =
r
r/2
2 + 1 2(2) , for r = 2, 4, 6, . . ..
(This includes the one point at innity.) You may assume this.
(a) Show that C is smooth.
(b) What is the genus of C?
(c) Determine the Zeta function ZC (t) of C. Express your answer as a rational function of t in the
simplest form possible.
2. Consider the curve C dened in Exercise #1. Find a basis for 1C .
118
120
PART IV
Projective and Polar Spaces

19. Classical Ane and Projective Spaces
For every eld F and integer n 0, we dene classical ane n-space over F as the
incidence system An (F ) formed by the ane subspaces of a vector n-space over F . Thus
the points, lines, planes, etc. of ane n-space are formed by the cosets of vector k-subspaces
for k = 0, 1, 2, . . . respectively. The eld F may be replaced by any skeweld K in this
construction, in which case one takes just the subspaces of the left vector space K n over
K (or alternatively, the right vector space K n ); again, however, this remark does not yield
any new nite examples since every nite skeweld is commutative, hence a eld. Of more
interest to us, and for similar reasons as described in Section 8, are the projective spaces
which we now introduce.
Let V be an n-dimensional vector space over a eld F . The incidence structure formed
by the subspaces of V of dimension k = 1, 2, 3 . . . , n1 form the objects called points,
lines, planes, . . . , hyperplanes of classical projective (n1)-space over F , denoted
PV = Pn (F ); in general a projective (k1)-subspace1 of PV is given by a vector k-subspace
of V . The incidence relation between subspaces is the natural inclusion relation. Again,
the eld F may be replaced by any skeweld K.
In the case of a nite eld F = Fq , there are only nitely many vector k-subspaces U
V for each k {1, 2, 3, . . . , n}. One may specify any such subspace as U = u1 , u2 , . . . , uk
where u1 , u2 , . . . , uk V are linearly independent. There are
q n 1 choices of u1 V ........ {0}.
Then for each such u1 there are
q n q choices of u2 V ........ u1 .
And for each choice of u1 , u2 there are
q n q 2 choices of u3 V ........ u1 , u2 ;
etc. Thus the number of possible ordered lists of k linearly independent vectors u1 , u2 , . . . ,
uk V , giving an ordered basis for some vector k-subspace of V is
(q n 1)(q n q)(q n q 2 ) (q n q k1 ).
A similar count shows that the number of choices of ordered basis for a given k-subspace
U V is
(q k 1)(q k q)(q k q 2 ) (q k q k1 ).
1
The projective dimension of a subspace PU PV , is one less than the vector space dimension of U V . Unless
we specify projective dimension, then we refer to the usual vector space dimension. For example if V is a 3dimensional vector space, then its projective dimension is 2, i.e. PV is a projective 2-space, otherwise known as
a projective plane.
122
IV. PROJECTIVE AND POLAR SPACES
Therefore the number of vector k-subspaces of the n-space V is given by the Gaussian
coecient
n n
(q n 1)(q n q)(q n q 2 ) (q n q k1 )
:= k
=
.
(q 1)(q k q)(q k q 2 ) (q k q k1 )
k
k q
For example projective 3-space over Fq , denoted P3 (Fq ), has

q4 1
4
=
= (q 2 + 1)(q + 1) points;
q1
1 q

(q 4 1)(q 4 q 2 )
4
= 2
= (q 2 + 1)(q 2 + q + 1) lines; and
(q 1)(q 2 q)
2 q

4
(q 4 1)(q 4 q 2 )(q 4 q 3 )
= (q 2 + 1)(q + 1) planes.
= 3
3
3
2
(q 1)(q q)(q q )
3 q
In the case of the eld F2 , projective 3-space has 15 points, 35 lines, and 15 planes. We
may represent this geometry by the Hasse diagram
...........................................................................................................................................................................................................................................................
.........
......
.........
................................................ ............... ....... ..................
. .................................................................................................................................................................. ...................................................................................................................................................
....
............. ..................... ........................................................................... ................................. .. .................................................................................
.. ... ... . ............. .. ............................... .................. ..... .................... ........... ...................... ............
...
... .............................. ..................................................................................................................................................................... .................................................................................................................................................................................................
... ....
....................................................................................... .......
.................................. .. ..............................................
........ .............
.................................
............................................................................................
. ..... . .. ..
...............
... .....
. .............. ......... .......................................................................
............ ...................................... .....
..... ......
.. ..
..................................... ........................ ............................................................................................................................................................................................. ..... ....................................................................................................................................................................................
.
.
.
........
.
.
..... ........ .............. ..
..
......................................................................... .......................................................... ........... ................................ . ...................................................................
.
.
.
....
.
.
.
.
.
.
.
.
.
......... ....... ..
.
3
.... ....... ......... ..................... ............................... ............. ...... ....... ............. ............................ ....................
...
......
....
........... .. .. .............................................................. ................................................................. .......... .......................................................................................................................................
.... .................. .. .......
2 ............ ...............
...................... ..........................................
.. ............. .. ...... .. ...... ......... ... ... .. .. ..... ................................................... .. ......................... ............... ................ .............
..................... .......................................
........................ .................... ....................... .... .......... .......... ........................................................................... .................................................................................................................
. ......
.
.
.... ..... .....
... .....
.
.
.
.
.
.. .
. ..
...... .. ......
..... ..... ..................................
....... ......
... .....
..... ..... .......................................... ...................................................................................................................................................................................................................................................................
......... .........
........ ........................
...... ............. ....
............ ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
........... ..
...... ... ............ ............................................
................................... ........................... ..................................................................................................................................
.
.. ......
....
... ... ............. .......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..... ..... ...........
............. ............... ................................................... ....................................... ....... .................... ............................... ......................................................................................
... ......................
......
... ......... ............
..... .. .. ... ......
............................... .................................
............. ...... ....... ................................................. ..................... .................. ......................... ..................................
........... ..... .. ...
.........
...... .... .........
................. ...... ................
...........
................... .................. ........ .........................
... ..... ... ....
... ........... ... ..... ........
..... ....
..........
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
................. ......... ........... .......... .................................... ............. ................................................................... .....................................................................................................
.
.
.
.
.
.. ..... .. .... .....
. ........ .. ....
.... ...... . ..
.... ..
..... .
.... ..
.......... ........... ............. ......... ............ ......... ........... .............. ........... ............. ............................. ........... ................. ............... ............... ........... .................. .............. ................. ................. ................... ..................................... .................. .................. ............................... ....................................... ........... .................... ................... .............................................
........... .. ....... .. ....................... .. ............ ............ .......... .. ......... ............. ............... ............................ .............. ................ ............................ .......... ............... .............. .......... .. .......... ............. ................. .............. .......... ............. .. ............. ................. ................................................ ............... ..........................................
............. .. ............... ............................. ................... ......... ............ ....................... .................. ................ ........................................... ........................................................... ....... ...................................... ............. ............. ............................... .............. ........... ................................................. ................. ...................................................................... .............
............... .............. ................... ........ ...... .......... ................... ... ............... .... ..................... ................................................................ ...... ...................................................... .......... ............................................. ............. .......... ................ ............. ............................ ......................... .......................................................... ............. ....................
............... ................. ... ........................... ........ .................................. .... ................. ........................ ................................................................... ....... .................................................... .................................................. ........... ............................ .................. ................ ............... .......................................................................................................... ... .................
... .......... . ................ ............................... .............. .................. ..... ................. ........................ .................................................................... ............. .................................... .................... ....................... ......................... ............... ........................... .............................. ........................................................................ .. ..................
..... ..... .. ...... ........ .. .................. ................... .... ................... ........ . ...... .................... ...... ..... ..... ..... ............. ....... ....... .. ..................... ........... ....... ..... ................
... ............ ... ............... .......................................... ...............................
... ...... ........
.. . ... . .. ... .......... .. ...... .....
......................... ................... ....... ............................ ..................................... ............................ ..................... ........ ............................................... ... ........... ......................................... .......................... .................................... .......
................
... ..... ....... ............................................................................. .................................... ..... .......................................... ............................................................................................................................................................................................................ ......................................................................................................................................................................................................................................................................... ................. .........................
.
.
.
.
.
.
.
.
.
.
.
.
.
... ..... .......
.
.
.
.
.
.
.
.
.
.
.
.
.. .. .... .... .. .... ...... ... ......... ............ .... ............... . ............ ...... .............. .............. ............. ............ ........... ............ .... . . ... .... .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... .......... ....... ................ .......................... ................................................................................................................. .................................................................................................................................................................................................................................................................................................................................................................................................................... ............................................................................................................................................................ .........................................
... .. ...... ..... ... ........ .. ............ ........................................ ............................
..... ... ... ... ..... ........... .... ...... ..... ... ..... .... ... ....... ... .... ..... ..... ....... ........ ..... ........... ... .... .. ......... . ... . ... .. . .
... ... .... ....... .. .. ........... ......... ...... ... ...................................................... ......................................................... ....................................................................................................................................................................................................................................................................................................................................................................................................................................... .................................. .................... .......................... ...
..... ....... .... .. ........ ... .......... ..... ............... ............... ......................... ..... ........ ......... . ...... ... .. . .
... ... .......
.. . . . ... . ........ .. ...... .... .. ..... ...... ..... ... .. .. ...
... .. .. .... .............................................................................................................................................................................................................. .................................................................... ................................................................................................................................................................................................................................................................................................................................................................................................... ............................................. ........... ..... ....
... ... .. ...
... . .. . .. . ... ...... .. ... ... .... ....... .... .
... . .... ... .... ... ... ......... ...... .. ....... ......... .............................. ....... ...... ........ ... .... .. ... ... .... . ... . .
... ... .. .... .................................. ................................................................. ................................................................................................................................................................. ............................................................................................................................................................................................................................................................................................................................................................................................................................................................... ............................ ............ ......... .....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....................... . ...................... .... ......................... ........... ..... .......... .... ........ ......... ....... ..... .. ..... .. ..
... .. . ... ..................
.
............................ ............. ............................................... .................................................
... . .... ..
........
. .. . . . . . . . . .. . . .
............................... ... ...... .......................................... .................................................................... ............................. ......................................................................................... ........... ..... ... ..
........ ............................... ............................................................................................................................... ................................................................................................................................................ ..................................................................... ......................................................................................................................................................................................................................................................................................................................................................................................... .................... .... ...
.
.
.
.............. ... .... ... .. ......... ... .... .. ...... ... . .. ..... ....... .. ......... .... ....... ......... ... . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
........ .................... .....
.
.
.
.
.
.
.
.
.
.
............ ..
.
.
.
.
.
.
.
.
.
.
.
.
. .
....... ............ ....... ............................. ....... .......................................... ................................................ .................................................................................................. ............................................................................................................................................................................................................................................................................................................................................................................................................................................................... ...
. . . . . . . . . .. . . . . . . .. . . . . . .. ... . . . .
........ ......... .......... ......... ............. ........ ........... .......... .......
....... .........
.... . . . .. . . . . . . .
...... ...........
.............
...........
.............
....................... ...................................... ......................................... ...................................... ...................................... ...................................... ...................................... .................................. ......................................
...........
........
....
15
planes
19.1
Figure
P (F )

35
lines
15
points
in which the bottom row lists the 15 points in lexicographic order

(0, 0, 0, 1), (0, 0, 1, 0), (0, 0, 1, 1), (0, 1, 0, 0), . . . , (1, 1, 1, 1);
the top row lists the 15 planes in lexicographic order
T
(0, 0, 0, 1) , (0, 0, 1, 0) , (0, 0, 1, 1) , (0, 1, 0, 0) , . . . , (1, 1, 1, 1) .

Here we denote the plane

a0
a1
(x0 x1 x2 x3 ) : x0 x1 x2 x3 = a0 x0 + a1 x1 + a2 x2 + a3 x3 = 0
a2
a3
123
by the 1-space (a0 , a1 , a2 , a3 )T of column vectors; this notation closely parallels the notation of Section 6 for coordinatizing points and lines of P2 (F ). The middle row of Figure 19.1
lists the 35 lines in lexicographic order of their Pl
ucker coordinates (these coordinates for
lines will be dened in Section 22). Edges in this graph indicate incidences between points
and lines, and between lines and planes. For clarity, we have not shown the incidences
between points and planes; but these are implied by the edges shown (for example the rst
point P = (0, 0, 0, 1) is incident with the second plane = (0, 0, 1, 0)T as shown by the
three paths of length 2 in the graph above; these three paths correspond to the three lines
incident with both P and ). Note that two distinct lines , m may be related in either
of two ways: either they meet in a point (in which case they also lie in some plane); or
they are disjoint, in which case we say and m are skew. However one does not speak of
two distinct lines as being incident; recall that incidence is inclusion, and no line can be
contained in another unless they are equal. In P3 (F2 ),
every point is incident with 7 lines and 7 planes;
every line is incident with 3 points and 3 planes; and
every plane is incident with 7 points and 7 lines.
More than this, the subgraph of Figure 19.1 consisting of all points and lines incident with
a given plane, is of course just the incidence graph of the projective plane P2 (F2 ):
19.2 Figure
Incidence
Graph of
P2 (F2 )
................................................................... ....................................................
......
......
......
.........
.........
.........
.................... .......... ...
.. . .
...
..... .....
..... .....
..... .....
.......... ......... .......... ............................ .......... .... ..... .......... .... ..... .......... .... .....
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
. .......
. .. ...
. .. ...
. ..
. ...... ....... .......
.. ........
.. ...
.. ... ............................................................ ............................. ............
...
....... . ............. .
....... .
... .........
... ..
... ...
...
... ....
... ....
....................... ................................ .......................... ............ ....
...
.
.
.
.
.
.
.
...
.
. ... ...
... .............
. ..... ...
.
... ..........
... ....
...
..................... ......................................... .............................. ..... ....
......... ...
... ...
.
. . . . . . .. . . .
...
... ... .............. .... ............. .... ....................... .... ......................................... ............... ....
...
... .... ....... ... .... ....... ... .... ....... ............... ....... ............... ............................
.
.
.
.
.
... .. ......
.
.
.
.
.
.
.
.
.
.
.
.
.
.. ... .......
.
.
... ... .......
.
.
. ............. ..... ......... ......................... ....
.
.
.
.
.
.
.
...........
.
.
.
.
.
.
...........
...........
............ ..
....... ....
....... .....
...........
.
.......
.
.
.
.
.
.
...
...
....
..
..
..
Similarly, the subgraph of Figure 19.1 consisting of all lines and planes incident with a
given point, also gives the incidence graph of the projective plane P2 (F2 ) shown in Figure
19.2. To see why this should be true, observe that for every point, i.e. 1-space P < V ,
there is a one-to-one correspondence between subspaces of V containing P , and subspaces
of the vector 3-space V /P . Under this correspondence, the lines and planes containing P
correspond to points and lines of V /P .
The subgraph of Figure 19.1 consisting of all points and planes incident with a given
line , gives the incidence graph shown in Figure 19.3. This incidence structure is known as
a generalized digon of order 2. (The explanation for this name will be given in Section 29.)
Here each of the three points on is incident with each of the three planes containing .
19.3 Figure
Incidence Graph of a
Generalized Digon
of order 2
........................
......
........
.......
......... ..
. .
... .. ..
...... ... .
..... .................
... .. ....
..... .. ..
.. .... ........ ..... .... ...... ........... ..... ....
..
..
... ......... .. ......... ....
..
.... .. ...... .. ...... .... ..
...
.............
......
......
..
...
.
.
.
.
.
.
.
.
.
...
.
..
.
.
.
.
.
.
... ..... .......... ... ........... .... .....
...
..
.
. ......
. ...
.
.
.
.
..
.
.
.
.
.
.
.
... ..... ................ ... ................ ..... ...
.. ... ......
.
.
.
.
.
.
...... ... ...
... .. ...
... .. ..
...... ... ..
... ............
..........
.......
............
......
......
...
A generalized digon of order n is an incidence structure with n+1 points and the same
number of blocks, such that every point is incident with every block; its incidence graph
is therefore a complete bipartite graph Kn+1,n+1 .
124
The space Pn (F ) is self-dual. In this case duality interchanges points with hyperplanes,
lines with subspaces of codimension 2; and in general, vector k-subspaces of V with vector
(n+1k)-subspaces. (See Exercise #4.)
A hyperplane of Pn (F ) is isomorphic to Pn1 (F ); by deleting this hyperplane, one is
left with an ane space An (F ). Conversely, if one starts with an ane space An (F ) and
adds a point at innity for every parallel class of lines of the ane space (also naturally
dening other subspaces at innity as natural subsets of these new points), one obtains the
projective space Pn (F ) as the projective completion of the original ane space. This
generalizes the process of completing an ane plane to a projective plane as described in
Section 7. To see this, note that the points of Pn (F ) are 1-spaces (x0 , x1 , x2 , . . . , xn ),
partitioned into those points with x0 = 0 (this being a typical hyperplane, a projective
(n1)-subspace) and those with x0 = 0 (which may be normalized by the scalar factor
1
x0 so as to be uniquely expressible in the form (1, a1 , a2 , . . . , an ) with a1 , a2 , . . . , an F
(and the latter points form an ane n-space). In the case F = Fq this partition of the
points of Pn (Fq ) gives rise to the identity
qn
+ (q n1 + + q + 1) .
q n + q n1 + + q + 1 =

n+1
n
n
n1
points
of
points of P (Fq )
(Fq )
1
1 q points of P
q
An (Fq )
A homogeneous polynomial f(X0 , X1 , . . . , Xn ) F [X0 , X1 , . . . , Xn ] of degree d yields a
well-dened point set consisting of all points (x0 , x1 , . . . , xn ) such that f(x0 , x1 , . . . , xn ) =
0 (a hypersurface of degree d); here we may take the coordinates xi F (which gives
the so-called F -rational points of the surface) or more generally, xi F , the algebraic
closure of F . The same processes of homogenization and dehomogenization as described
in Section 7 for the case n = 2, allow us to form the projective completion of hypersurfaces
in ane n-space, and the ane parts of hypersurfaces in projective n-space. For example
the surfaces x2 + y 2 + z 2 = 1 and x2 = 1 + y 2 + z 2 in ane 3-space have the same
projective completion: the rst equation homogenizes to X 2 + Y 2 + Z 2 = W 2 , which then
dehomogenizes (by setting Z = 1) to x2 + y 2 + 1 = w2 .
Every semilinear transformation of V = F n maps k-dimensional subspaces to kdimensional subspaces, giving a collineation (i.e. automorphism) of the projective geometry. In fact, every collineation of Pn1 (F ) is induced by a semilinear transformation. Thus
the full collineation group of Pn1 (F ) is isomorphic to P Ln (F ). An ordered frame
in Pn1 (F ) is an ordered (n+1)-tuple (P0 , P1 , . . . , Pn ) of points, no n of which lie in a
hyperplane. We state (but do not prove):
19.4 Fundamental Theorem of Projective Geometry. The full collineation
group of Pn1 (F ) is isomorphic to P Ln1 (F ). Its normal subgroup P GLn (F ) is
regular (i.e. sharply transitive) on ordered frames.
125
We associate to projective planes, generalized digons, projective 3-spaces, and general

projective spaces the following diagrams respectively:
projective plane:
points lines
..............................................
(type A2 )
generalized digon:
points blocks
(type A1 A1 )
projective 3-space:
points lines planes

...........................................................................................
projective n-space:
(type A3 )
points lines planes

hyperplanes
............................................................................................................. ...................

n types of objects
(type An )
The extent of the association between these Coxeter-Dynkin diagrams and the corresponding classes of geometries, is too far-reaching to completely describe here; but a few
observations will serve to illustrate the nature of this association. The symmetry of the
An diagram for n 2, which interchanges the nodes representing k-spaces with (n+1k)spaces, illustrates the fact that the dual of a projective n-space is again a projective n-space.
Given an object U in a geometry (say), the geometry induced on all objects of incident
with U is called the residual geometry of U, or simply the residue of U. Its diagram
is obtained from the diagram for by removing the node corresponding to the type of U,
together with all edges from that node. For example deleting either of the end nodes of the
A3 diagram gives the A2 diagram; this illustrates the fact that objects in P3 (F ) incident
with a given point or plane, have the structure of a projective plane. Deleting the middle
node of the A3 diagram leaves the A1 A1 diagram; this illustrates the fact that the points
and planes of P3 (F ) incident with a given line, have the structure of a generalized digon
as described above. More generally given an n-space V with a k-subspace U V , the
residue of U has diagram Ak1 Ank1 (assuming appropriate allowances are made for
the special cases k {1, n}). The objects of this geometry are the subspaces of U, forming
the subgeometry PU Pk1 (F ) of type An1 , and the subspaces of V containing U.
The latter subspaces correspond to subspaces of V /U , which therefore form a subgeometry
isomorphic to P(V /U ) Pnk1 (F ). Moreover every object in the Ak1 -subgeometry is
incident with every object in the Ank1 -subgeometry, as is represented by the fact that
there are no edges between the corresponding subgraphs in the Ak1 Ank1 diagram.
Exercises 19.
n
.
k q

limq1 n
= n
,
k q
k
1. Show that the number of ane k-subspaces of An (Fq ) is qnk
2. Show that the limit of the Gaussian coecient is

the binomial coecient. We regard projective (n1)-space as the analogue of an (n1)-simplex, which is the incidence system formed
126
by all subsets of an n-set. Note that a projective plane is thus an analogue of a 2-simplex or triangle,
which is the unique 2-(3, 2, 1) design. Also under this analogy, the automorphism group P GL(n, F ) of
projective (n1)-space is the analogue of the symmetric group Sn , which is the automorphism group
of the (n1)-simplex. The (n1)-simplex is often called projective (n1)-space over the eld of
order 1, (even though there is no eld of order 1) or thin projective (n1)-space, while the other
projective (n1)-spaces are thick. Compare with comments in Exercise #6.3.
3. Show that the points and hyperplanes of projective (n1)-space over Fq form a 2-(v, k, ) design where
n1
n2
n
; k=
; =
.
v=
1 q
1 q
1 q
Remarks. Again letting q 1, this gives a 2-(n, n1, n2) design, i.e. an (n1)-simplex. It has n
points, and its blocks (i.e. hyperplanes) are the (n1)-subsets.
4. Let V be an n-dimensional vector space over a eld F , and let V be the vector space of all linear
functionals V F ; thus V is also n-dimensional. Show that there is a one-to-one correspondence
between subspaces of V and subspaces of V which maps each U V to its annihilator
U = { V : (U ) = 0}.
Show moreover that dim(U ) = n dim U and that the map U U reverses inclusion: we have
U W V i U W . Conclude that Pn1 (F ) is self-dual.
Hint. Represent elements of V and V as row and column vectors of length n, respectively. To
evaluate a linear functional at a vector, is simply to right-multiply a row vector by a column vector.
Imitate the proof of Theorem 6.5.
5. Let H be the n (2n 1) matrix whose columns are the nonzero vectors in Fn
2 , i.e. the points of
Pn1 (F2 ). Show that H is the parity check matrix for a [2n 1, 2n n1, 3] binary code. (See Section 13 for the case n = 3.) Verify that this is a perfect 1-error correcting code.
Remark.
This is the family of binary Hamming codes.
20. Axioms
At this point the reader may wonder whether the concept of a projective space should really
be dened axiomatically in terms of natural incidence properties, as we did for projective
planes; and then the classical examples constructed as we have done above, starting with
a vector space over an arbitrary eld (or skeweld). However every projective space of
projective dimension n 3 is necessarily classical. For example here are some natural
candidates as axioms for projective 3-space, in which the objects are points, lines and
planes; and where three incidence relations are given (between points and lines; between
points and planes; and between lines and planes).
(S1) Any two distinct points are incident with exactly one line.
(S2) Any two distinct planes meet in exactly one line.
(S3) Given any line and any point P not on , there exists a unique plane incident
with both P and .
(S4) Every plane incident with a given line is also incident with every point on .
(S5) Any two distinct lines meet in a point, i they lie in a common plane.
(S6) There exists a set of ve points, of which no four lie in a common plane.
127
20. Axioms
Axiom (S3) asserts that every line or plane may be identied with the set of its points, in
such a way that incidence of a line and plane is the same thing as inclusion of the corresponding point sets. These potential axioms are somewhat redundant (see Exercise #20.1).
However they do include all that we naturally require of a projective 3-space. In particular
these axioms imply that for every plane , the incidence structure formed by the points
and lines on is exactly that of a projective plane; and dually, for every point P , the
incidence structure formed by the lines and planes containing P is also that of a projective
plane.
It is possible to axiomatize projective n-space in a style similar to that above, having
projective k-subspaces for k = 0, 1, 2, . . . , n as undened concepts, and with an incidence
relation between subspaces of dimensions j and k for all j = k, and introducing axioms
similar to (S1)(S6) above. An alternative approach, which we prefer, is the following
approach of Veblen and Young (1910).
A projective n-space is a linear space (P, L) satisfying the following axioms, where
we dene a subspace to be any subset of P which contains every line through two of its
points.
(V1) If P and Q are distinct points then there is exactly one line containing both P
and Q.
(V2) There exist three noncollinear points.
(V3) Every line has at least three points.
(V4) The maximum length of a chain of subspaces is n + 2.
.......
.... ..
... .....
.
.
.
...
...
...
....
...
....
...
...
.
.
...
......
.
.
...
.
.. ........
.
.
..
.
.......
.
.
.
.
....... .................
..
.
.
.
...
.
.
...........
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.......
..
.......
...... .....
.... ......................
.
.
.
.
.
...... ...
...
...... ...
...............
........
....
(V5) (Veblen-Pasch Axiom) Every line meeting two sides of

a triangle, but none of its vertices, must also meet the
third side.
Axiom (V1) is included in the denition of a linear space but we have listed it again for
emphasis. In (V4), a chain of subspaces of length n has the form S1 S2 Sn
where each Si is a subspace. For example in projective 3-space every maximal chain has
the form
{P } P
where P is a point lying on a line in a plane . Every subspace S has a well-dened
projective dimension k 1, where k+2 is the maximum length of any chain of subspaces
contained in S. Here the empty subspace has projective dimension 1, points have
projective dimension 0, lines have projective dimension 1, planes have projective dimension
2, etc. with the full point set P being the unique subspace of projective dimension n.
A projective 2-space is the same thing as a projective plane; and we have constructed
non-classical examples of these. However for n 3, every projective n-space is classical!
20.1 Theorem. Every projective n-space for n 3 is isomorphic to Pn (K) for some
skeweld K.
128
We present here just a few key ideas from the proof of Theorem 20.1. Let (P, L) be a
projective n-space where n 3. We rst argue that
(20.2)
Every plane of (P, L) is isomorphic to P2 (K) for some skeweld K.
Let be any plane in (P, L). Consider a point-line conguration in as shown below; we
must show that the three points P, Q, R are collinear. Let S be a point outside , and let
be a point on SB distinct from S and B.
B
= OB
SB , P = AB
SP , R
= BC
SR; these points exist by the Veblen-Pasch
Let B
axiom.
S
S ...........

B

B
.
.......
.... .....
...
...
.
.
.
...
..
...
....
...
....
.
.
...
..
.
.
...
..........
.
.
...
. .......
.
.
.
......
..
.
.
.
...... ....................
..
.
.
.
.
.
.
...
.
........
.
..
.
.
.
.
.
.
.
.
.
.
.
.
..
.......
.....
.
.
..
.
.
.
.
.
.
.
.
.
.
.
....... ....
. ...............
.
.
.
.
.
....... ...
................
..........
........
.
. .
.. ..
.. .....
...
..
.
....

....
... ........... .....
.
..
. .......
.
...
.............
...
...
...
........
...
...... ..
...... ....
...
.
.
.
.
.
.
...
...
.
.
.
.
.
.
.
..
.
...
.
.
.
.
.
.
.
...
..
.
.
.
.
.
.
.
..................................................................................................

B

B

B
.....
.... ...
.... .....
.
.
...
.
...
....
...
...
...
...
.
.
..
...
.
.
........
.
...
.
.
. ......
.
...
.
.
.
.......
..
.
.
.
...... ...................
..
.
.
.
.
.
...
.
.
........
.
.
..
.
.
.
.
.
.
.
.
..
.
.
.
.......
..
.........
....... ....
.... .......................
.
.
.
.
....... ....
...............
....... ...
.......
.
P

R
are collinear; this is because they lie in two distinct planes, as

The points P, A and B
shown:
S
.....
.. ...
... ......
...
...
.
.
....
.
...
....
..
....

..
.
.
.
...........
.
.
....
..
...
.
.
.
.
.
.
.
........
.
.
.
.
...
. ......................
...
...
...
... . ...... ................................. ......
.............
...

... . ......
.
.. ....
... ..
.........
P

B
B
....
.......
........................... .... B
B
. ..
.
...
.......
..
........
...
.
.....
........
O.........................................................................................................
... .. A
A ........ ....
... ..
... .
.....
...
P
Q are also collinear since they lie in two distinct planes as shown:
Now points P, R,
.....
B
... ..
.. .
... .....
...
...
......
.
...
.
.. ....
.... ........... .....
.
.
........
.
.
.
.......
.
...
..........
...
... ........
...
... .......
...
.
.
......
...
..
.
......
.
...
..
......
...
....
....
.
.
....... ....... ....... ....... ....... ....... ....... .............
P

R
.....
B
.. ..
... ..
... .....
...
..
...... ....
.
.
.
...... ..

.......
...
........
..
.
.
... .......
...
.
.
.
.
.
.
.
... .......
....
......
...
......
...
...
......
...
...
....
...
.
.
........ ....... ....... ....... ....... ........ ....... .............
A......................
P

R
129
21. Codes
Finally the points P, Q, R are collinear since they lie in two distinct planesthe original
plane and the plane shown:
S
......
.... ...
.... ....
.
.
.
...
..
...
....
...
....
...
....
.
.
..
...
.
.
..
.
.
.
..... ....... ....... ....... .......... ....... ....... ...........
.
.
.
.
...
..
...........
.
.
.. .......................
..
.
.
................
...
................
.
.
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.
...
... .................
.................
P

R
By the converse of Desargues Theorem (see comments at the end of Section 11) the plane
is isomorphic to P2 (K) for some skeweld K. This proves (20.2).
Next we prove that all planes in (P, L) are isomorphic to P2 (K) for the same skeweld K:
(20.3)
Any two planes , in (P, L) are isomorphic.
Consider rst the case that is a line; then and lie in a common projective solid
U (a projective 3-subspace). Let P U be a point not in ; then every line of U
through P meets in a unique point, and also meets in a unique point. This gives a
bijection from points of to points of . This bijection maps lines of to lines of :
every plane of U passing through P meets both and in lines, which demonstrates
the required correspondence. So
= , which proves (20.3) in the case is a line.
Now in the general case it is easy to nd a sequence of at most two intermediate planes
1 , 2 of (P, L) such that
= 2
= ;
= 1
here shares a line with 1 , which shares a line with 2 , which shares a line with . For
example if and are disjoint (the worst case) take a ag (P, ) in and a ag (P , )
in ; then set 1 = P and 2 = P . The other cases are similar, so (20.3) holds in
general.
From here it is not hard to complete the proof that (P, L) itself is isomorphic to
Pn (K).
Exercises 20.
1. Show that Axioms (S1)(S6) above are not independent; that is, deduce one of these six statements
from the other ve.
2. Show that the Veblen-Pasch Axiom (V5) follows from (S1)(S6), but is independent of (V1)(V4).
21. Codes
In Section 13 we found the
p-rank of the incidence matrix of an arbitrary projective plane
of order n, assuming p n, i.e. p divides n exactly once. More generally, however, the
p-rank of a projective plane is not uniquely determined by its order, but rather depends
130
on the particular structure of the plane. A good rule-of-thumb is that among several
nonisomorphic objects with the same parameters, the nicer examples (in particular the
more classical examples, or those with larger automorphism groups) tend to have lower
p-ranks. In particular a nice formula is available for the p-rank of every classical projective
plane of order q = pr . We have waited until now to derive this formula since the same
technique applies more generally to the incidence matrix of points versus hyperplanes of
Pn (Fq ).
21.1 Theorem. Let A be the N N incidence matrix of points versus hyperplanes

= (q n+1 1)/(q 1). Then the p-rank of A
of Pn (Fq ) where q = pr and N = n+1
1
q
is

r
p+n1
+ 1.
n

Check: By Theorem 13.3 the p-rank of P2 (Fp ) is p+1

+1; and Theorem 21.1 conrms
2
this. Our proof of Theorem 21.1 relies on a theorem of Lucas, which we introduce by
looking at the pattern of even and odd numbers in Pascals Triangle:
1
1 1
1 2 1
1 3 3 1
1 4 6 4 1
1 5 10 10 5 1
1 6 15 20 15 6 1
1 7 21 35 35 21 7 1
. . . etc. . . .
Most students will quickly recognize the emerging pattern as one that appears in Sierpin
skis Triangle. The question of whether the binomial coecient m

k is even or odd is most
readily answered by rst expressing both m and k in binary as
m = m0 +2m1 +4m2 +8m3 + ;
k = k0 +2k1 +4k2 +8k3 +
where mi , ki {0, 1}.
Then
(21.2)
mj
j0
kj
mod 2
which vanishes whenever kj > mj for some j, since 01 = 0. The product (21.2) is eectively nite since m and k have only nitely many nonzero binary digits, beyond which
131
21. Codes
all remaining factors are simply 00 = 1. Also if k

/ {0, 1, 2, . . . , m} then by denition
m
= 0; but then the right side of (21.2) also vanishes. The relation (21.2) generalk
izes to arbitrary primes, and also to multinomial coecients. We recall the denition of
multinomial coecients:
!

m!
m
k0 !k1 !kn ! , if m, k0 , . . . , kn 0 and k0 +k1 + +kn = m;
=
k0 , k1 , , kn
0,
otherwise.
21.3 Theorem (Lucas). Let p be prime, and express the non-negative integers
m, k0 , k1 , . . . , kn in base p as
"
"
mj pj ;
ki =
kij pj
m=
j0
j0
where mj , kij {0, 1, 2, . . . , p1}. Then

#
mj
m
mod p.
k0j , k1j , , knj
k0 , k1 , , kn
j0
Proof.
(21.4)
By the Multinomial Theorem,

m
(X0 + X1 + + Xn )
k0 , k1 , ..., kn

m
X0k0 X1k1 Xnkn .
k0 , k1 , , kn
There are only nitely many nonzero terms in the latter sum, namely those for which
k0 , k1 , . . . , kn 0 and k0 +k1 + +kn = m. As a special case of (21.4), we rst observe
that
(21.5)
(X0 + X1 + + Xn )p X0p + X1p + + Xnp mod p.
This follows from the fact that the multinomial coecient

p
p!
=
k0 , k1 , . . . , kn
k0 !k1 ! kn !
is divisible by p unless one of the kj s is p and all other kj s are zero. Now (21.4) gives
(X0 + X1 + + Xn )m =
(X0 + X1 + + Xn )p
mj
j0
j0
(X0p + X1p + + Xnp )mj mod p
132
by repeated application of (21.5), and so

(X0 +X1 + +Xn )m

$
#

j
mj
k pj k pj
X0 0j X1 1j Xnknj p
k0j , k1j , , knj
j0 k0j , k1j , ..., knj

$

k00 +k01 p+k02 p2 + k10 +k11 p+k12 p2 +

2
mj
X1
Xnkn0 +kn1 p+kn2 p +
k0j ,k1j ,,knj X0
j0
mod p.
k00 , k01 , ...;
k10 , k11 , ...;
kn0 , kn1 , ...
Here each kij {0, 1, 2, . . . , p1} and so by the uniqueness of the base p expansion of
the integers m, k0 , . . . , kn , we may equate coecients with those in (21.4) to obtain the
result.
We are also interested in counting the number of terms in the multinomial expansion
(21.4).
21.6 Lemma. The number of (n+1)-tuples (k0 , k1 , . . . , kn ) of non-negative integers
such that k0 +k1 + +kn = m is

m+n
.
n
Proof. The (n+1)-tuples (k0 , k1 , . . . , kn ) satisfying the required constraints, correspond
bijectively to n-subsets of {0, 1, 2, . . . , m+n1} via
%
&
(k0 , k1 , k2 , . . . , kn ) k0 , k0 +k1 +1, k0 +k1 +k2 +2, . . . , k0 +k1 + +kn1 +n1 .

of Pn (Fq ), where
Let A be the N N incidence matrix of points versus hyperplanes

n+1
N=
1

q
= 1+q+q 2 + +q n 1 mod p.
By Exercise #19.3, every row and column of A has sum

n
= 1+q+q 2 + +q n1 1 mod p;
1 q
therefore the following matrix identity holds mod p :
...
...
...
.... 0 0
.... 1 1
.... 0 0
0
1
1
.
.
.
.....................................................
.....................................................
......................................................
..
..
1 .......
0 .......
T
T
B
=
.
=
B
B 0.. .......
.
. ...
.. ....
B
.
.
.
AJ
A
A
.
.
.
. ...
. ...
. ...
0 .....
1 .....
0 .....
133
21. Codes
where the matrices
1 00
B = 1..
. I
1
...
..
.
........................................................
...
..
..
..
...
N
...
..
...
.
1
1
and J = ..
.
1
1
1
..
.
1
..
.
1
1
..
.
1
have size (N +1)(N +1) and N N respectively, and IN denotes the N N identity
matrix. This yields
rankp A = 1 + rankp (AJ ) = 1 + rankp (J A).
The matrix J A is easier to describe algebraically than A, as we now observe. Every
nonzero vector a = (a0 , a1 , . . . , an ) Fn+1
determines a hyperplane Ha consisting of all
q
n+1
such that
points x=(x0 , x1 , . . . , xn ) Fq
a0 x0 + a1 x1 + + an xn = 0.
Note that Ha = Ha whenever = 0; therefore the hyperplane Ha is determined by the
. The (x, a)-entry of J A is
1-space a Fn+1
q
!
(a0 x0 + a1 x1 + + an xn )q1 =
0,
if the point x lies in Ha ;
1,
otherwise.
We consider therefore the q n+1 q n+1 matrix M with rows and columns indexed by vectors
, and whose (x, a)-entry is (a0 x0 + a1 x1 + + an xn )q1 . This matrix is larger
x, a Fn+1
q
than J A itself; in particular it has an extra row and column of zeroes, corresponding to
x=0 and a=0. It also has many repeated rows and columns, since for all = 0 the rows
of M indexed by x and x coincide; also the columns of M indexed by a and a coincide.
Evidently J A is obtained from M by deleting duplicate rows and columns, as well as
deleting a row of zeroes and a column of zeroes, so that
(21.7)
rankp A = 1 + rankp (J A) = 1 + rankp M.
The matrix M can be conveniently expressed in terms of another q n+1 q n+1 matrix
and columns indexed by (n+1)-tuples k =
U with rows indexed by vectors x Fn+1
q
n+1
; the (x, k)-entry of U is
(k0 , k1 , . . . , kn ) {0, 1, 2, . . . , q1}
ux,k = xk00 xk11 xknn .
Also dene the q n+1q n+1 diagonal matrix D with rows and columns indexed by (n+1)tuples k, {0, 1, 2, . . . , q1}n+1; the (k, )-entry of D is

dk, =

q1
k,
k0 , k1 , . . . , kn
134
where we use the Kronecker delta symbol

1, if k = , i.e. (k0 , k1 , . . . , kn ) = (0 , 1 , . . . , n );
k, =
0, otherwise.
Now the (x, a)-entry of U DU T is
$
ux,k dk, ua, =
k,
$
k
dk,k ua,k ux,k
k0 , k1 , ..., kn

q1
ak0 ak1 aknn xk00 xk11 xknn
k0 , k1 , . . . , kn 0 1
= (a0 x0 + a1 x1 + + an xn )q1
and so
(21.8)
M = U DU T .
We now observe that

(21.9)
the square matrix U is nonsingular.

For suppose a vector v = ck : k {0, 1, 2, . . . , q1}n+1 satises U vT = 0; then
$
ck0 ,k1 ,...,kn xk00 xk11 xknn = 0
k0 , k1 , ..., kn
. This forces all coecients to vanish, i.e. ck = 0, which

for all x = (x0 , x1 , . . . , xn ) Fn+1
q
proves (21.9).
Combining (21.7) through (21.9), the p-rank of A equals
the number
of (n+1)-tuples
q1
k = (k0 , k1 , . . . , kn ) such that the multinomial coecient k0 , k1 , ..., kn is not divisible by p.

The answer is found using Lucas Theorem: Consider the base p representations
q1 = (p1) + (p1)p + (p1)p2 + + (p1)pr1 ;
ki = ki0 +ki1 p+ki2 p2 + +ki,r1 pr1 ,
We have

q
1
p
k0 , k1 , . . . , kn
kij {0, 1, 2, . . . , p1}.

#
p
1
i p
k0j , k1j , . . . , knj
0j<r
i for all j {0, 1, 2, . . . , r1} we have k0j +k1j + +knj = p1.

choices of (n+1)-tuple (k0j , k1j , . . . , knj )
By Theorem 21.6, for each j there are p1+n

n
r
n+1
such that k0j +k1j + +knj = p1; this gives exactly p1+n
{0, 1, 2, . . . , p1}
n
22. The Pl
ucker Map
135
choices of (k0 , k1 , . . . , kn ) {0, 1, 2, . . . , q1}n+1 such that p k0 , kq1

. The con1 , ...,kn
clusion of Theorem 21.1 follows.
Exercises 21.
1. A frame in Pn (F ) is a set of n+2 points, no n+1 of which lie in a hyperplane. (Thus for example a
frame in P2 (F ) is a quadrangle: 4 points with no three collinear; see also Theorem 19.4.) If A is a
point-hyperplane incidence matrix for Pn (F2 ), show that the rows of A corresponding to a frame, form
a basis for the row space of A over F2 .
Hint.
Use Theorem 21.1.
22. The Pl
ucker Map
The Pl
ucker map introduces coordinates for the k-subspaces of a nite-dimensional vector
space over an arbitrary eld F . We introduce this topic by discussing the coordinatization
of lines in P3 (F ). This important special case leads naturally to the Klein correspondence
(Section 25).
22.1 Lemma. Let F be an arbitrary eld. For every 2-subspace U < F 4 there is a
unique (up to nonzero scalar multiple) skew-symmetric 4 4 matrix over F with row
space equal to U. Such a matrix is given by uTv vTu where {u, v} is a basis for U.
Proof. Let U = {u, v} < F 4 be a 2-subspace. There exists F 4 such that uT = 0
and vT = 1, so (uTv vTu) = u lies in the row space of uTv vTu; similarly v lies in its
row space. Thus the row space of uTv vTu contains U = u, v and so must equal U.
If {.
u, v.} is also basis for U then there exists a unique A GL2 (F ) such that u
., .
v are
the rows of LA, where L is the 2 4 matrix with rows u, v. Now
0
. = LA
u
. v. v. u
1
T

T
1
0 1
T T
T
T
A L = (det A)L
L = (det A) u v v u
0
1 0
so the matrix uTv vTu changes only by a nonzero scalar multiple.

Finally let M be any skew-symmetric 4 4 matrix whose row space is U; we must
show that kM = xTy yTx for some nonzero k F and some basis {x, y} of U. We may
suppose
0
1
a b
c d
1 0
M =
a c 0 e
b d e 0
136
for some a, b, c, d, e F since some entry of M is nonzero, and we may multiply M by a

nonzero scalar to make this entry 1. (The argument is similar if the nonzero entry of M
is in any other position.) Since the rank of M is 2, we have
0
1 b
0 = det 1 0 d = e ad + bc
a c e
and so M = xTy yTx where x, y are the rst two rows of M.
If = u, v is any line (i.e. vector 2-space) of F 4 then the corresponding 4 4 matrix
as in Lemma 22.1 has the form
0

(01 , 02 , 03 , 12 , 13 , 23 ) := 01
02
03
01
0
12
13
02
12
0
23
03
13
T
T
=u vv u
23
0
where we abbreviate an arbitrary skew-symmetric 4 4 matrix by the 6-tuple of its abovediagonal entries; here
u
ij = ji = det i
vi
uj
vj

= ui vj uj vi .
ucker coordinates of the line . Denote by

We refer to (01 , 02 , 03 , 12 , 13 , 23 ) as the Pl
S the vector 6-space of all skew-symmetric 4 4 matrices over F . By Exercise #1, every
nonzero element of S has rank 2 or 4; accordingly we distinguish the points of S as having
either rank 2 or rank 4. The Pl
ucker map is the bijection
: {lines of P3 (F )} {rank 2 points of PS}
T
= u, v (01 , 02 , 03 , 12 , 13 , 23 ) = u v v u

using our abbreviation of elements of S as 6-tuples.
137
22. The Pl
ucker Map
22.2 Theorem. (i) The Pl

ucker map is a bijection from the set of lines in P3 (F )
and the set of rank 2 points of PS = P5 (F ).
(ii) The Pl
ucker coordinates of are also determined (to within a nonzero scalar
multiple) by any two distinct planes containing (as above) by the formula
ij = ji = ak bm am bk
where (i, j, k, m) is an even permutation of (0, 1, 2, 3).
(iii) A point z = (z0 , z1 , z2 , z3 ) lies in i

z0
0
23 13 12

0

23
03
02 z1
= 0.
13 03
0
01
z2
12 02 01
0
z3
(iv) A plane = (c0 , c1 , c2 , c3 )
0
01
0
01
02 12
03 13
contains i

c0
02 03
12 13 c1
= 0.
0
23
c2
23 0
c3
(v) The Pl
ucker coordinates satisfy 01 23 02 13 + 03 12 = 0.
Proof. Conclusion (i) has already been veried. Consider now an arbitrary line = x, y
as above. A point z = (z0 , z1 , z2 , z3 ) lies on i the matrix
x0
y0
z0
x1
y1
z1
x2
y2
z2
x3
y3
z3
has rank 2, which is to say that all four 3 3 submatrices (formed by deleting one column
at a time) are singular; for example this implies that
x1
0 = det y1
z1
x2
y2
z2
x3
y3 = 23z1 13 z2 + 12 z3 .
z3
By similarly expanding determinants for all 3 3 submatrices, we obtain (iii). Also since
the columns of xTy yTx lie in , they must satisfy the condition (iii); this yields (v).
In view of (iii) we are led to consider the linear transformation R : S S dened by
0
c
R : 01
c02
c03
c01
0
c12
c13
c02
c12
0
c23
c03
0
c13
c23

c23
c13
0
c12
c23
0
c03
c02
c13
c03
0
c01
c12
c02
.
c01
0
138
Thus for every line = x, y we have

(22.3)
= ker R(xTy yTx) = row space of xTy yTx.
Here we may identify row and column space via the transpose map, since xTy yTx is
skew-symmetric; similarly
T
ker R(x y y x) = left null space of R(x y y x)

is identied with the right null space via the transpose map. A plane contains i
i (xTy yTx)T = 0, which yields (iv).
Assuming = = , where , F 4 are linearly independent, from
(22.3) we have
T
, = ker R( ) = row space of

and so
= , = ker( ) = row space of R( ).
Since S has a unique member (up to nonzero scalar multiple) with row space equal to ,
we have
T
T
T
T
R( ) = k(x y y x)
for some nonzero k F . This yields (ii).
As promised, we now describe the more general Pl
ucker map which coordinatizes vector k-subspaces of an n-space; that is, projective
(k1)-subspaces of Pn1 (F ). Associated
n
n
to a k-subspace U F one obtains an k -tuple of coordinates ui1 i2 ik F for all subucker coordinates,
scripts satisfying 1 i1 < i2 < < ik n. The resulting vector of Pl
which may be computed from any choice of basis for U, is well-dened (i.e. independent
of the choice of basis) up to
nonzero scalar multiple. This vector determines a point (U)
ucker map. The image of this map in genin PN 1 (F ) where N = nk ; here is the Pl
eral is a (typically small) subset of the points of PN 1 (F ) known as a Grassmannian
or Grassmann variety (or rather the set of F -rational points of this variety, if F is not
algebraically closed). In the case k = 2 and n = 4 the Pl
ucker map associates to every line
(vector 2-subspace) < F 4 the point
() = (ij )0i<j3
in P5 (F ). In this case the Grassmann variety is the point set in P5 (F ) dened by the single
equation 01 23 02 13 +03 12 = 0, which is a quadric. More general Grassmann varieties
(for larger values of n and k) are not quadrics, but can be described as intersections of
quadrics.
139
23. Quadratic Forms
The Pl
ucker map in the full generality we are describing here is most conveniently
described in the language of exterior algebra (see Appendix A4): one simply maps a
/k
/k n
U
F . This gives a wellvector k-subspace U F n to the vector 1-subspace
/k
dened map from the projective (k1)-subspace
PU to the point (U) = P( U) in

/
k
Pk ( F ) = PN 1 (F ) where N = nk . If one species U by choosing a k n matrix A
whose row space is U, then the coordinates of (U) are the k k minors of A; there are N of
these. Although the choice of matrix A is not uniquely determined by U, dierent choices
of A will give the same N -tuple of coordinates of (U), up to nonzero scalar multiple. This
generalizes the case for k = 2 and n = 4 described above,
wherethe Pl
ucker coordinates

of = u, v are just the 2 2 minors of the matrix uv11 uv22 uv33 uv44 .
Exercises 22.
1. Let F be an arbitrary eld. Show that every skew-symmetric matrix overF has the form BABT where
0 1
or [0]. Conclude that
B is invertible and A is block-diagonal with blocks of the form either 1
0
every skew-symmetric matrix has even rank.
Hint. Let M be a skew-symmetric matrix over F . If A is nonzero, show that a sequence of elementary
operations can be performed simultaneously on the
rows and columns of M , transforming it to a
0 1
block-diagonal matrix with upper-left corner 1
. Apply induction.
0
2. Show that for every skew-symmetric 4 4 matrix M = (cij : 0 i, j 3), we have det M =
(c01 c23 c02 c13 + c03 c12 )2 . Conclude that the image of the Pl
ucker map of Theorem 22.2 consists
of all points of PS satisfying the necessary condition of conclusion (v).
Hint. Use Exercise #1.
23. Quadratic Forms

After studying subspaces of a projective space (these being the solutions of linear systems,
hence varieties of degree 1) the next natural objects of study are quadrics (varieties of
degree 2), these being the zeroes of quadratic forms. For example, quadrics in the classical
projective planes are simply conics, which we have previously considered. Insight into the
geometry of more general quadrics is founded upon a solid understanding of quadratic
forms. In this section we lay the algebraic foundations for quadratic forms; and in the
next section we study more geometric questions concerning the associated quadrics. Some
of this terminology was previously introduced in Section 14; here it is presented in more
generality.
Let V = F n where the eld F is for the moment arbitrary. A quadratic form on V
is a map Q : V F dened by a homogeneous polynomial of degree 2 in the coordinates.
Equivalently, for all x, y V and c F we have
(Q1) Q(cx) = c2 Q(x);
(Q2) Q(x + y) = Q(x) + Q(y) + B(x, y)
where B(x, y) is a symmetric bilinear form. Note that the bilinear form B is uniquely
determined by the quadratic form Q using the polarization identity (Q2). Every quadratic
form, and its associated bilinear form, may be written as
T
Q(x) = xAx ;
B(x, y) = x(A + A )y
140
for some nn matrix A = aij over F .

Let V and W be vector spaces over a eld F , and let Q : V F and Q : W F be
quadratic forms. An isometry from (V, Q) to (V , Q ) (or from Q to Q ) is a vector space
isomorphism : V V such that Q ((v)) = Q(v) for all v V . (Strictly speaking,
such a map is a linear isometry; more general isometries are invertible semilinear transformations preserving the quadratic form.) The relation of isometry is the most natural
equivalence relation on the set of quadratic forms. The isometries from (V, Q) to itself
form a group, called the orthogonal group O(V, Q). By virtue of (Q2) we see that every isometry O(V, Q) is also an isometry of B, i.e. B((x), (y)) = B(x, y) for all
x, y V . The converse fails in characteristic 2 as we observe below.
A typical vector space isomorphism : V V has the form x xM where M
GLn (F ). For Q(x) = xAxT we have
T T
Q((x)) = Q(xM) = xMAM x
so that is an isometry from Q to the quadratic form Q dened by the matrix MAM T . We
say that two n n matrices A, B are congruent if B = MAM T for some M GLn (F ).
This is an equivalence relation on the set of all n n matrices over F , and congruent
matrices represent isometric quadratic forms.
For each subspace U V we dene U = {v V : B(u, v) = 0 for all u U}.
The radical of U is U U; in particular the radical of V is simply V . We say that
the bilinear form B is nondegenerate if V = 0; in this case elementary linear algebra
shows that dim U + dim U = dim V for all U V . In general however the subspace
U need not be disjoint from U; the subspaces U and U may overlap considerably, and
even coincide. Note that B is nondegenerate i the matrix A + AT is nonsingular. A point
u is called singular if Q(u) = 0; or nonsingular if Q(u) = 0. A subspace U V
is totally singular if Q(u) = 0 for all u U. We say that a subspace U is totally
isotropic if B(u, u ) = 0 for all u, u U, i.e. U U . Every totally singular subspace is
totally isotropic; the converse holds only in characteristic = 2. We say Q is degenerate
if the radical V contains a singular point; otherwise Q is nondegenerate. Thus if B is
nondegenerate, so is Q; the converse holds in odd characteristic. If Q is nondegenerate
then
01
1
by preceding remarks, every totally singular subspace has dimension at most 2 dim V .
A maximal totally singular subspace is a totally singular subspace U which is not
contained in any larger totally singular subspace.
Assume V is a vector space with basis {v1 , v2 , . . . , vn } over a eld F . The discriminant of a quadratic form Q : V
the determinant of the Gram
F , denoted disc Q, is
T
matrix B(vi , vj ) : 1 i, j n . Assuming Q(x) = xAx , we have
T
disc Q = det(A+A ).
The discriminant of Q is dened only up to multiplication by a nonzero square in F , since
a change of basis for V (using a change-of-basis matrix M, say) replaces A by MAM T ,
and we obtain
T
T
T
det(M(A + A )M ) = (det M)2 det(A + A ).
23. Quadratic Forms
141
With this understanding, the discriminant depends only on the isometry type of Q. Note
that if F is a nite eld of characteristic 2 then every element of F is a square so the
discriminant contains no useful information. However if F = Fq where q is odd then half
the nonzero elements of F are squares and the other half are nonsquares; so assuming
Q is nondegenerate, the value of disc Q contains one bit of information, i.e. square or
nonsquare.
Let V1 , V2 , . . . , Vr be vector spaces over F with corresponding quadratic forms Qi :
Vi F . The orthogonal direct sum of the Vi s is the vector space V = V1 V2 Vr
equipped with the quadratic form
Q(v1 + v2 + + vr ) = Q1 (v1 ) + Q2 (v2 ) + + Qr (vr ) for vi Vi .
Such an orthogonal direct sum is often written V = V1 V2 Vr . If Qi is dened
by a square matrix Ai , then Q is dened by the block-diagonal matrix A1 A2 Ar
and
disc Q =
disc Qi =
det(Ai +ATi ) .
1ir
1ir
23.1 Real Quadratic Forms. Consider a quadratic form Q : Rn R given by Q(x) =

xAxT where A is a real nn matrix. We may suppose A is symmetric; otherwise replace
A by its symmetric part 12 (A + AT) without changing the value of Q(x) (because xATxT =
xAxT). Recall that every real symmetric matrix is diagonalizable by an orthogonal change
1
of basis, so that A = MDM T where M T = M and D = diag(1 , 2 , . . . , n ). Now
Q(xM) = 1 x21 + + n x2n .
By a further change of basis we may assume all i {1, 0, 1}; for if i = 0 we replace
xi |i |1/2 xi . Finally we may permute coordinates to see that Q(x) is isometric to
x21 + + x2k x2k+1 x2m
where 0 k m n. The number of isometry classes of quadratic forms on Rn

is the number of pairs of integers (k, m) satisfying these constraints, namely n+2
=
2
1
2 (n+1)(n+2). The form Q is nondegenerate i A is nonsingular, i m=n. There are n+1
isometry classes of nondegenerate quadratic forms on Rn , ranging from positive denite
(k=m=n) to negative denite (k=0, m=n). In the nondegenerate case Q(x) = x21 + +
x2k x2k+1 x2n has maximal totally singular subspaces of dimension min{k, nk}
0n1
; an example of such a subspace is the subspace spanned by the vectors ei +ek+1 for
2
i = 1, 2, . . . , min{k, nk}, where {e1 , . . . , en } is the standard basis of Rn .
23.2 Odd Characteristic. Suppose F has characteristic dierent from 2 (possibly characteristic zero). We will assume (as we may, for the same reasons as in the real case) that
142
A is symmetric. We may recover the quadratic form Q from the associated bilinear form
since
B(x, x) = Q(2x) 2Q(x) = 4Q(x) 2Q(x) = 2Q(x)
from which we obtain
Q(x) = 12 B(x, x).
From this identity it follows that isometries of the bilinear form B are the same as isometries
of the quadratic form Q: if GL(V ) satises B((x), (y)) = B(x, y) for all x, y V
then Q((x)) = 12 B((x), (x)) = 12 B(x, x) = Q(x) and the converse follows from (Q2).
Also in this case a subspace U V is totally isotropic i it is totally singular. Moreover
Q is nondegenerate i B is, i A is nonsingular.
23.3 Even Characteristic. Suppose F has characteristic 2. In this case we cannot
assume A is symmetric; when convenient, however, we may assume A is upper triangular.
We obtain B(x, x) = 2Q(x) = 0 and so the associated bilinear form is alternating; in
other words the matrix A+AT associated to B is skew-symmetric with zero diagonal. By
Exercise #22.1, the matrix A+AT has even rank, so it cannot be nonsingular unless n is
odd. We cannot recover the quadratic form from the bilinear form. For example when
n = 2, the two forms Q1 (x1 , x2 , x3 ) = x21 +x2 x3 and Q2 (x1 , x2 , x3 ) = x2 x3 give rise to the
same degenerate bilinear form B(x, y) = x2 y3 + x3 y2 . Both have radical (1, 0, 0) but in
the case of Q1 this radical is nonsingular, whereas in the case of Q2 the radical is singular.
Thus Q1 is nondegenerate whereas Q2 is degenerate. Note that the quadric dened by Q2
(i.e. the conic dened by the zeroes of Q2 ) is a pair of intersecting lines (0, 1, 0) and
(0, 0, 1) , a clearly degenerate conic; the point (1, 0, 0) of intersection is the singular
radical. On the other hand Q1 denes a nondegenerate conic; in this case the nonsingular
radical (1, 0, 0) is the nucleus of the conic.
Most questions about general quadratic forms reduce to questions about the nondegenerate case. This reduction follows from the following observation.
23.4 Lemma. Let V be a vector space over a eld F , and let Q : V F be a
degenerate quadratic form whose radical V contains a totally singular subspace U.
Then Q naturally induces a quadratic form on V /U .
. : V /U F by v + U Q(v). Note that this is wellProof. We naturally dene Q
dened: for all v V and u U,
Q(v + u) = Q(v) + Q(u) + B(v, u) = Q(v).
. satises the properties (Q1) and (Q2), using the similar properOne easily checks that Q
ties for Q.
23. Quadratic Forms
143
Classication of quadratic forms proceeds by induction using the following lemma.

23.5 Lemma. Let V be a vector space over a eld F of nite dimension n, and let
Q : V F be a quadratic form. Suppose x is a singular point not contained in the
radical V . Then
(a) There exists a singular point y such that V = x, y x, y and
Q(ax + by + z) = ab + Q(z)
for all a, b F and z x, y .
(b) The radical of x, y equals V . In particular Q is nondegenerate i the
restriction of Q to the (n2)-space x, y is nondegenerate.
A pair of vectors {x, y} such that Q(x) = Q(y) = 0 and B(x, y) = 1, as above, is a
hyperbolic pair.
Proof of Lemma 23.5. Since x
/ V , there exists y V such that B(x, y) = 0. For all
c F,
Q(cx + y) = c2 Q(x) + Q(y) + cB(x, y) = Q(y) + cB(x, y),
and since B(x, y) = 0, there is a unique c F such that Q(cx + y) = 0. Set y = cx+y.
1
Now x and y are singular points with B(x, y ) = 0. Set y = B(x, y ) y so that
Q(x) = Q(y ) = 0; B(x, y ) = 1. With no loss of generality, we may assume that y = y;
thus
Q(ax + by) = a2 Q(x) + b2 Q(y) + abB(x, y) = ab
for all a, b F . Clearly x, y x, y = 0 and so
V = x, y x, y .
If z, x, y then
Q(ax + by + z) = Q(ax + by) + Q(z) + B(ax + by, z) = ab + Q(z)
so (a) holds. Conclusion (b) is clear.
Every quadratic extension E F of nite elds1 (so that [E : F ] = 2) naturally gives
rise to a quadratic form E F , as we now describe. There exists a nonidentity eld
automorphism of E, which we denote z z, xing every element of F (in other words the
extension E F is Galois). The norm map
N : E F,
1
z zz
This much is true more generally for any separable extension of degree 2. It is the eld-theoretic analogue
of the fact from group theory that any subgroup of index 2 is normal.
144
is a quadratic form with no absolute points (since if z = 0 then N (z) = zz = 0). If we

choose a basis {, } for E over F , then
N (r + s) = (r + s)(r + s) = ar2 + brs + cs2
where aX 2 + bX + c F [X] is irreducible of degree 2. This fact has a useful converse, as
we now observe:
23.6 Lemma. A quadratic polynomial f(X) = aX 2 + bX + c Fq [X] is irreducible
over Fq i there exists a basis {, } for Fq2 over Fq , such that N (r + s) = ar2 +
brs + cs2 .
Proof. There exists Fq2 such that f() = 0. Since the norm map is surjective (see
Appendix A1), there exists Fq2 such that N () = a. Now
N (r s) = aN(r s) = a(r s)(r s) = ar2 + brs + cs2
as required. The converse follows by the preceding remarks.
We now consider the isometry group O(V, Q) of a quadratic form Q : V F , which
is typically quite large. The primary source for this assertion is the following result.
23.7 Theorem (Witt). Let Q : V F be a quadratic form, and let : U (U)
be an isometry between two subspaces of V , i.e. Q((u)) = Q(u) for all u U. Then
the following two conditions are equivalent.
(i) (U V ) = (U) V .

(ii) extends to an isometry of
V to itself, i.e. there is an isometry : V V
such that the restriction U = .
The proof of this result, given below, is rather technical; we recommend that on rst
reading, the student skip it. However, the reader is asked to consider some of the important
implications of this result, including the following.
23.8 Corollary. Let Q : V F be a nondegenerate quadratic form.
(a) Any two maximal totally singular subspaces have the same dimension.
(b) Any two singular points lie in the same number of maximal totally singular
subspaces.
23. Quadratic Forms
145
Proof. Let U, U V be maximal totally singular subspaces, and assume dim U dim U .
Let : U U be any injective linear transformation. Since Q vanishes on both U and U ,
is an isometry. By Witts Theorem 23.7 there exists an isometry : V V extending .

Since U is maximal, so is (U)
= (U) U , and this forces equality: (U) = U . In
particular dim U = dim (U) = dim U so (a) holds.
If x and y are singular points then we have an isometry : x y dened by
(cx) = cy for all c F . By Witts Theorem this extends to an isometry : V V
which induces a bijection between the maximal totally singular subspaces containing x,
and those containing y. This gives (b).
Proof of Theorem 23.7. Assuming (ii) then (i) must follow since the isometry : V V
must preserve the radical V . Thus we assume (i) and seek to prove (ii). We rst show
that without loss of generality,
(23.9)
U V and (U) V .
To show this, let W V such that V = W (UV ) = W ((U)V ); such a

subspace exists by Exercise 4. It follows that
U +V =U W
(U) + V = (U) W.
and
We extend to the map

U W (U) W,
u + w (u) + w
for all u U, w W . This is an isometry since

Q((u) + w) = Q((u)) + Q(w)
since B((u), w) = 0
= Q(u) + Q(w)
= Q(u + w)
since B(u, w) = 0
so we may replace U by U + V , and (U) by (U) + V if necessary, so that both U and

(U) contain the radical V . Furthermore we may assume the containment U V is
proper; otherwise the argument above applies with W = 0. So we may assume that (23.9)
holds.
We proceed to verify the conclusion by induction on k = dim(U/V ). The initial case
k = 0 was settled above; so now we take k 1. Let H < U be a subspace of codimension 1
containing V , so that the subspace (H) < (U) of codimension 1 is isometric to H and
also contains V . We may assume that
(23.10)
xes every element of H.
146

Otherwise by the inductive hypothesis, the restriction H : H (H) extends to an
1
isometry : V V . The isometry = : U (U) does x every element of H.
If we can extend to an isometry : V V , then the isometry : V V extends
: U (U). So it suces to assume (23.10) holds.
Now both U and (U) contain H as a subspace of codimension 1 xed elementwise
by . There exists z U such that
U = H z;
(U) = H (z).
We assume (z) = z; otherwise the conclusion follows. For all v H we have

B(z(z), v) = B(z, v) B((z), v) = B((z), (v)) B((z), v) = 0
since (v) = v; thus H y where y = z(z).
(23.11)
For all x U we have (x) x y;
this follows since x U = H y where xes every element of H. We may now assume
that
(23.12)
U, (U) y . In particular, y y .
For suppose U y . Then there exists x U such that

0 = B(x, y) = B(x, z) B(x, (z))
= B((x), (z)) B(x, (z)) = B((x)x, (z))
where (x) x y, so that (U) y . Now there exists a subspace W such that
y = H W . It follows that V = U W . (Since U W U y = H we have
U W H W = 0.) The mapping V V dened by u + w (u) + w (for u U,
w W ) maps U to (U) and is an isometry since
Q((u) + w) = Q((u)) + Q(w) + B((u), w)
= Q(u) + Q(w) + B(u, w)
= Q(u + w)
since W y . Thus we may assume (23.12). Next we may assume
(23.13)
(U) = U = y < V .
For if (U) = U then we may write y = U S and an extension of to y is given by

u + s (u) + s
147
23. Quadratic Forms
for all u U, s S; this is an isometry since

Q((u) + s) = Q((u)) + Q(s) + B((u), s)
= Q(u) + Q(s) + B(u, s)
since u(u) y S
= Q(u + s).
Replacing U by y gives (23.13) in this case.
On the other hand if (U) = U then we choose vectors d, e such that U = H d and
(U) = H e. There exists a subspace W such that y = (U + (U)) W . Consider
the subspace S = W d+e. Since U + (U) = U d+e = (U) d+e, we have
y = U S = (U) S.
As before, an isometry extending to y is given by u + s (u) + s for all u U, s S.
Once again we may replace both U and (U) by y ; so in any case we may assume that
(23.13) holds.
Since B(z, (z)) = B(z, zy) = B(z, z), we have
Q(y) = Q(z(z)) = Q(z) + Q((z)) B(z, (z)) = 2Q(z) B(z, z) = 0.
There
exists y H such that {y, y } is a hyperbolic pair, by Lemma 23.5 applied

to QH ; similarly there exists y H such that {(y), y } is a hyperbolic pair. Now
V = H y, y = H (y), y
where y, y = (y), y = H. An extension of to V is given by
ay + by + u a(y) + by + u
for all a, b F , u H; this is an isometry since
Q(ay + by + u) = ab + Q(u) = Q(a(y) + by + u).
Exercises 23.
1. Classify all isometry classes of quadratic forms on Cn , as we have done for Rn . Justify your answer.
2. Let Q : V F be a quadratic form. Show that if O(V, Q) is an isometry and (U ) = U for some
subspace U V , then (U ) = U .
3. Let Q : V F be a quadratic form with associated bilinear form B, and suppose u V such that
B(u,v)
Q(v) = 0. Dene Ru : V V by v v Q(v) v. Show that Ru is an isometry xing every element
of u and mapping u u.
4. Let V be an n-dimensional vector space over F and let U, U V be k-dimensional subspaces. Show
that there exists a subspace W V such that V = U W = U W .
148
Hint: Let v1 , . . . , vr be a basis for U U . Extend to a basis v1 , . . . , vr , vr+1 , . . . , vk for U and a

, . . . , vk for U . Show that the union of the preceding bases gives a basis for
basis v1 , . . . , vr , vr+1
U + U , and adjoin new vectors vk+1, . . . , vn to obtain a basis for V . Consider the subspace spanned

, . . . , vk +vk , vk+1 , . . . , vn .
by vr+1 +vr+1
5. Let Q : V F be a quadratic form on a 2-dimensional vector space V . Show that either 0, 1, 2 or
all of the points of V are singular. Give examples to show that each of these four possibilities may
actually occur. Can all these possibilities occur for every choice of eld F ? Explain.
24. Quadrics and Polar Spaces

Let V be an n-dimensional vector space over a eld F . A quadric in PV = Pn1 (F ) may
be dened as the set of singular points of a quadratic form Q : V F . Such a quadric is
nondegenerate if Q is. Note that a quadric in a projective plane is simply a conic. Three
familiar quadrics in real projective 3-space are the elliptic and hyperbolic quadrics, and the
quadratic cone, as shown below. Note that maximal subspaces lying on the quadric (i.e.
maximal totally singular subspaces) are points in the case of the elliptic quadric; or lines
in the case of the hyperbolic quadric or the cone.
x21
Elliptic Quadric
+ x22 + x23 x24 = 0
Hyperbolic Quadric
x1 x2 + x3 x4 = 0
x22
Cone
+ x3 x4 = 0
These three quadrics are distinguished by the fact that the elliptic quadric contains no
lines; the cone contains innitely many lines, all passing through a common point (the
vertex of the cone); and the hyperbolic quadric contains two innite families of mutually
%
disjoint lines.& (see below). In fact the hyperbolic quadric contains a family of lines s :
s R {} (called a regulus) dened by
s = (1, 0, s, 0), (0, s, 0, 1) for s R;
and = (0, 1, 0, 0), (0, 0, 1, 0)
&
%
which partitions its points; and a second regulus s : s R {} , described as the
alternate regulus to the rst regulus, given by
s = (1, 0, 0, s), (0, s, 1, 0) for s R;
and = (0, 1, 0, 0), (0, 0, 0, 1)
149
The two reguli (plural of regulus) are as shown:
Second
Regulus
First
Regulus
Topologically, the elliptic and hyperbolic quadrics in P3 (R) are homeomorphic to the sphere
S 2 and the torus T 2 respectively.
If we disregard the equation x21 + x22 + x23 + x24 = 0 (which has no real solutions,
yielding the empty quadric) then every nondegenerate quadric in P3 (R) is equivalent to
one with equation either
x21 + x22 + x23 x24 = 0
or
x21 + x22 x32 x24 = 0.
The latter equation denes a hyperbolic quadric since an invertible linear change of variables
(x1 , x2 , x3 , x4 ) = (y1 +y2 , y3 +y4 , y1 y2 , y3 y4 )
transforms the equation x21 + x22 x23 x24 = 0 into the equation y1 y2 + y3 y4 = 0.
Just as a single equivalence class of nondegenerate conics in the real projective plane
gives rise to several inequivalent real ane (i.e. Euclidean) plane conics, namely ellipses,
parabolas and hyperbolas, so also our two nondegenerate quadrics in P3 (R) give rise to
many dierent quadrics in Euclidean 3-space A3 (R). For example the ane hyperbolic
paraboloid z = x2 y 2 , when homogenized by the addition of a fourth variable w, becomes
x2 y 2 zw = 0; then setting
(x, y, z, w) = (x1 , x3 , x4 x2 , x2 +x4 )
transforms the equation to x21 +x22 x23 x24 = 0. If we then set (x1 , x2 , x3 , x4 ) = (X, Y, Z, 1)
then we obtain the one-sheeted hyperboloid Z 2 = X 2 + Y 2 1.
We see from the classication above that Pn1 (R) has n2 equivalence classes of (nonempty) nondegenerate quadrics; note that this number tends to as n . By contrast,
in Pn1 (Fq ) there are just two types of nondegenerate quadrics for n even (the elliptic and
hyperbolic quadric) and one type of nondegenerate quadric for n odd (the parabolic type).
The main property of R which accounts for this dierence is the fact that R is an ordered
eld, whereas nite elds are not ordered.
150
We proceed to classify quadrics over nite elds. Here some aspects of the discussion
are more delicate for elds of even characteristic, and the beginning student may prefer to
ignore this case on rst reading. But elds of even characteristic are of such signicance
and interest that the extra eort to understand them proves rewarding. Regardless of
the characteristic we require a more complete description of the quadratic form and its
relationship to an associated bilinear form.
Note that singular points are simply points on the associated quadric; more generally,
totally singular subspaces are simply projective subspaces contained in the quadric.
This leads to the desired classication of quadratic forms on Fnq :
24.1 Theorem. Let V be an (n+1)-dimensional vector space over Fq , and let Q :
V Fq be a nondegenerate quadratic form. Fix a, b, c Fq such that the quadratic
polynomial aX 2 + bX + c Fq [X] is irreducible over Fq . If q is odd, x a nonsquare
Fq .
If n=2k then after a suitable linear change of coordinates, Q is given by either
(H) x1 x2 + x3 x4 + + x2k1 x2k ; or
(E) x1 x2 + x3 x4 + + x2k3 x2k2 + ax22k1 +bx2k1 x2k +cx22k .
If n = 2k+1 then after a suitable linear change of coordinates, Q is given by either
(P) x1 x2 + x3 x4 + + x2k1 x2k + x22k+1 ; or
(P ) x1 x2 + x3 x4 + + x2k1 x2k + x22k+1 . (This case arises only for q odd).
Quadratic forms as in (H) and (E), and the associated quadrics, are of hyperbolic and
elliptic type respectively. Quadrics as in (P) or (P ), and the associated quadrics, are of
parabolic) type. Denoting the quadratic forms in (P) and (P ) by Q and Q respectively,
we observe that Q is isometric to Q since
Q(x1 , x2 , x3 , x4 , , x2k1 , x2k , x2k+1 ) = Q(x1 , x2 , x3 , x4 , , x2k1 , x2k , x2k+1 )
and so the forms Q and Q have the same singular points; accordingly their corresponding
quadrics have the same geometric properties.
Proof of Theorem 24.1. If n = 1 then Q(x1 ) = cx21 where c Fq is nonzero. If c = t2 for
some nonzero t Fq then the change of variable y1 = tx1 yields cx21 = y12 and (P) holds.
Otherwise q is odd and c Fq is a nonsquare so that c = t2 for some nonzero t Fq ; but
then the change of variable y1 = tx1 yields cx21 = y12 so (P ) holds.
Suppose n = 2. If there is a singular point x then by Lemma 23.5 the quadratic
form may be written as Q(x1 , x2 ) = x1 x2 after a suitable coordinate change, so that
(H) holds. So we may assume there is no singular point. By Lemma 23.7 there exists a
basis {, } for Fq2 over Fq such that the norm map N : Fq2 Fq , z z q+1 satises
N (r + s) = ar2 + brs + cs2 . We may similarly identify V with Fq2 , and choose a basis
151
{ , } for Fq2 over Fq , in such a way that Q : V F satises Q(x1 , x2 ) = N (x1 +x2 ).
A change of basis from { , } to {, } shows that Q(x1 , x2 ) is equivalent to ar2 +brs+cs2 ,
and so (E) holds.
Finally suppose n 3. By the Chevalley-Warning Theorem A2.3 (Appendix 2), the
number of solutions (x1 , x2 , . . . , xn ) Fnq to the equation Q(x) = 0 is divisible by p where
q = pe ; since the zero vector is a solution, there must exist a nonzero solution. Thus there
exists a singular point u. By Lemma 23.5 there exists another singular point v such
that the restriction of Q to the (n2)-space u, v is nondegenerate, and
Q(x1 u + x2 v + z) = x1 x2 + Q(z)
for all x1 , x2 Fq and z u, v . The result follows by induction.
24.2 Corollary. Let Q : V Fq be a nondegenerate quadratic form on a nitedimensional vector space over a eld Fq where q is odd. Then the isometry type of Q is
uniquely determined by the quantities q, n= dim V , and disc Q. Assuming n = 2k, the
form Q is hyperbolic or elliptic according as (1)k disc Q is a square or a nonsquare.
Proof. Suppose rst that n = 2k. By Theorem 24.1, to within an isometry the matrix of
B is given by either

(H) 01 10 01 10 01 10
(k summands), or
a b/2

(k summands, where aX 2 +bX+c is irreducible
(E) 01 10 01 10 b/2
c
over Fq ).
The determinant of this matrix is (1)k in case (H), or (1)k (b2 4ac)/4 in case (E).
Since aX 2 +bXY + cY 2 is irreducible, its discriminant b2 4ac is a nonsquare in Fq and
the result follows.
Now suppose n = 2k+1. Again by Theorem 24.1, to within an isometry the matrix of
B is given by either

(P) 01 10 01 10 [1] (k+1 summands), or

(P ) 01 10 01 10 [] (k+1 summands)
where Fq is a nonsquare. The determinant of this matrix is (1)k in case (P), or
(1)k in case (P ). One of these values is a square and the other is a nonsquare, so once
again the result follows.
152
24.3 Theorem. Let V be an n-dimensional vector space over Fq , and let Q : V F

be a nondegenerate quadratic form. Suppose x is a singular point. Then Q naturally
induces a nondegenerate quadratic form on the (n2)-space x /x. This form has
the same type (hyperbolic, parabolic or elliptic) as the original space (unless n = 2
in which case x /x = 0).
Proof.
Since x is the radical of x , by Lemma 23.4 the map

. : x /x F,
Q
v + x Q(v)
is a well-dened quadratic form induced naturally by Q. There is no loss of generality in

assuming Q is given by one of the four canonical forms listed in Theorem 24.1; in particular
Q(x) = x1 x2 + Q1 (x3 , x4 , . . . , xn )
using coordinates (x1 , x2 , . . . , xn ) for V relative to some basis e1 , e2 , . . . , en , where Q1 has
the same type as Q (H, E, P or P ) but in 2 fewer variables. By Witts Theorem 23.7 there
is no loss of generality in assuming that x = e1 . Then x has basis {e1 , e3 , e4 , e5 , . . . , en }
and x /x has basis {ei +e1 : i = 3, 4, 5, . . . , n}. The quadratic form induced on x /x
is therefore Q1 .
24.4 Theorem. Let V1 and V2 be even-dimensional vector spaces over Fq . Then

the orthogonal direct sum V1 V2 is
(i) hyperbolic, if V1 and V2 are either both hyperbolic or both elliptic; or
(ii) elliptic, if one of V1 , V2 is hyperbolic and the other elliptic.
Proof.
Let dim Vi = 2ki . For q odd we have disc Q = (disc Q1 )(disc Q2 ) and so
(1)k1 +k2 disc Q = (1)k1 disc Q1 (1)k2 disc Q2
where (1)ki disc Qi = 1 or 1 according as Qi is hyperbolic or elliptic, and similarly

for Q. The result follows in this case.
Hence we may assume q is even. It clearly suces to assume k1 = k2 = 1 with both
Q1 and Q2 elliptic; the general case then follows by taking orthogonal direct sums with
suciently many hyperbolic lines. We may assume
Q1 (x1 , x2 ) = ax21 + bx1 x2 + cx22 ,
Q2 (y1 , y2 ) = ay12 + by1 y2 + cy22
where at2 +bt+c is irreducible over Fq ; note that abc = 0 in this case. Since
(24.5)
153
Q(x1 , x2 , y1 , y2 ) = ax21 +bx1 x2 +cx22 + ay12 +by1 y2 +cy22

= (x1 +y1 )(ax1 +ay1 +bx2 ) + (x2 +y2 )(cx2 +cy2 +by1 )
which has the form u1 u2 + u3u4 where the ui s are the latter parenthesized quantities in
(24.5) and the linear change of variables (x1 , x2 , y1 , y2 ) (u1 , u2 , u3 , u4 ) is nonsingular.
So (24.5) is a quadratic form of hyperbolic type as required.
Let Q : V Fq be a nondegenerate quadratic form where V is n-dimensional over
Fq . It is customary to denote the isometry type of (V, Q) as
+
(q) if n=2k and Q is hyperbolic;
O2k
(q) if n=2k and Q is elliptic; or

O2k
O2k+1 (q) if n=2k+1, in which case Q is parabolic.

We summarize the statement of Theorem 24.4 as
(q) O2k
O2k
(q) O2(k+k ) (q) .
24.6 Theorem. The number of

singular points and maximal totally
singular subspaces for each type of
nondegenerate quadratic form over
Fq are as listed in the table to the
right.
Proof.
For each a Fq , let Nk (a) be the set of vectors (x1 , x2 , . . . , x2k ) F2k
q such that
x1 x2 + x3 x4 + + x2k1 x2k = a.
It is easy to see that |Nk (a)| = |Nk (1)| for all nonzero a Fq using the bijection
Nk (1) Nk (a),
(x1 , x2 , x3 , x4 , . . . , x2k1 , x2k ) (x1 , ax2 , x3 , ax4 , . . . , x2k1 , ax2k ).
Writing nk = |Nk (0)|, it follows that |Nk (a)| =

(24.7)
q2k nk
q1
for a = 0. We claim that in fact
|Nk (0)| = nk = q 2k1 +q k q k1 ; and consequently |Nk (a)| = q k1 (q k 1)

for all nonzero a Fq .
154
Vectors in Nk (0) are partitioned into those solutions with x1 = 0 and x2 Fq arbitrary,
0, the
with nk1 choices of the remaining coordinates; and those solutions with x1 =
coordinates x3 , x4 , . . . , x2k arbitrary and x2 Fq uniquely determined. This gives
nk = qnk1 + (q 1)q 2k2 .
Taking either n0 = 1 or n1 = 2q1 as the initial case, (24.7) follows easily by induction.
We may assume the nondegenerate quadratic form Q is the standard representative
+
of its isometry class as given in Theorem 24.1. In the case of O2k
(q) the form Q(x) =
x1 x2 + x3 x4 + + x2k1 x2k has
nk 1
q k 1 k1
+ 1)
=
(q
q1
q1
singular points, arising from the nk 1 nonzero vectors in Nk (0). For O2k
(q) the form
2
2
2
Q(x) = x1 x2 + + x2k3 x2k2 + ax2k1 + bx2k1 x2k + cx2k (with at + bt + c irreducible
over Fq ) has
|Nk1 (0)| + (q 2 1)|Nk1 (1)| = q 2k1 q k + q k1
singular vectors, counting separately those with (x2k1 , x2k ) = (0, 0) or = (0, 0), giving
q 2k1 q k + q k1 1
q k1 1 k
=
(q + 1)
q1
q1
as the required number of singular points. Similarly in the O2k+1 (q) case Q(x) = x1 x2 +
+ x2k1 x2k + x22k+1 has
|Nk (0)| + (q 1)|Nk (1)| = q 2k
2k
1
singular vectors and hence q q1
singular points.
In each of the three cases we denote by mk the number of maximal totally singular
subspaces. This number is found by rst counting in two dierent ways the number of
pairs (x, U) where U is a totally singular k-space containing a point x. Consider rst
+
the case O2k
every such U is k-dimensional. There are mk choices of U, each
k(q)
in which
k
containing 1 q = (q 1)/(q 1) points. On the other hand there are
q k 1 k1
q
+1
q1
choices of singular point x; and for each, the totally singular k-subspaces U containing
x are in one-to-one correspondence with the mk1 totally singular (k1)-subspaces of the
+
(q)-space x /x. Thus
O2(k1)
q k 1 k1
qk 1
mk =
q
+ 1 mk1
q1
q 1
155
which yields the recurrence relation mk = (q k1 + 1)mk1 . Given that an O2+ (q)-space
has 2 singular points as its maximal totally singular subspaces, we have m1 = 2 and so by
induction
mk = 2(q + 1)(q 2 + 1) (q k1 + 1)
as claimed.
In the O2k
(q) case, maximal totally singular subspaces have dimension k1 and there
k
qk1 1
are q1 q + 1 choices of the singular point x, so the same reasoning gives
q k1 1 k
q k1 1
mk =
q + 1 mk1
q1
q 1
whence mk = (q k +1)mk1 in this case. Since O2 (q) has {0} as its unique maximal totally
singular subspace, we have m1 = 1 and so by induction
mk =
(q i + 1) = (q 2 + 1)(q 3 + 1) (q k + 1).
2ik
Finally in the O2k+1 (q) case, maximal totally singular subspaces have dimension k
and there are (q 2k 1)/(q 1) choices of the singular point x, so
q 2k 1
qk 1
mk =
mk1
q1
q1
which gives mk = (q k + 1)mk1 . Since the O3 (q)-quadric is a conic with q+1 points as its
maximal totally singular subspaces, we have m1 = q+1 so by induction
mk = (q + 1)(q 2 + 1) (q k + 1).
The incidence system formed by all totally singular subspaces with respect to a hy+
(q) quadric, are
perbolic quadratic form over Fq , i.e. the subspaces contained in an O2k
described by Coxeter-Dynkin diagrams of type Dk :
O4+ (q) quadric:
(type D2 )
O6+ (q) quadric:

(type D3 )
lines of type I
lines of type II
planes of type I
..
.....
.....
.....
.
.
.
.
....
..........
.....
.....
.....
.....
.....
points
planes of type II
156
O8+ (q) quadric:

(type D4 )
+
O2n
(q)
quadric:
(type Dn )
solids of type I
.
.....
.....
points lines..................
..........................................................
.....
.....
.....
...
solids of type II
n-spaces
(n3)- (n2)- ............ of type I
.
.
.
spaces spaces.........
points lines solids
............................................................................................................. .................................................................................
.....
.....
..... n-spaces
... of type II

n types of objects
As with the diagrams of type An for projective n-spaces, the Dn diagrams are endowed with
many layers of meaning. For example, the evident symmetry of the diagram represents
the existence of an automorphism interchanging the two types of totally singular n-spaces.
More insight gleaned from the Dn diagrams will emerge as we proceed to examine some
particular cases.
...
...
...
...
.....
................
................
................
24.8 The O4+ (q)-Quadric. The O4+ (q) quadric consists of
..
..
..
.........................
...
...
...
.
.....
................
................
................
........................
(q+1)2 points and 2(q+1) lines, forming a grid as shown.
...
...
...
.
.
.
................
.........
.............
.....
................
................
.
.
.
.
q+1
.
.
.
.
There are two types of lines, each type forming a regulus. Any
...
...
...
...
.. .. ..
.. lines
two lines in the same regulus are skew; but every line in one
.. .. ..
..
....
....
....
....
.
.
.
regulus meets every line in the opposite (alternate) regulus.
.....
...............
...............
............... .........
............
...
...
...
...
We say that two lines are incident if they meet in a point, i.e.

q+1 lines
if they are from opposite reguli. (This is a dierent denition
of incidence than the relation of inclusion used in projective
space!) Since every line of one regulus is incident with every line of the alternate regulus,
the geometry is simply the generalized digon that we have encountered previously. It is no
coincidence that the D2 diagram coincides with the A1 A1 diagram previously used for
this geometry.
The student may at rst wonder why the diagram contains no node representing
points of the quadric. In fact such a node is not needed, for reasons that we hope will
become clear in time. For now the reader may observe that the geometry (in this case,
generalized digon) formed by the two families of lines of the quadric has as its incidence
graph a complete bipartite graph Kq+1,q+1 , and that points of the quadric are adequately
represented as the (q + 1)2 edges of this graph.
+
+
24.9 The O2n
(q)-Quadric. In just the same way, the O2n
-quadric contains two types
2
n1
+ 1) of each type; this accounts
of totally singular n-spaces, in fact (q + 1)(q + 1) (q
for the factor of 2 appearing in the formula from Theorem 24.6. The geometry of the
quadric contains n types of objects: the totally singular subspaces of dimension 1, 2, . . . ,
n2, and the two types of totally singular n-subspaces. It does not matter which type
157
of totally singular n-subspace is designated as type I (and the other as type II) since
an automorphism of the geometry interchanges these two types. Every totally singular
(n1)-space U lies in exactly one totally singular n-subspace of each type, corresponding
to the two singular points of U /U O2+ (q). Accordingly we say that two totally singular
n-subspaces are incident if they intersect in such an (n1)-subspace, in which case the
two n-subspaces have opposite type. Again, the totally singular (n1)-subspaces do not
appear among the n types of objects of our geometry, because they are already represented
as incident pairs of totally singular n-subspaces.
For objects of dimension k = k in our geometry, incidence is inclusion just as in
the case of projective spaces. Given a totally singular k-subspace U, the residue of U
is the subgeometry formed by the objects incident with U. This geometry has diagram
obtained from the Dn diagram by deleting the corresponding node. Consider rst the case
k n2, in which the resulting diagram is of type Ak1 Dnk . The objects of this
residual geometry are the subspaces of U, forming the projective space PU Pk1 (Fq ); and
the totally singular subspaces containing U. The latter correspond to the totally singular
+
(q), these forming a geometry of type Dnk . Furthermore
subspaces of U /U O2(nk)
every object of the Ak1 -geometry is incident with every member of the Dnk -geometry,
as is represented by the fact that there are no edges joining the corresponding subgraphs.
In the case k = n, the residue of a totally singular n-space U form a geometry of type
An1 , as we should expect since this is the diagram obtained from the Dn diagram by
removing the node corresponding to U. In fact the objects of the residual geometry are the
subspaces of U of dimension 1, 2, . . . , n2; and the totally singular n-subspaces U < V
incident with U. As we have said, such subspaces U are completely determined by the
intersections U U, these being all the hyperplanes of U; so the residue of U is clearly
isomorphic to PU Pn1 (Fq ) as expected.
We devote Section 25 to the study of the important special case of the O6+ (q)-quadric
(the Klein quadric). The other particularly fascinating special case is that of O8+ (q)quadric, the triality quadric, so-called because it admits a symmetry of order 3 mapping
singular
totally singular
totally singular
singular
points
solids of type I
solids of type II
points
while mapping totally singular lines to totally singular lines. Regrettably, we will probably not have time to explain how such an automorphism is dened. Unlike the duality
automorphism of projective spaces, which exists in every dimension, the triality automorphism is unique to the O8+ (q)-quadric. Together with the automorphisms interchanging the
two types of totally singular solids, we see that the automorphism group of the geometry
induces the full symmetry group S3 of the D4 diagram.
24.10 Finite Classical Polar Spaces. We describe three types of nite classical polar
spaces embedded in a nite-dimensional vector space V over F = Fq . A polarity of PV
is a map of order 2 (so that (X ) = X) interchanging points with hyperplanes, lines
158
with subspaces of codimension 2, etc., while preserving (or rather, reversing) incidence:
X1 X2 i X2 X1 . A subspace X V is absolute if X X . All objects of
the associated polar space, are nonzero subspaces of V which are absolute with respect to
the associated polarity. The three types of polarities (orthogonal, symplectic and unitary)
give rise to the three types of polar spaces. Although the orthogonal polar spaces are the
most complicated of the three to dene, we describe them rst since most people nd them
closer to their experience.
Orthogonal Polar Spaces. Let Q : V F be a quadratic form, with associated
bilinear form B; thus for all x, y V and a, a F ,
Q(x + y) = Q(x) + Q(y) + B(x, y);
B(ax+a x , y) = aB(x, y) + a B(x , y).
We require the form Q to be nondegenerate; the meaning of this requirement is recalled
below. The orthogonal polarity is dened by
X = {v V : B(x, v) = 0 for all x X}, where X V .
Consider rst the case that q is odd. Here we require that B be nondegenerate, i.e.
V = 0. The objects of the polar space are the nonzero subspaces X V such that
X X . These are just the nonzero subspaces lying in the associated quadric.
For general q we require only that Q be nondegenerate, which means that V
contains no singular points. The objects of the polar space are the (nonzero) totally
singular subspaces X V , i.e. those which satisfy Q(x) = 0 for all x X. These
are simply the nonzero subspaces contained in the associated quadric. (For q odd this
description is equivalent to that given in the preceding paragraph. But for q even, the
bilinear form B is alternating, so the nonzero subspaces satisfying X X give the
symplectic polar space described below. In particular for q even, the absolute points
with respect to B include all points of PV , not just the points of a quadric.)
In most cases, we dene two objects of the polar space to be incident if, as
+
(q)-quadrics,
subspaces of V , one properly contains the other. In the case of O2n
however (see Section 24.9), we make an exception: two totally singular subspaces of
dimension n2 are incident i they intersect in a subspace of dimension n2 1. Moreover
in this case, totally singular subspaces of dimension n2 1 are not properly considered
as objects of the polar space; instead they are identied as incident pairs of totally
singular n-spaces.
+
(q), of parabolic type
We have orthogonal polar spaces of hyperbolic type O2n
O2n+1(q), and of elliptic type O2n (q), according to the type of the corresponding
+
(q) belong to the Coxeter-Dynkin
quadratic form. Orthogonal polar spaces of type O2n
diagram of type Dn :
points lines solids
.............................................................................................................
(n3)spaces
n-spaces
...
(n2)- ............. of type I
.
spaces ........
........................................................................
.....
.....
.....
.....
.....
n-spaces
of type II
159
as described above. Polar spaces of type O2n+1(q) belong to the Coxeter-Dynkin

diagram of type Bn or Cn :
points lines planes
.............................................................................................................
(n1)-
spaces
n-spaces
..............................................................................................................
(The distinction between the Coxeter-Dynkin diagrams of type Bn and Cn arises when
considering root systems and Lie algebras, but is not an issue here.) The meaning of
the double bond between totally singular (n1)-spaces and totally singular n-spaces,
(q) belong to the Coxeterwill be explained in Section 28. Polar spaces of type O2n
Dynkin diagram of type Bn1 :
points lines planes
.............................................................................................................
(n2)-
spaces
(n1)-
spaces
..............................................................................................................
An invertible linear transformation g GL(V ) is a (linear) isometry of the quadratic form Q if Q(xg ) = Q(x) for all x V . The set of all such isometries is the
+
(q), O2n+1 (q) or O2n

(q) according to the type of the
orthogonal group, denoted O2n
quadratic form, and the eld order q. This group acts on the polar space. But not
necessarily faithfully; for q odd, the normal subgroup {I} of order 2 acts trivially.
+
(q) case, only half the isometries are type-preserving; the other
Moreover in the O2n
half interchange the two types of maximal totally singular subspaces.
Symplectic Polar Spaces. Let B : V V F be an alternating bilinear form;
thus for all x, x , y V and a, a F ,
B(ax+a x , y) = aB(x, y) + a B(x , y);
B(x, x) = 0 and B(y, x) = B(x, y).
For every subspace X V we dene
X = {v V : B(x, v) = 0 for all x X}.
We require that B be nondegenerate; that is, V = 0. This condition requires that
dim V = 2n be even, as follows from Exercise #22.1 by considering a matrix dening B. We refer to such a nondegenerate alternating bilinear form as simply a symplectic form. There is a version of Witts Theorem 23.7 for symplectic forms. This
has numerous consequences similar to Corollary 23.8; for example, all maximal totally
isotropic subspaces have the same dimension (in this case, n). The symplectic polarity is the map X X . The symplectic polar space Spn (q) has as its objects
the (nonzero) totally isotropic subspaces with respect to B, i.e. those nonzero subspaces X V such that X X . This polar space belongs to the Coxeter-Dynkin
diagram of type Bn (shown above).
An invertible linear transformation g GL(V ) is a (linear) isometry of the
alternating form B if B(xg , y g ) = B(x, y) for all x, y V . The set of all such isometries
160
is the symplectic group Spn (q). This group acts on the symplectic polar space. But
not necessarily faithfully; for q odd, the normal subgroup {I} of order 2 acts trivially.
Unitary Polar Spaces. Here we must assume the eld order q = q02 . The quadratic
extension Fq Fq0 has an automorphism of order two given by x xq0 . Let H :
V V F be a Hermitian form; thus for all x, x , y V and a, a F ,
H(ax+a x , y) = aH(x, y) + a H(x , y);
H(y, x) = H(x, y)q0 .
For every subspace X V we dene
X = {v V : H(x, v) = 0 for all x X}.
Note that H is linear in its rst argument, but semilinear (conjugate linear) in its
second argument:
H(x, ay+a y ) = aq0 H(x, y) + (a )q0 H(x, y );
so we say H is sesquilinear, i.e. 1 12 -linear. We require that H be nondegenerate;
that is, V = 0. There is also a version of Witts Theorem 23.7 for Hermitian
forms. So once
all maximal totally isotropic subspaces have the same dimension,
1
0 magain,
namely n = 2 where m = dim V . The unitary polarity is the map X X . The
unitary polar space Un (q0 ) has as its objects those nonzero subspaces X V such
that X X . This polar space belongs to the Coxeter-Dynkin diagram of type Bn
(shown above).
An invertible linear transformation g GL(V ) is a (linear) isometry of the
Hermitian form H if H(xg , y g ) = H(x, y) for all x, y V . The set of all such isometries
is the unitary group Un (q0 ). This group acts on the unitary polar space. The normal
subgroup {I : q0 +1 =1} of order q0 +1 acts trivially.
In each case if one wants the full automorphism group of the given polar space,
one must consider semilinear maps rather than just linear maps. Moreover one must
consider not only isometries (which actually preserve the relevant form) but similarities,
which scale the form by a nonzero constant; thus for example in the orthogonal case
Q(xg ) = cg Q(x) for all x V , where cg F
q may depend on g but not on x.
Polar spaces, like projective spaces, may be dened axiomatically using just points
and lines; and just as one nds with projective spaces (Section 20), with exceptions only
in small dimensions, all nite polar spaces turn out to be the classical ones listed above.
We feel the beginning student should study the classical examples before worrying about
to the axiomatic foundations.
Exercises 24.
1. Let V be the vector 4-space over a eld F , and let Q : V F be the determinant map.
(a) Show that Q is a quadratic form.
(b) Nonzero elements of V have rank 1 or 2, and elements of PV are distinguished as points of rank 1
and rank 2 accordingly. Show that the quadric dened by Q consists of all points of rank 1.
25. The Klein Correspondence
161
(c) Show that the corresponding quadric is hyperbolic, and that one regulus has lines parameterized
by the (1-dimensional) row space of the underlying matrix; and the alternate regulus has lines
parameterized by the (1-dimensional) column space of the underlying matrix.
2. Extend Theorem 24.6 by counting also the number of totally singular lines in each case (hyperbolic,
parabolic and elliptic).

Let F be a eld. Recall that every line in P3 (F ) is specied by a 6-tuple of Pl
ucker coordinates (01 , 02 , 03 , 12 , 13 , 23 ) which is determined by up to nonzero scalar multiple.
The Pl
ucker coordinates satisfy
(25.1)
01 23 02 13 + 03 02 = 0
and the Pl
ucker map (01 , 02 , 03 , 12 , 13 , 23 ) is a bijection between the set of lines
3
of P (F ) and the point set of the Klein quadric, which is the quadric in P5 (F ) dened
by (25.1). Note that for a nite eld F = Fq , the Klein quadric is simply the hyperbolic
quadric, otherwise known as the O6+ (q)-quadric. Every line of the Klein quadric lies in
exactly two planes of the quadric, one of each type. Two distinct planes of the quadric
(necessarily of opposite type) are incident if they intersect in a line of the quadric. Two
distinct planes which are not incident must be disjoint (if they are of opposite type) or
intersect in a point (if they are of the same type).
The Pl
ucker map described above extends to an isomorphism from the entire A3 geometry of projective 3-space, to the D3 -geometry of the Klein quadric, as indicated by
the following diagram. (Although this works for arbitrary elds, we indicate here the
number of objects of each type in the corresponding geometry according to our previous
counts. It is comforting to see that these numbers agree!)
This isomorphism of geometries, denoted here by , is known as the Klein correspondence. Its explicit description is conveniently described in the language of exterior algebra.
162
Let V be a vector 4-space over F , and consider the exterior algebra of V which is the 16dimensional algebra
/
V =
2 /k
k0
V =F V
/2
/3
/4
dened as in Appendix A4; here the summands have dimension 1, 4, 6, 4, 1 respectively.

We take {e0 , e1 , e2 , e3 } as a basis for V , so that
/2
V has basis {e0 e1 , e0 e2 , e0 e3 , e1 e2 , e1 e3 , e2 e3 };
/3
V has basis {e0 e1 e2 , e0 e1 e3 , e0 e2 e3 , e1 e2 e3 }; and
/4
V has basis {e0 e1 e2 e3 }.
/k
V for k = 1, 2, 3 respectively
The points, lines and planes of PV are mapped to points of
/k
via the Pl
ucker map U
U. In the case k = 2 this map is not surjective; its image is
/2
the Klein quadric which we denote K P

V . This quadric is dened by the quadratic
form
"
/2
Q(v) = v01 v23 v02 v13 + v03 v12 where v =
vij ei ej
V.
0i<j3
The associated bilinear form B(u, v) = Q(u + v) Q(u) Q(v) satises

u v = B(u, v) e0 e1 e2 e3
where u, v
/2
V.
We are now ready to dene the Klein correspondence.

The Klein Correspondence
For a point x < V , dene
&
%
/2
(x) = x V = v V : x v = 0 .
For a line = x, y < V , dene
/2
() = = = x y K.
For a plane U = x, y, z < V , dene
/2
(U) = U = x y, x z, y z.
We proceed to show that has the required properties. For any line = x, y < V where
"
"
x = 0i3 xi ei and y = 0i3 yi ei , we have
() = (x, y) = v

xi xj

ei ej
where v =

0i<j3 yi yj
"
163
which agrees with the image () under the Pl

ucker map as dened previously. By Section 22, maps lines of PV bijectively to points of the Klein quadric K. Also by Section 22
/2
and remarks from Appendix A4 identifying V with skew-symmetric 4 4 matrices, we
see that points of the Klein quadric are those represented as pure wedge products x y
for some linearly independent x, y V . The following observation is useful.
25.1 Lemma.
equivalent.
Let and be two lines of V . Then the following four conditions are
(i) and meet in a point of V .

(ii) and lie in a plane of V .
(iii) The points (), ( ) K lie on a line of K.
(iv) The points (), ( ) K are perpendicular relative to B.
Proof. Write = x, y, = x , y . Both (i) and (ii) are equivalent to the condition
that x, y, x , y are linearly dependent; this in turn is equivalent to the condition that
x y x y = 0
by (A4.5). Since x y x y = B(x y, x y ) e0 e1 e2 e3 , the latter condition is
equivalent to the points x y, x y K being perpendicular; this is condition (iv).
Finally (iii) is equivalent to (iv) since every line of K is totally singular and hence totally
isotropic; conversely any two singular points which are perpendicular, span a totally singular line.
We are ready to verify that the Klein correspondence is an isomorphism from the
A3 -geometry of PV to the D3 -geometry of K. A point x of V lies in a line = y, y of
V , i {x, y, y } is linearly dependent, i x y y = 0, i () = y y lies in (x).
Let , U < V be a line and a plane respectively. If U then meets every line

U, so the singular point () is perpendicular to every point ( ) of (U); thus
(), (U) is totally singular. But (U) is a maximal totally singular subspace so the
point () must lie in (U). Conversely if () lies in the totally singular subspace (U)
then () meets every point ( ) of (U), so meets every line U, so U.
Let x, U < V be a point and a plane of V respectively. If x lies in U then there
exist lines , < V such that = x and , = U. Then (x) (U) contains
the totally singular line in K spanned by () and ( ), so (x) is incident with (U).
Conversely, if (x) and (U) are incident then (x) (U) contains a singular point,
which we know has the form () for some line < V such that x U, so x is
incident with U.
164
The Klein correspondence is immensely useful in understanding many features of the

geometry of projective 3-space. For example recall that a spread of P3 (Fq ) is a collection
of q 2 +1 lines which partition the (q 2 +1)(q+1) points. This is equivalent to a set of q 2 +1
mutually skew lines of P3 (Fq ); and by Lemma 25.1, such a set of lines is equivalent to a
set of q 2 +1 points on the Klein quadric, no two of which are perpendicular; i.e. no two on
a line of the quadric. Such a point set is called an ovoid of the Klein quadric.
25.2 Example: Ovoids in O6+ (3). The quadratic form on F63 dened by Q(v) = v12 +
v22 + +v52 v62 has discriminant 1, and since (1)3 disc Q = 1 is a square, Q is hyperbolic.
(Note that the form v12 + v22 + + v62 is elliptic over F3 and therefore unsuitable for our
present purpose.) The associated bilinear form is
B(u, v) = Q(u + v) Q(u) Q(v) = 2(u1 v1 + u2 v2 + + u5v5 u6 v6 ) .

Consider the set O consisting of the 53 = 10 points of the form (13 02 |0); i.e. the last
coordinate is zero, and the rst ve coordinates include three 1s and two 0s. Note that
O consists of singular points, no two of which are perpendicular since 1s in distinct points
overlap in either 1 or 2 positions. Also |O| = 33 +1 so O is an ovoid.
Likewise the set O consisting of 25 = 10 points of the form (14 0| 1); here the
last coordinate is 1 and the rst ve coordinates include four 1s and one 0. The point
set O is also an ovoid; for example

B (111101), (111102) = 2(1 + 1 + 1 + 1 + 0 2) = 1 = 0;

B (111101), (111011) = 2(1 + 1 + 1 + 0 + 0 1) = 1 = 0;

B (111101), (111012) = 2(1 + 1 + 1 + 0 + 0 2) = 2 = 0.

By the Klein correspondence, both O and O give rise to spreads of P3 (F3 ), and therefore
translation planes of order 32 = 9. These two planes are non-isomorphic, as we might
expect from the fact that O spans only a 4-dimensional subspace of codimension 2, whereas
O spans the entire O6+ (3)-space. In fact the translation planes associated to O and O are
the classical plane and the Hall plane of order 9, respectively.
25.3 Example: Ovoids from Slope Matrices [33]. Consider a set of 22 slope matrices for a spread over Fq , say
a
Mi = i
ci

bi
,
di
i = 0, 1, 2, . . . , q 2 1
where Mi Mj is nonsingular whenever i = j. The associated spread has components

i = (1, 0, ai , bi ), (0, 1, ci , di ),
= (0, 0, 1, 0), (0, 0, 0, 1).
165
By the Klein correspondence we obtain the points of the associated ovoid as

(1, 0, ai , bi ) (0, 1, ci , di ) = e01 +ci e02 +di e03 ai e12 bi e13 +(ai bi bi ci )e23
= (1, ci , di , ai , bi , ai bi bi ci );
(0, 0, 1, 0) (0, 0, 0, 1) = e23 = (0, 0, 0, 0, 0, 1)
with respect to the hyperbolic quadratic form
Q(v1 , v2 , . . . , v6 ) = v1 v6 v2 v5 + v3 v4 .
For example if one takes the explicit spread matrices for the Hall plane of order 9, as given
in Example 2.5, and constructs the associated ovoid as shown above, one may change
coordinates so as to obtain the quadratic form of Example 25.2, in such a way that our
ovoid gives the ovoid denoted O in Example 25.2.
25.4 Example: Classical (regular) ovoids in O6+ (q). We may decompose any O6+ (q)space as O4 (q) O2 (q); simply take any elliptic line and its perp. By Theorem 24.6,
the O4 (q)-subspace contains q 2 +1 singular points; and these must form an ovoid, since an
O4 (q)-space contains no totally singular lines. We show that these ovoids do indeed give
rise to the classical translation planes of order q 2 .
As explained in Section 3, the classical planes of order q 2 may be constructed as
follows. Let t2 + t + c Fq [t] be an irreducible quadratic polynomial over Fq . A regular
spread is dened by the slope matrices
ct st

,
s, t Fq
since these form a matrix representation of Fq2 as described in Appendix A1.2. The
corresponding ovoid O, constructed as in the previous example, consists of the points
(0, 0, 0, 0, 0, 1);
(1, ct, st, s, t, s2 st+ct2 ) for s, t Fq
lying in x, y where x = (0, 1, 1, 1, 0, 0) and y = (0, c, 0, 0, 1, 0). Since

Q(ax + by) = Q(0, a+bc, a, a, b, 0) = a2 +ab+cb2
we have x, y O2 (q) and O x, y O4 (q).
25.5 Example: p 1 mod 4. The quadratic form Q(v) = v12 + v22 + + v62 has
associated bilinear form B(u, v) = 2(u1 v1 + u2 v2 + + u6 v6 ) and discriminant 26 . In
order for Q to be hyperbolic, we require that (1)3 disc Q = 64 be a nonzero square in
166
Fq , i.e. q 1 mod 4. The following construction works only for elds Fp of prime order
p 1 mod 4. Consider the set of all 6-tuples (a1 , a2 , . . . , a6 ) of integers such that
a21 + a22 + + a26 = 6p;
a1 a2 a6 1 mod 4.
In [44] it is shown that there are exactly p2 +1 such vectors; and that they give an ovoid
O in O6+ (p).
p = 5 the ovoid O has 6 points of the shape (51 , 15 ) (i.e. one 5 and ve 1s)
For
and 63 = 20 points of shape (33 , 13 ), for a total of 26 = 52 +1 points. After reducing
mod 5, these points have shape (0, 15 ) and (13 , 23 ). We check that these points form
an ovoid:

B (011111), (101111) = 2(0+0+1+1+1+1) 3 0 mod 5;

B (011111), (111222) = 2(0+1+1+2+2+2) 1 0 mod 5;

B (011111), (211122) = 2(0+1+1+1+2+2) 4 0 mod 5;

B (111222), (211122) = 2(2+1+1+2+4+4) 3 0 mod 5.

Similarly for p = 13 the ovoid O has
20
30
60
60
points
points
points
points
of
of
of
of
type
type
type
type
(53 , 13 );
(7, 5, 14 );
(52 , 33 , 12 ); and
(7, 33 , 12 )
for a total of 170 = 132 +1 points.

Exercises 25.
1. Construct an explicit ovoid in O6+ (17) by the method of Example 25.5.
26. Ovoids and Spreads of Projective Space

Here we consider the terms ovoid and spread as they are used in the context of projective
space. In Section 27 we will encounter distinct, albeit related, usages of these terms in
the context of polar spaces, particularly quadrics; this will be explained later. We begin
with ovoids of projective 3-space.
A k-arc of P3 (F ), or simply an arc, is a set O of k points, no three of which are
collinear.
26.1 Theorem. An arc in P3 (Fq ) for q > 2 has at most q 2 +1 points.
For q > 2, a (q 2 +1)-arc of P3 (Fq ) (the maximum possible size) is called an ovoid of the
projective 3-space. The failure of the indicated bound in the case q = 2 is discussed in
Exercise #1.
167
26.2 Theorem. Let O be an ovoid of P3 (Fq ), and

let P O. Then there is a unique plane P (the
tangent plane at P ) such that P O = {P }. Every
plane meets O either in a single point (i.e. is a
tangent plane) or in an oval of .
Proofs of Theorems 26.1 and 26.2 are included below, only in the case q is odd; for proofs
in the general case see [55], [3]. First we describe the known examples of ovoids. For all q,
an elliptic quadric in P3 (Fq ) is an ovoid. Indeed if O is an O4 (q) quadric then |O| = q 2 +1
by Theorem 24.6; and if a line meets O in three points then is totally singular by
Exercise #23.5, a contradiction. Barlotti [3] showed that if every oval arising as a plane
section of O is a conic, then O is an elliptic quadric; by Segres Theorem 12.14, it follows
that for q odd, every ovoid is an elliptic quadric. For q = 2r there is another innite
family of ovoids known, the Suzuki-Tits ovoids which arise for odd exponent r 3, i.e.
q {8, 32, 128, 512, . . .}. It is not known whether every ovoid in projective 3-space for q
even must be either an elliptic quadric or a Suzuki-Tits ovoid. This question has stood
for several decades as one of the reigning open problems in nite geometry. The most
signicant progress in this direction to date, is a theorem of Brown [8] showing that every
ovoid having at least one conic as a plane section, is an elliptic quadric. All ovoids for
q 32 are known, using computer results which rely on the classication of ovals for these
small values of q.
Proof of Theorems 26.1 and 26.2 for q odd. Let O be a k-arc in P3 (Fq ). We may assume
k 2, and let the points P, Q O be distinct. By the restriction on q, every plane meets
O in at most q + 1 points. There are exactly q + 1 planes containing the line P Q, each of
which contains at most q1 points of O other than P and Q; this gives
|O| 2 + (q + 1)(q 1) = q 2 + 1
which is the conclusion of Theorem 26.1. Henceforth we assume k = |O| = q 2 + 1.
We proceed using counting arguments similar to those arising in the proof of Proposition 18.1. Let mk be the number of planes of P3 (Fq ) meeting O in exactly k points for
k {0, 1, 2, . . . , q+1}. The total number of planes is
(26.3)

4
mk =
= (q 2 + 1)(q + 1).
3 q
0kq+1
Counting in two dierent ways the number of pairs (P, ) where is a plane and P O
gives
168
(26.4)
kmk = (q 2 + 1)(q 2 + q + 1).
0kq+1
Here we have used the fact that each of the q 2 + 1 points of O lies in exactly q 2 + q + 1
planes. Next we count in two dierent ways the number of ordered triples (P, Q, ) with
any plane, and the points P, Q O distinct:
$
(26.5)
k(k1)mk = q 2 (q 2 + 1)(q + 1).
0kq+1
Here we have used the fact that there are q 2 (q 2 + 1) ordered pairs of points (P, Q) in O,
and each such pair determines a line P Q which lies in exactly q+1 planes. Finally we
count in two dierent ways the number of ordered 4-tuples (P, Q, R, ) with any plane,
and the points P, Q, R O distinct:
$
(26.6)
k(k1)(k2)mk = q 2 (q 4 1).
0kq+1
Here we use the fact that O contains q 2 (q 2 + 1)(q 2 1) ordered triples (P, Q, R) of points,
each of which determines a unique plane. From (26.5) and (26.6) we obtain
$
k(k1)(q+1k)mk = (q1)
0kq+1
k(k1)mk
0kq+1
2
k(k1)(k2)mk
0kq+1
= (q1)q (q +1)(q+1) q 2 (q 4 1)
= 0.
We must have mk = 0 for all k {2, 3, . . . , q}; otherwise the left side in the preceding
equality would be positive. Now the equations (26.3)(26.5) become
0
0
1
1
0
m0
(q 2 +1)(q+1)
q+1 m1 = (q 2 +1)(q 2 +q+1)

q(q+1)
q 2 (q 4 1)
mq+1
which has the unique solution (m0 , m1 , mq+1 )=(0, q 2 +1, q(q 2 +1)).
26.7 Inversive Planes from Ovoids. Let O be an ovoid in P3 (Fq ). Recall that P3 (Fq )

has 41 q = (q 2 +1)(q+1) planes, q 2 +1 of which are tangent to O; the remaining q(q 2 +1)
planes meet O in ovals. It is not hard to see that the incidence structure (O, C) formed by
the q 2 + 1 points of O and the q(q 2 + 1) ovals in O, form an inversive plane of order q
(see Exercise #2.3). An inversive plane arising in this way is called egglike. Note that
this construction mimics the construction of the real inversive plane in Exercise #2.3(a).
169
A signicant open problem in nite geometry is to classify nite inversive planes up

to isomorphism. This question can be divided into two parts:
(I) Must every nite inversive plane be egglike (i.e. arise from an ovoid in a nite
projective 3-space)?
(II) What are the ovoids in nite projective 3-space? (This will tell us what the
egglike inversive planes are.)
In the direction of (I), it is still unknown whether every nite inversive plane has primepower order. Since an inversive plane of order n yields an ane plane of order n by
Exercise #2.3(b), and hence a projective plane of order n by Section 7, the prime power
conjecture for nite projective planes would imply the prime power conjecture for nite
inversive planes. But it may be possible to prove that every nite inversive plane has prime
power order, without proving that every nite projective plane has prime power order is
reversible; this is because the converse of Exercise #2.3(b) might not hold.
It is known [24] that every nite inversive plane of even order is egglike (in particular
its order must be a power of 2), answering question (I) armatively in the even order case;
but in this case question (II) is open, as we have explained. By contrast, question (II) is
settled in the odd order case as we have said, but in this case question (I) is open! So the
original question of classifying nite inversive planes remains open, both in the even order
case and in the odd order case, but for dierent reasons!
Traditionally, ovoids are dened more generally in projective n-space; and then it is
proved that they do not exist for n > 3! For n = 2 an ovoid is simply an oval. Nu said
about ovoids.
Recall that a spread of a 2k-dimensional vector space V over F , is a collection S
consisting of k-dimensional subspaces of V , which induce a partition of the nonzero vectors
of V . We now interpret this denition projectively: A spread of P2k1 (F ) is a partition
of the points into projective (k1)-subspaces. In particular for F = Fq we must partition

2k
1)/(q 1) points into subspaces each having k1 q = (q k 1)/(q 1)
the 2k
1 q = (q
points, so the number of such spread members (i.e. components) is
(q 2k 1)/(q 1)
= q k +1
(q k 1)/(q 1)
as we saw in Section 3. A translation net is a collection of mutually disjoint projective
(k1)-subspaces; thus a spread is a translation net having q k +1 members, the maximum
possible number. Recall that every eld extension E F of degree k gives rise to a spread
of P2k1 (F ) by taking as components all the one-dimensional E-subspaces of E 2 = E E,
interpreted as F -subspaces of dimension k. Such spreads are called regular; and this term
is justied by a geometric condition which posits the existence of large numbers of reguli
in the spread. (For once the word regular, so over-used in mathematical contexts, is
170
etymologically motivated.) Rather than explain this here in full generality, we focus on
the case of projective 3-space.
Let V be a 4-dimensional vector space over a eld F .
...........
....
....
...
Consider any two skew (i.e. non-intersecting) lines , in
.....
....
......
.....
...
..
.....
....
.....
.
.
3

.
.
.....

.
...
P
PV = P (F ). If P is any point not on then there is
.....
.... ...........
.....
..........
.....
......
..
.....
.
.

.
.
.
.
.
.....
..
....
a unique line m through P meeting both and . This is
.....
......
...
..... ..........
....
.........
....

.....
.
.
.
....
.
. .....
.
not hard to see since P, and P, are necessarily distinct
.
.
.
....
.
.....
..
.
.
.
...
.
.
.....
...
.
.
.
.
.
.
.
.....
..
.
.
.
.
.....
.
planes, which therefore meet in the required line m. We call
.....
.....
m
.....
m the transversal to and through P .
A collection of lines is mutually1 skew if any two distinct lines in the collection are
skew. Let , , be three mutually skew lines. Every point P lies on a unique line
meeting both and ; and as P varies over all points of , this process generates the
. , ) of all lines transversal to each of the original three lines , , , i.e. meetfamily R(,
. , ) are
ing each of the lines , , . Now by the preceding comments, the lines of R(,
. , ) be distinct; then R(m,
.
m , m )
themselves mutually skew. Now let m, m , m R(,
.......
....
....
...
...
... .......... .
... ....
....
.......
..
.. . ....
......
..... ........... .....
.
.
......
.
.
...... .......
....... ... .
......
.... ........... .....
........... ...
.... .. .
..... .................
.
.
... ... .. ..
.
.
.. .... ... ......... ....
....
.... ................
.
...................
.. .. .
.
.
..
..
.. ......... .... ..........
.... ........................
........
...
...... ........
.....
. ........... .....
.
.
.
.
.
.
.
.
.
.
............
..
.. ...............
....
....
..
... ......................
.....
....
...
. .......... .... ....
.....
.
.
.
.
.
.
.
.
.
.
.
.
.
...........
.
.
...
..
...
.........
....
.....
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.....
...
....
..
...

.
lines of R(, , )
.......
......
......
.....
......
.....
......
........
....... ......... .....
........
.
.
..
........
.
........ .......... .................. .....
..........
.............
.. ........ .... .
.
.
.
.
.
.
.
.
.
.
.
...
. ..
..........
.......... ......... .................... ...................... .
............
........ ...
.. ............
.
.
.
.
.
.
.
..................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.........
.. ....... ..........
.............. ........
.... ........................ ...................... ...........
.................
..... ..
. ............
. ...................
.
.
.
.
.
.
.
....................
.
. ......... ... ..........
.. ......................
...........................
....
.. ..................... ........................
....
. .................
.... .......................................... ......
.
.
.
.
.
.
..
...
........................................ .....................
....
.....
......................
...
.........
....
...
...
.....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
......
........
...
...
.....
.
..
.
....
...
m

lines of

R(, , )
m m

. , )
lines of R(,
is likewise a family of mutually skew lines containing the original three lines , , ; we
denote
(26.8)
.
R(, , ) = R(m,
m , m )
and we call this the regulus generated by the original three lines , and . The rst
family of mutually skew lines,
(26.9)
. , ),
R(m, m , m ) = R(,
is also a regulus; and the two reguli (26.8) and (26.9) are alternate, i.e. opposite to
one another. We proceed to show that every member of one regulus is incident with
every member of the alternate regulus; and that these two reguli are the ruling lines of a
hyperbolic quadric in PV .
1
Some would say pairwise skew; see the footnote on p.19.
171
We may coordinatize the lines of our reguli as follows: First choose bases {e0 , e1 }
for and {e2 , e3 } for . Now {e0 , e1 , e2 , e3 } is a basis for V ; and there is no loss of
generality in assuming this to be the standard basis {(1, 0,0, 0),
. . . , (0, 0, 0, 1)}. Eva b

idently = (1,
0, a,1 b),
(0, 1, c, d) where the 22 matrix c d is non-singular. We
a b
0
may take c d = 0 1 , after performing a change of basis for if necessary; thus
. , ) passing through a typical point
= (1, 0, 1, 0), (0, 1, 0, 1). The unique line of R(,
(s, t, 0, 0) is the line m = (s, t, 0, 0), (0, 0, s, t) shown:
...
..
........................
m
......................................... .......(s,t,s,t)
(1,0,1,0)....
.................
............................................
...
..
..
...................(0,1,0,1)
.................
...
...
...............
.
....
....
............
...
..
..
...
...
.
....
.
.
.
.
.
...
.
...
..
......................
..
...
..
..............................
...
................... ...(0,0,s,t) ....
(0,0,1,0)......
....................
..
...................
.
.....................(0,0,0,1)
..
...
...
..
...
... ..............................
.
....
.
....
.
...
.
.
.
...
....
.
.
...
...
...
...
...
...
...
.
.
.
.
.
.
..........................
..
..
.. .............................................. (s,t,0,0) ....
(1,0,0,0)......
................................................... (0,1,0,0)
..........................
....
...
.......
..
.
...
...
..
...
.
.
...
lines of
R(, , )

.
lines of R(, , )
All points of m are singular with respect to the hyperbolic quadratic form
Q(x) = x0 x3 x1 x2 .
. , ) consists of all lines of the form
The regulus R(,
(s, t, 0, 0), (0, 0, s, t),
(0, 0) = (s, t) F 2
while the alternate regulus R(, , ) consists of all lines of the form
(s, 0, 0, t), (0, t, s, 0),
(0, 0) = (s, t) F 2 .
All of the preceding discussion of reguli is greatly claried and simplied using the
Klein correspondence. Let , , be three mutually skew lines in P3 (F ). The Klein
correspondence maps these three lines to three mutually nonperpendicular points P, P , P
of the O6+ (q)-quadric. The plane = P, P , P spanned by these three points must
therefore be nondegenerate, which means that intersects the Klein quadric in a conic.
The q+1 points of this conic, including P, P , P , correspond (by the Klein correspondence
to the regulus R(, , ) generated by the three original lines. Moreover since O3 (q)
is nondegenerate, the plane is also nondegenerate and disjoint from ; we have F 6 =
= . The plane intersects the Klein quadric in another conic, whose q+1
. , ). All
points correspond (by the Klein correspondence) to the alternate regulus R(,
172
of this follows easily from the known properties of the Klein correspondence, using the fact
that every point of is perpendicular to every point of .
A spread S of P3 (F ) is regular if for all choices of three distinct lines , , S,
we have the entire regulus R(, , ) S. By the Klein correspondence, this means that
the corresponding ovoid O (justiably called a regular ovoid) in the Klein quadric has
the property that for any three of its points P, P , P , all q+1 singular points of the plane
P, P , P must also belong to O. It is not hard to see that every such ovoid lies in an
O4 (q)-subspace, and so the translation plane dened by any regular spread is isomorphic
to A2 (Fq2 ) (see Example 25.4).
Now let be a spread of P3 (F ). If contains all the lines of some regulus, then clearly
these lines may be replaced by the lines of the alternate regulus (since the alternate regulus
covers the same points as the original regulus). This gives a new spread, and thereby a new
translation plane which will typically not be isomorphic to the plane dened by . For
example the regular spread in P3 (F3 ) contains many reguli. Replacing one of these by its
alternate regulus gives a spread which is not regular; this denes the Hall plane of order 9.
To describe this process in terms of an ovoid O of the Klein quadric K: If O contains a
conic (which has the form K for some nondegenerate plane ) then by deleting this
conic and replacing it by the conic in K, we obtain a new ovoid.
Exercises 26.
1. What is the maximum size of an arc in P3 (F2 ), and what is the structure of an arc attaining this
maximum size?
Hint.
Consider the complement of a plane.
2. Let C be the binary code spanned by the planes of P3 (Fq ).

(a) What is the dimension of C? (Use Theorem 21.1.)
(b) If O is an ovoid in P3 (Fq ), show that the collection of tangent planes {P : P O}, in the notation
of Theorem 26.2, is a linearly independent subset of C.
(b) Assuming that q is even, prove a stronger condition: that the tangent planes as in (b) actually
form a basis for C.
27. Ovoids and Spreads of Polar Spaces

I personally think of ovoids and spreads of polar spaces as certain kinds of tilings or
packings, and this is how I prefer to introduce them. Consider the problem of tiling the
gure shown (which has been subdivided into eight unit squares) using 21 dominoes.
.........................................
...
..
...
.
...
.. . . . . . ... . . . . . ......................
..
...
.
.
...
.
....
...
...
..
..
..
..
..
...
......................
..
..
..
..
..
...
.
.
.
.
.
.
..............................................................
173
We require that the entire gure be covered, with no overlapping dominoes; so clearly 4
dominoes will be required. This problem has exactly four solutions, as shown:
..........................................
...
..
...
..
...
..
.
.
.... . . . . . .. . . . . . ......................
..
..
...
...
...
...
..
..
..
..
..
....................
...
.
.
...
..
..
....
..
..
..
.
..
..........................................................
..........................................
...
..
...
..
...
..
.
.
.... . . . . . .. . . . . . ......................
..
..
...
...
...
...
..
..
..
..
..
....................
...
.
.
...
..
..
....
..
..
..
.
..
..........................................................
..........................................
...
..
...
..
...
..
.
.
.... . . . . . .. . . . . . ......................
..
..
...
...
...
...
..
..
..
..
..
....................
...
.
.
...
..
..
....
..
..
..
.
..
..........................................................
..........................................
...
..
...
..
...
..
.
.
.... . . . . . .. . . . . . ......................
..
..
...
...
...
...
..
..
..
..
..
....................
...
.
.
...
..
..
....
..
..
..
.
..
..........................................................
This problem may be represented graphically as follows. Consider the bipartite graph
with 18 vertices corresponding to the 10 domino positions and the 8 cells of the gure.
Edges in the graph indicate which dominoes cover which cells, as shown:
.............................
.. . ..
.... . . .... . . ................
... . . ...
... .. .. ...
............
... .. .. ..
. . . .
......................................
.............................
.. . ..
.... . . .... . . ................
... . . ...
... .. .. ...
............
... .. .. ..
. . . .
......................................
.............................
.. . ..
.... . . .... . . ...............
... . . ...
... .. .. ...
............
... .. .. ..
. . . .
......................................
.............................
.. . ...
.... . . .... . . ...............
.. .. .. ...
... . . ..
............
.. .. .. ..
.. . . .
....................................
.............................
.. . ...
.... . . .... . . ................
.. .. .. ...
... . . ..
............
.. .. .. ..
.. . . .
....................................
.............................
.. . ...
.... . . .... . . ...............
.. .. .. ...
... . . ..
............
.. .. .. ..
.. . . .
....................................
.............................
.. . ...
.... . . .... . . ...............
.. .. .. ...
... . . ..
............
.. .. .. ..
.. . . .
....................................
.............................
.. . ...
.... . . .... . . ................
.. .. .. ...
... . . ..
............
.. .. .. ..
.. . . .
....................................
.............................
.. . ..
.... . . .... . . ................
... . . ...
... .. .. ...
............
... .. .. ..
. . . .
.....................................
.............................
.. . ..
.... . . .... . . ...............
... . . ...
... .. .. ...
............
... .. .. ..
. . . .
.....................................
.
..
.......
........
......
.....
.........
........
......
.....
...........
.......
....... .........
...... ...........
.......
.. ...........
...... ...........
.... ..........
.. ....
... ....
.. .....
.....
.... ...
.......
......
..... ..
......
......
......
...
......
...
......
... ....
... .....
.. ...... .............
...... .......... ....
..... ............
......
..... ....
...... ...........
...
...
......
.
.
.
.
.
.
.
.....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
.
.
.
.
..
.
.
.
.
.
.
.........
...
.
......
.....
...
.
..
.......
........
.... ......
.........
..
...... ....
...
...
...... .....
....
...... ..........
...... .........
..... ........... ....
..... ...
................ .....
........
...
...
........
...
.....
......
......
.....
.....
........
....
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
...
......
...
..
.....
....
.
..
...
. ....
.. ..
... ............. ...
..... ...........
..
...... ..........
... ........... ........
... ............
... ...........
...
... ..........
... .......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
.
.
.
.
.
.
.
.
.
.
........
.......
........
......
......
....
......
...
.
.
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
........... ...
.....
.......
....
.. .......
. ......
.
.. ....
.
.
.
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.. .......
.
.
.
.
.
.
.
....
.
.
......
......
.....
......
.
.
..
.
.
...
... ....... .... ..
.
.
.
.
.
.
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
.
... ...
.
.
.
.
.
.
.
.
.
.
.
.
.....
...
......
......
...... ... ..
......
..
.
.
.
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.. .....
.
.
.
.
.
.
...... .. . ......
...... .. ....
...... ....
..... .. .......
...
.....
..
..
.
.
.... ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.....
.
.
.
.
.
.
.
.
.
.
.
..... .....
...... ...
................
...... .....
...............
.... ......
.......
.
.
.
....
.
.
.
.
.
.
.
.
.
..
...
...
..
..
.
..
............................
... . ...
... . . .... . . ................
.. . . ...
... .. .. ..
............
.. . . ..
... .. .. ...
....................................
............................
... . ...
... . . .... . . ................
.. . . ...
... .. .. ..
............
.. . . ..
... .. .. ...
.....................................
............................
... . ...
... . . .... . . ................
.. . . ...
... .. .. ..
............
.. . . ..
... .. .. ...
.....................................
...........................
... .. ...
.... . . .... . . ................
... .. .. ...
.
............
..... .. .. ...
. . . .
..........................................
............................
... . ...
... . . .... . . ................
.. . . ...
... .. .. ..
............
.. . . ..
... .. .. ...
....................................
............................
... . ...
... . . .... . . ................
.. . . ...
... .. .. ..
............
.. . . ..
... .. .. ...
.....................................
............................
... . ...
... . . .... . . ................
.. . . ...
... .. .. ..
............
.. . . ..
... .. .. ...
.....................................
...........................
... .. ...
.... . . .... . . ................
... .. .. ...
.
............
..... .. .. ...
. . . .
..........................................
The problem of tiling the gure with dominoes, is equivalent to nding a subset S of the
top 10 vertices, so that each of the 8 bottom vertices is adjacent to exactly one member of
S in the graph . A related (or perhaps, dual) problem is that of nding a subset O of the
bottom 8 vertices, such that each of the top 10 vertices is adjacent to a unique member of
O in the graph . This problem has two solutions:
.........................................
..
...
...
...
.....
.
.. . . . . . .. . . . . . .....................
..
..
...
...
..
..
...
..
...
...
...
.
..
...................
..
....
..
..
..
..
..
.
...
................................................................
.........................................
..
...
...
...
.....
.
.. . . . . . .. . . . . . .....................
..
..
...
...
..
..
...
..
...
...
...
.
..
...................
..
....
..
..
..
..
..
.
...
................................................................
We refer a solution S of the tiling problem, as a spread; and a solution O of the dual
problem, as an ovoid.
Likewise, given a polar space P, we may consider the bipartite graph whose two types
of vertices correspond to the points and maximal subspaces of P. A spread (think: tiling)
of the polar space is a subset of the set of maximal subspaces of P, such that every point
of P lies in a unique member of S (so that S partitions the points of the polar space).
An ovoid of P is a collection of points of P, such that every maximal subspace contains
a unique point of O. The ovoids of O6+ (q)-quadrics introduced in Section 25 are in fact
174
examples of ovoids in our currently used sense of the term, as we shall see. First let us
examine the situation for real quadrics.
The hyperbolic quadric Q in P3 (R) contains two spreads, these being the two reguli
of the spread (the two families of lines ruling the quadric). It has many ovoids. For any
plane , either is tangent to the quadric, in which case Q is a pair of intersecting
lines; or Q is a conic meeting every line of Q in a unique point. In the latter case
Q is an ovoid of Q, and we obtain 20 ovoids of Q in this way. However there are many
more ovoids: Since Q is a grid formed by two families of lines, each parameterized by S 1 =
P1 (R) = R {}, the point-line incidence structure of Q is isomorphic to (P, L) where
P = S 1 S 1 = {(x, y) : x, y S 1 } and lines L are point sets of the form x = constant
or y = constant. Ovoids are point sets O P of the form Of = {(x, f(x)) : x S 1 }
0
where f : S 1 S 1 is any bijection. Since |S 1| = 20 , there are 22 = 2 such ovoids.
The conics arise just for those bijections f expressible as fractional linear transformations.
. . . . . . . . . .
...................................................................................................................................
. . . . . . . . . .
...................................................................................................................................
.. .. .. .. .. .. .. .. .. ..
.............................................................................................................................
... ... ... ... ... ... .. .. ... ...
.........................................................................................................................
.. .. .. .. .. ... . . .. ..
.............................................................(x,f(x))
..........................
. .
. . . . . .
f(x)...........................................................................................................................................................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...................................................................................................................
.. .. .. .. .. .. .. .. .. ..
.........................................................................................................................
.. .. .. .. .. .. .. .. .. ..
.......................................................................................................................................
.. .. .. .. .. .. .. .. .. ..
...................................................................................................................................
. . . . . .. . . . .
..
The elliptic quadric in P (R) contains a unique ovoid and a unique spread; both are
simply the entire point set of the quadric. Every (nonempty) nondegenerate quadric Q in
Pn1 (R) is can be dened by an equation of the form
x21 + + + x2m x2m+1 x2n = 0
0 1
where 1 m n2 ; an example of an ovoid in Q is the set of points of Q lying in the
subspace em , em+1 , . . . , en where {e1 , . . . , en } is the standard basis of Rn .
Let us now determine the size of an ovoid or of a spread, for quadrics over nite elds.
27.1 Theorem. In any nite quadric, the size
of an ovoid or of a spread is listed in the accompanying table. These values also give the maximum
size of any set S of mutually disjoint maximal
totally singular subspaces, and of any set O of
mutually non-perpendicular singular points.
It may at rst come as a surprise that the size of an ovoid and of a spread coincide. An
explanation for this fact is available using the language of m-systems; see [59] for details.
175
Proof of Theorem 27.1. Let Q : V Fq be a nondegenerate quadratic form. Every

collection S of mutually disjoint maximal totally singular subspaces satises
|{sing. pts.}|
|{pts. in each max. tot. sing. subspace}|
k1
(q
+1)(q k 1)/(q1)
+
= q k1 +1, in case O2k

(q);
k 1)/(q1)
(q
k
(q +1)(q k1 1)/(q1)
=
in case O2k
(q);
= q k +1,
k1
(q
1)/(q1)
(q 2k 1)/(q1)
in case O2k+1 (q)

= q k +1,
(q k 1)/(q1)
|S|
and equality holds i S is a spread.

Now let O be a collection of mutually nonperpendicular singular points; equivalently,
O is a set of singular points, no two on a line of the quadric, and therefore no two in
the same maximal totally singular subspace. By similar arguments as in the spread case
(by considering the graph described above) the upper bound for |O| occurs when every
maximal totally singular subspace contains a unique point of O. It therefore suces to
assume O is an ovoid.
For each isometry type of Q we count in two dierent ways the number of pairs (x, U)
where U is a maximal subspace of the quadric containing a point x O, thus:

(27.3) {U V : U max. tot. sing.} = |O|{U V : U max. tot. sing., x U}

= |O|{max. tot. sing. subsp. of x /x}
where x is a typical singular point.
+
+
(q) then x /x O2(k1)
(q) so by Table 24.6, (27.3) gives
If Q : V F is of type O2k
2(q+1)(q 2 +1) (q k1 +1) = |O| 2(q+1)(q 2 +1) (q k2 +1)
(q) then x /x O2(k1)

(q)
and so |O| = q k1 +1 as required. Similarly if Q is of type O2k
and so
(q 2 +1)(q 3 +1) (q k +1) = |O| (q 2 +1)(q 3 +1) (q k1 +1)
whence |O| = q k +1. Finally if Q is of type O2k+1(q) then x /x O2k1

(q) and so
(q+1)(q 2 +1) (q k +1) = |O| (q+1)(q 2 +1) (q k1 +1)

which also gives |O| = q k +1.
27.4 Examples of ovoids. An ovoid in a quadric of type O2+ (q), O3 (q) or O4 (q) consists of the entire quadric in each case: two singular points, a conic, or an elliptic quadric
176
in P3 (Fq ) respectively. Ovoids in the Klein quadric O6+ (q) consist of q 2 +1 mutually nonperpendicular points, and are equivalent to spreads of P3 (Fq ) via the Klein correspon
dence; examples of these were given in Section 25. Every O2k
(q)-quadric is embedded in
an O2k+1 (q)-quadric as a hyperplane section, which in turn is similarly embedded in an
+
(q)-quadric. Therefore any ovoid in O2k

(q) is automatically an ovoid in O2k+1 (q)
O2k+2
+
and in O2k (q); an example of this principle gives the embedding of the O4 (q)-quadric as
an ovoid in the Klein quadric O6+ (q). Similarly every ovoid O5 (q) is automatically an ovoid
in O6+ (q).
Consider the standard quadratic form Q(x) = x21 + + x28 which is nondegenerate
for all odd q, dening an O8+ (q) quadric. In this case an ovoid O consists of q 3 +1 singular
points, no two of which are perpendicular
to the standard dot product. An
8
with respect
3
example for q = 3 is given by the set of 2 = 28 = 3 +1 points of the form (02 16 ), i.e.
points spanned by vectors with two 0s and six 1s. This example lies in the nondegenerate
hyperplane (1, 1, . . . , 1) and so it is in fact an ovoid in an O7 (3)-quadric. An example for
q = 5 is given by:
8
7 = 56 points of the shape (016 3);
8
4 4
4 = 70 points of the shape (1 2 )
for a total of 126 = 53 +1 points. Similar constructions exist for every prime p. For p = 2
we need 23 +1 = 9 points but the quadratic form must be replaced by a nondegenerate
one.
To construct an ovoid in O8+ (2) (the unique ovoid in this space, up to isometry) we proceed as follows. Recall that the weight of a vector x F92 , denoted wt(x) {0, 1, 2, . . . , 9},
is the number of nonzero coordinates. The subspace V = (1, 1, 1, . . . , 1) < F92 consists
of all even weight vectors, i.e. vectors of weight 0, 2, 4, 6 or 8. Dene the quadratic form
Q : V F2 by
$
xi xj .
Q(x) =
1i<j9
Note that if wt(x) = k, then Q(x) k2 mod 2 so x is singular i k {0, 4, 8}. In order
to show that Q is nondegenerate, we show that for every nonzero vector u V there exists
x V such that Q(x+u) = Q(x), i.e. B(u, x) = 0. Since the argument depends only on
wt(u) it suces to consider
u = (111100000), x = (100010000) so that wt(x) = 2, wt(x+u) = 4 so that Q(x) =
1 = 0 = Q(x+u);
u = (111111110), x = (100000001) so that wt(x) = 2, wt(x+u) = 8 so that Q(x) =
1 = 0 = Q(x+u).

Thus Q is nondegenerate. The number of singular points x is 94 + 98 = 135. By

Theorem 24.6 we see that hyperbolic and elliptic quadratic forms have

23 + 1
24 1
21
= 135
and
24 + 1
23 1
21
= 119
177
singular points respectively; so Q must be hyperbolic. Let O be the collection of all

9 = 23 +1 points spanned by vectors of weight 8, so that Q is hyperbolic. Let O be the set
of all points x such that wt(x) = 8; then wt(x+x ) = 2 for all x = x in O, so that O is
an ovoid.
It is known [62] that the O2n

(q) quadric has no ovoids for n 3; and [28] that the
O2n+1(q) quadric has no ovoids for n 4. Ovoids are known [19] to exist in O8+ (q) for all
prime values of q, as well as for all q 1 mod 3; see [33]. Ovoids of O8+ (q) (when they
exist) are equivalent to spreads of O8+ (q), by a triality automorphism; see Section 24.9. It
is not known whether the O8+ (25) quadric (the smallest open case) has any ovoids. The
+
(q) has any ovoids for
most signicant open problem in this area is whether or not O2n
n 5. The most signicant progress in this direction is the following, which for example
+
+
proves the nonexistence of ovoids in O10
(2r ) and O10
(3r ). A cap in a polar space is a
collection of points, no two collinear.
27.5 Theorem [6]. Every cap O in a nondegenerate quadric in Pn (Fpr ) has size

p+n3
r
|O| p+n1
+1.
n
n
In particular ovoids cannot exist unless

p+n3
.
p
n/2 > p+n1
n
n
Proof of Theorem 27.5. Actually we present here a proof of a slightly weaker bound.
Let k = |O|. For each P O the hyperplane P V contains P and no other point
of O; therefore the point-hyperplane incidence matrix A of Pn (Fpr ) has a k k identity
submatrix with rows indexed by points P O and columns indexed by hyperplanes P
for P O. Since the p-rank of any matrix is bounded below by the p-rank of any of its
submatrices, we have

r
p+n1
+1
|O| = k rankp A =
n
by Theorem 21.1. The slight improvement of this bound stated in the Theorem, arises
from the observation that the k k identity submatrix actually lies in a submatrix
of
A, having
rows
indexed by just the singular points. The rank of this submatrix is
p+n1
p+n3
r
+ 1; see [6].
n
n
Exercises 27.
+
1. Let O be an ovoid in O2n
(q), and let x
be a singular point not in O. Show that the hyperplane x
+
intersects O in exactly qn2 + 1 points, which give an ovoid in x / x
O2n2
(q). Thus conclude
+
+
that if O2n2 (q) has no ovoids, then neither does O2n (q).
178
28. Generalized Quadrangles

The symmetric group S6 has 75 elements of order 2, which fall into three conjugacy classes:
6
= 15 such elements;
(C1) (12), (13), . . . , (56):
2 6
(C2) (12)(34), (12)(35), . . . , (34)(56):

3 4 = 45 such elements;

4
1 6
(C3) (12)(34)(56), (12)(35)(46), . . . , (16)(25)(34):
6 2
2 = 15 such elements.
Thus classes (C1) and (C3) consist of odd permutations; class (C2) consists of even permutations. Consider the incidence structure (P, L) with point and line sets given by the
elements of order 2 in classes (C1) and (C3) respectively; and where a point P lies
on a line L precisely when and commute. One checks that this is a partial linear
space, with three points on each line, and three lines through each point, as shown:
28.1 Figure.
The Generalized
Quadrangle of
Order 2
This is the unique generalized quadrangle of order 2. It is embedded in the projective

plane of order 4, as follows: Let O
be a hyperoval in P2 (F4 ), let P be the 21 6 = 15
points not in O, and let L be the 62 = 15 secants to O. Then (P, L) is isomorphic to the
structure dened above.
A generalized quadrangle is a point-line incidence structure satisfying the following:
(GQ1) Every line has s + 1 points, for
some s 1.
(GQ2) Every point lies on t + 1 lines,
for some t 1.
(GQ3) If P is a point not on a line
, then there is a unique line
through P meeting .
We call the pair (s, t) the order of the generalized quadrangle; or if s = t we say
simply a generalized quadrangle of order s. From the axioms it is clear that the point-line
179
dual of a generalized quadrangle of order (s, t) is a generalized quadrangle of order (t, s).
The example above is the unique generalized quadrangle of order 2, up to isomorphism; in
particular since it is unique, it is self-dual (isomorphic to its dual). Usually one requires
s, t 2, giving the so-called thick generalized quadrangles; those with s = 1 or t = 1 are
thin. Indeed a generalized quadrangle of order (s, 1) is just a 2-net of order s + 1, i.e. a
grid with (s + 1)2 points and 2(s + 1) lines. A generalized quadrangle of order (1, t) is
a dual grid with two sets of t + 1 points, and (t + 1)2 lines joining every point in one
set with every pair in the other set. Some examples of thin generalized quadrangles are
shown:
A thin generalized quadrangle

(i.e. a grid) of order (4, 1)
A thin generalized quadrangle

(i.e. a dual grid) of order (1, 3)
...
.....
...................
..........
...........................................................................................................
.............
.
.
.
.............
........ .........
........ ......
...... .......
... .... .................. ............
... .... ........ ....... ..... ...
......... ..
... ..........
.. .
.. ...
.
.
............. .................. ........ ....... .................. ...............
.....................................................................................................................................
.
.
.
.
.
.
.
.
.
.
.
.
.
.............. ...... ...... ................. ....
.
.
.
.
..... ........... .... .. .......... .....
... ... ................... .......
........ ................... ........
............................ ..............................
.
.
.
........ ................. .......... ......... ................. .......
................................................................................................................
............. ................... ........ ....... ................... ...............
.
.
.. ...
...............
........ ...
... .... .............. ..... ...
.......... ......................... ............
.
.
........ .......
...... ........
..............
................
.....................................................................................................................
..........
............
......
...
......
......
.......
....
......
.....
......
........
...... ......... .....
........
..........
..
........
........ ........ ............... ......
..
...........
..........
......... ...... ................ ...
.
..........
.
.
.
...........
.......... ......
... .... .
.............
.. ............... ................ ...
.. ..... ....
...
.
..............
..... ....................... ..... ................... ...............
..............
.
.
.
....................
.
.
.
.
.
.......
.. ........... ......
..
..... ................................ ...... ....................... .....................
.
.
....................
.
.
................
.....
........................
.. .........
...
... .......................... .....................
.. .................................
...............
.
.........................
.....
....
. ........................
. ............
.....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....................
.
.
...........
...
...
....
.....
...
...
.....
...
....
..
.....
....
..
...
..
..
...
.....
....
..
...
..
.................................
...............................
.
.. ..............
....
..
..
.
..
..
....
...
...
..
..
..
.
..
....
..
...
..
.
...
..........
... ................................................. ..
.
.
..
.
.
.
.
.
.
.
.
.
........
..
..
A thin generalized
quadrangle of
order 1, i.e. a
quadrangle
Just as a projective plane of order 1 is a triangle, so a generalized quadrangle of order 1

is a quadrangle. So a projective plane may be reasonably viewed as a generalized triangle.
In Section 29 we will describe generalized polygons from this perspective.
28.2 Proposition. A generalized quadrangle of order (s, t) has exactly (s+1)(st+1)
points and (t + 1)(st + 1) lines.
Proof.
Let P be a point. The set of all points is a disjoint union

{P } P1 P2
where P1 is the set of all points distinct from P but collinear with P ; and P2 is the set of
all remaining points. We have
|P1 | = (t + 1)s
since each of the (t + 1) lines through P has s points of P1 , with no point counted twice.
Every point R P2 is collinear with exactly t + 1 points in P1 , one on each of the lines
through R, by (GQ3). Every point Q P1 is collinear with st points in P2 : thats s
points of P2 on each of the t lines through Q other than the line P Q. So
|P2 | =
|P1 |st
= s2 t
t+1
180
and the total number of points is

1 + |P1 | + |P2 | = 1 + (t + 1)s + s2 t = (s + 1)(st + 1).
The remaining conclusion follows by the dual argument.
Just as there is an innite family of classical projective planes coordinatized by elds,
so also there are three innite families of generalized quadrangles formed by classical polar
spaces of rank 2 (i.e. having just two types of objects, points and lines) and their duals (so
altogether six families). The more general polar spaces were dened in Section 24.10.
28.3 The O5 (q) and Sp4 (q) Quadrangles. We verify that the points and lines contained
in an O5 (q)-quadric form a generalized quadrangle of order q. Certainly every line has
q + 1 points. If x is any singular point then by Theorem 24.3, x /x O3 (q) which is a
projective plane having q + 1 singular points (the points of a conic). These q + 1 singular
points correspond to q + 1 totally singular lines in x containing x. This veries (GQ1)
and (GQ2).
Let x be a singular point, and let be a totally singular line not containing x.
Then x is a hyperplane not containing , so this hyperplane x meets in a single point
y = x . The totally singular line x, y is the unique line of the quadric containing
x and meeting . Thus (GQ3) holds. This gives the O5 (q) quadrangle, historically
denoted Q(4, q) since it is formed by the points and lines of a quadric in P4 (Fq ).
Recall that by Theorem 24.6, the O5 (q)-quadric has (q 2 + 1)(q + 1) points and the
same number of lines, in agreement with Proposition 28.2. The O5 (2) quadrangle is the
unique generalized quadrangle of order 2 given as our rst example above. Not surprisingly,
the isometry group of the quadratic form of type O5 (2) is isomorphic to S6 .
The O5 (q) quadrangle is self-dual i q is even. For arbitrary q, the dual of the O5 (q)
quadrangle is the symplectic quadrangle Sp4 (q), historically denoted W3 (q) or simply W (q). This is simply the rank 2 symplectic polar space described in Section 24.10.
An explicit duality between these two generalized quadrangles is given by the Klein correspondence . Recall that maps the lines of P3 (Fq ) to the points of the Klein quadric. It
is not hard to check (using the exterior algebra by which was dened) that if we restrict
to just the lines of an Sp4 (q)-quadrangle, the corresponding points of the quadric lie in
an O5 (q)-hyperplane.
28.4 The O6 (q) and U4 (q) Quadrangles. Recall that the O6 (q)-quadric contains
points and lines, but no planes. These points and lines form a quadrangle of order (q, q 2 );
the proof is similar to the previous case. This is the O6 (q) quadrangle, also known as
the Q(5, q) quadrangle. Again by Theorem 24.6, the O6 (q)-quadric has (q + 1)(q 3 + 1)
points and (q 2 + 1)(q 3 + 1) lines, in agreement with Proposition 28.2.
The dual of the O6 (q) quadrangle is a generalized quadrangle of order (q 2 , q), which
also has a classical construction. Here the points and lines are those lying in the Hermitian
181
surface X0q+1 + X1q+1 + X2q+1 + X3q+1 = 0 in P3 (Fq2 ). This is the unitary quadrangle
U4 (q), also known as H(3, q). It is simply the rank 2 unitary polar space of type U4 (q)
described in Section 24.10.
28.4 The U5 (q) Quadrangles. The points and lines of P4 (Fq ) lying in the Hermitian
variety X0q+1 + X1q+1 + + X4q+1 = 0 form a generalized quadrangle of order (q 2 , q 3 ),
known as the unitary quadrangle U5 (q); or historically, H(4, q). This example also
appeared in Section 24.10. Its dual is a generalized quadrangle of order (q 3 , q 2 ), which
however is not readily described other than as the dual of the quadrangle naturally arising
from the Hermitian variety.
The Coxeter-Dynkin diagram for generalized quadrangles is the following diagram,
whose symmetry (like that of the An diagram for projective spaces) indicates that the
class of generalized quadrangles is closed under duality:
generalized quadrangle:
points lines
............................................................................................
(type B2 )
This diagram is the rst member of the innite sequence of diagrams, including the following:
points lines
B2 polar space:
............................................................................................
B3 polar space:
points lines planes

.........................................................................................................................................
Bn polar space:
points lines planes

.............................................................................................................

(n1)-
spaces
n-spaces
..............................................................................................................

n types of objects
(q), Sp2n(q), U2n (q) and U2n+1(q) are represented

Polar spaces of type O2n+1(q), O2n+2
by the latter Bn diagram. Consider for example the points, lines and planes of the O7 (q)quadric, with diagram B3 shown above. Given a singular point x, the totally singular
lines and planes containing x form a geometry isomorphic to the totally singular lines and
planes of x /x O5 (q) by Theorem 24.3. This agrees with the prediction of the CoxeterDynkin diagram, that the residue (Section 19) of the point x should be a geometry of
type B2 (the diagram formed by deleting the node for points from the B3 diagram). The
residue of any plane is a geometry of type A2 , i.e. just a projective plane. And the residue
of a line is a geometry of type A1 A1 , i.e. a generalized digon.
This can be used for counting objects of each dimension in the associated polar
space. For example, we know that the O7 (q)-quadric has (q 6 1)/(q1) points and
(q+1)(q 2 +1)(q 3 +1) planes, by Theorem 24.6. Let N be the number of lines in the quadric.
Then the number of incident point-line pairs (x, ) in the quadric is (q+1)N . But for
182
each singular point x, the quotient x /x O5 (q) is a generalized quadrangle of order q. There are (q+1)(q 2 +1) totally singular lines through x, corresponding to the
(q+1)(q 2 +1) points of the O5 (q) quadrangle. Thus
(q + 1)N =
q6 1
(q + 1)(q 2 + 1) ,
q1
which yields N = (q 2 +1)(q 6 1)/(q1) = (q+1)(q 2 +1)(q 4 +q 2 +1).

Next we give some constructions of nonclassical generalized quadrangles.
28.6 Example: Tits Quadrangles. Let 0 be a plane in P3 (Fq ), and let O be an oval
in 0 . Consider the point-line incidence structure (P, L) with (q 2 +1)(q+1) points given
by
(i) the q 3 points of P3 (Fq ) outside 0 (the ane points),
(ii) the q(q+1) planes P3 (Fq ) such that 0 is a tangent line of O, and
(iii) one additional point denoted ;
and (q 2 +1)(q+1) lines given by
(iv) the q 2 (q+1) lines of P3 (Fq ) such that 0 is a point of O, and
(v) the q+1 points of O.
Incidence between points of type (i)(iii) and lines of type (iv)(v) is the natural containment, with the exception that the point of type (iii) is incident with all lines of type
(v) and no lines of type (iv). It is straightforward to check that (P, L) is a generalized
quadrangle of order q. This quadrangle [25], discovered by Tits, is denoted T2 (O). If O is
a conic (as must be the case when q is odd, by Segres Theorem), then T2 (O) is isomorphic
to the O5 (q) quadrangle. Otherwise T2 (O) is a new generalized quadrangle. The smallest
nonclassical T2 (O) arises for q = 8, where O is taken to be a pointed conic (Section 12).
Similarly, we may take S0 P3 (Fq ) to be a solid (i.e. hyperplane) of P4 (Fq ), and O
an ovoid of the projective 3-space S0 . Consider the point-line incidence structure (P, L)
with (q 3 +1)(q+1) points given by
(i) the q 4 points of P3 (Fq ) outside S0 (the ane points),
(ii) the q(q 2 +1) solids S P4 (Fq ) such that S S0 is a tangent plane to O in S0 ,
and
(iii) one additional point denoted ;
and (q 3 +1)(q 2 +1) lines given by
(iv) the q 3 (q 2 +1) lines of P4 (Fq ) such that S0 is a point of O, and
(v) the q 2 +1 points of O.
Incidence between points of type (i)(iii) and lines of type (iv)(v) is the natural containment, with the exception that the point of type (iii) is incident with all lines of type (v)
183
and no lines of type (iv). Then (P, L) is a generalized quadrangle [25] of order (q, q 2 ), also
discovered by Tits, and denoted T3 (O). If O is an elliptic quadric (as must be the case
when q is odd, by Barlottis Theorem), then T3 (O) is isomorphic to the O6 (q) quadrangle.
Otherwise T3 (O) is a new generalized quadrangle. The smallest nonclassical T3 (O) arises
for q = 8, where O is taken to be a Suzuki-Tits ovoid.
28.7 Example: The Ahrens-Szekeres Quadrangles. The construction of Ahrens
and Szekeres [1] gives nonclassical generalized quadrangles of order (q1, q+1) for any
prime power q. Among the known generalized quadrangles, these parameters (s, t) (or
their duals; see Paynes construction in the following example) in which s and t are not
required to be powers of the same prime, or in fact prime powers at all. The construction
takes dierent forms for q even and q odd. In both cases we denote the resulting generalized
quadrangle by AS(q) (although for most even values of q it is not uniquely determined by
the parameter q).
For q even, the construction is a variation on Tits T2 (O) construction. Let 0 be a
plane in P3 (Fq ), and let O be a hyperoval in 0 , so that |O|=q +2. Consider the point-line
incidence structure (P, L) whose q 3 points are all the points of P3 (Fq ) outside 0 (i.e. the
ane points), and whose q 2 (q+2) lines are the lines of P3 (Fq ) such that 0 is a
point of O. This gives AS(q) for q even.
For q odd, x a symplectic polarity of P3 (Fq ). Recall that the (q 2 +1)(q+1) points
of this projective 3-space, and the (q 2 +1)(q+1) absolute lines, form an Sp4 (q) quadrangle.
Fix one point P of this quadrangle, and consider the incidence structure formed by the q 3
points of P3 (Fq ) not contained in the plane P , and the q 2 (q+2) lines given by
(i) the q 3 +q 2 absolute lines (i.e. = ) not containing P , and
(ii) the q 2 lines of P3 (Fq ) such that P = {P }.
This gives AS(q) for q odd.
Note that AS(2) is just a dual grid of order (1, 3); and AS(3) is isomorphic to U4 (2),
the unique generalized quadrangle of order (2, 4). The smallest new quadrangle obtained
by this construction is AS(4), the unique generalized quadrangle of order (3, 5).
28.8 Example: Paynes Quadrangles of Order (q+1, q1), q even. Let 0 be a
plane in P3 (Fq ) where q is even, and consider a hyperoval in 0 partitioned as O {M, N }
where O is a q-arc. (It is known that for even q > 4, every q-arc extends to a unique
hyperoval. So we must choose simply a hyperoval with two distinguished points M, N .)
Consider the point-line incidence structure (P, L) whose q 2 (q+2) points are given by
(i) the q 3 points of P3 (Fq ) outside 0 (i.e. the ane points); and
(ii) the 2q 2 planes of P3 (Fq ) containing exactly one of {M, N };
and whose q 3 lines are the lines of P3 (Fq ) such that 0 is a point of O.
Payne [51] showed that this gives a generalized quadrangle of order (q+1, q1). In some,
but not all cases, it is dual to AS(q).
184
The following table lists the smallest known generalized quadrangles. For a more
extensive table, together with further information on these examples, including explicit
incidence matrices, see [47].
28.9 Table. Generalized Quadrangles of Small Order
order
(2, 2)
(2, 4)
(3, 3)
(3, 3)
(3, 5)
(3, 9)
(4, 2)
(4, 4)
(4, 6)
(4, 8)
no. of points
15
27
40
40
64
112
45
85
125
165
no. of lines
15
45
40
40
96
280
27
85
175
297
Name
O5 (2), Sp4 (2)
U4 (2), AS(3)
O5 (3)
Sp4 (3)
AS(4)
U4 (3)
O6 (2)
O5 (4)
AS(5)
U5 (2)
There are many open questions regarding existence and classication of generalized quadrangles with small parameters, for example: Does there exist a generalized quadrangle of
order 6? or (4, 11)? Is the generalized quadrangle of order (4, 6) unique?
A spread of a generalized quadrangle, is a collection of lines which partition the point
set. Dually, an ovoid of a generalized quadrangle is a subset O of the points such that
every line meets O in a unique point. In the notation of our rst example, the generalized
quadrangle of order 2 has {(12), (13), (14), (15), (16)} as an ovoid and
{(12)(36)(45), (13)(25)(46), (14)(23)(56), (15)(34)(26), (16)(24)(35)}
as a spread.
28.10 Proposition. In a generalized quadrangle of order (s, t), every ovoid has
st + 1 points and every spread has st + 1 lines.
Proof. Since a spread partitions the (s + 1)(st + 1) points into lines each of size s + 1,
every spread must consist of
(s + 1)(st + 1)
= st + 1
s+1
lines. The dual argument gives the size of an ovoid.
29. Generalized Polygons and Buildings
185
Here we summarize briey what is known about existence and classication of ovoids
and spreads in the classical generalized quadrangles:
The O5 (q) quadrangle always has regular ovoids arising from the embedding of
the O4 (q)-quadric in O5 (q). Other examples of ovoids are known only in characteristic 2 or 3. Spreads do not exist unless q is even, when the generalized
quadrangle is self-dual (in which case spreads arise from spreads by duality).
The O6 (q) quadrangle has spreads but no ovoids.
The U5 (q) quadrangle has no ovoids. No spreads exist for q = 2, and for q > 2
the question of their existence is an open problem.
Of course each of the statements above can be dualized. For example the Sp4 (q) quadrangle
contains regular spreads for every q; indeed every regular spread of projective 3-space
consists of totally isotropic lines with respect to some alternating bilinear form.
It is known that a thick generalized quadrangle with s and t both nite, must have
s t2 and t s2 . A particularly tantalizing open problem asks whether a generalized
quadrangle can have 1 < s < and t innite. The answer is no for s = 2 (by a oneparagraph elementary proof [12, p.86]), for s = 3 (a nontrivial result of Brouwer [7]), and
for s = 4 (a deep model-theoretic result of Cherlin [15]). For larger values of s the problem
remains open.
Exercises 28.
1. Show that the collinearity graph of a generalized quadrangle is strongly regular, and nd its parameters
(v, r, , ) in terms of the order (s, t) of the generalized quadrangle. (See Exercise#4.4 where the
relevant terms from graph theory are dened.)

We give a graph-theoretic denition of generalized polygons, which requires the following
terminology from graph theory. Let be an ordinary graph (so has no loops or multiple
edges). A path of length n from v0 to vn in is a sequence (v0 , v1 , v2 , . . . , vn ) of vertices
in such that vi1 is adjacent to vi for all i = 1, 2, . . . , n; and where v0 , v1 , . . . , vn1 are
distinct. Such a path is closed if n > 0 and v0 = vn ; evidently this requires n 3. We
say is connected if for any two of its vertices, there exists a path from one vertex to
the other. The distance from vertex v to vertex v is the length of the shortest path from
v to v in . This distance is nite whenever is connected; otherwise it is undened (or
innite). This denes a metric d(v, v ) on the vertices of , i.e.
(D1) d(v, v ) 0, and equality holds i v = v ;
(D2) d(v , v) = d(v, v );
(D3) d(v, v ) d(v, v ) + d(v , v )
186
for all vertices v, v , v . The diameter of is the maximum distance between any two of
its vertices (or if is not connected). The girth of is the smallest n for which there
exists a closed path (v0 , v1 , . . . , vn ) having n 3 distinct vertices (so v0 , v1 , . . . , vn1 are
distinct and vn = v0 ). We agree that the girth of is , if no such closed path exists.
We say is bipartite if its vertices can be partitioned into two subsets V1 and V2 , such
that every edge of has one endpoint in V1 and the other endpoint in V2 . Note that
is bipartite i every closed path in has even length. Moreover if is connected and
bipartite, then the choice of partition V1 V2 is unique, up to interchanging V1 and V2 .
Also is complete bipartite if every vertex in V1 is joined to every vertex in V2 (in
addition to no edges having both endpoints in Vi for the same i).
Recall that every point-line incidence structure (P, L) is equivalently described by
its incidence graph , the graph with vertex set P L and where each point-line pair
(P, ) is joined by an edge in , i the pair (P, ) is incident in (P, L). It is understood
here that P and L are disjoint sets; and that in one clearly distinguishes which vertices
correspond to points, and which vertices correspond to lines. Note that the dual incidence
structure (L, P) yields the same incidence graph, but with the labelling of vertices (as
points and lines) reversed. If is connected then the only point-line incidence structures
corresponding to are (P, L) and (L, P).
A point-line incidence structure (P, L) is a generalized n-gon (where n 2) if its
incidence graph satises the following:
(GP1) The set of vertices of is partitioned into points and lines. Every edge of
joins a point with a line; thus is bipartite.
(GP2) has girth n and diameter 2n.
(GP3) Every line has s + 1 2 points, and every point is on t + 1 2 lines.
The order of the generalized polygon is the pair (s, t), or simply s if s = t. If s = 1 or t = 1
then the generalized polygon is thin. If s, t 2 then the generalized polygon is thick.
The Coxeter-Dynkin diagram for a generalized n-gon is
n
..................................................
points
lines
(type I2 (n))
which reduces to A1 A1 , A2 , B2 or G2 when n = 2, 3, 4, 6 respectively. (The diagram is

simplied by drawing an (n2)-fold bond, rather than an n-fold bond.)
For every n 2, an n-gon satises the conditions (GP1)(GP3), giving the unique
generalized n-gon of order 1, which is thin. This example has exactly n points and n lines,
and consists of simply a 2n-cycle, i.e. a closed path of length 2n. The thick examples
are however more rare and much more interesting.
Consider rst the case n = 2. Here we have a point set and a line set (we should
really call them blocks) such that every point is incident with every block; equivalently,
is a complete bipartite graph. To see this, note that is bipartite by (GP1). If there
187
exists a point P and a line which are not incident, then since the distance from P to is
odd, we must have d(P, ) 3, contrary to the diameter of being 2. Thus a generalized
2-gon of order (s, t) is the same thing as a generalized digon (Section 19) with s + 1 points
and t + 1 lines.
Now consider the case n = 3. We claim that a generalized triangle (i.e. a generalized
3-gon) is simply a projective plane (or possibly a triangle, i.e. a thin projective plane,
whose order is 1) and we must have s = t, the order of the plane. We now justify these
conclusions. If two points P, Q are not joined by any line, then d(P, Q) > 2; and since
d(P, Q) is even, we have d(P, Q) 4, contrary to the diameter of being 3. If P and Q
lie on two distinct lines and m then (P, , Q, m, P ) is a closed path of length 4, violating
the girth of being 6. This proves the rst projective plane axiom (P1), and (P2) follows
dually. Note that the nite girth assumption rules out the closed congurations listed in
Proposition 9.1(i)(iv), leaving only cases (v) and (vi) of that Proposition. In fact (GP3)
allows only the cases of a triangle or a projective plane.
Next consider the case s = 4. We claim that a generalized 4-gon is simply a generalized
quadrangle as dened in Section 28. Let P be a point not on a line , so that d(P, ) > 1.
Since d(P, ) is odd and the diameter of is 4, we must have d(P, ) = 3. This means
there is a path (P, m, Q, ) from P to in . If there is another line m joining P to ,
as shown, then we have a closed path (P, m, Q, , Q , m , P ) of length 6, which is less than
the girth 8, a contradiction. This veries our claim.
P m
m
Q
...........
. .....

.....
... .. .......
.......
..
.......
.......
....................
...
.
..
..
....... ....... ....
...
........

... ..............
...........
.
.
.
.
.
....
....... ....
.......
........
It is interesting to note that thick generalized n-gons only exist for n {2, 3, 4, 6, 8},
and are apparently quite rare for n {6, 8}.
29.1 Theorem (Feit and Higman [27]).
for n {2, 3, 4, 6, 8}.
Finite thick generalized n-gons exist only
The known generalized hexagons (n = 6) and generalized octagons (n = 8) are as follows:

Generalized hexagons of type G2 (q) having order q for every prime power q, and
their duals (these geometries are however self-dual for q = 3r ). Each such generalized hexagon has (q 6 1)/(q1) points and the same number of lines. The smallest
example, G2 (2), having 63 points and 63 lines, is the unique generalized hexagon of
order 2.
188
Generalized hexagons of type 3 D4 (q) having order (q, q 3 ) for every prime power q.
These have (q+1)(q 8 +q 4 +1) points and (q 3 +1)(q 8 +q 4 +1) lines. Also their duals.
The smallest example, of order (2, 8), has 819 points and 2457 lines.
Generalized octagons of type 2 F4 (q) having order (q, q 2 ) for q = 22e+1. These have
(q+1)(q 3 +1)(q 6 +1) points and (q 2 +1)(q 3 +1)(q 6 +1) lines. Also their duals. The
smallest example, of order (2, 4), has 1755 points and 2925 lines.
These families are named for the exceptional groups G2 (q), 3 D4 (q), 2 F4 (q) of Lie type;
see [63] for details. The smallest examples are explicitly provided in [47]. Despite much
eort, no one has a clue why these should be the only generalized hexagons and octagons.
In particular why should there not exist modications of the families listed above, in the
same way that the standard families for n = 3, 4 admit modications. It is still an open
question whether G2 (3) is the unique generalized hexagon of order 3.
Even from a graph-theoretic viewpoint, generalized polygons (thought of in terms of
their incidence graphs) are very interesting as they provide examples of sparse but highly
connected graphs. The competing goals of being sparse but highly connected are desirable
in the design of ecient communication networks, or for constructing good LDPC (low
density parity check) codes. There is more than one way to formally express the condition
sparse but highly connected; and having large girth and small diameter does this quite
nicely, although it is very hard to nd explicitly constructible families of graphs which
achieve these conditions.
Beginning in the 1950s, Jacques Tits dened a class of incidence structures known
as buildings [57], [64], a notion which includes both projective spaces and polar spaces
as special cases. A building of rank r has r types of objects. A building of rank 2
is simply a generalized polygon. In the general case, the structure of the building is
described by a Coxeter-Dynkin diagram of the appropriate type, having r nodes, one node
representing each type of object in the geometry. Edges of the graph specify the type of
generalized polygons arising as rank 2 residues. A thin geometry corresponding to the
given Coxeter-Dynkin diagram, is known as a Coxeter complex. A building (in the strict
sense) is a thick geometry corresponding to the given diagram, which is a union of certain
thin subgeometries (known as the apartments of the building) with the same diagram.
Thus for example, any projective plane may be formed as a union of triangles:
......
..............
...... .............
......
.......
.
.
.
.
.
......
.....
.....
.......
...
...
...
...
...
...
...
...
...
...
...
...
.
........
..............
.
.
.
.
.
..
.
.......
.
.
.
......
.....
.......
......
....... ............
.......
thin A2
geometry
(triangle)
..
....
..........
..........
.............
.............
......
.... ...
.... ..
...........
..........
....... ...............
.......
............ ............ ...
... .............. .... .............. .... .............. .... ...................... .... .............. ................................. ....
..
.. .. ...... .. .. ...... .. .. ................. ................. .......... ..........
...... .. ........... .. .................... ......... ..
...... ..
...
... ..
.
....... ....... ....... .................... ....... .. ..
....... ..
.. ...
.. ... ... ............ ....................... ................................. ....................... ... .... ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
.
.
.
.
.
.. ..
... .
.. .
.. ...
...
.......... ...
... .. ......................... ...................................... ..........................
.
.
..
.
.. .......... ......................... .. ............... .. ........ .... ............ .... ...
..
. ... ....... ....... . ... ... . . ... .. . ... . .
...
.. .................................................... ........................... .......... .... .... .......... .... .... .......... .... ....
..... ....
..... ....
..... ....
... ................. ...... ....... ...... ....... ............
.........
.........
.........
.......
........
........
.
.
.
.........
thick A2 geometry
(projective plane)
189
Another instructive example is the Coxeter diagram

............................
..
....................
.1 )
(type A
whose buildings are innite trees. Take for example a k-regular tree where k 2: this is a
(necessarily innite) connected graph with no cycles (closed paths), in which every vertex
has exactly k neighbours. This graph is bipartite; so we may choose an arbitrary vertex v
to represent objects of type 1, and then any other vertex v has type 1 or 2 according as
the unique path from v to v has even or odd length. Such a tree is the incidence graph
.1 -geometry. For k = 2 the geometry is thin, consisting of just an innite path.
of an A
.1 whose apartments are
For k 3 the geometry is thick; it is in fact a building of type A
innite paths contained in the tree.
k = 2:
k = 3:
Buildings were invented to study groups of Lie type. The groups of Lie type are
classied by their Coxeter-Dynkin diagrams, and each such group is best understood by
means of the natural geometry on which it acts: a building of the corresponding type.
Thus for example, the group P GLn+1(F ) corresponds to the Coxeter-Dynkin diagram of
type An and has Pn (F ) as its associated building.
Consider a Coxeter-Dynkin diagram with r nodes labelled 1, 2, . . . , n, and edges labelled mij . Also set mii = 1. The group dened by the presentation
4
5
W = s1 , s2 , . . . , sr : (si sj )mij = 1
is the Coxeter group dened by the given diagram. It acts regularly on the objects of
the associated Coxeter complex (i.e. thin building). In particular, the number of objects
190
in the Coxeter complex equals |W |. We illustrate with four Coxeter-Dynkin diagrams as

examples:
29.2 Example.
1
2
.....................................
Type A2

mij
2 1
1
=
3
3
1
......................
..............................................................
...................
................
..................
...............................
.
.
.
.
.
.................
.
...
.............
2 1 ......................................
............
..............
.
.
.
.
.
.
.
.
.
.
.....
.............. 1
.
.
.
.
.
.
.
.
1 2.................
.............
............
....................
............
...........
..........
.......
..........
.........
..........
.......
.........
... ...
.. ...
... ...
........
.. ...
.
..
... ..
......
......
........
.
.......
.
.
......
1 2 ..................
........
...........
............
.
.
.
.
.
.
............
. .
...................
............
................................
...................
...................................................................................................
.......
1
= 2
5
4
W = 1 , 2 : 12 =22 =(1 2 )3 =1
= {1, 1 , 2 , 1 2 , 2 1 , 1 2 1 }
W
= S3 via 1 (12), 2 (23)
29.3 Example. Type A1 A1

1
2

1 2
mij =
2 1
4
5
W = 1 , 2 : 12 =22 =(1 2 )2 =1
= {1, 1 , 2 , 1 2 }
Coxeter
complex
1 2 ..............................................................................
..................
= 2 1......................................
.................2
..........
...........
..
..........
......
..........
.....
..........
........
........
.....
......
.....
...........
............
..
.........
......
.
.........
.
.....
.
..........
.
....
.........
......
..............
.....
...........
...........
.
.............
.
.
.
1 ............................................................................................
....................
Coxeter
complex
W is elementary abelian
1 2
.1
29.4 Example. Type A
..............................................
4
5
W = 1 , 2 : 12 =22 =1 is innite dihedral; 1 2 has innite order
Coxeter complex:
1 2 1 2
1 2
2 1
2 1 2 1
1
............................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
1 2 1
2
2 1 2
1
29.5 Example. Type A3
1
2
3
...........................................................................................
4
W = 1 , 2 , 3 : 12 =22 =32 =(1 2 )3 =(1 3 )2 =(2 3 )3 =1
W
= S4 via 1 (12), 2 (23), 3 (34)
mij = 3
2
5
The corresponding Coxeter complex is a triangulated 2-sphere:
3 2
1 3
3 1
191
This is in fact the geometry of the 3-simplex formed by all its subsimplices (Exercise #19.2). The 4 points, 6 lines, and 4 planes of this geometry are the subsets of
{1, 2, 3, 4} of size 1, 2, and 3 respectively. These give the 14 vertices of the triangulation. The 36 edges correspond to the incident pairs of objects (12 incident point-line
pairs, 12 incident point-plane pairs, and 12 incident line-plane pairs). The 24 triangular faces correspond to the triples (P, , ) such that P . We check that
14 36 + 24 = 2, the Euler characteristic of the 2-sphere.
The Coxeter complex (or diagram, or matrix, or group), in turn, gives for each choice
of eld F a building, as a thickened-up version of the Coxeter complex. It also gives
a presentation of the corresponding group G of Lie type dened over F , by means of
generators and relations. In addition to the r generators of the Coxeter group (where r is
the rank of the building), one generates G by means of certain subgroups whose elements
are parameterized by F , and these elements must satisfy certain new relations as prescribed
by the Coxeter group.
Coxeter gave a beautiful characterization of those diagrams for which W is nite.
And in these cases, the associated Coxeter complex is a triangulated (r1)-sphere, as
the examples above suggest. Buildings corresponding to these diagrams are known as
spherical buildings. Thus generalized n-gons are examples of spherical buildings. Tits,
who rst dened buildings axiomatically, also showed that with only certain exceptions of
small rank, spherical buildings are classical: they are the buildings associated with groups
of the corresponding Lie type. Of course the rank 2 buildings (generalized polygons) are
far from classied, however. And even though higher rank buildings may in some sense
be known, there remain many open problems regarding these geometries (for example,
questions of existence of ovoids and spreads).
Although weve exhausted the time available this semester, we have really just arrived
at the beginning of a very big subject that would require another course or two to introduce
properly!
192
Exercises 29.
1. For a generalized hexagon of order (s, t), determine
(a) the number of points and lines;
(b) the size of an ovoid and the size of a spread.
Justify your answers.
2. For a generalized octagon of order (s, t), determine
(a) the number of points and lines;
(b) the size of an ovoid and the size of a spread.
Justify your answers.
194
Appendix A1: Finite Fields

The ring of integers modulo a prime p is a eld, denoted Fp . For every r 1 there exists
a eld of order q = pr , which is unique up to isomorphism. This eld, denoted Fq , is an
extension of Fp of degree r. It may be constructed as the quotient ring Fp [X]/(f(X))
where f(X) Fp [X] is irreducible over Fp of degree r. Such an irreducible polynomial
f(X) of degree r exists; and although not unique, the quotient eld Fp [X]/(f(X)) is
unique up to isomorphism.
The automorphism group of Fq is = {1, , 2 , . . . , r1 }, a cyclic group of order r
where
: Fq Fq ,
x xp .
The xed eld of is just Fp , so that the extension Fq Fp is Galois; more generally the
xed eld of k is Fpd where d = gcd(k, r).
........
{0} of any nite eld is cyclic: we have F
The multiplicative group F
q = Fq
q =
{1, , 2 , . . . , q2 } for some F
q which is called a primitive element. In particular
Fp [] = Fq , which says that generates the eld extension Fq Fp and so its minimal
polynomial over Fp , call it f(X) = Irr,Fp (X), is of degree r. But the condition that is
a primitive element is stronger, and the resulting polynomial f(X) is called a primitive
polynomial.
A1.1 Example: The field of order 16. Let f(X) = X 4 + X + 1 F2 [X]. This
polynomial is irreducible. (If necessary this can be checked by considering all possible
linear and quadratic factors, since altogether there are only six of these.) We denote by
a zero of f(X) in F16 , so that
4 = +1;
10 = 4 + 2
5 = 2 +;
11 = 3 + 2 +;
6 = 3 + 2 ;
12 = 4 + 3 + 2
7 = 4 + 3
= 3 ++1;
= 2 ++1;
= 3 + 2 ++1;
13 = 4 + 3 + 2 + = 3 + 2 +1;
8 = 4 + 2 + = 2 +1;
14 = 4 + 3 +
9 = 3 +;
15 = 4 + = 1.
= 3 +1;
Since F16 = {0, 1, , 2 , 3 , . . . , 14 }, the element is primitive. Minimal polynomials for
196
Appendix A1: FINITE FIELDS
each of the elements of F16 are listed as follows:
Irr,F2 (X)
, 2 , 4 , 8
X 4 +X+1
3 , 6 , 9 , 12
5 , 10
X 4 +X 3 +X 2 +X+1
X 2 +X+1
7 , 11 , 13 , 14
X 4 +X 3 +1
1
0
X+1
X
k
Since F
16 is cyclic of order 15, it has (15) = 8 generators; these are where gcd(k, 15) =
1. The minimal polynomials of these elements, namely X 4 +X+1 and X 4 +X 3 +1, are the
two primitive polynomials of degree 4 over F2 . The elements 3 , 6 , 9 , 12 are imprimitive
but they are algebraic of degree 4, and so their minimal polynomial X 4 +X 3 +X 2 +X+1
is also irreducible over F2 ; this makes altogether three irreducible polynomials of degree 4
over F2 . The automorphism group of F16 is = {1, , 2 , 3 } and the xed eld of 2 is
F2 [ 5 ] = F4 . The diagram of subelds, and the diagram of subgroups of , are as shown:
F16 =F2 []
F4 =F []
.2
F2
1
....
..
..
..
...
...
.
...
...
..
...
..
...
.
.............
...........
........
........
......
......
......
.....
.
.
......
.
.
...
......
......
......
......
......
......
......
.
.
.
.
.
.
.
... ...
.....................................................................................................................
2
.... .........
......
......
.
.
.
.
.
......
.....
......
.....
......
......
.
.
......
.
.
.
...
......
.
.
.
.
...... .
..
.
.
.
.
.
........
.
................
.
..........
....
..
..
..
...
...
.
...
...
..
...
..
...
.
The arrows show the Galois correspondence between each subgroup of Aut F16 and its
xed subeld.
Consider an extension E F of nite elds, so that E = Fqn and F = Fq for some
prime power q, where n = [E : F ] 1. The group of all F -automorphisms of E is the Galois
group
G = G(E/F ) = {1, , 2 , . . . , n1 }
where : E E, x x . (This generalizes the case F = Fp considered above.) The
norm and trace maps of the extension E F are the maps

2
n1
2
n1
xg = x1++ ++
= x1+q+q ++q
NE/F : E F, x
gG
and
TE/F : E F,
x

gG
xg = x + x + x + + x
n1
= x + xq + xq + + xq
n1
197
The trace map TE/F : E F is F -linear and surjective; in particular

TE/F (x + y) = TE/F (x) + TE/F (y).
The fact that TE/F is surjective follows from the fact that as a polynomial of degree
q n1 < q n , the trace TE/F (x) cannot vanish at every element of E; and choosing x E
such that TE/F (x) = 0, we see that TE/F (cx) = cTE/F (x) takes on all values in F as we
vary c F . The norm map NE/F : E F is multiplicative, i.e.
NE/F (xy) = NE/F (x)NE/F (y).
It is also surjective, which we justify as follows. Since NE/F (0) = 0, it suces to consider
the restriction of NE/F to the multiplicative group E , a cyclic group of order q n 1.
Since this map is x xm where the exponent m = 1+q+q 2 + +q n1 = (q n 1)/(q1)
divides |E |, the image is therefore the subgroup of order q 1, i.e. F as required.
A1.2 Matrix Representation of Fields. Let E F be a eld extension of degree
n 1, and x a basis {1 , . . . , n } for E over F . For every E the multiplication map
: E E, x x is F -linear and so may be represented as an n n matrix over F .
Clearly + = + and = so the set of maps { : E} (or just as
well, the corresponding set of n n matrices) is a ring, in fact a eld, isomorphic to E via
.
For example consider the extension C R with basis {1, i}; a typical element = a+bi
gives rise to the multiplication map
a+bi(x + yi) = (axby) + (bx+ay)i
which in matrix form appears as
x y
This gives a subring

a b
= axby
b a

R=
b a

: a, b R

bx+ay .
R22
=
where R22 denotes the ring
of all 22 matrices over R; moreover an isomorphism C R
a b
is given by a+bi b a .
More generally, consider any eld extension E F of degree n and write E = F []
=
F [X]/(f(X)) where f(X) F [X] is a monic polynomial of degree n which is irreducible
over F , and f() = 0. We may write
f(X) = X n + an1 X n1 + + a1 X + a0 ,
ai F.
198
The F -linear transformation : E E, x x maps

1 ;
2 ;
2 3 ;
..
.
n2 n1;
n1 n = a0 a1 an1 n1
and so the matrix of relative to the basis {1, , 2 , . . . , n1 } is given by
0
1
0
0
0
0
0
1
0
0
0
0
1
0
.
..
..
..
..
A=
..
...
.
.
.
.
.
0
0
0
0
1
a0 a1 a2 a3 an1
The matrix A is known as the companion matrix of f(X). Since generates the
extension E F , the matrix A generates a subring R F nn where F nn is the ring of
=
R is determined by A.
all n n matrices over F ; moreover an isomorphism E
With this identication of E with R, the norm and trace maps
NE/F : E F,
TE/F : E F
of the extension E F , dened as above, become simply the familiar determinant and
trace maps
det : R F,
tr : R F
of matrix theory.
Let us revisit the extension F16 F2
matrix
0
0
A=
0
1
described in Example A1.1. In this case the

1
0
0
1
0
1
0
0
0
0
1
0
generates a subring

R = F2 [A] = a+bA+cA2 +dA3 : a, b, c, d F2 F44
;
2
=
R is given by
an isomorphism F16
a+b+c 2 +d 3 a+bA+cA2 +dA3 .
199
We see that nite extensions E F may be generally represented as rings of nn

matrices; and in principle this allows us to perform any desired computations in E explicitly. However in practice, the representation of E as a quotient ring F [X]/(f(X)) is
much easier to implement, either by computer or by hand, since it represents elements of
E as polynomials of degree < n, i.e. n-tuples over F ; whereas n n matrices require n2
elements of F to represent every element of E. The matrix representation is primarily of
conceptual value.
A1.3 Proposition. Consider a nite eld Fq . Recall that its nonzero elements F
q
form a cyclic multiplicative group of order q 1.

1
(i) Suppose q is odd. Then F
q has 2 q1 squares and the same number of nonsquares. The element 1 is a square i q 1 mod 4. Every nonzero square has
exactly two square roots in Fq .
(ii) If q is even then every element of Fq has a unique square root in Fq .
Proof. The set of nonzero squares is just the image of the group homomorphism :
F
q Fq , x x . Note that Fq is cyclic of order q1, which is odd (or even) according
as q is even (or odd). The result follows from elementary properties of cyclic groups. In
particular for q odd, the unique element 1 of order two is a square, i F
q has an element
of order 4, i 4 divides |Fq | = q 1.

Suppose q = 2r is even, so that F = Fq is an extension of K = F2 . The map
F F,
x x2 + x
is K-linear. Denote the image of this map

C0 = {x2 + x : x Fq } Fq
which is a K-subspace of codimension 1. Also denote the complementary set
C1 = Fq ........ C0 .
Note that |C0 | = |C1 | =
1 respectively.
q
2
. Elements of C0 and C1 are called elements of class 0 and class
200
A1.4 Theorem. Consider a quadratic polynomial f(X) = aX 2 + bX + c Fq [X]

with a = 0.
(i) Suppose q is odd and let = b2 4ac. Then f(X) has two distinct linear factors,
or one repeated linear factor, or is irreducible over Fq , according as is a nonzero
square, or zero, or a nonsquare. If is a square then the zeroes of f(X) are
b
r1 , r2 =
Fq .
2a

(ii) Suppose q is even. If b = 0 then f(X) = a X + ac )2 has a double zero c/a.
Now suppose b = 0. Then f(X) has two distinct zeroes, or is irreducible over Fq ,
according as ac
b2 is of class 0 or 1 respectively. If r Fq is a zero then the other
zero is r + ab .
Proof. The case of odd characteristic is well-known and the proof is the same as in
characteristic zero. So consider q = 2e . The case b = 0 is clear, so assume b = 0. The
condition for an element r Fq to be a zero of f(X) is that
u2 + u =
ac
,
b2
where u =
ar
.
b
Consider the trace map T = TFq /F2 : Fq F2 , t t+t2 +t4 + +t2

T (u2 ) = u2 + u4 + + x2
e1
e1
. Note that
+ x = T (u)
since tq = t for all t Fq . So a necessary condition for f(X) to have zeroes in Fq is that
T
i.e.
ac
b2
ac
b2
= T (u2 ) + T (u) = 2T (u) = 0,
must be of class 0. If r is indeed a zero of f(X) then clearly

f(X) = aX 2 + bX + c = a X + r X + r + ab .
By convention we agree that 00 = 1.

A1.5 Lemma.
Let k {0, 1, 2, . . . , q1}. Then

aFq
a =
0,
if k < q1;
1, if k = q1.
201
Proof. For k = q1 all terms in the sum are 1 except the term 0q1 = 0, so the sum is

q1 = 1 (in Fq ). Now suppose k {0, 1, 2, . . . , q2} and denote Sk = aFq ak . For
every F
q the map F F, x x is a permutation so
Sk =
ak =
aFq
(a)k = k
aFq
ak = k Sk .
aFq
k
Now (k 1)Sk = 0 for all F
q . Since the polynomial X 1 has at most k q2
k
zeroes there exists F
q such that = 1, and this forces Sk = 0.
A1.6 Theorem (Chevalley-Warning). Consider a system of polynomials fi (X1 ,

. . . , Xn ) Fq [X1 , X2 , . . . , Xn ] for i = 1, 2, . . . , k of total degree i deg fi (X1 , . . . , Xn )
< n. Let S be the set of all simultaneous zeroes of the fi s in Fnq , i.e.

S = (a1 , a2 , . . . , an ) Fnq : fi (a1 , a2 , . . . , an ) = 0 for all i = 1, 2, . . . , k .
Then |S| is divisible by p, where q = pr .
Proof. We rst extend Lemma A1.5 by observing that if k1 , k2 , . . . , kn 0 with
(q1)n, then

ak11 ak22 aknn =
aki = 0;
a1 ,a2 ,...,an Fn
q

i
ki <
1in aFq
this holds since at least one of the exponents ki < q1. Now consider the polynomial

1 fi (X1 , X2 , . . . , Xn )q1 Fq [X1 , X2 , . . . , Xn ]
f(X1 , X2 , . . . , Xn ) =
1ik
ki < (q1)n. Clearly

1, if (a1 , a2 , . . . , an ) S;
f(a1 , a2 , . . . , an ) =
0, otherwise.
of total degree (q1)
By considering each monomial term in the multinomial expansion of f(X1 , X2 , . . . , Xn ),

we have

f(a1 , a2 , . . . , an ) = 0 Fq .
a1 ,a2 ,...,an Fn
q
Since the sum has |S| ones and all other terms are zero, the result follows.
Let f : Fq Fq . Then f can be uniquely expressed as a polynomial in Fq [X]
of degree less than q, for example using Lagrange interpolation. A polynomial f(X)
202
Fq [X] of degree less than q is called a permutation polynomial if the function Fq

Fq represented by f is bijective. We give Dicksons condition for a polynomial to be a
permutation polynomial.
First observe that if f(X) Fq [X] is any polynomial, we may divide f(X) by X q X
to obtain f(X) = (X q X)g(X) + r(X) where deg r(X) < q. Then r(X) is the unique
polynomial of degree less than q which represents the same function Fq Fq as the original
f(X). We call r(X) the reduction of f(X) mod (X q X).
A1.7 Theorem (Dicksons Criterion). Let f(X) Fq [X] be a polynomial of
degree less than q. Then f(X) is a permutation polynomial i the following two
conditions hold:
(i) f(X) has exactly one zero in F ; and
(ii) for every k {1, 2, . . . , q2}, the reduction of f(X)k mod (X q X) has degree at
most q2.
Proof. First suppose f(X) is a permutation polynomial, so that (i) holds. For each
k {0, 1, 2, . . . , q2} write

bki X i mod (X q X)
f(X)k
0i<q
where bki Fq . By Lemma A1.5 we have

f(a)k =
bki
ai = bk,q1
0=
aF
0i<q
aFq
for k = 0, 1, 2, . . . , q2.
1
Conversely, suppose (i) and (ii) hold. For each a Fq let na = |f (a)|, which is the
number of solutions of f(x) = a for x Fq . But
(A1.8)

aFq
na a =
0,
for k = 0, 1, 2, . . . , q2;
1, for k = q1
ak .
aFq
The validity of (A1.8) follows from (i) for k = q1, and from (ii) for k = 0, 1, 2, . . . , q2
by Lemma A1.5. For each b Fq consider the polynomial

cbk X k Fq [X].
gb (X) = 1 (X b)q1 =
0k<q
For each a Fq we see that

gb (a) = ab =
1, if a = b;
0, otherwise.
Clearly
cbk
0k<q
203
(na 1)ak = 0
aFq
since the inner sum vanishes by (A1.8); but after interchanging the order of summation
we obtain

(na 1)
cbk ak =
(na 1)ab = nb 1 Fq
0=
aFq
0k<q
aFq
which says that nb 1 mod p for all b Fq . Since the nb s are non-negative integers
whose sum is q, this clearly implies that nb = 1 Z for all b Fq .
Appendix A2: Groups

A2.1 Permutation Groups
Let X be a set. The collection of all bijections X X is a group under composition,
called the symmetric group on X, denoted Sym X. Elements of Sym X are called permutations of X. Permutations are often written using standard cycle notation, especially
when X = {1, 2, . . . , n}, in which case we abbreviate Sn = Sym{1, 2, . . . , n}. The action
of a permutation Sym X on X is denoted either using superscripts, as in x x
(right action); or using the usual function notation x (x) (left action), depending
on context. We prefer right action, unless this results in too many nested subscripts and
superscripts. Multiplication in Sym X is left-to-right composition in the case of right actions, so that (x ) = x ; or right-to-left composition in the case of left actions, so that
( )(x) = ( (x)).
A permutation group on X is a subgroup G X. More generally given an arbitrary group G, a permutation action or permutation representation of G on X is
a homomorphism : G Sym X. Such an action is faithful if is injective, in which
case G
= (G) so that we may identify G with the permutation group (G) Sym X. If
G acts on X via , we denote the action of an individual element g G by x x(g) or
simply x xg if the choice of action is clear from context. The degree of G is |X|, the
number of points being permuted.
Let G be a permutation group on a set X, or more generally a group acting on a
set X. For each x X, the stabilizer of x is the subgroup
Gx = {g G : xg = x}.
The orbit of x is the subset
xG = {xg : g G} X.
It is well-known that the orbits of G form a partition of X; and that for every x X the
index of the stabilizer equals the size of the orbit:
A2.2 Theorem. If G acts on X and x X, then [G : Gx ] = |xG |.
Assume G is a finite group acting on a finite set X, and let x1 , . . . , xn X be representatives of the distinct orbits of G on X. The set of points of X xed by g G is denoted
by
FixX (g) = {x X : xg = x}.
Appendix A2: GROUPS
205
A2.3 Theorem. The average number of points of X xed by elements of G, equals

n, the number of orbits. That is,
1
|G|
|FixX (g)| = n.
gG
Proof. Let S be the set of pairs (x, g) with x X, g G and xg = x. We count in two
dierent ways the number of such pairs. On the one hand
|S| =
|FixX (g)|.
gG
On the other hand, for each x xG

i the number of elements of G xing x is
|Gx | =
thus
|S| =
|G|
|G|
;
=
G
|x |
|xG
i |
|Gx |
xX
1in
xxG
i
|Gx |
|G|
|xG
i |
G
1in xx
i
|xG
i |
1in
|G|
|xG
i |
|G|
1in
= n|G|.
A permutation group G (or more generally, an action of a group G) on X is transitive
if there is only one orbit; that is, if for all x, x X there exists g G such that xg = x .
We also say G is sharply transitive, or regular, if the choice of g G mapping x x
is unique. Note that G permutes X regularly i |G| = |X| and every point x X has
trivial stabilizer Gx = 1. We say that G is doubly transitive or 2-transitive on X if
G is transitive on ordered pairs in X, i.e. if for all pairs (x, y) and (x , y ) with x = y and
x = y in X, there exists g G such that (xg , y g ) = (x , y ). If moreover the choice of
g G is unique, we say G is sharply 2-transitive on X.
206
Appendix A2: GROUPS
The following is often useful in determining the size and structure of an automorphism
group.
A2.4 Theorem. Let G be a group acting on a set X, and suppose H G is a
transitive subgroup. Then for every element x X, the stabilizer Gx satises
G = Gx H = {kh : k Gx , h H}.
Proof. Let g G. Since H is transitive, there exists h H mapping x xg . Then h
1
1
1
maps xg x, i.e. xgh = x. Thus gh Gx and g = (gh )h Gx H.
Let : G Sym X and : G Sym Y be two permutation representations of the

same group G, on possibly dierent sets X, Y . We say these two permutation representations are equivalent if there exists a bijection f : X Y such that f(x(g) ) = f(x)(g)
for all x X and g G. Unravelling this denition, we see that two actions of G are
equivalent i one is obtained from the other by simply renaming the points that are being
permuted. This is the natural equivalence relation for permutation groups and actions,
just as isomorphism is the natural equivalence relation for abstract groups.
For example, let G be the symmetry group of the square in the Euclidean plane as
shown, so that G is dihedral of order 8:
.
D..... ....
.... D
..............................c
.....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
....
3................
.. .. 2
.... ...
...
...
....
..
...
..
..
....
...
.....
..
.
..
. . . ....
...
.......
...
...
...
...
...
. ....
.
.
..
.
...
..
.
..
.
....
...
...
..
.....
.. .....
.... ....
. . ..
... ... .
.............................................................................
.....
..
.
....
.
....
1.
.....
Then G has several natural actions, including an action on the set of vertices X =
{1, 2, 3, 4}; and an action on the set of four sides Y = {a, b, c, d}; and an action on the
set of diagonals Z = {D, D }. No two of these actions are equivalent. For example, let g
be the reection in the diagonal D; then the action of g on vertices is given by (g) = (13),
whereas the action on edges is given by (g) = (ab)(cd). Since (g) is a transposition
whereas (g) is a product of two disjoint transpositions, the two actions are inequivalent.
The third action is not even faithful; four of the elements of G act trivially on Z (i.e.
they both diagonals) whereas the other four act as (D D ). Note that the two diagonals
and the other two axes of symmetry divide the square into eight isosceles triangles, as
shown:
........................................................................................................
...................................... ....
..................
...
..............................
......................
...
......................
......................
............................
...
........
.......................................... ...........
...
.................................................. .
.
.........................................
...
.
.
.
.
.
.
.
.
.
.
.
.
..
... .......................
..
............. ................................................
.
.
.
...
.......................
..........
..................
..
...................
.............
.. .........................................
.........
... ...............................
.............................................................................................
Appendix A2: GROUPS
207
Then G regularly permutes these eight triangles.

Every group G has a regular permutation action on the elements of X = G by rightmultiplication; this is Cayleys Representation Theorem. More generally, let H G be
any subgroup, and consider the set H\G = {Hg : g G} of right cosets of H in G.
Then G acts transitively on X = H\G by right-multiplication; moreover every transitive
permutation action of G is equivalent to such an action for some choice of subgroup H G.
Indeed, given a transitive permutation action of G on X, let x X and H = Gx ; then
the action of G on X is equivalent to the action of G on Gx \G. In the case Gx = 1, we
recover the regular representation.
A2.5 Linear Groups
Let F be a eld. The general linear group GLn (F ) is the multiplicative group consisting
of all invertible nn matrices over F . Its centre is the subgroup of all scalar matrices, i.e.
nonzero scalar multiples of the identity matrix; thus Z(GLn (F )) is naturally isomorphic
to F , the multiplicative group of all nonzero scalars. The projective general linear
group is the quotient group
P GLn (F ) = GLn (F )/Z(GLn (F )).
Its elements may be thought of as invertible matrices, but with two such matrices identied
whenever one is a nonzero scalar multiple of the other. The group P GLn (F ) has a natural
permutation action on the set of all 1-dimensional subspaces of F n , since multiplication by a
nonzero scalar xes every 1-dimensional subspace. Similarly for each k {1, 2, . . . , n1},
the group P GLn (F ) has a natural permutation action on the set of all k-dimensional
subspaces of F n . We easily check that this action is faithful.
Now suppose F = Fq . The number of invertible n n matrices over Fq is
|GLn (Fq )| = (q n 1)(q n q)(q n q 2 ) (q n q n1 )

since if A = v1 v2 vn is such a matrix, where v1 , . . . , vn are the columns of A, then
there are
q n 1 choices of v1 , i.e. any nonzero vector of length n;
q n q choices of v2 for each v1 , i.e. any vector not a multiple of v1 ;
q n q 2 choices of v3 for each (v1 , v2 ), i.e. any vector not in v1 , v2
;
etc.; and nally
q n q n1 choices of vn for each (v1 , v2 , . . . , vn1 ), i.e. any vector not in v1 , v2 , . . . , vn1
.
Consequently
|P GLn (Fq )| =
n
1
q1 (q
1)(q n q)(q n q 2 ) (q n q n1 ).
208
Appendix A2: GROUPS
For example we see that |P GL2 (F2 )| = 6. Since this group faithfully permutes
the three 1-dimensional subspaces of F22 (the 21 2 = 3 points of P1 (F2 )) we must have
P GL2 (F2 )
= S3 . In fact GL2 (F2 )
= S3 since the centre of GL2 (F2 ) is trivial.
In the same way, |P GL2 (F3 )| = 24. From the action of P GL2 (F3 ) on the four points
2
of P (F3 ) we see that P GL2 (|F3 )
= S4 .
Similarly, |P GL2 (F4 )| = 60 and P GL2 (F4 ) permutes the 5 points of P2 (F4 ). Since S5
has a unique subgroup of order 60, we obtain P GL2 (F4 )
= A5 , the alternating group of
degree 5.
Let V be a vector space over a eld F . A map T : V V is semilinear if there
exists an automorphism Aut F such that
T (au + bv) = a T (u) + b T (v)
for all u, v V ; a, b F.
The case = 1 gives simply a linear transformation as a special case. If Aut C is

complex conjugation, we call T simply a conjugate linear transformation. The group
of all semilinear transformations V V is denoted L(V ), or Ln (F ) if V = F n . Again
we write
P L(V ) = GL(V )/Z
where Z P L(V ) is the normal (although not central) subgroup consisting of all scalar
maps v v where 0 = F . We have
|Ln (Fq )| = r|GLn (Fq )|;
|P Ln (Fq ) = r|P GLn (Fq )|
where q = pr and the orders of GLn (Fq ) and P GLn (Fq ) are listed above. Every T L(V )
is uniquely representable in the form
v Av
where A GL(V ) and Aut F acts coordinatewise on V ; thus (v1 , v2 , . . . , vn ) =
(v1 , v2 , . . . , vn ).
An ane linear transformation T : V V is a map of the form
TA,a (v) = Av + a
for some A GL(V ) and a V . The group of all ane linear transformations V V
is the ane general linear group denoted AGL(V ), or AGLn (F ) if V = F n . The
right-to-left composition law on AGL(V ) is
TA,a TB,b = TAB, Ab+a .
It follows that AGLn (F ) is isomorphic to the subgroup

A aT
n
: A GLn (F ), a F
< GLn+1(F )
0 1
Appendix A2: GROUPS
209
aT
is an isomorphism. (Here we think of F n as column vectors
and the map TA,a A
0 1
of length n.) We can go one step further and dene AL(V ) as the group of ane
semilinear transformations
v Av + a
where A GL(V ), Aut F and a V . Note that
|AGLn (Fq )| = q n |GLn (Fq )|;
|ALn (Fq )| = rq n |GLn (Fq )|
with notation as above.

One often abbreviates GLn (q) = GLn (Fq ) and similarly for the other matrix groups
over Fq .
A2.6 Direct Products
Let K and G to be two groups. We assume for now that both K and G are multiplicative.
The (external) direct product of K and G is the group
H = K G = {(k, g) : k K, g G}
with componentwise multiplication
(k, g)(k , g ) = (kk , gg ).
Note that we may identify K with the subgroup {(k, 1) : k K}, and identify G with
the subgroup {(1, g) : g G}. With this identication, we observe that the subgroups K
and G are complementary, i.e. H = KG = {hk : h K, k G} (recall the identication
of k with (k, 1) and g with (1, g)) and K G = 1 (so that every g H can be uniquely
expressed as h = kg for k and g as above). Moreover these two subgroups are normal and
they commute with each other: kg = gk for all k K and g G.
Conversely, given a group H, in order to recognize H as the direct product of two
subgroups K, G H, we require that H = KG, K G = 1, and K commutes with G (in
particular both K and G are normal subgroups). We then write H = KG = K G, the
(internal) direct product of K and G.
This construction is entirely analogous to the construction of (internal and external)
direct sums of vector spaces.
A2.7 Semidirect Products
Here we generalize the notion of a direct product. Let K and G be groups, and suppose
that G acts on K. In this context (as in Section A2.1) G permutes the set of elements of
K; but here we require that the resulting permutations of K are actually automorphisms
of K. This means that each g G determines a map K K denoted by k k g such
that
(h1 h2 )k = hk1 hk2 ; hk1 k2 = (hk1 )k2
210
Appendix A2: GROUPS
for all k, k1 , k2 K; g, g1 , g2 G. (Note that the data we are given includes not only
groups K and G but also a choice of homomorphism G Aut(K).) Dene the (external)
semidirect product of K and G as
H = K G = {(k, g) : k K, g G}
where the product in H is dened by
(k1 , g1 )(k2 , g2 ) = (k1g2 k2 , g1 g2 )
for all ki K, gi G. If you have never done this before, you should check that
this actually does dene a group; most importantly, this product is associative. Again
{(k, 1) : k K} is a subgroup (actually a normal subgroup) which we identify with K;
and {(1, g) : g G} is a subgroup (although not in general normal) which we identify
with G. Note that K and G do not typically commute with each other; indeed
1
(1, g) (k, 1)(1, g) = (k g , 1)

so that the original action of G on K which was given, is realized as the action by conjugation in the group H. It is important to realize that the data required to construct the
group H includes not only the groups K and G, but also the choice of action of G on K.
In particular if one chooses the trivial action, one obtains simply a direct product as a
special case.
Reversing our viewpoint, suppose we are given a group H and two subgroups K, G H
such that K is normal and every element h H is uniquely expressible as h = kg where
k K, g G (i.e. H = KG with K G = 1). Then H is the (internal) semidirect product
of K and G.
As an example, the group of ane linear transformations on V is
AGL(V ) = V GL(V )
where V is identied as the normal subgroup consisting of translations v v + a, and
GL(V ) is identied as the subgroup consisting of linear transformations v Av. Similarly,
AL(V ) = V L(V ).
Also the group of semilinear transformations on V is
L = GL(V ) Aut F.
As a nal example, consider a cyclic group K = {1, x, x2 , . . . , xn1 } of order n, and let
G = {1, y} be a group of order 2. Then any semidirect product of K by G is either a direct
1
1
product (in which x commutes with y) or a dihedral group (in which xy = y xy = x ).
Appendix A3: Algebras and Representations

An algebra over a eld F is a vector space over F which also a ring with identity (and
certain assumptions of compatibility between the ring and vector space structures are
required to hold, including distributivity of multiplication over addition). Although an
identity element 1 A is not assumed in every case, we will only consider algebras with
identity. Formally, we begin with a vector space A over a eld F . We assume that A also
has a multiplicative structure: xy A whenever x, yA and that A has two distinguished
elements 0, 1 A such that
x + 0 = 0 + x = 1x = x1 = x
for all x A. If no confusion arises, we write simply 0, 1 instead of 0, 1 (although these
elements are distinct from the scalars 0, 1 F ). Moreover we assume that the following
statements hold for all r, s F and x, y A:
(rx)y = r(xy) = x(ry);
(rs)x = r(sx).
In this case we call A an algebra over F . We say A is commutative or non-commutative
according as its multiplication is commutative or not (regardless of the addition in A, which
is always commutative).
A3.1 Example: Matrix Algebras. The set F nn consisting of all n n matrices over
a eld F is an algebra over F . This algebra is not commutative when n 2. We usually
denote 0 (the zero matrix) and 1 (the identity matrix) simply by 0 and I.
A3.2 Example: Polynomial Algebras. The polynomial ring R = F [X1 , X2 , . . . , Xn ]
consisting of all polynomials in X1 , X2 , . . . , Xn with coecients in F , is an algebra over
F . Here 0 and 1 are just the constant polynomials 0 and 1. It is a graded ring, i.e.
R=
Rk ,
Rk R Rk+ .
k0
Here Rk is the subspace of all k-homogeneous polynomials. By Lemma 21.6 its dimension

is n1+k
.
k
A3.3 Example: Group Algebras. Let G be a nite multiplicative group, and let F
be a eld. The group algebra of G over F is the set F G consisting of all formal linear
combinations of elements of G with coecients in F . Typical elements a, b F G have the
form

ag g;
b=
bg g
a=
gG
gG
212
Appendix A3: ALGEBRAS AND REPRESENTATIONS
where ag , bg F ; addition and multiplication in F G are then dened by

a+b=
(ag + bg )g;
ab =
gG
ag bh gh =

ag bg1 x x.
xG gG
g,hG
Note that the group algebra F G is commutative i the group G is itself abelian.
For example consider G = S3 = {(1), (12), (13), (23), (123), (132)}. In this case
F S3 = {a(1) + b(12) + c(13) + d(23) + e(123) + f(132) : a, b, c, d, e, f F }.
An example of multiplication in this algebra is given by

(12) 2(13) (1) + (12) (23) + (123) = (1) + (12) (13) 2(23) + 2(123) 3(132).
Here the zero element 0 F G is the linear combination of group elements in which all
coecients are zero; it is usually denoted simply 0. Also the algebra identity 1 F G is
just the group identity, usually denoted simply by 1, or in this case (1).
A skewfield (or division ring1) is a ring with identity, in which every nonzero
element is a unit (i.e. is invertible). A field is a commutative skeweld. A division
algebra over a eld F is an algebra over F which is a skeweld (i.e. division ring).
A3.4 Theorem (Wedderburn). Every nite skeweld is a eld.
A3.5 Example: The Real Quaternions. The algebra of real quaternions is the 4dimensional algebra over R given by
H = {a + bi + cj + dk : a, b, c, d R}
where
i2 = j 2 = k 2 = ijk = 1;
ij = k;
ji = k;
jk = i;
kj = i;
ki = j;
ik = j.
This is the easiest example of a non-commutative skeweld.

Let A be an algebra over F . A (left) A-module is a vector space M over F together
with a left action of A on M, written as left-multiplication (ax M whenever a A,
x M) such that
a(x + y) = ax + ay;
1
Caution: The term division ring is sometimes used in a dierent sense by nite geometers; see [31].
213
(a + b)x = ax + bx;
a(bx) = (ab)x;
1x = x
for all a, b A; x, y M. Given an A-module M, an F -subspace N M is an Asubmodule if ax N whenever a A, x N .
A3.6 Example: The Natural Module for F nn . Let F nn be the algebra of n n
matrices over a eld F . Then F n (the n-dimensional vector space consisting of column
vectors of length n over F ) is a module for F nn , in which matrices act on the left by the
usual matrix multiplication.
A3.7 Example: The Regular Module. Let A be an algebra over F . Then A may
be regarded as a module over itself, where the left action of A on itself is the usual
multiplication in A. This module is called the (left) regular A-module. Note that a
submodule of the left regular module, is the same thing as a left ideal of A.
Let A be an algebra over F , and let M be a nonzero A-module. We say M is simple
(or minimal or irreducible) if its only A-submodules are (0) and M itself. We say M is
semisimple if it satises the equivalent conditions (i) and (ii) of the following.
A3.8 Proposition. Let A be an algebra over F , and let M be a nite-dimensional
A-module. Then the following two conditions are equivalent.
(i) For every submodule U M there exists a complementary ideal U M with
M = U U .
(ii) There exist simple submodules U1 , U2 , . . . , Ur M such that M = U1 U2
Ur .
A3.9 Example: Modules for F nn . It is easy to see that the natural module F n for
F nn is simple. The regular module is semisimple:
F nn
= U1 U2 Un
where Ui F nn is the submodule (i.e. left ideal) consisting of all n n matrices having
zeroes outside of the i-th column.
Let G be a nite multiplicative group and F a eld. A (linear) representation of
G over F is a homomorphism : G GLn (F ). Given such a representation, the vector
space F n has the structure of a module over the group algebra F G, where a typical element
g G acts on v F n as v (g)v, and more generally

a=
ag g F G
gG
214
acts on F n as
v
ag (g)v .
gG
Conversely, an action of the group algebra F G on F n gives rise to a representation G

GLn (F ), so the two notions of a representation of G, and a module for F G, are really
equivalent. A useful sucient condition for semisimplicity of F G-modules is the following.
A3.10 Theorem (Maschke). Let G be a nite group, and let F be a eld whose
characteristic does not divide |G|. (Thus either char F = 0 or char F = p |G|.) Then
every F G-module is semisimple.
Proof. Let M be an F G-module and let U M be a submodule. Choose any complementary subspace W M, so that M = U W as F -vector spaces. We cannot assume that
W is an F G-submodule; if it were, we would be done. Let P : M U be the projection
onto U along W ; thus P : M U is the unique F -linear transformation satisfying
P x = x i x U;
P x = 0 i x W ;
P2 = P.
Dene the F -linear transformation
P
: M U;
v
1 1
h P hv.
|G|
hG
Note that the scalar |G| is invertible in F , by the hypothesis on the characteristic of F .
We rst show that
(A3.11)
P
(av) = a(P
v) for all a F G, v V .
It suces to verify (A3.11) in the case a = g G; but in this case

1 1
h P hgv
P
(gv) =
|G|
hG
1 1
gk P kv
=
|G|
kG
1 1
k P kv
g
=
|G|
where k = hg
kG
= g(P
v)
and so (A3.11) follows in the general case from linearity. Next we check that
(A3.12)
215
P
2 = P
.
Indeed, for all v M we have

P
2 v =
1 1
1
h P hk P kv
2
|G|
h,kG
1 1 1
h hk P kv
=
|G|2
h,kG
since P kv U implies that hk P kv U also, and so the latter element is xed by P .

Thus
1 1
P
2 v =
k P kv
|G|2
h,kG
1
1
|G|k P kv
2
|G|
kG

1
1
k P kv
=
|G|
=
kG
= P
v
which proves (A3.11). Thus P
is a projection operator. We show that it is in fact a
projection onto U:
(A3.12)
P
v = v i v U.
First observe that if v U then

1 1
h P hv
P
v =
|G|
hG
1 1
h hv
=
|G|
hG
since hv U is xed by P ; thus P

v = v. Conversely, suppose that P
v = v; then since
P
v U it follows that v U. Thus (A3.12) holds. Finally
(A3.12)
M = U U where U = ker P is an F G-submodule.
If v U = ker P then for all a F G we have

P (av) = a(P v) = 0
by (A3.11), so av U as well. Thus U M is an F G-submodule, and the remaining
assertions in (A3.12) follow from elementary properties of projections.
216
A3.13 Corollary. Let G be a nite group, and let F be a eld of order not dividing
|G|. (Thus either char F = 0 or char F = p |G|.) Then for every left ideal I F G
there is a complementary left ideal I F G satisfying F G = I I .
Apply Theorem A3.10 to the regular module F G.
Proof.
Let A be an algebra over F . Elements of the centre Z(A) are called central elements
of A. An idempotent in A is an element e A such that e2 = e.
Now suppose that A is commutative and that the regular module of A is semisimple.
We exhibit a one-to-one correspondence1 between the idempotents of A, and the ideals
of A. First suppose I A is an ideal. By assumption there exists a complementary ideal
I A such that A = I I . (Note that all ideals are two-sided since A is commutative.)
In particular
(A3.14)
1 = e + e
for unique elements e I, e I . From (A3.14) we obtain e = e2 + e e where e e

I I = (0), so e2 = e and similarly (e )2 = e . Right-multiplying (A3.14) by an arbitrary
x A gives
(A3.15)
x = ex + e x
where ex I and e x I ; thus the map x ex is the projection of A onto I along I ;

similarly x e x is the projection of A onto I along I. Thus I = Ae and I = Ae .
Evidently e is the unique idempotent generator of I so the correspondence I e is
bijective. The essential features of this correspondence are summarized as follows.
A3.16 Theorem. Let A be a commutative algebra over F whose regular module is
semisimple. Then there is a one-to-one correspondence between idempotent elements
e A and ideals I A, such that I = Ae. Moreover if I1 = Ae1 and I2 = Ae2 where
each ei is idempotent, then
(i) I1 I2 = I1 I2 = Ae1 e2 and
(ii) I1 + I2 = A(e1 +e2 e1 e2 )
where e1 e2 and e1 +e2 e1 e2 are the unique idempotent generators of the ideals listed
in (i) and (ii) respectively.
1
A similar conclusion holds under the weaker hypothesis that A is semisimple, meaning that its Jacobson
radical is (0), but A is not necessarily commutative. In this case there is a one-to-one correspondence between
the (two-sided) ideals of A and the central idempotents of A; see e.g. [32]. We will not require this stronger
version.
217
Proof. We have already established the indicated one-to-one correspondence. Consider

ideals Ii = Aei for i = 1, 2 where e1 and e2 are idempotent. Then
(e1 e2 )2 = e21 e22 = e1 e2 ;
(e1 +e2 e1 e2 )2 = e21 +e22 +e21 e22 +2e1 e2 2e21 e2 2e1 e22 = e1 +e2 e1 e2
so these two elements are idempotent. Clearly
I1 I2 I1 I1 = Ae1 e2 .
But if x I1 I2 then x = e1 x = e1 e2 x Ae1 e2 , so equality holds in (i). For (ii) it is
clear that
I1 + I2 A(e1 +e2 e1 e2 );
to check equality, note that for all x A we have
(xe1 )(e1 +e2 e1 e2 ) = x(e21 +e1 e2 e21 e2 ) = xe1
so that I1 A(e1 +e2 e1 e2 ). A similar argument shows that I2 A(e1 +e2 e1 e2 ), so in
fact
I1 + I2 A(e1 +e2 e1 e2 )
also, and so the equality (ii) holds.
A3.17 Example: Diagonal Matrices. Let A be the ring of n n diagonal matrices
over F . [Of course A
= F n = F F F (n summands).] Then A has 2n ideals,
corresponding to the 2n idempotent elements, these being the diagonal matrices with
diagonal entries {0, 1}. To be even more concrete, lets take n = 4 and consider the
ideals
a 0 0 0
1 0 0 0
0 b 0 0
0 1 0 0
where e1 =
I1 = Ae1 =
: a, b F
;
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 a 0 0
0 1 0 0
where e2 =
I2 = Ae2 =
: a, b F
.
0 0 1 0
0 0 b 0
0 0 0 0
0 0 0 0
Then
0 0 0 0
0 0 0 0
0 a 0 0
0 1 0 0
I1 I2 = I1 I2 = Ae1 e2 =
:
a
F
where e1 e2 =
0
0
0
0
0 0 0 0
0 0 0 0
0 0 0 0
and
a 0 0 0
0 b 0 0
I1 +I2 = A(e1 +e2 e1 e2 ) =
: a, b, c F
0 0 c 0
1 0 0 0
0 0 0 0
0 1 0 0
where e1 +e2 e1 e2 =
.
0 0 1 0
0 0 0 0
Appendix A4: Exterior Algebra

Let V be an n-dimensional vector space over F with basis e1 , e2 , . . . , en . The k-th exterior

k
V of dimension nk with basis given by the symbols
power of V is the vector space
ei1 ei2 eik ,
1 i1 < i2 < < ik n.
(The symbol is read as wedge.) We extend this notation by dening v1 v2 vk

k
V for all v1 , v2 , . . . , vk V subject to the rules:
(A4.1) The vector v1 v2 vk is linear in each argument if the other arguments
are xed, i.e.
v1 v2 (avi +bvi ) vk
= a(v1 v2 vi vk ) + b(v1 v2 vi vk ) .
(A4.2) The vector v1 v2 vk vanishes if any two of its arguments coincide. Thus
if vi = vj for some i = j then v1 v2 vk = 0.
From (A4.1) and (A4.2) we obtain
(A4.3) The vector v1 v2 vk is replaced by its negative, if any two of its arguments
are interchanged. Thus for all i = j we have
v1 v2 vj vi vk = (v1 v2 vi vj vk ).
The identity (A4.3) follows from (A4.1) and (A4.2) by simply expanding
0 = v1 v2 (vi +vj ) (vi +vj ) vk
(v1 v2 vi vi vk ) (v1 v2 vj vj vk ).
Conversely we may derive (A4.2) from (A4.1) and (A4.3), assuming char F = 2. (If
char F = 2 then nothing can be deduced from (A4.3) since the assertion (A4.3) says
merely that a vector equals itself.) The properties above suce to evaluate an arbitrary
wedge product v1 v2 vk ; however if one requires a more explicit denition then

vij ej where vij F ; then
the following will serve. First expand vi =
1jn
(A4.4)
v1 v2 vk

v1j1 v1j2

v2j1 v2j2
.
=
..
.
.
1j1 <j2 <<jk n .

vkj1 vkj2

v1jk

v2jk
. ej1 ej2 ejk .
..
. ..

vkjk
219
Appendix A4: EXTERIOR ALGEBRA
Consider the matrix M whose rows are the

coordinates of v1 , v2 , . . . , vk with respect to
e1 , e2 , . . . , en . This kn matrix M has nk submatrices of size kk, whose determinants
are simply the coecients in (A4.4). Recall that M is singular i all of its kk submatrices
are singular; thus
(A4.5)
Vectors v1 , v2 , . . . , vk V are linearly dependent i v1 v2 vk = 0.
0
k

1
V = V ; also
V = F . If v
V and w V then v w
We naturally identify
k+
V is well-dened; here one simply identies
(v1 v2 vk ) (w1 w2 w ) = v1 v2 vk w1 w2 w .
When k = 0 or = 0, the corresponding factor v or w is an empty wedge product, which
k
V = 0
by denition is simply the scalar 1. Also it follows from the denitions that
whenever k > n.
The exterior algebra of V is the algebra
V =
k
k0
V =F V
2
3
n
with respect to the wedge product . This is a graded algebra of dimension

n
= 2n .
k
0kn
k
V
By graded we mean simply the fact that multiplying (i.e. wedging) elements of
k+

V , as previously mentioned.
with elements of V , gives elements of
2
The inexperienced reader is encouraged to think of V as F n , and to identify V with

the n2 -dimensional space Skewn (F ) of all skew-symmetric n n matrices over F . An
actual isomorphism is given by
2
V Skewn (F ),
v w v TwwTv.
Note that an n n matrix of the form v Tw has rank 1 (unless v = w = 0, in which case
v Tw = 0 has rank 0) and that the skew-symmetric matrix v TwwTv has rank either 0 or
2, according as v and w are or are not linearly dependent. In general, a skew-symmetric
n n matrix is not necessarily of the form v TwwTv; elements of Skewn (F ) can have any
rank in 0, 2, 4, . . . , 2k where n {2k, 2k + 1}. However, Skewn (F ) is spanned by its rank 2
elements.
The preceding remarks are intended to help the reader remember that not every vector
k
k
V has the form pure form v1 v2 vk where v1 , v2 , . . . , vk V ; rather,
V
in
is spanned by elements of this pure form.
Appendix A5: Coding Theory

Coding theory, or the theory of error-correcting codes, concerns how information may best
be represented for the purpose of transmission over a noisy channel, or stored in imperfect media, in such a way that a limited number of errors introduced during transmission/reception, or storage/retrieval, may be corrected. To formalize these goals inevitably
leads us to notions of nite geometry. Coding theorists use many of the standard constructions of nite geometry in the construction of good codes. In this study, where nite
geometry is our primary interest, we shall have more use for the reverse process in which
coding theory is used to study nite geometry. For a more comprehensive introduction to
coding theory, see e.g. [65].
An alphabet is simply a nite set A of symbols, possibly the Roman alphabet
{a, b, c, . . . ,z} or the set of 256 ASCII characters; or more typically for our use, the elements of a nite eld, especially the binary alphabet F2 = {0, 1} whose elements we
call bits. A code of length n over an alphabet A is a subset C An . We refer to C as a
q-ary code where q = |A|. Elements of An are called words (or bitstrings if A = {0, 1}),
and elements of the subset C are called codewords. The (Hamming) distance between
two words w, w An , denoted (w, w ) {0, 1, 2, . . . , n}, is the number of coordinates in
which they dier. Note that is a metric on An . The minimum distance of C is
d = min (v, w).
w = v
in C
Denote the closed ball of radius r centred at a vector w Fnq by

Br (w) = {v Fnq : (v, w) r}.
A code C is e-error correcting if the balls Be (w) centred at the codewords w C are
mutually disjoint. In this case a received word v Be (w) is uniquely decoded as w C,
at least in principle. (Actually nding the closest codeword w C to a given word v may
be dicult in practice.)
A5.1 Proposition. A code C is e-error correcting i its minimum distance is at least
2e + 1.
Proof. If C is not e-error correcting then there exist w = w in C such that Be (w)Be (w )
contains some vector v Fnq ; but then (w, w ) (w, v) + (v, w ) e + e = 2e. Clearly
the converse of this argument also holds.
We say that C is e-error detecting if Be (w) Be1 (w ) = for all w = w in C.
This says that if at most e symbols in a transmitted word are altered during transmission,
the recipient can be sure that such an error occurred; but cannot in general correct it; see
221
Appendix A5: CODING THEORY
Example A5.6 below. Clearly every code with minimum distance d is

and d1
-error correcting.
2
d
2
-error detecting,
Note that

n
n
2
(q 1) + +
(q 1)r .
|Br (v)| = 1 + n(q 1) +
2
r
An elementary counting argument using this yields
A5.2 Theorem (Sphere-Packing Bound; Hamming Bound). If C is an e-error
correcting q-ary code of length n, then
|C|
1 + n(q1) +
n
2
qn

.
(q1)2 + + nr (q1)r
A perfect code is one for which equality holds in the Sphere-Packing Bound; in this case
the balls Be (w) centred at codewords w C partition the set An of all words.
Usually A = Fq is a nite eld, and subspaces C Fnq are called linear codes. A
linear code of length n and dimension k is called an [n, k]-code. The (Hamming) weight
of a word (i.e. vector) v Fnq is the number wt(v) {0, 1, 2, . . . , n} of nonzero coordinates
of v. Note that the Hamming distance between two words v, w Fnq is given by
(v, w) = wt(w v).
It follows that the minimum distance of a linear code C coincides with the minimum
weight:
d = min wt(w).
w = 0
in C
A linear code with minimum weight d is also called an [n, k, d]-code.

Let G be a k n matrix over Fq having linearly independent rows (so in particular
k n). The row space of G is an [n, k]-code
C = {xG : x Fkq }.
We call G a generator matrix for the code C. In this case the injective map Fkq Fnq ,
x xG gives an encoding of all q-ary words of length k, as words of length n. We say that
C has information rate equal to nk . This is the fraction of the transmitted symbols which
contain useful information; the remaining symbols may serve to provide error correction.
222
An [n, k]-code may be specied as the row space of a k n matrix G as above, or as

the right null space of an (nk) n matrix H having linearly independent rows:
}.
C = {y Fnq : Hy T = 0 Fnk
q
Such a matrix H is called a parity check matrix for C. Given a received word y Fnq , we
call the vector Hy T Fnk
the (error) syndrome of y. The syndrome is useful in error
q
detection and correction: certainly a nonzero syndrome indicates that an error occurred
during transmission. To see the utility of syndromes in error correction, see Example A5.8
below.
The fundamental goal of coding theory is to nd codes with the following desirable
properties:
High information rate;
High error-correcting capability (i.e. large minimum distance);
Ecient encoding and decoding algorithms should be available.
These goals tend to be conicting in nature; for example one can increase the errorcorrecting capability, at the cost of reducing the information rate and thereby increasing
the cost of transmission, as Theorem A5.2 shows. Another bound which shows this tradeo between minimum distance and information rate is the following.
A5.3 Theorem (Singleton Bound). If C is a linear [n, k, d]-code over Fq then
k n d + 1.
Proof. Let H be an (nk) n parity check matrix for C. Since H has rank n k, it has a
set of nk linearly independent columns; we may suppose that the rst nk columns of H
are linearly independent, otherwise permute the columns of H appropriately. (Permuting
the coordinates of C yields once again an [n, k, d]-code.) Now we may assume H is in
reduced row-echelon form; otherwise replace H by its reduced row-echelon form, and this
will not change its right null space. Thus

H = Ink X
(nk)k
where I is the (nk) (nk) identity matrix, and X Fq

the matrix

G = X T Ik
. It is easy to check that
is a generator matrix for C. Its rows have weight at most n k + 1, so that d n k +1.
Codes for which the Singleton Bound is attained, are called MDS codes. (This is
an acronym for Maximum Distance Separable, which is such an unfortunate term that
223
everyone just calls them MDS codes.) Although both bounds A5.2 and A5.3 describe the
optimal possible tradeo between information rate and minimum distance, neither bound
implies the other. Thus there exist MDS codes which are not perfect; and there exist
perfect codes which are not MDS.
The weight enumerator of a code C of length n is the polynomial
AC (z) =

vC
xnwt(v) y wt(v) =
Ad xnd y d
0dn
where Ad is the number of codewords of weight d. The essential information conveyed by

the weight enumerator is the list of integer values A0 , A1 , . . . , An , known as the weight
distribution of C. The reason for representing this list of values as a single polynomial
is motivated by Theorem A5.4 below. In the case C is a linear code, we dene the dual
code by
C = {w Fnq : w v = 0 for all v C}
where w v = wv T. Thus the dual of an [n, k]-code is an [n, nk]-code. If C has generator
matrix G and parity check matrix H, then C has generator matrix H and parity check
matrix G.
The MacWilliams relations (Theorem A5.4) show that the weight distribution of either
of these two codes (C or C ) determines the weight distribution of the other.
A5.4 Theorem (MacWilliams). Let C be a linear [n, k]-code over Fq . Then the
weight enumerator of the dual code is given by
AC (x, y) = q k AC (x + (q1)y, x y).
Before proving Theorem A5.4 we consider an example.
A5.5 Example. Consider the binary [5, 2]-code C = {00000, 11100, 10011, 01111}, which
has weight enumerator
AC (x, y) = x5 + 2x2 y 3 + xy 4 .
Its dual is the binary [5, 3]-code
C = {00000, 01100, 00011, 01111, 11010, 11001, 10110, 10101}
whose weight enumerator is
AC (z) = x5 + 2x3 y 2 + 4x2 y 3 + xy 4 .
224
The MacWilliams relations are expressed as the identity

AC (x, y) = 14 AC (x + y, x y),
i.e.
x5 + 2x3 y 2 + 4x2 y 3 + xy 4 =
1
4
(x+y)5 + 2(x+y)2 (xy)3 + (x+y)(xy)4
which may be veried directly. In passing we note that C C = {00000, 01111}. This is
typical in that while the subspaces C, C Fnq have complementary dimension, they are
not in general complementary subspaces (unlike the situation for inner product spaces).
Proof of Theorem A5.4. A character of the additive group of Fq is a map : Fq C
such that (a + b) = (a)(b) for all a, b Fq . (Here C is the multiplicative group of
nonzero complex numbers.) In particular the map (a) = 1 for all a Fq is the principal
character; every other character of Fq is nonprincipal. Let be a nonprincipal character
of Fq and observe that

(a) = 0.
aFq
For all u Fnq dene
g(u) =
(u v)xnwt(v) y wt(v) .
vFn
q
Then

g(u) =
nwt(v) wt(v)
(u v)x
uC vFn
q
uC

(u v) xnwt(v) y wt(v) .
vFn
uC
q
The innermost sum is q k for v C and vanishes otherwise; therefore

g(u) = q k
xnwt(v) y wt(v) = q k AC (x, y).
uC
vC
For v = (v1 , . . . , vn ) Fnq we have

wt(v) = wt(v1 ) + + wt(vn ) where wt(vi ) =
0, if vi = 0;
1, if vi = 1
and so
g(u)=
(u1 v1 + +un vn )xnwt(v1 )wt(v2)wt(vn ) y wt(v1 )++wt(vn)
v1 ,...,vn Fq
(u1 v1)x1wt(v1 ) y wt(v1 ) (u2 v2)x1wt(v2 ) y wt(v2 ) (un vn)x1wt(vn ) y wt(vn )
v1 ,...,vn Fq
1in vi Fq
(ui vi )x1wt(vi ) y wt(vi ) .
225
The innermost sum equals x + (q1)y if ui = 0, and x y if ui = 0; thus

nwt(u)
(x y)wt(u) .
g(u) = x + (q1)y
Summing over u C gives the required result.
Consider the problem of transmitting a single hexadecimal digit, or equivalently, a

bitstring of length 4. Such a message may be transmitted as a sequence of four bits, but
this scheme will provide no error-correcting capability: if the message 0100 is transmitted,
and the third bit is altered due to static so that the word 0110 is received, the error cannot
be detected by the receiver. The following examples show how we can do better.
A5.6 Example: Parity Check Code. A slightly better scheme would be to add a single
parity check bit to every word; thus message words and the corresponding codewords are
as listed in the following table. This code detects single errors but corrects no errors.
Message
Word
Codeword
Message
Word
Codeword
0000
0001
0010
0011
0111
0000 0
0001 1
0010 1
0011 0
0111 1
1000
1001
1010
1011
1111
1000 1
1001 0
1010 0
1011 1
1111 0
This code has an information rate of 45 = 80%, meaning that 4 out of every 5 bits transmitted carry useful information; 1 out of every 5 bits transmitted supply redundancy useful
in checking the validity of the information transmitted.
This code has generator matrix and parity check matrix given by
1
0
G=
0
0
0
1
0
0
0
0
1
0
0
0
0
1
1
1
,
1
1

H= 1 1 1 1 1 .
A5.7 Example: 3-Repetition Code. The idea behind this code is to simply send
every bit three times, and to require the receiver to use a simple majority rules approach
to decoding. This code is not only 1-error detecting, but in fact 1-error correcting. This
226
advantage comes at the cost: 2 out of every 3 bits transmitted are redundant bits, and the
information rate is only 13 33%.
Message
Word
Codeword
Message
Word
Codeword
0000
0001
0010
0011
0111
000 000 000 000

000 000 000 111
000 000 111 000
000 000 111 111
000 111 111 111
1000
1001
1010
1011
1111
111 000 000 000

111 000 000 111
111 000 111 000
111 000 111 111
111 111 111 111
111
0 0 0
G=
000
000
000
111
000
000
000
000
111
000
000
0 0 0
,
000
111
101
0 1 1
0 0 0
0 0 0
H=
0 0 0
0 0 0
000
000
000
000
101
011
000
000
000
000
000
000
000
000
101
011
000
000
000
0 0 0
0 0 0
0 0 0
.
0 0 0
0 0 0
101
011
A5.8 Example: The [7, 4, 3] binary Hamming Code. The following scheme is superior to the previous example in that it also allows single bit errors to be corrected, yet
it has a substantially higher information rate of 47 57%. It is in fact a perfect code,
although not an MDS code.
Message
Word
Codeword
Message
Word
Codeword
0000
0001
0010
0011
0100
0101
0110
0111
0000 000
0001 111
0010 110
0011 001
0100 101
0101 010
0110 011
0111 100
1000
1001
1010
1011
1100
1101
1110
1111
1000 011
1001 100
1010 101
1011 010
1100 110
1101 001
1110 000
1111 111
1
0
G=
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
1
1
1
1
0
1
1
1
1
,
0
1
H= 0
1
0
1
0
0 1
1 0
1 0
1
0
1
1
1
0
1
1.
1
227
Note that the columns of H consist of the numbers one through seven, written in binary.
This allows for very easy decoding. Note that a message string x F42 is encoded as a
codeword y = xG F72 . Suppose that this word is transmitted and a word y F72 is
received. The receiver rst computes the syndrome Hy T F32 .
If the syndrome is zero, then y is a legitimate codeword. We assume no bit errors
occurred during transmission (otherwise at least three bit errors occurred, which
we deem unlikely). We recover x F42 as the rst four bits of y .
If the syndrome is nonzero, then the syndrome is the binary representation of
some i {1, 2, . . . , 7}. First switch the i-th bit of y (i.e. add 1 mod 2 in the i-th
coordinate) to obtain the unique closest codeword to y . We assume the resulting
codeword is actually y; so as before, we recover x F42 as the rst four bits of y .
For example consider the message word x = 1101, which is encoded as y = xG = 1101001.
Suppose this codeword is transmitted but that due to interference, the string y = 1111001
is received. We compute the syndrome

0
Hy T = 1 .
1
This is the binary representation of 3, so switch the third bit of y to recover y = 1101001
and thereby we nd x = 1101.
Appendix A6: Invariant Theory

Let R = C[X1 , X2 , . . . , Xn ], the algebra of all polynomials in X1 , X2 , . . . , Xn with com
plex coecients. For every invertible matrix A = aij i,j GLn (C) and polynomial
f(X1 , . . . , Xn ) R we dene

(Af)(X1 , . . . , Xn ) = f((X1 , . . . , Xn )A) = f( i ai1 Xi , i ai2 Xi , . . . , i ain Xi ).
It is easy to see that A(f + g) = Af + Ag and A(fg) = (Af)(Ag). The map f Af is
therefore an automorphism of the algebra R. Moreover (AB)(f) = A(Bf) for all A, B
GLn (C) since
A(Bf)(X1 , . . . , Xn ) = (Bf)((X1 , . . . , Xn )A)
= f((X1 , . . . , Xn )AB)
= ((AB)f)(X1 , . . . , Xn ).
This means that we have a (left) action of GLn (C) on R.
Let : G GLn (C) be a complex linear representation of a group G. Recall that
this simply means that is a group homomorphism from G to GLn (C). Then G also acts
on R, where the action of g G on a polynomial f is given by
f (g)f.
The set of invariants of G (which of course depends on the choice of representation ) is
the set of all polynomials xed by every element of G:
RG = {f R : (g)f = f for all g G}.
(Unfortunately this conicts with the use of superscripts for G-orbits; but the notation is
by now too standard to adjust.) Note that RG is a subring of G; it is called the ring of
invariants of G. The main question is to determine RG . A rst observation is that every
A GLn (C) maps k-homogeneous polynomials to k-homogeneous polynomials, for each
k 0. Writing Rk R for the subspace of all k-homogeneous polynomials (with the zero
polynomial included), then

R=
Rk
k0
in which every summand is preserved by the action of G. By Section A3.2 we have

(A6.1)
dim Rk =
n+k1
k
We obtain
RG =

k0
RG
k
Appendix A6: INVARIANT THEORY
229
where RG
k is the space of all k-homogeneous polynomials xed by every element of G. Thus
it suces to determine each of the summands RG
k.
A6.2 Example: The Orthogonal Group. Consider the real orthogonal group
G = {A GLn (R) : AAT = I}
in its natural action on Cn , i.e. we have simply the inclusion mapping : G GLn (C),
A A. Since orthogonal transformations preserve the Euclidean norm, it is clear that G
xes the polynomial
(X1 , X2 , . . . , Xn ) = X12 + X22 + + Xn2 .
It follows that the ring of invariants RG contains the subring generated by , i.e.
RG C[] = {f(X12 + X22 + + Xn2 ) : f(T ) C[T ]}.
It turns out that equality holds: RG = C[].
A6.3 Example: The Symmetric Group. Consider the symmetric group G = Sn in
its natural permutation representation of degree n. Here every Sn is represented by
the corresponding permutation matrix, thus:

() = i ,j i,j .
It is clear that the elementary symmetric polynomials
e1 (X1 , X2 , . . . , Xn ) = X1 + X2 + + Xn ,

e2 (X1 , X2 , . . . , Xn ) = 1i<jn Xi Xj = X1 X2 + X1 X3 + + Xn1 Xn ,
..
.
ek (X1 , X2 , . . . , Xn ) =
1i1 <i2 <<ik n Xi1 Xi2
Xik ,
..
.
en (X1 , X2 , . . . , Xn ) = X1 X2 Xn
are invariant under coordinate permutations. Therefore the ring of invariants contains the
subring generated by e1 , e2 , . . . , en , thus:
RG C[e1 , e2 , . . . , en ].
Once again, although it is not obvious, equality holds: RG = C[e1 , e2 , . . . , en ]. For example,
the polynomials
pk (X1 , X2 , . . . , Xn ) = X1k + X2k + + Xnk
230
are clearly invariants; according to our assertion, these may be expressed as polynomials
in e1 , e2 , . . . , en . Indeed, we nd that
p1 = e1 ;
p2 = e21 2e2 ;
p3 = e31 3e1 e2 + 3e3 ;
etc.
These examples are suggestive of the following general result: whenever G is nite, the
ring of invariants is nitely generated. (The same conclusion holds more generally whenever
the representation is unitary, i.e. with respect to some basis of Cn the matrices (g) are
unitary; in particular this happens whenever the subgroup (G) GLn (G) is a compact
Lie group. The orthogonal group of Example A6.2 satises this condition.) The conclusion
here is that there exists a nite list of polynomials 1 , 2 , . . . , m C[X1 , . . . , Xn ] such that
RG = C[1, 2 , . . . , m ].
A list of such polynomials k (X1 , . . . , Xn ) for k = 1, 2, . . . , m, with m as small as possible,
form what is called a fundamental set of invariants. Given G and , algorithms are
known (using Gr
obner basis methods) for explicitly determining such a fundamental set
of invariants. Such an algorithm is presented, for example, in [20], [60]. Although it is not
our present purpose to describe such an algorithm here, nevertheless we outline one of the
key ingredients: a criterion for testing whether a known list of invariants 1 , . . . , m is a
fundamental set of invariants. A graded subring of R is a subring S R satisfying
S=
Sk
k0
where Sk = S Rk ; for example that the ring of invariants of G is a graded subring. We

dene the Hilbert series of a graded subring S R by
HS (t) =

dim Sk tk .
k0
A6.4 Theorem (Molien). Let : G GLn (C) be a representation of a nite

group G. Then the Hilbert series of the ring of invariants is given by
HRG (t) =
1
1
.
|G|
det(I t(g))
gG
231
In order for Moliens Theorem to be useful, we must have some means of computing Hilbert
series of some commonly arising subrings of R. Recall that polynomials g1 , g2 , . . . , gm
C[X1 , X2 , . . . , Xn ] are algebraically dependent if there exists a nonzero polynomial
f(T1 , T2 , . . . , Tm ) C[T1 , T2 , . . . , Tm ] such that f(g1 , g2 , . . . , gm ) = 0. Otherwise, we say
that g1 , g2 , . . . , gm are algebraically independent.
A6.5 Theorem. Let 1 , 2 , . . . , m R be algebraically independent homogeneous
polynomials of degree d1 , d2 , . . . , dm respectively, and let S = C[1 , 2 , . . . , m ] R,
the subring generated by 1 , 2 , . . . , m . Then the Hilbert series of S is given by
HS (t) =
1
(1
td1 )(1
td2 ) (1
td m )
em
Proof. The k-homogeneous component Sk S is spanned by {1e1 2e2 m
: ei 0,

e1 e2
em
ei = k}. In fact this is a basis for Sk : the polynomials 1 2 m cannot be linearly
dependent, for otherwise the polynomials 1 , 2 , . . . , m would be algebraically dependent.
So dim Sk is the number of m-tuples (e1 , e2 , . . . , em ) of non-negative integers satisfying
e1 +e2 + +em = k; but this is also the coecient of tk in
1
(1td1 )(1td2 ) (1tdm )
= (1+td1 +t2d1 +t3d1 + )(1+td2 +t2d2 +t3d2 + ) (1+tdm +t2dm +t3dm + ).
A6.6 Example: The Trivial Group. Let G = 1 acting on R = C[X1 , . . . , Xn ], so that

the ring of invariants is the full polynomial ring: RG = R. By (A6.1) and the Binomial
Theorem we obtain

n + k 1
1
k
(dim Rk )t =
.
tk =
HRG (t) = HR (t) =
(1 t)n
k
k0
k0
This result agrees with the prediction of Moliens Theorem A6.4. Note that in this case
a fundamental set of invariants is given by X1 , X2 , . . . , Xn , each of degree 1, so that
Theorem A6.5 gives the same result.
A6.7 Example: Even Polynomials. Let n = 1 and consider the group G = {1, g} of

order two, acting on R = C[X] by sign changes; here (g) = 1 . Clearly the ring of
invariants is simply the subring of all even polynomials in X, i.e.
RG = C[X 2 ] = {f(X 2 ) : f(T ) C[T ]}.
232
k
G
Clearly RG
k is one-dimensional (spanned by X ) whenever k = 0, 2, 4, 6, . . .; and Rk = 0
whenever k is odd. Thus
HRG (t) = 1 + t2 + t4 + t6 + =
1
.
1 t2
Again since a fundamental set of invariants consists simply of (X) = X 2 of degree 2, the
same result is given by Theorem A6.5.
A6.8 Example: The Symmetric Group S3 . Consider the group G = S3 acting
naturally on R = C[X, Y, Z] by means of coordinate permutations. This is the special case
of Example A6.3 with n = 3, where we claim that a fundamental set of invariants consists
of the homogeneous polynomials
e1 (X, Y, Z) = X+Y +Z;
e2 (X, Y, Z) = XY +XZ+Y Z
and e3 (X, Y, Z) = XY Z.
We show how Moliens Theorem can be used to verify this result. The polynomials e1 , e2
and e3 are algebraically independent, which can be seen as follows. Suppose there exists a
nonzero polynomial
h(T1 , T2 , T3 ) = cT1i1 T2i2 T3i3 + C[T1 , T2 , T3 ]
such that h(e1 , e2 , e3 ) = 0. Here c = 0 and T1i1 T2i2 T3i3 is the lex-highest monomial appearing
in h; that is, for every monomial T1j1 T2j2 T3j3 appearing in h, either
i1 > j1 ; or
i1 = j1 and i2 > j2 ; or
i1 = j1 and i2 = j2 and i3 j3 .
Then the coecient of X i+j+k Y j+k Z k in h(e1 , e2 , e3 ) is c = 0; in particular, h(e1 , e2 , e3 )
cannot vanish. By Theorem A6.5, the subring S = C[e1 , e2 , e3 ] R has Hilbert series
HS (t) =
1
.
(1 t)(1 t2 )(1 t3 )
By Moliens Theorem, the Hilbert series of RG is

HRG (t) =
2
1
1
3
1
+
.
=
+
3
2
3
6 (1 t)
(1 t)(1 t ) 1 t
(1 t)(1 t2 )(1 t3 )
We conclude that S = RG is the full ring of invariants. To see this, note that S RG and so
G
the k-homogeneous components satisfy Sk RG
k for all k 0; however dim Sk = dim Rk
by comparing coecients of tk in the corresponding Hilbert series; therefore equality must
hold as claimed.
For further reading on the main themes of this course, the student is encouraged to
consult additional sources as listed below:
Algebra: [49], [32]
Incidence geometry: [11]
Classical groups and their geometry: [61] Projective geometry: [21], [12]
Coding Theory: [65], [14]
Projective planes: [31], [4]
Combinatorics: [13], [14]
Generalized quadrangles: [52]
Finite geometry: [25]
Generalized polygons: [63]
Bibliography
[1] R.W. Ahrens and G. Szekeres, On a combinatorial generalization of 27 lines associated
with a cubic surface, J. Austral. Math. Soc. 10 (1969), 485492.
[2] A. Balog and T.D. Wooley, Sums of two squares in short intervals, Canad. J. Math.
52 (2000), 673694.
[3] A. Barlotti, Unestensione del teorema di Segre-Kustaanheimo, Bull. Un. Mat. Ital./
10 (1955), 498506.
[4] A. Beutelspacher, Projective planes, pp.107136 in [11].
[5] A. Blokhuis, On the size of a blocking set in P G(2, p), Combinatorica 14 (1994),
111114.
[6] A. Blokhuis and G.E. Moorhouse, Some p-ranks related to orthogonal spaces, J.
Algebraic Combinatorics 4 (1995), 295316.
Available at http://www.uwyo.edu/moorhouse/pub/orthog.pdf
[7] A.E. Brouwer, A non-degenerate generalized quadrangle with lines of size four is
nite, pp.4749 in Advances in Finite Geometries and Designs, ed. J.W.P. Hirschfeld
et. al., Oxford Univ. Press., Oxford, 1991.
[8] M. Brown, Ovoids of P G(3, q), q even, with a conic section, J. London Math. Soc.
62 (2000), 569582.
[9] R.H. Bruck and H.J. Ryser, The nonexistence of certain nite projective planes,
Canad. J. Math.1 (1949), 8893.
[10] A.A. Bruen, Blocking sets in nite projective planes, SIAM J. Appl. Math. 21
(1971), 380392.
[11] F. Buekenhout, editor, Handbook of Incidence Geometry: Buildings and Foundations,
Elsevier, Amsterdam, 1995.
[12] P.J. Cameron, Projective and Polar Spaces, QMW Maths Notes 13, London, 1991.
A second edition (2000) is available at http://www.maths.qmul.ac.uk/~pjc/pps/
[13] P.J. Cameron, Combinatorics: Topics, Techniques, Algorithms, Cambridge Univ.
Press, Cambridge, 1994.
[14] P.J. Cameron and J.H. van Lint, Designs, Graphs, Codes and their Links, Camb.
Univ. Press, Cambridge, 1991.
234
Bibliography
[15] G. Cherlin, Locally nite generalized quadrangles with at most ve points per line,
Discrete Math. 291 (2005), 7379.
[16] S.S. Chern and P. Griths, Abels theorem and webs, Jahresberichte der Deut. Math.
Ver. 80 (1978), 13110; also, Corrections and addenta to our paper: Abels theorem
and webs, same Journal, 83 (1981), 7883.
[17] W. Cherowitzo, -Flocks and Hyperovals, Geom. Dedicata 72 (1998), 221246.
[18] W. Cherowitzo, Hyperoval Page, website.
http://www-math.cudenver.edu/~wcherowi/research/hyperoval/hypero.html
[19] J.H. Conway, P.B. Kleidman, and R.A. Wilson, New families of ovoids in O8+ , Geom.
Dedicata 6 (1988), 157170.
[20] D.A. Cox, J.B. Little and D. OShea, Ideals, Varieties, and Algorithms, 3rd ed.,
Springer-Verlag, New York, 2007.
[21] H.S.M. Coxeter, Projective Geometry, 2nd ed., University of Toronto Press, 1974.
[22] T. Czerwinski and D. Oakden, The translation planes of order twenty-ve, J. Combin.
Theory Ser. A 59 (1992), 193217.
[23] P. Deligne, La conjecture de Weil, Publ. Math. IHES 43 (1974), 273307.
[24] P. Dembowski, Mobiusebenen gerader Ordnung, Math. Ann. 157 (1964), 179205.
[25] P. Dembowski, Finite Geometries, Springer Verlag, Berlin, 1968.
[26] U. Dempwol, Translation planes of order 27, Des. Codes and Crypt. 27 (1994),
105121.
[27] W. Feit and G. Higman, The nonexistence of certain generalized polygons, J. Algebra
1 (1964), 114131.
[28] A. Gunawardena and G.E. Moorhouse, The nonexistence of ovoids in O9 (q), Europ.
J. Combinatorics 18 (1997), 171173.
Available at http://www.uwyo.edu/moorhouse/pub/9dim.pdf
[29] D.R. Hughes, Generalized incidence matrices over group algebras, Ill. J. Math. 1
(1957), 545551.
[30] D.R. Hughes, Collineations and generalized incidence matrices, Trans. Amer. Math.
Soc. 86 (1957), 284296.
[31] D.R. Hughes and F.C. Piper, Projective Planes, Springer Verlag, New York, 1973.
[32] T.W. Hungerford, Algebra, Springer-Verlag, New York, 1974.
[33] W.M. Kantor, Ovoids and translation planes, Canad. J. Math. 34 (1982), 11951207.
[34] N.M. Katz, An overview of Delignes proof of the Riemann hypothesis for varieties
over nite elds, Proc. Symp. Pure Math. 28, Amer. Math. Soc., Providence, R.I.,
1976, pp.275305.
Bibliography
235
[35] N. Koblitz, Introduction to Elliptic Curves and Modular Forms, 2nd ed., SpringerVerlag, New York, 1993.
[36] C.W.H. Lam, The search for a nite projective plane of order 10, Amer. Math.
Monthly 98 (1991), 305318.
[37] C.W.H. Lam, G. Kolesova and L. Thiel, A computer search for nite projective planes
of order 9, Discrete Math. 92 (1991), 187195.
[38] C.W.H. Lam, L. Thiel and S. Swiercz, The nonexistence of nite projective planes of
order 10, Canad. J. Math. 41 (1988), 1117-1123.
[39] E. Landau, Handbuch der Lehre der Verteilung der Primzahlen, Bd. 2, Teubner,
Leipzig-Berlin, 1909.
[40] S. Lie, Bestimmung aller Fl
achen, die in mehrfacher Weise durch Translationsbewegung einer Kurve erzeugt werden, Arch. f
ur Math. Bd. 7, Heft 2 (1882), 155176.
[41] L. Lunelli and M. Sce, K-archi completi nei piani proietivi desarguesiani di rango 8 e
16, Centro Calcoli Numerici, Politecnico di Milano, 1958.
[42] G.E. Moorhouse, Bruck nets, codes, and characters of loops, Designs, Codes and
Cryptography 1 (1991), 729.
Available at http://www.uwyo.edu/moorhouse/pub/loop1.pdf
[43] G.E. Moorhouse, Ovoids from the E8 root lattice, Geom. Dedicata 46 (1993), 287
297.
Available at http://www.uwyo.edu/moorhouse/pub/e8.pdf
[44] G.E. Moorhouse, Ovoids and translation planes from lattices, in Mostly Finite Geometries, ed. N. L. Johnson, Marcel Dekker Inc., 1997, 127134.
Available at http://www.uwyo.edu/moorhouse/pub/iowa1.pdf
[45] G.E. Moorhouse, Ranks of nets, Quasigroups and Related Systems 14 (2006), 6172.
Available at http://www.uwyo.edu/moorhouse/pub/qrs.pdf
[46] G.E. Moorhouse, Projective planes of small order, technical report.
Available at http://www.uwyo.edu/moorhouse/pub/planes/
[47] G.E. Moorhouse, Generalised polygons of small order, technical report.
Available at http://www.uwyo.edu/moorhouse/pub/genpoly/
[48] G.E. Moorhouse, On projective planes of order less than 32, in Finite Geometries,
Groups, and Computation, ed. A. Hulpke et. al., de Gruyter, Berlin, 2006, pp.149162.
Available at http://www.uwyo.edu/moorhouse/pub/wyoming.pdf
[49] G.E. Moorhouse, Abstract Algebra I, Lecture Notes for Math 5550, University of
Wyoming, revised 2003.
Available at http://www.uwyo.edu/moorhouse/handouts/algebra.pdf
[50] J. di Paola, On minimum blocking coalitions in small projective plane games, SIAM
J. Appl. Math. 17 (1969), 378392.
236
Bibliography
[51] S.E. Payne, Nonisomorphic generalized quadrangles, J. Algebra 18 (1971), 201212.

[52] S.E. Payne, A new innite family of generalized quadrangles, Congressus Numerantium 49 (1985), 115128.
[53] S.E. Payne and J.A. Thas, Finite Generalized Quadrangles, Pitman, Boston, 1984.
[54] T. Penttila, G.F. Royle and M.K. Simpson, Hyperovals in the known projective planes
of order 16, J. Combinatorial Designs 4 (1998), 5965.
[55] B. Qvist, Some remarks concerning curves of the second degree in a nite plane,
Ann. Acad. Sci. Fenn., no. 134 (1952), pp.127.
[56] G. Royle, Projective planes of order 16, technical report.
Available at http://people.csse.uwa.edu.au/gordon/remote/planes16/
[57] R. Scharlau, Buildings, pp.477645 in [11].
[58] J.-P. Serre, A Course in Arithmetic, Springer Verlag, New York, 1973.
[59] E.E. Shult and J.A. Thas, m-Systems of polar spaces, J. Combin. Theory Ser. A 68
(1994), 184204.
[60] B. Sturmfels, Algorithms in Invariant Theory, Springer-Verlag, Vienna, 1993.
[61] D.E. Taylor, The Geometry of the Classical Groups, Heldermann Verlag, Berlin, 1992.
[62] J.A. Thas, Ovoids and spreads of nite classical polar spaces, Geometriae Dedicata
10 (1981), 135144.
[63] J.A. Thas, Generalized polygons, pp.383431 in [11].
[64] J. Tits, Buildings of Spherical Type and finite BN-pairs, Springer-Verlag, New York,
1974.
[65] J.H. van Lint, Introduction to Coding Theory, 2nd ed., Springer-Verlag, Berlin, 1992.
[66] H.A. Wilbrink, A note on planar dierence sets, J. Comb. Theory Ser. A 38 (1985),
9495.
Index
absolute . . . . . . . . . .
absolutely irreducible . . . .
ane
general linear group . . .
linear transformation . .
plane . . . . . . . . .
semilinear transformation
space . . . . . . . . .
algebra . . . . . . . . . .
algebraically dependent . . .
algebraically independent . .
alternate regulus . . . . . .
alternating bilinear form . . .
Andre
neareld . . . . . . . .
plane . . . . . . . . .
quasield . . . . . . .
antiag . . . . . . . . . .
antiag collineation . . . . .
antipodal . . . . . . . . .
apartment . . . . . . . . .
arc . . . . . . . . . . . .
augmentation map . . . . .
automorphism . . . . . . .
automorphism group . . . .
axis . . . . . . . . . . . .
. 61, 69, 158

. . . . 109
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 208
. 208
. . 9
. 209
. 121
. 211
. 231
. 231
148, 170
142, 159
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . 18
16, 18
. . 18
. . 51
. . 57
. . 40
. 188
68, 166
. 100
. . 4
. . 4
. . 57
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . 60
53, 105
. . 50
. 220
. 220
. 186
. 110
. 220
. 220
. . 5
. 105
. . 85
. 188
cap . . . . . . . . . . .
central element . . . . . .
centre . . . . . . . . . .
chain of subspaces . . . .
character . . . . . . . .
Cherowitzo hyperoval . . .
Chevalley-Warning Theorem
circulant matrix . . . . .
class 0 . . . . . . . . . .
class 1 . . . . . . . . . .
classical ane plane . . . .
classical projective plane . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Baer collineation . .
Baer subplane . . .
Bezouts Theorem . .
binary alphabet . . .
binary code . . . .
bipartite graph . . .
birational equivalence
bit . . . . . . . .
bitstring . . . . . .
block . . . . . . .
blocking set . . . .
Bruck-Ryser Theorem
building . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
177
216
. 57
126
224
. 73
201
. 94
199
199
. 9
. 35
closed
ball . . . . . . . . . . . . . . 220
conguration . . . . . . . . . . . 50
path . . . . . . . . . . . . . 185
code . . . . . . . . . . . . . . . . 220
codeword . . . . . . . . . . . . . 220
collinearity graph . . . . . . . . . . . 22
collineation . . . . . . . . . . . . . 53
commutative algebra . . . . . . . . 211
companion matrix . . . . . . . . . . 198
complete bipartite graph . . . . . . . 186
component . . . . . . . . . . . . . . 13
cone . . . . . . . . . . . . . . . . 148
congruent . . . . . . . . . . . . 85, 140
conic . . . . . . . . . . . . . . . . 67
conjugate . . . . . . . . . . . . . 100
conjugate linear transformation . . . . 208
correlation . . . . . . . . . . . . . . 60
Coxeter complex . . . . . . . . . . 188
Coxeter group . . . . . . . . . . . 189
Coxeter-Dynkin diagram . 125, 155, 181, 189
curve . . . . . . . . . . . . . . 42, 109
cyclic 3-net . . . . . . . . . . . . . . 29
degenerate
closed conguration .
bilinear form . . . .
quadratic form . . .
quadric . . . . . .
degree of a representation
dehomogenize . . . . .
Desargues Theorem . . .
Desarguesian plane . . .
design . . . . . . . . .
diameter . . . . . . . .
Dicksons Criterion . . .
dierence set . . . . . .
dierential 1-form . . . .
digon . . . . . . . . .
direct product . . . . .
discriminant . . . . . .
distance (in a graph) . .
division algebra . . . . .
division ring . . . . . .
double translation surface
doubly transitive . . . .
dual . . . . . . . . . .
dual code . . . . . . .
duality . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 51
140
140
148
204
. 42
. 65
. 66
12, 40
. 186
. 202
. . 90
. 114
. . 5
. 209
86, 140
. 185
. 212
. 212
. . 23
. 205
. . 4
. 223
. . 37
.
.
.
.
.
.
.
.
e-error correcting . . . . . . . . . . 220

e-error detecting . . . . . . . . . . 220
egglike . . . . . . . . . . . . . . . 168
238
elation . . . . . . . . . . . . . . . . 57
elementary divisor . . . . . . . . . . 76
elementary symmetric polynomials . . 229
elliptic curve . . . . . . . . . . . . 114
elliptic quadric . . . . . . . . . 148, 150
equivalent linear representations . . . . 56
equivalent permutation representations
206
error syndrome . . . . . . . . . . . 222
Euler characteristic . . . . . . . . . 110
exponential sum . . . . . . . . . . . 30
exterior
algebra . . . . . . . . . . . . 218
point . . . . . . . . . . . . . . 69
power . . . . . . . . . . . . . 218
external direct product . . . . . . . 209
external semidirect product . . . . . 210
faithful representation . . . . .
Feit-Higman Theorem . . . . .
nite eld . . . . . . . . . .
xed point free collineation . . .
ag . . . . . . . . . . . . .
ag collineation . . . . . . . .
frame . . . . . . . . . . . .
fundamental set of invariants . .
Fundamental Theorem
of Projective Geometry . .
of Projective Plane Geometry
Galois eld . . . . .
Galois plane . . . .
Gaussian coecient .
general linear group .
generalized
digon . . . . .
elation . . . . .
hexagon . . . .
homology . . .
incidence matrix
perspectivity . .
n-gon . . . . .
octagon . . . .
polygon . . . .
quadrangle . . .
triangle . . . .
generator matrix . .
genus . . . . . . .
girth . . . . . . .
graded ring . . . .
graded subring . . .
Gram matrix . . . .
Grassmann variety .
Grassmannian . . .
grid . . . . . . . .
.
.
.
.
.
.
.
.
. 204
. 187
. 195
. . 57
. . 51
. . 57
124, 135
. . 230
.
.
.
.
.
.
. . . 124
. . . 63
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 66
. 66
122
207
. 123
. . 57
. 187
. . 57
. 100
. . 57
. 186
. 188
. 186
. 178
. 187
. 221
110, 115
. . 186
211, 219
. . 230
. . 140
. . 138
. . 138
. . 179
group algebra
. . . . . . . . . . . 211
Hadamard matrix . . . .
Hall Multiplier Theorem .
Hall plane . . . . . . .
Hamming
code . . . . . . . .
distance . . . . . .
weight . . . . . . .
Hasse-Minkowski Theorem
Hasse-Weil bound . . . .
Hermitian
curve . . . . . . .
form . . . . . . .
unital . . . . . . .
Hilbert series . . . . . .
Hilbert symbol . . . . .
Holy Grail . . . . . . .
homogeneous coordinates
homogenize . . . . . .
homology . . . . . . .
hyperbolic pair . . . . .
hyperbolic quadric . . .
hyperoval . . . . . . .
hyperplane . . . . . . .
hypersurface . . . . . .
idempotent . . . . . . .
incidence
graph . . . . . . .
matrix . . . . . . .
relation . . . . . .
structure . . . . . .
incident . . . . . . . .
information rate . . . .
interior point . . . . . .
internal direct product . .
internal semidirect product
invariant . . . . . . . .
inversive plane . . . . .
involution . . . . . . .
irreducible . . . . . . .
isometry . . . . . . . .
isomorphism . . . . . .
. . . . . . . 90
. . . . . . . 95
. . . . . . . 15
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
78, 126, 226

. . . 220
. . . 221
. . . . 87
. . . 111
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 113
. 160
. . 62
. 230
. . 87
. . 39
. . 35
. . 42
. . 57
. 143
148, 150
. . . 68
. . 121
. . 124
.
.
.
.
.
.
.
.
.
.
. . . . . . 216
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
60, 123
. . 3
. . 3
. . 3
3, 158
. 221
. . 69
. 209
. 210
. 228
12, 168
. . 60
109, 213
140, 158
. . . 4
.
.
.
.
.
.
.
.
.
.
.
.
35, 51
join . . . . . . . . . . . . . . .
kernel . . . . . . .
Klein correspondence
Klein quadric . . . .
knot . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 18
161
161
. 69
Latin square . . . . . . . . . . . . . 18
left action . . . . . . . . . . . . . 204
left module . . . . . . . . . . . . . 212
239
Legendre symbol . . . . . . . . . .
line . . . . . . . . . . . . . . . .
line at innity . . . . . . . . . . .
linear
code . . . . . . . . . . . . . .
representation . . . . . . . . .
space . . . . . . . . . . . . .
loop . . . . . . . . . . . . . . . .
low density parity check (LDPC) codes
Lucas Theorem . . . . . . . . . .
Lunelli-Sce hyperoval . . . . . . . .
MacWilliams relations . . . . . .
Maschkes Theorem . . . . . . .
MDS code . . . . . . . . . . .
meet . . . . . . . . . . . . .
minimum distance . . . . . . .
minimum weight . . . . . . . .
Minkowski sum . . . . . . . . .
M
obius plane . . . . . . . . . .
model . . . . . . . . . . . . .
module . . . . . . . . . . . .
Moulton plane . . . . . . . . .
multinomial coecient . . . . . .
Multinomial Theorem . . . . . .
multiplicity of intersection . . . .
multiplier . . . . . . . . . . .
mutually orthogonal Latin squares
natural module . . . .
net . . . . . . . . .
noncommutative algebra
nondegenerate . . . .
nonprincipal character .
norm . . . . . . . .
nucleus . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 86
3, 5
. 40
221
213
. 5
. 16
188
131
. 72
. 223
. 214
. 222
35, 51
. 220
. 221
. . 23
. . 12
. . 9
. 212
. . 11
. 131
. 131
. . 48
. . 95
. . 19
. . . 213
. . . . 20
. . . 211
67, 140, 148
. . . 224
. . . 196
. . . . 69
o-polynomial . . . . . . . . . . . . . 71
orbit . . . . . . . . . . . . . . . 204
ordered frame . . . . . . . . . . . 124
ordered quadrangle . . . . . . . . . . 63
orthogonal
direct sum . . . . . . . . . . . 141
group . . . . . . . . . . . 140, 159
Latin squares . . . . . . . . . . . 18
polar space . . . . . . . . . . . 158
polarity . . . . . . . . . . . 62, 158
quadrangle . . . . . . . . . . . 180
Ostrom-Dembowski-Wagner Theorem . . 65
oval . . . . . . . . . . . . . . . . . 68
ovoid
of a generalized quadrangle . . . . 184
of a polar space . . . . . . . . . 173
of projective 3-space . . . . . . . 166
. . . . . . . . . . 164
of a quadric
p-adic numbers . . . . .
Pappian plane . . . . .
Pappus Theorem . . . .
parabolic quadric . . . .
parallel lines . . . . . .
parity check bit . . . . .
parity check matrix . . .
partial linear space . . .
partial spread . . . . .
Pascals Theorem . . . .
passant . . . . . . . .
path . . . . . . . . .
Payne hyperovals . . . .
perfect code . . . . . .
permutation
action . . . . . . .
group . . . . . . .
polynomial . . . . .
perspectivity . . . . . .
planar
collineation . . . .
dierence set . . . .
ternary ring . . . .
Pl
ucker coordinates . . .
Pl
ucker map . . . . . .
pointed conic . . . . . .
points at innity . . . .
polar space . . . . . . .
polarity . . . . . . . .
polarization . . . . . .
p-rank . . . . . . . . .
primitive element of a eld
primitive polynomial . .
principal character . . .
projective
completion . . . . .
dimension . . . . .
general linear group .
line . . . . . . . .
plane . . . . . . .
space . . . . . . .
pure wedge product . . .
q-ary code . .
quadrangle . .
quadratic form
quadric . . .
quasield . .
quaternion . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . 90
. . 66
44, 64
. 150
. . 10
. 225
. 222
. . 5
. . 21
. . 46
. . 68
. 185
. . 73
. 221
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . 57
. . 90
. . 66
. 136
. 136
. . 71
. . 40
. 158
61, 157
. 139
. . 77
. 195
. 195
. 224
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
40, 124
. 121
. 207
. 112
. . 35
121, 126
163, 219
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 220
. . 63
85, 139
. 148
. . 16
. 212
204
204
202
. 57
radical . . . . . . . . . . . . . . . 140
rank of a building . . . . . . . . . . 188
240
rank of a web . . . . . . .
rational point . . . . . .
rationally congruent matrices
regular
dierential form . . .
hyperoval . . . . . .
module . . . . . . .
ovoid . . . . . . . .
permutation group . .
spread . . . . . . . .
reguli . . . . . . . . . .
regulus . . . . . . . . .
repetition code . . . . . .
representation . . . . . .
residual geometry . . . . .
residue . . . . . . . . .
Riemann hypothesis . . . .
Riemann sphere . . . . .
right action . . . . . . .
ring of invariants . . . . .
. . . . . . 25
. . . . . 109
. . . . . . 85
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
scalar matrix . . . . . . . .
secant . . . . . . . . . . .
Segre hyperoval . . . . . . .
Segres Theorem . . . . . .
self-dual . . . . . . . . . .
self-referential . . . . . . .
semidirect product . . . . .
semilinear transformation . .
semisimple . . . . . . . . .
sesquilinear form . . . . . .
sharply divide . . . . . . .
sharply transitive . . . . . .
similar matrices . . . . . . .
similarity . . . . . . . . .
simple module . . . . . . .
Singleton bound . . . . . .
singular point . . . . . . .
skew . . . . . . . . . . .
skeweld . . . . . . . . . .
slope matrices . . . . . . .
Smith normal form . . . . .
smooth curve . . . . . . . .
sphere-packing bound . . . .
sphere with handles . . . . .
spherical building . . . . . .
spread
of a generalized quadrangle
of a polar space . . . . .
of a projective space . . .
of a vector space . . . .
squarefree part . . . . . . .
stabilizer . . . . . . . . . .
. . . . 115
. . . 71, 72
. . . . 213
165, 172, 185
. . . . 205
. . . . 115
. . 148, 169
. . 148, 170
. . . . 225
. . 204, 213
. . . . 125
. . 125, 157
. . . . 112
. . . 42, 110
. . . . 204
. . . . 228
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 207
. . 68
. . 72
. . 74
. 4, 38
. 240
. 210
. 208
. 213
. 160
. . 79
. 205
. . 86
. 160
. 213
. 222
209, 140
. . 123
. 66, 212
. . . 14
. . . 76
. . 109
. . 221
. . 109
. . 191
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
184
173
169
. 13
. 87
204
strongly regular graph

submodule . . . . .
subplane . . . . . .
subspace . . . . . .
Suzuki-Tits ovoid . .
symmetric group . .
symplectic
form . . . . .
group . . . . .
polar space . . .
polarity . . . .
quadrangle . . .
syndrome . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 22
213
. 51
126
167
204
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
159
160
159
159
180
222
tangent . . . . . . . . . . . . . . . 68
thick . . . . . . . . . . 40, 126, 179, 186
thin . . . . . . . . . . . 40, 126, 179, 186
totally isotropic subspace . . . . . . 140
totally singular subspace . . . . . . . 140
trace . . . . . . . . . . . . . . . 196
2-transitive permutation group . . . . 205
transitive permutation group . . . . . 205
translation
complement . . . . . . . . . . . 17
group . . . . . . . . . . . . . . 17
net . . . . . . . . . . . . . 21, 169
plane . . . . . . . . . . . . . . 13
transversal . . . . . . . . . . . . . 170
transversely . . . . . . . . . . . . . 23
triality . . . . . . . . . . . . . . 157
triality quadric . . . . . . . . . . . 157
triangular collineation . . . . . . . . . 57
unital . . . .
unitary
group . .
polar space
polarity .
quadrangle
. . . . . . . . . . . . 62
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 160
. 160
62, 160
. 181
Veblen-Pasch axiom . . . . . . . . . 126

web . . . . . . . . .
Wedderburns Theorem
wedge product . . . .
weight . . . . . . . .
weight distribution . .
weight enumerator . .
Weil conjectures . . .
white whale . . . . .
Witts Theorem . . . .
Wyoming planes . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 24
212
218
221
223
223
112
. 31
144
105
Zeta function . . . . . . . . . . . . 111

Incidence Geometry

Uploaded by

Copyright:

Available Formats

Incidence Geometry

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Incidence Geometry

Uploaded by

Copyright:

Available Formats

ii

Note on First Corrected Edition

II. Ane Planes

Denitions and Examples . . . . . . . . . . . . . . . . . . . . . . . . . 9

III. Projective Planes

Denitions and Examples . . . . . . . . . . . . . . . . . . . . . . . . 35

IV. Projective and Polar Spaces

Classical Ane and Projective Spaces . . . . . . . . . . . 121

Finite Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

Aut = (P Q)(mn), (P R)(n)

1. Definitions and Examples

[Q] = {, n},

[m] = {P, R},

II. AFFINE PLANES

2. Definitions and Examples

II. AFFINE PLANES

II. AFFINE PLANES

3.3 Example: Regular Spreads.

Consider distinct i, j {0, 1, 2, . . . , q n 1} and suppose

then x(Mi Mj ) = 0. Since Mi Mj is nonsingular, this forces x = 0 and so Vi Vj = {0}.

Let F = F3 . The nine matrices

is an automorphism of E xing every element of F , and mapping squares to squares. Dene

if y is a square (possibly zero);

II. AFFINE PLANES

such that the subspaces of the form

together with the subspace {(0, y) : y Q} form a spread of V .

II. AFFINE PLANES

4. Let G = M1 , M2  GL2 (F5 ) be the subgroup generated by

5. Let M0 , M1 , . . . , Mqn 1 be a set of n n slope matrices over Fq with M0 = 0, M1 = I. Let K be the

4. Latin Squares and Nets

4. Latin Squares and Nets

is dened by the pair of orthogonal Latin squares

II. AFFINE PLANES

Conversely, any set of k 2 mutually orthogonal Latin squares of order n species an

4. Latin Squares and Nets

corresponds to the 4-net

II. AFFINE PLANES

5. Nets and Webs

for i = 1, 2. We will assume that i (0) = 0; otherwise simply translate i appropriately

5. Nets and Webs

5.1 Example. Fix c F with c

II. AFFINE PLANES

5.2 Example. Fix c F with c

5. Nets and Webs

a general point w W is uniquely determined by any two of its coordinates

3 (u(s, t)) + 4 (v(s, t)) = 1 (s) + 2 (t) S.

II. AFFINE PLANES

Dene the coordinate functions xi : F 2 F by

By Theorem 5.5 we may choose a basis for V0 of the form

41 (v(s, t)) 42 (v(s, t)) 4m (v(s, t))

5. Nets and Webs

Consider the four parameterized smooth curves

This gives the j-th coordinate in the vector relation

1 (s) + ()3 (u(s, t)) + ()4 (v(s, t)) = 0

1 (s) + ()3 (u) + ()4 (

1 (s) + ()3 (u) + ()4 (

II. AFFINE PLANES

5. Nets and Webs

II. AFFINE PLANES

Aut = (P Q)(mn), (P R)(n)

4. Let G = M1 , M2 GL2 (F5 ) be the subgroup generated by

1 (s) + ()3 (u) + ()4 (

1 (s) + ()3 (u) + ()4 (

point (x, y, 1)

and (1, 1, 0).