Poly 08 Feb

POLYNOMIALS IN FINITE GEOMETRY
Peter Sziklai
20 February, 2008
ii
Foreword
-1
iii
Special foreword
Before the real Foreword, let me say some words about this book. It is ready in
the sense that this is the topic that I wanted to write about and this is how I
wanted to write about it. Naturally, this book is not finished in the sense that (i)
it is about an active field of mathematics, which changes rapidly; (ii) if you have
worked on a book for years then it is hard to stop it: every day you can have a
new idea how to slightly improve it; (iii) no editor/referees have seen it yet, hence
there may (must!) be typos, inaccuracies, missing citations, line overflow errors...
left. It would be nice to increase the number of figures as well.
However, I think that in the current state of this volume it is possible to decide
about its values and shortcomings. Although I understand that it can be evaluated
by this current version, in the spirit of (i) and (ii) above I will keep an updated
version on the secret, unlinked webpage
http://www.cs.elte.hu/ sziklai/poly.html , (where I will enlist all the
changes performed, too), just for pleasure. When a decision of publication is
reached, the editors of the publisher can decide between this and a possibly updated version.
Foreword
A most efficient way of investigating combinatorially defined pointsets of projective

spaces over finite fields is associating polynomials to them. This technique was first
used by Jamison and Bruen, then, followed by several people, became a standard
method; nowadays, the contours of a growing theory can be seen already.
The polynomials we use should reflect the combinatorial properties of the pointset,
then we have to be equipped with enough means to handle our polynomials and get
an algebraic description about them; finally, we have to translate the information
gained back to the original, geometric language.
The first investigations in this field examined the coefficients of the polynomials,
and this idea proved to be very efficient. Then the derivatives of the polynomials
came into the play and solving (differential) equations over finite fields; a third
branch of results considered the polynomials as algebraic curves. The idea of associating algebraic curves to pointsets goes back to Segre, recently a bunch of new
applications have shown the strength of this method. Finally, due to Gacss recent
results, dimension arguments on polynomial spaces have become fruitful.
This book is an attempt to collect and classify the most interesting ways of applying polynomials in finite geometry.
We focus on combinatorially defined (point)sets of projective geometries. They
are defined by their intersection numbers with lines (or other subspaces) typically,
like arcs, blocking sets, nuclei, caps, ovoids, etc. Here we do not want to give a
iv
complete account on them, however some of the best results will be presented.
In order to show a wider scope of different applications, some other fields are
gently touched like group theory, graph theory, etc.
This book is divided into chapters. The first one (Background) contains selected
tools for a finite geometer, from the basic facts to some theory of polynomials over
finite fields; proofs are only provided when they are short or interesting for us. We
provide slightly more information than the essential background for the second
chapter. After the basic facts (Section 5) the most useful representations of affine
and projective spaces are presented (6), then we introduce our main tool, the Redei
polynomial associated to pointsets (7,8). The coefficients of Redei polynomials are
elementary symmetric polynomials themselves, what we need to know about them
and other invariants of subsets of fields or spaces is collected in Section 9. The
multivariate polynomials associated to pointsets can be considered as algebraic
varieties, so we can use some basic parts of algebraic geometry (10). Finally, in
Section 11 some background needed for stability results is presented.
The second (and main) chapter contains several results of finite Galois geometry,
where polynomials play a main role. We start with general results on intersection
numbers of planar pointsets (12,13). Then we turn to special cases as arcs, maximal arcs, unitals, semiovals, untouchable sets. (14-17). The next highlight is the
topic of directions (18,19) and blocking sets (20-22). Section 23 shows the new
resultant method for stability results. Affine blocking sets and nuclei are considered in sections 24,25. Section 26 introduces an interesting mixed representation.
Stability theorems for flocks, some basic facts for spreads and a nice result on
ovoids is proved in Sections 27,28. Finally some non-geometrical applications are
collected in Section 29.
The last chapter contains hints and solutions for the exercises.
Contents
-1
Special foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . .
iii
Foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
iii
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . .
Definitions, basic notation . . . . . . . . . . . . . . . . . . . . . . .
Background
Finite fields and polynomials . . . . . . . . . . . . . . . . . . . . .
5.1
Some basic facts . . . . . . . . . . . . . . . . . . . . . . . .
5.2
Self-dual normal bases . . . . . . . . . . . . . . . . . . . . .
5.3
Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.4
Differentiating polynomials . . . . . . . . . . . . . . . . . .
5.5
Polynomials vanishing at many points, Alons Combinatorial Nullstellensatz and the Ball-Serra refinement . . . . . .
5.6
Additive, linearized polynomials . . . . . . . . . . . . . . .
16
5.7
The random-like behaviour . . . . . . . . . . . . . . . . . .
17
5.8
Lacunary polynomials . . . . . . . . . . . . . . . . . . . . .
19
Representations of affine and projective geometries . . . . . . . . .
22
6.1
Subspaces, subgeometries . . . . . . . . . . . . . . . . . . .
24
6.2
Transformations . . . . . . . . . . . . . . . . . . . . . . . .
25
The Redei polynomial and its derivatives
. . . . . . . . . . . . . .
26
7.1
The Redei polynomial . . . . . . . . . . . . . . . . . . . . .
26
7.2
Differentiation in general . . . . . . . . . . . . . . . . . . .
30
7.3
Hasse derivatives of the Redei polynomial . . . . . . . . . .
32
7.4
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . .
33
Univariate representations . . . . . . . . . . . . . . . . . . . . . . .
35
vi
Contents
10
11
2
8.1
The affine polynomial and its derivatives . . . . . . . . . . .
36
8.2
The projective polynomial . . . . . . . . . . . . . . . . . . .
37
8.3
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . .
38
Symmetric polynomials and invariants of subsets . . . . . . . . . .
39
9.1
The Newton formulae . . . . . . . . . . . . . . . . . . . . .
39
9.2
The invariants uf , vf and wf . . . . . . . . . . . . . . . . .
43
9.3
Arbitrary subsets . . . . . . . . . . . . . . . . . . . . . . . .
47
9.4
The generalized formulae . . . . . . . . . . . . . . . . . . .
51
9.5
Resultants . . . . . . . . . . . . . . . . . . . . . . . . . . . .
54
Elements of algebraic geometry . . . . . . . . . . . . . . . . . . . .
61
10.1
Conditions implying linear (or low-degree) components . . .
65
Finding the missing factors, removing the surplus factors . . . . . .
68
Polynomials in geometry
71
12
Lines meeting a pointset . . . . . . . . . . . . . . . . . . . . . . . .
71
12.1
Prescribing the intersection numbers . . . . . . . . . . . . .
71
12.2
A general result . . . . . . . . . . . . . . . . . . . . . . . . .
72
12.3
Another immediate corollary . . . . . . . . . . . . . . . . .
73
Sets with constant intersection numbers mod p . . . . . . . . . . .
74
13.1
Sets with intersection numbers 0 mod r . . . . . . . . . . .
76
13.2
Small and large super-Vandermonde sets . . . . . . . . . . .
80
14
Arcs, Segre . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
85
15
Maximal arcs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
89
15.1
Existence: hyperovals
. . . . . . . . . . . . . . . . . . . . .
89
15.2
Non-existence . . . . . . . . . . . . . . . . . . . . . . . . . .
93
15.3
Embeddability . . . . . . . . . . . . . . . . . . . . . . . . .
96
16
Unitals, semiovals . . . . . . . . . . . . . . . . . . . . . . . . . . . .
98
17
Sets without tangents: where Segre meets Redei . . . . . . . . . . . 102
13
18
17.1
The proof using Segres method . . . . . . . . . . . . . . . . 103
17.2
Untouchable sets in Galois planes of even order . . . . . . . 109
Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
18.2
When D is (contained in) a subgroup of GF(q) . . . . . . . 112
When D is (contained in) a subgroup of q+1 1 GF(q 2 ) . 115
18.3
Sets determining
18.4
Small blocking sets of Redei type are linear . . . . . . . . . 116
18.1
q+1
2
directions: affine linear pointsets
. 115
Contents
19
vii
Over prime fields: the Gacs method . . . . . . . . . . . . . . . . . . 122

19.1
Permutation polynomials . . . . . . . . . . . . . . . . . . . 124
19.2
Directions in the plane . . . . . . . . . . . . . . . . . . . . . 125
19.3
Linear combinations of three permutation polynomials . . . 131
20
Projective linear pointsets . . . . . . . . . . . . . . . . . . . . . . . 134
21
Blocking sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
22
21.1
One curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
21.2
Three new curves . . . . . . . . . . . . . . . . . . . . . . . . 144
21.3
Three old curves . . . . . . . . . . . . . . . . . . . . . . . . 147
21.4
Small blocking sets . . . . . . . . . . . . . . . . . . . . . . . 149
Multiple blocking sets . . . . . . . . . . . . . . . . . . . . . . . . . 154

22.1
23
24
25
A t (mod p) result . . . . . . . . . . . . . . . . . . . . . . . 155
Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
23.1
Stability theorems for small blocking sets and the SzonyiWeiner method . . . . . . . . . . . . . . . . . . . . . . . . . 164
23.2
Stability theorems for blocking sets: the prime case . . . . . 167
23.3
Stability theorems for blocking sets: the non-prime case . . 170
23.4
Stability theorem for sets of even type . . . . . . . . . . . . 176
23.5
Number of lines meeting a (q + 2)-set . . . . . . . . . . . . 179
Jamison, Brouwer-Schrijver, affine blocking sets . . . . . . . . . . . 180

24.1
Applications of the punctured Nullstellensatz . . . . . . . . 184
24.2
Redei polynomials in three indeterminates . . . . . . . . . . 189
24.3
Redei polynomials in many indeterminates
24.4
Affine blocking sets . . . . . . . . . . . . . . . . . . . . . . . 195
. . . . . . . . . 191
Nuclei and affine blocking sets . . . . . . . . . . . . . . . . . . . . . 198

25.1
Lower nuclei . . . . . . . . . . . . . . . . . . . . . . . . . . 201
25.2
Internal nuclei . . . . . . . . . . . . . . . . . . . . . . . . . 203
25.3
Higher dimensions . . . . . . . . . . . . . . . . . . . . . . . 203
26
Mixed representations . . . . . . . . . . . . . . . . . . . . . . . . . 205
27
Flocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
28
27.1
Partial flocks of the quadratic cone in PG(3, q) . . . . . . . 208
27.2
Partial flocks of cones of higher degree . . . . . . . . . . . . 211
Spreads and ovoids . . . . . . . . . . . . . . . . . . . . . . . . . . . 214

28.1
Spreads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
28.2
Ovoids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
viii
Contents
29
Non-geometrical applications . . . . . . . . . .
29.1 Normal factorizations of Abelian groups
29.2 The Paley graph . . . . . . . . . . . . .
29.3 Representing systems . . . . . . . . . .
29.4 Wielandts visibility theorem . . . . . .
29.5 Burnsides theorem . . . . . . . . . . . .
29.6 Jaegers conjecture . . . . . . . . . . . .
29.7 Graphs containing p-regular subgraphs .
29.8 Blokhuis proof of Bollobas theorem . .
29.9 Alon-F
uredi . . . . . . . . . . . . . . . .
29.10 Chevalley-Warning . . . . . . . . . . . .
29.11 Cauchy-Davenport . . . . . . . . . . . .
29.12 Another application . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
219
219
220
221
222
222
223
225
225
226
226
227
228
Hints and short solutions to the exercises

231
1
Of Chapter I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
2
Of Chapter II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
Glossary of concepts
241
Notation
243
Index
244
References
245
1. Introduction
Introduction
This one is not intended to be an introductory textbook about finite geometries,

finite fields nor polynomials. There are very good books of these kinds available,
e.g. Ball-Weiner [23] for a smooth and fascinating introduction to the concepts of
finite geometries, the three volumes of Hirschfeld and Hirshfeld-Thas [77, 78, 79]
as handbooks and Lidl-Niederreiter [90] for finite fields. Still, the interested reader,
even with a little background, may find all the definitions and basic information
here (the Glossary of concepts at the end of the volume can help) to enjoy this
interdisciplinary field in the overlap of geometry, combinatorics and algebra.
We would like to use a common terminology. Parts and reformulations of results
of some interest for non-geometers will be emphasized.
Bruen and Fisher called the polynomial technique as the Jamison method and
summarized it in performing three steps: (1) Rephrase the theorem to be proved
as a relationship involving sets of points in a(n affine) space. (2) Formulate the
theorem in terms of polynomials over a finite field. (3) Calculate. (Obviously, step
3 carries most of the difficulties in general.) In some sense it is still the method,
we will show several ways how to perform steps 1-3.
We have to mention the book of Laszlo Redei [103] from 1970, which inspired a
new series of results on blocking sets and directions in the nineties.
There are a few survey papers on polynomial methods, by Blokhuis [35, 36], Szonyi
[122], Ball [7], I included here some of their material. The form of a book allows
to show the algebraic background in more details; also to insert the most recent
results.
The typical theories in this book have the following character. Define a class of
(point)sets (of a geometry) in a combinatorial way (which, in general, means restrictions on the intersections with subspaces); examine its numerical parameters
(usually the spectrum of sizes in the class); find the minimal/maximal values of the
spectrum; characterize the extremal entities of the class; show that the extremal
(or other interesting) ones are stable in the sense that there are no entities of
the class being close to the extremal ones.
There are some fundamental concepts and ideas that we feel worth to put into
light all along this treatise:
an algebraic curve (or envelope) containing all the interesting lines of a
pointset;
the Lemma of tangents with its generalizations and analogues;
resultants as sophisticated tools improving the simple use of Bezouts theorem;
examination of (lacunary) coefficients of polynomials;
Introduction
considering (sub)spaces generated by polynomials.
Apology. There are some theorems being proved more than once in this book,
these multiple proofs were included in order to show how the different ideas and
techniques can lead to the same result; sometimes they even serve as touchstones
of a new method.
Also, this book contains a few proofs being longer than the usual ones. They
were included to show some elaborate series of ideas that lead to practical use of
polynomial techniques.
There are exercises, like a hundred and twenty. One type of them is for the necessary technicalities where no further idea is needed (so boring but useful), another
type is where an argument analogous to the preceding ones can solve the question
(so good to practice), and a third type which needs minor new ideas (so the interesting ones). We provide short solutions for a great part of them at the end of
the volume.
Please!
There are side comments on the margins. They might help reading, the editors/referees are kindly asked to decide whether they should be (i) left as they
are, or (ii) included into the normal text, or (iii) the number of them can be
increased (I would be happy to add more).
Acknowledgements
Most of this book is about the work of Simeon Ball, Aart Blokhuis, Andras Gacs,
Tam
as Sz
onyi and Zsuzsa Weiner. They, together with the author, contributed in
roughly one half of the references; also their results (and sometimes even their texts
with minor modifications: thanks to them for letting me do so) form an important
part of this volume. Not least, I always enjoyed their warm, joyful, inspirating and
supporting company in various situations in the last some years. I hope that the
selection and the choice of the topics covered in this book makes them happy as
well, and that being the topic of a book is at least as funny as writing it. Most
of them have seen a preliminary version of it, I am grateful for all the suggestions
they made.
Above all I am deeply indebted to Tamas Szonyi, from whom most of my knowledge
and most of my enthusiasm for finite geometries I have learned.
It was G
abor Korchm
aros who suggested (and not only to me) some seven years
ago to write a book like this. Probably any of us, or maybe someone else, could
have done it. Of course this one is about the way how I can see the topic.
Last but not least I am grateful to all the colleagues and friends who helped me
in any sense in the last some years: researchers of Ghent (and among them my
multiple coauthor Leo Storme), Potenza, Naples, Barcelona, Eindhoven and, of
course, Budapest.
3. Definitions, basic notation
Definitions, basic notation
We will not be very strict and consistent in the notation (but at least well try to
be). However, here we give a short description of the typical notation we are going
to use.
If not specified differently, q = ph is a prime power, p is a prime, and we work in the
Desarguesian projective (or affine) space PG(n, q) (AG(n, q), resp.), each space coordinatized by the finite (Galois) field GF(q). The n-dimensional vectorspace over
GF(q) will be denoted by V(n, q) or simply by GF(q)n . When discussing PG(n, q)
and the related V(n + 1, q) together then for a subspace dimension will be meant
projectively while vector space dimension will be called rank. A field, which is not
necessarily finite will be denoted by F.
In general capital letters X, Y, Z, T, ... will denote independent variables, while
x, y, z, t, ... will be elements of GF(q). A pair or triple of variables or elements in
any pair of brackets can be meant homogeneously, hopefully it will be always clear
from the context and the actual setting.
We write X or V = (X, Y, Z, ..., T ) meaning as many variables as needed; Vq =
(X q , Y q , Z q , ...). As over a finite field of order q for each x GF(q) xq = x holds,
two different polynomials, f and g, in one or more variables, can have coinciding
values everywhere over GF(q). In this case we ought to write f g, as for
univariate polynomials f (X), g(X) it means that f g (mod X q X) in the ring
GF(q)[X]. However, as in the literature f g is used in the sense f and g are
equal as polynomials, we will use it in the same sense; though simply f = g and
f (X) = g(X) may denote the same.
Throughout this book we mostly use the usual representation of PG(n, q).
This means that the points have homogeneous coordinates (x, y, z, ..., t) where
x, y, z, ..., t are elements of GF(q). The hyperplane [a, b, c, ..., d] of the space have
equation aX + bY + cZ + ... + dT = 0.
When PG(n, q) is considered as AG(n, q) plus the hyperplane at infinity, then we
will use the notation H for that (ideal) hyperplane. If n = 2 then H is called
the line at infinity ` .
According to the standard terminology, a line meeting a pointset in one point will
be called a tangent and a line intersecting it in r points is an r-secant (or a line of
length r).
This book is about combinatorially defined (point)sets of (mainly projective or
affine) finite geometries. They are defined by their intersection numbers with lines
(or other subspaces) typically. The most important definitions and basic information are collected in the Glossary of concepts at the end of this book. These are:
blocking sets, arcs, nuclei, spreads, sets of even type, etc.
Warning. In this book a curve is allowed to have multiple components, so in fact
the curves considered here are called cycles in a different terminology.
rank=dim+1
Introduction
Chapter 1
Background
5
5.1
Finite fields and polynomials

Some basic facts
Here the basic facts about finite fields are collected. For more see [90].
For any prime p and any positive integer h there exists a unique finite field (or
Galois field) GF(q) of size q = ph . The prime p is the characteristic of it, meaning
a + a + ... + a = 0 for any a GF(q) whenever the number of as in the sum is
(divisible by) p. The additive group of GF(q) is elementary abelian, i.e. (Zp , +)h
while the non-zero elements form a cyclic multiplicative group GF(q) ' Zq1 , any
generating element (often denoted by ) of it is called a primitive element of the
field.
For any a GF(q) aq = a holds, so the field elements are precisely the roots of
X q X. (Lucas theorem implies, see below, that) we have (a + b)p = ap + bp for
any a, b GF(q), so x 7 xp is a field automorphism. GF(q) has a (unique) subfield
GF(pt ) for each t|h; GF(q) is an ht -dimensional vectorspace over its subfield GF(pt ).
i
The (Frobenius-) automorphisms of GF(q) are x 7 xp for i = 0, 1, ..., h 1,
t
forming the complete, cyclic automorphism group of order h. Hence x 7 xp fixes
the subfield GF(pgcd(t,h) ) pointwise (and all the subfields setwise!); equivalently,
t
(X p X)|(X q X) iff t|h.
P
One can see that for any k not divisible by (q 1), aGF(q) ak = 0. From this,
P
k
if f : GF(q) GF(q) is a bijective function then
xGF(q) f (x) = 0 for all
k = 1, ..., q 2. See also Dicksons theorem.

We will frequently use Lucas theorem when calculating binomial coefficients nk
in finite characteristic, so modulo p: let n = n0 + n1 p + n2 p2 + ... + nt pt , k =
5
so if a 6= 0 then
aq1 = 1
and X q1 1
is the root
polynomial
of
GF(q)
write GF(q) =
{ 0 , 1 , ..., q2 }
and consider
Pq2 ik
=
i=0
k(q1) 1 .
k
1
Mn
I. Background

k0 + k1 p + k2 p2 + ... + kt pt , with 0 ni , ki p 1, then nk nk00 nk11 ... nktt
(mod
p). In particular, in most cases we are interested in those values of k when
n
k is non-zero in GF(q), so modulo p. By Lucas theorem, they are precisely the
elements of
Mn = {k = k0 + k1 p + k2 p2 + ... + kt pt : 0 ki ni }.
Exercise 5.1. Prove Lucas theorem!
2
We define the trace and norm functions on GF(q n ) as Trqn q (X) = X +X q +X q +

n1
2
n1
... + X q
and Normqn q (X) = XX q X q ...X q , so the sum and the product of
all conjugates of the argument. Both maps GF(q n ) onto GF(q), the trace function
is GF(q)-linear while the norm function is multiplicative.
Exercise 5.2. Show that Tr and Norm are in some sense unique, i.e. any GF(q)linear function mapping GF(q n ) onto GF(q) can be written in the form Trqn q (aX)
with a suitable a GF(q n ) and any multiplicative function mapping GF(q n ) onto
GF(q) can be written in the form Normqn q (X a ) with a suitable integer a. (See
Exercise 5.27 and also Section 8.1 for Tr).
5.2
Self-dual normal bases
GF(q n ) is a vector-space over its subfield GF(q), so several bases can be chosen.
2
n1
There are some natural choices like normal bases of form {, q , q , ..., q },
the conjugates of a certain primitive element. Another nice requirement is selfduality (with respect to the inner product ha, bi = Trqn q (ab)), i.e. the inner
product of two basis elements bi and bj is Trqn q (bi bj ) = 0 if i 6= j and 1 if i = j.
The following theorem answers the question whether self-dual normal bases exist.
Result 5.3. [89] For a prime power q and n > 1 a self-dual normal basis of GF(q n )
over GF(q) exists if and only if either n is odd, or n 2 (mod 4) and q is even.
5.3
Polynomials
Here we summarize some properties of polynomials over finite fields. Given

a field F, a polynomial f (X1 , X2 , ..., Xk ) is a finite sum of monomial terms
ai1 i2 ...ik X1i1 X2i2 Xkik , where each Xi is a free variable, ai1 i2 ...ik , the coefficient of
the term, is an element of F. The (total) degree of a monomial is i1 + i2 + ... + ik
if the coefficient is nonzero and otherwise. The (total) degree of f , denoted
by deg f or f , is the maximum of the degrees of its terms. These polynomials
form the ring F[X1 , X2 , ..., Xk ]. A polynomial is homogeneous if all terms have
the same total degree. If f is not homogeneous then one can homogenize it, i.e.
transform it to the following homogeneous form: Z deg f f ( XZ1 , XZ2 , ..., XZk ), which
is a polynomial again (Z is an additional free variable).
5. Finite fields and polynomials
P
Given f (X1 , ..., Xn ) =
ai1 ...in X1i1 Xnin F[X1 , ..., Xn ], and the elements
x1 , ..., xn F then one may substitute them into f : f (x1 , ..., xn ) =
P
ai1 ...in xi11 xinn F; (x1 , ..., xn ) is a root of f if f (x1 , ..., xn ) = 0.
A polynomial f may be written as a product of other polynomials, if not (except
the algebraic
in a trivial way) then f is irreducible. If we consider f over F,
closure of F, and it still cannot be written as a product of polynomials over F

2
then f is absolutely irreducible. E.g. X + 1 GF(3)[X] is irreducible but not
absolutely irreducible, it splits to (X + i)(X i) over GF(3) where i2 = 1.
X 2 + Y 2 + 1 GF(3)[X, Y ] is absolutely irreducible. Over the algebraic closure
every univariate polynomial splits into linear factors.
In particular, x is a root of f (X) (of multiplicity m, see below) if f (X) can be
written as f (X) = (X x)m g(X) for some polynomial g(X), m 1.
Over a field any polynomial can be written as a product of irreducible polynomials
(factors) in an essentially unique way (so apart from constants and rearrangement).
Let f : GF(q) GF(q) be a function. Then it can be represented by the linear
combination
X
x GF(q) f (x) =
f (a)a (x),
aGF(q)
where
this is Lagrange
interpolation
a (X) = 1 (X a)q1
is the characteristic function of the set {a}. In other terms it means that any
function can be given as a polynomial of degree q 1. As both the number of
functions GF(q) GF(q) and polynomials in GF(q)[X] of degree q 1 is q q , this
representation is unique.
Let now f GF(q)[X]. Then f , as a function, can be represented by a polynomial
f of degree at most q 1, this is called the reduced form of f . The degree of f will
be called the reduced degree of f .
both
a
vectorspace
of
dim = q over
GF(q)
multiplicity of
a
root
may
change
when
reducing f
Proposition 5.4. For any (reduced) polynomial f (X) = cq1 X q1 + ... + c0 ,

X
xk f (x) = cq1k0 ,
xGF(q)
where k = t(q 1) + k0 , 0 k0 q 2. In particular,
f (x) = cq1 .
xGF(q)
Proof:
xk f (x) =
P Pq1
x
i=0
ci xi+k =
Pq1
i=0
ci
xi+k = cq1k0 .
Exercise 5.5. If f is bijective (permutation polynomial) then the reduced degree of

f k is at most q 2 for k = 1, ..., q 2.

We note that (i) if p 6 |{t : f (t) = 0}| then the converse is true;
see also Exercise 9.13, which

is Dicksons
I. Background
(ii) it is enough to assume it for the values k 6 0 (mod p).

Exercise 5.6. If q = 2h then the quadratic equation aX 2 + bX + c = 0 has no
= 1.
solution iff Tr2h 2 ac
b2
Lets examine GF(q)[X] as a vector space over GF(q).
Exercise 5.7. G
acs [67] For any subspace V of GF(q)[X], dim(V ) = |{deg(f ) : f
V }|.
q = ph
Exercise 5.8. [56] Consider the space S spanned as S = h(X a) : a GF(q)i.

Ph1
Ph1
Write = i=0 i pi , where 0 i < p and let M = { i=0 i pi : 0 i i }.
Then S has a basis {X r : r M }.
In several situations we will be interested in the zeros of (uni- or multivariate)
polynomials. Let a = (a1 , a2 , ..., an ) be in GF(q)n . We shall refer to a as a point
in the n-dimensional vector
P space V(n, q) or affine space AG(n, q). Consider an f
in GF(q)[X1 , ..., Xn ], f = i1 ,i2 ,...,in X1i1 Xnin .
def
multa f = m
m 1
f (a)=fa (0)=0
We want to define the multiplicity of f at a. It is easy if a = 0 = (0, 0, ..., 0). Let

m be the largest integer such that for every 0 i1 , ..., in , i1 + ... + in < m we have
i1 ,i2 ,...,in = 0. Then we say that f has a zero at 0 with multiplicity m.
For general a one can consider the suitable translate of f , i.e. fa (Y1 , ..., Yn ) =
f (Y1 + a1 , Y2 + a2 , ..., Yn + an ), and we say that f has a zero at a with multiplicity
m if and only if fa has a zero at 0 with multiplicity m.
5.4
Differentiating polynomials
Pn
0
Given a polynomial f (X) = i=0 P
=
ai X i , one can define its derivative X f = fX
n
0
0
i1
f in the following way: f (X) = i=0 iai X . Note that if the characteristic p
divides i then the term iai X i1 vanishes; in particular deg f 0 < deg f 1 may
i
f or f (i) or f 00 , f 000 etc. If a is a
occur. Multiple differentiation is denoted by X
root of f with multiplicity m then a will be a root of f 0 with multiplicity at least
m 1, and of multiplicity at least m iff p|m. Also if k p then a is root of f with
multiplicity at least k iff f (i) (a) = 0 for i = 0, 1, ..., k 1.
We will use the differential operator = (X , Y , Z ) (when we have three varii
i
i
ables) and maybe i = (X
, Yi , Zi ) and probably iH = (HX
, HYi , HZ
), where
7.3).
The
only
properties
Hi stands for the i-th Hasse-derivation
operator
(see

we need are that Hj X k = kj X kj if k j (otherwise 0); Hj is a linear opPj
erator; Hj (f g) = i=0 Hi f Hji g; a is root of f with multiplicity
at least k iff
i+j
Hi f (a) = 0 for i = 0, 1, ..., k 1; and finally Hi Hj = i+j
H
.
i
We might use the following differential equation:

V F = XX F + Y Y F + ZZ F = 0,
where F = F (X, Y, Z) is a homogeneous polynomial in three variables, of total
degree n. Let F (X, Y, Z, ) = F (X, Y, Z) = n F (X, Y, Z), then
nn1 F (X, Y, Z) = ( F )(X, Y, Z, ) = (XX F + Y Y F + ZZ F )(X, Y, Z).
It means that if we consider V F = 0 as a polynomial equation then
( F )(X, Y, Z, ) = 0 identically, which holds if and only if p divides n = deg(F ).
If we consider our equation as (V F )(x, y, z) = 0 for all (x, y, z) GF(q)3 , and
deg(F ) is not divisible by p, then the condition is that F (x, y, z) = 0 for every
choice of (x, y, z), i.e. F hY q Z Y Z q , Z q X ZX q , X q Y XY q i, see later.
5.5
Polynomials vanishing at many points, Alons Combinatorial

Nullstellensatz and the Ball-Serra refinement
We will quite often get into a situation when our polynomials have many roots,
sometimes they vanish almost everywhere in their domain. It was Bruen and later
Alon who started to explore this situation; here we show some results of this kind.
An incredible number of wonderful applications were found later, we will see some
of the geometrical ones and in the last section some more non-geometrical ones
as well. Recently Ball and Serra achieved a new improvement on Alons Nullstellensatz, (which promises new applications, in fact they found some already), we
show that as well. Therefore this Section is essentially based on [1] and [22].
Exercise 5.9. Let S be a subset of GF(q)2 and f GF(q)[X, Y ], such that f (aY +
b, Y ) = 0, for all (a, b) S. If |S| > deg(f ) then f (X, Y ) 0.
Exercise 5.10.
Show that G(x1 , ..., xn ) = 0 for all (x1 , ..., xn ) GF(q)n if and
only if G is of form g1 (X1 , ..., Xn )(X1q X1 )+...+gn (X1 , ..., Xn )(Xnq Xn ).
Similarly show that G(x1 , ..., xn ) = 0 for all (x1 , ..., xn ) (GF(q) )n if and
only if G h(X1q1 1), ..., (Xnq1 1)i.
Show that all (x1 , ..., xn ) GF(q)n are t-fold zeros of G(X1 , ..., Xn ) if and
only if G is an element of the ideal
Jt = Jt (X1 , ..., Xn ) = h(X1q X1 )i1 (X2q X2 )i2 ...(Xnq Xn )in : i1 +i2 +...+in = ti.
Similarly show that all (x1 , ..., xn ) (GF(q) )n are t-fold zeros of
G(X1 , ..., Xn ) iff G h(X1q1 1)i1 (X2q1 1)i2 ...(Xnq1 1)in : i1 +
i2 + ... + in = ti.
Theorem 5.11. Let f GF(q)[X1 , ..., Xn ] satisfy f (0, 0, ..., 0) 6= 0 and f (a) = 0 for
all a 6= 0. Then deg(f ) n(q 1).
Jt
10
I. Background
Proof Let f (0) = 6= 0 and g(X1 , ..., Xn ) =

vanishes everywhere hence it is in J1 .
a generalization
Qn
i=1 (1
Xiq1 ). Then
1
f
Exercise 5.12. Let t be a positive integer, and let the polynomial f

GF(q)[X1 , ..., Xn ] satisfy f (0, 0, ..., 0) 6= 0 and multa f t for all a 6= 0. Then
deg(f ) (n + t 1)(q 1).
Exercise 5.13. Give an example when in Exercise 5.12 deg(f ) = (n + t 1)(q 1)
holds, for n + t 1 q + 1.
Exercise 5.14. Suppose that the curves C
1 , ..., Cm cover AG(2, q) \ {(0, 0)} but neiP
ther of them contains the origin. Then i deg(Ci ) 2(q 1).
another
generalization
Theorem 5.15. Let A, B GF(q) . Let f (X, Y ) GF(q)[X, Y ] satisfy (i) f (0, 0) 6=
0 and (ii) f (a, b) = 0 whenever a A or b B. Then f can be written as f = g+h,
where h J1 (X, Y ), deg(g) deg(f ) and deg(f ) |A| + |B|.
Proof: Lets reduce f modulo (X q X) and modulo (Y q Y ), this gives f = g+h
with degX (g) q 1, degY (g) q P
1, h J1 (X, Y ), deg(h) deg(f ) and clearly
deg(f ) deg(g). Write g(X, Y ) = ci (Y )X i .
As g(x, b) = 0 for any x GF(q), b B andQdegX (g), degY (g) q 1, it follows
that g(X, b) = 0, so each ci (b) = 0. Hence bB (Y b) divides g, and similarly
Q
aA (X a) divides g.
Note that it gives another proof of Theorem 5.11 for n = 2, with A = B = GF(q) .
In applications we often need the homogeneous version of Exercise 5.10. Note
that e.g. X q Y XY q = Y (X q X) X(Y q Y ). If f (X, Y ), a homogeneous
polynomial in X and Y , of total degree d, vanishes everywhere on GF(q)2 then
f h(X q X), (Y q Y )i, so f (X, Y ) = f1 (X, Y )(X q X) + f2 (X, Y )(Y q Y ),
where f1 and f2 can be chosen to be homogeneous polynomials of total degree
d q. Now the terms of low degree must disappear, so Xf1 + Y f2 = 0. Hence Y
divides f1 ; let f1 (X, Y ) = Y g(X, Y ), then
f (X, Y ) = g(X, Y )(X q Y XY q ).
Alons Nullstellensatz
Hilberts Nullstellensatz is the fundamental theorem stating that if F is an algebraically closed field, and f ; g1 , ..., gm are polynomials in F[X1 , ..., Xn ], where f
vanishes over all common zeros of g1 , ..., gm , then
Pm there is an integer k and polynoIn the special case m = n,
mials h1 , ..., hm in F[X1 , ..., Xn ] so that f k = i=1 hi gi .Q
where each gi is a univariate polynomial of the form sSi (Xi s), a stronger
conclusion holds, by Alons result [2].
11
Theorem 5.16. Let F be an arbitrary field, and let f = f (X1 , ..., Xn ) be a

polynomialQin F[X1 , ..., Xn ]. Let S1 , ..., Sn be nonempty subsets of F and define
gi (Xi ) = sSi (Xi s). If f vanishes over all the common zeros of g1 , ..., gn
(that is; if f (s1 , ..., sn ) = 0 for all si Si ), then f hg1 , ..., gn i and there are
polynomials h1 , ..., hn F[X1 , ..., Xn ] satisfying deg(hi ) deg(f ) deg(gi ) so that
f=
n
X
hi gi .
i=1
Moreover, if f ; g1 , ..., gn lie in R[X1 , ..., Xn ] for some subring R of F then there
are polynomials hi R[X1 , ..., Xn ] as above.
As a consequence of the above one can prove the following:
Theorem 5.17. Let F be an arbitrary field, and P
let f = f (X1 , ..., Xn ) be a polyn
nomial in F[X1 , ..., Xn ]. Suppose that deg(f ) =Q i=1 ti , where each ti is a nonn
negative integer, and suppose the coefficient of i=1 Xiti in f is nonzero. Then, if
S1 , ..., Sn are subsets of F with |Si | > ti , there are s1 S1 , s2 S2 , ..., sn Sn so
that f (s1 , ..., sn ) 6= 0.
To prove Theorem 5.16 we need the following lemma.
Lemma 5.18. Let P = P (X1 , ..., Xn ) be a polynomial in n variables over an arbitrary field F. Suppose that the degree of P as a polynomial in Xi is at most ti
for 1 i n, and let Si F be a set of at least ti + 1 distinct members of F. If
P (X1 , ..., Xn ) = 0 for all n-tuples (x1 , ..., xn ) S1 ... Sn then P 0.
Proof: We apply induction on n. For n = 1, the lemma is simply the assertion
that a non-zero polynomial of degree t1 in one variable can have at most t1 distinct
zeros. Assuming that the lemma holds for n 1, we prove it for n (n 2). Given a
polynomial P = P (X1 , ..., Xn ) and sets Si satisfying the hypotheses of the lemma,
let us write P as a polynomial in Xn ; that is,
P =
tn
X
Pi (X1 , ..., Xn1 )Xni ,
i=0
where each Pi is a polynomial with Xj -degree bounded by tj . For each fixed (n1)tuple (x1 , ..., xn1 ) S1 ...Sn1 , the univariate polynomial P (x1 , ..., xn1 , Xn )
vanishes for all xn Sn , and is thus identically 0. Thus Pi (x1 , ..., xn1 ) = 0 for
all (x1 , ..., xn1 ) S1 ... Sn1 . Hence, by the induction hypothesis, Pi 0 for
all i, implying that P 0.
Proof of Theorem 5.16 Define ti = |Si | 1 for all i. By assumption,
f (x1 , ..., xn ) = 0
(x1 , ..., xn ) S1 ... Sn
(1)
12
I. Background
For each i, 1 i n, let

gi (Xi ) =
(Xi s) = Xiti +1
sSi
ti
X
gij Xij .
j=0
Pti
(2)
gij xji .
Observe that, if xi Si then gi (xi ) = 0; that is, xtii +1 = j=0
Let f be the polynomial obtained by writing f as a linear combination of monomials and replacing, repeatedly, each occurrence of Xiui (1 i n), where ui > ti , by
a linear combination of smaller powers of Xi , using the relations (2). The resulting
polynomial f is of degree at most ti in Xi , for each 1 i n, and is obtained from
f by subtracting from it products of the form hi gi , where the degree of each polynomial hi F[X1 , ..., Xn ] does not exceed deg(f )deg(gi ) (and where the coefficients
of each hi are in the smallest ring containing all coefficients of f and g1 , ..., gn .)
Moreover, f(x1 , ..., xn ) = f (x1 , ..., xn ), for all (x1 , ..., xn ) S1 ... Sn , since the
relations (2) hold for these values of x1 , ..., xn . Therefore, by (1), f(x1 , ..., xn ) = 0
for every n-tuple P
(x1 , ..., xn ) S1 ... Sn and hence, by the Lemma, f 0. This
n
implies that f = i=1 hi gi , and completes the proof.
Proof of Theorem 5.17 We may assume
Q that |Si | = ti + 1 for all i. Suppose the
statement is false, and define gi (Xi ) = sSi (Xi s). By Theorem 5.16 there are
Pn
polynomials
j)
i=1 ti deg(gj ) so
Pnh1 , ..., hn F[X1 , ..., Xn ] satisfying deg(hQ
n
that f = i=1 hi gi . By assumption, the coefficient of i=1 Xiti on the left hand
side is nonzero, and hence so is the coefficient
of this monomial on the right hand
Q
side. However, the degree of hi gi = hi sSi (Xi s) is at most deg(f ), and if
there are any monomials of degree
) in it they are divisible by Xiti +1 . It
Qn deg(f
ti
follows that the coefficient of i=1 Xi on the right hand side is zero, and this
contradiction completes the proof.
For nice applications see Section 29.
Now we present the punctured version of Alons Nullstellensatz by Ball and Serra
[22], which states that if f vanishes at nearly all, but not all, of the common zeros
of some polynomials g1 (X1 ), ..., gn (Xn ) then every I-residue of f , where the ideal
I = hg1 , ..., gn i, has a large degree. As a consequence we prove a converse of the
corollary to Alons Nullstellensatz. The corollary to Alons Nullstellensatz states
that if f has a term of maximum degree X1r1 ...Xnrn , where ri = |Si | ti and ti > 0
for all i, then a grid D1 ...Dn containing the points of the grid S1 ...Sn where
f does not vanish, satisfies |Di | ti . The converse, which will follow as a corollary
to the punctured version of Alons Nullstellensatz, states that if D1 ... Dn is
a grid containing the points of the grid S1 ... Sn where f does not vanish,
then f has a term X1r1 ...Xnrn , where ri satisfies |Si | 1 ri |Si | |Di | for all
i. Furthermore, Ball and Serra extend Alons Nullstellensatz to functions which
13
have multiple zeros at the common zeros of g1 , g2 , ..., gn and prove a punctured
version of this generalised version.
The following corollary is slightly more general than Theorem 5.17. Note that
under the hypothesis there is always at least one point of the grid where f does
not vanish.
Corollary 5.19. If f F[X1 , X2 , ..., Xn ] has a term of maximum degree X1r1 ...Xnrn ,
where ri = |Si | ti and ti 1 for all i, then a grid which contains the points of
S1 ... Sn where f does not vanish, has size at least t1 ... tn .
Proof: Suppose that there is a grid M1 ... Mn , where |Mj | < tj for some j,
containing all the points S1 ... Sn where f does not vanish. Let
Y
ej (Xj ) =
(Xj mj ).
mj Mj
The polynomial f ej is zero at all points of S1 ...Sn and has a term of maximum
|M |
degree X1r1 ...Xnrn Xj j . Note that rj + |Mj | < |Sj | and ri < |Si | for i 6= j. By
Pn
Theorem 5.16 the polynomial f ej = i=1 gi hi for some polynomials hi . The terms
of maximum degree in f ej have degree in Xi at least |Si | for some i, which is a
contradiction.
Punctured Combinatorial Nullstellensatz

In Alons Combinatorial Nullstellensatz, Theorem 5.16, the function f was assumed to have zeros at all points of the grid S1 S2 ... Sn . In the case that
there is a point in S1 S2 ... Sn where f does not vanish a slightly different
conclusion holds. The following can be thought of as a punctured version of Alons
Combinatorial Nullstellensatz.
Let F be a field and let f be a polynomial in F[X1 , X2 , ..., Xn ]. For i = 1, ..., n, let
Di and Si be finite non-empty subsets of F, where Di Si , and define
Y
Y
gi (Xi ) =
(Xi si ), and li (Xi ) =
(Xi di ).
si Si
di Di
Theorem 5.20. If f vanishes over all the common zeros of g1 , g2 , ..., gn except at
least one element of D1 D2 ... Dn , where it is not zero, then there are
polynomials h1 , h2 , ..., hn F[X1 , X2 , ..., Xn ] satisfying deg(hi ) deg(f ) deg(gi )
and a non-zero polynomial w, whose degree in Xi is less than |Si | and whose total
degree is at most deg f , with the property that
f=
n
X
i=1
hi gi + w,
14
I. Background
and
w=u
n
Y
gi
l
i=1 i
for some non-zero polynomial u.

Note that the theorem gives the lower bound deg f
Proof:
We can write
f=
n
X
Pn
i=1 (|Si |
|Di |).
gi hi + w,
i=1
for some polynomials hi of degree at most the degree of f minus the degree of gi ,
and a polynomial w, where the degree of w in Xi is less than the degree of gi and
the overall degree of w is at most the degree of f . For each i the polynomial f li
has zeros on all common zeros of gi , by assumption, and hence so does wli . By
Alons Nullstellensatz there are polynomials vi with the property that
wli =
n
X
gi vi .
i=1
However the degree of Xj in wli , for j 6= i, is less than the degree of gj (Xj ) and so
wli = gi vi . Thus gi divides wli . Note that li divides gi , so this divisibility implies
gi /li divides w. Hence
n
Y
gi
w=u
l
i=1 i
for some polynomial u and u is not zero since 0 6= f (d1 , d2 , ..., dn ) =
w(d1 , d2 , ..., dn ) for some di Di .
The following corollary is a converse of the corollary to Alons Nullstellensatz,
Corollary 5.19.
Exercise 5.21. If D1 ... Dn is a grid containing all the points of the grid
S1 ...Sn where f does not vanish, then f has a term X1r1 ...Xnrn , where |Si |1
ri |Si | |Di |.
Combinatorial Nullstellensatzen with multiplicity
In this section we shall consider polynomials that have zeros of multiplicity, see
the end of Section 5.3. The following proof of Theorem 5.22 is based on the proof
of Theorem 1.3 in [53].
Let T be the set of all non-decreasing sequences of length t on the set {1, 2, ..., n}.
For any T , let (i) denote the i-th element in the sequence .
15
Let F be a field and let f be a polynomial in F[X1 , X2 , ..., Xn ]. Suppose that

S1 , S2 , ..., Sn are arbitrary non-empty finite subsets of F and define
Y
gi (Xi ) =
(Xi si ).
si Si
Theorem 5.22. If f has a zero of multiplicity t at all the common zeros

of g1 , g2 , ..., gn thenPthere are polynomials h in F[X1 , X2 , ..., Xn ], satisfying
deg(h ) deg(f ) i deg(gi ), such that
X
f=
g (1) ...g (t) h .
T
Proof: This can be proved by double induction on n and t. If n = 1 and f has

a zero of degree t for all si Si then f = g(X1 )t h(X1 ) for some polynomial h.
If t = 1 then the theorem is Alons Nullstellensatz, Theorem 5.16. For the details
see [22].
Theorem 5.22 has the following corollary.
Exercise 5.23. Let F be a field and let f be a polynomial in F[X1 , X2 , ..., Xn ] and
suppose that f has a term X1r1 X2r2 ...Xnrn of maximal degree. If S1 , S2 , ..., Sn are
nonempty subsets of F with the property that for all T , the set of non-decreasing
sequences of length t on {1, 2, ..., n}, there exists an i such that
X
|Si | > ri ,
i
then there is a point a = (a1 , a2 , ..., an ), with ai Si , where f has a zero of

multiplicity at most t 1.
Note that this statement with t = 1 is the original corollary to Alons Nullstellensatz that has proven so useful. Specifically, if ri < |Si | for all i then there is a point
(a1 , a2 , ..., an ), with ai Si , where f does not vanish. When t = 2 the hypothesis
is that ri < |Si | for all i except possibly one, when i = 1 say, for which r1 < 2|S1 |.
The conclusion is that there is a point (a1 , a2 , ..., an ), with ai Si , where f has a
simple zero (of multiplicity one) or does not vanish. When t = 3 the hypothesis is
that ri < |Si | for all i except possibly two values, 1 and 2 say, for which r1 < 3|S1 |
and r2 < 2|S2 |.
The following is a version of Theorem 5.20 for functions that have many zeros
with multiplicity.
Let F be a field and let f be a polynomial in F[X1 , X2 , ..., Xn ]. For i = 1, ..., n, let
Di and Si be finite non-empty subsets of F, where Di Si , and define
Y
Y
gi (Xi ) =
(Xi si ), and li (Xi ) =
(Xi di ).
si Si
di Di
16
I. Background
Theorem 5.24. If f has a zero of multiplicity at least t at all the common zeros
of g1 , g2 , ..., gn , except at at least one point of D1 D2 ... Dn where it has a
zero of multiplicity less than t,
Pthen there are polynomials h in F[X1 , X2 , ..., Xn ],
satisfying deg(hi ) Pdeg(f ) i deg(gi ), and a non-zero polynomial u satisfying
n
deg(u) deg(f ) i=1 (deg(gi ) deg(li )), such that
X
f=
g (1) ...g (t) h + u
n
Y
gi
l
i=1 i
Moreover, if there is a point of D1 D2 ... Dn where f is non-zero, then for

any j,
n
X
deg(f ) (t 1)(|Sj | |Dj |) +
(|Si | |Di |).
i=1
5.6
Additive, linearized polynomials
A polynomial f GF(q)[X] is additive if f (U +V ) = f (U )+f (V ) or, alternatively,

f (x + y) = f (x) + f (y) for all x, y GF(q). It is GF(pe )-linear if it is additive and
for all GF(pe ) GF(q), f (X) = f (X) or f (x) = f (x) for each x GF(q).
If all the exponents in f are powers of pe then f is obviously GF(pe )-linear.
We remark that GF(p)-linearity and additivity are equivalent.
Note that the polynomial X 6 + X 4 + X 2 + X GF(3)[X] is additive (GF(3)-linear)
but the exponents are not powers of 3. This cannot occur if the degree is at most
q 1.
Theorem 5.25. A polynomial f GF(q)[X] of degree at most q 1 is GF(pe )-linear
if and only if all of its exponents are powers of pe .
Proof: The if part is obvious. Consider GF(q) as an he -dimensional vectorspace
over GF(pe ), then f is a GF(pe )-linear transformation of it. The number of such
h 2
transformations is (pe )( e ) . The polynomials of degree at most q 1 and with
all exponents being powers of pe are among these transformations, the number of
h 2
h
these polynomials is q e = (pe )( e ) .
Exercise 5.26. (Segre-Bartocci) A linear map f : GF(q) GF(q) is nonsingular if

Ph1
i
and only if for the coefficients of its unique polynomial form f (X) = i=0 ai X p ,
a0
a1
... ah1
ap
h1
the matrix A =
2
ap
h2
ap
0
2
ap
h1
...
ap
h2
...
ap
h3
...
ap
0
.
.
.
h1
a1p
h1
a2p
h1
is nonsingular.
17
Note that the determinant of the Segre-Bartocci matrix is always an element of

GF(p).
Exercise 5.27. A polynomial f GF(q)[X] of degree q 1 maps to the subfield
GF(pe ) and is GF(pe )-linear if and only if it is of the form f (X) = Trqpe (aX)
with some a GF(q).
Theorem 5.28. Fundamental Theorem of Additive Polynomials. Let f (X) F[X]
be a separable polynomial with roots {x1 , ..., xm }. Then f (X) is additive if and
only if S = {x1 , ..., xm } is a(n additive) subgroup of (F, +).
Proof:
sider
The additive subgroup direction is obvious. To show the other, conf (X) =
m
Y
(X xi ).
i=1
If y S then f (X +s) = f (X). Now for any y put g(X) = f (X +y)f (X)f (y).
Here deg g < deg f = m and g(xi ) = 0 for i = 1, ..., m, hence g(X) = 0 and f is
additive.
5.7
The random-like behaviour
In this section we prove a lemma, which is a generalization of a result of Szonyi.

It is interesting in itself, and it is used in several applications as well. In fact this
lemma is a consequence of the character sum version of Weils estimate. In order
to formulate it, we need
Definition 5.29. Let f1 (X), ..., fm (X) GF(q)[X] be given polynomials. We say
s
that their system is dpower independent, if no partial product fis11 fis22 ...fijj (1
j m; 1 i1 < i2 < ... < ij m; 1 s1 , s2 , ..., sj d 1) can be written as a
constant multiple of a d-th power of a polynomial.
s
Equivalently, one may say that if any product fis11 fis22 ...fijj is a constant multiple of
a d-th power of a polynomial, then this product is trivial, i.e. for all the exponents
d|si , i = 1, ..., j. Now
Lemma 5.30. Let f1 , ..., fm GF(q)[X] be a set of dpower independent polynomials, where d|(q 1), d, m 2. Denote by N the number of solutions
{x
GF(q) : fi (x) is a d-th power in GF(q) for all i = 1, ..., m}. Then |N dqm |
Pm
q i=1 deg fi .
Note that this lemma implies that, under some natural conditions, one can solve
a system of equations
d (fi (X)) = i
(i = 1, ..., m),
separable: splits
into deg(f ) distinct linear factors
18
I. Background
where the i -s are d-th complex roots of unity, and d is a multiplicative character
of order d. So the d-th power behaviour can be prescribed if the polynomials are
independent.
It can be interpreted as being a d-th power is like a random event of probability
1
d.
Some words about the condition d|(q 1): if d and q 1 are co-primes, then every
element is a d-th power in GF(q). If g.c.d.(d, q 1) = d1 and we write d = d1 d2
and g.c.d.(d2 , q 1) =g.c.d.(d1 , d2 ) = 1, then the lemma can be applied with d1 ,
as d-th and d1 -th powers are the same in this case.
We remark that Sz
onyi [126] proved this lemma for d = 2. [114] contains a general
bound; the proof below is a modified version of a lemma for linear polynomials in
Babai, G
al and Wigderson [5].
We need the character sum version of Weils estimate:
Result 5.31. ([90], Thm. 5.41) Let f (X) be a polynomial over GF(q) and r the
number of distinct roots of f in its splitting field. If e is a multiplicative character
(of order e) of GF(q) and f (X) 6= cg(X)e , then
X
|
e (f (x)) | (r 1) q.
xGF(q)
q1
Proof of Lemma 5.30: First note that we use the definition (x) = d (x) = x d .
Let {0 = 1, 1 , 2 , ..., d1 } be the set of d-th complex roots of unity. Let h(Z) =
Z d 1
d1
Z1 = 1 + Z + ... + Z Q ; then h(1) = d, h(j ) = 0 for j = 1, ..., d 1 and
m
h(0) = 1. Define H(x) = i=1 h((fi (x))).
If x is a solution then H(x) = dm , if x is a root of some fi then H(x) = 0 or
H(x) = dm1 . In the remaining
cases H(x) = 0. Hence,
Pm
P if N denotes the number
of solutions and D := i=1 deg fi then the sum S = xGF(q) H(x) satisfies
N dm S N dm + Ddm1 .
()
H(x) is a product of sums of d terms each. Lets expand the product to the sum of
dm terms. Let denote the set of the dm functions : {1, ..., m} {0, ..., d 1},
which will serve to index this sum. Now
S=
m
XY
((fi (x)))(i) =
xGF(q) i=1
so x f (x) = 1
0
(f (x)),
xGF(q)
Qm
def
where (f (x) = i=1 fi (x)(i) . Let 0 (i) = 0 for all i and = \ {0 }. After
switching the order of summation and separating the term corresponding to 0 ,
this (main) term will be q. For the error term R = S q we have

X X

|R|
(f (x)).

xGF(q)
19
By our assumption for the d-th power independence of {f1 , ..., fm } and as each (i)
is at most d 1, we can use Theorem 5.31 (note that f have at most D distinct
roots in the splitting field), so the inner sum has absolute value (D1) q. Hence
|R| dm (D 1) q and from () we get |N dqm | (D 1) q + D/d D q
(as if D/d > q then Lemma 5.30 clearly holds).

The bounds in Result 5.31 can be strengthened when f (X) and are quadratic.
Exercise 5.32. Let q be an odd prime power, f (X) = aX 2 + bX + c GF(q)[X]
with a 6= 0, and let denote the quadratic character GF(q) {1, 1, 0}. Then

X
(a),
if b2 4ac 6= 0;
2
(ax + bx + c) =
(q 1)(a), if b2 4ac = 0.
xGF(q)
5.8
Lacunary polynomials
A polynomial is called lacunary if an interval is missing from its terms, i.e. some
consecutive coefficients happen to be zero. A polynomial of GF(q)[X] is fully reducible if it splits to linear factors over GF(q). These two requirements proved to
be hard to satisfy simultaneously, there are a few results stating that fully reducible lacunary polynomials (the extent of lacunarity must be specified) should
have some particular feature.
The theory of lacunary polynomials is treated in Redeis book [103], where the
following two problems are considered:
dei Let d|q 1, d > 1. Determine the fully
Problem 5.33. Problem 1 of Re
q1
reducible polynomials f (X) = X d + g(X) for which deg g q1
d2 , X 6 |g, and f
has no multiple factors.
and
a
third
problem
as
a
tool for Problem
2
dei Determine the fully reducible polynomials

Problem 5.34. Problem 2 of Re
f (X) = X q + g(X), f (X) GF(q)[X] \ GF(q)[X p ], for which deg g q+1
2 .
In geometric applications a modified version of Problem 2 will be crucial, where f
is allowed to be a polynomial of X p :
it is meaningful
if only q 6= p
Problem 5.35. Problem 2 Determine the fully reducible polynomials f (X) =

X q + g(X), f (X) GF(q)[X], for which deg g q+1
2 .
Problem 1 is rather easy:
Theorem 5.36. (Redei, [103]) If d > 2 then the only solutions of Problem 1 are the
q1
q1
Euler-binoms X d , where = u d for a u GF(q) . If d = 2 and q 1
(mod 4) then there are other solutions as well, namely the four polynomials
so if 1
square
is
20
I. Background

q1
4
q1

1 X 4 ,
where 2 = 1.
Lets consider Problem 2. The conditions in it do not seem natural. The following
lemma ([103], Satz 18) shows that the degree of g is always at least q+1
2 unless
f (X) = X q X. The degree of a polynomial h is denoted by h .
Lemma 5.37. Let s be a power of p with 1 s < q and suppose that
X q/s + g(X) GF(q)[X] \ GF(q)[X p ]
in particular s=1
q+1
gives g
2
is fully reducible over GF(q). Then either s = 1 and g(X) = X or

g
q+s
.
s(s + 1)
Proof: A zero of X q/s + g of multiplicity k is a zero of (X q/s + g)0 = g 0 of

multiplicity at least k 1 and a zero of X q X with multiplicity 1. Hence
X q/s + g | ((X q/s + g)s (X q X))g 0 = (g s + X)g 0 .
By assumption g 0 6= 0 so either g s = X and hence s = 1 or the right hand side
is non-zero and q/s sg + g 1.
At this stage one can solve Problem 2 in the q = p prime case.
Theorem 5.38. [103] If q = p 6= 2 is a prime then the solutions of Problem 2 are
precisely the four polynomials

p1
p1
f (X) = (X + a) (X + a) 2 (X + a) 2 , = 1, = 0, 1.
It means that either f has one single and p1
2 double roots, or it has one root of
p1
multiplicity p+1
and
other
single
roots.
2
2
this translation
changes
a p+1
2
as well because
from X p we get
an extra ap = a
Proof: As the translation x 7 x+a does not change the problem, we can assume
p+1
p1
that in g(X) = a0 X 2 + a1 X 2 + ... + a p+1 the coefficient a1 is zero. From
2
the proof of Lemma 5.37 (s = 1) one can see that f (X)|(g(X) + X)g 0 (X); as
their degrees are equal, one is constant times the other. Comparing the leading
coefficients we get
a20
a2
f (X) = 0 (X p + g(X)) = (g(X) + X)g 0 (X).
2
2
In particular
()
21
(i) g(X) + X and g 0 (X) are fully reducible;

(ii) g(X) + X divides X p X hence its roots are all single.
p+3
a0 6= 0, a1 = 0
In () one can observe that the coefficients of X p1 , ..., X 2 are all zero. From
this, comparing coefficients we get that a2 = a3 = ... = a p+3 = 0, so g(X) =
p+1
p1
a0 X 2 + a p1 X + a p+1 . Now g 0 (X) = a20 X 2 + a p1 will be reducible only

2
2
2
if a p1 = a20 , 0. Now using full reducibility and comparing the coefficients the
2
statement of the theorem follows.
The following generalization of Lemma 5.37 is easy:
Exercise 5.39. [32] Suppose that f (X) = X q g(X) + h(X) GF(q)[X] (q = pn , p
prime) is fully reducible over GF(q), (g, h) = 1. Then either f (X) = a(X q X);
f (X) = aX q + b = (aX + b)q ; or q = p and max(g , h ) p+1
2 ; or n 2 and
[ n+1
]
max(g , h ) p 2 .
The best (currently available) result of this kind is the following (up to my knowledge):
Theorem 5.40. [39] Suppose that f (X) = X q g(X) + h(X) GF(q)[X] (q = pn , p
prime) is fully reducible over GF(q), (g, h) = 1. Let k = max(g , h ) < q. Let e be
maximal such that f is a pe -th power. Then we have exactly one of the following
cases:
(1) e = n and k = 0;
(2) e 2n/3 and k pe ;
(3) 2n/3 > e > n/2 and k pne/2 32 pne ;
(4) e = n/2 and k = pe and f (X) = aTrqq (bX + c) + d or f (X) =
aNormqq (bX + c) + d for suitable constants a, b, c, d, e;
l
m
p
(5) e = n/2 and k pe 14 + (pe + 1)/2 ;
(6) n/2 > e > n/3 and k pn/2+e/2 pne pe /2, or, if 3e = n + 1 and p 3
then k pe (pe + 1)/2;
(7) n/3 e > 0 and k pe d(pne + 1)/(pe + 1)e;
(8) e = 0 and k (q + 1)/2;
(9) e = 0, k = 1 and f (X) = a(X q X).
22
I. Background
The following lemma will be required, for example to prove Theorem 18.14.
Lemma 5.41. Let s be a power of p with 1 s < q and suppose that
f = X q/s + g GF(q)[X] \ GF(q)[X p ]
is fully reducible over GF(q). If 3(g 0 ) < 2q/s sg then
2
f h1, X, X s , X s , ..., X q/s iGF(q) .

Proof:
We have the same divisibility as in Lemma 5.37. Let m(X) be defined by

(X q/s + g)m = (X + g s )g 0 .
()
Now m = sg q/s+(g 0 ) . Differentiating (), we get another equation; combining

them we conclude to
(X q/s + g)(g 00 m g 0 m0 ) = (m 1)(g 0 )2 .
The degree of the right hand side is at most m + 2g 0 sg q/s + 3(g 0 ) < q/s,
so g 00 m g 0 m0 = 0, implying (m 1)g 0 = 0 and hence m = 1. It follows that g 0 is
constant and from the differential equation () it is an easy task to deduce that f
is in the required span.
Representations of affine and projective geometries
In order to profit from the Desarguesian structure of an affine or projective geometry, a few ways of representation or coordinatization is used. In each of them the
following questions should be answered:
How can the subspaces be seen or handled?
How can one calculate the intersection or span of two subspaces?
How can the subgeometries and other nice subsets, like quadrics, Hermitian surfaces, etc. be seen or handled?
How can transformations (collineations, polarities) be represented?
How can general pointsets be handled?
standard
representation
of AG(n, q)
The standard representation of AG(n, q) is just GF(q)n = V(n, q), the ndimensional vectorspace over GF(q), where the k-subspaces of AG(n, q) are represented by the k-dimensional affine subspaces (i.e. translates of linear subspaces)
of GF(q)n , for k = 0, 1, ..., n, and incidence is the natural one.
6. Representations of affine and projective geometries

standard
homogeneous
representation
of PG(n, q)
23
In the standard homogeneous representation of PG(n, q) the k-subspaces of

PG(n, q) are represented by the (k+1)-dimensional linear subspaces of GF(q)n+1 =
V(n + 1, q), for k = 0, 1, ..., n, and incidence is the natural one. In particular,
points can be represented by nonzero vectors (v1 , ..., vn+1 ) and another vector
(w1 , ..., wn+1 ) represents the same point if hvi = hwi, i.e. there is a GF(q) for
which wi = vi for i = 1, ..., n + 1.
When coordinatizing, an inner product should be chosen and used, in these two
representations the easiest is the standard inner product.
Hence a hyperplane, i.e. a 1-codimensional subspace can be represented by any
(nonzero) vector [v1 ,P
..., vn+1 ] being orthogonal to it(s points), i.e. for any point
u of the hyperplane
ui vi = 0; and another vector [w1 , ..., wn+1 ] represents the
same hyperplane if hvi = hwi.
Exercise 6.1. Calculate the number of k-dimensional subspaces of V(n, q), AG(n, q)

a
a1
1)...(q ab+1 1)
and PG(n, q). You may want to use the notation ab q = (q 1)(q
(q b 1)(q b1 1)...(q1)
i+1
and i = 1 q .
As AG(n, q) can be imagined as an n-dimensional vectorspace over GF(q), and so
is GF(q n ), it is quite natural to identify them. The incidence structure of AG(n, q)
will be recognised in the following way: three points A, B, C are collinear if for
the corresponding elements a, b, c of GF(q n ) there exists a GF(q) such that
c = (1 )a + b or equivalently (a b) = a c. Taking (q 1)th powers kills
the coefficient GF(q) and we get (a b)q1 = (a c)q1 . That is why we can
identify the directions with the n1 -th roots of unity in GF(q n ), which are the
(q 1)-st powers, and then one can say that A and B determine the direction
(a b)q1 .
q-binomials
affine
big field
representation
In this and in the next representation, when an inner product is needed, the usual
one is hx, yi = Trqn q (xy) (or Trqn+1 q (xy)); we will see that this is the most
natural one.
As PG(n, q) can be imagined as an (n + 1)-dimensional vectorspace over GF(q)
modulo multiplication by non-zero scalars, it is quite natural to identify it with
GF(q n+1 ) /GF(q) , i.e. the elements x and y of GF(q n+1 ) represent the same
point of PG(n, q) if and only if x = y for some GF(q). Note that in this
case xq1 = y q1 as q1 = 1 for all GF(q) ; this is again the common trick
to eliminate factors from a subfield. As usually the hyperplanes are represented
by the same set as the points (referring to the geometrically self-dual structure),
we will make it so. The incidence structure of PG(n, q) will be recognized in the
following way: the point A will be incident with the hyperplane H iff for the
corresponding elements a and h of GF(q n+1 ) we have Trqn+1 q (ah) = 0. For this
the most convenient way is the use of a trace-orthogonal normal base (if exists, as
for example in the planar case), see Section 5.2.
projective
big field
representation
24
I. Background
Suppose that is a primitive element of GF(q n+1 ), i.e.

GF(q n+1 ) = {1 = 0 , 1 , 2 , ..., q
cyclic projective
representation
It
is
common
to
imagine
PG(n, q) as the
hyperplane
at
infinity, i.e. set
of directions of
AG(n + 1, q).
n+1
}.
Then for the subfield GF(q) = {1 = 0 , n , 2n , ..., (q2)n }. Now one can
see what modulo GF(q) means: PG(n, q) is represented as the factor group of
these two cyclic groups; for calculations the convenient representation is either
{ 0 , 1 , 2 , ..., n 1 } or { 0 , q1 , 2(q1) , ..., (n 1)(q1) }, so points (and the
hyperplanes) are represented by the n -th roots of unity in GF(q n+1 ), which are
the (q 1)-st powers. In the first one the map i 7 i+1 (mod )n , in the second one the map i(q1) 7 (i+1)(q1) (mod q n+1 1) gives a cyclic (Singer-)
automorphism, being regular on both the points and the hyperplanes. This representation is very useful if the cyclic structure, or some very regular substructure
of the space is examined.
6.1
Subspaces, subgeometries
It is vital to be able to describe subspaces and subgeometries in affine and projective spaces. In the standard representations they are quite obvious however.
and = x1
x1 = x2 x2
In the affine big field representation the points x of a hyperplane are Trqn q (ax) +
b = 0, where a GF(q n ) and b GF(q). A suitable linear combination of k hyPnk+1
j
perplane polynomials gives an equation j=0 j X q + = 0 whose zeros correspond to an (nk +1)-dimensonal subspace, the intersection of the k hyperplanes.
In particular, lines are given by X q X + = 0, and for a line joining x1 and
x2 we have = (x1 x2 )q1 . The non-zero (q 1)-th powers are n1 -th roots
of unity in GF(q n ), so we see the one-to-one correspondence between the n1 -th
roots of unity in GF(q n ) and the n1 direction of lines in AG(n, q).
In the projective representation we have a slightly more complicated situation.
Given a set of indeterminates {Xi : i = 0, ..., n} the hyperplanes of PG(n, q) (i.e.
subspaces of V(n + 1, q) of rank n) are given by linear homogeneous equations of
the form
n
X
ci Xi = 0,
()
i=0
splitting poly of
GF(q n+1 )
observe the link

with the cyclic
representation!
where (c0 , c1 , ..., cn ) is a point of PG(n, q). The points of PG(n, q) are subspaces of
rank 1 in V(n+1, q) which in GF(q n+1 ) are given by the sets of zeros of equations of
n
n1
the form X q = uX where uq +q +...+q+1 = 1. This is a necessary and sufficient
n+1
condition on u for the polynomial X q uX to divide X q
X and hence be
a polynomial that splits completely into distinct linear factors over GF(q n+1 ).
Hence we got again that it makes sense to refer to the points of PG(n, q) as
(q n + q n1 + ... + q + 1)-st roots of unity in GF(q n+1 ). In GF(q n+1 ) the polynomial
i
Trqn+1 q ( q X) = q X + q
i+1
X q + ... + q
i+n1
Xq
n1
+ q
i+n
Xq
6. Representations of affine and projective geometries
25
splits completely into distinct linear factors over GF(q n+1 ), has degree q n and
is linear over GF(q). Hence we choose = to be a fixed primitive element
of GF(q n+1 ) and consider the hyperplane Xi = 0 of PG(n, q) as the equation
i
Trqn+1 q ( q X) = 0, over GF(q n+1 ), and in general the hyperplane (*) as the
equation
n
X

i
Trqn+1 q (
ci q ) X
= 0.
i=0
| {z }
c
In other words, the points x of a hyperplane are Trqn+1 q (cx) =

n
P
i
i
0 = cx
cq 1 xq 1 . Writing z = xq1 and a = cq1 we get
i=0
Pn
(q i 1)/(q1) (q i 1)/(q1)
z
.
i=0 a
A
suitable
linear
combination
of k hyperplane polynomials gives an equation
Pnk+1
j
+
=
0
whose
zeros correspond to an (n k + 1)-dimensional
Z
j
j=0
subspace, the intersection of the k hyperplanes. In particular lines are given by
Z q+1 Z + = 0 (where there exist relations between and depending on
the direction), and for a line joining z1 and z2 (viewed as (q 1)-th powers) we
have = (z1q+1 z2q+1 )/(z1 z2 ).
For example, the lines of PG(3, q) represented in GF(q 4 ) are obtained by looking
at the set of zeros of polynomials whose zeros are zeros of two such hyperplane
2
polynomials and we conclude that these have the form L(Z) := Z q + cZ q + eZ
for some c and e in GF(q 4 ). These polynomials must have q 2 distinct zeros in
4
2
2
2
2
GF(q 4 ) and hence divide Z q Z. The polynomial Lq cq Lq (eq cq +q )L
4
(mod Z q Z) has degree q and q 2 zeros and is therefore identically zero. Equating
coefficients gives the following necessary and sufficient conditions that
cq+1 = eq eq
6.2
+q+1
and eq
+q 2 +q+1
= 1.
()
Transformations
As is known, the automorphism group of PG(n, q) and AG(n, q) are PL(n + 1, q)

and AL(n, q), respectively. Here we describe them in the non-standard representations as well.
Theorem 6.2. Any automorphism : AG(n, q) AG(n, q), : x 7 (x), where
the point x is represented in the standard way, can be written in the form (x) =
M x + v, where M is a non-singular linear map , v V(n, q) gives a translation
and : GF(q) GF(q) is a field automorphism.
Theorem 6.3. Any automorphism : PG(n, q) PG(n, q), : x 7 (x), where
the point x is represented with homogeneous coordinates, can be written in the form
(x) = M x , where M is a non-singular linear map and : GF(q) GF(q) is a
field automorphism.
i.e.
M GL(n, q)
Frobeniusautomorphism,
possibly =id
i.e. M GL(n +
1, q)
Frobeniusautomorphism,
possibly =id
26
I. Background
So we have to understand what a linear transformation, a translation and a field

automorphism means in the big field representation r : V(n, q) GF(q n ).
Translation is easy: adding the vector v to each point of V(n, q) is the very same
as adding its representative r(v) to each element of GF(q n ).
Field automorphism is also easy: the automorphism of GF(q) is just among the
automorphisms of GF(q n ), and it commutes with r, i.e. it is the same to use it
in the big field representation.
The interesting one is the linear map. Let its matrix be M = (mij ), so M x =
Pn1
y, yi = j=0 mij xj . On the other hand, a polynomial f (X) = c0 X + c1 X q +
2
n1
c2 X q +...+cn1 X q , with c0 , ..., cq1 GF(q n ) defines a map GF(q n ) GF(q n ),

which is linear over GF(q). One can check that the number of matrices is equal
2
to the number of these GF(q)-linear polynomials (i.e. q n ). See Section 5.6 about
polynomials being linear over a subfield.
Given M , the corresponding f (X) depends on the base chosen in GF(q n ). If a
2
n1
base of the form {, q , q , ..., q } was chosen, then to find fM we have
Pn1
P
P
Pn1 Pn1
n1
n1
qk
qk
qk
r(M x) =
=
=
and
k=0 yk
k=0
j=0 mkj xj
j=0 xj
k=0 mkj
Pn1 Pn1
P
P
j+i
j+i
n1
n1
q
q
fM (r(x)) = i=0 ci j=0 xj
= j=0 xj i=0 ci
so to find the coefficients c0 , ..., cn1 we have to solve the linear equation
q0
q0
P
qk
q1
q2
q n1
1
q
2
q
2
q
3
q
n1
3
q
4
q
...
...
...
0
q
1
q
...
n2
c0
c1
c2
cn1
mk0
Pk
qk
k mk1
P
qk
k mk2
P
mk,n1 q
= M>
1
q
2
q
...
n1
q
The assumption that M is a non-singular linear map is equivalent to the following:

fM (X) has no root in GF(q n ) .
7
7.1
The Redei polynomial and its derivatives

The Redei polynomial
Generally speaking, a Redei polynomial is just a (usually multivariate) polynomial

which splits into linear factors. We use the name Redei polynomial to emphasize
that these are not only fully reducible polynomials, but each linear factor corresponds to a geometric object, usually a point or a hyperplane of an affine or
projective space.
Let S be a pointset of PG(n, q), S = {Pi = (ai , bi , ..., di ) : i = 1, ..., |S|}.
The (Redei-)factor corresponding to a point Pi = (ai , bi , ..., di ) is Pi V = ai X +
bi Y + ... + di T . This is simply the equation of hyperplanes passing through Pi .
When we decide to examine our pointset with polynomials, and if there is no
7. The Redei polynomial and its derivatives
27
special, distinguished point in S, it is quite natural to use symmetric polynomials

of the Redei-factors. The most popular one of these symmetric polynomials is the
Redei-polynomial, which is the product of the Redei-factors, and the *-polynomial,
which is the (q 1)-th power sum of them.
I have not found

any good name
for it yet
Definition 7.1. The Redei-polynomial of the pointset S is defined as follows:

S
R (X, Y, ..., T ) = R(X, Y, ..., T ) :=
|S|
Y
(ai X + bi Y + ... + di T ) =
i=1
|S|
Y
Pi V.
i=1
The points (x, y, ..., t) of R, i.e. the roots R(x, y, ..., t) = 0, correspond to hyperplanes (with the same (n+1)-tuple of coordinates) of the space. The multiplicity of
a point (x, y, ..., t) on R is m if and only if the corresponding hyperplane [x, y, ..., t]
intersects S in m points exactly.
Given two pointsets S1 and S2 , for their intersection

RS1 S2 (X, Y, ..., T ) = gcd RS1 (X, Y, ..., T ) , RS2 (X, Y, ..., T )
the
fundamental
property
of
R
edei
polynomials
holds, while for their union, if we allow multiple points or if S1 S2 = , we have

RS1 S2 (X, Y, ..., T ) = RS1 (X, Y, ..., T ) RS2 (X, Y, ..., T ).
Definition 7.2. The *-polynomial of S is
GS (X, Y, ..., T ) = G(X, Y, ..., T ) :=
|S|
X
(ai X + bi Y + ... + di T )q1 .
i=1
If a hyperplane [x, y, ..., t] intersects S in m points then the corresponding m terms

will vanish, hence G(x, y, ..., t) = |S|m (in other words, all m-secant hyperplanes
will be solutions of G(X, Y, ..., T ) |S| + m = 0).
The advantage of the *-polynomial (compared to the Redei-polynomial) is that it
is of lower degree if |S| q. The disadvantage is that while the Redei-polynomial
contains the complete information of the pointset (S can be reconstructed from
it), the *-polynomial of two different pointsets may coincide. This is a hard task
in general to classify all the pointsets belonging to one given *-polynomial.
The *-polynomial of the intersection of two pointsets does not seem to be easy
to calculate; the *-polynomial of the union of two pointsets is the sum of their
*-polynomials.
The next question is what happens if we transform S. Let M GL(n + 1, q) be a
linear transformation. Then
R
M (S)
(V) =
|S|
Y
i=1
(M Pi ) V =
|S|
Y
i=1
Pi (M > V) = RS (M > V).
modulo the characteristic
obviously
28
I. Background
For a field automorphism , R(S) (V) = (RS )() (V), which is the polynomial RS
but all coefficients are changed for their image under .
Similarly GM (S) (V) = GS (M > V) and G(S) (V) = (GS )() (V).
The following statement establishes a further connection between the Redei polynomial and the *-polynomial.
Lemma 7.3. (G
acs) For any set S,
RS (GS |S|) = (X q X)X RS + (Y q Y )Y RS + ... + (T q T )T RS .
In particular, RS (GS |S|) is zero for every substitution [x, y, ..., t].
Proof:
Trivial induction on |S|.
We remark that
X G(X, Y, ..., T ) =
|S|
X
ai (ai X + bi Y + ... + di T )q2 .
i=1
q2
Note that
= for all 0 6= GF(q). Compare this to the derivative of the
Redei polynomial, see below.
Next we shall deal with Redei-polynomials in the planar case n = 2. This case
is already complicated enough, it has some historical reason, and there are many
strong results based on algebraic curves coming from this planar case. Most of
the properties of Redei-surfaces in higher dimensions can be proved in a very
similar way, but it is much more difficult to gain useful information from them.
Let S be a pointset of PG(2, q). Let LX = [1, 0, 0] be the line {(0, y, z) : y, z
GF(q), (y, z) 6= (0, 0)}; LY = [0, 1, 0] and LZ = [0, 0, 1]. Let NX = |S LX | and
NY , NZ are defined similarly. Let S = {Pi = (ai , bi , ci ) : i = 1, ..., |S|}.
Definition 7.4. The Redei-polynomial of S is defined as follows:
coefficientpolynomials
R(X, Y, Z) =
|S|
Q
i=1
the
fundamental
property
of
R
edei
polynomials
(ai X + bi Y + ci Z) =
|S|
Q
Pi V = r0 (Y, Z)X |S| + r1 (Y, Z)X |S|1 + ... + r|S| (Y, Z).
i=1
For each j = 0, ..., |S|, rj (Y, Z) is a homogeneous polynomial in two variables,

either of total degree j precisely, or (for example when 0 j NX 1) rj
is identically zero. If R(X, Y, Z) is considered for a fixed (Y, Z) = (y, z) as a
polynomial of X, then we write Ry,z (X) (or just R(X, y, z)). We will say that R is
a curve in the dual plane, the points of which correspond to lines (with the same
triple of coordinates) of the original plane. The multiplicity of a point (x, y, z)
on R is m if and only if the corresponding line [x, y, z] intersects S in m points
exactly.
29
Remark 7.5. Note that if r = 1, i.e. [x, y, z] is a tangent line at some (at , bt , ct )
S, then R is smooth at (x, y, z) and its tangent at (x, y, z) coincides with the only
linear factor containing (x, y, z), which is at X + bt Y + ct Z.
Exercise 7.6. Let S be the pointset of the parabola X 2 Y Z in PG(2, q). Prove
q1
that GS (X, Y, Z) = X q1 if q is even and GS (X, Y, Z) = (X 2 4Y Z) 2 if q is
odd. What is the geometrical meaning of it?
Exercise 7.7. Let S be the pointset of the parabola X 2 Y Z in PG(2, q). Prove
that
Y
q1
q+1
RS (X, Y, Z) = Y
(tX + t2 Y + Z) = Y (Z q + Y q1 Z C q1 Y 2 Z 2
2
tGF(q)
C q3 X 2 Y
q3
2
where Ck =
q1
2
C q5 X 4 Y
2
1
k+1

2k
k
q5
2
q3
2
... C1 X q3 Y Z 2 C0 X q1 Z),
are the famous Catalan numbers.
Remark. If there exists a line skew to S then w.l.o.g. we can suppose that LX S =
and all ai = 1. If now the lines through (0, 0, 1) are not interesting for some
reason, we can substitute Z = 1 and now R is of form
R(X, Y ) =
|S|
Y
(X + bi Y + ci ) = X |S| + r1 (Y )X |S|1 + ... + r|S| (Y ).
i=1
This is the affine Redei polynomial. Its coefficient-polynomials are rj (Y ) =

j ({bi Y + ci : i = 1, ..., |S|}), elementary symmetric polynomials of the linear
terms bi Y + ci , each belonging to an affine point (bi , ci ). In fact, substituting
y GF(q), bi y + ci just defines the point (1, 0, bi y + ci ), which is the projection of
(1, bi , ci ) S from the center at infinity (0, 1, y) to the line (axis) [0, 1, 0].
One may ask what happens with the affine Redei polynomial if the pointset
{(bi , ci )} is translated by the vector (u, v) or enlarged by . Then RS+(u,v) (X, Y ) =
RS (X + uY + v, Y ) and RS (X, Y ) = |S| RS ( 1 X, Y ).
Exercise 7.8. Suppose that S = {(x, f (x)) : x GF(q)} AG(2, q), i.e. the graph
of a function Y = f (X), where f (X) = an X n + an1 X n1 + ... + a0 . Express the
Redei polynomial of S, or some coefficient if it, from the ai -s.
How can one see the points of S from its Redei polynomial?
1. They can be seen from the factors of R(X, Y, Z);
2. Let {1, , 2 } be a base of GF(q 3 ) over GF(q). Then if b + c 2 is a root of
R(X, , 2 ) then (1, b, c) is a point of S. (You cant see the points (0, b, c)
this way.)
30
I. Background
3. If S is an affine pointset then, using the affine Redei polynomial R(X, Y )
and the GF(q 2 )-representation of AG(2, q), with GF(q 2 ) \ GF(q), now
R(X, ) has roots exactly a + b where (a, b) are the points of S.
7.2
Differentiation in general
Here we want to introduce some general way of differentiation. Give each point
Pi the weight (Pi ) = i for i = 1, ..., |S|. Define the curve
R0 (X, Y, Z)
|S|
X
i=1
R(X, Y, Z)
.
ai X + bi Y + ci Z
()
If i = ai then R0 (X, Y, Z) = X R(X, Y, Z), and similarly, i = bi means Y R

and i = ci means Z R.
Theorem 7.9. Suppose that [x, y, z] is an m-secant with S [x, y, z]
{Pti (ati , bti , cti ) : i = 1, ..., m}.
(a) If m 2 then R0 (x, y, z) = 0. Moreover, (x, y, z) is a point of the curve R0

of multiplicity at least m 1.
(b) (x, y, z) is a point of the curve R0 of multiplicity at least m if and only if
for all the Ptj S [x, y, z] we have tj = 0.
(c) Let [x, y, z] be an m-secant with [x, y, z] [1, 0, 0] 6 S. Consider the line
[0, z, y] of the dual plane. If it intersects R0 (X, Y, Z) at (x, y, z) with inPm t
tersection multiplicity m then j=1 atj = 0.
j
for help
see Section 10
Proof: (a) Suppose w.l.o.g. that (x, y, z) = (0, 0, 1) (so every ctj = 0).
Substituting Z = 1 we have R0 (X, Y, 1). In the sum () each term of
P
R(X,Y,1)
i6{t1 ,...,tm } i ai X+bi Y +ci will contain m linear factors through (0, 0, 1), so, after
expanding it, there is no term with (total) degree less than m (in X and Y ).
Consider the other terms contained in
m
X
i{t1 ,...,tm }
Here
X
R(X, Y, 1)
R(X, Y, 1)
RS[0,0,1] (X, Y, 1)
= S[0,0,1]
tj
.
ai X + bi Y
atj X + btj Y
R
(X, Y, 1) j=1
R(X,Y,1)
RS[0,0,1] (X,Y,1)
is non-zero in (0, 0, 1). Each term
RS[0,0,1] (X,Y,1)
atj X+btj Y
contains at
least m 1 linear factors through (0, 0, 1), so, after expanding it, there is no term
with (total) degree less than (m 1) (in X and Y ). So R0 (X, Y, 1) cannot have
such a term either.

S[0,0,1]
31
(b) As R
(X, Y, 1) is a homogeneous polynomial in X and Y , of total
degree (m 1), (0, 0, 1) is of multiplicity exactly (m 1) on R(X, Y, 1), unless
S[0,0,1] 0
R
(X, Y, Z) happens to vanish identically.
Consider the polynomials
RS[0,0,1] (X,Y,1)
.
atj X+btj Y
They are m homogeneous polynomials
in X and Y , of total degree (m 1). Form an m m matrix M from the coefficients. If we suppose that atj = 1 for all Ptj S [0, 0, 1] then the coefficient of
X m1k Y k in
RS[0,0,1] (X,Y,1)
,
atj X+btj Y
so mjk is k ({bt1 , ..., btm } \ {btj }) for j = 1, ..., m
and k = 0, ..., m 1. So M is the elementary

Q symmetric matrix (see in Section 9
on symmetric polynomials) and | det M | = i<j (bti btj ), so if the points are all
S[0,0,1]
distinct then det M 6= 0. Hence the only way of R

(X, Y, 1) = 0 is when
j tj = 0.
In order to prove (c), consider the line [0, z, y] in the dual plane. To calculate its
intersection multiplicity with R0 (X, Y, Z) at (x, y, z) we have to look at R0 (X, y, z)
and find out the multiplicity of the root X = x. As before, for each term of
P
R(X,y,z)
i6{t1 ,...,tm } i ai X+bi y+ci z this multiplicity is m, while for the other terms we
have
m
i{t1 ,...,tm }
X
RS[x,y,z] (X, y, z)
R(X, y, z)
R(X, y, z)
tj
.
= S[x,y,z]
ai X + bi y + cz
atj X + btj y + ctj z
R
(X, y, z) j=1
R(X,y,z)
Here RS[x,y,z]
is non-zero at X = x. Now RS[x,y,z] (X, y, z) =
(X,y,z)
btj y + ctj z.
Qm
j=1
atj X +
S[x,y,z]
(X,y,z)
Each term R
atj X+btj y+ctj z is of (X-)degree at most m 1. We do know that the
S[x,y,z]
Pm
(X,y,z)
degree of j=1 tj R
atj X+btj y+ctj z is at least (m 1) (or it is identically zero), as
the intersection multiplicity is at least m1. So if we want intersection multiplicity
m then it must vanish, in particular its leading coefficient
m
Y
j=1
atj )
m
X
tj
j=1
atj
= 0.
Remark. If i = ai , i.e. we have the partial derivative w.r.t X, then each
tj
atj
are
equal to 1. The multiplicity in question remains (at least) m if and only if on the
corresponding m-secant [x, y, z] the number of affine points (i.e. points different
from (0, z, y)) is divisible by the characteristic p.
In particular, we may consider the case when all (P) = 1.
Consider
X
R(X, Y, Z)
= |B|1 ({b1 X + b2 Y + b3 Z : b B}).
R1 =
b1 X + b2 Y + b 3 Z
bB
32
I. Background
For any 2-secant [x, y, z] we have R1 (x, y, z) = 0. It does not have a linear component if |B| < 2q and B is minimal, as it would mean that all the
lines through a point are 2-secants. Somehow this is the prototype of all
the derivatives of R. E.g. if we coordinatize s.t. each b1 is either 1 or 0, then
P
R(X,Y,Z)
1
, which is a bit weaker in the sense that it contains
X
R = bB\LX b1 X+b
2 Y +b3 Z
the linear factors corresponding to pencils centered at the points in B LX .
Substituting aQtangent line [x, y, z], with B [x, y, z] = {a}, into R1 we get
R1 (x, y, z) = bB\{a} (b1 x + b2 y + b3 z), which is non-zero. It means that R1
contains precisely the 2-secants of B. In fact an m-secant will be a singular
point of R1 , with multiplicity at least m 1.
A similar argument shows, that in general, if
Rt = |B|t ({b1 X + b2 Y + b3 Z : b B}),
then for any t + 1-secant [x, y, z] we have Rt (x, y, z) = 0. It does not have a
linear component if |B| < (t + 1)q and B is minimal, as it would mean that all
the lines through a point are t + 1-secants. Somehow this is the prototype
of all the t-th derivatives of R. E.g. if we coordinatize s.t. each b1 is either
t
1 or 0, then X
R = |B|t ({b1 X + b2 Y + b3 Z : b B \ LX }), which is a bit
weaker in the sense that it contains the linear factors corresponding to pencils
centered at the points
in B LX . Substituting a t-secant line [x, y, z] into Rt we
Q
get Rt (x, y, z) = bB\[x,y,z] (b1 x+b2 y +b3 z), which is non-zero. It means that R0
contains all the t + 1-secants but none of the t-secants of B. In fact Rt contains
all the m-secants with multiplicity at least m t.
7.3
Hasse derivatives of the Redei polynomial
The next theorem is about Hasse derivatives of R(X, Y, Z). (For its properties see
Section 5.4.)
Theorem 7.10. (1) Suppose [x, y, z] is an r-secant line of S with [x, y, z] S =
rij
i
{(asl , bsl , csl ) : l = 1, ..., r}. Then (HX
HYj HZ
R)(x, y, z) =
X
y, z)
R(x,
asm1 asm2 ...asmi bsmi+1 ...bsmi+j csmi+j+1 ...csmr ,
m1 <m2 <...<mi
mi+1 <...<mi+j
mi+j+1 <...<mr
{m1 ,...,mr }={1,2,...,r}
y, z) = Q
where R(x,
l6{s1 ,...,sr } (al x + bl y + cl z), a non-zero element, independent from i and j.
(2) From this we also have
X
0i+jr
rij
i
y, z)
(HX
HYj HZ
R)(x, y, z)X i Y j Z rij = R(x,
r
Y
(asl X+bsl Y +csl Z),
l=1
constant times the Redei polynomial belonging to [x, y, z] S.
33
rij
i
(3) If [x, y, z] is a ( r + 1)-secant, then (HX
HYj HZ
R)(x, y, z) = 0.
rij
i
(4) If for all the derivatives (HX
HYj HZ
R)(x, y, z) = 0 then [x, y, z] is not
an r-secant.
(5) Moreover, [x, y, z] is a ( r + 1)-secant iff for all i1 , i2 , i3 , 0 i1 + i2 + i3 r

i 1 i 2 i3
the derivatives (HX
HY HZ R)(x, y, z) = 0.
(6) The polynomial
X
rij
i
(HX
HYj HZ
R)(X, Y, Z)X i Y j Z rij
0i+jr
vanishes for each [x, y, z] ( r)-secant lines.

(7) In particular, when [x, y, z] is a tangent line to S with [x, y, z] S =
{(at , bt , ct )}, then
(R)(x, y, z) = ( (X R)(x, y, z), (Y R)(x, y, z), (Z R)(x, y, z) ) = (at , bt , ct ).
If [x, y, z] is a ( 2)-secant, then (R)(x, y, z) = 0. Moreover, [x, y, z] is a
( 2)-secant iff (R)(x, y, z) = 0.
Proof: (1) comes from the definition of Hasse derivation and from asl x + bsl y +
i1 i2 i3
csl z = 0; l = 1, ..., r. In general (HX
HY HZ R)(X, Y, Z) =
X
Y
am1 am2 ...ami1 bmi1 +1 ...bmi1 +i2 cmi1 +i2 +1 ...cmi1 +i2 +i3
(ai X+bi Y +ci Z).
m1 <m2 <...<mi
1
mi +1 <...<mi +i
1
1
2
mi +i +1 <...<mi +i +i
1
2
1
2
3
|{m1 ,...,mi +i +i }|=i1 +i2 +i3
1
2
3
i6{m1 ,...,mi1 +i2 +i3 }
(2) follows from (1). For (3) observe that after the r-th derivation of R still
remains a term asi x + bsi y + csi z = 0 in each of the products. Suppose that for
some Q
r-secant line [x, y, z] all the r-th derivatives are zero, then from (2) we get
r
that l=1 (asl X + bsl Y + csl Z) is the zero polynomial, a nonsense, so (4) holds.
Now (5) and (7)
Qrare proved as well. For (6) one has to realise that if [x, y, z] is an
r-secant, still l=1 (asl x + bsl y + csl z) = 0.
Or: in the case of a tangent line
R =
|S|
X
j=1
7.4
(Pj V)
Y
i6=j
Pi V =
|S|
X
j=1
Pj
Y
(Pi V).
i6=j
Examples
In this section we compute the polynomials (curves) for some well-known pointsets. The
computations are also used to illustrate several results and ideas of this book. In the cases
34
I. Background
when the pointset is a blocking set, we use the notation f = (f1 , f2 , f3 ) and g = (g1 , g2 , g3 )
for certain curves (see Section 21.2) satisfying R = f (X, Y, Z) = g (Y q Z Y Z q , Z q X
ZX q , X q Y XY q ).
Example 7.11.
The line [a, b, c]. Its Redei polynomial is a(Y q Z Y Z q )+b(Z q X ZX q )+c(X q Y XY q ).
Example 7.12.
A conic. For the Redei polynomial of the parabola X 2 Y Z see Exercise 7.7.
Example 7.13.
The projective triangle. Let q be odd, B = {(1, 0, 0); (0, 1, 0); (0, 0, 1)} {(a2 , 0, 1);
(1, a2 , 0); (0, 1, a2 ) : a GF(q) }. Then
R(X, Y, Z) = XY Z((Z)
(X q X)Y Z[Z
q1
2
q1
2
(Y )
q1
2
q1
2
)(X
(X Y XY )Z[Z
q1
2
(Y )
q1
2
(X)
q1
2
(Z q X ZX q )Y [(Y )
2
q1
2
)((Y )
] + (Y q Y )XZ[(X)
(Z q Z)XY [(Y )
q
q1
2
q1
2
(X)
q
q1
2
q1
2
q1
2
q1
2
q1
2
)=
]+
]=
]+(Y ZY Z )X[(X)
q1
2
q1
2
(X)
q1
2
q1
2
q1
2
(Y )
q1
2
].
Note that g = 0 iff [x, y, z] = [ , 1, 0] or [1, 0, ] or [0, , 1], so for the 2-secants.
Example 7.14.
The sporadic almost-Redei blocking set. The affine plane of order 3 can be embedded into
PG(2, 7) as the points of inflexion of a non-singular cubic. The 12 lines of this plane cover
each point of PG(2, 7), so in the dual plane they form a blocking set of size 12 = 3(7+1)/2,
but its maximal line-intersection is only 4 = (12 7) 1. A characterization of it can be
found in [69].
A representation of this blocking set is the following: its affine part is U =
{(x, x6 + 3x3 + 1, 1) : x GF(7)} {(0, 1, 1)}, the infinite part is D =
{(1, 0, 0), (1, 1, 0), (1, 2, 0), (1, 4, 0)}. Now
R(X, Y, Z) = X 10 Y 2 X 10 Z 2 X 7 Y 5 2X 7 Y 3 Z 2 + 3X 7 Y Z 4 X 4 Y 8 + X 4 Z 8 + XY 11 +
XY 9 Z 2 + 4XY 7 Z 4 + XY 3 Z 8
from which we get f =
(X 3 Y 2 X 3 Z 2 Y 5 2Y 3 Z 2 + 3Y Z 4 , X 4 Y + XY 4 + XY 2 Z 2 + 4XZ 4 , X 4 Z + XY 3 Z)
and g = (XY 2 Z, X 3 Z + 2Y 3 Z, X 3 Y Y 4 + 3Z 4 ).
Note that g(x, y, z) = (0, 0, 0) iff (x, y, z) {[1, 0, 0]; [1, 2, 0]; [1, 4, 0]} and all three are
2-secants.
Example 7.15.
]+
8. Univariate representations
35
The Baer-subplane.
In PG(2, q) the standard embedding of a Baer subplane is {(a, b, 1) : a, b GF( q)}
{(m, 1, 0) : m GF( q)} {(1, 0, 0)}.

Now its Redei polynomial is
Y
R(X, Y, Z) =
a,bGF(
(X q X)[Y Z
q)
aGF(
Z] + (Y q Y )[X
q
(Y Z Y Z )X
(aX + bY + Z)
Z XZ
+ (Z X ZX )Y
(aX + Y )X =
q)
] + (Z q Z)[XY
q
+ (X Y XY )Z
Y]=
Here the equation of the blocking set is

X
(c
b cb
)+Y
(a
b ab
)+Z
(a
c ac
) = 0.
Example 7.16.
2
In PG(2, q 3 ) let U = {(x, x + xq + xq , 1) : x GF(q 3 )}, and D the directions determined

by them, |D| = q 2 + 1.
Now for B = U D we get
2
f = ( Xq Z + Xq
Y qZ Xq
X q
and
Y Zq + Y q Z Y Zq , Xq Z + Xq
g = ( X q , X q + X q
+1
Xq Y Xq
2
Y q + Y q , Xq
q+1
q+1
Z q + XZ q ,
2
Y q XY q )
Z q + Z q ).
Example 7.17.
In PG(2, q 3 ) let U = {(x, xq , 1) : x GF(q 3 )}, and D the directions determined by them,
D = {(1, aq1 , 0) : a GF(q 3 ) }, |D| = q 2 + q + 1.
Then, after the linear transformation (1, aq1 , 0) 7 (1 aq1 , aq1 , 0), where is a
(q 1)-st power, we have
rNZ = (X Y )q
+q+1
(X Y )q
+q+1
Example 7.18.
The
Hermitian
curve.
In PG(2, q), the Hermitian curve, which is a unital, {(x, y, z) :
x q+1 + y q+1 + z q+1 = 0} in PG(2, q) has the following Redei polynomial:

R(X, Y, Z) = (X
q+1
+Y
q+1
+Z
q+1 q q+1
Xq
q+1
Yq
q+1
Zq
q+1
Univariate representations
Here we describe the analogue of the Redei polynomial for the big field representations.
36
8.1
root polynomial
I. Background
The affine polynomial and its derivatives
After the identification AG(n, q) GF(q n ), described in Section 6, for a subset

S AG(n, q) one can define the root polynomial
Y
X
BS (X) = B(X) =
(X s) =
(1)k k X |S|k ;
sS
direction
polynomial
and the direction polynomial

Y
X
F (T, X) =
(T (X s)q1 ) =
(1)k
k T |S|k .
sS
linearized
polynomials
Here k and
k denote the k-th elementary symmetric polynomial of the set S and
{(X s)q1 : s S}, respectively. The roots of B are just the points of S while
F (x, t) = 0 iff the direction t is determined by x and a point of S, or if x S and
t = 0.
If F (T, x) is viewed as a polynomial in T , its zeros are the n1 -th roots of unity,
moreover, (x s1 )q1 = (x s2 )q1 if and only if x, s1 and s2 are collinear.
In the special case when S = Lk is a k-dimensional affine subspace, one may think
that BLk will have a special shape.
We know that all the field automorphisms of GF(q n ) are Frobenius-automorphisms
m
x 7 xq for i = 0, 1, ..., n 1, and each of them induces a linear transformation of AG(n, q). Any linear combination of them, with coefficients from
GF(q n ), can be written as a polynomial over GF(q n ), of degree at most q n1 .
These are called linearized polynomials. Each linearized polynomial f (X) induces
a linear transformation x 7 f (x) of AG(n, q). Whats more, the converse is
also true: all linear transformations of AG(n, q) arise this way. Namely, distinct
linearized polynomials yield distinct transformations as their difference has degree q n1 so cannot vanish everywhere unless they were equal. Finally, both
the number of n n matrices over GF(q) and linearized polynomials of form
2
n1
c0 X + c1 X q + c2 X q + ... + cn1 X q , ci GF(q n ) is (q n )n . This is the same
argument as in Section 5.6.
Here we show (what we did already in 6.1) that
Proposition 8.1. (i) The root polynomial of a k-dimensional subspace of AG(n, q)
containing the origin, is a linearized polynomial of degree q k ;
(ii) the root polynomial of a k-dimensional subspace of AG(n, q) is a linearized
polynomial of degree q k plus a constant term.
Proof: (i) For k = 0 the statement is obvious, B0 = X. Any one-dimensional
linear subspace is of form {a : GF(q)} = aGF(q) for some a GF(q n ) .
The root polynomial of the one-dimensional subspace corresponding to GF(q) is
BGF(q) = X q X, so the root polynomial of aGF(q) is BGF(q) (a1 X), which is a
linearized polynomial of degree q.
8. Univariate representations
37
Now if L is a subspace of dimension k > 1 then let K be a subspace of L of

dimension k 1. By induction, the root polynomial BK is a linearized polynomial
inducing a linear map with kernel K. The image of L is one-dimensional, BK (L) =
M . It is easy to check that BL (X) = BM (BK (X)).
(ii) Any subspace is a translate of a linear subspace; consider any subspace v + L
with v GF(q n ), its root polynomial is BL (X v). As BL is a linear map, we
have BL (X v) = BL (X) BL (v).
A more compact form of the equation of a hyperplane is Trqn q (aX) + b = 0,
where a, b GF(q n ), a 6= 0. Indeed, for the root polynomial
BH (X) =
n1
X
ci X q + b
i=0
n
q
q
q
of the hyperplane H, the polynomial cn1 BH
cq+1
n1 (X X)cn2 BH has degree
n2
n1
at most q
and vanishes in all the q
points of H, hence it is identically zero.
i
Equating the coefficients of X q for 0 i n 2 implies the trace function form
above.
Now we examine the derivative(s) of the affine root polynomial (written up with
a slight modification). Let S GF(q n ) and consider the root and direction polynomials of S [1] = {1/s : s S}:
B(X) =
(1 sX) =
sS
F (T, X) =
(1)k k X k ;
(1 (1 sX)q1 T ) =
sS
(1)k
k T k .
P
n
For the characteristic function of S [1] we have |S|(X) = sS (1sX)q 1 .
P
P
n
1
, we have (X X q )B 0 = B(|S| sS (1
Then, as B 0 (X) = B(X) sS 1sX
n
n
sX)q 1 ) = B, after derivation B 0 + (X X q )B 00 = B 0 + B0 , so B 0 (B)0
and (as B 0) we have BB 0 B 2 0 .
8.2
The projective polynomial
After the identification PG(n, q) GF(q n+1 )/GF(q)

, described in Section 6, for
Q
a subset S PG(n, q) one can define C0 (X) = sS Tr(sX) =
= X |S| (
sS
s)
Y
sS

(X q1 )n1 (sq1 )n1 +...+(X q1 )1 (sq1 )1 +X q1 sq1 +1 .
Here is the
proof of
Exercise 5.2!
38
I. Background
So it is enough to consider C(X) =

P
Q n1
(
(X q1 )i (sq1 )i ) + 1 =
sS

n1
P i i
Y )+1 =
(
Q
S [q1]
q1
i=0
i=0
n1

P i n1 i
(
Y
) + n1 , where Y =
Q
1 S [q1]
i=0
Here the direction polynomial is

F (U, X) =
(1 sX(1 sX)q1 U ).
sS
For a value x, the linear factors of F (U, x) have zeros of the form u1 = sx(1
2
sx)q1 which satisfies 1+v+v q+1 +v q +q+1 +...+v n1 = 0 for v = u1 . Moreover,
s1 x(1 s1 x)q1 = s2 x(1 s2 x)q1 if and only if 1/x, s1 and s2 are collinear, so
there is a one-to-one correspondence between the zeros v and the n1 direction
of lines through 1/x.
8.3
Examples
This section is about the polynomials for some well-known pointsets, like in a former
part. We leave them as exercises.
Affine sets
For an affine line we have the equation X q c0 X + b = 0 as we have seen already.
Exercise 8.2. Calculate the polynomials of the following affine pointsets.
For an ellipse, i.e. the q + 1 points of a conic contained in AG(2, q).
An affine Baer-subplane: A = AG(2, q) = {(a, b) : a, b GF( q)} ! {a + b : a, b
2
GF( q)} GF(q ) .
Projective sets
Exercise 8.3. Calculate the polynomials of the following projective pointsets.
A conic.
The projective triangle., see Section 7.4.
A Baer-subplane PG(2, q) of PG(2, q) in standard position.
q+1
The cyclic embedding of the Baer-subplane: GF(q 3 ) = hwi, B = hwq
The cyclic (q q + 1)-arc

The Hermitian curve.
i.
9. Symmetric polynomials and invariants of subsets
9
9.1
39
Symmetric polynomials and invariants of subsets

The Newton formulae
In this section we recall some classical results on symmetric polynomials. For more
information and the proofs of the results mentioned here, we refer to [135].
The multivariate polynomial f (X1 , ..., Xt ) is symmetric, if f (X1 , ..., Xt ) =
f (X(1) , ..., X(t) ) for any permutation of the indices 1, ..., t. Symmetric polynomials form a (sub)ring (or submodule over F) of F[X1 , ..., Xt ]. The most famous
particular types of symmetric polynomials are the following two:
Definition 9.1. The k-th elementary symmetric polynomial of the variables
X1 , ..., Xt is defined as
X
k (X1 , ..., Xt ) =
X i1 X i2 X ik .
{i1 ,...,ik }{1,...,t}
0 is defined to be 1 and for j > t j = 0, identically.

Given a (multi)set A = {a1 , a2 , ..., at } from any field, it is uniquely determined by
its elementary symmetric polynomials, as
t
X
i (A)X ti =
i=0
t
Y
(X + aj ).
j=1
Definition 9.2. The k-th power sum of the variables X1 , ..., Xt is defined as
k (X1 , ..., Xt ) :=
t
X
Xik .
i=1
The power sums determine the (multi)set a bit less than the elementary symmetric polynomials. For any fixed s we have
s
X
s
i=0
i (A)X si =
t
X
(X + aj )s
j=1
but in general it is not enough to gain back the set {a1 , ..., at }. Note also that
in the previous formula the binomial coefficient may vanish, and in this case it
hides i as well.
One may feel that if a (multi)set of field elements is interesting in some sense then
its elementary symmetric polynomials or its power sums can be interesting as well.
E.g.

0, if j = 1, 2, ..., q 2, q ;
A = GF(q): j (A) = j (A) =
1, if j = q 1.
q1 (GF(q)) =
Q
= 1
GF(q)
is Wilsons theorem
40
I. Background
If A is an additive subgroup of GF(q) of size pk : j (A) = 0 whenever p 6 |j < q 1

holds. Also j (A) = 0 for j = 1, ..., pk 2, pk .
If A is a multiplicative subgroup of GF(q) of size d|(q 1): j (A) = j (A) = 0 for
j = 1, ..., d 1.
The fundamental theorem of symmetric polynomials: Every symmetric polynomial
can be expressed as a polynomial in the elementary symmetric polynomials.
There are several proofs for this theorem, the most well-known one uses induction
for the exponent-series of the terms after ordering the terms in a lexicographic
way; one may formulate the proof in the language of initial ideals (see [72]), or
there is another one using Galois-theory, etc.
According to the fundamental theorem, also the power sums can be expressed
in terms of the elementary symmetric polynomials. The Newton formulae are
equations with which one can find successively the relations in question. Essentially
there are two types of them:
kk = 1 k1 2 k2 + ... + (1)i1 i ki + ... + (1)k1 k 0
(N1)
t+k t+k1 1 + ... + (1)i t+ki i + ... + (1)t k t = 0.
(N2)
and
In the former case 1 k t, in the latter k 0 arbitrary. Note that if we define
i = 0 for any i < 0 or i > t, and, for a fixed k 0, 0 = k, then the following
equation generalizes the previous two:
k
X
(1)i i ki = 0.
(N3)
i=0
Exercise 9.3. Prove the Newton identities by differentiating

B(X) =
Y
sS
(1 + sX) =
|S|
X
i X i .
i=0
Symmetric polynomials play an important role

Q when we use Redei polynomials, as
e.g. expanding the affine Redei polynomial i (X + ai Y + bi ) by X, the coefficient
polynomials will be of the form k ({ai Y + bi : i}).
Pq1
Magic. (G
acs) Let f (X) = i=0 ci X i be a polynomial from GF(q)[X]. Consider
the following (affine Redei-) polynomial of its graph:
Y
(X aY f (a));
aGF(q)
41
then its
part of degree q 1 is precisely the homogenized form of
Phomogeneous
q1
f , i.e. i=0 ci X i Y q1i .
Proof: Let k be the coefficient of X q1k Y k in the Redei-polynomial, we prove
by induction that k = cq1k . Let GF(q) = {a1 , ..., aq } and bi = f (ai ).
P
We know that 0 = bi = cq1 and
X
X
k = (1)k+1
ai1 aik
bi
i1 <...<ik
i6={i1 ,...,ik }
Pq
if 1 k q 1. Hence k = (1)k+1 k (a1 , ..., aq ) i=1 bi + k1 , where k1 is
defined as k1 , with bi replaced by ai bi .
Pq1
Let f(X) = (c0 + cq1 )X + i=2 ci1 X i ; then f(ai ) = ai bi . By induction we may
assume that k1 is the coefficient of X qk in f(X). Hence we obtain k = k1 =
cqk1 for k < q 1; finally q1 = cq1 + q2 = c0 .
Corollary 9.4. Expanding the Redei-polynomial

Y
(X aY f (a)) =
q
X
rk (Y )X qk ,
k=0
aGF(q)
for k = 1, ..., q 2 we get degY (rk ) k 1; equality holds iff cqk 6= 0.

Proof: The first part is easy: rk (Y ) = (1)k k ({(aY + f (a)) : a GF(q)}) so
the leading coefficient is just k (GF(q)) = 0. The second part, the case of equality
follows from the previous statement.
Given S = {x1 , x2 , ..., xn }, the following determinant is

determinant of S:

1
...
1
1
x1
x2
...
xn
2
x
x22
...
x2n
V dM (x1 , x2 , ..., xn ) = .1
.
.
xn1 xn1 ... xn1
1
called the Vandermonde

Y

(xi xj )
=

i<j

The P -adic (P = pe , p prime) Moore-determinant of S is

x1

x2
...
xn
xP

P
P
x
...
x

n
1
2
n
2
2
Y
X
xP 2

P
P
x2
...
xn

=
1
M RDP (x1 , ..., xn ) =
i xi

..

(1 ,...,n )PG(n1,P ) i=1
.

xP n1 xP n1 ... xP n1
1
it can be called
k,1(a1 ,...,aq ; b1 ,...,bq)
42
I. Background
Note that this formula gives the value of the determinant up to a non-zero constant only, one way to find the exact value is choosing each homogeneous vector
(1 , 2 , ..., n ) in such a way that its first non-zero coordinate is 1.
The elementary symmetric determinant of S is

1
1
...
1

1 (S \ {x1 })
1 (S \ {x2 })
...
1 (S \ {xn })

Y
2 (S \ {x1 })
2 (S \ {x2 })
...
2 (S \ {xn })
=
(xi xj )

..

i<j

.
(S \ {x }) (S \ {x }) ... (S \ {x })
n1
n1
n1
Exercise 9.5. Give a unified proof for the determinant formulae! (Hint: consider
x1 , ..., xn as free variables.)
Exercise 9.6.
(a) Prove that V dM (x1 , x2 , ..., xn ) 6= 0 iff {x1 , ..., xn } are pairwise distinct.
(b) Prove that M RDpe (x1 , x2 , ..., xn ) 6= 0 iff {x1 , ..., xn } are independent over
GF(pe ).
(c) Prove the following version of Wilsons theorem:
Y
e
y
= (1)n M RDpe (x1 , ..., xn )p 1 .
yhx1 ,...,xn iGF(pe )
y6=0
Exercise 9.7. Prove the following general form of the elementary symmetric determinant: Given S = {x1 , x2 , ..., xn , xn+1 , ..., xm },

1
1
...
1

1 (S \ {x1 })
(S
\
{x
})
...
(S
\
{x
})
n
1
2
1

Y
2 (S \ {x1 })
2 (S \ {x2 })
...
2 (S \ {xn })
(xi xj ),
=

.

..
1i<jn

(S \ {x }) (S \ {x }) ... (S \ {x })
n1
n1
n1
so somehow the elements xn+1 , ..., xm do not count.

Exercise 9.8. Prove that
Y
(T + 1 X1 + 2 X2 + ... + n1 Xn1 + n ) =
GF(q)n
T
Tq
2
Tq
..
.
..
.
n
Tq
X1
X1q
X2
X2q
...
...
X1q
X2q
...
X1q
X2q
...
Xn1
q
Xn1
1
1
q
Xn1
q
Xn1
X1
X1q
X2
X2q
...
...
Xn1
q
Xn1
1
1
X1q
..
.
X2q
...
q
Xn1
X1q
n1
X2q
n1
...
n1
q
Xn1
usually we ask
det = 0 or not
9.2
43
The invariants uf , vf and wf
In this section,
Pn which is extracted from [135] mostly, we consider a fixed polynomial
fQ(X) = i=0 ci X i GF(q)[X]
of degree n. Put sk = k ({f (x) : x GF(q)}), so
Pq
qk
(X
+
f
(x))
=
s
X
.
xGF(q)
k=0 k
Let u = uf be the smallest positive integer k with sk 6= 0 if such k exist,
otherwise set u = .
Let v = vf denote the number of distinct values of f (so v 2 if we assume
that f is not a constant function). If v = q then f is called a permutation
polynomial (PP).
Let w = wP
f be the smallest positive integer with pk = k ({f (x) : x
GF(q)}) = xGF(q) f (x)k 6= 0 if such k exists, otherwise set w = .
One can find another description of u and w in Proposition 9.9.

Firstly we study the relations among u, v, w, n, q. Then we shall investigate the
number Mf of c GF(q) such that f (X) cX is a PP.
With any family S of q elements of GF(q), counted possibly with multiplicity, we
may associate the same invariants u = uS , v = vS , w = wS as above; obviously
the order of the values is irrelevant. However, n measures the complexity of this
set (ordered in a certain sense) somehow. If g(X) is a permutation polynomial
then f (g(X)) has the same invariants u, v and w as f ; and the same is true if f is
replaced by af (X) + b with a, b GF(q), a 6= 0 (and b = 0 if v = 1).
The similar invariants can be defined for a (multi)set S of cardinality 6= q. Note
that if |S| < q then the multiset S (q |S|){0} of size q has the same invariants
u and w as S, while v, the number of distinct elements of S, increases by one iff
0 6 S.
If |S| > q and v = q then for S 0 = S \ GF(q) we have uS 0 = uS and wS 0 = wS if
any of these values is less than q 1.
Q
Proposition 9.9. (a) q u is the degree of X q aGF(q) (X f (a)). This polynomial is equal to c GF(q) iff f (X) = c is a constant polynomial. Hence
v 2 implies 1 u < q and v = 1 implies u = q or u = according to
f = 0 or f = c 6= 0.
P

Q

q1
(b) q 1 w = deg
= deg ( aGF(q) (X f (a)) )0 .
aGF(q) (X f (a))
(c)
P
ak f (a) = cq1k for 0 k < q 1 and aGF(q) aq1 f (a) =
(c0 + cq1 ). Hence n = q 1 iff u = 1 P
iff w = 1. For arbitrary f we have
that f has reduced degree q l 2 iff aGF(q) ak f (a) = 0 for 0 k l.
(We will see it later again.)
aGF(q)
Mf
44
I. Background
(d) From (c) it is clear that w is the smallest positive integer k such that the
reduction of f (X)k mod X q X has degree q 1, if such k exists, and
otherwise w = .
Note that if v 2 then (c) implies that one can change the order of the values
such that the polynomial will have degree q 1 or q 2 (in fact one transposition
is enough); while the invariants u, v, w remain the same.
Lemma 9.10. (a) If 1 k q and k < u + w then pk = (1)k1 ksk .
(b) If ksk 6= 0 for some 1 k q 1 then w is the smallest k with this property,
otherwise w = .
For (a) use the Newton identity, for (b) see (d) below.
Here we simply enlist some of the relations known for u, v, w, n, q, most of them
are easy to prove (an asterisk denotes some of the non-obvious ones):
Proposition 9.11.
(a) v = 1 n = 0 or n = , u = q in the first and u = in the latter
case;
(b) if n 1 then v q/n and hence v [ q1
n + 1]. v = q/n w = or
q1
n = 1. v = [ n + 1] implies n|q 1 or w = . If g.c.d.(n, p) = 1 then
the case n|q 1 can occur only. If n < p then f is of form a(X b)n + c
with a 6= 0.
(c) n = q 1 u = 1 w = 1;
(d) If w < then w < q (since f (a)q = f (a)); and p 6 |w (since pkp =
(pk )p ). Compare to the section on Vandermonde sets, 9.3;
(e) w < w < v;
(f) w
q1
n ,
with equality iff n|q 1;
(g) w = for each a GF(q) the number na of preimages of a is

divisible by p, see Exercise 9.17.
(h) w = n p, v q/p (if n 1);
(i) if f (X) = X p X then w = , n = p, v = q/p;
(j) if f (X) = X n then w = v 1;
(k) n|q 1 v q 1;
(l) w < u w, with equality iff p 6 |u;
(m) w = p|u or u = ;
(n) () w < w < 2(q v) or v = q;

(o) () u
q1
n ,
45
with equality iff n|q 1;
(p) 2 v < q u + v q;
(q) v < q v < q
q1
n
(if n 1);
(r) v < q w 2q/3 1 if w < , (this bound is probably not optimal);

(s) v < q v + w 4q/3 1 if w < , (this bound is probably not
optimal);
(t) v < q v + w q if v > q p, v > 1.
The proof of (e) is obvious from the first equality of Proposition 9.9 (b): if g(X) =
P
q1
then for every b, which is not in the range of f (X), we
aGF(q) (X f (a))
have f (b) = 0. For another proof (and a generalization) see Exercise 9.14.
Proposition 9.12. The following properties are equivalent if 1 n < q:
(a) v = q, i.e. f is a permutation polynomial;
(b) u = q 1;
(c) u > q q/n;
(d) u > q v;
(e) v > q
q1
n ;
(f) w = q 1;
(g) 2q/3 1 < w < ;
(h) q
q+1
n
< w < ;
(i) q u w < ;
(j) u >
q1
2
and w < .
Exercise
q1 then either (i) f is a permutation polynomial
P 9.13. Prove that if wf P
and x f (x)q1 = 1; or (ii) x f (x)q1 = 0 and f assumes every value 0 mod p
times.
A general form of Proposition 9.11 (e) is the following:
Exercise 9.14. (P. Das) Let f (X1 , ..., Xn ) be a polynomial, its value set (range) is
V (f ) = {f (1 , ..., n ) : (1 ,P
..., n ) GF(q)n }. Define wf as the smallest positive
integer k such that
f (1 , ..., n )k 6= 0. Prove (using a Vander(1 ,...,n )GF(q)n
monde determinant) that |V (f )| wf + 1.

We defined Mf as the number of elements in C = {c GF(q) : f (X) cX is a
permutation polynomial}. An alternative definition can be given with the help of
the following
Mf
46
I. Background
(b)
Proposition 9.15. Let D = Df = { f (a)f
: a 6= b GF(q)}, the set of difference
ab
quotients (or directions), determined by (the graph of ) f . Then C = GF(q) \ D.
Later we will use the notation N = Nf = |D| for the number of directions determined by f . The obvious connection is
Mf + Nf = q.
For more on directions we refer to Section 18. Note that Mf is a bit less invariant
than u, v, w are; in general, if f is changed for a f (g(X)) + b X + c, Mf remains
the same only if g(X) is a linear polynomial (and not an arbitrary permutation,
as in the earlier cases, but there the term bX was not allowed).
nk
Let nk denote the reduced degree of f (X)k . Here we summarize some properties
of Mf :
1. If f (X) +QcX is a permutation polynomial
then for 1 k < q 1, in the
Pq
notation aGF(q) (X aY f (a)) = k=0 rk (Y )X qk (see Magic), we have
rk (c) = 0. So if rk (Y ) 6= 0 then degY (rk ) Mf .
2. n1 < q Mf .
3. If Mf 1) q Mf is the number of (n 1)-st powers in GF(q) , so
q1
q1
q Mf = g.c.d.(n1,q1)
. In particular, for n = q/p we have Mf = q p1
,
while n = q yields Mf = q q 1.
6. Mf 1 if 1 < n4 < p = q.
7. Mf n 1 if n 6 |q and n4 q.
8. Mf q 1 with equality iff f has reduced degree 1.
9. Mf q
q1
n1
if 1 < n < q.
10. If Mf < q 1 then Mf q
q 1. See Section 21.
11. (Lov
asz-Schrijver) If q = p then Mf p3
2 . If Mf =
up to a linear transform; see Theorem 19.1 as well.
p3
2
then f (X) = X
p+1
2

12. (Redei) If Mf >
q3
2
47
then
(i) if f (X) cX is not a PP then the number of preimages of each value

f (a) ca is divisible by p;
(ii) Mf q
q1
p1
and p|(Mf + 1);
(iii) if n < q then f 0 (X) = f 0 (0).

(iv) in fact much more is known: in this case, after assuming f (0) = 0, f
is linear over GF(pe ) for some e|h (so GF(pe ) is a subfield of GF(q = ph ));
for details see [43] and Theorem 18.14.
In Section 9.4 we shall see that Mf is somewhat related to double symmetric
polynomials and double power sums as well.
9.3
Arbitrary subsets
Now we consider the more general case when the number of elements is not necessarily q.
Definition 9.16. Let 1 < P
t < q. We say that T = {y1 , ..., yt } GF(q) is a
Vandermonde-set, if k = i yik = 0 for all 1 k t 2.
Vandermonde-sets were first defined and studied in [70]. This part is a generalization from [120].
Here we do not allow multiple elements in T . Observe that the power sums do not
change if the zero element is added to (or removed from) T . Note that in general
the Vandermonde property is invariant under the transformations y ay + b
(a 6= 0) if and only if p|t; if p 6 |t then a constant term tbk occurs in the power
sums. (It may help in some situations: we can translate T to a set with 1 = 0
if needed.)
If p|t then a t-set cannot have more than t 2 zero power sums (so in this case
Vandermondeness means w = wT = t 1). This is an easy consequence of the fact
that a Vandermonde-determinant of distinct elements cannot be zero: consider the
product
1
1
1
...
1
y1
2
y1
.
.
.
t1
y1
y2
2
y2
t1
y2
...
...
...
yt
2
yt
t1
yt
1
1
.
.
.
1
it cannot result in the zero vector.

Note that we have already seen that for a set of q elements the value w can be .
Recall that in that case the set has to contain multiple elements, in fact we had
w = all the multiplicities are divisible by the characteristic p.
48
I. Background
Exercise 9.17. Prove that if 1 < t q, |T | = t and multiplicities are allowed

then wT = all the multiplicities of the elements of T are divisible by the
characteristic p.
t1
The proof above, with slight modifications, shows that in general a t-set cannot
have more than t 1 zero power sums (so for a Vandermonde-set wT is either t 1
or t). If the zero element does not occur in T then consider the product
y1
2
y1
.
.
.
t1
y1
t
y1
y2
2
y2
t1
y2
t
y2
...
...
...
...
yt
2
yt
t1
yt
t
yt
1
1
1
.
.
.
1
it cannot result in the zero vector as the determinant is still non-zero. If 0 T

then remove it and we are again in the zero-free situation.
If for a set T of cardinality t we have that k (T ) = 0 for k = 1, ..., t 1, so wT = t
then such a set can be called a super-Vandermonde set. Note that the zero element
is never contained in a super-Vandermonde set (removing it, for the other t 1
elements all the first t 1 power sums would be zero, which is impossible). The
same argument gives the first examples of super-Vandermonde sets:
Example 9.18. If T is a Vandermonde set, containing the zero element, then T \{0}
is a super-Vandermonde set. In particular, if T is a Vandermonde set and |T | = t
is divisible by the characteristic p, then for any a T , the translate T a is a
Vandermonde set, containing the zero element.
In the next proposition we characterize the Vandermonde-property.
Proposition 9.19. Let T = {y1 , ..., yt } GF(q). The following are equivalent
(i) T is a Vandermonde set, i.e. wT = t 1;
Qt
0
(ii) the polynomial f (Y ) = i=1 (Y yi ) is of the form Y t g(Y )p +aY +b (where
0 t0 p 1, t0 t mod p);
Pt
(iii) for the polynomial (Y ) = i=1 (Y yi )q1 , tY q1 + (Y ) has degree
q t; moreover
Pt
(iv) for some Q = ps , t < Q, the polynomial tY Q1 i=1 (Y yi )Q1 has
degree Q t.
Proof: The coefficients of are the power sums of the set T , so (i) and (iii)
are clearly equivalent. (i) (iv) is similar. The equivalence of (i) and (ii) is an
easy consequence of the Newton formulae relating power sums and elementary
symmetric polynomials.

t + (Y ) = (Y )
if p|t
49
Note that for the function in (iii), t + (Y ) is the characteristic function of

T , that is it is 1 on T and 0 everywhere else. (i) means that a Vandermonde set
is equivalent to a fully reducible polynomial of form g p (Y ) + Y t + cY . (In the
important case when p|t we have g p (Y ) + Y .)
And now we characterize the super-Vandermonde-property.
Proposition 9.20. Let T = {y1 , ..., yt } GF(q). The following are equivalent
(i) T is a super-Vandermonde set, i.e. wT = t;
0
(ii) the polynomial f (Y ) = ti=1 (Y yi ) is of the form Y t g(Y )p + c (where

0 t0 p 1, t0 t mod p);
Pt
(iii) for the polynomial (Y ) = i=1 (Y yi )q1 , tY q1 + (Y ) has degree
q t 1; moreover
Pt
(iv) for some Q = ps , t < Q, the polynomial tY Q1 i=1 (Y yi )Q1 has
degree Q t 1.
Proof:
Very similar to the Vandermonde one above.
Here come some really motivating examples for Vandermonde sets.

Example 9.21. Let q be a prime power.
(i) Any additive subgroup of GF(q) is a Vandermonde set.
(ii) Any multiplicative subgroup of GF(q) is a (super-)Vandermonde set.
(iii) For q even, consider the points of AG(2, q) as elements of GF(q 2 ). Any q-set
corresponding to the affine part of a hyperoval with two infinite points is a
Vandermonde set in GF(q 2 ).
(iv) Let q be odd and consider the points of AG(2, q) as elements of GF(q 2 ) and a
q +1-set A = {a1 , ..., aq+1 } in it, intersecting every line in at most two points
(i.e.
position, i.e.
P an oval or (q + 1)-arc). Suppose that it is in a normalized
ai = 0. Then A is a super-Vandermonde set in GF(q 2 ).
Proof: (i) Suppose T is an additive subgroup of size t in GF(q). We want to prove
that Proposition 9.19 (ii) is satisfied, that is f (Y ) = yT (Y y) has only terms
of degree divisible by p, except for the term Y . By Theorem 5.25 if we prove that
f is additive, hence GF(p)-linear, then this implies that f has only terms of degree
a power of p.
Consider the polynomial in two variables F (X, Y ) = f (X) + f (Y ) f (X + Y ).
First of all note that it has full degree at most t and that the coefficient of X t and
Y t is zero. Considering F as a polynomial in X, we have
F (X, Y ) = r1 (Y )X t1 + r2 (Y )X t2 + + rt (Y ),
50
I. Background
where ri (Y ) (i = 1, . . . t) is a polynomial in Y of degree at most i (and deg(rt )

t 1). Now F (X, y) 0 for any y T (as a polynomial of X), so all ri -s have
at least t roots. Since their degree is smaller than this number, they are zero
identically, so we have F (X, Y ) 0, hence f is additive.
(ii) Suppose T is a multiplicative subgroup of size t in GF(q). Then the polynomial
f (Y ) = ti=1 (Y yi ) is of the form Y t 1 so Proposition 9.20 (ii) is satisfied, we
are done.
(iii) Let {x1 , ..., xq } GF(q 2 ) correspond to the affine part of the hyperoval H
and 1 and 2 be (q + 1)-st roots ofP
unity corresponding to the two infinite points.
q
Consider the polynomial (X) = i=1 (X xi )q1 . For any point x out of the
hyperoval every line through x meets H in an even number of points, and since
(x xi )q1 represents the slope of the line joining the affine points x and xi ,
we have that (x) = 1 + 2 for any x
/ {x1 , ..., xq }. There are q 2 q different
choices for such an x, while the degree of is at most q 2, so (X) 1 + 2
identically (that is, all coefficients of are zero except for the constant term), so
by Proposition 9.19 (iv), we are done.
(iv) A short proof is that by Segres theorem 14.9 such a pointset is a conic if q is
odd, so affine equivalent to the unit circle { GF(q 2 ) : aq+1 = 1}, which is a
multiplicative subgroup.
Exercise 9.22. Given an additive subgroup G

Q (GF(q), +), with base
G = hg1 , ..., gt iGF(P ) , write up its root polynomial gG (Y g) as in Proposition
9.19 (ii).
ForQa multiplicative subgroup H = hi (GF(q) , ), |H| = t, its root polynomial
is hH (Y h) = Y t 1.
Note that Proposition 9.19 (iv) implies that if T GF(q1 ) GF(q2 ) then T is a
Vandermonde-set in GF(q1 ) if and only if it is a Vandermonde-set in GF(q2 ).
There are other interesting examples as well.
i
Example 9.23. Let q = q0t1 , then in GF(q) T = {1} { q0 : i = 0, ..., t 2} for

some element GF(q) satisfying Trqq0 ( k ) = 1 for all k = 1, ..., t 1.
As:
t2
X
i=0
( q0 )k =
t2
X
( k )q0 = Trqq0 ( k ) = 1.
i=0
Note that such exists for several triples (t, q0 , q), here I enlist some values; -
means that such does not exist, while x means that the only element with the
property above is 1 GF(q0t1 ), see Exercise 9.24 :

q0 t
3
4
5
7
8
9
3
x
4
x
x
6
x
x
8
x
x
x
9
x
-
x
x
-
10
x
x
11
-
13
-
14
x
?
x
x
12
x
x
51
?
x
x
15
x
?
16
17
18
x
?
?
Exercise 9.24. Prove that if g.c.d.(q0 , t) 6= 1 then the only element with the
property above is 1 GF(q0t1 ).
Exercise 9.25. Let T = {y1 , ..., yt } be a super-Vandermonde set. Prove that
y
y

y

1
1
1
1
1
1 = t.
y2
y3
yt
Note that it is easy check the previous condition for a multiplicative subgroup;
also that any element of T can play the role of y1 , so in fact we have t conditions.
For applications see for instance Section 13, Proposition 13.9 and its consequence,
Theorem 13.12.
9.4
The generalized formulae
Following G
acs in this section, we define two functions in the independent variables
a1 , ..., at and b1 , ..., bt over any field K, where t is a positive integer. Let k and l
be two non-negative integers, (k, l) 6= (0, 0), k, l t.
Definition 9.26. The elementary symmetric polynomial of order (k, l) is defined as
X
k,l (a1 , ..., at ; b1 , ..., bt ) =
ai1 ai2 aik bj1 bj2 bjl .
{i1 ,...,ik }; {j1 ,...,jl }{1,...,t}
{i1 ,...,ik }{j1 ,...,jl }=
In other words, k,l is the coefficient of X tlk Y k in the Redei polynomial

Qt
Qt
k l
i=1 (X + ai Y + bi ); or the coefficient of X Y in
i=1 (ai X + bi Y + 1). It shows
again that k,l s determine the set {a1 , ..., at ; b1 , ..., bt }.
For k = l = 0, we define 0,0 (a1 , ..., at ; b1 , ..., bt ) := 1 identically.
Definition 9.27. The power sum of order (k, l) is defined to be
k,l (a1 , ..., at ; b1 , ..., bt ) :=
t
X
aki bli .
i=1
Evaluating 0,l or k,0 , 00 may occur, which is defined to be 1.
52
I. Background
Note that for k = 0, 0,l (a1 , ..., at ; b1 , ..., bt ) = l (b1 , ..., bt ), the l-th elementary
symmetric polynomial of the bi -s, and k,0 (a1 , ..., at ; b1 , ..., bt ) = k (a1 , ..., at ) similarly.
Taking l = 0 or k = 0 in the second definition, we get the corresponding power
sum of the ai s or bi s, respectively.
The functions in the previous two definitions are somewhat parallelly symmetric;
they do not change value if we take the same permutation of the first t and second t
variables simultaneously. A direct analogue of the fundamental theorem mentioned
in the previous section does not seem to be true for such functions, but the Newton
formulae can be generalized:
Theorem 9.28. For fixed k and l, define 0,0 := k + l and a,b = 0 for any a > t,
b > t or a + b > t. The following holds:
l
k X
X
r+s
(1)
r=0 s=0

r+s
r,s kr,ls = 0.
r
(NG)
Proof: It is easy to see that on the left hand side we have a homogeneous polynomial of degree k + l with monomials ari bsi ai1 aikr bj1 bjls , where all
indices are different and 0 r k, 0 s l.
If r and s are both positive, then we can get such a term from three summands:
r,s kr,ls , r1,s k+1r,ls
kr,l+1s
the
and r,s1
. Hence
coefficient of such
r+s1
r+s1 r+s1
a monomial is (1)r+s r+s
+
(1)
(
+
) = 0. (Note that if
r
r1
r
the monomial in question exists, then necessarily k + 1 r t, l + 1 s t and
k + l + 1 r s t, hence r,s1 and r1,s are not zero by definition. Similar
remarks will hold for the next two cases.)
If r = 0, s > 1, then there are two summands giving the monomial
in question:

= 0
0,s k,ls and 0,s1 k,l+1s , so the coefficient is (1)s 0s + (1)s1 s1
0
again. The s = 0, r > 1 case is similar.
The {r, s} = {0} and {r, s} = {0, 1} cases are the same, so what is left is to show
that the coefficient of a monomial of type ai1 aik bj1 bjl is also zero. We get
this term k + l times from 0,0 k,l , k times from 1,0 k1,l and l times from
0,1 k,l1 , so its coefficient is certainly zero (here the binomial coefficients are all
1).
Remark 9.29. Taking l = 0, the theorem coincides with (N3).

Remark 9.30. There is a slightly more general formula:
k X
l
X
r=0 s=0
f (r, s)r,s kr,ls = 0,
(NG )
53
were 0,0 is defined to be 1 and the function f (r, s) is defined as follows. Let a and
b be two fixed field elements, f (0, 0) = ak + bl, and for (r, s) 6= (0, 0),

r+s1
r+s1
r+s
f (r, s) = (1)
a+
b .
(NG )
s
r
Proof: Looking through the proof of 9.28, it is easy to see that the function
f (r, s) has to satisfy the following equations: f (0, 0) + kf (1, 0) + lf (0, 1) = 0;
f (r, 0) + f (r + 1, 0) = 0 for all r 1; f (0, s) + f (0, s + 1) = 0 for all s 1; and
f (r, s) + f (r 1, s) + f (r, s 1) = 0 for all r 1, s 1. Now defining f (1, 0) = a,
f (0, 1) = b, it can be seen by induction on k + l that the general solution for
such a function is (NG ).
(NG ) seems to be much stronger, but (NG) has turned out to be sufficient for
all applications found so far. In many applications, the set U = {(ai , bi ) : i =
1, ..., t} will be the graph (or almost a graph) of a polynomial.
For the graph of
P
a polynomial, the double power sum k,l becomes x xk f (x)l , the meaning of
which can be understood through Proposition 5.4.
For a demonstration of the strength of using Redei polynomials together with
these formulae see 15.10.
The invariant Wf
For any polynomial f , let
X
Wf = min{k + l :
xk f (x)l 6= 0} = min{k + l : k,l (a1 , ..., aq ; f (a1 ), ..., f (aq )) 6= 0}
xGF(q)
(where {ai : i} = GF(q)). Here k and l are non-negative integers; for k = 0 or

l = 0, x0 or f (x)0 is defined to be the polynomial 1. By the remarks at the end
of the previous section, Wf is the smallest k + l, for which xk f (x)l has reduced
degree p 1. Wf is very similar to wf above. The second and third statements of
the next lemma can be implicitly found in [91].
Lemma 9.31. (G
acs [66]) (i) For any of the transformations mentioned after Proposition 9.15 for Mf , the value of Wf remains unchanged;
(ii) For any polynomial f , Wf + Nf p + 1;
(iii) For any non-linear polynomial, Wf p1
2 with equality if and only if f (x) is
p+1
2
2
affinely equivalent to x or x
;
p1
(iv) For any polynomial of degree between 2 and p+1
2 , Wf
3 , unless it is
p+1
affinely equivalent to x2 or x 2 .
Nf :
number
of
determined
directions
54
I. Background
A polynomial of the form xk f (x)l will be called a double power of f , the degree of
xk f (x)l is defined to be k + l.
We will need the following result of A. Biro:
Theorem 9.32. (A. Bir
o [28]) Suppose the graph of f GF(p)[X] is contained in
p1
p1
the union of two intersecting lines. Then Wf = p1
2 or Wf = 3 or Wf 4 .
p1
p1
There is an example with 4 > Wf > 5 .
Exercise 9.33. Prove the equalities or compute the value of the symmetric
functions of the interesting sets below:
k (GF(q)) = 0 if k = 1, ..., q 2 or q and = 1 if k = q 1;
tk = 0 if (q 1) 6 |k and = 1 if (q 1)|k;
k =
tGF(q)
P
{t1 ,...,tk ;tk+1 ,...,tk+l }GF(q)
all distinct
t1 t2 ...tk t2k+1 t2k+2 ...t2k+l =?;
j (S) =?
SGF(q)
|S|=k
9.5
Resultants
In this section the classical notion of resultant is reconsidered. The author does
not know who found the generalization considered below. It appears in a theorem
and its corollary by Green and Eisenbud [72], and probably already in Coolidge
[58]; also in Sz
onyi [125] where it was first applied in finite geometry. The method
was refined later in Weiner [137].
As is known from classical algebra, given two polynomials f (X) = a0 X k +a1 xk1 +
... + ak1 X + ak and g(X) = b0 X l + b1 xl1 + ... + bl1 X + bl , not both a0 and
b0 being zero, then their greatest common divisor has degree at least one iff the
determinant of the following matrix of size (l + k) (l + k)
a0
b0
0
...
0
0
0
...
0
R0 (f, g) = R0 =
a1
a2
a0
a1
0
a0
.
.
.
ak
ak1
ak2
ak
ak1
..
.
..
...
...
..
.
...
0
0
b1
b2
b0
b1
.
.
.
bk
bk1
..
.
...
a0
bl1
a1
bl
0
.
.
.
0
...
bl1
bl
...
...
..
.
...
...
...
.
.
.
..
ak1
ak
0
0
...
bl
0
bl1
bl
55
is zero. This matrix is sometimes called the Sylvester matrix Syl(f, g) of f and g;
Q
the determinant of it is the resultant of f and g and is equal to al0 bk0 i,j (i j ) =
Q
Q
bk0 j f (j ) = al0 i g(i ), where {i : i = 1, ..., k} and {j : j = 1, ..., l} are the
roots of f and g, resp., in the algebraic closure. Note also that the resultant is a
homogeneous polynomial of degree l in the variables a0 , ..., ak and of degree k in
b0 , ..., bl .
We will frequently need the following
Theorem 9.34. Let f (X) = a0 X k + a1 xk1 + ... + ak1 X + ak and g(X) = b0 X l +
b1 xl1 + ... + bl1 X + bl , not both a0 and b0 being zero. If their greatest common
divisor has degree m then the next matrix is non-singular, moreover, their greatest
common divisor has degree at least m + 1 iff the determinant of the following
matrix of size (k + l 2m) (k + l 2m)
Rm (f, g) = Rm =
a0
a1
a2
0
a0
a1
0
0
a0
.
.
.
alm1
alm
alm2
alm1
..
.
alm3
alm2
.
.
.
.
.
.
.
.
.
ak+l2m1
...
...
...
...
..
.
...
...
...
0
0
0
b0
b1
b2
0
b0
b1
a0
a1
blm1
blm
blm2
blm1
..
.
akm
bk+l2m1
bk+l2m2
...
...
...
..
.
...
...
.
.
.
..
.
.
.
.
...
0
0
0
b0
0
b0
0
0
b0
.
.
.
blm
is zero. (ai and bj are considered to be zero if they are out of the range 0 i
k, 0 j l.)
Note that when the greatest common divisor of the two polynomials is supposed
to be large then the matrix Rm is small.
Proof: Let r(X) be the greatest common divisor of f and g. Denote by c the
quotient f /r and let d = g/r. We can suppose that r is monic (of degree m), so
c(X) = a0 X km + c1 X km1 + c2 X km2 + ... + ckm
and
d(X)
= b0 X lm + d1 X lm1 + d2 X lm2 + ... + dlm .
In other words, this means that
For these polynomials we have r = f /
c = g/d.
f d g
c = 0 or equivalently
f (d b0 X lm ) g(
c a0 X km ) = a0 X km g b0 X lm f,
()
lm
km
56
I. Background
and this polynomial equation can be interpreted as a system of linear equations

for the coefficients d1 , ..., dlm , c1 , ..., ckm .
More precisely, compute the coefficient of X k+lm1 , X k+lm2 , ..., X m in (), it
yields the system of linear equations
Rm (d1 , ..., dlm ,
c1 , ...,
ckm )T =
(a0 b1 b0 a1 , a0 b2 b0 a2 , ..., a0 bk+l2m b0 ak+l2m )T .
It is important to note that the solutions of the system of linear equations correspond to solutions of the polynomial equation, if we know in advance that
the greatest common divisor r has degree at least m. Indeed, if (d1 , ..., dlm ,
c1 , ..., ckm ) is a solution of our system of linear equations, then the polynomials
c(X) = a0 X km + c1 X km1 + ... + ckm and d(X)

= b0 X lm + d1 X lm1 +
... + dlm will have the property that f d g

c has degree less than m. Since this
polynomial is divisible by r, it must be identically zero.
Cramers rule
Now the polynomial equation (and hence the system of linear equations) has a
unique solution if r has degree exactly m. To see this, first observe that f /r =
c, g/r = d is a solution of the polynomial equation of degree exactly k m, this
means that there is at least one solution. (Remember that r was supposed to be
monic.) Suppose that there are two solutions of the equations, yielding d0 , d00 and
c0 , c00 . Then from f d0 g
c0 = 0 and f d00 g
c00 = 0 we get f (d0 d00 ) g(
c0 c00 ) = 0.
0
00
Now the degree of c c is less than k m; so we get a contradiction after dividing
c0 c00 ) we have that c and d are relatively primes, so e.g. d
by r: in c(d0 d00 ) = d(
0
00
divides (d d ), but the latter has smaller degree so d0 = d00 , and similarly c0 = c00 .
Therefore the solution is unique if deg(r) = m, det Rm 6= 0, and the solution can
be obtained by Cramers rule.
If the degree of r is strictly bigger, say m0 > m, then det Rm is zero, since the
polynomial equation f d g
c = 0 cannot have a unique solution. Indeed, we can
c of degree l m0 and k m0 resp., and then multiply both by the
determine d,
same arbitrary monic polynomial of degree m0 m.
Note that if a0 = b0 = 0 then both det Rm and the right hand side of the equation
() is zero.
In the applications the coefficients aj and bj are polynomials in Y (or homogeneous polynomials in (Y, Z) say), which satisfy deg(aj ), deg(bj ) j. Then, by the
remark on Cramers rule above, the coefficients of d and c will be quotients of two
polynomials, where the denominator is det Rm (Y ). Multiply each coefficient of d
and c by det Rm (Y ) to obtain the polynomials dm (X, Y ) and cm (X, Y ). Then the
following holds.
P
Result
9.35. (Sz
onyi [125]) Suppose that f (X) =
ai (Y )X ni and g(X) =
P
bi (Y )X n1i satisfy deg(ai ), deg(bi ) i and not both a0 and b0 are zero. Then

deg det Rm (Y )
57
the determinant of Rm (Y ) in Theorem 9.34 has degree at most (l m)(k m)

as a polynomial in Y . The total degree of the polynomial cm (X, Y ) is at most
(l m + 1)(k m), and the total degree of dm (X, Y ) is at most (l m)(k m + 1).
Proof: To estimate the degree of the determinant Rm (Y ), observe that the degree
of the (i, j)-th entry (i, j = 1, ..., k + l 2m) is at most i j if we are on the lefthand side, and it is at most i j + l m if we are on the right-hand side of
Rm (Y ). So each term in the expression of the determinant has Y -degree at most
(l m)(k m).
The solutions d1 , ..., dlm ,
c1 , ...,
ckm can be obtained using Cramers rule;
d0 = b0 , c0 = a0 . If we put the right hand side of the equation in place of the j-th
column, then the degree will be bigger than that of Rm by j or j l + m according
as j is in the left or right part of the matrix, from which the second part of the
result follows.
Note that the elements of Rm (Y ) depend on ai (Y ), i k m, and on bj (Y ),
j l m. This is also true for the polynomials am (X, Y ) and cm (X, Y ). Hence
in Result 9.35 it is enough to assume that deg(ai ) i, when i k m, and
deg(bj ) j, when j l m.
How can these observations be applied? Consider two homogeneous polynomials
f (X, Y, Z) and g(X, Y, Z), f (a, b, c) = 0 iff the line [a, b, c] has some (good) property Pf , and g(a, b, c) = 0 iff the line [a, b, c] has some (good) property Pg . We want
to find/count the infinite points (0, c, b) through which there pass > m affine
lines being good in both sense, i.e. for which deg gcd(f (X, b, c), g(X, b, c)) > m
for a certain value m. From Theorem 9.34 it follows that these are all roots of
Rm (Y, Z) = Rm (f (X, Y, Z), g(X, Y, Z)). See Theorem 12.1 for a nice illustration.
cm (X, Y ), dm (X, Y ) seem to appear in the reasonings above
The polynomials c, d,
as by-products; however at some point, e.g. in stability results, they play a crucial
role. The idea will be the following. Like before, we are given f (X, Y, Z) and
g(X, Y, Z), where g is responsible for some bad property of lines (and f for some
f
good property). So we are interested in gcd(f,g)
but this is somewhat meaningless.
f (X,b,c)
Well, then we can calculate gcd(f (X,b,c),g(X,b,c))
, but it does not say anything about
the change of the situation when we alter (b, c). Suppose that there is a so-called
typical index m on the line [1, 0, 0], meaning that for most of the points
(0, b, c), the degree of gcd(f (X, b, c), g(X, b, c)) is exactly m. Then, like in Theorem
9.34, construct Rm (Y, Z) = Rm (f (X, Y, Z), g(X, Y, Z)) and the other polynomials
c(X, Y, Z), cm (X, Y, Z). We have estimates on their degree, and we do know that
f (X,b,c)
for all typical points (so for most of the points of [1, 0, 0]) gcd(f (X,b,c),g(X,b,c))
=
c(X, b, c). So c(X, Y, Z) is zero for most of the (a, b, c)-s we are interested in. The
only problem is that we gained c(X, Y, Z) from Cramers rule, so it is a fraction
with denominator R(Y, Z). Hence if we prefer polynomials / algebraic curves (as we
usually do), we have to consider c(X, Y, Z)Rm (Y, Z) = cm (X, Y, Z) instead. Note
58
I. Background
that in principle we know it precisely, it can be calculated from the coefficient

polynomials ai (Y, Z) and bj (Y, Z) as above.
For illustrations see the proof of Theorem 15.17 in [137].
We want to refine the statement that if the greatest common divisor has degree
bigger than m then det Rm is zero; we will show that as the degree gets bigger,
the determinant becomes more and more zero.
More precisely, we know that if for the fix value Y = y the degree of the greatest
common divisor of f (X, y) and g(X, y) is bigger than m, then the determinant of
Rm (y) is 0; hence (Y y) is a factor of det Rm (Y ). The refinement is the following
Theorem 9.36. For Y = y, if deg(gcd(f (X, y), g(X, y))) = m then
det(Rm (f (X, y), g(X, y))) is non-zero. If deg(gcd(f (X, y), g(X, y))) m + t then
(Y y)t | det(Rm (f (X, Y ), g(X, Y ))).
The homogeneous version is:
P
P
Theorem 9.37. Let f (X, Y, Z) = ai (Y, Z)X ki and g(X, Y, Z) = bi (Y, Z)X li
be homogeneous polynomials in three variables and assume that not both a0 and b0
are zero. For (Y, Z) = (y, z), if deg(gcd(f (X, y, z), g(X, y, z))) = m 0 then
det(Rm (f (X, y, z), g(X, y, z)))
is non-zero. If deg(gcd(f (X, y, z), g(X, y, z))) m + t then
(zY yZ)t | det(Rm (f (X, Y, Z), g(X, Y, Z))).
For the proof the next lemma is needed.
Lemma 9.38. Let M (Y, Z) be an n n matrix with entries being polynomials
depending on Y and Z and s 1. Assume that for the fix value y and z, (zY yZ)s
divides all (n 1) (n 1) subdeterminants of M (Y, Z), then (zY yZ)s+1 is a
factor of det M (Y, Z).
Proof: Let M (Y, Z) be the transpose of the matrix obtained by replacing each
element mij of M (Y, Z) by (1)i+j times the corresponding (n 1) (n 1) subdeterminant of M (Y, Z). Note that M (Y, Z)M (Y, Z) = (det M (Y, Z))I, where I
is the n n identity matrix; hence det M (Y, Z) = (det M (Y, Z))n1 . By assumption the elements of M (Y, Z) are divisible by (zY yZ)s , therefore (zY yZ)sn
divides det M (Y, Z) = (det M (Y, Z))n1 and so the lemma follows.
Proof of Theorem 9.37 As before, we want to determine the coefficients of the
polynomials d(X)
and c(X), for that f (X, y, z)d(X)
g(X, y, z)
c(X) = 0 holds,
59
deg c = k m and deg d = l m. Again this equation can be interpreted as a
system of linear equations with matrix Rm (y, z), for the coefficients of c and d.
Since the degree of the greatest common divisor of f (X, y, z) and g(X, y, z) is
f (X,y,z)
m + t, the polynomials d and c are the products c(X) = gcd(f (X,y,z),g(X,y,z))
(X t +
f (X,y,z)
(X t + u1 X t1 + + ut ), where
u1 X t1 + + ut ) and d(X)
=
gcd(f (X,y,z),g(X,y,z))
every ui can be chosen freely. This means that the (k + l 2m) (k + l 2m)
matrix Rm (y, z) has rank k +l2mt. Hence the (k +l2mt+1)(k +l2m
t + 1) subdeterminants of det Rm (y, z) are all 0. As det Rm (Y, Z) is a homogeneous
polynomial, (zY yZ) divides all (k + l 2m t + 1) (k + l 2m t + 1)
subdeterminants of det Rm (Y, Z). The result follows by applying Lemma 9.38 t
times.
Later we will see how Theorem 9.37 can be used and this is one of the observations
that lead to improvement on several results.
Lets say some words about the connections with Bezouts theorem 10.6. Applying Theorem 9.37 with m = 0 for the homogeneous polynomials f (X, Y, Z) and
g(X, Y, Z), and using the notation my,z = deg gcd(f (X, y, z), g(X, y, z)), i.e. the
common intersection multiplicities with the line [0, z, y], we have that the number of common points of the two curves (even if counted with these multiplicities)
is at most
X
my,z deg det R0 (f, g) = kl.
B
ezout
(y,z)
In fact here the common intersection multiplicities with horizontal lines, i.e. lines
through (1, 0, 0) are added up.
It becomes more interesting when we fix a typical multiplicity m and we want to
estimate the number of (y, z)-s for which f (X, y, z) and g(X, y, z) has more than m
common roots. If this excess is ty,z , i.e. m+ty,z = deg gcd(f (X, y, z), g(X, y, z)),
then Theorem 9.37 gives
excess sum
Theorem 9.39.
X
ty,z deg det Rm (f, g) = (deg(f ) m)(deg(g) m).
(y,z)
ty,z 0
As an application lets consider the case when we are interested in g.c.d.(X q

X, f (X)), so the distinct factors of f . If we look for more than m factors, we need
60
I. Background
det
1
0
0
1
.
.
.
0
0
0
0
..
1
0
.
.
.
.
.
.
.
1
...
...
..
.
...
...
.
.
.
...
0
0
a0
a1
akm1
akm
1
0
0
a0
aq+k2m1
akm2
akm1
..
.
aq+k2m2
...
...
..
.
...
...
.
.
.
..
.
.
.
.
...
0
0
a0
0
a0
0
0
a0
.
.
.
=0
akm
where the -1s do not appear in the lower left corner if q + k 2m 1 < q 1 i.e.
k/2 < m. In this case the determinant is just the determinant of the lower right
submatrix. In any case, if the coefficients ai are polynomials of Y -degree at most
i, then the degree of the determinant above is at most (q m)(k m).
The second very important example is when we work on g.c.d.(f (X), f 0 (X)), i.e.
we look for the multiple factors of f . If we want more than m multiple factors, we
need the determinant of
a0
a1
B
B
.
B
.
B
.
B
B a
B km2
B a
B km1
B
B
B
B
B
B
.
.
B
.
B
B
B
.
.
@
.
a2k2m2
0
...
...
..
.
...
...
..
.
0
0
ka0
(k 1)a1
0
ka0
a0
a1
(m + 2)akm2
(m + 1)akm1
(m + 3)akm3
(m + 2)akm2
..
.
... akm (2m + 2 k)a2k2m2 (2m + 3 k)a2k2m3
1
...
0
...
0
C
C
C
..
C
.
C
C
... ka0 0
0
C
C
...
ka0
0
C
C
.
C
.
C
.
C
C
C
..
C
.
ka0
C
C
C
.
.
.
.
A
.
.
...
(m + 1)akm1
to be zero. Again, if the coefficients ai are polynomials of degree at most i, then

the degree of this determinant is at most (k m)(k m 1).
Additive/linearized polynomials, P -adic resultant
P = pe
Let f (X), g(X) be monic separable GF(pe )-linear polynomials (i.e. each of them
splits into distinct linear factors), and let S = {x1 , ..., xn } and U = {t1 , ..., tm } be
bases for the GF(pe )-vectorspaces of their roots (see Theorem 5.28), respectively.
The P -adic resultant of them is defined as
RP (f, g) =
Example:
pe = 3,
n = m = 1,
x1 = a, t1 = b
f (X)=X 3 a2 X
g(X)=X 3 b2 X
R(
RP
f (X) g(X)
,
) =
X
X
(a2 b2 )2
2
2
(f, g)=b a
Theorem 9.40.
(2) RP (f, g) =
M RD(x1 , ..., xn , t1 , ..., tm )

.
M RD(x1 , ..., xn )M RD(t1 , ..., tm )
e
(1) RP (f, g)p
g(X)
= R( f (X)
X , X );
M RD(f (t1 ),...,f (tm ))

M RD(t1 ,...,tm )
1 ),...,g(xn ))
= (1)nm M RD(g(x
;
M RD(x1 ,...,xn )
(3) RP (f, g) = 0 the GF(pe )-vectorspaces hx1 , ..., xn i and ht1 , ..., tm i have a
non-trivial intersection.
10. Elements of algebraic geometry

Proof:
10
61
Easy (using the Moore determinant formula).
Elements of algebraic geometry
A typical method in Galois geometry is to associate an algebraic curve to a pointset

of the plane or the space. The classical example of this method is Segres theory
of complete arcs or the theory of Redei polynomials.
The usual idea is that investigating a pointset in PG(2, q) we are trying to collect
all the interesting (deviant, irregular) lines in an envelope (so a curve in the
dual plane, or a dual curve), where the meaning of interesting depends on the
special properties of the pointset. E.g. in the case of arcs Segre introduced a dual
curve containing all the tangent lines. He proved the existence of that curve using
Menelaos theorem only; in some other cases we need the explicite form of a curve.
For finding something similar to it we often write up the Redei-polynomial of the
pointset and we consider the Redei polynomial, or another polynomial derived
from it, as an algebraic curve.
For using a very little from the enormous theory of algebraic geometry we collect
a few results here concerning algebraic curves. Some parts of this section are taken
from Sz
onyi [128]. We use a slightly simplified terminology, the interested reader
can find several thousands pages of introductory books on algebraic geometry, we
mention [64], [107] or the recent excellent Hirschfeld-Korchmaros-Torres [81].
An algebraic plane curve defined over a field F is just a polynomial f (X, Y ) in
two variables or a homogeneous polynomial f (X, Y, Z) in three variables, with
coefficients from F. For a constant F, f is considered to be the same curve.
The irreducible components of the curve f correspond to the irreducible factors
of the polynomial f . We may quite often switch back and forth between bivariate
(affine) polynomials and their homogenized 3-variate (projective) forms. Defined
over F, the curve can be viewed over any field G containing F. The points of f over
G are just the pairs (x, y) or homogeneous triples (x, y, z) for which f (x, y) = 0 or
f (x, y, z) = 0, resp., where x, y, z G (so (x, y) AG(2, G) or (x, y, z) PG(2, G)).
Obviously, the same f can have more zeros (points) over a larger field.
Exercise 10.1. Let F = GF(q). How many different curves (projectively inequivalent homogeneous polynomials) of degree n are there?
Exercise 10.2. Let S be a set of
ppoints in AG(2, q). Then there exists a two-variable
polynomial of degree at most 2|S| 1, so that it vanishes at every point of S.
One can write the affine curve f (X, Y ) in the form
f (X, Y ) = f0 + f1 (X, Y ) + f2 (X, Y ) + ... + fn (X, Y ),
minor abuse
notation
of
62
I. Background
where fi is a homogeneous polynomial in two variables of total degree i. Now

the point (0, 0) has multiplicity k on f (X, Y ) if k is maximal such that f can be
written as
f (X, Y ) = fk (X, Y ) + fk+1 (X, Y ) + ... + fn (X, Y ),
so f0 = f1 = ... = fk1 = 0. Multiplicity of another point P (a, b) can be
defined translating it to the origin, i.e. we can examine f (X + a, Y + b) instead.
The tangents of f (X, Y ) at (0, 0) are the k linear factors of fk (X, Y ) above,
tangents at another
they may well be defined over the algebraic closure F;
point can be defined translating it to the origin. A point is simple if it has
multiplicity 1. At a simple (projective) point (x, y, z) the (unique) tangent line
is [(X f )(x, y, z), (Y f )(x, y, z), (Z f )(x, y, z)]. Equvalently, (x, y, z) is a singular
point iff ((X f )(x, y, z), (Y f )(x, y, z), (Z f )(x, y, z)) = (0, 0, 0).
As the following result shows, a curve can have only a restricted number of points
with multiplicity more than one.
Result 10.3. [64, p.118] Let F be a projective plane curve of degree n without
P mP (mP 1)
multiple components. Then for the multiplicities mP of its points
2
P F

n
2 holds.
One can define the intersection multiplicity (denoted by I(f g; P ) or I(P ; f g))
of two curves f and g at a point P (x, y, z) of the plane. The direct definition is
rather complicated in the general case; here is a list of some easy facts which will
be enough for our purposes:
(a) If at least one of f and g does not go through P then obviously I(f g; P ) = 0.
(b) If P is a simple point on both f and g and the tangents at P are distinct
then I(f g; P ) = 1. In particular, if f =
6 g are linear, meeting at P , then
I(f g; P ) = 1.
(c) P is not on a common component of f and g I(f g; P ) < .
(d) I(f g; P ) = I(g f ; P ).
(e) I(f g; P ) = I(f g + hf ; P ).
(f) If g is of degree 1 (i.e. a line) then I(f g; P ) can be calculated easily: one
can express one of the variables from g(X, Y, Z) = 0 and substitute it into f ;
then the resulting homogeneous polynomial in two variables will vanish at P ,
the multiplicity of this root will be the intersection multiplicity. Note that, as
a univariate polynomial of degree n has at most n roots (and the same holds
for homogeneous polynomials in two variables), this intersection multiplicity
63
is at most the degree of f , except when after the substitution f vanishes

identically, which means that g was a factor of f . In this case (and generally
when f and g have a common factor through P ) we say I(f g; P ) = .
(g) Note that if g is linear and it happens to be a tangent of f at P then
I(f g; P ) 2.
(h) I(f (g1 g2 ); P ) = I(f g1 ; P ) + I(f g2 ; P ).
(b) A more general form of (b) is: I(f g; P ) mP (f )mP (g), where mP is the
multiplicity of P on the curve in question.
We note that there is a unique intersection multiplicity satisfying these properties
(and being invariant under projective transformations), so they can be used to
define intersection multiplicity. Also note that these properties can be used to
actually compute the intersection multiplicity of two curves. Let us briefly recall
how this works for the origin. Since any point can be transformed to the origin,
this approach works for other points, too.
We wish to compute I(P ; F G), where P = (0, 0). Let f (X) = F (X, 0),
g(X) = G(X, 0) and d(X) = g.c.d.(f (X), g(X)). Then we can write d(X) =
a(X)f (X) + b(X)g(X) by the Euclidean algorithm. Put f1 (X) = f (X)/d(X),
g1 (X) = g(X)/d(X). Then f1 (X)a(X) + g1 (X)b(X) = 1, which means that

f1 (X) g1 (X)
G(X, Y )
R(X, Y )
=
b(X) a(X)
F (X, Y )
U (X, Y )
is an invertible linear transformation. Hence the ideals (F, G) and (H, U ) are the
same, so I(P ; F G) = I(P ; H U ). Since Y divides R(X, Y ) = f1 (X)G(X, Y ) +
g1 (X)F (X, Y ), so we can use property (h) of the intersection multiplicity. This
process can be continued and after some steps we can recognize that a certain
intersection multiplicity is 0, using property (a).
As an illustration the following useful lemma can be proved, which can be found
in [107], Lemma 9.2, see also p. 87 in [107].
Lemma 10.4. Let P be the origin.
(1) If I(P ; F Y ) = r, I(P ; G Y ) = s and r s, then I(P ; F G) s.
(2) If r > s and P is a simple point of F , then I(P ; F G) = s.
Proof: We only prove (1). In determining the intersection multiplicity with the
approach indicated above, then we have X s |d(X). So when we compute I(P ; H U )
with H = Y H1 , we get I(P ; Y H1 U ) = I(P ; Y U ) + ..., where ... 0. Now put
U = U1 (X) + Y U2 (X, Y ). Then I(P ; Y U ) = I(P ; Y U1 ). Since the terms of U
that do not contain Y come from similar terms of F and G, and d(X) is divisible
by X s , X s divides U1 (X). But then I(P ; Y U1 (X)) I(P ; Y X s ) = s. (2) is
64
I. Background
left as an exercise.
Exercise 10.5. Prove case (2) above!

In the applications we need bounds on the common points of two curves, and
bounds on the number of points on a single curve.
Theorem 10.6. (Bezout) If f and g has no common component then the number of their common points, counted with intersection multiplicities, is at most
always equality holds.
deg(f ) deg(g). Over the algebraic closure F
The proof of Bezouts theorem is based on resultants, compare to Section 9.5.

Now let F = GF(q). How many points can a curve f GF(q)[X, Y, Z] of degree n
have over GF(q)? Denote this number by Nq = Nq (f ).
Theorem 10.7. Hasse-Weil, Serre For an absolutely irreducible non-singular algebraic curve f GF(q)[X, Y, Z] of degree n we have
|Nq (f ) (q + 1)| gb2 qc (n 1)(n 2) q.

In the theorem g denotes the genus of f , we do not define it here. Note that Nq
counts the points of f with multiplicities, also that for GF(q) GF(q1 ) we have
Nq (f ) Nq1 (f ).
It happens that some absolutely irreducible component of f cannot be defined over
GF(q) (but it still has some GF(q)-rational points, i.e. points in PG(2, q)). Then
the following bound can be used:
Exercise 10.8. Prove that for an absolutely irreducible algebraic curve
f GF(q)[X, Y, Z] of degree n, that cannot be defined over GF(q), we have
Nq (f ) n2 .
In some cases the Hasse-Weil bound can be changed for a better one, for example
when q = p is a prime number.
Theorem 10.9. St
ohr-Voloch [111] For an irreducible algebraic curve f
GF(q)[X, Y, Z], q = ph , of degree n with Nq rational points over GF(q) we have
(i) if n > 1, not every point is an inflexion and p 6= 2 then Nq 12 n(q + n 1);
(ii) if n > 2, not every point is an inflexion and p = 2 then Nq 12 n(q + 2n 4);
(iii) if q = p and n > 2 then Np 2n(n 2) + 25 np;
(iv) if q = p, 3 n 12 p and f has s double points then Np 25 n(5(n2)+p)4s.
In (i) the condition is automatically satisfied if q = p is a prime.
65
Exercise 10.10. Let f (X) be a polynomial of degree at most 4 q, (q odd), which

assumes square elements of GF(q) only. Prove that f = g 2 for a suitable polynomial
g(X).
10.1
Conditions implying linear (or low-degree) components
In the applications it is typical that after associating a curve to a certain set (in
principle), the possible linear components of the curve have a very special meaning
for the original problem. Quite often it more or less solves the problem if one can
prove that the curve splits into linear factors, or at least contains a linear factor.
Here some propositions ensuring the existence of linear factors are gathered. The
usual statement below considers the number of points (in PG(2, q)) of a curve.
We will use two numbers: for a curve C, defined by the homogeneous polynomial
f (X, Y, Z), Mq denotes the number of solutions (i.e. points (x, y, z) PG(2, q))
for f (x, y, z) = 0, while Nq counts each solution with its multiplicity on C.
Exercise 10.11. Barlotti-bound Prove that if a curve of order n has no linear
factors over GF(q) then Nq (n 1)q + n.
Exercise 10.12. Prove that a curve of degree n defined over GF(q), without linear
components, has always Nq (n 1)q + n2 points in PG(2, q). Hint: let k be
maximal such that every tangent of the curve contains at least k points of the the
curve (counting without multiplicity, in PG(2, q)). Show that (i) Nq (n 1)q + k;
(ii) Nq (n 1)q + (n k).
Conjecture 10.13. We conjecture that a curve of degree n defined over GF(q),
without linear components, has always Nq (n 1)q + 1 points in PG(2, q).
For n = 1, 2, q + 1, q 1 it would be sharp as the curves X 2 Y Z, X q+1 +
Y q+1 + Z q+1 and X q1 + Y q1 ( + )Z q1 (where , , + 6= 0) show.

It can be called the Lunelli-Sce bound for curves, see Section 25.
Note that the conjecture is true
(i) if there exists a line skew to the curve and (q, n) = 1, see Corollary 25.6;
(ii) if n q + 1 then q + 1 + (n 1)(n 2) q nq q + 1 proves it by Weils

bound Theorem 10.7;
(iii) if the curve has a singular point in PG(2, q);
(iv) if n q + 2.
The statement (ii) can be proved by induction: if C has more points then it cannot
be irreducible, so it splits to the irreducible components C1 C2 ... Ck with
degrees n1 , ..., nk ; if each Ci had (ni 1)q + 1 points then in total C would have
Pk
i=1 (ni q q + 1) = nq k(q 1) < nq q + 1 points. So at least one of them,
Mq Nq
66
I. Background
Cj say, has more than nj q q + 1 points. By Exercise 10.8 Cj can be defined over
GF(q) and Weil does its job again.
For (iii) the Barlotti-bound, recounted looking around from a singular point, will
work.
The truth of the conjecture would also mean that the counterexamples for the
Lunelli-Sce cenjecture are not pointsets of curves, see Section 25, Corollary 25.6.
Let Cn be a curve of degree n defined over GF(q), with at least qn affine points
(counted with multiplicity). (In fact its enough to have qn points, see below).
We want to prove that it has a linear component, then repeat everything for n 1
and (n 1)q. The next proposition, which may be omitted in the final version,
deals with curves of high degree as well.
tq 2 < nq;
(t + 1)q n + (t +
1) 0
Proposition 10.14. A curve of degree n defined over GF(q), without linear components, has always Nq nq points in AG(2, q). More precisely, if tq < n (t + 1)q
then Nq max{tq 2 , (q + 1)(n t 1)}.
(For t = 0 it gives Nq (n 1)q + n 1.)
The key idea is that if you find a line intersecting Cn in at least n + 1 points then
it is a (linear) component by Bezout. If not then Cn is a (k, n)-arc and one can
estimate its cardinality.
Let tq < n (t + 1)q and suppose that Cn is a curve with
Nq > max{tq 2 , (q + 1)(n t 1)}
points. Then Nq > tq 2 , so there is a point P of maximal multiplicity m t + 1.
Consider a tangent line ` through P , then I(Cn `, P ) m + 1, so there are at
most n m 1 further points on ` \ P . As there exist at least m tangents through
P , we have
Nq m + (q + 1)(n m) m = (q + 1)(n m) (q + 1)(n t 1),
which is a contradiction.
The following lemma is a generalization of a result by Szonyi. For d = 1 it can be

found in Sziklai [113], which is a variant of a lemma by Szonyi [124].
Lemma 10.15. Let Cn , 1 d < n be a curve of order n defined over GF(q), not
containing a component defined over GF(q) of degree d. Denote
by N the number
d(d1) q
1
of points of Cn in PG(2, q). Choose a constant d+1 + (d+1)(q+1) . Assume that
n q 1 + 1. Then N n(q + 1).
67
1+d(d1) q
1
+ (d+1)q , n < q d + 2, N < nq. Here = d1
It works also with > d+1
can be written (that is often needed) when d 6 q.

Proof: Suppose first that Cn is absolutely irreducible. Then Weils theorem ([136],
[77]) gives N q + 1 + (n 1)(n 2) q n(q + 1). (The latter inequality, being
quadratic in n, has to be checked for n = d + 1 and n = q 1 + 1 only.)

If Cn is not absolutely irreducible, then it can be written as Cn =P
Di1 ... Dis ,
s
where Dij is an absolutely irreducible component of order ij , so j=1 ij = n. If
2
Dij can not be defined over GF(q), then it has at most Nij (ij ) ij (q + 1)
points in PG(2, q) (see Ex. 10.8). If Dij is defined over GF(q), then the Weil-bound
implies again that Nij ij (q + 1). Hence
N=
s
X
j=1
N ij
s
X
ij (q + 1) = n(q + 1).
j=1
For applications see Section 11, Theorem 23.1, Theorem 23.2 and Theorem 27.8.
Exercise 10.16. Prove that if q = p is prime and > 52 then in the theorem above
n ( 12 15 )p + 2 is enough for N n(p + 1).
In the applications quite often happens that we do not know the number of points
of a certain curve but we know the numbers of intersections with a pencil of lines
(each intersection counted with intersection multiplicity).
Lemma 10.17. Let C, 1 d < n be a curve defined by the homogeneous polynomial
F GF(q)[X, Y, Z], deg(F ) = n. Suppose that C does not contain a component
defined over GF(q) of degree d. Assume that C does not contain (multiple components nor) components with vanishing partial derivative with respect to X. Let
`1 , ..., `q+1 be the lines of the pencil centered at P (1, 0, 0). Denote by N0 the number of intersection points of C and the lines {`1 , ..., `q+1 }, each point on `j counted
with the intersection multiplicity of `j and
C (but we do not count P even if it is
d(d1) q
1
on C). Choose a constant d+1
+ (d+1)(q+1) . Assume that n q 1 + 1.
Then N0 n(q + 1) + n(n 1).
Proof: The only difference here, compared to the situation of Lemma 10.15, is
that N0 can be bigger than N , the number of points of C. As before we can count
componentwise. The difference N0 N comes from points where the tangent of
an irreducible component Gij of the curve is horizontal, that is the intersection
points of Gij and X Gij . If X Gij is not zero and deg Gij = ij then by Bezouts
theorem ij (i
Pj 1) is an upper bound for the number of points in Gij X Gij . So
N0 N + j ij (ij 1) n(q + 1) + n(n 1) .
Exercise 10.18. (De Beule - G
acs) Let I consist of all the linear combinations of the
monomials X p+1 , Y p+1 , Z p+1 , X p Y, XY p , Y p Z, Y Z p , Z p X, ZX p . Suppose that f
p+1
is a homogeneous quadratic polynomial for which f 2 is in I. Then f is reducible.
68
11
I. Background
Finding the missing factors, removing the surplus

factors
Here we treat a very general situation, with several applications in future.

Qq
Given A = {a1 , ..., aq } GF(q), all distinct, let F (X) = i=1 (X ai ) be their
root polynomial. We would like to find Q
the missing elements {aq+1 , ..., aq } =
q
GF(q) \ A, or, equivalently, G (X) = i=q+1 (X ai ). Obviously, G (X) =
q
X X
q
F (X) , so F (X)G (X) = X X. Expanding this, and introducing the elementary
symmetric polynomials
j = j (A), k = k (GF(q) \ A),
we get X q X =
(X q 1 X q1 + 2 X q2 ... q1 X q )(X 1 X 1 + 2 X 2
... 1
X ),
from which j can be calculated recursively from the k -s, as the coefficient of
X qj , j = 1, ..., q 2 is 0 = j + j1
1 + ... + 1 j1 + j ; for example
1 = 1 ; 2 = 12 2 ; 3 = 13 + 21 2 3 ; etc.
(1)
Note that we do not need to use all the coefficients/equations, it is enough to do

it for j = 1, ..., . (The further equations can be used as consequences of the fact
that the ai -s are pairwise distinct, see e.g. Theorem 25.7 and Lemma 25.8.)
The moral of it is that the coefficients of G (X) can be determined from the
coefficients of F (X) in a nice way.
***
Let now B = {b1 , ..., bq+ } GF(q) be a multiset of elements of GF(q), and let
Qq+
F (X) = i=1 (X bi ) be their root polynomial. We would like to find
Q the surplus
elements {bk1 , ..., bk } = B \ GF(q), or, equivalently, G(X)

= i=1 (X bki ).
q
Obviously, G(X)
= XFq(X)
X , so F (X) = (X X)G(X). Suppose that q 2.
Expanding this equation and introducing the elementary symmetric polynomials
j = j (B),
k = k (B \ GF(q)),
we get X q+ 1 X q+1 + 2 X q+2 ... 1 X q+1 X q ...
... q+1 X q+ = (X q X)(X
1 X 1 +
2 X 2 ...
1 X
) =
= X q+
1 X q+1 +
2 X q+2 ...
1 X q+1
X q + terms of lower degree.
if q 1 then
its slightly more
complicated
From this
j can be calculated even more easily then in the previous case:
k = k for all k = 1, ..., .
(2)
11. Finding the missing factors, removing the surplus factors
69
***
In both case suppose now that instead of the elements {ai } or {bj } we have
(for example) linear polynomials ci Y + di and a set S GF(q) such that for each
y S the set Ay = {ci y + di : i} consists of pairwise distinct elements of GF(q),
or, similarly, the multiset By = {ci y + di : i} contains GF(q). Then the k -s in the
reasonings above become polynomials in Y , with degY (k ) k. Now one cannot
speak about polynomials k (Y ) (or
k (Y ), resp.) as there is no guarantee that the
missing values (or the surplus values) for different y-s can be found on lines. So
y (X) or
first we define k (y) (or
k (y), resp.), meaning the coefficient of X k in G
Gy (X), so the elementary symmetric function of the missing (or surplus) elements
when substituting Y = y S. However, the equations for the k -s or
k -s are still
valid. So one may define the polynomials analogously to (1):
def
def
1 (Y ) = 1 (Y ); 2 (Y ) = 12 (Y ) 2 (Y );
def
3 (Y ) = 13 (Y ) + 21 (Y )2 (Y ) 3 (Y ); etc.
or analogously to (2):
def
k (Y ) = k (Y ) for all k = 1, ...,

with the help of them. Note that from the defining equations it is obvious that
degY k (Y ) k and degY
k (Y ) k.
Now we can define the algebraic curve
def
G (X, Y ) = X 1 (Y )X 1 + 2 (Y )X 2 ... 1
(Y )X (Y )
or in the other case

def
G(X,
Y ) = X
1 (Y )X 1 +
2 (Y )X 2 ...
1 (Y )X
(Y ).
As before, for each y S we have that the roots of G(X, y) are just the missing
(or the surplus) elements of Ay or By , resp. Our aim is to factorize G (X, Y ) or
G(X,
Y ) into linear factors X (i Y + i ). To do so, observe that G (X, Y ) has
many points in GF(q) GF(q): for any y S we have solutions of G (X, y) = 0,
i.e. the missing values after substituting Y = y in the linear polynomials ci Y +di ,
so after determining the sets Ay . So G (X, Y ) has at least |S| points.
A similar reasoning is valid for G(X,

Y ). If it splits
P into irreducible components
G = G1 G2 Gr , with deg Gi = degX Gi = i ,

i = , then for any y S,
the line Y = y intersects Gi in i points, counted with intersection multiplicity.
So the number of points on Gi is at least i |S| i (i 1), where the second
term stands for the intersection points of Gi and X Gi , where the intersection
multiplicity with the line Y = y is higher than the multiplicity of that point on
Gi . So,
Punless some Gi has zero partial derivative w.r.t. X, we have that G has at
least
i |S| i (i 1) |S| ( 1) points.
70
I. Background
Now we can use Lemma 10.15, Lemma 10.17 (or any similar result) repeatedly,
with d = 1, it will factorize G(X, Y ) into linear factors of the form X (j Y + j )
if deg G(X, Y ), which is at most in our case, is small enough, i.e. if < q and
1+ 4 q
1+ 4 q
|S| > max{ q , 12 }(q+1) in the first and |S| > max{ q , 12 }(q+1)+1
in the second case.
It means, that one can add linear polynomials i Y +i in the first case such that
for any y S, the values {ci y + di } {j y + j } = GF(q). In the second case we
have a weaker corollary: for any y S, the values {ci y + di } \ {j y + j } = GF(q),
which means that adding the new lines j Y + j with multiplicity= 1 then
S GF(q) is covered exactly once. (What we do not know in general, that these
lines were among the given q + lines, so whether we could remove them.)
Finally, these lines (or similar objects), covering S GF(q) usually have some
concrete meaning when applying this technique; this book contains several such
applications, see Theorem 27.4, Exercise 27.6, etc.
The arguments above are easy to modify when we change some of the conditions,
for example when ai or bi is allowed to be some low degree (but non-linear) polynomial of Y .
Exercise 11.1. Use the second (surplus) case above to prove that a blocking
set B PG(2, q) with |B AG(2, q)| = q + 1 affine points always contains a(n
affine) point that is unnecessary (i.e. it can be deleted without violating the blocking
property).
Exercise 11.2. Let fi (T ), i = 1, ..., q be polynomials of degree at most d, and
suppose that their graphs {(t, fi (t)) : t GF(q)} are pairwise distinct. Prove that
if < ... then one can find fq+1 (T ), ..., fq (T ), each of degree at most d such that
the graphs of these q polynomials partition the affine plane.
Exercise 11.3. Let fi (T ), i = 1, ..., q be polynomials, each from a subspace U
of GF(q)[T ] with 1 U , and suppose that their graphs {(t, fi (t)) : t GF(q)} are
pairwise distinct. Prove that if < ... then one can find fq+1 (T ), ..., fq (T ), each
from U , such that the graphs of these q polynomials partition the affine plane.
Chapter 2
Polynomials in geometry
12
12.1
Lines meeting a pointset

Prescribing the intersection numbers
Suppose that a function m : L N is given, where L is the set of lines of PG(2, q).
The problem is to find conditions, necessary and/or sufficient, under which we can
find a pointset S such that |S `| = m(`) for all ` L.
Note that one can pose the similar question for any hypergraph.
If A denotes the incidence matrix of the plane, m = (m(`1 ), m(`2 ), ..., m(`q2 +q+1 ))
is the weight-vector, then the problem is reduced to finding a (characteristic
vector) v such that Av = m. As A is non-singular, we have v = A1 m. Now
one can turn the question around: for which m will v be of the required type, for
example a non-negative, integer or 0-1 vector?
It is easy to compute that
A1 =
1 T
1
A
J.
q
q(q + 1)
We also know that the eigenvalues of A are q+1 (with multiplicity 1 and eigenvector
(1, 1, ..., 1)); q and q (their multiplicities depend on the order of points and
lines we choose).
In most cases m is not given, we know some of its properties only. It means that
a certain set M of weight-vectors is given (for example, the set of all vectors with
each coordinate from a small fixed set of integers, say {0, 1, 2}); and we want to
know some property (for example, the possible Hamming-weight, i.e. the number
of nonzero coordinates) of (0-1) vectors v satisfying Av M .
71
72
12.2
II. Polynomials in geometry
A general result
Metsch conjectured [95] and Weiner [130] proved the following

Theorem 12.1. Let B PG(2, q) be a pointset of size b on r lines of a pencil
centered at P 6 B. Then the number of lines meeting B in at least one point is at
most
b(q + 1) (r(b r + 1) 1).
It can be reformulated as: the points of B determine at least r(b r + 1) 1
lines, where a line containing t 2 points of B is counted with multiplicity t 1.
First of all observe that there are point sets for which the given bound is sharp.
Assume that r 1 is the order of a subplane in PG(2, q) and let B be a proper
subset of containing r collinear points (i.e. a subline), hence it blocks all the lines
of . So the number of lines meeting B is ((r1)2 +(r1)+1)+|B|(q+1r), where
the first part is the number of lines of , the second is the number of tangents
through the points of B. The number of lines meeting B through a point P of
\ B is r and so for such a point the bound in the conjecture is sharp.
Note that this statement is stronger than the following well known result of Jamison [84] and Brouwer and Schrijver [49].
Theorem 12.2. (Jamison, Brouwer and Schrijver) A blocking set in AG(2, q) contains at least 2q 1 points.
Proof: Assume to the contrary that there is a blocking set B in AG(2, q) so that
|B| 2q 2. Embed AG(2, q) in PG(2, q) and let P be and ideal point. Now the
value r in Theorem 12.1 above is q and so the total number of lines meeting B is
at most 1 + qq + (|B| q)(q + 1 q) q 2 + q 1; which is a contradiction, since
B blocks all the q 2 + q affine lines.
and many other
Note that there are blocking sets of size 2q 1, e.g. the points of two intersecting
lines.
Exercise 12.3. Prove Jamisons theorem using the Combinatorial Nullstellensatz
(Theorems 5.16, 5.17).
Exercise 12.4. [123] If the smallest blocking set of AG(2, q) has size r then in
PG(2, q) the smallest blocking set, which contains two different minimal blocking
sets, is of size r + 2, and vice versa.
Note that it implies a kind of stability of blocking sets of size 2q.
There are blocking sets of size less than 2q 1 in some non-Desarguesian affine
planes of order q, see [54]; which shows that a purely combinatorial proof cannot
be given for neither Theorem 12.2 nor Theorem 12.1. To prove Theorem 12.1 the
following lemma is essential.
12. Lines meeting a pointset
73
Exercise 12.5. (Blokhuis and Brouwer, [44]) Let B be a blocking set in PG(2, q)
and let P be an essential point of B. Then there are at least 2q + 1 |B| tangents
through P .
Lemma 12.6. Let ` = [1, 0, 0] be called now the line at infinity in PG(2, q) and let
S be a point set in PG(2, q)\` . Assume that through the ideal point (0, z, y) there
pass t (affine) lines meeting S. Denote by nt+h the P
number of ideal points through
qt
that there pass exactly t + h lines meeting S. Then h=1 hnt+h (|S| t)(q t).
Proof: For the points of S write {(ai , bi , 1)} and consider the three-variable
Q|S|
Redei polynomial of S, that is R(X, Y, Z) =
i=1 (X + ai Y bi Z) =
P|S|
|S|j
. Recall that deg rj = j. It follows from the fundamental
j=0 rj (Y, Z)X
property of the Redei polynomial, that degX gcd(R(X, y, z), X q X) = t.
For the polynomials R and X q X construct the matrix Rt (Y,P
Z) introduced
qt
in Section 9.5. By Theorem 9.39, for the excess sum in question h=1 hnt+h
deg(det Rk (Y, Z)) (|S| t)(q t).
Proof of Theorem 12.1: For the line at infinity ` choose an m-secant of

B, m > 0, passing through P . Note that now the line at infinity meets
B. Again denote by n(r1)+h the number of ideal points through that there
pass exactly (r 1) + h affine lines meeting B. Let us sum up the number
of affine
through the ideal
points, in total we get at most
lines meeting B P
q(r1)
qm+ (q + 1 m)(r 1) + h=1
hn(r1)+h ; where the first part corresponds
to the points of ` B, the second to the points of ` \ B. The result follows from
Lemma 12.6 immediately as 1+qm+(q+1m)(r1)+(bm(r1))(q(r1)) =
b(q + 1) (r(b r + 1) 1).
It has a nice (combinatorial) corollary in the 3-space:

Corollary 12.7. [95] Suppose B is a set of b q + 1 points in PG(3, q). Then B
meets at most q 2 + q + 1 + bq 2 lines, with equality if and only if B is a planar
blocking set.
Exercise 12.8. Formulate and prove the analogue of the corollary in the plane
(where a combinatorial proof can be given easily).
12.3
Another immediate corollary
An m-fold blocking set in AG(2, q) is a set of points intersecting each line in at

least m points. Blokhuis ([31]) showed that an m-fold blocking set S in AG(2, q),
where (m, q) = 1, has at least (m + 1)q 1 points. Later Ball ([6]) extended this
74
result to arbitrary m, he showed that if e(m) is the maximal exponent such that
pe(m) |m, then |S| (m + 1)q pe(m) , see Corollary 24.25.
In this subsection we show that Theorem 9.39 immediately implies a lower bound
on the size of an m-fold blocking set in AG(2, q). In general, except when m|q,
this result is weaker than Balls result.
Corollary 12.9. (Sz
onyi-Weiner [130]) The size of an m-fold blocking set in AG(2, q)
is at least (m + 1)q m.
Proof: Assume to the contrary that there exists an affine m-fold (not necessarily
minimal) blocking set B of size (m + 1)q m 1. Let ` be an (m + k)-secant,
k 0, where |B| (m + k) 6= qm. Such a line ` can be chosen, since counting
the points of B on the lines through a point of B and on the lines through an
affine point not in B shows that the intersection numbers of B with lines take
at least two different values. Change the coordinate system, so that ` is the ideal
line and () B. Now B contains at least m points from each line, except from
the old line at infinity that is skew to B. Denote this line by `0 and by (y`0 ) the
ideal point of it in this new coordinate system. Let U = B \ ` = {(ai , bi )}i and
Q|B|(m+k)
consider the Redei polynomial of U , that is R(X, Y ) = i=1
(X + ai Y
P|B|(m+k)
bi ) = j=0
rj (Y )X |B|(m+k)j . By the properties of the Redei polynomial,
degX gcd(R(X, y`0 ), (X q X)m ) = m(q 1) and for any (y) ` \ (B (y`0 )),
degX gcd(R(X, y), (X q X)m ) = mq. For the polynomial R and (X q X)m
and for the value s = max(degX R, qm) m(q 1), construct the matrix Rs
introduced in Section 9.5. By Result 9.34, the determinant of this matrix is not
zero. Furthermore, similarly as it is in the proof of Lemma 12.6, one can show that
deg(det Rs ) m(q m k 1). This is a contradiction, since by Theorem 9.39,
m(q m k) deg(det Rs ).
13
Sets with constant intersection numbers mod p
In [34, 40] the following is proved:

Proposition 13.1. Let S be a pointset in AG(2, q), suppose that every line intersects
S in 1 (mod p) points, or is completely disjoint from S. Then |S| q p + 1.
Proof: Counting the points of S on the lines through some fixed point s S we
have |S| 1 (mod p). After the AG(2, q) GF(q 2 ) identification define
X
f (X) =
(X s)q1 ,
sS
it is not identically zero as the coefficient of X q1 is 1. Note that for x GF(q 2 )

the value of (x s)q1 depends on the direction of the line joining x and s. If
13. Sets with constant intersection numbers mod p
75
x S then every direction will occur with multiplicity divisible by p, hence all the
points of S are roots of f , which is of degree q 1. The biggest value 1 mod p
below q is q p + 1.
There are examples of sets like in the statement above, e.g. some (1 mod p)
collinear points, or a projective subplane of order < q completely contained in

AG(2, q).
Exercise 13.2. Prove that in Proposition 13.1 it does not make any difference if S
is allowed to be a multiset.
We remark that the projective case is totally different: there are very big pointsets
in PG(2, q) with 1 (mod p) -secants only, e.g. the plane itself, or a unital, etc.
Exercise 13.3. (Blokhuis) Prove the following generalization: Let S be a pointset
in AG(n, q), n 1, and suppose that every hyperplane intersects S in 1 (mod p)
points, or is completely disjoint from S. Then |S| q p + 1.
Exercise 13.4. Let S be a pointset in AG(2, q) and suppose that for some
1 r p 1, every line intersects S in r (mod p) points, or is completely
disjoint from S. Then |S| q p + r. Why does this statement become not very
interesting after proving if ?
The situation is different if we consider the projective plane.
Theorem 13.5. Given a pointset S = {(ai , bi , ci ) : i = 1, ..., s} = {(ai , bi , 1) : i =
1, ..., s1 } {(aj , bj , 0) : j = s1 + 1, ..., s} PG(2, q), the following are equivalent:
(i) S intersects each line in r mod p points for some fixed r;
(ii) G(X, Y, Z) =
P|S|
i=1 (ai X
(iii) for all 0 k + l q 1,
+ bi Y + ci Z)q1 0;
k+l
k
6= 0 mod p, we have
P|S|
i=1
aiq1kl bki cli = 0.
here 00 = 1

Ps1 k l
(iv) for all 0 k + l q 2, k+l
i=1 ai bi = 0
kP 6= 0 mod p, we have
s
q1m m
and for all 0 m q 1,
a
b
=
0.
i
i=1 i
Proof: Note that if each line intersects S in r mod p points then |S| r (mod p).
So let r be defined by |S| r (mod p). If each line [x, y, z] intersects S in r mod
p points then (mod p) |S| r 0 terms (ai x + bi y + ci z)q1 will be 1 in G(x, y, z)
hence G(x, y, z) = 0. As deg G q 1 we have (i)(ii). One can turn it around:
if G(x, y, z) = 0 then the number of terms (ai x + bi y + ci z)q1 with nonzero (i.e.
=1) value should be zero mod p, so (ii)(i).
count |S| from a

point not in S
76
For the rest consider the coefficient of X q1kl Y k Z l in G (for 0 k + l q 1),

it is

|S|
q 1 k + l X q1kl k l
ai
bi ci = 0
k+l
k
i=1
if (ii) holds and vice versa. Finally (iii)(iv) is obvious.
Many interesting pointsets (small blocking sets, unitals, maximal arcs, even
pointsets, in particular (0, 2, t)-arcs and hyperovals) have constant modulo p intersection numbers with lines; we may give the name (generalized) Vandermonde
set to such sets. (We recall here Section 9.3.)
the
rest
does
not count in the
power sum
Take the affine part of a Vandermonde-set, i.e. points with ci 6= 0 and suppose
that all its points are written as (ai , bi , 1). After the AG(2, q) GF(q 2 ) identification this point becomes ai + bi for some generator of GF(q 2 ). Substituting
(1, , Z) into G we get
X
0 = G(1, , Z) =
(ai ,bi ,1)S

q2
X
k=0
Z q1k
((ai + bi ) + Z)q1 +
(ai ,bi ,1)S
(aj + bj )q1 =
(aj ,bj ,0)S
(ai + bi )k
(ai + bi )q1
(ai ,bi ,ci )S
which means that the affine part of a (generalized) Vandermonde set, considered
as a set in GF(q 2 ), has power sums equal to zero for exponents 1, ..., q 2. (The
last, constant term is just G(1, , 0) = 0.)
13.1
Sets with intersection numbers 0 mod r
Here we continue the examination of sets of PG(2, q) intersecting every line in a

constant number of points mod p. This section is based on [15]. For the more
general k-dimensional case see Section 26. The proofs in both sections are similar
and are streamlined and then generalised versions of the proof in [14].
Theorem 13.6. Let 1 < r < q = ph . A pointset S PG(2, q) which is incident with
0 mod r points of every line has |S| (r 1)q + (p 1)r points and r must divide
q.
Proof: Let us first see that r divides q. By counting the points of S on lines
through a point not in S we have that |S| = 0 mod r. By counting points of S on
lines through a point in S we have |S| = 1 + (1)(q + 1) mod r and combining
these two equalities we see that q = 0 mod r.
77
Assuming |S| < r(q + 1) (for if not the theorem is proved) there is an external
line to S, so we can view S as a subset of GF(q 2 ) ' AG(2, q) and consider the
polynomial
R(X, Y ) =
(X + (Y b)q1 ) =
|S|
X
j (Y )X |S|j .
j=0
bS
For all y, b and c GF(q 2 ) the corresponding points of AG(2, q) are collinear if and
only if (y b)q1 = (y c)q1 and each factor X + (y b)q1 of R(X, y) divides
X q+1 1 whenever y 6= b.
For y S we have
R(X, y) = X(X q+1 1)r1 g1 (X)r ,
and for y 6 S
R(X, y) = g2 (X)r .
In both cases j (y) = 0 for 0 < j < q and r does not divide j. The degree of j is
at most j(q 1) and there are q 2 elements in GF(q 2 ), hence j 0 when 0 < j < q
and r does not divide j. So
R(X, Y ) = X |S| +r X |S|r +2r X |S|2r +...+q X |S|q +q+1 X |S|q1 +...+|S| .
For all y GF(q 2 ) we have
R
(X, y) =
Y
X (y b)q2
X + (y b)q1
!
R(X, y).
bS
In all terms the denominator is a divisor of X q+1 1 so multiplying this equality

by X q+1 1 we get an equality of polynomials and we see that
R(X, y) | (X q+1 1)
R
(X, y),
Y
or even better
R(X, y)Gy (X) = (X q+1 1)
R
(X, y) =
Y
0
0
(X q+1 1)(r0 X |S|r + 2r
X |S|2r + ... + q0 X |S|q + q+1
X |S|q1 + ...). ()
Here G = Gy is a polynomial in X of degree at most q + 1 r. The term of highest

degree on the right-hand side of () that has degree not 1 mod r is of degree |S|
0
and has coefficient q+1
, where 0 is differentiation with respect to Y .
First examine y 6 S. As R(X, y) is an r-th power, any non-constant term in G,
with degree not 1 mod r would give a term on the right-hand side of degree > |S|
and not 1 mod r, but such a term does not exist. Hence every term in G has degree
0
1 mod r except for the constant term which has coefficient q+1
.
78
For any natural number and i = 1, . . . , r 2 the coefficient of the term of degree
|S| i(q + 1) r (which is not 0 or 1 mod r) on the right-hand side of () is
0
0
+ (i+1)(q+1)+r
i(q+1)+kr
and must be zero. However if (r 1)(q + 1) + r > |S| then (r1)(q+1)+r 0

0
and we have i(q+1)+r
= 0 for all i = 1, . . . , r 2. Now consider the coefficient of
0
the term of degree |S| r. On the right hand side of () this has coefficient r
0
(since q+1+r = 0). The only term of degree zero mod r in G is the constant
0
term which is q+1
. The coefficient of the term of degree |S| r in R(X, y) is
r . Hence
0
0
r q+1
= r
for all y 6 S.
()
If y S then q+1 (y) = 1 and if y 6 S then q+1 (y) = 0. Let
Y
f (Y ) =
(Y y).
yS
2
Then f q+1 = (Y q Y )g(Y ) for some g GF(q 2 )[Y ] of degree at most |S| 1
(the degree of q+1 is at most q 2 1). Differentiate and substitute for a y S
and we have f 0 (y) = g(y). Since the degree of f 0 and g are less than |S| we have
0
g f 0 . Now differentiate and substitute for a y 6 S and we get q+1
f = f 0.
0
0
0
Thus for y 6 S we have r f /f = r and so (f r ) (y) = 0. The polynomial
(f r )0 has degree at most r(q 1) + |S| 2, which is less than q 2 |S| if
r q 2r. So from now on let |S| = (r 1)q + r. The polynomial (f r )0 0
and so f r is a p-th power. Hence f p1 divides r .
If p 2 then (p 1)(r 1)q + r(p 1) > r(q 1) and so r 0.
However the polynomial whose terms are the terms of highest degree in R(X, Y )

is (X + Y q1 )|S| which has a term X (r1)q Y r(q1) since |S|
r = 1. Thus r has
a term Y r(q1) which is a contradiction. Therefore p 1.
Corollary 13.7. A code of dimension 3 whose weights and length have a common
divisor r and whose dual minimum distance is at least 3 has length at least (r
1)q + (p 1)r.
A maximal arc in a projective plane is a set of points S with the property that
every line is incident with 0 or r points of S, see Section 15. Apart from the trivial
examples of a point, an affine plane and the whole plane, that is where r = 1, q
or q + 1 respectively, there are examples known for every r dividing q for q even,
see e.g. Denniston [59].
Corollary 13.8. There are no non-trivial maximal arcs in PG(2, q) when q is odd.

Proof:
79
A maximal arc has (r 1)q + r points, see Exercise 14.1.
After Theorem 13.5 it is quite natural to return to Vandermonde-like properties

(see also Section 9.3). To end this section and to kind of prepare for the next one,
here is an application of the Vandermonde property from [70]:
Proposition 13.9. Let q be even and suppose that K is a (q +t, t)-arc of type (0, 2, t)
having t points on the line at infinity, {(y1 ), (y2 ), ..., (yt )}, say. If the y-axis is also
a t-secant, then the set {y1 , y1 , ..., yt } is a Vandermonde-set.
Proof: First of all note that () cannot be in K, since it would contradict the
easy observation that through any point of K there is one t-secant and q 2-secants.
Let U = {(ai , bi ) : i = 1, ..., q} be the affine part of K. Suppose
Qq a1 = ... =
at = P
0 and consider the affine Redei polynomial R(X, Y ) = i=1 (X + ai Y +
q
bi ) = j=0 rj (Y )X qj , of U . According to the fundamental property of the Redei
polynomial and to the properties of K, for a fixed Y = y 6= yi , the polynomial
R(X, y) is a square (each affine line through (y) intersects K in an even number
of points, hence each root of R(X, y) has even multiplicity), while for Y = yi
(i = 1, ..., t), R(X, yi ) = X q X. This means that the coefficient polynomial
rq1 (Y ) is 1 on the set T = {y1 , ..., yt } and zero elsewhere (note that since the
characteristic
Pt is two, we have 1 = 1). This is also true about the polynomial
(Y ) = i=1 (Y yi )q1 , so since these polynomials have degree at most q 1,
they have to be the same. On the other hand (using that a1 = ... = at = 0), the
Y -degree of R and hence also of rq1 is at most q t, so using Proposition 9.19
(iii), we are done.
Exercise 13.10. Let B PG(2, q) be a pointset, |B| = q +k, with every intersection
number being 1 mod p and suppose that B` = k. Prove that B\` AG(2, q),
considered as a subset of GF(q), is a Vandermonde-set.
We note that much more is true if k q+1
2 : such a set, which is a blocking set
of Redei type, is always a (translate of a) subspace of GF(q 2 ), considered as a
vectorspace over a suitable subfield (hence an additive subgroup of GF(q 2 )); see
Theorem 18.14.
Exercise 13.11. Let B PG(2, q) be a pointset, |B| = q +k, with every intersection
number being 1 mod p. Let `1 and `2 be two lines such that |`1 B| = |`2 B| = k
and B `1 `2 = . W.l.o.g. let `1 = [1, 0, 0] the X-axis and `2 = [0, 0, 1] the line
at infinity. Prove that `1 B and `2 B are Vandermonde sets!
Theorem 13.12. [70] The t-secants of a (q +t, t)-arc of type (0, 2, t) are concurrent.
Proof: Suppose to the contrary that (after transformation) the line at infinity
and the two axes are t-secants.
|K| = q + t, `
line |`K| = 0, 2
or t
80
Let T = {(y1 ), ..., (yt )} be the intersection of the arc and the line at infinity.
According to Proposition 13.9, T is a Vandermonde-set. On the other hand, the
affine transformation switching the two affine coordinates (that is (u, v) (v, u))
(which extends to the ideal line as (y) (1/y)) interchanges the two axes, while
the set T is replaced by T 0 = {1/y1 , ..., 1/yt } (note that (0) and () cannot be in
T ). It follows
again by Proposition 13.9, T 0 is also a Vandermonde set. So we
P that
k
have that i yi = 0 for any 1 k t 2 and for any q t + 1 k q 2.
Pt
Now consider the polynomial (Y ) = i=1 (Y yi )q1 . According to Propositions
13.9 and 9.19, it has degree q t P
and it has q t different roots (the complement
of T ). On the other hand, since i yik = 0 for any q t + 1 k q 2, it is
divisible by Y t , which means that 0 is a root of multiplicity at least t. This is a
contradiction.
13.2
Small and large super-Vandermonde sets
If in Proposition 9.20(ii) we write Y t f ( Y1 ) then we get a polynomial of degree t

and its roots are { y1 : y T }. Hence a super-Vandermonde set is equivalent to a
fully reducible polynomial of form g p (Y ) + Y t , t > p deg g.
Lets explore this situation. Firstly, if q = p is a prime then the only possibility is
f (Y ) = Y t + c, i.e. a transform of the multiplicative group {y : y t = 1}, if it exists
(so iff t|q 1).
If f (Y ) = g p (Y ) + Y t is a fully reducible polynomial without multiple roots then
we can write it as Y q Y = f (Y )h(Y ). Now we may use the trick I have learnt
from G
acs: differentiating this equation one gets
1 = tY t1 h(Y ) + f (Y )h0 (Y ).
Substituting a root y1 of f we get h(y1 ) =
1
ty1t1
= 1t y1qt . Suppose that t > 2q ,
then h(Y ) = 1t Y qt holds for more values than its degree hence it is a polynomial
identity implying a contradiction unless q t = 1. As t = 2q is impossible (it would
imply p = 2 and f would be a power), we have that either t = q 1 (and then
h(Y ) = Y so f (Y ) = Y q1 1) or t q1
2 .
For describing small and large super-Vandermonde sets we need to examine the
coefficients of the original equation Y q Y = f (Y )h(Y ) carefully. What does small
and large mean? We know that any additive subgroup of GF(q) forms a Vandermonde set, so removing the zero element from it one gets a super-Vandermonde
set. The smallest and largest non-trivial additive subgroups are of cardinality p
and q/p, respectively. Note that the super-Vandermonde set, derived from an additive subgroup of size p, is a transform of a multiplicative subgroup. This motivates
that, for our purposes small and large will mean of size q/p,
resp.
81
Theorem 13.13. [120] Suppose that T GF(q) is a super-Vandermonde set of size

|T | < p. Then T is a (transform of a) multiplicative subgroup.
Proof: Since t < p the polynomial f (Y ) is of the form f (Y ) = Y t b0 . As f (Y ) is
a fully reducible polynomial without multiple roots, it implies that b0 has precisely
t distinct t-th roots, t|q 1 and T is a coset of a multiplicative subgroup.
Theorem 13.14. [120] Suppose that T GF(q) is a super-Vandermonde set of size

|T | > q/p. Then T is a (transform of a) multiplicative subgroup.
Proof: Let us write Y q Y = f (Y )h(Y ), where f (Y ) = Y t + bmp Y mp +
b(m1)p Y (m1)p +...+bp Y p +b0 and h(Y ) = Y qt +aqt1 Y qt1 +...+a2 Y 2 +a1 Y .
Consider the coefficient of Y 1 , Y 2 , ..., Y q in this equation. We get
Y 1 : 1 = a1 b0
Y j : aj = 0 if 2 j t and j 6= 1 (mod p)
Y j : aj = 0 if t + 1 j 2t and j 6= 1, t + 1 (mod p)
Y j : aj = 0 if 2t + 1 j 3t and j 6= 1, t + 1, 2t + 1 (mod p) and so on, generally
Y j : aj = 0 if kt + 1 j (k + 1)t and j 6= 1, t + 1, ..., kt + 1 (mod p).
Y p+1 : ap+1 b0 + a1 bp = 0
Y 2p+1 : a2p+1 b0 + ap+1 bp + a1 b2p = 0
generally
Y kp+1 : akp+1 b0 + a(k1)p+1 bp + ... + ap+1 b(k1)p + a1 bkp = 0, for k = 1, 2, ..., m.
Y t+1 : a1 + b0 at+1 + bp atp+1 + b2p at2p+1 + ... + bmp atmp+1 = 0
The indices of coefficients a are of the form t kp + 1. Since t kp + 1 < t and
t kp + 1 6= 1 (mod p) (because t 6= 0 (mod p) is true) these coefficients are 0.
So the equation is of the form
Y t+1 : a1 + b0 at+1 = 0.
Y 2t+1 : at+1 + b0 a2t+1 + bp a2tp+1 + ... + bmp a2tmp+1 = 0
The indices j of coefficients aj are t < j < 2t. These coefficients are 0 if j 6= 1, t + 1
(mod p). It means 2t + 1 6= 1 (mod p) so 2t 6= 0 (mod p) which means p 6= 2. The
other condition 2t + 1 6= t + 1 (mod p) is satisfied by any t. Hence
Y 2t+1 : at+1 + b0 a2t+1 = 0 if p 6= 2.
Similarly
Y 3t+1 : a2t+1 + b0 a3t+1 + bp a3tp+1 + ... + bmp a3tmp+1 = 0.
The indices are between 2t and 3t here. The coefficients are 0 if 3t+1 6= 1, t+1, 2t+1
(mod p). It gives only one new condition: 3t + 1 6= 1 (mod p) so 3t 6= 0 (mod p)
which means p 6= 3. The two other conditions has occurred earlier: p 6= 2 and t 6= 0
(mod p).
82
Y 3t+1 : a2t+1 + b0 a3t+1 = 0 if p 6= 2, 3.

Generally
Y lt+1 : a(l1)t+1 + b0 alt+1 + bp altp+1 + ... + bmp altmp+1 = 0, for l = 1, 2, ..., n 1.
The indices are of the form t kp + 1 and they are between (l 1)t and lt. Hence
the coefficients a are 0 if lt + 1 6= 1, t + 1, ..., (l 1)t + 1 (mod p). It gives (l 1)t
conditions:
lt + 1 6= 1 (mod p) so p 6= l;
lt + 1 6= t + 1 (mod p) so p 6= (l 1);
lt + 1 6= 2t + 1 (mod p) so p 6= (l 2);
and so on
lt + 1 6= (l 2)t + 1 (mod p) so p 6= 2; finally
lt + 1 6= (l 1)t + 1 (mod p) so t 6= 0, which is true.
Hence generally we get
Y lt+1 : a(l1)t+1 + b0 alt+1 = 0 if p 6= 1, 2, ..., l.
In particular, substituting l = n 1 into this equation we get
Y (n1)t+1 : a(n2)t+1 + b0 a(n1)t+1 = 0 if p 6= 1, 2, ..., n 1.
The greatest index of a coefficient a can be q t 1.
(n 1)t < q 1 and nt q 1 because of the definition of n.
It means that (n 1)t q t 1 so (n 1)t + 1 q t.
It implies that a(n1)t+1 (which occurred in the previous equation) does not exist.
So we have two possibilities:
Case 1. (n 1)t + 1 = q t, so t =
Y
(n1)t+1
q1
n
and the equation is of the form
: a(n2)t+1 + b0 = 0. (Hence we can write 1 instead of a(n1)t+1 .)
Case 2. (n 1)t + 1 > q t, so the equation is of the form

Y (n1)t+1 : a(n2)t+1 = 0 if p 6= 1, 2, ..., n 1. We will now prove that it leads to
a contradiction.
Substituting a(n2)t+1 = 0 into the equation
Y (n2)t+1 : a(n3)t+1 + b0 a(n2)t+1 = 0, we get a(n3)t+1 = 0.
We can substitute this again into the equation
Y (n3)t+1 : a(n4)t+1 + b0 a(n3)t+1 = 0, and we get a(n4)t+1 = 0.
Substituting this in a decreasing order we get
Y t+1 : a1 + b0 at+1 = 0 so a1 = 0.
Hence 1 = a1 b0 , so a1 6= 0, Case 2 implied a contradiction. It means that Case 1
will occur, so t = q1
n if p 6= 1, 2, ..., n1. In other words t|q 1 if p 6= 1, 2, ..., n1.
Hereafter, we can write 1 instead of aj if j = (n 1)t + 1, and 0 if j > (n 1)t + 1.
Y (n1)t+1 : a(n2)t+1 + b0 = 0 so a(n2)t+1 = b0 .
83
Substituting this into the equation

Y (n2)t+1 : a(n3)t+1 + b0 a(n2)t+1 = 0, we get
Y (n2)t+1 : a(n3)t+1 + b0 b0 = 0 so a(n3)t+1 = b0 2 .
Substituting this in a decreasing order we get
Y lt+1 : a(l1)t+1 = (b0 )nl for l = n 1, n 2, ..., 1. Finally
Y t+1 : a1 = (b0 )n1 .
Substituting this into 1 = a1 b0 , we get 1 = (b0 )n1 b0 so 1 = (b0 )n .
We are going to examine the equation that belongs to Y lt+kp+1 . First we write up
Y (n1)t+p+1 : a(n2)t+p+1 + bp + b2p a(n1)t+p+1 + ... = 0
We have already seen that the coefficients a occurring in this equation are 0,
because these are the same as in the equation of Y (n1)t+1 . So
Y (n1)t+p+1 : a(n2)t+p+1 + bp = 0.
Similarly
Y (n1)t+2p+1 : a(n2)t+2p+1 + b2p = 0, generally
Y (n1)t+kp+1 : a(n2)t+kp+1 + bkp = 0 for k = 1, 2, ..., m.
On the other hand
Y lt+p+1 : a(l1)t+p+1 + b0 alt+p+1 + bp alt+1 = 0, for l = 1, 2, ..., n 1.
Generally we get
Y lt+kp+1 : a(l1)t+kp+1 + b0 alt+kp+1 + bp alt+(k1)p+1 + ... + bkp alt+1 = 0 for l =
1, 2, ..., n 1 and k = 1, 2, ..., m.
In particular, if l = 1 the equation is of the form
Y t+kp+1 : akp+1 + b0 at+kp+1 + bp at+(k1)p+1 + ... + bkp at+1 = 0.
Lemma 13.15. bp = b2p = ... = bmp = 0.
Proof: We prove it by mathematical induction.
Step 1. First we prove that bp = 0. Consider the equation
Y (n2)t+p+1 : a(n3)t+p+1 + b0 a(n2)t+p+1 + bp a(n2)t+1 = 0.
We have seen that
Y (n1)t+p+1 : a(n2)t+p+1 + bp = 0 so a(n2)t+p+1 = bp and
()
Y (n1)t+1 : a(n2)t+1 = b0 .
Substituting these into the equation (), we get
Y (n2)t+p+1 : a(n3)t+p+1 b0 bp b0 bp b0 = 0 so a(n3)t+p+1 = 2b0 bp . Generally
we can write
Y lt+p+1 : a(l1)t+p+1 + b0 alt+p+1 + bp alt+1 = 0 for l = n 1, n 2, ..., 1.
()
Substituting
Y (l+1)t+p+1 : alt+p+1 = (1)nl1 (n l 1)bnl2
bp and
0
84
Y (l+1)t+1 : alt+1 = (b0 )nl1 into the equation (), we get

Y lt+p+1 : a(l1)t+p+1 = (1)nl (n l)bnl1
bp for l = n 1, n 2, ..., 1. If l = 0
0
it means
Y p+1 : ap+1 b0 + a1 bp = 0.
( )
Substituting
Y t+p+1 : ap+1 = (1)n1 (n 1)bn2
bp and
0
Y t+1 : a1 = (b0 )n1 into ( ), we get
Y p+1 : (1)n1 nbn1
bp = 0.
0
In this equation 1 6= 0 (mod p), n 6= 0 (mod p) and b0 6= 0 (mod p) (from the
equation a1 b0 = 1). It means that bp = 0.
Step 2. Suppose bp = b2p = ... = b(s1)p = 0. We show that bsp = 0. Consider
Y (n2)t+sp+1 : a(n3)t+sp+1 + b0 a(n2)t+sp+1 + bsp a(n2)t+1 = 0.
(?)
We have seen that

Y (n1)t+sp+1 : a(n2)t+sp+1 + bsp = 0 so a(n2)t+sp+1 = bsp and
Y (n1)t+1 : a(n2)t+1 = b0 .
Substituting these into the equation (?), we get
Y (n2)t+sp+1 : a(n3)t+sp+1 b0 bsp b0 bsp = 0 so a(n3)t+sp+1 = 2b0 bsp .
Generally we can write
Y lt+sp+1 : a(l1)t+sp+1 + b0 alt+sp+1 + bsp alt+1 = 0 (??) for l = n 1, n 2, ..., 1.
Substituting
Y (l+1)t+sp+1 : alt+sp+1 = (1)nl1 (n l 1)bnl2
bsp and
0
Y (l+1)t+1 : alt+1 = (b0 )nl1 into the equation (??), we get
Y lt+sp+1 : a(l1)t+sp+1 = (1)nl (n l)bnl1
bsp for l = n 1, n 2, ..., 1.
0
If l = 0 it means
Y sp+1 : asp+1 b0 + a1 bsp = 0.
(? ? ?)
Substituting
Y t+sp+1 : asp+1 = (1)n1 (n 1)bn2
bsp and
0
Y t+1 : a1 = (b0 )n1 into (? ? ?), we get
Y sp+1 : (1)n1 nbn1
bsp = 0.
0
In this equation 1 6= 0 (mod p), n 6= 0 (mod p) and b 6= 0 (mod p) (from the
equation a1 b0 = 1). It means that bsp = 0.
So we have got bp = b2p = ... = bmp = 0. It means that f (Y ) is of the form

f (Y ) = Y t + b0 , and that t|q 1 so t =
q1
n
and (b0 )n = 1. Hence
14. Arcs, Segre
85
q1
f (Y ) = Y n + b0 , where (b0 )n = 1 if p 6= 1, 2, ..., n 1 (mod p). So the roots of

f (Y ) are the elements of a coset of a multiplicative subgroup of order t.
14
Arcs, Segre
A (k, n)-arc in a projective plane is a set of k points such that each line intersects
it in at most n points. It is complete if it cannot be extended to a (k + 1, n)-arc.
Considering (k, n)-arcs, Barlotti [24] showed that for 1 < n < q + 1, k qn q + n
and equality can only hold when n divides q.
Exercise 14.1. Verify Barlottis bound. Observe that in case of equality each line
intersects the (k, n)-arc in either 0 or n points and so n|q.
In this section we consider the case n = 2, so simply k-arcs. Note that, by Exercise
14.1, (q + 2)-arcs exist only if q is a power of 2. So if q is odd then arcs are of
size q + 1. In some sense Segres theory was a starting point of modern Galois
geometry.
Exercise 14.2. Let the lines `1 , `2 and `3 form a triangle in PG(2, q). Prove that
there exists a set {m1 , ..., mq1 } of lines (mi 6= `j ), covering the non-vertex points
of the triangle if and only if q is even.
The Segre curve is not used in general as a polynomial in three variables, but
as an object with general, local or global, algebraic-geometric properties. Strictly
speaking it is not obvious why to include the Segre method to this book. I
decided to treat it here as well because of (i) historical reasons; (ii) it inspired
directly or indirectly the outburst of the polynomial methods; (iii) there are results
where Segre-like arguments can be used together with other polynomial methods,
see e.g. Section 17.
A generalization of the theorem of Menelaos is the following.
Let T be an arbitrary triangle in PG(2, q). The points on the sides of T may
be identified by coordinates as follows. Take T as fundamental triangle (i.e. the
vertices of T are A1 (1, 0, 0), A2 (0, 1, 0) and A3 (0, 0, 1)), and place the unit point
(1, 1, 1) arbitrarily (not on a side of T of course). Points on the side A2 A3 are of
form (0, 1, c), points on the side A3 A1 are of form (a, 0, 1), and points on the side
A1 A2 are of form (1, b, 0). These coefficients will be called the coordinates of
the points. Note that these three points are collinear (so they are contained in an
algebraic curve of degree one) precisely when abc = 1.
More generally, we have the following:
this is Menelaos
86
Theorem 14.3. A necessary and sufficient condition that 3n points, distributed in

three n-tuples on the three sides of T (some possibly coinciding with each other,
but none with any vertex of T ) are contained in an algebraic curve C of degree n,
is that the product of their coordinates is
n
Y
i=1
ai
n
Y
i=1
bi
n
Y
ci = (1)n
().
i=1
If a point P appears m times in an n-tuple then we require that at P the intersection multiplicity of C and the corresponding line is precisely m. (It also implies
that C does not contain any side of the triangle.)
Proof: We are looking for a curve of form
X
f (X, Y, Z) =
ijk X i Y j Z k .
i+j+k=n
n1
free
2
coefficients
Let {Ps = (1, bs , 0) : s = 1, ..., n} be the given (multi)set of points on A1 A2 .

Then F (X, Y, 0) must factor to d3 (b1 X Y )(b2 X Y ) (bn X Y ); similarly
F (0, Y, Z) = d1 (c1 Y Z)(c2 Y Z) (cn Y Z) and F (X, 0, Z) = d2 (a1 Z
X)(a2 Z X) (an Z X), where d1 , d2 and d3 are suitable non-zero constants. These equations determine all the coefficients ij0 , i0k , 0,jk . In particular, n00 = d3 b1 b2 bn = d2 (1)n , 0n0 = d1 c1 c2 cn = d3 (1)n and
00n = d2 a1 a2 an = d1 (1)n . The product of these equations gives exactly ().
On the other hand, if () holds, lets fix one of the constants, say d1 = 1, then d2 , d3
and all the coefficients ij0 , i0k , 0,jk are determined. The remaining coefficients
ijk , i, j, k 6= 0 can be chosen arbitrarily.
As it is used frequently, we formulate the dual version of the above, which is then
a generalization of the theorem of Ceva.
Let T be an arbitrary triangle in PG(2, q). The lines through the vertices of T
may be identified by coordinates as follows. Take T as fundamental triangle (i.e.
the vertices of T are A1 (1, 0, 0), A2 (0, 1, 0) and A3 (0, 0, 1)), and place the unit
point (1, 1, 1) arbitrarily (not on a side of T of course). Lines through the vertices
of T other than the sides will have an equation of the form X2 = bX3 , X3 =
cX1 and X1 = aX2 , with a, b, c GF(q) . This coefficient will be called the
coordinate of the line. Note that these three lines, i.e. [0, 1, b], [c, 0, 1] and [1, a, 0]
are concurrent (so they are contained in a pencil, i.e. an algebraic envelope of
degree one) precisely when abc = 1. More generally, we have the following dual
version:
Ceva: n = 1
Result 14.4. A necessary and sufficient condition that 3n lines, distributed in three
n-tuples through the three vertices of T (some possibly coinciding with each other,
but none with any side of T ) are contained in an algebraic envelope C of degree n
containing no side of T , is that the product of their coordinates is (1)n .
14. Arcs, Segre
87
We are especially interested in the following extension :

Result 14.5. (Segre) Let us consider any number k ( 3) of lines L = {L1 , ..., Lk }
in PG(2, q) no three of which are concurrent (that is a dual k-arc), and on each
line Li a set Gi of n points, not necessarily distinct, but each on one line of L
only. In order that there exists an algebraic curve C of degree n, containing each
Gibut not containing any line of L it is necessary and sufficient that each of the
k
3 triplets of sets Gi satisfies the condition in the previous Result 14.3.
Finally, we remark that if k > n, then the curve C is unique, for if C and C
both satisfy the requirements above, the same holds for any linear combination of
them. Now form a linear combination C that contains an extra point P on one of
the lines Li . Since C intersects the line Li in too many points it will contain the
line Li as a factor, and since it introduces more intersections with the other lines
as well, all lines of L will factor into C but C only has degree n and k > n.
Let us also underline that the curve does not contain any point which is on two
lines of the dual k-arc A.
The most general form of this theorem we are aware of is the following.
Result 14.6. [134] Let us consider any number k ( 3) of lines L = {L1 , ..., Lk }
in PG(2, q), and on each line Li a set Gi of n points, not necessarily distinct, but
each on one line of L only. Suppose that for each J {1, 2, ..., k} of the following
type there is a curve CJ of degree n intersecting the lines Lj : j J at the given
points Gj with the required multiplicities:
(a) {Lj : j J} is a triangle;
(b) {Lj : j J} are all concurrent.
Then there exists an algebraic curve C of degree n, intersecting each Li in the given
points Gi with the required multiplicities. In this case, if k > n then the curve C is
unique.
Exercise 14.7. Prove unicity, so the last sentence of the result.
Note that the result does not hold any more if we require (b) for a bounded number
of concurrent lines of L only. But if the hypothesis holds for any t concurrent lines
of L, t n + 2, then it holds in general.
Note also that if we know that a curve C of degree n contains some points
Ps = (as , bs , cs ), s = 1, 2, ... then each f (as , bs , cs ) = 0 gives a homogeneous linear
equation for the coefficients ijk , i + j + k = n. If the curve is unique then the
system of equations has a unique solution (in the projective sense, so up to a
constant multiplier). The matrix of the system of linear equations contains entries
like ais bjs cks .
There are many applications of the results of this section. See Section 17 on sets
without tangents, Section 16 on unitals, etc.
We begin with an easy application of Cevas theorem.
This argument is
useful for distinct points Ps
only.
88
Proposition 14.8. (Bichara, Korchmaros) Let K PG(2, q) be a set of q +2 points,

and suppose that A1 , A2 , A3 K are such that any line meeting {A1 , A2 , A3 }
intersects K in exactly two points. Then q is even.
Proof: Choose A1 , A2 , A3 as the base points of a coordinate frame.
Construct a (q 1) 3-as matrix M , whose rows are indexed by K \ {A0 , A1 , A2 },
whose columns are indexed by A0 , A1 , A2 . The element of M determined by (P, Ai )
will be the coordinate i (P ) of the line P Ai . Cevas theorem gives that the product
of the three elements in one row of M will be 1. Therefore the product of all
elements of M will also be 1. On the other hand, in each column there are q 1
pairwise different nonzero element of GF(q). This means that in every column we
see each nonzero element of GF(q). By Wilsons theorem their product is (1),
whence 1 = (1)3 follows. This implies that q is even.
Of course the most celebrated application is Segres theorem on arcs.
Theorem 14.9. Let S PG(2, q) be a (q + 1)-arc, i.e. a set of q + 1 points, no three
of them being collinear. If q is odd then S is a conic.
Proof: ([23]) Suppose w.l.o.g. that S contains the points (1, 0, 0), (0, 1, 0) and
(0, 0, 1). Let the tangents at these points be Z = cY , X = aZ and Y = bX,
respectively. Let {Si | i = 1, 2, . . . , q 2} be the other q 2 points of S.
Let (1, bi , 0) be the point that is the intersection of the line (0, 0, 1), Si with the
line Z = 0. Let (0, 1, ci ) be the point that is the intersection of the line (0, 0, 1), Si
with the line X = 0 and let (ai , 0, 1) be the point that is the intersection of the
line (0, 1, 0), Si with the line Y = 0. Since S is an oval, the set
{ai | i = 1, 2, . . . q 2} {a}
contains every non-zero element of GF(q) and similarly for the bi s and the ci s.
The lines Y = bi X, X = ai Z and Z = ci Y have a point in common (namely Si ),
hence ai bi ci = 1.
The product of all the non-zero elements of GF(q) is 1. Therefore
!
!
!
q2
q2
q2
Y
Y
Y
1 =
ai a =
bi b =
ci c,
i=1
i=1
q2
Y
i=1
!
ai bi ci
abc = 1
i=1
hence abc = 1.
Note that if 1 = 1, so when q is even then we got a(n algebraic) proof for the
(combinatorial) fact, that the tangents of a (q +1)-arc are concurrent (and so their
15. Maximal arcs
89
intersection point, called the nucleus, can be added to the arc, which results in a
(q + 2)-arc or hyperoval).
Let f (X, Y, Z) := cXY + acY Z + XZ. The conic defined from f = 0 contains
the points (1, 0, 0), (0, 1, 0) and (0, 0, 1) and it follows from the fact that abc = 1
that it has tangents Y = cX, X = bZ and Z = aY at these points.
We have to show that f (the conic defined by f ) contains another arbitrary point
S of S and then we have finished. Let {P, Q, R} = {(1, 0, 0), (0, 1, 0), (0, 0, 1)}. Let
f1 be the quadratic form that defines the conic f1 = 0 that contains the points P ,
Q and S and whose tangents coincide with the tangents of S at those points. In
the same way define f2 for the points P , R and S.
Let tP be the tangent at the point P . The forms f , f1 and f2 modulo tP have a
double zero at P , so we can multiply by a scalar so that the three forms coincide
at all the points of tP . The forms f1 and f2 modulo tS share a double point at S
and coincide at tS tP , so they coincide at all the points of tS . In the same way
the forms f and f1 coincide at all the points of tQ and the forms f and f2 coincide
at all the points of tR . But then f1 and f2 also coincide at tQ tR . So in all f1
and f2 also coincide on two lines and point not on these lines. Hence the quadratic
form f1 f2 is zeros on two lines and another point and is therefore identically
zero. So f1 = f2 and in the same way f = f1 = f2 . Hence the conic defined by the
equation f = 0 contains S.
15
Maximal arcs
After Barlottis bound (Exercise 14.1), (k, n) arcs of size k = qn q + n are

called maximal. In the Desarguesian projective plane of order q, Denniston [59]
constructed maximal arcs for every divisor of q, when q is even. Ball, Blokhuis and
Mazzocca [13] proved that for q odd, there is no maximal arc, see Theorem 15.12.
Their proof was simplified in Ball-Blokhuis [14] and then generalized in [15], as we
have seen in Theorem 13.6.
15.1
Existence: hyperovals
Here we give some information about the case n = 2 (hence q = 2h ), i.e. about
hyperovals. For other even values of n, Denniston, Thas and Mathon constructed
examples.
Consider a hyperoval Hf of PG(2, q), with oval polynomial f (X), i.e. Hf =
{(1, 0, 0), (0, 1, 0)} {(x, f (x), 1) : x GF(q)}; we require f (0) = 0, f (1) = 1.
Exercise 15.1. Hf is a hyperoval if and only if f (X) is a permutation polynomial
(s)
and Fs (X) = f (X+s)f
is a permutation polynomial for all s GF(q), with
X
or o-polynomial
N.B.:
Fs (0) = f 0 (s)
90
Fs (0) = 0.
q > 2 even
Exercise 15.2. Prove that an oval polynomial is of form

f (X) = a2 X 2 + a4 X 4 + ... + aq2 X q2 .
It
does
not
mean that we
can characterize
monomial
hyperovals...
If f (X) = X r is a monomial then everything becomes easier.

Exercise 15.3. f (X) = X r is an oval polynomial if and only if (r, q 1) = 1,
(X)
(r 1, q 1) = 1 and f (X+1)f
is a permutation polynomial.
X
k
Exercise 15.4. In particular, f (X) = X 2 is an oval polynomial of PG(2, 2h ) if

and only if (k, h) = 1.
It is equivalent
to saying that f
is additive.
q = ph
It may happen that the graph of f has the property that the group of translations
of AG(2, q) acts transitively on it, i.e. for any c GF(q), f (X) + f (c) = f (X + c).
In this case Hf is called a translation oval. Then, from Exercise 15.1 we have that
Hf is a translation oval if and only if f (X) is an additive permutation polynomial,
and also f (X)/X is a permutation polynomial.
Recall Theorem 5.25, saying that every additive function GF(q) GF(q) is a
2
h1
linearized polynomial a0 X + a1 X p + a2 X p + ... + ah1 X p . See also Exercise
5.26.
Ph1
i
Exercise 15.5. The map f (X) = i=0 ai X p is an additive permutation of GF(q)
such that f (X)/X
is also a permutation, if and
only if for the Segre-Bartocci-like
a1
... ah1
ap
p
... aph2
h1
2
2
ap
p
p2
h
h2 ah1 ... ah3 = 2 1 1 holds.
matrix A() = det
.
h1
h1
h1
p
p
p
a1
a2
...
Proposition 15.6. (Payne, Hirschfeld) The oval polynomial of a translation oval is
k
of form f (X) = X 2 (and (k, h) = 1 of course).
Proof: [97] We know that the oval polynomial (being additive) is of the form
f (X) = a2 X 2 + a4 X 4 + a8 X 8 + ... + aq/2 X q/2 and, by Exercise 15.1 now f (X)/X
is a permutation polynomial. We have to prove that only one ai can be nonzero.
In Exercise 15.5 deg A() = 2h 1. It is easy to see that the coefficient of
Ph
t , 0 < t < 2h 1 can be calculated as follows. Let t = i=1 ti 2i1 , ti {0, 1};
and let tj1 , ..., tjr be precisely the binary digits of t being equal to zero. Let Mt be
the r r principal minor of A() obtained by selecting the rows and columns of
indices j1 , ..., jr and then substituting = 0. Then det Mt is the coefficient of t .
15. Maximal arcs
91
0
a1
... ah1
ap
0
...
aph2
h1
2
2
ap
p
p2
h2 ah1 ... ah3
Hence we need that each k k principal minor of A =
.
.
.
h1
ap1
h1
ap2
... 0
0 < k < 2h1 1 is singular.
Now for 2 k < h let {i1 , ..., ik } {1, ..., h}. By induction on k one can prove
that ai2 i1 ai3 i2 aik ik1 ai1 ik = 0, where the indices are modulo h.
The coefficient of 0 is clearly det A = 1. Expanding det A and using the observation of the previous paragraph on cyclic product, we see that at most (hence
precisely) one ai is nonzero.
Exercise 15.7. If 1 < r, r1 , r2 , r3 < q 1, rr1 1 (mod q 1), (r 1)(r2 1) 1

(mod q 1), r + r3 1 (mod q 1), then the hyperovals HX r , HX r1 , HX r2 and
HX r3 are projectively equivalent.
Exercise 15.8. HX r is regular (conic + nucleus) iff r = 2, q/2 or q 2.
We remark that in PG(2, 2h ) non-regular hyperovals exist iff h 4. As for h = 5
and for h 7 there exist translation hyperovals; for n = 4 Lunelli and Sce found
Hf with
f (X) = (a2 X 7 + e7 X 6 + e3 X 5 + e1 3X 4 + eX 3 + e7 X 2 + eX)2 , where e4 = e + 1;
finally for h = 6 we have
f (X) = X 62 + X 30 + X 24
+e21 (X 60 +X 58 +X 54 +X 52 +X 46 +X 44 +X 40 +X 38 +X 34 +X 16 +X 14 +X 10 +X 8 +X 4 )
+e42 (X 50 + X 48 + X 36 + X 32 + X 26 + X 20 + X 18 + X 12 + X 6 ), where e6 = e + 1.
For h = 1, 2, 3 it is easy to check that every hyperoval is regular.
Proposition 15.9. Segre In PG(2, 2h ) f (X) = X 6 is an oval polynomial iff h is
odd.
6
+1
Proof: By Exercise 15.3 we have to check whether (X+1)
= X 5 + X 3 + X is
X
5
bijective in the case when h is odd. Suppose that for s 6= t s + s3 + s = t5 + t3 + t,
from this, dividing out by (s + t) and introducing u = s2 + t2 , v = st we get the
quadratic u2 + u(v + 1) + v 2 + v + 1 = 0. Dividing by (v + 1)2 and introducing
1 2
1
u
w = v+1
we get (w + v+1
) + (w + v+1
) + 1 = 0. By Exercise 5.6 there is no
2
solution of Y + Y + 1 if Tr2h 2 (1) = h = 1, so when h is odd.
92
We demonstrate the strength of using the Redei polynomial together with the
generalised Newton formulae (NG) from Section 9.4 by giving a short proof for a
theorem of Glynn.
It is easy to see that a set of q + 2 points is a hyperoval if and only if it meets
every line in an even number of points. (Through a point of the set in question
every line contains at least one more point, but 1 + (q + 1) 1 = q + 2, so every
line has to meet it in 0 or 2 points.) For even more on even sets (sets of even type)
see e.g. Section 23.4. Glynn gave an equivalent description of oval polynomials as
follows.
Theorem
15.10. (Glynn) A polynomial f (x) is an oval polynomial if and only if

k+l P k
x
f (x)l = 0 for every 1 k + l q 1, k + l odd, {k, l} =
6 {0, q 1}, and
Pl
q1
f (x)
= 1.
Proof: (G
acs) First suppose H = {(t, f (t)) : t} {(0), ()} is a hyperoval in
PG(2, q) and let U = {(t, f (t)) : t GF(q)} be its affine part. Consider the affine
Redei polynomial
R(X, Y ) =
Y
tGF(q)
=
Pk,l
t1 tkf(s1) f(sl)
t1 ,...,sl
all distinct
X + tY f (t) =
q
X
ri (Y )X qi
i=0
of U . H is a hyperoval precisely if every line meets it in an even number of points.

For lines with direction y 6= 0 this is equivalent to the condition that R(X, y) is a
square. Since q is even, and deg(rj ) j, this is equivalent to rj (Y ) = 0 identically
for all 1 j q 3, j odd, and rq1 (Y ) = Y q1 1 (since R(X, 0) = X q X).
Considering the coefficients of the rj s, we get k,l = 0 for all 1 k + l q 1,
k + l odd, except for k = q 1 or l = q 1, when it is 1 (note that since the
characteristic is 2, we have 1 = 1). This implies (k + l)k,l = 0 for all k and l,
except for 0,q1 = q1,0 = 1.
From this, by induction on k + l and using (NG), we get the desired condition.
On the other hand, suppose f (x) is a polynomial satisfying our conditions and let
H = {(t, f (t)) : t} {(0), ()}.
Again induction on k + l and (NG) give that k,l = 0 for all 1 k + l q 1,
k + l odd, except for k = q 1 or l = q 1, when it is 1. This implies that for
any fixed y 6= 0, R(X, y) is a square, so all lines with slope not in {0, } meet
H in an even number of points. This is also true about vertical lines, since f is a
polynomial and () is in H , so what is left is to show that horizontal lines meet
H in one affine point, i.e. f is bijective.
Q
Note that the bijectivity of f is equivalent to R(X, 0) = t (X + f (t)) = X q + X.
Now R(X, 0) = X q + 0,1 X q1 + ... + 0,q = X q + X + g(X)2 (since 0,k = 0 for
odd k, except for k = q 1, when it is 1). But H 0 (X, 0) = 1, so R(X, 0) cannot
have a multiple root, i.e. R(X, 0) = X q X.
15. Maximal arcs
93
Denniston-arcs
Still in GF(q), q = 2h , choose which makes T 2 +T +1 an irreducible polynomial.
Consider the quadratic curves
F (X, Y, Z) = X 2 + XY + Y 2 + Z 2
GF(q).
For = 0 it defines the point (0, 0, 1) only, while for other values these are irreducible conics contained in AG(2, q). The nucleus of each F ( 6= 0) is (0, 0, 1).
As the additive group of GF(q) is elementary abelian, it has subgroups of every
size 2k , 0 k h.
Exercise 15.11. (Denniston [59]) Let H be a multiplicative subgroup of GF(q) of
size n = 2k , 0 k h. Define
[
SH =
( the points of F ).
H
Prove that SH is a (qn q + n, n)-arc (so a maximal arc).
15.2
Non-existence
We saw in Section 13.1 a more general statement and its proof. Here we show the
proof by Ball and Blokhuis from [14] of the following result:
Theorem 15.12. Ball, Blokhuis, Mazzocca [13] In PG(2, q), with q odd and n < q,
maximal (k, n)-arcs do not exist.
First we can assume that B is a maximal (k, n)-arc contained in AG(2, q) (external
lines always exist if n < q), with |B| = k = (n 1)q + n. As usual, we represent
AG(2, q) by GF(q 2 ) so B GF(q 2 ).
For simplicity assume that 0
6 B and let B [1] = {b1 : b B}. Define the
polynomial
Y
X
B(T ) =
(1 bT ) =
(1)i i T i ,
i=0
bB
where i is the i-th elementary symmetric polynomial of the set B, so i = 0 if

i > k. Also define the polynomial
F (S, T ) =
Y
bB
q1
(1 (1 bT )
S) =
(1)i
i S i ,
i=0
where
i is the i-th elementary symmetric polynomial of the set of polynomials
{(1 bT )q1 : b B}, so degT
i i(q 1).
94
Lemma 15.13. (i) If t0 GF(q 2 ) \ B [1] , then F (S, t0 ) is an n-th power.

(ii) If t0 B [1] , then F (S, t0 ) = (1 S q+1 )n1 .
Proof:
n|k.
(i) When t0 = 0, the polynomial f (s, 0) =
bB (1
S) = (1 S)k and
When t0 6= 0, then t1
is a point not contained in the arc, whence every line
0
through t1
0 contains a number of points of B that is either 0 or n. In the multiset
q1
{(t1
: b B}, every element therefore occurs with multiplicity n, so that
0 b)
in F (S, t0 ) every factor occurs exactly n times.
(ii) Every line passing through the point t1
0 contains exactly n 1 other points
q1
of B; hence the multiset (t1
: b B consists of every (q + 1)-st root of
0 b)
unity repeated n 1 times, together with the element 0. This gives
F (S, t0 ) =
q1 q1
(1 (t1
t0 S) = (1 tq0
0 b)
S q+1 )n1 = (1 S q+1 )n1 .
bB
Corollary 15.14. (i) The functions

i are identically zero for 0 < i < q unless n|i.
(ii) The polynomial B divides
n , which is not identically zero; the degree of
n /B
is at most q 2n.
Proof: From the form of F in both cases of the lemma, it follows that
i (t0 ) = 0
for all t0 GF(q 2 ) with 0 < i < q unless n|i; since the degree of
i is at most
i(q 1) < q 2 , these functions are identically zero. Therefore the first coefficient of
F that is not necessarily identically zero is
n .

k
It is immediate that
n (0) = n . With q = ph , n = pe ,
k = (n 1)q + n = (p 1)ph+e1 + ... + (p 1)ph + pe ;

hence by Lucas theorem nk 1 (mod p), so is not identically zero. On the
other hand, the coefficient of S n in (1 S q+1 )n1 is zero; so
n (t0 ) = 0 for
t0 B [1] . Equivalently, B divides
n . Hence the degree of
n /B is at most
n(q 1) nq + q n = q 2n.
Corollary 15.15.
(i)
F (S, T ) = 1 +
k
X
(1)i
in S in +
i=1
(ii)
BF (S, T ) = B + B
k
X
(1)i
i(q+1) S i(q+1)
(mod T T q ).
i=1
k
X
i=1
(1)i
in S in
(mod T T q ).
15. Maximal arcs
95
Proof: By the previous corollary, the first coefficient of F that is not necessarily
identically zero is
n . Since, for all t0 GF(q 2 ), the function
j vanishes unless n|j
2
or (q + 1)|j, it follows that T T q |
j . If n does not divide j, then
j still vanishes
2
2
[1]
for t0 GF(q ) \ B
, and since B|
n , the divisibility relation T T q |
n
j
follows.
Proof of Theorem 15.12: We shall show that
n2 is a p-th power. This, together
with (a) B|
n and (b)
n is not identically zero, leads to a contradiction for p 6= 2.
Calculating the derivative of B(T ) gives
B 0 (T ) =
X
bB
XX
b
B(T ) = (
bi+1 T i )B(T ).
1 bT
i=0
bB
Since bq = b, it follows that

2
XX
X qX
X
2
(T T )(
bi+1 T i ) =
bi T i =
(1 bT )q 1 .
q2
bB i=0
bB i=0
bB
P
2
Now, bB (1 bT )q 1 takes the value 1 for T = t0 B [1] and is zero for all
other elements of GF(q 2 ). Since
q+1 takes the same values and they are both of
P
2
degree q 2 1 it follows that
q+1 = bB (1 bT )q 1 . This gives
2
(T T q )B 0 =
q+1 B.
Differentiating it and multiplying by B gives that
0
BB 0 B 2
q+1
(mod T T q ).
However, differentiating F (S, T ) with respect to T , it follows that

T F (S, T ) = (
k
X
X b(1 bT )q2 S
)F
(S,
T
)
=
(1)i
i0 S i .
1 (1 bT )q1 S
i=0
bB
Multiplying this by (1 S q+1 ) and putting T = t0 GF(q 2 ), the expression in

brackets becomes a polynomial in S. Hence
F (S, t0 )|(1 S q+1 )T F (S, t0 ).
Define the quotient of this division to be Rt0 (S). It follows that
0
t (S)S 2n +
Rt0 (S) =
n0 (t0 )S n + R
q+1
(t0 )S q+1 ,
0
96
t (S) is an n-th power (considered as a function of S). Abusing notation

where R
0
we define the polynomial R(S, T ) with the property that for t0 GF(q 2 ) R(S, t0 ) =
Rt0 (S). Then we have that
F R = (1 S q+1 )T F
(mod T T q ).
Multiplying it by B,
qq/n+1
X

(1)i B
in S in R (1 S q+1 )BT F
(mod T T q ).
()
i=1
Equating the coefficients of S q+1+n we get

0
0
q+1+n
B + B
n0
q+1
B
n = B 0
n
Hence
0
B
q+1+n
= (B
n )0
(mod T T q ).
2
(mod T T q ).
Equating successively the coefficients of S i(q+1)+n in () for 1 < i < n 1 gives

0
0
B
i(q+1)+n
B
(i1)(q+1)+n
(B
n )0
(mod T T q ).
This implies that B p1 divides

n , which gives a contradiction for p 6= 2, since the
degree of
n is at most n(q 1) and it is not identically zero.
15.3
Embeddability
There are several improvements on Barlottis bound, when n is not a divisor of

q. Lunelli and Sce [94] showed that k (n 1)q + n 3 and if q is large enough
compared to n, then k (n 1)q + 8n/13.
Let K be an arbitrary (qn q + n , n)-arc in PG(2, q). When n divides q, a
natural question is whether K is incomplete if is small enough, that is whether
it can be completed to a maximal arc. For = 1, this was shown by Thas [132].
Ball, Blokhuis [12] showed it for < n/2 when q/n > 3, for < 0.476n when
n = q/3 and for < 0.381n when n = q/2. This result was improved by Hadnagy
and Sz
onyi [73], namely they proved it for 2n/3, if q/n is large enough.
When n = 2, (k, n)-arcs are simply called k-arcs and maximal arcs (of size k =
q + 2) are called hyperovals. In this case Segre [106] showed that a (q + 2 )-arc,
q, can be extended to a hyperoval. When 4 < q is a square, this result
is sharp, since there are complete (q + 1 q)-arcs, see Boros and Szonyi [48],
Fisher, Hirschfeld and Thas [63] and Kestenband [85]. The only known example of
(q + 1 q)-arcs comes from the cyclic representation of the plane: observe that
15. Maximal arcs
97
q 2 + q + 1 = (q + q + 1)(q q + 1) and take every (q + q + 1)-th point in
the Singer-cycle (i.e. a point orbit of a subgroup of size (q q + 1) in the cyclic

representation), it takes some calculations to prove that they form a complete arc.
It is nice to remark that if you take every (q q + 1)-th point instead then you
get the (q + q + 1) points of a Baer subplane PG(2, q).

Note that these examples exist for odd square q-s as well, and it is conjectured that
also in the odd case they are the second largest complete arcs, however the best
known bound is q 21 q + 5 by Hirschfeld and Korchmaros. (For a more detailed

description of the known results see e.g. [82].) We remark that Hirschfeld and
Korchm
aros were also able to prove the stability of the second largest complete
arc in the even square case:
Result 15.16. [80] A complete arc of PG(2, q), q = 22e , e 3 has size either q + 2,
or q q + 1, or at most q 2 q + 6.
The proofs of the Segre or the Hirschfeld-Korchmaros theorems are based on the
profound examination of the properties of an algebraic curve associated to the arc.
Theorem 15.17. (Weiner) Assume that K is a (qpe q + pe , pe )-arc in PG(2, q),
q = ph , p prime. Suppose also that q/4 and pe < q. Then K can be

embedded in a maximal arc.
After some hesitation we omit its (nice but lengthy) proof. For e = 1 see Szonyi
[125], his method was improved by Weiner, who proved it for general e in [137].
For large n = p, the results of Ball-Blokhuis and Hadnagy-Szonyi turn out to be
better, while for small p Theorem 15.17 is stronger. This and Theorem 15.12 of
Ball, Blokhuis and Mazzocca on the non-existence of maximal arcs for p > 2, yield
an upper bound on the size of a (k, pe )-arc.
The following result and proof by Weiner contains some of the key ideas of the
proof of Theorem 15.17; it is similar to Segres idea as it associates an(other)
algebraic curve (envelope) to the arc. The adventage is that this proof seems to
be easier and there is a nice way to generalize it to (k, pe )-arcs of PG(2, q) to get
Theorem 15.17. We will use Theorem 9.37.
Theorem 15.18. Any arc in PG(2, q), q even, of size greater than q q + 1 can
be embedded in a hyperoval.
Proof: ([137]) Assume to the contrary, that K is not a hyperoval, hence |K| =
q + 2 , where 1 q. If there is a point, not in the arc, through that there

pass only 0- and 1-secants, then adding this point to the arc we still get an arc.
Repeating this process until there is no more such a point, we obtain an arc K.
We show that K is a hyperoval.
First of all observe that there are exactly 1-secants passing through a point of
K, hence the total number of 1-secants is (q + 2 ) 6= 0.
98
Define the index of a point as the number of 1-secants through it. Let s be the
index of a point P on some 0-secant `, we show that s is at most .
R(X, y, z)
is
fully reducible
Let ` be the line X = 0 (so [1, 0, 0]) and R(X, Y, Z) be the Redei polynomial
of K. For a fixed (0, z, y) `, consider a root x of R(X, y, z), its multiplicity
is 1 or 2 according as the line [x, y, z] is a 1- or a 2-secant. If x is a root of
multiplicity 2 then its multiplicity as a root of X R(X, y, z) is still at least 2, so
deg(g.c.d.(R(X, y, z), X R(X, y, z))) is exactly (q+2)(the number of 1-secants
through the point (0, z, y)).
For some s now write up the resultant R(Y, Z) = Rs (R(X, Y, Z), X R(X, Y, Z))
as in Section 9.5, its (homogeneous) degree is s(s 1). We have to give an upper
bound for the points on ` with index less than s: if ty,z denotes the number of
1-secants through (0, z, y) then
X
min{s ty,z , 0}
s(s 1).
(0,z,y)`
So counting the 1-secants through the points of ` we get at least (q + 1)s s(s 1),
that is at most (q + 2 ); from which s or s q + 2 follows. Note that
there can be no point with index at least q + 2 , since such a point would have
index q + 2 exactly and it could have been added to K. Furthermore, since
through each point outside K there passes at least one 0-secant, the argument
above implies that the index of any point is at most .
Hence the 1-secants form a dual ((q + 2 ), )-arc and so when q, the
contradiction follows by Barlottis bound 14.1.
In general, probably q is not the right order of magnitude for , except for the
case n = pe = 2. However if we want a bound for not depending on pe , the best
order of magnitude we can get is q.

Combining Theorem 15.17 with the result of Ball, Blokhuis and Mazzocca Theorem
15.12 on non-existence of maximal arcs, the next corollary follows:
Corollary 15.19. A (k, pe )-arc in PG(2, q), q = ph , p > 2 prime, has size less than
qpe q + pe 14 q.
16
Unitals, semiovals
A unital is a 2 (q 3 + 1, q + 1, 1)-design. We use the same name for a pointset U

of PG(2, q 2 ) if it, together with its (q + 1)-secants, forms a unital in the previous
sense. It means that we have |U | = q 3 + 1 points, and every line intersects U in
either 1 or q + 1 points, the size and this intersection property is equivalent to the
16. Unitals, semiovals
99
definition. It is also obvious by elementary counting that there is a unique tangent

line through each point of a unital. Hence unitals are minimal blocking sets (of
maximum possible size, in fact) and semiovals as well.
The absolute points of a unitary polarity of PG(2, q 2 ) form a unital, it is called
the classical unital or a Hermitian curve. This curve is projectively equivalent to
X1q+1 + X2q+1 + X3q+1 = 0.
Note that for a Hermitian curve, the tangent lines of collinear points are concurrent, i.e. they meet in the polar point of that line. We will see below that this
property characterizes Hermitian curves, also giving us an opportunity to show an
application of Result 14.6 and the Segre technique.
Various characterizations of a Hermitian curve have been given. Lefèvre-Percsy
and independently Faina and Korchmaros proved the following:
Result 16.1. Let H be a unital in PG(2, q 2 ). If every (q + 1)-secant of H meets H
in a Baer subline then H is a Hermitian curve.
Hirschfeld, Storme, Thas and Voloch gave a characterization in terms of algebraic
curves:
Result 16.2. In PG(2, q 2 ), q 6= 2, any algebraic curve of degree (q+1), without linear
components, and with at least q 3 +1 points in PG(2, q 2 ) must be a Hermitian curve.
Thas [133] proved the following
Theorem 16.3. Let H be a unital in PG(2, q 2 ). If the tangents of H at collinear
points of H are concurrent, then H is a Hermitian curve.
The next lemma was the first step in Thas proof:
Lemma 16.4. Let H be a unital in PG(2, q 2 ) with the property that the tangents of
H at collinear points of H are concurrent. Let P1 , P2 , P3 be non-collinear points
of H and let Li be the tangent of H at Pi . Choose the coordinates in such a way
that L1 L2 = (0, 0, 1), L2 L3 = (1, 0, 0), L1 L3 = (0, 1, 0). If P1 = (0, b, 1),
P2 = (1, 0, c) and P3 = (a, 1, 0) then (abc)q+1 = 1.
The proof is a nice example of the Segre-like arguments, see the first part of
Theorem 14.9. We leave it as an exercise.
Exercise 16.5. Prove the lemma of Thas above!
In his original proof Thas showed, using this lemma, that every (q + 1)-secant
meets H in a Baer subline, then by Result 16.1 H is classical. Here we follow the
proof of [134], which uses Result 16.2 instead.
Proof of Theorem 16.3, sketch. Let L1 , L2 and L3 be non-concurrent tangents
of H at the tangent points P1 , P2 and P3 . By the lemma and Segres Theorem
100
14.3, there is an algebraic curve C 0 intersecting each Li at Pi with intersection

multiplicity q + 1. Next, let Q 6 H and let M1 , M2 , ..., Mq+1 the tangents of H
through Q. Then the tangent points Q1 , Q2 , ..., Qq+1 are collinear, let ` be the line
containing them. The curve of degree q + 1, consisting of ` counted q + 1 times,
intersects Mi at Qi with intersection multiplicity q + 1. Now from Result 14.6 it
follows that there exists a curve C of degree q + 1 intersecting each tangent of H at
its tangent point with intersection multiplicity q + 1. As C contains all points of H,
it has at least q 3 +1 points. If C contains a linear component L, then L contains all
the q 3 + 1 points of H, a contradiction. Now by Result 16.2 of Hirschfeld, Storme,
Thas and Voloch, if q 6= 2 then C is a Hermitian curve. As |C| = |H| = q 3 + 1,
C = H and H is a Hermitian curve. The case of q = 2 is easy to check.
A semioval is a set of points such that there is a unique tangent at each point.
A semioval is regular if all non-tangent lines intersect it in either 0 or a constant
number a of points.
A semioval is called almost oval if there are exactly q 1 2-secants through each
of its points. This means that an almost oval semioval C has q + a points and
through each point of C there is a unique tangent and a unique (a + 1)-secant
(a > 1). For a = 1 these sets were called seminuclear sets by Blokhuis and Bruen
in [45].
In [34] Blokhuis characterizes almost oval semiovals.
Theorem 16.6. (Blokhuis) Let C be an almost oval semioval in PG(2, q), |C| =
q+a. Then either q = a+2 and C consists of the symmetric difference of two lines,
with one further point removed from each line, or q = 2a + 3 and C is projectively
equivalent to the projective triangle, that is the set defined in Example 18.2.
The proof of this theorem is based on two propositions, the first one can be proved
using the Redei polynomial.
Proposition 16.7. Let ` be an (a + 1)-secant of an almost oval semioval C, |C| =
q + a. Then the tangents of C at the different points of ` C are concurrent.
Proof: Choose coordinates so that ` = ` and let U = C \ `. Let U = {(ai , bi ) :
i = 1, . . . , q 1} and consider the affine Redei polynomial R(X, Y ) of U . For an
(m) ` C we have that R(X, m)|(X q X). Since degX (R) = q 1, for these
(m)s the quotient (X q X)/R(X, m) can be computed; it has to be X r1 (m).
This means that the line Y = mX + r1 (m) is the tangent of C at (m). Since
deg(r1 ) = 1, r1 (m) = am + b, so the tangents are the lines Y = mX + am + b.
These lines pass through (a, b).
The following generalization of Proposition 16.7 can be proved in exactly the same
way as the original one.
16. Unitals, semiovals
101
Proposition 16.8. Let U be a set of q 1 points in AG(2, q), which determines at

most q 1 directions. Then there is a unique point w such that U {w} determines
the same directions as U .
Proof: Take an infinite point (m) not determined by U . Then there is a unique
line tm with slope m which does not meet U . These lines pass through a point w;
this can be proved exactly as in the proof of Proposition 16.7. (The details are left
as an exercise.)
This result can be generalized further, namely for sets of size less than q 1. This
will be discussed in Theorem 23.1.
Not much is known about semiovals, Thas proved lower and upper bounds for
their size (and nonexistence results for their higher dimensional analogues). For
regular semiovals Blokhuis and Szonyi [40] showed that a has to divide q 1 and
the points not on the semiovals are on 0 or a tangents. Thus in this case there is
a dual (regular) semioval. For regular semiovals in PG(2, q) more can be said.
Proposition 16.9. [40] Let S be a regular semioval in PG(2, q). Then S is a unital,
or the tangents at collinear points of S are concurrent.
Proof:
(Hint:) Try to copy the proof of the previous two propositions.
We have seen above that unitals with the property that tangents at collinear points
are concurrent were characterized by Thas as classical unitals (i. e. Hermitian
curves).
Using the results of Blokhuis and Szonyi, and applying an approach similar to
Thas above, G
acs in [68] could give a nice proof of Segre-type for the following.
Theorem 16.10. In PG(2, q) any regular semioval is either an oval or a unital.
To end this section we show an easy non-existence result.

Theorem 16.11. For q odd there exist no semioval of size q + 2, i.e. a set S
PG(2, q) with the property that through each point of S there pass exactly one
tangent.
Proof: Observe that through any point of S there pass exactly one 3-secant, and
the tangent and 3-secant lines together cover the points of the plane, as q + 2 is
odd, hence counting the points of S on the lines through any point P 6 S we
have to find an odd number of odd secants. Therefore they form a dual blocking
set of size q + 2 + q+2
3 . It is minimal and does not contain a dual line (pencil),
102
hence by Sz
onyis Corollary 21.21, every dual line intersects it in 1 mod p points.
But there are intersections of size 2 (with pencils centered at the points of S),
contradiction.
17
Sets without tangents: where Segre meets Redei
This section is based on [42], [41] and [130]. Sets without tangents were introduced
by Blokhuis, Seress and Wilbrink [42], who called them untouchable sets. As a set
of size q + 1 always has a 1-secant (through each point of it), an untouchable set
has at least q + 2 points. For q odd, Blokhuis, Seress and Wilbrink
proved that the
size of a set without tangents in PG(2, q) is at least q + 14 2q + 2. For q even, the

plane PG(2, q) always has a hyperoval, which is an untouchable set of minimum
cardinality; here the question is to find the size of the second smallest untouchable
set. When all lines intersect a set in an even number of points, then such a set
is a codeword in the dual code of the plane. For codewords of small weight, see
Chapter 6 of the book [3] by Assmus and Key. Note that the cardinality q + 4 can
be realized in the planes of order 4, 8 and 16, but probably in no other planes.
All known examples have at least q + q points, see Korchmaros, Mazzocca [88].
They constructed q + t-sets of type (0, 2, t) for t q. More examples and a proof
that the t-secants are concurrent for such sets can be found in Gacs, Weiner [70],
see Theorem 13.12.
As mentioned in [42], there are untouchable sets of size 2q, and this is best possible
for q = 3, 5. Smaller examples were also constructed, the best one for odd qs has
size 2q q/p, where q = pn . For more constructions in the case q odd, and for
planes of small order, we refer to the original paper [42]. For q even see some
examples from [41] at the end of this section.
The method introduced in the paper by Blokhuis, Seress and Wilbrink is to associate a curve to an untouchable set, whose points correspond to the lines intersecting the set in at least 3 points. This is a nice application of Segres Lemma
of tangents. Their bound follows from estimating the number of singular points of
this curve. For q even, the construction of the curve still works, but typically we
have no information on the number of its singular points. In the next section we
briefly recall the construction of these curves.
The Segre-curve, here containing exactly the > 2-secants of a pointset is, intuitively, the Redei-curve minus the points corresponding to 2-secants (if there are
no tangents). The result we are going to prove here is the following.
Theorem 17.1. Let S be a set of points in PG(2, q), without tangents. Then
(a) if q is odd then |S| > q + 2 +
1
8 q;
17. Sets without tangents: where Segre meets Redei

(b) if q is even and S has an odd line then |S| > q + 2 +
103
q
1
6 q.
In Section 17.2 we show an improvement by the Szonyi-Weiner resultant method.

In the case when q is even they proved that a set without tangents, having an odd
line, is of size > q + 3b qc 7 if q > 16. Their proof is very short, and elegant,
but it is based on other strong results of their stability method.
17.1
The proof using Segres method
Curves associated to untouchable sets using Segres method

Let S be a set without tangents in PG(2, q), 2|q, |S| = q + 2 + . Throughout this
section wepsuppose that q > 62 + 5 + 1, at some points we only need that is
less than q/6. Since |S| is close to q + 2, and the set has no tangents, almost all
lines intersect it in 0 or 2 points. We wish to show that the lines intersecting S in
more than 2 points are contained in an algebraic envelope of degree . For this we
will need the generalization of the theorem of Ceva, see Theorems 14.5,14.3.
Lets remember that if k > n, then the envelope C of Segre is unique. Let us also
underline that the envelope does not contain any 2-secant of the k-arc A.
Lemma 17.2. Let T be any triangle contained in S, (S is a set without tangents
and |S| = q + 2 + ), with the property that the sides of this triangle are 2-secants.
(a) Suppose that q is even. Consider the i-secants through the vertices of T ,
each with multiplicity (i 2). Then these lines are contained in an algebraic
envelope of degree .
(b) Suppose that q is odd. Consider the i-secants through the vertices of T , each
with multiplicity 2(i 2). Then these lines are contained in an algebraic
envelope of degree 2.
Proof: Looking at S from a vertex of T we see that the sum of the multiplicities
of the i-secants through this vertex is just . Consider a (q 1 + ) 3 matrix
whose rows are indexed by the points of S \ T , and the columns are indexed by the
vertices A0 , A1 , A2 of T . As the notation suggests, choose T as the fundamental
triangle. Put the coordinate of the line P Ai in position (P, Ai ) of this matrix.
Then the product of the coordinates in each row will be 1. In each column we
see the coordinates of the i-secants (i 1) times each, and every other element
of GF(q) once. Since the product of the elements of GF(q) is 1, the product of
the remaining elements is (1)q+ . This simply means that the i-secants with
multiplicity (i 2) satisfy the condition in Result 14.3 if q is even (i.e. 1 = 1),
but they do not satisfy it when q is odd. However, by the argument above, the
i-secants with multiplicity 2(i 2) will satisfy the required condition, we pay for
it by taking square, i.e. the resulting curve will be of degree 2.
104
Note that when is small then most of the lines will be 2-secants or skew to S.
We want to collect, as usual when using Segre-curves, the rare, irregular lines
in one curve (envelope).
Lemma 17.3. If 62 + 5 + 1 < q, then there exists an algebraic envelope of degree
in the even and 2 in the odd case, containing all the i-secants of S with i > 2,
but containing no 2-secants of S.
Proof: On the points of S we form a graph as follows: vertices of are the
points of S, two points are adjacent if the line joining them is an i-secant with
i > 2. Since the number of i-secants through a point of S is at most , there are
at least q + 1 lines through the points which are 2-secants. Hence the degree
of a point is at most 2. One can show that there is a set A S of at least
3 independent points, for which Result 14.5 can be applied, and one can glue
together the locally obtained curves if (2 + 1)(3 + 1) > q.
The proof in the q = odd case
We have a curve C of degree 2 with the property that it intersects each pencil
of S in any i-secant line ` of it with multiplicity 2(i 2). As for each of the i
points (pencils) on `, C has a multiple intersection in `, it follows that ` cannot
be a simple point of C. We are going to use Result 10.3, stating that a curve
without multiple components can only have a very limited number of multiple
points: for a projective plane curve F of degree n without multiple components
P
mP (mP 1)
n2 holds.
P F
2
So we get rid of the multiple components: write C = E D2 , where E no longer
contains multiple components. The essential observation is that E is of positive
(even) degree as otherwise D would be a curve of degree with the property that
it intersects each pencil of S in every one of its i-secant lines ` with multiplicity
(i 2), but we have seen that
P such a curve does not exist. As C fully intersects
every pencil P S (since `3P 2(i` 2) = 2), the same holds for E. It follows
from the fact that the intersection numbers are still even (we subtracted an even
number from 2(i 2)) and if an the intersection number is positive for one pencil
P containing ` then ` E so the intersection number is positive (and 2) for all
i` points on P . As a consequence, ` is a multiple point of E.
For ` E let (P ) denote the multiplicity of ` on E and for a pencil P S let
(`, P ) be the intersection multiplicity of E and the pencil P at `. Then (`, P )
(`), where > holds if and only if the pencil P is a tangent of E at `. Also the
number of pencils P S that are tangents at ` is at most (`). Let the degree of
E be 2m; since E fully intersects each pencil P S, we get
XX
P S `3P
(`, P ) = 2m(q + 2 + ).
105
Lets change the order of summation and use the following estimate:
X
(`, P ) (`)( (`)) + 2m(`) 3(P ).
()
P `
This estimate is obtained as follows: if the pencil P is a tangent (here we

mean tangent in the algebraic sense, so this neither implies or is implied by
|E the pencil P | = 1) of E at `, then we know no better than that (`, P ) 2m,
and this happens at most (`) times. For the remaining points of ` we have
(`, P ) = (`). Finally we used 2m 2. Using Result 10.3 we obtain
X
(`)((`) 1) 2m(2m 1).
()
È
We now combine () and (), where in () we restrict ourselves to those

P È
that are on a point of S. For those `, we have that (`) 2, so certainly ` (`)
2m(2m 1). Combining this with () we get
(2m 1)3 q + 2 + .
Since 2m 2, we get q 62 4 2 < 62 + 5 + 1, a contradiction.
The proof in the q = even case: Segre and Redei hand in hand
Here the idea is to combine the above method with a Redei polynomial approach
introduced for (k, n)-arcs in [125]. Besides the Segre-curve of the previous sections, another curve will be associated to the untouchable set, with many similar
properties as the Segre-curve, implying that they have many points in common.
Finally, we can show that the two curves have a common linear component and
derive a contradiction from this fact, if the size of the set is not far from q +2. This
second method does not work when the set is contained in the dual of the code
generated by the lines of PG(2, q), that is when each line intersects our untouchable
set in an even number of points.
Now another curve will be associated to the untouchable set. This method relies
on the paper [125]. Again, the geometric statements will be proved and we will
use the algebraic results of Section 9.5.
Consider a line ` which will be the line at infinity. Let U = S \ ` = {(ai , bi ) : i =
1, . . . , |U |} and write the Redei polynomial as
R(X, Y ) =
|U |
Y
i=1
(X + ai Y bi ) =
|U |
X
j=0
rj (Y )X |U |j .
106
Note that degY (rj ) j. First let us try to interpret in terms of R(X, y) the fact
that through (y) there pass exactly s odd lines. (Here an odd line is an i-secant
with i odd. Note that i 3 automatically.) From now on the index of a point will
be the number of odd lines passing through it.
Lemma 17.4. The point (y)
/ S has index s if and only if the greatest common
divisor of R(X, y) and X R(X, y) has degree exactly |U | s.
Proof: Since the characteristic is 2, for (y)
/ S the polynomial R(X, y) has roots
with even multiplicity, and roots corresponding to the odd lines. If we consider
X R(X, y), then the roots of even multiplicity remain with at least this multiplicity
and the multiplicity of the other roots decreases by 1. In other words, the greatest
common divisor of R(X, y) and X R(X, y) has degree exactly |U | s.
Bounds for the number of odd lines is also needed in the proof.
Lemma 17.5. The total number of odd lines is at most |S|/3 and at least q + 1.
Look
around
from a point not
in the blocking
set!
Proof: As we have seen in the previous part, the number of i-secants, with i 3,
through any point of S is at most . Therefore, the total number of odd lines is
at most |S|/3. For the lower bound note that when |S| is even then the number
of odd lines passing through any point is even, hence the odd secants form a dual
even set. When |S| is odd, the number of odd lines passing through a point is odd,
so they form a dual blocking set. It is well-known that a blocking set of PG(2, q)
is of size at least q + 1.
The main idea of the proof is that for a fixed Y = y the polynomial
b(X, y) = R(X, y)/ gcd(R(X, y), X R(X, y))
can be expressed using the coefficients r1 (y), ... of R(X, Y ). Choose a parameter s
and consider the polynomial b = bs of degree s in X, which is obtained by solving
the system of linear equations in the proof of Theorem 9.34 (s is not fixed yet!).
We will try to choose a typical index s for the line at infinity; that is an s for
which through most of the points there pass exactly s odd lines. The next lemma
says that such a typical index exists.
Proposition 17.6. Assume that ` intersects S in 0 or 2 points. Then there is a
unique s = s(`) such that s and at least q q points on ` have index at
most s and at least 56 q q of them have index exactly s.

Proof: Since there are at most |S|/3 q/2 odd lines, the average index of a
point is at most /2. Hence there are at least q+1
2 points on the line at infinity
with index at most . Go down from and stop at the largest s , where there
107
are points of index s. Then the determinant of R = Rs in Theorem 9.34 is not

identically zero as a polynomial in Y , and we know the following from Theorems
9.34 and 9.35:
(A) There are at most s(s 1) points for which det R = 0.
(B) The points whose index is less than s are among these points, and det R 6= 0
for points of index s.
Hence there are at least q+1
2 s(s 1) points that have index exactly s. Using
the same trick, the first t > s for which there can be points of index t must satisfy
at most s the determinant of Rt in
t(t 1) q+1
2 , since for the points of index p
Theorem 9.34 must vanish. This implies t q/2. Since the average index is at
most |S|/(3(q + 1)), there are at most |S| 2/(3 q) points having index larger
p

than s. This is certainly at most 2( q + 1)/3 q, since q/6. We

already saw in (A) that there are at most s(s 1) q/6 points of index less than
s, so the result follows.
Proposition 17.7. Let s be the typical index for the line ` at infinity. Then there
is a curve c` (X, Y, Z) with the following properties:
(1) degX (c` ) s, deg(c` ) s2 ;
(2) c` (x, y, 1) = 0 implies that [y, 1, x] is an odd line, if the point (1, y, 0) has
index s.
Proof: Denote the typical index by s. By Lemma 17.4 this means that the g.c.d.
of R(X, Y ) and X R(X, Y ) has degree degX R s. Consider c` as a homogeneous
polynomial constructed in the proof of Theorem 9.34. By Result 9.35, the polynomial c(X) det R = c` has X-degree s and total degree at most s2 . If a fixed point
(1, y, 0) of the line at infinity has index s, then the odd lines through it correspond
to the roots of c`s (X, y). This also shows (2).
The two curves have a common component
In this section we would like to choose the line at infinity to be a 2-secant of S, so
that the typical index of it s > 0 holds. Next we show that such 2-secants exist.
Lemma 17.8. Assume that there is a line intersecting S in an odd number of points.
Then there exists a 2-secant of S, such that the typical index for it is not zero.
Proof:p The envelope C(X0 , X1 , X2 ) obtained by Segres method has degree less
than q/6 and it contains all the odd secants of S, hence by Bezouts theorem
108
p
any pencil with vertex
must be a component of C. So
p having index at least q/2 p
there are at most q/6 points with index at least q/2.
p
Choose a 2-secant ` that contains no point with index at least q/2. (Such a 2secant exists, since
p through a point of S there passpat least (q +1) 2-secants and
there are only q/6 points with index at least q/2.) Assume that the typical
index on ` is zero. Note that there are no points of ` \ S with index greater than
the typical index of p
`. (Since by the proof of Proposition 17.6 such a point would
have index at least q/2.) Hence each point of ` \ S has index zero and the odd
lines intersect ` in the points of S `. So there can be at most 2 odd lines in
total, which contradicts our lower bound for the number of odd lines, see Lemma
17.5.
Let the line at infinity be a 2-secant of S and assume that for the typical index
s > 0 holds. Then the curve C(X0 , X1 , X2 ) obtained by Segres method has X2 degree (which is its total degree). Substitute X0 = Y , X1 = 1, and X2 = X in C
to get the curve g(X, Y ). According to the results in the second section, this curve
contains all odd lines. Similarly, let us denote by c(X, Y ) the curve constructed in
the previous section.
Lemma 17.9. The curves c and g have a common component.
Proof: The curve c has at least ( 56 q q)s points, which correspond to odd
lines. These points are also points of the curve g. If the two curves had no common
component, then Bezouts theorem would give that
5
s( q q) s2 ,
6
which is impossible, since s .
Now we are able to prove our main theorem.
Theorem 17.10. The size
pof an untouchable set in PG(2, q), q even, having odd
lines, is at least q + 1 + q/6.
Proof: Assume to the contrary, that there existspan untouchable set in PG(2, q),
having odd lines, with size q + 2 + , where < q/6 1. Note that then 62 +
5 + 1 < q. Consider a common (absolutely irreducible) component of c and g
guaranteed by the previous lemma. Let the X-degree of this component be u,
which is its total degree. Then the component has at least ( 65 q q)u points.
Using the Weil bound one immediately sees that this component has to be linear.
But that implies the existence of a point with at least ( 56 q q) odd lines through
it. This is a contradiction, since such a set would have at least 3( 56 q q) points.
17.2
109
Untouchable sets in Galois planes of even order
Here we improve Theorem 17.1 in the case when q is even. The proof is based on
Theorem 23.22.
First we show that a not too large untouchable set is very close to be a set of even
type, hence we can apply Theorem 23.22 to such sets.
Lemma 17.11. Let U be a set without tangents in PG(2, q), q even. Assume that
U has q + 2 + points, then the number of odd-secants is at most |U |/3.
Proof: Through any point of U there pass at most odd-secants. An odd-secant
contains at least 3 points of U , therefore the total number of odd-secants is at
most |U |/3.
Theorem 17.12. (Sz

onyi-Weiner [130]) The size of an untouchable set U in
PG(2, q), 16 < q even, having odd lines, is larger than q + 3b qc 7.
Proof: Assume that |U | q + 3b qc 7. By Lemma 17.11, the number of odd
secants, of U is less than (b qc + 1)(q + 1 b qc). By Theorem 23.22, we
can construct a set of even type from U by modifying d q+1

e 1 points. If P is
a modified point, then through P there pass at least q + 1 (d q+1

e 1) oddsecants of U . But then counting the points of U on the lines through P we get
that |U | > 2(q b qc), when P U , and |U | > 3(q b qc), when P 6 U ; which
is a contradiction.
Concluding remarks
In this section first we give examples for small sets without tangents but having
odd lines. First of all note that the union of two non-disjoint hyperovals H1 and
H2 is always such a set of size 2(q + 2) |H1 H2 |. In [86] it was showed that in
PG(2, 16) there exist two hyperovals (a Lunelli-Sce and a regular one) intersecting
each other in 8 points; hence there is a non-even type, untouchable set of size 28
in PG(2, 16). By the construction method introduced in [70], the above set can
be used to obtain non-even type, untouchable sets in PG(2, 16h ), (h 2), of size
7
h
h1
and of size 74 16h .
4 16 16
For untouchable sets in planes of even order we were lucky in the sense that both
Segres method and Redeis method worked. The Redei method can be generalized
to the following situation: let S be a set having only 0-secants and i-secants, where
i p, q = ph . Then the size of S is at least qp q + p, and in case of equality
S is a maximal arc. By Theorem 15.12 of Ball, Blokhuis and Mazzocca [13], such
an arc does not exist if p is odd. Even in that case, one can show that |S| has
to be at least qp q + p + with 4 q/2, if S has an i-secant for which i is
110
not divisible by p. For this, one has to copy more or less the entire proof in [125].
The reason why the result is weaker than for our case p = 2, is that the method
based on Segres lemma of tangents does not seem to work in this case. Similarly,
instead of i p one can suppose that i pr and the situation is more or less the
same in this case, too.
When pr > 4 q/2, then a better bound can be obtained by simple counting arguments:
Exercise 17.13. Let S be a set of size nq + n q + which intersects every line in
either 0 or at least n points for some n = pr . Then n.
18
Directions
The theory of directions determined by a pointset is one of the most prominent

ones in finite geometry. There are several applications of it, within geometry (e.g.
blocking sets of Redei type), in group theory (Hajos-Redei type results), for permutations, etc. One can see the original question from at least four points of view.
1st form
2nd form
3rd form
Consider a pointset U = {Pi (ai , bi ) : i = 1, ..., |U |} AG(2, q). One can say that
U determines the direction/infinite point (m) ` , where m GF(q) {} if
b b
there are points Pi and Pj such that m = ajj aii . The set of determined directions
will be denoted by D, its cardinality by N = |D|. When |U | > q then D = ` (i.e.
N = q +1) by the pigeon hole principle, as through any (m) ` there are q affine
lines, so at least one of them will contain at least two points of U , determining
(m); this shows that only |U | q is interesting. The most examined case is when
|U | = q, we are going to discuss it in details, but some results concerning smaller
U can be found in this book, like Theorem 23.1, etc. The general feeling is that if
|U | is close to q then it can be extended to a q-set determining the same directions,
while small pointsets
U can behave as they want. Note that a random pointset
p
of size at least 2q log(q) determines every direction with high probability.
Suppose U is a set of q points which does not determine all directions. Then
applying an affine transformation, we can achieve that the direction of vertical
lines is not determined. But this means that our point set in question can be
considered as the graph of a function f (X), over the underlying field. Since over
a finite field every function can be represented by a polynomial, we may assume
f GF(q)[X]. (Note that lines correspond to linear polynomials.) We will say that
a polynomial f determines a direction, if its graph determines it, which means that
(y)
: x 6= y}. This was the
the set of determined directions is Df = D = { f (x)f
xy
original question Redei considered, he tried to determine the number of difference
quotients a polynomial can have.
It is easy to verify that D = {c GF(q) : f (X) cX is not bijective }, so a third
formulation of our problem is to look for polynomials, for which there are many
18. Directions
111
c-s such that f (X) cX is a permutation. Compare with the definition of Mf in

Section 9.
In his book [103] Redei proved deep theorems about fully reducible, lacunary
polynomials. As an application of his algebraic results, he considered the problem
of determining the possible values N = Nf = |Df | can take, that is, how many
different directions a point set can determine. For an arbitrary prime-power q, he
proved that for Nf < q+1
2 , Nf cannot take all values, it has to be an element of
the union of some intervals.
For q = p prime, he proved much more, namely that unless U is a line (and hence
p+1
N = 1), N is at least p+1
2 . Later Megyesi realised that the case 2 is impossible,
so their result can be formulated as follows:
Theorem 18.1. (Redei and Megyesi [103]) If a set of p points in AG(2, p) is not a
line, then it determines at least p+3
2 directions.
This bound is sharp, as the (unique) Example 18.2, constructed by Megyesi, shows,
see below. Unicity is proved by Lovasz and Schrijver, see Theorem 19.1. For the
proof of Theorem 18.1, which is a classical lacunary polynomial argument, see the
first part of Theorem 18.14.
Now we present a series of examples constructed by Megyesi [103] for arbitrary
prime-powers.
Example 18.2. For a d|q 1, 1 < d < q 1, let G be the multiplicative subgroup
of GF(q) of order d. Set U = {(x, 0) : x G} {(0, x) : x
/ G}, that is put G and
the complement of G on the x and y axes, respectively. An easy calculation shows
that the number of determined directions is q + 1 d.
Perhaps it is useful to give Example 18.2 in the graph form as well. In general, a
polynomial giving the example on two lines with N = q + 1 d is XG , where
G is the characteristic function of G, the multiplicative subgroup of order d. A
q+1
2 .
polynomial having N = q+3
2 is f (X) = X
We remark that the Megyesi example belonging to N = q+3
2 , together with the
determined directions as points at infinity, is called the projective triangle. It is a
set of size 23 (q + 1) in PG(2, q).
After the publication of Redeis book, it turned out that the problem of determining the possible values of N and looking for sets determining a small number
of directions has an important application: a certain type of blocking sets (called
blocking sets of Redei type) in PG(2, q) of size q + N is equivalent to a set of q
points in AG(2, q) determining N directions.
Exercise 18.3. Show that for a blocking set B PG(2, q) and a line `, |B \ `| q
holds, and in case of equality B \`, considered as a pointset of AG(2, q) = PG(2, q)\
`, determines (as determined directions) the necessary points of B `.
4th form
112
In the general case today everything is known for N < q+3

2 : Ball, Blokhuis,
Brouwer, Storme and Sz
onyi [43], and then the beautiful improved proof of Ball
[9] determined all possible values, see Theorem 18.14.
18.1
When D is (contained in) a subgroup of GF(q)
Given D ` = GF(q) {}, let FD denote the set of all polynomials f determining directions from D, i.e. for which Df D.
Lemma 18.4. [56] Let D be a non-trivial subgroup of the multiplicative group
GF(q) . Then the polynomials in FD , with the operation of composition, form a
permutation group of the ground set GF(q).
As (0) 6 D, each f FD is a permutation. Let f, g FD , then
(g(y)) g(x)g(y)
= f (g(x))f
, so the product of two elements of D,
g(x)g(y)
xy
hence also in D.
Proof:
f (g(x))f (g(y))
xy
3/2-transitive:
transitive
and
all
stabilizers
have orbits of
equal size.
3/2-trans.
primitive or
Frobenius
group.
Exercise 18.5. Prove that the permutation group FD is

lin
3/2-transitive on GF(q), moreover, even the subgroup FD
= {f (X) = aX + b :
a D, b GF(q)} FD is 3/2-transitive on GF(q).
Exercise 18.6. [56] Prove that if D GF(pe ) for some subfield GF(pe ) < GF(q)
lin
, i.e. the only functions determining directions from D are the
then FD = FD
linear ones with slope from D.
As the multiplicative group GF(q) is cyclic, its subgroups are exactly of the form
D = {x : xd = 1} for some d|(q 1). We are going to prove the following theorem
of Carlitz, McConnel, Bruen and Levinger [56]:
Theorem 18.7. If Df D = {x : xd = 1} for some d|(q 1) then f is of the form
j
j
f (X) = a + bX p , where a GF(q), b D and m = q1
d divides (p 1).
Proof: ([56]) Let V denote the vectorspace of functions GF(q) GF(q), or, equivalently, the space of polynomials in GF(q)[X] of degree q 1. The element g of
FD acts on f V as f g (X) = f (g(X)).
This proof will go in four steps.
The first is about the linear space of homomorphisms V V , which commute
with elements of the permutation group FD ; that is let
HomFD (V, V ) = { : V V | ( f (g(X) ) = (f )(g(X)) for all f V, g FD }.
Define also the monomial functions i (X) = X id for i = 0, 1, ..., m = q1
d . Finally
recall a (X) = 1 (X a)q1 , the characteristic function of the set {a}, see
Section 5.3.
18. Directions
113
Step 1. If HomFD (V, V ) then (0 ) can be written as a linear combination ofP

the monomials i , i = 0, ..., m. Conversely, given any linear comm
bination P i=0 i i , there is a unique element HomFD (V, V ) such that
m
(0 ) = i=0 i i .
To prove this we need the following
Exercise 18.8. The dimension of HomFD (V, V ) over GF(q) is just the number of
the orbits of (FD )0 (the stabilizer of 0) on GF(q), i.e. m + 1.
Ta
For a GF(q) define the translation
Ta (X) = X
P
Pa. Then 0 T=a a . As any
f V has the form f (X) = aGF(q) f (a)a (X) = aGF(q) f (a)0 (X), for any
HomFD (V, V ) and f V we have
X
X
(f ) =
f (a)(T0 a ) =
f (a)(0 )Ta ,
aGF(q)
aGF(q)
hence is uniquely determined by the image of 0 .

To finish Step 1 we have to prove that (0 ) is some linear combination of the
Pq1
lin
monomials i , i = 0, ..., m. Let (0 )(X) = i=0 i X i and f (X) = bX FD
f
(so b D). Then 0 (X) = 0 (bX) = 0 (X), so
(0 )(X) = (f0 )(X) = (0 )f (X) =
q1
X
i (bX)i .
i=0
Pq1
i
i
i
Thus
i=0 i (1 b )X = 0, hence all the coefficients i (1 b ) are zero for
i = 0, 1, ..., q 1. As b was an arbitrary element of D, it means that i = 0 unless
d|i and we are done.
Step 2. We prove that for each k = 0, 1, ..., m, Uk = h(X a)kd : a GF(q)i is an
FD -module (i.e. for any g FD , f Uk also f g Uk ).
To see this observe that if is in HomFD (V, V ) then (V ) is an FD -module. By
Step 1 there exists a in HomFD (V, V ) for which (0 )(X) = X kd . Further
(a )(X) = (T0 a )(X) = ((0 ))Ta (X) = (X a)kd .
Step 3. Recall Exercise 5.8. There we defined S = h(X a) : a GF(q)i
Pn1
and M = { i=0 i pi : 0 i i }. Since d|(q 1) we have (d, p) = 1 and
1, (d 1) Md . In Step 2 we have seen that Sd is an FD -module. Thus, for any
g FD and r Md , gr (X) = r (g(X)) = g(X)r is a polynomial in Sd . We are
going to use this.
Lemma 18.9. Let f FD . Then f (X) = a + bX t , where a = f (0), b = f (1) f (0)
and td d (mod q 1).
Pn1
i pi
i=0
0 i < p
r (X) = X r
114
d1
reduction
Xq X
after
mod
Proof: Let a = f (0). Since the translation Ta is in FD , (X) = Ta (f (X)) =

f (X) a is also in FD and (0) = 0. As 1, (d 1) Md , the functions
1 =
(x)
d1
and
=
(after
reduction)
are
in
S
.
Now,
F
implies
that
=
d
D
1
x
(x)(0)
d
d
D for x 6= 0. From the definition of D, it follows that X = (X) =
x
(X)(X)d1 . But X d can be a product of two polynomials of degree d only
if (X) = bX t and (X)d1 = b0 X dt , where bb0 = 1 and 0 < t < d. Then
d (X) = bd X td = X d , if and only if td d (mod q 1) and bd = 1. Further,
b = (1) = f (1) a = f (1) f (0).
Step 4. We now show that the only possible choices for t above are t = pj . W.l.o.g.
we may assume that f FD has the form f (X) = X t . FD is a group containing
the translations Ta . Thus for any fixed 6= 0, h(X) = f (T (X)) = (X )t is
0
in FD . By Step 3, h(X) = a + bX t = (X )t , where a = h(0) = ()t and
0
b = h(1) h(0). We have the equation (X )t (a + bX t ) = 0 as it is of degree
< q 1. Hence t = t0 , b = 1 and (X )t = X t + ()t . Since was arbitrary,
this shows that f (x + ) = (x + )t = xt + t for all x, GF(q), so f is an
automorphism of GF(q), hence t = pj and by Step 3, d(pj 1) 0 (mod q 1).
Exercise 18.10. Prove that if D = {x GF(q) : xd = } for some fixed for which
q1
j
j
d = 1, then FD consists of the functions f (X) = a + bX p where q1
d |(p 1),
a GF(q) and b D.
q = ph
The theorem above has some further applications in geometry. In Section 18.4
we will see that pointsets of size q determining at most q+1
directions can be
2
classified: they are GF(pe )-linear pointsets in AG(2, q), where GF(pe ) is a subfield
of GF(q). It also means that the set D of determined directions is a GF(pe )-linear
pointset of (the line at infinity ') PG(1, q), generated by he + 1 points. (For the
definition of generation see Section 20.) So one can use that we are in the situation
that {(xd ) : x GF(q)} h(a1 ), ..., (am+1 )iGF(pe ) for some a1 , ..., am+1 GF(q).
One can also try to alter the proof above a little bit to prove the following
Conjecture 18.11. If Df D = {x : xd = 1} {0} for some d|(q 1) then f is of
j
j
the form f (X) = a + bX p , where a GF(q), b D and m = q1
d divides (p 1).
It follows from Theorem 18.1 of Redei and Megyesi that if q = p is a prime then
the Conjecture holds and f is a linear function.
Exercise 18.12. Suppose that a set of disjoint parabolas {Y = ai X 2 + bi : i =
1, ..., q} partition AG(2, q). Then (i) the bi -s are all distinct; (ii) writing ai = f (bi ),
the directions determined by the function f are minus non-squares and possibly (0);
finally (iii) if q = p is a prime and we do not allow a line as a degenerate parabola
then all ai -s are equal and they are the vertical translates of the same parabola.
18. Directions
18.2
115
When D is (contained in) a subgroup of
q+1
1 GF(q 2 )
Well, it almost never happens, if we mean proper subgroups, see Theorem 29.5. For
this being meaningful we have to put our pointset U into the affine plane AG(2, q)
which is identified with GF(q 2 ) and the directions with the (q +1)-th roots of unity.
Theorem 29.5 says that, given 1 < d < q + 1, if U determines directions from one
coset of the multiplicative subgroup of order d, then U is a line.
Aart Blokhuis has recently asked the opposite question: is it possible that U does
not determine any direction of the multiplicative subgroup of order d (unless U is a
line)? Such (positive or negative) results would give rise several applications. One
may feel that such sets U probably do not exist. For d = q+1
2 Blokhuis theorem in
[29], so Theorem 29.5 shows this. The next possible case d = q+1
3 is already open
as far as I know.
18.3
Sets determining
q+1
2
directions: affine linear pointsets
Linear pointsets has gained an important role in the theory of blocking sets. First
we give the definition of affine linear pointsets:
Definition 18.13. A pointset S AG(n, q) is called GF(pe )-linear if
(i) GF(pe ) is a subfield of GF(q), and
(ii) there is an affine space AG(n0 , q) containing AG(n, q) such that S is a oneto-one projection of a subgeometry AG(t, pe ) AG(n0 , q) from a suitable subspace
(vertex) V onto PG(n, q).
Algebraically this means that if we suppose that S contains the origin and has size
|S| = (pe )t , then one can choose t points (vectors) v1 , ..., vt of
q) such that S
PAG(n,
t
is the vectorspace spanned by them over GF(pe ), i.e. S = { i=1 i vi : 1 , ..., t
GF(pe )}.
Suppose that we have an affine pointset S with the suspect that it is GF(pe )-linear
in the affine sense. W.l.o.g. suppose that the origin is in S. Then S is GF(pe )-linear
iff (i) (a1 , b1 ), (a2 , b2 ) S implies (a1 + a2 , b1 + b2 ) S and (ii) (ca1 , cb1 ) S for
all c GF(pe ). Now changing the representation to the GF(q 2 )-one, we conclude
that S is GF(pe )-linear iff substituting Y = GF(q 2 ) \ GF(q) into its affine
Redei polynomial R(X, Y ), R(X, ) contains terms with exponents being powers
of pe only. As R was defined over GF(q), it is equivalent to saying that all the
X-exponents of R(X, Y ) are powers of pe .
Everything is similar in AG(n, q) for bigger n. But it gets much harder if S is
non-affine, see Section 20.
so e|h
116
18.4
Small blocking sets of Redei type are linear
Here we present a short proof of the following theorem of Blokhuis, Ball, Brouwer,
Storme and Sz
onyi [43], which was refined and turned to its current beautiful form
by Ball [9]:
Theorem 18.14. Let |U | = q be a pointset in AG(2, q), q = ph , p prime, and let
N be the number of directions determined by U . Let s = pe be maximal such that
every line intersects U in a multiple of s points. Then one of the following holds:
(i) s = 1 and q+3
2 N q + 1;
(ii) GF(s) is a subfield of GF(q) and qs + 1 N q1
s1 ;
(iii) s = q and N = 1.
Moreover, if s 3 then U is a GF(s)-linear pointset.
Proof: Before proving the lower bounds, note first that the upper bounds are
trivial. Let P be a point of U . If s > 1 then the points of U lie on lines incident
with P and with a direction in D. Therefore
N (s 1) + 1 q.
j (Y )
=
j ({ai Y bi :
i = 1, ..., q})
The degree of a polynomial g is denoted by g . We will require Lemma 5.37 by

Redei on lacunary polynomials, and Lemma 5.41 as well.
Let U = {(ai , bi ) : i = 1, ..., q} be our set determining the directions D. We
can assume that () 6 D (so one may think about U as a graph of a function
f (X) : GF(q) GF(q)), N = |D| and s = pe is maximal such that every line
intersects U in a multiple of s points. The (affine) Redei polynomial is
R(X, Y ) =
q
Y
(X + ai Y bi ) =
i=1
q
X
j (Y )X qj .
j=0
The polynomial in one variable R(X, y) has a repeated factor if and only if (y) D.
Hence for (y) 6 D
R(X, y) = X q X,
non-determined
directions
and j (y) = 0 for j = q and j = 1, 2, ..., q 2.

The overall degree of R(X, Y ) is q and (so) the degree of j (Y ) is at most j, and
the coefficient of Y j in j (Y ) is j ({ai }) = j (GF(q)). Hence for j = 1, 2, ..., q 2
the degree of j (Y ) is at most j 1 and moreover, from the previous paragraph,
has at least q N distinct zeros. So the polynomial j (Y ) = 0 whenever j q N .
Therefore
N
1
X
q
R(X, Y ) = X +
qj (Y )X j .
j=0
determined
directions
Let y D. The line c = yX Y is incident with exactly k points of U if and
18. Directions
117
only if (X + c) is a factor of R(X, y) with multiplicity exactly k. By assumption

this k is always a multiple of s, hence R(X, y) is an s-th power. We can write
R(X, y)1/s = X q/s + g(X) for some g(X) = gy (X); since s was chosen to be
maximal there exists a y D such that gy0 (X) 6= 0. It follows from Lemma 5.37
q+s
. Let N0 be maximal
that, for any y D such that g 0 (X) 6= 0 we have g s(s+1)
such that qN0 is not identically 0. We have shown that
N0
so 1=...=
q(N +1)=0
0
so
k,l ({a1 , .., aq ;
b1 , .., bq }) = 0
k+l q(N0+1)
q+s
N0 N 1.
s(s + 1)
This completes the proof of Redei and Megyesi for the case s = 1. In the remainder
of the proof s 2.
see Thm 18.1
If j 6= 1 and s does not divide j then j (y) = 0 for all y GF(q) so j (Y ) = 0.

Therefore
N0 /s
X
R(X, Y ) = X q +
qjs (Y )X js + q1 (Y )X.
j=0
In what follows the Hasse derivatives are taken with respect to Y . The k-th Hasse
derivative of R(X, Y ) is HYk (R(X, Y )) =
=
as1 as2 ...ask
(X+at Y bt ) =
t6=si
(s1 ,s2 ,...,sk )
X
(
ask
as1
)...(
)R(X, Y ).
X + as1 Y bs1
X + ask Y bsk
(s1 ,s2 ,...,sk )
(1)
For all (ai , bi ) and y D
N0 /s
q
X + ai y bi | R(X, y) (X X) = X +
qjs (y)X js .
j=0
Note that q1 (y) = 0 for y D as R(X, y) is an s-th power.

If we evaluate the equation (1) for y D and multiply by the polynomial above
k times then we get an equality of polynomials in X the right-hand side of which
contains a factor R(X, y) which divides the left-hand side. That is
N0 /s
R(X, y) | (X +
qjs (y)X js )k HYk (R)(X, y).
(2)
j=0
The polynomial
HYk (R)(X, y) =
Hasse
derivatives
N0
X
HYk (qj )(y)X j
j=0
has degree at most N0 so the right-hand side of (2) has degree at most (k + 1)N0 .
118
PN0 /s
The polynomial X + j=0
qjs (y)X js is not zero as this would imply s = 1.
If (k + 1)N0 q 1 then the left hand side of (2) has degree larger than the
right-hand side and we conclude that
HYk (qjs )(y) = 0 for all 0 j N0 /s.
Let us assume that N q/s.
Now N0 N 1 q/s 1 and (k + 1)N0 sN0 q s < q 1 whenever
k s 1. Hence for k s 1 the k-th Hasse derivatives of the polynomials qjs
are zero when evaluated at y D. We shall show that they are in fact identically
zero when js = N0 .
It follows immediately from the definition of Hasse derivatives that HYs1 (f ) is an
s-th power for any polynomial f . A zero of an s-th power is a zero of multiplicity
at least s. Hence there are at least sN zeros of the polynomial HYs1 (qN0 ).
However the degree of this polynomial is less than q N0 < sN0 < sN . Hence
HYs1 (qN0 )(Y ) 0. But Hs2 (qN0 ) is then an s-th power, it also has at least
sN zeros and is therefore identically zero. We continue by induction. Assume that
Hsi (qN0 ) 0 for i = 1, 2, ..., j 1. Then Hsj (qN0 ) is an s-th power and
if j < s then it has at least sN zeros and is therefore identically zero. Finally for
j = s we have that qN0 itself is an s-th power. However qN0 is zero for all
y 6 D and so has at least s(q N ) zeros which is far greater than its degree. But
qN0 6 0 by assumption, a contradiction.
Hence N q/s + 1, the lower bound is proved.
Now we are going to prove that GF(s) is a subfield. Without loss of generality we
can assume that (0, 0) U .
Note that
HYk (R)(X, y)0 = HYk (q1 )(y),
where f 0 is the derivative with respect to X, and note that the polynomial
q1 (Y ) =
q
Y
(ai Y bi ).
i=1
Moreover, y is a zero of q1 exactly k times if and only if the line joining y to

(0, 0) contains exactly k points, other than (0, 0), of U .
Let D D be such that for all y D the point (y) is joined to (0, 0) by an
s-secant. Then HYk (q1 )(y) = 0 for all k = 0, 1, ..., s 2 and HYs1 (q1 )(y) 6= 0.
Hence HYs1 (R)(X, y) = 0.
Counting points of U on lines through the origin
|D \ D |
q
q 1 N (s 1)
1.
s
s
18. Directions
Hence |D | N
119
q
s2
+ 1.
Let Q(X), a polynomial of degree at most sN0 q, be the quotient in the divisibility
N0 /s
R(X, y) | (X +
qjs (y)X js )s1 HYs1 (R)(X, y),
j=0
where y D and so HYs1 (R)(X, y) 6= 0. Differentiating with respect to X we

get
R(X, y)Q0 (X) =
N0 /s
(X+
N0 /s
qjs (y)X js )s1 HYs1 (q1 )(y)(X+
j=0
qjs (y)X js )s2 HYs1 (R)(X, y).
j=0
The right-hand side of this equation has degree at most (s 1)N0 < q while if
Q0 6= 0 then the left-hand side has degree at least q. We conclude Q0 = 0 and
N0 /s
HYs1 (R)(X, y)
HYs1 (q1 )(y)(X
qjs (y)X js ).
j=0
Extracting s-th roots and incorporating any constant multiple in the quotient Q
we have
N0 /s
X
R1/s Q1/s = X +
qjs (y)X js .
j=0
Differentiating again with respect to X we see R1/s (Q1/s )0 + (R1/s )0 Q1/s = 1.

Now ((R1/s )0 Q1/s ) N0 /s 1 + N0 q/s < q/s and the degree of R1/s is
q/s so (Q1/s )0 = 0. Therefore Q1/s and (R1/s )0 are constant and hence for some
GF(q)
N0 /s
X
R(X, y) = (X +
qjs (y)X js )s .
j=0
Now it is easy to check that since

N0 /s
q
R(X, y) = X +
qjs (y)X js
j=0
we have
R(X, y) h1, X s , X s , ..., X q/s , X q iGF(q) .

In particular this implies that q is a power of s and equivalently that GF(s) is a
subfield of GF(q).
120
If j 6= si for some i then j (y) = 0 for all y (GF(q) \ D) D . By comparing

the degree of j with the number of zeros of j we see that j 0 whenever
j 1 q N + N q/s2 + 1.
Let y D\D be such that R(X, y) GF(q)[X s ]\GF(q)[X sp ]. Then R(X, y)1/s =
q+s
. However by the
X q/s + g(X) and by Lemma 5.37 we know that g s(s+1)
3
previous paragraph if g 6= q/s then g < q/s which contradicts this. Hence
g = q/s2 and (g 0 ) < q/s3 . If s > 2 then Lemma 5.41 implies that
2
R(X, y) hX s , X s , ..., X q/s , X q iGF(q) .

In the final part of the proof we show that U is a GF(s)-linear set.
Let Dt D be the subset of elements of D for which y Dt implies
R(X, y) GF(q)[X t ] \ GF(q)[X tp ].
Counting points of U on lines incident with a point of U gives
|
Dt |
t>s
q 1 N (s 1)
q s2
.
ts
s(t s)
S
If s 6= si for some i then qj (y) = 0 for all y GF(q) \ ( t>s Dt ). By degrees
qj 0 for j
qs2
s(ts) .
Let t > s be minimal such that Dt 6= and let y Dt . Write

R(X, y) = X q + g(X) + h(X)t
where g(X) has non-zero terms of degree sj for some j. We have that
(ht ) <
q s2
.
s(s t)
If t is not a power of s then si < t < si+1 for some i. Extracting t-th roots and
following the proof of Lemma 5.37 we get
R1/t |(X + g + ht )h0 .
For 1 j i we have that
j
X q/s + g 1/s + ht/s = 0
(mod R1/t ).
The polynomial g 1/s has degree at most q/sj+1 and has terms of degrees that are
powers of s. We can reduce g in the divisibility modulo R1/t to a polynomial in
2
i+1
hX, X s , X s , ..., X q/s
, ht/s , ..., ht/s i.
18. Directions
121
Note that
th
q s2
q
i+1 .
s(t s)
s
So the right-hand side of the divisibility has degree less than q/t after the reduction
of g and is therefore zero. However h0 6= 0 and so
2
i+1
ht hX, X s , X s , ..., X q/s
, ht/s , ..., ht/s i
which is impossible since t is not a power of s.

If t = si for some i > 1 then we go through the same argument replacing h0 by (g 1/t + h)0 and concluding that some linear combination of
2
i+1
i1
X, X s , X s , ..., X q/s , h, hs , ..., hs
is zero. It follows from this that
2
h hX, X s , X s , ..., X q/s i.

2
Hence for all y GF(q) we have R(X, y) hX, X s , X s , ..., X q i.

Let GF(q 2 ) \ GF(q) and map AG(2, q) to GF(q 2 ) by (a, b) 7 a + b. The
images of the points of U under this map are zeros of
R(X, ) =
q
Y
(X + ai bi ) hX, X s , X s , ..., X q iGF(q2 ) .
i=1
Now R(X1 + X2 , ) = R(X1 , ) + R(X2 , ) and R(X, ) = R(X, ) for all

GF(s).
We will show a more geometric proof for this theorem, see Corollary 21.27.
This result has an analogue for arbitrary dimensional k-blocking sets of Redei
type:
Theorem 18.15. Storme, Sziklai [108] Let U AG(n, q), |U | = q k , and let D H
k1
be the set of directions determined by U . If D q+3
+q k2 +q k3 +...+q 2 +q,
2 q
e
e
then U is a GF(p )-linear set for some subfield GF(p ) of GF(q).
Note that this theorem is sharp, as the cone (cylinder), with base the q affine
points of a projective triangle in a plane 6 H , and with a (k 2)-dimensional
k1
vertex subspace contained in H \ , has q k points and determines q+3
+
2 q
k2
k3
2
q
+q
+ ... + q + q + 1 directions. For a strong improvement of this result
consult Theorems 24.14 and 24.19.
122
19
Over prime fields: the Gacs method
For the q = p prime case, as we have seen in Theorem 18.1, U AG(2, p) determines either 1 direction (i.e. U is an affine line), or at least p+3
2 directions. It was
possible to characterize equality here already in 1981:
Theorem 19.1. (Lov
asz and Schrijver [91]) For every prime p > 2, up to affine
transformation there is a unique set of p points in AG(2, p) determining p+3
2 directions.
q+3
Choosing d = q1
2 in the construction of Megyesi we get an example with N = 2 ,
so this should be the unique set in the theorem above. Note that the analogue over
R, i.e. the union of three halflines {(a, 0, 1), (1, a, 0), (0, 1, a) : a 0} PG(2, R)
is still a blocking set. It is also worth to write the (affine part of the) Megyesi
q+1
example as the graph of the function X 2 . The analogue of it over R is the
absolute value function |X| with determined directions D = {(m) : 1 m 1}.
Proof: W.l.o.g. we can assume that (0, 0), (1, 0) and (0, 1) are in our affine
pointset. There are several ways available to prove
Qp this theorem. One of them
is to write up the Redei polynomial R(X, Y ) = i=1 (X + ai Y bi ) for the points
{(ai , bi ) : i} = U of our pointset. Then substituting a determined direction y, one
can see that we are in the situation of Theorem 5.38. Considering the possible
multiplicities of roots (i.e. the distribution of the points of U on the affine lines
through (y) ), observing that each point of U is contained in a collinear triple of U
if p 7 (for p = 3, 5 everything is easy), and using some elementary geometrical
arguments we get the result.
Note that here Theorem 5.38 has done most of the job.
Exercise 19.2. Make the geometric argument precise in the proof above.
Also after p+3
2 , there is a big gap in the possible values of N , the number of
determined directions. The main result here is the following:
Theorem 19.3. G
acs [67] For every prime p, besides lines and the example characterized by Lov
asz and Schrijver, any set of p points in AG(2, p) determines at
least [2 p1
3 ] + 1 directions.
One may recall Megyesis construction 18.2, there N = q + 1 d where d|q 1 is
q+3
the size of a multiplicative subgroup of GF(q). So after q + 1 q1
2 = 2 one gets
q1
q1
q + 1 3 = 2 3 + 2(whenever 3|q 1), so for primes the bound in Theorem
19.3 is one less than a possible sharp result.
The examples of Megyesi are all contained in the union of two lines. This property
was characterized by Sz
onyi. For primes the result can be formulated as follows:
19. Over prime fields: the G

acs method
123
Theorem 19.4. (Sz

onyi [121]) Suppose U is a set of p points in AG(2, p), p prime,
which is contained in the union of two lines and denote by N the number of
determined directions. If 1 < N < p 1, then N = p + 1 d for a d|p 1, and
after linear transformation, U = {(x, 0) : x K} {(0, x) : x
/ K} , where K is
the union of some cosets of the multiplicative subgroup of GF(p), of order d.
For another description of pointsets covered by two lines see Theorem 9.32 of Biro
from [28] as well, which also shows the limits of the method we are using.
From now on our underlying field will be GF(p), where p is a prime. For any
polynomial f , N = Nf will denote the number of directions f determines. As
it was pointed out in the section on symmetric polynomials, if f and g are affine
transforms of each other (that is f (X) = ag(bX +c)+dX +e, where a, b, c, d, e F,
a, b 6= 0), or if f and g are bijective and f 1 = g, then Nf = Ng and Wf = Wg .
Our next duty is to recall the parameter Wf on polynomials over finite fields
(introduced in Section 9.4), which has a close relationship with N as Wf + Nf
p + 1 (see Lemma 9.31), and which will help us to take the advantages of the
algebraic terminology:
Wf = min{k + l :
xk f (x)l 6= 0}.
xGF(p)
Here k and l are non-negative integers; for k = 0 or l = 0, X 0 or f (X)0 is defined to

be the polynomial 1. Remember that Wf is the smallest k + l, for which X k f (X)l
has reduced degree p 1.
Throughout this section the result of Exercise 5.7 will be used without mentioning
it; i.e. for any subspace V of GF(q)[X], dim(V ) = |{deg(f ) : f V }|.
Theorem 19.3 will be proved, in a slightly more general form, as Theorem 19.6.
We will follow the arguments of Ball-Gacs-Sziklai [17] in the following sections.
There are some natural ways to generalize Theorem 19.3. One is to consider the
problem in AG(2, q) for any prime-power q. After the classification result in [43, 9]
(see Theorem 18.14) for the case N < q+3
2 , one should ask, whether Theorem 19.3
is true in general and is there a gap in the possible values of N after q+3
2 . A proof
is known for the case q = p2 only.
Proposition 19.5. (G
acs-Lov
asz-Szonyi) For any prime p, in AG(2, p2 ) the only
p2 +3
point set determining 2 directions is the one coming from Example 18.2. There
are no examples for
p2 +3
2
<N <
p2 +p
2
+ 1.
This result is sharp, there is aconstruction by Polverino, Szonyi and Weiner [102]
q+ q
for a pointset determining 2 + 1 directions, whenever q is a square.
124
19.1
Mf
Permutation polynomials
Let p be a prime. We say that f GF(p)[X] is a permutation polynomial if the

mapping x 7 f (x) is a permutation of GF(p), and by Mf we denoted the number
of elements a GF(p) for which f (X) + cX is a permutation polynomial. The
connection with the directions determined by f (X) is straightforward. Namely,
the set of parallel lines defined by the equation cX1 + X2 = a, where a runs
through the elements of GF(p), all contain exactly one point of the graph of f if
and only if f (X) + cX is a permutation polynomial. Thus the direction of these
lines is not a direction determined by f . Hence Mf + Nf = p.
Therefore the Redei-Megyesi result 18.1 says that if Mf (p 1)/2 then f (X) =
cX +d for some c, d GF(p); in other words, the graph of f , {(x, f (x)) : x GF(p)}
is a line.
Similarly Theorem 19.3 says that if Mf 2d p1
6 e + 1, then (f (X) (cX +
d))(f (X) (bX + e)) = 0 for some b, c, d, e GF(p); in other words, the graph of
f is contained in the union of two lines. The new method, invented by Gacs was
very fruitful, it became the key technique of other results as well, see below.
In the Megyesi examples 18.2, G is a multiplicative subgroup of GF(p) of size d
and f (X) = XG (X), where G (X) is the characteristic function of G. Here if
1 < d < q 1 then Mf = d 1.
In Theorem 19.4 Sz
onyi proved that if the graph of f is contained in the union of
two lines and Mf 2, then the graph of f is affinely equivalent to a generalised
example of Megyesi detailed above. In the generalised Megyesi example G can
be replaced by a union of cosets of a multiplicative subgroup of GF(p). In the
generalised example the value of Mf is again d 1 for some divisor d of p 1.
Thus, the above results imply that, either Mf 2d p1
6 e, f is affinely equivalent
p+1
to X 2 , or f is linear.
In Theorem 19.6 and on, we will follow [17] when proving a version of the planar
Gacs result. Then in Theorem 19.15 a 3-dimensional version will be stated. We
p1
2
shall prove that if there are more than (2d p1
6 e + 1)(p + 2d 6 e)/2 2p /9 pairs
2
(c, d) GF(p) with the property that x 7 f (x) + cg(x) + dx is a permutation of
GF(p) then there are elements a, b, e GF(p) such that f (x) + ag(x) + bx + e = 0,
for all x GF(p); in other words the graph of (f, g), {(x, f (x), g(x)) | x GF(p)},
is contained in a plane.
The situation is analogous to the planar case. The set of parallel planes defined by
the equation dX1 + X2 + cX3 = a, where a runs through the elements of GF(p), all
contain exactly one point of the graph of (f, g) if and only if f (X) + cg(X) + dX
is a permutation polynomial.

acs method
19.2
125
Directions in the plane
In [67] it was proved that if Mf > (p 1)/4 and Wf 2d p1

6 e + 2 then the graph
of f is contained in the union of two lines.
Let
k (Y ) =
(f (x) + xY )k .
xGF(p)
Its a simple matter to check (see Section 5.1), that if x 7 f (x) + ax is a permutation then k (a) = 0 for all 0 < k < p P
1. Since the polynomial k (Y ) has degree
at most k 1 (the coefficient of Y k is xGF(p) xk = 0) it is identically zero for
all 0 k 1 < Mf , unless Mf = p 1 in which case f is linear. Hence if f is not
linear then Wf 1 Mf .
The original proof of Theorem 19.3 [67] showed that if Mf 2d p1
6 e + 1, then the
graph of f is contained in the union of two lines.
In this section we shall prove the following, slightly stronger theorem.
Theorem 19.6. [17] If Mf (p 1)/6 and Wf 2d p1
6 e + 2 then the graph of f
is contained in the union of two lines.
The values Wf and Mf are invariant under affine transformations and inversion.
Replacing f by its inverse is the transformation which switches coordinates, in
other words if we switch coordinates then the graph of f , {(x, f (x)) | x GF(p)},
becomes the graph of f 1 . Let E(f ) denote the set of all polynomials that can be
obtained from f by applying affine transformations and inversions.
Let (f i ) be the degree of the polynomial f i modulo X p X. Unless stated
otherwise all equations are to be read modulo X p X.
P
Note that for any polynomial g of degree less than p the sum xGF(p) g(x) is
equal to the coefficient of X p1 of g.
Lemma 19.7. If 3 f (p 1)/2 then Wf (p + 1)/3.
Proof: Write p 1 = af + b with 0 b < f . The degree of f (X)a X b is p 1,
so we have Wf a + b.
If f = 3 then a + b (p 2)/3 + 1 = (p + 1)/3.
If (p + 1)/3 f (p 1)/2 then a + b = 2 + p 1 2f (p + 1)/3.
If (p + 1)/4 f (p 1)/3 then a + b = 3 + p 1 3f 3 + p 1 3(p + 1)/4 =
(p + 1)/4 + 1 (p + 1)/3 for p 11.
If 4 f (p + 1)/4 then a + b (p b 1)/f + b p/f + (bf b 1)/f
p/f + f 2. This is at most (p + 1)/3 if and only if the quadratic inequality
3(f )2 (p + 7)f + 3p 0 is satisfied. For p 20, the inequality is satisfied for
both f = 4 and f = (p + 1)/4, so it holds for all values between 4 and (p + 1)/4.
126
For p < 20 a case by case analysis suffices to show that a + b (p + 1)/3.

Note that for f = 2 we have Wf = (p 1)/2 and Mf = 0 and for f = 1 we have
Wf = p 1 and Mf = p 1.
Lemma 19.8. If f = (p+1)/2 then either Wf (p+5)/4 or f is affinely equivalent
p+1
to X 2 .
Proof: After applying a suitable affine transformation we can suppose that
p+1
f (X) = X 2 + g(X) where g (p 3)/2.
If g 1 then by applying another linear transformation we can subtract g from
p+1
f and hence f is affinely equivalent to X 2 .
Suppose g 2. Write (p 3)/2 = ag + b with 0 b < g and consider the
polynomial
a+1
X a + 1 p+1
a+1 b
f (X) X =
X i 2 +b g(X)a+1i .
i
i=0
We claim that the only term in the sum that has a term of degree X p1 (modulo
p+1
p+1
X p X) is g(X)a X 2 +b . Let r(X) = g(X)a+1i X i 2 +b (modulo X p X), a
typical term in the sum (note that all the binomial coefficients are non-zero). If
i is even then r(X) = g(X)a+1i X b+i , which has degree (a + 1 i)g + b + i =
p1
(p 3)/2 + g (g 1)i < p 1. If i 6= 1 is odd then r(X) = g(X)a+1i X 2 +i+b ,
which has degree (a + 1 i)g + (p 1)/2 + i + b = p 2 + g (g 1)i (p 3)/4 then a = 1 and b = (p 3)/2 g < (p 3)/4
and so a + b < (p + 1)/4.
Let s = d(p 1)/6e.
We will assume from now on that Wf 2s + 2. By the definition of Wf the sum
X
xk f (x)
xGF(p)
has no term of degree X p1 , for all k = 0, 1, . . . , 2s, and therefore the degree of f
is at most p 2s 2. By Lemma 19.7 and Lemma 19.8 the degree of f is at least
(p + 3)/2.
Lemma 19.9. There is polynomial in h E(f ) with one of the following properties.
Either
(i) for all i such that 1 i 2s, (hi ) h + i 1 and (h2 ) = h + 1, or

acs method
127
(ii) for all i such that 1 i 2s, (hi ) (h2 ) + i 2 and (h3 ) = (h2 ) + 1,
and h has no root in GF(p).
Proof:
Let
d(f ) = max{(f i ) i | 1 i 2s}
and let d = d(f1 ) be maximal over all polynomials in E(f ). The fact that f
(p + 3)/2 implies that d (p + 1)/2.
The coefficient of Y p1dj in (Y ) is
f which, by the definition of d, is non-zero for at least
xGF(p) x
one j where 1 j 2s. Hence (Y ) 6 0.
Let
(Y )
P
p1d
j
p1d (Y ).
p1dj j
If for all a such that f (X) + aX is a permutation polynomial we have (a) =

0 (a) = 00 (a) = 0 then (Y a)3 divides (Y ) and since Mf (p 1)/6 the
degree of , = p 1 d 3Mf (p 1)/2 which isnt the case.
Since 0 < p 1 d < p 1 we have already seen that (a) = 0, so either 0 (a) 6= 0
or 00 (a) 6= 0 for some a.
Let f2 be the inverse of the function f (X) + aX.
If
0 6= 0 (a) = (d + 1)
x(f + ax)p2d
xGF(p)
p2d
then
6= 0 and so f2 d + 1. By the maximality of d,
zGF(p) f2 (z)z
i
f2 = d + 1 and so (f2 ) i f2 1. If (f22 ) f2 then let f3 = f2 + cX where
c is chosen so that (f32 ) f3 + 1 and f3 is not a permutation polynomial. Note
that f32 = f22 + 2cXf2 + c2 X 2 .
If
0 6= 00 (a) = (d + 1)(d + 2)
x2 (f + ax)p3d
xGF(p)
then zGF(p) (f2 (z))2 z p3d 6= 0 and so (f22 ) d + 2. By the maximality of d,

(f22 ) = d + 2 and so (f2i ) i (f22 ) 2. If (f23 ) (f22 ) then let f3 = f2 + cX
where c is chosen so that (f33 ) (f32 ) +1 and f3 is not a permutation polynomial.
P
Finally, let e be an element not in the image of f3 and let f4 = f3 e. Then f4

has no root in GF(p).
Lemma 19.10. There is a polynomial in h E(f ) for which there exist polynomials
F , G and H, where H 2 = F 1 = G = r s 2, (F, G) = 1 and
F h + Gh2 = H.
128
Note that this implies that h satisfies the conditions of Lemma 19.9 (i).
Proof: Let h be a polynomial satisfying the conditions of Lemma 19.9. Since
Wh 2s + 2 we have (hi ) p 2s 3 + i.
Define subspaces of the vector space of polynomials of maximum degree p 1
j = {F h + Gh2 | F j, G j 1},
where j s 1. If there are polynomials F and G such that F h + Gh2 = 0 then
since h has no root F + Gh = 0 which is impossible since (hG) is at least 3s and
at most 5s 3 < p 1. Thus the dimension of j is 2j + 1.
Since Wh 2s + 1 and 2(j + 1) 2s, the sum over GF(p) of the evaluation of the
product of any two elements of j is zero, hence the sum of the degrees of any two
elements of s1 is not equal to p1. The maximum degree of any element of s1
is p s 3 and so only half of the degrees in the interval [s + 2, . . . , p 1 (s + 2)]
can occur. But dims1 = 2s 1 > (p 1 (s + 2) (s + 1))/2 and so there is
an element H of degree at most s + 1 in s1 .
Let H be of minimal degree, so (F, G) = 1.
If h satisfies case (i) of Lemma 19.9 then (h2 ) = h + 1 and r = G = F 1.
Moreover F h2 + Gh3 = Hh and (h3 ) h + 2 implies H r + 2.
If h satisfies case (ii) of Lemma 19.9 then (h3 ) = (h2 ) + 1 h + 2 and (h4 )
(h2 ) + 2. Let F = r + 1 and so G r. The equation F h3 + Gh4 = Hh2 implies
H r + 2. If G r 1 then F h2 + Gh3 = Hh implies r + 2 + h H + h =
r + 1 + (h2 ) and so (h2 ) = h + 1. But then F h + Gh2 = H implies G = r.
Either way we have r = G = F 1 H 2.
Let r1 = h+aX and F1 = F 2aXG, G1 = G and H1 = H a2 X 2 G+aXF . Then
F1 r1 + G1 r12 = H1 and we can choose a so that H1 has degree r + 2. Now when
we look at r+1 for r1 we find F1 , G1 and H1 as required. Note that (F, G) = 1
implies (F1 , G1 ) = 1.
We wish to prove r = 0. So let us assume r 1 and define i to be such that
(i 2)r + 1 s < (i 1)r + 1 for r 2 and i = s for r = 1. Note that r s 2
implies i 3 and that s + r 1 2s i if i = 3 or i = s and also if both i 4
and r 2, since r (s 1)/2 and i (s 1)/2.
Lemma 19.11. There is a polynomial h E(f ) and a polynomial G, where G =
r s 2, such that for all j = 2, . . . , i, there is an Fj and an Hj with the property
that (Fj , G) = 1,
Fj h + Gj1 hj = Hj ,
Fj (j 1)(r + 1), Hj (j 1)r + j and Hi = (i 1)r + i.
Proof: Let r1 satisfy the conditions of Lemma 19.10. We start by proving that
there is an h E(f ) for which (hi1 ) h + i 2.

acs method
129
If (r1i1 ) r1 + i 3 then let h = r1 + aX. Choose a so that hi1 =

Pi1 i1
ij1 j
r1 has degree at least h + i 2 while at the same time
j=0
j (aX)
the degree of F 2aXG is r + 1 and the degree of H a2 X 2 G + aXF is r + 2.
We will prove the lemma by induction. Lemma 19.10 implies that for j = 2 we
can take F2 = F and H2 = H.
Define Fj = (Fj1 F + Hj1 G) and Hj = HFj1 . It can be checked by induction, multiplying by Gh and using Gh2 = H F h, that
Fj h + Gj1 hj = Hj .
The degrees satisfy Fj (j 1)(r + 1) and Hj (j 1)r + j and (Fj , G) = 1,
since (F, G) = 1 by Lemma 19.10 and (Fj1 , G) = 1 by induction.
Now (hi1 ) h + i 2 and the equation Fi1 h + Gi2 hi1 = Hi1 implies that
Fi1
(i 2)(r + 1) and so Fi1
= (i 2)(r + 1). Finally Hi = HFi1 implies
Hi = (i 1)r + i.
Let h satisfy the conditions of Lemma 19.11. Note that this implies that h satisfies
the conditions of Lemma 19.10 and Lemma 19.9 (i). Define
j = {Ah + Bhi | A j, B j + 1 i}.
Note that Hi (i1)r+i1 and that (i 1)r + i 1 s + r + i 2 2s 1.
Lemma 19.12. For j 2s 1 all polynomials of j have degree at least Hi and
those of degree at most p 2 h are multiples of Hi .
Proof: If Ah + Bhi = 0 then, since h has no root in GF(p), A + Bhi1 = 0. The
degree of Bhi1 is at most p 4 and at least (p + 3)/2 and so A = B = 0. Thus
the dimension of j is 2j + 3 i.
Suppose that j contains a polynomial C of degree n but no polynomial of degree
n + 1. Then j+1 contains a polynomial of degree n + 1, X C for example, and
a polynomial of degree one more than the maximum degree of an element of j .
However dimj+1 = dimj +2, so n is unique. Moreover, the polynomials of degree
n + 1 in j+1 are multiples of a polynomial of degree n in j .
Since j 2s1, j contains no element of degree p1h . Now Hi (i1)r+i1
and is a polynomial of degree less than p 1 h . It is not a multiple of any
polynomial in j for j < (i1)r+i1, since if it were there would be a non-constant
polynomial K and polynomials A and B with the property that (KA)h+(KB)hi
(i1)r+i1 , with (KA) (i 1)r + i 1 and (KB) (i 1)r, which would be a
constant multiple of Hi . This is not possible since (Fi , G) = 1. Thus all polynomials
in j of degree at most p 2 h are multiples of Hi and in particular have degree
at least Hi .
130
The following lemma contradicts the previous one which implies that our assumption that r 1 was incorrect.
Lemma 19.13. There is a non-zero polynomial of degree less than Hi in j for
some j 2s 2.
Proof:
Suppose r 2 and so i s. Let
= {Ah + B2 h2 + . . . + Bi1 hi1 + Chi | A s 1, Bj r 1, C s i}.

Since Wh 2s + 1 the sum of the degrees of any two elements of is not equal
to p 1. The maximum degree of any element of is p s 3 and so only
half of the degrees in the interval [s + 2, . . . , p 1 (s + 2)] can occur, in other
words at most b(p 4 2s)/2c 2s 2 of the degrees in this interval occur. If
dim = (i 2)r + 2s i + 1 then there is a polynomial
E = Ah + B2 h2 + . . . + Bi1 hi1 + Chi
in of degree at most s + 2 ((i 2)r + 2s i + 1 (2s 2)) = s (i 2)r + i 1.
If dim < (i 2)r + 2s i + 1 then E = 0 non-trivially. Either way there is
a polynomial E with not all A, Bj , C zero where E s (i 2)r + i 1.
Substituting Gj1 hj = Hj hFj we have
Gi2 E = Gi2 Ah + CGi2 hi +
i1
X
Bj Gi1j (Hj hFj )
j=2
and rearranging
G
i2
i1
X
Bj G
i1j
Hj = (G
j=2
i2
i1
X
Bj Fj Gi1j )h + CGi2 hi .
j=2
Checking the degrees on the right-hand side we see that the left-hand side is a
polynomial in j for some j 2s 2.
The degree of the left-hand side is at most max{s + i 1, ir r + i 2} which is
less than Hi = (i 1)r + i.
If r = 1 then take i = s and define as above. There is a polynomial E in of
degree at most s + 1 and the degree of Gi2 E is at most 2s 1 which is the degree
of Hs . If we have equality then by Lemma 19.12 the polynomial
(Gs2 A
s1
X
Bj Fj Gs1j )h + CGs2 hs
j=2
is a constant multiple of Fs h + Gs1 hs which implies CGs2 is a constant multiple

of Gs1 which it is not since one has degree s 2 and the other s 1.

acs method
131
We can now prove Theorem 19.6.

Proof: By the previous lemmas there exist polynomials h E(f ) and F of degree
1 and H of degree 2 such that h2 + F h = H. Thus (h + F/2)2 = H + F 2 /4. All
values of H + F 2 /4 are squares and so H + F 2 /4 = (aX + b)2 . Hence (h + F/2
aX b)(h + F/2 + aX + b) = 0 and the graph of h (and so the graph of f too) is
contained in the union of two lines.
19.3
Linear combinations of three permutation polynomials
Let M (f, g) be the number of pairs (a, b) GF(p)2 for which f (X) + ag(X) + bX
is a permutation polynomial. Let
X
W (f, g) = min{k + l + m |
xk f (x)l g(x)m 6= 0}.
xGF(p)
Let again s = d p1
6 e. Before we prove the main result of this section we need the
following lemma
Lemma 19.14. [17] If M (f, g) > (2s + 1)(p + 2s)/2 then W (f, g) 2s + 2 or there
are elements c, d, e GF(p) such that f (x) + cg(x) + dx + e = 0 for all x GF(p).
Proof:
Let k (Y, Z) =
xGF(p) (f (x)
+ g(x)Y + xZ)k .
By Section 5.1, if f (X)+ag(X)+bX is a permutation polynomial then k (a, b) = 0

for all 0 < k < p 1. Write
Y
k =
j (Y, Z),
where each j is absolutely irreducible. Then
j = k k.
Let Nj be the number of solutions of j (a, b) = 0 in GF(p) for which f (X) +

ag(X) + bX is a permutation polynomial.
If j GF(p)[Y, Z], for some in an extension of GF(p), and j 2 then by
Theorem 10.9(i) Nj j (p + j 1)/2.
Suppose j = 1 and there are at least (p + 1)/2 pairs (a, b) for which j (a, b) = 0
and f (X) + ag(X) + bX is a permutation polynomial. Let j = Y + Z + .
If 6= 0 then there are (p + 1)/2 elements b GF(p) with the property that
f (X) (b + )g(X) + bX = f (X) g(X) + b(X ) is a permutation
polynomial. By Redei and Megyesis theorem mentioned in the introduction, this
implies that f (X) g(X) is linear and hence there are elements c, d, e GF(p)
such that f (x) + cg(x) + dx + e = 0 for all x GF(p). If = 0 then there are
(p + 1)/2 elements a GF(p) with the property that f (X) X + ag(X) is a
permutation polynomial. The set of p points {(f (x) x, g(x)) | x GF(p)}
132
may not be the graph of a function but it is a set of p points that does not
determine at least (p + 1)/2 directions. Thus it is affinely equivalent to a graph of
a function that does not determine at least (p 1)/2 directions and so by Redei
and Megyesis theorem, it is a line. Hence, there are elements c, d and e with the
property that c(f (x) x) + dg(x) + e = 0 for all x GF(p). Thus, either there
are elements c, d, e GF(p) such that f (x) + cg(x) + dx + e = 0 for all x GF(p)
or Nj (p 1)/2.
Suppose
Z] forP
any in any extension of GF(p). The polynomials
P j 6nGF(p)[Y,
p
j =
nm Y Z m and
j =
nm
Y n Z m have at most (j )2 zeros in common
by Bezouts theorem. However if (y, z) GF(p)2 and j (y, z) = 0 then
j (y, z) = 0.
Hence
Nj (j )2 j (p + j 1)/2,
whenever j (p 1)/2.
Thus if k 6 0 and k (p1)/2 then N (k ), the number of solutions of k (y, z) =
0 in GF(p) for which f (X) + ag(X) + bX is a permutation polynomial, satisfies
N (k )
Nj
j (p + j 1)/2 k(p 1)/2 +
1X 2
(j )
2
1 X 2
k(p 1)/2 + (
j ) = (k(p 1) + k 2 )/2.
2
By hypothesis k 0 or
(2s + 1)(p + 2s)/2 < Nk (k(p 1) + k 2 )/2,
which gives k 2s + 2. Now

kl
k X
X
k
k l X klm
x
f (x)l g(x)m Y m Z klm ,
k (Y, Z) =
l
m
m=0
l=0
xGF(p)
and so W (f, g) 2s + 2.
Theorem 19.15. [17] If M (f, g) > (2s + 1)(p + 2s)/2 then there are elements
c, d, e GF(p) such that f (X) + cg(X) + dX + e = 0.
Proof: If p = 3 and M (f, g) > (2s + 1)(p + 2s)/2 = 15/2 then there is a c such
that f (X) + cg(X) + bX is a permutation polynomial for all b GF(p), which can
only occur if there is a constant e such that f (X) + cg(X) + e = 0.
So suppose p 5 and that there are no elements c, d, e GF(p) with the property
that f (X) + cg(X) + dX + e = 0.
Clearly Wf +ag W (f, g) for all a GF(p) and W (f, g) 2s + 2 by Lemma 19.14.

acs method
133
There is an a1 GF(p) with the property that

Mf +a1 g M (f, g)/p (p 1)/6.
By Theorem 19.6 there are constants c, d, c0 , d0 GF(p) with the property that
(f + a1 g + cX + d)(f + a1 g + c0 X + d0 ) = 0
so the graph of (f, g), the set of points {(x, f (x), g(x)) | x GF(p)}, is contained
in the union of two planes.
By Redei and Megyesis theorem, since we have assumed that the graph of f + a1 g
is not a line, Mf +a1 g (p 1)/2 and so there is an a2 6= a1 with the property
that
Mf +a2 g (M (f, g) (p 1)/2)/(p 1) (p 1)/6.
Thus the graph of (f, g) is contained in the union of two other planes, different
from the ones before. The intersection of the two planes with the two planes is
four lines and so the graph of (f, g) is contained in the union of four lines.
Similarly, since (M (f, g) (p 1))/(p 2) (p 1)/6 and (M (f, g) 3(p
1)/2)/(p 3) (p 1)/6, there is an a3 and an a4 with that property that
Mf +a3 g (p1)/6 and Mf +a4 g (p1)/6 and so the graph of (f, g) is contained
in two other distinct pairs of planes. The four lines span three different pairs of
planes and so the graph of (f, g) is contained in the union of two lines and hence
a plane, which is a contradiction.
There is an example when q is an odd prime (power) congruent to 1 modulo 3
with M (f, g) = 2(q 1)2 /9 1 where the graph of (f, g) is not contained in a
plane, which shows that the bound is the right order of magnitude.
Let E = {e GF(q) | e(q1)/3 = 1} {0}. Then the set S =
{(e, 0, 0), (0, e, 0), (0, 0, e) | e E} is a set of q points. If , the plane defined by
X1 + aX2 + bX3 = c,
is incident with (e, 0, 0) for some e E then c E. Likewise if it is incident with
(0, e, 0) for some e E then a/c E and if it is incident with (0, 0, e) for some
e E then b/c E.
If is incident with two points of S then either a E, b E or a/b E. Thus
if a, b and a/b are not elements of E then and all the planes parallel to are
incident with exactly one point of S. There are 2(q 1)2 /9 such sets of parallel
lines.
If we make a change of coordinates so that {x1 = x | x GF(q)} is one such set of
parallel planes then there are functions f and g for which S = {(x, f (x), g(x)) | x
GF(q)}. Each other set of parallel lines with the above property corresponds to
a pair (a, b) such that f (X) + ag(X) + bX is a permutation polynomial. Thus
134
M (f, g) = 2(q 1)2 /9 1. Explicitly the functions f and g can be defined by

f (x) = R (x)x and g(x) = H (x)x, where H is the characteristic function of
H = {t3 | t GF(p)} and is a primitive third root of unity.
20
Projective linear pointsets
In Section 18.3 we defined affine linear pointsets. The general geometric definition
in the projective case is the following:
Definition 20.1. A pointset B PG(n, q) is called GF(pe )-linear if
(i) GF(pe ) is a subfield of GF(q), and
(ii) there is a projective space PG(n0 , q) containing PG(n, q) such that B is a
projection of a subgeometry PG(t, pe ) PG(n0 , q) from a suitable subspace
(vertex or center) V to PG(n, q).
Note that here dim V = n0 n 1 and the projection is not necessarily one-to-one.
Projective linear pointsets are more complicated then affine ones. One way of the
algebraic description is that B is GF(pe )-linear iff one can choose t + 2 points
(vectors) v 0 , v 1 , ..., v t , v t+1 of PG(n, q) such that B is the span of them over
GF(pe ), i.e.
(1) their homogeneous coordinates are chosen in such a way that v 0 = v 1 + v 2 +
... + v t+1 ;
(2) v 1 , ..., v t+1 are independent over GF(pe );
t+1
X
(3) B = hv 0 , v 1 , ..., v t+1 iGF(pe ) = {
i v i : 1 , ..., t+1 GF(pe ), (1 , ..., t+1 ) 6= (0, 0, ..., 0)}.
i=1
(As usual, for any GF(q) and any vector u, the point u represents the same
e t+1
point). So we get (p p)e 11 points, possibly counted with multiplicities.
In this case some points may well coincide. Lets examine the structure of multiple points! For any point u B consider the (homogeneous) t + 1-tuples
Pt+1
Lu = {(1 , ..., t+1 )} PG(t, pe ) defining it with u = i=1 i v i . Obviously any
Lu is a projective subspace of PG(t, pe ) (so all the multiplicities are of the form
(pe )i+1 1
pe 1
for i {0, 1, ..., t}), and {Lu : u B} is a partition of PG(t, pe ).
Exercise 20.2. Prove that, because of (2) above, most of the subspaces Lu are in
fact points of PG(t, pe ).
Define the matrix
U = (v T1 , v T2 , ..., v Tt+1 ),
20. Projective linear pointsets
135
then B = hv 0 , v 1 , ..., v t+1 iGF(pe ) is the image of PG(t, pe ) = {(1 , ..., t+1 ) 6=
(0, 0, ..., 0) : i GF(pe )} under the map 7 U T .
Pt+1
Pt+1
Two linear combinations P =
i v i define coinciding
i=1 i v i and Q =
i=1P
t+1
points if there exists an GF(q) such that P = Q, so i=1 (i i )v i = 0.
It can happen if the v i -s are dependent over GF(q) (as usually they are). Consider
the ((n + 1) (t + 1)) matrix U defined above. Let W be the projective subspace
of PG(t, q) consisting of nonzero vectors w for which U wT = 0. (It may be the
empty set if n t.) If W = then all the points of B are distinct.
Note that, counting without multiplicities, the number of points of B satisfies
|B|
(pe )t+1 1
;
pe 1
and the author conjectures that if GF(pe ) is the maximum subfield of linearity
then
(pe )t+1 1
(pe )t + (pe )t1 + 1 |B|
pe 1
holds as well. Intuitively it would mean that we cannot lose more than a (t 2)dimensional subspace, collapsing into one point.
Theorem 20.3. The Redei polynomial of B is
t+1
X
(
i v i )V =
R(X, Y, ..., T ) =
(1 ,...,t+1 )PG(t,pe ) i=1
= det
te
te
(v 2 V)p
sgn()
Sym({1,2,...,t+1})
v1 V
v2 V
...
v t+1 V
e
e
e
(v 1 V)p (v 2 V)p ... (v t+1 V)p
2e
2e
2e
(v 1 V)p (v 2 V)p ... (v t+1 V)p
..
.
(v 1 V)p
te
t+1
Y
e (i)
(v i V)(p
i=1
= M RDpe (v 1 V, v 2 V, ..., v t+1 V).
... (v t+1 V)p
Proof: To prove this first observe that both R and the determinantPis of degree
t+1
1 + pe + p2e + ... + pte . Hence it is enough to prove that each factor ( i=1 i v i )V
of R appears in the determinant as well.
W.l.o.g. suppose that 1 6= 0. Multiply the first column of the determinant by the constant 1 , then successively add 2 (the second column), ..., t+1 (the (t + 1)-th column) to the first column, this processPdoes not change
the determinant essentially.
first column is
Pt+1
Pt+1 Now the
e
te
t+1
( ( i=1 i v i )V, (( i=1 i v i )V)p , ..., (( i=1 i v i )V)p
)> , so each enPt+1
try in it is divisible by the factor ( i=1 i v i )V, hence the same holds for the
136
determinant as well.
Suppose that a a hyperplane x = [x, y, ..., t] contains precisely one point P =
Pt+1
e
i=1 i v i point of B (i GF(p )). It means that
0 = Px =
t+1
X
i v i x.
i=1
Pt+1
je
Take the (pe )j -th power of this equation, it is i=1 i (v i x)p = 0, meaning that
linear combination of the columns of the determinant above (after substituting
V = x), with the same i -s, result in the zero vector, hence the value of the
determinant is zero.
Exercise 20.4. Prove the other direction, i.e. if the determinant
Pt+1 is zero for some
substitution V = x = (x, y, ..., t) then there is a point P = i=1 i v i B on the
hyperplane [x, y, ..., t].
e k+1
Intuitively, if a hyperplane [x, y, ..., t] contains (p p)e 11 points of B, as the intersection is a linear set generated by some u0 , u1 , ..., uk+1 B, then it means that
there are k + 1 independent equations for the columns of the determinant above
(i.e. k + 1 independent vectors = (1 , ..., t+1 ), each coming from a uj , expressed
from the v i -s), hence the rank of the matrix is (t + 1) (k + 1).
Exercise 20.5. Prove that it implies that (x, y, ..., t) is a point of R(X, Y, ..., T ) with
e k+1
multiplicity (p p)e 11 precisely (as it has to be).
If for example pte = q then B is a blocking set with respect to hyperplanes, since
for any hyperplane [x, y, ..., t] the Redei polynomial R(x, y, ..., t) vanishes: the first
and the last rows of the determinant above are identical after the substitution.
Note that now R(X, Y, ..., T ) may contain multiple factors. (Removing all but one
copies of a multiple point the blocking property remains intact.)
21
Blocking sets
A blocking set with respect to k-dimensional subspaces is a pointset meeting every

k-subspace. As a blocking set plus a point is still a blocking set, we are interested
in minimal ones (with respect to set-theoretical inclusion) only. Note that in a
(projective) plane the only interesting case is k = 1.
In any projective plane of order q the smallest blocking set is a line (of size q + 1).
In PG(2, q) there exist minimal blocking sets of size 23 q; the projective triangle
of size 3(q + 1)/2 if q is odd and the projective triad (which is a linear pointset
in fact) of size 3q/2 + 1 if q is even. In general, in PG(n, q) it is easy to construct
21. Blocking sets
137
a blocking set with respect to k-dimensional subspaces; it is straightforward to

prove that the smallest example is a subspace of dimension n k (so consisting
nk+1
of q q1 1 q nk points), this example is called trivial. Another easy one is a
cone, with a planar blocking set as a base and an (n k 2)-dimensional subspace
as vertex; if the base was of size 23 q then the blocking set will be of size 32 q nk
roughly. A blocking set with respect to k-dimensional subspaces of PG(n, q) is said
to be small if it is smaller than 23 (q nk + 1), in particular in the plane it means
that |B| < 3(q + 1)/2.
A most interesting question of the theory of blocking sets is to classify the small
ones. A natural construction (blocking the k-subspaces of PG(n, q)) is a subgeometry PG(h(n k)/e, pe ), if it exists (recall q = ph , so 1 e h and e|h).
It is one of the earliest results concerning blocking sets, due to Bruen [51], that a
nontrivial blocking set of a projective plane of order q is of size q + q + 1, and
equality holds if and only if it is a Baer subplane (i.e. a subgeometry of order q).
It is easy to see that the projection of a blocking set, w.r.t. k-subspaces, from a
vertex V onto an r-dimensional subspace of PG(n, q), is again a blocking set, w.r.t.
the (k + r n)-dimensional subspaces of PG(r, q) (where dim(V ) = n r 1 and
V is disjoint from the blocking set).
A blocking set of PG(r, q), which is a projection of a subgeometry of PG(n, q), is
called linear. (Note that the trivial blocking sets are linear as well.) Linear blocking
sets were defined by Lunardon, and they were first studied by Lunardon, Polito
and Polverino [93], [98].
Conjecture 21.1. The Linearity Conjecture. In PG(n, q) every small blocking set,
with respect to k-dimensional subspaces, is linear.
There are some cases of the Conjecture that are proved already.
Theorem 21.2. For q = ph , every small minimal non-trivial blocking set w.r.t.
k-dimensional subspaces is linear, if
(a) n = 2, k = 1 (so we are in the plane) and
(i) (Blokhuis [33]) h = 1 (i.e. there is no small non-trivial blocking set at
all);
(ii) (Sz
onyi [123]) h = 2 (the only non-trivial example is a Baer subplane
with p2 + p + 1 points);
(iii) (Polverino [99]) h = 3 (there are two examples, one with p3 + p2 + 1
and another with p3 + p2 + p + 1 points);
(iv) (Blokhuis, Ball, Brouwer, Storme, Szonyi [43], Ball [9]) if p > 2 and
there exists a line ` intersecting B in |B `| = |B| q points (so a
blocking set of Redei type);
138
(b) for general k:

(i) (Sz
onyi and Weiner) [129] if h(nk) n, p > 2 and B is not contained
in an (h(n k) 1)-dimensional subspace;
(ii) (Storme-Weiner [110] (for k = n 1), Bokler and Weiner [138]) h = 2,
q 16;
(iii) (Storme-Sziklai [108]) if p > 2 and there exists a hyperplane H intersecting B in |B H| = |B| q nk points (so a blocking set of Redei
type).
There is an even more general version of the Conjecture. A t-fold blocking set w.r.t.
k-subspaces is a pointset which intersects each k-subspace in at least t points.
Multiple points may be allowed as well.
Conjecture 21.3. The Linearity Conjecture for multiple blocking sets: In PG(n, q)
any t-fold blocking set B, with respect to k-dimensional subspaces, is the union
of some (not necessarily disjoint) linear pointsets B1 , ..., Bs , where Bi is a ti -fold
blocking set w.r.t. k-dimensional subspaces and t1 + ... + ts = t; provided that t and
|B| are small enough (t T (n, q, k) and |B| S(n, q, k) for two suitable functions
T and S).
Note that there exists a ( 4 q +1)-fold blocking set in PG(2, q), constructed by Ball,
Blokhuis and Lavrauw [19], which is not the union of smaller blocking sets. (This
multiple blocking set is a linear pointset.)
First we study 1-fold (planar) blocking sets.
As an appetizer, we present here Blokhuis theorem, which was a real breakthrough
at 1994. It was conjectured by Jane di Paola in the late 1960s.
Theorem 21.4. (Blokhuis [33]) In PG(2, p), p prime, the size of a non-trivial blocking set is at least 3(p + 1)/2.
Proof 1: Let B = U D, U = B AG(2, p) = {(ai , bi ) : i = 1, ..., p + s} with s 1
and D = B \ AG(2, p) = {(), (y1 ), ..., (yks )}, so |B| = p + 1 + k, and suppose
that B is a minimal non-trivial blocking set. Consider the affine Redei polynomial
R(X, Y ) =
p+s
Y
i=1
(y) 6 D
(X + ai Y + bi ) =
p+s
X
rj (Y )X p+sj
j=0
of U , where degY rj (Y ) j and r0 (Y ) = 1. If (y) 6 B then the affine lines of slope

y should be blocked by U . It means that R(X, y) vanishes for all X = x GF(p)
hence R(X, y) = (X p X)t(X) for some t(X) of degree s. Consequently, R(X, y)
does not contain terms of exponent s + 2, s + 3, ..., p 1, so the polynomials
rs+1 (Y ), ..., rp2 (Y ) vanish for all y 6 D, so for p k + s values. It means that
21. Blocking sets
(yr ) D
139
rs+1 (Y ) = ... = rpk+s1 (Y ) = 0 identically, so the terms X k+1 , ..., X p1 are all
missing even from R(X, Y ).
Now consider a point at infinity from D, i.e. some yr . As R(X, yr ) still do not
contain the terms X k+1 , ..., X p1 , we have
R(X, yr ) = X p g(X) + h(X), deg(g) = s, deg(h) k.
It is almost the situation of Exercise 5.39, excepting that g(X) and h(X) may
have a common factor. Dividing out the common factors we get X p g1 (X)+h1 (X),
which is still totally reducible, deg(g1 ) s, deg(h1 ) k. As s k, Exercise 5.39
3
gives k p+1
2 , so |B| = p + k + 1 2 (p + 1). The only cases we have to exclude
p
p
are (i) X g1 (X) + h1 (X) = (aX + b) and (ii) X p X | X p g1 (X) + h1 (X). Both
are impossible as (i) would imply that B contains a whole line, while (ii) would
mean that X p X | R(X, yr ) so the point (yr ) could be deleted without loss of
the blocking property, contradicting the minimality of B.
lacunary,
isnt it?
Proof 2: Let B = {(ai , bi ) : i = 1, ..., p + k} {()}. The affine Redei polynomial

R(X, Y ) =
p+k
Y
(X + ai Y + bi )
i=1
vanishes for all X = x, Y = y, x, y GF(p). Hence R(X, Y ) is in the ideal

h(X p X), (Y p Y )i and we can write R(X, Y ) = (X p X)F (X, Y ) + (Y p
Y )G(X, Y ), where f and g are of total degree k. Let R0 denote the part of R that
is homogeneous of total degree p + k, and let F0 and G0 denote the parts of F and
G resp. that are homogeneous of total degree k. Now R0 = X p F0 + Y p G0 . As the
equation is homogeneous lets put Y = 1, and define h(X) = R0 (X, 1), f (X) =
F0 (X, 1) and g(X) = G0 (X, 1).
Q
So f (X) = i (X + ai ) = X p f (X) + g(X) is a fully reducible polynomial with
f , g k. Using Exercise 5.39 after dividing out the common factors of f and g
we can finish the argument as in Proof 1.
These proofs seem to be very similar, the first treats the case s 1, the second the
case s = 0. The little philosophical difference led to different trains of thought,
as we will see later. This theorem somehow became a probe of new ideas examining
blocking sets, at least two more proofs will be presented, see Theorem 21.4 and
Corollary 21.22.
One can formulate the chase for minimal (nontrivial) blocking sets (in PG(2, q))
in an algebraic way as follows. Consider the polynomial ring GF(q)[X, Y, Z] and
its subset GF(q)[X, Y, Z]hom , the homogeneous polynomials of any degree. The
fully reducible polynomials form the multiplicative sub-semigroups R and Rhom
so
Q R0 (X, Y ) =
(X + ai Y )
140
in them. Let GF(q)[X, Y, Z]0 and GF(q)[X, Y, Z]hom,0 denote the sets (ideals) of
polynomials vanishing everywhere in GF(q) GF(q) GF(q).
Both GF(q)[X, Y, Z]0 and GF(q)[X, Y, Z]hom,0 , as ideals, can be generated by
three polynomials from R (and Rhom , resp.), for example GF(q)[X, Y, Z]0 =
h(X q X); (Y q Y ); (Z q Z)i and GF(q)[X, Y, Z]hom,0 = h(Y q Z Y Z q ); (Z q X
ZX q ); (X q Y XY q )i. Note also that for any a, b, c GF(q) the polynomial
a(Y q Z Y Z q ) + b(Z q X ZX q ) + c(X q Y XY q ) is still totally reducible. (It
is the Redei polynomial of the pointset consisting of the points of the line [a, b, c].)
The blocking set problem is now equivalent to finding minimal polynomials, w.r.t.
divisibility, as partial order, in
(Rhom GF(q)[X, Y, Z]hom,0 ).
The trivial blocking sets, as we have seen, correspond to the minimal polynomials
a(Y q Z Y Z q ) + b(Z q X ZX q ) + c(X q Y XY q ).
21.1
One curve
So let |B| = q + k be our blocking set. We often suppose that |B| < 2q. Recall the
Redei polynomial of B:
R(X, Y, Z) =
(ai X + bi Y + ci Z) =
q+k
X
rj (Y, Z)X q+kj .
j=0
(ai ,bi ,ci )B
Definition 21.5. ([47], [123]) Let C be the curve of degree k defined by

f (X, Y, Z) = r0 (Y, Z)X k + r1 (Y, Z)X k1 + . . . + rk (Y, Z).
Note that as deg(rj ) = j (or rj = 0), the polynomial f (X, Y, Z) is homogeneous
of degree k indeed.
Lemma 21.6. If the line LX [1, 0, 0] contains the points {(0, bij , cij ) : j = 1, ..., NX }
then
N
X
Y
Y
rNX (Y, Z) = (
as )
(bij Y + cij Z) | R(X, Y, Z);
as 6=0
j=1
rNX (Y, Z) | f (X, Y, Z);

so f can be written in the form f = rNX f, where f(X, Y, Z) is a homogeneous
polynomial of total degree = X-degree = k NX . In particular, if LX is a Redei
line then f = rNX . One can write R(X, Y, Z) = rNX (Y, Z)R(X,
Y, Z) as well.
Proof: obvious from the definitions: rNX |ri i. Indeed, rNX contains the X-free
factors of R; NX is the smallest index j for which rj is not identically zero. As, by
21. Blocking sets
141
Q
definition, rj is gained from R = (ai X + bi Y + ci Z) by adding up all the partial
products consisting of all but j (bi Y + ci Z) factors and j non-zero ai factors,
each of these products will contain all the factors of rNX , so rNX |rj j.
Note that the curve rNX consists of NX lines on the dual plane, all passing through
[1, 0, 0].
On the other hand if k < q then
X
Y
q
f = HX
R=
as1 as2 ...asq
(aj X + bj Y + cj Z).
{s1 ,s2 ,...,sq }
j6{s1 ,s2 ,...,sq }
Obviously it is enough to sum for subsets {Ps1 , Ps2 , ..., Psq } B \ LX .

If one coordinatizes B such that each ai is either 0 or 1, then
Y
X
f = rNX f = rNX
(X + bj Y + cj Z).
J{1,2,...,q+k}
|J|=kNX
ai 6=0 iJ
jJ
Also
q+kNX
kNX
rNX = HX
R = HX
f.
The next proposition summarizes some important properties of the Redei polynomial and of this curve.
Theorem 21.7. ([123])
(1.1) For a fixed (y, z) (where (0, z, y) 6 B), the element x is an r-fold root of
Ry,z (X) = R(X, y, z) if and only if the line with equation xX + yY + zZ = 0
intersects B in exactly r points.
(1.2) Suppose Ry,z (X) = 0, i.e. (0, z, y) B. Then the element x is an (r 1)
fold root of R(X,
y, z) if and only if the line with equation xX + yY + zZ = 0
intersects B in exactly r points.
(2.1) For a fixed (0, z, y) 6 B the polynomial (X q X) divides Ry,z (X).
Moreover, if k < q 1 then Ry,z (X) == (X q X)f (X, y, z) for every
(0, z, y) 6 B; and f (X, y, z) splits into linear factors over GF(q) for these
fixed (y, z)s.
(2.2) If the line [0, z, y] (where (0, z, y) 6 B) meets f (X, Y, Z) at (x, y, z) with
multiplicity m, then the line with equation xX + yY + zZ = 0 meets B in
exactly m + 1 points.
This theorem shows that the curve f has a lot of GF(q)-rational points and helps
us to translate geometric properties of B into properties of f .
142
Proof: (1.1) and (1.2) are straightforward from the definition of the Redei polynomial. The multiplicity of a root X = x is the number linear factors in the
product defining R(X, Y, Z) that vanish at (x, y, z), which is just the number of
points of B lying on the line
Q [x, y, z]. The first part of (2.1) follows from (1.1) and
the well-known fact that xGF(q) (X x) = X q X. The rest of (2.1) is obvious.
To prove (2.2) note that if the intersection multiplicity is m, then x is an (m + 1)fold root of Ry,z (X). Now the assertion follows from (1.1).
The facts given in Theorem 21.7 will be used frequently without further reference.
The next lemma shows that the linear components of f (or the curve C defined by
f = 0) correspond to points of B which are not essential.
Lemma 21.8. ([123])
(1.1) If a point P (a, b, c) B \ LX is not essential, then aX + bY + cZ divides
f(X, Y, Z) (as polynomials in three variables).
(1.2) Conversely, if NX < q + 2 k and aX + bY + cZ divides f(X, Y, Z), then
(a, b, c) B \ LX and (a, b, c) is not essential.
(2.1) If a point P (0, b, c) BLX is not essential, then X q X divides R(X,

c, b)
(as polynomials in three variables).
(2.2) Conversely, if X q X divides R(X,

c, b), then (0, b, c) cannot be an essential
point of B.
Proof: (1.1): Take a point Q(0, z0 , y0 ) 6 B. For this Q(0, z0 , y0 ) there are at
least two points of B on the line P Q, hence (aX + by0 + cz0 ) divides f(X, y0 , z0 ).
In other words, the line L : aX + bY + cZ and C have a common point for
(Y, Z) = (y0 , z0 ). This happens for q + 1 NX values of (y0 , z0 ), so Bezouts
theorem implies that L is a component of C.

(1.2): Conversely, if aX +bY +cZ divides f(X, Y, Z), then for every Q(0, z0 , y0 ) 6
B the line through Q and (a, b, c) intersects B in at least two points. If (a, b, c)
/ B,
then |B| 2(q + 1 NX ) + NX . Putting |B| = q + k gives a contradiction. Hence
(a, b, c) B. Since every line through (0, z0 , y0 ) 6 B, contains at least two points
of B, the point (a, b, c) cannot be essential.
(2.1) and (2.2) can be proved in a similar way as (1.1) and (1.2).
If the line [1, 0, 0] is a tangent, or if B is a small blocking set, then the previous
lemma simply says that there are no linear components of f if |B| < 2q. Note that
also in Segres theory there is a lemma corresponding to this one (see [77], Lemmas
10.3.2 and 10.4.), and it plays an important role in proving the incompleteness of
arcs.
21. Blocking sets
143
Recall also a lower bound on the number of GF(q)-rational points of certain components of f , see Blokhuis, Pellikaan, Szonyi [47].
Lemma 21.9. ([47]) (1) The sum of the intersection multiplicities I(P, f `P )
over all GF(q)-rational points of f is at least deg(f )(q + 1) deg(f)NX , where `P
denotes the line through P and (1, 0, 0) (the horizontal line). If g is a component
of f , then the corresponding sum for g is at least deg(g)(q+1)deg(
g )(NX ), where
g0 = g.c.d.(g, rNX ) and g = g0 g.
(2) Let g(X, Y, Z) be a component of f (X, Y, Z) and suppose that it has neither
multiple components nor components with zero partial derivative w.r.t. X. Then
the number of GF(q)-rational points of g is at least
deg(g)(q + 1) deg(
g )(NX + deg(
g ) 1)
Proof: Let g = g0 g, where g0 contains the product of some linear components
(hence g0 |rNX ) and g has no linear component; s = deg(g), s = deg(
g ). First note
that the linear components of rNX all go through (1, 0, 0) while f does not. For
any fixed (Y, Z) = (y, z), for which (0, z, y) 6 B, the polynomial f (X, y, z) is the
product of linear factors over GF(q), hence the same is true for every divisor g of
f . So the number of points, counted with the intersection multiplicity of g and the
horizontal line at that point, is at least s(q + 1 NX ) + deg(g0 )(q + 1). To count
the number of points without this multiplicity we have to subtract the number of
0
(see [47]); Bezouts theorem then gives the result. Note
intersections of g and gX
0
are counted once if
also that in this counting the common points of g and gX
the intersection multiplicity I(P ; g `P ) is not divisible by p, and the points with
intersection multiplicity divisible by p are not counted at all. Hence we have at
least s(q + 1 NX ) + (s s)(q + 1) s(
s 1) points of g.
These elementary observations already yield interesting results on blocking sets.
We mention without a proof that Lemma 21.9, combined with the Weil-estimate on
the number of rational points of a curve gives the result of Bruen |B| q + q + 1.
We repeat a lemma of Blokhuis and Brouwer.
Proposition 21.10. ([44]) There are at most k 2 k + 1 lines that meet B in at least
two points.
Proof: Exercise 12.5, with |B| = q + k, gives that the total number of tangents
is at least (q + k)(q + 1 k), which means that there are at most k 2 k + 1 lines
intersecting B in at least two points.
Now we are ready to prove Blokhuis theorem 21.4 in the prime case.
Theorem 21.11. (Blokhuis [33]) In PG(2, p), p prime, the size of a non-trivial
blocking set is at least 3(p + 1)/2.
144
Proof: Take a component g = g (of degree s) of f. Since p is prime, it cannot

have zero partial derivative with respect to X. Therefore it has at least s(p + 1)
s(NX + s 1) points by Lemma 21.9. On the other hand, again since p is prime,
it cannot be non-classical with respect to lines. Therefore, by the Stohr-Voloch
theorem 10.9, it has at most s(p + s 1)/2 GF(p)-rational points. This implies
s(p + 1) s(NX + s 1) s(p + s 1)/2,
which means s (p + 5 2NX )/3. In particular, if |B| < p + 1 + 2(p + 3)/3 and
we choose LX to be a tangent, then the curve f must be (absolutely) irreducible.
Now Lemma 21.9 can be applied to f itself and it says that f has at least (k
1)(p + 1) (k 1)(k 1) points. On the other hand, the previous lemma shows
that it can have at most k 2 k + 1 points over GF(p). Solving the inequality
pk k(k 1) k 2 k + 1 implies k (p + 2)/2.
Let us repeat the interesting observation that for |B| < p + 1 + 2(p + 3)/3 the
curve f itself must be irreducible. A second proof of a slightly weaker bound will
be given in Proposition 21.20. In case of a prime-power q the same argument shows
that for |B| p + (p + 3)/3 each component of f whose partial derivative with
respect to X is not identically zero must be non-classical with respect to lines.
However, the existence of such non-classical components can be excluded by using
a pair or triple of curves, as we shall see soon.
21.2
Three new curves
In this subsection we introduce three nice curves. We use the notation V =

(X, Y, Z); Vq = (X q , Y q , Z q ) and = V Vq = ((Y q Z Y Z q ), (Z q X
ZX q ), (X q Y XY q )). Let B be a minimal blocking set of PG(2, q). Since
R(X, Y, Z) vanishes for all homogeneous (x, y, z) GF(q) GF(q) GF(q), we
can write it as
R(X, Y, Z) = g = det(V, Vq , g) =
(Y q Z Y Z q ) g1 (X, Y, Z) + (Z q X ZX q ) g2 (X, Y, Z) + (X q Y XY q ) g3 (X, Y, Z),
where g1 , g2 , g3 are homogeneous polynomials of degree k 1 in three variables
and g = (g1 , g2 , g3 ). Note that g is not determined uniquely, it can be changed
by g0 = g + g0 V for any homogeneous polynomial g0 GF(q)[X, Y, Z] of total
degree k 2, if k < q; and for g0 = g + g0 V + g00 Vq for arbitrary homogeneous
polynomials g0 of degree k 2 and g00 of degree k q 1.
Why is this a most natural setting? For example observe that if B is the line
[a, b, c] then R = (a, b, c) .
Now one can define f = V g. Then f (Vq V) = f1 (X q X) + f2 (Y q Y ) +
f3 (Z q Z) = R, and f1 , f2 , f3 are homogeneous polynomials of degree k. If k < q
then, by this decomposition of R, f is determined uniquely. Conversely, if for
21. Blocking sets
145
some g0 also f = V g0 holds then g0 = g + gV for some homogeneous polynomial

g of degree k 2.
We also remark that, as V (V g) = 0, we have Vf = 0. For another proof see
21.16.
If k q then f is not necessarily unique in the decompositon of R. But if we
choose f = V g for some g then 21.16 remains valid (otherwise it may happen
that V f is not the zero polynomial).
The following lemma summarizes some fundamental properties of g.
Proposition 21.12. (1.1) If a point P (a, b, c) B is not essential, then there exists
an equivalent g0 = g + g0 V (or g0 = g + g0 V + g00 Vq ) of g such that aX +
bY + cZ divides gi0 (X, Y, Z), i = 1, 2, 3 (as polynomials in three variables).
(1.2) Conversely, if NX < q +2k and aX +bY +cZ divides each gi (X, Y, Z), i =
1, 2, 3, then (a, b, c) B and (a, b, c) is not essential.
(2) If B is minimal then g1 , g2 and g3 have no common factor.
Proof: (1.1) In this case R0 = R/(aX + bY + cZ) still vanishes everywhere, so it
can be written in the form R0 = g0 , so (g0 (aX + bY + cZ) g) = 0.
(1.2) Now aX + bY + cZ divides R as well, so (a, b, c) B. Deleting it, the
g1 (X,Y,Z)
q
Redei polynomial of the new pointset is (Y q Z Y Z q ) aX+bY
+cZ + (Z X
g2 (X,Y,Z)
g3 (X,Y,Z)
q
q
ZX q ) aX+bY
+cZ + (X Y XY ) aX+bY +cZ , so it remains a blocking set.
(2) Such a factor would divide R as well, which splits into linear factors. Then for
a linear factor see (1.2).
We want to evaluate R along a line [a, b, c] of the dual plane (so we examine the
lines through (a, b, c) of the original plane). We use the notation

(X, Y, Z)
= (bZ cY, cX aZ, aY bX).
[a,b,c]

In general f (X, Y, Z)
= f (bY cZ, aY, aZ) = f (bX, aX cZ, bZ) =

[a,b,c]
f (cX, cY, aX bY ), where e.g. f (bY cZ, aY, aZ) can be used if a 6= 0 etc.
Theorem 21.13. (1)

R
= ag1
[a,b,c]
[a,b,c]

+ bg2
[a,b,c]

+ cg3

[a,b,c]

(ZY q Z q Y )

[a,b,c]

(the last factor should be changed for (XZ q X q Z)
if a = 0 and b 6= 0
[a,b,c]

and for (Y X q Y q X)
if a = b = 0 and c 6= 0, normally these factors
[a,b,c]
are identical when restricting to aX + bY + cZ = 0).
146

(2) (a, b, c) B if and only if ag1
[a,b,c]

+ bg2
[a,b,c]

+ cg3
= 0. It means that
[a,b,c]
if one considers this equation as an equation in the variables a, b, c then the

points of B are exactly the solutions of it.
(3) If (a, b, c) B then consider

+ cg3
+ bg2
a(g1

R
[a,b,c]
[a,b,c]
[a,b,c]

q
q

=
(ZY
Z
Y
)

aX + bY + cZ
[a,b,c]
(aX + bY + cZ)
[a,b,c]
[a,b,c]
(4) Suppose (a, b, c) 6 B then if a line

[x, y, z] through (a, b, c) is an r-secant of

B, then (x, y, z) is a root of ag1
+ bg2
+ cg3
with multiplicity
[a,b,c]
[a,b,c]
[a,b,c]
r 1.
Proof:
Easy calculations. For instance to prove (2) we simply need

R
= ag1
+ bg2
+ cg3
= 0.
[a,b,c]
[a,b,c]
[a,b,c]
[a,b,c]
See Example 7.4 for showing the use of (2) above: there we get that the equation
of the canonical Baer subplane is
Ga,b,c (X, Y, Z) = X
(c
b cb
)+Y
(a
b ab
)+Z
(a
c ac
) = 0,
meaning that the Baer subplane is just {(a, b, c) PG(2, q) : Ga,b,c (X, Y, Z) 0}.
The map [x, y, z] 7 [g1 (x, y, z), g2 (x, y, z), g3 (x, y, z)], acting on the lines, is a
remarkable one.
Proposition 21.14. Let [x, y, z] be a tangent line to B at the point (at , bt , ct ) B.
Then [g1 (x, y, z), g2 (x, y, z), g3 (x, y, z)] is also a line through (at , bt , ct ), different
from [x, y, z].
If [x, y, z] is a secant line then [g1 (x, y, z), g2 (x, y, z), g3 (x, y, z)] is either [x, y, z]
or meaningless (i.e. [0, 0, 0]).
Obviously, if g(x, y, z) = [0, 0, 0] then [x, y, z] is a 2-secant as the 1-st derivatives are 0.
Proof:
Recall Theorem 7.10, here we have

(at , bt , ct ) = ( (X R)(x, y, z), (Y R)(x, y, z), (Z R)(x, y, z) ) =
( yg3 (x, y, z) + zg2 (x, y, z), xg3 (x, y, z) zg1 (x, y, z), yg1 (x, y, z) xg2 (x, y, z) ).
Now the scalar product with (g1 (x, y, z), g2 (x, y, z), g3 (x, y, z)) vanishes.
21. Blocking sets
147
or:
(at , bt , ct )g(x, y, z) = (R)(x, y, z)g(x, y, z) = ((x, y, z)g(x, y, z))g(x, y, z) = 0.
Here (x, y, z) 6= g(x, y, z) as their cross product is (at , bt , ct ).
If [x, y, z] is a secant line then there are more than one components of R going
through (x, y, z) (see Theorem 7.10) hence
0 = (R)(x, y, z) = (x, y, z) g(x, y, z).
or: we can use Theorem 21.13.

We also calculate R for future use: R = ( g) + Vq g.
The following is true as well. We will see (Theorem 21.19, Corollary 21.23) that
if B is small, then R can be written in the form
X q (Y gXY + ZgXZ ) + Y q (XgY X + ZgY Z ) + Z q (XgZX + Y gZY ),
where the polynomials gXY , ... contain only exponents divisible by p. Surprisingly
or not, gXY = gY X , ..., as 0 = Xf1 + Y f2 + Zf3 = XY (gXY + gY X ) + Y Z(gY Z +
gZY ) + ZX(gZX + gXZ ). One can take X Y , Y Z , Z X of both sides.
Now g = (gY Z , gZX , gXY ), which is a kind of natural choice for g.
Or: gXY = Y f1 = X f2 ,..., or similarly (if p 6= 2) g = f and f = V(f );
g = 1R (qH R).
21.3
Three old curves
In this section we will present the method using three old algebraic curves. As
an application we show Sz
onyis result [123] that blocking sets of size less than
3(q + 1)/2 intersect every line in 1 modulo p points. This immediately implies
Blokhuis theorem for blocking sets in PG(2, p).
Let now B be a minimal blocking set of PG(2, q). Since R(X, Y, Z) vanishes for all
(x, y, z) GF(q) GF(q) GF(q), we can write it as
R(X, Y, Z) = (X q X)f1 (X, Y, Z)+(Y q Y )f2 (X, Y, Z)+(Z q Z)f3 (X, Y, Z)=Wf ,
where f = (f1 , f2 , f3 ) and deg(fi ) k as polynomials in three variables. Note that
f1 here is the same as the polynomial f defined in Definition 21.5 and examined
in Section 21.1; while f2 and f3 behave very similarly.
Proposition 21.15. (Lov
asz, Sz
onyi) Let [x, y, z] be a tangent line to B at the point
(at , bt , ct ) B. Then
f (x, y, z) = (f1 (x, y, z), f2 (x, y, z), f3 (x, y, z)) = (at , bt , ct )
as homogeneous triples.
148
Proof:

Recall Theorem 7.10, here we have
((X R)(x, y, z), (Y R)(x, y, z), (Z R)(x, y, z)) = (f1 (x, y, z), f2 (x, y, z), f3 (x, y, z))
Or:
(at , bt , ct ) = (R)(x, y, z) = ((W f ))(x, y, z) = ((W)f +(f )W)(x, y, z) =
I f (x, y, z) + 0 = f (x, y, z).
Lemma 21.16. If k < q 1 then V f = Xf1 + Y f2 + Zf3 = 0. (Hence R = Vq f
as well).
Proof: If [x, y, z] is a tangent then by 21.15 V f vanishes, and by the end of
Theorem 7.10 it also vanishes if [x, y, z] is at least a 2-secant. As the degree is less
than q we are done. Or: it is just Theorem 7.10 (6) with r = 1.
Note also that f = qH R; and f = ( f )V. Moreover, in general i1
H f =
(iH f )V. It follows from the derivation of f V = 0.
From Theorem 7.10 (5) one can see that each of the curves f1 , f2 , f3 go through the
point (x, y, z) of the dual plane corresponding to a secant line [x, y, z]. Where are
the other (extra) points of e.g. f1 ? They are exactly the points of rNX of Lemma
21.6, so points on factors corresponding to points with ai = 0.
If one fixes (Y, Z) = (y, z) then R(X, y, z) is divisible by (X q X). If R(X, y, z) 6=
0, so if (0, y, z) 6 B LX then for an (x, y, z) GF(q) GF(q) GF(q) if the line
with equation xX +yY +zZ = 0 intersects B in at least two points (cf. Proposition
21.7 (2.2)) then f1 (x, y, z) = 0. One can repeat the same reasoning for f2 , f3 and
this immediately gives the following lemma:
Lemma 21.17. ([123]) The curves fi have almost the same set of GF(q)-rational
points. The exceptional points correspond to lines intersecting LX , LY or LZ in a
point of B.
Proof: Since this observation is crucial, a direct proof is also included. Consider
the Redei polynomial R(X, Y, Z). For an element (x, y, z) GF(q) GF(q)
GF(q) we get f1 (x, y, z) = X R(x, y, z) and similarly f2 (x, y, z) = Y R(x, y, z).
Since R is a product of linear factors and R(x, y, z) = 0, X R(x, y, z) = 0 if and
only if there are two linear factors vanishing at (x, y, z), or if R(X, y, z) = 0 (i.e.
(0, z, y) B). The similar statement holds for Y R, hence the two derivatives are
zero for the same values (x, y, z), except in the cases described in the statement.
Lemma 21.18. ([123]) If k < q 1 then the polynomials f1 , f2 and f3 cannot have a
common factor. Moreover, e.g. f1 and f2 have a common factor g iff (0, 0, 1) B
and g = Z.
21. Blocking sets
149
Proof: Such a common factor must divide R(X, Y, Z), hence it must be divisible
by ai X + bi Y + ci Z for some i. Lemma 21.8 (2) gives (N = 1, k q 2) that the
point (ai , bi , ci ) can be deleted, a contradiction.
Suppose that g is a common factor of f1 and f2 , then from Xf1 + Y f2 + Zf3 = 0
we have g|Zf3 .
Therefore, (f1 , f2 , f3 ) is a triple of polynomials (curves) having no common factor
(component), but they pass through almost the same set of GF(q)-rational points.
Using Bezouts theorem it immediately gives Lemma 21.10 back.
21.4
Small blocking sets
Lemmas 21.17 and 21.18 can also be used to show that all the components of
f have identically zero partial derivative with respect to X. Note that for any
component h of f the total degree of h is the same as its degree in X.
Theorem 21.19. ([123]) If k (q+1)/2 and g(X, Y, Z) is an irreducible polynomial
0
= 0.
that divides f1 (X, Y, Z), then gX
Proof: Suppose to the contrary that g is a component of f1 with nonzero partial
X-derivative, denote its degree by deg(g) = s. By Lemma 21.9 the number of
GF(q)-rational points on g is at least s(q + 2 NX s). Since these points are
also on f2 , Bezouts theorem gives s(q + 2 NX s) sk, since by Lemma
21.18, if f2 and g has a common component (i.e. g itself) then it cannot be a
component of f3 and one can use Bezout for g and f3 instead. This immediately
implies q + 2 k + NX + s and from NX + s k it follows that k (q + 2)/2, a
contradiction.
Note that it implies that all the X-exponents appearing in f1 are divisible by p
(as rNX does not involve X); and a similar statement holds for the Y -exponents
of f2 and for the Z-exponents of f3 . Lets define e, the (algebraic) exponent of B,
e
e
as the greatest integer such that f1 GF(q)[X p , Y, Z], f2 GF(q)[X, Y p , Z] and
e
f3 GF(q)[X, Y, Z p ]. By the Theorem e 1.
X
Proposition 21.20. If q = p is a prime and |B| < p + 2p+4N
, then the curve f1
3
is irreducible (and similarly for f2 , f3 ).

Proof: Suppose to the contrary that e.g. f1 is not irreducible, and let g be a
component of f1 of degree at most (k NX )/2. The proof of Theorem 21.19 gives
X
p + 2 k + NX + deg(g) 3k/2 + NX /2, that is 2(p+2)N
k.
3
The following corollary, due to Szonyi, generalizes the similar result of Redei on
blocking sets of Redei type.
150
Corollary 21.21. ([123]) If B is a blocking set of size less than 3(q + 1)/2, then
each line intersects it in 1 modulo p points.
Proof: Take a line ` and coordinatise such that ` LX B = . If ` = [x, y, z]
then rNX (y, z) 6= 0. Since all the components of f1 contain only terms of exponent
(in X) divisible by p, for any fixed (Y, Z) = (y, z) the polynomial f1 (X, y, z) =
rNX (y, z)f1 (X, y, z) itself is the p-th power of a polynomial. This means that at
the point P (x, y, z) the horizontal line (i.e. through P and (1, 0, 0)) intersects
f1 (X, Y, Z) with multiplicity divisible by p (and the same is true for f1 ), so by
Theorem 21.7 the line [x, y, z] intersects B in 1 modulo p points.
Note that now we have |B| 1 (mod p). Of course, this theorem also implies
Blokhuis theorem in the prime case.
Corollary 21.22. (Blokhuis [33]) If q = p is a prime, then |B| 3(q + 1)/2 for the
size of a non-trivial blocking set.
Corollary 21.23. If B is a blocking set of size less than 3(q + 1)/2,

then the X-exponents in f1 , the Y -exponents in f2 and the Zexponents in f3 are 0 (mod p)e ; moreover all the exponents appearing in
R(X, Y, Z), f1 , f2 , f3 ; rNX (Y, Z), rNY (X, Z), rNZ (X, Y ), are 0 or 1 (mod p)e .
Proof: The first statement is just Theorem 21.19. From this the similar statement
follows for fi : the X-exponents in f1 , the Y -exponents in f2 and the Z-exponents
in f3 are 0 (mod p)e .
e
Consider a term aX p +1 Y Z of Xf1 in the identity Xf1 + Y f2 + Zf3 0. It

should be cancelled by Y f2 and Zf3 , which means that it should appear in either
one or both of them as well with some coefficient. It cannot appear in both of them,
e
0 e
0 e
as it would imply exponents like X p +1 Y p +1 Z p +1 , but the exponents must
add up to k + 1, which is 2 mod pe , a contradiction. So this term is cancelled by its
e
0 e
negative, for example contained in Y f2 , then it looks like aX p +1 Y p +1 Z ,
where the exponents, again, add up to k + 1, which is 2 mod pe , hence 0
e
0 e
0 e
(mod pe ), so the original term of f1 was of form aX p Y p +1 Z p .
For rNX (Y, Z), rNY (X, Z) and rNZ (X, Y ) recall that they are also homogeneous
polynomials of total degree 1 (mod pe ) and for instance f1 = rNX f1 and deg f1 =
degX f1 , so in f1 the terms of maximal X-degree have 0 or 1 mod pe exponents
(as terms of f1 ), on the other hand they together form rNX X kNX .
Finally R = X q f1 + Y q f2 + Z q f3 so R has also 0 or 1 mod pe exponents only.
Note that in fi other exponents can occur as well. Comparing the exponents one
can find Y Y f1 + ZZ f1 = XX f1 + Y Y f1 + ZZ f1 = 0 as well.
21. Blocking sets
151
The (geometric) exponent eP of the point P can be defined as the largest integer
for which each line through P intersects B in 1 mod peP point. It can be proved
(e.g. [39]) that the minimum of the (geometric) exponents of the points in B is
equal to e defined above.
Theorem 21.24. [118] Let B be a blocking set with exponent e. If for a certain line
|` B| = pe + 1 then GF(pe ) is a subfield of GF(q) and ` B is GF(pe )-linear.
Proof: Choose the frame such that ` = LX and (0, 0, 1); (0, 1, 0); (0, 1, 1) ` B.
Consider f = f1 , now rNX (Y, Z) is a homogeneous polynomial of (total) degree pe +
e
e
e
e
1, with exponents 0, 1, pe or pe + 1, so of form Y p +1 + Y Z p + Y p Z + Z p +1 .
e
p
pe
As
Q rNX (0, 1) = rNX (1, 0) = rNX (1, 1) = 0 we have rNX = Y Z Y Z =
(a,b)PG(1,pe ) (aY + bZ).
Now we can disclose one of our main goal: to get as close as we can to the proof
of the conjecture that every small blocking set is linear.
By the following proposition, a blocking set with exponent e has a lot of (pe + 1)secants (so nice substructures). Similar arguments can be found in [37].
Proposition 21.25. Let P be any point of B with exponent eP .
(1) (Blokhuis) There are at least (q k + 1)/peP + 1 secant lines through P .
(2) Through P there are at most 2(k 1)/peP 1 long secant lines, i.e. lines
containing more than peP + 1 points of B (so at least q/peP 3(k 1)/peP +
2 (peP + 1)-secants).
(3) There are at most 4k 2peP 4 points Q B \ {P } such that P Q is a long
secant.
(4) There are at least q 3k + 2pe + 4 points in B with (point-)exponent e.
Proof: (1) was proved by Blokhuis using lacunary polynomials. To prove (2)
denote by s the number of (peP + 1)-secants through P and let r be the number of
( 2peP + 1)-secants through P . Now speP + 2rpeP + 1 q + k. From (1) s + r
(qk+1)/peP +1, so q/peP (k1)/peP +r+1 s+2r q/peP +(k1)/peP hence
r 2(k 1)/peP 1 and s q/peP (k 1)/peP +1r q/peP 3(k 1)/peP +2.
For proving (3) subtract the number of points on (peP +1)-secants through P from
|B|, it is q + k (q/peP 3(k 1)/peP + 2)peP 1 = 4k 2peP 4. There is
at least one point P B for which eP = e. On the pe + 1-secants through it (by
(2)) we find at least 1 + pe (q/pe 3(k 1)/pe + 2) points, each of exponent e, it
proves (4).
Recall that there are at least q+1k tangent lines through P , so at most k secants.
We also know from Sz
onyi [123] that q/pe +1 k q/pe +q/p2e +2q/p3e +... Now
152
almost all line-intersections of B are GF(pe )-linear (in fact they are isomorphic
to PG(1, pe ) in the non-tangent case).
Corollary 21.26. [118] For the exponent e of the blocking set, e|h (where q = ph ).
Proof: By Proposition 21.25 B has a lot of short secants. By Theorem 21.24 these
intersections are all isomorphic to PG(1, pe ), so GF(pe ) is a subfield of GF(ph ) =
GF(q).
Now we can give a very short proof for Theorem 18.14 in the case when pe > 13.
Corollary 21.27. [118] Small blocking sets of Redei type, with pe > 13, are linear.
Proof: Suppose LZ is the Redei-line, O = (0, 0, 1) B, eO = e, and take any
P B \ LZ , with eP = e, and any GF(pe ). Claim: P (affine point!) is also in
B. If OP is a short secant then it is obvious.
Consider the short secants through P , there are at least q/pe 3(k 1)/pe + 2.
e
e
4
Most of them, at least q/pe 3(k 1)/pe + 2 4k2p
pqe 7k3k/p
pe 1
pe 1 , say
{ì : i I}, contain at least two points Q1 , Q2 B, such that OQ1 and OQ2 are
short secants.
For any of them, say ì , take ì , it contains P . If all of {ì : i I} were long
e
secants then they would contain at least 2pe ( pqe 7k3k/p
pe 1 ) > q + k points of B,
contradiction if pe > 13. Say ` is a short secant, then P B ` and eP = e
as well.
Let U0 be the set of affine points of B with exponent e. Now we have that U0 is
invariant for magnifications from any center in U0 and with any scale GF(pe ),
so it forms a vectorspace over GF(pe ). As its size is q 3k + 2pe + 4 |U0 | q we
have |U0 | = q and it contains all the affine points of B.
Consequences
The bounds for the sizes of small blocking sets are now the following.
Corollary 21.28. Let B be a minimal blocking set of PG(2, q), q = ph , of size
|B| < 3(q + 1)/2. Then there exists an integer e, called the exponent of B, such
that
1 e|h,
and
e
1+(pe +1)(q+1) (1+(pe +1)(q+1))2 4(pe +1)(q 2 +q+1)
+1
q + 1 + pe d q/p
e
|B|
.
pe +1
2
If |B| lies in the interval belonging to e and pe 6= 4 then each line intersects B in
1 modulo pe points. Most of the secants are (pe + 1)-secants, they intersect B in a
pointset isomorphic to PG(1, pe ).
21. Blocking sets
153
These bounds are due to Blokhuis, Polverino and Szonyi, see [99, 123], and asymptotically they give q + pqe pq2e + pq3e ... |B| q + pqe + pq2e + 2 pq3e + ... . Note
that for q = p2s and q = p3s , where s is a prime, the lower bound is sharp:
|B| q + q/ps + 1 and |B| q + q/p2s + 1, resp.
The 1 mod pe property was established by Szonyi; our Theorem 21.26 shows that
only a very few of the intervals of Szonyi, Blokhuis, Polverino contain values from
the spectrum of blocking sets, i.e. only those with e|h. The linearity of short secants
is Theorem 21.24, on their number see Proposition 21.25.
Let S(q) denote the set of possible sizes of small minimal blocking sets in PG(2, q).
Corollary 21.29. Let B be a minimal blocking set of PG(n, q), q = ph , withrespect
to k-dimensional subspaces, of size |B| < 32 (q nk + 1), and of size |B| < 2q nk
if p = 2. Then
|B| S(q nk );
if p > 2 then ((|B| 1)(q nk )n2 + 1) S((q nk )n1 ).
If p > 2 then there exists an integer e, called the exponent of B, such that
1 e|h,
for which every subspace that intersects B, intersects it in 1 modulo pe points. Also
|B| lies in an interval belonging to some e0 e, e0 |h. Most of the k-dimensional
subspaces intersecting B in more than one point, intersect it in (pe + 1) points
precisely, and each of these (pe + 1)-sets is a collinear pointset isomorphic to
PG(1, pe ).
Most of this was proved by Sz
onyi and Weiner in [129]. Consider the line determined by any two points in a (pe + 1)-secant k-subspace, this line should contain
pe + 1 points. Then the technique of [129] can be used to derive a planar minimal
blocking set (in a plane of order q nk ) with the same exponent e: firstly embed
PG(n, q) into PG(n, q nk ) where the original blocking set B becomes a blocking set
w.r.t. hyperplanes, then choose an (n 3)-dimensional subspace PG(n, q nk )
not meeting any of the secant lines of B and project B from onto a plane
PG(2, q nk ) to obtain a planar minimal blocking set, for which Theorem 21.24
and Proposition 21.25 can be applied, implying e|h(n k).
Now in PG(n + 1, q) PG(n, q) build a cone B with base B and vertex V
PG(n + 1, q) \ PG(n, q); then B will be a (small, minimal) blocking set in PG(n +
1, q) w.r.t. k-dimensional subspaces. The argument above gives e|h(n + 1 k), so
e | g.c.d.(h(n k), h(n + 1 k)) = h.
Exercise 21.30. Let B PG(2, q) be a double blocking set of size 2q + k. Then
R(X, Y, Z) = W F(X, Y, Z) WT ,
where F(X, Y, Z) = (fij (X, Y, Z))33 .
154
22
Multiple blocking sets
In this section we give a generalization of the Redei-polynomial approach of blocking sets. Some of the statements below (and much more) can be found in [37], some
others in [62].
Let B = {(ai , bi , ci )} be a t-fold blocking set in PG(2, q), having possibly weighted
(i.e. multiple) points, |B| = tq + k.
We use the very same Redei polynomial as we did before.
R(X, Y, Z) =
|B|
Y
(ai X + bi Y + ci Z)
i=1
|B|
X
X |B|j rj (Y, Z)
j=0
fj1 j2 j3 (X, Y, Z)(X q X)j1 (Y q Y )j2 (Z q Z)j3 ,
()
0j1 ,j2 ,j3 t

j1 +j2 +j3 =t
where deg(rj ) j, j = 0, . . . , |B|, and each fj1 j2 j3 , if present, is a homogeneous

polynomial of degree k, for 0 j1 , j2 , j3 t, j1 + j2 + j3 = t. The latter equation
above follows from the fact that this Redei polynomial is zero t times everywhere
in PG(2, q).
The polynomials fj1 j2 j3 satisfy a (low degree) polynomial equation:
Lemma 22.1. If k + t < q then
X
X j1 Y j2 Z j3 fj1 j2 j3 (X, Y, Z) = 0
0j1 ,j2 ,j3 t

j1 +j2 +j3 =t
and
R(X, Y, Z) =
fj1 j2 j3 (X, Y, Z)X j1 q Y j2 q Z j3 q .
0j1 ,j2 ,j3 t

j1 +j2 +j3 =t
Proof: For the second equation observe that in () the left hand side is homogeneous of degree tq + k so all the terms on the right hand side with degree 6= tq + k
must vanish.
As a special case, consider the lowest degree terms of (), they are of degree k + t,
and they must disappear after summing them, this proves the first equation.
j1 q j2 q j3 q
As a corollary note that fj1 j2 j3 = HX
HY HZ R as polynomials,
and also for

j1 j2 j3
every x, y, z GF(q) we have fj1 j2 j3 (x, y, z) = HX HY HZ R (x, y, z).
22. Multiple blocking sets
155
We would like to understand what the (GF(q)-rational points of the) curves fj1 j2 j3
mean. For ft00 it is obvious: (x, y, z) is a point of it if and only if the line [x, y, z]
is either an ( t + 1)-secant (this is the typical case), or if [x, y, z] intersects the
line [1, 0, 0] in a point of B. Similar statements hold for f0t0 and f00t .
Proposition 22.2. If the line [x, y, z] is an ( t + 1)-secant then fj1 j2 j3 (x, y, z) = 0
for all 0 j1 , j2 , j3 t, j1 + j2 + j3 = t.
Proof: Fix 0 j1 , j2 , j3 t, j1 + j2 + j3 = t and some (t + 1)-secant line
j1 j2 j3
HY HZ R (x, y, z) = 0 and
[x, y, z]. Recall Theorem 7.10 (3) stating that HX

j1 j2 j3
also fj1 j2 j3 (x, y, z) = HX
HY HZ R (x, y, z).
22.1
A t (mod p) result
Let B be a minimal weighted t-fold blocking set in PG(2, q), with |B| = tq + t + k,
where 2t + k < q + 2. From now on we are going to use a more affine point of
view, still based on [37] and [62].
Assume that the line l is an m-secant to B. Consider PG(2, q) as the affine
plane AG(2, q) with l as the line at infinity. Assume that B l = D =
{(), . . . , (), (y1 ), . . . , (yms )}, where () is a point of weight s of B (1 s t),
where some of the other points of D might be multiple points of B, and that
U = B \ D = {(ai , bi ) : i = 1, . . . , tq + t + k m}, where U is a multiset when B
has affine multiple points.
Now we redefine the Redei polynomial associated to the t-fold blocking set B,
(with () deleted).
Definition 22.3 (The Redei polynomial of the set B).
R(X, Y ) =
ms
Y
(Y yj )
j=1
ms
Y
j=1
tq+t+km
Y
(X + ai Y bi )
i=1
(Y yj )
tq+t+km
X
X tq+t+kmi ri (Y )
(2.1)
i=0
= (X q X)t f0 (X, Y ) + (X q X)t1 (Y q Y )f1 (X, Y )

+ + (Y q Y )t ft (X, Y ),
(2.2)
where deg(ri ) i, i = 0, . . . , tq + t + k m, and deg(fi ) k + t s, i = 0, . . . , t.

Choose a point (b, m), b 6= bj , m 6= yj . Consider R(X, m) = (X q X)t f0 (X, m).
By the properties of the Redei polynomial, the line Y = mX + b intersects U in
more than t points iff X = b is a root of R(X, m) with multiplicity t+1 iff (b, m)
156
is a point of the algebraic curve f0 (X, Y ). Considering R(b, Y ) = (Y q Y )t ft (b, Y )

instead, we get that the line Y = mX + b intersects U in more than t points iff
(b, m) is a point of the algebraic curve ft (X, Y ).
Therefore, these two algebraic curves f0 and ft have essentially the same set of
GF(q)-rational points.
If m = yj or b = bj , and the line Y = mX + b intersects U in more than t points,
then still f0 (b, m) = ft (b, m) = 0. Because of the above mentioned divisibilities,
f0 (b, m) = 0 or ft (b, m) = 0 does not imply that Y = mX + b intersects U in more
than t points.
In this section we will assume that there is no line contained in B. As the following
theorems will show, this is no restriction when 2t + k < q + 2.
Theorem 22.4. Let B be a minimal weighted t-fold blocking set of PG(2, q), with
|B| = tq + t + k, where 2t + k < q + 2, containing a line `. Then B is the sum of
the line ` and the minimal weighted (t 1)-fold blocking set B , obtained from B
by reducing the weight of every point P of ` by one.
Proof:
Since ` B, |` B| q + 1.
If |` B| q + t, then after reducing the weight of every point of ` by one, a new

weighted set B is obtained which still intersects every line in at least t 1 points.
Since B is a minimal weighted t-fold blocking set, also B is a minimal weighted
(t 1)-fold blocking set.
Assume now that q+1 |B`| < q+t. Reduce again the weight of every point on `
by one, and add a minimal number of simple points P1 , . . . , of ` back so that again
a weighted (t1)-fold blocking set B is obtained, hence |B `| = t1. We need to
add at most t1 points P1 , . . . to achieve this, hence |B | tq+t+k(q+1)+t1 =
(t 1)q + 2(t 1) + k. A particular feature of a point Pi , i = 1, . . ., is that the line
` is the only (t 1)-secant to B passing through Pi .
Finally, we show that through Pi , there pass at least two (t 1)-secants, hence the
above case cannot occur. Now we choose our coordinate system in such a way that
() B, Pi is an affine point (a, b), and (` `) 6 B and () has multiplicity
s. Suppose that |` B | = m and write up the Redei polynomial. Since B is a
(t 1)-fold blocking set, we get that:
R(X, Y ) =
ms
Y
j=1
tq+t+k(ms)
(Y yj )
(X + ai Y bi )
i=1
= (X q X)t1 f0 (X, Y ) + (X q X)t2 (Y q Y )f1 (X, Y )

+ + (Y q Y )t1 ft1 (X, Y ),
(2.3)
where deg(fi ) |B | q(t 1) s = 2(t 1) + k 1, i = 0, . . . , t 1.
157
The argument before this theorem shows that if a line Y = mX + b intersects B

in more than (t 1) points, then (b, m) is a point of the curve f0 . Each line except
` through the point Pi (a, b) intersects B in at least t points. These lines are points
of the line X + aY b in the dual plane. Hence X + aY b intersects f0 in at least
q 1 points (we do not see the vertical line here). Since degf0 < q 1, Bezouts
theorem implies that the line X + aY b is a component of f0 . Suppose that ` is
the line ` = Y + m0 X + b0 . Then f0 (b0 , m0 ) = 0 and since ` ` 6 B , ` intersects
B in at least t points. This is a contradiction, hence q + 1 |B `| < q + t does
not occur.
As the next example shows, the above theorem is sharp.
Example 22.5. Let S be the set of points lying on the lines of a dual hyperoval in
PG(2, q), q even. Then S is a ( 2q +1)-fold blocking set of size ( 2q +1)q+( 2q +1) (each
point in S has multiplicity one). Note that now t = 2q + 1, k = 0 and 2t + k = q + 2.
If we delete a line of S, then the resulting point set is not a 2q -fold blocking set.
Remark 22.6. Theorem 22.4 has several important applications.
(1) It first of all shows that when characterizing minimal weighted t-fold blocking
sets of size tq + t + k, where 2t + k < q + 2, in PG(2, q), it is possible to assume
that they do not contain any lines.
(2) Moreover, also when proving the t (mod p) result for a minimal weighted t-fold
blocking set B, |B| = tq + t + k, where 2t + k < q + 2, it is possible to assume that
there are no lines contained in B. For, if there is a line ` contained in B, Theorem
22.4 shows that it is possible to reduce the weight of every point of ` by one in
order to obtain a new minimal weighted (t 1)-fold blocking set B . Proving the t
(mod p) result for B is now reduced to proving the (t 1) (mod p) result for B .
(3) We are also now able to characterize weighted minimal t-fold blocking sets
of size tq + t and to exclude the existence of minimal t-fold blocking sets of size
tq + t + 1.
Theorem 22.7. A weighted t-fold blocking set B in PG(2, q), of size |B| = tq + t,
where 2t < q + 2, is a sum of t lines.
There does not exist a weighted minimal t-fold blocking set B in PG(2, q) of size
|B| = tq + t + 1, 2t + 1 < q + 2.
Proof: Suppose that tq + t |B| tq + t + 1. Then counting the incidences of
the points of B with the lines through a point R not in B, we have that through R
all the lines are t-secants if |B| = tq + t and there is exactly one (t + 1)-secant and
q t-secants if |B| = tq + t + 1. Counting the incidences of the points of B with the
lines through a point R0 B, we get that R0 lies on at least one line ` completely
contained in B, when |B| = tq + t and R0 lies on at least one line ` completely
contained in B or R0 lies on q + 1 (t + 1)-secants, when |B| = tq + t + 1. This latter
158
case means that R0 is not minimal, hence we can assume that each point of B lies
on at least one line completely contained in B.
Now the t points of any t-secant (which must exist) and Theorem 22.4 show that
B contains the sum of t lines, which is a t-fold blocking set already, of size tq + t.
Remark 22.8. One can observe now that a weighted t-fold blocking set in PG(2, q),
of size tq + t, where 2t < q + 2, intersects every line in t (mod p) points; also that
through any point of it there pass at least q + 1 t t-secants.
By the theorem above, from now on we can (and will) assume that k 2 for the
minimal blocking sets considered.
In this section, from now on, we assume that B does not contain any line, hence
|B| = tq+t+k tq+t+2, and we suppose that |B| < tq+(q+3)/2. (Note that since
k 2, we still have 2t + k < q + 2.) Furthermore, we choose our coordinate system
so that ` is a t-secant and the point () in B has multiplicity s, where 1 s t.
Lemma 22.9. The polynomial
Proof:
Qts
j=1 (Y
yj ) divides f0 (X, Y ).
By (2.1),
R(X, Y ) =
tq+k
X
i=0
ri (Y )
ts
Y
(Y yj ) X tq+ki .
j=1
Qts
So every coefficient polynomial of a term X tq+ki is divisible by j=1 (Y yj ). By
Qts
Qts
(2.2), the high degree part j=1 (Y yj ) X tq+k + + rk (Y ) j=1 (Y yj ) X tq
must be equal to X tq f0 (X, Y ), when one Q
compares the X-degrees of the two
ts
expressions (2.1) and (2.2) for R(X, Y ). So j=1 (Y yj ) divides f0 (X, Y ).
If X = 0 intersects U in the, possible

weighted, points (0, bj ), j = 1, . . . , z, then a
Q
similar argument shows that bj (X bj ) divides ft (X, Y ), where the product is
taken over the values bj , according to their weights.
Theorem 22.10. Let B be a minimal weighted t-fold blocking set of PG(2, q), with
|B| = tq + t + k < tq + (q + 3)/2 and k 2. Then every point of B lies on at least
q + 1 k t different t-secants.
Proof: Let P = (a, b) U and suppose () B, |l B| = t. Assume that P
lies on more than k + t different lines sharing at least t + 1 points with B. Then
159
more than k of those lines intersect l inQa point not belonging to B. Each of these
ts
latter lines defines a point of f0 (X, Y )/ j=1 (Y yj ). More precisely, they define
Qts
intersection points, in the dual plane, of the algebraic curve f0 (X, Y )/ j=1 (Y
Qts
yj ) = 0 with the line X + aY b = 0. The polynomial f0 (X, Y )/ j=1 (Y yj )
has at most degree Q
k, so by Bezouts theorem, the linear term X + aY b is a
ts
factor of f0 (X, Y )/ j=1 (Y yj ).
Consider a line through P with slope m 6= yj , m 6= , so that we can use the
arguments above.
Plugging Y = m into R(X, Y ), we get
R(X, m) =
ts
Y
j=1
(m yj )
tq+k
Y
(X + ai m bi ) = (X q X)t f0 (X, m).
i=1
The fact that X + aY b is a linear factor of f0 means geometrically that the lines
through P with slope m 6= yj , m 6= , intersect U in at least t + 1 points.
Assume that a line ` through P with slope m = yj or m = is a t-secant. Then
replace the line at infinity l by a new line at infinity l

not intersecting ` in a
point of B, and containing exactly t points of B.
as line at infinity, we use the same notations, but in

In the affine plane with l
combination with the symbol * to make the distinction with the notations used
in the original affine plane. The theorem of Bezout will now lead to the desired
contradiction.
Through P , there pass at least q + 1 t lines that intersect B in at least t + 1

points. Whenever a line contains at least t + 1 affine points of B, it defines a
point of f0 (see the reasoning at the beginning of this section). This implies that
if the coordinates of P in the new coordinate system are equal to (a , b ), then
the line X + a Y b = 0 meets the curve f0 in at least q + 1 t t points.
We subtracted t from q + 1 t since t of those lines through P could contain a
. Then such a line through P , not containing a point

point of B belonging to l
at infinity of B, and containingQat least t + 1 affine points (ai , bi ) defines a point
ts
of the polynomial f0 (X , Y )/ j=1 (Y yj ). This latter polynomial has degree
at most k < (q + 3)/2 t.
Then, if q + 1 2t (q + 3)/2 t > k deg(f0 ), Bezouts theorem implies that
this linear component X + a Y b is a component of f0 , so also the line `
through P intersects U in at least t + 1 points, since its point at infinity in the
new coordinate system does not lie in B, and so the observations made before this
theorem apply. Note that the condition q + 1 2t (q + 3)/2 t implies that
t (q 1)/2.
We have proved that all lines through P B are at least (t + 1)-secants to B;
this contradicts the minimality of B. So a point of B lies on at most k + t lines
sharing more than t points with B.
160
Corollary 22.11. Let B be a weighted t-fold blocking set of PG(2, q), with |B| =
tq + t + k < tq + (q + 3)/2 and 2t < q + 2. Assume that P is an essential point of
B. Then there are at least q + 1 k t different t-secants through P .
Proof: Delete the non-essential points of B one-by-one until a minimal t-fold
blocking set B 0 is obtained. By Theorem 22.10 and Remark 22.8, there will be at
least q + 1 (|B 0 | tq) different t-secants of B 0 through P . Now if we add back the
points of B \ B 0 , then through P , we will see at least q + 1 (|B 0 | tq) |B \ B 0 |
t-secants to B.
Let B be a minimal weighted t-fold blocking set of size tq + t + k, where t + k <

(q + 3)/2 and k 2. Recall the definition of the Redei polynomial from the
beginning of this section, the following lemma is straightforward.
Lemma 22.12. If a line Y = mX + b intersects B U in more than t points, then
f0 (b, m) = = ft (b, m) = 0.
We will also need

Lemma 22.13. The algebraic curve f0 (X, Y ) = 0 does not have linear components
depending on the variable X.
Proof: Such a linear component depending on X should have the form
X + aY b = 0. The proof of Theorem 22.10 then shows that the point P = (a, b)
is a non-essential point of B; which contradicts the minimality of B.
Lemma 22.14. If B is minimal, then the polynomials f0 , . . . , ft cannot have a

common divisor different from Y yj .
Proof: Such a polynomial would divide R(X, Y ); so would be linear. This can
only be of the form Y yj .
We now come to the main theorem of this section: the proof of the t (mod p)
result.
161
Theorem 22.15. Let B be a minimal weighted t-fold blocking set in PG(2, q), q = ph ,
p prime, h 1, with |B| = tq + t + k, t + k < (q + 3)/2, k 2. Then every line
intersects B in t (mod p) points.
Proof: By Remark 22.6, it is possible to assume that B does not contain any
lines. We will assume that the line at infinity intersects B in t points.
Qts
Let h(X, Y ) be an absolutely irreducible component of f0 (X, Y )/ j=1 (Y yj ) of
degree larger than one. Similar arguments as in the case t = 1 imply that X h 0.
If Y = m 6= yi , we obtain R(X, m) = (X q X)t f0 (X, m), having t (mod p)
solutions since f0 (X, m) is a p-th power. So every line Y = mX + b, not containing
a point of B at infinity, intersects B in t (mod p) points.
For every line ` of which we are not yet sure that it intersects B in t (mod p)
points, it is possible to find a new line at infinity intersecting B in t points and
intersecting ` in a point not belonging to B. Repeating the previous arguments
now shows that also ` intersects B in t (mod p) points.
The next corollary follows from Theorem 22.10 and Remark 22.8.
Corollary 22.16. Let B be a weighted t-fold blocking set in PG(2, q), q = ph , p
prime, h 1, with |B| = tq + t + k, t + k < (q + 3)/2, 2t < q + 2. Assume that
all the points of B on the line ` are essential. Then ` intersects B in t (mod p)
points.
When each line intersects B in t (mod q) points, then the characterization of B

is immediate.
Proposition 22.17. Let B be a minimal weighted t-fold blocking set in PG(2, q) of
size tq + t + k, where t + k < (q + 3)/2, k 2. Assume that each line intersects B
in t (mod q) points. Then B is a sum of t (not necessarily different) lines.
Proof: Let ` be a line of PG(2, q) not contained in B. Let P ` \ B. Since all
the lines, different from `, through P contain at least t points of B, ` contains at
most t + k points of B.
Every point R of B lies on at least one line containing more than t points of B, so
on a line ` containing at least t + q points of B. Since t + k < t + q, the preceding
paragraph implies that ` is contained in B. By Theorem 22.4, B is the sum of this
line ` and a (t 1)-fold blocking set B intersecting every line in (t 1) (mod q)
points. Repeating the above argument shows that B is a sum of t lines.
These results are enough to give a determine a lower bound on the size of a minimal
weighted t-fold blocking set B in PG(2, q), q = ph , p prime, h 1.
162
We again assume that B does not contain any lines, for it is trivially possible to
construct a minimal weighted t-fold blocking set in PG(2, q) by taking a sum B of
t lines. Then |B| = t(q + 1).
Theorem 22.18. Let B be a minimal weighted t-fold blocking set in PG(2, q), q = ph ,
p prime, h 1, with |B| = tq + t + k, t + k < (q + 3)/2, containing no lines.
Assume that h(X, Y ) is a component of f0 , which can be written as h(X, Y ) =
e
e
0
g(X p , Y ) with gX
6 0. Then k pq+p
e +1 t + 1.
We omit the details now, the interested reader may see Proposition 3.6 in [37] for
all the ideas needed.
23
Stability
We start with a result of [113], which is a generalization of the main result of [124].
Let D be a set of directions in AG(2, q). A set U AG(2, q) is called a D-set if U
determines precisely the directions belonging to D.
Theorem 23.1. Let U be a D-set of AG(2, q) consisting of q points, where
q and |D| < (q + 1)(1 ), 1/2 1. Then U is incomplete, i.e. it can

be extended to a D-set Y with |Y | = q.
Proof: In the proof we follow the method of Section 11.
Let U = {(ai , bi ) : i = 1, ..., q }, suppose () D.
If (y) 6 D then the values ai y bi are all distinct elements of GF(q). Define the
Redei polynomial of U as
R(X, Y ) =
q
Y
q
X
i=1
j=0
(X + ai Y bi ) =
rj (Y )X qj .
Then deg(rj ) j. Let Ry (X) = R(X, y), then Ry |(X q X) if and only if the
elements in A(y) = {ai y + bi : i = 1, ..., q } are pairwise distinct, that is,
when (y) 6 D. (We define A(Y ), a set of linear polynomials, in the similar way.)
In this case let j = j (A(y)) be the j-th elementary symmetric polynomial of the
elements in A(y), and j = j (A(y)) = j (GF(q) \ A(y)) be the j-th elementary
symmetric polynomial of the remaining elements. Note that j = (1)j rj , and by
induction, using the recursive formula
j = j
j1
X
k jk
,
k=1
hence in principle j can be expressed by the i -s, so by the ri -s (i = 0, ..., j),

which means that j can be considered as a polynomial in Y . Now the recursive
formula shows that degY (j ) j as well. (See also Section 25 on nuclei.)
23. Stability
163
Define the polynomial f (X, Y ) = X 1 X 1 + 2 X 2 +... + (1) . Here

f is of total degree and if (y0 ) 6 D, then R(X, y0 )f (X, y0 ) = X q X. For such
y0 -s, we have
Y
f (X, y0 ) =
(X ),
GF(q)\Ay0
so the curve F defined by f (X, Y ) = 0 has precisely distinct simple points (x, y0 ).
So F has at least
N (q + 1 |D|) > (q + 1)
simple points in PG(2, q).
Now, using Lemma 10.15 with the same , we have that F has a linear component
X + aY b over GF(q). Then ay + b 6 Ay if (y) 6 D. Let U = U {(a, b)}, then
R (X, y) = R(X, y)(X + ay b) divides X q X for all (y) 6 D, as X q X =
R(X, y)f (X, y) and (X + aY b)|f (X, Y ). Hence U does not determine any
directions not in D, so U is also a D-set. Repeating this procedure we end up
with a D-set consisting of q points.
Comparing this to Theorem 18.14 one can see that if U AG(2, q) determines
N q+1
2 directions and U is of size q , with small then still we know the
structure of U .
The analogue of Theorem 23.1 is the following version of the results of [69, 47]:
Theorem 23.2. Let U AG(2, q) be a pointset of size |U | = q+, where q1

1
1
and 2 + 4q 1. Suppose that there are more than (q + 1) points on ` ,
through which every affine line contains at least one point of U ; let the complement
of this pointset on ` be called D. Then one can find points of U , such that
deleting them the remaining q points will still block all the affine lines through the
points of ` \ D.
Proof: Let U = {(ai , bi ) : i = 1, ..., q + }, suppose () D. Define the Redei
polynomial of U as
R(X, Y ) =
q+
Y
q+
X
i=1
j=0
(X + ai Y bi ) =
rj (Y )X q+j .
Then deg(rj ) j. Let Ry (X) = R(X, y), then (X q X)|Ry if and only if GF(q)
A(y) = {ai y + bi : i = 1, ..., q + } for the multiset A(y), that is, when (y) 6 D.
Similarly, let A(Y ) = {ai Y + bi : i = 1, ..., q + }, a set of linear polynomials. In
this case let j = j (A(y)) be the j-th elementary symmetric polynomial of the
elements in A(y), and
j =
j (A(y)) = j (A(y) \ GF(q)) be the j-th elementary
symmetric polynomial of the extra elements. Note that j = (1)j rj , and like
in Section 11, we have
j = j , and we can define
def
j (Y ) = j (Y ) = (1)j rj (Y ).
nuclei of U P
for some P`
164
Define the polynomial f (X, Y ) = X

1 X 1 +
2 X 2 +... + (1)
. Here
f is of total degree and if (y0 ) 6 D then R(X, y0 ) = (X q X)f (X, y0 ). For such
y0 -s, we have
Y
f (X, y0 ) =
(X ),
Ay0 \GF(q)
so the curve F defined by f (X, Y ) = 0 has precisely distinct simple points (x, y0 ).
So F has at least
N (q + 1 |D|) > (q + 1)
simple points in PG(2, q).
Now, using Lemma 10.15 with the same , we have that F has a linear component
X + aY b over GF(q). Then ay + b has multiplicity at least two in A(y) if
(y) 6 D. Now (X q X)(X + ay b) divides R(X, y) for all (y) 6 D, as R(X, y) =
(X q X)f (X, y) and (X + aY b)|f (X, Y ). Suppose that the point (a, b) 6 U .
Then counting the points of U on the lines connecting (a, b) to the points of ` \D,
we find at least 2|` \ D| q + 1 + points (at least 2 on each), a contradiction.
Hence (a, b) U , and one can delete (a, b) from U . Repeating this procedure we
end up with a set consisting of q points and still not determining any direction in
` \ D.
Usually it is difficult to prove that, when one finds the surplus element(s), then
they can be removed, i.e. they were there in the original set. Here the meaning
of a non-essential point (i.e. each line through it is an 2-secant) helped.
see Section 12
A = the incidence matrix of
the plane
Exercise 23.3. Prove that if S1 , S2 PG(2, q) are two pointsets, with characteristic
vectors vS1 , vS2 and weight (or line-intersection) vectors mS1 = AvS1 , mS2 =
AvS2 , then
||mS1 mS2 ||2 = ||vS1 vS2 || + q ||vS1 4 vS2 ||
pP
where ||x|| =
x2i and 4 denotes symmetric difference.
Note that it means that the (euclidean) distance of the line-intersection vectors of
any two pointsets is at least q.

The next sections are taken from Szonyi-Weiner [130], which is a collection of
wonderful stability results, most (but not all) of them based on the new resultant
method of the authors.
23.1
Stability theorems for small blocking sets and the Sz

onyiWeiner method
If we delete a few (say, ) points from a blocking set, then we get a point set
intersecting almost all except of at most a few (q) lines. A point set which is
close to be a blocking set is a point set that intersects almost all lines. A stability
23. Stability
165
type question asks whether all point sets having only a few 0-secants can be
obtained from a blocking set by deleting a few points of it. Of course, if we only
had a few 0-secants, then by adding one-one point to each of these lines (hence in
total still only a few points), we obtain a blocking set. So we have to be careful
by choosing sensible bounds in such stability theorems.
The next result is about the stability of the smallest blocking sets, that is, about
the lines.
Result 23.4. (Erd
os and Lov
asz, [61]) A point set of size q + 1 in PG(2, q), with
less than q q q skew lines always contains at least q q points from a line.
Note that this result is sharp, since if we delete q points from a Baer subplane,
then the resulting set will have q q q skew lines exactly and it has at most
q + 1 collinear points.
By the result of Blokhuis and Brouwer (Exercise 12.5) we know that through an
essential point of a blocking set B there pass at least 2q + 1 |B| tangents. This
result helps to estimate the number of 0-secants we get by deleting an essential
point from a small blocking set. So if we delete essential points from a small
blocking set then we get at least 2q skew lines. The main result of this section is
the following theorem.
Theorem 23.5. Let B be a point set in PG(2, q), q 81, of size less than 23 (q + 1).
Denote the number of 0-secants of B by , and assume that it is at most 13 q q,
q2
when |B| < q + q and at most 3(|B|q)
otherwise.
Then B can be obtained from a blocking set by deleting at most
2
q
points of it.
Note that the assumptions in this theorem are chosen such that
q
3(|B|q)
will both hold whenever |B| <
3
2 (q
1
3q q
and
+ 1).
When the size of B is large, the theorem above is immediate.

Lemma 23.6. Assume that for the size of B,
2
q
3(|B|q)
, q
2
most q points
Proof:
5
4q
|B| <
3
2 (q
+ 1) holds and
81. Then B can be obtained from a blocking set by deleting at

of it.
Since any two lines intersect, we can add at most
points to B so that we
2
q
get a blocking set B . This blocking set will have size at most |B| + 6(|B|q)
. This
5
3
5
function takes its maximum at |B| = 4 q when |B| is in the interval [ 4 q, 2 q + 1]
(and q 81). Hence |B 0 | 23
12 q and so by Result 12.5, through each of its essential
1
point there pass at least 12
q tangents. Since 43 q, this shows that indeed we
4
1
could have added at most 3 q/ 12
q = 16 points to B to obtain a blocking set of
3
size at most 2 q + 17. Again by Result 12.5, through each of its essential point
there pass at least 12 q 16 tangents, hence again Result 12.5 implies that indeed
0
166
we needed less than 5 points (here we use again that q 81). Repeating this
argument once more, we see that we need at most 2 ( 2
q ) points.
As the above lemma shows, Theorem 23.5 is weak when the size of B is in the
upper part of the given interval. The aim of the next two sections is to improve
on this result, when |B| is relatively large.
To prove Theorem 23.5 we need the following two lemmas.
Lemma 23.7. Let S be a point set of size less than 2q in PG(2, q), q 81, and
assume that the number of external lines of S is less than (q 2 q)/2.
(1) Denote by s the number of external lines of B passing through a point P .
Then (2q + 1 |S| s)s .
(2) If |S| 54 q and 31 qq, then through any point there are at most
1
2q+1|S| + 2 or at least 2q + 1 |S| 2q+1|S| 2 external lines to S.
Proof: Let ` be a line intersecting S in k > 0 points and let () S. Furthermore, assume that |S| k 6= q. This can be done, otherwise every line that
intersects S would intersect it in k points. Since |S| < 2q, counting the points
of S through a point in S, we get that k = 2. Hence S is a hyperoval, but this
contradicts our assumption on .
Assume that there is an ideal point different from () through that there pass
t affine lines intersecting S \ ` in at least 1 point. Denote by nt+h , the number
of ideal points different from () through P
that there pass (t + h) affine lines
qt
intersecting S \ ` . Hence by Lemma 12.6, h=1 hnt+h (|S| k t)(q t).
Suppose that P is a point of ` \ S and assume that through P there pass q t
affine lines not intersecting S \ ` . Denote by r(qt)h , the number of ideal points
different from () through that there pass (q t) h affine skew
lines. Through
Pqt
these points there pass t+h affine lines intersecting S\` , hence h=1 hr(qt)h =
Pqt
h=1 hnt+h and from above this is at most (|S| k t)(q t). Hence by counting
the number of skew lines through the points of ` \ S, we get a lower bound on
:
qt
X
(q + 1 k)(q t)
hr(qt)h .
(2.4)
h=1
Pqt
Pqt
Using
h=1 hr(qt)h =
h=1 hnt+h (|S| k t)(q t) and substituting
s = (q t), we get the first part of the lemma:
(2q + 1 |S| s)s .
(2.5)
To prove the second part of the lemma, we estimate the discriminant of (2.5) (from
2
below) by (2q + 1 |S|) ( 2q+1|S|
+ 1) (here we use the fact that 31 q q).
23. Stability
Hence s
2q+1|S|
167
+
1
2
or s 2q + 1 |S|
2q+1|S|
12 .
Lemma 23.8. Assume that |B| 54 q. Let N be the set of points through that there
are at least 2q + 1 |B| 2q+1|B|

12 external lines to B. Then B N is a
blocking set.
Proof: Assume to the contrary that there exists a line ` external to B, so that
12 external
through each point of ` there pass less than 2q + 1 |B| 2q+1|B|
lines of B. Then by Lemma 23.7 (2), through each of these points there pass at
most 2q+1|B|
+ 12 external lines (including `). Counting the external lines of B
through the points of `, we get an upper bound on .
1 + (q + 1)(
1
(|B| q)
q+1
)=+
+1
2q + 1 |B| 2
2q + 1 |B|
2
(2.6)
q
This is a contradiction, since 3(|B|q)
. Hence on each of the external lines,
there is at least one point through that there pass at least 2q+1|B| 2q+1|B|
12
external lines.
Proof of Theorem 23.5: By Lemma 23.6, we may assume that |B| 45 q. We

construct the point set B 0 , by adding the points through that there pass at least
2q + 1 |B| 2q+1|B|
12 external lines to B. Since 31 q q and |B| 54 q,
11
q skew lines through such points. Counting the external lines
there pass at least 36
11
11
through these points, we see at least 11
36 q + ( 36 q 1) + ( 36 q 2) + . . . 0-secants,
1
which shows that there were less than d 4 qe such points (here we use that q 81).
Hence |B 0 | < 3q/2. By Lemma 23.8, B 0 is a blocking set. Let be the minimum
number of points we need to add to B in order to obtain a blocking set B .
(|B | |B 0 | < 3q/2). By Result 12.5, through each essential point of this blocking
set (such are the points of B \ B) there pass at least 2q + 1 |B| q/2
tangents, i.e. external lines to B. Hence in total B has at least q/2 external lines,
which shows that 2/q.
23.2
Stability theorems for blocking sets: the prime case
In this section we will improve on the stability theorem of the previous section,
when the order of the plane is prime. Blokhuis ([33]) proved that a blocking set
with less than 32 (q + 1) points must contain a line. We are going to show that if B
is a point set with |B| 23 (q + 1) ... that has at most (q + 1) 0-secants, then it
contains a huge part of a line. Here can even be cq, where c is a small constant.
The proof is motivated by [33].
168
Lemma 23.9. Let B be a point set in PG(2, q), |B| < 32 (q + 1). Assume that there
are at most (q + 1) skew lines to B. Then the total number of 1-secants of B is
at least (q + 1)(2q |B| 2). Hence there is a point P of B so that there are at
least 32 (2q |B| 2) 1-secants through P .
Proof: Take a 0-secant ` of B. If there is no such line then B is a blocking set and
(by Result 12.5) through any essential point of B there pass at least 21 (q + 1) 1
tangents. Let the points of ` be denoted by P1 , . . . , Pq+1 and let i be the number
of 0-secants, i be the number of tangents through Pi . Looking at B from Pi one
gets that q (i + i ) (|B| i )/2, which implies that 2i + i 2q |B|.
Summing over all i we get that (q + 1)(2q |B|) 2(q + 1) + , from which
(q + 1)(2q |B| 2) follows. On the other hand, if we add up the number
of tangents at the points of B, we get , so there is a point which has at least the
average number of tangents.
p
Theorem 23.10. Let be the integer part of 2(q + 1) 1. Let B be a set of
points of PG(2, q), q = p prime, that has at most (q + 1) 0-secants for some
< 14 (q 6). Suppose that |B| < 23 (q + 1 ). Then there is a line that contains
at least q 2 points of B.
Proof: Choose the coordinate system in such a way that () is a point of B
with at least 23 (2q |B| 2) tangents, one of them being the line at infinity. Let
U = {(ai , bi ) : i = 1, . . . , |B|1} be the affine part of B. The 0-secants of B can be
written as Y = mj X + bj , j = 1, . . .. Consider the polynomial a(x, y) of smallest
degree , p
which vanishes at the points (bj , mj ), j = 1, . . . , (q + 1). By Exercise
10.2, 2(q + 1) 1. Now write up the polynomial
!
Y
R(X, Y ) =
(X + ai Y bi ) a(X, Y ).
i
The first product is the Redei polynomial of U . This polynomial R vanishes for
every (x, y), hence it can be written as
R(X, Y ) = (X q X)f (X, Y ) + (Y q Y )g(X, Y ),
where deg(f ), deg(g) |B| 1 q + . As in Blokhuis [33], consider the terms
of highest degree of this equation and substitute Y = 1 in it. Then we get a
polynomial equation
Y

r (X) =
(X + ai ) a (X) = X q f (X) + g (X),
where X q - g (X). We may suppose that f and g are coprime, since otherwise
we could divide by their greatest common divisor and obtain an equation of the
23. Stability
169
same type with smaller degrees. Denote by s the maximum of the degrees of f
and g after this division. Now we can continue copying Blokhuis proof. The roots
of r (X) in GF(q) are also roots of Xf (X) + g (X). The multiple roots of r (X)
in GF(q) are also roots of X q (f (X))0 + (g (X))0 . The roots not in GF(q) are roots
of a (x). Hence
r (X)|(Xf (X) + g (X))((f (X))0 g (X) (g (X))0 f (X))a (X).
(2.7)
If the polynomials on the right hand side of (2.7) are non-zero, then comparing
the degrees gives q + s s + 1 + 2s 2 + , that is s (q + 1 )/2. Since
s |B| 1 q + , it gives that |B| 23 (q + 1) 32 ; which is a contradiction.
The third term on the right hand side of (2.7) cannot be the zero polynomial, since
the terms of highest degree of a(X, Y ) formed a homogeneous polynomial and so
(Y 1) cannot be a factor of it.
If the first term on the right hand side of (2.7) is the zero polynomial then r (X)
is divisible by (X q X). Since a (X) has degree atQmost , the remaining q
factors of (X q X) must come from the product (X + ai ). Geometrically this
would imply that through the point () there passed at most + 1 tangents,
which contradicts the choice of (). (Here we use that + 1 < 32 (2q |B| 2).)
If the second term is zero, then, since f and g are coprime, f (X)|(f (X))0 and
similarly g (X)|(g (X))0 . Hence (f (X))0 = (g (X))0 = 0. For q = p prime, it
implies that either |B| 2q + 1 (which is not possible by our upper bound on
|B|) or aX q + b divides r (X). Since aX q + b = (aX + b)q , and at most of these
factors can come from a (X), it implies that there is a line ` (through ()) that
contains at least q+1 points of B. Finally, assume that |`B| = q+1k, k .
Then the 0-secants pass through the k missing points of `. Since |B| 32 q +1 32 ,
the number of 0-secants is at most k(q ( 32 q + 1 32 q 1 + k)) 12 k(q + 1).
Hence k 2.
Note that there is no restriction on how small can be. Chosing to be slightly
1
less than q+1
one gets Theorem 21.4 of Blokhuis.
Remark 23.11. (1) The condition < 14 (q 6) is not really necessary. Indeed,
fix the size of B. Determine the largest so that |B| < 32 (q + 1 ). Then
< 14 (q 6) will automatically hold if |B| > q/2 (a more exact bound comes
from |B| < 23 (q + 1 )). On the other hand, if a set B has size cq for some
c < 1 then the number of 0-secants is at least (1 c)q(q + 1). This can be seen by
counting 0-secants through points of a fixed 0-secant. Hence our theorem gives a
non-trivial bound only if (1 c)q(q + 1) < (q + 1), where |B| = cq = 32 (q + 1 ).
Roughly speaking this gives that for a fixed c < 1, the value of has to be smaller
than (1 32 c)2 q/2, which gives the equation 4c2 + 6c 9 = 0 for the critical
value of c. The positive root of the equation is (3 + 45)/4. Since it is bigger

than 12 , this shows that = (q 6)/4 is never reached. Let us underline that
170
the result gives

a non-trivial stability theorem also for sets of size cq, c < 1 (but
c > (3 + 45)/4 0.927).

(2) However, even when |B| = q + 1, one cannot expect only missing points.
Indeed, if we take an affinely regular -gon K, inscribed in an ellipse of AG(2, q),
and the set of q + 1 directions N not determined by K, then the number of
0-secants od B = K N is exactly (q ), so when q, then this is smaller

than q. For example, if = (q + 1)/2, then (q ) is roughly q/2. The bound on
in Theorem 23.10 does not allow this value, but it shows that the combinatorial
argument at the end of the proof cannot really be improved.
23.3
Stability theorems for blocking sets: the non-prime case
As mentioned before, in this section we are going to improve on Theorem 23.5,

when the size of our point set is relatively large. The proof is guided by the ideas of
the paper [123], where it is shown that each line intersects a small minimal blocking
set in 1 mod p points. In case of an almost blocking set B, a very similar argument
to those in [123] will show that B contains a point set intersecting almost every
line in 1 mod p points and so its size cannot be too large. Finally, using resultants
(see Section 9.5), we show that we can add a few points to B so that we obtain a
blocking set.
Theorem 23.12. Let B be a point set in PG(2, q), q = ph , h > 1, q 81, and
1
pq. Assume that
suppose that the number of 0-secants, , of B is at most 1000
(p+3)
3
|B| < 2 (q + 1) 2
2. Then B can be obtained from a blocking set by deleting
at most 2
q points of it.
q
Note that when |B| > q + 1000
3 p , then this theorem is an improvement on Theorem
23.5.
To prove the theorem our main aim is first to show that B can be embedded in a
blocking set of size less than 23 (q + 1), and then the result will follow immediately.
For the points (av , bv , cv ) of B, consider the three-variable Redei polynomial.
RB (X, Y, Z) =
|B|
Y
v=1
(cv X + av Y bv Z) =
|B|
X
rj (Y, Z)X |B|j .
(2.8)
j=0
Note that rj (Y, Z) is a homogeneous polynomial of degree j. As in the case of

the two variable Redei polynomial, this polynomial also encodes the intersection
multiplicities of B and the lines:
For a fixed (z, y, 0) `, the element x GF(q) is an r-fold root of R(X, y, z) if
and only if the line with equation zY = yX + xZ intersects U in exactly r points.
Hence the line wY = uX + vZ is a 0-secant of B, if and only if R(u, v, w) 6= 0.
The following lemma is similar to Exercise 10.2.
23. Stability
171
Lemma 23.13. There exists athree-variable homogeneous polynomial a

(X, Y, Z)
over GF(q), of degree at most 2 1, such that for each skew line wY = uX + vZ
to B, where w, u, v GF(q), a
(u, v, w) = 0.
Delete each linear component (ci X + ai Y bi Z) of a

and add the corresponding projective point (ai , bi , ci ), if it was not in B, to B. We will denotethe new
2 1.
polynomial by a
1 and the new set by B1 . Hence |B1 | < 3(q + 1)/2 p+1
2
First of all, note that if B1 has at most one 0-secant, then by putting one point
on this line, we obtain a blocking set of size less than 32 (q + 1) containing B. So
from now on we assume that B1 has at least two 0-secants, and we choose our new
coordinate system in such a way, that the lines Z = 0 and X = 0 are skew to B1 .
Note that, in this new coordinate system, B1 is an affine point set. Let a (X, Y, Z)
be the polynomial a
1 (X, Y, Z) in this new coordinate system. From now on we will
substitute Z = 1, hence we will consider the two variable (affine) Redei polynomial
RB1 (X, Y ) and the two variable polynomial a (X, Y ). By the construction of a ,
the polynomial RB1 a vanishes for all (x, y) GF(q) GF(q). If there is a factor
d(X, Y ) of a (X, Y ), so that (RB1 ad )(x, y) = 0 for every (x, y) pair, then we delete
this factor from a . We repeat this process until there is such factor.
Remark 23.14. Hence we
obtained a polynomial a(X, Y ), such thatt := dega =
deg
a (|B1 | |B|) 2 1, deg(RB1 a) < 3(q + 1)/2 p+1
2 1, and
2
a(X, Y ) has no linear component. Furthermore, RB1 (x, y)a(x, y) = 0 for every
pair (x, y) GF(q) GF(q), and a is minimal in the sense that this property will
not hold if we delete any factor of a.
Hence, we can write RB1 a as:
RB1 (X, Y )a(X, Y ) = (X q X)f (X, Y ) + (Y q Y )g(X, Y ),
p+1
2
(2.9)
where deg(f ), deg(g) |B1 | q+deg(a) = |B1 | q + t < (q + 1)/2

2.
Denote by C(X, Y ) the product of the common factors (with multiplicity) of
f (X, Y ) and g(X, Y ).
RB1 (X, Y )a(X, Y ) = C(X, Y )((X q X)f(X, Y ) + (Y q Y )
g (X, Y )),
(2.10)
where f and g have no common factors. Furthermore, deg(f), deg(

g
) and deg(C)
are at most |B1 | q+deg(a) = |B1 | q + t < (q + 1)/2 p+1
2. By the
2
minimality of a, the polynomials C and a cannot have a common factor, hence
C(X, Y )|RB1 (X, Y ).
Corollary 23.15. All the factors of C(X, Y ) have multiplicity one and are of form
(X + ak Y bk ). Furthermore, the points (ak , bk , 1) corresponding to the linear
factors of C(X, Y ) are in B1 .
172
Let C denote the set of these points, hence |C| =degX C. From above, C B1 .
Denote the point set B1 \C by B2 . Note that B2 is an affine point set and construct
the Redei polynomial RB2 (X, Y ) of B2 . Then:
RB2 (X, Y )a(X, Y ) = (X q X)f(X, Y ) + (Y q Y )
g (X, Y ),
(2.11)
As the next lemma shows, through each point of C, there pass a few 1-secants only.
Lemma 23.16. If (X +ak Y bk ) is a factor of C(X, Y ) then the number of 1-secants
of B1 through (ak , bk , 1) ( B1 ) is at most t + 1.
Proof: We have seen that (X + ak Y bk ) cannot divide a(X, Y ). Hence by
Bezouts theorem, a and (X + ak Y bk ) have at most t common points.
For any value y, x = bk ak y is an at least 2-fold root of the right hand side of
equation (2.10), and so this also holds for the left hand side. Hence from above,
there are at least q t values y, such that x = bk ak y is an at least 2-fold root of
RB1 (X, y). For these values, the line Y = yX + x (through (ak , bk , 1)) intersects
B1 in at least two points. Now the lemma follows, since the lines through (ak , bk , 1)
are either of type Y = yX + x or vertical.
Lemma 23.17. For any (x, y) GF(q) GF(q), if f(x, y) = 0, then g(x, y)a(x, y) =
0.
Proof: Suppose that for a fixed Y = y, x is a root of f(X, y). Then, by equation
(2.11), the intersection multiplicity of RB2 (X, Y )a(X, Y ) and the line Y = y in
(x, y) is at least two. Now assume that a(x, y) 6= 0, then by the properties of Redei
polynomials we have that the line Y = yX + x intersects B2 in at least two points.
Hence the intersection multiplicity of the line X = x and RB2 (X, Y ) in (x, y) is
also at least two, and so by equation (2.11), g(x, y) = 0.
Lemma 23.18. Let ri (X, Y ) be irreducible

polynomials over the algebraic closure
P
ri
of GF(q) dividing f. If X
6= 0 then i degX ri pq .
Proof: Let GF(q) denote the algebraic closure of GF(q), and assume that
degX ri = si . Then for any i, the sum of the intersection multiplicities I(P, ri `P )
over GF(q) GF(q) is exactly si q, where `P denotes the horizontal line through P .
First we show that the above sum over GF(q)GF(q) is less than si qsi pt. Assume
to the contrary that the sum of these intersection multiplicities is at least si qsi pt.
To get the number of points of ri over GF(q) GF(q) without multiplicities, we
ri
have to subtract the number of intersections of ri and X
. Hence by Bezouts
theorem, we get that ri has at least si q si pt si (si 1) GF(q)-rational points.
By Lemma 23.17, these points are also on ga. Now we show that ri 6 |
g a. Since f
23. Stability
173
and g have no common factors, if ri |

g a, then ri |a. Hence ri also divides the right
hand side of equation (2.11) and so it divides (Y q Y )
g (X, Y ). Again as f and g
q
have no common factors, ri |(Y Y ) and so ri = Y m, m GF(q); but this is not
possible since ri |a and a has no linear component. Hence we can apply Bezouts
theorem on the polynomials ri and ga:
q + 1 (p 1)
2).
2
2
2 si ; which is a
After simplifying the inequality, we get that (q + 1)/2 p+1
2
contradiction since si degx f < (q + 1)/2 p+1

2.
2
si q si pt si (si 1) si (
Hence the sum of the intersection multiplicities I(P, ri `P ) over (GF(q) \ GF(q))
GF(q) is at least si pt. For any fixed Y = y, RB2 (X, y) splits into linear factors,
hence f (X, y) can have at most t roots that are not from GF(q). So the sum of the
intersection multiplicities
I(P, f `P ) over (GF(q) \ GF(q)) GF(q) is at most tq,
P
from which pt i si tq follows.
Hence for a fixed Y = y, f(X, y) is almost a p-th power.
f(X, y) = (wy (X))p uy (X),
(2.12)
where degX uy q/p. Using equations (2.11) and (2.12), we get that for any
y GF(q),
RB2 (X, y)a(X, y) = (X q X)f(X, y) = (X q X)(wy (X))p uy (X).
(2.13)
As the next lemma shows, this equation helps to bound the size of B2 .
Lemma 23.19. The size of B2 is less than q + 7 pq .
Proof: We will estimate the sum S of the intersection multiplicities I(P, f rP ),
where rP are the horizontal lines through those points P GF(q) GF(q), where
these intersection multiplicities are at least p.
For a fixed value y GF(q), f(X, y) = (wy (X))p uy (X) splits into linear factors
over GF(q), hence the sum S of the above intersection multiplicities is at least
q(degX (f) maxy deg(uy )). By Lemma 23.18 and by equation (2.13), this is at
least q(|B2 | + t q q/p).
Now we will give an upper bound for S. First of all note that
I(P, f rP ) = I(P, RB2 rP ) + I(P, a rP ) 1,
for all points P . By Bezouts theorem, the sum of the intersection multiplicities
I(P, a rP ) over all points P GF(q) GF(q) is at most qdega = qt. To give an
174
upper bound for the sum of the intersection multiplicities I(P, RB2 rP ) (when
for each P , I(P, f rP ) p), we will distinguish between the points P according
as I(P, RB2 rP ) (p + 2)/2 and I(P, RB2 rP ) < (p + 2)/2.
By the fundamental properties of the Redei polynomial, the first case considers
those points P (x, y) that correspond to lines Y = yX +x intersecting B2 in at least
(p+2)/2 points. Hence for these points P , the sum of the intersection multiplicities
I(P, RB2 rP ) is exactly the number of incident point-line pairs, where the point
lies in B2 and the line is a (p + 2)/2-secant. The number of (p + 2)/2-secants
through a point of B2 is at most 2(|B2 | 1)/p, hence the number of these incident
point-line pairs is at most 2|B2 |(|B2 | 1)/p.
Now we bound the sum of the intersection multiplicities I(P, RB2 rP ) over the
points P , for that I(P, RB2 rP ) < (p + 2)/2 (and I(P, f rP ) p). For these
points, I(P, a rP ) p/2 and so I(P, a f) p/2. Hence by Bezouts theorem the
number of such points P is at most (degfdega)/(p/2). For each of these points P ,
I(P, RB2 rP ) (p + 1)/2, hence the sum of these intersection multiplicities is at
p+1
most p+1
p degf dega = p (|B2 | q)t. Hence:
q
2|B2 |(|B2 | 1) p + 1
q(|B2 | + t q ) S
+
(|B2 | q)t + qt.
p
p
p
Subtract q(|B2 | + t q pq ) from both sides of the above inequality.
0
q
2|B2 |(|B2 | 1) p + 1
+
(|B2 | q)t q(|B2 | q )
p
p
p
When q = ph , h > 2, (using that |B2 |

that |B2 | < q + 7 pq .
3
2 q),
this inequality yields immediately
So now we only consider the case q = p2 . For |B2 |, the right hand side of the
above inequality is a quadratic expression which is positive when |B2 | = q and
negative when |B2 | = q + 7 pq (the lemma is only interesting if q + 7 pq < 23 q (since
|B2 | 32 q and so we may suppose that p > 13). Divide the above inequality by
2
p , so the coefficient of the quadratic term is 1 and the constant term is ((p +
1)qt + q 2 p + q 2 )/2. It follows from the previous argument that one of the roots of
this quadratic expression is less than q + 7 pq < 32 q; hence the other is larger than
((p + 1)qt + q 2 p + q 2 )/3q. This is larger than 23 q, when p > 13 and since |B2 | 32 q,
we get that |B2 | < q + 7 pq .
Lemma 23.20. The number of 0-secants of B2 , 0 , is at most
1
40 q pq.
1
Proof: Originally B had at most 1000
pq 0-secants. To obtain B2 we added some
points and deleted the points of C. By Lemma
23.16, after deleting the points
q
of C, we may get at most |C|(t + 1)
q
2
2
1000 pq
new 0-secants. Hence 0
23. Stability
q
2
2
1000 pq
175
1
1000 pq
<q
1
2000 pq
1
+ q 1000
pq.
Lemma 23.21. The number of 0-secants of B2 through a point is either at most

0
1
0
1
2q+1|B2 | + 50 p or at least 2q + 1 |B2 | ( 2q+1|B2 | + 50 p).
Proof: By Lemma 23.7, if through a point P there pass s external lines of B2 ,
then (2q + 1 |B2 | s)s 0 . Estimate the discriminant (from below) by
(2q + 1 |B2 |) (
2 0
2
+ p).
2q + 1 |B2 | 50
Note that, 2q + 1 |B2 | 0.5q, and so we get that s

s 2q + 1 |B2 |
( 2q+1|B
2|
(2.14)
0
2q+1|B2 |
1
50 p
or
1
50 p).
Proof of Theorem 23.12. First we show that B can be embedded in a blocking set
of size less than 32 (q + 1).
Construct the point set B1 (see the beginning of this section). As we have already
noticed there, if B1 has only at most one 0-secant then adding one point to B1 ,
we get a blocking set of size less than 23 (q + 1). Otherwise construct the point set
B2 .
By Lemma 23.19, B2 has less than q + 7 pq points. Hence if the number of 0-secants
1
0 of B2 (which is at least the number of 0-secants of B) is at most 21
pq, then
Theorem 23.5 finishes our proof.
1
Finally, we show that 0 is indeed smaller than 21
pq. Assume to the contrary
1
0
that > 21 pq. By the construction of B2 , all the 0-secants of B2 that are not
1
skew to B pass through one of the points of C. There are at least 0 1000
pq
such 0-secants, hence there exists a point P C through that there pass at least
pq
0
|C| 1000|C| 0-secants of B2 . We show that this contradicts Lemma 23.21. To this,
note that |B2 | 39
0-secants through the
40 q. Otherwise counting the number of
1 2
1
points of a skew line to B2 , we get more than 40
q 40
q pq skew lines; which is
a contradiction by Lemma 23.20.
pq
0
0
1
First we show that |C|
1000|C|
> 2q+1|B
+ 50
p. We may assume, that 2q +
2|
41
1
1 |B2 | dq, where 40 d > 2 . Recall that B1 = B2 C and |B1 | 32 q, so
0
p
0
1
> dq
|C| (d 21 )q. Hence it is enough to show that (d 1 )q 1000(d
+ 50
p.
1
)
2
1/2
1
1
1
1
Now 0 > 21
pq; hence we need to see that 21
( d(d
> 50
. After
1 )
1000(d 12 )
2)
differentiating the left hand side of the above inequality, one can see that it is
41
decreasing in the interval ( 12 , 41
40 ]. So for d = 40 , it takes its minimum, which is
1
larger than 50 .
To get a contradiction with Lemma 23.21, now we only have to show that the
number of 0-secants of B2 through any point of C is less than 2q + 1 |B2 |
176

0
( 2q+1|B
+
2|
1
50 p).
By Lemma 23.16, there are at most t + 1 <
through any point of C. Using Lemma 23.20 and that |B2 |

1
0
+ 50
p).
t + 1 < 2q + 1 |B2 | ( 2q+1|B
2|
2pq
1000 0-secants
3
2 q, we get the
2
Hence by Theorem 23.5, we can add at most 21
p < p points to B2 , so that we
obtain ablocking set. Note that we constructed B2 from B by adding at most
deg
a<
2 points to B and deleting some points. Hence in total we can add at
most 2 + p points to B in order to get a blocking set. This blocking set will
have size less than 23 (q + 1) (here we assume that > 1, see the beginning of this
section). Hence similarly as at the end of the proof of Theorem 23.5, we get that
2
q .
23.4
Stability theorem for sets of even type
A set of even type U is a point set intersecting each line in even number of points.
By counting the points of S on the lines through a point of S and on the lines
through a point not in S, one can see immediately that q must be even. Hence
from now on in this section we will assume that 2|q. The smallest sets of even
type are the hyperovals, they have q + 2 points. All other examples known to
the authors have at least q + q points; see Korchmaros, Mazzocca [88]. They
constructed (q + t)-sets of type (0, 2, t) for t q. More examples and a proof

that the t-secants are concurrent for such sets can be found in Gacs, Weiner [70].
It is easy to see that the symmetric difference of any two sets of even type is a set
of even type.
A set of almost even type is a point set having only a few odd-secants. If we delete a
point from a set of even type U then the new point set will have q + 1 odd-secants,
and of course this will be also the case when we add a point to U . If we modify
points then we will get at least (q + 1 ( 1)) and at most (q + 1) odd-secants.
In this section we are going to prove the following stability theorem.
Theorem 23.22. Assume that the point set H in PG(2, q), 16 < q even, has
odd-secants, where < (b qc + 1)(q + 1 b qc). Then there exists a unique set
H0 of even type, such that |(H H0 ) \ (H H0 )| = d q+1

e.
Remark 23.23. Note that a complete arc of size q q +1 has ( q +1)(q +1 q)

odd-secants, which shows that Theorem 23.22 is sharp, when q is a square. (Since
the smallest sets of even type are hyperovals.) For the existence of such arcs see
[48], [63] and [85].
Let ` be the line at infinity intersecting the point set M in even number of
points, furthermore let M \ ` = {(av , bv )}v and M ` = {(yi )}i . Consider the
23. Stability
177
following polynomial:
|M\` |
g(X, Y ) =
(X+av Y bv )q1 +
v=1
(Y yi )q1 +|M| =
yi M`
q1
X
ri (Y )X q1i ,
i=0
(2.15)
Note that degri i.
Lemma 23.24. Assume that the line at infinity contains even number of points of
M. Through a point (y) there pass s odd-secants of M if and only if the degree of
the greatest common divisor of g(X, y) and X q X is q s.
Proof: To prove this lemma we only have to show that x is a root of g(X, y) if
and only if the line Y = yX + x intersects M in even number of points. To this,
for the (x, y) pairs, one has to count the parity of zero and non-zero terms.
Remark 23.25. Assume that the line at infinity is an even-secant and suppose also
that there is an ideal point, different from (), with s odd-secants. Let nh denote
the number of ideal points different from (), through that there pass s h oddsecants
Ps1 of the point set M. Then Lemma 23.24 and Corollary 12.6 imply that
h=1 hnh s(s 1).
Lemma 23.26. Let M be a point set in PG(2, q), 16 < q even, having less than
(b qc + 1)(q + 1 b qc) odd-secants. Then the number of odd-secants through any
point is at most b qc + 1 or at least q b qc.

Proof: Pick a point P with s odd-secants and let ` be an even-secant of M
through P . (If there was no even-secant, then the lemma follows immediately.) By
Remark 23.25, counting the number of odd-secants through ` \ (), we get that:
qs s(s 1) .
This is a quadratic inequality for s, where the discriminant is larger then q + 1
2(b qc + 2). Hence s q + 1 (b qc + 2).
Proposition 23.27. Let M be a point set in PG(2, q), 16 < q even, having less than
(b qc + 1)(q + 1 b qc) odd-secants. Assume that through each point there pass at
most 3b qc odd-secants. Then the total number of odd-secants of M is at most
b qcq q + 2b qc + 1.
Proof: Assume to the contrary that > b qcq q + 2b qc + 1. Pick a point

P and let ` be an even-secant of M through P . Assume that there are s oddsecants through P . If there is a point Q on this even-secant through which there
178
pass at least s odd-secants, then choose the coordinate system so that Q is ().
Then by Remark 23.25, counting the number of odd-secants through `, we get a
lower bound on :
(q + 1)s s(s 1) .
Since < (b qc + 1)(q + 1 b qc), from the above inequality we get that s q + 1 b qc, but by the assumption of the

proposition the latter case cannot occur.
Now we show that through each point there are at most b qc odd-secants. The
above argument and Lemma 23.26 show that on each even-secant there is at most
one point through that there pass b qc + 1 odd-secants and through the rest of
the points there are at most b qc of them. Assume that there is a point R with
b qc + 1 odd-secants. Since > b qc + 1, we can find an odd-secant ` not through

R. From above, the number of odd-secants through the intersection point of an
even-secant on R and ` is at most b qc. So counting the odd-secants through the
points of `, we get at most (q b qc)(b qc 1) + (b qc + 1)b qc + 1, which is
a contradiction, so there was no point with b qc + 1 odd-secants.
This means that the odd-secants form a dual b qc-arc, hence (b qc 1)(q +
1) + 1, which is a contradiction again; whence the proof follows.
Proof of Theorem 23.22 By Lemma 23.26, through each point there pass either
at most b qc + 1 or at least q b qc odd-secants. Consider the points through
that there pass at least q b qc odd-secants. If such a point was in H then delete
it, otherwise add it to H. Denote this new set by H0 . Note that the number of
modified points is less than 2b qc, hence through each point there pass at most
3b qc odd-secants of H0 . Denote by 0 , the number of odd-secants of H0 . Note

that when we modify (delete or add) a point then each odd-secant through that
point will become an even-secant and vice-versa. Hence, if through a point there
passed s q b qc odd-secants of H, then at least s 3b qc of them will be an
even secant of H0 . So through this point there will pass at most q + 1 (s 3b qc)
new odd-secants. Since, when q > 16, s q + 1 (s 3b qc), the number of odd
secants 0 of H0 is at most . By Proposition 23.27, 0 b qcq q + 2b qc + 1.

Our first aim is to show that H0 is a set of even type.
Let P be an arbitrary point with s odd-secants, and let ` be an even-secant
through P . Assume that there is a point on ` with at least s odd-secant through
it. Then as in Proposition 23.27, counting the number of odd-secants through `,
we get a lower bound on 0 :
(q + 1)s s(s 1) 0 .
This is a quadratic inequality for s, where the discriminant is larger then (q + 2
0
0
0
+q
+q
+q
2 q+1
). Hence s < q+1
or s > q + 2 q+1
, but by the construction of H0 , the
latter case cannot occur.
23. Stability
179
Now we show that there is no point through that there pass at least
0 +q
q+1
odd-
0 +q
q+1
secants. On the contrary, assume that T is a point with

s odd-secants. We
choose our coordinate system, so that the ideal line is an even-secant through T
and T 6= (). Then from above, through each ideal point, there pass less than
0
+q
) odd-secants. First we show that there exists an ideal point through that
s( q+1
there pass exactly (s 1) odd-secants. Otherwise by Remark 23.25, 2(q 1)
s(s 1); but this is a contradiction since by Lemma 23.26, s b qc + 1. Let ()

be a point with (s 1) odd-secants. Then as before, we can give a lower bound on
the total number of odd-secants of H0 :
(s 1) + qs s(s 1) 0
0
+q
), it follows that s <
Bounding the discriminant (from below) by (q +22 q+1
0 +q
q+1
+q
. This is a contradiction, since by assumption, the latter case
or s > q + 2 q+1
cannot occur and the first case contradicts our choice for T .
0
+q
Hence through each point there pass less than q+1
odd-secants. Assume that ` is
0
an odd-secant of H . Then summing up the odd-secants through the points of `
0
1
we get that 0 < (q + 1) q+1
+ 1, which is a contradiction. So H0 is a set of even
type.
e. As
To finish our proof we only have to show that |(H H0 ) \ (H H0 )| = d q+1
we saw at the beginning of this proof, the number of modified points is smaller
than 2b qc. On one hand, if we construct H from the set H0 of even type, then
we see that (q + 1 ( 1)). Solving the quadratic inequality we get that
 q + 1 b qc, but from above this latter case cannot happen.
On the other hand, (q + 1). From this
and
the previous inequality (and from
b qc(b qc1)
. Hence the theorem follows.

b qc), we get that q+1 q+1 +
q+1
For applications of Theorem 23.22 see the next section and also Section 17.2.
23.5
Number of lines meeting a (q + 2)-set
Stability of hyperovals was already studied by Blokhuis and Bruen. A hyperoval is

not only nice in the sense that it intersects each line in even number of points, but
it is also extremal in the sense that considering point
sets of size q + 2, a hyperoval
intersects the least number of lines, that is q+2
. The next result shows that a
2
(q + 2)-set intersecting a bit more than q+2
lines
can always be constructed by
2
modifying a hyperoval a little bit.
Result 23.28. (Blokhuis, Bruen [44]) Let H be a point set in PG(2, q), q even,
of size
q + 2. Assumeq that the number of lines meeting H in at least 1 point is
q+2
+ , where 2 . Then H is a hyperoval or there exist two points P and Q,
2
so that (H \ P ) Q is a hyperoval.
180
Now we will try to improve on this result. Hence we will consider (q + 2)-sets
intersecting a bit more than q+2
+ 2q lines.
2
Theorem 23.29. Let H be a point set in PG(2, q), 16 < q even, of size q + 2.
Assume that the number of lines meeting H in at least 1 point is q+2
+ , where
2
< 41 (b qc + 1)(q + 1 b qc). Then there exists a set H of even type, such that
4
|(H H ) \ (H H )| = d q+1
e.
Proof: First we show that almost all lines meeting H intersect it in 2 points. Let
li denote the number of lines intersecting H in i points. We will do the standard
counting arguments to get a lower bound on l2 . That is:
q+1
X
i=1
q+1
X

li =

q+2
+
2
(2.16)
ili = (q + 2)(q + 1)
(2.17)
i(i 1)li = (q + 2)(q + 1)
(2.18)
i=1
q+1
X
i=2
Calculating ((2.17)-(2.16)), and then subtracting it twice from (2.18), we get

q+1
X

(i 1)li =
i=2
q+1
X

q+2
(i 2)(i 1)li = 2
(2.19)
(2.20)
i=3
Pq+1
Hence from (2.20), we
have that
i=3 (i 1)li 2 and so using (2.19), we

have that l2 q+2
3.
Hence
the
total number of odd-secants is at most

2
q+2
q+2
23.22 finishes our proof.
+
3)
=
4.
So
Theorem
2
2
24
Jamison, Brouwer-Schrijver, affine blocking sets
Here we reconstruct the short proof of Brouwer and Schrijver [49] for the theorem
on (1-)blocking sets in AG(n, q). It was proved independently and in a more general
form by Jamison [84].
Theorem 24.1. In AG(n, q) a blocking set B with respect to hyperplanes has at least
|B| n(q 1) + 1 points.
24. Jamison, Brouwer-Schrijver, affine blocking sets
181
Note that the bound is sharp: n concurrent lines form such a blocking set (but
there are several other examples as well, no classification of the minimal blocking
sets can be hoped for).
Proof: We can assume that the origin 0 = (0, 0, ..., 0) is in B, let B 0 = B \{0}. B 0
blocks all hyperplanes not through 0, hence, as any such hyperplane has equation
of form w1 X1 + w2 X2 + ... + wn Xn = 1 for some w1 , ..., wn , not all zero, we have
that the polynomial
Y
(b1 Y1 + b2 Y2 + ... + bn Yn 1)
F (Y1 , ..., Yn ) =
bB 0
vanishes at every (w1 , ..., wn ) 6= 0. It means that for each i we have (Yiq1
Qn
1)|F (Y1 , ..., Yn ), so i=1 (Yiq1 1)|F (Y1 , ..., Yn ). As F (0) 6= 0, its degree, so |B 0 |,
is at least n(q 1).
After removing the origin from B (w.l.o.g. it was in there), the remaining points
block all hyperplanes not through 0. After dualisation, as Jamison did, one can
formulate Theorem 24.1 like:
Theorem 24.2. In AG(n, q) if a set B of nonzero hyperplanes cover AG(n, q) \ {0}
then it has at least |B| n(q 1) hyperplanes.
In fact Jamison proved the following more general result:

Theorem 24.3. Any covering of AG(n, q) \ {0} with k-dimensional affine subspaces
not through the origin, contains at least q nk 1 + k(q 1) subspaces; where
0 k < n. Moreover, this bound is always sharp.
Proof: Suppose to the contrary that S is a family of nonzero k-subspaces covering
AG(n, q) \ {0} and |S| < q nk 1 + k(q 1). We represent AG(n, q) as GF(q n ).
Let p(X) GF(q n )[X] be the product of all root polynomials pL (X) as L ranges
though the k-subspaces of S. As p(x) = 0 for all x GF(q n ) , we have
p(X) = (X q
1)t(X)
(1)
for some t(X). Let aj be the coefficient of the term of degree j(q n 1) in p(X)
and bj the coefficient of the term of degree j(q n 1) in t(X). From (1) we obtain
the recursive relations
a0 = b0
and
aj = bj1 bj for j > 0.
(2)
Since 0 is not covered, 0 6= p(0) = a0 so b0 6= 0. We want to show that aj = 0
for j 1. From this and (2), bj = b0 6= 0 for all j, which is a contradiction as a
polynomial can have finitely many nonzero terms only.
nonzero
planes
hyper-
182
By |S| < q nk 1 + k(q 1) the degree of p(X) is strictly less than

q k (q nk 1 + k(q 1)) (k + 1)q n (k + 1)q k < (k + 1)(q n 1);
whence aj = 0 for all j k + 1. We are going to prove aj = 0 for 1 j k.
As p(X) is the product of root polynomials (each a linearized polynomial plus a
constant term), we deduce that if aj 6= 0 then
j(q 1) = c0 + c1 q + ... + ck q k ,
(3)
c0 + ... + ck |S| < q nk 1 + k(q 1).
(4)
where each ci 0 and
is for constant
factors
from
some
root
polynomial
We want to prove thatP

such an expansion (3), (4) cannot hold. Suppose that it
exists, choose one with i ci minimal. Then it is easy to see that
ci q 1 for 0 i k 1.
(5)
As all terms in (3) are nonnegative,

ck jq nk 1,
(6)
since otherwise jq nk q k > j(q n 1).

P
Suppose j = 1. If both in (5) and (6) equalities hold then i ci = k(q1)+q nk 1,
contradicting (4). So one of the inequalities must be strict. Consequently,
k
X
ci q i < (q 1) + (q 1)q + ... + (q 1)q k1 + (q nk 1)q k = q n 1,
i=0
contradicting (3).
Suppose j > 1. Let m be the least integer such that j q m . By induction on j it
is easy to see that j q j1 , hence 0 < m j 1 k 1. Considering (3) modulo
q m we obtain
q m j c0 + c1 q + ... + cm1 q m1
(mod q m ).
As m 1 k 1 we can apply (5) here and conclude that the sum on the right
hand side lies between 0 and q m 1, so it is not only a congruence but actually
an equality. Thus we may write
j(q n 1) = q m j + cm q m + ... + ck q k .
If ci = q 1 when m i k 1 and ck = jq nk 1, then
cm + ... + ck = (k m)(q 1) + jq nk 1 (k m)(q 1) + q nk 1 + mq nk
183
> (k m)(q 1) + q nk 1 + m(q 1) = q nk 1 + k(q 1).

This contradicts (4), so either one of the inequalities in (5) must be strict for some
i between m and k 1, or inequality (6) must be strict. In either case
c0 +c1 q +...+ck q k < q m j +(q 1)q m +...+(q 1)q k1 +(jq nk 1)q k = jq n j,
and this contradicts (3). Therefore the representation (3) is impossible if |S|
q nk 1 + k(q 1). Consequently, aj = 0 for all j > 0.
Note that even if |S| = q nk 1 + k(q 1), then aj is nonetheless 0 for all j > 1.
Of course, since a covering does exist in this case, a1 may be nonzero. Therefore
a1 is the critical coefficient.
For k = n 1 Bruen [53] generalized it in the following way for t-covers (i.e. when
each point is covered at least t times):
Theorem 24.4. Any t-covering of AG(n, q) \ {0} with hyperplanes not through the
origin, contains at least (n + t 1)(q 1) hyperplanes. Dually:
Any t-fold blocking set of AG(n, q) (with respect to the hyperplanes) contains at
least (n + t 1)(q 1) + 1 points.
Proof: Let B be a set of t-covering hyperplanes, defined by the linear equations
Q|B|
fi = 0, i = 1, ..., |B|. Then i=1 fi is a polynomial not vanishing at the origin but
vanishing t times elsewhere. Now Exercise 5.12 finishes the proof. For the dual
statement suppose w.l.o.g. that the origin is contained in the blocking set, and use
the standard duality in PG(n, q) containing AG(n, q).
We mention without proof that Bruen generalized Theorem 24.3 for t-coverings:
(x0 , x1 , ..., xn ) 7
[x0 , x1 , ..., xn ]
which is similar
to Jamisons
Theorem 24.5. Any t-covering of AG(n, q)\{0} with k-dimensional affine subspaces
not through the origin, contains at least
tq nk 1 +
k1
X
bi
i=0
subspaces, where q k t =
Pk1
i=0
bi q i , 0 bi < q, kt q k .
Exercise 24.6. [53] Let be a dual translation plane of order q 2 , with (dual) kernel
of order q. Then any blocking set of contains at least q 2 + 2(q 1) points.
See also Theorem 27.1.
affine plane
184
24.1
Applications of the punctured Nullstellensatz
In this section, which is based on [22], we show applications to the punctured

Nullstellensatz by Ball and Serra (see Section 5.5).
Consider two lines `1 and `2 of a projective plane over a field F and finite nonintersecting subsets of points Si of ì . Let A be a set of points with the property
that every line joining a point of S1 to a point of S2 is incident with a point of A.
If we asked ourselves how small can A be then obviously we could simply choose
A to be the smaller of the Si and clearly we can do no better. If, however, we
impose the restriction that one of the lines joining a point P1 of S1 to a point
P2 of S2 is not incident with any point of A then it is not so obvious how small
can A can be. We will see that at least |S1 | + |S2 | 2 points are needed, which is
clearly an attainable bound, for example take A to be (S1 S2 ) \ {P1 , P2 }. The
next theorem provides the bound to arbitrary dimension and to sets that have not
just one point incident with the lines joining a point of S1 to a point of S2 , but a
fixed number t of points.
Let F be an arbitrary field.
Theorem 24.7. Let `1 , `2 , ..., `n be n concurrent lines spanning PG(n, F) and let
Si be a subset of points of ì for each i not containing the point of intersection
of the lines ì . Let A be a set of points with the property that every hyperplane
hs1 , s2 , ..., sn i, where si Si is incident with at least t points of A except those
hyperplanes hd1 , d2 , ..., dn i, where di is an element of a subset Di of Si for all
i, which may be incident with less than t points of A. If there is a hyperplane
hd1 , d2 , ..., dn i, where di is an element of a subset Di for all i, which is incident
with no point of A, then for all j
|A| (t 1)(|Sj | |Dj |) +
n
X
(|Si | |Di |).
i=1
Proof: Let H be a hyperplane that meets the lines ì in a point of Si but is

not incident with any point of A. Apply a collineation of PG(n, F) that takes
`1 , `2 , ..., `n to the axes of AG(n, F), the affine space obtained from PG(n, F) by
removing the hyperplane H, and takes the point H ì to the point hei i, where
i
ei = (0, ..., 0, 1, 0, ..., 0).

We can then assume that A is a subset of AG(n, F), the affine space obtained from
P G(n, F) by removing the hyperplane H. The hyperplane H is defined by the
equation Xn+1 = 0.
Let 0 Ti be the subset of F with the property that s1 Ti \ {0} if and only if
hsei + en+1 i is a point of Si . Note that the line ì , after applying the collineation,
is hei , en+1 i and |Ti | = |Si |. Let 0 Ei be the subset of F with the property that
185
d1 Ei \ {0} if and only if hdei + en+1 i is a point of Di . Define

!
!
n
Y
X
f (X1 , X2 , ..., Xn ) =
ai Xi 1 .
aA
i=1
Pn
The affine hyperplanes i=1 ti Xi = 1, where ti Ti are not all zero, are the affine
hyperplanes spanned by points s1 , s2 , ..., sn , where si Si . By hypothesis there
are t points of A incident with these hyperplanes, unless ti Ei for all i, and so
f has a zero of multiplicity t at (t1 , t2 , ..., tn ), unless ti Ei for all i.
However 0 Ei for all i and F (0, 0, ..., 0) = (1)|A| , so there is an element of
T1 T2 ... Tn where f does not vanish. Theorem 5.24 implies that for all j
n
n
X
X
|A| = deg(f ) (t1)(|Tj ||Ej |)+ (|Ti ||Ei |) = (t1)(|Sj ||Dj |)+ (|Si ||Di |).
i=1
i=1
The above theorem also holds for any multi-set A.

The condition that there is a hyperplane that is not incident with a point of A is
essential. If we do not impose this condition then for t < n any collection of t of
the sets Si would be appropriate for A.
Sn
For t = 1 the bound is tight: take A = i=1 (Si \ Di ).
Theorem 24.7 has some corollaries. The following theorem, which, after Theorem
24.7, can be formulated as an exercise, is due to Bruen [53]. It was initially proven
for t = 1 by Jamison (Theorem 24.1); see also the independent proof found by
Brouwer and Schrijver [49].
It is nice to prove Theorem 24.4 again, using Theorem 24.7:
Exercise 24.8. If every hyperplane of AG(n, q) is incident with at least t points of
a set of points A, then A has at least (n + t 1)(q 1) + 1 points.
The bound in this theorem can be improved slightly in many cases when t q as
was proven in [31]. In Theorem 24.11 we shall investigate when this improvement
applies to the more general Theorem 24.10 below.
Firstly let us look at a consequence of Theorem 24.7 for projections.
Theorem 24.9. If there are m 1 points x1 , x2 , ..., xm1 of PG(n, F) that project
m collinear points S1 onto m collinear points S2 then there is a further point xm
which also projects S1 onto S2 .
Proof: Suppose that there is no such point xm which also projects S1 onto S2 .
There are m lines `1 , ..., `m that join a point of S1 to a point of S2 but are not
incident with any of the points x1 , x2 , ..., xm1 . The points of S1 and S2 are all
186
contained in the same plane and so any two lines ì and `j are incident. If they
are not all incident with a common point xm then we can choose m 2 points
y1 , ..., ym2 such that yi is incident with ì but is not incident with `m and ym2 is
the intersection of the lines `m2 and `m1 . (We may have to relabel the lines to
ensure that `m is not incident with the intersection of the lines `m2 and `m1 .)
The set A = {x1 , x2 , ..., xm1 , y1 , ..., ym2 } has the property that every line, except
`m , that joins a point of S1 to a point of S2 is incident with a point of A, which
contradicts Theorem 24.7.
The following theorem is almost the dual of Theorem 24.7. The proof is shorter
as here we do not need an initial transformation.
Theorem 24.10. Let A be a set of hyperplanes of AG(n, F) and let Di be a nonempty proper subset of Si , a finite subset of F. If every point (s1 , s2 , ..., sn ), where
si Si , is incident with at least t hyperplanes of A except at least one point of
D1 D2 ... Dn , which is incident with no hyperplane of A, then for all j
|A| (t 1)(|Sj | |Dj |) +
n
X
(|Si | |Di |).
i=1
Proof:
Define
f (X1 , X2 , ..., Xn ) =
n
X
!
ai Xi
!
an+1
i=1
where each
Pnfactor in the product corresponds to a hyperplane, defined by the
equation i=1 ai Xi = an+1 , in A. By hypothesis the polynomial f has a zero of
multiplicity t at all the points of S1 S2 ... Sn except at least one point of
D1 D2 ... Dn where it is not zero. By Theorem 5.24 the bound follows.
If Si = GF(q) and Di = {0} then Theorem 24.10 implies that a set of hyperplanes
A with the property that every non-origin point of AG(n, q) is incident with at
least t hyperplanes of A is a set of at least (n + t 1)(q 1) hyperplanes, which
dualising gives Bruens theorem (Exercise 24.8) again.
We end this section by proving the following theorem which is similar to Theorem
24.10 but in which there are translations of the set of hyperplanes of AG(n, q), not
incident with the origin, which also cover most, but not all, of the points of the
grid S1 ... Sn .
For any Fn and A, a finite subset of Fn , define A + = {a + |a A}.
In the following a punctured grid is a set of points (S1 ... Sn ) \ (D1 ... Dn ),
where Di is a proper non-empty subset of some subset Si F.
187
Theorem 24.11. Let be a set of vectors of Fn with the property that if

and only if there is a finite punctured grid G , punctured at the origin and whose
dimensions do not depend on , with the property that every point of G is incident
with at least t hyperplanes defined by equations of the form
b1 X1 + ... + bn Xn = 1,
for some b A + . Let m be minimal such that for all
Y
(X s) = 1 + X m r (X)
sS1 \D1
for some polynomial r , where G is the punctured grid (S1 ... Sn ) \ (D1 ...
Dn ). If there are non-negative integers j and k with the property that either
k j min{t 1, m 1, |{1 |(1 , ..., n ) (A)}| 1}
or
k + 1 j min{t 1, m 1, |{1 |(1 , ..., n ) }| 1},
and
Pn
i=1 (|Si |
then
|A|
n
X
|Di |) + (t 1)(|S1 | |D1 |) + k

j

6= 0
(|Si | |Di |) + (t 1)(|S1 | |D1 |) + k + 1.
i=1
Proof:
Suppose that |A| =
Pn
i=1 (|Si |
|Di |) + (t 1)(|S1 | |D1 |) + k. Let

!
!
n
Y
X
f (X1 , X2 , ..., Xn ) =
(ai + i )Xi 1 .
aA
i=1
The degree of f is |A| 1 + , where = 0 if = a for some a A and = 1 if

not. By hypothesis the polynomial f has a zero of multiplicity t at all the points
of S1 S2 ... Sn except at least one point of D1 D2 ... Dn where it is
non-zero. Let
Y
Y
gi (Xi ) =
(Xi si ) and li (Xi ) =
(Xi di ).
si Si
di Di
By Theorem 5.24
f (X1 , ..., Xn ) =
X
T
g (1) ...g (t) h + u1
n
Y
gi
l
i=1 i
188
for some polynomial u1 of degree at most (t1)(|S1 ||D1 |)+k 1+. Since there
is a point, the origin, of D1 D2 ... Dn where f is non-zero, the polynomial
u1 is non-zero at this point. The polynomial in one variable
f (X, 0, ..., 0) = g1t h + u2
g1
,
l1
for some polynomial h and polynomial u2 of degree at most deg u1 .

By hypothesis f has a zero of multiplicity t at all the points (s1 , 0, ..., 0), where
s1 S1 \ D1 . Therefore f (X, 0, ..., 0) is divisible by
t
g1
= (1 + X m r (X))t .
l1
Let A1 be the multiset where a1 appears as an element of A1 the number of times
it appears as the first coordinate of an element of A. Thus
f (X, 0, ..., 0) = g1t h + u2
g1
,
l1
can be written as
Y
((a1 + 1 )X 1) = (1 + X m r )t (u3 + hl1t ),
a1 A1
for some polynomial u3 of degree at most deg u2 (t 1)(|S1 | |D1 |), which is
at most k 1 + . Since 0 D1 we can write hl1t = X t h2 for some polynomial
h2 . Thus, the coefficient of X j in the right hand side of the equation above is zero
for all j for which k + j min{m 1, t 1}. On the left-hand side of this
equation the coefficient of X j is a polynomial in 1 of degree
at most j where the
term j1 , if it appears in the polynomial, has coefficient |A|
j .
If the number of 1 which appear as first coordinate in the vectors in is more
than j, then the coefficient of X j , which is a polynomial in 1 of degree at most
j, must be identically zero, and so the binomial coefficient is zero.
The following corollary is a slight generalisation of a result of Blokhuis (Theorem
2.2) from [31], see Section 25 as well for this and the next one.
Corollary 24.12. A set A of points of AG(n, q) with the property that every hyperplane is incident with at least t points of A, has size at least
(n + t 1)(q 1) + k + 1,
provided that there exists a j with the property that k j min{t 1, q 2} and

n t + k + 1
6= 0.
j
189
Proof: Apply Theorem 24.11 with = A, Si = GF(q) and Di = {0} for all
i = 1, 2, ..., n. Note that for all i,
Y
(X s) = X q1 1,
sSi \Di
so m = q1 and that |{1 |(1 , ..., n ) (A)}| = |{a1 |(a1 , ..., an ) A}| = q.
We conclude that if there are non-negative integers j and k with the property that
k j min{t 1, q 2} and

(n + t 1)(q 1) + k
n t + k + 1
=
6= 0,
j
j
then |A| (n + t 1)(q 1) + k + 1.
The following corollary is from Blokhuis [31] for (t, q) = 1 and [8] in general.
Exercise 24.13. A set of points of AG(2, q) with the property that every line is
incident with at least t points of A has size at least (t + 1)q (t, q).
24.2
Redei polynomials in three indeterminates
The next two sections come from Ball-Lavrauw [21].

In most of the applications of Redei polynomials we have a set of points with
certain intersection properties with hyperplanes. Now we shall look at how to use
Redei polynomials in situations where we have a set of points that have certain
intersection properties with smaller dimensional subspaces. The obvious place to
start is in three dimensional affine or projective space, so let A be a pointset of
AG(3, q) and define
Y
R(T, S, U ) =
(T + xS + yU + z).
(x,y,z)A
We define polynomials j in two indeterminates, of total degree at most j, by

writing
|A|
X
R(T, S, U ) =
j (S, U )T |A|j .
j=0
Let a and b be any elements of GF(q) and consider the factors of the polynomial
Y
R(T, aU + b, U ) =
(T + (ax + y)U + bx + z).
(x,y,z)A
The point (x, y, z) is incident with the line defined by the hyperplanes aX +Y =
and bX + Z = if and only if ax + y = and bx + z = if and only if the
190
corresponding factor in R(T, aU + b, U ) is T + U + . Note that for a fixed

a and b these q 2 lines are parallel. Identifying the affine point (x, y, z) with the
projective point (x, y, z, 1) then the line defined by the (hyper)planes aX + Y =
and bX + Z = is incident with point (1, a, b, 0). Indeed, every affine line
incident with (1, a, b, 0) can be defined by the intersection of two planes of the
form aX + Y = and bX + Z = , for some and , and this line is incident
with k points of A if and only if T + U + occurs as a factor of R(T, aU + b, U )
with multiplicity k. Let us fix a situation to see how we can make use of this.
Let A be a set of q 2 points that does not determine every direction, so we can
assume there is a point (1, a, b, 0) whose corresponding parallel class of q 2 lines
are each incident with exactly one point of A. Each factor in R(T, aU + b, U ) is
distinct and we have
Y
2
R(T, aU +b, U ) =
(T +U +) = T q ((U q U )q1 +1)T q +(U q U )q1 T.
,GF(q)
However, we also have that

2
R(T, aU + b, U ) =
q
X
j (aU + b, U )T |A|j
j=0
and hence for 1 j q 2 q 1, j (aU + b, U ) 0.

So we have that j (S, U ) 0 mod S aU b from which it follows that
S aU b | j (S, U ).
If there are more pairs (a, b) than the degree of j for which this holds then
j (S, U ) 0. So in our situation let N be the number of directions (1, a, b, 0)
not determined by two points of A. Then
R(T, S, U ) = T
q2
2
qX
N
q2 j (S, U )T j .
j=0
When we evaluate both S = s and U = u we are looking at the intersection

properties of A with planes. Indeed, a factor T + has multiplicity k in R(T, s, u)
if and only if the plane sX + uY + Z = is incident with k points of A. However,
2
R(T, s, u) = T q + g(T ), where deg g q 2 N , and moreover factorises into linear
factors in GF(q)[T ]. In [20] Ball and Lavrauw prove the following
Theorem 24.14. If A is a set of q 2 points of AG(3, q) that does not determine at
least pe q directions for some e then every plane meets A in 0 mod pe+1 points.
By Theorem 18.15 of Storme and Sziklai we know that if the number of directions
determined is less than q(q + 3)/2 then the set A is GF(s)-linear for some subfield
191
GF(s) of GF(q). In the case when q is prime this means that A is a plane. Theorem
24.14 says that in the case when q is prime and we have a set of points that
determine less than q 2 + 2 directions then every plane contains 0 mod q points.
It would be interesting to know if this implies that A is a cone. Some immediate
corollaries of Theorem 24.14 relate to ovoids of the generalised quadrangles T2 (O)
or T2 (O).
Corollary 24.15. [20] Let O be an ovoid of the generalised quadrangle T2 (O). The
planes of AG(3, q) are incident with zero mod p points of O.
Corollary 24.16. [20] Let O be an ovoid of the generalised quadrangle T2 (O) containing the point (1). The planes of AG(3, q) are incident with zero mod p points
of O.
In the case when O is a conic the generalised quadrangle T2 (O) is isomorphic to
Q(4, q). Since we can choose any point of Q(4, q) to be the point (1), the corollary
says that every elliptic quadric meets an ovoid of Q(4, q) in 1 mod p points or
not at all. (See also Theorem 28.2 in Section 28.2) In fact one can eliminate the
possibility that an elliptic quadric and an ovoid are disjoint but that does involve
slightly more work, see [18]. A short counting argument leads to the following
theorem.
Theorem 24.17. When q is prime an ovoid of Q(4, q) is an elliptic quadric.
Although the Redei polynomial considered in [18] is the same as we have used
here, they used a more roundabout argument involving the Klein correspondence
to deduce the fact that for nearly all j the identity j (aU, U a) 0 holds for all
a GF(q). Now we see that from j (aU +b, U ) 0, one only needs to substitute U
with U a and put b = a2 to obtain the same equivalence. The previous theorem
can be pushed a little further. De Beule and Metsch recently proved the following.
Theorem 24.18. When q is prime a set of q 2 + 2 points that blocks every line of
Q(4, q) is an elliptic quadric together with a point.
24.3
Redei polynomials in many indeterminates
Let us prove the natural generalisation of Theorem 24.14 to AG(n, q) by using

Redei polynomials in n variables.
Theorem 24.19. If A is a set of q n1 points of AG(n, q) that does not determine
at least pe q directions for some e then every hyperplane meets A in 0 mod pe+1
points.
Proof: Let A AG(n, q) = PG(n, q) \ where is defined by the equation
X0 = 0. Let D be the set of directions determined by A, i.e.
D = {P |P hQ, Ri, Q, R A}.
192
Define the Redei polynomial as

R(T, X) =
n
X
T+

ai Xi ,
i=1
(a1 ,...,an )A
where X = (X1 , ..., Xn ), and define the polynomials j (X) by writing

n1
qX
R(T, X) =
Tq
n1
j (X).
j=0
Note that the total degree of j is at most j. Let P = (0, y1 , y2 , ..., yn ) \ D

be a direction not determined by A and since P must have a non-zero coordinate,
we may assume that
Pn ym = 1, for some m. Now consider the Redei polynomial
R(T, X) modulo i=1 yi Xi . We get

R(T, X)
T+
n
X
(ai am yi )Xi
mod
i=1
(a1 ,...,an )A
n
X
yi X i .
i=1
For all GF(q)n1 , the line defined by the n 1 equations Xi yi Xm = i X0 ,

where i 6= m, contains the point P , which is a direction not determined by A, and
so contains exactly one point of A. Hence there is exactly one (1, a1 , ..., an ) A
such that ai am yi = i for all i 6= m and we have

R(T, X)
T+
n
X
i Xi
mod
yi X i .
i=1
i=1
i6=m
GF(q)n1
n
X
The above polynomial is linear over GF(q) in T and so

R(T, X) T q
n1
n2
X
qn1 qk (X)T q + qn1 (X) mod
n
X
yi X i
i=1
k=0
Pn
n1
and we conclude that j (X) 0 mod
and
i=1 yi Xi , whenever j 6= q
n1
k
q for some k. Hence for each P = (0, y1 , y2 , ..., yn ) \ D
j 6= q
n
X
yi Xi | j (X)
i=1
in these cases and so j (X) 0 whenever deg(j ) j < | \ D|. By hypothesis

| \ D| pe q and so we can write
R(T, X) = T
q n1
n1
qX
j=pe q
j (X)T q
n1
193
Now let x GF(q)n be any vector in n coordinates and let d be maximal such
d
d+1
that R(T, x) GF(q)[T p ] \ GF(q)[T p ]. We wish to prove that d e + 1. Write
d
R(T, x) = S(T )p , so S GF(q)[T ]\GF(q)[T p ] and importantly T S 6= 0. Moreover
S(T ) = T q
n1
/pd
+ S1 (T ),
where deg(S1 ) q n1 /pd qped . Since S(T ) is the product of linear factors
S(T ) | (T q T )T S.
The degree of the right-hand side of this divisibility is less than q + deg(S1 )
q + q n1 /pd qped . If d e, the degree of the left-hand side, which is the degree
of S and equal to q n1 /pd , will be greater than the degree of right-hand side and
we conclude that the right-hand side of the divisibility must be zero. However this
implies that T S is zero, a contradiction, hence d e + 1.
We have shown that for all x GF(q)n ,
R(T, x) =
Y
(1,a1 ,...,an )A
T+
n
X
ai xi
e+1
GF(q)[T p
i=1
e+1
and so all its factors T + occur with multiplicity
Pn a multiple of p . Hence there
e+1
are a multiple of p
points of A satisfying i=1 ai xi = , or, in other words,
e+1
there
are
0
modulo
p
points of A on the hyperplane defined by the equation
Pn
x
X
=
.
This
concludes
the proof.
i=1 i i
Now the following improvement to Theorem 24.19 is given with a shorter and
simpler proof from [11]. The main difference here is the use of GF(q) GF(q n1 )
as a model for AG(n, q) in place of the natural GF(q)n .
Theorem 24.20. Let q = ph and 1 pe < q n2 . If there are more than pe (q
1) directions not determined by a set S of q n1 points in AG(n, q) then every
hyperplane meets S in 0 modulo pe+1 points.
In [11] sets of points that reach or almost reach the bound (for some special values
of e) are constructed.
If f is a function in n 1 variables over the finite field GF(q) then the set of
points S = {(x, f (x))|x GF(q)n1 } is a set of q n1 points of AG(n, q) and the
set of directions determined by the function f is defined to be the set of directions
determined by S. We call S the graph of f , for obvious reasons. Note that for any
set S of q n1 points which does not determine all the directions, by applying an
affine transformation so that the projective point (0, ..., 0, 1) is a non-determined
direction, we can construct a function whose graph is the set S.
Proof: Let us consider the points of AG(n, q) as a set of elements of GF(q)
GF(q n1 ) and S a set of q n1 points. We can assume, after making an affine
194
transformation if necessary, that the hyperplane with first coordinate zero does not
contain exactly q n2 points of S. Then the set of directions not determined by S
consists of projective points (1, m) where m runs through some set N GF(q n1 ).
Let x = (x1 , x2 ) and y = (y1 , y2 ) be two points of AG(n, q) with the property that
x1 6= y1 and there is an element GF(q n1 ) such that x1 x2 = y1 n y2 .
Then, (x1 y1 ) = x2 y2 and the projective point (x1 y1 , x2 y2 ) = (1, ).
Thus, (1, ) is the direction determined by the points x and y.
For all m N the projective point (1, m) is not determined by S and so the set
{x1 mx2 | x = (x1 , x2 ) S} consists of distinct elements, and hence all elements,
of GF(q n1 ).
Define a polynomial r in two variables and polynomials j , in one variable of
degree at most j(q 1), by
r(X, Y ) =
q1
(X (x1 Y x2 )
)=
n1
qX
Now r has the property that for all m N ,

Y
r(X, m) =
(X (x1 m x2 )q1 ) =
(X q1 ) =
GF(q n1 )
xS
= X(X (q
n1
j=0
xS
n1
j (Y )X q
1)/(q1)
1)q1 = X q
+ Xq
n1
(q n1 1)/(q1)
+ ... + X.
n1
Thus, for all m N and 1 j < (q

1)/(q 1) we have j (m) = 0. However,
|N | > pe (q 1) deg(j ) for all j pe and since a polynomial has at most as
many zeros as its degree, j is identically zero for all 1 j pe .
Let cj be coefficient of the term of degree j(q 1) in the polynomial j (Y ). Then
Y
(X
xq1
)
1
n1
qX
cj X q
n1
= Xq
n1
+ cpe +1 X q
n1
pe 1
+ ... + cpn1 .
j=0
xS
Now x1 GF(q) and so xq1

= 1 unless x1 = 0. If we write t for the number of
1
points of x S with first coordinate zero then
Y
xS
(X
xq1
)
1
q n1 t
= X (X 1)
n1
qX
j=0

n1
t
(1)j X q j .
j

Comparing the above two equations we see that t
pr = 0 mod p for all 0 r e
and so t = 0 mod pe + 1, by Lucas Theorem. Thus the number of points of S
incident with the hyperplane with first coordinate zero is 0 modulo pe +1. However,
this hyperplane was chosen arbitrarily among the hyperplanes that do not contain
exactly q n2 points of S. Hence, every hyperplane is incident with 0 modulo pe+1
points of S.
195
Corollary 24.21. Let q = ph and 1 pe q t < q n2 , where e, t Z. If there are more

than pe q t (q 1) directions not determined by a set S of q n1 points in AG(n, q)
then every (n 1 s)-dimensional subspace, where 0 s t, meets S in 0 modulo
pe+1 q ts points.
Proof: Let n1s be a (n 1 s)-dimensional subspace which does not contain
q n2s points of S.
We first prove that there is a hyperplane containing n1s which does not contain
exactly q n2 points of S. So we assume s 1 since otherwise there is nothing to
prove. If every subspace of dimension ns which contains n1s contains exactly
q ns1 points of S then
|S| = q n1 = n + (q ns1 n)(q s+1 1)/(q 1),
where n is the number of points of S contained in ns1 . This implies that
n = q n1 which by assumption it is not. Therefore we can find a chain of subspaces
ns1 ns ... n1 with the property that |S \ n1 | 6= q n2 . Thus
n1 meets the subspace at infinity in a subspace consisting entirely of determined
directions.
There are (q n q n1 )/(q ns q ns1 ) = q s subspaces of dimension n s that
contain n1s and are not contained in n1 and so one of them meets the subspace at infinity in a subspace that contains at least pe q ts (q 1) non-determined
directions. Applying Theorem 24.20 to this subspace, the corollary follows.
24.4
Affine blocking sets
This part is from Ball [8].

The following proposition will be analysed later to provide lower bounds on the
sizes of multiple blocking sets with respect to hyperplanes in AG(n, q). For the
moment we satisfy ourselves with proving that for t-fold blocking sets of AG(n, q),
which have at most (n+t1)q n points, a set of linear equations has a nontrivial
solution. The degree of a polynomial f will be denoted f .
Proposition 24.22. Let S, a set of (t + n 1)(q 1) + k 1 points in AG(n, q),
be a t-fold blocking set with respect to hyperplanes. Then there exist i(q1) with
0 = 1 such that
bj/(q1)c
X
i=0

nq q n + k t i(q 1)
i(q1) = 0 (mod p)
j i(q 1)
for all j such that k 1 j < t.
196
Proof: Let S, a set of (t + n 1)(q 1) + k 1 points in AG(n, q), be a t-fold

blocking set with respect to hyperplanes and assume 0 S. Define
Y
F (X1 , ..., Xn ) =
(s1 X1 + ... + sn Xn + 1),
sS
then F = |S| 1 = (t + n 1)(q 1) + k 2. Write F (X) = W (X) + V (X),

where V Jt and all terms divisible by X1c1 q X2c2 q Xncn q with c1 + ... + cn = t
appear only in V . Note that W (0) 6= 0 as F (0) 6= 0. By hypothesis we have that
X1t F Jt and hence that X1t W Jt . Moreover, considering the possible terms
that can appear in X1t W it follows that X1t W = (X1q X1 )W1 where W1 Jt1 .
Continuing for X2 , ..., Xn we see that we can write
W (X) = U (X)
n
Y
(Xiq1 1),
i=1
where U (t 1)(q 1) + k 2 and U (0) 6= 0.

On every hyperplane X1 = c there exist at least t points of S and therefore
(X1q1 1)t divides
Y
F (X1 , 0, ..., 0) = (X1q1 1)U (X1 , 0, ..., 0) + (X1q X1 )t V (X1 ) =
(s1 X1 + 1),
sS
where V (X1 , 0, ..., 0) = (X1q X1 )t V (X1 ). Let S be a multi-subset of the multiset

S [1] := {s1 |(s1 , ..., sn ) S} of size |S| tq such that the multi-set S [1] \ S contains
each element of GF(q) repeated exactly t times. Define the symmetric functions
j by
nqqn+kt
Y
X
(aX + 1) =
j X j .
aS
j=0
Put X1 = X, s1 = a and F (X, 0, ..., 0) = (X q1 1)t f (X), let V (X, 0, ..., 0) =

v(X) and U (X, 0, ..., 0) = (X q1 1)t1 u(X) and rewrite the above as
X
f (X) = u(X) + X t v(X) =
j X j ,
where u k 2. The coefficient of X j implies that j = 0 whenever k 1 j < t.
For all GF(q) the set S = {(s1 + , s2 , ..., sn )|(s1 , s2 , ..., sn ) S} is a t-fold
blocking set with respect to hyperplanes of AG(n, q) and we can make a shift of
()
the other n 1 coordinates to ensure that 0 is in S . Let j denote the j-th
then it follows that () = 0
symmetric function of the set S = {a + |a S},
j
whenever k 1 j < t. Now
()
j
(a1 + )(a2 + ) (aj + ) =
j
X
r
|S|
r=0
jr
r jr ,
197
where the first sum extends over all possible combinations of a1 +, a2 +, ..., aj +
in S . Hence for k 1 j < t
k1
X
r=0

nq q n + k t r
r jr = 0 (mod p).
jr
This is zero when reduced modulo q since it is valid for all GF(q).
Therefore
bj/(q1)c
X
i=0

nq q n + k t i(q 1) r
i(q1)+r = 0 (mod p)
j i(q 1) r
where 0 r < q 1. We are specifically interested in getting equations involving

0 , since this is non-zero and so we conclude that for k 1 j < t
bj/(q1)c
X
i=0

nq q n + k t i(q 1)
i(q1) = 0 (mod p).
j i(q 1)
Theorem 24.23. For t < q a t-fold blocking set with respect to hyperplanes in
AG(n, q) has at least (t + n 1)(q 1) + k points provided
there exists a j such

that k 1 j < t and the binomial coefficient knt
6= 0 (mod p).
j
Proof: In fact this is an almost trivial consequence of Proposition 24.22. Assume
that there is a t-fold blocking set with (t+n1)(q1)+k1 points. Since t < q we
have that j < q 1 and there is only one term in the sum. Lucas theorem allows
us to ignore the terms
topof the binomial coefficient divisible by q, and we
in the
knt
deduce that knt
=
= 0 (mod p), which gives a contradiction if a j
0
j
j
in the condition of the theorem exists.
Corollary 24.24. For t < q a t-fold blocking set with respect to hyperplanes in
n
AG(n, q) has at least (t + n 1)q n + 1 points provided t1
6= 0 (mod p).
Corollary 24.25. For t < q a t-fold blocking set with respect to hyperplanes in
AG(n, q) has at least (t + 1)q pe(t) points (where e(t) is maximal such that pe(t)
divides t).
Let k = t pe(t) + 1 and j = t 1, write t = t0 pe(t) where p 6 |t0 . Then

e(t)
+pe(t) 1
p
1
0 = (t02p
= t02
1 (mod p), which is non-zero.
t1
1)pe(t) +pe(t) 1
Proof:
e(t)
Now we are well prepared for the next section on nuclei and affine blocking sets
as well.
198
25
Nuclei and affine blocking sets
The theory of nuclei shows a nice application of the GF(q 2 )-representation.

Definition 25.1. P 6 B is a t-fold nucleus of B PG(2, q) if all the lines through
P intersect B in at least t points.
N (B)
A 1-fold nucleus is called nucleus simply. The motivation of this notion is that
the nucleus of an oval in a plane of even order is a nucleus in this sense as well.
Obviously, if B has a nucleus then |B| t(q + 1). The set of nuclei is sometimes
denoted by N (B).
It is also interesting to dualise this concept. Given L, a set of lines in PG(2, q), a
line ` 6 L is a t-fold nuclear line of it if the lines of L cover each point of ` at least
t times. It means that if you consider L as a totally reducible algebraic curve, then
` intersects it in each point (of `) with multiplicity at least t.
Theorem 25.2. If B PG(2, q), |B| = q + 1 and it is not a line then it has at most
q 1 nuclei.
Proof 1: There exists a line disjoint from B so w.l.o.g. B AG(2, q) and no infinite
point canPbe a nucleus. Using the GF(q 2 )-representation of the affine plane, define
f (X) = bB (X b)q1 . If x is a nucleus then the set of (x b)q1 is exactly
the complete set of (q + 1)-st roots of unity, so their sum is 0, hence x is a root of
f , which is of degree q 1.
Proof 2: Consider the following polynomial defined by B:
X
F (X, Y, Z) =
(b1 X + b2 Y + b3 Z)q1 ;
(b1 ,b2 ,b3 )B
we will prove that if (a1 , a2 , a3 ) is a nucleus of B then (a1 X + a2 Y +

a3 Z)|F (X, Y, Z). Then it follows that
Y
(a1 X + a2 Y + a3 Z) | F (X, Y, Z),
(a1 ,a2 ,a3 )N (B)
which is of degree q 1.
Take a line [v1 , v2 , v3 ] meeting B in precisely one point. Then F (v1 , v2 , v3 ) = 0 as
of the q + 1 terms of the sum one is zero and the other q are all 1.
Now w.l.o.g. assume that a1 6= 0 and write F (X, Y, Z) = (a1 X + a2 Y +
a3 Z)g(X, Y, Z) + G(Y, Z) with G(Y, Z) homogeneous of degree q 1 or identi3w
is a line through (a1 , a2 , a3 ) and
cally zero. Since [u, v, w] with u = a2 v+a
a1
hence contains a unique point of B, so F (u, v, w) = 0 and a1 u + a2 v + a3 w = 0.
Hence G(v, w) = 0 for all v, w GF(q). So G cannot have degree q 1 hence it is
25. Nuclei and affine blocking sets
199
identically zero.
Proof 3: This proof uses the Segre-trick:
Lemma 25.3. (Segre, Korchm
aros) Let B PG(2, q) a pointset of size q + 1 and
A1 , A2 , A3 three non-collinear nuclei of B. Write Bi = Aj Ak B, where {i, j, k} =
{1, 2, 3}. Then the three points B1 , B2 , B3 B are collinear.
Proof of the lemma Let A1 , A2 , A3 be the base points of the coordinate frame,
let B1 = (0, 1, b1 ), B2 = (b2 , 0, 1), B3 = (1, b3 , 0), where b1 b2 b3 6= 0. Apply the
Segre-trick for the lines joining Ai to to points of B \ {B1 , B2 , B3 }, so build a
matrix of size (q 2) 3, with rows indexed by the points of B \ {B1 , B2 , B3 }
and with columns indexed by A1 , A2 and A3 . The entry in position (P, Ai ) is
the coordinate of the line P Ai , so if the equation of the line is X3 = 1 X2
or X1 = 2 X3 or X2 = 3 X1 then this i will appear in the matrix. By Cevas
theorem, the product of the three elements in any row is 1. On the other hand,
in each column all but one elements of GF(q) appears, the missing ones are just
the coordinates of the lines Ai Bi . As the product of the elements of GF(q) is 1
by Wilsons theorem, the (product of all the entries) times b1 b2 b3 is on the one
hand = b1 b2 b3 , on the other hand = (1)3 . So b1 b2 b3 = 1, hence by Menelaos
theorem B1 , B2 and B3 are collinear.
Now we are ready for Proof 3 (which is due to Blokhuis and Mazzocca): Again
we can restrict ourselves to AG(2, q). Define a map f : N (B) GF(q) in the
following way: take an arbitrary P N (B) and put f (P ) = 1. If Pi N (B),
let Bi = P Pi B (which is unique). Now Bi = i P + (1 i )Pi for a suitable
i
i GF(q) \ {1}. Then put f (Pi ) = i1
. Denote the affine coordinates like
1 2
1 2
1 2
P (p , p ), Pi (pi , pi ) and Bi (bi , bi ). It is easy to check that in AG(3, q) the points
(p1 , p2 , f (P )), (p1i , p2i , f (Pi )) and (b1i , b2i , 0) are collinear.
The Segre-Korchm
aros lemma implies that if B 0 B is on the line Pi Pj then
0
Bi , Bj , B and hence (p1i , p2i , f (Pi )), (p1j , p2j , f (Pj )) and (b01 , b02 , 0) are collinear. In
particular f is injective, so |N (B)| q 1.
A more general form of it is the following
Theorem 25.4. (Blokhuis) If B AG(2, q), |B| = t(q + 1) + k 1, k < q, then the
number of its t-fold nuclei is at most k(q 1), supposing that t+k1
6 0 (mod p)
k
holds.
Proof: As in the previous proof, after identifying AG(2, q) and GF(q 2 ), define the
polynomial
Y
F (X, T ) =
(T (X b)q1 ).
bB
lift to
3-space!
200
P
(Note that
(X b)q1 , used in the previous proof, is just the coefficient of
t(q+1)+k2
T
.) For a t-fold nucleus x, the multiset {(x b)q1 : b B} contains
every (q + 1)-st root of unity at least t times, so the polynomial F (x, T ) in one
variable is divisible by (T q+1 1)t for each t-fold nucleus x.
Let j (X) denote the j-th elementary symmetric polynomial of the terms (X

b)q1 . As a polynomial of X, its degree is j(q 1), with equality if |B|
is
j
nonzero in GF(q); for j = k this is the condition in the statement. As F (X, T ) =
P|B|
j
|B|j
, in case of X = x being a t-fold nucleus, we have
j=0 (1) j (X)T
F (x, T ) = (T q+1 1)t (T k1 + ...).
On the right hand side the coefficient of T t(q+1)1 is zero as k < q, on the left hand
side this coefficient is (1)k k (x). So every t-fold nucleus x is a root of k (X),
which is of degree k(q 1).
A nice refinement and re-frasing is the following

Theorem 25.5. (Ball) If B is a t-fold blocking set of AG(2, q), e(t) is the maximal
exponent for which pe(t) |t then |B| (t + 1)q pe(t) .
Jamison
For e(t) = 0, i.e. when g.c.d.(t, q) = 1, it was proved by Blokhuis. In particular,

when t = 1 it is Jamisons result 12.2.
Proof: First lets see the connection between nuclei and blocking sets. If B is a
t-fold blocking set then all the points in its complement are t-fold nuclei of B.
Suppose to the contrary that B is a blocking set of size (t + 1)q pe(t) 1 =
t(q + 1) + (q t pe(t) ) 1, so by the previous theorem (k = q t pe(t) ) it has
at most (q t pe(t) )(q 1) nuclei, which is less then the size
of the complement
of B, so B cannot be a t-fold blocking set provided t+k1
6= 0 modulo p. Write
k
t = pe(t) c where p 6 |c. Using Lucas theorem and calculating modulo p we have

t(q + 1) + k 1
t(q + 1) 1

=

tq + q pe(t) 1
=
tq + t 1

e(t)

q pe(t) 1
q 2pe(t) + pe(t) 1
q/p
2 pe(t) 1
=
=
,
cpe(t) 1
(c 1)pe(t) + pe(t) 1
c1
pe(t) 1
which is non-zero mod p as, using base p, each digit of q/pe(t) 2 is (p 1) except
the last one which is (p 2), but the last digit of c 1 cannot be p 1 as p 6 |c.
Corollary 25.6. If K is a (k, n)-arc of AG(2, q), and e is the maximal exponent
such that pe |n then |K| (n 1)q + pe .
201
Proof: Let B be the complement of K in AG(2, q). It is a t = (q n)-fold blocking

set and the previous theorem gives the bound.
Lunelli and Sce conjectured that the size of a (k, n)-arc K in PG(2, q) is k
(n 1)q + 1. It was disproved it by Hill and Mason, their counterexample is the
disjoint union of some Baer subplanes; but if g.c.d.(n, q)=1 and there exists a line
skew to K then the previous Corollary shows that the Lunelli-Sce bound is true.
25.1
Lower nuclei
Now suppose that B AG(2, q), |B| = t(q + 1) k + 1, and we want to investigate
its t-fold lower nuclei, so the points from which B looks like a (|B|, t)-arc. The
result is quite similar to the case of nuclei:
Theorem 25.7. If B AG(2, q), |B| = t(q + 1) k + 1, then the number of its

t-fold nuclei is at most k(q 1), supposing that t(q+1)
6 0 (mod p) holds.
k
As we did before, lets identify AG(2, q) and GF(q 2 ), so B GF(q 2 ). Let E be
the multiset of (q + 1)-st roots of unity in GF(q 2 ), each with multiplicity t. Let
B(X) = {(X b1 )q1 , (X b2 )q1 , ..., (X b|B| )q1 } where B = {b1 , ..., b|B| }.
Then the point represented by x GF(q 2 ) is a t-fold (lower) nucleus of B if and
only if it is not in B and B(x) E as a multiset.
First we prove two lemmas.
Lemma 25.8. Let S E, |S| = t(q + 1) r, i = i (S), i = i (E \ S). Then
r
X
i r+1i = 0.
i=0
Q
Q
Q
Proof: (X q+1 1)t = aE (X a) = bS (X b) (E\S) (X ) =
Pt(q+1)r
Pr
(1)i i X t(q+1)ri j=0 (1)j j X rj . The coefficient of X t(q+1)u in
i=0
this expression, u = 1, 2, ..., q 2, is
r
X
i ui = 0.
(u)
i=0
From (u), (u = 1, ..., r), u can be expressed as

u
u1
X
i=0
i ui .
(u )
202
Now the next, (r + 1)-st equality gives the desired relation among the elementary
symmetric polynomials, as no more i can appear (note that the upper limit is r
instead of r + 1):
r
X
i r+1i .
(r + 1)
i=0
We remark that the lemma can be proved in a shorter way, but the equations
above will be needed; this way one can see that (r + 1) is the first (i.e. minimal
degree) non-trivial condition that can be proved.
a
Pj
Lemma 25.9. For j 1 l=0 (1)l a+l1
l
jl = 0 holds.
Proof: Since (X + 1)a =
follows that
a+l1
l
l=0
1 = (X + 1)a (X + 1)a =
(1)l X l and (X + 1)a =
Xj
j=0
(1)l
l=0
a+l1
l

l=0
a
l
X l , it

a
,
jl
which immediately gives the lemma.

Now we are ready for the
Proof of Theorem 25.7: Let
G(X) =
k1
X
i (B(X))ki (B(X)).
i=0
If x is a tfold nucleus then Lemma 25.8 applies, that is

G(x) =
k1
X
i (B(x))ki (B(x)) = 0.
i=0
We state that G(X) has degree k(q1). All the nuclei are roots of it. We determine
the coefficient of X k(q1) . Since k1 < q, all the equations (u) hold for u = 1, ..., k.
Let zi denote the coefficient of X i(q1) in
i ((X b1 )q1 , (X b2 )q1 , ..., (X b|B| )q1 ).
Then for i = 1, ..., k 1
zi =
i1
X

zj
j=0

a
.
ij
By Lemma 25.9 for i = 1, ..., k 1

zi = (1)i+1

a+i1
.
i
203
Then for our polynomial (k):

zk =
k1
X

zj
j=0
a
kj

=
k1
X
j=0
from which
k1
zk = (1)
j+1
(1)
a+j1
j

a
kj

t(q + 1)
6= 0
k
by the assumption of the Theorem. So the polynomial is of degree k(q 1) exactly,

it is not the zero polynomial, hence it has at most so many roots.
25.2
Internal nuclei
A point of B not contained in a collinear triplet is usually called an internal

nucleus. The set of internal nuclei is denoted by IN(B). Further result on internal
nuclei can be found e.g. in [139], we only state a nice result of Wettl here. The
proof is left to the reader as an exercise.
Theorem 25.10. (Wettl) Let q be odd, B be a set of q + 1 points in PG(2, q). Then
IN(B) is contained in a conic.
Proof: (Hint: Show that there is a unique tangent at each P IN(B). Apply
Segres lemma for a triangle A1 , A2 , A3 IN(B)).
25.3
Higher dimensions
Now we generalize the previous results to higher dimensions. Here we identify

AG(n, q) with GF(q n ).
Definition 25.11. P 6 B is a t-fold nucleus of B PG(n, q) if all the lines through
P intersect B in at least t points.
A 1-fold nucleus is called nucleus simply. Obviously, if B has a nucleus then |B|
tn1 .
Theorem 25.12. If B AG(n, q), |B| = n1 and it is not a hyperplane then it
has at most q 1 nuclei.
Exercise 25.13. Modify Proof 1 and Proof 2 of Theorem 25.2 as needed!
A more general form of it is the following
204
Theorem 25.14. Let B PG(n, q), |B| = t(q + 1) + k 1, k < q, suppose that
there are exactly i points of it on the ideal hyperplane PG(n, q) \ AG(n, q). Then
the number of its t-fold nuclei contained in AG(n, q) is at most (k + r)(q 1) for
any r 0 such that the binomial coefficient t+ki1
6 0 (mod p) holds.
k+r
It is a generalisation of Blokhuis [31]. For r = 0 this is in [112].
Exercise 25.15. Modify the proof of Theorem 25.4 as needed!
From 25.12, so in the case t = k = 1 we have that there are q 1 nuclei only.
However, if B intersects every hyperplane of PG(n, q), Blokhuis and Mazzocca
proved the following.
Theorem 25.16. If B of size n1 intersects every hyperplane of PG(n, q) then it
has at most q n1 q n2 nuclei; moreover there exist sets attaining this bound.
In the extremal case when r = q n2 1 we have that q n2 |n1 i or we would

have been able to find a smaller r for which the binomial coefficient was non-zero.
It follows then that i n3 (mod q n2 ) or in words that every hyperplane meets
B in n3 (mod q n2 ) points. From this it can be deduced that the classification
of n1 -sets with q n1 q n2 nuclei is equivalent to the classification in the case
of (q + 1)-sets in PG(2, q) having q 1 nuclei, see ... .
As in the planar case we have the following consequence for affine blocking sets.
Theorem 25.17. (Ball) If B is a t-fold blocking set of AG(n, q), e(T ) is the maximal
exponent for which pe(t) |t then |B| (t + 1)q pe(t) .
For e(t) = 0, i.e. when g.c.d.(t, q) = 1, it was proved by Blokhuis. In particular,
when t = 1 it is Jamisons result.
Lower nuclei
Now suppose that B AG(n, q), |B| = tn1 k + 1, and we want to investigate
its t-fold lower nuclei, so the points from which B looks like a (|B|, t)-cap. The
result is quite similar to the case of nuclei:
Theorem 25.19. Let B PG(n, q), |B| = t(q + 1) k + 1, k < q, suppose that
there are exactly i points of it on the ideal hyperplane PG(n, q) \ AG(n, q). Then
the number of its t-fold lower nuclei contained in AG(n,
q) is at most (k + r)(q 1)
ti
for any r > 0 such that the binomial coefficient k+r
6 0 (mod p) holds.
26. Mixed representations
26
205
Mixed representations
This part is from [15]. It is a continuation of Section 13.1, where the planar case
was arranged. Here we see the high dimensional generalization, and what makes
it work is the following mixed representation of the space.
Let k 4. If we view GF(q)k3 GF(q 2 ) as the (k 1)-dimensional vector space
over GF(q) then the points of AG(k1, q) can be viewed as a = (1, a1 , a2 , . . . , ak2 )
where ak2 GF(q 2 ) and the other ai are elements of GF(q). In the quotient space
of w0 = (1, 0, . . . , 0, y0 ), w1 = (0, 1, 0, . . . , 0, y1 ),. . . , wk3 = (0, . . . , 0, 1, yk3 ) the
Pk3
point a is given by (0, . . . , 0, y0 ak2 + i=1 ai yi ). As in Section 13.1, the points
a and b quotient to the the same point in PG(1, q) if and only if (y0 ak2 +
Pk3
Pk3
q1
= (y0 bk2 + i=1 bi yi )q1 if and only if hw0 , w1 , . . . , wk3 , ai =
i=1 ai yi )
Pk3
hw0 , w1 , . . . , wk3 , bi. Note that y0 ak2 + i=1 ai yi = 0 if and only if a
hw0 , w1 , . . . , wk3 i.
Theorem 26.1. A set of points S in PG(k 1, q) which is incident with 0 mod r
points of every hyperplane has at least (r 1)q + (p 1)r points, where 1 < r <
q = ph and k 4.
Proof: Either there is a co-dimension 2 subspace that is incident with no points
of S or |S| q 2 + q + 1 (from which the theorem follows) or every co-dimension
3 subspace is incident with a point of S (which if were the case either |S| > q 3 or
every co-dimension 4 subspace is incident with a point of S.)
Counting points of S on hyperplanes containing this co-dimension 2 subspace
either |S| r(q + 1) (and hence the thereom is proved) or there is a hyperplane
incident with no points of S. Moreover |S| = 0 mod r.
Let s be the greatest common divisor of the non-trivial intersections that S has
with the co-dimension 2 subspaces. If s > 1 then take a co-dimension 2 subspace
incident with ms points of S and by induction each hyperplane containing this
co-dimension 2 subspace contains at least (s 1)q + (p 1)s points of S and so
|S| ((s1)q +(p1)sms)(q +1)+ms > (r +1)q. Thus there is a co-dimension
2 subspace incident with just one point of S. Counting points of S on hyperplanes
containing this co-dimension 2 subspace we see that |S| = 1 + (q + 1)(1) mod r.
Combining this with |S| = 0 mod r we have that q = 0 mod r.
We can view S as a subset of GF(q)k3 GF(q 2 ) ' AG(k 1, q) and consider the
polynomial
R(X, Y ) =
Y
aS
(X + (Y0 ak2 +
k3
X
i=1
ai Yi )q1 ) =
|S|
X
j (Y )X |S|j .
j=0
For all y = (y0 , y1 , . . . , yk3 ) GF(q 2 )k2 the points w0 = (1, 0, . . . , 0, y0 ), w1 =

(0, 1, 0, . . . , 0, y1 ),. . . , wk3 = (0, . . . , 0, 1, yk3 ), a and b span the same hyperplane
Pk3
Pk3
if and only if (y0 ak2 + i=1 ai yi )q1 = (y0 bk2 + i=1 bi yi )q1 6= 0.
206
Suppose that W = hw0 , w1 , . . . , wk3 i is a (k 3)-dimensional subspace is incident

with t points of S. By hypothesis every (k 2)-dimensional subspace contains a
multiple of r points of S and so
R(X, y) = X t (X q+1 1)rt0 g(X)r ,
where t0 = t mod r.
For all y GF(q 2 )k2 the polynomial j (y) = 0 whenever 0 < j < q and r does
not divide j. However j has degree at most j(q 1) and so, for 0 < j < q and r
does not divide j, the polynomial j 0.
Let |S| = (r 1)q + r.
For future reference note that if t = 0 then q+1 (y) = 0. If t = 1 then q+1 (y) = 1
and r (y) = 0 since the degree of g in this case is at most 1.
Let 0 be differentiation with respect to any one of the variables Yi . As in the proof
of the planar case, Theorem 13.6, we have
R(X, y) | (X q+1 1)
R
(X, y).
Y
If |W S| = 0 then R(X, y) is an r-th power and in exactly the same way as in

the proof of Theorem 13.6 we have
0
0
kr q+1
= kr
.
Let
f (Y ) =
(Y0 ak2 +
k3
X
ai Yi ).
i=1
aS
If |W S| > 0 then f (y) = 0 and if |W S| = 0 then q+1 (y) = 0, therefore

f q+1 = 0 for all y GF(q 2 )k2 . So there are polynomials ri for i = 1, . . . , k 2
such that
k2
X q2
f q+1 =
(Yi Yi )ri (Y ),
i=1
and the degree of ri is at most deg f +deg q+1 q 2 |S| 1.

Now choose any (y1 , . . . , yk3 ) GF(q 2 )k3 and b S and set y0 = bk2
Pk3
i=1 bi yi . Then b W so |W S| 1. Differentiate f q+1 with respect to Yi and
evaluate at y. Since |W S| 1 it follows that f (y) = 0 and so q+1 f 0 = ri .
If |W S| > 1 then f 0 (y) = 0 and hence ri (y) = 0 = f 0 (y). If |W S| = 1
then q+1 = 1 and ri (y) = f 0 (y). Thus the polynomial ri + f 0 has at least
|S|q 2(k3) zeros but is a polynomial of degree at most |S| 1 over GF(q 2 ) in
k 3 indeterminates and is either zero or has at most (|S| 1)q 2(k3) zeros. Thus
ri = f 0 and
k2
X q2
f
f q+1 =
(Yi Yi )
.
Y
i
i=1
27. Flocks
207
Now again differentiate f q+1 but now evaluate for y GF(q 2 )k2 where |W S| =
0. Then q+1 (y) = 0 and
0
q+1
= f 0 /f.
0
0
Combining this with r q+1
= r
we have that (f r )0 = 0.
Returning to the case where |W S| = 1 we have that r (y) = f (y) = 0 and so

(f r )0 = 0. And in the case |W S| = t > 1 the polynomial f has a zero of degree
t and so again (f r )0 = 0. The polynomial (f r )0 is zero for all y GF(q 2 )k2
but is a polynomial of degree at most |S| + r(q 1) < q 2 . Thus (f r )0 0. This
holds for which ever indeterminate Yi we choose to differentiate with respect to
p
and so we conclude that f r GF(q 2 )[Y0p , . . . , Yk3
]. Hence f p1 divides r .
If p 2 then (p 1)(r 1)q + r(p 1) > r(q 1) and so r 0. However
the polynomial whose terms are the terms of highest degree in R(X, Y0 , 0, . . . , 0)

r(q1)
is (X + Y0q1 )|S| which has a term X (r1)q Y0
since |S|
r = 1. Thus r has
a term Y r(q1) which is a contradiction. Therefore p 1.
Corollary 26.2. A code whose weights and length have a common divisor r and
whose dual minimum distance is at least 3 has length at least (r 1)q + (p 1)r.
27
Flocks
Theorem 27.1. (Fisher and Thas 1979, Orr 1973) Given an elliptic quadric in
PG(3, q) and a set of q 1 pairwise disjoint conics partitioning all but two of its
points, then the q 1 planes of those conics must contain a common line (that
misses the quadric).
Proof: Use stereographic projection from one of the uncovered points, mapping
all the other points onto the affine plane AG(2, q) in such a way that the other
uncovered point is mapped to the origin. All the conics become pairwise disjoint
circles of the affine plane (or the inversive plane over GF(q)). One can imagine
them as conics in the plane having two conjugate imaginary points at infinity in
common. Now the theorem can be translated saying that a family of q 1 disjoint
circles covers AG(2, q) \ {0} if and only if all circles are centered at the origin.
Now we identify AG(2, q) \ {0} and GF(q 2 ) \ {0} as usual; the points are exactly
2
the roots of Z q 1 1 = 0.
The j-th circle is the set {z : (z aj )(z aj )q = rj }, where aj GF(q 2 ), rj
Qq1
GF(q) . Set cj (Z) = (Z aj )(Z aj )q rj , then the theorem says j=1 cj (Z) =
Zq
1 if and only if all aj = 0.
Exercise 27.2. Prove that this translation is correct!
208
Qq1
Qq1
Now j=1 cj (Z) = j=1 (Z q+1 aj Z q )+ terms of degree at most (q +1)(q 2)+1.
Equating terms of degree greater than (q + 1)(q 2) + 1 one sees that the claim
Qq1
Qq1
2
is j=1 Z q (Z aj ) = Z q 1 if and only if all aj = 0, i.e. j=1 (Z aj ) = Z q1 if
and only if all aj = 0.
Exercise 27.3. Generalize the theorem in the following way: If q 1 norm sets
{z GF(q n ) : Normqn q (z aj ) = rj }, where aj GF(q n ), rj GF(q) , partition
the points of GF(q n ) = AG(n, q) \ {0}, then all aj must be 0.
We are going to use norm sets in Section 29.12.
27.1
Partial flocks of the quadratic cone in PG(3, q)
A flock of the quadratic cone of PG(3, q) is a partition of the points of the cone
different from the vertex into q irreducible conics. Associated with flocks are some
elation generalised quadrangles of order (q 2 , q), line spreads of PG(3, q) and, when
q is even, families of ovals in PG(2, q), called herds. In [109] Storme and Thas
remark that this idea can be applied to partial flocks, obtaining a correspondence
between partial flocks of order k and (k + 2)-arcs of PG(2, q), and constructing
herds of (k + 2)-arcs. Using this correspondence, they can prove that,
for q > 2
even, a partial flock of size > q q 1 if q is a square and > q 2 q if q is a

nonsquare, is extendable to a unique flock.
Applying this last result, Storme and Thas could give new and shorter proofs of
some known theorems, e.g., they can show directly that if the planes of the flock
have a common point, then the flock is linear (this originally was proved by Thas
relying on a theorem by D. G. Glynn on inversive planes, and is false if q is odd).
Here we prove the following
Theorem 27.4. Assume that the planes Ei , i = 1, ..., q intersect the quadratic
1
cone C PG(3, q) in disjoint irreducible conics. If < 14 (1 q+1
) q then one can
find additional planes (in a unique way), which extend the set {Ei } to a flock.
Proof: Let C be the quadratic cone C = {(1, t, t2 , z) : t, z GF(q)} {(0, 0, 1, z) :
z GF(q)} {(0, 0, 0, 1)} and C = C \ {(0, 0, 0, 1)}. Suppose that the planes Ei
intersect C in disjoint conics, and Ei has the equation X4 = ai X1 + bi X2 + ci X3 ,
for i = 1, 2, ..., q .
Define fi (T ) = ai + bi T + ci T 2 , then Ei C = {(1, t, t2 , fi (t)) : t GF(q)}
{(0, 0, 1, ci )}. Let k (T ) = k ({fi (T ) : i = 1, ..., q }) denote the i-th elementary
symmetric polynomial of the polynomials fi , then degT (k ) 2k. As for any fixed
T = t GF(q) the values fi (t) are all distinct, we would like to find
Q
Xq X
,
i (X fi (t))
27. Flocks
209
the roots of which are the missing values GF(q) \ {fi (t) : i = 1, ..., q }.
We are going to use the technique of Section 11. In order to do so, we define
the elementary symmetric polynomials j (t) of the missing elements with the
following formula:
Xq X =

X q1 (t)X q1+2 (t)X q2...q (t) X 1 (t)X 1+2 (t)X 2... (t) ;
from which j (t) can be calculated recursively from the k (t)-s, as the coefficient
(t)1 (t) + ... + 1 (t)j1 (t) + j (t); for

of X qj , j = 1, ..., q 2 is 0 = j (t) + j1
example
1 (t) = 1 (t); 2 (t) = 1 (t)2 2 (t); 3 (t) = 1 (t)3 + 21 (t)2 (t) 3 (t);
etc. Note that we do not need to use all the coefficients/equations, it is enough to
do it for j = 1, ..., .
Using the same formulae, obtained from the coefficients of X qj , j = 1, ..., , one
can define the polynomials
1 (T ) = 1 (T ); 2 (T ) = 1 (T )2 2 (T ); 3 (T ) = 1 (T )3 +21 (T )2 (T ) 3 (T );
(1)
up to . Note that degT (j ) 2j. From the definition

X q 1 (T )X q1 +...q (T ) X 1 (T )X 1 +2 (T )X 2 ... (T )
is a polynomial, which is X q X for any substitution T = t GF(q), so it is
X q X + (T q T )(...). Now define
G(X, T ) = X 1 (T )X 1 + 2 (T )X 2 ... (T ),
(2)
from the recursive formulae it is a polynomial in X and T , of total degree 2

and X-degree .
For any T = t GF(q) the polynomial G(X, t) has roots in GF(q) (i.e. the
missing elements GF(q) \ {fi (t) : i = 1, ..., q }), so the algebraic curve G(X, T )
has at least N q distinct points in GF(q) GF(q). Suppose that G has no
component (defined over GF(q)) of degree 2. Lets apply the Lemma with d = 2
1
and < 21 (1 q+1
); (as 2 q) we have
q N 2(q + 1),
which is false, so G = G1 G2 , where G1 is an irreducible factor over GF(q) of degree
at most 2. If degX G1 = 2 then degX G2 = 2, which means that G1 has at
most q + 1 and G2 has at most ( 2)q distinct points in GF(q) GF(q) (at most
2 for each T = t GF(q)), contradiction (as G has at least q).
210
Both G1 and G2 , expanded by the powers of X, are of leading coefficient 1. So G1

is of the form G1 (X, T ) = X fq+1 (T ), where fq+1 (T ) = aq+1 + bq+1 T +
cq+1 T 2 . Let the plane Eq+1 be defined by X4 = aq+1 X1 + bq+1 X2 +
cq+1 X3 .
The plane Eq+1 intersects C in {(1, t, t2 , fq+1 (t)) : t GF(q)}
{(0, 0, 1, cq+1 )}. Now we prove that for any t GF(q) the points
{(1, t, t2 , fi (t)) : i = 1, ..., q } and (1, t, t2 , fq+1 (t)), in other words, the
values
f1 (t), ..., fq (t); fq+1 (t) are all distinct.But this is obvious
from

X q 1 (t)X q1 + 2 (t)X q2 ... q (t)
X fq+1 (t) | X q X.
Now one can repeat all this above and get fq+2 , ..., fq , so we have
G(X, T ) =
q
Y
(X fi (T ))
q+1
and the values fi (t), i = 1, ..., q are all distinct for any t GF(q). The only remaining case is t = : we have to check whether the intersection points Ei C
on the plane at infinity X1 = 0, i.e. the values c1 , ..., cq ; cq+1 , ..., cq are all
{z
}
| {z } |
distinct (for we know it). (Note that if q planes partition the affine part of C
then this might be false for the infinite part of C .) From (1), considering the
leading coefficients in each defining equality, we have
1 ( ) = 1 (); 2 ( ) = 1 ()22 (); 3 ( ) = 1 ()3+21 ()2 ()3 ();
etc., so

X q 1 ()X q1 + 2 ()X q2 ...q () X q

1 ( )X q1 + 2 ( )X q2 ...q ( ) , which completes the proof.
Xq X
Exercise 27.5. Prove that if q = p is a prime then in Theorem 27.4 the condition
1
< 14 q can be changed for the weaker < 40
p + 1 (so the result is much
stronger).
Exercise 27.6. Assume that the planes Ei , i = 1, ..., q + intersect the quadratic
cone C PG(3, q) in disjoint irreducible conics that cover the cone minus its
1
vertex. If < 41 (1 q+1
) q then one can find planes (in a unique way), such
that if you remove the points of the irreducible conics, in which these planes
intersect C, from the multiset of the original cover then every point of C (except
the vertex) will be covered precisely once.
Exercise 27.7. Let P be the parabola Y = X 2 in AG(2, q). Suppose that the secant
lines Y = ai X + bi , i = 1, ..., q3
2 and the tangent Y = a0 X + b0 = 0 meet P in
q 2 distinct points. Find the formula of the missing secant Y = X + (i.e.
express , with the ai , bi -s).
***
27. Flocks
27.2
211
Partial flocks of cones of higher degree
Using the method above one can prove a more general theorem on flocks of cylinders with base curve (1, T, T d ). This is from [116].
Theorem 27.8. For 2 d 6 q consider the cone {(1, t, td , z) : t, z

GF(q)} {(0, 0, 1, z) : z GF(q)} {(0, 0, 0, 1)} = C PG(3, q) and let
C = C \ {(0, 0, 0, 1)}. Assume that the planes Ei , i = 1, ..., q , Ei 63 (0, 0, 0, 1),
intersect C in pairwise disjoint curves. If < b d12 qc then one can find additional planes (in a unique way), which extend the set {Ei } to a flock, (i.e. q
planes partitioning C ).
The proof (see below) starts like in the quadratic case. Using elementary symmetric
polynomials we find an algebraic curve G(X, Y ), which contains the missing
planes in some sense. The difficulties are (i) to show that G splits into factors,
and (ii) to show that each of these factors corresponds to a missing plane. For
(i) we use our Lemma 10.15. For (ii) we have to show that most of the possible
terms of such a factor do not occur, which needs a linear algebra argument on
a determinant with entries being elementary symmetric polynomials; this matrix
may be well-known but the author could not find a reference for it.
Proof of Theorem 27.8. Suppose that the plane Ei has the equation X4 = ai X1 +
bi X2 + ci X3 , for i = 1, 2, ..., q .
Define fi (T ) = ai + bi T + ci T d , then Ei C = {(1, t, td , fi (t)) : t GF(q)}
{(0, 0, 1, ci )}. Let k (T ) = k ({fi (T ) : i = 1, ..., q }) denote the k-th elementary
symmetric polynomial of the polynomials fi , then degT (k ) dk.
We proceed as in the quadratic case and so we define the polynomials
1 (T ) = 1 (T ); 2 (T ) = 1 (T )22 (T ); 3 (T ) = 1 (T )3+21 (T )2 (T )3 (T ); ...
(1 )
up to . Note that degT (j ) dj. From the definition

X q 1 (T )X q1 +...q (T ) X 1 (T )X 1 +2 (T )X 2 ... (T )
is a polynomial, which is X q X for any substitution T = t GF(q), so it is of

the form X q X + (T q T )(...). Now define
G(X, T ) = X 1 (T )X 1 + 2 (T )X 2 ... (T ),
(2)
from the recursive formulae it is a polynomial in X and T , of total degree d

and X-degree .
For any T = t GF(q) the polynomial G(X, t) has roots in GF(q) (i.e. the
missing elements GF(q) \ {fi (t) : i = 1, ..., q }), so the algebraic curve G(X, T )
has at least N q distinct points in GF(q) GF(q). Suppose that G has no
212
component (defined over

GF(q)) of degree d. Lets apply the Lemma with a
1+d(d1) q
1
suitable d+1 + (d+1)q < d1 , n = deg G d d1 q d + 32 , we have
q N dq < q,
which is false, so G = H1 G1 , where H1 is an irreducible factor over GF(q) of degree
at most d. If degX H1 = dX 2 then degX G1 = dX , which means that H1
has at most q + 1 + (dX 1)(dX 2) q and G1 has at most ( dX )q distinct

points in GF(q) GF(q) (at most dX for each T = t GF(q)), so in total G has
q N ( dX + 1)q + 1 + (dX 1)(dX 2) q,

a contradiction if 2 dX
q + 1, so degX H1 = 1.
One can suppose w.l.o.g. that both H1 and G1 , expanded by the powers of X, are
of leading coefficient 1. So H1 is of the form H1 (X, T ) = X fq+1 (T ), where
fq+1 (T ) = aq+1 + bq+1 T + cq+1 T d + q+1 (T ),
where q+1 (T ) is an error polynomial with terms of degree between 2 and d1.
At the end of the proof we will show that q+1 and other error polynomials are
zero.
Now one can repeat everything for G1 , which has at least ( 1)q distinct points
in GF(q) GF(q) (as H1 has exactly q and H1 G1 has at least q). The similar
reasoning gives G1 = H2 G2 , where H2 (X, T ) = X fq+2 (T ) with fq+2 (T ) =
aq+2 + bq+2 T + cq+2 T d + q+2 (T ). Going on we get fq+3 , ..., fq (where
for j = q + 1, ..., q we have fj (T ) = aj + bj T + cj T d + j (T ), where j (T )
contains terms of degree between 2 and (d 1) only). Hence
G(X, T ) =
q
Y
(X fi (T )).
q+1
For any t GF(q) the values f1 (t), ..., fq (t) are all distinct, this is obvious from

X q 1 (t)X q1 + 2 (t)X q2 ... q (t) (X fq+1 (t))...(X

fq (t)) = X q X.
For j = q + 1, ..., q let the plane Ej be defined by X4 = aj X1 + bj X2 + cj X3 .
We are going to prove that {Ej : j = 1, ..., q} is a flock.
First we check the case t = : we have to check whether the intersection points
Ei C on the plane at infinity X1 = 0, i.e. the values c1 , ..., cq ; cq+1 , ..., cq are
| {z } |
{z
}
all distinct (for we know it). (Note that even if q planes partition the affine part
27. Flocks
213
of C then this might be false for the infinite part of C .) From (1 ), considering
the leading coefficients in each defining equality, we have
1 ( ) = 1 (); 2 ( ) = 1 ()22 (); 3 ( ) = 1 ()3+21 ()2 ()3 ();
etc., so

X q X = X q1 ()X q1+2 ()X q2...q () X 1 ( )X 1+

2 ( )X 2 ... q ( ) ,
which we wanted to prove.
Now we want to get rid of the j s, i.e. we are going to prove that q+1 , ..., q =
0. Let s be the maximal T -exponent appearing in any of q+1 , ..., q , so each
j (T ) = dj T s + ... (for j = q + 1, ..., q; also 2 s d 1 and there exists a
dj 6= 0). In the equation
G(X, T ) = X 1 (T )X 1 + 2 (T )X 2 ... (T ) =
(X aq+i bq+i T cq+i T d q+i (T )),
i=1
the coefficient of X j T d(j1)+s , j = 1, ..., is zero on the left hand side (i.e. the
coefficient of T d(j1)+s in j , it can be seen by induction from (1 ) for instance),
and it is
j1 ( \{cq+1 })dq+1 + j1 ( \{cq+2 })dq+2 +...+ j1 ( \{cq })dq
on the right hand side. Hence we have a system of homogeneous linear equations
for dq+1 , ..., dq with the elementary symmetric determinant

1
1
...
1

1 ( \ {cq+1 })
1 ( \ {cq+2 }) ... 1 ( \ {cq })

2 ( \ {cq+1 })
2 ( \ {cq+2 }) ... 2 ( \ {cq }) =

..

.

1 ( \ {cq+1 }) 1 ( \ {cq+2 }) ... 1 ( \ {cq })
Q
1i<j (cq+i cq+j ), which is non-zero as the ci s are pairwise distinct.

Hence the unique solution is dq+1 , ..., dq = 0 and fj (T ) = aj + bj T + cj T d for
each j = 1, ..., q.
Our final and the last missing argument we need is that for j = 1, ..., q the plane
Ej intersects C in {(1, t, td , fj (t)) : t GF(q)}{(0, 0, 1, cj )}, so these intersections
are pairwise disjoint, E1 , ..., Eq is a flock of C.
214
28
Spreads and ovoids
In this section we show, very briefly, how spreads and ovoids can be handled by
polynomials. The first, very basic part is from Ball [7], while the second one is
based on [10].
28.1
Spreads
A 1-spread of PG(3, q) is a collection of q 2 + 1 lines that partitions the space.

Consider the representation of PG(3, q) with (q 3 + q 2 + q + 1)-th roots of unity (i.e.
(q 1)-th powers) in GF(q 4 ) . (Hyper)planes are given by the zeros of equations
of the form
2
2
aq +q+1 X q +q+1 + aq+1 X q+1 + aX + 1 = 0,
where a is also a (q 3 + q 2 + q + 1)-th root of unity. Lines are given by the zeros of
polynomials of the form
L (X) = X q+1 X + ,
where and satisfy certain conditions which we shall calculate. All the points
3
2
x of PG(3, q) satisfy xq +q +q+1 1 = 0 so
2
Lq X q+1 (X q
+q 2 +q+1
1) + q XLq ( q q
= (q q + q q
+q+1
)X ( q
+1
+q
+q
)L
1)
is identically zero since it is a polynomial of degree at most 1 and has q + 1 distinct

zeros corresponding to the points on the line L . By manipulating the coefficients
we have that L is a line precisely when
q
+q 2 +q+1
= 1 and q+1 = q q
+q+1
If q +1 = 1 then = 0, if q +1 6= 1 then there are q + 1 possibilities for .

This gives (q 3 + q)(q + 1) + q 2 + 1 = (q 2 + 1)(q 2 + q + 1) lines as needed, so these
restrictions are sufficient as well as necessary.
Symplectic spreads
For every point of PG(3, q), viewed as a (q 3 + q 2 + q + 1)-st root of unity in
GF(q 4 ), let
2
(X) = (q )q
+q+1
Xq
+q+1
+ (q )q+1 X q+1 (q )X + 1
be a polynomial over GF(q 4 ) whose zeros correspond to the points of a hyperplane.

It is easy to verify that
()q+1 = ()q+1 and () = 0
28. Spreads and ovoids
215
and it follows from this that defines a symplectic polarity on PG(3, q). The
planes (X) and (X) intersect in the line given by the zeros of the equation
X q+1 +
(q+1 q+1 )q
q+1 q+1
X
= 0.
The line joining the points and is given by the zeros of the equation X q+1
X + = 0 where
=
q+1 q+1
q+1 q+1
and =
.
The two lines coincide whenever q = and the lines for which this condition
holds are the totally isotropic lines.
A symplectic 1-spread is a spread whose elements are totally isotropic lines.
28.2
Ovoids
Generalised quadrangles
A generalised quadrangle is a polar space of rank 2 and consists of points and lines
which have the following properties. (Q1) Two points lie on at most one line (Q2)
If L is a line, and p a point not on L, then there is a unique point of L collinear
with p. (Q3) No point is collinear with all others. The axioms (Q1)-(Q3) are selfdual; the dual of a generalised quadrangle is a generalised quadrangle. Let Q be
a finite generalised quadrangle. Each line is incident with 1 + s points and each
point is incident with 1 + t lines, for some s and t, and we say Q is a generalised
quadrangle of order (s; t). If s = t then Q is said to have order s. An ovoid of a
generalised quadrangle is a set of points O such that each line contains exactly
one point of O. A spread of a generalised quadrangle is a set S of lines such that
each point is incident with exactly one line of S. An ovoid O and a spread S of Q
satisfy
|O| = |S| = st + 1.
The set of lines dual to the ovoid O form a spread in the generalised quadrangle
dual to Q. The set of points dual to the spread S form an ovoid in the generalised
quadrangle dual to Q.
Let q = ph for some prime p and integer h. Let Sp(4, q) denote the symplectic
generalised quadrangle of order q. The points of Sp(4, q) are the points of PG(3, q)
and the lines are the totally isotropic lines of a symplectic polarity. Recall that a
symplectic polarity is induced by an alternating bilinear form b. An alternating
bilinear form on a vector space V satisfies b(v, v) = 0 for all v V . This implies
b(v; w) = b(w; v) for all v, w V . (Expand b(v + w, v + w) = 0.) Hence if the
characteristic is 2 then any alternating form is symmetric.
216
Let O(5, q) denote the generalised quadrangle of order q whose points are the points
of a non-singular quadric in PG(4, q) and whose lines are the lines contained in
that quadric. Later we will examine ovoids in O(5, q). To define O(5, q) one may
choose the quadratic form Q(x) = x0 x4 + x1 x3 + x22 on V(5, q); in this case note
that any ovoid containing (0; 0; 0; 0; 1) may be written in the form
O(f ) = {(0; 0; 0; 0; 1)} {(1, x, y, f (x, y), y 2 xf (x, y)) : x, y GF(q)}.
In fact, due to the following lemma, ovoids in O(5, q) are dual to spreads in Sp(4, q).
Lemma 28.1. O(5, q) is the dual of Sp(4, q).
Proof: Let O+ (6, q) be the Klein quadric of lines of PG(3, q). The image of the
lines of Sp(4, q) is the intersection of O+ (6, q) with a hyperplane PG(4, q), which
is O(5, q). The lines of Sp(4, q) incident with a given point form a pencil of lines
in a plane and therefore their images on O(5, q) lie on a line.
The geometry Sp(4, q) in GF(q 4 )

We recall the geometry of PG(3, q) represented in GF(q 4 ), see Section 6.1. Condition (), that we will use, can be found there.
Let be an element of GF(q 4 ) satisfying q
bilinear form defined by
2
= 1. Let b be the alternating

2
b(X, Y ) := Tr(Y q X) = Y q X + q Y q X q Y X q q Y q X q
and note that b(X, Y ) = b(Y, X). The map y 7 b(X, y) = Tr(y q X) = 0 maps
y to its symplectic hyperplane and defines a symplectic polarity. Let x and y be
2
two orthogonal elements of GF(q 4 ), b(x, y) = 0, and let L(X) = X q +cX q +eX be
the line that joins them. By elimination from the equations L(x) = 0 and L(y) = 0
we can deduce that
2
2
2
2
(xq y y q x)e = xq y q y q xq and (xq y y q x)c = (xq y y q x)
and
(c + q ecq )(xq y y q x) = b(x, y) = 0.
The totally isotropic lines of the polarity defined by b(X, Y ) have c = ecq where
= 1q , as well as the necessary restrictions. In the case when c = 0 there are
q 2 + 1 lines where each line is given by the set of zeros of an equation of the form
2
2
X q + eX = 0 where eq +1 = 1. In the case c is non-zero let d = c1 and we find
that e = dq1 and
3
2
dq +q 1 dq +1 + 1 = 0.
(+)
For each d satisfying this equation there is a totally isotropic line which is given
2
by the set of zeros of an equation of the form dX q + X q dq X = 0. The points
28. Spreads and ovoids
217
of Sp(4, q) are the points of PG(3, q) and for this reason we take as before the
points to be the (q 3 + q 2 + q + 1)-st roots of unity (alternatively the non-zero
(q 1)-st powers) in GF(q 4 ). Therefore we replace the indeterminate X by U
where U = X q1 . It now follows that the lines of Sp(4, q) are given by the zeros
(all (q 3 +q 2 +q +1)-st roots of unity) of equations U q+1 +e = 0 for each e satisfying
2
eq +1 = 1 and
dU q+1 + U dq = 0
(++)
for each d satisfying (+).
Remark. The geometry PG(1, q 2 ) has as points the subspaces of rank 1 in V(2, q 2 ).
In GF(q 4 ) they are the given by sets of zeros of equations of the form
2
X q + eX = 0
2
where eq +1 = 1. Hence the lines defined by the sets of zeros of equations of the
form U q+1 + e = 0 are skew and together they form a Desarguesian spread R
of Sp(4, q). A Desarguesian spread is equivalent to a regular spread. The set of
points in the generalised quadrangle O(5, q) dual to a regular spread of Sp(4, q)
is an elliptic quadric. Hence we need to prove that a spread of Sp(4, q) meets the
spread R in 1 modulo p lines.
Remark. The equations (**) and c = ecq imply that
c2 = 1 eq+1 (eq
+1
1).
When q is even we can take square roots and parameterize the lines using (q 3 +q 2 +
q + 1)-st roots of unity. Moreover when q is even we can assume that = 1 since
the alternating form is also symmetric. Thus we have that the totally isotropic
lines of Sp(4, q) are given by the zeros of equations of the form
U q+1 + (e(q
+q+2)/2
+ e(q+1)/2 )U + e = 0
and one can check that if the point x lies on the line parameterized by e then e
lies on the line parameterized by x2q . Hence we see that Sp(4, q) is self-dual when
q is even, a fact first noted by Tits in 1962.
Ovoids of O(5, q)
Theorem 28.2. An ovoid in O(5, q) meets an elliptic quadric in 1 modulo p points.
This result was proved for q even by Bagchi and Sastry in 1987, who proved that
an ovoid meets not only an elliptic quadric but also a Tits ovoid in an odd number
of points.
Recall that q = ph for some prime p and integer h. In this section we prove that
a spread of Sp(4, q) has 1 modulo p lines in common with the regular spread R.
This regular spread is entirely arbitrary.
218
Proof of the theorem. Let S be a spread of Sp(4,q) and let the sets D and E be
such that for d D the line
dU q+1 + U dq = 0
is in S and for e E the line U q+1 + e = 0 is in S. Clearly |D| + |E| = q 2 + 1. The
aim will be to show that |D| = 0 modulo p and then the result will follow.
The bilinear form b(X, Y ) can be rewritten for the points of PG(3, q) by replacing
X q1 by U and Y q1 by V . Hence for a fixed point u in PG(3, q) the zeros of the
polynomial
2
2
(u, V ) := uq+1 V q+1 + uq +q+1 V uV q +q+1
are the points that are orthogonal to u, i.e. lie on the symplectic hyperplane
through u. Let v be the point of Sp(4, q) (i.e. PG(3, q)) that is the intersection of
the line of Sp(4, q) of the form
dV q+1 + V dq = 0
with the plane (u, V ) = 0, assuming that the line is not contained in the plane.
We can calculate directly or check by substitution that
v q = u(duq+1 + u dq )q1 .
Similarly, if v is the point of intersection of the line of Sp(4, q) of the form V q+1 +
e = 0 with the plane (u, V ) = 0 then
v q = 1 ue(uq+1 + e)q1 .
2
The coefficient of V q +q in (u, V ) is minus the sum of all the points in the plane
(u, V ) = 0 and is zero. Likewise the sum of all the points on any line is minus the
coefficient of V q in the equation of this line which is also zero. Hence the sum of
all points lying in an affine plane is also zero. Thus the sum of all the points of
intersection of the spread S with the plane (u, V ) = 0 is zero and we have
X
X
X
X
0=
v=
vq =
u(duq+1 + u dq )q1 +
1 ue(uq+1 + e)q1 .
eE
dD
Note that of course one of the lines of the spread contains the point u and the
term in the sum corresponding to this line will be zero. The polynomial
X
X
U (dU q+1 + U dq )q1
1 U e(U q+1 + e)q1
dD
eE
is zero for all points of PG(3, q) and since its degree is only q 2 it is identically
zero. However the coefficient of U q is |D| and hence |D| = 0 modulo p.
29. Non-geometrical applications
29
219
Non-geometrical applications
There is an extensive number of applications of polynomials throughout mathematics, here I show some which are close to my combinatorial-algebraic taste.
E.g. the (unfortunately still unpublished) book of Babai and Frankl [4] contains
a chapter devoted to spaces of polynomials. I learned some of the following applications from Tam
as Sz
onyis paper [127], some others from Alons paper [1]. I am
also grateful for the advice of Gy. Karolyi.
29.1
Normal factorizations of Abelian groups
Let G = Zp Zp (we are going to use the additive notation) and suppose that
G = A + B with A, B G, (0, 0) A, B and for each g G there is a unique way
to write it as g = a + b, a A, b B. Then we say that G = A + B is a normal
factorization of G.
Theorem 29.1. In every normal factorization of G = Zp Zp , either A or B is a
subgroup.
Proof: [91] One can identify G and AG(2, p) in the obvious way; then the subgroups of order p correspond to lines through the origin. The key idea is to show
that no direction is determined by both A and B.
Suppose to the contrary that some direction (suppose without loss of generality
that the horizontal direction, i.e. (0)) is determined by both A and B. Let (g)
be the first coordinate of g G, and a primitive p-th root of unity. Then
X
aA
(a)
X
bB
(b)
X
aA,bB
(a+b)
=p
p1
X
i = 0,
i=0
as each element of G occurs precisely once in the form a + b. So one of the two
factors of the product above must be zero, suppose this is the first one (corresponding to A). It means that the set of numbers {(a) : a A} contain all the
residues modulo p, hence A does not determine the horizontal direction. In general
it means that for any direction (m), at least one of A and B does not determine
(m). So either A or B determines at most p+1
2 directions, hence, by Theorem 18.1,
it is a line. As they contain the origin, that line is a subgroup.
The stability version of this result is the following: let G = Zp Zp and A, B

G, (0, 0) A, B, |A| > p , |B| > p , where cp for a small constant c.
Suppose that if for some g G g = a1 + b1 = a2 + b2 , ai A, bi B then a1 = a2
and b1 = b2 . (We may say that A + B G is a partial normal factorization of G.)
220
Exercise 29.2. In every partial normal factorization of G = Zp Zp , either A or

B can be extended to a subgroup of order p, if ... (determine!).
Exercise 29.3. (Redei) Let G = Zp Zp and let A G, |A| = p. Assume that A

is not a subgroup. Then there are at most p1
2 subsets B for which |B| = p, 0 B
and G = A + B holds.
Note that the factorizations of Abelian groups has a rather complicated theory;
neither this statement above can be extended to groups of order pn , n > 2.
A motivating problem for Lov
asz and Schrijver [91] was the following application.
It was conjectured by van Lint (and proved in the prime case by van Lint and
Williams) that if X GF(q 2 ), 0, 1 X, |X| = q and the difference of any two
elements of X is a square element of GF(q 2 ), then X = GF(q).
Exercise 29.4. Prove this conjecture in the case q = p prime, using the representation AG(2, p) GF(p2 ) and Theorem 18.1.
The conjecture for general q was proved by Blokhuis [29] with polynomials. His
proof and his method was generalized for the analogous statement concerning d-th
powers instead of squares of GF(q 2 ) in [113]:
Theorem 29.5. If 1 < d|(q + 1), X GF(q 2 ), 0, 1 X, |X| = q and the difference
of any two elements of X is always a d-th power of GF(q 2 ), then X = GF(q).
Exercise 29.6. Prove either of these results using Theorem 18.14.

Again, the q-subsets of GF(q 2 ) above (or the q-cliques of the Paley graph Pq2 ,d ,
see below) are quite stable:
Exercise 29.7. [113] If 1 < d|(q + 1), X GF(q 2 ), 0, 1 X, |X| = q , where
< (1 d1 ) q. Suppose that the difference of any two elements of X is always a

d-th power of GF(q 2 ), then X GF(q) GF(q 2 ).
29.2
The Paley graph
The Paley graph Pq has GF(q) as vertex set, and (a, b) is an edge if and only if ba
is a square element of GF(q). It is undirected iff 1 is a square, i.e. q = 4k + 1.
Lets examine the automorphisms of Pq . Note that the maps x 7 a2 x + b, a, b
GF(q), a 6= 0 are always automorphisms.
Let f : GF(q) GF(q) be an automorphism and consider the difference quotients
f (y)f (x)
. If (x, y) was an edge then both the denominator and the numerator
yx
are squares, while if (x, y) was not an edge then both the denominator and the
221
numerator are nonsquares; in both cases the difference quotient, i.e. the direction
determined by (x, f (x)) and (y, f (y)) is a square. In fact we have proved that f is
an automorphism if and only if f is bijective and the directions it determines are
all squares in GF(q).
Hence f determines at most
applied.
q1
2
directions and Theorems 18.1 and 18.14 can be
(0) cannot be
determined as f
is a permutation
So for example if q = p we immediately have that f is linear, so of form x 7 a2 x+b.

The non-prime case is more difficult.
Exercise 29.8. Prove that the automorphism group of the Paley graph Pq (q = ph )
j
is {f (X) = a2 X p + b : a GF(q) , b GF(q), j = 0, ..., h 1}
Exercise 29.9. Prove that the Paley graph Pq2 has clique number (Pq2 ) = q.
Exercise 29.10. Generalize the notion of Paley graph, using d-th powers instead of
squares.
Pq,d
Exercise 29.11. Formulate Theorem 29.5 in the language of Paley graphs.
29.3
Representing systems
Another application in Redeis book [103] (Section 37) is about common representing systems of subgroups. Let G be a group, H a subgroup of it, then R G is
a representing system of H if R intersects each coset of H in precisely one element.
RH = G,
|R| |H| = |G|
Let G = Zp Zp again, and let H1 , ..., Hk be its subgroups of order p. We want to

find a common representing system R, so for which (0, 0) R (it can be achieved
w.l.o.g.), |R| = p (this is obvious) and R + Hi = G (by definition). Note that any
further subgroup (different from H1 , ..., Hk ) will do, so we require that R is not a
subgroup. In the usual representation (as above), every Hi is a line through the
origin; and R is a representing system for Hi if r1 6= r2 R (r1 r2 ) 6 Hi , in
other words when the directions determined by R are different from the slopes of
the given lines Hi . Hence, by 18.1, there is no such common representing system
p1
if k > p1
2 , and if k = 2 then R can be described:
Theorem 29.12. Suppose that the subgroups H1 , ..., Hk G = Zp Zp have a
common representing system R, (0, 0) R, which is not a subgroup. Then k p1
2 ,
p1
and if k = 2 then, after a suitable change of the two factors Zp , R is the set
described in Theorem 19.1, and the subgroups Hi are lines of form {(x, mx) : x
GF(p)}, where each (m) is a direction not determined by R.
Note that it does not mean that for any set of ( p1

2 or less) subgroups there is a
common representing system.
so a change of
coordinates
222
29.4
Wielandts visibility theorem
The following theorem was proved first by Wielandt using very complicated methods; later Blokhuis and Seidel [38] has pointed out that it was an easy consequence
of Redeis theorem.
Theorem 29.13. (Wielandt) Let G be a permutation group acting on the points of
AG(2, p), p prime, containing all the translations. Let G0 denote the stabilizer of
the origin. Let S be the set of k lines through the origin, 1 k p1
2 . If G0 maps
the union of the lines in S, as a pointset, onto itself, then any g G0 maps the
lines of S to lines (of S).
Proof: [38] Let g G0 and lets denote by (v) the translation by the vector
v GF(p) GF(p); (g is also a map GF(p)2 GF(p)2 ). Let x and y be points on
a line ` of S. The vector (x y) shows the slope of `. Translate the point (x y)
by y, then let g act on it, finally translate it by g(y): we get

(g(y)) g (y) (xy) = (g(y)) g (x) = (g(y)) (g(x)) = g(x)g(y).
we want g(`) to
be a line ` S
Note that the two translations do not change the directions, and g turned the
direction of x y to the direction of a(nother) line of S.
It means that for any line ` in S, the direction determined by any two points on
g(`), coincides with the slope of one of the lines in S. Hence g(`) determines at
most p1
2 directions, so by Theorem 18.1 it is a line.
Note that the permutations of G do not necessarily preserve the lines. Naturally,
as G = G0 T (where T is the group of translations), all the elements of G map
the lines in S into lines. To make the theorem above (which sounds somewhat
peculiar) more motivated, we remark that the translations form a regular, but
(in general) not normal subgroup T of order p2 , and the lines through the origin
are subgroups of T , each of order p.
29.5
Burnsides theorem
Dress, Klin and Muzichuk [60] and, independently, Ott [96] has realized that a
famous theorem of Burnside can be proved in an elementary way, using Redeis
result (in fact the prime case was re-proved independently by Muzichuk and also
by others a few times). Here, for historical reason, we just quote the theorem,
without proof, for the details see [60].
Theorem 29.14. (Burnside) Let G be a transitive permutation group of degree p
(p prime). Then either G is doubly transitive or it is isomorphic to a subgroup of
the affine transformation group {x 7 ax + b : a, b GF(p), a 6= 0}.
223
Bereczky.
We also mention an application of Blokhuis Theorem 21.4, given by A.
His result is about the existence of fixed-point free p-elements of a permutation
group (a p-element is a permutation whose order is a power of p).
Bereczky [25]) Let p be an odd prime, a 1. If p + 1
Theorem 29.15. (A.
3
b < 2 (p + 1) then every transitive permutation group of degree pa b contains a
fixed-point free p-element.
29.6
Jaegers conjecture
In [2] Alon and Tarsi use Redei polynomials to prove Jaegers conjecture for q non
prime. Jaegers conjecture states that when q 5, for all matrices M GL(n, q)
there exists a vector y GF(q)n with the property that neither y nor M y have
any zero coordinate. (Jaeger conjectured this for q = 5 only.)
This problem can be formulated in many equivalent ways:
the points of the vectorspace V(n, q) cannot be covered by two sets of n
independent linear hyperplanes;
independent:
they intersect
in {0} only
given two bases of V(n, q), {vi : i =P1, ..., n} and

P {wi : i = 1, ..., n} then there
exists a vector u V(n, q), u =
i vi =
i wi , with each i , j being
nonzero coordinates;
PG(n 1, q) cannot be covered by (the union of the faces of) two simplices.

1 1
1 1
, while for q = 3 the matrix
are
1 0
1 1
counterexamples. Also if the conjecture is false for some (q, n0 ) then it is false for
any (q, n), n n0 .
Note that for q = 2 the matrix
Let us see how Redei polynomials can be used to tackle this problem. If the
conjecture is not true then there is a matrix M GL(n, q) with the property
that M y has a zero component, for all y GF(q)n with no zero component.
The set of n points A that is made up of the rows of M is a set of n points in
PG(n 1, q) that spans the whole space and has the property that the hyperplane
y1 X1 + y2 X2 + ... + yn Xn = 0 is incident with a point of A for all y GF(q)n with
no zero component. In other words, if we define a Redei polynomial
Y
R(Y1 , Y2 , ..., Yn ) :=
(x1 Y1 + x2 Y2 + ... + xn Yn )
(x1 ,x2 ,...,xn )A
then for all vectors y which have no zero component R(y1 , y2 , ..., yn ) = 0. This
algebraic property of R implies that we can write
R(Y1 , Y2 , ..., Yn ) = (Y1q1 1)f1 + (Y2q1 1)f2 + ... + (Ynq1 1)fn ,
()
add
a
zero
column and an
independent
row
to
the
counterexample
matrix M
224
where f1 , f2 , ..., fn GF(q)[Y1 , Y2 , ..., Yn ] are polynomials of degree at most n

(q 1). The matrix M is non-singular
and so has an inverse M 1 = (aij ). If we
Pn
make the substitution Yi = j=1 aij Zj we have
Z1 Z2 ...Zn =
n X
n
X
i=1
aij Zj
q1

1 fi .
j=1
Now on the right-hand side when we try to find a term Zj1 Zj2 ...Zjq1 from one
P
q1
of the
aij Zj
terms, as we must since the left-hand side is Z1 Z2 ...Zn , the
coefficient is a multiple of (q 1)!. If q is not prime then this is always zero and
we have a contradiction.
We can explore this situation a little bit. First note that if q is a prime we have
Z1 Z2 ...Zn =
n
X
i=1

aij1 aij2 ...aijq1 Zj1 Zj2 ...Zjq1 fi =
X
j1 <j2 <...<jq1
Zj1 Zj2 ...Zjq1
j1 <j2 <...<jq1
n
X

aij1 aij2 ...aijq1 fi .
i=1
A typical term of fi is
Yi1 Yi2 Yinq+1 =
n
X
ai1 j Zj
j=1
n
X
j=1
n

X

ai2 j Zj
ainq+1 j Zj ,
j=1
where i1 , ..., inq+1 are not necessarily pairwise distinct. What is the coefficient of
Z1 Z2 Znq+1 for instance on the right hand side? It is
X
ai1 (1) ai2 (2) ainq+1 (nq+1) .
Sym(1,2,...,nq+1)
Lets examine the Redei polynomial of the counterexamples in the q = 2 and q = 3

cases above! They are
R(Y1 , Y2 ) = (Y1 + Y2 )Y1 = (Y11 1)Y1 + (Y21 1)Y1 = Y11 Y1 + Y21 Y1
and
R(Y1 , Y2 ) = (Y1 + Y2 )(Y1 Y2 ) = (Y12 1)1 + (Y22 1)(1) = Y12 1 + Y22 (1).
In () consider the terms of degree (n q + 1). They are f1 + f2 + ... + fn , hence
f1 +f2 +...+fn 0 and R(Y1 , Y2 , ..., Yn ) = Y1q1 f1 +Y2q1 f2 +...+Ynq1 fn , ()
Hence n q 1
or contradiction
225
where f1 , ..., fn are homogeneous polynomials of degree n q + 1. We claim that

if (y1 , ..., yn ) is a zero-free point which is orthogonal to at least two rows of the
matrix M then fj (y1 , ..., yn ) = 0 for each j = 1, ..., n. Let e.g. j = 1 and consider
1 ) = R(Y1 , y2 , ..., yn ) = (Y q1 1)f1 (Y1 , y2 , ..., yn ). Since R
splits into linear
R(Y
1
it
factors, so does f1 (Y1 , y2 , ..., yn ) as well; and as Y1 = y1 is a multiple root of R,
is a root of f1 (Y1 , y2 , ..., yn ) (with multplicity decreased by one).
Consequently each fi vanishes on the same set of points S (GF(q) )n .
29.7
Graphs containing p-regular subgraphs
Theorem 29.16. Alon-Friedland-Kalai [1] For any prime p, any loopless graph
G(V, E) with average degree bigger than 2p 2 and maximum degree at most
2p 1 contains a p-regular subgraph.
The most interesting case is p = 3, see [1] for the details.
Proof: Let (av,e )vV,eE be the incidence matrix of G, i.e. av,e = 1 or 0 as v e
or not. Associate a variable Xe to each edge e, and define the polynomial
!
Y
X
Y
F =
1(
av,e Xe )p1
(1 Xe )
vV
eE
eE
over GF(p). deg F = |E| as (p 1)|V | < |E| by assumption. The coefficient of
Q
|E|+1
6= 0. Hence, by Theorem 5.17, there is a vector x with
eE Xe is (1)
coordinates xe {0, 1} such that P
F (xe : e E) 6= 0. This x is not the zero vector
as F (0) = 0. Also for each v V , eE av,e xe = 0 mod p since otherwise F would
vanish at x. So in the subgraph consisting of the edges with xe = 1 every degree
is precisely p (as the maximum degree is 2p 1).
29.8
Blokhuis proof of Bollobas theorem
Theorem 29.17. Let A1 , ..., Ah and B1 , ..., Bh be collections of sets

with i : |Ai | =
r, |Bi | = s and Ai Bj = if and only if i = j. Then h r+s
s .
Proof: [30] W.l.o.g. we can assume that every set is a subset of R. Define the
polynomials
Y
Y
ai (X) =
(X ), i = 1, ..., h;
bj (X) =
(X ), j = 1, ..., h.
Ai
Bj
By the properties of resultants, see Section 9.5, and the conditions on the sets Ai
and Bj , we get that R(ai , bj ) 6= 0 if and only if i = j. Let c(X) = cs X s +...+c1 X+c0
226
be an arbitrary polynomial of degree s, then for a fixed i we have that R(ai , c) is

a homogeneous polynomial of degree s in the coefficients (considered as variables)
c0 , ..., cs of c. We claim that the polynomials R(ai , c), i = 1, ..., h are independent
in the r+s
polynomials of degree r in
s -dimensional vectorspace of homogeneous

r+s
the variables
c
,
...,
c
(and
hence
h
).
To
prove
this
independence, suppose
0
s
s
P
that i i R(ai , c) = 0. Then substituting bj for the polynomial c we get j = 0.
29.9
Alon-F
uredi
Theorem 29.18. Let F be a field. Suppose that in Fd the union of the hyperplanes
Hi , i = 1, ..., m cover the vertices of the hypercube {0, 1}d except the origin. Then
m d.
Pd
Proof: Let the equation of Hi be j=1 aij Xj 1 = 0. Then
F (X1 , ..., Xd ) =
m X
d
Y
i=1
aij Xj 1
j=1
vanishes in every x = (x1 , ..., xd ), xj {0, 1}, x 6= (0, 0, ..., 0), but F (0, 0, ..., 0) 6= 0.
Now
m X
d
d

Y
Y
aij Xj 1 (1)d+m
(Xj 1)
G(X1 , ..., Xd ) =
i=1
j=1
j=1
vanishes on {0, 1} . Theorem 5.17 states that if X1 X2 Xd is a leading term then

G cannot vanish on a set of form S1 S2 ... Sd , with each |Si | > 1. Hence
X1 X2 Xd is not a leading term and m d.
Note that in case of the equality m = d the coefficient of X1 X2 Xd must be
zero in
d X
d
d

Y
Y
G(X1 , ..., Xd ) =
aij Xj 1
(Xj 1)
i=1
j=1
j=1
hence Per(aij ) = 1.
29.10
Chevalley-Warning
This theorem of Chevalley and Warning holds for any finite field; for simplicity
we prove it over GF(p) only.
Theorem 29.19. Let f1 , ..., fm be polynomials from GF(p)[X1 , ..., Xn ]. If n >
P
m
i=1 deg(fi ) and the polynomials have a common zero c = (c1 , ..., cn ) then they
have another common zero.

Proof:
227
Suppose that the statement is false and define
f (X1 , ..., Xn ) =
m
Y
n

Y
(1 fi (X1 , ..., Xn ))p1
i=1
j=1
(Xj ),
GF(p)\{cj }
where 6= 0 is chosen such that f (c) = 0.

Observe that f (x) = 0 for any x GF(p)n . It holds for x = c; by assumption there
p1
exists some fj for which
= 0. Similarly, as for some
Q fj (x) 6= 0 hence 1 fj (x)
i xi 6= ci , so we have GF(p)\{cj } (Xi ) = 0 and then f (x) = 0.
Qn
Let ti = p 1 for i = 1, ..., n; then the coefficient of i=1 Xiti in f is 6= 0.
Hence by Theorem 5.17 with Si = GF(p) for all i, we have that there exists an
s GF(p)n for which f (s) 6= 0, a contradiction.
A Chevalley-Warning type application of the punctured Nullstellensatz

This nice result of Ball-Serra [22] can be now posed as an exercise.
Exercise 29.20. Let f1 , f2 , ..., fm be polynomials of GF(q)[X1 , X2 , ..., Xn ] and let
d = |D1 | + ... + |Dn |, where Di is the set of elements c of GF(q) where there exists
a common zero of f1 , f2 , ..., fm with i-th coordinate c. If d 6= 0, in other words if
the polynomials f1 , f2 , ..., fm have a common zero, then
m
X
i=1
29.11
deg(fi )
nq d
.
q1
Cauchy-Davenport
Theorem 29.21. Let A, B (Zp , +) be two subsets of size |A| = k, |B| = l. Then
|A + B| min{k + l 1, p}.
Proof: Let Zp = (GF(p), +). If k +l > p then everything is obvious: for any g G
the sets A and g B will intersect, so g A + B. Suppose that |A + B| k + l 2
and
Q choose C G such that A + B C and |C| = k + l 2. Define F (X, Y ) =
cC (X + Y c), it vanishes on A B. Here deg F = (k 1) + (l 1) and the
coefficient of X k1 Y l1 is k+l2
6= 0 (as k + l 2 < k + l 1 p). It contradicts
k1
Theorem 5.17.
The very similar proof of the Eliahou-Kervaire-Bollobas-Leader theorem is
228
Exercise 29.22. Alon [1] Let p (r, s) denote the smallest n for which the triple
(r, s, n) satisfies the Hopf-Stiefel condition, i.e. nk 0 (mod p) for every n r <
k < s.
If A and B are nonempty subsets of the vectorspace V(d, p), |A| = r, |B| = s, then
|A + B| p (r, s).
Dias da Silva and Hamidoune proved that for A Zp , |A| = k, p a prime, |A+A|
=
|{a1 + a2 : a1 6= a2 A}| min{2|A| 3, p}, settling a problem of Erdos and
Heilbronn. The proof below contains a slight simplification by Karolyi.
Suppose to the contrary that a set GF(p) C A+A,

with |C| = 2k 4. Consider
f (X, Y ) = (XY )2
(X+Y c) = (XY )2 (X+Y )2k4 + terms of lower degree.
cC
Then f (a1 , a2 ) = 0 for all a1 , a2 A. deg f = 2k 2, the coefficient of X k1 Y k1

2(2k4)!(k1)
2k4
2k4
is 2k4
k3 2 k2 + k1 = (k1)!(k1)! , which is nonzero if 0 < 2k 4 < p
(otherwise we are done). Thus, by Theorem 5.17 we can find a1 , a2 A such that
f (a1 , a2 ) = 0, contradiction.
29.12
Another application
Let H be a fixed graph. The classical problem from which extremal graph theory
has originated is to determine the maximum number of edges a graph on n vertices
can have without containing a copy of H. This maximum value is the Turan
number of H and is customarily denoted by ex(n, H).
It is particularly interesting to determine the Turan numbers when H is bipartite,
as in most cases even the order of magnitude is open. The Zarankiewicz problem
concerns the complete bipartite graph, so ex(n, Kt,s ), where t s.
Kov
ari, T. S
os and Tur
an gave the upper bound for arbitrary fixed t:
1
ex(n, Kt,s ) ct,s n2 t ,

where ct,s > 0 is a constant depending on t and s. The right hand side is conjectured to give the correct order of magnitude. However the best general lower
bound, obtained by the probabilistic method, is
c0 n2
s+t2
st1
ex(n, Kt,s ),
where c0 > 0 is an absolute constant, but for all 2 t s

the lower bound is always of lower order of magnitude.
s+t2
st1
>
1
t
holds; so
The optimality of the upper bound (up to a constant factor) is proved for t = 2, 3
and all s t. The incidence graphs of projective planes demonstrate this order of
229
magnitude for t = 2. In this

case, however, even the asymptotic order of magnitude
3/2
is known: ex(n, K2,s ) = s1
+ O(n4/3 ) (F
uredi [65]).
2 n
The optimality for t = 3 was established by Brown [50]. Later F
uredi proved that
Browns construction is asymptotically optimal: ex(n, K3,3 ) = (n5/3 ).
The construction of Brown is the following unit distance graph in a 3dimensional affine space: the vertices of our graph will be the points of AG(3, q).
Let k1 , k2 be such that E : X 2 + k1 Y 2 + k2 Z 2 = 1 is an elliptic quadric in
AG(3, q). Then the vertices (x, y, z) and (u, v, w) are connected by an edge iff
(x u)2 + k1 (y v)2 + k2 (z w)2 = 1.
Proposition 29.23. This graph has n5/3 edges and does not contain a K3,3 .
Proof: E intersects the ideal plane in the conic X 2 + k1 Y 2 + k2 Z 2 = 0, so it has
q 2 + 1 (q + 1) = q 2 q affine points and the same is true for all its translates.
The neighbours of the vertex are on a translate of E, so the number of edges is
1 3 2
3 5/3
.
2 q (q q) (q )
Let U (u1 , u2 , u3 ), V (v1 , v2 , v3 ) and W (w1 , w2 , w3 ) be three points and EU , EV , EW
the corresponding quadrics, i.e. the (equations of the) neighbours of U, V and W
resp., so (X u1 )2 + ... = 1, (X v1 )2 + ... = 1 and (X w1 )2 + ... = 1. Subtract
the first one from the second and the third, we get two linear equations which do
not coincide, so they define either a line or an empty set. As EU does not contain
a line, this line (if exists), intersects it in at most two points, so U, V and W has
at most 2 common neighbours.
The idea of this construction is that we define a surface around each point of
the space, such that (i) they are translates of each other (so it is easy to handle
them); (ii) they have many points (so the graph will contain many edges); and
(iii) any three translates intersect in a bounded number of points (so any triple of
vertices has a bounded number of common neighbours). The next result, due to
Koll
ar, R
onyai and Szab
o [87] generalizes the same idea.
Let the set of vertices of our graph be the elements of GF(q t ), and recall that
t1
t
Normqt q (a) = a aq ... aq
= a(q 1)/(q1) . Now two vertices a and b are
adjacent iff Normqn q (a + b) = 1. The number of solutions of Normqn q (x) = 1
t
t
1
1
1
is qq1
in GF(q t ) so the number of edges is at least 12 q t ( qq1
1) 12 (q t )2 t .
Theorem 29.24. This graph Gq,t contains no subgraph isomorphic to Kt,t!+1 .
1
Corollary 29.25. For t 2 and s t! + 1 we have ex(n, Kt,s ) ct n2 t , where

ct > 0 depends on t only, we may choose ct = 2t . For every t and s t, the
inequality holds with c = 1/2 for infinitely many values of n.
230
The Corollary follows from the Theorem and from the fact that there is a prime
power between 12 n1/t and n1/t . The union of [ qnt ] disjoint copies of Gq,t will have
the appropriate number of edges.
The Theorem is a direct consequence of the following: if d1 , d2 , ..., dt are distinct
elements of GF(q t ) then the system of equations
t1
Normqt q (X + d1 ) = (X + d1 )(X q + dq1 )...(X q
t1
+ dq1
Normqt q (X + d2 ) = (X + d2 )(X q + dq2 )...(X q
t1
+ dq2
t1
+ dqt
t1
)=1
)=1
..
.
Normqt q (X + dt ) = (X + dt )(X q + dqt )...(X q
t1
)=1
has at most t! solutions x GF(q ). They prove it by considering a more general

system of equations:
Lemma 29.26. Let K be a field and aij , bi K for 1 i, j t such that aij1 6= aij2
if j1 6= j2 . Then the system of equations
(X1 a11 )(X2 a21 )...(Xt at1 ) = b1
(X1 a12 )(X2 a22 )...(Xt at2 ) = b2
..
.
(X1 a1t )(X2 a2t )...(Xt att ) = bt
has at most t! solutions (x1 , x2 , ..., xt ) K t .
In the proof of the Lemma first polynomials fj = (X1 a1j )(X2 a2j )...(Xt
atj ) are defined, then the regular map F : K t K t by F (X1 , X2 , ..., Xt ) :=
(f1 , f2 , ..., ft ). Then the Lemma states that |F 1 (b)| t! holds for every b K t .
It is obvious for b = 0. The rest of the proof involves some commutative algebra
and we omit it.
Remark 29.27. One gets Theorem 29.24 by putting
K = GF(q t ), aij = dqj
i1
, bj = 1.
Chapter 3
Hints and short solutions to the

exercises
1
Of Chapter I
Exercise 5.1
Q
i
In the polynomial ring GF(q)[X] the equality (1 + X)n = i=0 (1 + X)ni p =

Q
P
Q
P
i
i
p1
(1 + X p )ni holds. In other words k=0 nk X k = i=0 ki =0 nkii X ki p =

Pi=0
2
p1
n0 n1
k0 +k1 p+k2 p +...
and the result follows from comparison of
k0 ,k1 ,...=0 k0 k1 ...X
coefficients.
Exercise 5.2
q n 1
If is a generator element of GF(q n ) then q1 generates GF(q). Now let N :

GF(q n ) GF(q) be a multiplicative map with N () = GF(q), then N ( k ) =
k . If =
q n 1
q1 a
then N ( k ) = Norm(( k )a ).
Exercise 5.5
P
xGF(q) f (x)k is on the one hand the coefficient of X q1 , on the other hand it
P
is aGF(q) ak .
Exercise 5.6
If b = 0 then the equation is solvable as x 7 x2 is an automorphism. Multiplying
through by a/b2 and introducing Y = aX/b we get Y 2 + Y + ac/b2 = 0. As Tr is
additive we have Tr(Y 2 )+Tr(Y )+Tr(ac/b2 ) = 0. As, by definition, Tr(Y 2 ) = Tr(Y ),
if the equation is solvable then Tr(ac/b2 ) = 0.
231
232
Hints and solutions
Conversely, if Tr(ac/b2 ) = 0 then the equation is solvable: the sets {y 2 + y : y

GF(2h )} and {z : Tr2h 2 (z) = 0} are identical; is obvious and either of them is
the image space of a linear mapping with one-dimensional kernel.
Exercise 5.7
Let {deg(f ) : f V } = {d1 , d2 , ..., dn }, where d1 > d2 > ... > dn . Choose a monic
fi with deg(fi ) = di , i = 1, ..., n. We prove that {f1 , ..., fn } is a base of V . Let
g V , then define g0 = g, and for k = 1, ..., n let ak be the coefficient of X dk in
gk1 , finally letPgk = gk1 ak fk . Observe that deg(gk ) < dk and gk V . Now
gn = 0, so g = ai fi hf1 , ..., fn i.
Exercise 5.8
(X a) =
(a)r X r ; use Lucas theorem (Exercise 5.1).

P
For the converse prove that aGF(q) aq1+r (X a) = (1)r1 r X r unless
= q 1 and r = 0 or q 1.
r=0
Exercise 5.9
If f (aY + b, Y ) = 0 then f (X,
Q Y ) = 0 modulo X aY b, and hence X aY
b | f (X, Y ). It follows that (a,b)S (X aY b) | f (X, Y ). Since the degree of
the left hand side is |S|, the result follows.
Exercise 5.13
Let A = {Pi : i = 1, ..., n+t1} be an arc in PG(n1, q). Then fi (X) = Pi X = 0
is the equation of a hyperplane Hi through the origin in AG(n, q). Any point of
AG(n, q) \ 0 is covered at most (n 1) times by these hyperplanes, because of the
arc property.
Now fi (X)q1 1 vanishes precisely in the points of AG(n, q) \ Hi ; hence for
Qn+t1
(fi (X)q1 1), at any point of AG(n, q) \ 0 at least (n + t 1) (n 1) = t
i=1
factors will be zero. This is a polynomial of degree (n + t 1)(q 1).
Exercise 5.14
Let Ci be defined by the polynomial fi (X, Y ), then consider
fi and use 5.12.
Exercise 5.21
Q
Q
Let gi (Xi ) = si Si (Xi si ), and li (Xi ) = di Di (Xi di ). By Theorem 5.20
Pn
Q n gi
we can write f = i=1 hi gi + w, and w = u i=1 li for some non-zero polynomial
u, and the degree in Xi of w is less than |Si |.
Exercise 5.23
Suppose that f has a zero of degree at least t at all elements of S1 S2 ... Sn .
By Theorem
, ..., Xn ] with the property
P 5.22 there are polynomials h F[X1 , X2P
that f = T g (1) ...g (t) h , and deg h deg(f ) i deg(gi ). On the right
1. Of Chapter I
233
Q
|Si |
hand side of this equality the terms of highest
P degree are divisible by i Xi
for some . So there is a for which ri i |Si | for all i , a contradiction.
Exercise 5.26
f is singular iff f (X) = 0 has more than one solution. For any solution x GF(q)
k
h
one can consider the pk -th power 0 = f (x)p . Using xp = x and re-ordering, one
gets the row of the matrix corresponding to this exponent. Hence for any solution x,
2
h1
the vector (x, xp , xp , ..., xp )> is a solution of the homogeneous matrix equation.
The other direction is similar.
Exercise 5.27
As in the proof of Theorem 5.25, the if part is obvious. Consider GF(q) as an he dimensional vectorspace over GF(pe ), then f is a GF(pe )-linear map. The number
h
of such maps is (pe ) e . The polynomials of form Trqpe (aX) are among these maps,
h
the number of these polynomials is q = (pe ) e .
Exercise 5.32
P
Multiply the sum by (4a2 ) = 1, we get (a) xP
((2ax + b)2 (b2 4ac)).
2
Replacing 2ax + b by y and b 4ac by d yields (a) y (y 2 d). The case d = 0
is clear, for d 6= 0 we have to count the solutions of Y 2 d = Z 2 . It defines a
we
hyperbola with q 1 affine points. If d is a nonsquare then for q1
2 values y
q+1
2
have (y d) = 1 and for the other 2 it is 1. If d is a square then for y = d
2
we have (y 2 d) = 0; for q3
2 values y we have (y d) = 1 and for the other
q1
2 it is 1.
Exercise 6.1
The number of k-dimensional (linear) subspaces in V(n, q) is
n
k q.
nd
The number of k-dimensional (affine) subspaces in AG(n, q) is q
The number of k-dimensional projective subspaces in PG(n, q) is
n
k q.
n+1
k+1
.
q
Exercise 7.6
X q1 means that (1, 0, 0) is the nucleus in the even case, every line except those
through the nucleus, intersects the parabola in 0 mod 2 points. Adding (1, 0, 0) to
S the *-polynomial becomes zero.
In the odd case the tangents of the conic are the solutions [x, y, z] of X 2 4Y Z.
Exercise 9.5
Each factor of the right hand side should divide the determinant. The degree of
the two sides is equal.
Exercise 9.7
234
Hints and solutions
Similar to Exercise 9.5. The variables are x1 , ..., xn only.

Exercise 9.13
Q
Let F (T ) = x (T f (x)). The Newton-Waring-Girard formulae (Ex. 9.3, N1-3)
imply that all the coefficients, i.e. the elementary symmetric polynomials
i are
P
zero if i q 2 and p 6 |i. Hence F (T ) = G(T )p + cT , where c = x f (x)q1 . Here
c = 0 means case (ii), while c 6= 0 implies F 0 (T ) = c 6= 0, so F has no multiple
root, f is a permutation polynomial, F (T ) = T q T , c = 1.
Exercise 9.14
Let V (f ) = {a1 , ..., aK } and mi = |f 1 (ai )| the multiplicity of ai . If wf <
then
m1 + m2 + ... + mK = 0
a1 m1 + a2 m2 + ... + aK mK = 0
a21 m1 + a22 m2 + ... + a2K mK = 0
...
w
w
w
a1 f m1 + a2 f m2 + ... + aKf mK = ,
where 6= 0. If K < wf + 1 then the first K equations has the trivial solution
m1 = ... = mK = 0 only, contradicting the last one.
Exercise 9.22
Use Exercise 9.8!
Exercise 9.25
Consider
1t
y1
2
y1
.
.
.
t1
y1
1
y2
2
y2
...
...
...
1
yt
2
yt
t1
y2
...
t1
yt
1
1
.
.
1
.
.
0
0
. = . ,
hence the determinant should be zero:

V dM (y1 , y2 , ..., yt ) ty2 y3 yt V dM (y2 , y3 , ..., yt ) = 0. So
dM (y1 ,y2 ,...,yt )
3 )(y1 yt )
t = y2 y3Vy
= (y1 y2 )(yy21yy
.
t V dM (y2 ,y3 ,...,yt )
3 yt
Exercise 10.1

The number of coefficients is n+2
. So the coefficient vectors form the points
2

n+2
of an 2 1 dimensional projective space over GF(q).
Exercise 10.2
Each point (u, v) S gives a linear equation for the coefficients of the desired
polynomial
p f . Hence we have a homogeneous system of |S| such equations. When
deg f 2|S| 1 then the number of coefficients is larger than |S|.
2. Of Chapter II
235
Exercise 10.8
P
Consider f f (q) , where f (q) denotes
(ijk )q X i Y j Z k when f (X, Y, Z) =
ijk
P
ijk X i Y j Z k ; then use Bezouts theorem.
ijk
Exercise 10.10
Consider the curve F (X, Y ) = f (X) Y 2 , it has roughly 2q points (a bit less in
fact). By Weil it cannot be irreducible. Write it as the product of two factors, it is
not too difficult to see that it must be of form f (X)Y 2 = (g(X)Y )(h(X)+Y ),
and then g = h and f = g 2 .
Exercise 10.11
Stand in a point P of the curve and look around! If P has multiplicity m 1 and
there is no linear component then on each line we can see at most n m points,
counted with intersection multiplicities, plus the one we are standing in. It gives
m + (q + 1)(n m) = qn + n qm q(n 1) + n.
Exercise 10.12
Note that 1 k n 1. (i) If we have a k-secant line ` and P ` C then looking
around from P we get the bound Nq (n 1)q + k. (ii) We say that a point P can
see the tangency at Q if Q is on the curve and (one of) the tangent line(s) at Q goes
through P . (P = Q is allowed.) Now the points of the curve can see at least Nq k
tangencies, so there is at least one point P of the curve seeing at least k tangencies.
Counting the points of the curve looking around from P we lose at least one
(from the total number) at each tangency that P can see, which gives (n1)q+nk
as an upper bound. Finally min{(n 1)q + k, (n 1)q + n k} (n 1)q + n2 .
Exercise 10.16
Use Theorem 10.9 by St
ohr and Voloch!
Exercise 10.18
Suppose that f is irreducible. By a coordinate transformation, which only permutes the elements of I, one can achieve f (X 0 , Y 0 , Z 0 ) = X 02 Y 0 Z 0 . But now
p+1
f 2 6 I, contradiction.
Of Chapter II
Exercise 12.3
Let {(ai , bi ) : i = 1, ..., k} be an affine blocking set and suppose that (a1 , b1 ) =
Qk
(0, 0). Consider F (X, Y ) = i=2 (ai X+bi Y +1). As {(ai , bi ) : i = 2, ..., k} blocks all
the lines not through the origin, F (x, y) = 0 for all (x, y) 6= (0, 0). Now G(X, Y ) =
236
Hints and solutions
Qk
+ 1) (X q1 1)(Y q1 1) vanishes on GF(q) GF(q), hence

cannot be (one of) its leading term(s) by Theorem 5.17. Hence k 1
i=2 (ai X + bi Y
q1 q1
X
Y
2q 2.
Exercise 12.5
If P B is essential with t tangents through it then choose the coordinate system
so that P ` and ` is a tangent to B. Putting one point on each tangent
except ` results in an affine blocking set of size |B| 1 + t 1, which is, by
Theorem 12.2, at least 2q 1, hence t 2q + 1 |B|.
Exercise 13.2
If x S is of multiplicity m then from x in each direction one can see 1m mod p
further points. HenceP
(1) the number of points (with
P multiplicity) is m+(1m)(q
1) P
1 (mod p); (2) sS (x s)q1 = (1 m) q+1 =1 = 0, so again x is a root
of sS (X s)q1 of degree q 1.
Exercise 13.3
n = 1 is obvious, n = 2 is Proposition 13.1. Suppose that the statement is proved
for n 1 2 and consider a (multi)set S in AG(n, q). Let P H be a point
at infinity (in other words, a direction) not determined by S. Then the quotient
geometry at P is ' PG(n 1, q), the points of S go into points of a multiset
S 0 PG(n 1, q) in a natural way without any further coincidence; there is a
hyperplane (i.e. the quotient of H ) disjoint from S 0 , so S 0 AG(n 1, q); finally
every hyperplane of AG(n 1, q) intersects S 0 in either 0 or 1 mod p points, so by
the (induction) hypothesis we are done.
Exercise 13.4
Counting the points of S on the lines through some fixed point s S we have
|S| r (mod p). After the AG(2, q) GF(q 2 ) identification define
X
f (X) =
(X s)q1 ,
sS
it is not identically zero as the coefficient of X q1 is r. If x S then every direction

will occur withPthe same multiplicity r 1 mod p, hence for each point s S,
f (s) = (r 1) q+1 =1 = 0, so they are roots of f , which is of degree q 1. The
biggest value r mod p below q is q p + r. However, a set of size q + 1 always
has a 1-secant, so the corollary is that r = 1.
Exercise 14.1
See the proof of Exercise 10.11.
Exercise 14.2
Let ì be the axis Xi = 0 and mj the line X2 = aj X1 + bj . So for the intersections
with the axes we have GF(q) = {aj : j = 1, ..., q 1} = {bj : j = 1, ..., q 1} =
2. Of Chapter II
237
{ bjj : j = 1, ..., q 1}. By Wilsons theorem (Section 9), we get 1 =

Q
j
bj =
Q aj
j
bj
Q
aj
Qj
j bj
aj =
= 1.
If q is even then the lines {X2 = aX1 + a2 : a GF(q) } suffice. In fact, if

{ì } {mj } is a dual hyperoval then it works. Another solution is the following: let
P be a point not on the triangle. Then the q2 lines through P , not containing the
vertices of the triangle, plus the line joining the three missing non-vertex points
on `1 , `2 and `3 form such a set.
Exercise 14.7
Let Q = L1 L2 and let F and G two curves through the given points. As neither F
nor G contains Q, a suitable linear combination H of them does, hence H contains
the given points and Q as well. Then L1 is a component Q
of H, so H intersects
k
each of the other lines Li also in at least n + 1 points, so i=1 Li divides H. As
deg H = n < k, it follows that H = 0 and F = G.
Exercise 15.1
The condition is equivalent to saying that there is no collinear triple of points on
the graph of f .
Exercise 15.2
As f (X) is a permutation polynomial, deg f q 2. Calculate Fs (0) = f 0 (s), it
vanishes for every s GF(q) by Exercise 15.1.
Exercise 15.3
X r is a permutation polynomial iff (r, q 1) = 1. F0 (X) = X r1 gives (r 1, q
r
r
sr
1 (X/s+1) 1
1) = 1 similarly. For s 6= 0 Fs (X) = (X+s)
= sr1
, which is a
X
X/s
permutation polynomial if and only if
(Y +1)r 1
Y
is.
Exercise 15.4
Use Exercise 15.3.
Exercise 15.11 Its size is 1 + (n 1)(q + 1), 1 for F0 and q + 1 for the others. The
lines through (0, 0, 1) intersect each F in one point. For the other affine lines of
equation aX + Y + bZ, b 6= 0, calculate the intersections and use that H is an
additive subgroup.
Exercise 17.13
P
Let us count now the weighted index of points, that is let i(P ) =
ktn+k (P ),
where tn+i (P ) denotes the number of (n + i)-secants through P . If t0 (P ) = q/n,
then the index of P is just . In general, 0 i(P ) (mod n). In particular,
for < n, the indices are always at least . On the other hand, adding up indices
along a 0-secant gives the total (weighted number) of lines longer than n. For a
238
Hints and solutions
point of S, the index is , hence the total (weighted) number of long lines is at
most |S|/(n + 1) < (q + 1). This is a contradiction.
Exercise 18.3
Stand into a point P ` \ B and look around, you have to see at least one (affine)
point of B on each line 6= `.
In case of equality, if a direction P ` is determined by two affine points of B \ `
then, as there are q affine lines through P and at least one of them contains at
least two of the q affine points of B, there exists an affine line through P which is
not blocked by the affine points of B hence P should be added. A non-determined
point means precisely one point per affine line through it, hence it is unnecessary
to put it to B.
Frobenius: transitive, not regular, g 6=id has

1 fixed points
Exercise 18.6
A 3/2-transitive group is either primitive or a Frobenius group. As D GF(pe ),
lin
FD , it
we have that GF(pe ) is a block of FD , hence FD is not primitive. As FD
lin
follows that FD = FD .
Exercise 18.8
Use Exercise 18.5.
Exercise 18.12
(i) If bi = bj then the parabolas meet at X = 0. (ii) The equation
ai aj
bi bj
= 1/x2
p+1
2
must not have a solution x GF(q). (iii) f defines

directions, hence its
graph is a line. If it is non-horizontal then it intersect the horizontal axis.
Exercise 23.3
Use that the eigenvalues of the incidence matrix are q + 1 and q.

Exercise 24.6
Assume that 0 is in the blocking set, and is the dual translation plane with
respect to 0. The dual
is a translation plane of order q 2 with kernel of order q.
Each line of
can be regarded as a plane of AG(4, q) and by Jamison or Theorem
24.5 with n = 4, k = 2, t = 1 we are done.
Exercise 24.8
Let l1 , l2 , ..., ln be n lines of PG(n, q) incident with the same point x of A and
spanning PG(n, q). Let H be the hyperplane which is incident with no point of A
and set Si = li \ {x} and Di = li H. Theorem 24.7 implies |A| 1 (t 1)(q
1) + n(q 1).
Exercise 24.13
The binomial coefficient in Corollary 24.12 with n = 2 and j = k = t (t, q) is

1(t,q)
, which is non-zero by Lucas Theorem.
t(t,q)
2. Of Chapter II
239
Exercise 27.5
Use the St
ohr-Voloch bound (Exercise 10.16).
Exercise 27.6
def
Use
k (T ) = k (T ) instead of k (T ) for k = 1, ..., in the proof; everything
remains the same, like in Section 11.
Exercise 27.7
The secant joining (c, c2 ) and (d, d2 ) is Y = (c + d)X cd. If theP
missing points
are (, 2 ) and (, 2 ) then we need = + and = . As zGF(q) z = 0
Q q3 1
Q
P
2
.
and zGF(q) z = 1, we have = ai and =
i=1 bi
Exercise 29.3
By the preceding, every such set B must be a subgroup, hence a line through the
origin. G = A + B implies that every line parallel to B must intersect A, hence
the direction of B is not determined by A. By Theorem 18.1, there are at most
p1
2 such directions.
Exercise 29.4
For x 6= y GF(p2 ), observe that x y is a square in GF(p2 ) iff (x y)q1 is a
square in the multiplicative subgroup of the (p + 1)-th roots of unity ( GF(p2 ) ).
So after the GF(p2 ) AG(2, p) identification, X AG(2, p) determines at most
p+1
2 directions, hence, by Theorem 18.1, it is a line containing 0 and 1.
Exercise 29.8
Use Theorem 18.7.
Exercise 29.20
Qm
Define f = i=1 (1 fi (X1 , X2 ..., Xn )q1 ) and note that f is non-zero only precisely at the common zeros of f1 , fP
zero then Theorem
2 , ..., fm . If there is a common
Pn
m
5.20 implies that deg f = (q 1) i=1 deg(fi ) nq i=1 |Di |.
Exercise 29.22
Lets identify V(d,Q
p) with GF(pd ). Let C = A + B, suppose |C| = n < p (r, s).
Define f (X, Y ) = cC (X + Y c), then f vanishes on A B. By the definition
of p (r, s) there exists some n r < k < s such that nk 6 0 (mod p). So the
coefficient of X nk Y k in f is nonzero, and (as |A| = r > n k, |B| = s > k), by
Theorem 5.17 it is a contradiction.
240
Hints and solutions
Chapter 4
Glossary of concepts
Here one can find the most important definitions, in alphabetical order.
A (k, n)-arc of PG(2, q) is a pointset of size k, meeting every line in at most n
points. An arc is a (k, 2)-arc. A (k, n)-arc is complete if it is not contained in a
(k + 1, n)-arc.
A blocking set (w.r.t. lines) is a pointset meeting every line. In general, a blocking
set in PG(n, q) w.r.t. k-dimensional subspaces (sometimes it is called an (n k)blocking set) is a pointset meeting every k-subspace. A t-fold blocking set meets
every k-subspace in at least t points. A point P of B is essential if B \ {P } is no
longer a blocking set, i.e. there is a 1-secant k-space through P . B is minimal if
every point of it is essential. A blocking set B of PG(2, q) is small if |B| < 32 (q + 1),
in general, a blocking set B in PG(n, q) w.r.t. k-dimensional subspaces is small if
|B| < 32 q nk + 1.
Given B PG(2, q), the point P 6 B is a t-fold nucleus of B if all the lines
through P intersect B in at least t points. P 6 B is a t-fold lower nucleus of B if
all the lines through P intersect B in at most t points. So from a nucleus B seems
to be a t-fold blocking set while from a lower nucleus B seems to be a (|B|, t)-arc.
If P is a point of B then the similar notions are called internal nuclei.
A blocking set B PG(n, q), with respect to k-dimensional subspaces, is of Redei
type, if it has precisely q nk points in the affine part AG(n, q) = PG(n, q) \ H .
A semioval is a pointset of PG(2, q) with the property that there is a unique tangent
line at each point of it. It is regular, if it is a pointset of type (0, 1, a), i.e. all the
secants are of the same length a.
A k-spread of PG(n, q) is a partition of the space into k-dimensional subspaces.
241
242
A subgeometry of = PG(n, q) is a copy of some 0 = PG(n0 , q 0 ) embedded in it,
so the points of 0 are points of and the k-dimensional subspaces of 0 are just
the intersections of some k-subspaces of with the pointset of 0 . It follows that
GF(q 0 ) must be a subfield of GF(q).
The type of a pointset of PG(2, q) is the set of its possible intersection numbers
with lines. In particular, an arc is a set of type (0, 1, 2), a set of even type is a
pointset with each intersection numbers being even, etc.
A unital is a pointset of PG(2, q 2 ) of size q 3 + 1 intersecting each line in either 1
or q + 1 points. It has a unique 1-secant through each point of it so it is a minimal
blocking set.
An untouchable set is a set without tangents.
Chapter 5
Notation
V(n, F) denotes the n-dimensional vector space with coordinates from the field F.
If F = GF(q) then we write V(n, q) instead.
AG(n, F) denotes the n-dimensional affine space with coordinates from the field F.
If F = GF(q) then we write AG(n, q) instead.
PG(n, F) denotes the n-dimensional projective space with coordinates from the
field F. If F = GF(q) then we write PG(n, q) instead.
a
a1
ab+1
1)...(q
1)
= (q 1)(q
(the q-binomials or Gaussian binomials, the
(q b 1)(q b1 1)...(q1)
number of b-dimensional linear subspaces of V(a, q)).
a
b q
If the order q of a plane or space is fixed we write

q i+1 1
i
i1
i = i+1
+ ... + q + 1.
1 q = q1 = q + q
2
Trqn q (X) = X + X q + X q + ... + X q

GF(q).
2
Normqn q (X) = XX q X q X q
n1
n1
is the trace function from GF(q n ) to
is the norm function from GF(q n ) to GF(q).
In some cases will denote a primitive element of GF(q n ) (and sometimes it is

2
n1
nice to assume that (, q , q , ..., q ) form a trace-orthogonal base over GF(q)
if possible, see 5.2).
Jt is the ideal h(X1q X1 )i1 (X2q X2 )i2 ...(Xnq Xn )in : 0 i1 +i2 +...+in = ti in
GF(q)[X1 , ..., Xn ] of polynomials vanishing everywhere with multiplicity at least t.
H is the hyperplane at infinity in the projective space PG(n, q) when an affine
part is fixed, i.e. H = PG(n, q) \ AG(n, q). When n = 2, it is called the line at
infinity ` .
243
244
Mf : given the polynomial f GF(q)[X], the number of elements a GF(q) for
which f (X) + aX is a permutation polynomial.
(y)
Df : given the polynomial f GF(q)[X], Df = { f (x)f
: x 6= y GF(q)}, the
xy
set of directions determined by the graph of f .
Nf = |Df |.
wf : given the polynomial f GF(q)[X], wf = min{k :
xGF(q)
Wf : given the polynomial f GF(q)[X], Wf = min{k + l :
N (B) denotes the set of (external) nuclei of the pointset B.

IN(B) denotes the set of internal nuclei of the pointset B.
f (x)k 6= 0}.
xGF(q)
xk f (x)l 6= 0}.
Bibliography
[1] N. Alon, Combinatorial Nullstellensatz, Combin. Probab. Comput. 8
(1999), no. 1-2, 729.
[2] N. Alon, M. Tarsi, A nowhere zero point in linear mappings, Combinatorica 9 (1989), 393-395.
[3] E. F. Assmus, Jr, J. D. Key, Designs and codes, Cambridge University
Press, 1992.
[4] L. Babai, P. Frankl, Linear algebra methods in combinatorics, unpublished preliminary version.
l and A. Wigderson, Superpolynomial lower bounds
[5] L. Babai, A. Ga
for monotone span programs, Combinatorica 19 (1999), 301319.
[6] S. Ball, Partial unitals and related structures in Desarguesian planes, Designs, Codes and Cryptography, 15 (1998), 231236.
[7] S. Ball , Polynomials in finite geometries, in Surveys in Combinatorics,
1999, LMS Lecture Note Series 267 Cambridge University Press 1999, pp.
1735.
[8] S. Ball, Intersection sets in Desarguesian affine spaces, European J. Combin., 21 (2000), 441446.
[9] S. Ball , The number of directions determined by a function over a finite
field, J. Combin. Th. Ser. A, 104 (2003), 341350.
[10] S. Ball, On ovoids of O(5, q), Adv. Geom., 4 (2004) 17.
[11] S. Ball, On the graph of a function in many variables over a finite field,
Des. Codes Cryptogr., to appear
[12] S. Ball, A. Blokhuis, On the incompleteness of (k, n)-arcs in Desarguesian planes of order q where n divides q, Geom. Dedicata, 74, (1999),
325332.
245
246
Bibliography
[13] S. Ball, A. Blokhuis, F. Mazzocca, Maximal arcs in Desarguesian

planes of odd order do not exist, Combinatorica 17 (1997), 3141.
[14] S. Ball, A. Blokhuis An easier proof of the maximal arc conjecture, Proc.
Amer. Math. Soc., 126 (1998), no. 11, 33773380.
cs, P. Sziklai, Zs. Weiner, On linear
[15] S. Ball, A. Blokhuis, A. Ga
codes whose weights and length have a common divisor, Advances in Mathematics 211 (2007), 94104.
cs, On the graph of a function over a prime field whose
[16] S. Ball, A. Ga
small powers have bounded degree, preprint.
cs, P. Sziklai, On the number of directions determined by
[17] S. Ball, A. Ga
a pair of functions over a prime field, J. Combin. Theory Ser. A, to appear.
[18] S. Ball, P. Govaerts and L. Storme, On ovoids of parabolic quadrics,
Designs, Codes and Cryptography 38 (2006), 131145.
[19] S. Ball, A. Blokhuis and M. Lavrauw, Linear (q +1)-fold blocking sets
in PG(2, q 4 ), Finite Fields Appl. 6 (2000), 294-301.
[20] S. Ball, M. Lavrauw, On the graph of a function in two variables over a
finite field, J. Algebraic Combin. 23 (2006), 243253.
[21] S. Ball, M. Lavrauw, How to use Redei polynomials in higher dimensional
spaces, Le Matematiche (Catania) 59 (2004), 3952 (2006).
[22] S. Ball, O. Serra, Punctured combinatorial Nullstellensatzen, (2007),
manuscript.
[23] S. Ball, Zs. Weiner, An introduction
http://www-ma4.upc.es/ simeon/IFG.pdf
to
finite
geometry,
[24] A. Barlotti, Sui {k, n}-archi di un piano lineare finito, Boll. Un. Mat.
Ital. 11 (1956), 553556.
Bereczky, Fixed-point-free p-elements in transitive permutation groups,
[25] A.
Bull. London Math. Soc. 27 (1995), 447452.
[26] T. Beth, D. Jungnickel and H. Lenz, Design theory, 2nd Edition, Cambridge Univ. Press, 1999
[27] A. Beutelspacher, Blocking sets and partial spreads in finite projective
spaces, Geom. Dedicata 9 (1980), 425449.
, On polynomials over prime fields taking only two values on the
[28] A. Biro
multiplicative group, Finite Fields and Their Appl. 6 (2000), 302-308.
Bibliography
247
[29] A. Blokhuis, On subsets of GF(q 2 ) with square differences, Indag. Math.

46 (1984), 369-372.
[30] A. Blokhuis, Solution of an extremal problem for sets using resultants of
polynomials, Combinatorica 10 (1990), 393396.
[31] A. Blokhuis, On multiple nuclei and a conjecture of Lunelli and Sce, Bull.
Bel. Math. Soc. Simon Stevin 3 (1994), 349-353.
[32] A. Blokhuis, Blocking sets in Desarguesian Planes, in: Paul Erdos is
Eighty, vol. 2, (1996) 133155. eds.: D. Miklos, V.T. Sos, T. Szonyi, Bolyai
Soc. Math. Studies.
[33] A. Blokhuis, On the size of a blocking set in PG(2, p), Combinatorica 14
(1994), 273276.
[34] A. Blokhuis, Characterization of seminuclear sets in a finite projective
plane, J. Geom. 40 (1991), 1519.
[35] A. Blokhuis, Extremal problems in finite geometries, Bolyai Society Mathematical Studies 3 (1991), 111135.
[36] A. Blokhuis, Polynomials in finite geometries and combinatorics, in Surveys in Combinatorics, LMS Lecture Notes Series 187 Cambridge University
Press 1993, pp. 3552.
sz, L. Storme and T. Szo
nyi, On multiple block[37] A. Blokhuis, L. Lova
ing sets in Galois planes, Adv. Geom. 7 (2007), 39-53.
[38] A. Blokhuis and J.J. Seidel, Remark on Wielandts visibility theorem,
Lin. Alg. and its Appl., 71 (1985), 29-30.
nyi, Lacunary polynomials, multiple
[39] A. Blokhuis, L. Storme and T. Szo
blocking sets and Baer subplanes, J. London Math. Soc. (2) 60 (1999), 321
332.
nyi, Note on the structure of semiovals in finite
[40] A. Blokhuis and T. Szo
projective planes, Discr. Math. 106/107 (1992), 6165.
nyi, Zs. Weiner, On sets without tangents on Galois
[41] A. Blokhuis, T. Szo
planes of even order, Designs, Codes and Cryptograpy 29 (2003), 9198.
Seress, H. A. Wilbrink, On sets of points without
[42] A. Blokhuis, A.
tangents, Mitt. Math. Sem. Univ. Giessen 201 (1991), 3944.
nyi, On
[43] A. Blokhuis, S. Ball, A. Brouwer, L. Storme and T. Szo
the number of slopes determined by a function on a finite field, J. Comb.
Theory Ser. (A) 86 (1999), 187196.
248
Bibliography
[44] A. Blokhuis and A. E. Brouwer, Blocking sets in Desarguesian projective planes, Bull. London Math. Soc. 18 (1986), 132134.
[45] A. Blokhuis, A.A. Bruen, The minimal number of lines intersected by
a set of q + 2 points, blocking sets and intersecting circles, J.Comb. Theory
(A), 50 (1989), 308315.
[46] A. Blokhuis, K. Metsch, Large minimal blocking sets, strong representative systems and partial unitals, in: Finite Geometries, (F. De Clerck et
al., eds), Cambridge Univ. Press, Cambridge, 1993, 3752.
nyi, Blocking sets of almost Redei
[47] A. Blokhuis, R. Pellikaan, T. Szo
type, J. Comb. Theory Ser. A 78 (1997), 141150.
nyi, On the sharpness of the theorem of B. Segre, Com[48] E. Boros, T. Szo
binatorica 6 (1986), 261268.
[49] A. E. Brouwer and A. Schrijver, The blocking number of an affine
space, J. Combin. Theory Ser. A 24 (1978), 251253.
[50] W.G. Brown, On graphs that do not contain a Thomsen graph, Canad.
Math. Bull. 9 (1966), 281-289.
[51] A. A. Bruen, Baer subplanes and blocking sets, Bull. Amer. Math. Soc. 76
(1970), 342344.
[52] A. A. Bruen, Blocking sets in finite projective planes, SIAM J. Appl. Math.
21 (1971), 380392.
[53] A.A. Bruen, Polynomial multiplicities over finite fields and intersection
numbers, J. Combin. Th. Ser. A 60 (1992), 1933.
[54] A. A. Bruen and M. J. de Resmini, Blocking sets in affine planes, In
Combinatorics 81, 18 of Ann. Discrete Math., North-Holland, AmsterdamNew York (1983), 169175. (Rome, 1981)
[55] A.A. Bruen and J.C. Fisher, The Jamison method in Galois geometries,
Designs, Codes and Cryptography 1 (1991), 199205.
[56] A.A. Bruen and B. Levinger, A theorem on permutations of a finite
field, Can. J. Math. XXV. (1973), 1060-1065.
[57] A. A. Bruen and J. A. Thas, Blocking Sets, Geom. Dedicata 6 (1977),
193203.
[58] J.L. Coolidge, A treatise on algebraic plane curves, Dover Publications,
Inc., New York 1959.
Bibliography
249
[59] R. H. F. Denniston, Some maximal arcs in finite projective planes, J.

Combinatorial Theory 6 (1969), 317319.
[60] A.W.M. Dress, M.H. Klin and M.E. Muzichuk, On p-configurations
with few slopes in the affine plane over Fp and a theorem of W. Burnside,
Bayreuther Math. Schriften 40 (1992), 7-19.
s, L. Lova
sz, Problems and results on 3-chromatic hypergraphs
[61] P. Erdo
and some related questions, Infinite and finite sets (Colloq., Keszthely, 1973;
dedicated to P. Erdos on his 60th birthday), Vol. II, pp. 609627, Colloq.
Math. Soc. Janos Bolyai, Vol. 10, North-Holland, Amsterdam, 1975.
[62] S. Ferret, L. Storme, P. Sziklai, Zs. Weiner, A t mod p result on
weighted multiple (nk)-blocking sets in PG(n, q), Innovations in Incidence
Geometry, to appear.
[63] J. C. Fisher, J. W. P. Hirschfeld, J. A. Thas, Complete arcs on planes
of square order, Ann. Discrete Math. 30 (1986), 243250.
[64] W. Fulton, Algebraic Curves: An Introduction to Algebraic Geometry,
reprint, Addison Wesley, 1989.
redi, Tur
[65] Z. Fu
an-type problems, in: London Math Soc. Lecture Note Series
166 (ed. A.D. Keedwell), Cambridge University Press, 1991, 253-300.
cs, On the size of the smallest non-classical blocking set of Redei type
[66] A. Ga
in PG(2, p), J. Combin. Th. Ser. A. 89 (2000), 43-54.
cs, On a generalization of Redeis theorem, Combinatorica 23 (2003),
[67] A. Ga
no. 4, 585598.
cs, On regular semiovals in PG(2, q), J. Alg. Comb. 23 (2006), 7177.
[68] A. Ga
cs, P. Sziklai and T. Szo
nyi, Two remarks on blocking sets and
[69] A. Ga
nuclei in planes of prime order, Designs, Codes and Cryptography, 10 (1997),
29-39.
cs, Zs. Weiner, On (q + t, t)-arcs of type (0, 2, t), Designs, Codes
[70] A. Ga
and Cryptography 29 (2003), 131139.
[71] D. Goss, Basic structures of function field arithmetic. Ergebnisse der Mathematik und ihrer Grenzgebiete (3) 35. Springer-Verlag, Berlin, 1996.
[72] M.L. Green, Generic initial ideals, Six lectures on commutative algebra
(Bellaterra, 1996), 119186, Progr. Math. 166, Birkhuser, Basel, 1998.
Hadnagy, T. Szo
nyi, On the embedding of large (k, n)-arcs and partial
[73] E.
unitals, Ars Combinatoria 65 (2002), 299308.
250
Bibliography
[74] U. Heim, Proper blocking sets in projective spaces, Discrete Math. 174
(1997), 167176.
[75] U. Heim, Blockierende Mengen in endliche projektiven Raumen, Mitt. Math.
Semin. Giessen 226 (1996), 4-82.
[76] J. W. P. Hirschfeld, Ovals in desarguesian planes of even order, Ann.
Mat. Pura Appl. (4) 102 (1975), 79-89.
[77] J. W. P. Hirschfeld, Projective geometries over finite fields, Clarendon
Press, Oxford, 1979, 2nd edition, 1998, 555 pp.
[78] J. W. P. Hirschfeld, Finite projective spaces of three dimensions, Oxford
University Press, Oxford, 1985, 316 pp.
[79] J. W. P. Hirschfeld, J. A. Thas, General Galois geometries, Oxford
University Press, Oxford, 1991, 407 pp.
ros, Arcs and curves over a finite
[80] J. W. P. Hirschfeld and G. Korchma
field, Finite Fields Appl. 5 (1999), 393-408.
[81] J. W. P. Hirschfeld, G. Korchmros and F. Torres, Algebraic Curves
over a Finite Field, Princeton University Press.
[82] J.W.P. Hirschfeld and L. Storme, The packing problem in statistics,
coding theory and finite projective spaces: update 2001. Developments in
Mathematics 3, Kluwer Academic Publishers. Finite Geometries, Proc. 4th
Isle of Thorns Conference (2000), 201-246.
nyi, Constructions of large arcs and blocking
[83] J. W. P. Hirschfeld, T. Szo
sets in finite planes, European J. Comb. 12 (1991), 499-511.
[84] R.E. Jamison, Covering finite fields with cosets of subspaces, J. of Comb.
Th. Ser. A 22 (1977), 253266.
[85] B. C. Kestenband, A family of complete arcs in finite projective planes,
Colloq. Math., LVII (1987), 5967.
[86] J. D. Key, M. J. de Resmini, Small sets of even type and codewords, J.
Geom. 61 (1998), no.1-2, 83104.
r, L. Ro
nyai, T. Szabo
, Norm-graphs and bipartite Turan num[87] J. Kolla
bers, Combinatorica 16 (1996), 399406.
ros and F. Mazzocca, On (q + t)-arcs of type (0, 2, t) in
[88] G. Korchma
a desarguesian plane of order q, Math. Proc. Camb. Phil. Soc. 108 (1990),
445-459.
Bibliography
251
[89] A. Lempel and M.J. Weinberger, Self-complementary normal bases in

finite fields, SIAM J. of Discrete Math. 1 (1988), 193198.
[90] R. Lidl and H. Niederreiter, Finite fields. Addison-Wesley, Reading,
Mass., 1983. (Cambridge University Press, 1988). (Second Edition, Cambridge University Press, 1997.)
sz and A. Schrijver: Remarks on a theorem of Redei, Studia
[91] L. Lova
Scient. Math. Hungar 16 (1981), 449-454.
[92] G. Lunardon, Normal spreads, Geom. Dedicata 75 (1999), 245261.
[93] G. Lunardon, Linear k-blocking sets, Combinatorica 21 (2001), 571581.
[94] L. Lunelli, M. Sce, Considerazioni aritmetiche e risultati sperimentali sui
{K, n}q -archi, Ist. Lombardo Accad. Sci. Rend. A 98 (1964), 352.
[95] K. Metsch, Blocking sets in projective spaces and polar spaces, Combinatorics 2002, Topics in Combinatorics: Geometry, Graph Theory and Designs
(2002), 204219.
[96] U. Ott, Group assotiation schemes and Schur rings, Le Matematiche (Catania) XLVII (1992), 281294.
[97] Stanley E. Payne, A complete determination of translation ovoids in finite
Desarguian planes, Atti Accad. Naz. Lincei Rend. Cl. Sci. Fis. Mat. Natur.
(8) 51 (1971), 328331.
[98] P. Polito, O. Polverino, On small blocking sets, Combinatorica 18
(1998), 133137.
[99] O. Polverino, Small minimal blocking sets and complete k-arcs in
PG(2, p3 ), Discrete Math. 208/9 (1999), 469476.
[100] O. Polverino, Small blocking sets in PG(2, p3 ), Designs, Codes and Cryptography 20 (2000), 319324.
[101] O. Polverino, L. Storme, Small minimal blocking sets in PG(2, q 3 ),
Eur. J. Comb. 23 (2002), 8392.
nyi, Zs. Weiner, Blocking sets in Galois planes of
[102] O. Polverino, T. Szo
square order, Acta Sci. Math. (Szeged) 65 (1999), 737748.
dei, L
[103] L. Re
uckenhafte Polynome u
ber endlichen K
orpern, Akademiai
Kiad
o, Budapest, and Birkhauser Verlag, Basel, 1970 (English translation:
Lacunary polynomials over finite fields, Akademiai Kiado, Budapest, and
North Holland, Amsterdam, 1973).
252
Bibliography
[104] H. Sachar, The Fp -span of the incidence matrix of a finite projective plane,
Geometriae Dedicata 8 (1979), 407415.
[105] B. Segre, Lectures on modern geometry, Edizioni Cremonese, Rome, Italy,
1961.
[106] B. Segre, Introduction to Galois geometries (ed. J.W.P. Hirschfeld), Atti
Accad. Naz. Lincei Mem. Cl. Sci. Fis. Mat. Natur I 8 (1967), 133-236.
[107] A. Seidenberg, Elements of the theory of algebraic curves, Addison-Wesley,
Reading, Mass., 1968.
[108] L. Storme and P. Sziklai, Linear pointsets and Redei type k-blocking
sets in PG(n, q), J. Alg. Comb. 14 (2001), 221228.
[109] L. Storme, J. A. Thas, k-arcs and partial flocks, Linear Algebra Appl.
226/228 (1995), 33-45.
[110] L. Storme and Zs. Weiner, On 1-blocking sets in PG(n, q), n 3. Des.
Codes Cryptogr. 21 (2000), 235251.
[111] K-O. St
ohr and J. F. Voloch, Weierstrass points and curves over finite fields,
Proc. London Math. Soc., 52 (1986) 119.
[112] P. Sziklai, Nuclei of pointsets in PG(n, q), Disc. Math., 174 (1997), 323-327.
[113] P. Sziklai, Subsets of GF(q 2 ) with d-th power differences, Disc. Math.,
208-209 (1999), 547-555.
[114] P. Sziklai, A lemma on the randomness of d-th powers in GF(q), Bull. Belg.
Math. Soc. 8 (2001), 95-98.
[115] P. Sziklai, Partial flocks of the quadratic cone, J. Combin. Th. Ser. A, 113
(2006), 698-702.
[116] P. Sziklai, Flocks of cones of higher degree, J. Algebraic Combin., 25
(2007), 233238.
[117] P. Sziklai, A conjecture and a bound on the number of points of a plane
curve, Finite Fieds Appl., 14 (2008), 4143.
[118] P. Sziklai, On small blocking sets and their linearity, J. Combin. Th. Ser.
A, to appear.
nyi, Blocking sets and algebraic curves, Rend. Circ.
[119] P. Sziklai and T. Szo
Mat. Palermo 51 (1998), 7186.
ts, Vandermonde and super-Vandermonde sets,
[120] P. Sziklai and M. Taka
submitted.
Bibliography
253
nyi, Combinatorial problems for Abelian groups arising from geom[121] T. Szo
etry, Periodica Polytechnica, 19 (1991) 197212.
nyi, Some applications of algebraic curves in finite geometry and
[122] T. Szo
combinatorics, in: Surveys in Combinatorics (R. A. Bailey, ed.), Cambridge
Univ. Press, Cambridge, 1997, 198236.
nyi, Blocking sets in Desarguesian affine and projective planes, Finite
[123] T. Szo
Fields and Appl. 3 (1997), 187-202.
nyi, On the number of directions determined by a set of points in an
[124] T. Szo
affine Galois plane, J. Combin. Th. Ser. A 74 (1996), 141146.
nyi, On the embeddability of (k, p)-arcs, Designs, Codes and Cryp[125] T. Szo
tography 18 (1999), 235246.
nyi, Note on the existence of large minimal blocking sets in Galois
[126] T. Szo
planes, Combinatorica 12 (1992), 227-235.
nyi, Some recent applications of Redeis theory of lacunary polyno[127] T. Szo
mials (in Hungarian), Polygon 5 (1995), 2, 4978.
nyi, On some applications of curves in finite geometry, Socrates lec[128] T. Szo
ture notes, Potenza, 2001.
nyi, Zs. Weiner, Small blocking sets in higher dimensions, J. Com[129] T. Szo
bin. Th. Ser. A 95 (2001), 88-101.
nyi, Zs. Weiner, On stability theorems in finite geometry,
[130] T. Szo
manuscript 2006.
nyi, Zs. Weiner, Large minimal blocking sets, manuscript, 2002.
[131] T. Szo
[132] J. A. Thas, Some results concerning {(q +1)(n1), n}-arcs and {(q +1)(n
1) + 1, n}-arcs in finite projective planes of order q, J. Combinatorial Theory
Ser. A 19 (1975), 228232.
[133] J. A. Thas, A combinatorial characterization of Hermitian curves, J. Alg.
Combin. 1 (1992), 97-102.
[134] J.A. Thas, P.J. Cameron, A. Blokhuis, On a generalization of a theorem of B. Segre. Geom. Dedicata 43 (1992), 299305.
[135] G. Turnwald, A new criterion for permutation polynomials, Finite Fields
and Appl. 1 (1995), 6482.
[136] A. Weil, Sur les Courbes Algebrique et les varietes qui sen deduisent,
Actualites Scientifiques et Industrielles 1041, Herman & Cie, Paris, 1948.
254
Bibliography
[137] Zs. Weiner, On (k, pe )-arcs in Desarguesian planes, Finite Fields and Appl.
10 (2004), 390-404.
[138] Zs. Weiner, Small point sets of PG(n, q) intersecting each k-space in 1
modulo q points, Innovations in Incidence Geometry, 1, (2005), 171180.

[139] F. Wettl, On the nuclei of a finite projective plane, J. of Geometry 30
(1987), 157163.

Poly 08 Feb

Uploaded by

Copyright:

Available Formats

Poly 08 Feb

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Poly 08 Feb

Uploaded by

Copyright:

Available Formats

POLYNOMIALS IN FINITE GEOMETRY

A most efficient way of investigating combinatorially defined pointsets of projective

Definitions, basic notation . . . . . . . . . . . . . . . . . . . . . . .

Finite fields and polynomials . . . . . . . . . . . . . . . . . . . . .

Some basic facts . . . . . . . . . . . . . . . . . . . . . . . .

Self-dual normal bases . . . . . . . . . . . . . . . . . . . . .

Additive, linearized polynomials . . . . . . . . . . . . . . .

The random-like behaviour . . . . . . . . . . . . . . . . . .

Representations of affine and projective geometries . . . . . . . . .

The Redei polynomial and its derivatives

The Redei polynomial . . . . . . . . . . . . . . . . . . . . .

Hasse derivatives of the Redei polynomial . . . . . . . . . .

The affine polynomial and its derivatives . . . . . . . . . . .

The projective polynomial . . . . . . . . . . . . . . . . . . .

Symmetric polynomials and invariants of subsets . . . . . . . . . .

The Newton formulae . . . . . . . . . . . . . . . . . . . . .

The invariants uf , vf and wf . . . . . . . . . . . . . . . . .

The generalized formulae . . . . . . . . . . . . . . . . . . .

Elements of algebraic geometry . . . . . . . . . . . . . . . . . . . .

Conditions implying linear (or low-degree) components . . .

Finding the missing factors, removing the surplus factors . . . . . .

Lines meeting a pointset . . . . . . . . . . . . . . . . . . . . . . . .

Prescribing the intersection numbers . . . . . . . . . . . . .

Another immediate corollary . . . . . . . . . . . . . . . . .

Sets with constant intersection numbers mod p . . . . . . . . . . .

Sets with intersection numbers 0 mod r . . . . . . . . . . .

Small and large super-Vandermonde sets . . . . . . . . . . .

Sets without tangents: where Segre meets Redei . . . . . . . . . . . 102

The proof using Segres method . . . . . . . . . . . . . . . . 103

Untouchable sets in Galois planes of even order . . . . . . . 109

When D is (contained in) a subgroup of GF(q) . . . . . . . 112

When D is (contained in) a subgroup of q+1 1 GF(q 2 ) . 115

Small blocking sets of Redei type are linear . . . . . . . . . 116

directions: affine linear pointsets

Over prime fields: the Gacs method . . . . . . . . . . . . . . . . . . 122

Permutation polynomials . . . . . . . . . . . . . . . . . . . 124

Directions in the plane . . . . . . . . . . . . . . . . . . . . . 125

Linear combinations of three permutation polynomials . . . 131

Projective linear pointsets . . . . . . . . . . . . . . . . . . . . . . . 134

Blocking sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

One curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

Three new curves . . . . . . . . . . . . . . . . . . . . . . . . 144

Three old curves . . . . . . . . . . . . . . . . . . . . . . . . 147

Small blocking sets . . . . . . . . . . . . . . . . . . . . . . . 149

Multiple blocking sets . . . . . . . . . . . . . . . . . . . . . . . . . 154

A t (mod p) result . . . . . . . . . . . . . . . . . . . . . . . 155

Stability theorems for blocking sets: the prime case . . . . . 167

Stability theorems for blocking sets: the non-prime case . . 170

Stability theorem for sets of even type . . . . . . . . . . . . 176

Number of lines meeting a (q + 2)-set . . . . . . . . . . . . 179

Jamison, Brouwer-Schrijver, affine blocking sets . . . . . . . . . . . 180

Applications of the punctured Nullstellensatz . . . . . . . . 184

Redei polynomials in three indeterminates . . . . . . . . . . 189

Redei polynomials in many indeterminates

Affine blocking sets . . . . . . . . . . . . . . . . . . . . . . . 195

Nuclei and affine blocking sets . . . . . . . . . . . . . . . . . . . . . 198

Lower nuclei . . . . . . . . . . . . . . . . . . . . . . . . . . 201

Internal nuclei . . . . . . . . . . . . . . . . . . . . . . . . . 203

Higher dimensions . . . . . . . . . . . . . . . . . . . . . . . 203