University of Sussex

Luca Giuzzi

Hermitian varieties over finite fields

Thesis submitted for the award of the
degree of Doctor of Philosophy
November 2000
I hereby declare that this thesis has not been submitted, either in
the same or different form, to this or any other University for a
degree.

Signature:
Acknowledgements
This research work has been funded by an INDAM scholarship of the Italian ‘Istituto
Nazionale di Alta Matematica’ and by a bursary of the Engineering and Physical Sciences Re-
search Council, which covered tuition fees. I am most grateful to these organisations for their
support.
Among the several people which have helped me during the work on this thesis I would
like to explicitly thank my supervisor, Prof. J.W.P. Hirschfeld, both for his constant advice on
the research and for having taught me how to write scientific prose. I would like to thank also
Professors G. Korchmáros and A. Cossidente from Potenza for very helpful discussions on the
topics investigated here.
I would like to express once more my gratitude towards my parents for their continuous
support and understanding.
Finally, I wish to thank in a special way Jennifer Huynh for her kindness and our excellent
friendship, and all the other people that, in one way or another, have been near me during all
these years.
For Angling may be said to be so much like the
Mathematicks, that it can ne’er be fully learnt;
at least not so fully, but that there will still be
more new experiments left for the tryal of other
men that succeed us.
Izaak Walton, [WC76]
Summary

In this work, submitted for the award of the title of Doctor of Philosophy, we have investigated
some properties of the configurations arising from the intersection of Hermitian varieties in a
finite projective space.
Chapter 1 introduces some background material on the theory of finite fields, projective
spaces, Hermitian varieties and classical groups.
Chapter 2 deals with the 2-dimensional case. In Section 2.1, we present the the point-line
classification of the intersections, due to Kestenband. In Section 2.2, we determine the full
linear collineation group stabilising any of the configurations of 2.1 and we prove that if two
configurations have the same point-line structure, then they are in fact projectively equivalent.
A new and simplified proof of the group theoretical characterization of the Hermitian curve as
√
the unital stabilised by a Singer subgroup of order q − q + 1 closes the chapter in Section 2.3.
In Chapter 3 we study the 3-dimensional case. In Section 3.1 we determine what incidence
configurations fulfill the combinatorial properties required in order to be the intersection of
Hermitian surfaces. Section 3.2 presents some further general remarks on linear systems of
Hermitian curves and extensive computations on 4 × 4 Hermitian matrices. In Section 3.3, we
produce models that realize all the possible intersection configurations in dimension 3.
Chapter 4 is organized in two independent sections. In Section 4.1 we provide a general
formula to determine the list of possible sizes of Hermitian intersections in PG(n, q). The for-
mula itself has been obtained by studying the geometry of the set H of all singular Hermitian
hypersurfaces of PG(n, q). Such a set is endowed with the structure of an algebraic hypersur-
face of PG(n2 + 2n, q) of degree n + 1; the locus of the singular points of H is analyzed
in detail. In Section 4.2 we introduce some computer code in order to explicitly compute the
intersection configurations arising in PG(n, q).

iv
Contents

Introduction 1

1 Preliminary results 3
1.1 Permutation Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Transitivity and regularity . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.3 Similarities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.4 Actions and representations . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Finite fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Definitions and models . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.2 Automorphisms and extensions . . . . . . . . . . . . . . . . . . . . . 7
1.2.3 Trace and norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Projective spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.1 Incidence structures . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.2 Algebraic constructions . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3.3 Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.4 Singer cyclic groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4 Polynomials and matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.5 Varieties over finite fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5.1 Polynomials as functions over projective spaces . . . . . . . . . . . . . 18
1.5.2 Algebraic sets and ideals . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.5.3 Number of rational points: zeta functions . . . . . . . . . . . . . . . . 20
1.5.4 Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.5.5 Non-singular varieties . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.6 Unitary Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.6.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.6.2 Subspaces and forms . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.6.3 Hermitian groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.6.4 Generators of the unitary group . . . . . . . . . . . . . . . . . . . . . 26

v
CONTENTS

1.6.5 Morphisms and automorphisms . . . . . . . . . . . . . . . . . . . . . 26

1.7 Hermitian Forms and Hermitian varieties . . . . . . . . . . . . . . . . . . . . 27
1.7.1 Hermitian matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.7.2 Hermitian forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.7.3 Hermitian hypersurfaces: definitions . . . . . . . . . . . . . . . . . . . 30
1.7.4 Hermitian sub-varieties and points . . . . . . . . . . . . . . . . . . . . 32
1.7.5 Hermitian varieties and unitary groups . . . . . . . . . . . . . . . . . . 33
1.7.6 Polarities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.7.7 Subspaces in Hermitian varieties . . . . . . . . . . . . . . . . . . . . . 36
1.7.8 Incidence properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

2 The 2-dimensional case: Hermitian curves 37

2.1 Classification of intersections in dimension 2 . . . . . . . . . . . . . . . . . . 37
2.1.1 Incidence classification . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.1.2 Outline of the proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.2 Groups of the intersection of two Hermitian curves . . . . . . . . . . . . . . . 42
2.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.2.2 A non-canonical model of PG(2, q) . . . . . . . . . . . . . . . . . . . 43
2.2.3 Equations for the non-singular Hermitian curve . . . . . . . . . . . . . 44
2.2.4 Groups preserving the intersection of two Hermitian curves . . . . . . 45
Class I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Class II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Class III . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Class IV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Class V . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
Class VI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
Class VII . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.3 A group-theoretic characterization of Hermitian curves as classical unitals . . . 58
2.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.3.2 A result on classical unitals . . . . . . . . . . . . . . . . . . . . . . . . 58
2.3.3 Proof of Theorem 2.3.1 . . . . . . . . . . . . . . . . . . . . . . . . . . 59

3 The 3-dimensional case: Hermitian surfaces 61

3.1 Description of incidence configurations in dimension 3 . . . . . . . . . . . . . 61
3.1.1 Some general remarks on cones . . . . . . . . . . . . . . . . . . . . . 63
3.1.2 Pencils whose degenerate surfaces have all rank 3 . . . . . . . . . . . . 65
3.1.3 Pencils whose degenerate surfaces have all rank 2 . . . . . . . . . . . . 68
3.1.4 Pencils whose degenerate surfaces have ranks 2 and 3 . . . . . . . . . . 71
3.1.5 Pencils with r1 (Γ) ≥ 1 . . . . . . . . . . . . . . . . . . . . . . . . . . 73

vi
CONTENTS

3.2 Hermitian matrices and polynomials . . . . . . . . . . . . . . . . . . . . . . . 73

3.2.1 Some general considerations on Hermitian pencils . . . . . . . . . . . 78
Further linear algebra observations . . . . . . . . . . . . . . . . . . . . 79
3.2.2 Case I: 3 or 4 distinct roots . . . . . . . . . . . . . . . . . . . . . . . . 81
3.2.3 Case II: 2 distinct roots . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.2.4 Case III: 1 root . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.2.5 Case IV: some notes when the factorisation contains irreducibles . . . . 87
3.3 Construction of the incidence configurations . . . . . . . . . . . . . . . . . . . 90
3.3.1 Pencils with degenerate surfaces of rank 3 only . . . . . . . . . . . . . 90
3.3.2 Pencils with degenerate surfaces of rank 2 only . . . . . . . . . . . . . 92
3.3.3 Pencils whose degenerate surfaces have rank 2 and 3 . . . . . . . . . . 93

4 General results 97
4.1 The cardinality formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.1.2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.1.3 The determinantal variety H . . . . . . . . . . . . . . . . . . . . . . 101
4.1.4 Action of PGL(n + 1, q) on H . . . . . . . . . . . . . . . . . . . . . 106
4.1.5 The order formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.1.6 Some further remarks on H . . . . . . . . . . . . . . . . . . . . . . . 110
4.2 Explicit computations and algorithms . . . . . . . . . . . . . . . . . . . . . . 112
4.2.1 The computer code: general remarks . . . . . . . . . . . . . . . . . . . 112
4.2.2 Initialization:
main and param . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.2.3 Auxiliary functions:
Projective and Helpers . . . . . . . . . . . . . . . . . . . . . . 114
4.2.4 Preliminary computations:
Prelim.gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
4.2.5 Unitary group construction:
Unitary.gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.2.6 Orbit computation:
Orbits.gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.2.7 Intersection and results output:
Post-orb.gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.2.8 Generators of the projective unitary group . . . . . . . . . . . . . . . . 138
Quasi-symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
Ishibashi’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
Construction of the matrices . . . . . . . . . . . . . . . . . . . . . . . 142

vii
CONTENTS

4.2.9 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

Bibliography 147

viii
List of Figures

1.1 Veblen-Young configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1.2 Desargues configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Pappus configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.1 Possible configurations for the 2-dimensional case . . . . . . . . . . . . . . . . 41

ix
List of Tables

2.1 Possible intersection numbers for non-degenerate Hermitian Curves. . . . . . . 38

2.2 Minimal polynomials corresponding to given rank sequences in the 2-dimensional
case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.3 Canonical forms for 3 by 3 Hermitian matrices. . . . . . . . . . . . . . . . . . 40
2.4 Equations for the non-degenerate Hermitian curve . . . . . . . . . . . . . . . . 44

3.1 Possible intersection numbers for Hermitian surfaces: non-degenerate pencil. . 62

3.2 Possible intersection numbers for Hermitian surfaces: degenerate pencil; r3 6=
0, r4 = 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.3 Possible intersection numbers for Hermitian surfaces: degenerate pencil; r2 6=
0, r3 = r4 = 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.4 Possible intersections E between a cone and a non degenerate Hermitian sur-
face; vertex not in the intersection. . . . . . . . . . . . . . . . . . . . . . . . . 64
3.5 Indices for the intersection of a cone and a Hermitian surface. . . . . . . . . . . 65
3.6 Possible incidence classes for two non-degenerate Hermitian surfaces: Γ con-
tains degenerate surfaces of rank 3 only. . . . . . . . . . . . . . . . . . . . . . 68
3.7 Possible incidence classes for two non-degenerate Hermitian surfaces: Γ con-
tains degenerate surfaces of rank 2 only. . . . . . . . . . . . . . . . . . . . . . 70
3.8 Indices for the intersection of Hermitian surfaces of rank 2 and of rank 3. . . . 71
3.9 Possible incidence classes for two non-degenerate Hermitian surfaces: Γ con-
tains degenerate surfaces of ranks 2 and 3. . . . . . . . . . . . . . . . . . . . . 74
√
3.10 Canonical forms for the 3-dimensional case: MH (x) splits over GF( q); 2 or
3 distinct roots. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
√
3.11 Canonical forms for the 3-dimensional case: MH (x) splits over GF( q); 1 or
4 roots. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
√
3.12 Canonical forms for the 3-dimensional case: MH (x) does not split over GF( q);
degree 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

4.1 Intersection numbers for non-degenerate Hermitian varieties in dimension 4. . . 110

4.2 Dimension of the varieties O ≥t for small n . . . . . . . . . . . . . . . . . . . . 111

x
Introduction

A projective space PG(n, q) admits at most three types of polarity: orthogonal, symplectic and
unitary. The absolute points of an orthogonal polarity constitute a non-degenerate quadric in
PG(n, q); for a symplectic polarity, all the points of PG(n, 2t ) are absolute; the locus of all
absolute points of a unitary polarity is a non-degenerate Hermitian variety.
Non-degenerate Hermitian varieties are unique in PG(n, q) up to projectivities. However,
two distinct Hermitian varieties might intersect in many different configurations. Our aim in
this thesis is to study such configurations in some detail.
In Chapter 1 we introduce some background material on finite fields, projective spaces,
collineation groups and Hermitian varieties.
Chapter 2 deals with the two-dimensional case. Kestenband has proven that two Hermitian
curves may meet in any of seven point-line configurations. In Section 2.1, we present this
classification. In Section 2.2, we verify that any two configurations belonging to the same class
are, in fact, projectively equivalent and we determine the linear collineation group stabilizing
each of them. Such a group is usually quite large and is transitive on almost all the points of
the intersection. A subset U of PG(2, q) such that any line of the plane meets U in either 1 or
√
q + 1 points is called a unital. A Hermitian curve is a classical unital. However, there exist
non-classical unitals as well. In Section 2.3 we present a short proof of a characterisation of the
√
Hermitian curve as the unital stabilized by a Singer subgroup of order q − q + 1.
In Chapter 3, we describe the point-line-plane configurations arising in dimension 3 from
intersecting two Hermitian surfaces. Our approach consists first in determining some combi-
natorial properties the configurations have to satisfy and then in actually constructing all the
possible cases. Section 3.1 presents the list of all possible intersection classes; after some more
technical results in Section 3.2, in Section 3.3 we construct linear systems of Hermitian sur-
faces yielding the wanted configurations for any class. In this chapter we deal with intersections
√
which contain at least q + 1 points on a line.
Chapter 4 is divided into two independent sections: in Section 4.1 we study the determinan-
√
tal variety of all the (n + 1) × (n + 1) Hermitian matrices as a hypersurface of PG(n2 + 2n, q).
From the study of such a variety we are able to determine the list of all possible intersection
sizes for any dimension n. In Section 4.2 we present some computer code in order to produce
pencils of Hermitian varieties in PG(n, q). This code, however, is able to provide useful re-

1
sults only for small values of n and q. Some possible improvements, both from the theoretic
standpoint and the computational one, are suggested.

2
Chapter 1

Preliminary results

The aim of this first chapter is to recall some known results that will be used throughout the
thesis.

1.1 Permutation Groups

Our main reference for the theory of finite permutation groups is [Wie64]. For general results
on group theory, [Rob80] has been used.

1.1.1 Definitions
Definition 1.1.1. Let X be a non-empty set. A permutation of X is a bijective mapping of X
into itself. The set of all permutations of X, together with map composition forms a group SX ,
the symmetric group on X.

Definition 1.1.2. A permutation group G on a set X, or an X-permutation group, is any sub-

group of SX . The degree of a permutation group G is the cardinality of the set X it acts upon.

Definition 1.1.3. Let G be an X-permutation group and define † as the equivalence relation on
X given by
x † y ⇐⇒ ∃σ ∈ G : x = σ(y).
The equivalence classes of † are the orbits of G on X. The orbit of any x ∈ X will be denoted
by the symbol xG .

1.1.2 Transitivity and regularity

Definition 1.1.4. A permutation group G is transitive on X if and only if it has only one orbit,
namely X itself.

3
CHAPTER 1. PRELIMINARY RESULTS
1.1. PERMUTATION GROUPS

Definition 1.1.5. For any X-permutation group G and for any Y ⊆ X, the stabilizer of Y in G
is the set
StG (Y ) := {σ ∈ G : ∀y ∈ Y : σ(y) ∈ Y }.
A group G is semi-regular if for any x ∈ X, StG (x) = 1; G is regular if it is semi-regular and
transitive.

Definition 1.1.6. Let n be a positive integer. A permutation group G is 1-transitive on X if and

only if it is transitive on X. A group G is n-transitive on a set X if and only if for any x ∈ X,
StG (x) is (n − 1)-transitive on X \ {x}.

Lemma 1.1.7. Assume that G is a n-transitive group on X. Then,

(i) for any x ∈ X, StG (x) is (n − 1)-transitive on X \ {x};

(ii) given two tuples lx = (x1 , . . . , xn ), ly = (y1 , . . . , yn ) of elements of X, there exists a

permutation σ ∈ G such that σ(lx ) = ly .

Definition 1.1.8. A group G is strictly n-transitive on X if it is n-transitive and, given any two
different n-tuples lx , ly of distinct elements of X, there exists exactly one σ ∈ G such that
σ(lx ) = ly .

Lemma 1.1.9. Let G be a permutation group on a set X. Then, for any x ∈ X,

|G| = |xG ||StG (x)|.

1.1.3 Similarities
Definition 1.1.10. Let G be an X-permutation group and H be a Y -permutation group. A
similarity between G and H is a pair (α, β), where

(i) α is an isomorphism between G and H;

(ii) β is a bijection between X and Y ;

(iii) for any σ in G and for any x ∈ X,

β(xσ ) = (β(x))α(σ) .

Two permutation groups are similar if there is a similarity between them.

The sets X and Y have to be isomorphic in order for G and H to be similar; still, |X| = |Y |
and G ' H is not sufficient for G and H to be similar.

4
CHAPTER 1. PRELIMINARY RESULTS
1.2. FINITE FIELDS

1.1.4 Actions and representations

Definition 1.1.11. Let G be a group and let X be a non-empty set. A right action of G on X is
a function ρ : X × G → X such that

ρ(x, g1 g2 ) = ρ(ρ(x, g1 ), g2 ).

In an analogous way, a left action of G on X is defined as a function λ : G × X → X such that

λ(g1 g2 , x) = λ(g1 , λ(g2 , x)).

Definition 1.1.12. Let G be any group and let X be a set. Any group homomorphism γ :
G → SX is a permutation representation of G on X and it induces a right action on X via the
mapping (x, g) → γ(g)x.
The cardinality of X is the degree of the representation γ. If ker γ = {1}, then the repre-
sentation is faithful; a representation is transitive if its image in SX is transitive; γ is regular if
its image in SX acts regularly on X.

1.2 Finite fields

The reference text for most of the definitions and results presented in this section is [LN97].

1.2.1 Definitions and models

Definition 1.2.1. A ring (R, +, ·) is an integral domain if, for all x, y ∈ R, xy = 0 implies
x = 0 or y = 0; if (R \ {0}, ·) is a group, then the ring R is a division ring. A division ring R
in which the group R? = (R \ {0}, ·) is Abelian is a field; a division ring in which the group R?
is non-commutative is a skew-field.

Theorem 1.2.2 (Wedderburn, 1905). Every finite division ring is a field.

Definition 1.2.3. The characteristic of a ring (R, +, ·) is the smallest positive integer char(R) =
n such that for all r ∈ R:
nr := r| + .{z
. . + r} = 0.
n times
If no such an n exists, R is said to have characteristic 0.

Definition 1.2.4. A subfield K of a field (F, +, ·) is a subset of F which is closed under the
field operations. The field (F, +, ·) is called an extension (field) of (K, + |K , · |K ). The symbol
K ≤ F is used to denote that F is an extension field of K or, equivalently, that K is a subfield
of F .

5
CHAPTER 1. PRELIMINARY RESULTS
1.2. FINITE FIELDS

Any field contains itself, the empty set and the set consisting of its zero only as subfields.
Those subfields are called trivial.

Definition 1.2.5. A field is prime if it does not contain any proper subfield.

Lemma 1.2.6. Any subfield has the same characteristic as its parent field.

Definition 1.2.7. Let K be a subfield of F and assume f ∈ F . The extension of K by f is the

smallest subfield K(f ) of F such that,

K ∪ {f } ⊆ K(f ) ⊆ F.

Definition 1.2.8. Given a field K, its algebraic closure K is the smallest field containing K
such that any polynomial
Xn
ki xi ∈ K[x]
i=0

splits to linear factors in K[x].

Lemma 1.2.9. Let Z be the ring of integers, p a prime number and let pZ be the ideal generated
by p in Z; then, Z/(pZ) is a finite field containing p elements.

Proof. Let Fp := {0, . . . , p − 1} ⊂ Z, and let φ be the projection mapping

½
Fp → Z/(pZ)
φ:
x → x + (pZ).

Define, for any b, c ∈ F ,

b + c := φ−1 (φ(b + c)); b · c := φ−1 (φ(bc)).

The structure (Fp , +, ·) is closed under its operations and every non-zero element admits an
inverse. It follows that F is a field.

Definition 1.2.10. The field (Fp , +, ·) constructed in the previous lemma is called the Galois
field with p elements. It will be denoted by the symbol

GF(p) := (Fp , +, ·).

Definition 1.2.11. Let (F, +, ·) and (G, ⊕, ¯) be two fields. A morphism between F and G is
a mapping φ : F → G preserving the algebraic structure; that is, for all p, q ∈ F :

(i) φ(p−1 ) = φ(p)−1 ,

(ii) φ(−p) = ªφ(p),

6
CHAPTER 1. PRELIMINARY RESULTS
1.2. FINITE FIELDS

(iii) φ(pq) = φ(p) ¯ φ(q),

(iv) φ(p + q) = φ(p) ⊕ φ(q).

A necessary condition for a morphism to exist between two fields F and G is that char(F ) =
char(G).
Monomorphisms, epimorphisms, isomorphisms and automorphisms are defined the usual
way.

Theorem 1.2.12. Any field K contains a prime subfield F which is either isomorphic to the
field of the rational numbers Q or to GF(p) for some prime p, according as the characteristic
of K is 0 or p.

Theorem 1.2.13. Let F be a finite field. Then, the cardinality of F is pn , where p = char(F ) is
a prime number and n is an integer. The integer n is called the degree of F over its prime field.

As a matter of fact, the converse is true as well.

Theorem 1.2.14. For every prime p and for every integer n there exists a finite field F of order
pn . All fields of given order q = pn are isomorphic to the splitting field of the polynomial
(xq − x) over GF(p). This field F will be written as

GF(q) := GF(p)[x]/(xq − x).

Definition 1.2.15. Given any finite field (F, +, ·), its multiplicative group F ? is cyclic. A gen-
erator of F ? is a primitive element of F .

1.2.2 Automorphisms and extensions

Lemma 1.2.16. Let F be a finite extension field of GF(q). Then, there exists an integer m > 0
such that F = GF(q m ). This integer is the degree of F over GF(q) and is written as

[F : GF(q)] := m.

Definition 1.2.17. Let GF(q m ) be an extension field of GF(q), and let α ∈ GF(q m ). The
m−1
elements α, αq , . . . , αq are called the conjugates of α with respect to GF(q).

Definition 1.2.18. An automorphism of GF(q m ) over GF(q) is a field automorphism of GF(q m )

which fixes all the elements of GF(q). The group of all these automorphisms is the Galois group
of GF(q m ) over GF(q) and it is denoted by the symbol

Gal(GF(q m ) : GF(q)).

Lemma 1.2.19. For any t ∈ GF(q),

tq − t = 0.

7
CHAPTER 1. PRELIMINARY RESULTS
1.2. FINITE FIELDS

Theorem 1.2.20. The elements of Gal(GF(q m ) : GF(q)) can be described as the automor-
phisms σ0 , σ1 , . . . , σm−1 given by
½
GF(q m ) → GF(q m )
σj : j
x → σj (x) = xq .
The automorphism σ1 which generates the Galois group is called the Frobenius automorphism.

Lemma 1.2.21. Let K be a field and let σ be one of its automorphisms. Then, the set FixK σ is
a subfield of K. This subfield is called the fixed field of σ in K.

Corollary 1.2.22. The following equality holds:

| Gal(GF(q m )) : GF(q))| = m = [GF(q m ) : GF(q)].

Definition 1.2.23. An automorphism σ is involutory if

(i) σ is not the identity;

(ii) σ 2 is the identity.

For any prime-power q, the field F = GF(q 2 ) admits one and only one non-identity auto-
morphism over GF(q), namely the mapping sending x → xq . Since | Gal(F : GF(q))| = 2,
this is an involutory automorphism of F .

Lemma 1.2.24. Let K be a field and let σ be an involution of K. Then, there exist a subfield
K 0 ≤ K and an element i ∈ K such that,

(i) [K : K 0 ] = 2;

(ii) K 0 is fixed by σ;

(iii) K = K 0 (i).

1.2.3 Trace and norm

Let F and K be finite fields with K ≤ F . From Theorem 1.2.14, it is possible to assume
without loss of generality that F = GF(q m ), K = GF(q) with q, m integers and q a prime
power.

Definition 1.2.25. The trace over K of an element x ∈ F is given by

X m−1
TF/K (x) := σ(x) = x + xq + . . . + xq .
σ∈Gal(F :K)

The trace of x over the prime subfield of F is the absolute trace of x and is simply denoted as
TF (x).

8
CHAPTER 1. PRELIMINARY RESULTS
1.2. FINITE FIELDS

Theorem 1.2.26. For any x, y ∈ F and f, g ∈ K, the trace TF/K satisfies the following
properties.

(i) TF/K (x + y) = TF/K (x) + TF/K (y);

(ii) TF/K (f x) = f TF/K (x);

(iii) TF/K is a linear transformation from F onto K, where both F and K are viewed as vector
spaces over K;

(iv) TF/K (f ) = mf ;

(v) TF/K (xq ) = TF/K (x).

Theorem 1.2.27. Let K be a finite field, and let F a finite extension of K. The linear trans-
formations from F into K, both seen as vector spaces over K, are exactly the mappings Lt for
t ∈ F given by ½
F → K
Lt :
x → TF/K (tx).
Also, Lt 6= Lt0 when t 6= t0 .

Theorem 1.2.28 (Composition of traces). Assume K ≤ F ≤ E to be all finite fields. Then, for
any x ∈ E,
TE/K (x) = TF/K (TE/F (x)).

Definition 1.2.29. The norm over K of an element x ∈ F is given by

Y m
NF/K (x) := σ(x) = x(q −1)/(q−1) .
σ∈Gal(F :K)

The norm of x over the prime subfield of F is the absolute norm of x and is simply denoted by
NF (x).

Theorem 1.2.30. For any x, y ∈ F and f, g ∈ K, the norm NF/K satisfies the following
properties.

(i) NF/K (xy) = NF/K (x)NF/K (y);

(ii) NF/K maps F onto K and F ? onto K ? ;

(iii) NF/K (f ) = f m ;

(iv) NF/K (xq ) = NF/K (x).

Theorem 1.2.31 (Composition of norms). Assume K ≤ F ≤ E to be all finite fields. Then, for
any x ∈ E:
NE/K (x) = NF/K (NE/F (x)).

9
CHAPTER 1. PRELIMINARY RESULTS
1.3. PROJECTIVE SPACES

1.3 Projective spaces

Good references for the results presented in this section are [Dem68], [KSW73] and [BR98].

1.3.1 Incidence structures

Definition 1.3.1. A triple (P, L, I) is a linear space if

(i) P, L are non-empty sets;

(ii) I is a symmetric relation on (P × L) ∪ (L × P );

(iii) for all distinct x, y ∈ P there exists exactly one element l := xy of L such that (x, l) ∈ I
and (y, l) ∈ I.

The elements of P are the points of (P, L, I); those of L are called the lines of the incidence
structure. The relation I is named incidence relation.

Definition 1.3.2. In a linear space (P, L, I) two lines l, m intersect if there exist a point x ∈ P
such that (x, l) ∈ I and (x, m) ∈ I. The notation

l ∩ m := {x ∈ P : (x, l) ∈ I and (x, m) ∈ I}

will be used.

Definition 1.3.3. Three distinct points x, y, z in a linear space (P, L, I), are called collinear if
(x, yz) ∈ I.

Definition 1.3.4. A mapping φ from a linear space (P, L, I) to a linear space (P 0 , L0 , I 0 ) is a

collineation (or isomorphism) if

(i) φ(p) ∈ P 0 , for all p ∈ P ;

(ii) φ(l) ∈ L0 , for all l ∈ L;

(iii) φ is bijective on P ;

(iv) the incidence relation is preserved, that is,

(p, l) ∈ I ⇐⇒ (φ(p), φ(l)) ∈ I 0 .

Definition 1.3.5. Let (P, L, I) be a linear space. A subset U of P is a linear set if, given any
two distinct points x, y ∈ U ,
t ∈ xy ⇒ t ∈ U.
The fact that U is a linear subset of P is written as U ≤ P .

10
CHAPTER 1. PRELIMINARY RESULTS
1.3. PROJECTIVE SPACES

Lemma 1.3.6. Assume (P, L, I) to be a linear space. For any linear subset U of P , let

(i) LU := {l ∈ L : ∃ x, y ∈ U such that x 6= y, (x, l) ∈ I, (y, l) ∈ I};

(ii) IU := I |U ×LU ∪LU ×U .

Then, (U, LU , IU ) is a linear space.

Definition 1.3.7. Given two linear spaces (P, L, I) and (P 0 , L0 , I 0 ), we say that (P 0 , L0 , I 0 ) is a
subspace of (P, L, I) with support P 0 if

(i) P 0 ≤ P ;

(ii) L0 = LP 0 ;

(iii) I 0 = IP 0 .

Definition 1.3.8. Given a linear space (P, L, I) and a set V ⊆ P , the linear set V spanned by
V is given by \
V := {T ≤ P : V ⊆ T }.

Lemma 1.3.9. For any V ⊆ P , the linear set spanned by V is characterised as follows:

V := {x ∈ P : ∃h, k ∈ V such that h 6= k, (x, hk) ∈ I}.

Definition 1.3.10. A projective space (P, L, I) is a linear space in which the following axioms
are satisfied:

(i) (Veblen-Young) If a, b, c, d ∈ P are distinct points, then

ab ∩ cd 6= ∅ ⇒ ac ∩ bd 6= ∅;

(ii) any line is incident with at least three points;

(iii) there are at least two lines.

A finite projective space is a projective space in which the set of points is finite.

Definition 1.3.11. A projective plane (P, L, I) is a projective space in which any two lines
intersect.

Definition 1.3.12. The dimension of a projective space P = (P, L, I) is the number

dim P := inf{|U | : U ⊆ P and U = P } − 1.

11
CHAPTER 1. PRELIMINARY RESULTS
1.3. PROJECTIVE SPACES

b
a

d
c

Figure 1.1: Veblen-Young configuration

Definition 1.3.13. Six distinct points xi , yi with i ∈ {1, 2, 3} in a linear space (P, L, I) consti-
tute a Desargues configuration if

(i) there exists c ∈ P such that (xi , cyi ) ∈ I and (yi , cxi ) ∈ I for all i;

(ii) no three of the points c, x1 , x2 , x3 and c, y1 , y2 , y3 are collinear;

(iii) the points pij := xi xj ∩ yi yj are collinear.

Definition 1.3.14. A projective space (plane) P is Desarguesian if, for any choice of six points
satisfying (i) and (ii) of the Desargues configuration, condition (iii) holds.

Definition 1.3.15. Six distinct points gi , hi for i ∈ {1, 2, 3} in a linear space (P, L, I) constitute
the Pappus configuration if

(i) the gi are all collinear;

(ii) the hi are all collinear;

(iii) the line G := g1 g2 meets the line H := h1 h2 in a point c that is distinct from any of the gi
and any of the hi ;

(iv) given qij := gi hj ∩ gj hi , then (q12 , q13 q23 ) ∈ I.

Definition 1.3.16. A projective space (plane) P is Pappian if, for any choice of six points
satisfying conditions (i), (ii) and (iii) of Pappus configuration, condition (iv) holds as well.

12
CHAPTER 1. PRELIMINARY RESULTS
1.3. PROJECTIVE SPACES

y
1

x1
p
23
c y2
x2
p
x 12
3
y
3

p
13

Figure 1.2: Desargues configuration

Theorem 1.3.17. Any projective space of dimension at least 3 is Desarguesian.

Still, there exist non-Desarguesian projective planes.

Theorem 1.3.18 (Hessenberg). A finite projective plane P is Pappian only if it is Desarguesian.

1.3.2 Algebraic constructions

Theorem 1.3.19. Let K be a division ring and let V be a (left) vector space over K of dimension
n ≥ 3. Then, the triple (P, L, ⊆) where

(i) P is the set of all 1-dimensional (left) vector subspaces of V ,

(ii) L is the set of all 2-dimensional (left) vector subspaces of V ,

(iii) ‘⊆’ is the natural inclusion,

is a Desarguesian projective space of dimension n − 1.

Definition 1.3.20. The symbol P(V ) is used to denote a projective space constructed from a
vector space V as in Theorem 1.3.19.

Theorem 1.3.21 (Projective derivation of a vector space). Let V be a vector space of dimension
n ≥ 3 over a field K. Define

(i) 0 := (0, . . . , 0) ∈ V ;

(ii) V ? := V \ {0};

13
CHAPTER 1. PRELIMINARY RESULTS
1.3. PROJECTIVE SPACES

G
g 3

g2
g1

q 13
c q
12 q 23

h
1
h 2 h 3
H
Figure 1.3: Pappus configuration

(iii) Vi be the set of i-dimensional subspaces of V ;

(iv) P := V ? /K ? := {T \ 0 : T ∈ V1 };

(v) L := { KX? : X ∈ V2 }.

Then, PV := (P, L, ∈) is a (n − 1)-dimensional Pappian projective space which is isomorphic

to P(V ).

Definition 1.3.22. Let PVn be the projective derivation of a vector space V of dimension n over
a field K. Then, V is the underlying vector space of (P, L, ∈). The projection map
½ ?
V → V ? /K ?
P:
x → [x] := K ? x

is called the projectivisation of V .

Theorem 1.3.23. Let V be a vector space over a division ring K; then P(V ) is Pappian if and
only if K is a field.

Definition 1.3.24. The symbol PG(n, K) := PK n+1 ' P(K n+1 ) denotes the Pappian projec-
tive space obtained by projectivisation from a vector space of dimension n + 1 over K.

Definition 1.3.25. Let (P, L, I) = PG(n, K) be a Pappian projective space. For any given t
such that −1 ≤ t ≤ n, the symbol PGt (n, K) denotes all projective subspaces of PG(n, K)
with dimension t. In particular PG−1 (n, K) = ∅, PG0 (n, K) = P and PGn (n, K) = {P }.

14
CHAPTER 1. PRELIMINARY RESULTS
1.3. PROJECTIVE SPACES

Theorem 1.3.26 (First representation theorem). Given any Desarguesian projective space P of
dimension n, there exist a division ring K and a vector space V of dimension n + 1 over K
such that P is isomorphic to P(V ).

Corollary 1.3.27. Let P be a Pappian projective space of dimension n. Then, there exists a
field K such that P ' PG(n, K).

Proof. By Theorem 1.3.18, P is Desarguesian; Theorem 1.3.26 guarantees that P = P(V ),

with V a vector space of dimension n over a division ring K; by Theorem 1.3.23, K is a field.
Hence, P ' PG(n, K).

Corollary 1.3.28. Let P be a finite projective space of dimension at least 3. Then, P is Pappian.

Proof. From Theorem 1.3.17, P is Desarguesian; hence, P = P(V ) for some vector space V
over a finite division ring K; by Theorem 1.2.2, K is a field. It follows that P is isomorphic to
PG(n, q) := PG(n, K). Hence, P is Pappian.

Theorem 1.3.29 (Tecklenburg). A Desarguesian finite plane is Pappian.

This result can be proved as in Corollary 1.3.28 or in a more direct geometric way, as done
in [Tec87]. Due to Theorem 1.3.23, this is equivalent to Wedderburn’s result.

1.3.3 Morphisms
Definition 1.3.30. Let V be a vector space over a field K and let σ be an automorphism of K.
A semi-linear automorphism of V with companion automorphism σ is a bijection θ from V into
V such that, for all x, y ∈ V and g ∈ K:

(i) θ(x + y) = θ(x) + θ(y);

(ii) θ(gx) = g σ θ(x).

Theorem 1.3.31 (Second representation theorem). For any collineation φ of a Desarguesian

projective space (P, L, I) = P(V ), there exists a semi-linear automorphism θ of V that induces
φ, in the sense that for all x ∈ P ,

φ(x) := {θ(t) : t ∈ x}.

Definition 1.3.32. Let P = P(V ) be a Desarguesian projective space. A projectivity of P is a

collineation of P induced by a linear map of the underlying vector space V .

Definition 1.3.33. Let φ be a collineation of the projective space PG(n, K). A point p ∈
PG(n, K) is a centre of the collineation φ if

(i) φ(p) = p;

15
CHAPTER 1. PRELIMINARY RESULTS
1.3. PROJECTIVE SPACES

(ii) any line L through p is fixed by φ.

A hyperplane H of PG(n, K) is an axis of the collineation φ if φ|H is the identity.

Theorem 1.3.34. A non-trivial collineation φ of PG(n, K) has at most one centre and one axis;
furthermore, φ has a centre if and only if φ has an axis.

Definition 1.3.35. A collineation φ is central (or axial) if it has a centre. A central collineation
φ whose centre is incident with its axis is called an elation; if the centre of φ is not incident with
its axis, then φ is called a homology.

Definition 1.3.36. The set of all collineations of the Pappian projective space PG(n, K) is
denoted by the symbol
PΓL(n + 1, K);
the set of all projectivities of PG(n, K) is written as

PGL(n + 1, K).

Theorem 1.3.37. For all integers n ≥ 1 and for all fields K, the set PΓL(n + 1, K) together
with mapping composition constitutes a group. The set of all projectivities of PG(n, K) is a
subgroup of PΓL(n, K).

Theorem 1.3.38. Let K be a field, n ≥ 1 and GL(n + 1, K) be the group of all non-singular
matrices of dimension (n + 1) × (n + 1) over K. Then,

PGL(n + 1, K) ' GL(n + 1, K)/K ? .

Definition 1.3.39. The stabilizer of a line of PG(n, K) in PGL(n + 1, K) is the affine linear
group AGL(n, K).

1.3.4 Singer cyclic groups

Singer cyclic groups have been introduced in [Sin38].

Definition 1.3.40. A Singer cyclic group of a projective space Π is a collineation group that is
cyclic and transitive on the points of Π. A generator of a Singer cyclic group is a Singer cycle.

Definition 1.3.41. A projective space Π is cyclic if it admits a Singer cyclic group.

Lemma 1.3.42. Any Desarguesian projective space PG(n, q) is cyclic.

Lemma 1.3.43. Any Singer cyclic group is transitive on the set of the lines of PG(2, q).

Lemma 1.3.44. Any Singer cycle of PG(n, q) is conjugate to a diagonal linear transformation
in PGL(n + 1, q n+1 ).

16
CHAPTER 1. PRELIMINARY RESULTS
1.4. POLYNOMIALS AND MATRICES

1.4 Polynomials and matrices

The results in this section are taken from [Lan80] and [HH70].

1.4.1 Definitions
Definition 1.4.1. Let K be a field. By the symbol Mat(n, K) we denote the ring of all n ×
n matrices with entries in K. For any prime power q, the ring Mat(n, GF(q)) is written as
Mat(n, q).

For any given matrix M ∈ Mat(n, K), we write its transposed matrix, obtained interchang-
ing rows and columns of M , as M ? .

Definition 1.4.2. Two matrices A, B ∈ Mat(n, K) are equivalent if there exist a matrix C ∈
Mat(n, K), such that

(i) det C 6= 0;

(ii) A = CBC ? .

Definition 1.4.3. Given any matrix M ∈ Mat(n, K), its characteristic polynomial CM (x) is

CM (x) := det(M − xIn ),

where In is the identity matrix in Mat(n, K).

P
Definition 1.4.4. For any matrix M ∈ Mat(n, K) and any polynomial p(x) = pi xi ∈ K[x],
the valuation of p(x) at M is X
p(M ) := pi M i .

Lemma 1.4.5. Let M ∈ Mat(n, K); the set

J(M ) := {p ∈ K[x] : p(M ) = 0}

is an ideal of the ring K[x].

Since K[x] is a principal ideal domain, there exist a generator element for J(M ) that is
unique up to multiplication by elements of K ? .

Definition 1.4.6. The minimal polynomial of M ∈ Mat(n, K) is the monic generator MM (x)
of the ideal J(M ) of K[x].

Lemma 1.4.7. For any M ∈ Mat(n, K),

(i) deg CM (x) = n;

17
CHAPTER 1. PRELIMINARY RESULTS
1.5. VARIETIES OVER FINITE FIELDS

(ii) CM (x) ∈ J(M );

(iii) MM (x) divides CM (x).

Definition 1.4.8. The null space of a matrix M ∈ Mat(n, K) is the set Null M of all vectors
x ∈ K n such that
M x? = 0.
Equivalently, the null space of M is the kernel of the homomorphism induced by M . The null
space of a polynomial p(x) with respect to a matrix M is the set

NullM p(x) := {y ∈ K n : p(M )y ? = 0}.

1.5 Varieties over finite fields

The few general algebraic geometry results presented in this section are from [Ful69]. Ref-
erences for the GF(q)-case are [Hir98b] and our notes [Hir98a]; the conventions adopted are
those of [Hir98b].

1.5.1 Polynomials as functions over projective spaces

For any field K, we shall denote by Rn the ring Rn (K) := K[X1 , . . . , Xn ] of polynomials in
n indeterminates with coefficients in K. This ring is the free ring generated by the symbols
X1 , . . . , Xn over K.

Definition 1.5.1. Let K be a field; a form f of degree r in n + 1 variables is a homogeneous

polynomial f ∈ K[X0 , . . . , Xn ] of degree r.

Definition 1.5.2. We denote the set of all forms in n + 1 variables over a field K by Rn (K) ⊆
K[X0 , . . . , Xn ]; the set of all forms in n + 1 variables of given degree r over K is written as
r
Rn (K).

Let f ∈ Rn (K) be a form, and let V be an n + 1-dimensional vector space over K; for any
vector v = (v0 , . . . , vn ) ∈ V ' K n+1 , the value of f at v is the scalar

f (v) := f (v0 , . . . , vn ).
r
Remark 1.5.3. Neither Rn (K) nor Rn (K) are subrings of Rn [X0 ], since the former is not
closed under the sum and the latter under the product. On the other hand, (Rn (K)? , ·) is a
r
monoid and (Rn (K) ∪ {0}, +) is a group.

Lemma 1.5.4. Let P = PG(n, K) = PK n+1 be a projective space and consider a form f ∈
Rn (K). Then, for all v ∈ K n+1 \ {0}, f (v) = 0 implies

f (z) = 0, for all z ∈ Pv.

18
CHAPTER 1. PRELIMINARY RESULTS
1.5. VARIETIES OVER FINITE FIELDS

Proof. Since z ∈ Pv, there exists a λ ∈ K ? such that z = λv. Denote by r be the degree of f ;
then,
f (λv) = λr f (v) = 0.

1.5.2 Algebraic sets and ideals

Definition 1.5.5. Let K be a field and assume F := {f1 , . . . , fk } ⊆ Rn (K). The set of zeros
determined by F is

V(F) := V(f1 , . . . , fk ) := {Pv ∈ PK n+1 : f1 (v) = f2 (v) = . . . = fk (v) = 0}.

The ideal of F in Rn (K) is denoted by

I(F) := I(f1 , . . . , fk ).

Lemma 1.5.6. Let F, F0 ⊆ Rn (K), with I(F) = I(F0 ). Then,

V(F) = V(F0 ).

Remark 1.5.7. Lemma 1.5.6 shows that the set V(F) does not depend on the list of forms F
but only on the ideal F generates in Rn (K).
Definition 1.5.8. Let K be a field and assume F = {f1 , . . . , fr } to be a set of forms in Rn (K).
The projective variety defined by F is the pair

F(F) := F(f1 , . . . , fr ) := (V(F), I(F)).

The set I(F) is the ideal of F(F).

Definition 1.5.9. The intersection of two varieties F(F) and F(G) is the projective variety

F(F) ∩ F(G) := F(F ∪ G).

Definition 1.5.10. A sub-variety of a variety F(F) is a projective variety F(G) such that
F(G) ∩ F(F) = F(G).
Lemma 1.5.11. Let F(F) and F(G) be two projective varieties. Then,

F(F) ∩ F(G) = (V(F) ∩ V(G), I(F ∪ G)).

Definition 1.5.12. Let F = {f1 , . . . , fr } ⊆ Rn (K) be a set of forms over a field K. A K-

rational point of F(F) is an element Pv ∈ V(F) ∩ PG(n, K).
n+1
Let K denote the algebraic closure of K. A point of F(F) is any P t ∈ PK such that
f (t) = 0, for all f ∈ F. The set of all points of F(F) is

V(F).

For any subfield K 0 of K, a K 0 -rational point P t of F(F) is a point of PG(n, K 0 ) ∩ V(F).

19
CHAPTER 1. PRELIMINARY RESULTS
1.5. VARIETIES OVER FINITE FIELDS

Definition 1.5.13. An algebraic set Y in PG(n, K) is a set of points of PG(n, K) such that
there exists F ⊆ Rn (K), with
Y = V(F).
The ideal of the set Y is defined as
Ĩ(Y ) := I(F).

1.5.3 Number of rational points: zeta functions

Definition 1.5.14. Let X = V(F) be a projective variety over a finite field GF(q). A point P t
of X of degree i is a point that is GF(q i )-rational, but is not GF(q j )-rational for any j < i.
A closed point of degree i is a set
i−1
P t := {P tσ : σ ∈ Gal(GF(q i ) : GF(q))} = {P t, P tq , . . . , P tq },

where P t is a point of X of degree i.

Definition 1.5.15. A divisor on a curve X = V(F) is an element of the free group generated
by all its closed points.

The group of all divisors of X is written as Div(X).

Definition 1.5.16. A divisor D is a K-divisor if all its components are K-rational points.

Remark 1.5.17. A divisor D on a curve X = V(F) is a formal sum

X
D= nP t P t,
P t∈V(F)

where

(i) nP t ∈ Z;

(ii) nP t = 0, for all but a finite number of P t;

(iii) if P t is a point of X of degree i, then nP t = nP t0 , for all P t0 ∈ P t.

Definition 1.5.18. Let X = V(F) be a projective curve; its zeta function is the formal series
X
ζX (T ) := T deg(D) .
D∈DivK (X)

Lemma 1.5.19. For any projective curve X = V(F) defined over GF(q), let

(i) Ni be the number of points of X that are GF(q i )-rational;

(ii) Ms be the number of effective GF(q)-divisors on X of degree s;

20
CHAPTER 1. PRELIMINARY RESULTS
1.5. VARIETIES OVER FINITE FIELDS

(iii) Bj be the number of closed points of degree j.

Then,

(i)
∞
X
ζX (T ) = 1 + Ms T s ;
s=1

(ii) X
Ni = jBj ;
j|i

(iii)
∞
Y ∞
X
j −Bj
ζX (T ) = (1 − T ) = exp( Ni T i /i).
j=1 i=1

1.5.4 Dimension and algebraic sets

Lemma 1.5.20. The family of the algebraic sets of PG(n, K) is closed under finite union and
intersection; both ∅ and PG0 (n, K) are algebraic sets.

Definition 1.5.21. The Zariski topology on PG0 (n, K) is the topology whose open sets are the
complements of algebraic sets.

Remark 1.5.22. Any sub-variety F(G) of a projective variety F(F) is closed in F(F).

Definition 1.5.23. A non-empty subset Y of a topological space Ξ is irreducible if it cannot be

expressed as the union Y1 ∪ Y2 of two proper subsets, each one of which is closed in Y with the
induced topology.
The empty set is not irreducible.

Definition 1.5.24. The dimension of a topological space Ξ is the supremum of all the integers i
such that there exists a chain
Z0 ⊂ Z1 ⊂ . . . ⊂ Zn = Ξ
of distinct irreducible closed subsets.

Definition 1.5.25. The dimension of a variety F(F) is the topological dimension of its point set
V(F), when endowed with the Zariski topology.

Remark 1.5.26. The null form 0 ∈ Rn (K) defines the projective variety

F(0) := (PG0 (n, K), {0}).

Its topological dimension is n, the same as the incidence dimension of PG(n, K). This is the
only variety of dimension n in PG(n, K).

21
CHAPTER 1. PRELIMINARY RESULTS
1.6. UNITARY GROUPS

Definition 1.5.27. Let F(F) be a variety. By

F r (F)

we denote the set of all r-dimensional sub-varieties of F(F).

Definition 1.5.28. An hypersurface of PG(n, K) is a projective variety of dimension n − 1 in

PG(n, K). A 3-dimensional projective variety is a projective surface; a 2-dimensional projec-
tive variety is a projective curve.

1.5.5 Non-singular varieties

Definition 1.5.29. Let F = {f1 , . . . , fk } ⊆ Rn (K). The Jacobian matrix associated to F at a
point p ∈ K n+1 is the k × (n + 1) matrix JF (p) given by
¯
∂Fi ¯¯
(JF (p))i(j+1) := ,
∂Xj ¯p

where

(i) 0 ≤ j ≤ n;

(ii) 1 ≤ i ≤ k;

(iii) ∂Fi /∂Xj is the formal derivative of Fi with respect to Xj .

Definition 1.5.30. A variety F(F) is non-singular at a point p ∈ V(F) if the rank of its Jacobian
matrix JF (p) at p is maximal, that is

rankJF(p) = (n − dim F(F)).

A variety F(F) is non-singular if it is non-singular at each of its points.

Definition 1.5.31. Assume Px to be a non-singular point of a variety F(F). The tangent space
to F(F) at Px is the subspace of PG(n, K) generated by the points corresponding to the pro-
jection of the rows of the Jacobian matrix JF (x).

1.6 Unitary Groups

The results in this section are taken from [Die71] and from the notes of the 1999 Socrates
Summer school [Kin99].

22
CHAPTER 1. PRELIMINARY RESULTS
1.6. UNITARY GROUPS

1.6.1 Definitions
Let V be a vector space over a field K and assume σ to be either an involution of K or the
identity mapping; denote by K 0 the subfield of K fixed by σ.

Definition 1.6.1. A sesquilinear form f defined on V is a mapping V × V → K such that, for

any choice of x, x1 , x2 , y, y1 , y2 in V and λ in K,

(i) f (x1 + x2 , y) = f (x1 , y) + f (x2 , y);

(ii) f (x, y1 + y2 ) = f (x, y1 ) + f (x, y2 );

(iii) f (xλ, y) = λσ f (x, y);

(iv) f (x, yλ) = f (x, y)λ.

Definition 1.6.2. Let f and g be two sesquilinear forms which are respectively defined over the
vector spaces X and Y . Then, f and g are equivalent if there exists an isomorphism T of X
into Y such that, for any x, y ∈ X,

f (x, y) = g(T x, T y).

Definition 1.6.3. A sesquilinear form f defined on V is non-degenerate if, for any non-zero
vector x ∈ V , there exists a vector y ∈ V such that

f (x, y) 6= 0.

The form f is reflexive if, for any x, y ∈ V ,

f (x, y) = 0 ⇐⇒ f (y, x) = 0.

Definition 1.6.4. A sesquilinear form f is Hermitian if, for any x, y ∈ V ,

f (x, y) = f (y, x)σ .

From now on we assume f to be a sesquilinear form on V which is both non-degenerate and

reflexive.

Definition 1.6.5. A transformation u of V unitary with respect to the form f is any bijective
linear transformation of V which preserves f . That is, for any x, y ∈ V ,

f (x, y) = f (u(x), u(y)).

Lemma 1.6.6. The set of all transformations of V unitary with respect to the form f form a
subgroup Uf (n, K) of the general linear group GL(n, K).

23
CHAPTER 1. PRELIMINARY RESULTS
1.6. UNITARY GROUPS

Definition 1.6.7. Let θ be an automorphism of the field K. A unitary semi-similarity of V

(corresponding to the form f and the automorphism θ) of multiplier ru is a collineation u :
V → V such that, for any x, y ∈ V ,

f (u(x), u(y)) = ru f (x, y)θ .

Lemma 1.6.8. The set of all unitary semi-similarities of V forms a subgroup ΓUf (n, K) of the
group ΓL(n, K) of all collineations of V .

Definition 1.6.9. A semi-similarity u is linear if the automorphism θ associated with u is the

identity of K.

Lemma 1.6.10. The set of all linear semi-similarities is a subgroup GUf (n, K) of the general
linear group GL(n, K); indeed,

GUf (n, K) = ΓU(n, K) ∩ GL(n, K).

Lemma 1.6.11. Let f , g be two equivalent sesquilinear forms. Then, the unitary groups induced
by f and g are all isomorphic; that is, Uf (n, K) ' Ug (n, K), ΓUf (n, K) ' ΓUg (n, K) and
GUf (n, K) ' GUg (n, K).

Definition 1.6.12. Let α be in K ? . The dilation of V with coefficient α is the mapping

½
V → V
hα :
x → xα.

Lemma 1.6.13. The set H(n, K) := {hα : α ∈ K ? } of all dilations of V is a normal subgroup
of ΓL(n, K).

Definition 1.6.14. The group of all projective collineations of the projective space PV , obtained
by derivation from V , is the quotient group

PΓL(n, K) := ΓL(n, K)/H(n, K).

The canonical projection from ΓL(n, K) in PΓL(n, K) is denoted as

½
ΓL(n, K) → PΓL(n, K)
P:
x → xH(n, K).

Definition 1.6.15. The projective collineation groups

PΓUf (n, K), PGUf (n, K), PUf (n, K)

are defined as the images in PΓL(n, K) via the projection P of

ΓUf (n, K), GUf (n, K), Uf (n, K).

24
CHAPTER 1. PRELIMINARY RESULTS
1.6. UNITARY GROUPS

1.6.2 Subspaces and forms

Lemma 1.6.16. A necessary and sufficient condition for the existence of a unitary transforma-
tion u, mapping a subspace X ≤ V into Y ≤ V , is that the restriction of the form f to X is
equivalent to the restriction of the same form f to Y .

Definition 1.6.17. A subspace X of V is orthogonal to Y ≤ V with respect to the form f if and

only if, for all x ∈ X and y ∈ Y ,
f (x, y) = 0.
The largest subspace of V orthogonal to X is denoted as

X ⊥ := {v ∈ V : ∀x ∈ X, f (v, x) = 0}.

Definition 1.6.18. A space X is isotropic if X ∩ X ⊥ 6= {0}; it is totally isotropic if X ⊆ X ⊥ .

Definition 1.6.19. An hyperbolic plane X is a non-isotropic plane of V containing at least one

isotropic vector.

Lemma 1.6.20.

(i) If X and Y are two totally isotropic subspaces of V and dim X = dim Y , then there exists
an u ∈ Uf (n, K), such that u(X) = Y .

(ii) There exists an integer ν such that any totally isotropic subspace of V is contained in a
totally isotropic subspace of maximal dimension ν.

Definition 1.6.21. A quasi-symmetry (with hyperplane H) is a unitary transformation u fixing

point-wise a non-isotropic hyperplane H in V .

Definition 1.6.22. A hyperbolic transformation is a unitary transformation u ∈ Uf (n, K) fixing

all vectors of a subspace Y ≤ V of dimension n − 2 which is orthogonal to an hyperbolic plane.

Remark 1.6.23. The projective image PX of an hyperbolic plane X is a hyperbolic line. A pro-
jective hyperbolic transformation is, hence, a unitary transformation fixing all points belonging
to a projective hyperplane (PX)⊥ .

1.6.3 Hermitian groups

Definition 1.6.24. A Hermitian sesquilinear form g is a trace form if, for all v ∈ V , the element
g(v, v) ∈ K can be written as a trace over K 0 ; that is, for any v ∈ V , there exists a λv ∈ K such
that
g(v, v) = TK/K 0 (λv ).

25
CHAPTER 1. PRELIMINARY RESULTS
1.6. UNITARY GROUPS

Lemma 1.6.25. If the characteristic of K is not 2, then all Hermitian sesquilinear forms over
K are trace forms. If K has characteristic 2, then the only trace forms are the alternating ones.

Definition 1.6.26. The standard Hermitian product on the vector space V is the Hermitian
sesquilinear form h·, ·i, given by
½
V ×V → K P
h·, ·i :
(u, v) → ui viσ .

For any Hermitian form g equivalent to the standard Hermitian product, the unitary groups
Ug (n, K), GUg (n, K), ΓUg (n, K) and PUg (n, K), PGUg (n, K), PΓUg (n, K), are denoted
omitting the reference to g; that is, they are simply written as U(n, K), GU(n, K), ΓU(n, K)
and PU(n, K), PGU(n, K), PΓU(n, K).
For any prime power q, the notation U(n, q) := U(n, GF(q)) will be adopted as well.

1.6.4 Generators of the unitary group

Theorem 1.6.27. Let f be a non-alternating symplectic form. Then, either the unitary group is
Uf (n, K) = U(2, 4), or Uf (n, K) is generated by the quasi-symmetries of V ' K n .

Theorem 1.6.28. Let ν be the dimension of a maximal totally isotropic subspace X of V . Then,

(i) if ν ≥ 1, all the transformations in Uf (n, K) are products of hyperbolic transformations;

(ii) if ν ≥ 1 and n ≥ 3, all non-isotropic lines are the intersection of two hyperbolic planes,
except in the case where Uf (n, K) coincides with the orthogonal group O(3, 3).

1.6.5 Morphisms and automorphisms

Define Z(n, K) as the centre of GL(n, K).

Lemma 1.6.29. For any sesquilinear form f , the following isomorphism relations are satisfied:

(i) PGL(n, K) ' GL(n, K)/Z(n, K);

(ii) PΓUf (n, K) ' ΓUf (n, K)/H(n, K);

(iii) PGUf (n, K) ' GUf (n, K)/Z(n, K).

Theorem 1.6.30. Suppose g to be a Hermitian form, let n ≥ 3 and assume charK to be odd.
Then, all automorphisms of the unitary group Ug (n, K) can be written as

φ(u) = χ(u)ug ,

where

26
CHAPTER 1. PRELIMINARY RESULTS
1.7. HERMITIAN FORMS AND HERMITIAN VARIETIES

(i) g ∈ ΓUg (n, K);

(ii) χ is an homomorphism of Ug (n, K) into its centre.

Theorem 1.6.31 (Walter). Let K be a field of odd characteristic with more than 3 elements and
assume n ≥ 5. Then, any automorphism of the projective unitary group PUf (n, K) is obtained
from an automorphism of Uf (n, K) by way of quotienting.

1.7 Hermitian Forms and Hermitian varieties

The main references for this section are [Seg67], [Seg65] and [Hir98b].

1.7.1 Hermitian matrices

Assume K to be a field with an involutory automorphism σ and let K 0 be the fixed field of σ in
K.

Definition 1.7.1. Let H ∈ Mat(n, K). Then, H is

(i) Hermitian with respect to the automorphism σ, if

H? = Hσ;

(ii) anti-Hermitian with respect to the automorphism σ, if

H ? = −H σ ;

(iii) unitary with respect to the automorphism σ, if

H σ = (H ? )−1 ;

(iv) anti-orthogonal with respect to the automorphism σ, if

H σ = H −1 .

Lemma 1.7.2. The conjugate of a Hermitian matrix via a unitary matrix is a Hermitian matrix.

Lemma 1.7.3. Let H ∈ Mat(n, K) be a Hermitian matrix; then, det H ∈ K 0 .

Proof. Since det H = det H ? ,

d := det H = det H ? = det H σ = (det H)σ = dσ .

Hence, d ∈ FixK (σ) = K 0 .

27
CHAPTER 1. PRELIMINARY RESULTS
1.7. HERMITIAN FORMS AND HERMITIAN VARIETIES

Definition 1.7.4. Let V be a vector space over K, and assume H ∈ Mat(n, K) to be a matrix,
Hermitian with respect to the automorphism σ. The σ-Hermitian form over V defined by H is
the form ½
V ×V → K
h:
(x, y) → xσ Hy ? .
Lemma 1.7.5. Assume q to be a square and let H ∈ Mat(n, q) be a Hermitian matrix. Then,
√
αH is a Hermitian matrix for any α ∈ GF( q).

Definition 1.7.6. A polynomial f (x) in K[x] is Hermitian if and only if, for any integer n and
any Hermitian matrix H ∈ Mat(n, K), f (H) is a Hermitian matrix.

Remark 1.7.7. All polynomials with coefficients in the subfield K 0 of K are Hermitian.

Lemma 1.7.8. The set of all Hermitian matrices of Mat(n, q) is a vector space of dimension n2
√
over GF( q).

Proof. A Hermitian H matrix can be given by providing the entries in its upper triangular part;
√
any entry on the main diagonal, that is of the form Hii , is an element of GF( q); entries above
√
the main diagonal are elements of GF(q); hence, they have the form a + ²b with a, b ∈ GF( q)
√
and ² a fixed element of GF(q) \ GF( q).
A direct count shows that exactly
1
n + 2( n)(n − 1) = n2
2
√
choices of elements of GF( q) have to be made in order to determine H.

Definition 1.7.9. Two matrices A, B ∈ Mat(n, K) are Hermitian-equivalent if there exists a

non-singular matrix C ∈ Mat(n, K) such that

A = C σ BC ? .

Theorem 1.7.10 (Equivalence of Hermitian matrices). Let H ∈ Mat(n, K) be a Hermitian

matrix of rank n − t. Then, H is Hermitian-equivalent to a matrix J of the form

J := diag(j0 , . . . , jn−t , 0, . . . , 0 ),
| {z }
t times

where ji ∈ K 0 for any i.

1.7.2 Hermitian forms

Lemma 1.7.11 (Representation theorem for Hermitian forms). Let V be a vector space of
dimension n over the field K and let H be a Hermitian matrix in Mat(n, K). Then, the σ-
Hermitian form over V defined by H is a sesquilinear Hermitian form in the sense of Definition

28
CHAPTER 1. PRELIMINARY RESULTS
1.7. HERMITIAN FORMS AND HERMITIAN VARIETIES

1.6.4. Conversely, given a sesquilinear Hermitian form h over a vector space V of dimension
n, there exists a Hermitian matrix H ∈ Mat(n, K) such that h is the σ-Hermitian form over V
defined by H.
Thanks to Lemma 1.7.11, it is possible to identify σ-Hermitian forms and sesquilinear Her-
mitian forms.
Remark 1.7.12. Over an arbitrary field K, there might exist different classes of Hermitian
forms, depending on the involutory automorphism σ of K which has been chosen. On the other
hand, if K is a finite field GF(q), then there is an involutory automorphism of K if and only if
q is a square. In this case, the involution is unique and it is associated with the subfield of K
√
with index 2, that is GF( q).
√
Let q be a prime power p2n . The unique involutory automorphism σ of GF(q) over GF( q)
will be denoted by the conjugation sign. Namely, for x ∈ GF(q):
√
x := xσ = x q .

Definition 1.7.13. Let r ≥ 1 be an integer and assume V to be a vector space of dimension n

over the field K. A mapping h : V n → K is a homogeneous polynomial mapping of degree r if
r
there exists a polynomial f ∈ Rn (K) such that
½ n
V → K
h:
(p1 , p2 , . . . , pn ) → f (p1 , p2 , . . . , pn ).
Such a polynomial f is said to represent the mapping h.
For a vector space V over an arbitrary field K, a σ-Hermitian form h usually cannot be
represented by a polynomial mapping. For instance, there is no polynomial in R2n (C) which
represents the Hermitian form induced over the vector space Cn by the standard Hermitian
product between complex numbers. However, if K is finite, the following theorem is true.
Theorem 1.7.14. Let V be a vector space of dimension n over GF(q), with q a square,
√
and
q+1
let h be a Hermitian form defined over V . Then, there exists a polynomial f ∈ R2n which
represents h.
√
Proof. The only involution of GF(q) is the Frobenius automorphism σ : t → t q . Theorem
1.7.11 guarantees that there exists a Hermitian matrix H such that the form h is represented by
H. Then, ½
V × V → K√
h:
(x, y) → x q Hy ? .
By computing the expansion of the vector/matrix product, this yields
n
X n
X √
q
h(x, y) = [yj xi Hji ],
j=1 i=1

whence the result.

29
CHAPTER 1. PRELIMINARY RESULTS
1.7. HERMITIAN FORMS AND HERMITIAN VARIETIES

1.7.3 Hermitian hypersurfaces: definitions

This subsection deals only with Hermitian forms which are representable by polynomials. How-
ever, the assumption that the field K is finite is not necessary.

Definition 1.7.15. The Hermitian hypersurface defined by the Hermitian form h is the algebraic
variety
H(h) := F(f (x, x)),
where f is a polynomial representing h and x = (X0 , . . . , Xn ). The Hermitian hypersurface
defined by the matrix H is the Hermitian hypersurface

H(H) := H(h),

where h is the Hermitian form induced by the matrix H. The algebraic variety F(0), induced
by the trivial Hermitian form, is not considered to be a Hermitian variety

Remark 1.7.16. A Hermitian form h over GF(q) is not a Hermitian form over GF(q i ) for i > 1.
However it makes sense to consider the GF(q i )-rational points of the GF(q)-Hermitian variety
H(h).

Theorem 1.7.17. The set of all Hermitian hypersurfaces of PG(n, q) can be endowed with the
√
structure of a PG(n2 + 2n, q).

Proof. The correspondence in Lemma 1.7.11 allows us to identify Hermitian forms with Hermi-
tian matrices. Let H, H 0 be two non-zero Hermitian matrices defining the same Hermitian hy-
persurface, and call f, f 0 the Hermitian polynomials associated with them. Since F(f (x, x)) =
F(f 0 (x, x)), the ideals generated by f (x, x) and f 0 (x, x) have to be the same, that is

I(f (x, x)) = I(f 0 (x, x)).

It follows that there exists an α ∈ GF(q)? such that

f (x, x) = αf 0 (x, x).

Then, by construction,
H = αH 0
√
and α ∈ GF( q)? . An immediate computation shows the converse, namely that if H = αH 0
√
with α ∈ GF( q), then H(H) = H(H 0 ).
By Lemma 1.7.8, the Hermitian matrices in Mat(n + 1, q) constitute a vector space V over
√
GF( q) with dimension (n + 1)2 .

30
CHAPTER 1. PRELIMINARY RESULTS
1.7. HERMITIAN FORMS AND HERMITIAN VARIETIES

Distinct Hermitian hypersurfaces of PG(n, q) are in one-to-one correspondence with el-

ements of the set U of all Hermitian polynomials modulo GF(q)? ; the set U , in turn, is in
one-to-one correspondence with the quotient space
√
V ? / GF( q)? ,
√
which is isomorphic to PG(n2 + 2n, q) and this proves the theorem.

Definition 1.7.18. The rank of a Hermitian hypersurface

H(H) := F(xHx? )

is the rank of the matrix H.

Lemma 1.7.19. A point x on a Hermitian hypersurface H(H) is singular if and only if x belongs
to the null-space of H, that is
xH = 0.

Lemma 1.7.20. A Hermitian hypersurface is non-singular if and only if the matrix associated
with it has full rank, that is, it is a non-singular matrix.

Lemma 1.7.21. Let H and H 0 be two Hermitian-equivalent matrices. Then, the hypersurfaces
H(H) and H(H 0 ) are projectively equivalent.

Lemma 1.7.22. For any k ≤ n+1, there exists exactly one Hermitian hypersurface in PG(n, q)
of rank k up to projectivities.

Proof. Theorem 1.7.10 and Lemma 1.7.21 imply that any non-singular Hermitian hypersurface
in PG(n, q) is projectively equivalent to the one generated by the diagonal matrix

M := diag(m0 , . . . , mk , 0, . . . , 0),
√ √
where m0 , . . . , mk are in GF( q). Since the norm from GF(q) onto GF( q) is surjective,
there exist elements ti ∈ GF(q), such that

ti ti = mi .

Let T be the matrix

T := diag(t0 , . . . , tk , 0, . . . , 0).
Then, M is Hermitian-equivalent via T to the block matrix

diag(Ik , 0, . . . , 0 ),
| {z }
n + 1 − k times

where Ik is the k × k identity matrix.

31
CHAPTER 1. PRELIMINARY RESULTS
1.7. HERMITIAN FORMS AND HERMITIAN VARIETIES

Definition 1.7.23. The canonical Hermitian hypersurface of rank k in PG(n, q) is the Hermi-
tian hypersurface
Πn−k−1 Uk,q
induced by the matrix diag(Ik+1 , 0, . . . , 0) ∈ Mat(n + 1, q). The canonical non-singular Her-
mitian hypersurface in PG(n, q) is the non-singular Hermitian hypersurface

Un := Π−1 Un,q := H(I),

where I is the identity of Mat(n + 1, q).

A Hermitian hypersurface which is projectively equivalent to a Πn−1 U0,q is a hyperplane

√
repeated q + 1 times; a Hermitian hypersurface projectively equivalent to a Πn−k−1 Uk,q is a
Hermitian cone having as vertex a Πn−k−1 ∈ PGn−k−1 (n, q) and base a Uk,q .

Definition 1.7.24. The canonical Hermitian norm of rank k in PG(n, q) is the form
k
X k
X √
q+1
hk,n (x) := xi xi = xi .
i=0 i=0

With Definition 1.7.24, we can write

(i) Un := F(hn,n (x));

(ii) Πn−k−1 Uk := F(hk,n (x)).

1.7.4 Hermitian sub-varieties and points

Lemma 1.7.25. The intersection of a Hermitian H variety with a subspace Πr ∈ PGr (n, q) of
dimension r not completely included in H is a Hermitian variety of Πr .

Corollary 1.7.26. The intersection of a Hermitian variety with a line is either a line, a Baer
subline or a point.

Proof. The corollary follows by observing that Hermitian varieties in dimension 1 are either
Baer sublines or lines (if completely degenerate), while a Hermitian variety in dimension 0 is a
point.

The following theorems deal with the cardinality of the set of the points of a (possibly
singular) Hermitian variety H.

Theorem 1.7.27. The zeta function of the non-singular Hermitian curve U2,q is
√ √
(1 + qT )q− q
ζU2,q (T ) = .
(1 − T )(1 − qT )

32
CHAPTER 1. PRELIMINARY RESULTS
1.7. HERMITIAN FORMS AND HERMITIAN VARIETIES

Theorem 1.7.28. The number of GF(q)-rational points of the non-singular Hermitian variety
Un,q is
µ(n, q) := [q (n+1)/2 + (−1)n ][q n/2 − (−1)n ]/(q − 1).

Corollary 1.7.29.
√
(i) µ(1, q) = q + 1;
√
(ii) µ(2, q) = q q + 1;
√
(iii) µ(3, q) = (q + 1)(q q + 1).

Theorem 1.7.30. The singular Hermitian variety Πm Uk,q contains:

(i) q m + q m−1 + . . . + 1 = (q m+1 − 1)/(q − 1) singular points;

(ii) q m+1 µ(k, q) non-singular GF(q)-rational points;

(iii) η(m, k, q) = (q m+1 − 1)/(q − 1) + q m+1 µ(k, q) total GF(q)-rational points.

1.7.5 Hermitian varieties and unitary groups

The stabilisers of a Hermitian variety Un,q in the collineation groups of PG(n, q) are unitary
groups. In fact, the following lemma holds true.

Lemma 1.7.31. The stabiliser in PGL(n + 1, q) of the set of the GF(q)-rational points of Un,q
is PGU(n + 1, q). The stabiliser of the same set in PΓL(n + 1, q) is PΓU(n + 1, q).

Definition 1.7.32. The group PγU(n + 1, q) is the group of all collineations of PG(n, q) sta-
bilizing the point set of Un,q which are associated either with the identity of the field GF(q) or
with its involutory automorphism.

Lemma 1.7.33 (Orders of groups). Assume p be a prime, and let q = ph , with h even; define
m
Y
m(m−1)/2
λ² (m, r) := r (ri − ²i ).
i=1

Then, the collineation groups have the following orders:

(i) | GL(n + 1, q)| = λ1 (n + 1, q);

(ii) | PGL(n + 1, q)| = (q − 1)−1 λ1 (n + 1, q);

(iii) PGU(n + 1, q) ≤ PγU(n + 1, q) ≤ PΓU(n + 1, q);

√
λ−1 (n+1, q)
(iv) | PGU(n + 1, q)| = √
q+1
;

33
CHAPTER 1. PRELIMINARY RESULTS
1.7. HERMITIAN FORMS AND HERMITIAN VARIETIES

(v) PGU(n + 1, q) < PγU(n + 1, q) ≤ PΓU(n + 1, q);

(vi) | PγU(n + 1, q) : PGU(n + 1, q)| = 2;

(vii) | PγU(n + 1, q) : PΓU(n + 1, q)| = h/2;

√
λ−1 (n+1, q) √
(viii) | PGL(n + 1, q) : PGU(n + 1, q)| = λ1 (n+1,q)
( q − 1);

(ix) If q = p2 , with p prime, then PγU(n + 1, q) = PΓU(n + 1, q).

Corollary 1.7.34. The number of distinct non-singular Hermitian hypersurfaces in PG(n, q) is

√ n+1
q {n(n+1)/2} Y qi − 1
√ .
q − 1 i=1 q i/2 − (−1)i

Proof. Since all non-singular Hermitian hypersurfaces are projectively equivalent, the number
is the index

| PΓL(n + 1, q) : PΓU(n, q)| = | PGL(n + 1, q) : PGU(n + 1, q)|,

whence the result.

1.7.6 Polarities
References for this subsection are [BC66], [Bae52] and [Cam95].

Definition 1.7.35. The dual of a Desarguesian projective space P = PG(n, q) is the incidence
structure P? = (J, P, I ? ) defined as follows:

(i) J is the set PGn−1 (n, q) of all hyperplanes of P;

(ii) P is the set of the points of P.

(iii) for j ∈ J and p ∈ P , j is I ? -incident with p if and only if p ∈ j.

For any n, the dual of PG(n, q) is isomorphic to PG(n, q).

Definition 1.7.36. A correlation is an inclusion reversing permutation of the subspaces of

PG(n, q).

Definition 1.7.37. A sesquilinear form f (x, y) represents a correlation θ if and only if, for all
x ∈ PG(n, q),
θ(x) = {y ∈ PG(n, q) : f (x, y) = 0}.

Lemma 1.7.38. Any correlation θ is represented by a non-degenerate sesquilinear form f such

that, for all x, y ∈ PG(n, q),

f (x, y) = 0 implies f (y, x) = 0.

34
CHAPTER 1. PRELIMINARY RESULTS
1.7. HERMITIAN FORMS AND HERMITIAN VARIETIES

Definition 1.7.39. A polarity is a correlation of order 2.

A polarity can be seen as a collineation between the projective space PG(n, q) and its dual
PG(n, q)? .

Definition 1.7.40. Let θ be a polarity represented by a sesquilinear form f . Assume f to be

induced by the bilinear mapping l(x, y) with companion automorphism σ. Then, there are three
possibilities for θ:

(i) The automorphism σ is the identity, f (x, y) = f (y, x) and there exists a z ∈ PG(n, q)
such that f (z, z) 6= 0; the polarity θ is called orthogonal.

(ii) The automorphism σ is the identity and f (x, x) = 0 for any z ∈ PG(n, q); the polarity θ
is called symplectic.

(iii) The automorphism σ is an involution and f (x, y) = f (y, x)σ ; the polarity θ is called
unitary.

Definition 1.7.41. Let φ be a polarity of PG(n, q). The polar space of a point p ∈ PG(n, q) is
the hyperplane φ(p).

Definition 1.7.42. Two points x, y ∈ PG(n, q) are conjugate with respect to the polarity φ if
and only if x ∈ φ(y) and vice-versa. A point x is self-conjugate or absolute if x ∈ φ(x).

Lemma 1.7.43. Let K be a field of characteristic different from 2. Then, the non-degenerate
quadrics of PG(n, K) are precisely the sets of absolute points with respect to the orthogonal
polarities of PG(n, K).

Lemma 1.7.44. The absolute points with respect to an orthogonal polarity of PG(n, 2t ) are
precisely those of a distinguished hyperplane H. Such an hyperplane is itself absolute if and
only if the dimension n is odd.

Lemma 1.7.45. A polarity is unitary if and only if there exists a line in PG(n, q) containing
one non-absolute and at least 3 absolute points. The non-degenerate Hermitian hypersurfaces
of PG(n, K) are precisely the sets of absolute points with respect to the unitary polarities of
PG(n, K).

Remark 1.7.46. Let f be a non-degenerate sesquilinear form. For any non-self-conjugate point
x ∈ PG(n, q), it is possible to assume

f (x, x) = 1.

35
CHAPTER 1. PRELIMINARY RESULTS
1.7. HERMITIAN FORMS AND HERMITIAN VARIETIES

1.7.7 Subspaces in Hermitian varieties

Lemma 1.7.47. Assume x and y to be two points of Un . The line xy is a sub-variety of Un if
and only if x and y are mutually conjugate.

Corollary 1.7.48. A necessary and sufficient condition for a t-dimensional subspace Πt ∈

PGt (n, q) to be a sub-variety of Un,q is that all its points are mutually conjugate.

Observe that this corollary implies that any subspace Π completely contained in a non-
singular Hermitian variety Un is included in the tangent space of the variety at any of its points.

Theorem 1.7.49. Let n be an integer, and assume t to be the integral part of (n − 1)/2. Then,
Un contains linear subspaces of dimension t and no higher.

Theorem 1.7.50. Let

ψ(n, t, q)
be the number of subspaces in PGt (n, q) contained in Un,q . Then,
Qt 2j+1
(i) ψ(2t + 1, t, q) = j=0 (q + 1);
Qt 2j+3
(ii) ψ(2t + 2, t, q) = j=0 (q + 1).

Corollary 1.7.51. The number of m-dimensional projective subspaces contained in a Un,q is

n+1
Y n−m
Y
√
N (m, Un,q ) := [( q)i − (−1)i ]/ [q i+1 − 1].
i=2m−n i=0

1.7.8 Incidence properties

Definition 1.7.52. A unital in PG(2, q) is a set of points U such that any line of the plane meets
√
U in either 1 or q + 1 points. A line l which intersects U in one point is tangent to U ; if l
√
intersects U in q + 1 points, then l is a chord of U .

Lemma 1.7.53. The set of GF(q)-rational points of a non-degenerate Hermitian curve is an

unital in PG(2, q). This unital is called classical.

For further information on unitals see also Chapter 12 of [Hir98b].

36
Chapter 2

The 2-dimensional case: Hermitian curves

In this chapter, first we introduce Kestenband’s classification of point-line configurations arising

from the intersection of Hermitian curves. Later, we show that these configurations are in fact
projectively unique and we compute the stabiliser of each of them in the full linear collineation
group PGL(3, q). A group theoretical characterisation of the Hermitian curve concludes the
chapter.

2.1 Classification of intersections in dimension 2

In [Kes81], the possible point-line configurations arising from the intersection of two Hermitian
curves in PG(2, q) are described. The classification is done with linear algebra techniques.

2.1.1 Incidence classification

Definition 2.1.1. Let H = H(H) and H0 = H(H 0 ) be two distinct Hermitian varieties. The
Hermitian pencil Γ generated by H and H0 is the set
√
Γ := {H(λH + µH 0 ) : λ, µ ∈ GF( q)}.

The base locus of Γ is the set \

Γ∩ := T.
T ∈Γ

The result of [Kes81] is based upon the following important observation.

Lemma 2.1.2. The intersection of any two distinct Hermitian varieties H and H0 is the base
locus of the Hermitian pencil Γ they generate.

Proof. Let E = H ∩ H0 and assume H = H(H), H0 = H(H 0 ) and x ∈ E. Then,

xHx? = xH 0 x? = 0.

37
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.1. CLASSIFICATION OF INTERSECTIONS IN DIMENSION 2

It follows that, for any λ, µ:

x(λH + µH 0 )x? = 0;
hence, E ⊆ Γ∩ . Conversely, since H, H0 ∈ Γ, if x ∈ Γ∩ , then x ∈ E and the lemma follows.
√
We may observe that the proof of Lemma 2.1.2 does not require Γ to be the GF( q)-linear
system generated by H and H0 . In fact, any linear system generated by H and H0 would do;
√
however, the GF( q)-linear system is the largest linear system containing H1 and H2 whose
elements are all Hermitian curves. Such a linear system can be seen as a line in the projective
√
space PG(n2 + 2n, q) considered as the set of all Hermitian varieties over PG(n, q).
Thanks to Lemma 2.1.2, it is sufficient to classify all Hermitian pencils in PG(n, q) in order
to have a description of all possible incidence configurations.
For any Hermitian pencil Γ, let ri (Γ) be the number of Hermitian varieties in Γ of rank i.
√
Clearly, the sum of all ri ’s is q.

Lemma 2.1.3. The cardinality of the intersection of two Hermitian curves H1 and H2 depends
only upon the indices ri (Γ) of the linear system they generate. The possible intersection num-
bers are as in Table 2.1.

r1 r2 k = |E|
√
0 3 ( q + 1)2
√
1 1 q+1
√
0 2 q+ q+1
0 1 q+1
1 0 1
√
0 0 q− q+1

Table 2.1: Possible intersection numbers for non-degenerate Hermitian Curves.

This lemma can be found in Kestenband [Kes81]. However, it is a special case of the general
result of Theorem 4.1.26. The main result of [Kes81] is the following theorem.

Theorem 2.1.4. Let H be a non-degenerate Hermitian matrix in Mat(3, q) and let M(x) and
C(x) be its minimal and characteristic polynomials. Then, the intersection E of the canonical
Hermitian curve U2 and H(H) belongs to one of the following seven classes.
√
(i) Class I: M(x) = C(x) = (x−α)(x−β)(x−γ) with α, β, γ distinct elements of GF( q):
√
• |E| = ( q + 1)2 ;
• the points are as in Figure 2.1.a;
√
• r3 (Γ) = q − 2; r2 (Γ) = 3.

38
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.1. CLASSIFICATION OF INTERSECTIONS IN DIMENSION 2

√
(ii) Class II: M(x) = H(x) = (x − α)(x − δ)2 with α, δ distinct elements of GF( q):
√
• |E| = q + q + 1;
• the points are as in Figure 2.1.b;
√
• r3 (Γ) = q − 1; r2 (Γ) = 2.

(iii) Class III: M(x) = C(x) = (x − α)p(x) with p(x) polynomial of degree 2 irreducible
√
over GF( q):

• |E| = q + 1;
• the points are as in Figure 2.1.c;
√
• r3 (Γ) = q; r2 (Γ) = 1.

(iv) Class IV: M(x) = C(x) = (x − λ)3 :

• |E| = q + 1;
• the points are as in Figure 2.1.d;
√
• r3 (Γ) = q; r2 (Γ) = 1.
√
(v) Class V: M(x) = (x − α)(x − β) with α, β distinct elements of GF( q):
√
• |E| = q + 1;
• the points constitute a Baer subline of PG(2, q);
√
• r3 (Γ) = q − 1; r2 (Γ) = 1.

(vi) Class VI: M(x) = (x − λ)2 :

• |E| = 1;
√
• r3 (Γ) = q; r2 (Γ) = 0.

(vii) Class VII:

√
• |E| = q − q + 1;
• no three points of E are collinear;
√
• r3 (Γ) = q + 1.

Observe that considering only linear systems containing the canonical Hermitian curve U2
does not hamper generality, since it is always possible to reduce to one of those via a projectivity.

39
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.1. CLASSIFICATION OF INTERSECTIONS IN DIMENSION 2

r1 r2 MH (x)
0 3 (x − α)(x − β)(x − γ)
1 1 (x − α)(x − β)
0 2 (x − α)2 (x − β)
0 1 (x − α)p(x), (x − α)3
1 0 (x − α)2
0 0 p(x)

Table 2.2: Minimal polynomials corresponding to given rank sequences in the 2-dimensional
case.

2.1.2 Outline of the proof

Theorem 2.1.4 is proven by constructing suitable canonical forms for Hermitian matrices in
Mat(3, q) modulus Hermitian equivalence and considering the various cases arising individu-
ally. This is done via the following lemma, which provides ‘canonical’ forms for Hermitian
matrices.

Lemma 2.1.5. Let H be a Hermitian matrix in Mat(3, q) and assume MH (x) to be its minimal
polynomial. Then, H is Hermitian equivalent to one of the matrices in Table 2.3.

MH (x) conditions canonical form

 
α 0 0
(x − α)(x − β)(x − γ)  0 β 0 
 0 0 γ 
α 0 0
(x − α)(x − β)  0 β 0 
 0 0 γ 
√
α 0 0
(x − α)(x − δ)2 (β − γ)2 + 4a q+1
=0  0 β a 
√
q
 0 a γ 
√
α 0 0
(x − α)p(x) (β − α)(γ − α) 6= a q+1  0 β a 
√
q
 0 a√q γ 
√ √
λ c 0√
(x − λ)3 a q+1
+c q+1
=0  c λ a q 
0 a λ

Table 2.3: Canonical forms for 3 by 3 Hermitian matrices.

In fact, the intersections might be described by just considering the position of two specially
chosen curves in Γ. Since this is the basic tool for the projective classification that we present

40
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.1. CLASSIFICATION OF INTERSECTIONS IN DIMENSION 2

q+1

q+1

q+1

q+1
√ √
2.1.a: ( q + 1)2 points 2.1.b: q + q + 1 points
C

q +1

q-1

q+1 q

2.1.c: q + 1 points 2.1.d: q + 1 points

Figure 2.1: Possible configurations for the 2-dimensional case

in the next section, it is worth to provide here a list of these suitable ‘special’ curves for Γ
belonging to the various classes.
For any of the classes I-VII, let H be a curve in Γ different from U2 . Then, H might be
chosen as follows:

(i) Class I:

• H is a Hermitian cone with vertex not in U2 ,

• each generator of H is a chord of U2 ;

(ii) Class II:

• H is a Hermitian cone with vertex in U2 ,

• each generator of H is a chord of U2 ;

(iii) Class III:

41
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

• H is a Hermitian cone with vertex outside U2 ,

• two generators of H are tangent to U2 , all the others being chords;

(iv) Class IV

• H is a Hermitian cone with vertex in U2 ,

• one generator of H is tangent to U2 , all the others being chords;

(v) Class V
√
• H is a doubly degenerate Hermitian curve, that is a line counted q + 1 times,
• H is a chord of U2 ;

(vi) Class VI

• H is a doubly degenerate Hermitian curve,

• H is tangent to U2 ;

(vii) Class VII

• H is a non-singular Hermitian curve.

Definition 2.1.6. A point of the intersection E is special if it is either the vertex of a Hermitian
cone in Γ or the only common point of E with a generator of a Hermitian cone in Γ.

There are

• no special points in classes I, V, VI and VII;

• one special point in classes II and IV;

• two special points in class III.

2.2 Groups of the intersection of two Hermitian curves

2.2.1 Introduction
As presented in Section 2.1, Kestenband’s paper [Kes81] classifies the point-line configurations
arising from the intersection of two Hermitian curves. Nothing is said about projective equiva-
lence of the configurations in a given class nor about the linear collineation groups preserving
the structures. This is the aim of the present section. This work has been submitted for publi-
cation as [Giub]. I would like to thank Professor G. Korchmáros for helpful discussions on the
topic.

42
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

Let H1 and H2 be Hermitian curves and denote by E their intersection. As before, let Γ be
√
the GF( q)-linear system generated by H1 and H2 and define for any i = 1, 2, 3, the set Γi as
the subset of Γ which contains all the curves of rank i. Clearly, ri (Γ) = |Γi |.
Our main results are the following theorems.

Theorem 2.2.1. Each of the seven classes I-VII consists of pairwise projectively equivalent
Hermitian intersections.

Theorem 2.2.2. The linear collineation group Aut(E) preserving a Hermitian intersection E
acts transitively on both the special and non-special points of E. The abstract structure of
Aut(E) depends on the class containing E and is given in Theorems 2.2.9, 2.2.15, 2.2.18, 2.2.22,
2.2.24, 2.2.26, 2.2.28.

2.2.2 A non-canonical model of PG(2, q)

Definition 2.2.3. For any i ≤ j, a projective subspace Π of order pi is in canonical position in
PG(n, pj ) if and only if
Π ∩ PG(n, pj ) = PG(n, pi ),
that is, Π is the subspace of PG(n, pj ) coordinatized over GF(pi ).

In order to compute the collineation group fixing a Hermitian intersection of class VII, a
model of PG(2, q) that does not lie in canonical position in PG(2, q 3 ) is needed. The results
here presented are from [CK98] and [CK97].
Let F = GF(q), where q = ph and p is odd. Consider the field G = GF(q 3 ) as a cubic
extension of F; furthermore take b as a primitive (q 2 + q + 1)-th root of unity over G. The linear
collineation β of PG(2, G) given by

 X → bX
β: Y → bq+1 Y

Z→Z

has clearly order q 2 + q + 1 and fixes each vertex of the fundamental triangle X∞ Y∞ Z∞ of
PG(2, G).
Let B = hβi; the point orbit of E = (1, 1, 1) under B is the set
2 +q+1
Π = {(c, cq+1 , 1) : cq = 1, c ∈ G}.

Such a set induces a subgeometry in PG(2, G), whose points are the points of Π and whose
lines are the lines of PG(2, G) intersecting Π in at least 2 (and hence in q + 1) points. In fact,
this subgeometry is a projective plane, see [CKT99] and [CK98].

43
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

Theorem 2.2.4 ([CK98], Proposition 1). The subgeometry Π is isomorphic to PG(2, F). More
precisely, Π is a projective subplane of PG(2, E) lying in non-classical position. The lines of Π
have equation [tx + tq+1 y + z = 0], with t running on the (q 2 + q + 1)-th roots of unity and they
form the line-orbit of [x + y + z = 0] under the group B.

It can be proven, see [CKT99], that the collineation

 2
 X → bX + Y + bq +1 Z
2
κ: Y → bq +1 X + bY + Z
 2
Z → X + bq +1 Y + bZ

maps PG(2, q) into Π. Observe that β is a Singer cycle of Π which is represented in diagonal
form.
In order to simplify the notation, the symbol (i) will be used to denote the point (bi , bi(q+1) , 1)
of Π. Similarly, [i] will indicate the line of Π of equation [bi X + bi(q+1) Y + Z = 0].

2.2.3 Equations for the non-singular Hermitian curve

There is only one non-singular Hermitian curve up to projectivities in PG(2, q); however, when
describing intersection configurations, it is useful to have different models for the curves. Table
2.4 presents the standard equations that will be used.
√ √ √
(M 1) X √q+1 + Y q+1
√
+ Z√ q+1 = 0
(M 2) X q+1
√
+YZ√
q
+ Y q√Z = 0 √
(M 3) XY √q − X q√Y + ωZ √q+1 = 0 ω q−1 = −1
(M 4) XY q + Y Z q + ZX q = 0

Table 2.4: Equations for the non-degenerate Hermitian curve

All the models (M1)-(M3) of the Hermitian curve are GF(q)-equivalent, that is, there exists
a linear transformation in PGL(3, q) that maps one equation into the other.
The model (M1) is the one induced by the canonical Hermitian form and corresponds to
the identity matrix. Equations (M2) and (M3) allow to consider easily the affine points of the
Hermitian curve H: in (M2), the line at infinity l∞ : [Z = 0] is tangent to H at the point
Y∞ = (0, 1, 0); in (M3), the polar of the point Z∞ = (0, 0, 1) is the line l∞ and the intersection
√
between l∞ and H is a subline belonging to PG(2, q) ⊆ PG(2, q).
Let H be the curve in the plane Π corresponding to the equation (M4). We want to prove
that H is a Hermitian curve.
√
Lemma 2.2.5. The stabiliser of H in the group B is a subgroup K of order q − q +1 generated
√
by β (q+ q+1) .

44
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

Proof. The mapping β transforms the equation of H

√ √ √
q q q
H : XY +YZ + ZX =0

into
√ √ √ √ √
Hβ : (bq q+q+2 q+1
)[XY q
+YZ q
+ ZX q
] = 0;
hence, the group K stabilises H. Conversely, suppose that β l stabilises H. Then,
√ √
b(q q+q+1)l
= b(q+1)l = b ql
.

The result follows.

Lemma 2.2.6. The set H is a Hermitian curve of Π.

Proof. We prove that H is a classical unital. First, we verify that the mapping
½ √
(i) → [iq q]
ϕ: √
[i] → (iq q)

is a non-degenerate polarity. Since Π is a cyclic plane, it is enough to show that ϕ sends lines
through (0) to points incident with [0] and vice-versa. The line [i] is incident with the point
√ √ √
(0) if and only if bi + bi(q+1) + 1 = 0 that is biq q + bi(q+1)q q + 1 = 0. As (iq q) is the
image point of [i], the first assertion follows. A similar argument proves the converse. A direct
computation shows also that the set of all self-conjugate points of ϕ coincides with H. The
classification of polarities of PG(2, q) implies that the polarity ϕ is either orthogonal or unitary.
Hence, in order to get the result, it remains to prove that the former possibility cannot actually
occur. It is well known, see Lemmas 1.7.43 and 1.7.44, that the set of all self-conjugate points
of an orthogonal polarity is a (non-degenerate) conic for q odd and a line for q even. On the
other hand, no collineation group preserving either a conic or a line contains a cyclic subgroup
√
of order q − q + 1. Hence, Lemma 2.2.5 rules out the possibility for ϕ not to be unitary.

It follows that the model (M4) is not GF(q)-equivalent to (M1); however it is GF(q 3 )-
equivalent.

2.2.4 Groups preserving the intersection of two Hermitian curves

Let Aut(E) denote the linear collineation group preserving the Hermitian intersection E of H1
and H2 . As a subgroup of PGL(3, q), the group Aut(E) acts in a natural way as a permutation
group on the set of all Hermitian curves in the linear system Γ. More precisely, the three
(possibly empty) subsets Γi ⊆ Γ are invariant under the action of Aut(E).
Our approach to computing Aut(E) is to take H1 from Γ3 and H2 from Γ1 or from Γ2 , when
Γ1 is empty, and determine its subgroup G consisting of all linear collineations preserving both

45
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

H1 and H2 . This will be done by using several properties of the linear collineation group
PSU(3, q) including the classification of all maximal subgroups of PSU(3, q), see [Har26],
[Hof72], [Mit11].
The subgroup G turns out to be quite large and always transitive on the set of all non-
special points of E. Finally, in order to obtain the whole Aut(E) we will also need some direct
computations depending on the particular properties of E and the possible actions of Aut(E) on
the pencil Γ.

Class I

Let E be a Hermitian intersection in class I. A non-singular Hermitian curve H1 in the pencil Γ

can be assumed in its canonical form (M1),
√ √ √
q+1 q+1 q+1
H1 : X +Y +Z = 0.

Since the collineation group preserving H1 acts transitively on the points outside H1 , a Hermi-
tian cone in Γ can be assumed to have its vertex V in the origin O = (0, 0, 1). In particular,
both H1 and H2 are associated to a diagonal matrix, and this holds true for every curve in the
pencil. Under these assumptions, the three Hermitian cones in Γ are of the form
√ √
q+1 q+1
1. H2 : λX +Y = 0;
√ √
q+1 q+1
2. H3 : (λ − 1)Y + λZ = 0;
√ √
q+1 q+1
3. H4 : (1 − λ)X +Z = 0;
√
with λ ∈ GF( q)? \ {1}. A priori, the properties of E may depend on λ. However, as the
next Theorem 2.2.7 states, different choices of λ provide projectively equivalent Hermitian
intersections.

Theorem 2.2.7. Hermitian intersections in class I are projectively equivalent.

√
Proof. Let λ, λ̄ ∈ GF( q)? \ {1}. Then, there are elements u, v ∈ GF(q)? such that

√
q+1 λ−1 √
q+1 (λ − 1)λ̄
u = , v = .
λ̄ − 1 (λ̄ − 1)λ
Let γ be the linear collineation represented by the non-singular matrix
 
u 0 0
g =  0 v 0 .
0 0 1
√ √
q+1 q+1
The projectivity γ sends H2 and H3 to the Hermitian cones of equations λ̄X +Y =0
√ √
and (λ̄ − 1)Y q+1 + λ̄Z q+1 = 0. This proves the theorem.

46
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

Our next aim is to determine the abstract structure and the action of Aut(E) on the point-set.
To do this, the following lemma is needed.

Lemma 2.2.8. The collineation group G which preserves both the curve H1 and the cone H2
consists of all collineations

t(², η) : (X, Y, Z) → (²X, ηY, Z),

√ √
q+1 q+1
with ² =η = 1. In fact,

G ' C√q+1 × C√q+1 ,

and it acts on the point-set of E as a regular permutation group.

Proof. The linear collineations t(², η) preserve both the curves H1 and H2 . In fact,

t(², η)(H1 ) =
√ √ √ √ √ √
q+1 q+1 q+1 q+1 q+1 q+1
(²X) + (ηY ) +Z =X +Y +Z =
H1 ;

likewise t(², η)(H2 ) = H2 . To show the converse, let γ be a linear collineation of PG(2, q)
preserving both H1 and H2 . Since γ fixes the vertex (0, 0, 1) of the Hermitian cone H2 , the
non-singular unitary matrix associated to γ is a block diagonal matrix
 
a b 0
 c d 0 ,
0 0 1
√
with b = c q . Since
√ √
q+1 q+1
λ(aX + bY ) + (cX + dY ) =
√ √ √ √ √ √ √ √ √
T[(λ(a q b) + c q d)X q
Y ] + (λb q+1
+d q+1
)Y q+1
+ (λa q+1
+c q+1
)X q+1
,

the condition on γ to preserve H2 yields

√
q
(i) b(λa + d) = 0;
√ √ √ √
q+1 q+1 q+1 q+1
(ii) (λa +b ) = λ(λb +d ) 6= 0.
√
From (i), we have either b = 0 or b 6= 0 and d = −λa q . In the latter case,
√ √
q+1 q+1
b = −λa
√ √ √ √
and λa q+1 + b q+1 = 0, against (ii). This shows that b = 0. Then, (ii) implies a q+1 = d q+1 .
√
Also, from c q+1 = b = 0 it follows that c = 0. Hence, the collineation γ is indeed t(a, b) and
G ' C√q+1 × C√q+1 .

47
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

Note that E = H1 ∩ H2 has no point on the fundamental triangle of PG(2, q). On the other
hand, the group G does not fix any point outside the fundamental triangle. It follows that the
√
G-orbit of a point P ∈ H1 ∩ H2 has size ( q + 1)2 , the same as G and E. Hence, since E is
preserved by G, E coincides with the orbit of P under G, and G acts regularly.

Theorem 2.2.9. The linear collineation group Aut(E) preserving a Hermitian intersection in
√
class I acts transitively on the points of E. Furthermore, Aut(E) has order 3( q + 1)2 and

Aut(E) ' (C√q+1 × C√q+1 ) o S3 .

Proof. For any b, c, d ∈ GF(q) such that

√
q+1 1 √
q+1 (1 − λ)2 √
q+1
b =− , c =− , d = λ(1 − λ),
λ λ
the group Σ ∼
= S3 generated by the linear collineations

σ1 : (X, Y, Z) → (Z, dX, cY ),

σ2 : (X, Y, Z) → (bY, b−1 X, Z)

is a subgroup of the normaliser of G in Aut(E). In fact, hG, Σi = G o Σ: all collineations in
G fix the curves in the pencil Γ; hence, G ∩ Σ = 1. Furthermore, Σ preserves Γ2 , that is the set
{H2 , H3 , H4 } and, in particular, Σ is a subgroup of Aut(E). To show that hG, Σi = Aut(E), let
τ ∈ Aut(E). Since Σ induces the full symmetric group on Γ2 , there exists σ ∈ Σ such that στ
preserves each of the Hermitian cones H2 , H3 , and H4 . By virtue of the fact that the vertices
of these Hermitian cones are also the vertices of the fundamental triangle, it turns out that στ is
associated to a diagonal matrix  
η 0 0
 0 µ 0 .
0 0 1
√ √ √
As στ fixes H2 , we have η q+1 = µ q+1 . Also, µ q+1 = 1, because στ fixes H3 as well. This
shows that στ ∈ G, whence Aut(E) = hG, Σi. The claim now follows from the above results
together with Lemma 2.2.8.

Class II

Let E be a Hermitian intersection in class II. A non-singular Hermitian curve H1 in the pencil
Γ is assumed in the canonical form (M2),
√ √ √
q+1 q q
H1 : X +YZ + ZY = 0,

while a Hermitian cone with vertex Y∞ = (0, 1, 0), say

√ √
q+1 q+1
H2 : λX +Z = 0,

48
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

is chosen to generate Γ together H1 . One more Hermitian cone belongs to Γ, namely

√ √ √
q q q+1
H3 : λY Z + λZY −Z = 0.

Its vertex is the point at infinity X∞ = (1, 0, 0).

Theorem 2.2.10. Hermitian intersections in class II are projectively equivalent.
Proof. Let Γ and Γ̄ be two pencils which define Hermitian intersections belonging to class II,
and let Γ = hH1 , H2 i. Without loss of generality, we may assume that Γ̄ is generated by H1
together with the Hermitian cone of equation
√ √
q+1 q+1
λ̄X +Z = 0,
√
where λ̄ ∈ GF( q)? . Arguing as in the proof of Theorem 2.2.7, choose an element u ∈ GF(q)?
√
such that u q+1 = λ/λ̄. The linear collineation
√
q+1
ξ : (X, Y, Z) → (uX, u Y, Z).
√ √ √
q+1 q+1 q
sends H2 and H3 to the Hermitian cones of equations λ̄X +Z = 0 and λ̄Y Z +
√ √
λ̄ZY q − Z q+1 = 0, whence the claim follows.

We now determine the structure and the action on E of the linear collineation group Aut(E)
preserving E. By Theorem 2.2.10, we may assume without loss of generality λ = −1. Hence,
the vertex of H3 is the point Y∞ = (0, 1, 0).
Lemma 2.2.11. A linear collineation γ belongs to Aut(E) if and only if

γ(a, c, d) : (X, Y, Z) → (aX, Y + cZ, dZ),

with
√
(i) d ∈ GF( q)? ;

(ii) T(c) = 1 − d;

(iii) N[a] = d2 .
Proof. The collineation γ fixes the vertices (1, 0, 0) and (0, 1, 0) of both Hermitian cones in Γ.
Hence, it is represented by a non-singular matrix of the form
 
a 0 b
g =  0 1 c .
0 0 d
Since γ preserves H2 ,
√ √
q+1 q+1
γ(H2 ) = λ(aX + bZ) + (dZ) =
√ √ √ √ √ √
q+1 q+1 q+1 q+1
λa X + (b + d) Z + T[ab q XZ q
] = H2 .

Hence, given that a 6= 0, the following are necessary conditions for g in order to represent γ:

49
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

(i) b = 0;
√ √
q+1 q+1
(ii) a =d .

On the other hand, the image of H3 under γ is

√ √
γ(H3 ) = λ[T((Y + cZ)(dZ) q ] − (dZ) q+1
=
√ √ √ √ √
q q q q+1 q+1
λT[d Y Z + cd Z ] − (dZ )=
√ √ √ √ √
q q
λT[d Y Z ] + (T[cd q ] − d q+1
)Z q+1
.
√
q
Hence, d 6= 0 and γ(H3 ) = H3 , which together imply d = d, that is (i), and dT[c] − d2 = d,
that is (ii).

Lemma 2.2.12. Let G be the subgroup of Aut(E) preserving H1 . Then, we can write Aut(E) =
C√q−1 G, where C√q−1 is the cyclic group consisting of all collineations

φ(d) : (X, Y, Z) → (X, Y, dZ)

√
with d ∈ GF( q)? .

Proof. The action on H1 of a collineation γ which is represented by a matrix g as in Lemma

2.2.11 is as follows:
√ √ √
γ(H1 ) = d2 X q+1
+ dT[Y Z q
] + T[c]Z q+1
.

In order for the image to be H1 , we clearly need d = 1 and T[c] = 0. The latter yields N[a] = 1;
hence, every element in Aut(H) can be written as a the product of an element in C√q−1 by an
element in G.
√ √
Lemma 2.2.13. The group G has order ( q + 1) q and it is isomorphic to the semidirect
√
product of an elementary Abelian normal subgroup of order q by a cyclic group of order
√
q + 1.
√
Proof. The collineations γ(1, c, 1) form an elementary Abelian subgroup E√q of order q.
Likewise, the set of the collineations of the form γ(a, 0, 1) with N[a] = 1 constitutes a cyclic
√
subgroup C√q+1 of order q + 1. The generic element of G is of the form g(a, c, 1), with the
before mentioned conditions on c and a. In fact,

g(a, c, 1) = g(1, c, 1)g(a, 0, 1),

which implies the result.

Lemma 2.2.14. The group G acts on the points of E distinct from (0, 1, 0) as a regular permu-
tation group.

50
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

Proof. We keep using the notation introduced in the previous lemma. Let ∆2 be the set of
all generators of H2 . The cyclic component C√q+1 consists of elations of centre X∞ and axis
[X = 0]. Since [X = 0] is not an element of ∆, but C√q+1 preserves H2 , we obtain that C√q+1
acts in a fixed-point free manner on ∆. It follows that C√q+1 is regular on the generators of
H2 . Likewise, the elementary Abelian subgroup E√q of G acts transitively on the set of all
generators ∆3 of H3 distinct from [Z = 0]. Since any point of E is obtained as the intersection
of an element of ∆2 with an element of ∆3 , it follows that G is transitive on the points of E
√ √
distinct from (0, 1, 0). Finally, observe that G has order q( q + 1) which is equal to the size
of E \ (0, 1, 0).

By virtue of the above lemmas, both the abstract structure and the action of Aut(E) are
completely determined.

Theorem 2.2.15. The linear collineation group Aut(E) preserving a Hermitian intersection
E in class II acts transitively on the points of E distinct from the special point. Furthermore,
√
Aut(E) has order q(q − 1) and

Aut(E) ' C√q−1 × (E√q o C√q+1 ).

Class III

Let E be a Hermitian intersection in class III. A non-singular Hermitian curve H1 in the pencil
Γ is assumed to be in form (M3), that is
√ √ √
q q q+1
H1 : XY −X Y + ωZ = 0,
√
with ω q−1 = −1. Since the collineation group preserving H1 is doubly transitive on the points
of H1 , the two generators of the Hermitian cone H2 in E may be assumed to be the tangent lines
to H1 at the points (0, 1, 0) and (1, 0, 0). Then, H2 has equation
√ √
q q
H2 : XY − uY X = 0,
√
with u q+1 = 1. Actually, u 6= 1. In fact, every generator of H2 different from the axes must
√
be a chord of H1 , and this occurs if and only if u 6= 1. Hence, in our setting, we have just q
pairwise distinct Hermitian intersections.

Theorem 2.2.16. Hermitian intersections in class III are projectively equivalent.

√
q−1
Proof. For every t ∈ GF (q) such that u = t , define the linear collineation

θt : (X, Y, Z) → ((1 − t)−1 X, Y, Z).

The image of H1 under θt is

√ √ √ √
θt (H1 ) = (1 − t)−1 XY q
− (1 − t q )−1 X q
Y + ωZ q+1
:= H̄1 .

51
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

Likewise,
√ √ √ √ √
θt (H2 ) = (1 − t)−1 XY q
− u(1 − t q )−1 Y X q
= XY q
+X q
Y.

On the other hand, since H1 = H̄1 +t(t−1)−1 H̄2 , the collineation θt maps E into the Hermitian
intersection Ē generated by H1 and H̄2 . For two distinct values of t, the resulting Hermitian
intersections do not coincide. In fact, if t ∈ GF(q) also satisfies the above condition, that is
√ √ √
t̄ q−1 = u, and (1 − t)(1 − t q )−1 = (1 − t̄)(1 − t̄) q−1 holds, then we have (1 − t)(1 − tu)−1 =
(1−t̄)(1−ut̄)−1 , and the latter relation implies t = t̄. This shows that the family parametrized by
√
t consists of q − 1 pairwise distinct Hermitian intersections which are projectively equivalent
√
to E. None of them coincides with E, as (1 − t)(1 − t q ) = 1 implies u = 1 which is currently
ruled out. Adding E to that family, we obtain all possible Hermitian intersections, and this
completes the proof.

To determine the abstract structure and the action of Aut(E), we need some further prelim-
inary results.

Lemma 2.2.17. The linear collineation group G preserving both H1 and H2 consists of all
collineations
√
γ(a) : (X, Y, Z) → (a q+1 X, Y, aZ),
√
q+1
δ(a) : (X, Y, Z) → (−a Y, X, aZ),
with a ∈ GF(q)? . The subgroup H = {γ(a)|a ∈ GF(q)? } is a cyclic normal subgroup of
G, and it acts regularly on the points of E distinct from (1, 0, 0) and (0, 1, 0). Furthermore, if
√
a ∈ GF( q)? and q is even, then δ(a) is an involution and G = hδ(a)i o H; if q is odd, then
δ(a) has order 4 and G = C2 o H.

Proof. Let g be a linear collineation preserving both H1 and H2 . Then, g preserves the fun-
damental triangle. More precisely, g fixes the origin Z∞ = (0, 0, 1) and either interchanges
the points Y∞ = (0, 1, 0) and X∞ = (1, 0, 0), or fixes them both. In the former case, g is
represented by a diagonal non-singular matrix
 
b 0 0
M1 (a, b) =  0 1 0  .
0 0 a
√ √
The collineation g preserves H2 if and only if GF ( q)? contains b. For b ∈ GF ( q)? , the
√
condition on g to preserve H1 is equivalent to a q+1 = b. Hence, if g fixes the vertices of the
√
fundamental triangle, then g = γ(a) with a suitable element a ∈ GF ( q)? . A similar argument
shows that if g interchanges the vertices X∞ and Y∞ , then g is represented by the non-singular
matrix  
0 b 0
M2 (a, b) =  1 0 0  ,
0 0 a

52
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

√
and a necessary and sufficient condition for g to preserve both H1 and H2 is b = −a q+1 . This
completes the proof of the first statement. The group H is isomorphic to the multiplicative
√
group of GF(q); hence, it is cyclic of order q − 1. Since δ(b)−1 γ(a)δ(b) = γ(a q ) for all
a, b ∈ GF (q)? , the group H is normal in G. As no non-trivial element in the subgroup H fixes
a point outside the fundamental triangle, the orbit of any point P ∈ E under H has size q − 1,
the same as E. It follows that the orbit of a P ∈ E is the whole of E. This completes the proof
of the second statement.
By direct computation, the square of δ is
√ √
δ(a)2 : (X, Y, Z) → (−a q+1
X, −a q+1
Y, a2 Z).
√
Hence, when q is even, for any a ∈ GF( q)? , the collineation δ(a) is an involution and G =
hδ(a)ioH. When q is odd, δ(a) has period 4, but δ(a)2 = γ(−1) ∈ H; hence, G = C2 oH.

We are in position to prove the following theorem.

Theorem 2.2.18. The linear collineation group Aut(E) preserving a Hermitian intersection E
in class III acts transitively on the points of E distinct from the two special points. Furthermore,
Aut(E) has order 2(q − 1) and

Aut(E) ' C2 o Cq−1 .

Proof. We prove that every linear collineation g preserving E belongs to the group G introduced
in the previous lemma. Actually, g preserves H2 ; hence, it suffices to prove that g also preserves
H1 . As we have already noticed in the proof of Lemma 2.2.17, the condition on g to preserve
H2 implies that g is represented by either one of the matrices M1 (a, b) or M2 (a, b), with a ∈
√
GF (q)? , b ∈ GF ( q)? . In the former case, g sends H1 to the Hermitian curve H̄1 of equation
√ √ √ √
q q q+1 −1 q+1
XY −X Y + (a b )ωZ = 0.
√
On the other hand, the pencil Γ generated by H1 and H2 contains H̄2 if and only if a q+1 b−1 =
√
1. This only occurs for b = a q+1 , that is for g ∈ G. A similar argument shows that the same
holds when g is represented by M2 (a, b).

Class IV

Let E be a Hermitian intersection in class IV. The non-singular Hermitian curve

√ √ √
q+1 q q
H1 : X +YZ +Y Z=0

given by the model (M2), together with the Hermitian cone H2 of vertex (0, 1, 0) and equation
√ √ √
q q q+1
H2 : ZX − λXZ = 0, λ =1

can be chosen to generate the pencil Γ.

53
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

Theorem 2.2.19. Hermitian intersections in class IV are projectively equivalent.

√ √
Proof. Let λ, λ̄ be elements of GF(q) with λ q+1 = λ̄ q+1 = 1. Choose a ∈ GF(q)? such that
√
a q−1 = (λ̄/λ). Arguing as in the proof of Theorem 2.2.10, it suffices to check that the linear
collineation
γ(a) : (X, Y, Z) → (aX, Y, Z)
preserves H1 and sends H2 to the Hermitian cone H̄2 of equation
√ √
q q
H̄2 : ZX − λ̄XZ = 0.

The previous lemma guarantees that in order to determine Aut(E) we may assume without
loss of generality λ = 1.

Lemma 2.2.20. The collineation group G which preserves both H1 and H2 consists of all
collineations 
 X → aX + cZ
t(a, c, f ) : Y → −acX + a2 Y + f Z

Z → Z,
√ ? √
with a ∈ GF( q) , c ∈ GF( q) and T[f ] = −N[c].

Proof. Via a direct computation, we have

√ √ √
t(H1 ) = a2 X q+1
+ a2 T[Y Z q
] + (c2 + f )Z q+1
,

t(H2 ) = aH2 .
Hence, any collineation t(a, c, f ) belongs to G. Conversely, let γ be a linear collineation pre-
serving H2 . Then, γ is an elation of centre (0, 1, 0) and axis [Z = 0]. It follows that γ is
associated to a non-singular matrix  
a 0 c
 d e f ,
0 0 1
√ ? √
with a ∈ GF( q) and c ∈ GF( q). Since γ preserves H1 as well, then d = −ac, e = a2 and
T[f ] = −N[c].

Lemma 2.2.21. The subgroup T of G consisting of all collineations of the form t(1, 0, f ) is
√ √
elementary Abelian of order q. In fact, T ' E√q , and G/T ' AGL(1, q). The group T is
the translation subgroup of G.

54
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

√
Proof. The order of t(1, 0, f ) is, for any suitable f , equal to the characteristic p of GF( q) and
any two elements in T commute; hence, T is an elementary Abelian group. Since T(f ) = 0
√ √
has q solutions, T has order q and T ' E√q . Let now Ḡ be the permutation group induced
by G on the set ∆ of all generators of H2 . We show that the kernel K of the permutation
representation G → Ḡ is T . It is immediate to verify that any element of T fixes all the lines
through (0, 1, 0); hence, T ≤ K. On the other hand, the generic line through (0, 1, 0) is of the
form
l : [αZ + X] = 0,
with α ∈ GF(q) ∪ {∞}. Any linear collineation t of G acts on l by transforming it into the line
(α + c)
lt : [ Z + X] = 0.
a
Hence, if a 6= 1 or c 6= 0, the collineation t(a, c, f ) fixes at most two generators of H2 . It follows
that T is the full kernel of the representation of G into Ḡ. The proof is completed by observing
that, according to Lemma 2.2.20, Ḡ acts on ∆ as the group of all permutations X → aX + c
√
with a 6= 0 and c ranging over GF( q).

The abstract structure and the action of the linear collineation group Aut(E) are given in the
following theorem.

Theorem 2.2.22. The linear collineation group Aut(E) preserving a Hermitian intersection E
in class IV acts transitively on the points of E distinct from the special point. Furthermore,
√
| Aut(E)| = q( q − 1) and
√
Aut(E)/E√q ' AGL(1, q).

Proof. The translation group T acts transitively on the common points of E and any affine
line through (0, 1, 0). Also, as seen in the proof of the previous lemma, G acts transitively
on the generators of H2 . This proves the transitivity of G on the points of E distinct from
(0, 1, 0). It remains to show that G coincides with Aut(E). Take g ∈ Aut(E) and let ḡ be the
permutation induced by g on the set ∆ of the generators of H2 . If g is in the kernel of the
permutation representation Aut(E) → Aut(E), then g is a translation. Let T 0 = hT, gi be the
group generated by T and g. Then, T 0 is again a translation group. Hence, no non-trivial element
in T 0 fixes an affine point. On the other hand, T 0 preserves the set of all common points of E
√
and one affine line through (0, 1, 0). This yields that T 0 has order at most q; hence T 0 = T ,
that is g ∈ T . The factor group Aut(E)/T induces on ∆ a permutation group containing G/T .
Since Aut(E)/T preserves H2 , it follows that Aut(E)/T consists of permutations

X → aX + b,
√ √
with a ∈ GF( q)? , b ∈ GF( q). This proves that G/T = Aut(E)/T and, therefore, G =
Aut(E).

55
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

Class V

Let E be a Hermitian intersection in class V. The non-singular Hermitian curve

√ √ √
q+1 q q
H1 : X +YZ +Y Z = 0,

given by (M2), together with the totally degenerated Hermitian cone of equation
√
q+1
H2 : X =0

can be chosen to generate the pencil Γ. From Theorem 2.1.4 and Corollary 1.7.26, every Her-
mitian intersection in class V is a Baer subline of PG(2, q). Arguing as in section 2.2.4 or,
alternatively, using classical results from finite geometry, see [Hir98b], the following theorems
can be proved.

Theorem 2.2.23. Hermitian intersections in class V are projectively equivalent.

Theorem 2.2.24. The linear collineation group Aut(E) preserving a Hermitian intersection E
√
in class V acts 3-transitively on the points of E and has order q 2 q(q − 1)2 (q + 1). Let AG(2, q)
be the affine plane whose infinite line contains E and let O be a point of AG(2, q). Then, the
subgroup K of Aut(E) fixing E point-wise is the semidirect product of the full translation group
T of AG(2, q) by the group of all dilatations of AG(2, q) with centre O. Furthermore,
√
Aut(E)/K ' PGL(1, q).

Class VI

Let E be a Hermitian intersection in class VI. The non-singular Hermitian curve

√ √ √
q+1 q q
H1 : X +YZ +Y Z = 0,

given by (M2), together with the totally degenerated Hermitian cone

√
q+1
H2 : Z =0

can be chosen to generate the pencil Γ. Every Hermitian intersection in class VI is reduced to a
single point of PG(2, q). Then, the following theorems hold.

Theorem 2.2.25. Hermitian intersections in class VI are projectively equivalent.

Theorem 2.2.26. The linear collineation group Aut(E) preserving a Hermitian intersection E
in class VI has order q(q + 1)(q − 1)2 and is isomorphic to AGL(2, q).

56
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.2. GROUPS OF THE INTERSECTION OF TWO HERMITIAN CURVES

Class VII

Let E be a Hermitian intersection in class VII. Then, [BS86], [Cos97], [Ebe85], [FHT86] and
√
[Kes89] have proven that E is the point-orbit of a Singer subgroup of order (q − q + 1) and its
√
points form a complete (q − q + 1)-arc in PG(2, q). In this case, we will use the model (M4)
for H1 :
√ √ √
H1 : XY q + Y Z q + ZX q = 0.
Observe that the collineation β maps the curve H associated with the equation (M4) into the
Hermitian curve of equation
√ √ √ √ √
H1β : bq q+1
XY q
+ bq− q+1
YZ q
+ ZX q
= 0.

Since H1 6= Hβ , we may choose Hβ as H2 . The intersection E = H1 ∩ H1β consists of the

points of Π that satisfy the following system of equations:
 √ √ √
 XY √q + Y Z √q + ZX √q = 0 √ √
(aq− q+1 − aq q+1 )Y Z q + (1 − aq q+1 )ZX q = 0

Z = 1.

It follows that
2 √ √
√
q aq +q+1 − aq q+1 √ aq(q− q+1) − 1
q
√ √
q (q− q+1 q−1
√
q
Y = −X q−
√
q+1 q
√
q+1
= X √
q(q−
√
q+1
= X (a − 1) = X .
a −a a −1
√ √
Hence, E is represented by all the points of the form (², ² q , 1) with ²q− q+1 = 1 and it is a set
√
of cardinality q − q + 1. From 2.1.4, it follows that E is an arc and that the intersection of H1
and H2 = H1β is in class VII.

Theorem 2.2.27. Hermitian intersections in class VII are projectively equivalent.

Proof. This theorem is a corollary to the known result that any two Singer subgroups of the
same order are conjugate under the full linear collineation group PGL(3, q) of PG(2, q), see
[Blo67], [Har26] and [Mit11].

Theorem 2.2.28. The linear collineation group Aut(E) preserving a Hermitian intersection E
in class VII is transitive on the points of E. Furthermore, Aut(E) contains a normal cyclic
√
subgroup of order q − q + 1 acting regularly on the points of E and

Aut(E) = C3 o Cq−√q+1 .
√ √
Proof. In the above model, the Singer subgroup S of order q− q+1 generated by γ = β q+ q+1
preserves E, as it preserves both H1 and H2 . The same holds true for the linear collineation
group E of order 3 generated by

τ : (X, Y, Z) → (Y, Z, X).

57
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.3. A GROUP-THEORETIC CHARACTERIZATION OF HERMITIAN CURVES AS
CLASSICAL UNITALS

Since E normalises S, the group G = hE, Si is the semidirect product of S by E. Hence,

G ' C3 o Cq−√q+1 .
√
To prove that Aut(E) = G it will be useful to regard E as a (q − q + 1)-arc. Let Λ be the
algebraic envelope associated to E, viewed as an algebraic curve in the dual plane of PG(2, q).
Clearly, Aut(E) is an automorphism group of Λ. For q even, Λ is a projectively equivalent
to a non-singular Hermitian curve H, see [Tha87]. The same holds for q odd, provided that
projective equivalence is replaced by birational equivalence, see [CK98]. In any case, Aut(E)
turns out to be isomorphic to a subgroup L of PGU(3, q). Since G ≤ Aut(E), L contains a
subgroup isomorphic to G. Then, the assertion follows from the classification of all maximal
subgroups of PGU(3, q), see [Har26], [Hof72] and [Mit11]. In fact, the subgroups of PGU(3, q)
√
which are the semidirect product of a cyclic group of order 3 by a cyclic group of order q− q+1
are all maximal in PGU(3, q).

2.3 A group-theoretic characterization of Hermitian curves

as classical unitals
The non-canonical model presented in 2.2.2 is used in order to provide a short proof of an
already known characterization of classical unitals [CEK00]. This work has been submitted for
publication [Giua].

2.3.1 Introduction
We recall some basic definitions. A unital in a Desarguesian projective plane PG(2, q) of square
√ √
order q, is a set U of q q + 1 points such that any line of PG(2, q) meets U in either 1 or q + 1
points. The absolute points of a unitary polarity of PG(2, q) form a unital which is called
the classical (or Hermitian) unital. The linear collineation group of PG(2, q) preserving a
classical unital is PGU(3, q). By a theorem due to Hoffer [Hof72] this group-theoretic property
characterises classical unitals: if a unital U is preserved by a collineation group isomorphic
to PSU(3, q), then U is classical. Cossidente, Ebert and Korchmáros [CEK00] showed that
Hoffer’s result holds true under some weaker assumption, namely for unitals preserved by a
√
Singer subgroup of PGL(3, q) of order q − q + 1. Their proof heavily depends on previous
results concerning cyclic partitions of PG(2, q) in Baer sub-planes. A different and shorter
proof of this result is the purpose of the present section.
Theorem 2.3.1. A unital U in PG(2, q) is classical if and only if it is preserved by a cyclic
√
linear collineation group of order q − q + 1.

2.3.2 A result on classical unitals

The following remarkable property of classical unitals is from [CEK00].

58
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.3. A GROUP-THEORETIC CHARACTERIZATION OF HERMITIAN CURVES AS
CLASSICAL UNITALS

Theorem 2.3.2 (Proposition 3.2 [CEK00]). Let q = pk . Every unital U of the projective plane
PG(2, q) meets every classical unital in r ≡ 1 mod p points.

Proof. Let l1 , . . . , lr be a set of lines of PG(2, q) and let ¯li be the characteristic vector of li . By
√
definition of unital, a line l intersects U in either 1 or q + 1 points. Let ū be the characteristic
vector of U. Then, the dot product between ū and any combination of lines can be evaluated as

ū · (¯l1 + . . . + ¯lr ) ≡ r mod p.

A theorem due to Blokhuis, Brouwer and Wilbrink [BBW91] (see also [AK92, 6.7.1]), states
that a classical unital H is a codeword in the GF(p)-code generated by the lines of PG(2, q).
This is to say that the characteristic vector h̄ of H is a combination of characteristic vectors
¯l1 , . . . , ¯lr of the lines of the projective space. Since the number of points of H is q √q + 1,
working modulus p,

1 = h̄ · h̄
= h̄ · (¯l1 + . . . ¯lr )
= h̄ · ¯l1 + . . . h̄ · ¯lr ,

and r ≡ 1 mod p. Now, the dot product between ū and h̄ can be computed as

ū · h̄ = ū · (¯l1 + . . . + ¯lr ) = ū · ¯l1 + . . . + ū · ¯lr ≡ r mod p.

The result follows.

2.3.3 Proof of Theorem 2.3.1

Here, we consider a Singer collineation group acting on the points and lines of the non-canonical
model Π of PG(2, q) in PG(2, q 3 ) as introduced in section 2.2.2.
Assume H to be the Hermitian curve of Π with equation (M4) and define K as its stabiliser
in B. Since B is Abelian, K is normal in B and it is possible to construct a quotient incidence
structure Π0 in the following way: define thick points as the point-orbits of Π under K, thick
lines as the orbits of H under B, and incidence as inclusion. Note that the factor group B/K
is a Singer group for Π0 , as it acts regularly on the set of thick-points as well as on the set of
thick-lines.
We need the following characterization of finite projective planes as incidence structures.

Lemma 2.3.3 (Proposition 3.2.3(m), [Dem68]). An incidence structure (P, L, I) is a projective

plane of order n if and only if

(i) |P | ≤ n2 + n + 1;

(ii) |L| ≥ n2 + n + 1;

59
CHAPTER 2. THE 2-DIMENSIONAL CASE: HERMITIAN CURVES
2.3. A GROUP-THEORETIC CHARACTERIZATION OF HERMITIAN CURVES AS
CLASSICAL UNITALS

(iii) every line in L contains at least n + 1 points;

(iv) any two distinct lines of L meet in at most 1 point.

√
Lemma 2.3.4. The incidence structure Π0 is a projective plane of order q.
√
Proof. Since the index [B : K] is equal to q + q + 1, the number of thick-points in Π0 is
√
q + q + 1. According to Lemma 2.2.5, the subgroup of B which preserves H is K; hence, we
√ √
have q + q + 1 thick lines as well. Furthermore, every thick line is incident with q + 1 thick
√
points, as the size of H is q q + 1. By Lemma 2.3.3, in order to prove that Π0 is a projective
√
plane of order q, it now suffices to verify that two thick lines share at most one thick point.
Since thick-lines are classical unitals, for q > 9 the assertion follows from the fact that the
√
number of common points of two distinct classical unitals in Π is at most ( q + 1)2 , smaller
√
than 2(q − q + 1). For q = 4, 9 a direct counting argument proves the assertion.

We are now in position to prove Theorem 2.3.1.

Proof of Theorem 2.3.1. For any n | q 2 + q + 1, all Singer subgroups of order n are conjugate
in P GU (3, q); hence, we may assume without loss of generality that U is a unital in Π which
√
is preserved by K. This hypothesis implies that U is the union of q + 1 point-orbits under K,
√
and hence, the unital U can be viewed as a set ∆ of q + 1 thick points in Π0 . In order to prove
Theorem 2.3.1, it remains to show that ∆ is actually a thick line, or, equivalently, that U meets
H and every image of H under the action of B; this is a corollary of Theorem 2.3.2.

60
Chapter 3

The 3-dimensional case: Hermitian

surfaces

This chapter deals with the point-line-plane incidence configurations arising from the intersec-
tion of two Hermitian surfaces. First, we consider which intersections fulfill some combinatorial
conditions in order to be possible. Later, we determine classes of matrices that allow us to di-
rectly construct such intersections. For the purposes of this chapter, we mean by Hermitian
cone a degenerate Hermitian surface of rank 3.

3.1 Description of incidence configurations in dimension 3

In this section we describe the possible point-line configurations arising from the intersection
of two Hermitian surfaces in PG(3, q). The argument here is purely combinatorial. For a
description of the geometric cases which can actually be realized in PG(3, q) and models of the
intersections, please see section 3.3.
The following result, which will be proven in Chapter 4 for dimension n, is needed.

Lemma 3.1.1. Let H1 , H2 be two distinct Hermitian surfaces; then, the size of the intersection
E = H1 ∩ H2 depends only upon the number and rank of the degenerate surfaces in the linear
system Γ = hH1 , H2 i. Define as ri the number of surfaces in Γ of rank i; then, the size of E for
the various possible values of ri is as in Tables 3.1, 3.2 and 3.3.

Definition 3.1.2. A linear system Γ is non-degenerate if it contains at least a non-degenerate

Hermitian surface.

Definition 3.1.3. We say that two points are conjugate with respect to the linear system Γ if
they are conjugate with respect to any non-degenerate surface in Γ.

61
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.1. DESCRIPTION OF INCIDENCE CONFIGURATIONS IN DIMENSION 3

r1 r2 r3 k = |E|
0 0 0 (q + 1)2
√ √
0 0 1 (q + q + 1)(q − q + 1)
0 0 2 (q 2 + 1)
0 0 3 q2 − q + 1
0 0 4 (q − 1)2
√
0 1 0 q2 + q q + q + 1
√
0 1 1 q2 + q q + 1
√ √
0 1 2 ( q + 1)(q q − q + 1)
√ √ √
0 2 0 ( q + 1)(q q + q − q + 1)
√
1 0 0 q q+q+1
√
1 0 1 q q+1

Table 3.1: Possible intersection numbers for Hermitian surfaces: non-degenerate pencil.

r1 r2 r3 k = |E|
√ √
0 0 q+1 q2 − q q + q + 1
√ √ √
0 1 q (q + q + 1)(q − q + 1)
√ √ √
0 2 q−1 ( q + 1)2 (q − q + 1)
√ √ √
0 3 q−2 q 2 + 2q q + q + 1
√
1 0 q q+1
√ √
1 1 q−1 q q+q+1

Table 3.2: Possible intersection numbers for Hermitian surfaces: degenerate pencil; r3 6=
0, r4 = 0.

r1 r2 k = |E|
√
0 q+1 2q 2 + q + 1
√
1 q q2 + q + 1
√
2 q−1 q+1

Table 3.3: Possible intersection numbers for Hermitian surfaces: degenerate pencil; r2 6=
0, r3 = r4 = 0.

62
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.1. DESCRIPTION OF INCIDENCE CONFIGURATIONS IN DIMENSION 3

3.1.1 Some general remarks on cones

Lemma 3.1.4. Assume r3 (Γ) ≥ 1 and r4 (Γ) ≥ 1. Take a cone C of vertex V in Γ. Let π be the
polar plane of V with respect to a non-degenerate surface H ∈ Γ, and define k = |C ∩ H| and
h = |P| where
P := C ∩ H ∩ π.
Then, if V 6∈ H,
√ √ √
k = q 2 + q q + q + 1 − qh;
if V ∈ H,
√ √ 1
k = q2 + q + 1 + (h − 1)( q − √ ).
q
Proof. All tangents to the surface H through V are of the form V P with P ∈ π ∩ H; hence, all
components of the cone C tangent to H are of the form V P 0 with P 0 ∈ P, while the remaining
√ √
(q q + 1 − h) lines are secant, that is intersecting H in q + 1 distinct points.
Hence,
√ √
k = h + (q q + 1 − h)( q + 1),
and the first part of the lemma is proven.
Assume now that V ∈ H. This implies that V ∈ π and π is tangent to H. Hence, π ∩ H
√
decomposes in q + 1 lines. On the other hand, π intersects C in either one point or in a
√
Hermitian cone, that is in q + 1 lines. Let j = (h − 1)/q be the number of lines of π in
common between C and H. Each of these lines is fully included in H ∩ C. All lines of C not on
√
π are secant to H, that is they intersect H in q + 1 points, one of those is V . It follows,
√ √
k= q(q q + 1 − j) + 1 − jq,

that is
√ √ 1
k = q2 + q + 1 + (h − 1)( q − √ ).
q

Corollary 3.1.5. Assume C to be a Hermitian cone of vertex V 6∈ H and let π be the polar plane
of V . As usual, denote by ri the number of Hermitian surfaces of rank i in the linear system Γ.
Then, the intersection between C and H belongs to one of the classes presented in table 3.4.

Proof. The proof is done directly by a counting argument.

Corollary 3.1.6. Assume that r1 (Γ) = r2 (Γ) = 0, while r3 (Γ), r4 (Γ) ≥ 1, and let C1 and C2 be
distinct cones in Γ of vertices respectively V1 and V2 . If V2 6∈ E, then V1 belongs to the polar
plane π of V2 with respect to any non-degenerate surface in Γ.

63
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.1. DESCRIPTION OF INCIDENCE CONFIGURATIONS IN DIMENSION 3

h r1 r2 r3 |U3 ∩ K| (r10 , r20 )

√
q− q+1 0 0 1 q2 + q + 1 (0, 0)
q+1 0 0 2 q2 + 1 (0, 1)
√
q+ q+1 0 0 3 q2 − q + 1 (0, 2)
√
( q + 1)2 0 0 4 q 2 − 2q + 1 (0, 3)
√
1 0 1 1 q2 + q q + 1 (1, 0)
√ √
q+1 0 1 2 q2 + q q − q + 1 (1, 1)
√ √
q q+1 1 0 1 q q+1 N.A.

Table 3.4: Possible intersections E between a cone and a non degenerate Hermitian surface;
vertex not in the intersection.
The column (r10 , r20 ) describes the type of planar configuration in E ∩ π, where π is the polar
plane of the vertex of C.

Proof. Let π be the polar plane of V2 with respect to a non-degenerate surface H ∈ Γ. Consider
√
the line l = V1 V2 . If V1 6∈ π, then l intersects E = C1 ∩ C2 in either q + 1 or q + 1 points, but
V2 6∈ E implies that the intersection in q + 1 points is not possible. On the other hand, a line
through the vertex of a cone intersects the cone in either 1 or q + 1 points – a contradiction. It
follows that V1 ∈ π.

Corollary 3.1.7. Let C1 and C2 be two Hermitian cones in a non-degenerate linear system Γ.
Assume that the parameters of Γ are of the form (0, r2 , r3 ), and let π be the polar plane of the
vertex of C2 with respect to any non-degenerate surface in Γ. Then, the intersection Σ of C1 with
π is a curve of a class corresponding to the parameters (r2 , r3 − 1).

Proof. The claim follows from the observation that π ∩ C1 is a degenerate curve. By looking up
the orders in Table 2.1 we complete the proof.

Lemma 3.1.8. The configurations given by Σ are uniquely determined up to projectivities by

the rank sequence (r1 , r2 , r3 ).

Proof. From Theorem 2.2.1, all plane intersections determining the same point-line configu-
ration are projectively equivalent. Furthermore, following [Kes81], we see that a point-line
configuration is uniquely determined by its cardinality k except in the case (r10 , r20 ) = (0, 1),
when k = q + 1.
The case k = q + 1, corresponds to the situation in which Γ contains a cone C whose
vertex V does not belong to E and the rank sequence is (0, 0, 2). A priori, there would be two
possibilities for E ∩ π:
√
(i) the intersection consists of q − 1 sublines, all disjoint;

(ii) the intersection consists of q sublines, all with a point in common.

64
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.1. DESCRIPTION OF INCIDENCE CONFIGURATIONS IN DIMENSION 3

In both cases, the lines containing the points of the intersection belong also to the cone of the
pencil whose vertex V 0 is in the polar plane π of V . On the other hand, in case (ii), such vertex
V 0 has to belong to H as well and the size of E would have to be q 2 + q + 1, while, from Table
3.4, we already know |E| = q 2 + 1; a contradiction. It follows that the surfaces intersect in π in
the configuration given by (i).

Lemma 3.1.9. Let Γ be a non-degenerate pencil of Hermitian surfaces and assume that there
is at least a cone C in Γ whose vertex V does not belong to the base locus E. Then, the vertices
of any other cone in Γ belong to the polar plane π of V with respect to any non-degenerate
surface H ∈ Γ.

Proof. The intersection of E with π corresponds to the configuration obtained as base locus
of a pencil of Hermitian curves with parameters (r10 , r20 ) = (r2 , r3 − 1). Hence, for r3 ≥ 2,
√
the section of any cone C 0 ∈ Γ different from C with the plane π is a set of q + 1 lines, all
concurrent in a point V 0 ∈ π. It follows that V 0 has to be the vertex of C 0 .

3.1.2 Pencils whose degenerate surfaces have all rank 3

In this subsection we describe the possible incidence configurations E when the pencil Γ is
assumed to contain only cones as degenerate surfaces.
The vertex of any cone Ci ∈ Γ is denoted with the letter Vi . By H we denote, as usual, a
generic non-degenerate surface in Γ.
For any cone C ∈ Γ, the indices si are defined in Table 3.5.

s1 number of components of C which are included in E

s2 number of components of C which are chords of E
s3 number of components of C tangent to E.

Table 3.5: Indices for the intersection of a cone and a Hermitian surface.

Lemma 3.1.10. Assume that Γ contains 4 cones C1 , . . . , C4 . Then, for any i = 1, . . . , 4, Vi 6∈ E;

√
furthermore, all components of any cone Ci intersect E in q + 1 points.

Proof. The cardinality of E is (q −1)2 . Assume that the vertex of a cone C1 in Γ does not belong
√
to E. Then, each line of C1 intersects H in either q + 1 or 1 points. Hence,
√ √
(q − 1)2 = s2 ( q + 1) + (q q + 1 − s2 );
√ √
this is equivalent to say that there are s2 = q q − q − 2 q lines of C1 which are chords of E
√
and q + 2 q + 1 lines which are tangent to E. Furthermore, since this configuration contain no
line, the vertex of none of the cones in Γ may belong to E.

65
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.1. DESCRIPTION OF INCIDENCE CONFIGURATIONS IN DIMENSION 3

Assume now that the vertex V of C belongs to Γ. Then,

√ √
(q − 1)2 − 1 = s1 (q) + s2 ( q) + (q q − s1 − s2 ).

This would yield

√ √ √
q(q − q − 1) = ( q − 1)[s1 ( q + 1) + s2 ],
√
a contradiction, since s1 and s2 are both natural numbers and ( q − 1) does not divide q(q −
√
q − 1). The result follows.

Lemma 3.1.11. Assume that Γ contains exactly 3 distinct cones and assume also that there is
√ √
at least a cone C ∈ Γ whose vertex V is not in E. Then, q(q − q − 1) components of C are
√
chords of H; furthermore, any cone in Γ whose vertex is in E contains q q − 1 chords and 2
tangents to H.

Proof. Since V is not in H, then

√ √
q 2 − q + 1 = s2 ( q + 1) + (q q − s2 + 1).

Hence,
√ √
q[q − q − 1] = s2 ( q)
√ √ √
and s2 = q(q − q − 1) components of C are chords of H; the remaining q + q + 1
components are tangent to the surface. On the other hand, if C 0 is a cone of Γ whose vertex
belongs to H, then
√
q 2 − q + 1 = s01 (q) + s02 ( q) + 1.
Hence,
√ √
q(q − 1) = ( q)s01 + s02 .
The result follows now by observing that V 6∈ H implies s01 = 0.

Lemma 3.1.12. Let Γ be a pencil with r3 (Γ) ≥ 2. Assume that the vertices V1 , V2 of any two
cones C1 , C2 in Γ belong to E; then, the line V1 V2 is included in E.

Proof. The line V1 V2 intersects E in at least two points. Since Γ is generated by C1 and C2 , this
implies that V1 V2 is a component of both C1 and C2 . The result follows.

Lemma 3.1.13. Assume r3 (Γ) = 3, r4 (Γ) ≥ 0. Then, either the vertex of at most one cone of Γ
belongs to E or the vertices of all three cones in Γ lie on the same line in E.

Proof. Assume that the vertices V1 and V2 of two cones C1 and C2 of Γ are points of E. Then,
according to Lemma 3.1.12, the line V1 V2 has to be included in E as well, and s01 = 1. On the
other hand, C1 and C3 generate Γ and the only lines completely in C3 are those through V3 . The
lemma follows.

66
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.1. DESCRIPTION OF INCIDENCE CONFIGURATIONS IN DIMENSION 3

√
Lemma 3.1.14. Let r3 (Γ) = 3 and assume r4 (Γ) ≥ 0. Then, E contains one line and q(q − 2)
sublines.
Proof. From the previous lemma, s01 = 1. Hence,
√
q 2 − q + 1 = (q + 1) + s02 ( q).
The result follows.
Lemma 3.1.15. Assume r3 (Γ) = 2. Then, one of conditions (i)-(iii) has to be satisfied.
Proof. Let C be a cone of Γ with vertex V .
If V 6∈ E, then
√ √
q 2 + 1 = s2 ( q + 1) + (q q + 1 − s2 ).
Hence,
√
s2 = (q q − q).
If V ∈ H, then
√
q 2 = s1 (q) + s2 ( q).
Hence,
√
s2 = q(q − s1 ).
It follows that there are three possibilities:
√
(i) the vertex of both cones in Γ belongs to E; then, s1 = s01 = 1 and s02 = s2 = q q − 1;
√
(ii) the vertex of a cone in Γ belongs to E, the other not; then, s1 = 0, s2 = q( q − 1) and
√
s02 = q q;
√
(iii) the vertex of none of the cones in Γ belongs to H; then, s2 = s02 = q( q − 1).

Lemma 3.1.16. Assume that r3 (Γ) = 1 and let C be the only cone in the pencil. If the vertex
√ √
V of C does not belong to E, then q[q − q + 1] lines of C are chords; if V ∈ E, then the
configuration E contains at least a line.
Proof. If the vertex V of C does not belong to H, then
√ √
q 2 + q + 1 = s2 ( q + 1) + (q q + 1 − s2 ).
Hence,
√ √
s2 = q[q − q + 1].
If the vertex V of C belongs to H, then
√
q 2 + q + 1 = s1 (q) + s2 ( q) + 1;
√ √
hence, s2 = q(q + 1 − s1 ). Since in this case s2 ≤ q q + 1, it follows s1 ≥ 1. On the other
√ √ √
hand, since s1 ∈ {1, q + 1}, we obtain s2 ∈ {q q, q q − q}.

67
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.1. DESCRIPTION OF INCIDENCE CONFIGURATIONS IN DIMENSION 3

Class c3 v s01 s02 s2 k

√
a 4 0 - - q q+1 (q − 1)2
√ √
b 3 0 - - q(q − q − 1) q2 − q + 1
√ √ √
c 3 1 0 q(q − 1) q(q − q − 1) q2 − q + 1
√ √ √
d 3 3 1 q(q − 2) q(q − q − 1) q2 − q + 1
√
e 2 0 - - q( q − 1) q2 + 1
√ √
f 2 1 0 q q q( q − 1) q2 + 1
√
g 2 2 1 q( q − 1) - q2 + 1
√ √
h 1 0 - - q(q − q + 1) q2 + q + 1
√
k-I 1 1 1 q q - q2 + q + 1
√
k-II 1 1 2 q(q − 1) - q2 + q + 1
√ √
k-III 1 1 q+1 q( q − 1) - q2 + q + 1

Table 3.6: Possible incidence classes for two non-degenerate Hermitian surfaces: Γ contains
degenerate surfaces of rank 3 only.

Table 3.6 presents all possible incidence configurations obtained as base locus of linear
systems Γ whose only degenerate surfaces are Hermitian cones. The first column contains the
total number c3 of cones in a given configuration. The number v is the number of cones in the
pencil Γ whose vertex belongs to the intersection. The numbers s1 and s2 are as in Table 3.5 for
a cone C whose vertex does not belong to E. The integers s01 and s02 are defined in the same way
for a cone C 0 whose vertex is assumed to be in E.

3.1.3 Pencils whose degenerate surfaces have all rank 2

This subsection deals with the case in which all degenerate surfaces in the linear system Γ have
rank 2, that is to say r3 (Γ) = r1 (Γ) = 0. We recall that the radical of a Hermitian form h is the
set of the points x of PG(n, q) such that for all y ∈ PG(n, q),

h(x, y) = 0.

Definition 3.1.17. The radical of a Hermitian surface H is the radical of the Hermitian form h
associated with H. We denote the radical of H with the symbol rad H.

Lemma 3.1.18. The radical of a Hermitian variety H is a subspace of PG(n, q). Furthermore,
the following equality holds:
dim rad H + rankH = n.

68
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.1. DESCRIPTION OF INCIDENCE CONFIGURATIONS IN DIMENSION 3

Proof. Consider the linear transformation φ induced by H. We have, dim ker φ+dim Im φ = n;
on the other hand, dim ker φ = dim rad H and dim Im φ = rankH. The result follows.

Lemma 3.1.19. Assume that the only degenerate surface C of Γ has rank 2; then, the radical of
√
C intersects any non-degenerate surface H ∈ Γ in either 1, q + 1 or q + 1 points.

Proof. Any plane which is a component of C, intersects H in either a Hermitian curve or in a de-
√
generate Hermitian curve, that is q+1 lines through a point. In the former case the intersection
√ √
consists of q q + 1 points; in the latter of q q + q + 1. Define v1 as the number of components
of C which intersect H in a degenerate curve and let v2 be the number of components secant to
√
H, that is, whose intersection is a non-singular curve. Obviously, v1 + v2 = q + 1.
Since C has rank 2, its radical l = rad C is a line of PG(3, n). Hence, the possible in-
√
tersection numbers with H are 1, q + 1 or q + 1. Let n = |l ∩ H|. We now analyse all
possibilities.
Class I.a: n = 1. The size of E is
√ √ √ √
q 2 + q q + q + 1 = v1 (q q + q) + ( q + 1 − v1 )q q + 1.

We obtain v1 = 1.
√
Class I.b: n = q + 1. The size of E is
√ √ √ √ √ √ √
q 2 + q q + q + 1 = v1 (q q + q − q) + ( q + 1 − v1 )(q q − q) + q + 1 =

Hence, v1 = 2.
Class I.c: l = q + 1. The size of E is
√ √ √ √
q 2 + q q + q + 1 = v1 (q q) + ( q + 1 − v1 )(q q − q) + q + 1.

Hence,
√
q 2 + q q = v1 q + q 2 − q
√
and v1 = (1 + q).

Corollary 3.1.20. Let Γ be a non-degenerate linear system of Hermitian surfaces with r2 (Γ) =
2. Assume C and C 0 to be the two degenerate surfaces in Γ. Then, the radicals of C and C 0 are
skew.

Proof. Assume V to be a point in the intersection of the radicals of C and C 0 . Then, V is

conjugate to any point of PG(3, q) with respect to the Hermitian form induced by a linear
combination of C and C 0 . It follows that V is in the radical of any surface of Γ, a contradiction.

Lemma 3.1.21. Let Γ, C and C 0 be as in corollary 3.1.20. Then, all components of the degen-
erate surfaces C and C 0 are secant to all the non-degenerate surfaces of Γ. Furthermore, the
radical of C is contained in a component of C 0 and vice-versa.

69
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.1. DESCRIPTION OF INCIDENCE CONFIGURATIONS IN DIMENSION 3

Class c2 v p1 p2 k
√ √
I-a 1 1 1 q q2 + q q + q + 1
√ √ √
I-b 1 q+1 2 q − 1 q2 + q q + q + 1
√ √
I-c 1 q+1 q+1 0 q2 + q q + q + 1
√
II-c 2 q+1 0 q+1 q2 + 1

Table 3.7: Possible incidence classes for two non-degenerate Hermitian surfaces: Γ contains
degenerate surfaces of rank 2 only.

Proof. Let n be the size of the intersection of the radical of C with a non-degenerate surface
√
H ∈ Γ. Then, n is either 1, q + 1 or q + 1. We now analyse the different possibilities.
Class II.a: n = 1. Then,
√
q 2 + 1 = v1 q + q 2 + q q + 1 > q 2 + 1,

a contradiction.
√
Class II.b: n = q + 1. Then,
√ √ √
q 2 + 1 = v1 q + ( q + 1)(q − 1) q + q + 1

a contradiction.
Class II.c: n = q + 1. Then,

q 2 + q = v1 q + q(q − 1) + q + 1 = v1 + q 2 + 1.

This case is actually possible and yields v1 = 0. It follows that all components of C are secant
to any non-degenerate surface H.

Table 3.7 presents all possible incidence classes E for two Hermitian surfaces H1 , H2 when
the pencil they generate contains degenerate surfaces of rank 2 only.
Here, the second column contains the total number c2 of degenerate surfaces in a given
configuration; v is the number of points in common between the radical of a degenerate surface
and the intersection configuration E.
We recall that a plane π is tangent E if it is tangent to any non-degenerate surface in Γ; we
say that π is secant (or transversal to) E if it intersects a non-degenerate surface of Γ in a non-
singular Hermitian curve. By p1 , the number of planes, components of a degenerate surface in
Γ, that are tangent E is denoted; p2 is the number of planes in such a surface that are secant E.

70
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.1. DESCRIPTION OF INCIDENCE CONFIGURATIONS IN DIMENSION 3

3.1.4 Pencils whose degenerate surfaces have ranks 2 and 3

This subsection describes the configurations E arising from the intersection of two Hermitian
surfaces, when the non-degenerate pencil Γ they generate is assumed to contain both Hermitian
cones and doubly-degenerate Hermitian surfaces; that is to say that both r2 (Γ) 6= 0 and r3 (Γ) 6=
0. We assume that C1 ∈ Γ is a degenerate surface of rank 2 and that C2 ∈ Γ has rank 3. The
symbols in Table 3.8 will be used. Observe that the indices s1 , s2 and s3 are not independent.
√
In fact, since C2 is a cone over a non-degenerate Hermitian curve, s1 + s2 + s3 = q q + 1.

s1 components of C2 included in E
s2 components of C2 secant to (being chords of) E
s3 components of C2 tangent to E
v1 planes of C1 tangent to E
v2 planes of C1 secant to E
l1 intersection between rad C1 and C2
l2 intersection between rad C2 and C1

Table 3.8: Indices for the intersection of Hermitian surfaces of rank 2 and of rank 3.

Lemma 3.1.22. Assume that a non-degenerate linear system Γ contains a surface C1 ∈ Γ of

rank 2 and a surface C2 ∈ Γ of rank 3. Then, the vertex of C2 does not belong to the radical of
C1 . Furthermore, the components of C1 and C2 intersect H as illustrated in Table 3.9, Class I.

Proof. Since C1 6= C2 , the surfaces C1 and C2 generate the linear system Γ. On the other hand,
Γ is non-degenerate by assumption; hence, the vertex of C2 cannot belong to the radical of C1 .
This proves the first part of the lemma.
We distinguish some different cases. First, we consider the information obtained by consid-
ering the intersection between the lines of C2 and E. There are two possibilities:
Class I.a: V2 6∈ E. Then, s1 = 0 and
√ √ √
q 2 + q q + 1 = s2 q + q q + 1.
√
Hence, s2 = q q and the intersection between the polar plane of the vertex of C2 and E ∩ C2
consists of a single point.
Class I.b: V2 ∈ E. Then,
√ √
q 2 + q q + 1 = s1 q + s2 q + 1.
√ √ √ √ √
Hence, s1 q + s2 = q q + q, with s1 + s2 ≤ q q + 1. This implies s2 = q(q + q − s1 ) and
√ √
q + q + 1 > s1 > q. On the other hand, the lines shared between C2 and E must lie in the

71
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.1. DESCRIPTION OF INCIDENCE CONFIGURATIONS IN DIMENSION 3

tangent plane τ at V to any non-degenerate surface H in Γ. Hence, they lie in the intersection
of the 2-dimensional cone C2 ∩ τ with the cone H ∩ τ . Such an intersection contains either 0,
√ √ √ √
1, 2 or q + 1 lines. It follows s1 = q + 1, s2 = q − 1 and s3 = q q − q − q + 1.
Now we focus on the intersection between the plane components of C1 and H. Let l1 be the
cardinality of the intersection between the radical of C1 and E. There are three actual possibili-
ties:
Class I.i: l1 = 1. Then,
√ √ √ √
q 2 + q q + 1 = v1 (q q + q) + ( q + 1 − v1 )(q q) + 1.
√
Hence, v1 = 0 and v2 = q+1, that is all the components of C1 are secant to any non-degenerate
surface H.
√
Class I.ii: l1 = q + 1. Then,
√ √ √ √
q 2 + q q + 1 = q[v1 (q + q − 1) + ( q + 1 − v1 )(q − 1) + 1] + 1.
√
Hence, v1 = 1, v2 = q and one of the components of C1 is tangent to any non-degenerate
surface H.
Class I.iii: l1 = q + 1. Then,
√ √ √ √
q 2 + q q + 1 = q[v1 ( q) + ( q + 1 − v1 )( q − 1) + 1] + 1.
√
Hence, v1 = q and v2 = 1.

Lemma 3.1.23. Let Γ be a linear system of Hermitian surfaces such that r3 (Γ) = 2 and r2 (Γ) =
1. Take C2 and C3 as the Hermitian cones of Γ and assume C1 to be the doubly-degenerate
surface in the linear system. Define r as the radical line of C1 . Then, either the vertices of both
C2 and C3 belong to E, or none of them does.

Proof. Let l1 = |E ∩ r|.

There are two possible classes with respect to the components of one of the cones, say C2 :
Class II.a: V2 6∈ E. Then,
√ √ √
q 2 + q q − q + 1 = s2 ( q + 1) + (q q + 1 − s2 ).
√
Hence, s2 = q(q − 1).
Class II.b: V2 ∈ E. Then,
√ √
q 2 + q q − q + 1 = s1 q + s2 q + 1.
√ √ √
If s1 = 0, then s2 = q q + q − q > q q + 1, impossible. Hence, s1 = 1 and the vertex of C3
√ √
has to belong to E as well. Furthermore, s2 = q q − q.

72
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.2. HERMITIAN MATRICES AND POLYNOMIALS

In fact, the intersection between the components of C1 and E can be realized in only two
ways, since most of the possibilities are ruled out by considerations on the order.
Class II.i: l1 = 1. Then,
√ √ √ √
q 2 + q q − q + 1 = v1 (q q + q) + ( q + 1 − v1 )(q q) + 1;

hence, v1 = −1, a contradiction, since v1 ≥ 0.

√
Class II.ii: l1 = q + 1 Then,
√ √ √ √
q 2 + q q − q + 1 = q[v1 (q + q − 1) + ( q + 1 − v1 )(q − 1) + 1] + 1;

hence, v1 = 0.
Class II.iii: l1 = q + 1 points. Then,
√ √ √ √
q 2 + q q − q + 1 = q[v1 ( q) + ( q + 1 − v1 )( q − 1)] + 1;
√
hence, v1 = q and v2 = 1.

3.1.5 Pencils with r1 (Γ) ≥ 1

√
A Hermitian surface of rank 1 is a plane π repeated q +1 times. The intersection configuration
E associated to a linear system Γ with r1 (Γ) = 1 is hence planar. In fact, a plane might meet a
non-degenerate Hermitian surface H in only two configurations:

(i) Class III-a: π is tangent to H; the intersection E is a Hermitian cone, that is a Hermitian
curve of rank 2;

(ii) Class III-b: π is secant to H; the intersection E is a non-degenerate Hermitian curve.

From Table 3.1, we have that in Class III-a, the rank sequence (r1 , r2 , r3 ) for Γ is (1, 0, 0); in
Class III-b, (r1 , r2 , r3 ) = (1, 0, 1) and there is a cone C in Γ whose vertex does not belong to π.

3.2 Hermitian matrices and polynomials

In this section, we provide some matrices which represent classes of Hermitian intersections.
This is useful in order to construct the possible intersection configurations in dimension 3.
The results are presented in Tables 3.10, 3.11, 3.12.

73
 Class c2 c3 l2 l3 s1 s2 v1 k
√ √
I-a-i 1 1 1 0 0 q q 0 q2 + q q + 1
√ √ √
I-a-ii 1 1 q+1 0 0 q q 1 q2 + q q + 1

degenerate surfaces of ranks 2 and 3.

√ √ √
I-a-iii 1 1 q+1 0 0 q q q q2 + q q + 1
√ √
I-b-i 1 1 1 1 q+1 q−1 0 q2 + q q + 1
√ √ √

74
I-b-ii 1 1 q+1 1 q+1 q−1 1 q2 + q q + 1
√ √ √
I-b-iii 1 1 q+1 1 q+1 q−1 q q2 + q q + 1
√ √ √
II-a-ii 1 2 q+1 0 0 q(q − 1)
1 q2 + q q − q + 1
√ √ √
II-a-iii 1 2 q+1 0 0 q(q − 1) q q2 + q q − q + 1
√ √ √
II-b-ii 1 2 q+1 1+1 1 q(q − 1) 1 q2 + q q − q + 1
√ √ √
II-b-iii 1 2 q+1 1+1 1 q(q − 1) q q2 + q q − q + 1
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.2. HERMITIAN MATRICES AND POLYNOMIALS

Table 3.9: Possible incidence classes for two non-degenerate Hermitian surfaces: Γ contains
 distinct roots.
MH (x) conditions canonical form
 
√
α 0 0 0
2 (η − ξ)2 + 4c q+1 = 0  0 β 0
 √
0  
(x − α)(x − β)(x − γ)
See Lemma 3.2.7  0 0 η c q+1 
0 0 c ξ
 
α 0 0 0
 0 β 0 0 
(x − α)(x − β)(x − γ)  
 0 0 γ 0 
0 0 0 δ
 
√ √
α 0 0 0

75
3 c q+1 + d q+1 = 0  0 β c 0 
(x − α)(x − β)  √ √ 
See Lemma 3.2.8  0 c q+1 β d q+1 
0 0 d β
 √ 
√ λ b q 0 0
(λ − µ)2 − 4b√ q+1 = 0  b µ 0 0 
(x − α)2 (x − β)2 (η − ξ)2 − 4d q+1 = 0  √ 
 0 0 η c q 
See Lemma 3.2.9
0 0 c ξ
 √ 
√
λ b q 0 0
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES

2 (λ − µ)2 − 4b q+1 = 0  b µ 0 0 
 
(x − α) (x − β)
See Lemma 3.2.10  0 0 β 0 
0 0 0 β

√
3.2. HERMITIAN MATRICES AND POLYNOMIALS

Table 3.10: Canonical forms for the 3-dimensional case: MH (x) splits over GF( q); 2 or 3
 roots.
MH (x) conditions canonical form
 
α 0 0 0
 0 β 0 0 
(x − α)(x − β)(x − γ)(x − δ) See Lemma 3.2.7  
 0 0 γ 0 
0 0 0 δ
 
α
√
a b c
See Lemmas
√
3.2.11, 3.2.14
√  a q α
4 q e 

(x − α) σ = (ad)
√
b − (ad)b q  √q √q d
 b
√
d√ α√ f 
T[(ad) q bσ] 6= 0

76
c q e q f q α
 
α
√
a b c
 a q α d e 
(x − α)3 See Lemmas 3.2.11, 3.2.14  √q √q 
 b f 
√
d√ α√
c q e q f q α
 
α
√
a b c
 a q α d e 
(x − α)2 See Lemma 3.2.11  √q √q 
 b f 
√
d√ α√
q q
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES

c e f q α

√
3.2. HERMITIAN MATRICES AND POLYNOMIALS

Table 3.11: Canonical forms for the 3-dimensional case: MH (x) splits over GF( q); 1 or 4
 degree 4.
MH (x) conditions canonical form
 
α 0 0 0
See Lemma 3.2.16  0 β 0 0 
(x − α)(x − β)p2 (x) √  
(ξ − ζ)2 + 4d q+1 ∈6 ¤  0 0 ζ c 
√
0 0 c q ξ
 
α 0 0 0
 0 λ d e 
(x − α)p3 (x)  √ 
 0 d q η f 
√ √
0 e q f q ξ
 
λ
√
c 0 0

77
See Lemma 3.2.17
√  c q µ d
 e 
(x − α)2 p2 (x) (λ + µ)2 + 4c q+1 =√
0  0 0 η
q+1 √
d 
(η − α)(ξ − α) 6= d
0 0 d q ξ
 
λ
√
c 0 c
See Lemma 3.2.18√  c q µ d e 
p2 (x)q2 (x) √  
ξ 2 − (λ + µ)ξ + d q+1 − c q+1 6= 0  0 0 η
√
d 
0 0 d q ξ
 
λ
√
c 0 c
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES

 c q µ d e 
p2 (x)2  
 0 0 η d 
√
0 0 d q ξ

√
3.2. HERMITIAN MATRICES AND POLYNOMIALS

Table 3.12: Canonical forms for the 3-dimensional case: MH (x) does not split over GF( q);
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.2. HERMITIAN MATRICES AND POLYNOMIALS

3.2.1 Some general considerations on Hermitian pencils

This subsection presents some general considerations that will be used later.
The first lemma proves that if we want to consider linear systems Σ in which any two
varieties intersect in the same configuration, then the dimension of Σ is at most 1. In the second
lemma, we verify that if the base locus E of a linear system Γ of Hermitian varieties has maximal
cardinality, then Γ is a maximal set of Hermitian varieties meeting in E. These results hold for
any dimension n.

Lemma 3.2.1. Let Σ be a linear system of Hermitian varieties. If any two distinct elements of
Σ meet in the same configuration, then dim Σ = 1.
√
Proof. The set Σ is a projective subspace of PG(n2 + 2n, q). Take three linearly indepen-
dent Hermitian varieties H1 , H2 , H3 ∈ Σ and let H1 , H2 and H3 be corresponding Hermitian
matrices. Assume also that

E := H1 ∩ H2 = H1 ∩ H3 = H2 ∩ H3 .

The three varieties H1 , H2 , H3 are all different; if it were H3 ⊆ H1 ∪ H2 , then we could write

H3 = (H3 ∩ H1 ) ∪ (H3 ∩ H2 ) = E ∪ E = E = H1 ∩ H2 .

This is a contradiction, since H1 6= H2 and E is not a Hermitian variety. Then, there exists a
t ∈ H3 \ E, belonging neither to H1 nor to H2 . The matrix H 0 given by

tH1 t?
H 0 = H1 − ( )H2
tH2 t?
identifies a Hermitian variety H0 := H(H 0 ) in the linear system Γ, defined by H1 and H2 . Such
variety H0 intersects H3 in t. Again, from the independence of H3 from H1 and H2 ,

H0 6= H3 ,

a contradiction.

Corollary 3.2.2. Let H1 , H2 and H3 be three independent Hermitian matrices such that

H(H1 ) ∩ H(H2 ) = H(H2 ) ∩ H(H3 ) = H(H1 ) ∩ H(H3 ).

√
Then, for any µ in GF( q),

H(H1 ) ∩ H(H2 ) ⊆ H(H1 ) ∩ H(H2 + µH3 ).

78
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.2. HERMITIAN MATRICES AND POLYNOMIALS

Proof. By hypothesis, we have

H(H1 ) ∩ H(H2 ) = H(H1 ) ∩ H(H2 ) ∩ H(H3 ).

On the other hand, a solution of the system of equations induced by the three Hermitian matrices
H1 , H2 and H3 is necessarily a solution of the system determined by H1 and H2 + µH3 . The
result follows.

Lemma 3.2.3. Assume that two Hermitian varieties H1 and H2 intersect in a configuration
E of maximal cardinality. Then, the only Hermitian varieties intersecting H1 in E are those
belonging to the linear system Γ defined by H1 and H2 .

Proof. Assume H3 to be a Hermitian variety intersecting H1 in E. Because of the maximality

of E, it is also H3 ∩H2 = E. Now, Corollary 3.2.2 affirms that any linear combination of H2 and
H3 intersect H1 in a superset of E; since E is maximal, such a superset has to be E again. By
the argument used in the proof of Corollary 3.2.1, this is impossible when H3 does not belong
to Γ, whence the result.

Further linear algebra observations

Now we present some linear algebra results. These results are straightforward generalizations
of those in [Kes81].

Lemma 3.2.4. Let H and J be two Hermitian matrices equivalent up to unitary transforma-
tions. Then, the minimal polynomials MH (x) and MJ (x) are the same.

Proof. By hypothesis, there exists a matrix U such that

H = U JU ?
−1
with U ? = U . It follows that H n = (U )J n (U ? ). Hence, for any polynomial p(x) = a0 +
a1 x + . . . + an xn such that p(J) = 0,

p(H) = a0 I + a1 (U JU ? ) + . . . + U J n U ? = U p(J)U ? = U 0U ? = 0.

Hence, MH (x) | MJ (x). By a similar argument, MJ (x) | MH (x) and the result follows.

Lemma 3.2.5 ([Kes81], Lemma 3). Let H be a Hermitian matrix and suppose that its minimal
polynomial MH (x) has a factorisation

p(x) = p1 (x)p2 (x) . . . pk (x)

√
in k > 1 factors co-prime in GF( q)[x]. Then, the null spaces corresponding to these factors
are mutually conjugate.

79
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.2. HERMITIAN MATRICES AND POLYNOMIALS

Proof. We work by induction on k. Observe that all pi (x) are Hermitian polynomials, since
√
their coefficients lie in the ground field GF( q).
Case k = 2: since p1 (x) and p2 (x) are co-prime, the Chinese Remainder Theorem implies
√
that there exist two Hermitian polynomials r1 (x), r2 (x) ∈ GF( q)[x] such that

1 = r1 (x)p1 (x) + r2 (x)p2 (x);

hence, the identity matrix I can be written as

I = r1 (H)p1 (H) + r2 (H)p2 (H).

Assume now v ? ∈ NullH p1 (x) and w? ∈ NullH p2 (x). Then,

v = v(p1 (H)r1 (H) + p2 (H)r2 (H)) = vp2 (H)r2 (H)

and, likewise,

w? = r1 (H)p1 (H)w? + r2 (H)p2 (H)w? = r1 (H)p1 (H)w? .

Hence, H being Hermitian,

vw? = wv ? = w(p1 (H)? r1 (H)? )v ? = w(r1 (H)p1 (H))v ? = wr1 (H)(p1 (H)v ? ) = w0 = 0.

Induction k ⇒ k + 1: Assume p(x) = p1 (x) . . . pk+1 (x), and for any 0 < i ≤ k define qi (x)
to be pi (x)pi+1 (x). The polynomial qk (x) is co-prime with all the pi (x) for i < k, hence, by the
inductive hypothesis, its null space is conjugate with all of theirs. In particular, both the null
space of pk (x) and that of pk+1 (x) are conjugate with the null space of any other pi (x), since
they are contained in NullH qk (x). By using the same argument on the factorisation of p(x)
given by
p(x) = p1 (x) . . . pk−2 (x)qk−1 (x)pk+1 (x),
the null space of pk+1 is proven to be conjugate with the null space of qk−1 , which contains
NullH pk (x). The result follows.

Lemma 3.2.6. Let H be an n × n Hermitian matrix such that

(i) its minimal polynomial M(x) has degree n;

√
(ii) M(x) splits in GF( q)[x];

(iii) M(x) has no double root.

Then, H is unitarily equivalent to a diagonal matrix.

80
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.2. HERMITIAN MATRICES AND POLYNOMIALS

Proof. Consider the matrix U consisting of the representation of H with respect to the base
formed by its normalized eigenvectors. This matrix is diagonal and, by Lemma 3.2.5, is unitarily
equivalent to H, whence the result.

We now produce some 4 × 4 Hermitian matrices. These matrices can be (and will be) used
to construct the linear systems corresponding to the cases as presented in section 3.1. Cases
√
I-III assume that the factorisation of the minimal polynomial splits over GF( q). Case IV
considers some of the possibilities when the minimal polynomial of a matrix contains higher
degree irreducible factors.

3.2.2 Case I: 3 or 4 distinct roots

Lemma 3.2.7. Let H be a 4 × 4 Hermitian matrix whose minimal polynomial has either the
form MH (x) = (x − α)(x − β)(x − γ)(x − δ) or MH (x) = (x − α)(x − β)(x − γ). Then, H
is unitarily equivalent to a diagonal matrix.

Proof. The first part of the lemma is a direct consequence of Lemma 3.2.6.
Consider now the case in which MH (x) has degree 3 and all its roots are distinct. The
characteristic polynomial CH (x) has degree 4 and is divisible by the minimal polynomial of H.
√
It follows that CH (x) splits in linear factors over GF( q) as well.
Assume the roots of CH (x) not to be the same as those of MH (x). Then,

CH (x) = (x − α)(x − β)(x − γ)(x − δ),

with α, β, γ and δ all distinct. Clearly, all the matrices (H − αI), (H − βI), (H − γI) and
(H − δI) have rank 3; then, MH (H) would have rank 1, a contradiction. It follows that

CH (x) = (x − α)(x − β)(x − γ)2 .

Since α, β and γ are all distinct, the null spaces corresponding to (x − α), (x − β) and (x − γ)
are mutually conjugate. Hence, H is unitarily equivalent to a matrix H 0 of the form
 
α 0 0 0
 0 β 0 0 
H0 =   0 0 σ d ,


0 0 d θ
√
where σ, θ ∈ GF( q) and d ∈ GF(q). By a direct computation, the characteristic polynomial
of H 0 is
√
CH 0 (x) = (x − α)(x − β)(x2 − (σ + θ)x + σθ − d q+1 ).
On the other hand, CH 0 (x) = CH (x); hence, the previous relation implies that
√
x2 − (σ + θ)x + σθ − d q+1
= (x − γ)2 ,

81
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.2. HERMITIAN MATRICES AND POLYNOMIALS

that is, the following system of equations has to be fulfilled:

½
(σ + θ)√= 2γ
σθ − d q+1 = γ 2 .

This yields
√
γ 2 − 2σγ + d q+1
+ σ 2 = 0;
√
the conclusion is that the norm of d is d q+1 = −(γ − σ)2 . On the other hand, the minimal
polynomial MH 0 (x) coincides with the minimal polynomial of H; hence, MH (H 0 ) = 0. The
matrix H 0 is block-diagonal. Assume H 0 = diag(α, β, T ) where T ∈ Mat(2, q), and define
W = MH (T ). Then, we have the following equalities
√ √
q+1 q+1
(i) W1,1 = (γ − σ)[(2γ − σ − α)(2γ − σ − β)+d )] + d (2γ − α − β);
√
q+1
(ii) W1,2 = d[(2γ − σ − α)(2γ − σ − β) + d + (2γ − α − β)(σ − γ)];
√
q
(iii) W2,1 = W1,2 ;
√
q+1
(iv) W2,2 = d (γ − α − β + σ) + (σ − α)(σ − β)(σ − γ).

A substitution and a simplification in W1,1 yield

W1,1 = (γ − σ)(γ − β)(γ − α),

whence γ = σ and, consequently, d = 0. It follows that H 0 is diagonal.

3.2.3 Case II: 2 distinct roots

Lemma 3.2.8. A non-degenerate 4 × 4 Hermitian matrix H with minimal polynomial of the
form MH (x) = (x − α)(x − λ)3 is unitarily equivalent to a matrix H 0 of the form
 
α 0 √ 0 0
 0 λ c q 0 
H0 =  0 c λ d q ,
√ 

0 0 d λ
√ √
q+1 q+1
where c, d ∈ GF(q) \ {0} and c +d = 0.

Proof. The argument develops exactly as in [Kes81], Lemma 7. In fact, it is enough to find four
non-self-conjugate, mutually conjugate vectors r, s, t, u such that

(i) Hr = λr + cs;
√
(ii) Hs = c q r + λs + dt;
√
(iii) Ht = d q s + λt;

82
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.2. HERMITIAN MATRICES AND POLYNOMIALS

(iv) Hu = αu.

The null space corresponding to the polynomial (x − α) has dimension 1 and, by Lemma 3.2.5,
it is conjugate to the space corresponding to (x − λ); hence, there exists a vector u with the
required properties. In order to find r, s, and t it is now enough to consider what happens for
3 × 3 matrices whose minimal polynomial is of the form (x − λ)3 . The result follows directly
from Table 2.3.

Lemma 3.2.9. A non-degenerate 4 × 4 Hermitian matrix H with minimal polynomial of the

form MH (x) = (x − α)2 (x − β)2 is unitarily equivalent to a matrix H 0 of the form
 
η c 0 0
√
 c q θ 0 0 
H0 =   0 0 ζ d ,

√
0 0 d q ξ
with
√
(i) (η + θ)2 + 4c q+1
= 0;
√
(ii) (ζ + ξ)2 + 4d q+1
= 0.

Proof. The two eigenspaces V1 , V2 corresponding to the eigenvalues α and β are mutually
conjugate with respect to H; hence, H admits a block diagonal representation H 0 . Furthermore,
the characteristic polynomial of H 0 can be written as
√ √
CH 0 (x) = (x2 − (η + θ)x + ηθ + c q+1
)(x2 − (ζ + ξ)x + ζξ + dq+1
),

whence we deduce the relation

√
q+1
4ηθ − 4c = (η + θ)2 .

From the latter, condition (i) follows. With a similar argument, it is possible to obtain condition
(ii).
An additional result is that

(η + θ) = 2α; (ζ + ξ) = 2β.

Lemma 3.2.10. A non-degenerate 4 × 4 Hermitian matrix H of the form

 
λ
√
b 0 0
 b q µ 0 0 
 
 0 0 β 0 
0 0 0 β

has minimal polynomial MH (x) = (x − α)2 (x − β) if and only if

83
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.2. HERMITIAN MATRICES AND POLYNOMIALS

√
(i) (λ − µ)2 − 4b q+1
= 0;

(ii) α = (λ + µ)/2.
· ¸
λ b
Proof. Let T = √ . In order for H to have minimal polynomial (x − α)2 (x − β), it is
b q µ
enough to require that the minimal polynomial of T is (x − α)2 . This is to say that:
(i) the discriminant of the equation E(x) = 0, given by
√
q+1
E(x) := (λ − x)(µ − x) − b ,
√
is zero, that is (λ − µ)2 = 4b q+1
;

(ii) E(α) = 0, that is α = (λ + µ)/2.

3.2.4 Case III: 1 root

Lemma 3.2.11. A Hermitian matrix H of the form
 
λ
√
a b c
 a q λ e 
 √q √q d 
 b d√ λ√ f 
√
c q e q f q λ
has characteristic polynomial (x − λ)4 if and only if its coefficients satisfy the following condi-
tions:
√ √ √ √ √ √
q+1 q+1 q+1 q+1 q+1 q+1
(i) a +b +c +d +e +f = 0;
√ √ √ √
(ii) T[(df ) q e + (bf ) q c + (ad) q b + (ae) q c] = 0;

(iii) det(H − λI) = 0, that is,

√ √ √ √ √ √
q+1 q+1 q+1
(af ) + (be) + (cd) = T[(bf ) q ae + (be) q cd + (adf ) q c].

Proof. The characteristic polynomial of H is

CH (x) =
x4 − 4λx3 +
√ √ √ √ √ √
[6λ2 − (a q+1
+b q+1
+c q+1
+d q+1
+e q+1
+f q+1
)]x2 +
√ √ √ √ √ √
[−4λ3 + 2λ(a q+1
+b q+1
+c q+1
+d q+1
+e q+1
+f q+1
)+
√ √ √ √
q q q q
− (T[(df ) e + (bf ) c + (ad) b + (ae) c])]x+
√ √ √ √ √ √
4 q+1 q+1 q+1 q+1 q+1 q+1
λ − ((a +b +c +d +e +f )λ2 +
√ √ √ √
(T[(df ) q e + (bf ) q c + (ad) q b + (ae) q c])λ+
√ √ √ √ √ √
q+1 q+1 q+1
[(af ) + (be) + (cd) − T[(bf ) q ae + (be) q cd + (adf ) q c]].

84
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.2. HERMITIAN MATRICES AND POLYNOMIALS

Conditions (i), (ii) and (iii) follow by direct comparison of the coefficients of this expression
with the expansion
(x − λ)4 = x4 − 4x3 λ + 6x2 λ2 − 4xλ3 + λ4 .

Lemma 3.2.12. An Hermitian matrix H given as in Lemma 3.2.11 has to satisfy the following
further conditions in order to have minimal polynomial (x − λ)2 :
√ √ √
q+1 q+1 q+1
(i) a +b +c = 0;
√ √ √ √ √ √
q+1 q+1 q+1 q+1 q+1 q+1
(ii) a =f ,b =e ,c =d ;
(2) √ √
q q
(iii) n1,2 := bd + ce = 0;
(2) √ (2)
q
(iv) n1,3 := ad + cf = 0, n1,4 = ae + bf = 0;
(2) √ √
(v) n2,3 := a q b + ef q
= 0;
(2) √
(vi) n2,4 := a q c + df = 0;
(2) √ √
(vii) n3,4 := b q c + d q e = 0;

(viii) H − λI 6= 0.

Proof. Let N = (H − λI). Then, N 2 is a Hermitian matrix

 (2) (2) (2) (2)

n1,1√ n1,2 n1,3 n1,4
 (2) q (2) (2) 
 n1,2 n2,2√ n2,3 n2,4 
 √ ,
 (2) q (2) q (2) (2) 
 n1,3 √ n2,3 √ n3,3√ n3,4 
(2) q (2) q (2) q (2)
n1,4 n2,4 n3,4 n4,4

where
(2) √ √ √
q+1 q+1 q+1
(a) n1,1 = a +b +c ,
(2) √ √ √
q+1 q+1 q+1
(b) n2,2 = a +d +e ,
(2) √ √ √
q+1 q+1 q+1
(c) n3,3 = b +d +f ,
(2) √ √ √
q+1 q+1 q+1
(d) n4,4 = c +e +f .

For i 6= j, the conditions (iii)-(vii) are immediate, while the results of Lemma 3.2.11 and
properties (a)-(d) yield conditions (i) and (ii).

85
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.2. HERMITIAN MATRICES AND POLYNOMIALS

Lemma 3.2.13. Let H be a 4 × 4 Hermitian matrix

 
α
√
b c d
 b q β f 
 √q √q e .
 c e√ γ√ g 
√
d q f q g q δ

Then H 2 = 0 if and only if

√ √ √
(i) α2 + b q+1
+c q+1
+d q+1
= 0;
√ √ √
(ii) β 2 + b q+1
+e q+1
+f q+1
= 0;
√ √ √
(iii) γ 2 + c q+1
+e q+1
+g q+1
= 0;
√ √ √
(iv) δ 2 + d q+1
+f q+1
+g q+1
= 0;
√ √
q q
(v) (α + β)b + ce + df = 0;
√
q
(vi) (α + γ)c + be + dg = 0;

(vii) (α + δ)d + f b + gc = 0;
√ √
q q
(viii) (β + γ)e + cb + fg = 0;
√
q
(ix) (β + δ)f + db + eg = 0;
√ √
q q
(x) (γ + δ)h + dc + fe = 0;

(xi) rankH = 2;

(xii) if α + β + γ + δ 6= 0, then cf = ed.

Proof. By direct computation, it is possible to obtain conditions (i)-(x). If c, f , d and e are all
non-zero, then by (vi) and (ix) together,
√ √
q
−cf (α + β + γ + δ) = f be + f dh + cb q d + ceh,

while from (viii) and (v),

√ √
q
−ed(α + β + γ + δ) = ef b + ehc + dcb + df h q .

It follows that either α + β + γ + δ = 0 or cf = ed.

Let Ψ be the linear transformation induced by H. Since Ψ2 = 0, it follows that Im Ψ ⊆
ker Ψ. Furthermore,
Im Ψ ' GF(q)4 /ker Ψ;
it follows that dim ker Ψ = dim Im Ψ = 2, and that the rank of H is 2.

86
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.2. HERMITIAN MATRICES AND POLYNOMIALS

Lemma 3.2.14. In order for a Hermitian matrix H, given as in Lemma 3.2.11, to have minimal
polynomial (x − λ)4 , one of the following coefficients has to be non-zero:
(3) √ √ √ √
(i) n1,2 = f [d q cf q−1
−f q
a + e q b];
(3) √ √ √
q−1
(ii) n1,3 = e[dce − e qb + f q
a];
(3) √ √
q−1
(iii) n1,4 = d[bed − d q c + af ],
√
(3) (3) q √ √ √ √
q q
(iv) n2,1 = n1,2 = f [dc − (f a) + eb q f q−1
];
(3) √ √ √
q−1
(v) n2,3 = c[ebc − c q d + a q f ];
(3) √ √ √
q−1
(vi) n2,4 = b[dcb − b q e + a q f ];
√
(3) (3) q √ √ √ √
(vii) n3,1 = n1,3 = e[f a q e q−1
− (eb) q
+ (dc) q ];
√
(3) (3) q √ √ √
q−1 q
(viii) n3,2 = n2,3 = c[f ac − (cd) + (eb) q ];
(3) √ √ √ √
(ix) n3,4 = a[d q ca q−1
− fa q
+ b q e];
√
(3) (3) q √ √ √ √
(x) n4,1 = n1,4 = d[(af ) q d q−1
− (cd) q
+ (be) q ];
√
(3) (3) q √ √ √ √
q q−1 q
(xi) n4,2 = n2,4 = b[f ab − (be) + (dc) q ];
√
(3) (3) q √ √ √ √
(xii) n4,3 = n3,4 = a[e q ba q−1
− (af ) q
+ dc q ].

If all of the coefficients in (i)-(xii) are zero, H 6= λI and the conditions of Lemma 3.2.12 are not
fulfilled, then MH (x) = (x − λ)3 .

Proof. Let N = H − λI. By considering the conditions in Lemma 3.2.11,

 √ (3) (3) (3)

−T[(df ) q e] n1,2 n1,3 n1,4
 (3) √ (3) (3) 
 n2,1 −T[(bf ) q c] n2,3 n2,4 
N3 =  (3) (3) √ (3) .
 n3,1 n3,2 −T[(ae) q c] n3,4 
(3) (3) (3) √
n4,1 n4,2 n4,3 −T[(ad) q e]

The result follows.

3.2.5 Case IV: some notes when the factorisation contains irreducibles
The symbols pn (x), qn (x) are used to denote distinct polynomials of degree n that do not split
√ √
over GF( q). Observe that any polynomial of degree 2 in GF( q)[x] splits in GF(q)[x].
√
Lemma 3.2.15. Let p4 (x) be an irreducible polynomial of degree 4 over GF( q). Then, p4
factorises into terms of degree at most 2 over GF(q).

87
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.2. HERMITIAN MATRICES AND POLYNOMIALS

Proof. The polynomial p4 (x) splits in GF(q 2 ). There are two possibilities:

(i) p4 (x) has at least a root in GF(q);

(ii) all the roots of p4 (x) are in GF(q 2 ).

√
In case (i), there exist α ∈ GF(q) such that p4 (α) = 0. Since, by hypothesis, α 6∈ GF( q), we
√ √
have α q 6= α and p4 (α q ) = 0. It follows that
√
p4 (x) = (x − α)(x − α q )q2 (x),

whence the result.

Assume now that p4 (x) does not have any root in GF(q), and let β ∈ GF(q 2 ) be one of
its roots. From Galois theory it is known that the orbit of β under the action of Gal(GF(q 2 ) :
√
GF( q)) consists of 4 elements. Hence, the roots of p4 are all distinct and the following
factorisation can be provided:
√ √
q
p4 (x) = (x − β)(x − β )(x − β q )(x − β q q ).

Then, there exist two polynomials r2 and s2 as follows:

r2 (x) = (x − β)(x − β q ) = x2 − xTGF(q2 ):GF(q) [β] + NGF(q2 ):GF(q) [β];

√ √ √ √
q
s2 (x) = (x − β )(x − β q q ) = x2 − xTGF(q2 ):GF(q) [β q
] + NGF(q2 ):GF(q) [β q
].
A straightforward computation shows that p4 (x) = r2 (x)s2 (x) and r2 (x), s2 (x) ∈ GF(q)[x].

Lemma 3.2.16. Let H be a 4 × 4 Hermitian matrix with minimal polynomial of the form
MH (x) = (x − α)(x − β)p2 (x). Then, H is unitarily equivalent to a block diagonal matrix H 0
of the form  
α 0 0 0
 0 β 0 0 
H0 = 
 0 0 ζ d 

√
0 0 d q ξ
where
√ √
(i) (ξ − ζ)2 + 4d q+1
is a non-square in GF( q);
√
(ii) there exists ω ∈ GF(q) \ GF( q) such that ζξ = N[ω] + N[d] and ζ + ξ = T[ω].

Proof. The null spaces corresponding to the polynomials (x−α), (x−β) and p2 (x) are mutually
conjugate; hence, H 0 has a block-diagonal structure. On the other hand, a 2 × 2 matrix
· ¸
ζ√ d
T =
d q ξ

88
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.2. HERMITIAN MATRICES AND POLYNOMIALS

√
has characteristic polynomial CT (x) = x2 −(ζ +ξ)x+(ζξ−d q+1 ). Since CT (x) is a polynomial
√
of degree 2 with coefficients in GF( q), it splits over GF(q). Assume ω to be one of its roots
√
in GF(q); it follows that ω q is a root of CT (x) as well. The condition that CT (x) does not split
√
over GF( q) is equivalent to require that
√ √
(ζ + ξ)2 − 4(ζξ − d q+1
) = (ξ − ζ)2 + 4d q+1

√ √
is a non-square in GF( q), whence (i). On the other hand, if CT (x) is irreducible over GF( q),
√ √
then ω ∈ GF(q) \ GF( q), and ω 6= ω q . This implies the equality

CT (x) = x2 − T[ω]x + N[ω],

√
that is (ζ + ξ) = T[ω] and (ζξ − d q+1 ) = N[ω].
By hypothesis, we assumed that MT (x) has not degree 1; it follows that MT (x) = CT (x),
which is the result.

Lemma 3.2.17. Let H be a 4 × 4 Hermitian matrix with minimal polynomial of the form
MH (x) = (x − α)2 p2 (x). Then, H is unitarily equivalent to a block diagonal matrix H 0
of the form  
λ
√
c 0 0
 c q µ 0 0 
H0 = 
 0 0 ζ d 

√
0 0 d q ξ
with
√
(i) (λ − µ)2 − 4c q+1
= 0;
√ √
(ii) (ξ − ζ)2 + 4d q+1
a non-square in GF( q).

Proof. The block form follows, as before, from the null spaces of (x − α) and p2 (x) being
mutually conjugate. Condition (i) is obtained as in Lemma 3.2.9; condition (ii) is proven as in
Lemma 3.2.16.

Lemma 3.2.18. Let H be a 4 × 4 Hermitian matrix with minimal polynomial of the form
MH (x) = p2 (x)q2 (x), p2 (x) 6= q2 (x). Then, H is unitarily equivalent to a block diagonal
matrix H 0 of the form  
λ
√
c 0 0
 c q µ 0 0 
H0 =   0 0 ζ d 

√
0 0 d q ξ
with
√ √
(i) (λ − µ)2 + 4d q+1
a non-square in GF( q);

89
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.3. CONSTRUCTION OF THE INCIDENCE CONFIGURATIONS

√ √
(ii) (ξ − ζ)2 + 4d q+1
a non-square in GF( q);
√ √
(iii) ξ 2 − (λ + µ)ξ + d q+1
−c q+1
6= 0.

Proof. Since the null spaces of p2 (x) and q2 (x) are mutually orthogonal, assertions (i) and (ii)
follow as in Lemma 3.2.16.
If it were p2 (x) = q2 (x), then
½
ξ+ζ = √
λ+µ √
q+1
ξζ − d = λµ − c q+1 .
Condition (iii) follows.

3.3 Construction of the incidence configurations

In this section we provide actual models which realize all Hermitian pencils which are non-
degenerate among the ones introduced in section 3.1.
Usually, we shall consider linear systems Γ generated by the canonical Hermitian surface
U3 and another non-degenerate surface H = H(H) and assume q to be odd.

3.3.1 Pencils with degenerate surfaces of rank 3 only

Class a: The variety H = H(H) where H is a matrix

 
α 0 0 0
 0 β 0 0 
H=  0 0 γ 0 


0 0 0 δ
√
with α, β, γ, δ distinct elements of GF( q) can be chosen to generate Γ together with U3 . The
minimal polynomial of H is MH (x) = (x − α)(x − β)(x − γ)(x − δ). All surfaces in the linear
system are associated with diagonal matrices.
Classes b and c: We choose a surface H(H) with H of the form
 
α
√
b 0 0
 b q β 0 0 
H=  0 0 γ 0 


0 0 0 δ
√
to generate Γ together· with U3 .¸Since we require that there exists exactly one λ ∈ GF( q) such
α b
that the block T = √ has rank 1, the following equation in λ has to admit only one
b q β
solution:
√
(α − λ)(β − λ) − b q+1 = 0.

90
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.3. CONSTRUCTION OF THE INCIDENCE CONFIGURATIONS

Hence,
√
(α − β)2 + 4b q+1
= 0.
Since T has never rank 0, necessarily b 6= 0. The minimal polynomial for H is MH (x) =
(x − λ)2 (x − γ)(x − δ). The cone corresponding to H − λI has equation
√ √ √ √ √ √ √
q+1 q+1 q
C : (α − λ)X + (β − λ)Y + bXY + b qX q
Y +Z q+1
+T q+1
=0

and the coordinates of its vertex V have to satisfy the following system of equations:
 √ √

 (α − λ)X√ q + bY q
=0
 √ √
(β − λ)Y + b X q = 0
q q


 Z=0

T = 0.
Hence, the homogeneous coordinates for V are of the form
√ √
V = ((λ − α) q , b q , 0, 0).

Finally, since, λ = (α + β)/2, we can rewrite this expression using only α, β and b and
√
V = ((β − α)/2, b q , 0, 0).
√ √
If V ∈ U3 , then ((β −α)/2) q+1 +b q+1 = 0. In this case, Γ belongs to class (c); otherwise,
√ √
when ((β − α)/2) q+1 + b q+1 6= 0, Γ is of class (b).
Class d: The case corresponding to class (d) can be realized by considering the intersec-
tion between the Hermitian cones C1 = H(H1 ) and C2 = H(H2 ) described by the following
matrices:    
0√ a b 0 0 0 0 0√
 a q α c 0   0 γ d a q 
H1 =  b
√ √ ; H = 
 0 d q δ b q ;
√ √ 
q
c q
β 0  2

0 0 0 0 a b 0
with a 6= b, a and b different from 0 and

(α + γ)(β + δ) − N(d + c) 6= 0.

Classes e, f and g: We consider surfaces H(H) where

 
α
√
b 0 0
 b q β 0 0 
H=  0 0 γ c ,

√
0 0 c q δ
√ √
with (α − β)2 + b q+1 = 0 and (γ − δ)2 + c q+1 = 0 to generate Γ together with U3 . The
minimal polynomial of H is of the form MH (x) = (x − λ)2 (x − µ)2 . As before, we may
compute the coordinates of the vertices of the two cones in the pencil. They are
√ √
V1 = ((β − α)/2, b q , 0, 0); V2 = (0, 0, (δ − γ)/2, c q ).

91
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.3. CONSTRUCTION OF THE INCIDENCE CONFIGURATIONS

√ √
Class (e) is realized when neither ((β − α)/2) + b q+1 = 0 nor ((δ − γ)/2) + c q+1 = 0;
class (f) corresponds to the case where only one of these equations is satisfied; we have class
(g) when both equations are satisfied.
Class h: For class (h) we shall follow a more geometrical approach: that is, we will choose
suitable surfaces in PG(3, q) and reconstruct the pencil Γ from them.
A intersection of class (h) can be constructed as follows: let H be a non-degenerate Hermi-
tian surface; take a point V 6∈ H let π be the polar plane of V with respect to H. Take as L the
√
Hermitian curve in π such that (H ∩ π) ∩ L is the cyclic q − q + 1-arc. Then, the pencil gen-
erated by the cone V L and H belongs to class (g). As a matrix H inducing the Hermitian form
associated with P L we may choose a block diagonal one of the form (0, B) where hH(B), U2 i
belongs to Kestenband’s class VII.
Class k: Class (k) has three sub-classes. Like it has been done for class (h), we shall provide
some geometric constructions for the varieties involved. We need a preliminary lemma.

Lemma 3.3.1. Two rank-2 Hermitian curves sharing their singular point might intersect in
√
either 1, 2 or q + 1 lines. In the latter case, they coincide.

Proof. A rank-2 Hermitian curve is constructed as a cone over a Baer subline of any line of
the plane not through its singular point. Two Baer sublines of a line in PG(2, q), unless they
coincide, either have 1 or two points in common. This implies the result.

In all the three cases (k-i), (k-ii) and (k-iii), the pencil Γ is generated by a non-degenerate
surface U3 and a cone C whose vertex V belongs to U3 . Let π be the tangent plane to U3 at V
and let l be a line of π not through V . Both U3 and C intersect l in a degenerate Hermitian curve,
√
that is sets of q + 1 lines through V . Now, Lemma 3.3.1 allows us to identify and construct
the three possible situations.

3.3.2 Pencils with degenerate surfaces of rank 2 only

Class I-a: The intersection E belongs to class I-a if and only if the linear system Γ contains
exactly one Hermitian surface C of rank 2. We may assume such a surface to be associated with
a Hermitian matrix H 0 of the form
 
α
√
b 0 0
 b q β 0 0 
H0 =  0 0 0 0 ,


0 0 0 0
√ √
with (α − β)2 + 4b q+1 a non-square in GF( q). We observe that the minimal polynomial of
H 0 is an irreducible of degree 2. The radical of C is the line [Z = T = 0]. Such a line is tangent

92
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.3. CONSTRUCTION OF THE INCIDENCE CONFIGURATIONS

to the non-degenerate Hermitian surface given by

√ √ √ √ √
U30 : XZ q
+ XT q
+Y q+1
+Z q
X +T q
X.

Class I-b Consider the surface C as introduced in the previous class. Its radical meets the
√
canonical Hermitian surface U3 in q + 1 points. Furthermore, the minimal polynomial of any
non-degenerate surface in the pencil generated by U3 and C is of the form MH (x) = p(x)(x−λ)
√
where p(x) is irreducible over GF( q) of degree 2.
Class I-c We consider the surface C of class I-a together with the non-degenerate surface
√ √ √ √
U300 : XZ q
+ ZX q
+YT q
+ TY q
=0

as generators of Γ. The radical of C is included in U300 .

Class II-c: Class II can be realized by considering as generators for a linear system Γ the
two singular surfaces induced by the matrices
   
α
√
b 0 0 0 0 0 0
 b q β 0 0   0 0 0 0 
H1 =  ; H2 =  
0 0 0 0   0 0 γ c ,
√
0 0 0 0 0 0 c q δ
√ √ √
where we assume (α−β)2 +4b q+1 and (γ−δ)2 +4b q+1 to be non-squares in GF( q). Clearly,
the minimal polynomial of any non degenerate surface in Γ is of the form m(x) = p(x)q(x)
√
with both p(x) and q(x) irreducibles of degree 2 over GF( q).

3.3.3 Pencils whose degenerate surfaces have rank 2 and 3

In this case, we shall consider as generators for the linear system Γ two of the degenerate
surfaces in the pencil.
Class I-a A linear system in class I-a can be generated by a Hermitian cone C of vertex V
and a rank 2 Hermitian surface D of radical l with V 6∈ D.
In class (I-a-i), the radical l of D intersects C in exactly one point. This is to say, since
V 6∈ D, that l is tangent to a non-degenerate plane section of C. Such a configuration can be
realized by taking for C the Hermitian cone of vertex (0, 0, 0, 1) given by the equation
√ √ √
q q q+1
C:X Y +Y X +Z =0

and, as D, the surface induced by a Hermitian matrix H of the form

 
α 0 0 b
 0 0 0 0 
H=  0 0 0 0 ,

√
b q 0 0 δ

93
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.3. CONSTRUCTION OF THE INCIDENCE CONFIGURATIONS

√ √ √
with (α − δ)2 − 4b q+1 a non-square in GF( q) and αδ − b q+1 6= 0. Via a direct computation,
we see that the radical of D has equation l : [X = T = 0]. Such a line intersects C only in the
point (0, 1, 0, 0); hence, D and C satisfy the required conditions.
Class (I-a-ii) is realized in a similar way: this time we need as radical for D a line l which
√
is intersecting C in q + 1 points. Consider as C the Hermitian cone of equation
√ √ √
q q q+1
C : XY −Y X + ωZ = 0,
√
with ω q−1 = −1. A model for D is provided by the Hermitian surface induced by a matrix of
the form  
0 0 0 0
 0 0 0 0 
H=  0 0 γ c ,

√
0 0 c q δ
√ √ √
with (α − δ)2 − 4b q+1 a non-square in GF( q) and γδ − c q+1 6= 0. In fact, the radical of D
√ √
is the line l : [Z = T = 0], which intersects C in points satisfying XY q − Y X q = 0, that is
√
in the points of a subline PG(1, q).
Class (I-a-iii) is possible from the combinatorial point of view; however, it cannot be real-
ized: the only lines completely contained in C are those through the vertex; hence, V should
belong to E, against the non-degeneracy hypothesis.
Class I-b: As for Class (a), we consider the pencil Γ as generated by the cone C and the
rank 2 surface D of radical l; this time we assume V ∈ D.

Lemma 3.3.2. Let Γ be a linear system of Hermitian surfaces and assume that there are at least
two degenerate surfaces C1 and C2 in Γ. Let l1 = rad C1 and l2 = rad C2 . If l1 ∩ l2 6= ∅, then all
surfaces in Γ are degenerate.

Proof. Assume P ∈ l1 ∩ l2 . Then, all partial derivatives of both the equation of C1 and the
equation of C2 are zero in P . From the linearity of the derivative, it follows that all partial
derivatives with respect to the equation of λC1 + µC2 are zero as well.

Lemma 3.3.2 yields that V 6∈ l.

No non-degenerate surface belongs to a pencil of class (I-b-iii). In fact, in (I-b-iii), the
radical l of D intersects the cone C in q + 1 points; this implies that V ∈ l, in contradiction with
Lemma 3.3.2.
In order to realize a pencil in class (I-b-i), we consider the Hermitian cone C of vertex
V = (0, 0, 0, 1) and equation
√ √ √
q q q+1
C:X Y +Y X +Z = 0,

94
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.3. CONSTRUCTION OF THE INCIDENCE CONFIGURATIONS

together with the rank 2 surface D induced by the matrix

 
0 0 0 b
 0 0 0 0 
H=  0 0 0 0 ,

√
b q 0 0 0
√
with b ∈ GF(q) \ GF( q). Clearly, the point (0, 0, 0, 1) belongs to D; furthermore, the radical
of D has equation l : [X = T = 0], and such a line l intersects C in the point (0, 1, 0, 0),
different from V .
Likewise, a pencil Γ generated by the cone C 0 of equation
√ √ √
C 0 : XY q
−Y q
X + ωZ q+1
= 0,
√
q−1
with ω = −1, and the surface D induced by a matrix of the form
 
0 0 0 0
 0 0 0 0 
H=  0 0 0 c ,

√
0 0 c q 0
√
with c ∈ GF(q) \ GF( q), belongs to class (I-b-ii), since the intersection of the radical l with
√
C is the set of points (X, Y, 0, 0) ∈ PG(2, q) with X, Y ∈ GF( q).
Class II: In classes (II-a-iii) and (II-b-iii) the radical l of D is a subset of E; hence, it has
to contain the vertex of any cone in the pencil Γ and, by Lemma 3.3.2, we obtain that all the
surfaces in the linear system are degenerate. We now investigate classes (II-a-ii) and (II-b-ii).
A model for class (II-a-ii) can be realized by considering the linear system Γ generated by
the two Hermitian cones C1 and C2 given by the following matrices H1 and H2 :
   
0 0 0 0 1 0 0 0
 0 β b 0   0 β b 0 
H1 =  √
 0 b q γ 0 
 ; H 2 = 
 0 b q γ c ,
√ 

0 0 0 δ 0 0 0 0
√
with βγ − b q+1 6= 0. Since we assume to be in odd characteristic, we see that the linear
system generated by C1 and C2 is clearly non-degenerate; it contains two rank 3 and one rank 2
surface D, given by D = C1 − C2 . Furthermore, none of the vertices of the cones belongs to the
intersection E. Hence, Γ is of class (II-a-ii).
We generate a member of class (II-b-ii) as follows. Let C1 = H(H1 ) and C2 = H(H2 ) be
the Hermitian cones induced by the matrices:
   
0√ a c 0 0 0 0 0√
 a q α d 0   0 γ v a q 
H1 = 
 c
√ √ ; H = 
 0 v q δ c q ,
√ √ 
q
d q
β 0  2

0 0 0 0 0 a c 0

95
CHAPTER 3. THE 3-DIMENSIONAL CASE: HERMITIAN SURFACES
3.3. CONSTRUCTION OF THE INCIDENCE CONFIGURATIONS

where both H1 and H2 are assumed to be of rank 3. The line [Y = Z = 0], joining the vertices
of C1 and C2 , is contained in E. Since we want the linear system Γ generated by C1 and C2 to
contain one rank 2 surface, the following further conditions, equivalent to require that C1 + C2
has rank 2, have to be imposed on the coefficients:
( √ √
q
(v + d) q =√ ac √q (α + γ)
q
(β + δ) = ac √q (d + v).
√
On the other hand, α, β, γ, δ ∈ GF( q) are fixed by the Frobenius involution; hence, the first
condition can be rewritten as
√ √
q+1 q+1
(β + δ)a = (α + γ)c .

96
Chapter 4

General results

The aim of this chapter is to present a formula to determine the possible sizes of the intersection
of two Hermitian varieties in dimension n. Furthermore, some GAP code to work with Hermi-
tian hypersurfaces is introduced. It is observed that most of the ‘interesting’ cases are out of
reach for such a straightforward code and some further improvements are suggested.

4.1 The cardinality formula

In this section we want to determine the possible size of the intersection of any two Hermitian
hypersurfaces in PG(n, q). In order to prove the result, the variety H of PG(n2 + 2n, q)
representing the singular Hermitian hypersurfaces of PG(n, q) will be studied. This result has
been presented at the international conference “Combinatorics 2000” in Gaeta. I’m thankful
to Prof. A. Cossidente and Dott. A. Siciliano for having first suggested me to investigate the
properties of this remarkable object.

4.1.1 Introduction
In [CS], the geometry of Hermitian matrices of order three over a Galois field is studied; Sec-
√
tions 2 and 3 deal with varieties arising in PG(8, q) from rank 1 and rank 2 Hermitian ma-
trices. More precisely, it is proven that the variety coming from rank 2 matrices is a cubic
hypersurface M37 whose singular points form the variety of all rank 1 Hermitian matrices.
In our work, we shall investigate the variety of the singular Hermitian matrices of order n
and use its properties in order to determine the possible intersection orders for any two Hermi-
tian hypersurfaces.

4.1.2 Preliminaries
Definition 4.1.1. For any matrix M ∈ Mat(n, q), the symbol Mij denotes the element in the
i-th row and j-th column of M ; the symbol M ij denotes the minor of M obtained by deleting

97
CHAPTER 4. GENERAL RESULTS
4.1. THE CARDINALITY FORMULA

the i-th row and the j-th column.

Let L be a list of the form L = {i1 j1 , i2 j2 , . . . , ik jk }. If all it and jt are different, then M L
is the minor of M obtained by removing all the it -th columns and jt -th rows from M ; if there
exist g, h such that ig = ih or jg = jh , then M L is the (n − k) × (n − k) null matrix; likewise
we define Mijhk = 0 if i = h or j = k.

Definition 4.1.2. Assume K, F to be two fields with [F : K] = 2, F = K(²), and let x be the
image of x ∈ F under the involution fixing K. Let
 
X00 X01 + ²Xg g
01 . . . X0n + ²X0n
 X01 + ²Xg X11 . . . X1n + ²Xg 
 01 1n 
M :=  .. .. 
 . . 
g
X0n + ²X0n ... Xnn

be a (n + 1) × (n + 1)-Hermitian matrix over F with respect to the automorphism induced by

the conjugation. The symbol HerK (n + 1, F) denotes the set of all such matrices.

We introduce a mapping ξ defined as follows:

½ 2
HerK (n + 1, F) → Kn +2n+1
ξ g
M → (X00 , X01 , X 01 , X11 , X02 . . . , Xnn ).

Definition 4.1.3. The determinantal polynomial for HerK (n + 1, F) is the polynomial Pn in

g
K[X00 , X01 , X01 , . . . , Xnn ] given by

g
Pn (X00 , X01 , X −1
01 , . . . , Xnn ) := det(ξ (M )).

Observe that the degree of the polynomial Pn is n+1. In the following theorem we compute
fij .
partial derivatives of Pn with respect to Xij and X

Theorem 4.1.4. With the notations of Definition 4.1.3,

(i)
∂P
= det(M ii );
∂Xii
(ii) for i 6= j,
∂P
= (−1)i+j (det(M ij ) + det(M ji ));
∂Xij

(iii) for i 6= j,
∂P
= (−1)i+j [² det(M ij ) + ²det(M ji )].
f
∂ Xij

98
CHAPTER 4. GENERAL RESULTS
4.1. THE CARDINALITY FORMULA

Proof. For i = j, consider the expansion of P as

n
X
P = (−1)i+t Mit det(M it )
t=0

and observe that for any i, h, k with (h, k) 6= (i, i):

g
∂Xhk
= 0.
∂Xii
Then, the formal derivative with respect to Xii is

∂P
=
∂Xii
Xn · ¸
i+t ∂Mit it ∂ det(M it )
(−1) det(M ) + Mit =
t=0
∂Xii ∂Xii
Xn fit
∂Xit ∂X ∂ det(M it )
(−1)i+t [ det(M it ) + ² det(M it ) + Mit ]=
t=0
∂Xii ∂Xii ∂Xii
Xn
∂ det(M it )
(−1)2i det(M ii ) + 0 + (−1)i+t Mit .
t=0
∂X ii

On the other hand, M it does not contain the term Xii ; hence,
n
X ∂ det(M it )
(−1)i+t Mit =0
t=0
∂Xii

and (i) follows.

Now assume i 6= j. The determinantal polynomial P can be written as
n
X
P = fit ) det(M it ).
(−1)i+t (Xit + ²X
t=0

Then, the partial derivative with respect to Xij is

X n
∂P ∂ det(M it )
= (−1)i+j det(M ij ) + (−1)i+t Mit .
∂Xij t=0
∂Xij

On the other hand, the term Xij appears only in the expression of Mij and of Mji . Hence,
expanding the determinant by the j-th row and the i-th column, we obtain
X n
∂P ∂ det(M jt )
= (−1)i+j det(M ji ) + (−1)i+t Mjt .
∂Xij t=0
∂Xij

We claim that n
X ∂ det(M jt )
j+t
(−1) Mjt = (−1)i+j det(M ij ).
t=0
∂Xij

99
CHAPTER 4. GENERAL RESULTS
4.1. THE CARDINALITY FORMULA

As a matter of fact,

Xn
∂ det(M jt )
(−1)j+t Mjt =
t=0
∂Xij
Xn P
j+t ∂ nk=0 (−1)i+k Mik
jt
det(M jt,ik )
(−1) Mjt =
t=0
∂Xij
Xn X n jt ij,ik
∂Mik jt ∂ det(M )
(−1)(i+j)+(k+t) Mjt [ det(M jt,ik ) + Mik ]=
t=0 k=0
∂X ij ∂X ij
n
X
i+j
(−1) [ (−1)j+t Mjt det(M jt,ij )]+
t=0
n X
X n
jt ∂ det(M jt,ik )
(−1)i+j [ Mjt Mik ] = (−1)i+j det(M ij ) + 0,
t=0 k=0
∂Xij

since M jt,ik does not contain any term in Xij .

Given that à !
∂Mij ∂Mji
= = ²,
f
∂ Xij fij
∂X
fij as
it is possible to write the derivative of P with respect to X

X n
∂P ∂ det(M it )
= (−1)i+j [² det(M ij ) + (−1)i+t Mit ].
∂Xfij ∂Xfij
t=0

In order to prove part (iii), the main claim is

Xn
∂ det(M it )
(−1)i+t Mit = ²det(M ji ).
f
∂ Xij
t=0

In fact, we have

Xn
∂ det(M it )
(−1)i+t Mit =
∂ fij
X
t=0
Xn X
n it
i+j
∂Mjk
(−1) [ Mit det(M jk,it )] =
fij
∂X
t=0 k=0
X n
i+j
∂Mjiit
(−1) [ Mjt det(M jt,ij )] = (−1)i+j ² det(M ji ),
f
∂ Xij
t=0

whence the theorem follows.

100
CHAPTER 4. GENERAL RESULTS
4.1. THE CARDINALITY FORMULA

4.1.3 The determinantal variety H

From now on, we shall assume to be working over a finite field GF(q), with q square, and write
Her(n + 1, q) for HerGF(√q) (n + 1, GF(q)). Observe that the polynomial Pn defines an algebraic
√
hypersurface Hn,q = F(Pn ) in PG(n2 + 2n, q) that we may call the determinantal variety of
Her(n + 1, q). The name is justified by the fact that a point x is in Hn,q if and only if ξ −1 (x) is
a singular Hermitian hypersurface in PG(n, q) or, equivalently, the pre-images under ξ of x in
Her(n + 1, q) are singular Hermitian matrices.
We need some further notation.
√
Definition 4.1.5. Let Hn,q be the hypersurface in PG(n2 + 2n, q) of all singular Hermitian
hypersurfaces of PG(n, q). Denote by

(i) the symbol Or,n,q := Or,n the set of points of Hn,q corresponding to Hermitian matrices
of rank r;
i
(ii) the symbol On,q := Oni the set of points of Hn,q with multiplicity i;

(iii) the symbol O≤r,n,q := O≤r,n the set of points of Hn,q corresponding to matrices of rank
at most r;
≥i
(iv) the symbol On,q := On≥i the set of points of Hn,q of multiplicity at least i;
0 √
(v) the symbol On,q the points of PG(n, q) not in Hn,q .

Theorem 4.1.6. The singular points of Hn,q are in 1–1 correspondence with Hermitian vari-
eties of PG(n, q) having rank at most n − 1, that is

≥2
On,q = O≤(n−1),n,q .

Proof. If H(H) is a Hermitian variety in PG(n, q) of rank at most n−1, then all its n×n minors
have determinant zero. Theorem 4.1.4 guarantees that all partial derivatives of the function
det(M ) can be expressed as combinations of suitable determinants of n × n minors of M . It
follows that all partial derivatives of det(M ) are 0 at the point ξ(H) of Hn,q corresponding to
H; hence, the point is singular.
Consider now a singular point T on Hn,q , and let H = ξ −1 (T ) be a Hermitian matrix in
Her(n + 1, q) which is the pre-image of T . Let H ij be any n × n minor of H. Then,

(i) if i = j,
∂P
det(H ii ) = (H) = 0;
∂Xii

101
CHAPTER 4. GENERAL RESULTS
4.1. THE CARDINALITY FORMULA

(ii) if i 6= j, the linear system

( ∂P
det(H ij ) + det(H ji ) = ∂X ij
=0
ij ji ∂P
² det(H ) + ² det(H ) = ∂ X g = 0
ij

consists of two independent equations in det(H ij ) and det(H ji ); hence,

det(H ij ) = det(H ji ) = 0.

It follows that all n × n-minors have determinant 0; hence, the rank of H is at most n − 1.
√
Lemma 4.1.7. Let H be a singular Hermitian matrix of rank r. If r < q, then the image of H
on Hn,q is at most a (n − r + 1)-ple point.

Proof. If r = n, the result follows directly from Theorem 4.1.4.

Assume now r < n. Then, there exists a linear transformation ϑ such that H is equivalent
to a matrix H 0 of the form
" #
0 (n−r+1)×(n−r+1)
H 0 := .
Ir×r
√
Let σ be a primitive element of GF( q)? and define T as the matrix
 
I(n−r+1)×(n−r+1)
 
 σ 
T :=  .
 ... 
r
σ

The linear system generated by T and H 0 contains, by construction, exactly one matrix of rank
√ √
r, r matrices of rank n and q + 1 − r − 1 = q − r non-singular matrices. It follows that
√
the line T H 0 of PG(n2 + 2n, q) is not totally contained in Hn,q . The degree of Hn,q is n + 1;
hence, Bezout’s theorem guarantees that T H 0 intersects Hn,q in at most n + 1 points, counted
with the proper multiplicities. Since the matrices of rank n correspond to non-singular points
of Hn,q , it follows that
mult(H) + r ≤ n + 1,
that is mult(H) ≤ (n − r + 1).
√
Corollary 4.1.8. If n ≤ q, then all matrices of rank n − 1 correspond to double points of
Hn,q and vice versa. That is,
2
On−1,n,q = On,q .

Proof. From Lemma 4.1.12, matrices of rank n − 1 correspond to points of Hn,q of multiplicity
at most 2. On the other hand, all matrices of rank at most n − 1 do correspond to singular points
of Hn,q ; hence, their multiplicity as points is at least 2. The result follows.

102
CHAPTER 4. GENERAL RESULTS
4.1. THE CARDINALITY FORMULA

Lemma 4.1.9. Let N be a generic matrix in Mat(n + 1, q) define D as the polynomial

D(N00 , N01 , N10 , . . . , Nnn ) := det(N ),

and assume
Dn,q := F(D).
√
Then, Hn,q is a section of Dn,q via a PG(n2 + 2n, q); in fact, there exists a linear mapping φ
√
such that Hn,q is the image under φ of the GF( q)-rational points of Dn,q .

g
Proof. Consider the determinant function Pn (X00 , X01 , X 01 , . . .) for a matrix X written as in
√
Theorem 4.1.4. Assume that the variables do vary in GF(q), rather than in GF( q); this is
equivalent to consider the GF(q)-rational points of the variety Hn,q . In this case, Pn can be
seen as the determinant function of a matrix of the same structure as X. The two equations
(
X01 + ²Xg 01 = α
√
qg
X01 + ² X01 = β

are independent over GF(q) and always admit a solution in X01 , X g 01 . Hence, there exists a
matrix N such that
g
P (X00 , X01 , X 01 . . .) = D(N00 , N01 , N10 , . . .).

On the other hand, given any matrix N ∈ Mat(n + 1, q), consider the following transfor-
mation φ, where i > j:
Nii → Xii
Nij − ²N
√
ji → Xij
Nji − ² q Nij → Xji .
The mapping φ is linear (hence rational) and transforms the polynomial D(N ) into P (X).
√ √
It follows that the GF( q)-rational points of Dn,q are transformed by φ into the GF( q)-
rational points of Hn,q and the GF(q)-rational points of Hn,q are mapped into the GF(q)-
rational points of Dn,q . The result follows.

The result of Lemma 4.1.9 can be summarized in the following diagram.

Mat(n O + 1, q) / Dn,q ⊆ PG(n2 + 2n, q)

O
⊆ φ

/ Hn,q √
Her(n + 1, q) ⊆ PG(n2 + 2n, q)
√
Corollary 4.1.10. For any i ≥ 2, the GF( q i )-rational points of Hn,q do correspond to the
√
GF( q i )-rational points of Dn,q .

Corollary 4.1.11. For any l > 0, the GF(q 2l )-rational points of Hn,q and Hn,ql are the same.

103
CHAPTER 4. GENERAL RESULTS
4.1. THE CARDINALITY FORMULA

We are now in position to generalise Lemma 4.1.7 to matrices of any rank r, lifting the
restriction r < q.

Theorem 4.1.12. Let H be a singular Hermitian matrix of rank r. Then the image of H on
Hn,q is at most a (n − r + 1)-ple point; that is,

Or,q ⊆ Oq≤n−r+1 .
√
Proof. If r ≤ q − 1, this is Lemma 4.1.7.
√ √
Otherwise, let k be an integer such that q k > r, and consider the GF( q k )-rational points
√ √
of Hn,q . Every GF( q)-rational point H of Hn,q is a GF( q k ) rational point. Furthermore,
the multiplicity of any point H is preserved under algebraic extensions. Let θ be a primitive
√ √
element of GF( q k )? . Since q k > r, all the powers θ, θ2 , . . . , θr are distinct. Let P be the
√
image via ξ 0 : Mat(n + 1, q k ) → PG(n2 + 2n, q k ) of the matrix
 
I(n−r+1)×(n−r+1)
 
 θ 
T :=  ... .
 
θr
√
The line HP intersects Hn,q in H and in at least r more of its GF( q k ) rational points; fur-
thermore, HP is not completely included in Hn,q ; hence the result follows, as before, from
Bezout’s theorem.

Observe that the matrix T constructed in the proof of Theorem 4.1.12 is not Hermitian.

Lemma 4.1.13. Let K be a field, and assume L, M ∈ Mat(n + 1, K); then, for all a, b ∈ K,

rank(aL + bM ) ≤ rank(L) + rank(M ).

Proof. Let θ and τ be the linear mappings induced by L and M . Since for all x ∈ V = Kn

(aλ + bµ)(x) = aλ(x) + bµ(x),

the image of (aθ + bτ ) is included in the direct sum of the images of the spaces θ(V ) and τ (V ).
The result follows from

rank(L) = dim Im θ; rank(M ) = dim Im τ.

Corollary 4.1.14. Let H1 , H2 be two Hermitian matrices in Mat(n + 1, q) of rank at most n/2.
Then, the line ξ(H1 )ξ(H2 ) determined by the images of H1 and H2 is completely included in
Hn,q .

104
CHAPTER 4. GENERAL RESULTS
4.1. THE CARDINALITY FORMULA

√
Proof. From Lemma 4.1.13, for any λ, µ ∈ GF( q),

rank(λH1 + µH2 ) ≤ rank(H1 ) + rank(H2 ) ≤ n < n + 1.

It follows that all matrices in the linear system Γ generated by H1 and H2 are singular.

Corollary 4.1.15. For any integer i < n/2, the envelope of secants to the variety O≤i,n is a
subvariety of O≤2i,n .

Lemma 4.1.16. Let M be an (n + 1) × (n + 1) matrix of the form

" #
0(n−r+1)×(n−r+1)
M := .
Ir×r

And let H be a generic (n + 1) × (n + 1) matrix. Then, µn−r+1 | det(λM + µH).

Proof. The proof is by induction on n − r.

If n − r = 0, then the first row of M can be assumed to be the zero vector. It follows that

det(λM + µH) =
n
X
(λMi0 + µHi0 ) det(λM i0 + µH i0 ) =
i=0
n
X n
X
i0 i0
µHi0 det(λM + µH ) = µ Hi0 det(λM i0 + µH i0 );
i=0 i=0

hence, µ | det(λM + µH).

Assume now that the lemma is true for n−r = k; it is claimed that it holds for n−r = k +1.
Observe that, if M is of the form in the hypothesis with k = n − r > 1, then M i0 is a matrix
satisfying the same hypothesis with difference between order and rank k 0 = n − r − 1. By
inductive reasoning, it follows that

µn−r | det(λM i0 + µH i0 );

hence, by the same argument as in the first part of the proof,

µn−r+1 | det(λM + µH)

and we get the result.

Corollary 4.1.17. The following inclusion is true:

n−r+1
Or,n,q ⊆ On,q

105
CHAPTER 4. GENERAL RESULTS
4.1. THE CARDINALITY FORMULA

Proof. The points of Or,n,q have at least multiplicity n − r + 1, whence

≥n−r+1
Or,n,q ⊆ On,q .

From Theorem 4.1.12,

≤n−r+1
Or,n,q ⊆ On,q
and, by definition of O,
≤n−r+1 ≥n−r+1 n−r+1
On,q ∩ On,q = On,q .

The main result of this subsection is the following theorem.

Theorem 4.1.18. Let M be a Hermitian matrix of rank r ≤ n. Then, the multiplicity of the
point ξ(M ) of Hn,q corresponding to M is exactly n − r + 1; furthermore,

n−r+1
Or,n,q = On,q .

Proof. From Corollary 4.1.17, we have that matrices of rank r correspond to points on H of
multiplicity n − r + 1.
As a matter of fact, we know that
n
X n
X
m
|Or,n,q | = |On,q | = |Hn,q |.
r=1 m=1

Thanks to Corollary 4.1.17,

n−r+1
|On,q | ≥ |Or,n,q |;
hence, all cardinalities have to be the same and, since the sets are all finite, this yields the
equality.

4.1.4 Action of PGL(n + 1, q) on H

The action of any linear collineation σ ∈ PGL(n + 1, q) can be lifted to the space PG(n2 +
√ √
2n, q). For any P ∈ PG(n2 + 2n, q), we define

τσ (P ) := ξ(σ(ξ −1 (P ))).
√
This induces an action of PGL(n + 1, q) on PG(n2 + 2n, q). In fact, the following corollary
is true.

Corollary 4.1.19. For any i such that 0 < i < n, the τ -action of the group PGL(n + 1, q) on
√
PG(n2 + 2n, q) is transitive on On,q
i
.

106
CHAPTER 4. GENERAL RESULTS
4.1. THE CARDINALITY FORMULA

Proof. All Hermitian hypersurfaces of given rank n − i + 1 lie in the same orbit under the action
of PGL(n + 1, q). It follows that PGL(n + 1, q) is transitive on
ξ −1 (On−i+1,n,q ) = ξ −1 (On,q
i
),
whence the result.
Lemma 4.1.20. Let Γ := {λH +µT } be a linear pencil of (n+1)×(n+1) Hermitian varieties,
and let ri (Γ) be the number of Hermitian varieties of rank i in Γ. Then, if rn+1 (Γ) 6= 0,
X
(n − i + 1)ri (Γ) ≤ n + 1.
Proof. If rn+1 (Γ) 6= 0, then the line HT is not completely included in Hn,q . By Bezout’s
theorem, this implies that HT intersects Hn,q in at most n + 1 points, counted with the proper
multiplicities. The result follows from Theorem 4.1.18.
Lemma 4.1.21. For r ≤ n, the variety Hr,q can be embedded in O≤r,n . Furthermore, this
embedding η can be realized in such a way as to have
√
Hr,q = η −1 (O≤r,n,q ∩ PG(r2 + 2r, q)).
Proof. Let M 0 be an r × r Hermitian matrix, and define M to be the (n + 1) × (n + 1) matrix
" 0 #
M
η(M 0 ) := M = .
0n+1−r
√
Then, M ∈ O≤r,n,q ∩ PG(r2 + 2r, q). Observe that, from Corollary 4.1.19, any two embed-
dings of Hr,q in O≤r,n,q are equivalent.
√ √
Corollary 4.1.22. For any r < n, the canonical subspace PG(r2 + 2r, q) of PG(n2 + 2n, q)
≥n−r+1
intersects Hn,q in a subset of its set of singular points On,q . Furthermore, such intersection
√
has the structure of an algebraic variety of degree r + 1 in PG(r2 + 2r, q).
≥n−r+1
Proof. The embedding η of Hr,q into On,q is a bijection when restricted to Hn,q ∩PG(n2 +
√
2n, q). This proves the result.
Theorem 4.1.23. With the same conditions as in Corollary 4.1.20, assume that there exist an
integer v such that rv+1 6= 0, whereas for any t > v, rt+1 = 0. Then,
X
(v − i + 1)ri (Γ) ≤ v + 1.
Proof. If v = n, then this is Corollary 4.1.20.
n,q
Otherwise, the line HT is completely included in O≤v+1 . Modulus a linear transformation
it is possible to assume that both H and T are of the form
" # " 0 #
Iv 0 T 0
H := T := .
0 0n+1−v 0 0n+1−v
Hence, both of them can be considered as points of Hv,q , with H a non-singular point there. It
is now possible to apply Lemma 4.1.20 on this variety and the the result follows.

107
CHAPTER 4. GENERAL RESULTS
4.1. THE CARDINALITY FORMULA

4.1.5 The order formula

The following lemma is from [Kes81].

Lemma 4.1.24 ([Kes81], Lemma 1). Let H1 and H2 be two (n + 1) × (n + 1) Hermitian

matrices, and let H1 , H2 be the corresponding Hermitian varieties. Then,
√
(i) any two varieties in the GF( q)-linear system Γ generated by H1 and H2 meet in the
same configuration;

(ii) for any p in PG(n, q), there exists H ∈ Γ such that p ∈ H.

Proof. Assume A = r1 H1 + r2 H2 , B = s1 H1 + s2 H2 . Then, since (r1 , r2 ) 6= λ(s1 , s2 ), the

linear system ½
xAxt = 0
xBxt = 0
is equivalent to ½
xH1 xt = 0
xH2 xt = 0,
whence H(A) ∩ H(B) = H1 ∩ H2 and the first part of the lemma is done.
Assume m1 = pH1 p? and m2 = pH2 p? . Since both H1 and H2 are Hermitian, m1 , m2 ∈
√
GF( q). It follows that H(H3 ) = H(m2 H1 − m1 H2 ) ∈ Γ and

pH3 p? = m2 pH1 p? − m1 pH2 p? = m2 m1 − m1 m2 = 0,

which is the expected result.

√
Definition 4.1.25. Let Γ be a GF( q)-linear system of Hermitian hypersurfaces in PG(n, q).
The rank sequence of Γ is the list (r1 (Γ), . . . , rn+1 (Γ)) where ri is the number of varieties of
rank i in Γ.

Using Theorem 4.1.23, it is possible to describe all the allowed rank sequences for Hermitian
hypersurfaces in PG(n, q). Thanks to the following generalization of Lemma 2, [Kes81], we
are able to settle the problem of determining the list of all possible orders of intersection.

Theorem 4.1.26. The size of the intersection of any two non-degenerate Hermitian varieties
H1 , H2 depends only on the rank sequence of the linear system Γ generated by them. This size
is the number η(Γ, q), given by

X n
1
η(Γ, q) := √ {(1 − q n+1 ) + ri (Γ)[(qµ(i − 1, q) + 1)(q n+1−i − 1)]}
q(q − 1) i=1
rn+1 (Γ)
+ √ µ(n, q).
q

108
CHAPTER 4. GENERAL RESULTS
4.1. THE CARDINALITY FORMULA

Proof. By Lemma 4.1.24 any two distinct elements of Γ intersect in the same configuration, say
E. Let k be the cardinality of this set.
Any point of PG(n, q) \ E belongs to exactly one hypersurface in H(Γ), while the points in
√
E belong to all q + 1 of them. Hence,
à !
X √
| PG(n, q)| = |X | − q|E|.
X ∈Γ

From Theorems 1.7.28 and 1.7.30 it follows that

n
1 X
ri (H, q)[(qµ(i − 1, q) + 1)(q n+1−i − 1)]
q − 1 i=1
√ q n+1 − 1
+rn+1 (H, q)µ(n, q) − qk = ,
q−1
whence the result.

As an example of Theorem 4.1.26 we determine the cardinality of all possible intersections

of non-degenerate Hermitian varieties up to dimension n = 4. For n = 2, the results might be
found in Chapter 2; for n = 3, they are in Chapter 3. For n = 4, first, we count the number of
points of a non-degenerate Hermitian hypersurface: this is
√
µ(4, q) = (q + 1)(q 2 q + 1).

Then, we tabulate the results by applying the formula of 4.1.26 with all possible values of the
rank sequence. The results are in Table 4.1. Observe that when r1 = 1, there are only two
possible intersection configurations – namely the one obtained when the rank 1 hypersurface C
in Γ is secant to any other variety in the pencil and the one obtained when C is tangent to any
other hypersurface. This is a general property. In fact, we prove the following corollary.

Corollary 4.1.27. Let Γ be a non-degenerate linear system of Hermitian hypersurfaces in the

projective space PG(n, q). Assume that r1 (Γ) = 1 and let C be such a rank 1 hypersurface.
Then, one of the following holds

(i) Γ contains only one the singular hypersurface C and C is tangent to all the non-degenerate
hypersurfaces of Γ;

(ii) Γ contains also a hypersurface K of rank n and the base locus of Γ is a non-degenerate
Hermitian hypersurface Un−1 in the hyperplane C.

Proof. From Theorem 4.1.23, we have that the rank sequence for the linear system Γ is either
(1, 0, . . . , 0, 0) or (1, 0, . . . , 0, 1). Assume the latter; then, a straightforward computation shows
that |E| = µ(n−1, q). The hyperplane C is either tangent to any non-degenerate hypersurface H

109
CHAPTER 4. GENERAL RESULTS
4.1. THE CARDINALITY FORMULA

r1 r2 r3 r4 k = |E|
√
0 0 0 0 q3 + q2 − q q + q + 1
0 0 0 1 q3 + q2 + q + 1
√
0 0 0 2 q3 + q2 + q q + q + 1
√
0 0 0 3 q 3 + q 2 + 2q q + q + 1
√
0 0 0 4 q 3 + q 2 + 3q q + q + 1
√
0 0 0 5 q 3 + q 2 + 4q q + q + 1
√
0 0 1 0 q3 + q2 + q + q + 1
√ √
0 0 1 1 q3 + q2 + q q + q + q + 1
√ √
0 0 1 2 q 3 + q 2 + 2q q + q + q + 1
√ √
0 0 1 3 q 3 + q 2 + 3q q + q + q + 1
√ √
0 0 2 0 q3 + q2 + q q + q + 2 q + 1
√ √
0 0 2 1 q 3 + q 2 + 2q q + q + 2 q + 1
√ √ √
0 1 0 0 q3 + q2 q + q2 + q q + q + q + 1
√ √ √
0 1 0 1 q 3 + q 2 q + q 2 + 2q q + q + q + 1
√ √ √
0 1 0 2 q 3 + q 2 q + q 2 + 3q q + q + q + 1
√ √ √
0 1 1 0 q 3 + q 2 q + q 2 + 2q q + 2q + 2 q + 1
√
1 0 0 0 q2 q + q + 1
√ √
1 0 0 1 q2 q + q q + q + 1

Table 4.1: Intersection numbers for non-degenerate Hermitian varieties in dimension 4.

in Γ or it cuts all of them transversally. If C is transversal to all non-degenerate hypersurfaces,

then its intersection with H is again a non-degenerate Hermitian surface in PG(n − 1, q). Such
a surface has size µ(n − 1, q) and the rank sequence for Γ has to be (1, 0, . . . , 0, 1). When the
rank sequence is (1, 0, . . . , 0), the other possibility for C happens – that is C is tangent to all
other varieties in Γ.

4.1.6 Some further remarks on H

≤t
For n ≤ 5, the dimension of the subvarieties On,q have been computed. The results are presented
in table 4.2. If n = 2, the result is in [CS]; for n > 2 a direct computation with the program
Macaulay2 [GS00] has been performed. For n > 5, computations have been aborted after
having been running for three days on a quad-Xeon 550 server.
The code that has been used is the following. It is worth to remark that defining the base
field to be 9, in this case, does not hamper generality, since all the manipulations performed are
formal reductions of the polynomials which are independent from the characteristic c > 0 of
the field.
Size.m2
KK=GF(9)
i=KK_0
ib=iˆ3

110
CHAPTER 4. GENERAL RESULTS
4.1. THE CARDINALITY FORMULA

n Hn,q O ≥2 O ≥3 O ≥4 O ≥5
2 7 4
3 14 11 7
4 23 20 15 8
5 34 31(?) ? 19 10

Table 4.2: Dimension of the varieties O ≥t for small n

R4=KK[ X00,
X01, X01t, X11,
X02, X02t, X12,
X12t, X22,
X03, X03t, X13,
X13t, X23,
X23t, X33,
X04, X04t, X14,
X14t, X24,
X24t, X34,
X34t, X44]
M4=matrix(
{
{X00, X01+i*X01t, X02+i*X02t,
X03+i*X03t, X04+i*X04t},
{X01+ib*X01t, X11, X12+i*X12t,
X13+i*X13t, X14+i*X14t},
{X02+ib*X02t, X12+ib*X12t,
X22, X23+i*X23t, X24+i*X24t},
{X03+ib*X03t, X13+ib*X13t,
X23+ib*X23t, X33, X34+i*X34t},
{X04+ib*X04t, X14+ib*X14t,
X24+ib*X24t,i X34+ib*X34t,X44}
}
HVar4= ideal (det M4)
S4Points2 = minors(M4,4)
S4Points3 = minors(M4,3)
S4Points4 = minors(M4,2)

dim S4Points2
dim S4Points3
dim S4Points4

≤2
We conjecture that dim Hn,q − dim On,q = 3 for all n and q.
1
The variety On,q has been extensively studied. In fact, it constitutes a cap in PG(n2 +

111
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

√
2n, q), as is proven in the following lemmas.

Lemma 4.1.28 (Wan). Let H be a Hermitian matrix of rank 2. Then, there exist two Hermitian
matrices M1 , M2 of rank 1 such that

H = M1 + M2 .

Lemma 4.1.29 (Lunardon,[Lun99]; Cossidente, Siciliano [CS]). Take Σ as the set of all Her-
√
mitian matrices of Mat(n + 1, q) of rank 1. Its image in PG(n2 + 2n, q) is a cap V(n+1),2 , the
Hermitian Veronesian.

In the setting of [CS], we may enunciate the following corollary.

Corollary 4.1.30. The envelope of secants to the Hermitian Veronesian V(n+1),2 is the algebraic
variety whose points correspond to all Hermitian matrices of rank at most 2.

4.2 Explicit computations and algorithms

In this section we present some simple GAP ([GAP99]) code for the computation of the orbits
of the projective unitary group PU(n + 1, q) in its action on PG(n, q).

4.2.1 The computer code: general remarks

We used the package [GS99] to model small projective spaces. In fact, quite a little part of the
functionality which is provided by that code has been used.
√
In this section q is, as usual, assumed to be a square, and p is defined as q. Note that p is
not necessarily a prime.
The program itself has been split into several files. This has been done for two different
reasons:

(1) ease the development of the code;

(2) provide a simple way to insert checkpoints in the program – in fact the output of the pre-
vious stages of a computation can be dumped to disk and reloaded later in order to resume
execution with some slightly changed parameters.

Point (2) has been our main concern: GAP4 does provide a facility to save and restore its
workspace; however, the command SaveWorkspace cannot be used inside a function or a
loop.
Our code is composed of the following units: Main, Param, Projective, Helpers,
Prelim, Unitary, Orbits and Post-comp.

112
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

4.2.2 Initialization:
main and param
The file main.gap contains the core of the program and it is the one that calls all modules in
the proper order.
main.gap

#####
# Computation of orbits of Hermitian varieties
# under the action of the
# Projective unitary group
#####

#--> Main file

RequirePackage("pg");
Read("Param.gap");
F:=GF(q);

Read("Projective.gap");

Read("Helpers.gap");

Read("Prelim1.gap");
Read("Prelim2.gap");

Read("Unitary.gap");

Read("Orbits.gap");

Read("Post-orb1.gap);
Read("Post-orb2.gap);
Read("Post-orb3.gap);
Read("Post-orb4.gap);
Read("Post-orb5.gap);
Read("Post-orb6.gap);

Print("\n Done\n");

The parameters that determine the computation, that is the dimension n of the space and the
square q are defined in Param.gap.

113
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

4.2.3 Auxiliary functions:

Projective and Helpers
Both the files Projective.gap and Helpers.gap contain functions to support the com-
putation. The functions in the file Projective.gap are meant to be re-usable for general
work on projective spaces; on the other hand, the code in Helpers.gap has been written just
for this program and it makes heavy use of global variables.
While it is agreed that usually global variables are to be avoided, their choice for this very
application has been dictated by our wish to simplify the syntax of some functions and increase
the readability of the code. Another consideration that had to be made is that GAP transfers
parameters to functions ‘by value’: all variables are copied whenever a function call happens.
This could be (and is) quite a problem in ‘memory starvation’ situations, when huge amounts
of data need to be transferred as arguments, as it may happen in our case.
The Projective.gap file is as follows.
Projective.gap

##############
# General Use Macros
##############

#General Procedures
SetMinus := function(A,B)
local x,T,U;
U:=A;
T:=IntersectionSet(A,B) ;
#This should speed things up
for x in T do
RemoveSet(U,x);
od;
return(U);
end;;

SetMinus1 := function(A,B)
local x,T,U;
U:=A;
for x in B do
RemoveSet(U,x);
od;
return(U);
end;;

Prod := function (l1,l2,l3)

114
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

local i,t,u;
t:=0;
u:=Length(l1);
for i in [1..u] do
t:=t+l1[i]*l2[i]*l3[i];
od;
return t;
end;;

#scalar product
Scal := function (l1,l2)
return Prod(l1,l2,[1,1,1,1]);
end;;

#Hermitian
Herm1 := function(l1,l2,v,f)
return(Prod(List(l1,t->t),List(l2,t->tˆf),v));
end;;

HermM := function(l1,M,f)
local l1q,idl;
idl := List(l1,t->tˆ0);
l1q := List(l1,t->tˆf);
return(Prod(l1q*M,l1,idl));
end;;

Herm := function(l1,l2,f)
return(Herm1(l1,l2,[1,1,1,1],f));
end;;
#Forms associated to Hermitian product
HermS := function(l1,f)
return(Herm(l1,l1,f));
end;;

HermS1 := function(l1,v,f)
return(Herm1(l1,l1,v,f));
end;;

#####################################
# Pr Space Functions Implementation
#####################################

# Some functions have three variants:

115
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

# function(V): Construct from vector space

# function1(n,K): Construct the object
# of dim n over K
# function2(W,K): Construct from
# the set W over K
##
# These functions are all quite verbose

#############################
# We want to construct a projective space P
# as a records of the following:
# P.points
# P.lines
# P.field
# P.size
# P.order
############################

##########################################
# Function for constructing a vector space
# of dimension n over K
##########################################

#Canonical base vector

CanonicBase:= function(i,n,K)
local k,L,gen,u,z;
gen:=Z(Size(K));
u:=genˆ0;
z:=gen*0;
L:=[];
for k in [1..n] do
if k=i then
Add(L,u);
else
Add(L,z);
fi;
od;
return(L);
end;;

#Return the n dimensional vector space over K

Vsp:= function(n,K)
local i,k,gen,u,z,L,U;
U:=[];
if n=0 then

116
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

return([]);
fi;
for i in [1..n] do
L:=CanonicBase(i,n,K);
Add(U,L);
od;
return(VectorSpace(U,K));
end;;

#Actual construction of the projective space

#
# As point representative we assume the
# vector in the equivalency class whose
# first non-zero entry is 1
Norma := function(v)
local w,u,i,j;
u:=Length(v);
for i in [1..u] do
if not(v[i]-v[i]=v[i]) then
return(List(v,x->x/v[i]));
fi;
od;
return(v);
end;;

# List of the points of the projective space.

# There are different functions for this
# task.
#
# The first one we provide is extremely
# slow.

Alt1PrPoints := function(V)
local W,L,U,P,G,x,y,q;
G:=V.field;
q:=Characteristic(G);
W:=Set(Elements(V));
P:=[];
RemoveSet(W, V.zero );
while Size(W)>0 do
x:=Elements(W)[1];
Add(P,Norma(x));
Print(Norma(x)," ",Size(W),"\n");
U:=VectorSpace([ x ], G);

117
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

L:=Set(Elements(U));
# Removing the zero from L provides quite
# a remarkable performance improvement
# at the beginning of the computation.
RemoveSet(L,U.zero);

W:=SetMinus1(W,L);
#Observe that we already know that
# $W\cap L=L$.
od;
Print("\n P:=",Set(P),"\n");
return(Set(P));
end;;

# This implementation is faster than the

# one above: the set command is used in
# order to reorder the list P

PrPoints := function(V)
local W,L,U,P,G,H,x,y,q,d,l,m;
G:=V.field;
d:=Dimension(V);
l:=Size(G);
m:=(lˆd-1)/(l-1);
q:=Characteristic(G);
W:=Set(Elements(V));
H:=[];
RemoveSet(W, V.zero );
for x in W do
Add(H,Norma(x));
H:=Set(H);
Print(Norma(x)," ",Size(H),"\n");
if Size(H)=m then
P:=Set(H);
Print("\n P:=",P,"\n");
return(P);
fi;
od;
P:=Set(H);
Print("\n P:=",P,"\n");
return(P);
end;;

%%%%%%%%%%
###########################

118
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

# When the full projective space is needed this

# is actually the best option: we do not normalize
# but just add vectors already in canonical form
# (still it takes some time to run)
###########################

FullPrPoints:= function(n,K)
local T,P,U,L,x,gen,u,z,i,y,a;
gen:=Z(Size(K));
u:=genˆ0;
z:=gen*0;
P:=[];
for x in [1..n+1] do
if x=1 then
L:=[u];
Print("L:=",L,"\n");
else
L:=CanonicBase(x,x,K);
# for a in [1..x-1] do
# Add(L,z);
# This is the old routine # od;
# Add(L,u);
# fi;
Print("L:=",L,"\n");
fi;
if x=n+1 then
L:=CanonicBase(x,x,K);
Add(P,L);
else
U:=Elements(Vsp(n+1-x,K));
for y in U do
T:=[];
for i in L do
Add(T,i);
od;
for i in y do
Add(T,i);
od;
Add(P,T);
P:=Set(P);
Print("T=",T," ",Size(P),"\n");
od;
fi;
od;
return(Set(P));

119
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

end;;

############################################
# Now we define the lines....
# Note that we can not use the ‘‘FullPrPoints’’
# function here.
############################################

#First case: arguments=(n,K).

#Note that it is cheaper to compute
# an intersection rather than to normalize
PrLines1 := function(n,K)
local W,L,U,x,y;
W:=FullPrPoints(n,K);
L:=[];
for x in Elements(W) do
RemoveSet(W,x);
for y in Elements(W) do
U:=IntersectionSet(W,
Set(Elements(VectorSpace([x,y],K))));
Add(L,U);
Print("\n Line U:=",U,"; size of L:=",
Size(Set(L)),"\n");
od;
od;
Print("\n Set of lines L:=",Set(L),"\n");
return(Set(L));
end;;

#Second case: the set of points is already given

PrLines2 := function(W,K)
local L,U,x,y;
L:=[];
for x in Elements(W) do
RemoveSet(W,x);
for y in Elements(W) do
U:=IntersectionSet(W,
Set(Elements(VectorSpace([x,y],K))));
Add(L,U);
Print("\n Line U:=",U,"; size of L:=",
Size(Set(L)),"\n");
od;
od;
Print("\n L:=",Set(L),"\n");

120
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

return(Set(L));
end;;

#Third case: we give the vector space

PrLines := function(V)
local W,L,K,U,x,y;
W:=PrPoints(V);
K:=V.field;
L:=[];
for x in Elements(W) do
RemoveSet(W,x);
for y in Elements(W) do
U:=IntersectionSet(W,Set(VectorSpace([x,y],K)));
Print("\n Line U:=",U,"; size of L:=",
Size(Set(L)),"\n");
Add(L,U);
od;
od;
Print("\n L:=",Set(L),"\n");
return(Set(L));
end;;

# Points and lines enable us to define

# all the subspaces of the linear structure
# This is a recursive function and I expect
# it to be very slow; however it is hardly
# ever used in this form.

PrSubSpaces := function(V,n)
local Sub,P,U,N2,K2,x,y;
Sub:=[];
if n=0 then
return(PrPoints(V));
fi;
if n=1 then
return(PrLines(V));
fi;
#We need to use PrPoints here
P:=PrPoints(V);
#Note that since n>1, N2 is already a list
# of points
N2:=PrSubSpaces(V,n-1);
for x in N2 do
U:=SetMinus(P,Set(x));
for y in U do

121
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

Add(x,y);
K2:=IntersectionSet(P,
Set(Elements(VectorSpace(x,V.field))));
Add(Sub,K2); #We avoid PrPoints again
Print("Adding K2:=",K2,"\n Size of Sub:=",
Size(Set(Sub)),"\n");
od;
od;
return(Set(Sub));
end;;

PrSubSpaces1 := function(m,K,n)
local Sub,P,U,N2,K2,x,y;
Sub:=[];
if n=0 then
return(FullPrPoints(m,K));
fi;
if n=1 then
return(PrLines1(m,K));
fi;
P:=FullPrPoints(m,K);
#Note that since n>1,
# N2 is already a list of Points....
N2:=PrSubSpaces1(m,K,n-1);
for x in N2 do
U:=SetMinus(P,Set(x));
for y in U do
Add(x,y);
K2:=IntersectionSet(P,
Set(Elements(VectorSpace(x,K))));
Add(Sub,K2);
Print("Adding K2:=",K2,"\n Size of Sub:=",
Size(Set(Sub)),"\n");
od;
od;
return(Set(Sub));
end;;

PrSubSpaces2 := function(W,K,n)
local Sub,P,U,N2,K2,x,y;
Sub:=[];
if n=0 then
return(W);
fi;
if n=1 then

122
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

return(PrLines2(W,K));
fi;
P:=W;
#Note that since n>1,
# N2 is already a list of Points....
N2:=PrSubSpaces2(W,K,n-1);
for x in N2 do
U:=SetMinus(P,Set(x));
for y in U do
Add(x,y);
K2:=Set(IntersectionSet(W,
Set(Elements(VectorSpace(x,K)))));
Add(Sub,K2);
Print("Adding K2:=",K2,"\n Size of Sub:=",
Size(Set(Sub)),"\n");
od;
od;
return(Set(Sub));
end;;

PrSubSpaces3 := function(W,L,K,n)
local Sub,P,U,N2,K2,x,y;
Sub:=[];
if n=0 then
return(W);
fi;
if n=1 then
return(L);
fi;
P:=W;
N2:=PrSubSpaces2(W,K,n-1);
for x in N2 do
U:=SetMinus(P,Set(x));
for y in U do
Add(x,y);
K2:=IntersectionSet(W,
Set(Elements(VectorSpace(x,K))));
Add(Sub,K2);
Print("Adding K2:=",K2,"\n Size of Sub:=",
Size(Set(Sub)),"\n");
od;
od;
return(Set(Sub));
end;;

123
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

#space+function
Variety1 := function(V,f)
local U,x;
U:=[];
for x in V do
if f(x)-f(x)=f(x) then
Add(U,x);
fi;
od;
return(Set(U));
end;;

#space+list of functions
Variety := function(V,F)
local U,x,y,t,l,f;
l:=Size(F);
for x in V do
y:=0;
for t in F do
if t(x)-t(x)=t(x) then
y:=y+1;
fi;
od;
if y=l then
#All conditions satisfied!
Add(U,x);
fi;
od;
return(Set(U));
end;;

###############

This is the listing of Helpers.gap.

Helpers.gap

###########
# Helper Functions
###########
HermV := function(X, f)
return(HermS(PtToVect(X),f));
end;;

TestHpoint := function(x,f)

124
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

return(HermV(x,2)-HermV(x,2)=HermV(x,2));
end;;

HermVM :=function(x,M,f)
return(HermM(PtToVect(x),M,f));
end;;

TestHMpoint := function(x,M,f)
return(HermVM(x,M,f)-HermVM(x,M,f)=HermVM(x,M,f));
end;;

CreateHermVar := function(M,Pt,f,ind)
local x, listH;
listH:=[];
for x in Pt do
if ind=1 then
Print(HermVM(x,M,f),"\t");
fi;
if (TestHMpoint(x,M,f)) then
Add(listH,x);
if ind=1 then
Print(" ...OK\t"); #,x,"\n");
fi;
else
if ind=1 then
Print(" ...no\t"); #,x,"\n");
fi;
fi;
od;
return(listH);
end;;

#############

4.2.4 Preliminary computations:

Prelim.gap
The module prelim.gap constructs some tables and functions to implement different repre-
sentations for Hermitian varieties:
√
(i) the representation as a point of PG(n2 + 2n, q),

(ii) the representation as an element of Mat(n + 1, q) and

(iii) the presentation of the set of all GF(q)-rational points of the variety.

125
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

The first step consists in constructing the projective space PG(n, q) via a library function and
defining the sets associated with the canonical non-degenerate Hermitian variety UnI.
As seen in Theorem 1.7.17, the set of all Hermitian hypersurfaces in PG(n, q) is a PG(n2 +
2n, p).
In order to establish a 1—1 correspondence between the set of all Hermitian varieties in
PG(n, q) and PG(n2 + 2n, p), the following algorithm is used.

Definition 4.2.1. For any point x ∈ PG(n, q). A normalised representative for x is a vector t
in the underlying vector space V such that

(i) Pt = x;

(ii) the first non-zero component of t is 1.

We denote the normalised representative of x as x̃.

For any i = 1, . . . n, define

(i) α := [(n + 1)(n + 2)]/2;

(ii) β(i) := [(n + 1)(n + 2) − (n − i)(n − i + 1)]/2 + 1;

(iii) γ(i) := β + [n(n + 1) − (n − i − 1)(d − i)];

and assume ² to be a primitive element of GF(q) and η a primitive element of GF(p) =

√
GF( q). Consider the following mapping between PG(n2 + 2n, p) and Her(n + 1, q):
½
PG(n2 + 2n, p) → Her(n + 1, q)
ϑ:
x → ρ(x̃) = M,

where the (n + 1) × (n + 1) matrix M has the following entries

(i) for i = j, Mii := ηx̃β(i) ;

(ii) for i < j,

Mij := ηx̃β(i)+j−i + ²x̃α+γ(i)+j−i

(iii) for j < i, Mji = Mij .

It is straightforward to verify that such a mapping is a bijection and that it preserves the linear
structure of PG(n2 + 2n, p). The gap code to implement this correspondence is as follows.
Prelim1.gap

#######
# Preliminary computations
#######

126
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

Print("Start Computations\n");
S:=ProjectiveGeometry(n,q);
P:=ProjectivePoints(S);
UnI:=CreateHermVar(Z(p)*IdentityMat(n+1),P,p,1);
UnIp:=Set(List(UnI,t->PtToVect(t)));
Print("UnI is now defined\n");

########
# Construction of the set of all
# Hermitian matrices of given
# size over a field
########

# We consider all Hermitian matrices:

# they form a PG(nˆ2+2n,p)
#
# Encoding:
# First n*(n+1)/2 entries -> entries over
# the ground field;
# Remaining entries -> "complex" part.
#

Print("Take the space of all Hermitian

Matrices: dim=",nˆ2+2*n,"\n");
MatricesSpace:=ProjectiveGeometry(nˆ2+2*n,p);
Print("Now, the points are computed...\n");
MatricesPoints:=ProjectivePoints(MatricesSpace);
SizOfM:=Size(MatricesPoints);
MatricesSet:=[];
#
Print("Start the cycle...\n");
# Basic cycle on points
offset:=(n+1)*(n+2)/2;
for x in MatricesPoints do
tempMatrix:=[];
for l in [1..n+1] do
Add(tempMatrix,[]);
od;
tempVect:=PtToVect(x);
#
# Actual construction of the matrix
#
for i in [0..n] do
#Real Entries start here!
refpos:=(n+1)*(n+2)/2-(n-i+1)*(n-i+2)/2+1;

127
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

#While complex ones start here!!

refpos2:=offset+n*(n+1)/2-(n-i)*(n-i+1)/2;
#
# Computing the rows....
#
for j in [0..n] do
# Diagonal entries are "real"...
if i=j then
tempMatrix[i+1][i+1]:=tempVect[refpos]*Z(p);
Print("coord=",refpos,"--> (",i+1,",",j+1,")\t",
tempVect[refpos],"\n");
fi;
# Upper triang. part: get "real" and "complex"
# part for each entry
if i<j then
tempMatrix[j+1][i+1]:=tempVect[refpos+j-i]*Z(p)+
tempVect[refpos2+j-i]*Z(q);
Print("coord=",refpos+j-i,";",refpos2+j-i,
"--> (",i+1,",",j+1,")\t",
tempVect[refpos+j-i],"+",
tempVect[refpos2+j-i],"i\n");
fi;
# Obtain the lower trian. by conjugation
if i>j then
tempMatrix[j+1][i+1]:=(tempMatrix[i+1][j+1])ˆp;
fi;
od;
od;
Print(tempMatrix,"\t",Size(MatricesSet)+1,"/",
SizOfM,"\n");
Add(MatricesSet,tempMatrix);
od;
Print("List of Matrices computed\n");

##########

Once all Hermitian matrices in Her(n + 1, q) have been constructed, it is possible to produce
two more lists: the list of all rational points of Hermitian varieties and the list of determinants
corresponding to given varieties. Since these lists are ordered objects, functions for converting
from one representation to the other are the most straightforward.
Prelim2.gap

#
# Create List of Hermitian Varieties

128
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

VarHermList := List(MatricesSet,
M->CreateHermVar(M,P,p,1));
DetList := List(MatricesSet,M->Determinant(M));

#
# Define functions for associating Matrices, Points
# and Varieties.
# ‘‘Helpers2.gap’’
#

MatToPt := function(M)
return(MatricesPoints[Position(MatricesSet,M)]);
end;;

PtToMat := function(V)
return(MatricesSet[Position(MatricesPoints,V)]);
end;;

VarToMat := function(V)
return(MatricesSet[Position(VarHermList,V)]);
end;;

MatToVar := function(V)
return(VarHermList[Position(MatricesSet,V)]);
end;;

VarToPt := function(V)
return(MatricesPoints[Position(VarHermList,V)]);
end;;

PtToVar := function(V)
return(VarHermList[Position(MatricesPoints,V)]);
end;;

###########

4.2.5 Unitary group construction:

Unitary.gap
The very simplest way for defining the Projective Unitary group is by considering the stabiliser
of the point set UnIp in PGL(n + 1, q). The group PGL(n + 1, q) is seen as the quotient group
of GL(n + 1, q) modulus its centre. For each of the equivalence classes PGL(n + 1, q) consists

129
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

of, we choose the matrix of determinant 1 as a representative. This is what the following code
aims to do. Observe that the action of PGL(n + 1, q) on the Hermitian curve is given by the
function DoHermActPGL().
Unitary.gap

#
# Start the actual computations....
# we want to implement PGL
#

Print("\n Take GL(",d+1,",",q,")... ");

GenL := GeneralLinearGroup(d+1,q);
Print("and get its Centre Z\n");
ZGenL:= Centre(GenL);
Print("Now define the Homomorphism
with Z(GL(",d+1,",",q,")) as its kernel\n");
Phom := NaturalHomomorphismByNormalSubgroup(GenL,
ZGenL);
Print("... and PGL is its epimorphic image\n");
PGenL:= Image(Phom);

#
# Helpers 3.
# Action of PGL on the representatives of Hermitian
# forms
#

# canonical preimage
GetPreImg:=function(t)
local pre1,pre2;
pre1:=Elements(PreImages(Phom,t));
pre2:=pre1[1]/Determinant(pre1[1]);
return(pre2);
end;;

# action of PGL on the set of matrices by conj

DoConjFromPGL:=function(M,t)
local pre;
pre:=GetPreImg(t);
return(preˆ(-1)*M*pre);
end;;

# action of PGL on the set of Hermitian forms

# as represented by Hermitian matrices

130
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

DoHermActPGL:=function(M,t)
local pre;
pre=GetPreImg(t);
return(TransposedMat(preˆp)*M*pre);
end;;

# Do a copy of the set of Matrices

ListTemp := Set(ShallowCopy(MatricesSet));

#
# Some further helper functions...
#

UnitTest1:=function(t)
return(TransposedMat(tˆp)*t);
end;;

UnitTest:=function(t)
local pre,com;
com:=DoConjFromPGL(IdentityMat(d+1),t);
return(Image(Phom,com));
end;;

UniHom:=GroupHomomorphismByFunction(PGenL,PGenL,
UnitTest);
UniKer:=Kernel(UniHom);
PGU:=PGenL/UniKer;

#########

4.2.6 Orbit computation:

Orbits.gap
The computation of the orbits under the action of the Projective Unitary Group is done via the
standard library function Orbits.
Orb-compute.gap

#
# Orbits computation
#

Print("Computing Orbits.... (way too slow...)\n");

StaL:=Stabilizer(PGenL,MatricesSet,DoHermActPGL);

131
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

Orb1:=Orbits(PGenL,MatricesSet,DoHermActPGL);
Print("Orbit Computation done; renormalizing...");
Orb:=List(Orb1,
t->IntersectionSet(Set(t),MatricesSet));
Print(" Done!\n");

#########

4.2.7 Intersection and results output:

Post-orb.gap
The first element in any single orbit is chosen as representative and some cardinalities are
printed out.
Post-orb1.gap

OrbL:=List(Orb,t->Size(t));
OrbR:=List(Orb,t->t[1]);
IdMat:=VarToMat(UnI);
for i in [1..Size(Orb)] do
if IdMat in Orb[i] then
OrbR[i]:=IdMat;
fi;
od;
OrbP:=List(OrbR, t->MatToPt(t));
Print("\n --- Orbits Under the action
of the (P)GL group --- \n");
Print(Orb,"\n");
Print("Number of Orbits :=",Size(Orb),"\n");
Print("Size of Orbits :=",OrbL,"\n");
Print("Set of Representatives :=", OrbR, "\n");
#Now Consider linear combinations...
IdPointPos:=Position(OrbP,MatToPt(IdMat));
IdPoint:=OrbP[IdPointPos];
SameInter:=[];
Print("Classes\t\t Sizes\n");

Next, there is some code in order to perform a consistency check. The MergeClass and
TryClass functions are used to merge all the orbits that lead to the same intersection con-
figuration. Lemma 3.2.1 guarantees that that MergeClass should act as the identity on the
orbits.
Post-orb2.gap

### ###

132
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

# Consistency check #
### ###

TryClass := function( n )
local a1, a2, a3, a4, x;
a4:=[n];
for x in Orb[n] do
for lamb in GF(p) do
a1 := lamb*IdMat+x;
a2 := VarToMat(CreateHermVar(a1,P,p,0));
a3 := PositionProperty(Orb,t-> a2 in t);
Print("From class: ",n, " -> To class: ", a3, "\n");
Add(a4,a3);
od;
od;
return(Set(a4));
end;;

SameInter:=[];
for i in [1..Size(Orb)] do
if not i=IdPointPos then
Add(SameInter,TryClass(i));
fi;
od;

### ###
# Merge Classes #
### ###

SameInterS:=Set(List(SameInter,t -> Set(t)));

TmpList:=[];
SameKindOfInter:=[];
for i in SameInterS do
iPos:=Position(SameInterS,i);
for j in SameInterS do
jPos:=Position(SameInterS,j);
Tmp:=IntersectionSet(i,j);
if not Tmp=Set([]) then
Tmp2:=Set(Union(i,j));
SameInterS[iPos]:=Tmp2;
SameInterS[jPos]:=Tmp2;
fi;
od;

133
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

od;
SameInterS:=Set(SameInterS);
Print("To Merge: ",SameInterS,"\n");
for i in SameInterS do
TmpOrb:=Union(List(i, t->Orb[t]) );
Add(TmpList,TmpOrb);
od;
Add(TmpList,[[IdMat]]);

Post-orb3.gap

SameKindOfInter:=Set(TmpList);
Print("Sizes of orbits =",List(Orb,t->Size(t)),
"\n");
Print("Sizes of classes =",
List(SameKindOfInter,t->Size(t)),"\n");

Now, the code computes the actual intersection with UnI and produces a list with the pos-
sible intersection sizes.
Post-orb4.gap

###
# Compute Actual intersection with U
###

IntersectMatWithU:=function(M)
local Var,Var1,t;
if M=[ IdMat ] then
return(UnIp);
fi;
Var:=MatToVar(M);
Var1:=Set(List(Var,t->PtToVect(t)));
return(IntersectionSet(UnIp,Var1));
end;;
SizeOfInters:=[];

# Check classes

IntWithUList:=[];
for x in SameKindOfInter do
Print("Orbit number ",
Position(SameKindOfInter,x),"\n");
TmpList:=[];
TmpWithUList:=[];

134
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

for y in x do
Add(TmpList,Size(IntersectMatWithU(y)));
Add(TmpWithUList,IntersectMatWithU(y));
od;
Add(IntWithUList,TmpWithUList);
Print("Sizes: ",TmpList,"\n");
Add(SizeOfInters,Set(TmpList));
od;
DiffConfList:=List(IntWithUList,t->Set(t));

# Check if some classes yield the same intersection

# and, if the case, merge them (ToUniteList)

TmpList:=[];
TmpList2:=[];
ToUniteListTmp:=[];
for i in DiffConfList do
Add(TmpList,Position(DiffConfList,i));
od;
l:=Size(TmpList);
for i in [1..l] do
for j in [0..l-1] do
k:=Position(TmpList,i,j);
if IsInt(k) then
Add(TmpList2,k);
fi;
od;
if not IsEmpty(TmpList2) then
Add(ToUniteListTmp,Set(TmpList2));
fi;
TmpList2:=[];
od;
ToUniteList:=Set(ToUniteListTmp);
TmpList:=[];
TmpList2:=[];
for i in ToUniteList do
TmpList2:=Union(List(i, t->SameKindOfInter[t]) );
Add(TmpList,TmpList2);
od;
MergedIntersections:=Set(TmpList);

The last two modules, Post-orb5 and Post-orb6 deal with the output and the saving of
the results. In Post-orb5, the program prints out a table which correlates size of orbits

135
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

(intersection classes) and the intersection sizes.

Post-orb5.gap

Print("--------------> Table <-----------------\n");

Print("Sizes of orbits =",
List(Orb,t->Size(t)),"\n");
Print("Sizes of classes =",
List(SameKindOfInter,t->Size(t)),"\n");
Print("Corr. sizes of int =",SizeOfInters,"\n");
Print("No. of distinct conf. per class:",
List(DiffConfList,t->Size(t)),"\n");
Print("Classes that yeld same intersections:",
ToUniteList,"\n");
Print("Corr. no. of varieties =",
List(MergedIntersections,t->Size(t)),"\n");
Print("Corr. sizes =",
List(ToUniteList,t->SizeOfInters[t[1]]),"\n");
Print("Size of U = ",Size(Unp),"\n");
Print("----------------------------------------\n");

The list Orb contains all the representatives for each of the possible intersection classes. Post-orb6
provides a function to dump the result of all computation on a file. The format is such as to make
it simple to reload the data, if need be.
Post-orb6.gap
DumpInfo:=function(filenm)
local i,j,k,t,kk;
PrintTo(filenm,"------> Status report <------\n\n");
AppendTo(filenm,"General Info:\n");
AppendTo(filenm,"\t p:=",p,"\n\t q:=",q,"\n\t d:=",
d,"\n");
AppendTo(filenm,"Space:= PG(",d,",",q,")\n\n");
AppendTo(filenm,"----> Table of sizes \n\n");
AppendTo(filenm,"Sizes of orbits =",
List(Orb,t->Size(t)),"\n");
AppendTo(filenm,"Sizes of classes =",
List(SameKindOfInter,t->Size(t)),"\n");
AppendTo(filenm,"Corr. sizes of int =",
SizeOfInters,"\n");
AppendTo(filenm,"No. of distinct conf. per class:",
List(DiffConfList,t->Size(t)),"\n");
AppendTo(filenm,"Classes that yield same
intersections:",ToUniteList,"\n");
AppendTo(filenm,"Corr. no. of varieties =",
List(MergedIntersections,t->Size(t)),"\n");

136
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

AppendTo(filenm,"Corr. sizes =",

List(ToUniteList,t->SizeOfInters[t[1]]),"\n");
AppendTo(filenm,"Size of U = ",Size(UnIp),"\n\n");
AppendTo(filenm,"----> Matrices in classes");
for i in SameKindOfInter do
AppendTo(filenm,"Class number: ",
Position(SameKindOfInter,i),"\n");
for j in i do

AppendTo(filenm,"\t M[",Position(i,j),"]= [\n");

for k in j do
AppendTo(filenm,"\t\t", k,"\n");
od;
AppendTo(filenm,"\t]\n;");
od;
AppendTo(filenm,"----\n");
od;
AppendTo(filenm,"\n\n ---> Configurations in
classes\n");
for t in [1..Size(SameKindOfInter)] do
i:=SameKindOfInter[t];
AppendTo(filenm,"Class number(s): ",
Elements(Filtered(ToUniteList,
hh -> t in hh)),"\n");
for j in i do
AppendTo(filenm," Case number:",
Position(i,j),"\n");
AppendTo(filenm," Points (");
kk:=IntersectMatWithU(j);
AppendTo(filenm,Size(kk),"):\n");
for k in kk do
AppendTo(filenm,"\t\t",k,"\n");
od;
AppendTo(filenm,"--\n");
od;
AppendTo(filenm,"------\n");
od;
end;;

#####

137
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

4.2.8 Generators of the projective unitary group

Direct computation of PU(n + 1, q) as the stabiliser of Un in PG(n, q) proves to be highly
unpractical for all but the most trivial cases. This is mostly due to the size of the objects involved
and constraints on memory available on the computers. Hence, some results on generating
sets for this group have been studied, in order to be able provide an explicit representation.
Such results have not been implemented in the code presented before. However, some explicit
computations have been carried out and we expect to use them in a new version of the program.
The results from Section 1.6 on quasi-symmetries will be used here.

Quasi-symmetries

Let Un be the standard Hermitian hypersurface Un in PG(n, q) = PV .

We recall that any quasi-symmetry σ of V fixes an hyperplane; hence, modulus a suitable
change of coordinates, σ can be represented as a diagonal matrix

Q(n+1)
α = diag(α, 1, 1, . . . , 1),

with αα = 1.
On the other hand, a direct computation shows that, by conjugation of Qα with the diagonal
matrix
T = diag(t, 1, 1, . . . , 1),
we get an equivalent transformation of the projective space PG(n, q); this transformation is
(n+1)
Qt2 α . It follows that there are as many distinct classes of quasi-symmetries as classes of
squares in the field K the vector space V is defined on. If K = GF(q), this means that there
are exactly two classes of quasi-symmetries.
√
Since GF( q) is included in the set of squares of GF(q), we may assume without loss of
generality the following:

(i) α ∈ {1, ²}, where ² is a fixed element of norm 1 which is not a square in GF(q);

(ii) the conjugates are computed modulus transformations in the Special Linear group SL(n +
1, q).

A direct computation for n = 1, 2 yields the following results.

(2)
(i) If n = 1 then, any quasi-symmetry is a conjugate of the diagonal matrix Dα = diag(α, 1);
hence, for any T ∈ SL(2, q):
· ¸
−1 t11 αt22 − t12 t21 t11 t12 (1 − α)
T AT = .
t21 t22 (α − 1) −t12 αt21 + t11 t12

138
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

(3)
(ii) If n = 2 then, the conjugate of Dα by T ∈ SL(3, q) is:
 
η1 t1,1 − η2 t2,1 + η3 t3,1 η1 t1,2 − η2 t2,2 + η3 t3,2 η1 t1,3 − η2 t3,3 + η3 t3,3
 η5 t2,1 − η4 t1,1 − η6 t3,1 η5 t2,2 − η4 t1,2 − η6 t3,2 η5 t2,3 − η4 t1,3 − η6 t3,3  ,
η7 t1,1 − η8 t2,1 + η9 t3,1 η7 t1,2 − η8 t2,2 + η9 t3,2 η7 t1,3 − η8 t2,3 + η9 t3,3

where
η1 = α(t2,2 t3,3 − t2,3 t3,2 ); η2 = t1,2 t3,3 − t1,3 t3,2 ;
η3 = t1,2 t2,3 − t1,3 t2,2 ; η4 = α(t2,1 t3,3 − t2,3 t3,1 );
η5 = t1,1 t3,3 − t1,3 t3,1 ; η6 = t1,1 t2,3 − t1,3 t2,1 ;
η7 = α(t2,1 t3,2 − t2,2 t3,1 ); η8 = t1,1 t3,2 − t1,3 t3,1 ;
η9 = t1,1 t2,2 − t1,2 t2,1 .

Definition 4.2.2. For any matrix T ∈ Mat(n, K) and for any d ∈ K, let T(x,y)=d be the matrix
in which the entry txy is replaced by the value d, and let T xy be the minor of T obtained by
deleting the x-th row and the y-th column from T .
(n+1)
Lemma 4.2.3. Let M be a conjugate of Dα via a matrix T ∈ SL(n + 1, q). Then,

Mij = det T(i,j)=tij α .

Proof. By expanding the row/column product,

(n+1)
(i) for i = 1, (Dα T )1j = αt1j ;
(n+1)
(ii) for i 6= 1, (Dα T )ij = tij .

Since
(T −1 )ij = (−1)i+j det T ij ,
a new computation yields
n+1
X
−1
(T Dα(n+1) T )ij = det T (i,1)
αt1j + (−1)i+j det T ik tkj = det(T(i,j)=tij α ).
k=2

139
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

Ishibashi’s theorem

For the odd characteristic case, it is possible to use a different result: a theorem by Ishibashi
[Ish85] provides a small set of generators for the unitary group.
Let f be an unitary form over a vector space V over GF(q) and assume that V has hyperbolic
rank at least 1. Denote by q(x) the Hermitian form associated to f and decompose V in the
direct product
V = H ⊥ L,
where H is a hyperbolic plane with isotropic base u, v. Furthermore, we may assume that

f (u, v) = λf (v, u),

and define the following five kinds of mappings:

(i) the isometry ∆ that fixes L and acts on H by mapping u → v, v → λu;

(ii) the isometry φ(²) that, for any ² ∈ GF(q), fixes L and sends u → ²u and v → ²−1 v;

(iii) the Eichler transformation (quasi-transvection) E(u, x), defined for any x ∈ L, that acts
on z ∈ V as

E(u, x) : z → z − λf (z, u)x + f (z, x)u − λf (z, u)q(x)u;

(iv) the symmetry τ (x), given, for any x ∈ V with q(x) 6= 0, by

τ (x) : z → z − f (z, x)q(x)−1 x;

(v) the permutation σ that fixes the hyperbolic plane H and permutes the vectors of a given
orthogonal base {z1 , . . . , zn−2 } of L in such a way as to have

z1 →σ z2 →σ . . . →σ zn−2 ,

where q(zi ) = q(zj ) for all i, j.

Theorem 4.2.4. For n > 2 odd, the group Uf (n, q) is generated by three isometries, namely:

(i) E(u, z1 ),

(ii) ∆σ,

(iii) φ(α), where α is a primitive element of GF(q)? .

From now on, we shall assume α to be a primitive element of GF(q)? .

140
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

Theorem 4.2.5. Assume that the restriction f |L is equivalent to the Hermitian form induced by
the identity matrix diag (1, . . . , 1, 1). Then, for n > 2, n even, the group Uf (n, q) is generated
by the three isometries

(i) E(u, z1 ),

(ii) ∆σ,

(iii) φ(α)τ (z1 − zn−2 ).

Theorem 4.2.6. Assume that the restriction f |L is equivalent to the Hermitian form induced by
a matrix diag (1, . . . , 1, α), with −2−1 = α2k+1 . Then, for n > 2, n even, the group Uf (n, q) is
generated by the three isometries

(i) E(u, z1 + zn−2 ),

(ii) ∆τ (z1 )σ,

(iii) φ(α)τ (z1 ).

Theorem 4.2.7. Assume that the restriction f |L is equivalent to the Hermitian form induced
by the matrix diag (1, . . . , 1, α), with −2−1 = α2k . Then, for n > 4, n even, the group Uf (n, q)
is generated by the three isometries

(i) E(u, z1 + zn−2 ),

(ii) ∆τ (z1 )σ,

(iii) φ(α)τ (z1 − zn−3 ).

The group Uf (4, q) is generated by

(i) E(u, z1 + z2 ),

(ii) ∆,

(iii) φ(α)τ (z1 ).

Remark 4.2.8. In the cases of Theorems 4.2.6 and 4.2.7, the group Uf (n, q) is the same as the
orthogonal group Of (n, q).

141
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

Construction of the matrices

Assume B = {u, v, z1 , . . . , zn−2 } be a fixed base of V , such that H = hu, vi is an hyperbolic

plane and let U be the matrix representing the Hermitian form f with respect to B.
Then, the transformations E, ∆ and σ have to satisfy the following conditions.

(i)  
0 1
 λ 0 
 
 1 0 
∆= ;
 . . 
 . 
0 1

(ii)  
² 0
 0 ²−1 
 
 1 
φ(²) =  ;
 ... 
 
1

(iii)
E(u, x) = I + A1 (x) − A2 (x),
where
A1 (x) = (−λxu? + ux? )U,
A2 (x) = λq(x)uu? U ;

(iv)
τ (x) = I − q(x)−1 xx? U ;

(v)  
1 0
 0 1 
 
 0 . . . 1 
σ= .
 . . 
 0 . 0 
1 ... 0

Furthermore, since it is possible to assume without loss of generality that the matrix U is
 
0 1
 1 0 
 
 1 0 
U = ,
 . .. 
 
0 1

142
CHAPTER 4. GENERAL RESULTS
4.2. EXPLICIT COMPUTATIONS AND ALGORITHMS

the transformation E(u, z1 ) can be written as

 
1 − 12 −1
 0 1 0 
 
 1 0 1 
 
 1 0 . . . 0 
E(u, z1 ) =  .
 0 1 
 
 ... 
 
0 1

4.2.9 Concluding remarks

The code introduced in this section is able to provide useful results only for small values of q
and low dimension n.
The main speed bottleneck of the program is in the computation of the orbits of the Projec-
tive Unitary Group seen acting on the set of all Hermitian matrices.
Some improvement can be attained by using the set of generators introduced in subsection
4.2 in order to define PGU(n + 1, q), albeit the best approach is definitely by providing directly
the list of the matrices in the group. However, even in this case the orbits function of GAP
seems to be too slow for getting meaningful results.
There are two main obstacles towards the solution of the problem of intersection of Hermi-
tian varieties in its full generality:

1. ‘combinatorial explosion’ of the number of possible cases, as Chapter 3 and the compu-
tation of intersection orders for Hermitian hypersurfaces in dimension 4 show;

2. lack of a general method to compute the groups except in the most simple cases.

In order to get a deeper understanding of the structures involved, it will be helpful to be able
to produce some examples for the high dimensional cases. In fact, we plan to rewrite the code
from scratch in C using the infrastructure provided by the gnump library to implement the finite
fields.

143
Bibliography

[AK92] E.F. Assmus and J.D. Key, Designs and their codes, Cambridge Tracts in Mathemat-
ics, vol. 103, Cambridge University Press, Cambridge, 1992.

[Bae52] R. Baer, Linear algebra and projective geometry, Academic Press, New York, 1952.

[BBW91] A. Blokhuis, A. Brouwer, and H. Wilbrink, Hermitian unitals are code words, Dis-
crete Math. 97 (1991), 63–68.

[BC66] R.C. Bose and I.M. Chakaravarti, Hermitian varieties in a finite projective space
PG(n, q 2 ), Canad. J. Math. 18 (1966), 1161–1182.

[Blo67] D.M. Bloom, The subgroups of PSL(3, q) for q odd, Trans. Amer. Math. Soc. 127
(1967), 150–178.

[BR98] A. Beutelspacher and U. Rosenbaum, Projective geometry: from foundations to ap-

plications, Cambridge University Press, Cambridge, 1998.

[BS86] E. Boros and T. Szőnyi, On the sharpness of a theorem of B. Segre, Combinatorica

6 (1986), 261–268.

[Cam95] P. J. Cameron, Projective and polar spaces, QMW Maths Notes/13, London, 1995.

[CEK00] A. Cossidente, G.L. Ebert, and G. Korchmáros, A group–theoretic characterization

of classical unitals, Arch. Math. 74 (2000), 1–5.

[CK97] A. Cossidente and G. Korchmáros, The Hermitian function field arising from a cyclic
arc in a Galois plane, Geometry, combinatorial designs and related structures (Cam-
bridge) (S.S. Magliveras J.W.P. Hirschfeld and M.J. de Resmini, eds.), Cambridge
University Press, 1997, pp. 63–68.

[CK98] , The algebraic envelope associated to a complete arc, Recent Progress in

Geometry (E. Ballico and G. Korchmáros, eds.), III, vol. 51, Rend. Circolo Mat.
Palermo, 1998, pp. 9–24.

144
BIBLIOGRAPHY

[CKT99] A. Cossidente, G. Korchmáros, and F. Torres, On curves covered by the Hermitian

curve, J. Algebra 216 (1999), 56–76.

[Cos97] A. Cossidente, Addendum to: “A new proof of the existence of (q 2 − q + 1)-arcs in

PG(2, q 2 )” [J. Geom. 53 (1995), no. 1-2, 37–40], J. Geom. 59 (1997), 32–33.

[CS] A. Cossidente and A. Siciliano, The geometry of Hermitian matrices of order three
over Galois fields, submitted.

[Dem68] P. Dembowski, Finite geometries, Springer–Verlag, Berlin, 1968.

[Die71] J. Dieudonné, La geometrie des groupes classiques, 3rd ed., Springer–Verlag, New
York, 1971.

[Ebe85] G.L. Ebert, Partitioning projective geometries into caps, Canad. J. Math. 37 (1985),
1163–1175.

[FHT86] J.C. Fisher, J.W.P. Hirschfeld, and J.A. Thas, Complete arcs in planes of square
orders, Ann. Discrete Math. 30 (1986), 243–250.

[Ful69] W. Fulton, Algebraic curves, The Benjamin/Cummings Publishing Company, Read-

ing, Massachusetts, 1969.

[GAP99] The GAP Group, Aachen, St Andrews, GAP – Groups, Algorithms, and Program-
ming, Version 4.1, 1999, (http://www-gap.dcs.st-and.ac.uk/˜gap).

[Giua] L. Giuzzi, A characterization of classical unitals, submitted.

[Giub] , Groups of the intersection of two hermitian curves, submitted.

[GS99] P. Govaerts and L. Storme, Projective geometry ‘pg1.0’, Department of Pure Math-
ematics and Computer Algebra, University of Ghent, 1999.

[GS00] D.R. Grayson and M.E. Stillman, Macaulay 2 version 0.8.60, 2000,
(http://www.math.uiuc.edu/Macaulay2).

[Har26] R.W. Hartley, Determination of the ternary collineation groups whose coefficients
lie in GF(2n ), Ann. of Math. 27 (1926), 140–158.

[HH70] B. Hartley and T.O. Hawkes, Rings, modules and linear algebra, Chapman and Hall,
London, 1970.

[Hir98a] J.W.P. Hirschfeld, Algebraic geometry over a field of positive characteristic, 1998,
Notes edited by L. Giuzzi.

145
BIBLIOGRAPHY

[Hir98b] , Projective geometry over finite fields, 2nd ed., Oxford Science Pubblica-
tions, Oxford, 1998.

[Hof72] A.R. Hoffer, On unitary collineation groups, J. Algebra 22 (1972), 211–218.

[Ish85] H. Ishibashi, Small systems of generators of isotropic unitary groups over finite fields
of characteristic not two, J. Algebra 93 (1985), 324–331.

[Kes81] B.C. Kestenband, Unital intersections in finite projective planes, Geom. Dedicata 11
(1981), no. 1, 107–117.

[Kes89] , A family of complete arcs in finite projective planes, Colloq. Math. 17

(1989), 59–67.

[Kin99] O.H. King, Classical groups, Notes of the Socrates intensive programme, Potenza,
1999.

[KSW73] H. Karzel, K. Sörensen, and D. Windelberg, Einführung in die Geometrie, Vanden-

hoeck and Ruprecht, Göttingen, 1973.

[Lan80] S. Lang, Linear algebra, Addison Wesley, Reading, Massachusetts, 1980.

[LN97] R. Lidl and H. Niederreiter, Finite fields, 2nd ed., Encyclopedia of Mathematics and
its Applications, vol. 20, Cambridge University Press, Cambridge, 1997.

[Lun99] G. Lunardon, Normal spreads, Geom. Dedicata 75 (1999), no. 3, 245–261.

[Mit11] H.H. Mitchell, Determination of the ordinary and modular ternary linear groups,
Trans. Amer. Math. Soc. 12 (1911), 207–242.

[Rob80] D.J.S. Robinson, A course in the theory of groups, Springer–Verlag, New York,
1980.

[Seg65] B. Segre, Forme e geometrie hermitiane con particolare riguardo al caso finito, Ann.
Mat. Pura Appl. 70 (1965), 1–201.

[Seg67] , Introduction to Galois geometries, Atti Accad. Naz. Lincei Mem. Cl. Sci.
Fis. Mat. Natur. Sez. I Mem. 8 (1967), 133–236.

[Sin38] J. Singer, A theorem in finite projective geometry and some applications to number
theory, Trans. Amer. Math. Soc. 43 (1938), 377–385.

[Tec87] H. Tecklenburg, A proof of the theorem of Pappus in finite Desarguesian affine

planes, J. Geom. 30 (1987), 172–181.

146
BIBLIOGRAPHY

[Tha87] J.A. Thas, Complete arcs and algebraic curves in PG(2, q), J. Algebra 106 (1987),
457–464.

[WC76] I. Walton and C. Cotton, The compleat angler, 5th ed., Richard Marriott, London,
1676.

[Wie64] H. Wielandt, Finite permutation groups, Academic Press, New York and London,
1964.

147