Professional Documents
Culture Documents
March 9, 2023
Abstract
The parameters of the AG codes on general linear groups are found. The hyperplane
sections having the minimum (or maximum) number of rational points are determined.
1 Introduction
In this article, we analyze the hyperplane sections of general linear groups over finite fields.
The main goal of our paper is to investigate the simplest instance of a family of linear error
correcting codes that we construct by using good filtrations. Let GLn (F ) denote the general
linear group of n × n matrices defined over a finite field F . To motivate our discussion, we
begin with the case of 2 × 2 matrices over F = F2 , the field with two elements. Let
a b
GL2 (F ) := : {a, b, c, d} ⊆ F and ad − bc = 1 . (1.1)
c d
The defining representation of GL2 (F ) is the two dimensional vector space V := F 2 . This
is a simple GL2 (F )-module. Let S(V ∗ ) denote the symmetric algebra on the dual vector
space V ∗ . By S r (V ∗ ) we denote the homogeneous part of degree r of S(V ∗ ). Every simple
GL2 (F )-module is contained in one of the spaces S r (V ∗ ) for some r ∈ {1, 2, . . . }. Let M2 (F )
denote the space of all 2 × 2 matrices with entries from F . Then we can identify the tensor
product V1 := S 1 (V ∗ ) ⊗ S 1 (V ) with the space of all homogeneous polynomial functions of
degree 1 on M2 (F ). The primary concern of our paper is the image, denoted by C, of the
following (F -linear) evaluation map:
ev1 : V1 −→ F |GL2 (F )|
f 7−→ (f (A)|A ∈ GL2 (F )),
1
where we use a fixed total order on the elements of the group GL2 (F ). After a straightforward
computation, we see that
|V1 | = 24 and |GL2 (F )| = 6.
Also, it is easy to check that
(1, 1, 1, 1, 0, 0), (1, 0, 0, 0, 1, 0), (0, 1, 0, 0, 0, 1), (0, 0, 1, 0, 1, 0),
(0, 0, 0, 1, 0, 1), (0, 0, 1, 1, 1, 1), (0, 1, 1, 0, 1, 1), (1, 0, 0, 1, 1, 1),
C= .
(1, 1, 0, 0, 1, 1), (1, 0, 1, 0, 0, 0), (0, 1, 0, 1, 0, 0), (1, 1, 1, 0, 0, 1),
(1, 1, 0, 1, 1, 0), (1, 0, 1, 1, 0, 1), (0, 1, 1, 1, 1, 0), (0, 0, 0, 0, 0, 0)
By inspection we see that the minimum Hamming weight that is achieved by a nonzero
vector in C is 2. Therefore, the parameters of our code are given by [6, 4, 2]2 . A code with
these parameters is a subcode of the dual of the repetition code of length 6 over F .
Now, let V denote the defining representation of GLn (F ), where n ≥ 2, and F is a finite
field with q elements. Then S 1 (V ∗ ) ⊗ S 1 (V ) can be identified with the space of all linear
functionals on the space of all n × n matrices with entries from F . We denote by C the
algebraic geometry code that is obtained by evaluating the elements of S 1 (V ∗ ) ⊗ S 1 (V ) on
the points of GLn (F ). Finally, we define, for a positive integer r, the q-analog of r is defined
n −1
by [r]q := 1 + q + · · · + q n−1 = qq−1 . As a convention, we set [0]q := 1. The factorial analog
is now given by [r]q ! := [r]q [r − 1]q · · · [2]q [1]q . In this notation, the first main theorem of our
article is the following.
Theorem 1. Let F = Fq . Let C denote the evaluation code that is obtained by evaluating
all homogeneous linear polynomials of the vector space Mn (F ) on the variety GLn (F ). Then
the parameters of C are given by
1. n = q ( 2 ) (q − 1)n [n]q !,
n
2. k = n2 ,
Corollary 1. Let C be the code that is defined by evaluating linear forms on GL2 (F ), where
F = Fq . Then the length (denoted by n), the dimension (denoted by k), and the minimum
distance (denoted by d) of C are given by the following formulas
1. n = q 4 − q 3 − q 2 + q,
2. k = 4,
3. d = q 4 − 2q 3 + q.
2
To prove Theorem 1, hyperplane sections of GLn (F ) with the greatest number of F -
rational points are identified. Another goal of our paper is to show that the hyperplane
sections of GLn (F ) with the fewest F -rational points are derived from double cosets of Borel
subgroups.
Let B denote the Borel subgroup of all upper triangular matrices in GLn (F ). The
opposite Borel subgroup to B is denoted by B − . In other words, B − is the Borel subgroup
of all lower triangular matrices in GLn (F ). It is well-known that there are only finitely many
(B − , B)-double cosets in GLn (F ). Chevalley’s big cell theorem [1, Part II, §1.9] tells us that
the geometric points of the double coset B − B is dense (in Zariski topology) in GLn (F̄ ),
where F̄ stands for an algebraic closure of F . The second main result of our paper identifies
the complement of this set as a hyperplane section.
Theorem 2. Let Y denote the (B − , B)-double coset B − B in GLn (F ). Then GLn (F ) \ Y is
a hyperplane section with the fewest F -rational points in GLn (F ).
The structure of our note is as follows. In the next section, we setup our notation. In
Section 3 we observe the fact that the number of F -rational points of a general hyperplane
section of GLn (F ), where F is a finite field is equal to the number of F -rational points of
a hyperplane section of GLn (F ) that passes through the origin of Mn (F ). In Section 4, we
prove our first main result. In Section 5, we prove our second main result. We conclude our
paper by discussing the defects of our GLn (F )-codes.
Since it is useful for our purposes, following [4], we review a well-known reformulation
of the q-ary linear codes. An [n, k, d]q system is a finite sequence P = (v1 , . . . , vn ), where
v1 , . . . , vn are not necessarily distinct vectors from an F -vector space V such that
• there is no hyperplane in V that contains all of the vectors v1 , . . . , vn (hence, n ≥
dimF V );
3
• k = dimF V ;
• d = n − maxH |P ∩ H|, where the maximum is taken over all hyperplanes H ⊆ V , and
the points are counted with multiplicities. (It is assumed that d ≥ 1.)
By [4, Proposition 1.1.4], there is a one-to-one correspondence ϕ between the set of classes
[n, k, d]q systems and the set of linear [n, k, d]q codes. Under this correspondence, ϕ(P) is
the code obtained by evaluating the linear functionals on V at the entries of P.
3 Hyperplane Sections
Let F be a field. Let n ∈ Z+ . Hereafter, unless otherwise stated, we assume that n ≥ 2. A
hyperplane H ⊂ Mn (F ) is the shift by an element C ∈ Mn (F ) of the kernel of a linear map
B ∨ : Mn (F ) → F . This means that there exists a matrix B ∈ Mn (F ), and a scalar c ∈ F ,
such that
Note that, unless the scalar c is zero, H does not contain the zero of Mn (F ). The purpose
of this section is to analyze the cardinality of the intersection,
H ∩ GLn (F ).
Proof. We begin with noting that the trace form is invariant under conjugation by the
elements of GLn (F ):
H = {A ∈ Mn (F ) : tr(EAB ⊤ E −1 ) = c}.
Let Mnr (F ) denote the set of all rank r matrices from Mn (F ). We proceed with the assump-
tion that B ∈ Mnr (F ). Then B ⊤ is an element of Mnr (F ) also.
The group GLn (F ) × GLn (F ) acts transitively on Mnr (F ) by the left-right multiplication,
4
In other words, there exists (D, E) ∈ GLn (F ) × GLn (F ) such that
⊤ −1 1r 0
DB E = ,
0 0
where 1r is the r × r identity matrix. Therefore, for H as in (3.1), there exist two invertible
matrices E and D such that every element A ∈ H satisfies
−1 1r 0
tr EAD = c.
0 0
Proof. We follow the notation of the (proof of) Lemma 1. Since the map A 7→ EAD −1 ,
A ∈ GLn (F ), is injective, we see that
It remains to show that we can replace the scalar c with 0. To this end, we will apply an
affine transformation to GLn (F ).
Let C denote the matrix −cer . The map sC : Mn (F ) → Mn (F ), Z 7→ C + Z is an affine
transformation of Mn (F ). It is easy to check that
|{X ∈ GLn (F ) : tr (Xer ) = c}| = |{X ∈ GLn (F ) : tr (sC (X)er ) = 0}| . (3.4)
5
Let
gn (F ) : tr (sC (X)er ) = 0}
ϕ : {X ∈ GLn (F ) : tr (sC (X)er ) = 0} −→ {sC (X) ∈ GL
X 7−→ sC (X).
Since sC (X) is a bijection, we see that ϕ is a bijection. Hence, in light of (3.4), we showed
that
gn (F ) : tr (sC (X)er ) = 0} .
|{X ∈ GLn (F ) : tr (Xer ) = c}| = {sC (X) ∈ GL (3.5)
4 Main Theorem
We are now ready to prove our main theorem. Let us recall its statement for convenience.
Let F = Fq . Let C denote the evaluation code that is obtained by evaluating all ho-
mogeneous linear polynomials of the vector space Mn (F ) on the variety GLn (F ). Then the
parameters of C are given by
1. n = q ( 2 ) (q − 1)n [n]q !,
n
2. k = n2 ,
6
3. d = q ( 2 )−1 (q − 1)n−1 ((q − 1)2 [n]q ! − [n − 2]q !).
n
Proof. Recall that γ(n, q) denotes the cardinality, |GLn (F )| = q ( 2 ) (q − 1)n [n]q !. This gives
n
us the length n. The dimension of our code is given by the dimension of the space of
homogeneous linear forms on Mn (F ). Clearly, this number is given by n2 . Finally, to find
the minimum distance, we use Proposition 1 and a result of R. Stanley. In his book [3,
Chapter 1, Exercise 196 (a) and (b)], Stanley observed the following fact.
For k ∈ {0, 1, . . . , n}, let fk (n) denote the number of matrices A = (aij ) ∈ GLn (F )
satisfying
a11 + a22 + · · · + akk = 0.
In other words, fk (n) is the number of F -rational points of the intersection of GLn (F ) with
the hyperplane H ⊂ Mn (F ) defined by the equation a11 +a22 +· · ·+akk = 0. Then, Stanley’s
formula is given by
1 k 1
k(2n−k−1)
fk (n) = γ(n, q) + (−1) (q − 1)q 2 γ(n − k, q) . (4.1)
q
Since q and n are fixed, we want to find k0 such that
We claim that k0 = 2.
q
We split our proof into two parts. First we will show that q−1 (f2 (n) − f1 (n)) ≥ 0 for
q
every n and q from {2, 3, . . . }. Then we will show that q−1 (f2 (n) − fj (n)) ≥ 0 for every
{n, q} ⊂ {2, 3, . . . }, and j ∈ {3, . . . , n}.
In the former case, we have
q
(f2 (n) − f1 (n)) = q 2n−3 γ(n − 2, q) + q n−1 γ(n − 1, q). (4.3)
q−1
Evidently, the right hand side of (4.3) is nonnegative.
In the latter case, we have
q 1
(f2 (n) − fj (n)) = q 2n−3 γ(n − 2, q) + (−1)j q 2 j(2n−j−1) γ(n − j, q). (4.4)
q−1
In (4.4), similarly to the previous case, if j is even, then there is nothing to do. We proceed
with the assumption that j is odd. Let j = 2m + 1 for some m ∈ {1, . . . , ⌊ n2 ⌋}. Then we
have
q
(f2 (n) − fj (n)) = q 2n−3 γ(n − 2, q) − q (2m+1)(n−m−1) γ(n − 2m − 1, q). (4.5)
q−1
7
Since we have the inequalities, (q − 1)n−2m−1 ≤ (q − 1)n−2 and [n − 2m − 1]q ! ≤ [n − 2]q !, our
task is reduced to a comparison of the terms q 2n−3+( 2 ) and q (2m+1)(n−m−1)+( 2 ) . But it
n−2 n−2m−1
It follows that the inequality in (4.5), hence the inequality in (4.4) always hold true. This
finishes the proof of our claim that k0 = 2. Let H0 ⊂ Mn (F ) denote the hyperplane defined
by {(aij ) ∈ Mn (F ) : a11 + a22 = 0}. Hence, the number of F -rational points of H0 is given
by f2 (n). In light of this observation and Proposition 1, we see that
d = n − max |GLn (F ) ∩ H|
H
= n − f2 (n)
1 1
= 1− γ(n, q) − (q − 1)q 2n−3 γ(n − 2, q)
q q
= q ( 2 )−1 (q − 1)n+1 [n]q ! − q ( 2 )−1 (q − 1)n−1 [n − 2]q !.
n n
5 B − × B-stable Boundary
The goal of this section is to prove that the hyperplane section of GLn (F ) with the minimal
number of F -rational points is given by a B − × B stable divisor of GLn (F ), where B stands
for a Borel subgroup of GLn (F ). To this end, we begin with setting up our algebraic group
theoretic notation. The basic reference for this material is Jantzen’s book, especially [1, Part
II, §1.9].
Let F be a field. Let G be a (split) connected reductive algebraic group defined over F .
We fix a (split) maximal torus T in G, and we fix a Borel subgroup B of G such that T ⊆ B.
The Weyl group of (G, T ) is denoted by W . It is given by NG (T )/T , where NG (T ) is the
normalizer of T in G. For w ∈ W , we denote by ẇ a representative of w in NG (T ). The
Bruhat decomposition of G is the following partitioning of G into double cosets of B:
G
G= B ẇB. (5.1)
w∈W
8
Then the Weyl group W becomes a partially ordered set with respect to inclusion order on
the Zariski closures of the double cosets:
This order is called the Bruhat-Chevalley order on W . It is a graded poset with the rank
function,
ℓ : W −→ Z
w 7−→ dim B ẇB − dim B.
The value of an element w ∈ W under ℓ is called the length of w. There is a unique maximal
length element w0 ∈ W . It has the following pleasant properties:
• B ẇ0 B = G,
• w02 = id,
• ẇ0 B ẇ0 ∩ B = T .
The Borel subgroup ẇ0 B ẇ0 , denoted by B − , is called the opposite of B (relative to W ). It
is easy to see that Bruhat decomposition is equivalent to the following decomposition:
G
G= B − ẇB. (5.2)
w∈W
We now focus on the most important linear algebraic group, namely, G := GLn (F ). The
group of all invertible upper triangular in G is a Borel subgroup. By abuse of notation we
denote it by B. The group of invertible diagonal matrices in G will be denoted by T . Clearly,
T is a maximal torus contained in B. The Weyl group of (G, T ) is naturally isomorphic to
the symmetric group Sn . The opposite Borel subgroup of B, which is denoted by B − ,
is theFBorel subgroup of lower triangular matrices in G. Then the Bruhat decomposition
G = w∈W B − wB is nothing but the LPU-decomposition of invertible matrices, where L
stands for the lower triangular matrices, P stands for the permutations, and U stands for the
upper triangular matrices. Note that we deliberately omitted the dots (on top of w) from
our notation, since the symmetric group Sn is actually a subgroup of G.
Every Weyl group comes with a (not necessarily unique) system of Coxeter generators.
In the case of the symmetric group Sn a Coxeter generating system is given by the set of
simple transpositions
9
the minimal number of simple transpositions that is need to express w as their products. It
follows from Matsumoto’s exchange condition [2] that, for every s ∈ S, we have
ℓ(w0 s) = ℓ(w0 ) − 1 and w0 s ≤ w0 .
In other words, the set Bw0 sB (s ∈ S) is a one codimensional subvariety of GLn (F ). We
note in passing that
Bw0 sB = w0 w0 Bw0 sB = w0 B − sB. (5.4)
Since w0 acts transitively on G, we see that dim Bw0 sB = dim B − sB for every s ∈ S.
Lemma 3. Let H0 denote the hyperplane H0 = {(aij )i,j=1,...,n ∈ Mn (F ) : a11 = 0}. Then
we have
G
H0 ∩ GLn (F ) = B − sB,
s∈S
10
which is an element of D also. In conclusion, we proved that
G
B − sB ⊆ D.
s∈S
It follows from (5.4) that B − sB is a B − × B stable prime divisor in GLn (F ). In fact, thanks
to Bruhat decomposition, we know that every B − × B stable prime divisor in GLn (F ) is
such an orbit closure for some s ∈ S. Since D is B − × B stable, and since it contains every
B − sB, where s ∈ S, we see that D is given by their union. This finishes the proof of our
assertion.
We are now ready to prove the second main result of our article.
Proof of Theorem 2. Recall from the proof of Theorem 1 that the number of F -rational
points of the intersection of GLn (F ) with the hyperplane H = {A = (aij ) ∈ Mn (F ) :
a11 + · · · akk = 0} is given by
1 1
fk (n) = γ(n, q) + (−1)k (q − 1)q 2 k(2n−k−1) γ(n − k, q) .
q
We view fk (n) as a Z-valued function. It is easy to check (along the lines of the proof of
Theorem 1) that the number mink∈[n] fk (n) is achieved by f1 (n). Lemma 3 shows that f1 (n)
F
is the number of F -rational points on the B − ×B-stable divisor s∈S B − sB. But this divisor
is precisely the complement of the “Chevalley’s big cell” B − B. This finishes the proof of our
theorem.
6 Final Remarks
Every linear [n, k, d]q code satisfies the Singleton bound, that is the inequality d ≤ n − k + 1.
Likewise, every linear [n, k, d]q code satisfies the Griesmer bound, P that isthe inequality
Pk d k d
i=0 ⌈ q i ⌉ ≤ n. We call the differences (n − k + 1) − d and n − i=0 ⌈ q i ⌉ the Singleton
defect and the Griesmer defect, respectively. An MDS code is a linear code whose Singleton
defect is 0. Such codes are among the most desirable codes. The GLn (F )-codes are far from
being an MDS code. For n = 2, we observe the following fact whose proof is straightforward.
Proposition 2. Let F = Fq . Let C denote the evaluation code that is obtained by evaluating
all homogeneous linear polynomials of the vector space M2 (F ) on the variety GL2 (F ). Then
the following statements hold:
1. The Singleton defect of C is given by q 3 − q 2 − 3.
2. The Griesmer defect of C is given by q − 1.
Over the binary field, we have the identification GLn (F2 ) = SLn (F2 ). Therefore, all of
our results automatically hold true for SLn (F2 ). For a non-binary field F , the parameters of
the SLn (F )-codes can be found in a similar way. In an upcoming manuscript we will present
similar calculations for the other classical groups.
11
Acknowledgement
This research was partially supported by a grant from the Louisiana Board of Regents
(Contract no. LEQSF(2021-22)-ENH-DE-26).
References
[1] Jens Carsten Jantzen. Representations of algebraic groups, volume 107 of Mathemati-
cal Surveys and Monographs. American Mathematical Society, Providence, RI, second
edition, 2003.
[4] Michael Tsfasman, Serge Vlăduţ, and Dmitry Nogin. Algebraic geometric codes: basic
notions, volume 139 of Mathematical Surveys and Monographs. American Mathematical
Society, Providence, RI, 2007.
12