Vector-Circulant Matrices over Finite Fields and Related

Codes
Somphong Jitman
arXiv:1408.2059v1 [math.RA] 9 Aug 2014

Abstract
A vector-circulant matrix is a natural generalization of the classical circulant matrix and
has applications in constructing additive codes. This article formulates the concept of a
vector-circulant matrix over finite fields and gives an algebraic characterization for this kind
of matrix. Finally, a construction of additive codes with vector-circulant based over F4 is
given together with some examples of good half-rate additive codes.

1 Introduction
Let q be a power of a prime number and n be a positive integer. Denote by Fq the finite field
of order q and Mn (Fq ) the Fq -algebra of all n × n matrices whose entries are in Fq . Given
α ∈ Fq \ {0}, a matrix A ∈ Mn (Fq ) is said to be α-twistulant [3] if

 
a0 a1 ... ak−2 an−1

 αan−1 a0 ... an−3 an−2 


A= αan−2 αan−1 ... an−4 an−3 
.
 .. .. .. .. .. 
 . . . . . 
αa1 αa2 ... αan−1 a0

Such a matrix is called circulant [resp., negacirculant ] matrix when α = 1 [resp., α = −1]. The
set of all n × n circulant [resp., α-twistulant, negacirculant] matrices over Fq is isomorphic to
Fq [x]/hxn − 1i [resp., Fq [x]/hxn − αi, Fq [x]/hxn + 1i] as commutative algebras [3].
Circulant matrices over finite fields and their well-known generalizations in the notion of
twistulant and negacirculant matrices have widely been applied in many branches of Mathematics.
Recently, they have been applied to construct circulant based additive codes [5] and double
circulant codes [4] with optimal and extremal parameters.
In Section 2, we generalize the concept of circulant matrix over the finite field Fq and call it a
vector-circulant matrix. The algebraic structure of the set of these matrices is investigated. This
generalization leads to a construction of vector-circulant based additive codes over F4 in Section
3. Examples are some optimal codes are also demonstrated here.

2 Vector-Circulant Matrices over Finite Fields

Given a vector λ = (λ0 , λ1 , . . . , λn−1 ) ∈ Fnq , let ρλ : Fnq → Fnq be defined by

ρλ ((v0 , v1 , . . . , vn−1 )) = (0, v0 , v1 , . . . , vn−2 ) + vn−1 λ

= (vn−1 λ0 , v0 + vn−1 λ1 , . . . , vn−2 + vn−1 λn−1 ). (2.1)
S. Jitman is with the Department of Mathematics, Faculty of Science, Silpakorn University, Nakhon Pathom
73000, Thailand (email: sjitman@gmail.com).

1
The map ρλ is called the λ-vector-cyclic shift on Fnq .
A matrix A ∈ Mn (Fq ) is said to be vector-circulant, or specifically, λ-vector-circulant if
 
a0 a1 · · · an−1
 ρλ (a0 a1 · · · an−1 ) 
 2

A =  ρλ (a0 a1 · · · an−1 )  =: cirλ (a0 , a1 , . . . , an−1 ).
 
 .. 
 . 
n−1
ρλ (a0 a1 · · · an−1 )

Clearly, a λ-vector-circulant matrix becomes the classical circulant [resp., α-twistualnt] matrix
when λ is the vector (1, 0, . . . , 0) [resp, (α, 0, . . . , 0) ].
Example 2.1. Consider the finite field F4 = {0, 1, α, α2 = 1 + α}. The matrices
 
1 α 0
 0 1 α  = cir(1,0,1) (1, α, 0)
α 0 α2

and
 
1 α 0 α
 α2 1 α α 
 2  = cir(α,0,0,1) (1, α, 0, α)
 α α 1 0 
0 α2 α 1

are 3 × 3 and 4 × 4 vector-circulant matrices, respectively. They are obviously not circulant.
From (2.1), it is easily verified that ρλ is an Fq -linear transformation corresponding to
 
0 1 0 ... 0
 0 0 1 ... 0 
 
Tλ =  ... .. .. .. ..  ,

 . . . . 
 0 0 0 ... 1 
λ0 λ1 λ2 . . . λn−1

i.e., ρλ (v) = vTλ , for all v ∈ Fnq . Consequently, for 1 ≤ i, ρiλ is an Fq -linear transformation
corresponding to Tλi . Hence,
n−1
X
cirλ (a0 , a1 , . . . , an−1 ) = ai cirλ (Ei+1 ), (2.2)
i=0

where Ei = (0, . . . , 0, |{z}

1 , 0, . . . , 0), for 1 ≤ i ≤ n.
ith
Observe that Tλ need not be invertible. For λ = (λ0 , λ1 , . . . , λn−1 ), the singularity of Tλ
depends on λ0 . By applying a suitable sequence of elementary row operations, Tλ is equivalent
to an n × n diagonal matrix diag(λ0 , 1, 1, . . . , 1). Then the next proposition follows.
Proposition 2.2. T(λ0 ,λ1 ,...,λn−1 ) is invertible if and only if λ0 6= 0.
The set of all n × n λ-vector-circulant matrices over Fq is denoted by Cirn,λ (Fq ). Consider
Mn (Fq ) as an algebra over Fq , our goal is to show that Cirn,λ (Fq ) is an abelian subalgebra of
Mn (Fq ). It follows directly from the linearity of ρλ that Cirn,λ (Fq ) is a subspace of the Fq -vector
space Mn (Fq ). Moreover, by application of (2.1), the set {cirλ (E1 ), cirλ (E2 ), . . . , cirλ (En )} can
be verified to be a basis of Cirn,λ (Fq ). To prove that Cirn,λ (Fq ) is a commutative ring, we need
the following lemma and corollary. For convenience, let Tλ0 denote the identity matrix In .

2
Lemma 2.3. i) Tλm = cirλ (ρm
λ (E1 )), for all integers 0 ≤ m.

ii) Tλi = cirλ (Ei+1 ), for all 0 ≤ i < n.

Proof. First, we prove i) by induction on m. By the definition, Tλ0 = In = cirλ (ρ0λ (E1 )). Clearly,
Tλ = cirλ (ρλ (E1 )). Assume that Tλk = cirλ (ρkλ (E1 )) for all positive integers k < m. Then

Tλk+1 = cirλ (ρkλ (E1 ))Tλ

 
ρkλ (E1 )
 ρk+1 (E1 ) 
 λ 
= ..  Tλ
 . 
ρk+n−1
λ (E1 )
 k+1 
ρλ (E1 )

 ρk+2
λ (E1 ) 

= .. 
 . 
ρk+n
λ (E1 )
= cirλ (ρk+1
λ (E1 )).

Hence i) is proved.
Note that, for all 1 ≤ i ≤ n, ρi−1
λ (E1 ) = Ei . Hence, ii) follows immediately from i).

The next corollary is a direct consequence of Lemma 2.3.

Corollary 2.4. Tλm ∈ Cirn,λ (Fq ), for all 0 ≤ m.
Theorem 2.5. Cirn,λ (Fq ) is a commutative subring of Mn (Fq ).
Proof. Since (Cirn,λ (Fq ), +) is an additive subgroup of Mn (Fq ) containing In = Tλ0 , it is sufficient
to show that Cirn,λ (Fq ) is closed under the usual multiplication of matrices. Let (a0 , a1 , . . . , an−1 ),
(b0 , b1 , . . . , bn−1 ) ∈ Fnq . Then

n−1
! n−1 
X X
cirλ (a0 , a1 , . . . , an−1 )cirλ (b0 , b1 , . . . , bn−1 = ai cirλ (Ei+1 )  bj cirλ (Ej+1 ) ,
i=0 j=0

by (2.2),
n−1
X n−1
X
= ai bj cirλ (Ei+1 )cirλ (Ej+1 )
i=0 j=0
n−1
X n−1
X
= ai b j T i T j , by Lemma 2.3,
i=0 j=0
n−1
X n−1
X
= ai bj T i+j (2.3)
i=0 j=0
∈ Cirn,λ (Fq ), by Corollary 2.4.

From (2.3), the commutativity is obvious.

Corollary 2.6. Cirn,λ (Fq ) is a commutative algebra.
Proof. It follows from the fact that Cirn,λ (Fq ) is an Fq -vector space and Theorem 2.5.
For λ = (λ0 , λ1 , . . . , λn−1 ), let λ(x) = λ0 + λ1 x + · · · + λn−1 xn−1 be the corresponding
polynomial representation of λ.
The next theorem is an algebraic characterization of Cirn,λ (Fq ).

3
Theorem 2.7. Cirn,λ (Fq ) is isomorphic to Fq [x]/hxn − λ(x)i as commutative algebras.
Proof. Defined ϕ : Cirn,λ (Fq ) → Fq [x]/hxn − λ(x)i by
n−1
X
cirλ (a0 , a1 , . . . , an−1 ) 7→ ai xi + hxn − λ(x)i.
i=0

It is easily seen that ϕ is an additive group isomorphism. Let (a0 , a1 , . . . , an−1 ), (b0 , b1 , . . . , bn−1 ) ∈
Fnq and α ∈ Fq . Then
ϕ(αcirλ (a0 , a1 , . . . , an−1 )) = ϕ(cirλ (αa0 , αa1 , . . . , αan−1 ))
n−1
X
= αai xi + hxn − λ(x)i
i=0
n−1
!
X
i n
=α ai x + hx − λ(x)i
i=0
= αϕ(cirλ (a0 , a1 , . . . , an−1 )). (2.4)
By (2.3) and Lemma 2.3, we then have
ϕ(cirλ (a0 , a1 , . . . , an−1 )cirλ (b0 , b1 , . . . , bn−1 ))
n−1
X n−1X
= ϕ( ai bj T i+j ), by (2.3),
i=0 j=0

X n−1
n−1 X
= ϕ( ai bj cirλ (ρi+j
λ (E1 ))), by Lemma 2.3,
i=0 j=0
n−1
X n−1
X
= ai bj ϕ(cirλ (ρi+j
λ (E1 ))), by (2.4),
i=0 j=0
n−1
X n−1
X
= ai bj xi+j + hxn − λ(x)i
i=0 j=0

n−1
! n−1 
X X
= ai xi + hxn − λ(x)i  aj xj + hxn − λ(x)i
i=0 j=0

= ϕ(cirλ (a0 , a1 , . . . , an−1 ))ϕ(cirλ (b0 , b1 , . . . , bn−1 )).

This completes the proof.

3 Vector-Circulant Based Additive Codes over F4

In this section, we restrict our study to the finite field of 4 elements F4 = {0, 1, α, α2 = 1 + α}.
A code of length n over F4 is defined to be a non-empty subset of Fn4 . A code C is said to be
additive if it is an additive subgroup of the additive group (Fn4 , +). Throughout, every code is
assume to be additive. It is know [5] that C contains 2k codewords for some 0 ≤ k ≤ 2n, and
can be defined by a k × n generator matrix, with entries from F4 , whose rows span C additively.
We regard an additive code of length n over F4 containing 2k codewords as an (n, 2k ) code. The
Hamming weight of v ∈ Fn4 , denoted wt(v), is defined to be the number of nonzero components
of v. The Hamming distance between u 6= v ∈ Fn4 is defined as wt(u − v). The minimum distance
of the code C, denoted by d(C), is the minimal Hamming distance between any two distinct
codewords of C. As C is additive, the minimum distance equals the smallest nonzero weight of
any codewords in C. An (n, 2k ) code with minimum distance d is called an (n, 2k , d) code.

4
 We focus on a construction of additive codes for the particular case k = n, or half-rate codes,
or equivalently, (n, 2n ) codes. It follows from the Singleton bound [2] that any half-rate additive
code over F4 must satisfy jnk
d≤ + 1.
2
An (n, 2n ) code C is said to be extremal if  it attains the equality in the Singleton bound, and
near-extremal if it has minimum distance n2 .
Given λ ∈ Fn4 , a λ-vector-circulant based additive code is defined to be an additive code
generated by a λ-circulant generator matrix of the following form:
 
a0 a1 · · · an−1
 ρλ (a0 a1 · · · an−1 ) 
 
 ρ2λ (a0 a1 · · · an−1 ) 
  =: cirλ (a0 , a1 , . . . , an−1 )
 .. 
 . 
n−1
ρλ (a0 a1 · · · an−1 )

Such a code is called a circulant based additive code if λ = (1, 0, . . . , 0) and it is called a α-
twistulant based additive code if λ = (α, 0, . . . , 0).
An advantage of this construction is that there are typically much more additive codes than
circulant based or twistulant based additive codes [5].
We implement a procedure in the computer algebra system Magma [1] to construct vector-
circulant based additive codes. Based on this construction, we search for half-rate additive codes
with highest minimum distances of length up to 13. The result is shown in Table 1; codes of
length 2 to 7 are extremal and codes of length 8 to 13 are near-extremal.

References
[1] Bosma W, Cannon J J, Playoust C. The Magma algebra system I: the user language.
Journal of Symbolic Computation, 24: 235–265 (1997)
[2] Danielsen L E, Parker M G. Directed graph representation of half–rate additive codes over
GF(4). http:// arxiv.org/abs/0902.3883v3 (preprint)
[3] Davis P J. Circulant Matrices. Chelesa publishing, New York, second edition, 1994
[4] Grassl M, Gulliver A. On circulant self-dual codes over small fields. Designs, Codes and
Cryptography, 52: 57–81 (2009)
[5] Gulliver T A, Kim J-L. Circulant based extremal additive self-dual codes over GF (4). IEEE
Transaction on Information Theory, 50: 359–366 (2004)

5
Table 1: Vector-circulant based (n, 2n ) codes C over F4 = {0, 1, α, α2 = 1 + α} generated by
cirλ (v) with highest minimum distances
n λ v d(C)
2 (1, 1) (α, 1) 2
3 (1, 0, α) (α, 1, 1) 2
4 (1, 0, 0, 1) (1, α, 1, 1) 3
5 (1, 0, 0, 0, α) (1, 0, α, 1, 1) 3
6 (1, 0, 0, 0, 0, 0) (α, α2 , α, 1, 1, 1) 4
7 (1, 0, 1, 0, 0, 0, 0) (0, 1, α, 1, 1, 1, 1) 4
8 (1, 0, 0, 0, 0, 0, 0, α) (0, α, α2 , α2 , 1, 1, 1, 1) 4
9 (1, 0, 0, 0, 0, 0, 0, 0, 1) (a2 , α, 1, 1, 1, 1, 1, 1, 1) 4
10 (1, 0, 0, 0, 0, 0, 0, 0, 0, 0) (0, α, α, 1, α, 1, 1, 1, 1, 1) 5
11 (1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0) (0, α, α2 , α, 1, 1, 1, 1, 1, 1, 1) 5
12 (1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) (0, 1, α2 , α2 , 1, α, 1, 1, 1, 1, 1, 1) 6
13 (1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) (0, α, α2 , 1, 1, α, 1, 1, 1, 1, 1, 1, 1) 6