Des. Codes Cryptogr.

(2014) 72:559–580
DOI 10.1007/s10623-012-9787-y

Codes over an infinite family of rings with a Gray map

Yasemin Cengellenmis · Abdullah Dertli ·

S. T. Dougherty

Received: 22 January 2012 / Revised: 20 November 2012 / Accepted: 12 December 2012 /

Published online: 1 January 2013
© Springer Science+Business Media New York 2012

Abstract Codes over an infinite family of rings which are an extension of the binary field
are defined. Two Gray maps to the binary field are attached and are shown to be conjugate.
Euclidean and Hermitian self-dual codes are related to binary self-dual and formally self-dual
codes, giving a construction of formally self-dual codes from a collection of arbitrary binary
codes. We relate codes over these rings to complex lattices. A Singleton bound is proved
for these codes with respect to the Lee weight. The structure of cyclic codes and their Gray
image is studied. Infinite families of self-dual and formally self-dual quasi-cyclic codes are
constructed from these codes.

Keywords Binary codes · Gray maps · Cyclic codes · Quasi-cyclic codes

Mathematics Subject Classification (2000) 94A15 · 13A99

1 Introduction

The rings of order 4 are of particular interest in terms of algebraic coding theory. This is
because they have natural Gray maps to the binary field which makes them of particular
interest in terms of constructing interesting binary codes. The four rings of order 4 are the
finite field of order 4 denoted F4 , the ring Z4 and the rings F2 +u F2 , with u 2 = 0 and F2 +v F2 ,
with v 2 = v. Codes over Z4 have been widely studied, in fact it was the realization that codes
over Z4 , together with their Gray map, could be used to understand certain binary codes that

Communicated by J. D. Key.

Y. Cengellenmis · A. Dertli
Department of Mathematics, Faculty of Science,
Trakya University, 22030 Edirne, Turkey

S. T. Dougherty (B)
University of Scranton, Scranton, PA18510, USA
e-mail: prof.steven.dougherty@gmail.com

123
560 Y. Cengellenmis et al.

began the study of codes over rings. Codes over Z4 have been generalized to codes over
Z2k together with an associated Gray map in [4]. Codes over F4 have been widely studied,
especially additive codes over F4 , which are related to quantum error-correction. The field
F4 can be naturally generalized to F2s . This field has a natural Gray map formed by viewing
F2s as a vector space over F2 . Codes over the ring F2 + u F2 were introduced in [7] and these
codes were generalized to the infinite family of rings Rk together with their associated Gray
maps in [11]. The Gray map was first introduced in [13] as a map from Zn4 → F2n 2 . This map
was the model for the definition of the Gray map given in [7]. It was generalized in [11] and
cyclic codes over this ring were defined in [12].
We make the standard assumption that by ring we mean a ring with unity. If A is a ring we
say that C is a code over A of length n if C is a subset of An , moreover if it is a submodule
we say that it is a linear code. The Hamming weight of a vector v is the number of non-
zero elements in v. The minimum weight, for a given weight, is the smallest of all non-zero
weights in the code.
In this paper, we generalize the ring F2 + v F2 to the infinite family of rings Ak . We define
Gray maps which map these rings to the Hamming space and we study codes over these rings
along with their binary images.

2 The ring Ak

We begin by defining an infinite collection of finite rings and codes over these rings. The
ambient space is equipped with canonical Gray maps which produce binary codes.
The rings are defined as follows. For integers k ≥ 1, let Ak = F2 [v1 , v2 , . . . , vk ]/vi2 =
vi , vi v j = v j vi , 1 ≤ i, j ≤ k. This ring
 is a finite commutative ring.
Let B ⊆ {1, 2, . . . , k} with v B = i∈B vi . In particular, v∅ = 1. It follows that each
element of Ak is of the form B∈Pk α B v B where α B ∈ F2 , and Pk is the power set of the
set {1, 2, 3, . . . , k}.
For A, B ⊆ {1, 2, . . . , k} we have that v A v B = v A∪B which gives that
 
   
αB vB · βC vC = α B βC v D .
B∈Pk C∈Pk D∈Pk B∪C=D
k
Lemma 2.1 The ring Ak has characteristic 2 and cardinality 22 .
Proof The factthat the characteristic is 2 follows immediately since the coefficients of v B
are in F2 . Let α B v B , with α B ∈ F2 . For each α B there are two choices and there are 2k
k
subsets of {1, 2, . . . , k}. This gives that the ring Ak has cardinality 22 . 

The ring Ak is not a local ring. For example, if k = 1 then v1  = 1 + v1  and both ideals
are maximal.
Lemma 2.2 The only unit in the ring Ak is 1.
 
Proof If a is any element in Ak then a =
 α B v B . a 2 = ( α
It follows that B vB ) =
2

(α B v B ) since the ring has characteristic 2. Then (α B v B ) =

2 2 αB vB =
2 2 α B v 2B ,
since every element is its own square in the binary field and vi = vi for all i. This gives that
2

a 2 = a for all α ∈ Ak . Then a(a + 1) = a + a = 0 for all a ∈ Ak . This gives that every
element is a zero divisor unless a + 1 = 0, that is a = 1. Hence 1 is the only unit. 

We note that no proper ideal has both vi and 1 + vi in it since then it would contain
vi + vi + 1 = 1 and the ideal would be Ak .

123
Codes over an infinite family of rings with a Gray map 561

Theorem 2.3 The ideal w1 , w2 , . . . , wk , where wi ∈ {vi , 1 + vi }, is a maximal ideal of

cardinality 22 −1 .
k


Proof The ideal v1 , v2 , . . . , vk  consists of all elements of the form α B v B with α∅ = 0,
so the cardinality is precisely half the cardinality of the ring. The other maximal ideals are
isomorphic to this ideal. Namely, given the ideal a1 , a2 , . . . , ak  simply map ai to vi . This
produces an isomorphism from the ideal a1 , a2 , . . . , ak  to the ideal v1 , v2 , . . . , vk . Since
they are isomorphic they have the same cardinalities. Since the ideal is a subgroup of the
additive group, the cardinality of an ideal must divide the cardinality of the ring. This ideal
consists of precisely half the elements of the ring so it must be maximal. 

Denote these maximal ideals by mi .
Lemma 2.4 Let mi be a maximal ideal as described in Theorem 2.3, then there are 2k such
ideals and mie = mi for all i and e ≥ 1. Hence its index of stability is 1. Moreover the direct
sum of any two of these ideals is Ak .
Proof For each i there are two choices, namely vi or 1 + vi . Hence there are 2k distinct
ideals. We know vi2 = vi and (1 + vi )2 = (1 + vi ), which gives that mie = mi for e ≥ 2. If
the ideals m j and m j  are distinct then for some i, vi ∈ m j and 1 + vi ∈ m j  . Then 1 is in
the direct sum and then the sum is Ak . 

k
Theorem 2.5 The ring Ak is isomorphic via the Chinese Remainder Theorem to F22 . Con-
sequently, the ring Ak is a principal ideal ring.
Proof Together with Theorem 2.3 and Lemma 2.4 we can apply the Chinese Remainder
2k
Theorem. We have that |Ak /mi | = ||A k|
mi | = 22k −1 = 2. Since mi is a maximal ideal of the ring
2

Ak we have that Ak /mi ∼

= F2 for all i. This gives the result.
The ring Ak is a principal ideal ring since it is the product via the Chinese Remainder
Theorem of chain rings, specifically the finite field F2 . 

Since the ring is a principal ideal ring, the ring is a Frobenius ring which allows us to
use the foundational results in [19]. We can now show how each ideal can be written as a
principal ideal.
Theorem 2.6 Let I = α1 , α2 , . . . , αs  be an ideal in Ak , then I is a principal ideal gener-
ated by the element which is the sum of all non-empty products of the αi , that is

 
I = αi .
A ⊆ {1, 2, . . . , s} i∈A
A = ∅
 
Proof Let β = A⊆{1,2,...,s},A =∅ i∈A αi . Then we have
   
αi β = αi αj + αj. (1)
A ⊆ {1, 2, . . . , s} − {i} j∈A A ⊆ {1, 2, . . . , s} j∈A

A = ∅ i∈A
That is αi is multiplied by the product of those sets not containing i and those sets containing
i. When αi is multiplied by the product of those sets containing i the product is unchanged.
Then each product appears twice except for αi which appears only once, since the other
occurrence would be αi (1) which does not occur since we do not take the empty product.
This gives that αi β = αi . Hence I ⊆ β. The other direction is immediate since β ∈ I .  

123
562 Y. Cengellenmis et al.

It is easy then to write the maximal ideals given in Theorem 2.3 as principal ideals.
Example 1 Since the ring is a principal ideal ring there are 16 ideals in A2 . These ideals
can be viewed as codes of length 1 over A2 . Let I x = x. The maximal ideals given in
Theorem 2.3 are I1+v1 v2 = 1 + v1 , 1 + v2 , I1+v1 +v1 v2 = 1 + v1 , v2 , I1+v2 +v1 v2 =
v1 , 1 + v2 , Iv1 +v2 +v1 v2 = v1 , v2 .

3 The Gray maps

We shall describe two canonical Gray maps and show that they are conjugate.
Recall that there exists a Gray map defined on A1 , namely φ1 : A1 → F22 by φ(a +bv1 ) =
(a, a + b), see [8] for details. For A1 this is realized as
0 → 00
1 → 11
v → 01
1 + v → 10.

We can extend this map inductively as follows. Each element of Ak can be written as α +βvk ,
where α, β ∈ Ak−1 . Then for k ≥ 2, we let φk : Ak → A2k−1 by
φk (α + βvk ) = (α, α + β).
2k
Then define k : Ak → F2 by 1 (γ ) = φ1 (γ ), 2 (γ ) = φ1 (φ2 (γ )) and
k (α) = φ1 (φ2 (. . . (φk−2 (φk−1 (φk (γ )) . . . ).
It follows immediately that k (1) = 1, the all one vector. We see that α + βvk =
α(1 + vk ) + (α + 
β)vk which gives insight as to why the Gray map isdefined as it is.
For an element α B v B let the Lee weight be defined as wt H (k ( α B v B )) where wt H
is the binary Hamming weight.
Theorem 3.1 The Gray map k is a linear distance preserving map.
Proof The map is the composition of linear functions and is therefore linear. It is weight
preserving by definition. 

We can define another Gray map as follows.
k k
List the elements of Pk lexicographically and use these to order F22 . Let k : Ak → F22 .
Define

k (v B ) = wE (2)
E⊆B

where F ∈ Pk and w E is a vector whose coordinates correspond to the elements of Pk . By

(w E ) F denote the element in the vector that corresponds to the coordinate F ∈ Pk . Then we
have

1 E⊆F
(w E ) F = (3)
0 otherwise.
 
Then k ( α B v B ) = α B k (v B ).

123
Codes over an infinite family of rings with a Gray map 563

For example w∅ is the all one vector and wvi has weight 2k−1 , since precisely half the
subsets contain i. Hence k (vi ) has weight 2k−1 since it is their sum as binary vectors. It
follows then that w L (v B ) = 2(k−|B|) for each B ⊆ {1, 2, . . . , k}.
Let σ1 , σ2 , . . . , σk be swap maps as in [11].
Theorem 3.2 For k ≥ 1, we have k = σk (σk−1 (. . . (σ1 (k )))).
Proof We prove the result by induction on k. It is true for A1 = F2 + v1 F2 since 1 (1) =
(1, 1), 1 (v1 ) = (1, 0) and 1 (1) = (1, 1), 1 (v1 ) = (0, 1). The rest follows from the
linearity of the maps.
Assume the statement is true for k − 1. The subsets of {1, 2, . . . , k} are written lexi-
cographically as B1 , B2 , . . . , B2k−1 , B1 , B2 , . . . , B2 k−1 where Bi ⊆ {1, 2, . . . , k − 1} and
Bi = Bi ∪ {k}.
Case 1: The set is Bi , i.e. k ∈ Bi . Then Bi ⊆ {1, 2, . . . , k − 1} and k (v Bi ) =
(k−1 (v Bi ), k−1 (v Bi )). Taking v Bi = v Bi + 0vk , we see that k (v Bi ) = (k−1 (v Bi ), k−1
(v Bi )). Since the assertion is true for k − 1 we have k−1 (v Bi ) = σk−2 (. . . (σ1 (k−1 (v Bi ))).
This gives the result.
Case 2: The set is Bi , i.e. k ∈ Bi . We write Bi = Bi ∪{k} where Bi ⊆ {1, 2, . . . , k}. We see
that k (v Bi ) = (k−1 (v Bi ), 02k−1 ). Then it is easily seen that k (v Bi ) = (02k−1 , k−1 (v Bi )).
By induction we have the result. 

As an example, we see that
2 (1) = (1, 1, 1, 1), 2 (v1 ) = (0, 1, 0, 1), 2 (v2 ) = (0, 0, 1, 1), 2 (v1 v2 ) = (0, 0, 0, 1)
and
2 (1) = (1, 1, 1, 1), 2 (v1 ) = (1, 0, 1, 0), 2 (v2 ) = (1, 1, 0, 0), 2 (v1 v2 ) = (1, 0, 0, 0).
The map −1 k is a ring isomorphism. Let w H (c) denote the Hamming weight of the vector
c and w L (c) be the Lee weight of the vector c. Then we have the following lemmas.
Lemma 3.3 Let C = −1 −1
k (C 1 , C 2 , . . . , C 2k ) and c ∈ C, with c = k (c1 , c2 , . . . , c2k ), ci ∈
Ci , then
k

2  
w H (c) = ck (4)
i=1 A⊆ {1, 2, . . . , 2k } k∈A
|A| = i
where the product is the Hadamard product, and

w L (c) = wt H (ci ). (5)

Proof The first statement comes from the standard inclusion-exclusion argument. The second
follows from the definition of Lee weight. 

Lemma 3.4 Let d H (C) and d L (C) be the minimum Hamming and Lee weights of the code
C. If C = −1
k (C 1 , C 2 , . . . , C 2k ) then d H (C) = d L (C) = min{d H (Ci )}.

Proof If ci ∈ Ci then −1 k (0, 0, . . . , ci , . . . 0) ∈ C and has the same Hamming and Lee
weight of ci . Therefore min{d H (Ci )} ≥ d H (C) and d L (C). For any vector c ∈ C its Lee
weight and Hamming is greater than or equal to the weight of its image under the Gray map.



123
564 Y. Cengellenmis et al.

4 Codes over Ak

4.1 Inner products

We define two inner-products in Ank . First we have the Euclidean

  inner-product: [w, u] =
w j u j . The second is the Hermition inner-product: [w, u] H = w j u j , where vi = 1+vi .
Then we define C ⊥ = {w | [w, u] = 0, ∀u ∈ C} and C H = {w | [w, u] H = 0, ∀u ∈ C}.
The ring Ak is a principal ideal ring and thus a Frobenius ring. Therefore, by the results
in [19] we have |C||C ⊥ | = |C||C H | = |Ak |n .
Theorem 4.1 Let I α B v B be an ideal in Ak then I ⊥
αB vB
= I1+ α B v B and I
H
αB vB
=
I1+ α B v B .
  
Proof We have seen in Lemma 2.2 that ( α B v B )(1 + α B v B ) = 0. Let α = αB vB .
If α + 1 = δγ with γ α = 0 then α ⊕ α + 1 ⊂ γ  ⊕ α but α ⊕ α + 1 = Ak . This
gives that α + 1 = Iα⊥ . The second result follows immediately from the first. 

Theorem 4.2 If I is an ideal of Ak then I ⊥ = I.
⊥
 
Proof If I = I then I =  α B v B  and 1 + α B v B ∈ I as well. This means 1 =
1 + α B v B + α B v B is in I as well and I = Ak . Of course A⊥ k = {0} so no ideal is equal
to its dual. 

The previous theorem does not hold for the Hermitian inner-product. For example, in
A2 , IvH1 = Iv1 . This means that there are Hermitian self-dual codes of length 1 but there are
no Euclidean self-dual codes of length 1.
Example 2 For k = 2, it is a straightforward computation to see that there are precisely 4
Hermitian self-dual codes of length 1. Namely, the codes Iv1 , Iv2 , I1+v1 , I1+v2 are Hermitian
self-dual codes. Their Gray images are therefore formally self-dual codes of length 4 that are
not self-dual.

4.2 Minimal generating sets

Consider the maps

i : Ak → Ak /mi
defined by

i (a) = a (mod mi ).
Then define

: Ak → Ak /m1 × · · · × Ak /m2k
by
(a) = (
1 (a),
2 (a), . . . ,
2k (a)). The map
−1 is the ring isomorphism given by
the Chinese Remainder Theorem.
Following the notation of [5], we give the following definition.
Definition
 1 Let w1 , . . . , wk be vectors in Ank . Then w1 , . . . , wk are independent if
α j w j = 0 implies that α j w j = 0 for all j.
We can modify the definition of modular independent given in [5] since the ring is isomor-
phic to the direct product of fields, so modular independence in a local ring can be replaced
with linear independence over fields.

123
Codes over an infinite family of rings with a Gray map 565

Definition 2 The vectors w1 , . . . , wk in Ank are modular independent if

i (w1 ), . . . ,
i (wk )
are linear independent for some i.

It is shown in [5] that a generating set that is both independent and modular independent
is a minimal generating set.
Consider the code C generated by the vector (v1 , 1+v1 ) over A1 . The code has 4 elements
and C = {(0, 0), (v1 , 1+v1 ), (v1 , 0), (0, 1+v1 )}. The vector (v1 , 1+v1 ) is independent and
modular independent. First if α(v1 , 1 + v1 ) = (0, 0) then of course α(v1 , 1 + v1 ) = (0, 0).
That is, it satisfies the condition in Definition 1 trivially. Moreover,
1 = (0, 1) which is
linearly independent over F2 and
2 (v1 , 1 + v1 ) = (1, 0) which is linearly independent over
F2 .
In this case, it is not possible to make a minimal generating set of the form
⎛ ⎞
Ik 1 B1 B2 B3
⎝ 0 (1 + v1 )Ik2 B4 B5 ⎠ . (6)
0 0 v1 Ik3 B6

Obviously, to do so we would need 2 vectors. For example, a generator set of that form
would be v1 (1, 0), (1 + v1 )(0, 1). While these vectors are independent they are not modular
independent. For example, their image under
1 is (0, 0), (0, 1) which are not linearly inde-
pendent and their image under
2 is (1, 0), (0, 0) which are not linearly independent. What
we have shown is that it is not always possible to put a minimal generating set into a matrix
of the form in Matrix 6. In fact, there is not a general form for the minimal generating matrix.
Consider another code generated by the vector (v1 , v1 ). This code has cardinality 2
and consists of (0, 0) and (v1 , v1 ). This vector is again clearly independent. The vector

1 ((v1 , v1 )) is not linearly independent but

2 ((v1 , v1 )) is linearly independent so that the
vector is a minimal generating set. This minimal generating set of one vector generates a
smaller code than the previous example. This leads us to make the following definition.
Let w = (α1 , α2 , . . . , αn ) be a non-zero vector then α1 , α2 , . . . , αn  is an ideal in Ak .
Let I (w) = |α1 , α2 , . . . , αn |.

Theorem
s 4.3 Let C be a code with minimal generating set w1 , w2 , . . . , ws , then |C| =
i=1 I (wi ).

Proof The summations αi wi are distinct when each αi wi is not zero and there are I (w)
choices for αi that make it a non-zero vector. 


4.3 MacWilliams relations

The important weight enumerator for this ring is the Lee weight enumerator. We shall find the
MacWilliams relations for this weight enumerator. To do this we shall use the MacWilliams
relations for the complete weight enumerator.
k
The ring Ak is isomorphic to F22 . As such its generating character can be determined
by the product of the generating
 character for 
the binary field. It follows that the generating
character for Ak is χ1 ( α B v B ) = (−1)wt L ( α B v B ) by the Chinese Remainder Theorem
k
isomorphism with F22 .
Index the following matrices by the elements of Ak .
Let

Ta,b = χa (b) = χ(ab) (7)

123
566 Y. Cengellenmis et al.

and let
(TH )a, b = χa (b) = χ(ab). (8)
Define the complete weight enumerator of a code C by:
  n (c)
cweC (X a ) = xi a (9)
c∈C

where n a (c) is the numer of occurances of the element a in the vector c.

Theorem 4.4 (MacWilliams Relations) Let C be a linear code over Ak then
1
cweC ⊥ (X a ) = cweC (T · X a ) (10)
|C|
and
1
cweC H (X a ) = cweC (TH · X a ). (11)
|C|
Proof The result follows from the results in [19]. 

2k 2k
The matrix T is a 2 by 2 matrix indexed by the elements of Ak . Let S be the 2k + 1 by
2k + 1 matrix where the columns are indexed by 0, 1, 2, . . . , 2k and represents the possible
Lee weights and the rows correspond  to an element of Lee weight i. Denote by α ∼ β if
wt L (α) = wt L (β). Define Sα,β = γ ∼β Tα,γ .
Note
 k that the first row of the matrix S is simpy the binomial coefficients. In Sk there
2 k
are elements with Lee weight i because the map is linear to the space F22 . We need
i
to show that ifα ∼ α  then Sα = Sα  , that is for each column β, we must show that

γ ∼β Tα,γ = γ ∼β Tα  ,γ .
We have
  
Tα,γ = (−1)wt L (αγ ) = (−1)wt (k (α)wt (k (γ )) .
γ ∼β γ ∼β γ ∼β

We know that k (αβ) = k (α)k (β) so wt (k (α)) = wt (k (α  )) and the result
follows.  n wt (c )
Define the Lee weight enumerator to be L C (y0 , y1 , . . . , y2k ) = c∈C i=0 yi L i =
 n−wt L (c) y wt L (c) . Then by the results in [19] we have the following theorem.
c∈C x

Theorem 4.5 Let C be a linear code over Ak then

1
L C (y0 , y1 , . . . , y2k ) = L(S · (y0 , y1 , . . . , y2k )).
|C|
Since there is only one unit in Ak the symmetrized weight enumerator is the complete
weight enumerator and since −1 = 1 the symmetric weight enumerator is also the com-
plete weight enumerator.
 The Hamming weight enumerator is defined in the usual way,
namely: WC (x, y) = c∈C x n−wt (c) y wt (c) . Collapsing the matrices T and TH give the same
MacWilliams relations for the Hamming weight enumerator, that is
1
WC ⊥ (x, y) = WC H (x, y) = WC (x + (|Ak | − 1)y, x − y). (12)
|C|

123
Codes over an infinite family of rings with a Gray map 567

5 Self-dual codes

Self-dual codes are an interesting and important class of codes. See [15] for a complete
description of these codes. They have interesting connections to number theory, designs and
invariant theory, see [14]. We define two types of self-dual codes.
Definition 3 A code C is Euclidean self-dual if C = C ⊥ . It is Hermitian self-dual if C = C H .
In either case it is said to be self-orthogonal if it is contained in its respective orthogonal.
j−k
We define  j,k : A j → A2k for j > k to be
 j,k = φk+1 ◦ φk+2 ◦ · · · ◦ φ j .
That is we apply the maps until we reach Ak .

Theorem 5.1 Let C be a self-dual code (Hermitian or Euclidean) over A j then j,k (C) is
a self-orthogonal (Hermitian or Euclidean respectively) code over Ak for j > k.
Proof We shall show the proof for Hermitian self-dual codes. The Euclidean case follows
similarly.   
We know that C = C H . We need to show that j,k (C H ) ⊆ j,k (C) H . Let j,k (c) ∈
 
 j,k (C ), this implies c ∈ C H . Then we have ci ui = 0 for all u ∈C. Then
H
  that 
j,k ( ci ui ) = j,k (0) = 0 which implies j,k (c) j,k (ui ) = 0 for all j,k (u) ∈
    
j,k (C). Then
 we have [ j,k (c), j,k (u)] H = 0 which gives j,k (c) ∈ j,k (C) . This
H

gives that j,k (C) is a Hermitian self-orthogonal code. 


Theorem 5.2 Let c1 , c2 , . . . , cs generate a self-dual code over Ak . Then c1 , c2 , . . . , cs gen-
erate a Hermitian (Euclidean) self-dual code over Ai for i > k.
Proof We shall prove it for Hermitian self-dual codes, the case for Euclidean self-dual codes
follows similarly. Let Ci be the code generated by c1 , c2 , . . . , cs over Ai . We assume that Ck
is a self-dualcode so |Ck ||CkH | = |Ak |n where n is the length of the code. It is immediate
that if v = αi ci and w = βi ci are orthogonal over Ak then they are orthogonal over
A j with j > k as well. Hence we have that C j ⊆ C H j .
All that is required then is to show that the cardinality is correct, namely that |C j | =
|C H 
j |. If I is an ideal in Ai equal to α1 , . . . , αt , then I = α1 , . . . , αt  in Ai+1 has
cardinality |I |2 . We shall use Theorem 4.3 to get the result. Let Ii, = I (c ) over Ai . Then
k
assuming that Ci is Hermitian self-dual we have, |Ci ||Ci | = (s =1 Ii, )2 = |Ai |n = (22 )n .
k
Then |Ci+1 ||Ci+1 | = (s =1 Ii+1, )2 = (s =1 |Ii, |2 )2 = ((22 )2 )n = (|Ai+1 |)n . Then by
induction we have that Ci is Hermitian self-dual for i > k. 

In this case, the code C  is called a lifted code of the original code.
Corollary 5.3 If C is a Hermitian (Euclidean) self-dual code over Ak then there exists
self-dual codes C  over A j for all j > k with  j,k (C  ) = C.
Proof Simply apply the previous theorem. 


5.1 Euclidean self-dual codes

Theorem 5.4 A code C is a Euclidean self-dual code if and only if C=C RT (C1 , C2 , . . . , C2k )
and each Ci is a binary self-dual code.

123
568 Y. Cengellenmis et al.

Proof Theorem 2.5 gives that the ring Ak is isomorphic via the Chinese Remainder Theorem
k
to F22 . It follows then from the result in [10] that the isomorphic image of Euclidean self-dual
codes is self-dual. 

Corollary 5.5 Euclidean self-dual codes exist if and only if the length is congruent to 0
(mod 2).
Proof By Theorem 5.4, we know that a Euclidean self-dual code exists if and only if there
is a binary self-dual code of that length. It is well known that binary self-dual codes exist if
an only if the length is even. 

Theorem 5.6 The image under the Gray map of a Euclidean self-dual code is a binary
self-dual code.
Proof Follows from the fact that the map is linear. 

Definition 4 A Euclidean self-dual code is said to be Type II if and only if the Lee weights
of every element in the code are congruent to 0 (mod 4).
Theorem 5.7 If C is a Type II Euclidean self-dual code then C = C RT (C1 , C2 , . . . , C2k )
and each Ci is a binary Type II code.
Proof Theorem 5.4 gives that the codes Ci must be self-dual. The fact that they are Type II
follows from the definition of the Gray map. 


5.2 Hermitian self-dual codes

Theorem 5.8 The code Ivi is a Hermitian self-dual code of length 1.

Proof Follows from Theorem 4.1 since IvHi = I1+vi and 1 + vi = 1 + (1 + vi ) = vi . 

 k−1

Notice that Ivi consists of all α B v B where i ∈ B. Hence its cardinality is 22 = 22 .
k

The following corollary is immediate.

Corollary 5.9 Hermitian self-dual codes exist over Ak for all lengths.
Proof There exists a Hermitian self-dual code of length 1 by the previous theorem so by
taking direct products of this code we get Hermitian self-dual codes of all lengths. 

Notice that the binary image of a Hermitian self-dual code need not be self-dual. For
example, take Iv1 in A1 , which is Hermitian self-dual. But its image under the Gray map is
(00, 01) which is neither self-orthogonal nor self-dual.
We will prove a theorem for Hermitian self-dual codes which is the analog of Theorem 5.4.
The ring Ak is isomorphic to A2k−1 via the Chinese Remainder Theorem. This is realized
as Ak = vk Ak−1 ⊕ (1 + vk )Ak−1 .
Consider the following, for vectors c1 , c2 , c1 , c2 over Ak−1 :

[vk c1 +(1+vk )c2 , vk c1 +(1+vk )c2 ] H = (vk (c1 )i +(1+vk )(c2 )i )(vk (c1 )i +(1+vk )(c2 )i )

= (1 + vk )(c2 )i (c1 )i + vk (c1 )i (c2 )i .

If this is 0 then it requires that [c2 , c1 ] = 0 and [c1 , c2 ] = 0. It follows that if C is a
Hermitian self-dual code over Ak , then C is isomorphic to D × D ⊥ , where D is any code
over Ak−1 .
The code D is isomorphic to E × F, where E and F are codes over Ak−2 . This gives
that D ⊥ = E ⊥ × F ⊥ . Hence we can write, with the proper arrangement of indices, C is
isomorphic to E × E ⊥ × F × F ⊥ . By induction we have the following theorem.

123
Codes over an infinite family of rings with a Gray map 569

Theorem 5.10 Let C be a Hermitian self-dual code over Ak , then, with the proper arrange-
ment of indices, C is isomorphic to
C1 × C1⊥ × C2 × C2⊥ × · · · × C2k−1 × C2⊥k−1
where Ci is any binary code.
Theorem 5.11 Let C be a Hermitian self-dual code over Ak of length n then k (C) is a
formally self-dual binary code of length 2k n.
Proof The results follows from the fact that the Gray map is distancing preserving, that is
the Lee weight satisfies the MacWilliams relations over Ak so the Hamming weight satisfies
the MacWilliams relations over the binary field. 

These theorems give a construction of formally self-dual binary codes. Namely, the
following.
Corollary 5.12 Let C1 , C2 , . . . , C2k−1 be arbitrary self-dual codes. Then
k (C RT (C1 , C1⊥ , C2 , C2⊥ , . . . , C2k−1 , C2⊥k−1 ))

(given the right ordering of indices) is a binary formally self-dual code of length 2k n.

6 Complex fields and lattices

We shall describe a connection between lattices and codes over the ring Ak . The techniques
of this section are similar to those in [1] where codes over Z2k are related to lattices in Rn .
The case when k = 1 is an example in a larger set of examples given in [16].
Let 1 , 2 , ..., k be distinct square free
√ integers with i ≡ 7 √
(mod 8).
√ Define Q√ i recur-
sively as follows: √ Q0 = Q and Qi+1 = Qi ( − i+1 ). Then Qk = Q( − 1 , − 2 , . . . , − k ).
Let ωi = −1+2 − . Define Oi recursively as follows: O0 = Z and Oi+1 = Oi [ωi+1 ]. We
have that Oi is the ring of integers of the field Qi . Let K = Qk and O K = Ok .
Each ωi satisfies the equation x 2 + x + i 4+1 . Notice that each i ≡ 7 (mod 8) so that
i +1
4 is an even integer.
Consider the canonical homomorphism ρ : O K → O K /2O K . Now the image of each ωi
satisfies the equation x 2 + x = 0.
Theorem 6.1 The ring O K /2O K is isomorphic to Ak .
Proof Define the map F : O K /2O K → Ak by
  
F αAωA = αAvA

where ω A = i∈A ωi and α A ∈ F2 .
The map is clearly a bijection and the fact that it is a homomorphism follows since ωi2 = ωi
in O K /2O K . 

√ √
Notice also that F(ωi ) = F(− 21 + 21 i i) = F(− 21 − 21 i i) = 1 + vi = vi , where i =
√
−1 in C. Hence complex conjugation corresponds to conjugation in Ak via the isomorphism
F.
A lattice  over K is an O K submodule of K n of full rank. Define its Hermitian dual to
be
∗ = {v ∈ K n | v · w ∈ O K , ∀w ∈ }

123
570 Y. Cengellenmis et al.


where v · w = vi wi . If  = ∗ we say that the lattice is unimodular and if  ⊆ ∗ we
say that lattice is integral.
Let C be a code in Ank . Define (C) = {v ∈ OnK | ρ(v) ∈ C}. The following is immediate
from the definitions since a linear code C is a submodule of Ans .

Lemma 6.2 Let C be a code over Ak then (C) is an O K lattice.

Next we shall relate the orthogonal of the code with that of the lattice.

Lemma 6.3 Let C be a code over Ak then (C H ) = 2(C)∗ .

Proof If v ∈ 2(C)∗ then 21 v · w ∈ Ok for all w ∈ (C). Then we have

1  vi 
v · w ∈ OK ⇒ wi ∈ O K ⇒ vi wi ∈ 2Ok ⇒ [ρ(v), ρ(w)] H = 0.
2 2
This gives that v ∈ (C H ). Then we have that 2(C)∗ ⊆ (C H ).
Let v ∈ (C H ) which implies ρ(v) ∈ C H and [ρ(v), ρ(w)] H = 0, for all w ∈ (C).
Then we have
  vi 1
[ρ(v), ρ(w)] H = 0 ⇒ vi wi ∈ 2Ok ⇒ wi ∈ O K ⇒ v · w ∈ OK .
2 2
This gives that v ∈ 2(C)∗ . Therefore (C H ) = 2(C)∗ . 

√
Lemma 6.4 Let C be a code over Ak then ( √1 (C))∗ = 2(C)∗ .
2
√
Proof Let v ∈ ( √1 (C))∗ , that is v · w ∈ O K for all w ∈ √1 (C). This implies ( √1 2)
2 √ 2 2
(v · w) ∈ Ok . Then we have that √1 v · 2w ∈ O K for all w ∈ √1 (C), that is for all
√ 2 √ 2
2w ∈ (C). Then we have √1 v ∈ (C)∗ which implies v ∈ 2(C)∗ . Therefore we
√ 2
have ( √1 (C))∗ ⊆ 2(C)∗ .
2 √ √
Now assume v ∈ 2(C)∗ , this implies v · w ∈ O K for all w ∈ 2(C) ∗
√ . That is
√1 v · w ∈ O K for all w ∈ (C), which implies v ∈ ( √1 (C))∗ . Then 2(C)∗ ⊆
2 2
( √1 (C))∗ and we have the result. 

2

Theorem 6.5 The code C over Ak is Hermitian self-dual if and only if √1 (C) is
2
unimodular.
√
Proof If C = C H then by Lemma 6.4 we have ( √1 (C))∗ = 2(C)∗ . Then by
2√
Lemma 6.3, we have (C H ) = 2(C)∗ which gives 2(C)∗ = √1 (C H ) which is
2
equal to √1 (C) √1 (C) = ( √1 (C))∗ .
by assumption. Therefore
2 2 2 √
Next assume ( √1 (C))∗ = √1 (C). Then ( √1 (C))∗ = 2(C)∗ by Lemma 6.4.
√ 2 2 2
Then 2(C)∗ = √1 (C H ) by Lemma 6.3. But √1 (C))∗ = √1 (C). Therefore we
2 2 2
have √1 (C H ) = √1 (C). Next we shall show that since √1 (C H ) = √1 (C) we
2 2 2 2
will have that C = C H . Let v ∈ C then there exists w ∈ (C) with ρ(w) = v. But
(C) = (C H ) which implies ρ(w) ∈ C H . This gives C ⊆ C H . The other direction is
identical. This gives that C = C H and C is Hermitian self-dual. 


123
Codes over an infinite family of rings with a Gray map 571

7 Bounds

For a code C we define the rank of C to be the minimum number of generators of C and
define the free rank of C as the maximum of the ranks of the Ak free submodules of C.
The standard Singleton bound states that for a length n code C over an alphabet
A, d H (C) ≤ n − log|A| (|C|) + 1. A code meeting this bound is said to be an MDS code. This
bound is a strictlty combinatorial bound. There is an algebraic version. In [17], it is proven
that for a code C over a finite principal ideal ring we have that d H ≤ n − r + 1 where r is
the rank of the code. A code meeting this bound is said to be an MDR code [6]. It follows
immediately that an MDR code is an MDS code only when the code is a free code, that is
when the rank of the code is equal to the free rank of the code. This implies that the code is
isomorphic to k direct copies of the ring.
Let C be a code over Ak with C = C RT (C1 , C2 , . . . , C2k ), where Ci is a binary code.
By Theorem 6.3 in [9] we have the following. If Ci is an MDR code for each i, then C =
C RT (C1 , C2 , . . . , C2k ) is an MDR code. If Ci is an MDS code of the same rank for each i,
then C = C RT (C1 , C2 , . . . , C2k ) is an MDS code.
This result follows from the following facts also proven in [9]:
k

2
|C| = |Ci |; (13)
i=1
rank(C) = max{rank(Ci )) | 1 ≤ i ≤ 2k }; (14)
C is a free code if and only if each Ci , is a free code of the same rank; (15)
d H (C RT (C1 , C2 , . . . , C2k )) = min{d(Ci ))}. (16)
The only binary MDS codes are the trivial codes and 1 and 1⊥ , where 1 is the all
one vector. Hence the only MDS codes are those that are formed by taking the Chinese
Remainder theorem of these codes. Note also that C RT (1n , 1n , . . . , 1n ) = 12k n . This gives
the following.
Theorem 7.1 The only non-trivial MDS codes over Ak are 1 which is simply the repetition
code and its orthogonal.
Proof If there was an MDS code C over Ak that was not described above with C =
C RT (C1 , C2 , . . . , C2k ) then by (15) the Ci would all have to have the same dimension
since MDS codes must be free. Then if one of the Ci were not MDS, i.e. its minimum dis-
tance was lower than n − r + 1 then by (16) the minimum distance of C would be too low
to be an MDS code. Hence each Ci must be MDS. 

The same is not true for MDR codes as in the following example.
Example 3 Consider the binary codes of length 7, E 7 and H7 , where E 7 is the even code of
length 7, that is 1⊥ and H7 is the binary Hamming code of length 7. The parameters of the
codes are [7, 6, 2] and [7, 4, 3] respectively. Consider the code C = C RT (E 7 , H7 ) over A1 .
The code has rank 6 by (14) and has minimum Hamming distance 2 by (16). Hence the code
is MDR since 7 − 6 + 1 = 2. The code is not MDS since it is not free. The image of this
code is a binary [14, 10, 2] code.
We can generalize this example to the following theorem.
Theorem 7.2 Let C = C RT (C1 , C2 , . . . , C2k ) with C j an MDS code. If rank(Ci ) ≤
rank(C j ) and d H (Ci ) ≥ d H (C j ) then C is an MDR code over Ak .

123
572 Y. Cengellenmis et al.

Proof By (14) the rank of C is the rank of C j and by (16) the minimum Hamming distance
of C is the minimum Hamming distance of C j . Hence the code is MDR. 


We can now generalize the bounds given for A1 in [6]. Define the minimum Lee weight
of a code C by d L (C). The following theorem follows directly from the Hamming weight
Singleton bound for binary codes.

Theorem 7.3 Let C be a code over Ak , then

d L (C) ≤ 2k n − log2 (|C|) + 1. (17)

Proof The image of the code is length 2k n, then the result follows since the binary image
has log2 (|C|) elements. 


We call a code meeting the bound in Theorem 7.3 a Maximum Lee Distance Seperable
(MLDS) Code.
Since the image of an MLDS code is a binary MDS code we know that the only MLDS
codes are the pre-images of the ambients space, or 1 or 1⊥ . We can now generalize the
algebraic bound. We denote the rank of C by rank(C) and the free rank of C by f-rank(C).
We begin with a lemma.

Lemma 7.4 If C is a code of length n over Ak , then

rank(C) + f-rank(C ⊥ ) = n.

Proof Let C = C RT (C1 , C2 , . . . , C2k ), then C ⊥ = C RT (C1⊥ , C2⊥ , . . . , C2⊥k ). It follows

from the Chinese remainder theorem that rank(C) = max{dim(Ci )} and f-rank(C ⊥ ) =
min{dim(Ci )⊥ }. The equation follows. 


The following setup is the same as in [6]. For a submodule D of Ank and a subset M ⊆
N := {1, 2, . . . , n}, we define
D(M) := {x ∈ D | supp(x) ⊆ M},
D ∗ := Hom R (D, R).
k |M|
It follows that D(M) = D ∩ V (M) is a submodule of V and |V (M)| = 22 . There exists
an isomorphism:
D∗ ∼
= D.
Moreover, there is an R-homomorphism as follows:
g : V −→ D ∗
; y  −→ ( ŷ : x  → [x, y]).
Since Ak is a Frobenius ring (injective module over itself) the map g is surjective. Since R
is a commutative Frobenius ring, we have the following proposition which appears in [17].

Proposition 7.5 Let C be a code of length n over Ak and M ⊆ N . Then there is an exact
sequence of R-modules:
inc g res
0 −→ C ⊥ (M) −→ V (M) −→ C ∗ −→ C(N − M)∗ −→ 0,
where the maps inc, res denote the inclusion map, restriction map, respectively.

123
Codes over an infinite family of rings with a Gray map 573

Let L(x) be the Lee weight of x. Define the minimum Lee weight of a code C by d L (C).
It is immediate that

L(x) ≤ 2k |supp(x)|. (18)

The following theorem is a direct generalization of Theorem 4.3 in [6]. There it is only
done for codes over rings of order 4 but the proof generalizes easily.

Theorem 7.6 Let C be a code of length n over Ak with minimum weight d L (C), then
 
d L (C) − 1
≤ n − rank(C). (19)
2k

Proof Consider the exact sequence in the above proposition, we replace C with C ⊥ giveing
the following exact sequence:

inc g res
0 −→ C(M) −→ V (M) −→ (C ⊥ )∗ −→ C ⊥ (N − M)∗ −→ 0. (20)

Then apply the duality functor ∗ = Hom R (−, R) and take an arbitrary subset M ⊆ N such
that
 
d L (C) − 1
|M| = .
2k

Since C(M)∗ = 0 from (18) and V (M)∗ ∼

= V (M), the exact sequence (20) implies the
following short exact sequence:

0 −→ C ⊥ (N − M) −→ C ⊥ −→ V (M) −→ 0.

We have that V (M) ∼

= R |M| is a projective module, then the above short exact sequence is
split, that is,

C⊥ ∼
= C ⊥ (N − M) ⊕ V (M).

It follows that
 
d L (C) − 1
f-rank(C ⊥ ) ≥ f-rank(V (M)) = |M| = .
2k

The theorem follows from Lemma 7.4. 



We call codes meeting the bound in Theorem 7.6 Maximum Lee Distance with respect to
Rank (MLDR) Code.
 
Theorem 7.7 Let C be a free MLDR code with d L (C)−1
2 k = d L (C)−1
2k
then C is an MLDS
code.
 
Proof If C is an MLDR satisfying d L (C)−1
2 k = d L (C)−1
2k
then d L (C)−1
2k
= n − rank(C).
This gives that d (C) = 2 n − 2 rank(C) + 1. If C is free then |C| = |A |rank(C) , then
L
k k
k
log2 (|C|) = 2k rank(C) and then C meets the bound given in Eq. 17. 


123
574 Y. Cengellenmis et al.

8 The structure of the cyclic codes over Ak

The ring Ak , like the ring Rk (see [11,12]), has the property that F2 is a subring for all k. In
contrast, the ring Z2k does not have this property. This allows for a simplified description of
polynomials over the rings Ak and Rk as apposed to the description for Z2k . Our description
of cyclic codes over Ak has a similar foundation as that for Rk as presented in [12]. However,
since Rk is a local ring and Ak is not, the description diverges significantly when describing
the ideals.
We define the following map t : At → At−1 where α ∈ At is mapped to α (mod vt ).
 of maps to be μk = 1 ◦ 2 ◦ · · · ◦ k . This means that μk
Define the following composition
takes an element of the form A⊆{1,...,k} c A v A to c∅ . We can extend this map to polynomials
in the natural way as follows:
 
μk ( p(x)) = μk ( αi x i ) = μk (αi )x i .

We say that a monic polynomial in Ak [x] is a basic irreducible polynomial if its projection
under μk is an irreducible polynomial in F2 [x]. Note that the term basic irreducible polyno-
mial refers to a polynomial in Ak [x] but the irreducibility refers to the fact that its projection
is not factorable over the binary field. Note that any polynomial in F2 [x] can be viewed as
a polynomial in Ak [x] since F2 is a subring of Ak . This means that if a polynomial p(x) is
irreducible in F2 [x] then it is irreducible in Ak [x] as well.
We let G R(Ak , s) = Ak [x]/ p(x) where p(x) is a basic monic irreducible polynomial
of degree s. If the polynomial p(x) is an irreducible polynomial over F2 of degree s. Then
F2 [x]/ p(x) is a field of order 2s .
Define A0 = F2 [x]/ p(x) and define

At = At−1 [vt ]/vt2 = vt , vt vi = vi vt . (21)

It follows immediately from the definition that any element in At is of the form α + βvt
where α and β are in At−1 .

Lemma 8.1 The element α + βvt in At is a unit if and only if α is a unit in At−1 .

Proof Take two elements α + βvt , γ + δvt ∈ At , then their product (α + βvt )(γ + δvt ) =
αγ + (αδ + βγ + βδ)vt . If α is a unit set γ = α −1 and δ = α −1 (βγ + βδ). This gives
that their product is 1 and so α + βvt is a unit for any β. If α is not a unit then it is a zero
divisor since the ring is finite. Let  be such that α = 0. Then (vt + 1)(α + βvt ) =
α(vt + 1) + β(vt + vt ) = 0, which gives that α + βvt is a zero divisor for all β. 


We define U (At ) be the group of units of the ring At . The previous lemma gives

|U (At )| = |U (At−1 )||At−1 |. (22)

We know that |A0 | = 2s , that is, it is the field of order s and so we have that |U (A0 )| =
t
2s − 1. This gives that |U (A1 )| = 2s (2s − 1). We also have that |At | = (2s )2 since there are
2t subsets of {v1 , . . . , vt } and 2s choices for each coefficient. Induction gives the following
result.

Theorem 8.2 The group of units U (At ) is the direct product of a cyclic group G of order
2s − 1, which is the multiplicative group of the finite field, and an abelian group H of order
(2s )2 −1 , with |U (At )| = (2s )2 −1 (2s − 1) = (22 )s − (22 −1 )s .
t t t t

123
Codes over an infinite family of rings with a Gray map 575

It follows that the elements of the group H are of the form 1 + α A v A where A ⊆
{1, 2, . . . , k} and α A ∈ G. It is immediate that the zero divisors are of the form α A v A , α A ∈
A0 , α∅ = 0.
We have that |A0 | = 2s and there are 2t − 1 non-empty subsets of {1, 2 . . . , t} which gives
that there are (2s )2 −1 = (22 −1 )s non-units in At .
t t

We summarize the results in the following.

t
Theorem
 8.3 The ring At has cardinality |At | = (22 )s . The units are of the form
α A v A , α A ∈ A0 , α∅ = 1 and there are (22 )s − (22 −1 )s of them. The non-units are
t t


of the form α A v A , α A ∈ A0 , α∅ = 0 and there are (22 −1 )s of them.
t

This theorem allows us to describe the ideals in the polynomial ring, as in the following
corollary.

Corollary 8.4 There is a bijective correspondence between the ideals of At and the ideals
of At . That is, any ideal in At is of the form β1 , β2 , . . . , β , where βi ∈ At .

Lemma 8.5 If x n − 1 = pi , where the polynomials pi ∈ Ak [x] are all basic irreducible,
pairwise co-prime polynomials with n is odd, then this factorization is the lift of the unique
factorization over F2 .

Proof Recall that the ring Ak contains the field F2 as a subring. We shall consider the
factorization of the polynomial x n − 1 over the binary field. The polynomial x n − 1 factors
uniquely as a product of pairwise co-prime irreducible polynomials in F2 [x]. Therefore,
 this
polynomial factors into irreducible polynomials over Ak . Namely, if x n − 1 = pi then
x n − 1 = μ(x n − 1) = μ( pi ). Hence the factorization is a lift of the unique factorization
over the binary field. 


Unlike the rings Rk , which are local rings (that is they have a unique maximal ideal), we
cannot apply Hensel’s lift to obtain a unique factorization of x n − 1 over Ak . However, we
can use the map μk to describe the ideals which correspond to cyclic codes. Consider the
map μk : Ak [x]/x n − 1 → F2 [x]/x n − 1. This map is a homomorphism. It follows
immediately that the image of an ideal is an ideal.

Theorem 8.6 Let n be odd and let p(x) be a divisor of x n − 1. The ideals in Ak [x]/x n − 1
are of the form
⎛ ⎞ ⎛ ⎞
    
p(x) + ⎝ α A v A ri (x)⎠ , ⎝ α B v B si (x)⎠ ,
i A⊂{1,2,...,k} i B⊂{1,2,...,k}
⎛ ⎞
  
..., ⎝ ⎠
αC vC qi (x) .
i C⊂{1,2,...,k}
 
Proof The image under μk of anything of the form i ( A⊂{1,2,...,k} α A v A ri (x)) is 0. Hence
these ideals are the preimages of the ideals of the form  p(x), where p(x) is a divisor of
x n − 1 over F2 . 


This theorem shows that over Ak , k ≥ 2 there are many cyclic codes since there is an
abundance of ideals in Ak .
We shall now give an alternate description of the structure of cyclic codes over Ak as
presented at ACCT 2012, Pomorie, Bulgaria [3]. Let m1 , m2 , . . . , m2k , be the maximal ideals

123
576 Y. Cengellenmis et al.

of Ak . Since the ring Ak is a principal ideal ring we can denote the single element that generates
mi by m i . We have shown in Theorem 2.6 how to construct this single element.
Let C be a linear code over Ak . It is immediate that
C = (m 1 )C1 ⊕ (m 2 )C2 ⊕ · · · ⊕ (m 2k )C2k , (23)
where Ci is a binary code. It follows that
C ⊥ = (m 1 )C1⊥ ⊕ (m 2 )C2⊥ ⊕ · · · ⊕ (m 2k )C2⊥k . (24)
Theorem 8.7 [3] Let C be a code over Ak and let Ci be the binary codes given in Eq. 23.
The code C is cyclic if and only if Ci is a cyclic code for all i.
Proof Take σ to be the cyclic shift and let v ∈ C and vi ∈ Ci with v = m 1 v1 + m 2 v2 +
· · · + m 2k v2k . It follows that
σ (v) = v = m 1 σ (v1 ) + m 2 σ (v2 ) + · · · + m 2k σ (v2k ). (25)
When each Ci is cyclic we have that σ (vi ) ∈ Ci for all i. Therefore, by Eq. 25 we have
σ (v) ∈ C.
If the code C is cyclic then σ (v) ∈ C and so by Equation 25 we have that σ (vi ) ∈ Ci for
all i. 

Corollary 8.8 [3] If a code C over Ak is cyclic then C ⊥ is cyclic.
Proof Since the orthogonal of a binary cyclic code is a cyclic code the result follows imme-
diately form the previous theorem. 

We can use this to give the alternate description of cyclic codes.
Theorem 8.9 [3] Let C be a cyclic code over Ak , then there exists a polynomial g(x) in
Ak [x] that divides x n − 1 that generates the code.
Proof Let C be a cyclic code over Ak with C = (m 1 )C1 ⊕ (m 2 )C2 ⊕ · · · ⊕ (m 2k )C2k . Take
gi (x) to be the generator of Ci . It follows that C has the form
m 1 g1 (x), m 2 g2 (x), . . . , m 2k g2k (x). (26)
Let D = m 1 g1 (x) + m 2 g2 (x) + · · · + m 2k g2k (x). We have that D ⊆ C. Notice that
m i m i = m i and m i m j = 0 if i = j. Then m i (m 1 g1 (x), m 2 g2 (x), . . . , m 2k g2k (x)) =
m i gi (x) which gives that C ⊆ D. This give that C = D and that C is generated by a single
element.
It is given that gi (x) divides x n − 1. Let ri (x) be the binary polynomial such
that gi (x)ri (x) = x n − 1. Then we have x n − 1 = (m 1 g1 (x) + m 2 g2 (x) + · · · +
m 2k g2k (x))(m 1 r1 (x) + m 2 r2 (x) + · · · + m 2k r2k (x)) recalling that m i m i = m i and m i m j = 0
for i = j. Finally, we have that x n − 1 = g(x)(m 1 r1 (x) + m 2 r2 (x) + · · · + m 2k r2k (x)).  
We can combine this result with the result in Theorem 8.6 and we have the following.
Corollary 8.10 Any ideal of the form
⎛ ⎞ ⎛ ⎞
    
p(x) + ⎝ α A v A ri (x)⎠ , ⎝ α B v B si (x)⎠ ,
i A⊂{1,2,...,k} i B⊂{1,2,...,k}
⎛ ⎞
  
..., ⎝ αC vC qi (x)⎠
i C⊂{1,2,...,k}

123
Codes over an infinite family of rings with a Gray map 577

where p(x) is a binary polynomial that divides x n − 1 can be rewritten as g(x) where g(x)
divides x n − 1 in Ak [x].
For a polynomial, p(x) = a0 +a1 x +. . . , ak x k recall that we define p(x) = ak +ak−1 x +
· · · + a0 x k .
Lemma 8.11 [3] If C is a cyclic code over Ak generated by g(x) then C ⊥ is a cyclic code
generated by (x n − 1)/g(x).
Proof Take C to be a cyclic code over Ak generated by g(x) where the code is given by
m 1 g1 (x), m 2 g2 (x), . . . , m 2k g2k (x) as given in Eq. 26. This gives that, as in Eq. 23, C =
(m 1 )C1 ⊕ (m 2 )C2 ⊕ · · · ⊕ (m 2k )C2k , where Ci is a binary code. We have that C ⊥ =
(m 1 )C1⊥ ⊕ (m 2 )C2⊥ ⊕ · · · ⊕ (m 2k )C2⊥k . We know that if gi (x) generates the binary cyclic code
Ci then there exists a polynomial h i (x) that generates C ⊥ , where h i (x) = (x n − 1)/gi (x).
The desired result follows by applying the ring isomorphism to these polynomials. 

Theorem 8.12 [3] If C = g(x) is a cyclic self-orthogonal code over Ak then g(x)g(x) =
x n − 1.
Proof Let C be a cyclic code over Ak generated by g(x) where the code is of the form
m 1 g1 (x), m 2 g2 (x), . . . , m 2k g2k (x) as given in Eq. 26. The isomorphism gives that each
gi (x) generates a binary self-dual code. It follows from [18] that gi (x)gi (x) = x n − 1.  
We can now use this alternate description of cyclic codes to get further results.

Theorem 8.13 Let x n − 1 ∈ F2 [x] be factored as x n − 1 = ri=1 ( pi (x))si where the pi (x)
are pairwise relatively prime non-zero polynomials over the field F2 . Then the number of
 k
cyclic codes over Ak is ri=1 (si + 1)2 .

Proof We know that the number of binary cyclic codes is ri=1 (si + 1). Then the result
follows from Eq. 23. 

Example 4 Let n = 2 and k = 2. Then x 3 −1 = (x +1)(x 2 +x +1). Then r = 2, s1 = s2 = 1,
so there are 256 cyclic codes of length 3 over A2 .
This example shows that there is a wealth of cyclic codes for all n, k over Ak .
Example 5 Let m1 , m2 , . . . , m2k be the maximal ideals of Ak . Take a polynomial g(x) ∈
F2 [x]] with g(x)g(x) = x n − 1. Then g(x)m1 ⊕ g(x)m2 ⊕ · · · ⊕ g(x)m2k is a self-dual
cyclic code of length n. Then each g(x) gives an infinite family of self-dual cyclic codes. For
example, if n = 4, let g(x) = 1 + x 2 .
Example 6 Let m1 , m2 , . . . , m2k be the maximal ideals of Ak . Let gi (x) F2 [x] be any divisor
of x n − 1. Then for the proper arrangement of indices we have that g1 (x)m1 ⊕ g1 (x)m2 ⊕
g2 (x)m3 + g2 (x)m4 · · · ⊕ g2k−1 (x)m2k is a cyclic Hermitian self-dual code of length n.
Both of these examples will be continued in Examples 7 and 8.

9 The Gray image of cyclic codes and quasi-cylic codes over Ak

Let a ∈ F22 n with a = (a0 , . . . , a2k n−1 ) = (a (0) |a (1) | . . . |a (2 −1) ), a (i) ∈ Fn2 for
k k

i = 0, 1, . . . , 2k − 1. Let σ ⊗2 be the map from F22 n to F22 n given by σ ⊗2 (a) =

k k k k

(σ (a (0) )| . . . |σ (a (2 −1) )) where σ is the usual shift (c0 , . . . , cn−1 )  → (cn−1 , c0 , . . . , cn−2 )
k

n
on F2 .

123
578 Y. Cengellenmis et al.

Proposition 9.1 Let σ be the cyclic shift on An1 , let 1 be the Gray map from An1 to F2n
2 , let
σ ⊗2 be as above. Then 1 σ = σ ⊗2 1 .
Proof Let c = (c0 , . . . , cn−1 ) = (r0 +v1 q0 , . . . , rn−1 +v1 qn−1 ) ∈ An1 with ri , qi ∈ F2 where
i = 0, 1, . . . , n − 1, we have σ (c) = (cn−1 , c0 , . . . , cn−2 ) = (rn−1 + v1 qn−1 , . . . , rn−2 +
v1 qn−2 ). If we apply 1 , we have
1 (σ (c)) = (rn−1 , r0 , . . . , rn−2 , rn−1 + qn−1 , r0 + q0 , . . . , rn−2 + qn−2 ).
On the other hand, 1 (c) = (r0 , . . . , rn−1 , r0 + q0 , . . . , rn−1 + qn−1 ) for every c and
σ ⊗2 (1 (c)) = (rn−1 , r0 , . . . , rn−2 , rn−1 + qn−1 , r0 + q0 , . . . , rn−2 + qn−2 ).
Therefore, we have that 1 σ = σ ⊗2 1 . 

Lemma 9.2 Let σ be cyclic shift on Ank , let k be Gray map from Ank to F2 , let σ 2k n ⊗2k be
as above. Then k σ = σ ⊗2 k .
k

Proof We shall prove the result by induction. We saw that the result is true for k = 1
in Proposition 9.1. Assume that the result is true for k − 1. Let c = (c0 , . . . , cn−1 ) with
ci = ai,k−1 + vk bi,k−1 where ai,k−1 , bi,k−1 ∈ Ak−1 for i = 0, 1, . . . , n − 1.
Then
k (σ (c)) = k (σ (c0 , . . . , cn−1 ))
= k ((an−1,k−1 + vk bn−1,k−1 ), . . . , (an−2,k−1 + vk bn−2,k−1 ))
= k ((an−1,k−1 , a0,k−1 , . . . , an−2,k−1 ) + vk (bn−1,k−1 , . . . , bn−2,k−1 ))
= (k−1 (an−1,k−1 , . . . , an−2,k−1 ), k−1 (an−1,k−1 + bn−1,k−1 , . . . ,
an−2,k−1 + bn−2,k−1 )).

Moreover, there exist s0 , s1 , . . . , sn−1 ∈ Ak−1 such that

k−1 ((an−1,k−1 , . . . , an−2,k−1 )) = k−1 (σ (s0 , . . . , sn−1 ))
and there exist t0 , t1 , . . . , tn−1 ∈ Ak−1 such that
k−1 (an−1,k−1 + bn−1,k−1 , . . . , an−2,k−1 + bn−1,k−1 )) = k−1 (σ (t0 , . . . , tn−1 )).
By assumption, we have
k (σ (c0 , . . . , cn−1 ))
(σ ⊗2 (k−1 (s0 , . . . , sn−1 )), σ ⊗2
k−1 k−1
= (k−1 (t0 , . . . , tn−1 )))
⊗2k
= (σ (k (c0 , . . . , cn−1 ))).


Theorem 9.3 The Gray image a cyclic code over Ak of length n is a quasi-cyclic code of
index 2k over F2 with length 2k n.
Proof Let C be a cyclic code over Ak of length n. This gives that σ (C) = C. If we apply
k , we have k (σ (C)) = k (C). From the previous lemma, k (σ (C)) = k (C) =
σ ⊗2 (k (C)). So k (C) is a quasi-cyclic code of index 2k over F2 with length 2k n.
k


Proposition 9.4 Let τs,l be the quasi-cyclic shift on An1 , let 1 be the Gray map from An1 to
2 , let 1 be as before. Then 1 τs,l = 1 1 .
F2n

123
Codes over an infinite family of rings with a Gray map 579

Proof Let e = (e0,0 , . . . , e0,l−1 , e1,0 , . . . , e1,l−1 , . . . , es−1,0 , . . . , es−1,l−1 ) with ei, j =
ri, j + v1 qi, j where i = 0, 1, . . . , s − 1 and j = 0, . . . , l − 1. We have that
τs,l (e) = (rs−1,0 + v1 qs−1,0 , . . . , rs−1,l−1 + v1 qs−1,l−1 , r0,0 + v1 q0,0 , . . . , r0,l−1
+v1 q0,l−1 , . . . , rs−2,0 + v1 qs−2,0 , . . . , rs−2,l−1 + v1 qs−2,l−1 ).
It follows that τs,l (e) = (rs−1,0 + v1 qs−1,0 , . . . , rs−2,l−1 + v1 qs−2,l−1 ). If we apply 1 , we
have
1 (τs,l (e)) = (rs−1,0 , . . . , rs−2,l−1 , r0,0 , . . . , r0,l−1 , . . . , rs−2,0 , . . . ,
rs−2,l−1 , rs−1,0 + qs−1,0 , . . . , rs−2,l−1 + qs−2,l−1 ).
On the other hand, we know that
1 (e) = (r0,0 , . . . , r0,l−1 , . . . , rs−1,0 . . . , rs−1,l−1 , r0,0
+q0,0 , . . . , r0,l−1 + q0,l−1 , . . . , rs−1,0 + qs−1,0 , . . . , rs−1,l−1 + qs−1,l−1 )
and
1 (1 (e)) = (rs−1,0 , . . . , rs−1,l−1 , . . . , rs−2,0 , . . . , rs−2,l−1 , rs−1,0 + qs−1,0 , . . . , rs−1,l−1
+qs−1,l−1 , . . . , rs−2,0 + qs−2,0 , . . . , rs−2,l−1 + qs−2,l−1 ).
This gives that 1 τs,l = 1 1 . 


Lemma 9.5 Let τs,l be quasi-cyclic shift on Ank and let k be the Gray map from Ank to F2 2k n

with k defined as above. Then

k τs,l = k k .
Proof The proof follows easily as above. 

Theorem 9.6 The Gray image of a quasi-cyclic code over Ak of length n with index l is a l
quasi-cyclic code of index 2k over F2 with length 2k n.
Proof Let C be a quasi-cyclic codes over Ak of length n with index l. That is τs,l (C) = C.
If we apply k , we have k (τs,l (C)) = k (C). From the previous Lemma, k (τs,l (C)) =
k (C) = k (k (C)). So k (C) is a l quasi-cyclic code of index 2k over F2 with length
2k n. 

Example 7 Continuing Example 5, consider the self-dual cyclic code of the form g(x)m1 ⊕
g(x)m2 ⊕ · · · ⊕ g(x)m2k with g(x)g(x) = x n − 1. Then k (C) is a binary self-dual quasi-
cyclic code of index 2k and length 2k n.
Example 8 Continuing Example 6, consider the cyclic Hermitian self-dual code g1 (x)m1 ⊕
g1 (x)m2 ⊕ g2 (x)m3 + g2 (x)m4 · · · ⊕ g2k−1 (x)m2k where gi (x) is a divisor of x n − 1 in F2 [x].
Then k (C) is a binary formally self-dual quasi-cyclic code of index 2k and length 2k n.

Acknowledgments The authors are grateful to Hamid Kulosman for helpful discussions.

References

1. Bannai E., Dougherty S.T., Harada M., Oura M.: Type II codes, even unimodular lattices, and invariant
rings. IEEE-IT 45(4), 1194–1205 (1999).
2. Cengellenmis Y.: On the Cyclic Codes over F3 + vF3 . Int. J. Algebra 4(6), 253–259 (2010).

123
580 Y. Cengellenmis et al.

3. Cengellenmis Y., Dougherty S.T.: Cyclic codes over Ak . In: Proceedings of ACCT2012, Pomorie, Bul-
garia.
4. Dougherty S.T., Fernandez-Cordoba C.: Codes over Z 2k , Gray maps and self-dual codes. Adv. Math.
Commun. 5(4), 571–588 (2011).
5. Dougherty S.T., Liu H.: Independence of vectors in codes over rings. Des. Codes Cryptogr. 51, 55–68
(2009).
6. Dougherty S.T., Shiromoto K.: Maximum distance codes over rings of order 4. IEEE-IT 47(1), 400–404
(2001).
7. Dougherty S.T., Harada M., Gaborit P., Solé P.: Type II Codes Over F2 + uF2 . IEEE Trans. Inf. Theory
45(1), 32–45 (1999).
8. Dougherty S.T., Gaborit P., Harada M., Munemasa A., Solé P.: Type IV self-dual codes over rings. IEEE-IT
45(7), 2345–2360 (1999).
9. Dougherty S.T., Kim J.L., Kulosman H.: MDS codes over finite principal ideal rings. Des. Codes Cryptogr.
50, 77–92 (2009).
10. Dougherty S.T., Kim J.L., Kulosman H., Liu H.: Self-dual codes over Frobenius rings. Finite Fields Appl.
16, 14–26 (2010).
11. Dougherty S.T., Yildiz B., Karadeniz S.: Codes over Rk , Gray maps and their binary images. Finite Fields
Appl. 17(3), 205–219 (2011).
12. Dougherty S.T., Yildiz B., Karadeniz S.: Cyclic codes over Rk , Gray maps and their binary images. Des.
Codes Cryptogr. 63(1), (2012).
13. Hammons A.R., Kumar P.V., Calderbank A.R., Sloane N.J.A., Solé P.: The Z4 -linearity of Kerdock,
Preparata, Goethals and related codes. IEEE Trans. Inf. Theory 40, 301–319 (1994).
14. Nebe G., Rains E.M., Sloane N.J.A.: Self-Dual Codes and Invariant Theory. Springer, Berlin (2006).
15. Rains E., Sloane N.J.A.: Self-Dual Codes in Handbook of Coding Theory. Elsevier, Amsterdam (1998).
16. Shaska T., Wijesiri S.: Codes over rings of size four, Hermitian lattices and corresponding theta functions.
Proc. Am. Math. Soc. 136(3), 849–857 (2008).
17. Shiromoto K.: Singleton bounds for codes over finite rings. J. Algebraic Comb. 12(1), 95–99 (2000).
18. Sloane N.J.A., Thompson J.G.: Cyclic self-dual codes. IEEE Trans. Inf. Theory IT-29 5, 364–366 (1983).
19. Wood J.: Duality for modules over finite rings and applications to coding theory. Am. J. Math. 121,
555–575 (1999).

123