On Codes over Z p s with the Extended Lee Weight

Filomat 30:2 (2016), 255–268 DOI 10.2298/FIL1602255O Published by Faculty of Sciences and Mathematics, University of Niš, Serbia Available at: http://www.pmf.ni.ac.rs/filomat On Codes over Zps with the Extended Lee Weight Zeynep Ödemiş Özgera , Bahattin Yıldızb , Steven T. Doughertyc a Department of Engineering Sciences, İzmir Kâtip Çelebi University, 35620 İzmir, Turkey of Mathematics, Fatih University, 34500 İstanbul, Turkey c Department of Mathematics, University of Scranton, Scranton, PA 18510, USA b Department Abstract. We consider codes over Zps with the extended Lee weight. We find Singleton bounds with respect to this weight and define MLDS and MLDR codes accordingly. We also consider the kernels of these codes and the notion of independence of vectors in this space. We investigate the linearity and duality of the Gray images of codes over Zps . 1. Introduction In the early history of coding theory, codes over finite fields were predominantly studied. The most common weight used for such codes was the Hamming weight, which is defined to be the number of nonzero coordinates. We will denote the Hamming weight by wH . Many encoding and decoding schemes as well as error correction algorithms are based on the Hamming distance. Codes over rings have been considered since the early seventies, however it was not until the beginning of the nineties that they became a widely popular research field in coding theory. In 1994, Hammons et al.([12]) solved a long standing problem in nonlinear binary codes by constructing the Kerdock and Preparata codes as the Gray images of linear codes over Z4 . This work started an intense activity on codes over rings. The rich algebraic structure that rings bring together with some better than optimal nonlinear codes obtained from linear codes over rings have increased the popularity of this topic. What started with the ring Z4 , later was extended to rings such as Z2k , Zpk , Galois rings, Fq + uFq , and various other rings. For codes over rings, weights other than the Hamming weight were considered. For example, in [12], the authors used the Lee weight on Z4 , which we will denote by wL and was defined as   0 if x = 0,    2 if x = 2, wL (x) :=     1 otherwise. The Gray map φL : Z4 → Z22 , 2010 Mathematics Subject Classification. Primary 94B05, 94B15 Keywords. extended Lee weight, Gray map, kernel, Singleton bound, MLDS codes, MLDR codes Received: 17 April 2013; Accepted: 18 March 2016 Communicated by Miroslav Ćirić Email addresses: zeynep.odemis.ozger@ikc.edu.tr (Zeynep Ödemiş Özger), byildiz@fatih.edu.tr (Bahattin Yıldız), prof.steven.dougherty@gmail.com (Steven T. Dougherty) Z. Ö. Özger et al. / Filomat 30:2 (2016), 255–268 256 with φL (0) = (00), φL (1) = (01), φL (2) = (11), φL (3) = (10) , Hamming distance), where the Lee turns out to be a nonlinear isometry from (Zn4 , Lee distance) to (F2n 2 distance on Z4 is defined as dL (x, y) := wL (x − y), x, y ∈ Z4 , and similarly, the Hamming distance on Z4 as dH (x, y) := wH (x − y), x, y ∈ Z4 . This means that if C is a linear code over Z4 of length n, size M and minimum Lee distance d, then φL (C) is a possibly nonlinear binary code with parameters [2n, M, d], where d is the minimum Hamming distance of φL (C). When extending the Lee distance from Z4 to the more general ring extensions, the homogeneous weight was mostly used. The homogeneous weight has a lot of advantages, which made them useful in constructing codes over rings. It is related to exponential sums (see [5] and [20] for example), making it easier to find bounds by using some number theoretic arguments such as the Weil bound. The homogeneous weight also gives rise to codes with high divisibility properties. Another extension of the Lee weight is also possible and has been used by different researchers. For example the weight wL on Z2s , defined by wL (x) := ( x if x ≤ 2s−1 , s 2 − x if x > 2s−1 , was used partly in [4], [6] and [23]. A simple Gray map for this weight maps codes over Z2s to (mostly) nonlinear binary codes. This extension was generalized to Zm as the Lee weight by letting wL (x) = min{x, m − x} in some works, however no Gray map has been offered for such a weight. In [22], the Lee weight on Z2s given above was generalized to the rings Zps and the Galois rings GR(ps , m), together with a simple description of a Gray map projecting codes over Zps to codes over the finite prime field Fp = Zp . In this work, we study codes over Zps together with this Lee weight from many angles such as Singleton bounds, independence, kernels and duality. The rest of the paper is organized as follows: In section 2, we recall the extended Lee weight, the Gray map and some properties for codes over Zps from [22]. In section 3, some bounds on codes over Zps concerning both length and size of the codes are given and MLDS and MLDR codes are defined accordingly. In section 4, the notions of kernel and independence are investigated. In section 5, some results about self-duality and self-orthogonality are found. 2. The Extended Lee Weight and Its Gray Map We recall that a new weight on Zps , a generalization of wL , was defined in [22] as follows:   x if x ≤ ps−1 ,    s−1 p if ps−1 < x ≤ ps − ps−1 , wL (x) :=     ps − x if ps − ps−1 < x ≤ ps − 1, where p is prime. Note that for p = 2 and s = 2 this reduces to the Lee weight for Z4 and for p = 2 and any s, this is the weight that was used briefly by Carlet in [4] and by Dougherty and Fernández-Córdoba in [6]. ps−1 We can define a Gray map from Zps to Zp just as was done for the homogeneous weight as follows: Z. Ö. Özger et al. / Filomat 30:2 (2016), 255–268 0 1 2 → → → · · ps−1 → ps−1 + 1 → ps−1 + 2 → · · ps−1 + ps−1 − 1 → 2ps−1 → 2ps−1 + 1 → · · 2ps−1 + ps−1 − 1 → 3ps−1 → · · (p − 1)ps−1 → (p − 1)ps−1 + 1 → · · ps − 2 → ps − 1 → 257 (000 · · · 000), (100 · · · 000), (110 · · · 000), (111 · · · 111), (211 · · · 111), (221 · · · 111), (222 · · · 221), (222 · · · 222), (322 · · · 222), (333 · · · 332), (333 · · · 333), ((p − 1) · · · (p − 1)), (0(p − 1) · · · (p − 1)), (000 · · · 0(p − 1)(p − 1)), (000 · · · 00(p − 1)). We simply put 1’s in the first x coordinates and 0’s in the other coordinates for all x ≤ ps−1 . If x > ps−1 then the Gray map takes x to q + φL (r), where φL is the Gray map for wL , q = (qqq · · · qqq) and q and r are such that x = qps−1 + r, which can be found by division algorithm. Here, 0 ≤ x ≤ ps − 1, 0 ≤ q ≤ p − 1, 0 ≤ r ≤ ps−1 − 1. Here by putting p = 2, we get the same Gray map given in [23] and [6], which is 0 1 2 → → → · · 2s−1 → 2s−1 + 1 → 2s−1 + 2 → · · 2s − 2 → 2s − 1 → (000 · · · 000), (100 · · · 000), (110 · · · 000), (111 · · · 111), (011 · · · 111), (001 · · · 111), (000 · · · 011), (000 · · · 001). As an example, when p = 3, s = 2 we get the extended Lee weight on Z9 is defined as   x if x ≤ 3,    3 if 3 < x ≤ 6, wL (x) :=     9 − x if 6 < x ≤ 8. 258 Z. Ö. Özger et al. / Filomat 30:2 (2016), 255–268 The Gray map takes Z9 to Z33 as follows: 0 1 2 3 4 5 6 7 8 → → → → → → → → → (000), (100), (110), (111), (211), (221), (222), (022), (002). We define the Lee distance on Zps as dL (x, y) := wL (x − y), x, y ∈ Zps . Note that this is a metric on Zps and by extending wL and dL linearly to (Zps )n in an obvious way, we get a weight and a metric on (Zps )n . ps−1 Theorem 2.1. The map φL : (Zps , dL ) −→ (Fp , dH ) is a distance preserving (not necessarily linear) map, where dL and dH denote the Lee and the Hamming distances respectively. The proof of this theorem can be found in [22] with the following corollary: Corollary 2.2. If C is a linear code over Zps of length n, size M and minimum Lee distance d, then φL (C) is a (possibly nonlinear) code over Fp of length nps−1 , size M and minimum Hamming distance d. The concepts of minimum Lee distance and minimum Lee weight are the same for linear codes over Zps . ps−1 m ps−1 A Gray map from GR(ps , m) to Fp can also be defined by extending φL : (Zps , dL ) −→ (Fp , dH ) (see [22], Section 3), which means that most of the work done in this paper is applicable to Galois rings. 3. Singleton Bounds for Codes Over Zps A Singleton bound for codes over a finite quasi-Frobenius ring is already given in [19] as an MDS bound. Since this result is given for any weight function, it can be specified for the extended Lee weight. Definition 3.1 (Complete weight). [19] Let R be a finite commutative quasi-Frobenius ring, and let V := Rn be a free module of rank n consisting of all n-tuples of elements of R. For every x = (x1 , · · ·, xn ) ∈ V and r ∈ R, the complete weight of x is defined by nr (x) := |{i |xi = r }| . Definition 3.2 (General weight function). [19] Let ar ,(0 ,)r ∈ R, be positive real numbers, and set a0 = 0. Then X ar nr (x) (1) w(x) := r∈R is called a general weight function. Note that when ar = 1, r ∈ R − {0}, w(x) gives the Hamming weight of x. The following theorem gives a Singleton bound for any finite quasi-Frobenius ring and any weight function. Z. Ö. Özger et al. / Filomat 30:2 (2016), 255–268 259 Theorem 3.3. [19] Let C be a code of length n over a finite commutative quasi-Frobenius ring R. Let w(x) be a general weight function on C, as in (1), and with maximum ar −value A. Suppose the minimum weight of the elements in C is d. Then $ % d−1 ≤ n − log|R| |C| , A where ⌊b⌋ is the integer part of b. Since Zps is a finite commutative Frobenius ring by letting w(x) = wL (x), we have ps−1 as the maximum ar −value. Applying these information to Theorem 3.3 we get the following: Theorem 3.4. Let C be a code of length n over Zps with minimum distance d. Then % $ d−1 ≤ n − logps |C| . ps−1 Codes meeting this bound are called MLDS (Maximum Lee Distance Separable) codes. In [18], another bound was found over Zl with a different generalization of the Lee weight. Now we will find a similar result for codes over Zps with wL (x) by the same method used. Definition 3.5 (Rank, Free Rank). Let C be any finitely generated submodule of Znps , that is isomorphic to Zps /pa1 Zps ⊕ Zps /pa2 Zps ⊕ · · · ⊕ Zps /pan−1 Zps , where ai are positive integers with pa1 |pa2 | · · · |pan−1 |ps . Then rank(C) := |{i |ai , 0 }| is called the rank of C and f ree rank(C) := |{i |ai = s }| is called the free rank of C. Any linear code over Zps has a generator matrix, which is permutationally equivalent to a matrix of the form:   A0,3 ··· ··· A0,s   Iδ0 A0,1 A0,2  0 pI pA1,2 pA1,3 ··· ··· pA1,s   δ1   0 p2 Iδ2 p2 A2,3 ··· ··· p2 A2,s   0    . 0 ··· ··· ··· ··· G =  · · · · · · (2)   · · · · · ·  ··· ··· ··· ··· ···     0 0 0 ··· ps−2 Iδs−2 ps−2 As−2,s−1 ps−2 As−2,s    ps−1 As−1,s 0 0 0 ··· 0 ps−1 Iδs−1 This means that after a finite number of permutations of columns and rows of the generator matrix, we can get a matrix of the form (2). A generator matrix G, which is of the form (2), is called a generator matrix in standard form. Then a code C over Znps is of type (ps )δ0 (ps−1 )δ1 · · · (p)δs−1 , and rank(C) = f ree rank(C) = δ0 + δ1 + · · · + δs−1 , δ0 . Let C⊥ , namely the dual of C, be defined as n o C⊥ := v ∈ Znps | hv, wi = 0 for all w ∈ C , Z. Ö. Özger et al. / Filomat 30:2 (2016), 255–268 where hv, wi = P 260 vi wi (mod ps ). The code C⊥ is isomorphic to Zps /ps−a1 Zps ⊕ Zps /ps−a2 Zps ⊕ · · · ⊕ Zps /ps−an−1 Zps . From [18], [15], [7], [8], [6], and the definitions above, the relationship between the rank of a code and its dual’s free rank can be given as follows: rank(C) + f ree rank(C⊥ ) = n. (3) For a submodule D ⊆ V := (Zps )n and a subset M ⊆ N := {1, 2, . . . , n}, we define n o D(M) := x ∈ D supp(x) ⊆ M , D∗ := HomZps (D, Zps ), where Hom(·, ·) is the hom functor, and supp(x) := {i ∈ N |xi , 0 } . From the fundamental theorem of finitely generated abelian groups, we have D∗  D. Shiromoto also gave the following basic exact sequence: Lemma 3.6. [18]Let C be a code of length n over Zl and M ⊆ N. Then there is an exact sequence as Zl -modules inc f res 0 → C⊥ (M) → V(M) → C∗ → C(N − M)∗ → 0 where the maps inc, res denote the inclusion map, the restriction map, respectively, and f is a Zl -homomorphism such that f : V → D∗ y → ( ŷ : x → x, y ). We can adjust Lemma 3.6 to our case: Lemma 3.7. Let C be a code of length n over Zps and M ⊆ N. Then there is an exact sequence as Zps -modules inc f res 0 → C⊥ (M) → V(M) → C∗ → C(N − M)∗ → 0, where the maps inc, res denote the inclusion map, the restriction map, respectively, and f is a Zps -homomorphism such that f : V → D∗ y → ( ŷ : x → x, y ). Note that for any x ∈ V, if supp(x) ⊆ M ⊆ N, then for any general weight function we have wt(x) ≤ ar |M|. In our case: wL (x) ≤ ps−1 |M| . So we have the following lemma for wL (x): Lemma 3.8. Let C be a code of length n over Zps , then C(M)∗ = 0 for any subset M ⊆ N such that |M| < d/ps−1 , where d is the minimum Lee weight of C. Z. Ö. Özger et al. / Filomat 30:2 (2016), 255–268 261 Proof. For any c , 0 ∈ C supp(c) ps−1 ≥ wL (c) ≥ d. (4) If |M| < d/ps−1 , then d > |M| ps−1 , (5) which means supp(c) ps−1 ≥ d > |M| ps−1 by (4) and (5). But this means supp(c) > |M|, i.e. supp(c) * M. So C ∩ V(M) = {0} and C(M)∗ = HomZps (C ∩ V(M), Zps ) = 0. By Lemma 3.8, we have the following bound: Theorem 3.9. Let C be a code of length n over Zps with the minimum Lee weight d. Then % $ d−1 ≤ n − rank(C). ps−1 Proof. We will follow the steps of Shiromoto in [18]. In the exact sequence of Lemma 3.7, replace C with C⊥ . Then the exact sequence transforms into the following one: f inc res 0 → C(M) → V(M) → (C⊥ )∗ → C⊥ (N − M)∗ → 0. (6) Apply ∗ = HomZps (·, Zps ) and take an arbitrary subset M ⊆ N such that $ % d−1 . |M| = ps−1 Since C(M)∗ = 0 by Lemma 3.8 and V(M)∗  V(M), the exact sequence (6) leads us to the following short exact sequence: 0 → C⊥ (N − M) → C⊥ → V(M) → 0. (7) V(M)  (Zps )|M| is a projective module. Hence (7) is a split, that is, C⊥  C⊥ (N − M) ⊕ V(M). Therefore % d−1 . f ree rank(C ) ≥ f ree rank(V(M)) = |M| = ps−1 $ ⊥ From (3) we have $ % d−1 n − rank(C) ≥ . ps−1 Codes meeting the bound above are called MLDR (Maximum Lee Distance with respect to Rank) codes. The following example illustrates a code which is both MLDS and MLDR.  Example 3.10. Let C be the linear code over Zp , whose generator matrix is G = [1]. So C = (0), (1), · · ·, (p − 1) , $ % $ % d−1 d−1 = log|R| |C| = logp1 |C| = 1, A = p − 1, d = 1, and n = 1. Hence = n − log|R| |C| = 0 and A p1−1 n − logp1 |C| = 0. Z. Ö. Özger et al. / Filomat 30:2 (2016), 255–268 262 4. Kernel and Independence of φL (C) For finite fields and vector spaces the notions of kernel and independence are strongly related (see [13]). In this section, we investigate the same notions for Gray images of linear codes over Zps . The kernel of a code C over Fp , where p is a prime, denoted by K(C), is defined as the set K(C) := {v |v ∈ C, v + C = C } . For further information we refer to [16]. Since φL (C) is a code over Fp (not necessarily linear), we can define n o K(φL (C)) := φL (v) v ∈ C, φL (v) + φL (C) = φL (C) . In [6], authors gave some results about K(φL (C)), φL -independence and modular independence over Z2s . We have similar results for Zps . First we define modular independence. We say that vectors v1 , v2 , . . . , vt are modular independent over t P Zps if αi vi = 0 then αi ∈ hpi for all i. The Gray images of modularly independent vectors on Zps might i=1 not be linearly independent on Zp . For counter example, we refer to [6]. A set of vectors in Zps is said to be φL -independent, if their Gray images are linearly independent over Zp . Lemma 4.1. Let G be the generating matrix of a linear code of type (ps )δ0 (ps−1 )δ1 · · · (p)δs−1 over Zps in standard form. n o Let vi,1 , vi,2 , . . . , vi,δi be the vectors of order ps−i . Then the vectors in the set αvi,j |1 ≤ α ≤ ps−i−1 are φL -independent ps−1 n in Fp . Proof. Let G be the generator matrix of the code in standard form. The Gray images of 1, 2, . . . , ps−1 form an upper triangular matrix and so the Gray image of the vectors in the first δ0 coordinates are linearly independent. All initial nonzero coordinates of submatrices pi Iδi form an uppertriangular matrix and their entries are all less than or equal to ps−1 . Therefore the other cases of the form pi Iδi form submatrix of the above mentioned upper triangular matrix. Hence they are also linearly independent. Theorem 4.2. Let v1 , v2 , . . . , vk be modular independent vectors in Znps . Then there exist modular independent ps−1 n vectors w1 , w2 , . . . , wk which are φL -independent in Fp such that hv1 , v2 , . . . , vk i = hw1 , w2 , . . . , wk i. Proof. Any set of modular independent vectors over Zps are permutationally equivalent to a set of vectors that form a generator matrix in standard form as shown in [15]. Therefore by Lemma 4.1 these vectors are φL -independent. The following proposition gives a restriction to the order of elements whose Gray images belong to K(φL (C)). Proposition 4.3. Let C be a linear code over Zps . If v ∈ C has order greater than p2 then K(φL (C)) does not contain φL (v). Proof. For the rest of the proof and the rest of the paper, let ai b j ck be the codeword of length i + j + k, whose first i entries are a, the next j entries are b, and the remaining k entries are c. Since ord(v) > p2 , v has a number i as its coordinate with ord(i) > p2 . We have the following three cases for i ∈ Zps with ord(i) > p2 : (i) If 0 < i < ps−1 then ord(i) = pk , k > 2, since ord(i)| Zps = ps . That means i = ps−k ui , where (ui , ps ) = 1, i.e., (ui , p) = 1. Since 0 < i < ps−1 φL (i) = 1i 0ps−1 −i , Z. Ö. Özger et al. / Filomat 30:2 (2016), 255–268 263 and since s − k ≤ s − 2, we have pi = ps−k+1 ui < ps . We know that i , ps−1 u j or i , ps−2 u j for any u j such that (u j , p) = 1. So by using division algorithm we can write i = qps−2 + r′ , 0 < r′ < ps−2 , pi = qps−1 + r, 0 < r = pr′ < ps−1 . Without loss of generality assume that i > r. Then, φL (i) + φL (pi) = = 1i 0ps−1 −i + qps−1 + 1r 0ps−1 −r q + 2r q + 1i−r qps−1 −i < φL (C), since r , 0, r − i , 0 and ps−1 − i , 0. Now assume i = r. Then, φL (i) + φL (pi) = = 1i 0ps−1 −i + qps−1 + 1i 0ps−1 −i q + 2i qps−1 −i < φL (C), since i , 0 and ps−1 − i , 0.  s−1 (ii) If ps−1 < i < ps − ps−1 then mp < i < (m + 1)ps−1 , where m ∈ 1, 2, 3, · · ·, p − 2 . Since ord(i) > p2 , i , ps−1 u j  s−2 or i , p u j for any u j ∈ 1, 2, 3, · · ·, p − 2, p − 1 . Let i = mps−1 + r, 0 < r < ps−1 , r = qps−2 + r′ , 0 < r′ < ps−2 . So pi = (mps−1 + r)p = pr = qps−1 + pr′ . Without loss of generality assume that r > pr′ . Then, φL (i) + φL (pi) = 1r 0ps−1 −r + mps−1 + qps−1 + 1pr′ 0ps−1 −pr′ = q + m + 2pr′ q + m + 1r−pr′ q + mps−1 −r < φL (C), since 0 < pr′ < ps−1 , r − pr′ , 0 and ps−1 − r , 0. (iii) If ps − ps−1 < i < ps then 0 < −i < ps−1 . So φL (−i) + φL (−pi) < φL (C) as we proved in the first case. We see that for each v ∈ Znps we have either φL (v) + φL (pv) < φL (C) or φL (−v) + φL (−pv) < φL (C). Hence either φL (v) + φL (C) , φL (C) or φL (−v) + φL (C) , φL (C) when ord(v) > p2 . So the Gray image of the code, which is generated by all vectors of C with order less than or equal to p2 should include K(φL (C)). Then we have the following corollary and lemmas, which generalize the results in [6]: Corollary 4.4. Let C be a linear code over Zps with generator matrix of the form (2).Then K(φL (C)) is contained in the Gray image of the code generated by the matrix:  s−2  s−2 s−2 s−2 ··· ··· ps−2 A0,s   p Iδ0 p A0,1 p A0,2 p A0,3   s−2 s−2 s−2 s−2 0 p Iδ1 p A1,2 p A1,3 ··· ··· p A1,s     s−2 s−2 s−2 0 0 p Iδ2 p A2,3 ··· ··· p A2,s      · · ·  . ··· 0 ··· ··· ··· ···     · · · · · · · · · · · · · · · · · · · · ·   s−2 s−2 s−2   0 0 0 · · · p I p A p A δs−2 s−2,s−1 s−2,s    0 0 0 ··· 0 ps−1 Iδs−1 ps−1 As−1,s 264 Z. Ö. Özger et al. / Filomat 30:2 (2016), 255–268 Lemma 4.5. Let C be a linear code over Zps and v, w ∈ C. Then we have φL (ps−1 v + w) = φL (ps−1 v) + φL (w) for each v, w ∈ C. Proof. Let vi , wi ∈ Zps be the ith coordinates of v, w respectively. Then by division algorithm we can write wi = qw ps−1 + rw , 0 ≤ qw ≤ p − 1, 0 ≤ rw < ps−1 , vi = qv p + rv , 0 ≤ qv < ps−1 , 0 ≤ rv < p. So ps−1 vi = ps−1 rv , where 0 ≤ ps−1 rv < ps . Therefore φL (ps−1 vi + wi ) = φL (ps−1 rv + qw ps−1 + rw ) = φL (ps−1 (rv + qw ) + rw ) = rv + qw ps−1 + 1rw 0ps−1 −rw = rv ps−1 + qw ps−1 + 1rw 0ps−1 −rw = φL (ps−1 rv ) + φL (qw ps−1 + rw ) = φL (ps−1 vi ) + φL (wi ). Applying this method coordinate-wise, the result follows. Theorem 4.6. Let C be a linear code over Zps with the generator matrix of the form (2). Then the Gray image of the code C′ generated by  s−1  p Iδ0  0   0    · · ·   · · ·   0  0 ps−1 A0,1 ps−1 Iδ1 0 ··· ··· 0 0 ps−1 A0,2 ps−1 A1,2 ps−1 Iδ2 0 ··· 0 0 ps−1 A0,3 ps−1 A1,3 ps−1 A2,3 ··· ··· ··· ··· ··· ··· ··· ··· ··· ps−1 Iδs−2 0 ··· ··· ··· ··· ··· ps−1 As−2,s−1 ps−1 Iδs−1 is a linear subcode of K(φL (C)). ps−1 A0,s ps−1 A1,s ps−1 A2,s ··· ··· ps−1 As−2,s ps−1 As−1,s              (8) Proof. Let v, w ∈ C, then ps−1 v ∈ C′ ⊆ C. Then φL (ps−1 v) ∈ φL (C′ ) and φL (w) ∈ φL (C). By Lemma 4.5 φL (ps−1 v + w) = φL (ps−1 v) + φL (w) ∈ φL (C), since ps−1 v, w ∈ C. This holds for every w ∈ C, which means φL (ps−1 v) + φL (C) ⊆ φL (C). Two different codewords will have different images. Therefore φL (ps−1 v) + φL (C) = φL (C), which tells us that φL (ps−1 v) ∈ K(φL (C)). Lemma 4.7. Let C be a linear code over Zps , λ ∈ Zps and v ∈ C such that φL (v) < K(φL (C)). Then φL (λv) ∈ K(φL (C)) if and only if ord(λv) = p. Proof. (=⇒)Suppose that ord(λv) = p, then φL (λv) ∈ K(φL (C)) by Theorem 4.6. (⇐=)Now assume φL (v) < K(φL (C)) and φL (λv) ∈ K(φL (C)). We have two cases. (i) If ord(v) > p2 and v = (v1 , v2 , · · ·, vn ), then there exists vi , 1 ≤ i ≤ n, such that ord(vi ) > p2 . Let ord(vi ) = pk with k > 2. Then vi = ps−k ui , where ui is a unit. By division algorithm, we have ui = qu p + ru , 0 ≤ qu ≤ ps−1 − 1, 0 < ru < p, vi = qv ps−1 + rv , 0 ≤ qv ≤ p − 1, 0 < rv < ps−1 , where ru , 0, since ui is a unit and rv , 0, since ord(vi ) > p2 . If φL (λvi ) ∈ K(φL (C)), then by Proposition 4.3 λ = pk−2 uλ or λ = pk−1 uλ , where uλ is a unit. For λ = pk−1 uλ we have ord(λvi ) = p, so φL (λvi ) ∈ 265 Z. Ö. Özger et al. / Filomat 30:2 (2016), 255–268 K(φL (C)) by Theorem 4.6. If λ = pk−2 uλ , then ord(λvi ) = p2 and λvi = ps−2 uλ ui = qu uλ ps−1 + ru uλ ps−2 , where 0 < ru uλ ps−2 < ps−1 . Without loss of generality assume that ru uλ ps−2 < rv , then we have φL (λvi ) + φL (vi ) = (qu + qv + 2)ru uλ ps−2 (qu + qv + 1)rv −ru uλ ps−2 (qu + qv )ps−1 −rv < φL (C), since ru uλ ps−2 , 0, rv − ru uλ ps−2 , 0, ps−1 − rv , 0. If ru uλ ps−2 = rv , then φL (λvi ) + φL (vi ) = (qu + qv + 2)rv (qu + qv )ps−1 −rv < φL (C), since ps−1 − rv , 0, rv , 0. (ii) If ord(v) = p2 and v = (v1 , v2 , · · ·, vn ), then there exists vi , 1 ≤ i ≤ n, such that ord(vi ) = p2 . Then vi = ps−2 ui , where ui is a unit. By division algorithm, we have vi = qv ps−1 + rv , 0 ≤ qv ≤ p − 1, 0 < rv < ps−1 , since ord(vi ) = p2 . If φL (λvi ) ∈ K(φL (C)), then by Proposition 4.3 ord(λvi ) = p2 or ord(λvi ) = p. If ord(λvi ) = p, we have φL (λvi ) ∈ K(φL (C)) by Theorem 4.6. If ord(λvi ) = p2 then λ is a unit and λvi = ps−1 q + r, 0 < r < ps−1 , r , 0, since ord(λvi ) = p2 . Without loss of generality assume that rv > r, then we have φL (λvi ) + φL (vi ) = (q + qv + 2)r (q + qv + 1)rv −r (qu + qv )ps−1 −rv < φL (C), since r , 0, rv − r , 0, ps−1 − rv , 0. If r = rv , then φL (λvi ) + φL (vi ) = (qu + qv + 2)rv (qu + qv )ps−1 −rv < φL (C), since ps−1 − rv , 0, rv , 0. In both cases φL (λv) + φL (v) < φL (C), whenever ord(λv) , p. Theorem 4.8. Let C be a linear code over Zps of type (ps )δ0 (ps−1 )δ1 · · · (p)δs−1 . If m = dim(K(φL (C))), then   s−1 s−1 s−1 s−1 s−1   X X X    X X δ , m∈ δ + 1, . δ + 2, · · ·, δ + δ − 2, δ + δ i i i i s−2 i s−2      i=0 i=0 i=0 i=0 i=0 Proof. By Theorem 4.6, the image of any codeword of order p is in K(φL (C)). If there is a codeword v of order greater than p2 , then φL (v) < K(φL (C)). Moreover, if φL (v) < K(φL (C)), then φL (λv) ∈ K(φL (C)) if and only if ord(λv) = p by Lemma 4.7. Otherwise φL (λv) + φL (v) < φL (C). So for φL (v) < K(φL (C)) and φL (λv) ∈ φL (C”) ⊆ s−3 P K(φL (C)) we have ord(λv) = p. This means we have the Gray images of first δi vectors of (8) as generators i=0 of K(φL (C)). Furthermore, we can show that the contribution of the Gray images of first s−3 P i=0 δi vectors of (2) to K(φL (C)) is restricted to that. To see this, let v be one these vectors in (2). Then ord(v) > p2 and φL (v) < K(φL (C)) by Proposition 4.3. For any scalar product of v, say λv, then φL (λv) ∈ K(φL (C)) if and only if ord(λv) = p by Lemma 4.7. If ord(v) = pk , k > 2, v = uv pk , this happens only when λ = ps−k−1 uλ , where uλ and uv are units. Therefore λv = ps−1 u, where u = uv uλ is a unit too. This shows that the only contribution of the Gray image of v to K(φL (C)) is its scalar products with the ps−1 u and their linear combinations. Also we know that the Gray image of the last δs−1 rows of (8) are generators of K(φL (C)) by Theorem 4.6. We don’t know whether each of the Gray images of δs−2 remaining vectors generate K(φL (C)) certainly. But we know that if their Gray images are not included in generators of K(φL (C)), the Gray image of their scalar products with pu, where u is a unit, are all included in K(φL (C)). Hence we can have at least the Gray image of the code generated by (8), and at most the Gray image of the code generated by  s−1  s−1 s−1 s−1 ··· ··· ps−1 A0,s   p Iδ0 p A0,1 p A0,2 p A0,3   ··· ··· ps−1 A1,s  0 ps−1 Iδ1 ps−1 A1,2 ps−1 A1,3    0 0 ps−1 Iδ2 ps−1 A2,3 ··· ··· ps−1 A2,s     · · ·  ··· 0 ··· ··· ··· ··· (9)   · · ·  ··· ··· ··· ··· ··· ···      0 0 0 ··· ps−2 Iδs−2 ps−2 As−2,s−1 ps−2 As−2,s    0 0 0 ··· 0 ps−1 Iδs−1 ps−1 As−1,s Z. Ö. Özger et al. / Filomat 30:2 (2016), 255–268 266 as K(φL (C)). Thus we have the following bound for m: s−1 P p i=0 δi m ≤p ≤p s−1 P i=0,i,s−2 δi · p2δs−2 , which means s−1 X δi ≤ m ≤ s−1 X δi + δs−2 . i=0 i=0 e be the code generated by matrix (9). Since K(φL (C)) is at most φL (C), e K(φL (C)) ⊆ φL (C). e So, Let C n o e : φL (c) + φL (C) = φL (C) . K(φL (C)) = c ∈ C e namely hv0 , v1 , · · ·, vk i = φL (C), e which means dim(φL (C)) e = Let {v0 , v1 , · · ·, vk } be the set of generators of φL (C), k + 1. Assume that dim(K(φL (C))) = k, and without loss of generality let K(φL (C)) = hv1 , · · ·, vk i. If e ⊆ φL (C), then we have v0 + vi ∈ φL (C) for all i = 1, · · ·, k, since vi ∈ K(φL (C)). But v0 + vi ∈ φL (C) e v0 ∈ φL (C) e for all i = 1, · · ·, k, that means v0 ∈ K(φL (C)) ⊆ K(K(φL (C))) ⊆ K(φL (C)), which is a contradiction. Hence s−1 P δi + δs−2 − 1. Therefore we have the following m, i=0   s−1 s−1 s−1 s−1 s−1   X X X   X X  . δ + δ m∈ δ + δ − 2, δ + 2, · · ·, δ + 1, δ , i s−2  i s−2 i i i     i=0 i=0 i=0 i=0 i=0 5. Linearity and Duality of φL (C) Self-dual codes are important since many of the best codes known are of this type. Numerous researchers have investigated their Gray images to find (not necessarily linear) codes with optimal or extremal parameters. Most of the best codes are nonlinear and they can be viewed as Gray images of linear codes. On the other hand, linearity makes things easier. Therefore it is also very important to know when the image φL (C) is nonlinear/linear. Also some researchers looked into when the images of self-dual codes are also self-dual. The aim of this section is to present some knowledge about these two topics for codes over Zps . Theorem 5.1. Let C be a linear code over Zps with the generating matrix of the form given in (2). If δi > 0 for any 0 ≤ i ≤ s − 3 then φL (C) is not linear. Proof. We have elements v ∈ C such that ord(v) > p2 by Proposition 4.3, so by Lemma 4.7 they are not in K(φL (C)), since ord(v) > p2 > p. Hence φL (C) is not a linear code overFp . Definition 5.2 (Free Code). A code C over Zps is said to be a free code if rank(C) = f ree rank(C). Theorem 5.3. Let C be a linear code over Zps . If p > 2 then the Gray image of a free code is not linear. Proof. If C is a free code, then it has a generating matrix of the form h i G = I δ0 A , where A is an δ0 × (n − δ0 ) matrix over Zps . Let vi be the ith row of G. Since every row of G is a codeword, if φL (C) is linear then −φL (v1 ) must be included in φL (C). But −φL (v1 ) = (−φL (1), −φL (v1,2 ), · · ·, −φL (v1,n )) < φL (C), because −φL (1) , −φL (x) for any x ∈ Zps . Z. Ö. Özger et al. / Filomat 30:2 (2016), 255–268 267 The image of a self-dual code C over Zps under the Gray map only has the cardinality of a self-dual code if p = 2 and s = 2, since a self-dual code should include exactly half of the ambient space, which means sn 2 ps−1 n = 2 . This implies s = ps−1 and hence p = s = 2. So for p > 2 we know that none of the self-dual codes has self-dual Gray image. However a code might have a self-dual Gray image if it is not self-dual. Example 5.4. Let C be the linear code over Z27 generated by " # 3 0 6 9 G= . 0 3 3 0 Since h(3, 0, 6, 9), (0, 3, 3, 0)i = 18 , 0 on Z27 , C is not self-dual code over Z27 . However D E φ (3, 0, 6, 9), φL (0, 3, 3, 0) = 0, D L E φ (3, 0, 6, 9), φL (0, 3, 3, 0) = 0, D L E φL (0, 3, 3, 0), φL (0, 3, 3, 0) = 0, on Z3 . The following theorem gives specific type of linear codes over Zps , whose Gray images are selforthogonal codes over Zps . δ1  δ2   δs−2
δ Theorem 5.5. Any linear code C over Zps of type ps−1 p s−1 has a Gray image that is a ps−2 · · · p2 self-orthogonal code. δ2  δs−2
δ  δ1   p s−1 , then it has a generating matrix of the form ps−2 · · · p2 Proof. If C is of type ps−1   ··· ··· pA1,s   pIδ1 pA1,2 pA1,3  0 2 2 2 ··· ··· p A2,s  p Iδ2 p A2,3     0  0 ··· ··· ··· ···  . G =  ··· ··· ··· ··· ···  · · ·    0 0 ps−2 Iδs−2 ps−2 As−2,s−1 ps−2 As−2,s   0   s−1 s−1 0 0 0 0 p Iδs−1 p As−1,s s−i2 s−i1 Let v = n(v1 , . . . , vn ), w = (w1 , .o. . , wn ) ∈ C are rows n of G with order p o and p , where i1 ≥ i2 ≥ 1. So each s i2 i2 i2 s i1 i1 i1 vk is in 0, p , 2p , . . . , p − p and each wk is in 0, p , 2p , . . . , p − p , where 1 ≤ k ≤ n. For any element m in Zps of order ps−e we have φL (m) = (q + 1)pe t (q)(ps−1−e −t)pe , D E where m = ps−1 q + r, 0 ≤ r = pe t < ps−1 , 0 ≤ q ≤ p − 1. We will consider φL (vk ), φL (wk ) instead of E D E D φL (v), φL (w) , since φL (v) = (φL (v1 ), . . . , φL (vn )), φL (w) = (φL (w1 ), . . . , φL (wn )), and therefore φL (v), φL (w) = E n D P φL (vi ), φL (wi ) . In both Gray images the number of successively repeated coordinates are divisible by i=1 a power of p (at least by p). So in coordinatewise product φL (vk ) · φL (wk ) = (vk,1 wk,1 , . . . , vk,ps−1 wk,ps−1 ) the coordinates will be repeated at least p times successively. So φL (vk ) · φL (wk ) = ((a1 )p , (a2 )p , . . . , (aps−2 )p ), where al is the lth repeating coordinate. Hence s−2 s−1 p p   X D E X pa j = 0, φL (vk ) · φL (wk ) = φL (vk ), φL (wk ) = i=1  ⊥ which means φL (C) ⊆ φL (C) . i j=1 Acknowledgement: The authors would like to thank the anonymous referees for their valuable comments and suggestions. Z. Ö. Özger et al. / Filomat 30:2 (2016), 255–268 268 References [1] J. Borges, C. Fernández, J. Rifà, Every Z2k -code is a binary propelinear code, Electronic Notes in Discrete Mathematics 10 (2001) 100–102. [2] J. Borges, C. Fernández, J. Rifà, Propelinear structure of Z2k -linear codes, Technical Report arxiv:0907.5287 (2009). [3] J. Borges, K. T. Phelps, J. Rifà, The rank and kernel of extended 1-perfect Z4 -linear and additive non- Z4 -linear codes, IEEE Trans. Inform. Theory 49(8) (2003) 2028–2034. [4] C. Carlet, Z2k -linear codes, IEEE Trans Inform Theory, 44 (1998) 1543–1547. [5] I. Constantinescu, W. Heise , A metric for codes over residue class rings of integers, Problemy Peredachi Informatsii 33 (1997) 22–28. [6] S. T. Dougherty, C. Fernández-Córdoba, Codes over Z2k gray map and self-dual codes, Adv. Math. Comm 5 (2011) 571–588. [7] S. T. Dougherty, H. Liu, Independence of vectors in codes over rings, Design Codes and Cryptography 51 (2009) 55–68. [8] S. T. Dougherty, K. Siromoto, MDR codes over Zm , IEEE Trans Inform Theory 46(1) (2000) 265–269. [9] S. T. Dougherty, J-L. Kim, H. Kulosman, MDS codes over finite principal ideal rings, Designs Codes and Cryptography 50 (2009) 77–92. [10] C. Fernández-Córdoba, J. Pujol, M. Villanueva, On rank and kernel of Z4 -linear codes, Lecture Notes in Computer Science 5228 (2008) 46–55. [11] C. Fernández-Córdoba, J. Pujol, M. Villanueva, Z2 Z4 -linear codes: rank and kernel, Design Codes and Cryptography 56(1) (2009) 43–59. [12] A. R. Hammons, V. Kumar, A. R. Calderbank, N. J. A. Sloane, P. Solé, The Z4 -linearity of Kerdock, Preparata, Goethals, and related codes, IEEE Trans Inform Theory 40 (1994) 301–319. [13] K. M. Hoffman, R. Kunze, Linear Algebra, (2nd Edition), Prentice-Hall Inc., Englewood Cliffs, New Jersey, 1971. [14] W. C. Huffman, Decompositions and extremal Type II codes over Z4 , IEEE Trans. Inform. Theory 44 (1998) 800–809. [15] Y. H. Park, Modular independence and generator matrices for codes over Zm , Design Codes and Cryptography 50(2) (2009) 147–162. [16] K. T. Phelps, J. Rifà and M. Villanueva, Kernels and p-kernels of pr -ary 1-perfect codes, Design Codes and Cryptography 37(2) (2005) 243–261. [17] K. T. Phelps, J. Rifà, M. Villanueva, On the additive Z4 -linear and non- Z4 -linear Hadamard codes: Rank and kernel, IEEE Trans. Inform. Theory 55(1) (2005) 316–319. [18] K. Shiromoto, A basic exact sequence for the Lee and Euclidean weights of linear codes over Zl , Linear Algebra and its Applications 295 (1999) 191–200. [19] K. Shiromoto, Singleton bounds for codes over finite rings, Journal of Algebraic Combinatorics 12 (2000) 95–99. [20] J. F. Voloch, J. L. Walker, Homogeneous weights and exponential sums, Finite Fields Appl 9 (2003) 310–321. [21] B. Yıldız, A Combinatorial construction of the Gray map over Galois rings, Discrete Mathematics 309(10) (2009) 3408–3412. [22] B. Yıldız, Z. Ödemiş Özger, Generalization of the Lee weight to Zpk , TWMS J. App.&Eng. Math. 2(2) (2012) 145–153. [23] B. Yıldız, Z. Ödemiş Özger, Linear codes over Z2s with the extended Lee weight, AIP Conf. Proc 1389 (2011) 621–624. DOI:10.1063/1.3636807.