11.07.2015 Views

On relations between certain exponential sums and multiple ...

On relations between certain exponential sums and multiple ...

On relations between certain exponential sums and multiple ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>On</strong> <strong>relations</strong> <strong>between</strong> <strong>certain</strong> <strong>exponential</strong><strong>sums</strong> <strong>and</strong> <strong>multiple</strong> Kloosterman <strong>sums</strong><strong>and</strong> some applications to coding theoryMarko MoisioAbstract. In this paper we consider some <strong>relations</strong> <strong>between</strong> <strong>multiple</strong> Kloosterman<strong>sums</strong> <strong>and</strong> <strong>certain</strong> <strong>exponential</strong> <strong>sums</strong> with monomial <strong>and</strong> binomial arguments,<strong>and</strong> some applications of these <strong>relations</strong> to coding theory.1. IntroductionLet F = F q denote the finite field with q elements <strong>and</strong> E = F qm be an extension ofdegree m>1ofF . In this paper we consider <strong>relations</strong> <strong>between</strong> <strong>certain</strong> <strong>exponential</strong><strong>sums</strong>S(f) := ∑ e E (f(x)),x∈E<strong>and</strong> (<strong>multiple</strong>) Kloosterman <strong>sums</strong>K m (a) :=∑x 1 ,...,x m ∈F ∗ e F (x 1 + ···+ x m + ax −11 ...x −1m ),where e E (resp. e F ) is the canonical additive character of E (resp. F ), f(X) ∈E[X], a ∈ F ,gcd(q, deg f(X)) = 1, <strong>and</strong> some applications of these <strong>relations</strong> tocoding theory.More precisely, we shall construct a class of binary irreducible cyclic codes <strong>and</strong>two classes of binary cyclic codes (by means of <strong>certain</strong> well known irreducible cyclic1


2codes), <strong>and</strong> estimate the weights of the words of these codes by using the Weil <strong>and</strong>the Deligne bounds, obtained by Weil in [12] <strong>and</strong> by Deligne in [1]:|S(f)| ≤(deg f(X) − 1)q m/2 , (1)|K m (a)| ≤(m +1)q m/2 . (2)It turns out e.g. that the weight distribution of the binary cyclic code constructedby means of two simplex codes of different length is closely related to the values ofa <strong>multiple</strong> Kloosterman sum over a field of characteristic 2.2. <strong>On</strong> <strong>relations</strong> <strong>between</strong> <strong>certain</strong> <strong>exponential</strong><strong>sums</strong> <strong>and</strong> <strong>multiple</strong> Kloosterman <strong>sums</strong>We assume that F , E, e F , e E <strong>and</strong> K m () are fixed as in the introduction. Wealso denote the canonical additive character of E (resp. F )simplybye if this doesnot cause any confusions.to F .Let Tr E/F <strong>and</strong> N E/F denote the trace <strong>and</strong> norm mappings, respectively, from ELet K be any finite field. We denote the multiplicative character group of K by ̂K<strong>and</strong> the identity element of that group by χ 0 .Ifψ is an additive character of K <strong>and</strong>χ ∈ ̂K, we denote the Gauss sum over K associated to these characters by G K (χ, ψ).If e is the canonical additive character of K, wedenoteG K (χ) =G K (χ, e). Bynotation e a (a ∈ K) we mean the character defined by e a (x) =e(ax) for all x ∈ K.We do not define χ(0), <strong>and</strong> consequently Gauss <strong>sums</strong> are calculated over K ∗ .Let ψ be a non-trivial additive character of K. The orthogonality <strong>relations</strong> ofcharacters, see [4, p. 195], implyψ(x) =1|K|−1∑G(χ, ψ)χ(x) ∀x ∈ K ∗ , (2.1)χ∈ ̂KWe shall also need the following two results:Theorem 2.1. Let d ||K| −1, <strong>and</strong>letH denote the subgroup of order d of ̂K.Then∑x∈K ∗ ψ(ax d )= ∑ χ∈HG(χ, ψ)χ(a) ( = ∑ χ∈HG(χ, ψ)χ(a) )


3for all a ∈ K ∗ .Proof. See [4, p. 217].Lemma. Let γ be a primitive element of K, a ∈ K <strong>and</strong> d ||K|−1. Then∑e(ax d )=dx∈K ∗|K|−1d∑−1e(aγ di )=d∑e(x).i=0x∈aProof. Obvious.Let d | q − 1<strong>and</strong>H (resp. H ′ ) be the subgroup of order d of ̂F (resp. Ê). Thesurjectivity of N E/F implies H ′ = {χ ◦ N E/F | χ ∈ H}.Denote N = N E/F . By Theorem 2.1 <strong>and</strong> the Davenport-Hasse theorem [4, p.197-199], we now have∑e E (ax d )= ∑ G E (χ ◦ N)χ(N(a)) = (−1) ∑ m−1 G F (χ) m χ(N(a)). (2.2)x∈E ∗ χ∈H χ∈HAssume now that d = q − 1 <strong>and</strong> consequently H = ̂F . Let x 1 ,...,x m−1 ∈ F ∗<strong>and</strong> denote b =N(a). It follows from (2.1) thate F (bx −11 ...x −1m−1 )= 1q − 1∑G F (χ)χ(x 1 ) ...χ(x m−1 )χ(b).χ∈ ̂FBy multiplying both sides of the preceding equation by e F (x 1 + ···+ x m−1 )<strong>and</strong>bysumming we obtain∑e F (x 1 + ···+ x m−1 + bx −11 ...x −1m−1 )= 1 ∑G F (χ) m χ(b).q − 1x 1 ,...,x m−1 ∈F ∗Thus we have provedTheorem 2.2.∑χ∈ ̂Fx∈E ∗ e(ax q−1 )=(−1) m−1 (q − 1)K m−1 (N(a)) ∀ a ∈ E ∗ .The surjectivity of N,the Weil bound <strong>and</strong> the Deligne bound, now imply


4Corollary 2.3.|K m (a)| ≤min{q m+12 − q m+12 − 1, (m +1)q m/2 } ∀ a ∈ F ∗ .q − 1This result is, of course, valid for any non-trivial additive character ψ of F <strong>and</strong>also for <strong>sums</strong>∑ψ(a 1 x 1 + ···+ a m x m−1 + bx −11 ...x −1m−1 ), ∀ a i,b∈ F ∗ .x 1 ,...,x m−1 ∈F ∗Mordell [10] proved that |K m (a)| ≤ q (m+1)/2 ,whenq is prime. Thus our estimategeneralizes <strong>and</strong> slightly improves this bound.Corollary 2.4. Let a ∈ E ∗ <strong>and</strong> d | q − 1. Then∑∣ e(ax d ) ∣ ≤ min{(d − 1)q m/2 +1,m( √ q − 1/ √ q)q m/2 }.x∈E ∗Proof. Let γ be a primitive element of E. DenoteT =< γ q−1 > <strong>and</strong> t =(q − 1)/d.We now have a partition = ⋃ t−1i=0 γdi T . By the Lemma <strong>and</strong> Theorem 2.2, wehave∑x∈E ∗ e(ax d )=d∑t−1∑i=0 x∈γ di T∑t−1e(ax) =(−1) m−1 d K m−1 (N E/F (aγ di )).i=0The claim follows now from the Deligne bound <strong>and</strong> the Weil bound.□Next we shall consider <strong>certain</strong> <strong>exponential</strong> <strong>sums</strong> with binomial arguments. Weneed the following result which is a generalization of a result of C.J. Moreno <strong>and</strong>O. Moreno [11, Theorem 9].We remark that the result can also be proved by using the results of R.J.McEliecein [8], where he used Gauss <strong>sums</strong> to determine the weight distributions of <strong>certain</strong>irreducible cyclic codes. We give, howewer, a direct proof for its simplicity <strong>and</strong>shortness. The proof rely on the 19th century theorem of Stickelberger, consideringthe values of <strong>certain</strong> Gauss <strong>sums</strong> [4, p. 202-203].


Assume 2 | m. Then there exists an intermediate field M of E over F satisfying[M : F ]=2.Theorem 2.5. Let a ∈ E ∗ <strong>and</strong> assume that m is even. If d | q +1,then⎧∑ ⎨ (−1) m/2 q m/2if ind a ≢ k (d),e(ax d )=⎩(−1) m/2−1 (d − 1)q m/2 if ind a ≡ k (d),x∈Ewhere k =0if(1) 2 | q; or2 ∤ q <strong>and</strong> m ≡ 0(4);or2 ∤ q, m ≡ 2(4)<strong>and</strong> 2 | (q +1)/d,<strong>and</strong> k = d/2 if(2) 2 ∤ q, m ≡ 2(4)<strong>and</strong> 2 ∤ (q +1)/d.5Proof. Let H be the subgroup of order d of ̂M. By (2.2) we have∑e(ax d )=(−1) ∑m/2−1x∈Eχ∈H ∗ G M (χ) m/2 χ(N(a)),where H ∗ := H \{χ 0 } <strong>and</strong> N := N E/M .Let χ ∈ H ∗ . Since ord(χ) | q + 1, we observe that Stickelbergers theorem isapplicable.Now, if 2 | q or 2 | m/2, then G M (χ) m/2 = q m/2 . To consider the remainingcases, we fix a generator of ̂M, sayλ, <strong>and</strong> denote t =(q 2 − 1)/d.Now χ = λ tj for some j ∈{1,...,d− 1}. Since ord(χ) =d/ gcd(d, j), we seethat (q +1)/ ord(χ) iseven,if(q +1)/d is even. Consequently, G M (χ) m/2 = q m/2 ,if (q +1)/d is even.Thusinthecase(1)wehave∑d−1∑e(ax d )=(−1) m/2−1 q m/2 λ tj (N(a)).x∈Ej=1Inthecase(2)(q +1)/ ord(χ) is even if <strong>and</strong> only if j is even. Thus∑d−1∑e(ax d )=(−1) m/2−1 q m/2 (−1) j λ tj (N(a)).x∈Ej=1By observing that, if γ is a primitive element of E then N(γ) is a primitiveelement of M, we easily obtain the result. □


6Proposition 2.6. Assume that m is even. Then⎧∑e(ax qm −1 ⎨ q m − 1 if Tr E/M (a) =0,q+1 )=⎩ − qm − 1x∈E ∗ q +1 K (1 NM/F (Tr E/M (a)) ) if Tr E/M (a) ≠0.Proof. Denote Tr = Tr E/M ,N=N E/M <strong>and</strong> d =(q m − 1)/(q +1). Ase E (ax d )=e M (Tr(a)N(x) q−1 ), we see that∑e E (ax d )= qm − 1 ∑eq 2 M (Tr(a)x q−1 ).− 1x∈E ∗ x∈M ∗The claim follows now from Theorem 2.2.□Theorem 2.7. Let a, b ∈ E, b ≠0<strong>and</strong> assume that m is even. Then⎧∑e(ax qm −1 ⎨ − 1 if Tr E/M (a) =0,q+1 +bx) =⎩x∈E(−1) m ∗ 2 −1 (−1) me(±c)q m 2 −1 q m 2 +12 + K 1 (h), if Tr E/M (a) ≠0,q +1where c = ab −(qm −1)/(q+1) , h =N M/F (Tr E/M (c)) <strong>and</strong> the ” − ” sign holds if <strong>and</strong>only if 2 ∤ q <strong>and</strong> m ≡ 2(4).Proof. Denote t =(q m − 1)/(q +1) <strong>and</strong>Tr=Tr E/M . If Tr(a) =0,thene(ax t )=e M (Tr(a)x t ) = 1 for all x ∈ E, <strong>and</strong> the claim follows. Assume that Tr(a) ≠0.Since the mapping x ↦→ bx is a permutation of E ∗ , it is enough to consider <strong>sums</strong>S := ∑x∈E ∗ e(cx t + x).Let us fix a primitive element of E, sayγ. Denote T =< γ q+1 >. Now we havea partition E ∗ = ⋃ j∈J γj T ,whereJ := {0,...,q}. Denote J ∗ = J \{k}, wherek =(q +1)/2 if2∤ q <strong>and</strong> m ≡ 2 (4), <strong>and</strong> otherwise k =0. NowS = ∑ e(cγ tj ) ∑j∈Jx∈γ j Te(x) =e(±c) ∑x∈γ k Te(x)+ ∑j∈J ∗ e(cγ tj ) ∑where the ” − ” sign holds if <strong>and</strong> only if 2 ∤ q <strong>and</strong> m ≡ 2(4).Since∑x∈γ j Te(x) = 1 ∑e(γ j x q+1 )q +1x∈E ∗x∈γ j Te(x),


7by the Lemma, it follows from Theorem 2.5 thatS = e(±c)((−1)m/2−1 q m/2+1 − 1)q +1+ (−1)m/2 q m/2 − 1( ∑ e(cγ tj ) − e(±c)).q +1j∈JThe claim follows now easily from the Lemma <strong>and</strong> Proposition 2.6.□Consider next the <strong>sums</strong>S(a, b) := ∑e(ax qm −1q−1+ bx) a, b ∈ E, q > 1.x∈E ∗Assume b ≠ 0. Again it is enough to consider <strong>sums</strong> S(c, 1), where c = ab −(qm −1)/(q−1) .Denote Tr = Tr E/F <strong>and</strong> N = N E/F . Assume that Tr(a) ≠0. Letγ be a primitiveelement of E, <strong>and</strong> denote T =< γ q−1 > <strong>and</strong> g =N(γ). Now∑q−2S(c, 1) = e(cg i ) ∑i=0x∈γ i TIt follows from the Lemma <strong>and</strong> Theorem 2.2 thatThus∑x∈γ i Te(x).e(x) =(−1) m−1 K m−1 (g i ).S(c, 1) = (−1) m−1 ∑ q−2i=0 e F (Tr(c)g i )K m−1 (g i )=(−1) m−1∑x 1 ,...,x m ∈F ∗ e F (x 1 + ···+ x m−1 +Tr(c)x m + x −11 ...x −1m−1 x m)= −1+(−1) m−1 ∑x 1 ,...,x m−1 ∈F ∗ e F (x 1 + ···+ x m−1 ) ∑The next two theorems follows now easily.x m ∈Fe F((Tr(c)+x−11 ...x −1m−1 )x m).Theorem 2.8. Let a, b ∈ E, b ≠0.Then∑{ − 1 if a + a q =0,e(ax q+1 + bx) =x∈E ∗ − qe F (−b q+1 (a + a q ) −1 ) − 1 if a + a q ≠0.


8Theorem 2.9. Let a, b ∈ E, b ≠0. Assume that m>2. Then{∑e(ax qm −1 − 1 if TrE/F (a) =0,q−1 +bx) =x∈E (−1) m−1 qK ∗ m−2 (− N E/F (b)Tr E/F (a) −1 ) − 1 if Tr E/F (a) ≠0.The discussion concerning <strong>sums</strong> S(a, b) is completed by a (trivial)Proposition 2.10. If q>2 then⎧∑e(ax qm −1 ⎨ q m − 1 if Tr E/F (a) =0,q−1 )=⎩ − qm − 1if Trx∈E ∗ E/F (a) ≠0.q − 1Remark. From the result of Theorem 2.8 it is easy to determine the distributionof the even cor<strong>relations</strong> of a set of PN-sequences, so called small Kasami set, bothin binary <strong>and</strong> non-binary cases. (c.f. [2], [6]).3. Three examplesLet us first consider some basic facts about the trace codes. For a more completedescription see [2].We restrict our considerations to the binary trace codes. Let K be the finitefield with 2 k elements <strong>and</strong> fix a primitive element of K, sayγ. Let Tr denote thetrace mapping from K to the prime field F 2 . Let P be an additive subgroup ofK[X] <strong>and</strong>n ∈ Z + . Define a trace code C(P ):={c(f) | f ∈ P }, wherec(f) =(Tr(f(1)), Tr(f(γ)),...,Tr(f(γ n−1 ))) <strong>and</strong> n ∈ Z + . It is easy to see that wt(c), the(Hamming) weight of a codeword c ∈ C(P ), is equal to 1 2 (n − ∑ n−1i=0 e K(f(γ i ))) [2].The next theorem is also from [2].Theorem 3.1. Set 2 k − 1=nN. The dual of the binary cyclic code B of length nwith zeros γ Ns 1,...,γ Ns uis the code C(P ), whereP = { ∑ ui=1 a ix Ns i| a i ∈ K}.Let q =2 r .ChooseK = F qm <strong>and</strong> P = {ax q−1 | a ∈ K}. It follows from Theorem3.1 that the dual of the binary cyclic code of length n =(q m −1)/(q −1) with zeroes


γ (q−1)2i , i =1, 2,... is the code C(P ). It is easy to see that ord n (2) = rm. ThusC(P ) is an irreducible cyclic code of dimension rm [5, p. 77-78]. Sincen−1∑i=0e(aγ (q−1)i )= 1q − 1∑x∈K ∗ e(ax q−1 )by the Lemma, it follows from Theorem 2.2 <strong>and</strong> Corollary 2.4, that|2wt(c) − qm − 1q − 1 |≤min{mq(m−1)/2 , √ qq (m−1)/2 − qm/2 − 1q − 1 }for all c ∈ C(P ) \{(0,...,0)}.In the case m =2,C(P ) is the dual of the Zetterberg code [7, p. 206]. This dualhas been studied at least in [9] <strong>and</strong> [3].9Let us assume that m is even. Denote d =(q m − 1)/(q +1), <strong>and</strong> choose K = F qm<strong>and</strong> P = {ax d + bx | a, b ∈ K}. The dual of the binary cyclic code of length q m − 1with zeroes γ d2i ,γ 2i , i =1, 2,... is the code C(P ).Denote P 1 = {bx | b ∈ K} <strong>and</strong> P 2 = {ax d | a ∈ K}. The weight of any nonzerocodeword of the code C(P 1 )isequaltoq m−1 . By Proposition 2.6 <strong>and</strong> theDeligne bound there is no codeword of that weight in the code C(P 2 ). Since C(P 1 )<strong>and</strong> C(P 2 ) are subgroups of C(P )wehaveC(P )=C(P 1 ) ⊕ C(P 2 ). Obviously|C(P 1 )| = q m , <strong>and</strong> by Proposition 2.6 |C(P 2 )| = q 2 . Thus |C(P )| = q m+2 . ByProposition 2.6, Theorem 2.7 <strong>and</strong> the Deligne bound, we know now that there are(1) q m − 1 codewords of weight q m /2 in the code C(P ),(2) q 2 − 1codewordsc whose weights satisfyq m − 1q +1 ≤|2wt(c) − qm +1|≤2 √ q qm − 1q +1 ,(3) (q 2 − 1)q m − q 2 +1codewords c whose weights satisfy|2wt(c) − q m +1|≤q m/2 +2 √ q qm/2 +1q +1 .We remark that C(P 1 )isthesimplexcodeoflengthq m − 1, <strong>and</strong> C(P 2 )isthecode constructed by pasting together d copies of each codewords of the dual of theZetterberg code of length q +1.


Let us assume that q>2<strong>and</strong>m>2. Denote d =(q m − 1)/(q − 1), <strong>and</strong> chooseK = F qm <strong>and</strong> P = {ax d + bx | a, b ∈ K}. NowC(P ) is the dual of the binary cycliccode of length q m − 1 with zeroes γ d2i ,γ 2i , i =1, 2,....Denote P 1 = {bx | b ∈ K} <strong>and</strong> P 2 = {ax d | a, b ∈ K}. Again C(P )=C(P 1 ) ⊕C(P 2 ), <strong>and</strong> |C(P 1 )| = q m . By Proposition 2.10 |C(P 2 )| = q. Thus|C(P )| = q m+1 .By Theorem 2.9, Proposition 2.10 <strong>and</strong> Corollary 2.3, we now know that there are10(1) q m − 1 codewords of weight q m /2,(2) q − 1 codewords of weight (q m − 1+(q m − 1)/(q − 1))/2.in the code C(P ).For the remaining (q − 1)q m − q + 1 non-zero codewords c ∈ C(P )itholdsthat(3) q − 1 ≤|2wt(c) − q m +1|≤min{q m+12 − q m+12 −qq−1+1, (m +1)q m/2 }.We remark that C(P 1 ) is again the binary simplex code of length q m − 1, <strong>and</strong>C(P 2 ) is the code constructed by pasting together d copies of each codewords of thebinary simplex code of length q − 1.References1. Deligne P. (1977) Applications de la formule des traces aux sommes trigonometriques.SGA 4 1/2 Springer Lecture Notes in Math 569: 168-232. Springer, NewYork.2. Honkala I. & Tietäväinen A. Codes <strong>and</strong> Number Theory. In Brualdi R.A., Huffman W. C. & Pless V. (ed) H<strong>and</strong>book of Coding Theory. ElsevierScience Publisher, Amsterdam. In Preparation.3. Lachaud G. & Wolfman J. (1990) The weights of the orthogonals of theextended quadratic binary Goppa codes. IEEE Trans. Inform. Theory 36:686-692.4. Lidl R. & Niederreiter H. (1984) Finite Fields. Cambridge Univ. Press,Cambridge.


5. van Lint J. H. (1982) Introduction to Coding Theory. Springer-Verlag, NewYork.6. Liu S. C. <strong>and</strong> Komo J. J. (1992) Nonbinary Kasami sequences over GF(p).IEEE Trans. Inform. Theory 38: 1409-1412.7. MacWilliams F. J. & Sloane N. J. A (1978) The Theory of Error-CorrectingCodes. North-Holl<strong>and</strong>, Amsterdam.8. McEliece R.J. (1974) Irreducible cyclic codes <strong>and</strong> Gauss <strong>sums</strong>. In: Hall M.Jr. & van Lint J.H. (ed) Combinatorics (Part 1): 179-196. MathematicalCentre Tracts 55, Mathematical Centre, Amsterdam.9. McEliece R. J. (1980) Correlation properties of sets of sequences derivedfrom irreducible cyclic codes. Inform. <strong>and</strong> Control 45: 18-25.10. Mordell L. J. (1963) <strong>On</strong> a special polynomial congruence <strong>and</strong> <strong>exponential</strong><strong>sums</strong>. In: Calcutta Math. Soc. Golden Jubilee Commemoration Volume,Part 1: 29-32. Calcutta Math. Soc., Calcutta.11. Moreno C. J. & Moreno O. (1994) The MacWilliams-Sloane conjecture onthe tightness of the Carlitz-Uchiyama bound <strong>and</strong> the weights of duals ofBCH codes. IEEE Trans. Inform. Theory 40: 1894-1907.12. Weil A. (1948) <strong>On</strong> some <strong>exponential</strong> <strong>sums</strong>. Proc. Nat. Ac. Sc. 34: 204-207.11

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!