You are on page 1of 21

FINITE FIELDS AND THEIR APPLICATIONS 4, 218—238 (1998)

ARTICLE NO. FF980211

An Upper Bound for the Extended Kloosterman Sums


over Galois Rings*
Abhijit G. Shanbhag- and P. Vijay Kumar

EE-Systems, University of Southern California, Los Angeles, California 90089-2565


E-mail: kumar@aditya.usc.edu

and

Tor Helleseth

Department of Informatics, University of Bergen, H~yteknologisenteret, N-5020 Bergen, Norway

Communicated by Igor Shparlinski

Received March 21, 1995; revised June 16, 1997

An upper bound for the extended Kloosterman sum over Galois rings is derived.
This bound is then used to construct new sequence families with low correlation
properties and alphabet size a power of a prime. ( 1998 Academic Press
Key Words: Kloosterman sum; Galois ring; Witt ring; exponential sum; CDMA
sequences; periodic correlation.

I. INTRODUCTION

I.1. Overview
For a fixed prime p and integers e, m, e52, and m51, let R denote the
e,m
Galois Ring of characteristic pe and containing pem elements. Let
t : R PC be an additive character of R and let f (x) be a non-
e,m e,m e,m
degenerate polynomial (i.e., no monomial term in the polynomial has degree

* This work was supported in part by the National Science Foundation under Grant Number
NCR-93-05017 and The Norwegian Research Council under Grant Numbers 107542/410 and
107623/420.
- Current address: Qualcomm Inc., Q-302G, 6455 Lusk Boulevard, San Diego CA 92121.

218
1071-5797/98 $25.00
Copyright ( 1998 by Academic Press
All rights of reproduction in any form reserved.
EXTENDED KLOOSTERMAN SUMS 219

which is a multiple of p) over R with weighted degree [4] D . Define the


e,m e,f
set of ¹eichmuller representatives T "T* XM0N where T* is a cyclic
e,m e,m e,m
subgroup of order pm!1 of R* . Working over Galois rings, Kumar et al.
e,m
[4], recently obtained an expression for an analog of the finite field Weil-
exponential sums, i.e., for sums of the form

+ t ( f (x))
e,m
x| Te,m

as a sum of D !1 complex numbers, each having magnitude Jpm. This


e,f
expression leads to the bound [4]

K x| Te,m
e,m K
+ t ( f (x)) 4(D !1)Jpm
e,f

which was used in [4] to design new efficient four- and eight-phase sequence
families for CDMA applications.
In this paper we obtain an upper bound for the extended Kloosterman sums
over Galois rings, i.e., exponential sums of the form

K ( f , f )" + t ( f (x)#f (x~1)),


e,m 1 2 e,m 1 2
x| T*e,m
where f (x), f (x) are polynomials over R . These sums leads to new
1 2 e,m
sequence designs for CDMA applications which in many instances offer
performance comparable to those of the families in [4].

I.2. Galois Ring Preliminaries


Some preliminaries on Galois rings are given below. For more details, the
reader is referred to [6, 10, 12, 5, 7, 1] and [2].
For every integer k51, let F denote the finite field with pk elements. Let
k
l, j, l'j51 be integers and k : Z l PZ j be the (mod pj) reduction map. In
j p p
particular, k : Z l PZ "F is the (mod p) reduction map. We extend k to
1 p p 1 j
Z l [x] in the natural way. A monic polynomial g (x) 3 Z l [x] is said to be
p l p
a monic basic irreducible if k (g (x))"g (x) is a monic irreducible polynomial
1 l 1
in Z [x]. A Galois ring GR(pl, k) of plk elements is simply a Galois extension
p
of Z l . We will write R "GR(pl, k) for short. It turns out that every such
p l,k
ring is isomorphic to the ring Z l [x]/( f (x)), where f (x) 3 Z l [x] is a monic
p p
basic irreducible of degree k. R* always contains a cyclic group of order
l,k
pk!1. In analogy with finite fields, we will call an element a primitive element
of the Galois ring R if it is a generator for this cyclic group. Let b denote a
l,k l,k
220 SHANBHAG, KUMAR, AND HELLESETH

k
primitive element in R . Let T "M0, 1, b , 2 , bp ~2N. It can be shown
l,k l,k l,k l,k
that every element z 3R has the p-adic expansion
l,k
z"z #pz #p2z #2#pl~1z , z 3T .
0 1 2 l~1 i l,k
Let s51 be an integer. Then R is an extension ring of R having Galois
l,ks l,k
group of order s generated by the automorphism q given by
k
k k k
q (x)"xp #pxp #2#pl~1x p ,
k 0 1 l~1
where x3 R has p-adic expansion
l,ks
x"x #px #2#pl~1x , x 3T .
0 1 l~1 i l,ks
We will abbreviate q ( ) ) to q( ) ). For x3 R , we define the trace
1 l,ks
¹ ( ) ) : R PR via
l,ks,k l,ks l,k
s~1
¹ (x)" + qi (x)
l,ks,k k
i/0
It is easily verified that

¹ (x)"¹ (¹ (x)).
l,ks,1 l,k,1 l,ks,k

For any t51, we will abbreviate and write ¹ ( ) ) in place of ¹ ( ) ).


l,t l,t,1
Let t , t : R PC, be the character defined by
l,k l,k l,k
t (z)"e(i2nb@pl )Tl,k (z), z 3R ,
l,k l,k
where b, 14b4pl!1, is a fixed integer relatively prime to p, i.e., (b, p)"1.
Note that for any integer s51,

t (y)"t (¹ (y)) ∀y3 R .


l,ks l,k l,ks,k l,ks
I.3. ¹he ºpper Bound
Let f (x), f (x) 3R [x] and assume moreover, that either k ( f )O0 or
1 2 e,m 1 1
k ( f )O0. Note that
1 2
+ t ( f (x)#f (x~1))" + t ( f (x)#f (x~1)).
e,m 1 2 e,m 2 1
x| T*e,m x| T*e,m
We, thus, assume w.l.o.g. for the remaining sections that k ( f )O0.
1 1
EXTENDED KLOOSTERMAN SUMS 221

Let w(x)3R [x] be non-degenerate. Consider the p-adic expansion of


e,m
w(x) as under

w(x)"w [x]#pw [x]#2#pe~1w [x]


0 1 e~1
obtained from the p-adic expansion of the coefficients of w( ) ). Let d be the
j
degree of w (x), 04j4e!1. The weighted degree D of w(x) is then
j e,w
defined [4] to be

D "maxMpe~1d , pe~2d , 2 , d N.
e,w 0 1 e~1
The principal result of the paper is
THEOREM 1. ¸et f (x), f (x) 3 R [x] be non-degenerate and have
1 2 e,m
weighted degree D and D respectively. Assume k ( f )O0, f I0. ¹hen
e,f1 e,f2 1 1 2

K x| T*e,m
2 K
+ t ( f (x)#f (x~1)) 4(D #D )Jpm.
e,m 1 e,f1 e,f2

I.4. Application to Sequence Design


Low correlation sequence families [11, 8] find use in CDMA applications.
Consider the set S of sequences defined via
D1,D2
S "MM¹ ( f (bt )#f (b~t ))N D D 4D , D 4D N
D1,D2 e,m 1 e,m 2 e,m t|Z e,f1 1 e,f2 2
where b is a generator of T* and f (x) 3R [x] has weighted degree
e,m e,m i e,m
D for i"1, 2. Let the set
e,fi
F LS
D1,D2 D1,D2
consist of a maximal family of pairwise, cyclically distinct sequences in
S with each sequence having period pm!1. Using Theorem 1, it is easy
D1,D2
to see that the maximum non-trivial correlation C (see [4]) of the family
.!9
F is upper bounded as under
D1,D2

C 41#(D #D )Jpm. (1)


.!9 1 2

When both D and D are less than Jpm!1, the size of the family F can
1 2 D1,D2
be shown to have the lower bound [9]

DF D1 @pye ~x
D5pm(D1`D2~x D2 @pye ~1). (2)
D1,D2
222 SHANBHAG, KUMAR, AND HELLESETH

TABLE 1
Four and Eight-Phase Sequence Designs

Weighted Degree Family Bound on


No. Alphabet (D , D ) Size C
1 2 .!9
1 Z (2, 2), (3, 1) 5¸3
4 41#4Jq
2 Z (3, 2), (2, 3) 5¸4
4 41#5Jq

3 Z (7, 1), (6, 2), (5, 3), (4, 4) 5¸7


8 41#8Jq
4 Z (7, 2), (6, 3), (5, 4), (4, 5) 5¸8
8 41#9Jq

Table 1 presents some sequence designs having alphabet size 4, 8


corresponding to p"2 and to e 3 M2, 3N. The different entries in the
table correspond to different choices of the weighted-degrees (D , D ) of
1 2
f (x) and f (x) respectively. The different choices for (D , D ) which result
1 2 1 2
in the same bounds for family size and maximum correlation are included.
For e.g., the eight-phase sequence family corresponding to (D , D )"
1 2
(4, 4) will consist of the pairwise cyclically distinct sequences from the
set

S "MM¹ ( f (bt )#f (b~t ))N D f (x)"c x#4c x3,


4,4 3,m 1 e,m 2 e,m t|Z i i,0 i,1
c 3 R , c 3 T , i"1, 2N.
i,0 3,m i,1 3,m

When the alphabet size is fixed, the sequence families listed in Table 1 offer
performance comparable to those constructed using weighted degree poly-
nomials [4]. In general, this will be true whenever

D #D #1 D D
1 2 "1# 1 # 2
pe pe pe

and in addition, pe P (D #D #1).


1 2

I.5. ¹he ¸ifting #


Let l, l52 be a fixed integer. Let G denote the set of all monic
1,m
polynomials in F [x]. In [4], Kumar et al. identify a factorial domain
m
(denoted by G ) within R [x] such that G and G are isomorphic
l,m l,m l,m 1,m
EXTENDED KLOOSTERMAN SUMS 223

multiplicative monoids. This isomorphism

# : G PG
1,l 1,m l,m

can be viewed as a lifting of polynomials over F to those over R .


m l,m
As discussed in [4], # is given by the composition of l!1 liftings # :
1,l j
R [x]PR [x], 14j4l!1. Each lifting # is defined by
j,m j`1,m j

# (g (x))"(!1)k(p~1)g (x), (3)


j j j`1

where

A B
p~1
g (xp)"q~1 < gL (xmi) ,
j`1 j
i/0
where gL is any element of R [x] satisfying k (gL (x))"g (x), m is a primi-
j j`1,m j j j
tive pth root of unity in the algebraic closure of the field Q of p-adic numbers
p
and k is the degree of g (x). Each such lifting # can be shown to be
j j
multiplicative, from which it follows that # is multiplicative. The subsets
1,l
G -R [x] are recursively defined by
j,m j,m

G "# (G ), 24j4l.
j,m j~1 j~1,m

It turns out that the polynomials in G are monic for all j, 14j4l.
j,m
We note:
1. #~1"k .
j j
2. For a monic irreducible polynomial h (x) 3 F [x] of degree k with
1 m
factorization

k~1
h (x)" < (x!bvqi ), (4)
1 1,mk
i/0
where 04v4pmk!2 is an integer and b is a primitive element in F , we
1,mk mk
have
k~1
h (x)"# (h (x))" < (x!bvqi ), (5)
l 1,l 1 l,mk
i/0
where b is the primitive element in R satisfying k (b )"b .
l,mk l,mk 1 l,mk 1,mk
Let gL (x) 3R [x] satisfy k (gL (x))"g (x). We now present Lemma 1
j j`1,m j j j
in [4]:
224 SHANBHAG, KUMAR, AND HELLESETH

LEMMA 1. ¹he coefficients MpL N and Mp N of the respective polynomials


j,n j`1,n
gL (x) and g (x)"# (g (x)) are related by:
j j`1 j j
p "
j`1,n

G
(!1)k(p~1)Mq~1(pL p )#pq~1[(!1)n(p~1)pL #G (pL , 2 , pL )]N
j,n j,pn 0 j,1 j,pn~1
if k5pn,
(!1)k(p~1)(q~1(pL p ))#q~1[G (pL , 2 , pL )] if k(pn,
j,n 1 j,1 j,k
where G ( ) ) and G ( ) ) are polynomials in several variables over Z j`1 .
0 1 p

II. AN ¸-FUNCTION

Let G denote the subset of G having degree k. Let g (x)"


e,m,k e,m e
+k (!1)ip xk~i 3G have roots Me Nk . Define j : G PC via
i/0 e,i e,m,k i i/1 e,m e,m

G
t ( + k ( f (e )#f (e~1 ))) if p 3 R*
j (g )" e,m i/1 1 i 2 i e,k e,m (6)
e,m e 0 otherwise.

Multiplicativity of j is evident. Next consider the formal power series


e,m

¸ (z)" + j (g )z$%'(ge) (7)


e,m e,m e
ge | Ge,m
and let the coefficients ¸ be defined by
e,ms
= d
+ ¸ zs"z log ¸ (z).
e,ms dz e,m
s/0
We now prove:
THEOREM 2. ¹he ¸-function ¸ (z) is a polynomial in z of degree
e,m
4D #D and can be factored as
e,f1 e,f2
De,f1`De,f2
¸ (z)" < (1!g z) (8)
e,m i
i/1
for some complex numbers g . ¹he exponential sums
i

f :
" + t ( f (x)#f (x~1)) (9)
e,ms e,ms 1 2
x| T*e,ms
EXTENDED KLOOSTERMAN SUMS 225

satisfy

De,f1`De,f2
f "¸ "! < gs . (10)
e,ms e,ms i
i/1
Proof. In the proof, we will abbreviate and write j( ) ) in place of j .
e,m
Since every polynomial g (x)3 G has a unique factorization as a product of
e e,m
monic, basic irreducibles in G , we can write
e,m

¸ (z)" < (1#j(h)z$%'(h)#j2(h)z2$%'(h)#2)


e,m
h| He,m

C D
1
" < ,
1!j(h)z$%'(h)
h| He,m
where H is the set of monic basic irreducibles in G . Note that
e,m e,m
H "# (H ) where H denotes the monic irreducibles in F [x].
e,m 1,e 1,m 1,m m
Then,

= d
+ ¸ zs"z log(¸ (z))
e,ms dz e,m
s/0
j(h) deg(h)
"z + z$%'(h)~1
[1!j(h)z$%'(h)]
h| He,m

" + deg(h)j(h)z$%'(h)[1#j(h)z$%'(h)#j2(h)z2$%'(h)#2].
h| He,m
Comparing coefficients of zs on both sides, we obtain that

¸ " + deg(h)[j(h)]s@$%'(h). (11)


e,ms
h| He,m,$%'(h) D s
Let s be an integer and h(x) 3 H , h(x)Ox, have degree rDs. Let
e,m

r
h(x)" + (!1)ih xr~i, h "1.
i 0
i/0
Clearly each of the r zeros of h(x) lies in T* . Further, given any zero x of
e,mr 1
h(x), the remaining zeros x , 24i4r of h(x) satisfy
i

x "qi~1 (x ), 24i4r.
i m 1
226 SHANBHAG, KUMAR, AND HELLESETH

Thus,
j(h)"t (¹ ( f (x )#f (x~1))).
e,m e,mr,m 1 1 2 1
From this and using (11) it follows that

¸ " + t (( f (x)#f (x~1))).


e,ms e,ms 1 2
x| T*e,ms
Next we show that for k'D #D
e,f1 e,f2
+ j(g )"0.
e
ge | Ge,m,k
Let g (x)"+ k (!1)ip xk~i, p "1 be the lifting under # of the finite
e i/0 e,i e,0 1,e
field polynomial g (x)"+ k (!1)ip xk~i, p "1. Then using Lemma
1 i/0 1,i 1,0
1 and the results of Section III-B in [4], we have for k5D
e,f1
k
+ f (e )"uq~r[pe~1p #J(p , 2 , p )],
1 i 1,De,f1 1,1 1,De,f1~1
i/1
where Me Nk are the roots of g (x), J is a multivariable polynomial over R ,
i i/1 e e,m
r is some integer and u is a unit in R . Furthermore, noting that Me~1Nk
e,m i i/1
are the roots of the reciprocal polynomial of g (x), we have
e
k
+ f (e~1)"u@p~1 q~r{[pe~1p #2#J@(p , 2 , p )],
2 i 1,k 1,k~De,f2 1,k 1,k~De,f2`1
i/1
where J@ is a multivariable polynomial over R , r@ is some integer and u@ is
e,m
a unit in R . Thus, for k!D 'D ,
e,m e,f2 e,f1

+ j(g )" + t (p )t (uq~r(J(p , 2 , p ))


e 1,m 1,De,f1 e,m 1,1 1,De,f1~1
ge | Ge,m,k p1,1, 2 , p1,k
#u@q~r(J@(p , 2 , p )))
1,k 1,k~De,f2`1
"0.

Thus we have shown that ¸ (z) is a polynomial in z of degree not


e,m
exceeding D #D . Since the constant term in ¸ (z) equals 1, we can
e,f1 e,f2 e,m
factor

De,f1`De,f2
¸ (z)" < (1!g z)
e,m i
i/1
EXTENDED KLOOSTERMAN SUMS 227

for some complex numbers g . Computing


i

A B
d De,f1`De,f2
z log < (1!g z)
dz i
i/1
and equating coefficients then gives (10). j

Let a, 14a4pe!1 be a fixed integer. More generally, we define j :


e,m,a
G PC via
e,m

G
t ( + k (a( f (e )#f (e~1 )))) if p 3R*
j (g )" e,m i/1 1 i 2 i e,k e,m (12)
e,m,a e 0 otherwise.

Corresponding to j , we define the formal series ¸ (z) and the coeffi-


e,m,a e,m,a
cients ¸ similar as above. We can carry out a computation identical to
e,ms,a
above to show that ¸ (z) is a polynomial in z. Note that when pDa, the
e,m,a
computations will effectively be carried out in a Galois ring of lower charac-
teristic. Thus, in the more general case, we obtain

De~la,f1`De~la,f2
¸ (z)" < (1!g z) (13)
e,m,a a,i
i/1
for some complex numbers g , where l "maxMiDpiDaN. Similarly, we have
a,i a
De~la,f1`De~la,f2
f "¸ "! < gs . (14)
e,ms,a e,ms,a a,i
i/1

III. MAGNITUDE OF THE ROOTS OF THE ¸-FUNCTION

Let q"pm. Our objective in this section is to show:


f when k ( f (x))O0, the complex numbers Mg N have magnitude
1 2 a,i
Jq or zero.
f when k ( f (x))"0 and j "maxMj Dk ( f (x))"0N, the complex num-
1 2 1 j 2
bers Mg N have magnitude Jq or zero if pe~j1 P a. If pe~j1 Da then precisely one
a,i
g is unity and the remaining Mg N have magnitude Jq or zero.
a,i a,i
We prove these results using the theory of function fields [13].

III.1. Relation to the Zeros of the ¹race Function


Let

Z "Mx 3 T* D¹ ( f (x)#f (x~1))"0N. (15)


e,s e,ms e,ms 1 2
228 SHANBHAG, KUMAR, AND HELLESETH

Then

pe~1
peDZ D" + + t (a ) ( f (x)#f (x~1))). (16)
e,s e,ms 1 2
a/0 x| T*e,ms
Using the analogue of Hilbert’s Theorem 90 for Galois rings, we have

peDZ D"D Mx3 T* , y3 R D q(y)!y"f (x)#f (x~1) (mod pe)ND.


e,s e,ms e,ms 1 2

Thus the ‘‘trace of exponential sums’’

pe~1
) " + + t (a ) ( f (x)#f (x~1)))
e,s e,ms 1 2
a/0 x| T*e,ms
can be expressed in two ways as

pe~1 De~la,f1`De~la,f2
) "pms!1# + + !gs (17)
e,s a,i
a/1 i/1
"D Mx3 T* , y3R D q(y)!y"f (x)#f (x~1) (mod pe)ND (18)
e,ms e,ms 1 2

where l is as defined in the previous section.


a
III.2. Connection with Places of a Function Field
The first step is to show that the number of solutions M(x, y)D 3T* ,
e,ms
y3R N of the equation
e,ms

q(y)!y"f (x)#f (x~1), (19)


1 2

is the same as that of a system of e equations over F , that will turn out to
m
define a cyclic extension of the rational function field F (x) of degree pe.
m
Certainly, by equating coordinates in the p-adic expansion of both sides of
(19), one can obtain a system of e equations over F that define an extension
m
of the rational function field F (x). Direct attempts at proving the extension
m
to be cyclic of degree pe ran however, into difficulties, which were circum-
vented by using the isomorphism between Galois rings and Witt rings having
finite field components.

III.2.1. ¼itt Ring Definition. The material for this subsection is taken
mostly from Jacobson [3, pp. 124—140]. The notation does however, differ in
places.
EXTENDED KLOOSTERMAN SUMS 229

Let X"Q[Mz N ], where Q denotes the field of rational numbers and the
i i
Mz N are a finite set of indeterminates. Given a"(a , a , 2 , a )3 Xe, we
i i 0 1 e~1
define

ap"(a p , ap , 2 , ap )
0 1 e~1
l l~1
a(l)"a p #pa p #2#pla , 04l4e!1,
0 1 l
"(ap)(l~1)#pla , 04l4e!1,
l
/(a)"(a(0), a(1), 2 , a(e~1)),

and for b"(b , b , 2 , b ) 3 Xe,


0 1 e~1
t(b)"(c , c , 2 , c ),
0 1 e~1
where

c "b
0 0
1 l l~1
c " (b ![c p #pc p #2#pl~1c p ]), 14l4e!1.
l pl l 0 1 l~1

It can be verified that / and t are inverse maps. Let addition = and
multiplication x of elements a, b3Xe be defined as follows:

a = b"t(/(a)#/(b)),
a x b"t(/(a) ) /(b)).

where #and ) denote the usual componentwise addition and multiplication


in Xe. It turns out that under these operations, Xe becomes a commutative
ring, known as the ring of Witt vectors of length e over X. It follows from the
addition and multiplication formulae that

(a = b)(l)"a(l)#b(l)
(a x b)(l)"a(l) ) b(l).

The theorem below is taken from Jacobson [3]. The proof is included, as
an intermediate step will prove useful in a subsequent subsection.
THEOREM 3. ¸et a, b 3 Xe and let

c"(c , c , 2 , c )"a = b"(a , a , 2 , a ) = (b , b , 2 , b ).


0 1 e~1 0 1 e~1 0 1 e~1
230 SHANBHAG, KUMAR, AND HELLESETH

¹hen the k-th component c of c is expressible in the form


k
c "A (a , b )
k k i j
where A (a , b ) is a polynomial in a , b with integer coefficients.
k i j i j
Proof. Clearly, c is so expressible. Let us assume the Theorem to be
0
true for c , 04j4k!1. Now
j
c(k)"a(k)#b(k)
implies
k k~1
pkc "a(k)#b(k)!(c p #pc p #2#pk~1c p )
k 0 1 k~1
"a(k)#b(k)!(cp)(k~1). (20)

Given a vector s 3Xe, to compute


k k~1
s(k)"s p #ps p #2#pks "(sp)k~1 (mod pk),
0 1 k
it is sufficient to compute s p , 04l4k!1 (mod p). But
l
a(k)#b(k)!(cp)(k~1)"(ap)(k~1)#(bp)(k~1)!(cp)(k~1) (mod pk)
"0 (mod pk),

since by the inductive hypotheses,

c p"Ap (a , b )"A (a p , b p ) (mod p).


l l i j l i j
Also, the RHS of (20) is again by the inductive hypothesis, a polynomial with
integer coefficients in a , b . It follows that
i j
1
c " Ma(k)#b(k)!(cp)(k~1)N"A (a , b ) (mod p)
k pk k i j

is also a polynomial with integer coefficients. j

It can similarly be shown that if a x b"d, that the components d of d are


j
given by

d "M (a , b ), 04l4e!1,
l l i j
where M are polynomials having integer coefficients.
l
EXTENDED KLOOSTERMAN SUMS 231

Next, let U be a commutative algebra over the finite field F . Let addition
m
and multiplication in Ue be defined by

a = b"c, a x b"d,

where

c "A (a , b ), (mod p) 04l4e!1,


l l i j
d "M (a , b ), (mod p), 04l4e!1.
l l i j
With these operations, Ue becomes the ring of Witt vectors over U.
III.2.2. ¼itt Ring, Galois Ring Isomorphism.
THEOREM 4. ¹he ring Fe of ¼itt vectors over the finite field F and the
m m
Galois ring R are isomorphic under the map !: R PFe given by
e,m e,m m
e~1
!(u)"(u , u p , 2 , u pe~1),
0 1
where

u"u #pu #2pe~1u , u 3T


0 1 e~1 i e,m
is the p-adic expansion of u.
Proof. Let u"+ e~1 piu , v"+ e~1 piv . We will show that
i/0 i i/0 i
!(u) = !(v)"!(u#v). Let w"!(u) = !(v). Then by the definition of Witt
ring addition,
i j
w "A (u p , v p ), 04l4e!1,
l l i j
where the polynomials A are as defined in the previous subsection. If
i e~1
u#v"z"+ e~1 z pi, z 3T , then !(z)"(z , zp , 2 , zp ). We have to show
l i/0 i i e,m 0 1 e~1
that z p "w , (mod p) 04l4e!1. Clearly, z "u #v (mod p). Let us
l l l 0 0 0
assume that z p "w (mod p), for 04l4k!1. We have the Galois ring
l l
equation

k k k
+ piz " + piu # + piv (mod pk`1).
i i i
i/0 i/0 i/0
Applying qk to both sides, gives
e,m
k k k k k k
+ piz p " + piu p # + piv p (mod pk`1).
i i i
i/0 i/0 i/0
232 SHANBHAG, KUMAR, AND HELLESETH

Therefore,

k k k~1 k~i i j
pkz p " + pi(u p ) p # + pi(v p ) p ! + piAip (u ip , v pj )
k i k~i i k~i
(mod pk`1).
k i i
i/0 i/0 i/0
(21)

Now a comparison of (20) and (21), gives us


k i j
z p "A (u p , v p )"w (mod p)
k k i j k
and so !(u) = !(v)"!(u#v). One can similarly establish that

!(u) x !(v)"!(u ) v),

thus proving the isomorphism. j

The possibility of a connection between Witt rings and Galois rings was
pointed out to us by H. Stichtenoth. J. Lahtonen working independently, was
also aware of a connection between the two.

III.2.3. Function Field Connection. By the isomorphism between Galois


and Witt rings, the number of solutions (x, y), x3 T , y 3R to the
e,ms e,ms
equation

q(y)!y"f (x)#f (x~1)


1 2
in the Galois ring is precisely the number of solutions
e~1
(!(x), !(y))"((x, 0, 0, 2 , 0), (y , y p , 2 , y p ))
0 1 e~1
where y"y #py #2#pe~1y , to the Witt ring equation
0 1 e~1
!(q(y)) > !(y)"!( f (x)#f (x~1)),
1 2
where > denotes subtraction in the Witt ring.
We have
2 e
!(q(y))"!(y p #pyp #2#pe~1yp )"(y p , y p , 2 , y p ).
0 1 e~1 0 1 e~1
It is not hard to show that if

!(q(y)) > !(y)"(b , b , 2 , b ),


0 1 e~1
EXTENDED KLOOSTERMAN SUMS 233

then
b "y p !y (mod p)
0 0 0
k`1 k 2 k~1
b "y p !y p #s (y , y p , y p , 2 , y p ) (mod p), 14k4e!1,
k k k k 0 1 2 k~1
where the s are polynomials with coefficients in F .
k 1
Let
f (x)"u (x)#pu (x)#2#pe~1u (x),
1 0 1 e~1
where u (x)"+ di u xj, u 3 T and similarly,
i j/0 i,j i,j e,ms
f (x~1)"v (x~1)#pv (x~1)#2#pe~1v (x~1),
2 0 1 e~1
where v (x~1)"+ di v x~j, v 3T .
i j/0 i,j i,j e,ms
Then

A B
e~1 di
!( f (x)#f (x~1))" + =pi + = (u xj, 0, 0, 2 , 0)
1 2 i,j
i/0 j/0

A B
e~1 di
= + =pi + = (v x~j, 0, 0, 2 , 0) ,
i,j
i/0 j/0
where = following the + denotes Witt ring addition. The Witt ring addi-
tion on the RHS can be carried out by treating u , v , x, x~1 as indeter-
i,j i,j
minates and pretending to work in the ring of Witt vectors over X"
Q[u , v , x, x~1]. In this situation,
i,j i,j
e~1 di e~1
/(!( f (x)#f (x~1)))" + pi + (u xj, u p xjp, 2 , u p xjp e~1 )
1 2 i,j i,j i,j
i/0 j/0
e~1 di e~1
# + pi + (v x~j, v p x~jp, 2 , v p x~jp e~1 )
i,j i,j i,j
i/0 j/0
"( f (x)#f (x~1), q( f (x)
1 2 1
#f (x~1)), 2 , qe~1( f (x)#f (x~1))),
2 1 2
where we have used ql( f ), ql( f ), 04l4e!1, to denote
1 2
e~1 e~1 di l
ql( f (x))" + piql(u (x))" + pi + u p xp l
1 i i,j
i/0 i/0 j/0
e~1 e~1 di l
ql( f (x~1))" + piql(v (x~j))" + pi + v p x~plj .
2 i i,j
i/0 i/0 j/0
234 SHANBHAG, KUMAR, AND HELLESETH

Let !( f (x)#f (x~1))"(a , a , 2 , a ). Then


1 2 0 1 e~1
e~1 e~2
/((a , a , 2 , a ))"(a , a p #pa , 2 , ap #pa p #2#pe~1a ).
0 1 e~1 0 0 1 0 1 e~1
It follows that

a "f (x)#f (x~1) (mod p),


0 1 2
1 k k~1
a " Mqk( f (x)#f (x~1))!(a p #pap #2#pk~1ap )N, (mod p),
k pk 1 2 0 1 k~1
14k4e!1.

By the theory of Witt rings, each component a , 04i4e!1 is guaran-


i
teed to be a polynomial in x, x~1. Note moreover, that the k-th component
a is only a function of f (x)#f (x~1) (mod pk`1).
k 1 2
We would now like to equate components on both sides of the equation

!(q(y)) > !(y)"!( f (x)#f (x~1))


1 2
and will consider two cases separately. In the first case, f (x~1)O0 (mod p).
2
In this case, one obtains a set of e equations of the form

z p !z "f (x)#f (x~1) (mod p)


0 0 1 2
z p !z "g (z , x, x~1) (mod p)
1 1 1 0
2"2
z p !z "g (z , z , ,z , x, x~1) (mod p),
e~1 e~1 e~1 0 1 2 e~2
where z "yp i, 04i4e!1 and the g are polynomials over R .
i i e,ms
In the second case, let j be the largest integer such that f (x~1)"pj1 h(x~1)
1 2
for some polynomial h(x) with coefficients in R . Here the system of
e,ms
e equations takes on the form

z p !z "f (x) (mod p)


0 0 1
z p !z "g (z , x) (mod p)
1 1 1 0
2"2
z p !z "g (z , z , ,z , x) (mod p)
j1~1 j1~1 j1~1 0 1 2 j1~2
z p !z "qj1 (h(x~1))#g (z , z , 2 , z , x) (mod p) (22)
j1 j1 j1 0 1 j1~1
z p!z "g (z , z , 2 , z , x, x~1) (mod p), j #14l4e!1.
l l l 0 1 l~1 1
EXTENDED KLOOSTERMAN SUMS 235

In either case, for 14j4e!1, we let F "F (x, z , 2 , z ) denote the


j m 0 j~1
jth extension field of the rational function field F (x) as defined by a corres-
m
ponding system of e equations. Let F "F (x) denote the rational function
0 m
field. It is then known (see [3], pp. 124—140) that F , 14j4e!1, is a cyclic
j
extension of dimension pj over F if and only if F is an Artin—Schreier
0 1
extension of F . Since f (x) is non-degenerate and k ( f (x))O0 by assump-
0 1 1 1
tion, F is always an Artin—Schreier extension of F .
1 0
In the next two subsections we consider separately the cases k ( f (x))O0
1 2
and k ( f (x))"0.
1 2
Case k ( f (x))O0. Here we claim that in (17)
1 2

Dg D"Jq or 0, ∀a, 14a4pe!1, ∀i, 14i4D #D . (23)


a,i e~la ,f1 e~la ,f2
Since the equation

z p !z "f (x)#f (x~1) (mod p)


0 0 1 2
defines an Artin—Schreier extension F over the rational function field F , it
1 0
follows from the non-degeneracy of f , f , that each of the places at zero and
1 2
at ‘‘infinity’’ of F is completely ramified in F . Since F is a cyclic
0 1 e~1
extension of F , it follows from the ramification theory for cyclic extensions
0
of function fields that F has a single (completely ramified) place of degree
e~1
1 above the places at zero and infinity of F .
0
The Riemann Hypothesis for function fields leads to the equations

A B
p e~1
pms!1# + !+ gs "pms#1!+ hs!(n #n ), all s51,
a,i i 0 =
a/1 i i
where n "n "1 is the number of places in F above the place at zero
0 = e~1
and the place at infinity respectively in F , and Mh N are complex numbers
0 i
having magnitude Jq. It follows from Newton’s identities that the set Mh N
i
and the non-zero elements in the set Mg N are identical, thus establishing (23)
a,i
and Theorem 1 for the case k ( f (x))O0.
1 2
Case k ( f (x))"0. As before, let j be the largest integer such that
j1 2 1
f (x~1) is expressible in the form f (x~1)"pj1h(x~1), for some polynomial
2 2
h over R . Here we note that when pe~j1 Da, aO0,
e,ms
¸ " + t (a( f (x)#f (x~1)))
e,ms,a e,ms 1 2
x| T*e,ms
" + t (a f (x))
e,ms 1
x| T*e,ms
236 SHANBHAG, KUMAR, AND HELLESETH

"!1# + t (a f (x)).
e,ms 1
x| Te,ms
In this case, we know from the results of [4] that !+ T t (a f (x)) is
x| e,ms e,ms 1
a sum of sth powers of complex numbers each having magnitude Jq (or zero).
Thus, for pj1!1 choices of a, !¸ is a sum of sth powers of complex
e,ms,a
numbers Mg N from which precisely one has magnitude 1 and the remaining
a,i
have magnitude Jq (or zero). We then have,

p e~j 1Da i
a,i A
) "pms!1!(pj1!1)# + !+ gs # + !+ gs . (24)
e,s
p e~j1 P a i
B
a,i A B
a 0
O

Since k ( f (x))"0, F is defined via


1 2 1
z p !z "f (x) (mod p).
0 0 1
F is an Artin—Schreier extension of F and so the place at infinity in F is
1 0 0
completely ramified in F . Furthermore, since k ( f (x))"0, we can com-
1 j1 2
pute the number of places of degree 1 in F above the place at zero in F by
j1 0
solving the equation

q(y)!y"0 (mod pj1), y3R .


j1,m
Clearly this equation has pj1 solutions. By Kummer’s theorem F has
j1
pj1 places of degree 1 above the place at zero. As an immediate consequence,
we note that the place at zero in F (x) is unramified (i.e. degree of ramifica-
m
tion"1) in the extension field F .
j1
Consider the j th equation, i.e., (22):
1
z p !z "qj1 (h(x~1))#g (z , z , 2 , z , x) (mod p),
j1 j1 j1 0 1 j1~1
where as above, h(x~1) is defined by f (x~1)"pj1 h(x~1). Let P be any
2 j1,0
place in F lying above the zero place in F . Since, k (h(x~1))O0 and f (x~1)"
j1 0 1 2
pj1 h(x~1) is non-degenerate, we can find as shown below, an element u 3 F
j1
such that

vP (qj1 (h(x~1))!(up!u))(0 and O0 (mod p).


j1,0

Since f is non-degenerate, it follows that h has degree d that is relatively


2
prime to p. If
d
h(x~1)" + c x~i, (mod p)
i
i/0
EXTENDED KLOOSTERMAN SUMS 237

then
d
qj1 (h(x~1))" + (c x~i)p j1
i
i/0

and one could choose

d
u" + M(c x~i)#(c x~i)p#2#(c x~i)p j1~1 N.
i i i
i/0

It can then be checked that

vP (qj1 (h(x~1))!(up!u))"vP (h(x~1))"!d(0 and


j1,0 j1,0
O0 (mod p).

From the Artin—Schreier extension theorem, it follows that each


place above zero in F is totally ramified in F . Thus, F has pj1
j1 j1`1 j1`1
places above the zero place in F . Further, from ramification theory
0
for cyclic extensions of function fields, it follows that the place at infinity
in F and each place above zero in F is completely ramified in F . It
0 j1 e~1
follows that F has precisely pj1 places above the place at zero and one
e~1
place above the place at infinity in F .
0
The Riemann Hypothesis for function fields then leads to the equation

A B A B
pe~1
pms!1!(pj1!1)# + !+ gs # + !+ gs
e~j1 D
a,i e~j
a,i
pO a i a/1,p 1 P
a t
a 0
"pms#1!+ hs!(n #n ),
i 0 =
i

where we now have n "pj1 and n "1 and again Mh N are complex numbers
0 = i
having magnitude Jq. Again from Newton’s identities the set Mh N and the
i
non-zero elements in the set Mg N are identical, thus leading to Theorem 1
a,i
for this case.

ACKNOWLEDGMENT

The authors acknowledge H. Stichtenoth for several helpful suggestions which considerably
shortened our proof.
238 SHANBHAG, KUMAR, AND HELLESETH

REFERENCES

1. S. Boz,tas, A. R. Hammons, Jr., and P. V. Kumar, 4-Phase Sequences with Near-Optimum


Correlation Properties, IEEE ¹rans. Info. ¹heory 38, No. 3, (May 1992), 1101—1113.
2. A. R. Hammons, P. V. Kumar, A. R. Calderbank, N. J. A. Sloane, and P. Solé, The
Z -Linearity of Kerdock, Preparata, Goethals and Related Codes, IEEE ¹rans. Info. ¹heory
4
40 (March 1994), 301—319.
3. N. Jacobson, ‘‘Lectures in Abstract Algebra,’’ Vol. III, van Nostrand, Princeton, 1964.
4. P. V. Kumar, T. Helleseth, and A. R. Calderbank, An Upper Bound for Weil Exponential
Sums over Galois Rings and Applications, IEEE ¹rans. Info. ¹heory 41 (March 1995),
456—468.
5. R. A. Liebler and R. A. Mena, Certain Distance-Regular Digraphs and Related Rings of
Characteristic 4, J. Combin. ¹heory Ser. A 47 (1988), 111—123.
6. B. R. MacDonald, ‘‘Finite Rings with Identity.’’ Marcel Dekker, New York, 1974.
7. A. A. Nechaev, The Kerdock Code in a Cyclic Form, Discrete Math. 1 (1989), 123—139.
8. D. V. Sarwate and M. B. Pursley, Cross-Correlation Properties of Pseudorandom and
Related Sequences, Proc. IEEE 68 (May 1980), 593—618.
9. A. G. Shanbhag, P. V. Kumar, and T. Helleseth, Improved Binary Codes and Sequence
Families from Z -Linear Codes, IEEE ¹rans. Info. ¹heory 42, (Sept.1996), 1582—1586.
4
10. P. Shankar, On BCH Codes over Arbitrary Integer Rings, IEEE ¹rans. Info. ¹heory 25
(May 1979), 480—483.
11. M. K. Simon, J. K. Omura, R. A. Scholtz, and B. K. Levitt, ‘‘Spread-Spectrum Com-
municatins,’’ Vol. I, Computer Science Press, Rockville, MD, 1985.
12. P. Solé, A Quaternary Cyclic Code and a Family of Quadriphase Sequences with Low
Correlation Properties, ¸ecture Notes Comput. Sci. 388 (1989), 193—201.
13. H. Stichtenoth, ‘‘Algebraic Function Fields and Codes.’’ Universitext. Springer-Verlag, New
York, 1993.

You might also like