A GAUSS SUM ESTIMATE IN ARBITRARY FINITE FIELDS

Jean Bourgain

Mei-Chu Chang

Summary. We establish bounds on exponential sums xFq (xn ) where q = pm , p prime, and an additive character on Fq . They extend the earlier work [BGK] to elds that are not of prime order (m 2). More precisely, a nontrivial estimate is 1 1 obtained provided n satises gcd (n, pq q for all 1 < m, |m, where 1 ) < p > 0 is arbitrary.

DES SOMMES DE GAUSS UNE ESTIMEE DANS DES CORPS FINIS ARBITRAIRES
n u Resum e. On etabli des bornes sur les sommes dexponentielles xFq (x ) o` m q = p , p est premier et est un caract` ere additif de Fq . Il sagit dune extension des r esultats de [BGK] pour un corps qui nest pas dordre premier, c.a.d. m 2. On 1 obtient une estim ee non-triviale pour tout n satisfaisant la condition pgcd (n, pq 1 ) < p q 1 pour tout 1 < m, |m et o` u > 0 est arbitraire.

Version fransaise abr eg ee Dans cette note nous d emontrons une extension des r esultats obtenus dans [BGK] t n j pour des sommes de Gauss xFq (x ) et plus generalement j1 u est =1 (g ), o` un caract` ere additif de Fq , g Fq dordre multiplicatif t t1 . Les r esultats de [BGK] traitent le cas o` u q = p est premier alors quici on consid` ere le cas g en eral q = pm . En usant de la m eme approche bas ee sur des propri et es combinatoires des ensembles sommes et produits, nous etablissons des estim ees non-triviales sous des hypoth` eses tr` es faibles (et essentiellement optimales). Si n satisfait la condition pgcd n, q1 < p q 1 pour tout 1 < m, |m p 1
Typeset by AMS-TEX

o` u > 0 est x e et arbitraire, on a lestim ee (xn ) < cq 1

x Fq

pour tout caract` ere additif non-trivial de Fq et o` u = () > 0. 1. Denote q = pm with p prime, m Z, m 1. Non-trivial subelds of Fq are of size p where 1 < m, |m. Denote m1 T r(x) = x + xp + . . . + xp the trace of x Fq . Let (x) = ep T r(x) , F q be a nontrivial additive character of Fq . Our aim is to extend certain estimates on exponential sums of the type (xn )
x Fq

(1.1)

and
j t1

(g j )

t1 t = ord(g )

(1.2)

obtained in [BGK] for prime elds (m = 1) to the general case (m 2) (in (1.2), we denoted ord(g ) the multiplicative order of g F q ). More precisely, it was shown in [BGK] that if q = p and gcd (n, p 1) < p1 ( > 0 arbitrary) in (1.1) resp. t t1 > p in (2.2) , then | xFq (xn )| < p1 (resp. | j t1 (g j )| < t1 p ), where = () > 0. The method involved in [BGK] as well as here is the sum-product approach, which permits us to establish non-trivial bounds in certain situations where classical methods such as Stepanovs do not seem to apply (see [KS] for details). Our main results are the following Theorem 1. Assume in (1.1) that n|(pm 1) and satises the condition gcd n, pm 1 < p q 1 for all 1 < m, |m p 1 (1.3)

where > 0 is arbitrary and xed. Then max

aFq xFq

(axn ) < cq 1

(1.4)

where = () > 0. and 2

Theorem 2. Assume in (1.2) that g F q and t t1 > q and for some > 0. Then again max
aFq j t1

max gcd (p 1, t) < q t

1<m |m

(1.5)

(ag j ) < cq t1

(1.6)

where = () > 0. Remark. The classical bound (xn ) (n 1)q 1/2

x Fq

(1.7)

becomes trivial for n > q 1/2 . The rst nontrivial estimate when n > q 1/2 was obtained 1 1 in [S], considering values of n up to p 6 q 2 . Condition (1.3) (and similarly (1.5)) has clearly to do with the presence of nontrivial subelds of Fq , which we do not want j to contain most of the multiplicative group {xn |x F q } ( and {g |j t} resp.). A condition of this form is obviously needed. 2. As pointed out earlier, we rely on the same approach as in [BGK]. The proof of Theorem 2 (which implies Theorem 1) will be based on the following two results. Proposition 3. Let A Fq and |A| > q . Let > > 0 and assume |A ( + S )| < q |A| whenever Fq and S Fq satises the condition |S | < q 1 20 and |S + S | + |S.S | < q |S |. (ax1 . . . xk ) < q |A|k .
x1 ,... ,xk A

(2.1)

(2.2) (2.3)

Then for some k = k () Z+ and = () > 0 max

a Fq

(2.4)

In (2.3), we denoted S + S = {x + y : x, y S } (resp. S.S = {x.y : x, y S }) the sum-set (resp. the product-set). For small > 0, condition (2.3) expresses the property that both S + S and S.S are not much larger than S . Hence it is important to understand the structure of such sets. The next result provides the required information. 3

Proposition 4. Assume S Fq , |S | > q and |S + S | + |S.S | < K |S |. Then there is a subeld G of Fq and F q such that |G| < K C |S | and where C = C ( ). Proposition 3 is essentially Theorem 3.1 in [BC]. The only dierence is that in [BC] we consider subsets of a ring R = Zgj instead of a eld Fq ; but the essentially general argument carries over verbatim to the present situation (in fact it simplies since the set R\R of non-invertible elements is trivial here). The proof of Theorem 3.1 in [BC] uses only the additive Fourier transform. We may again identify the set of additive characters of Fq with Fq , letting (x) = ep T r(x) ; where ranges in Fq . Proposition 4 appears in [BKT], as a byproduct of the proof of the sum-product theorem in prime elds. 3. With Proposition 3 and 4 at hand, the proof of Theorem 2 is rather straightforward. For simplicity, take t1 = t (considering the complete sum), in which case A = {g j : 0 j < t} is a multiplicative subgroup of F q . Assuming A satises conditions (2.1)-(2.3) from Proposition 3, the conclusion (2.4) is then simply max
a Fq xA

(2.5)

|S \G| < K C

(2.6)

ep (y ) = e

2iy p

(ax) < q |A|

(3.1)

which is (1.6). (To treat incomplete sums, i.e. t1 < t, some minor additional technicalities are involved). Assume that for some one has |A ( + S )| q |A|
with S satisfying (2.2), (2.3). Thus |S | > tq > q > q 2 if < 2 . 4

(3.2)

 Apply Proposition 4 to the set S with = 2 , K = q .

The subeld G satises by (2.5) and (2.2) |G| < q C |S | < q 1 20 +C () < q taking small enough. Hence G is nontrivial and |G| = p for some < m, |m. From (2.6) and (3.2) |A ( + G)| > q |A| q C () > implying that |{(s, s ) : 0 s, s t 1, g s g s G}| > Equivalently, we may write |{(s, s ) : 0 s, s t 1, g s g s +s G}| > 1 2 2 q t . 4 1 2 2 q t . 4 (3.5) 1 q |A| 2 (3.4) (3.3)

In particular there exist some s = 0 such that denoting 1 = (1 g s )1 |{s : 0 s t 1, g s 1 G}|

q 1

q 2 t.

(3.6)
1

p Let g = g0 t , where g0 is a generator of F q . Since by (3.3) x it follows from (3.6) that

= 1 for all x G ,

|{s : 0 s t 1, g0 t

q 1

(p 1)s

p = 1

}|

q 2 t.

= 1, or equivalently Therefore there is some 0 < s q 2 such that g0 t t|s(p 1). But then gcd (t, p 1) > q 2 t, violating assumption (1.5). References
[BGK]. J. Bourgain, A. Glibichuk, S. Konyagin, Estimates for the number of sums and products and for exponential sums in elds of prime order, to appear in J. London Math. Soc. [BKT]. J. Bourgain, N. Katz, T. Tao, A sum-product estimate in nite elds and their applications, GAFA 14 (2004), no. 1, 2757.

q 1

(p 1)s

[ BC]. J. Bourgain, M.-C. Chang, Exponential sum estimates over subgroups and almost subgroups of Z q , where q is composite with few factors, to appear in GAFA. [KS]. S. Konyagin, I. Shparlinski, Character sums with exponential functions and their applications, Cambridge UP, Cambridge, 1999. [S]. I. Shparlinski, Bounds on Gauss sums in nite elds, Proc. AMS, Vol. 132, no 10, 2817 2824. Institute for Advanced Study, Olden Lane, Princeton, N.J. 08540, U.S.A. E-mail address : bourgain@math.ias.edu Mathematics Department, University of California, Riverside, CA 92521 E-mail address : mcc@math.ucr.edu