Fibre Products, Character Sums,
and Geometric Goppa Codes
Serguei A. Stepanov
Abstract
The purpose of this paper is to construct some new families of smooth projective curves over a finite field Fq with extremely many Fq-rational points. The genus in every such family is considerably less than the number of rational points, and these two facts allow to produce sufficiently long geometric Gappa codes with very good parameters.
§1. Introduction
Let
Fqbe a finite field with
qelements. The linear space
F;can be provided with the structure of a metric space if we introduce
the
Hamming distance d( x, y)between
x= (
x1, ... Xn) EF;
and
y
=
(Y1, ... , Yn) EF;
as the number of coordinates in which
xand
y
differ:
d(x,y)
=
#{
i
11::;
i::;
n, x;-/- y;}.Each linear subspace
Cs;;;
F;
is called a
linear q-ary code of length n.The non-negative integers M =
ICI
and k = log
qICI
are called the
cardinality
and
log-cardinalityof the code
C,respectively (in fact,
227
228
Stepanovk
=
dimF
qC). The minimum distance of the code
C
is defined as
d
=
min{
d(x,y)I
x,yE
C,x-=f. y}.
A q-ary linear code
Cwith
parameters n, kand
dis called an
[n,
k, d]q-code.Elements of care called
code-words(or
code-vectors),and their components are called
positions( or
coordinates).Define
the
information rateR and
relative minimum distance8 of the code
C
as
R
=
k/n
and
8
=
d/n.
It is clear that O :S
R
'.S 1, 0 :S
8
'.S 1.
Let us briefly explain why the codes are called
error-correcting.The following question is essentially one of the central problems of
information theory. We consider
informationpresented as a very
long sequence of symbols from the
alphabet Fq,In this sequence
each symbol occurs with equal probability. This information is sent
to a receiver over a so-called
noisy channel.In the model that we
consider there is a small fixed probability
p that a symbol, which
is sent over the channel, is changed to one of the other symbols.
Such an event is called a
symbol-errorand
pis the
symbol-error probability.As a result a fraction
pof the transmitted symbols arrives
incorrectly at the receiver at the end of the communication channel.
The aim of coding theory is to lower the probability of error at the
expense of spending some of the transmission time or energy on
redundant symbols.
The basic idea of the theory can be explained in
the sentence that follows. When we read printed text we recognize a
printing error in a word because in our vocabulary there is only one
word that resembles (is sufficiently close to) the printed word.
In
block codingthe message is split into parts of
ksymbols. The
encoding
is an injective map from
Fqkto
F;(where
n � k).In other
words we take some [n,
k,
d]
q-code
C
and fix an embedding
, : F;-.'.::+ C �Ft.
The transmission is now
R
=
k/ntimes slower, which justifies the
term "information rate" for
R.
Instead of the part
zE
F,7
of the
message we transmit the corresponding word
x = 1(z) of length n.
On the end of the channel we obtain a distorted word
x'E
F; and
we transform it into the nearest code-word
x"E
C
(i.e. we decode
the message on the maximum likelihood basis). This transformation
Gappa Codes.
229
can be defined by some
decoding map( :
F;-+
C.If the number of
distorted symbols is at most
l d;lJ,
then
x"=
x,i.e. the decoding is
correct. The
maximum likelihood decodingis an ideal that is almost
unattainable. Usually we just give a map ( :
U
-+
C,
where
U
is the
union of all
ballsB
t(x)
= {
y E F;
I
d(x, y) St}
of radius t S
l d;lJ
centered in the elements
xE
C.
In this case
we speak about
decoding up to t(
or correcting
terrors). Usually
t
=
l d;lJ,
but sometimes it is less.
Example 1
(binary Hamming codes).Let
rbe a positive integer,
n
=
2
r- 1, and H
=
(aij) the unique
rx
nbinary matrix whose
columns are pairwise linearly independent over the field F2 (hence
the columns of Hare distinct and differ from the zero column). The
set of all solutions x
= (
x
1, ... ,
x
n) E
Ff of the system of linear
equations
H ·
XT=
0,
forms an (
n-
r)-dimensional subspace C
of
F;,
The linear code
C
is
called a
binary Hamming code.The Hamming code
Ccorrects single
errors. Indeed, let x be a code-vector and x'
=
x+e a received vector.
Assume that the
error-vector e= (
e
1 , ... , en)has a unique non-zero
component, say
Xi=
l. Multiplying the
parity-checkmatrix
Hby
the vector
x'=
x+
e,we obtain
H · (x
+
ef
=
H ·e
t= s
TThen the binary r-dimensional vector
s= (
s
1 , ... ,S
r) ( called a
syndrome
of
x')is equal to the i-th column of
H.Since we know the
matrix
H
we can compare the syndrome
swith columns of
H
and
find the unique position where the single error occurs. This means
that the minimum distance
d
of
C
is at least 3. It is easy to see that
in fact
d
=
3, so
C is a linear [n,
n-
r,3]z-code.
Now we note that all the unit balls B
1(
x)
centered in code-vectors
x E
C are disjoint, and also IB
1(
x)
I
=
1
+
n
=
2
r. In that case
L
IB
1(x)I
=
2
n-
rlB
1(x)I
=
2
n-
r2
r=
IF:I ,
xEC230
Stepanov so the codeC
provides a densest packing of the spaceF:f.
The codes with such a property are called perfect.Example 2 ( Reed-Solomon codes). Let P
= {
011, ... , an} � Fqbe a subset of cardinality n :::;: q. Consider a linear space
L(
m) of all polynomialsf
E Fq[u] in one variable u of degree at most m.Its dimension over Fq is dim
L(
m)=
m+
1. For n > m a non-zeropolynomial
f
(
u)
EL(
m) cannot vanish at all points ofP.
Moreover, it has at least ( n - m) non-zero values at points of the set P. Hence if n > m, the evaluation mapis injective and its image
C
is a linear [n, m+
1, n - m]q-code calleda Reed-Solomon code of degree m
(
or RS-code). The parameters of such a code satisfy the condition k+
d=
n+
1 and k=
m+
l can be freely chosen between 1 and n.The construction of Reed-Solomon codes can be extended as fol lows. Let X be a smooth projective curve of genus g
=
g(X) defined over a finite field Fq. We describe briefly the Goppa [7] constructionof linear [n,
k,
d]q-codes associated to the curveX.
Let { x1, ... , xn}be a set of Fq-rational points on
X
and set Do=
X1+ · · · +
XnLet
D
be a Fq-rational divisor onX.
We assume thatD
has support disjoint from Do, i.e. the points Xi occur with multiplicity zero in D.Denote by Fq(X) the field of rational functions on X and consider
the following vector space over Fq:
L(D)
= {
f
E Fq(X)*I
(!)
+
D � O} U {O} .The linear [n,
k,
d]q-codeC
=
C(D
0,D)
associated to the pair(Do, D)
is the image of the linear evaluation mapEv: L(D)-+
Ft
,
f � (f(x1), .. . , f(xn)) .Such a q-ary linear code is called a geometric Gappa code. If deg D <
Coppa Codes 231 Dually, denote by D(X) the Fq(X)-vector space of rational dif ferential forms on
X
and consider the following linear space overFq:
D(Do
-D)
= {
w
E D(X)*I
(w)
+
Do
-D �
0} u {O} . The linear mapdefines a linear [n,
k,
djg-codeC*
=
C*(D
0,D)
associated to the pair(Do, D) .
If degD
>
2g -2, the map Res is injective, so thatC*
�D(Do
-D)
�L(I{
+
Do
-D) ,
where J{ is a canonical divisor onX.
Each linear [n,k,
d]q-codeC
defines a pair of its relative parameters (8, R), where 8=
d/n is the relative minimum distance and R is the information rate ofC.
The points (8, R)
form the set of code points i�lin � [O, 1]2. Letu:n
denote the subset of limit points of �tin. In other terms, (8, R)
EU�in
if and only if there exists an infinite sequence of different linear codesCi
with distinct relative parameters8i
=
8(Ci) andRi
=
R(Ci)
such thatEm (8i,Ri)
=
(8, R) .
i-HX>
If 8 > 0 and R > 0 such a family of codes
Ci
is called asymptotically good. The structure of U!in can be described as follows (Aaltonen [1], Manin [9] ): there exists a continuous function a�n( 8) such thatmoreover, a�n(O)
=
1, a�n(8)=
0 for(q
-l)/q �
8 � 1, anda�n(8) decreases on the interval [O, (q-1)/q]. The function a�n(8) is unknown in general, but we are able to find a number of upper bounds and the Gilbert-Varshamov lower bound
where
232 Stepanov The Gilbert-Varshamov bound has a series of remarkable statistical properties. For example, it can be easily shown that the parameters of almost all linear codes lie on the curve R
=
1 - Hq( 8). For thisreason it was thought for a long time that a�n( 8)
=
Hg( 8). Onlyrecently this assumption was disproved (for
q
2: 49) with the help of very deep methods of algebraic geometry.It follows from the Riemann-Roch theorem that the relative pa rameters
R
=
k/n
and8
=
d/n
both forL-
and D-constructions satisfy ( see [24], [25] ).g-1
R>l-8---.
-
n
(1)
In order to produce a family of asymptotically good geometric Goppa codes for which R
+
8 comes above the Gilbert-Varshamov bounda�n( 8) 2: 1 - Hq( 8) ,
one needs a family of smooth projective curves with a lot of Fq
rational points compared to the genus. Examples of such families are provided by classical modular curves
X
0(N)
andX(N)
(Ihara [8], Tsfasman-Vladut-Zink [26], Moreno [10]), or by Drinfeld modular curves (Tsfasman, Vladut [25, Chapters 4.1 and 4.2]). So, if q=
pvis an even power of a prime
p,
there exists an infinite sequence of geometric Goppa codes Ci which gives the lower boundO'.�n
( 8) 2: 1 - b - (
ylq
- 1t
1The line
R
=
1- 8-(Jq-lt1 intersects the curveR
=
1-Hq(b) for
q
2: 49. A much easier proof of this result based on consideration of a sequence of (modified) Artin-Schreier coverings of the projective line IP'1 ( Fq) was recently proposed by Garcia and Stichtenoth [3]. Laterthey [4] discovered a still easier sequence of smooth projective curves with the same property.
In terms of algebraic geometry the problem on construction of asymptotically good codes can be reformulated as follows. Let Nq
(g)
denote the maximal number of Fq-ra.tional points on a smooth pro jective curve X over Fq of genus g
=
g(X), and. Nq(g)
A(q)
=
hm sup --g-+oo gGappa Codes
233
It follows from the Hasse-Weil bound [27] that
A(q)
s;
2ytq.
The Serre bound [16]
IN
q- (q
+
1)1
:s; _q[2Jq]
yields
A( q)
:s;
[2ytq]
A much stronger upper bound
A(q)
s;
ytq -1
was obtained by Drinfeld and Vladut [2] . This is the best possible
upper bound, and the construction of asymptotically good geometric
Goppa codes is reduced to the construction of a family of smooth
projective curves over F
q, for which
A
(q)
is close to the Drinfeld
Vladut bound. So if
q = p
11is an even power of a prime number
p,
the result of lhara [8], Tsfasman-Vladut-Zink [26] (see also Moreno
[10] and Garcia-Stichtenoth [3], [4]) implies
A(q)2:ytq-1.
If
q is an odd power of p, the result of Serre [16] provides the
existence of an absolute constant c > 0 such that
A(q)
2:
clogq.
In some cases the Serre lower bound was improved by Perret [15]
and Zink [28]. In particular, the Zink result yields
A(
q - q
3
) >
2(
q
2+
-
2
1)
In this paper we construct rather long geometric Goppa codes
coming from fibre products of superelliptic curves.
234 Stepanov
At first we consider a family of smooth projective curves )C given
over Fq by equations
(2)
where fi(u) are relatively prime square-free polynomials in Fq[u] of a special form. Every such curve is actually a fibre product of hy perelliptic curves. The main point of the paper is to calculate the genus g(Xs) (Lemma 1) and determine the number Nqv(Xs) of Fq
rational points (Lemma 4) on the curve X8• We show that the ratio
g(Xs)/Nq(Xs) is small enough, and deduce from (1) that the cor
responding geometric Goppa codes
C(D, Do)
andC*(D,
Do) have rather good parameters. For small values of s, these parameters are comparable with the parameters of codes on Artin-Schreier coverings introduced by Garcia and Stichtenoth [3], [4]. In particular, ifs=
1 then the codesC(D, Do)
andC*(D, Do)
have the same parameters as the codes on Hermitian curves (see [24, Sec. VII.3] ). Unfortu nately, the parameter .s in our construction is bounded by q112, and as a result the genus g(Xs) is bounded by(q
- 3)2,.fa-2+
1However, since the above upper bound is large enough for q 2 q0, the curves Xs provide sufficiently long geometric Goppa codes ( with
n
'.S q2y'q-I). Moreover, some modification of the polynomialsf,
(x)
(Ozbudak [11], [12] ) allows to construct linear [n,
k,
d]q-cocles overFq for any
n'.Sq(q -l)2
q.
It is necessary also to note that construction of linear codes on fibre products of some Kummer coverings (2) are closely related to the well-known combinatorial problem on configurations of lines in the finite plane Fq2 (Ozbudak [12], Ozbuclak and Thomas [14] ).
The genus g(Xs) can be easily c:alculated using the Hurwitz genus
formula. However, we prefer to use a slightly more complicated argument, which allows us to find explicitly a basis of the space
D(Do
-D).
This provides an easy way to write out generator maGappa Codes 235
Applying to curves Xs the Goppa constructions, we obtain the
following results (Stepanov [20], [21]).
Theorem 1. Let p
>
2 be a prime} v>
1 an even integer} and Fq a finite field consisting of q=
pv elements. For any positive integers s:S
q112 and r> (
sq1!2 - 3)2s-2 there exists a geometric Coppa[n, k, d]q-code C
=
C(D0, D) with r<
n :::; (2ql/2 _ 8 )q1;22s-t , k
2'.
r -(sql/2 -3)2s-2 ' d2'.n-r .Theorem 2. For p} q and s as before} and for any positive integer r > (sq112 - 3)2s-t there exists a geometric Coppa
[n,
k.d]q-code C*
=
C*(Do, D) withr _ (sqt/2 _ 3)2s-2 < n:::; (2ql/2 _ s)q1/22s-l , k
2'.
n -r+
(sq112 - 3)2s-2 ,d
2'.
r -(sq112 - 3)2s-lCorollary 1. The relative parameters R
=
k/n and 8=
d/n of the above codes satisfy( sql/2 _ 3)2s-2 R > l-8-
-
n
In particular} for n
=
(2q1l2 -s )q1!22s-l we have sq1/2 -3R > l-8---- 2(2 ql/2 -s )ql/2
By a suitable concatenation one gets reasonably good codes over
FP. Indeed, let k0 > 1 be an even number. Applying a linear
[n0, k0 , do]p-code C0 to an [n, k, d]q- code C
=
C(Do, D) over Fq,where q
=
pk0, we obtain an [n', k', d']P-code C' with parameters n'=
n0n, k'=
k0k, d'=
d0d .Let us denote by Ro
=
k0/n0 and 80=
d0/n0 the relative parameters of the code Co.236 Stepanov
Corollary 2.
For any positive integers n0>
l , s :=::; q112 andr
>
(sq112 - 3)2s-2 there exists a linear[n',
k', d']p-code C' with nor
<
n'=
n0n :=::; n0(2q112 - s )q1122s-l,k' 2'. ko(r
-
(sq112 - 3)2s-2), d' 2'. d0(
n-
r)Relative parameters R'
=
k' /n' and 8'=
d' /n' of the code C' satisfy R'+
8' 2'. Ro (;�--
(sql/2 � 3)2s-2)+
80( 1 - �)
Applying [no, ko, do]p-code C0 to an [n, k, d)q-code C*
=
C* (Do, D) we obtain the following result.Corollary
3. For any positive integers n0>
l , s :=::; q112 andr
>
(sq1l2 - 3)2s-2 there exists a linear[n",
k", d"Jp-code C" with no(r
-
(sq1l2 - 3)<
n" = non :=::; no(2q112 - s)q1122s-1, k" 2'. ko(n-
r+
(sq112 - 3)2s-2),d" :=::; do(r
-
(sq1/2 - 3)2s-1 )Relative parameters R"
=
k" /n" and 8"=
d" /n" of the code C" satisfy(
r (sq1/2 _ 3)2s-2 ) R"+
8">
Ro l - -n+ ---
n(
r(
sql/2 -3)2s-1 ) +80- -
---'----n nThe above results can be easily extended to the case of codes coming from fibre products of hyperelliptic curves over a finite field Fq with q
=
p11 elements, where v>
l is an odd number (Stepa.nov and Ozbudak [22] ).Theorem 3.
Let v>
l be an odd number, Fq a finite field of characteristic p>
2 consisting of q=
p11 elements, and s an integer such that1
< <
- -
s ---2p11+
4 p(v-1)/2(p+
1) _ 2Gappa Codes
Moreover} let r be an integer satisfying
2s-2 ( (p(v-l )/2 (p + 1 ) - 2)s -
4)
< r < 2spvThen there exists a linear
[n,
k , d]q-code C = C(D0, D) withparameters
r < n '.S 28pV,
k = r - 2s-2 ( (p(v-1 )/2 (p + l ) - 2)s - 4) , d ?_ n - r .
237
A similar result holds for codes coming from fibre products of superelliptic curves
zt
=
fi (u) ,
(3)
where µ ?. 2 is a positive divisor of q - 1. The following theorems
(Ozbudak [11] ) are based on the use of lower bounds for character sums obtained by Gluhov [6] and on the construction of geometric Goppa codes coming from superelliptic curves (Ozbudak and Gluhov
[13]).
Theorem 4.
Let T be a positive integer} Fq,. a finite field of characteristic p > 2 with qr elements} and µ ?. 2 a positive divisor
of q - 1 . Then:
(i) If T > 1 is an odd number} s an integer such that
1
<
S<
2µ( qr
+
1)
- - (
µ - l ) (q(r-1 )/2(q + l ) - 2) J
and r is an integer satisfying s-1
y
((µ - l ) (q(v-1)f2(q + 1 ) - 2)s - 2µ) < r < µsqr J
there exists a linear [n, k , d]qr -code C
=
C(D0, D) withparameters r < n '.S µsq7, s-1 k ?_ r
- Y
((µ - l ) (q(r-l)f2(q + 1 ) - 2)s - 2 µ) J d ?_ n - r ;238
Stepanov moreover) if r>
µs-1 ((µ - l)(
q(T-1)/
2(q + 1) -2)s -2µ)
i then s-1k
=
r- T
((µ - l)(
q(T-l)f2(
q+ 1) -2) -2µ)
(ii) If T
=
2 ( mod 4)
1 s an integer such that2
µ(
qT+
1)
1
<
s<
J- - (µ -l)(q
T/
2-l(q
2+ 1) -2)
and r an integer satisfying
s- 1
T
((µ - l)(q
T/
2-1(q
2+ 1) -2) -2µ) <
r<
µsqT !
there exists a linear
[n,
k, d]qr -code C=
C(Do, D) with parameters s-1k ;:::
r- T
((µ - l)(q
T/
2-1(q
2+ 1)-2)s -2µ) ,
d ;::: n-r ; moreover1 if r>
µs-1 ((µ - l)(q
T/
2-l(q
2+
1)-2)s -2µ)
i then s- 1k
=
r- T
((µ - l)(
qTf2-1(
q 2+
1)-2)s -2µ)
(iii) If T
=
0 ( mod 4)
1 s an integer such that2
µ(
qT+ l)
1
<
s<
1Coppa Codes
and r an integer satisfying s-1
T
((µ -l )(qTf-1(q2 + l)s -2q) -2µ)<
r<
µsqT }there exists a linear
[n,
k, d]qr -code C=
C(D0, D) with parameters r < n :S µsqT, s-1 k 2 r-T
((µ -l )(q7 !2-1(q2 + 1) -2q)s -2µ) , d 2 n-r ; moreover} if r>
µs-1 ((µ -l )(qT/2-l(q2 + 1) -2q)s-
2µ) J then s-1 k=
r - T
((µ-l )(q7/2-1(q2 + 1)-2q)s-2µ)239
Theorem 5 . Let T be a positive integer} Fq,. a finite field of
characteristic p
=
2 with q7 elements} and µ 2 2 a positive divisor
of q - l . Then:
(i) If T
>
l is odd} s an integer such thatl < s < - - ( 2µ(qT + l )
µ-l )(q( T-1)/2(q + l )-2q) J and r an integer satisfying
s-1
T
((µ-l )(q( T-1)/2(q + 1) -2q)s -2µ)<
r<
µsqT }there exists a linear
[n,
k, d]qr -code C=
C(Do, D) with parametersr < n :S µsqT, s-1
k 2 r
-T
((µ -l )(q( T-l)/2(q + 1) -2q)s-
2µ) , d 2 n-r ;240
Stepanov moreover, ifr >
µs-1 ((
µ- l)(qCr-1)/
2(q + 1) -2
q)s -2µ)
then s-1
k
=
r- Y
((µ - l)(q
(r-l)/
2(q + 1)-2q)s -2µ)
(ii} If T
=
2 (mod 4),
s an integer such that2
µ(
qT+
1)
1 <
- - (µ - l)(q
S<
r !/
2-l (q
2+1)-2q
2)
and r an integer satisfying
s
-1
y
((
µ- l)(q
T/
2-1 (q
2+ 1)-2q
2)s -2µ) <
r<
µSqT Ithere exists a linear
[n,
k, d]qr -code C=
C(Do, D) with parameters s-1
k �
r- Y
((µ - l)(q
7/2-1 (q
2+ 1) -2q
2)s -2µ) ,
d � n-r ; moreover, if r>
µs-1 ((
µ- l)(q
r/
2-1(q
2+ 1) -2
q2)s -2µ)
then s-1
k
=
r - Y
((
µ- l)(
qr f2-1 (
q2+
1) -2
q2)s -2µ)
(iii} If T
=
0 ( mod 4),
s an integer such that2
µ(
qT+
1)
1 <
S<
/
Gappa Codes
and r an integer satisfying
s-1
T
((µ - l ) (qT/2-1 (q2+
1)
-2q)s - 2µ)
<
r<
µsqT Jthere exists a linear
[n,
k,
d]qr -codeC
=
C(D
0,D)
withparameters s-1 k � r
- T
((µ - l ) (q7!2-1 ( q2+
1) -
2q)s - 2µ) , d � n - r ; moreover} if r>
µs-1 ((µ - l ) (qT /2-l (q2+
1) -
2q)s - 2µ) J then s-1 k=
r- T
((µ - l ) ( q7/2-1 (q2+
l ) - 2q) s - 2µ)24 1
The advantage of Theorems 4 and 5 is that the fibre products of
superelliptic curves provide a class of much longer linear codes than
the class of codes coming from fibre products of hyperelliptic curves.
Moreover, for odd
µ ,the fibre products of superelliptic curves over
a finite field of characteristic 2 gives a possibility to construct rather
long binary linear codes with very good parameters.
A similar construction of non-singular projective curves with a.
lot of F
q-rational points based on the use of fibre products of some
special Artin-Schreier curves was independently considered by van
der Geer and van der Vlugt [5]
§2. Notation and Lemmas
Let F'
qbe an algebraic closure of F
qand A/
+1
be the (
s+
1 )
242 Stepanov
Lemma 1 .
Let f1, . . . , fs be pairwise cop rime square-free manic polynomials in Fq[u] of the same odd degree m 2:1 ,
and Y the fibre product in _As+l given over Fq by equationsz;
=
fi ( u) , 1 :S i :S sThen the genus g
=
g(X) of the smooth projective model X of the curve Y isg
= (
ms - 3)2s-2+
1
Proof.
LetX
be a smooth projective model of the curveY.
Denoteby Vx the canonical valuation of the function field F'q(X), and by
D[X] the space of regular differential forms on
X.
The affine curveY
is easily seen to be smooth. IfY
is its projective closure, thenX
is a normalization of
Y
and we have the map 7/J :X
-tY,
whichis an isomorphism between
Y
and 7/J-1(Y). Hence it follows thatg = g(X) = g(Y).
The rational map ( u, z1, . . . , z .. ) 1-t u of the curve
Y
in A 1 determines a morphism 'P : X -t JfD1 of degree 2s, so that for u0 E A 1 ei
ther t.p-1(u
0 ) consists of 2s points of the form x'
=
(u0 , ±z1 , . . . , ±zs)in each of which vx,(t)
=
1 for a local parameter t at u0, or elset.p-1 ( u0 ) consists of 2s-l points of the form x?
= (
u0 , ±z1 , . . .
±
Z i-1, 0, ±zi+l, . . . , ±zs), and Vx11I (i)=
2.Let us consider the point at infinity u00 E JfD1 . If the coordinate on A.1 is denoted by u, then t
=
u-1 is a local parameter at u00•
If t.p-1 ( u00
)
were to consist of 2s pointsxt;,J,
then at each x00
=
xtl the function t would be a local parameter. Hence it would follow that Vx=(t)
=
1 and Vx=(fi(t))=
-m. But since m is odd,this contradicts the condition that Vx= (fi ( u))
=
2vx= ( Z i). Thust.p-1(u00
)
consists of r=
2s-l points xtl, 1 :S T :S r, with projectivecoordinates xtl
=
(0, 1, ±1, . . . , ±1, 0). It follows that X=
Y U { x�,l } U · · · U{xt,l } .
At any such pointx
00=
xtl we have Vx= (u)=
-2 and Vx= (z i)
=
-m.Let us now find a basis of the space D[X] over the field Fq. Any
Gappa Codes
differential forms w0
=
P0(
u) du andPi1 , ... ,i" ( u) du
W i1 , ... ,ic,-
=
'
Z i1 . . . Zi"
243
where i1, . . . , i17 are integers satisfying the condition 1 :::; i1 < · · · <
i17 :::; s and Pi1 , ... ,i" are polynomials in F'q [ u]. Indeed, the differential
form
du
w'
1.. 1 ' • • • ,i.
o-=
�---is regular at any point u0 E A 1 with the condition Z i ( u0)
#
0 for i E{i1, . . . , i17 }. Now if z i(u0 )
=
0 for a unique i E {i1, . . . , i17 }, then Ziis a local parameter at
xi'
=
(uo, ±z1, . . . , ±z i-1, 0, ±zi+I , . . . , ±zs) ,so that Vx11(z i)
=
1 and Vx11(u - uo)=
2. Therefore, Vx11(du)=
1 andagain wt ,.' .. ,i" is regular at uo. The form wb
=
du is �lso regular at any point u0 E A 1. Thus, the differential forms wb=
du and wi1, ... ,i,,.form a basis of the .F'q[u]-module D[Y] .
It remains to clarify which of the forms w0 and Wi1 , ... ,i" are regular
at points x�,l,
. . . ,
xt,l. Let x00 be one of these points. If t is a local parameter at x00, then u=
t-2u', Z i=
t-m zL where u' and<
areunits in the local ring Ox=· Therefore wt , ... ,i" = tm µ-3B i1, ... ,i"dt, with
ei1 l ... i a unit in O' CJ' x 00 ' hence (w� ; ) "1 l " " J " O"
=
(mO" - 3) . Xoo. Thus, thedifferential form
Pi1 , ... ,i" ( u) du
W i1 , ... ,ic,-
= ----
Z i1 . . . Zi"
is regular at x00 if and only if
v (P · (u)) > -(mO" - 3) Xoo i1 , . . . ,i cr
-This means that deg Pi1, ... , i"(u) :::; (m0" - 3)/2 and hence
deg P,, , . ,,. (u) :S {
ffi(J" - 4 2 ffi(J" - 3 2 if O"=
0 (mod 2) if O"=
1 (mod 2)The differential form w0
=
Po du is not regular at x00 for any non-zero polynomial Po E k"[u], so the regular differential formsI I
w i1 , ... ,i" ' uw i1 , ... ,i" '
n I
244
where 1 :S i
1< · · · < i
r;:S
sand
{
mcr -
4
n-
-
mcr - 3
2
2
if
er_ 0 ( mod 2)
if
er=
1 (
mod
2)
form a basis of the space D[X] over
F
q. Therefore
dim - D[X]
Fq=
!
2
o- = 1 l <i1 < . . ·<ia- < s(mcr - 2)
o-:O ( mod 2) --1
+-
2
o-=1 l < i1 < . . ·<ia- < s o- : l ( mo d 2) --(mcr - 1)
Stepanov cr=l o-:O ( mo d 2)(;) -t � (;)
o- :: 1 ( mo d 2)1
=
- (ms2
s-l - 2
8 -2
s-l
+
2)
2
and hence
9
=
g(X)
=
dimpq D[X]
=
(ms- 3)2
s-
2+
1
This completes the proof.
Let p be a prime number,
11a positive integer and Fq a finite field
with
q=
p11elements. The field Fq is a Galois extension of the prime
finite field F
pof degree
IIwith the cyclic Galois group of order
11 .The action of a generator 8 of this group on an element x E Fq is
given by the rule
O(x)=
xP.The map
1 v - 1
norm
v(x)
=
X . O(x) . . . ev- (x)=
X . xP . . . xPof F
qonto F
pis the
normof the element
x.Let x be a multiplicative character of the field Fp and x an element
of F
q. Set
Gappa Codes
245
and call Xv a multiplicative character of the field Fq induced by the character X.
Now let f be a square-free polynomial in the ring Fq [u] of degree m and
x
a non-trivial quadratic character of Fp. Consider the character sumSv(f)
=
L
Xv(J(u))=
L
x(normv (f(u)))and recall the well-known Weil bound [27] (see also Stepanov [19, Chapters 1 and 5] ):
I
Sv(f)l:S
2[
m
;
l]
q1 12The following result (Stepanov [18, Theorem 3] ) shows us that Weil's bound cannot be sharpened essentially in any extension Fq of the field Fp.
Lemma 2. Let Fq be a finite field with q
=
pv elements of characteristic p > 2 and Xv the character of Fq induced by a non-trivial quadratic character X of the field Fp. If v
>
1 then for the square-free polynomial f E Fq[u] given by{
U+
uP"/2 ifI/
=
Q ( mod 2) f(u)=
(u+
uP( v-1)/2 )(u+
Up( v + 1 )/2 ) ifI/ =
1 (mod 2) J we haveL
Xv(f(u))=
{ q - q ( 1 / 21)
1 / 2q
- 1
if v if v=
=
1 ( mod 2) 0 ( mod 2) u E FqProof. Let v > 1 be an even number. As far as uP"
=
u in Fq, then for any u E Fq we haveV V
normv (f(u))
=
II (u+
upv/2 t-1=
II (uPi-1+
upv/2+i-1 )i=l i=l
v/2 v/2
II (
uPi-1+
uPv/2+i-1 )II (
uPv/2+j-1+
uPj-1 )i=l j=l
246
StepanovTherefore,
L
Xv (J(u))
=
L
x(norm
v(J(u)))
=
q-
N ,
where N is the number of elements of the set A
= {
u E F
qI
f ( u)
=
0}.
Since f(u) = u(l
+
u
P"12 -
1 )we have A = {O} U B, where
v/2 1 }
B
= {
u E Fq I l
+
u
P-
=
0
is the set of roots of the polynomial 1
+
u
P"12 -1 in F
q
. Taking into
account the equality
gcd(p
v/2 - l, p
v- 1)
=
p
v/2 - 1 ,
we obtain from the Euler criterion that the number of roots of the
polynomial 1
+
u
P"12 -1 is equal to p
v/2 - l. In that case
N
=
IAI
=
1
+
IBI
=
1
+
(p
v/Z - 1)
=
q
1/2 ,
and hence
L
X
v(J(u)) = (q
l/2 - l)q
l/2 .
uEFqGappa Codes
Let now v > l be an odd number. In this case for any u E Fq
norm
v(f (
u))
=
II (
uPi- 1+
uP("- 1 ) /2+ i -1 ) ( uPi-1+
uP("+ 1 J /2+ i -1 ) i=l ( v- 1 ) / 2 V=
II (
uPi-1+
uP( v-1 ) /Z+ i- 1 )II (
uP i-1+
uP( ..,+ 1 ) /Z+ i -l ) i=l i= (v+l )/2 ( v- 1 ) / 2 V xII (
uPi- 1+
uPc"+ 1 ) /2+ i-1 )II (
uPi-1+
uP(..,+ 1 ) /2+ i -1 ) i=l i= (v+l )/2 (v+l ) / 2 (v- 1 ) /2=
II (
uPi-1+
uPc "- 1 ) /2+ i -1 )II (
uPc ..,+ 1 ) /2+; -1+
uPj-1 ) i=l j=l ( v - 1 ) / 2 (v+l ) / 2 xII (
uPi-1+
uPc "+ 1 )/2+ i - 1 )II (
uP"- 1 ) /2+; -1+
uP; -1 ) i=l j=l (v+l ) / 2 ( v - 1 ) / 2II (
pi -1 p( l-'-l )/2+i-l)2
II (
pi-1 pC "+1 )/2+i-l)2
=
u + u u + ui=l i=l
and hence
L
Xv(f(u))
=
L
x(norm
v(f(u)))
=
q -
N'
,
247
where N' is the cardinality of the set A =
{ u EF
qI
f ( u)O}.
Clearly
N'
=
l and therefore
L
Xv (f(u))
=
q - l
uEFq
This completes the proof.
Lemma 3.
LetF
p be a a prime finite field of characteristic p> 2,
F
q=
F
p"
be an extension ofF
p of even degree v> l
and A the setof roots in
Fq
of the polynomial248
Then:
Stepanov
(i) A is a subgroup of the additive group F/ of the field Fq; (ii) If
{
A1=
A,A
2, . . . , Ar} is the set of all cosets in F// A and{ a1 , a2 , . . . , ar} are distinct representatives of the cosets,
(
v/2Ji u)
= (
u+
ai)+ (
u+
ai )P(1 ::;
i ::;
r ) are pairwise coprime polynomials in Fq[u] ;(iii} r
=/
F//A/=
p11/2.(4)
Proof. The main point is (i). First of all we note that f (O)
=
0. Now, if a and /3 are zeros of f (u), thenf (a
+
/3)= (a +
/3)+ (a +
f3)P"12=
a + aPv/2+
(/3+
/3Pv/2 )=
f ( a) + f (/3)=
0 ,
so that a
+
/3 is also a root of the polynomial f ( u).
Thus A is a subgroup of F/ .To prove (ii) let us suppose that fi (u) and fi (u) for i -=/- j have a common root in Fq, say u
=
e.
In that caseand therefore This yields
(
v/2
ai
-
aj+
ai-
aj )P=
0 ,
and we find that ai
-
aj is a root of f (u), hence ai-
aj E A. But ai-
aj (/. A according to the choice of a1 , . . . , ar, and we arrive at a contradiction.Finally, since / A /
=
p11l2 we find thatCoppa Codes
249
This completes the proof.Lemma 4.
Let Fp be a prime finite field of characteristic p>
2,
Fq an extension of Fp of even degree v
>
1 and s:S
q112 a positive integer. Let Nq be the number of Fq-rational points of the affine curve Y given by equations(2)
with polynomialsv/2
fi(x)
=
(u
+
ai )
+
(u
+
ai)
Pdefined by
(
4) .
ThenProof. We have
Nq
=
L
( 1
+
Xv (fi (u)))· · ·
( 1
+
Xv (fs(u)))and hence
It follows from Lemma 2 and Lemma 3 that
Xv (fi(u))
= { �
if u E Ai if u E F q\
A
and since any two distinct sets
A
and Aj have no common elementwe obtain
Nq pv
+
E
(;)
(pv-
(Jpv/2) pv+
(2s - l)pv
_
s2s-lpv/2(2pv/2 _ S )pv/2 2s-l
=
(2ql/2 _ S )ql/22s-l This proves the lemma.250
Stepanov
Now let F
qbe a finite field of characteristic
p
>
2 with
q
=
p
velements, where v
>
I is an odd number. Consider again the smooth
projective curve X given over Fq by equations (2). Using the same
arguments as in the proof of Lemma 1 we obtain the following result
(Stepanov and Ozbudak [22]).
Lemma 5 .
LetJi, . . . ,
fsE Fq [u]
be pairwise coprime square-free manic polynomials of the same even degreem 2 4
and X the smooth projective curve inJP>
s + l defined overFq
viaThen the genus g
=
g(X) of the curve X isg
= (ms - 4)2
s-Z
+
1
Next, one can show that if v
=
l(mod 2) and
!(
u
) (
=
u + u
pC v-1 ) /2 ) (u + u
p( v+l ) /2 ),
then the polynomials
f(u
+
a) and f(u
+
/3)
are relatively prime for
any distinct a, /3
E Fq (Stepanov and Ozbudak [22]).
Lemma 6.
Let v>
I
be an odd integer,F
q a finite field ofcharacteristic p
>
2
and!(
) (
p( v-1 ) /2 )(
p(v+l )/2 )F [ ]
u
=
u + u
u + u
E
q UFor any distinct
a, /3 E F
q the polynomials f (u
+
a)
and f(u
+
/3)
are coprzme.
Using the result of Lemma 2 we can easily calculate the number
of Fq-rational points on the affine curve Y given by equations
Gappa Codes 251
Lemma 7.
Let v>
l be an odd integer, Fq a finite field of char acteristic p>
2 with q=
pv elements! a1, . . . , a5 distinct elements of FqJ and Nq the number of Fq-rational points on the affine curve Y defined over Fq by
Then
Proof. Since XvUi( u))
=
Xv(]( u + ai))=
1 for all u E Fq, lS
i
S
s,we have
Nq
=
L
(l+
Xv(f1 (u))) . .·
(1
+
XvUs (u)))uEFq
Now we turn into the consideration of fibre products of superellip tic curves (3) with polynomials f i(u), 1
S i S
s of a special form. At first we calculate the genus of the smooth projective model of such a product. The following result is a generalization of Lemma1 .
Lemma 8.
Let f1,j, . . . , fs,j E Fq [u] be pairwise coprime square free manic polynomials of the same degree mjJ for j=
l, 2. Letµ1, µ2 be positive integers! µ 2'.
2
a positive divisor of q-
l! andm
=
m1µ1+
m2µ2 2: µ+
l . Let Y be the fibre product in As+l given over Fq viaAssume that
(m,
µ)=
l or(m,
µ)=
µ. Then the genus g=
g(X) of the smooth projective model X of the curve Y isµs-1
9
= -
2- ((µ - l)s(m1+
m2) - (µ -1))
+
1
if
(m,
µ)=
l! and µs-1252
if (m,µ) = µ.
Stepanov
Let
µ1, µ2be positive integers. Consider the polynomials
f (u)
=
(1+
uq( r-l)/2 _1)µ1 (1+
uq( r+ 1 )/2 _l t2 'for
T=
1 ( mod 2), and
g(u)
=
(1
+
uqrf2-1 _1)µ1 (u+
uqr/2+1 _1) mu2 '(5)
(6)
for
TO (mod2). Then we have the following result ( Ozbudak
[1 1] ) .
Lemma 9. Let T
>
1 be a positive integer, and Fqr a finite field of characteristic p>
2
with q7 elements. Then:(i) If T
=
1(mod 2L
the polynomial f ( u+
a) and f ( u+
/3) are co prime in Fqr [u] for any distinct a, /3 E Fqr ;(ii) If T
=
2 (mod
4 ) and T>
2
} the polynomials g ( u+
a) andg( u
+
/3) are coprime in Fqr[u]
for any distinct a, /3 E Fqr ; (iii) If T _0 (
mod
4),
there exists an additive subgroup A C F/;. ofcardinality q2 such that the polynomials g( u + a) and g( u
+
/3)are coprzme in Fqr [u] for any a, /3 E Fqr with the condition a
-
/3 (j. A.Now we calculate the number of Fqr-rational points on the corre
sponding fibre products. The following is a modification of Gluhov's
result [6] .
Lemma 10. Let T
>
1 be a positive integer, Fqr a finite field of characteristic p>
2
with q7 elements, and µ a positivt: divisor of q - 1 . Let µ1, µ2 positive integers such that µ1+
µ2=
µ} and a1 , . . . , a5 distinct elements of Fqr . Then the following holds:(i) If T
=
l( mod 2)
andGappa Codes
the number Nq,,. (Y) of Fq,,. -rational points on the curve Y defined over Fq,,. by
(ii) If T
2 (mod 4),
T>
2,
andg(u)
=
(1+
uqr/2-1 _1 )µ1(1
+
uqr/2+1 _1 )µ2J
the number Nq,,.(Z) of Fq,,.-rational points on the curve Z defined over Fq,,. by
zf
=
g(u+
ai),253
(iii) If T
O (mod 4),
elements a1 , . . . a8 liem
distinct cosets of F/;./A, and -u=
( 1
+ uq r/2-1 _1 ) µ1( 1
+ uq r/2+1 _1 ) µ2 g()
1+
uq-I 1+
uq-I'
the number Nq,,.(
Z)
of Fq,,. -rational points on the curveZ
defined over Fq,,. by
zf
=
g ( U+
CYi),is Nqr(Z)
=
µ8q7•Similar results for fibre products of superelliptic curves over finite
fields
Fq,,.hold in characteristic
p=
2 (Ozbudak [1 1]).
Lemma 11. Let T
>
1 be a positive integer, Fq,,. a finite field ofcharacteristic p
=
2 with q7 elements, and f(u), g(u) polynomials defined by(
5)
and(
6) .
Then:(i) If T
=
1 (mod 2),
the polynomials f(u+
a) and f(u+
/3) are254
Stepanov {ii) If T=
2 (mod 4)
and T >2,
the polynomials g(u+
a) and g( u+
/3) are coprime in Fq,. [u] for any a, /3 E Fq,. such that a -/3 (/. Fq2;
{iii) If T
=
0 ( mod 4),
the polynomials g( u+
a) and g( u+
/3) are coprime in Fq,. [u] for any a, /3 E Fq,. such that a -/3 (/. Fq2 .The number of Fqr-rational points on the corresponding fibre prod
ucts is determined as follows.
Lemma 12. Let T >
l
be a positive integer, Fq,. a finite field ofcharacteristic p
=
2 with q7 elements, and µ a positive divisor ofq -
1 .
Let µ1, µ2 be positive integers such that µ1+
µ2=
µ, anda1 , . . . , as distinct elements of Fq,. . Then the following holds:
{i) If T
=
l (mod 2),
elements a1 , . . . , a8 lie in distinct cosets ofFqr
I
Fq , andJ'(u)
=
(
1
+
1
u + uq(r-1)/2_1q-t ) µ1(
_l_+_u __ _ uq(r+l)/2 _1q-t ) µ2'
the number Nqr (Y') of Fqr -rational points on the curve Y' defined over Fq,. byzf
=
J'(
u+
a;), is Nqr (Y')=
µ8 q7;(ii) If T
=
2 ( mod 4),
T>
2,
elements a1 , . . . , as lie in distinct cosets of Fq,./
Fq2 , andthe number Nqr (Z') of Fqr -rational points on the curve Z' defined over Fq,. by
Gappa Codes
255
(iii) If T
O ( mod 4)
J elementsa
1,. . . ,
CY8 lie in distinct cosets ofFqr
I
Fq2 J and II U=
( l
+ u qr/2-1 _1 ) µ1( 1
+ uq r/2+1 _1 ) µ2g ( )
1
+
uq-l1
+
uq-l'
the number Nqr (Z") of Fqr-rational points on the curve Z"
defined over Fq,. by