Introducing an Analysis in Finite Fields

Hélio M. de Oliveira and Ricardo M. Campello de Souza

Universidade Federal de Pernambuco UFPE

C.P.7800 - 50.711-970 - Recife - PE, Brazil
E-mail: {hmo,ricardo}@ufpe.br

Abstract - Looking forward to introducing an i= 0 1 2 3 | 4 indexes i mod 4

analysis in Galois Fields, discrete functions are 1 2 4 3 repeat.
x i
considered (such as transcendental ones) and Notice that (x) 2 0 and (i) 2  0 .
MacLaurin series are derived by Lagrange's
Interpolation. A new derivative over finite fields is Composite functions can also be generated as usual.
defined which is based on the Hasse Derivative and For instance, the function
is referred to as negacyclic Hasse derivative. Finite i= 0 1 2 3 4
field Taylor series and -adic expansions over GF(p), 1 4 1 3 2
p prime, are then considered. Applications to 3 x 4 x
exponential and trigonometric functions are 2
corresponds to  (mod 5) = 2  (3x  4 ).
presented. Theses tools can be useful in areas such
as coding theory and digital signal processing. Therefore inverse functions over GF(p) can be defined
-1
as usual, i.e., a function a(x) has an inverse a (x) iff
-1
1. INTRODUCTION a(x)  a (x)x mod p.
There are p! inversive GF(p)-functions. The inverse of
One of the most powerful theories in mathematics is -1
a(x)3x+4 is a (x)2x+2 mod 5.
the Real Analysis [1]. This paper introduces new tools Since the characteristic of the field is different from 2,
for Finite Fields which are somewhat analogous to we can define the odd and even component of a
classical Analysis. Finite fields have long been used in function, respectively, as:
Electrical Engineering [2], yielding challenging and a (x )  a(x ) and a(x )  a ( x ) .
exciting practical applications including deep space o(x ): e (x ):
2 2
communication and compact discs (CD). Such a An even (respectively odd) function has o(x)  0
theory is the basis of areas like algebraic coding, and
(respectively e(x)0).
is related, to various degrees, to topics such as x
cryptography, spread-spectrum and signal analysis by The 2 exponential function defined over GF(5) has
finite field transforms [3,4]. Some new and x= 0 1 2 3 4 x= 0 1 2 3 4
unexpected applications of this framework are o(x)= 3 3 3 2 2. e(x)= 3 4 1 1 4.
currently under developing which include bandwidth
efficient code division multiplex [5]. Hereafter we iii) k-Trigonometric functions
adopt the symbol := to denote equal by definition, the A trigonometry in Finite Fields was recently
symbol  to denote a congruence mod p and K[x] a introduced [6]. Let G(q) be the set of Gaussian
polynomial ring over a field K. integers over GF(q),i.e. G(q):={a+jb, a,b  GF(q)},
r 2
where q=p , p being an odd prime, and j =-1 is a
2. FUNCTIONS OVER GALOIS FIELDS quadratic non-residue in GF(q). With the usual
complex number operations, we have then that GI:=
For the sake of simplicity, we consider initially Galois <G(q),,> is a field.
Fields GF(p), where p is an odd prime.
Definition 1.
Functions: Let be an element of GF(p) of order N. Let  have multiplicative order N in GF(q). The GI-
i) One of the simplest mapping is the "affine function" valued k-trigonometric functions cosk(.) and sink(.) are:
which corresponds to a straight line on a finite field, 1
e.g. a(x)=3x+4, x GF(5). cos k ( i ): ( i k   ik ) ,
2
x= 0 1 2 3 4 i 1
a(x)= 4 2 0 3 1. sin k ( ): ( i k   ik ) i,k=0,1,...N-1.
2j
There exists p(p-1) linear GF(p)-functions. The properties of the unit circle hold, i.e.,
ii) Exponential function: i , i=0,1,...,N-1. 2 2
sin k (i )  cos k (i )  1 mod p.
e.g. letting =2, the exponential function 2x
i We consider now k-trigonometric functions with =3
corresponds to 2 (mod 5). an element of order 6 in GF(7). A pictorial
i= 0 1 2 3 4 indexes i mod 5 representation of the cos-function on this finite field is
1 2 4 3 1. presented in the sequel. There exists several k-cos:
It is also interesting to consider "shortened" functions {cosk(i)}, k=0,1,..,5. In order to clarify the symmetries,
in which indexes are reduced modulo p-k for 0<k<p. half of the elements of GF(p) are considered positive
and the other half assumes negative values. Therefore, each carrier (oscillatory behavior). We define  on
we represent the Galois field elements as the set GF(p) as
p ():=(p-1)/2. Clearly, cos1(i+())=-cos 1i.
The cosk(i) over GF(7) assumes the following values
(k,i=0,...,5): We shall see that any function from GF(p) to GF(p)
cos0(i)=1 1 1 1 1 1 (mod 7) defined by N | p-1 points can be written as a
polynomial of degree N-1 (i.e., N indeterminates), e.g.,
cos1(i)=1 4 3 6 3 4  1 -3 3 -1 3 -3 (mod 7) i 3
2  i 1 (mod 5) (indexes i mod 4).
cos2(i)=1 3 3 1 3 3 1 3 3 1 3 3 (mod 7) Generally there are N unknown coefficients, f(i)
2 N-1
 a0+a1i+a2i +...+aN-1i mod p, and such a
cos3(i)=1 6 1 6 1 6 1 -1 1 -1 1 -1 (mod 7)
decomposition corresponds to the (finite) MacLaurin
cos4(i)=1 3 3 1 3 3 cos2(i) series with the advantage that there are no errors.
cos5(i)=1 4 3 6 3 4 cos1(i)
3. DETERMINING THE SERIES
Cos 0 (i)
Given a function f(i), the only polynomial "passing
through" all pair of points {i, f(i)} can be found by
1
1 solving a Linear System over GF(p) [7]. For instance,
1
the 1-cos function over GF(5) leads to:
1 1
1 1 0 0 0   a 0  1 
1  
1 1 1 a1  0 (mod 5).
3
   
1 2 4 3 a 2  4
Cos1 (i)     
1 3 4 2 a 3  0
3

Another way to find out the coefficients of the

1 polynomial expansion is by using Lagrange's
interpolation formula [8],
(i 1)(i 2)(i  3)
-3 L3 (i)  1 
(0  1)(0  2)(0  3)
-1
(i  0 )(i  1)(i  3) (mod 5).
4
-3 (2  0 )(2  1)(2  3)
Cos 2(i) Therefore
cos1(i)  1  2i  i 2  2i 3 (mod 5).
Identically, the 1-sin function [6] can be expanded:
3 3
i= 0 1 2 3
sin1(i) 0 j3 0 j2,
1
1
3 2
sin 1(i)  j(i  i  2i) (mod 5).
3 3
Euler's formula over a Galois field proved in [6] can
also be verified in terms of series:
cos(i)+jsin(i)
(1  2i  i 2  2i3 ) (i3  i2  2i) i +12 mod 5
3 i
Cos (i)
3
(i mod 4).
1 Those results hold despite 1  2 mod 5!
The unicity of the series decomposition can be
1 established by:
-1
Proposition 1. Given a function f defined by its values
1
-1 f(x), xGF(p), there exist only one Maclaurin series
for f.
2 p-1
-1 Proof. Letting f(x)  a0+a1x+a2x +...+ap-1x (mod p)
Figure 1. Pictorial representation of cosk(i) in GF(7). then f(0)a0 and we have the following linear system

The "dough-nut ring" representation is well suitable

for handling with the cyclic structure of finite fields.
The envelope does not exists: It just gives some
insight on the concept of different "frequencies" on
 1 1 1  1  Conventional Hasse derivative considers polynomial
 2 2 2
2 3
 2 p 1  p d1
  modulo x  1 so that  0 . On the other hand,
2 3 p 1
 3 3 3  3  dx
 
       the fact that 
i 
  0 i<r does not allow to handle
( p  1) ( p  1) 2 ( p  1) 3  ( p  1) p 1  . r 
with negative powers of x.
 a1   f( 1 )  f( 0 ) 
   
 a2   f( 2 )  f( 0 )  In contrast with Hasse derivative which always
 a    f( 3 )  f( 0 )  (mod p) decreases the degree of the polynomial, we introduce a
 3    new concept of negacyclic Hasse derivative
     
 a   f(p  1 )  f( 0 )  considering the polynomial ring GF(p)[x] reduced
p-1
 p 1    modulo x +1 [11]. Thus, the degree of a(x) is deg
or V.a = f. Since V is a Vandermonde matrix, its a(x)=p-2. The derivative of a constant no longer
determinant is: vanishes and the degree of the polynomial function is
det[V ]  (x j  x i ) (mod p) where xi:=i.
 preserved! Over GF(7), for instance, we deal with
ij polynomials of degree 5 and assume
Therefore, det[V ] 0
 mod p concluding the proof. d 1 d ( x 6 ) 5
  x mod 7 (negacyclic)
dx dx
whereas
4. DERIVATIVES ON FINITE FIELDS d 1 d (x 7 )
  0 mod 7 (Hasse).
dx dx
Over a finite field, we are concerned essentially with
derivative of polynomials since other functions can be 5. -ADIC EXPANSIONS AND FINITE FIELD
th
expanded in a series. We denote by a(i)(x) the i- TAYLOR SERIES
derivative of the polynomial a(x). The concept of
classical derivative however presents a serious The classical Taylor series of a function a(x) is given
drawback, namely the derivatives of order greater or by
equal to the characteristic of the field vanish. Let a(1) ( x0 ) a( 2) ( x0 ) 2
N 1 a (x)  a (x0 )  ( x  x0 )  ( x  x 0 ) . . .
1! 2!
a(x)   ai x i where (i) ai  GF(p), x is a dummy (N 1 ) (x  x o )N 1
i0 a (x 0 )  Rn
(p) (i) (N  1)!
variable. Then the derivatives a (x)   and a (x)  0 N
where R  a (N ) ( ) (x  x o ) ( x 0    x )
(i  p) no matter the coefficients. A more powerful n
N!
concept of derivative over a finite field was introduced (Lagrange's remainder).
a long time ago (see [9]):
On a finite field, Taylor series developed around an
Definition 2. The Hasse derivative of a polynomial arbitrary point  GF(p) can be considered according
N 1
to
function a(x)   ai x i is defined by [1 ]
a ( x )  a ( )  a ( )(x   ) 
i0
N 1 i
i  a [2] ( )(x   )2 ...a [p 1] ( )(x   )p 1 .
[ r] 
a ( x):    ai x i  r with    0 for i<r. p
It is interesting that working with x  1  0 i.e.,
i 0 r r 
deg a(x)=p-1, we have
Clearly the classical Newton-Leibnitz derivative
yields a (p ) ( ) p
Rp  (x   )  0 (Section 4) and the
N 1 p!
a( r) ( x)   i(i  1). . . (i  r  1)ai xi  r . It follows series is finite (no rounded error, although is
i 0 unknown).
1 ( r) [ r] Now let the -adic expansion of a(x) be
that a (x)  a ( x) . 2 N 1
r! a (x )  b 0  b 1 (x   )  b 2 (x   ) ...b N 1 (x   )
The Hasse derivative has been successfully used on with b0=a(). It follows from Definition 2 that
areas where finite fields play a major role such as N 1 r 
i  [r ]
coding theory [10]. a [r ] (x ):    b i x i r so that a ( )   b r
r  r 
i r
In the following, we introduce a new concept of The unicity of the Taylor Finite Field series follows
derivative in a finite field. In order to exploit its cyclic directly from [12]:
structure, a polynomial ring GF(p)[x] is required. Let
1 be the constant function f(x)=1 xGF(p). Unicity. Let K be a field and K[x] be a ring of
polynomials over such a field. The adic expansion of
a polynomial a(x) over a ring K[x] is unique .
Any function a(x) over GF(p) of degree N-1 can be In order to introduce trigonometric functions, it is
written, for any element   GF(p), as possible to consider the complex j  1 since -1 is
N 1
[ i] i without error or a quadratic non-residue for p3 (mod 4) and pick ej i
a(x)   a (  )( x   ) so that new "e-trigonometric" sin and cos functions
i 0
are:
approximations. For instance, the 2x GF(3)-function 2 4
cos(i)=eexp(j.i)  1  4 i 5i (mod 7),
can be developed as 2 x  1  2 x  2 x 2 mod 3 3 5
(McLaurin series). sin(i)=mexp(j.i)  i  6 i  i (mod 7) .
Therefore,
(2 x )[1 ]  2  x ; (2 x )[2 ]  1  2 mod 3. It is straightforward to verify that cos(i) is even and
2! sin(i) is an odd function. The negacyclic Hasse
Letting 2 x  b 0  b 1 ( x   )  b 2 (x   )2 mod 3 derivative yields:
[1 ]
x
then 2  (b 0   b 1   2b 2 )  (b 1  2  b 2 ) x  b 2 x 2 sin i  cos(i) and cos[1 ] i  sin(i) .

with b =2 mod 3.
0 We have two kinds of trigonometric functions over a
The -adic expansions mod p=3 are: finite field: the k-trigonometric ones and these new e-
x 2 trigonometric functions. Unfortunately, cos(i) and
2 1 2 x  2x 0-adic
sin(i) do not lie on the unit circle, i.e.,
2 x  2  2(x  1)2 1-adic 2 2
x
2  1  (x  2 )  2 ( x  2 ) 2 2-adic. sin i  cos i 1  (mod p). The "number e" is not a
"natural" element of the field (in contrast with the real
Indeed,  case). However, the k-trigonometric case is defined
N 1
a(x )  a( ) using an element of the field and presents a lot of
 a [1 ] ( )  a [i ] ( )(x   )i 1 , so one

x  properties. Nevertheless, derivatives formulae of k-
i 2
can, by an abuse, interpret the Hasse derivative as a trigonometric functions are not related as usual
"solution" of a "0/0-indetermination" over GF(p): trigonometric derivatives.
? a(x )  a ( ) These results can be generalized so as to define
a [1] ( ) . hyperbolic functions over finite fields, p odd,
x   according to the series developments:
i GF(p)
6. APPLICATIONS: EXPONENTIAL AND i2 i 4 i 6 i p  3 and
TRIGONOMETRIC FUNCTIONS cosh (i): 1    . .. 
2! 4 ! 6 ! ( p  3 )!
Let us consider classical (real) series: i3 i5 i7 i p 2 .
sinh(i ):  i    ...
x2 x3 3! 5! 7 ! (p  2)!
ex 1 x   ... Indeed sinh(i)+cosh(i) exp(i).
2! 3!
x 2 x 4 x 6 +... 
cos(x )  1    We recall the convention  =0 ()GF(p) [7]. The
2! 4 ! 6! idea is to consider an extended Galois Field by
x 3 x 5 x 7 +... appending a symbol -  as done for the Real number
sin(x )  x   
3! 5! 7! set ([1], def.1.39). (x)GF(p):
x+(- )- x.(- )-and x/(-  )0.
As an example, we consider the ring GF(p)[x] with
p-1
x +10 so negacyclic Hasse derivatives can be used. Therefore, other functions such as an hyperbole 1/x
Since deg a(x)=p-2 we truncate the series yielding can also be defined.
x 2 3 4 5 i= 0 1 2 3 4
(e )7  1  x  4 x  6 x  5 x  x (mod 7),
2 4 1/i = - 1 3 2 4.
(cos x )7  1  4 x  5 x (mod 7),
3 5
(sin x )7  x  6 x  x (mod 7). Finally, it is interesting to consider "log" functions
The finite field Euler constant e and the exponential over a Galois field as the inverse of exponential. We
function can be interpreted as follows: have then
x [1 ] x
(e )p  (e )p , i= 0 1 2 3 4
log2 i = - 0 1 3 2.
x [r ] x
so that (e )p  (e ) p , r1.
1 1 1 1 log2(0) and log2(1) give the expected values.
e 1    ... (mod p). MacLaurin series cannot be derived as usual.
1! 2! 3! (p  2)! x
Moreover, the (e ) function does not have an inverse
Therefore e i  1  i  4 i 2  6 i 3  5i 4  i 5 mod 7.
so (lnx) is not defined, although logx exists, for any
i= 0 1 2 3 4 5
i primitive element .
e = 1 4 4 6 6 1.
7. INTEGRAL OVER A FINITE FIELD p 1 p 1
p 1 1
0 f .g ( )d    f (k ) g (k ) 
p 1
 a ib p 1 i
th k 0 i 0
We evaluate the sum of the n power of all the p
proof. Substituting the series of f and g in the sum and
elements of GF(p). It is also assumed that 00:1, changing the order of sums, we have
since that f(0)=a0. p 1 p 1 p 1

0  1  2  3...(p  1)  ?
 f (k )g (k )   ai b j  k i j .
k 0 i ,j 0 k 0
2 2 2 2 The proof follows from Lemma 1 and the fact that the
0  1  2  3 ...(p  1)  ?
-1
... inverse (p-1)  p-1 mod p.

0  1 p 1  2 p 1  3 p 1 ...(p 1 )p 1  ? In order to evaluate the integral over another "interval

Lemma 1. (nth power summation). of GF(p)", we consider:
p 1 N p 1
N
p  1 if n = p -1
x n
  . 0 f ( )d  :   f ( )  ai S N (i ) ,
x 0
0 otherwise .  0 i 0
N
proof. Letting S n : 1 n  2 n  3 n ...(p  1)n where S N (i): x i .
n=0,1,2,3...,(p-1). Trivial case n=0 apart, x 0
n  1  n
  1  n  1  A table with the values of SN(i) over GF(5) is showed
 S   S ...  S  p n 1  p  0
 1  1  2  2  n  n below.
so that np-1 implies S1S2...Sn0 (mod p).
i=0 i=1 i=2 i=3 i=4
S0(i) 1 0 0 0 0
Definition 3: The definite integral over GF(p) of a
function f from 0 to (p-1) is S1(i) 2 1 1 1 1
p 1 p 1 S2(i) 3 3 0 4 2
I  0 f ( )d  : f ( ). S3(i) 4 1 4 1 3
 0 S4(i) 0 0 0 0 4
Proposition 2. The integral of a function f presenting a
p 1
Another interesting result establishes a link between
series a(x )   ai x i is given by
Hasse derivatives and the Finite Field Fourier
i 0
p 1 Transform [13].
The Galois field function obtained by an image
 f (x )  (p  1)a p 1 mod p. element permutation of f is here referred to as an f-
x 0 permutation.
proof. Substituting the series of f in the integral
definition and applying Lemma 1, Proposition 4. The coefficients of the McLaurin series
p 1 p 1 p 1 p 1
expansion of a given signal f are exactly the inverse
I    ai  i   ai  x i 
finite field Fourier transform (FFFT) of an f-
 0 i 0 i 0 x 0 permutation.
p 1
  a i (p  1) i ,p 1 Proof. Given the 0-adic development a(x) of a discrete
i 0 function f, we consider then the finite field Fourier
where i,j is the Kronecker delta (1 if i=j, 0 otherwise). transform pair: a A where the FFFT is from GF(p)
p 1
to GF(p), and Ai=a(i)=
 a j  ij . Since a(x) is used
j 0
Corollary. If a function f defined GF(p) admits an
th to interpolate the discrete function f(.), for any given i
inverse, then the (p-1) coefficient of its series j i
there exist j{-  ,0,1,...,p-2} such that f( )=a( )=Ai.
vanishes.

The condition ap-1  0 mod p is a necessary condition

but not a sufficient one to guarantee that f is inversible.
In fact the integral vanishes since all the images are
distinct (the sum of all the elements of the field).
Proposition 3. Let f and g be functions defined for
every element of GF(p). Denoting by {ai} and {bi} the
coefficients of their respective MacLaurin series, then
8. CONCLUSIONS
The main point of this paper is to introduce the
background of new tools useful for Engineering
applications involving finite fields. A number of
discrete functions is analyzed including a brief look at
transcendental functions over Galois Fields. Finite
field (FF) derivatives are considered, which in turn
leads to Finite Field Taylor Series. A brief [13]- "Transform Techniques for Error-Control
introduction to FF integration is also presented. Codes", IBM J. Res. Develop., 23, n.3, pp. 299-314,
Clearly many other extensions and applications do May, 1979.
exist.

Acknowledgments- The authors wish to thank Prof.

V.C. da Rocha Jr. (Universidade Federal de
Pernambuco) for introducing us Hasse derivatives.

References

[1]- W. Rudin,"Principles of Mathematical Analysis",

McGraw-Hill, 1964.

[2]- R.J. McEliece, "Finite Fields for Computer

Scientists and Engineers", Kluwer Ac. Pub., 1987.

[3]- R.M. Campello de Souza, A.M.P. Léo and F.M.R.

Alencar, "Galois Transform Encryption", 2nd Int.
Symp. on Communication Theory and Applications,
Ambleside, England, 1993.

[4]- J.L. Massey, "The Discrete Fourier Transform in

Coding and Cryptography", Info. Theory Workshop,
ITW, 1998, CA: San Diego, Feb. 9-11.

[5]- H.M. de Oliveira, R.M. Campello de Souza and

A.N. Kauffman, "Efficient multiplex for band-limited
channels: Galois-Field Division Multiple Access",
WCC'99, Proc. of Coding and Cryptography, pp.235-
241, Paris, France, 1999.

[6]- R.M. Campello de Souza, H.M. de Oliveira, A.N.

Kauffman and A.J.A. Paschoal, "Trigonometry in
Finite Fields and a new Hartley Transform", IEEE
International Symposium on Information Theory, ISIT,
MIT Cambridge,MA, THB4: Finite Fields and Appl.,
p.293,1998.

[7]- S. Lin and D.J. Costello Jr., "Error Control

Coding: Fundamentals and Applications", NJ: Prentice
-Hall, 1983.

[8]- R.W. Hamming, "Numerical Methods for

Scientists and Engineers", McGraw-Hill, 1962.

[9]- J.L. Massey, N. von Seemann and P.A.

Schoeller,"Hasse Derivatives and Repeated-root
Cyclic Codes", IEEE International Symposium on
Information Theory, ISIT, Ann Arbor, USA, 1986.

[10]- V.C. da Rocha, Jr., "Digital Sequences and the

Hasse Derivative", Third Int. Symp. On
Communication Theory and Applications, Lake
Distric., 95, pp.299-306.

[11]- W.W. Peterson and E.J. Weldon Jr., "Error

Correcting Codes", Cambridge, Mass.: MIT press,
1972.

[12]- S. Lang, Linear Algebra, Addison-Wesley Pub.

Co., 1966.