Smooth Polynomials Over Finite Fields
Smooth Polynomials Over Finite Fields
Smooth Polynomials Over Finite Fields
1 Introduction
The security of many applications in public-key cryptography relies on the com-
putational intractability of finding discrete logarithms in finite fields. Examples
are the Diffie-Hellman key exchange scheme [7], E1 Gamal's cryptosystem [8], and
pseudorandom bit generators [3,10]. On the other hand, algorithms for comput-
ing discrete logarithms in finite fields depend on finding polynomials with all
of their irreducible factors with degree not greater than certain bound m - -
such polynomials that are the analogue of highly composite numbers are called
smooth polynomials. Thus quantitative characterizations of smoothness in ran-
dom polynomials over finite field are of relevance to cryptographic attacks; see
[14-16].
In different contexts, like computer algebra and error-correcting codes, knowl-
edge of the distribution of the largest irreducible factor of a random polynomial
over a finite field permits us a fine tuning of the stopping conditions in polyno-
mial factorization algorithms.
In this paper, we give a unified treatment of the asymptotic enumeration
of smooth polynomials over finite fields and quantify precisely the distribution
of largest irreducible factors. The results are expressed in terms of a familiar
number-theoretic function, the Dickman function, that is already known to un-
derlie the study of numbers with no primes larger than m; see [5, 6]. Our ap-
proach starts with an exact representation of enumeration problems by means
Smooth Polynomials over Finite Fields 227
2 Smooth polynomials
The Dickman function plays a central r61e in our results on smooth polynomials.
This classical number-theoretic function describes the distribution of the largest
prime divisor of a random integer [5, 6]. A survey on this topic is due to Hilde-
brand and Tenenbanm [12]. Our general reference for this paper is Tenenbaum's
book [18].
228 Panario, Gourdon, and Flajolet
E(a) = fa ~176
-e -- s ds.
8
where the big-Oh error term is uniform with respect to h, for ~(h) > 0 and
I (h)l < - .
PROOF. When ~(h) > 0, we have
1 1 (r 1 )__lr
m(e'/m - 1) = m s-~ m s"
Finally, the analyticity of r in Izl < 2~r implies that Rm(u) = O(1/m) uni-
formly for ~(u) > 0 and I~(u)l _< m~r. []
Theorem 1. The number of m-smooth polynomials of degree n over Fq satisfies
n n
PROOF. Let Z be the collection of all monic irreducible polynomials in Fq, and
[w] the degree of w E iT. The collection of monic polynomials with all irreducible
factors with degree smaller than or equal to m can be symbolically written as
Sin= H (1 + w + w 2 +...)= H (1-w)-l"
~ez, I~l<m ~ez, I~l<m
Let z be a formal variable. The substitution w ~-~ z I~l gives rise to the generating
function Sin(z) of m-smooth polynomials
fi( 1 ) I1' 1
P(z) = ~ = 1 - qz"
k=l
The number of m-smooth polynomial of degree n over lFq is given by Cauchy's
coefficient formula
where the contour C is chosen to be z = e -1/n+w, -~r < 0 < lr. The change of
variable z = e -h/" within the integral provides z n = e -1+in~ Thus, h = 1 -inO,
and the limits of integration are (1 + ni~r, 1 - nit). Therefore,
1 f 1-n'" (1) dh
Nq(n,m) = ~ri Jl+ni,~ Sm(e-h/n) - e~-h" (2)
An equivalent expression for Sin(z) that makes explicit the singularity at z = 1/q
can be obtained by taldng the logarithm and inverting summations. Indeed,
considering r ~ ] (z) = ~t~>rn IkzkJ, we have
k>m
)
= exp - ~ _ I~ z~ + - ~ - + - 5 - + . . .
k>m
= +
1
for Izl < -,
k>rn
q
and,
[z[~_l/q
o( qmG--1)
1 ) for j > 2.
where we may disregard the error term in the exponent since it is of smaller
order than the one in the statement of the theorem.
Substituting this estimate in (2) yields, for # - W, __ m
Set r - 1--e1- z 1 ' that is an analytic function in [z I < 27r. We can express
Z
the above number in terms of r as follows. First,
1 1 (r =1r 1
n(1 - e -h/n) - n n -h"
Second,
n ( 1 - - e - h / n )-- h § +0 .
Thus,
We treat separatedly the three integrals. The fact that e -E(z) is bounded in
the domain ~(z) _> 0 (see [1], w 5.1) entails that the contribution of the big-Oh
term in the integral is O(logn/m). Then, an integration by parts gives also a
small contribution of order O(log n/n) for the term containing r Finally,
we have
Nq(n,m) = qn 1 / -h eh dh + O
J 1 --in~r
We write
1 f l+in~e--E(l~h) e h dh = 1 ]rl+iOOe_E(l~h)
-~ e h d h - ]f -e--E(~h)
- e h dh,
27ri J l - - i n ~ h ~i Jl-ioo jE h
Smooth Polynomials over Finite Fields 231
where the integration domain s is the union of the two semi-vertical lines defined
by N(h) = 1, [~(h)[ > nrr. The last integral is O ( 1 / n ) as can be checked by
partial integration. Therefore,
( el-Pi~176 (l~))
g (n'm) = qn 1 ] -~ ehdh+O . (4)
To conclude the proof, it remains to show that the above integral is p ( n / m ) . The
Laplace transform ~(s) of the Dickman function satisfies (see [18], w p. 373)
s ~(s) = e -E(s). Thus,
e ~ dv. (5)
We now relate Equations (4) and (5). The change of variable # h = v in (4)
implies
= 1 re+i~176
(~) evn/mdv :
2?ri j 1-ioo
The theorem follows since p(u) < 1 / F ( u + l ) for all u > 0 ([18], w p. 366). []
The previous theorem shows that when m / l o g n -~ oc, the number of smooth
polynomials is given asymptotically by the Dickman function. In the sequel, we
extend the range of applicability of Theorem 1 to sublogarithmic values of m
with respect to n.
Note that we can restrict our attention to m < n since the case m = n
corresponds to the well-known enumeration of irreducible polynomials.
T h e o r e m 2. Let m < n, and k a positive integer such that k m < n and
mk /logn -+ oo. Then, the number of m-smooth polynomials of degree n over
]Fq satisfies
Nq(n,m)=qnp(n) (l+o(l-~)),
--_ 1 _r e -s + r e -S ds
rn m h
Thus,
R 2 (ph) = l--~e-2~h~/,2 ( h ) + o (e-2~h / m 3)
m2 ~
Expanding e -R-(~h) in (6), we have
e -m~(~'h) 1 lr l e -~h
n(1 - e-h~ ") = -h + -n hm r nm
Arguments similar to the ones employed in the previous theorem lead to the
conclusion that
(In order to improve on the error estimate, it would suffice to consider successive
terms in the expansion of e--Rm(~h).) I"1
The distribution of the largest degree among the irreducible factors of a random
polynomial over ]Fq underlies many problems dealing with polynomials over finite
fields. An instance is in the factorization problem. The joint distribution of the
two largest degrees D~ ], Din2] of the distinct factors of a random polynomial of
degree n in lFq provides the halting condition for the distinct-degree factorization
stage; see [9].
We first investigate the distribution of the largest degree D~ ] which is of
independent interest. The same analysis techniques are then applied in order to
produce the joint distribution of D~ ], D~ ].
The following theorem gives a local distribution for the largest degree D~ ] of a
random polynomial of degree n. We only sketch the proof since it is similar to
that of Theorem 1.
k=l
Thus, the generating function of polynomials for which D~ ] -- m is
qn -- 27ri Lm zn+l'
where the contour C is chosen to be z = e -1/n+ia, -~r </9 < 7r. As in Theorem 1,
the change of variable z = e -h/n within the integral gives
= 1 f l+n'~ Lm ( - ~ ) ehdh.
Pr(D~] m)= ~i al_nilr n
Lm = 1 - e -h m (9)
Pr(D~ ] = m ) - 1 1 fl+ni~re-E(#h)+O(i/m)e(l_#)hdh"
m 27ri Jl-ni~ n(1 -- e-h~ n)
_ ~g~ + O ,
1 fl+i~ e-E(.h/2)
gl (#) = ~ / a 1-io~ h e (1-t~)h dh;
Pr(D~ ] = m l , D ~ ] = m2) -- - -
1 (ml ~2) f logn ~
~lm2 g: ~, +okmlml],
where g2(#1,#2) is
PROOF. We only sketch the proof. With the same notations as in the proof of
the previous theorem, the generating function of polynomials for which D~ ] -- m
and DIn2] _< m / 2 is
Irazm (10)
L m ( z ) = S L m / 2 J (z) 1 - z m"
The generating function of polynomials with D[n] = ml and D[2n] = m2, m2 <
ml is
Lml,~2 (z) = Lm2 (z) _rm_~zff~' (11)
1 - z ml "
The behavior of the n t h coefficient of the generating functions in (10) and (11)
is then extracted like in Theorem 3. We briefly demonstrate the process for the
generating function of (11).
The estimate in (9) for Lm2 (z) and (11) entails
(e_~q
h) e-~(~2h)+~ e-m,he-~,h
Lml,m2 ~ 1 -- e - h m2 ml
Smooth Polynomials over Finite Fields 235
1 fl+ni~re-E(~2h)+O(1/m2)
m l m 2 P r ( n ~ ] = ml,n[2n ] = m2) = ~ Jl-,u~ - ~ = e_--_-~/n) e ( 1 - " ' - ~ 2 ) h d h .
An argument once more similar to the one in Theorem 1 completes the proof. []
We note t h a t it is possible to generalize the above theorem to the joint
distribution of the j t h largest distinct irreducible factors.
Acknowledgements. This work was supported in part by the Long Term Research
Project Alcom-IT (# 20244) of the European Union.
References