J
OURNAL DE
T
HÉORIE DES
N
OMBRES DE
B
ORDEAUX
J
ÖRG
M. T
HUSWALDNER
The complex sum of digits function and primes
Journal de Théorie des Nombres de Bordeaux, tome
12, n
o1 (2000),
p. 133-146
<http://www.numdam.org/item?id=JTNB_2000__12_1_133_0>
© Université Bordeaux 1, 2000, tous droits réservés.
L’accès aux archives de la revue « Journal de Théorie des Nombres de Bordeaux » (http://jtnb.cedram.org/) implique l’accord avec les condi-tions générales d’utilisation (http://www.numdam.org/conditions). Toute uti-lisation commerciale ou impression systématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit conte-nir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
The
Complex
Sum of
Digits
Function
and Primes
par
JÖRG
M. THUSWALDNERRÉSUMÉ. La notion de
développement q-adique
d’un entier, pourune base q
donnée,
segénéralise
dans l’anneau des entiers de GaussZ [i]
audéveloppement
d’un entier de Gauss suivant unecertaine base b ~
Z [i],
cedéveloppement
étant unique. Danscet
article,
on s’intéresse à la fonction 03BDb,désignant
la sommede chiffres dans le
développement
suivant la base b. On montreun résultat sur la fonction somme de chiffres pour les nombres
non
multiples
d’une puissancef-ième
d’un nombre premier. Onétablit aussi pour 03BDb un théorème du type Erdös-Kac. Dans ces
résultats, l’équidistribution
de 03BDbjoue
un rôle essentiel. Partantde
cela,
les démonstrations font alorsappel
à des méthodes decrible,
ainsiqu’à
une version du modèle de Kubilius.ABSTRACT. Canonical number systems in the
ring
of Gaussianintegers
Z[i]
are the naturalgeneralization
ofordinary q-adic
num-ber systems to
Z [i].
It turns out, that each Gaussianinteger
has aunique representation with respect to the powers of a certain base
number b. In this paper we
investigate
the sum ofdigits
function03BDb of such number systems. First we prove a theorem on the sum
of
digits
ofnumbers,
that are not divisibleby the
f-th
power of aprime.
Furthermore,
we establish an Erdös-Kac type theorem for03BDb . In all
proofs
theequidistribution of 03BDb
in residue classesplays
a crucial rôle.
Starting
from this fact we use sieve methods and aversion of the model of Kubilius to prove our results.
1. INTRODUCTION
Let
vq(n)
denote the sum ofdigits
of theq-adic
representation
of apos-itive
integer
n. Gelfond[5]
proved,
thatvq(n)
isequidistributed
in residueclasses modulo an
integer.
Inparticular,
he established thefollowing
result.For r, m E Z with
one has
The
exponent
~1
1 in the error term can becomputed explicitly
anddoes not
depend
on a, s, r and N. This result forms the basis for theapplication
of sievemethods,
in order toget
results on the sum ofdigits
of
prime
numbers. The first result of thistype,
also contained in Gelfond’spaper
[5],
is thefollowing.
LetSr,m(N) = In
N r(~n)}.
Thenthe number of elements of
8r,m(N),
being
not divisibleby
thef-th
power of aprime
isgiven by
with a constant
A2
1.Here (
denotes the Riemann zeta function.Re-cently,
Mauduit andS6xk6zy
[15]
proved
an Erd6s-Kactype
theorem forthe elements of the set
8r,m(N):
Let m, r E Z such that(1)
holds. Denote the normal lawby
~(a?)
and letw(n)
count the distinctprime
divisors ofn. Then
The aim of this paper is to extend these results to the sum of
digits
function of canonical number
systems
in thering
of Gaussianintegers
First wegive
a definition of these numbersystems.
Let b Eand fii
={O, 1,... ,
N(b) -
11,
whereN(b)
denotes the norm of b overQ.
If anynumber 1
E admits arepresentation
of the formfor c; E
{0,1, ... ,
N(b) -
11 (0 j
h)
and 0 forh ~
0,
then(b, N)
is called a canonical number
system
in7G~i~.
b is called the base of thisnumber
system.
InKitai,
Szab6[11]
it is shown that b can serve as a baseof a canonical number
system
in if andonly
ifThe sum of
digits
function of the numbersystem
(b,
N)
is definedby
Some
properties
of this function have beeninvestigated
in recent papers.An
asymptotic
formula of thesummatory
function of inlarge
circlesGittenberger
and Thuswaldner[6]
establishedasymptotic
formulas for themoments of
The notion of canonical number
systems
can be extended toarbitrary
number
fields, provided
thatthey
have a powerintegral
basis(cf.
Kovics[12]).
The bases of these numbersystems
are characterized inKovics,
Peth6
[13].
This characterization is notexplicit
anddepends
on theshape
of the
integral
basis of the number field. Some results on the sum ofdigits
function can be extended to the
general
case of canonical numbersystems
in number fields. The results
(3)
and(4)
can begeneralized
to numberfields,
whosering
ofintegers
is aunique
factorization domain. For furthergeneralizations
a new notion of the sum ofdigits
function with ideals asarguments
seems to be necessary. Since we also need the finiteness ofthe group of
unity
of thering
ofintegers
underconsideration,
we confineourselves to the Gaussian case.
Again
our results are derived from a result of thetype
(2).
The numberfield version of
(2)
is established in Thuswaldner[18].
Weonly
need the "Gaussian" case of this result: Let b = -~c ~ i be the base of a canonicalnumber
system
in with minimalpolynomial
We define the setsThen,
for m, r EZ,
a EZ [I]
and an ideal s ofZ [i],
satisfying
we have the estimate
A 1 is an
effectively computable
constantindependent
of r, a, s, and N. In Section 2 we willprovide
the necessary tools needed in theproofs
of our
theorems,
Section 3 is devoted to thegeneralization
of(3)
and inSection 4 the Erd6s-Kac
type
result(4)
is established for canonical numbersystems
inZ[I] .
2. AUXILIARY RESULTS
In this section we want to list various results from
algebraic
numbertheory,
that will be needed further on. First wegive
some estimates forsums over
prime
ideals of(in
fact,
these estimates remain valid forrings
of
integers
ofarbitrary
numberfields).
In Narkiewicz[17,
Lemmas7.3-7.6]
The sum is extended over all
prime ideals p
whose norm is less than x.Recall Abel’s
identity
(cf.
Hardy, Wright
[8,
Theorem421])
for
A(t) =
an and acontinuously
differentiable functionf (t).
Apply-ing
this to(7)
yields
the estimateFurthermore,
we have(cf.
Narkiewicz[17,
Chapter
7,
Exercise7])
We will also need the estimate
which can be
proved
in a similar way as Theorem 429 in[8].
Next we recall a result due to Hua
[9]
onexponential
sums over numberfields. Let
tr(z)
denote the trace of zover Q
and let D =2Z[i]
be thedifferent of
Q(i).
Then, writing
e(x)
=exp(27rix),
-where
runs over acomplete
residuesystem
in moduloc-1.
In order to prove our results we will need a version of
Selberg’s
sievemethod and a
generalized
Kubilius model in number fields. Theseobjects
are discussed in Kubilius[14,
Chapter
X]
(cf.
also Danilov[2]
for relatedresults).
In a moregeneral setting they
are studied inJuskys
[10]
andZhang
[20].
Moreover,
Thuswaldner[19]
gives
a survey onSelberg’s
sieveand the Kubilius model in number fields. We start with the statement of a
quantitative
version ofSelberg’s
sieve in(cf.
[19]).
Lemma 2.1. Let an E
(1 n N)
and p > 0. Let ~~, ... , p, beprime
idealsof
with ... p and set 11 = pi -" p,.is an ideal that divides il
(~~~
then we assume that there exists arepresentation
where X and R are real
numbers,
X >0,
and =for
divisors
of fl
with(Dl,
D2)
= 0(o
denotes the unitideal).
Furthermorefor
each
prime
ideal p we assume that 017(P)
1,
holds.Set
Then the estimate
holds
uniformly for
p >2,
w(c~)
is the numberof
distinctprime
idealfactors of -0,
andNow we sketch the construction of the
generalized
Kubilius model forZ[I].
Again
we refer to[19]
for details. Let the same notation as inLemma 2.1 be in force. Set
and,
for tlû,
. I r
Let A be the set
algebra generated by
the setsEt.
In order to make A to aprobability
space, we define the measureIf all conditions for the
application
of Lemma 2.1 arefulfilled,
it can bewith H as in Lemma
2.1,
1 Oll [
1 and82
1. Here[t, cJ
denotes the leastcommon
multiple
of t and D.Now we set
for each
Unfortunately, pEt
:= ~~ ingeneral
does not define aproba-bility
measure on,A,
because there may existempty
setsEt
withpositive
pEt.
Therefore we have to use a trick due toZhang
[20]
in order to construct a sequence ofindependent
random variables such that the distribution oftheir sum has
density
ut.Remark 2.1. The fact that
pEt
ingeneral
does not define a measure on,A was observed
by Zhang
[20].
Unfortunately,
many versions of Kubilius modelsproposed
sofar,
treatpEt
as a measure(cf.
for instance[14, 3, 19]).
This lacuna can be mended in many cases with
help
of a trick used inZhang
[20]
(cf.
also theproof
of Lemma 2.3 of thepresent
paper).
Wewant to mention
explicitely,
that the Erd6s-Kactype
theorems inMauduit-S6xk6zy
[15, 16]
are not affectedby
this lacuna. Theirproofs
can beadapted
by reasoning
in the same way as in thepresent
paper.Lemma 2.2.
Having
the sameassumptions
as in Lemma~.1,
suppose, thatR(N, t)
=0 (NA’)
and T = such thatA’+
7J 1. ThenEt :A 0
for
and N
large enough.
Proof. Using
(10)
weeasily
obtainOn the other
hand, by
theassumptions
in the statement of thepresent
lemma we have
Both constants cl, c2 are absolute.
By
the estimate(13)
the result follows.0 After these
preparations
we establish thefollowing
result(cf.
alsoElliot
[3,
Lemma3.5]).
Lemma 2.3.
Having
the sameassumptions
as in Lemma2.1,
suppose thatfor
N(~)
T~
for
some/3
> 0.Define
thestrongly
additivefunction
where
l(p)
is afunction from
to R.Define
theindependent
randomvariables
Wp
for
eachprime
divisor pof 11 by
Then
do not
differ by
more than/ ,
where
{3’
=min( 1, (3).
This holdsuniformly for
all sets F.Proof.
Consider the variablesIt is easy to see, that
P(E Wp
EF) _ p(n : g(an)
EF).
Thus it remainsto show
that it
and vapproximate
each other.Let E be the set
of all () 111,
for whichE
0.
Then v isagain
aprobability
measure on the setalgebra generated by
the elements of S.Following
Thuswaldner[19,
p.496]
weget
the estimateIt remains to treat the sets
Et
withN(t)
> T. Weget
from[19,
Lemma1]
(cf.
also[3,
Lemma2.3])
By
ourassumption
we haveApplying
(15)
to the sumscontaining
put in(16),
we arrive at(14)
(15)
and(17)
nowyield
the result. D3. PRIME POWERS
This section is devoted to the
generalization
of Gelfond’s result(3).
In-stead of theordinary
zeta function in our theorem the Dedekind zetafunc-tion of the number field
Q(i)
occurs.Theorem 3.1.
Suppose
m, r E Zfulfill
(5)
for
a base bof
a canonicalnumber
system
inZ[i].
LetT6 f(N)
be the numberof
elementsof
Ur,m(N)
that are not divisibleby
thef -th
powerof
aprime
(f > 2).
Then there is a constant 1L 1independent of
N,
such that the estimateholds.
Proof.
Define the functionwhere the s-sums run over all ideals of in the indicated range.
Setting
N1
=N~1-~»2,
, where A is as in
(6),
wesplit
the last double sum in twoparts:
For
IR21 ]
we obtain the estimateSince there are
0(rl) (E
>0,
arbitrary)
elements k in withN(k)
= r(cf.
[17,
Lemma4.2]),
weget
and,
henceKeeping
inmind,
thatNl
=N(1-À)/2,
weget
for c smallenough
In order to extract the main term, we
apply
(6)
toRi
to deriveSince
and
Setting p
=max(al, a2, a3)
1yields
the result. 04. AN
ERD6s-KAc
TYPE THEOREMNow we prove an Erd6s-Kac
type
theorem for thecomplex
sum ofdigits
function. This extends
(4)
to numbersystems
inZ[i].
Throughout
thissection
~b(x)
denotes the normal law. Furthermore we use the notationf (x) C 9O)
forf (z) =
0(g(x)).
Theorem 4.1.
Suppose
m, r E Zfulfill
(5)
for
a base bof
a canonicalnumbers
system.
in~(i).
Define
thefrequency
where
w(z)
counts the distinctprime
factors of
z EZ[i].
Thenholds
uniformly for X
and N > 8.Proof.
We will construct a measure, withrespect
to that the additivefunc-tion behaves like a sum of
independent
random variables. To this matter we willapply
thegeneral
Kubilius model defined in Section 2. In asecond
step
we will use theBerry-Esseen
Theorem(cf.
[1, 4])
to show theconvergence to the normal law.
Set t = exp and
let w, (z)
count the distinctprime
factors(
log log logN
p of z, whose norms
satisfy
N(b)
N(p)
t.LN(X)
shall denote thefrequency emerging
fromQN(X)
if onereplaces
by
wi (z) (We
will also use these functions with ideals asarguments.
In this casethey
count the distinctprime
ideal factors of theirargument.
Since is aprincipal
idealdomain we have
w(z)
= andwl(z)
=wl(z7G(il)
forany z E
Z[i]).
Now we want toapply
the Kubilius model:assign
the elements ofto the numbers aj of the model and set X =
IUr,m(N)I
and 17(q)
=N(q)-l.
For the
prime
ideals pl, ... , ps of the model we take allprime
ideals p withN(b) N(p)
t. The function S of the model turns out to beUsing
(8)
weget
the estimate S =log t(1
+o(l)).
Hence,
the conditions ofthe model allow T to be any power of N. Our first
goal
is theapplication
of Lemma 2.3. To this matter we have toshow,
that the setsEe
arenon-empty
up to a certain value ofN(~).
We do thisby using
Lemma 2.2.Hence,
wehave to estimate the remainder terms
R(N, t).
It is easy to see, that in our case the remainder termsR(N, t)
aregiven
by
In order to estimate the
R(N, t)
we use Hua’s result(11).
Let o be thedifferent of
7G~i~.
ThenThe sum runs over a
complete
residuesystem
inmod-ulo
c~-1
notcontaining
the element - 0(c~-1).
We will use this notation inthe rest of this paper.
Setting
for a E
yields
In the next
part
of ourproof
we need thefollowing
estimates(cf.
[18,
Lemmas 3.2 and
3.1]):
For 1 = 0 we haveMoreover,
forany ~
EQ(i)
wehave,
for :Using
the estimates(18)
and(19)
yields
Hence,
with A’ =max( ) , A)
we haveSet
11t’
andBy
(20)
and(21)
the conditions for theapplication
of Lemma 2.2 areful-filled. Since X = =
O(N)
by
(6),
weconclude,
thatEt =,4
0
for
N(t)
. With that we have established the conditions forthe
application
of Lemma 2.3. Define a collection ofindependent
randomvariables
by
Lemma 2.3 now
yields
where £’
indicates,
that the sum runs over allsquare-free
idealst,
whoseprime
factors p haveN(b)
N(p)
t.The estimate
(22)
holds for all t > 2 and formax(log t,
S) 1 8 .
Some calculationsyield
that the first error term in(22)
is foreach d E N.
Now we estimate the second error term. Recall that there are
(6
>0,
arbitrary)
ideals inZ [I]
whose norm isequal
to r. SinceI R (N, t)
I
=The last
step
is a consequence of(6).
This
implies,
together
with(22),
thatNow we turn to the second
step
of ourproof.
In order toapply
theBerry-Esseen Theorem on the convergence of a sum of
independent
randomvari-ables to the normal law
(cf.
~1, 4~),
we define the randomvariables
Zp
by
Let
Then,
by
(9)
we haveThe
Berry-Esseen
Theorem nowyields, again using
(9),
For the random variables
Wp
we deriveTogether
with(23)
thisyields
Since a number z with
Izl2
N has at mostlog log
N)
prime
factorsp with norms not contained in the interval
[N(b), t], we
canreplace
LN(X )
REFERENCES
[1] A. C. Berry, The accuracy of the Gaussian approximation to the sum of independent vari-ates. Trans. Amer. Math. Soc 49 (1941), 121-136.
[2] A. N. Danilov, On sequences of values of additive arithmetical functions defined on the set
of ideals of a field K of degree n over the field of rational numbers
(russian).
Leningrad Gos. Ped. Inst. Ucen. Zap 274 (1965), 59-70.[3] P. D. T. A. Elliott, Probabilistic number theory i. central limit theorems. Springer,
New-York, 1979.
[4] C. G. Esseen, Fourier analysis of distribution functions. A mathematical study of the Laplace-Gaussian law. Acta Math. 77 (1945), 1-125.
[5] A. O. Gelfond, Sur les nombres qui ont des propriétés additives et multiplicatives données. Acta Arithmetica 13 (1968), 259-265.
[6] B. Gittenberger and J. M. Thuswaldner, The moments of the sum of digits function in number fields. Canadian Math. Bull. 42 (1999), 68-77.
[7] P. J. Grabner, P. Kirschenhofer, and H. Prodinger, The sum- of- digits-function for complex bases. J. London Math. Soc., to appear.
[8] G. H. Hardy and E. M. Wright, An introduction to the theory of numbers. 2nd ed., Clarendon Press, Oxford, 1960.
[9] L. K. Hua, On exponential sums over an algebraic number field. Can. J. Math. 3 (1951), no. 1, 44-51.
[10] Z. Juskys, Limit theorems for additive functions defined on ordered semigroups with a reg-ular norm (russian). Lietuvos Matematikos Rinkinys 4 (1964), 565-603.
[11] I. Kátai and J. Szabó, Canonical number systems for complex integers. Acta Sci. Math.
(Szeged) 37 (1975), 255-260.
[12] B. Kovács, Canonical number systems in algebraic number fields. Acta Math. Hungar. 37
(1981), 405-407.
[13] B. Kovács and A. Pethö, Number systems in integral domains, especially in orders of
alge-braic number fields. Acta Sci. Math. (Szeged) 55 (1991), 286-299.
[14] J. Kubilius, Probabilistic methods in the theory of numbers. Amer. Math. Soc. Translations of Math. Monographs, No. 11, Providence, 1964.
[15] C. Mauduit and A. Sárközy, On the arithmetic structure of sets characterized by sum of
digits properties. J. Number Theory 61 (1996), no. 1, 25-38.
[16] C. Mauduit and A. Sárközy, On the arithmetic structure of the integers whose sum of digits is fixed. Acta Arith. 81 (1997), no. 2, 145-173.
[17] W. Narkiewicz, Elementary and analytic theory of algebraic numbers. Springer, Berlin, 1990.
[18] J. M. Thuswaldner, The sum of digits function in number fields: Distribution in residue classes. J. Number Th. 74 (1999), 111-125.
[19] ____, The fundamental lemma of Kubilius and the model of Kubilius in number fields.
Number Theory, Diophantine, Computational and Algebraic Aspects (Berlin) (K. Györy et. al., ed.), 1998, pp. 489-499.
[20] W.-B. Zhang, Probabilistic number theory in additive semigroups I. Analytic Number The-ory, Proceedings of a Conference in Honor of Heini Halberstam (Boston, Basel, Berlin) (B. C. Berndt et. al., ed.), vol. 2, 1996, pp. 839-885.
J6rg M. THUSWALDNER
Institut fur Mathematik und Angewandte Geometrie
Abteilung fur Mathematik und Statistik Montanuniversitat Leoben
FYanz-Josef-Strasse 18 A-8700 LEOBEN AUSTRIA