ON THE LIMITING DISTRIBUTION OF ADDITIVE ARITHMETIC
F U N C T I O N S
BY
P. D. T. A. ELLIOTTQ)
University of Colorado, Boulder, Colorado, USA
To Professor P.ErdSs on his sixtieth birthday1
L e t / ( n ) be a real-valued additive arithmetic function. Let a(x) and
fl(x)
be real-valued functions which are defined for all real numbers x >~ 1, and in such a way t h a tfl(x)>0.
For each real number z let N(x, z) denote the number of integers n not exceeding x, for which the inequality
/(n) - ~(x) <. z~(x)
is satisfied. Define the frequencies
n(n;/(n)-zc(x) -<< zfl(x))
= x - ! N ( x , z), (x >1 U.I n this paper we shall make certain restrictions upon the rate of growth of the renormalis- ing functions a(x) and
fl(x),
and then give necessary and sufficient conditions in order t h a t the above frequencies should converge weakly.For simplicity of exposition only , we shall assume that the function/(n) is strongly additive. I n other words, for each prime p and positive integer m the relation/(pro) =/(p) is satisfied.
No other assumptions will be made concerning the ]unction ](n).
I t also proves to be advantageous to consider frequencies which are defined in terms of a continuous parameter x.I n order to present our main result it is convenient to define independent random variables X~, one for each prime p, by
f(p)
with probability - 1X v = P
0 with probability 1 - -. 1 P (1) Partially supported by N.S.F. Grant NSF-GP-33026 •
5 4 P . D . T . A . E L L I O T T
THEOR~.M. Let fl(x) satis/y the conditions fl(x) ~ ~ , and sup Ifl(x)--fl(y)l=o(fl(x)), ( x ~ ) .
x89 <<_z
PROPOSITION A. There exists a real-valued/unction a(x), with the property that sup la(x)-ae(y)l=o(fl(x)), (x-+oo),
and such that the/requencies
~,~(n;/(n) - ~(x) < z/~(x)), (x/> 2), converge weakly as x-+ 0%
PROPOSITION B. There exists a/unction a(x) so that the distributions P( ~ X n - o~(x) <. zfl(x)), (x >12),
p~<x
converge weakly as x-+ oo. Moreover,/or each pair o/ positive real numbers e and u the condi- tion
Z 1- o, (x- oo)
z~<p<~x P lf(~)l > u~(x)
is satis/ied.
I/, in addition, the/.unction fl(x) is continuous/or x >~ 2, then each o/these two proposi- tions is also equivalent to the/ollowing proposition C.
PROPOSITION C. Set a = l +(log x) -1. Then there exists a/unction a(x) so that/or each pair o~ real numbers t and v the limit
exists, and is independent o/~. The/unction w(t) is continuous at t-~ O. Moreover, the lim- iting relation
1 ~ logp [itt(pq 1
lim l o ~ ~ p ~ exp =
9 1 + i T
is satisfied.
The /unctions a(x) which can occur in these propositions are determined uniquely up to the addition o/a/unction o/the/orm cfl(x) +o(fl(x)), c a constant. I n particular, i / A (and so B) i8 satisfied we may choose the same/unction a(x) in propositions A and B, and the limit laws will coincide. I/, moreover, fl(x) is continuous then we may choose the same/unction a(x) in all three o/the propositions. The limit law will then have a characteristic/unction o/the /orm exp (w(t)).
O N T H E L I M I T I N G D I S T R I B U T I O N O F A D D I T I V E A R I T H M E T I C F U N C T I O N S 5 5
Remarks. The addition to ~(x) of a function which is of the form eft(x) +o(fl(x)) merely convolutes the limit law with an improper law.
I f we assume t h a t fl(x) is a measurable function of x, t h e n the second of the two hy- potheses which we make upon its rate of growth is equivalent to the assertion t h a t for each positive real n u m b e r y, fl(x ~) Nil(x), as x-~ ~ . One can view fl(x) as a slowly oscillating func- tion of log x. For a study of the pertinent properties of measurable slowly oscillating func- tions we refer to the paper of v a n Aardenne-Ehrenfest, de Bruijn and K o r e v a a r [1].
Although we give a detailed proof of the theorem for strongly additive functions/(n) it is possible to prove t h a t the theorem is valid for an additive function / if and only if it is valid for the strongly additive function whose value coincides with the value of / on the prime numbers. The limit laws will then also coincide.
The theorem exhibits a connection between the theory of those Dirichlet series which possess Euler products, and the limiting behaviour of sums of independent r a n d o m vari-
ables.
I n particular, the present result includes the well-known theorem of Kubilius ([5]
Chapter 4, Theorem 4.1, p. 58), concerning the limiting behaviour of additive functions of class H.
We conclude this introduction with an historical example. L e t / ( n ) =co(n), the func- tion which counts the n u m b e r of distinct prime divisors of the integer n. Then Erd6s and K a c [3] proved t h a t as x-*
~x (n; ~o(n) - log log x ~< z 1V~og log x) --* a(z), where G(z) denotes the normal distribution, and which is defined b y
1 / '
G(z) = ~ | e-~W'dw.
V2~ J - ~
I t is easy to check t h a t the choices ~(x) = l o g log x, fl(x) = (log log x) :/~ fall within the scope of the above theorem. B y combining the equivalence of propositions A and B together with a well-known criterion of Gnedenko and Kolmogorov ([4] Chapter 5, Theorem 3, p. 130) we can assert the
COROLLARY. I n order that/or a strongly additive/unction/(n) the./requencie.s 9 z ( n ; / ( n ) - l o g log 9 < log 9
:),
Should converge to the normal law, it is both necessary and su/ficient that/or each positive number e the limiting relations
56 P. D. T. A. ELLIOTT
~ --~0,
1] f 2 ( p ~
log log x ~ ~ -~ 1, ( b o t h as x -~ oo )
<x, l f ( m J < e l / ~ P be satisfied.
Notation. I t is c o n v e n i e n t t o r e t a i n a n d e x t e n d t h e n o t a t i o n w h i c h was i n t r o d u c e d a b o v e . F o r a n y p r o p e r t y ... we d e f i n e t h e frequencies
rz(n; . . . ) = x - 1 ~ ' 1.
n~z
H e r e ' d e n o t e s t h a t t h e s u m m a t i o n is confined to t h o s e p o s i t i v e i n t e g e r s n for w h i c h t h e p r o p e r t y ... is v a l i d .
W e shall use Cl, c 2 .. . . t o d e n o t e p o s i t i v e c o n s t a n t s . These will be a b s o l u t e unless o t h e r - wise s t a t e d .
2. P r o o I o f t h e t h e o r e m
W e shall give a n e s s e n t i a l l y cyclic p r o o f of t h e t h e o r e m . I n o r d e r t o d o t h i s i t will b e c o n v e n i e n t t o i n t r o d u c e a m o d i f i e d f o r m of p r o p o s i t i o n C, n a m e l y :
P R o P o S I T I 0 N C 0. S~t O" O = 1 + (log x)-l, and s o = (70 + i~(ao - 1 ). T h e n there are/unctions
~(x), x>~2, and it(t), such that as x ~ o o :
\ ~ p t \ f l ( x ) / -
~ ( x ) / = if(t) + ~T h e / u n c t i o n it(t) is independent o / v , and continuous at the point t = 0 . Moreover, there is an interval I tl <~ t o about the origin t = 0 in which the limitinq relation
1 l o g p / i t l ( p ) ] 1 l i m 1 - o g x Z p S 0 e x p -
~_.~ \ fl(x) / 1 + iT
is v a l i d / o r every real number ~.
W e shall p r o v e t h e t h e o r e m b y e s t a b l i s h i n g t h e sequence of p r o p o s i t i o n s A - ~ C 0-~ B - ~ A a n d C ~ C o.
3. P r o o f t h a t A i m p l i e s C O
I n t h i s s e c t i o n we c o n s i d e r D i r i c h l e t series whose coefficients d e p e n d u p o n a r e a l p a r a - m e t e r x. W e s h o w t h a t i n f o r m a t i o n c o n c e r n i n g t h e b e h a v i o u r of t h e s e coefficients u n d e r a t r a n s f o r m a t i o n x -~ x v, w h e r e y is a p o s i t i v e r e a l n u m b e r , l e a d s t o i n f o r m a t i o n c o n c e r n i n g a c e r t a i n l i m i t i n g b e h a v i o u r of t h e D i r i c h l e t series w h i c h t h e y define~ a n d c o n v e r s e l y .
ON THE LIMITING DISTRIBUTION OF ADDTTIVE ARITHMETIC FUNCTIONS 5 7
L e t s = a + iT d e n o t e a c o m p l e x v a r i a b l e . W e shall set a0 = 1 + (log X) -1, a s i n P r o p o s i - t i o n C o .
F o r e a c h real n u m b e r t d e f i n e
g(n) =
e x p(it~(x)-l/(n))
so t h a t
g(n)
is a m u l t i p l i c a t i v e f u n c t i o n of n, w h i c h satisfiesIg(n) l
~< 1 for n = 1, 2 . . . . W e d e f i n e t h e a s s o c i a t e d D i r i c h l e t seriesG ( s ) = ~
g(n)n -~.
r~=l
T h i s series is a b s o l u t e l y c o n v e r g e n t in t h e h a l f - p l a n e a > 1.
L e t y be a r e a l n u m b e r . W e shall consider t h e c o n t o u r i n t e g r a l
o(s)ds.
W e b e g i n b y e x a m i n i n g t h e b e h a v i o u r of
J(x, y)
as x o oo, for a f i x e d v a l u e of y. W e shall n e e dL E M l g A 3.1.
Let the/requencies
~'x(n;/(n)
- ~(x) <<. z~(x)), (x >12),converge weakly to a distribution F(z). Let q~(t) denote the characteristic function o] F(z). We assert that as x--~ ~ we have
x_~J(x, y) exp ( _ it:t(x)~.~lq~(t) i/
y > Ofl(x)] [ 0 i t y<O.
Proo/,
SinceG(s)
converges a b s o l u t e l y for a > 1, we m a y a p p l y a s t a n d a r d t h e o r e m o f P e r r o n (see for e x a m p l e T i t c h m a r s h [6] C h a p t e r I X , p. 300), a n d d e d u c e t h a t if x u i s n o t a n i n t e g e r , t h e nJ(x, y)= ~ g(n),
n~xY
I f y < 0 t h e n
J(x,
y ) = 0 , a n d t h e s e c o n d of t h e t w o a s s e r t i o n s c o n t a i n e d in L e m m a 3.1 is i m m e d i a t e . Suppose, therefore, t h a t y > 0 . I n t h i s case, t h e s a m e t h e o r e m of P e r r o n as- sures t h a t we c a n o m i t t h e c o n d i t i o n t h a t x u be n o n - i n t e g r a l p r o v i d e d t h a t we a d d t o t h e s u m o v e r t h eg(n)
a t e r m w h i c h is h e r e a b s o l u t e l v b o u n d e d . C o n s i d e r t h e e x p r e s s i o nK = K(x,
y) = e x p- - ~ 1 I • g(n).
n<~xY
5 8 P. D. T. A.. EI-J.,IOTT
W e shall estimate this f u n c t i o n
K(x, y)
b y deforming it into an expression to which we can a p p l y t h e hypothesis of L e m m a 3.1. L e t us first replace the function ~(x) b y a(xY). Since, b y t h e hypothesis of proposition A, a ( x ) - g ( x ~)=o(fl(x))
as x-~ co, this changes t h e value of t h e sum K b y at m o s t2.,
This s u m in t u r n does n o t exceed
H e n c e
~ - x ) = o ( 1 ) ,
p.<. \ I. fl-(x) j ] + o ( 1 )
as x-~ oo. W e n e x t replace
fl(x)
b y fl(xv). This is a little m o r e complicated. L e t e be a posi- tive real number. Choose a real n u m b e r u, so large t h a t for all sufficiently large values ofx the inequality
v=,(n; I / ( n ) - ~<(~')I < ufl(#)) > 1 -
is satisfied. T h a t this can be d o n e is assured b y the second of t h e t w o assertions which oc- cur in proposition A of t h e theorem. F o r a particular value of x, let t h e integers n which are c o u n t e d in this last f r e q u e n c y be d e n o t e d b y nj, (j = 1, ..., r). W e write
K(x'y)=x-Ys~,exp= \-it t )l]-~] j/+x-":~:,2 (...)=Y~I+V~,
n * n f
say. W e can o b t a i n an u p p e r b o u n d for t h e second of these two sums at once b y ]Z2] ~< e.
I n each of t h e terms in Z 1 we replace fl(x) b y fl(x~). This will t h e n change the value of Z a b y n o t m o r e t h a n
x-~j~ exp ( - it {/(nJ)flT~(z*!}) - exp ( - i" ft(nj)- ~(~)l~ l
-<~-'2 Itll/(,~,)-,~(~)l 1 <ultllB(~)-B(x)l
j~i ~ ) ,6(x) o(1), (x-~oo).
W e h a v e n o w p r o v e d t h a t as x -~ oo
ON T H E L I M I T I N G D I S T R I B U T I O N O F A D D I T I V E A R I T H N I E T I C F U N C T I O N S 5 9
(101 1.
J=~
l
~(~) l /B y a d d i n g t o t h e s u m w h i c h occurs i n t h i s e q u a t i o n t h e a p p r o p r i a t e t e r m s , we see t h a t l i m s u p
K ( x , y ) - x -u ~
e x p ( " ' f ] ( n ) - - ~ ( ~ ) ] ' ~ [Since e c a n b e c h o s e n a r b i t r a r i l y s m a l l :
K(x, y) = f~_~ e~t~d~,xy (n ; / (n) - a(x ~) <~ zfl(x-v)) +
o(1 ),as x-+ ~ . H o w e v e r , t h e i n t e g r a l w h i c h a p p e a r s on t h e r i g h t h a n d side of t h i s e s t i m a t e is t h e c h a r a c t e r i s t i c f u n c t i o n of a f r e q u e n c y w h i c h converges w e a k l y t o
F(z)
as x-~ ~ . B y a s t a n d a r d t h e o r e m f r o m t h e t h e o r y of p r o b a b i l i t y we d e d u c e t h a t as x ~ co( _
its(x)]
=K(x,
u) ~~(0.
x-UJ(x, y)
e x p \ fl(x) ] T h i s c o m p l e t e s t h e p r o o f of L e m m a 3.1.To c o n t i n u e w i t h our p r o o f t h a t p r o p o s i t i o n A i m p l i e s C O i t is c o n v e n i e n t t o t r a n s f o r m t h e i n t e g r a l
J(x, y)
b y t h e s u b s t i t u t i o n ~-~ ( a 0 - 1 ) v . W e t h e n o b t a i n t h e r e p r e s e n t a t i o n= Z (~ o(s)
J(x,y) 2~,)_~r s
x ~ U ( a ~ w h e r e i t is n o w t o b e u n d e r s t o o d t h a t s = s o = % +iv(ao - 1).
D e f i n e
h(v) = ~ (o" o - - 1 ) .
~ 8 o
T h e n since x sou = x v e x p (y{s o - 1} log x) = x u e x p (y{1 + iT} (a0 - 1) log x) = x u e x p ((1 +
iv) y)
we c a n w r i t e-u J(x, y) = ~ e *~uh(v)
dr.
(ex)
It is pertinent at this point to note thatI O(so)l (~o- l) a(~) Ih(,)l- 2~1~0+ i,(~0-1)1 < Vt + ~*
w h e r e t h e ' c o n s t a n t ' 2(x) d e p e n d s u p o n x. F o r each f i x e d v a l u e of x t h e f u n c t i o n h(v) be- longs t o t h e class L2( - co, co).
6 0 P . D . T . A . E L L I O T T
I t will be convenient in w h a t follows to denote the fourier transform of a function h, b y h ^. I n fact the fourier integral involving h(v) can be proved to exist as an improper R i e m a n n integral.
We have so far proved t h a t
fJ I
say. Since for positive values of x >~ 1 and y the inequality
I J(x,
Y) I ~<cl x~ holds uniformly, the function which occurs on the right hand side of this equation belongs both to the class L( - cr oo) and L~( - co, ~ ) with respect to the variable y. We can thus apply a fourier inversion to obtain the relationito~(x)~
1 rh(v) exp -
\ 3(x) /
= ~J_ooe-'r~e(x, y)dy.
F o r each fixed value of y 4 0 , we have proved in L e m m a 3.1 t h a t { O ~ ( t ) i f Y > 0
~(x, y) -> if y < 0.
Moreover, I~(x, Y) I <
cl e-~
holds uniformly for all values of x >~ 1. We m a y therefore a p p l y Lebesgue's theorem on dominated convergence, to deduce t h a tlimh(z) exp{_ito~(x)~=l ~oo e-"~e-Vq~(t)dy = q~(t)
(3.2)x-~o \ fl(x) / 2~ J_~ 2~(1 + i~)"
L e t us examine this expression involving h(v). Let ~(s) denote the R i e m a n n zeta-function, which is defined for a > 1 b y
~(s) = ~ n -s.
n = l
This function is well-known to be everywhere analytic except for a simple pole with re- sidue 1 at the point s = 1. We shall only need its properties in the neighbourhood of s = 1.
We write h(T) in the form
~ao-
I$(So)}"
h(v) = G(8o) ~(So) -1 t 2--~s0
As x -~ ~ the expression inside the curly brackets has the estimate
%-1 1 %-1 1
2 ~ % ( 1 + ~ 2 ~ e o ( O ' o - l + i ' t ' ( O ' o - 1 ) ) 2 ~ ( l + i v ) "
O N T H E L I M I T I N G D I S T R I B U T I O N O F A D D I T I V E A R I T H M E T I C F U N C T I O N S 61 F r o m our limiting relation (3.2) we deduce t h a t
G(s0) ;(s0) -1 exp ( - it~(~)~
( x - ~ o o ) . (3.3) B o t hG(s)
a n d ~(s) possess E u l e r p r o d u c t s in t h e half-plane a > 1. I n t e r m s of these we we can writeF o r each p r i m e p define
O , ( x / = l o g (l + g ( p ) p - ' " ( 1 - p - 8 " I {1 - P-~'}I - {g(P) - 1}P -'~
Define also O(X ) --- E I0~ (x) l.
W e assert t h a t if t h e principal value of the logarithms are t a k e n , t h e n as x-~ 0% O ( x ) ~ 0 . I n f a c t if p is large, a n d we a p p l y T a y l o r ' s t h e o r e m , t h e n on the one h a n d
1 r162
H e n c e for all a b s o l u t e l y large n u m b e r s P , u n i f o r m l y in x ~> 2:
E ]0v(x)l = O( Y p - 2 ) =
O(p-1).
p > P p > P
On t h e o t h e r hand, for each fixed p r i m e p it is easy t o see t h a t 0p (x) -~ 0 as x ~ co. W e deduce t h a t
lira s u p @(x)=
O(P-1).
x--> oo
B u t P can be chosen a r b i t r a r i l y large, so t h a t @(x) ~ 0 as x ~ ~ , as was asserted.
A p p l y i n g this result, a n d m a k i n g use of (3.3) we see t b a t
e x p ( g ( p ) = 11 ~ - s . _
fi(X)i]
=~0(t)+o(1), a s x - ~ oo.H e r e t h e function q~(t) i s a characteristic function, and, so is :continuous f o r all v a l u e s of t, a n d in particular a t t h e p o i n t t = 0 .
This p r o v e s the v a l i d i t y of t h e first assertion m a d e in p r o p o s i t i o n Co, with #(t) =~0(t).
T o o b t a i n t h e second p a r t of p r o p o s i t i o n C o we c a r r y out t h e s a m e series of operations, b u t using
G'(s)=dG(s)/ds
in place ofG(s).
T h e integral corresponding toJ(x, y)
t h e n has a n a p p r o x i m a t e r e p r e s e n t a t i o n62 P . D . T. A . E L L I O T T - ~ g(n) log n.
This introduces an extra factor of y log x into the calculations, b u t no further complica- tions occur. Proceeding along the above lines we arrive at the asymptotic relation
O'(s0) ( a 0 - 1 ) 2 exp
( _ i t a ( x ) ] 1 fo ~
2~s 0
fl(x) ] ~ ~ e-'~Uye-Uq~ (t) dy
- ~(t) 27e(1 + i~) ~"
Since qg(t) is a characteristic function there is a proper interval about the origin, It ~<t 0 say, in which ~0(t) does not vanish. F o r values of t in the interval [t I ~< t o we a p p l y the above asymptotic relation together with the (genuine) asymptotic relation (3.3), to deduce t h a t
( a 0 - 1 ) G ' ( s ~ - 1
G(so) 1 + iv' (x ~ oo).
B y logarithmic differentiation of
G(s) :
o ' (~o) = _ Z g(p) p - ~" log p + o ( 1 ) , G(So)
a s x---> c o s o t h a t
- log Z-y. #(x) /
This leads at once to the validity of the second assertion contained in proposition Co, and we have completed a proof t h a t
A-+ C o.
4. Proof that C O implies B
I t is convenient to begin b y proving the second of the two assertions which we m a d e in proposition B.
We consider the second of the two limiting relations of Co, namely t h a t if I tl ~< t o then
1 ~v p-S~
log p exp(it/(p)] 1
log x \ f l - ~ ] -~ 1 + i v ' (x -~ ~ ) . F r o m the theory of the R i e m a n n zeta function, as x ~ oo we have
~"(So) 1
1 ~ p - S o l o g p ~ _ ( a 0 _ l ) ~ ~ l + i v log xso t h a t l o g x Z p - ' , ( 1 - g ( p ) ) l o g p - ~ 1 0 , ~ (x-~ oo).
O N T H E L I M I T I N G D I S T R I B U T I O N O F A D D I T I V E A R I T H M E T I C F U N C T I O N S 63 W e set ~ = 0, a n d t a k e real parts. I n this w a y we deduce t h a t for ] t I < to,
s(z) = fd~g 1 ~ ~ p - o . ( 1 - R e g(p))log p ~ 0 ' ( z ~ oo).
Here 1 - R e
g ( p ) = 2 ( s i n t[(p)/2fl(x)) ~.By
meansof the inequality Isin
m u I<mlsin u I
which is certainly valid for e v e r y positive integer m, a n d real n u m b e r u, we can e x t e n d the v a l i d i t y of this last limiting relation to hold for each real n u m b e r t. I t is convenient to note at this p o i n t t h a t since
I g(P)] = 1
S(x) < 2 ~ < 1
log x u n i f o r m l y for all x ~> 2.
L e t s a n d u be positive real numbers. Set T =
2/u.
T h e n we easily o b t a i n the chain of inequalities. T / ( p ) \
1 ~ log 1 ~
21ogp 1 sm , , i-8(-~-I
2 ~ log p 1 fo r
= l o g x z~<p<~11~ pC~ T sins
t](p)
dt1 ( T 1
Y: p-~~
l f o 8 ( x ) d t .B y applying Lebesgue's t h e o r e m on d o m i n a t e d convergence we see t h a t the integral which occurs on the e x t r e m e right h a n d end of this chain of inequalities is o(1) as x-~ oo. I n the range x ~ < p ~<x 1/~ we h a v e log p > e log x, a n d p - 1 < p - ~ . exp (I/e). I n particular, therefore, we have p r o v e d t h a t
lim ~ _l = 0.
x--~:r z~<p<x P If(P)l> ufl(x)
This is t h e second of t h e t w o limiting relations which are asserted to be valid in proposi- tion B.
W e shall n o w a p p l y this last relation to simplify the result t h a t
ito~ ( x ) ~
L e t e a n d u be positive real numbers. We shall u l t i m a t e l y allow t h e m decrease t o zero.
T h e n from t h e a b o v e results we can assert t h a t as x-* oo
64 P. D. T. &. ~.LI~OTT
P-'Ig(P)- ll~<2exp(l/e) E p-1=o(1).
<p<~xl le x~ <p<~xllg
If(v)l >ufl(x) If(v)l >u~(x)
On the other hand, whenever l i(P)[ ~
ufl(x)
is satisfied we can assert t h a t [g(P) - 11~<I/(v)fl(x)-' I <
u.Thus
Z p-~~ <~u ~ P-x=2u(-loge+o(1)),
Xe <p<~xlle ze < p<~xlle
If(P)[ <~ul~(x)
as x-+ c~. F r o m these last two estimates we deduce t h a t
lim sup ~
p-~"lg(p)-ll<-2uloge.
(4.1)X.--~oo xe < p ~ x l l e
Determine the unique integer k so t h a t 2~<
xXt~<~
2 k+l. Consider the right hand side of the following inequality:Y v- 'lg(v)-ll<2
~ > x l ! e m ~ k 2m < 1~<2 m + l
For each integer m the innermost sum has the value
p_,exp(_(qo_l)logp)<~2.,< ~2.,+ p_lexp (
m l o g 2 ~~=<p.<2m+l v~ l o g x /
~< m -1 c2 exp ( m log 2~ p~+i y log 2~
< cak-aJm
exp ( -l~g x ]dy.
i#g i
Here we have made use of the elementary estimate, which is uniform in all positive in=
tegers m :
._.,<v<2,.+ ~ ~ = ~ - - 1 log log 2m+~-log log 2m + 0 (l~g 1 2- ~ 1 = 0 ( 1 ) 9
The constants c z and c a are absolute, Summing over m = k, k + 1 . . . . we obtain the upper bound
2cak -1
logx/dy<~exp
l o g x / k l o g 2From the definition of k it follows t h a t k + 1 >~log
x/e
log 2, so t h a t if x is sufficiently large (in terms of e) the right hand side of this inequality will not be more than 4cae. Putting this inequality together with t h a t of (4.1) we see t h a tlira sup
~ p - r Ig(P) - I I < -2u log
e +4c3e.
(4.2);g--+ oo ~ : > x ~
In particular we deduce t h a t
x e < p ~ x
ON THE LIMITING DISTRIBUTION OF ADDITIV~E ARITHMETIC FUNCTIONS
p-llg(p)-ll<~e ~ p-r176 ), ( x ~ ) .
xe<p~<x We shall need this result later,
L e t us now examine the sum
Y p-'* (g(p) - 1).
p <~ x8
65
I f we replace s o b y 1 we change the value of this sum b y at m o s t
I (g(p)- 1) {p-~~ p-1}l < 2 ~ p-1 lexp ((s0-1) log v ) - 1 I.
p ~ x s p ~ X e
We note t h a t since each prime p does not exceed z , when x is large enough ] s 0 - 11 log p ~<
(1 + T2)89 < 1/2, provided only t h a t e is sufficiently small in terms of T. I n these circumstances l e x p t t s 0 - 1 ) log p ) - l l ~ ~ (]s o - l l l o g p ) m m ~ Z l s o - 1 I l o g p .
m=l *
Hence the error t e r m which we have presently introduced is not more t h a n 2 Is 0 - 11 ~ ~o -1 log p = 2 Is 0 - 11 (log ~ ~- 0 ( 1 ) ) < 3 (1 -~- T2)89
~x8
Here we have made use of another elementary estimate from the theory of numbers, namely
log
P=logy+O(1),
P < Y
which is valid for all real numbers y ~> 2.
P u t t i n g all of these inequalities together (with T = 0 ) leads to the following inequality
ito:(x ) ~ _/a(t)
lixm2u p e x p (p~<xP-1 (g(~o) -- l) -- ~ - j - ] < c 4 ( - - u l o g 8~-8 ).
valid for all sufficiently small b u t positive values of u and e. Letting u-~O + and t h e n e-~O + we arrive at the limiting relation
ira(x)]
e x p (p~x~9-1(g(V) - l) -- ~ - ~ - ] -->/.~(~), (x---> oo). (4.3) Consider now the distributions
P( 5 X~-o~(x)<~zfl(x)),
(x~>l).p<~x
Their associated characteristic functions ~(x t) have the form 5 - 742908
Acta mathematica
132. Imprim~ le 18 Mars 197466 P. D. T. A. ELLIOTT
~ ( x , t ) = e x p (
i t ~ ( x ) ] ~ ) ] ~<~
( 1 §p ( g ( p ) - 1 ) ) .
In a calculation very similar to t h a t made in 3.3 concerning the function h(v) one can prove t h a t
ira (x) )
q ~ ( x , t ) = e x p ( ~ p - i ( g ( p ) -
1 ) - f l - - ~ - + o ( 1 ) . From the limiting relation (4.3) we see at once that~(x, t)-~#(t), (x-~
co).
Since lu(t) is continuous at the point t = 0 , it must be a characteristic function, and the random variable
~(X)-I{ ~ X p - 0 ~ ( x ) } p~x
converges to its corresponding distribution.
This completes the first assertion of proposition B of the theorem, and also the proof t h a t
Co-+ B.
5. Proof that B implies A
In this section we shall make use of a representation theorem of Kubilius.
L]~MMA 5.1.
Let x be a real number, x >~ 2. Let r be a /urther real number in the range 2 <~ r <~ x. Let ~(n) be a strongly additive/unction. De/inc independent random variables ~, one
~=
/or each rational prime p, by
i(p with probability 1 P with probability 1 - 1_.
P Then there is a positive absolute constant so that the inequality
v,(n;Z ? ( P ) < ~ z ) = P ( Z ~ < ~ z ) + O ( e x p ( - - -
pln, p~r p~r
clog x]~
log
r ]]
holds uniformly/or all real numbers z, r
(2<-r ~ x), and/unctions j(n).
Proo/.
Kubilius proves this lemma in his monograph [5], Chapter 2, pp. 25-27. Our use of the real variable x where he has an integer n is not of great significance.I t is convenient to define distribution functions
G(x, z) = n ( n ; / ( n ) - a ( x ) << zfl(x)), and
ON T H E L I M I T I N G D I S T R I B U T I O N OF A D D I T I V E A R I T H M E T I C F1TNCTIONS 67
H(x, z) = P( ~ Xv - a(x) <. zfl(x)),
PQq
for x >~ 1, where the r a n d o m variables Xp are those which are i n t r o d u c e d in t h e formula- t i o n of the m a i n theorem.
L e t e be a positive real number, a n d let a(x) be a function so t h a t t h e t w o assertions of proposition B are valid. T h e n we o b t a i n at once t h e inequality
G(x, z) <~ %(n;/(n) - ~(x) <~ zfl(x), ~ p i n , xs < p <~ x, ]/(P)] > s2fl(x) )
+~x(~; ~pl n, x~ < p <x, I/(p)l > ~/~(x)).
The second of the two frequencies which occur on t h e right h a n d side of this inequality does n o t exceed
1 xe<p<.x p ]f(p)[ > e*~(x)
a n d b y t h e second p a r t of proposition B this sum is 0(1) as x-~ ~ . As for t h e first f r e q u e n c y on this same side, we note t h a t if n is an integer which is c o u n t e d in it, t h e n
Z /(P) -- o:(x) <<./(n) -- a(x) & 5 I/(P)] <~ (z + e) fl(x),
p]n,p ~x~ xe < p ~ x, pln
since n can h a v e at m o s t s -1 distinct prime divisors p which lie in t h e interval x ~ < p ~<x.
Hence we h a v e p r o v e d t h a t
G(x, z) <~ % ( n ; / ( n ) - ~ ( x ) <~ (z +~)fl(x)) +o(1), (x ~ co). (4.2) W e n o w a p p l y L e m m a 4.1 (Kubilius' representation theorem) with r = x ~, a n d replace t h e expression on t h e right h a n d side of the inequality b y
P ( ~ X~ - ~(x) < (z + e) fi(x)) + O(exp ( - c 8 - 1 ) ) + 0(1).
p ~ x ~
I n turn, the p r o b a b i l i t y which appears in this expression certainly does n o t exceed P ( ~ X ~ - ~ ( x ) < ~ ( z + 2 e ) f l ( x ) ) + P ( ] ~ X p ] > e f i ( x ) ) = H ( x , z + 2 e ) + P , ,
p ~ X x e < p ~ x
say. W e can majorise t h e p r o b a b i l i t y P1 b y choosing a positive real n u m b e r ~], a n d in- t r o d u c i n g n e w i n d e p e n d e n t variables Yp, defined b y
{o x , if Ix~l < nt~(x)
Y~ = if I Xp ] > ~/fl(x).
68 1,. D . T . A . E L L I O T T
Then
Pl<-..P(3Xp:4:Yp, x~<p<.~x)+P(I ~. Ypl>efl(x))=P~+Pa,
x s < p ~ x
say. We can estimate P2 at once b y applying the second p a r t of hypothesis B:
P~-<< ~ - = o(1), 1 (x-~ ~ ) .
X e < p ~ X p If(P)[ > ~/~(X)
To estimate Pa we apply a standard a r g u m e n t of Tchebycheff:
Pa <~ (eft(x)) -2
Expect ( ~. Yp)~ = (eft(x)) -~ (Var ( ~ Y~) § (Expect ~. rp)2}x$ < p ~ z x e < p <~x x s < p ~ z
=(eft(x))-2( ~
V a r Y p + ( ~. E x p e c t Y ~ ) 2}xS <p ~ x X e < p ~ x
I fcv)l <,~(x) IIcp)l <,~cx)
< e - e ~ 2 ( ~. 1 §
\xs<p <x p
= e - 2 ~ 2 ( 1 - 1 o g e + o ( 1 ) ) ~, ( z - , o o ) .
Altogether this proves t h a t if ~ and e are positive real numbers 0 < e < 1, then as x-~
P1 ~< e - ~ ( 1 - l o g e) ~ + o(1).
We have therefore proved t h a t the inequality
G(x, z) ~ H(x, z
+2e) +e-2~*(1 - l o g e) 2 + O(exp ( - c e - 1 ) ) +o(1)holds as x-~ ~ , for a n y fixed pair of positive real numbers e and
u,
0 < e < 1, and uniformly.for all real numbers z.
I n a precisely similar way we can obtain the inequality
G(x, z) >~ H(x,
z - 2 e ) - e - 2 ~ 2 ( 1 - l o g e)~+O(exp ( - c e - 1 ) ) + o ( 1 ) , (x-~ oo).We can express these two inequalities in a somewhat different manner.
We recall t h a t if F and G are two distribution functions then their
Ldvy-distance
~(F, G) is defined to be the infinum of those real numbers h for which the inequalities
F ( z - h ) - h <. G(z) < F(z+h)+h
hold uniformly for all real numbers z. This defines a metric on the space of distribution functions; and a sequence of distribution functions Fn, (n = 1, 2 .... ) will converge weakly
O N T H E L I M I T I N G D I S T R I B U T I O N O F A D D I T I V E A R I T H M E T I C F U N C T I O N S 6 9
to a distribution function F if and only if ~(F~, F)-~0 as n-~ ~ . I n these terms our last two inequa]ities can be expressed thus:
lim sup o(G(x, z), H(x, z)) ~ 2e + e-~/2 (1 - log e) 2 + 0(exp ( - ce-1)).
We let ~-+0 + and then e-+0 + to deduce t h a t
q(a(x, z), H(x, z))-~ O, (x ~ ~ ) .
I t is now clear t h a t if the distribution functions H(x, z) converge weakly as x-~ 0% then so will the distribution functions G(x, z), and to the same limit law.
This establishes the first part of proposition A, and it remains to verify t h a t sup ] a ( x ) - a(y)l = o(fl(x)), (x ~ o~).
x89
I n order to do this we make use of the fact t h a t a(x) occurs as a renormalising function which is restricted b y the fact t h a t the distributions H(x, z) converge. We shall need a part of the following result of Gnedenko and Kolmogorov ([4], w 25 Theorem 1, pp. 116- 121.).
LEMMA 4.3. I n order that/or some suitably chosen constants A~ the distributions o[ the s u m 8
~,~, + ... +~,~ - A ,
o[ independent in/initesimal random variables converge to a limit, it is necessary and su/[i- cient that there exist non-decreasing/unctions
M(u), (M( - ~ ) = 0), N(u), (N( + ~ ) = 0),
de/ined in the intervals ( - ~ , O) and (0, cr respectively, and a constant a >~O, such that (1) A t every continuity point o / M ( u ) and N(u)
kn
lim ~ P ( ~ , ~ < u ) = M ( u ) , ( u < 0 )
X-->o0 k ~ I
k n
lim ~ { P ( ~ n k < u ) - l } = N ( u ) : (u>O).
n-->oo k = 1
(2) lim l i m i n f z~dP(~n~ < z ) - = o ~,
e-~0 n oo k ~ l ( J l z l < e
together with a similar relation obtained by replacing 'lim in/' by 'lim sup'.
70 P . D . T . A . E L L I O T T
The constants A n may be chosen according to the/ormula
A n = ~ zdP(~,,k< z)
k = l z[<v
where i v are continuity points o / M ( u ) and N(u).
Remark. I n their theorem, G n e d e n k o a n d K o l m o g o r o v also determine the f o r m of the limit law in terms of t h e representation t h e o r e m for infinitely divisible distribu- tions of L 6 v y a n d Khinchine. W e do n o t state this p a r t of their result since we shall have no need of it.
I n our present circumstances we shall be interested in t h e variables {~nk, 1 < k < kn} = { ~ ( n ) - ' x ~ , p < n}, (n = 1, 2 . . . . ).
T h u s we shall be considering c u m u l a t i v e sums of i n d e p e n d e n t r a n d o m variables, so t h a t a n y possible limit law will belong to t h e class L of K h i n c h i n e (see for example G n e d e n k o a n d K o l m o g o r o v [4], Chapter 6, w167 29, 30). E a c h of the functions M(u) a n d N(u) which occur in the f o r m u l a t i o n of L e m m a 4.3 are t h e n a c t u a l l y continuous, so t h a t in t h e final assertion of t h a t l e m m a a n y (fixed) positive value of T m a y be taken.
Since t h e distributions H(x, z), (x ~> 2) a n d therefore H(n, z), (n = 1, 2, ...) are a s s u m e d to converge weakly, there exists a continuous f u n c t i o n k(u) ,defined for real n u m b e r s u > 0 , so t h a t
1 -> k(u), (n -+ oo). (4.4)
p ~ n p Ircp)l > up(n)
Assume n o w t h a t the distributions H(x, z) converge to a proper limit law. T h a t is to say, a law whose characteristic function is n o t of the f o r m exp (ict), c a constant. Consider t h e sequence of distributions H(x, z), n = 1, 2 . . . Then, b y L e m m a 4.3 we m a y choose
I(P) A(n) = z.
p < n p
Ifc~)l < rfl(n)
p r o v i d e d t h a t r is a fixed positive real number. Since the variables fl(n) -I { ~. X p -
A(n)}
p~<n
fl(n)-' { ~ X ~ - a(n)}
p ~ n
converge to the same proper law, an e l e m e n t a r y result f r o m the t h e o r y of probability (see
O N T H E L I M I T I N G D I S T R I B U T I O N O F A D D I T I V E A R I T H M E T I C F U N C T I O N S 71
for e x a m p l e G n e d e n k o a n d K o l m o g o r o v [4] C h a p t e r 2, T h e o r e m 2, p p . 42-44) i m p l i e s t h a t
a S n - - > ~
A ( n ) - a ( n ) +0.
fl(n) W e n e x t p r o v e t h a t as n -+
sup I~(y)- ~(n) l= o(fl(n)) (4.5)
] y - n ] ~ l
To d o t h i s i t suffices t o n o t e t h a t if y lies in t h e i n t e r v a l n - 1 < y < n + 1 t h e n G(n, z) a n d t h e f r e q u e n c y
vn(m; / ( m ) - o ~ ( y ) <-zfl(y)), ( = n - l y G ( y , z)) b o t h c o n v e r g e t o a c e r t a i n p r o p e r l i m i t law. H e n c e , we d e d u c e t h a t b o t h
fl(y)/fi(n) -~ 1,
I~(y)
- a(n)]/fl(n) -~ o,as n-+ cr T h e first of t h e s e l i m i t i n g r e l a t i o n s is of no p r e s e n t v a l u e t o us, b u t t h e s e c o n d is t h e r e l a t i o n w h i c h we w i s h e d t o e s t a b l i s h .
L e t x a n d y be r e a l n u m b e r s which s a t i s f y 2 <.x 1/2 <~y <~x. D e f i n e i n t e g e r s m a n d n so t h a t m ~<y < m + 1, a n d n - 1 < x ~<n. T h u s t h e i n e q u a l i t i e s 2 ~<m < n ~<x + 1 a r e satisfied.
As x-+ ~ we see f r o m p r o p e r t y (4.5) t h a t
~z(x) - a(y) = o~(n) - a(m) + o(fl(x) )
= A ( m ) - A ( n ) +o(fl(x)) I t is t h e n c o n v e n i e n t t o w r i t e
A (m) - A (n) =
/(P) + ~ /(P) Z /(P)- ~ + ~ + ~3,
If(p)l <Tfl(n) If(v) > Tfl(m) f ( p ) > ~:fl(n)
s a y . L e t e be a p o s i t i v e r e a l n u m b e r . T h e n if x is large e n o u g h fl(n) <<, (1 + e) fl(m), so t h a t b y c o n d i t i o n (4.4) of t h e p r e s e n t s e c t i o n
tv<m. lr(v)l<~(l+8)~(m) p p<~m,Js p l
< ~ ( n ) { -- k(~ + e~) + k(~) + o(1)}, (x -~ ~ ).
H e n c e (since f l ( n ) ~ fl(x) as x ~ ~ ) ,
l i m sup fl(x)-I I ~ I~< - ~ {k(~ + e v ) - k(~)).
i
W e l e t e ~ 0 + , a n d r e c a l l t h a t ~ is a p o i n t of c o n t i n u i t y of k(u). I n t h i s w a y we c a n p r o v e t h a t Z~ =o(fl(x)) as x-+ ~ .
72 P . D . T . A . ~LTJOTT I n exactly the same way we prove t h a t ]E s =o(fl(x)).
To consider Z1 we split the sum into two parts. I n t o the first part we p u t those terms involving primes p for which I/(P)[ > ~fl(n). The contribution to ZI which such terms make is at most
1
l,-+(n)l > ep(n)
and by the second condition of proposition B this expression is o(fl(x)) as x-~ ~ . On the other hand, those terms which remain in ]El contribute not more t h a n
Hence
eft(n) ~
~ 891
efl(n)(log 2 § (n-~ ~ ) .lim sup
(x)-ll LI <
~ log 2,x---~ oo
holds for every positive value of e. Letting e-~0 + we see t h a t •1
=O(fl(X)),
and t h a t~(x)-a(y)=o(fl(x)), (x---~oo) holds uniformly for all values of y in the interval x 1/2 <~y <~x.
This is the second assertion of proposition A, and so we have completed a proof t h a t B ~ A , in every case except t h a t of when the distributions H(x, z), and so G(x, z), converge to an improper law. I n this last case we are concerned with a form of the weak law of large numbers. This has been considered b y the author on another occasion [2], and the argu- ments and results given there guarantee the validity of the inference B t r u e ~ A true in this special case under hypotheses on ~(x) and fl(x) which are considerably weaker t h a n those which are assumed in the present theorem.
6. Proof that C O implies C, and conversely
I t is immediate t h a t C ~ C o. Assume, therefore, t h a t proposition C O is valid. Then b y the proofs of w167 3-5 so are propositions A and B, with in fact the same ~r as a possibility.
Moreover,/~(t) is the characteristic function of a limit law for the sums
fi(x) -~ ( ~ x , - ~(x)).
p ~ x
I t is easy to see t h a t the variables fl(x)-lX~, (2 ~ p ~x), are infinitesimal, and since t h e y are independent such a law must be infinitely divisible. I n particular/z(t) will be non-zero for all real values of t (Gnedenko and Kolmogorov [4] Theorem 2 of w 24, Chapter 4, and
O N T H E L I M I T I N G D I S T R I B U T I O N O F A D D I T I V E A R I T H M E T I C F U N C T I O N S 73 T h e o r e m 1 of w 17, C h a p t e r 3). T h u s n o n e of our earlier a r g u m e n t s need t o be restricted t o a n y p a r t i c u l a r i n t e r v a l of t-values, a n d t h e second assertion of proposition Co (and there- fore of C) holds for each real value of t.
W e n o w show h o w to o b t a i n t h e first of t h e t w o assertions in proposition C. More exactly, let log #(t) be defined continuously f r o m t h e principal value t a k e n a t t = 0. (Since
#(t) is a characteristic function it will be a continuous function of t). T h e n we shall p r o v e t h a t as x-* co
i fl(x) ] ~ - ~ I g/~(t).
F o r ease of p r e s e n t a t i o n let us d e n o t e t h e expression which occurs here on t h e left h a n d side of t h e a r r o w b y
w(x,
s o, t). T h e n t h e proof t h a tA ~ C o will
also yield t h a t e x pw(x, s o, t ) ~
/z(t) u n i f o r m l y o v e r a n y (fixed) b o u n d e d sets of real n u m b e r s 13[-4%, It] 4 t 0. T h e o n l y a d j u s t m e n t n e e d e d is t h a t one o b t a i n s a f o r m of L e m m a 3.1 in which t h e convergence is u n i f o r m o v e r a n y fixed i n t e r v a l of y-valuesO<ce<y4c7,
a n d a n y b o u n d e d i n t e r v a l of t-values. This is easily o b t a i n e d since u n i f o r m l y for such a n i n t e r v a l of y-values {a(x ~) - a(x)}fl(x)-X~0, as x-+ r W e can therefore assert t h a t for s u i t a b l y chosen integersn(x, 3, t)
we h a v ew(x,
So, t) = log/u(t) +2zein
(x, 3, t) +o(1)as x ~ co, u n i f o r m l y for 131 430, Itl 4 t 0. T h u s we can find a real n u m b e r x 0 so t h a t if x > x 0, t h e n w i t h these s a m e uniformities
I w(x, 80,
t ) - l o g / z ( t ) - 2 gin (x, 3,
t) I < 88T h e function
w(x, so, t)
is a continuous function of x. (Here fl(x) is n o w a s s u m e d to be a continuous function of x. T h e n u m b e r 80 also d e p e n d s c o n t i n u o u s l y u p o n x). This can be readily p r o v e d as follows. L e t P be a positive real n u m b e r . L e t x 1 be a real n u m b e r (x 1>/2), 81 = 1 + (log xl) -1 + iv(log xl) -1. T h e nso, t)_W(Xl ' sl, t ) 1 4 4 ~ p - ~ . + :
(p-S._:p-S,){exp( i t l ( p ) ~ _
1} I Iw( ,As x 1 -~ x we h a v e s 1 -~ So, fl(xl) ~ fl(x), a n d deduce t h a t
lim sup
Iw(x,
So,t) - w(xl, s,,
t)[ 4 4 ~ p - ~ ' .XI*-->X l0 > P
Letting P - * co w e s e e that uniformly f o r 131 4 %, I tl 4 t0, we have I (x, 3, tl - (xl, 3, t) l < X
74 1". D . T . A . E L L I O T T
p r o v i d e d o n l y t h a t z 1 is sufficiently near to x, a n d x > x 0. We can therefore assert t h a t t h e integers n(x, v, t) in fact do n o t d e p e n d u p o n the value of x. L e t us therefore write n(T, t) in place of n(x, T, t). T h e n as x-~ oo
w(x, s o, t) ~ log/~(t) + 2xdn(v, t)
u n i f o r m l y for IT I ~T0, I t ] ~<t0" Since w(x, so, t) is continuous in 7 a n d t a n d the convergence is suitably u n i f o r m the function lim w(x, s o, t) (x-~ ~ ) is also c o n t i n u o u s in 7 a n d t. I n view of t h e continuous definition of log/~(t) t h e integer n(v, t) m u s t be continuous, a n d so a c o n s t a n t for 171 ~<70, Itl ~<t 0. Therefore over this rectangle of values of 7 a n d t we have
n(7, t)=n(O, O)=0. This proves t h a t as x ~
w(x, s o, t) -~ log/~(t) u n i f o r m l y for a n y rectangle of (7, t)-values.
W e have n o w established t h e first assertion in proposition C, a n d so completed the proof of the theorem.
7. Concluding remarks
I t is clear from the a r g u m e n t s of w 3 t h a t the properties of :r a n d fl(x) which se assume are of a simple n a t u r e with respect to their b e h a v i o u r u n d e r t h e t r a n s f o r m a t i o n s x ~ x y (y >0). I n fact, for our purposes these functions are essentially a s y m p t o t i c a l l y in- v a r i a n t u n d e r such transformations. I t is quite possible to consider other renormalising functions a(x) a n d fl(x) whose b e h a v i o u r u n d e r these t r a n s f o r m a t i o n s is entirely different.
T h e n a t u r e of the resulting conditions which are necessary in order t h a t the frequencies G(x, z) should converge weakly are t h e n quite different. I n particular, the f u n c t i o n /(n) need no longer b e h a v e like a s u m of i n d e p e n d e n t r a n d o m variables.
I n a n o t h e r direction, we can view the t r a n s f o r m a t i o n s x-->x y as f o r m i n g a g r o u p 17 (with composition as a group law) which is isomorphic to t h e multiplicative g r o u p of po- sitive real numbers. Our use of fourier analysis with respect to t h e variable y can t h u s be viewed as fourier analysis u p o n t h e group F. Accordingly, we can ask w h e t h e r in certain circumstances one m i g h t n o t p r o f i t a b l y use groups of t r a n s f o r m a t i o n s other t h a n I ~ with which to operate.
W e i n t e n d t o r e t u r n to various such questions at a f u t u r e date.
Note: Since this paper was accepted for pubfication it has come to t h e notice of t h e a u t h o r t h a t a f o r m of necessary a n d sufficient condition in order t h a t proposition A be valid has been established, inter alia, in a paper of L e v i n a n d Timofeev (B. V. L e v i n a n d N. M.
Timofeev: A n analytical m e t h o d in probabilistic n u m b e r theory. T r a n s a c t i o n s of the Vla-
oN THE LIMITING DISTRIBUTION OF ADDITIVE AI~,ITIIMETIC FUNCTIONS 75 d i m i r S t a t e P e d a g o g i c a l I n s t i t u t e of t h e M i n i s t r y of Culture, R S F S R . p p . 56-150, see pp. 113-117). T h e m e t h o d t h a t t h e s e a u t h o r s use differs c o n s i d e r a b l y f r o m t h a t of t h e p r e s e n t p a p e r . I n p a r t i c u l a r t h e a b o v e m e t h o d is one w h i c h a p p l i e s q u i t e n a t u r a l l y i n m o r e g e n e r a l c i r c u m s t a n c e s , as is i n d i c a t e d i n t h i s section.
References
[1]. VAN AARDENNE-EHP~ENFEST, T., DE BRUIJN, N. G. & KOREVAAR, J., A note on slowly ascil- lating functions. Nieuw. Arch. Wislc., 23 (1949), 57-66.
[2]. ELLIOTT, P. D. T. A., On the law of large numbers for additive functions. Proceedings o]
the 1972 St. Louis con]erence on Analytic Number Theory, Amer. Math. Soc. Publ., no 24, 90-95.
[3]. ERDSS, P. & KAC, M., The Gaussian law of errors in the t h e o r y of additive number-theoretic functions. Amer. J. Math., 62 (1940), 738-742.
[4]. GNEDENKO, B. V. & KOLMOGOI~OV, A. N., Limit distributions for sums of independent ran- dom variables. Addison-Wesley, Reading (1954).
[5]. KUB~LIUS, J., Probabilistic Methods in the Theory of Numbers. Amer. Math. Soc. Transla- tions, Vol. 11 (1964).
[6]. TITCHMARSH, E. C., The Theory o] Functions. Oxford University Press, corrected second edition, (1952).
Received March 20, 1973