THE PROPINQUITY OF DIVISORS

P . ERDŐS AND R . R. HALL

Introduction
Let Í3 > log 3 -1 be fixed . Erdös [1 ] stated without proof that the sequence of
integers n having a divisors d, d' such that
d < d' < d(1+(log n)-a)
has asymptotic density 0 . It was also stated that if /3 < log 3 -1 then this density
is 1, but this claim has had to be withdrawn .
In this note we prove a result which is more precise than the former one quoted
above, particularly for the small d's (essentially those for which log d = o (log n)) .

THEOREM . Let e > 0 be fixed and set

.log log log log x)
q(x) = 3-(1 +e),/(2 log log x
and O(x,d) = gl(x)(logd)1-log3
Then the number of integers n < x having divisors d, d' such that
d < d' < d(1 +O(x, d))
is o(x) .
Alternatively, the sequence of integers n having divisors d, d' such that
d < d' < d(1 +O(n, d))
has asymptotic density 0 .

Remarks. (i) The two forms are equivalent because of the slow decrease of 11(n) .
(ii) If we restrict our attention to divisors d > xó (or > n') for any fixed S > 0, the
factor 0/log log log log x may be replaced by any function of x tending to infinity in
the definition of rl(x) . This will be made clear in the proof of the theorem . (iii) Since
the multiples of d(d + 1) have positive density, the theorem becomes false if O(x, d)
is any function of d alone, unless trivially O(x, d) 5 1/d . But it is not clear that our
q(x) is the most slowly decreasing function of x which will do .
The idea of the proof is to consider weighted sums over the integers n, the weight
being constructed so that only integers n with divisors d, d' close together have positive
weight, but in such a way that the sum is not dominated by n's and d's in which we
are not interested ; these are integers with either an abnormal number, or abnormal
distribution, of prime factors .

Proof of the theorem . Let n have divisors d and d' such that d < d' < d(1 +O(x, d)),
and suppose (d, d') = t . Since 0 is a decreasing function of d, we have that
(d/t) < (d'/t) < (d/t) (1 +0 (x, d/t)), so that n has a pair of such divisors which are
relatively prime . We therefore assume in what follows that (d, d') = 1 : actually all
we need is that dd'In .

Received 4 December, 1978 .

[BULL . LONDON MATH . SOC ., 11 (1979), 304-307]
 THE PROPINQUITY OF DIVISORS 305

We get a better result if we restrict to d > xó and we deal with this case first . Let

S = Z y " ) Y-B 1
n<x dd'jn
d>x ,
where Y,B denotes summation restricted by the condition d < d' < d(1 +0), n(n)
denotes the total number of prime factors of n (i.e. counted according to multiplicity)
and 0 < y < 1 . We have
S = y n(d) y n(d') yS2(m)
xő<d<xl/z d<d'<d(1+B) m<x/dd'

and to estimate the inner sum we use the formula, (valid for fixed y e (0, 2)),

E y"( m ) = C(y)x(log x)y - ' (1 + 0(1/log x)),

m<x
where
r1 (I-
C(Y) °
T(Y) P ) y ( 1- P ) - 1 .
This formula is well-known and is proved by contour integration . It holds for all x > 0
if we interpret log x as 1 say for x < e : and we make this convention throughout the
paper . Hence
yQ(á)
,7( y-1
S < x 2. . 2 (log 2 ) n
Y `d ),
d d d d<d'<d(1+B)

and we use the formula above again . Since 0(x, d) > 1/log x there is no difficulty
with the error term and
Yn`d) x y-'
S < x y_ 0(x, d) d (log ) (log d)y - '
d2

x Y-1
a
X11 (X)
(log x)v-log 3

xő<d<xl/~
y~(d) (log d 2

since for d > xó we have 0(x, d) <,, 0(x, x) . Now

'/x x v- 1
(log ) < E (log ni)y -'
d d2 m<xl/z/d

so that
x)y-log 3 (log m)y-1 E yn(a)
S <a VXl (x) (log m<xl/Z -á d<xl/Z/m

<óxyf(x)(logx)2y-1-log3 m - '(Iogm)v -1
m<xy_1/2-a

< ó xn (x) (I Og x) 3y - 1 - log 3

We choose y = 1/3, and deduce that

y_
31og log x-S2(n) Y_B
1 < xq(x) .
n<x MI n
d<xó

Let 1/á(x) -), oo arbitrarily slowly as x --> oo . It is a well-known result of Hardy and

E
306 P. ERDOS AND R . R . HALL

Ramanujan [5] that for all but o(x) integers n < x, we have

Ifl(n)-log log xl < O(x),/log log x .

We restrict the sum above to integers n for which this inequality is satisfied, and
deduce that the number of integers n with such a divisor dd' is

a x{o(1)+q(x)Y ( x)11iog log xl

By definition of q(x), this is o(x) as required . We notice that for these "large" d's the
factor /log log log log x in q(x) is not needed .
The proof is a little more difficult when we drop the condition d > xó . Consider
the sum
T = T - e z n(n, a) (log d) log 3 ,
n<x dd'In

where SZ(n, d) denotes the total number of prime factors of n which do not exceed d,
and 0 < z < 2. Notice that the condition d < d' < d(1 +B(x, d)) implies that
B(x, d) > 1/d, that is
d(logd)' °g s-1
> 1/n(x), or d > d&), say.
Then
Z52(dd', d) 3 ~ Z S2(m, d) .
T (log d)'°g
d<_xl1 2 d<d'<d(1+B) m<x/dd'

A result of Hall [4] states that if f : Z + -+ [0, 1] is multiplicative then

log log x \1 1 f(p) f(p 2 )

~, f(n) 5 eYx 1+O / rj 1- - 1+ - +
2
+
n<x ( log x ~ P<x ~ P ) ~ p p

x
and we deduce that ~ z' (' , d) < - (log d)Z -1 .
m<x/dd' d2
Substituting in the above, we obtain
zQ(d)
T < ~ - (log d)Z-1 +log 3
Zn(a ' d) .
d2 d<d'<d(1+B)

Provided 8(x, d) < 1, as we may assume, we have S2(d', d) , S2(d')-1 . Thus as before
z n(d)
T < xz -1 rl(x) ~-" - (log d) 2 z -1
d<x d
x

xz -1 q(x) Z ~
zn(a)
( log x) Z Z -1
- . (1-2z) d f t(log t) dt
2-sZ

(jog x)3z-1 +(1-2z) dt

xz-1rt(x)
t(log t) z-s z
á° X
We choose z = 1/3, and deduce that

E B 31og log d-4(n, a)

< xri(x)log log x .
n<xdd'in
 THE PROPINQUITY OF DIVISORS 3 07

We need the following lemma, which is an application of Theorem VI of Erdös [2] :

LEMMA . Let A > 0 be fixed, and let N(x, ~) denote the number of integers n < x
such that for some d, ~ < d < n, we have
I Sl(n, d) - log log dj > (1 +A),/(2 log log d . log log log log d) .
Then
lim lim sup x -1 N(x, ) = 0 .
{- 00 X- 00

Since N(x, ~) is a decreasing function of ~, if we substitute _ ~(x) where

~(x) -> oo as x -,, oo, then
lim x -1 N(x, l (x)) = 0
X- o0

and we apply this result with ~(x) = d o (x) . We restrict the sum above to integers n
not counted by N(x, do (x)), with A = s/2, and deduce that the number of integers
n < x with a divisor dd' of the required type is
N(x, d o (x))+xrl(x)log log x .3t1+x),/tz iae log xaoe log log log X>

By the definition of q(x) and the relation A = e/2, this is o(x) as required .

References
1. P. Erdös, "On some applications of probability to analysis and number theory", J. London Math .
Soc., 39 (1964), 692-696 .
2 . P. Erdös . "On the distribution function of additive functions", Ann . Math ., 47 (1946), 1-20.
3 . R . R. Hall, "The propinquity of divisors", J. London Math. Soc., (2) 19 (1979),35-40 .
4 . R . R. Hall, "Halving an estimate obtained from Selberg's upper bound method", Acta Arithmetica,
25 (1974), 347-351 .
5 . G . H. Hardy and S . Ramanujan, "The normal number of prime factors of a number n", Quarterly
J. of Math., 48 (1917), 76-92 .

Mathematical Institute,
Hungarian Academy of Sciences
and
Department of Mathematics,
University of York,
Heslington, York.