J . Austral . Math . Soc .

(Series A) 25 (1978), 4 79-485

ON SOME UNCONVENTIONAL PROBLEMS ON THE

DIVISORS OF INTEGERS

P . ERDOS and R . R . HALL

Dedicated to Professor K . Mahler for his 75th birthday

(Received 14 December 1977)

Communicated by J . H. Coates

Abstract

In this note a number of related problems about divisors are studied, and partial solutions
obtained by elementary means . The problems are rather unconventional and seem to suggest
interesting developments .

Subject classification (Amer. Math. Soc . (MOS) 1970) : 10 H 25 .

I, Introduction
In this note we study a number of related problems concerning the divisors of an
integer n . We denote these divisors by d and their number by T(n) ; they are labelled
in increasing order, thus 1 = d1 < d2 < . . . d, . = n . As usual v(n) denotes the number
of distinct prime divisors of n .
All the problems considered here were raised by one or the other of us at various
times : broadly speaking they are connected inasmuch as they are about relations
between divisors, often between d2 and dti+1, rather than arithmetic or analytic
properties of individual divisors .
To give an example of the problem we have in mind, consider the following
conjecture of P . Erdős that states the density of integers n which have two divisors
d1 < d2 < 2d1 is 1 . P . Erdős (1964) stated that he can prove this-unfortunately this
claim has to be withdrawn . More generally it was conjectured that the density of
integers n which have two divisors

d1 < d2 < dl(1 + (log n) -a), a < log 3 -1,

479


480 P. Erdős and R . R . Hall [21

is 1 . We only know that if true this is best possible, that is it does not hold for
a>log3-1 .
The following conjecture seems interesting : Denote by T+(n) the number of
integers k for which n has a divisor d satisfying 2 k ~< d < 2k+i Then T+(n)/T(n) --> 0 if
one disregards a sequence of density 0 . This conjecture if true of course implies that
for every e the density of integers n which have two divisors d, < d2 < (1 + e) d, is 1 .
The trouble is that at the moment we cannot attack this conjecture at all .

2
Let f(n) = card {i : (di,di+,) = 11 . Each prime divisor of n is an admissible di+i
so that f(n) v(n), with equality when n = plp2 . . . p v and pi >PIP2 . . . pi-, for 2 < i < v.
Thus the average order of f(n) is at least log log n ; we should like to determine the
average and maximum orders . Concerning the maximum order, we have the
following result .

THEOREM 1. For every e > 0 and x > xo (e),

max f(m) > (exp (log log x) 2-E) .

m<x

Next, let T k (n) denote the number of divisors of n of the form

d = t(t+ 1) . . . (t+k-1) .

In the case k = 2, an equivalent definition is T2 (n) = card {i : di+1 -di = 11 so that

T2(n) ~<f(n), with equality for a number like n = 2 .3 .7 .43 where

n = p1p2 . . . pu'pi+1 = plp2 • • • p i+ 1 .

It is easy to see that T2(m) = f(n) holds only for a finite number of n's . The average
order of 7- k (n) is a positive constant, indeed for k ~> 2, we have

+Q(xl1k)
Y Tk(n) _ (k-1)(k-1)1
n_<x

but the maximum order will be harder to determine . We have

THEOREM 2. For each k > 2, and every fixed A < ell k, we have Tk (n) > (log n) -4
infinitely often .

It is certain that T(n) > (log n)°, infinitely often for every c, but this may be very
difficult . Incidentally it is easy to see that the density of integers n for which
Tk (n) = r exists . Denote this density by ajr) . We have X-u ak (r) = 1,


[31 Problems on the divisors of integers 481

We can ask many questions about the function

1k (n) = min {t ,>1 : n i t(t + 1) . . . (t + k -1)},

and its restriction to the sequence of factorials . Plainly t72 _i ( m!) = 2, and we can
show that t m -2(m!) = O(m) (this is best possible, for example, if m = 2k) . What
can be said about t .-,(m!) ? It is true that for infinitely many values of n and
every 1 < i 5 n -1

(1) t; (n !) < ti-i(n) -1 ?

In particular we showed with Selfridge that (1) holds for n = 10 .
More generally let Fn be the smallest integer with F,,!-=0 (mod n) . Can one
characterize the integers n for which all 1 < i < Fn

(2) t i (n) < ti-,(n) -1 .

If Fn is very large (2) clearly cannot hold . What is the largest value of Fn for
which (2) holds? For how many n < x can (2) hold ? The maximum order of t k (n)
is easily settled since for primes p > k we have t k (p) = p+ 1-k . Here it is the
average and normal orders which are of interest . We have the following result .

THEOREM 3 .

1 log log log x

- E t 2 (n) << x
X n , x log log x
We conjecture that for some fixed a > 0, we can replace the right-hand side by
x(log x) -", indeed it is likely that any fixed a < log 2 will do . In view of the fact
that t 2 (p) = p-1, a> 1 is impossible .
Is it true that

(3) tz+i (n) = o ( t i (n)) ?

n=1 n=1

We have not even proved (3) for i = 2 .

Our final problem is rather different since it involves the divisors of two integers .
We say that m and n interlock, and we write m n, if every pair of divisors of n
are separated by a divisor of m, and conversely (with the exception that 1 and the
smallest prime factor of Inn obviously cannot be separated) . Thus 45 0 28, in view
of the pattern
45 15 9 5 3
28 14 7 4 2

We say that n is separable if there exists an m such that m A n, and we define A(x)
to be the number of separable n < x . We should like to prove the innocent-looking
16


482 P . Erdős and R . R . Hall [4]

relation A(x) = o(x), but have been unable to do so . In the opposite direction, we
have

THEOREM 4 . For a fixed c'> 0, and sufficiently large x, we have

A(x) > c' x/log log x.

We would like to mention two further conjectures concerning separable numbers .

Is it true that 2k is separable for almost all k? Notice that if k+1 is prime, this is
not possible for k,>4. Secondly, let N(k) be the product of the first 2k primes .
When can we have N(k) = mn, m An? k = 1, 2, 3, 4 are all possible, for k = 4,
m = 2 .5 .13 .19, n = 3 .7 .11 .17 . It seems likely that for large k this cannot happen .

3
In this section we prove our theorems and also state a few more problems .
To prove Theorem I put

nx =jjp, logx<p<(2--q)logx .

nx < x follows immediately from the prime number theorem . Put

y = [(log log x) 1-2, ]

and denote by Dl < D2 < . . . < D,, the divisors of n x having exactly y prime factors .
Clearly by the prime number theorem and a simple computation

(4) (log x)y < Dl < . . . < D, . < 2v(log x)v < (log x)v+l,

and

(5) r = (v(nx))
Y > (( I-2,q) log x ly/y!
J >exp(log x)2-3q)>
log log x

Now by (4) D i and Di+i are clearly consecutive divisors of n x and if (Di , D2+,) > I
then (Di, Di+l) > log x, so Di+i - Di > log x . Thus the number of indices i with
(Di, Di+i) > I is less than

2v(log x)v -i < 2r .

Hence finally from (5) (71 < 1E)

f(nx) > 2r > 8 exp (log log x) 2-3 V,

which completes the proof of Theorem 1 .


[51 Problems on the divisors of integers 483

At present we have no good upper bound for f(n) . It would be reasonable to

expect that for every e > 0
f(n) < exp ((log n)E) .

We are certain that the average off(n) is greater than any fixed power of log log n
but so far have not been able to prove it . Denote by A(u, v ; x) the number of
integers n < x for which a and v are consecutive divisors of n . Clearly,
x
(6) f(n) = E A(u, v ; x) .
n-1 1-<u<v<-x
(u, v)=1

The trouble is that it is very hard to estimate A(u, v ; x) . It can happen that
A(u, v ; x) = 0 because every n=- 0 mod ([u, v]) has a divisor u < d < v . We do not at
present know the number of these pairs ; it is not impossible that (6) is quite useless
for the estimation of En=, f(n) .
It is easy to see that for infinitely many n, f(n) = v(n) and it is not hard to show
that the density of the integers satisfying f(n) = v(n)-in fact f(n) < ( 1 +c) v(n)-
is 0 if c > 0 is sufficiently small . Perhaps f(n)lv(n) > oo if one disregards a sequence
of density 0 .
Assume that n is the product of k distinct prime factors . It is easy to see that
minf(n) = k, but we cannot at present determine maxf(n) and in fact we do not
even have a good estimation for it .
Denote by f(n), (f2 (n) = f(n)) the number of indices i for which (di+,1, d2+j) = 1
for every 0 < j1 <j2 < 1- 1 . Perhaps for every l > 2 the mean value off(n) is bounded .

PROOF OF THEOREM 2 . Let k be fixed, k > 2, and fix B, A < B < ell k. Put n == l .c.m .
(1, 2, . . ., y) . The prime number theorem implies y = (1 + a(l)) log n . Consider the
integers m < yB for which

(7) m-A-imodQ), i=1,2, . . .,k,

and Q runs through the primes and powers of primes ylk! < Q < yB. The number
of these m is by the well-known theorem of Mertens not less than (y > oo)

(8) yE(1-kEQ)=yB(1-k log log B-o(1))>eyB, E=e(B)

By a simple argument we obtain that if m satisfies (7) then 11k 1 (m+i) I n . Thus (8)
implies Theorem 2 .

PROOF OF THEOREM 3 . Let q be squarefree . We call the residue class h (mod q)

e-good if there exist integers r and d such that dl q, 1 < r < ed, (r, d) = 1, and
h=- -r-1(q/d)-1 (mod d) .


484 P. Erdbs and R . R . Hall [61

Let z < y < x, z--~ oo as x-> oo . We write the integer n < x in the form mq, where
the prime factors of q all lie in (z, y], and m has no prime factors in this range .
We assume from now on that q is squarefree . The number of integers n < x of the
form mq, q not squarefree is clearly O(x/z) ; the sum of their t 2's is O(x2lz) .
Now suppose that m is in an e-good residue class (mod q), and let r and d be the
pair of integers specified above . Let t = rmq/d. Then t + 1- 0 (mod d) and
t(t+ 1) - 0 (mod n) . Hence t2 (n) -< t<, en, and for these n, the sum of the t2 's does
not exceed Ex2.
For each q, we estimate the number of e-bad residue classes . Let p be a prime
factor of q . If h is e-bad, then h- -r-1(q/p)-1 (mode) where Ep < r <p . By the
Chinese remainder theorem, there are at most q(1- e)II ( q ) bad classes. Let us choose
y = x 1110 . Then the number of n < x such that n = mq and m is E-bad (mod q) is

X 1)
<< 11 1-i
q z<P-<Y p

and, summing over q, this is

log z E

<x H 1-£)<<x
z<P<-Y P (logy) ,
Hence

E t2(n) <x2 (1 +e+ ( 101og z)') I

nix z log x

and we set z = log x, E = 2(log log log x)/(log log x) . This gives the result stated .

PROOF OF THEOREM 4 . Let n < x be a squarefree number, P - (n) > (log x)A (P -(n)
is the least prime factor of n) . For any fixed íl, it is easily shown by Brun's method
that the number of such n is - e - 'Y x/(A log log x), where y is Euler's constant .
Write n = p, p2 . . . p,,, and for each prime p, let p' denote the next larger prime .
We consider whether m A n where m = pip' . . . pv . For each divisor phpz2 . .V paV of

n, m has the corresponding, larger, divisor p1 2a2 . . . p„012 and if this is always less
than the next larger divisor of n, we shall have m A n . A sufficient condition for
this is m/n < 8(n), where B denotes the smallest ratio, greater than 1, of two divisors,
of n . Choose a fixed K, 1'2 < K < 1 . It is well known that for p > p 0 (K), we have
p' < p + pK . Hence provided p 0(K) < (log x)A, as we assume, we have

m/n < 11 (1 +p K-1)

Ti ln

< (1-+(logx)A,-~)°(n)

< exp {2(log x)AK -a+ 1}



[71 Problems on the divisors of integers 485

since v(n) < 2 log x . We can choose a fixed A such that the right-hand side does not
exceed 1+(logx) _3 . It follows that if n cannot be interlocked, certainly we must
have B(n)<1+(logx)-3 . The ratio B(n) can be achieved with relatively prime
divisors of n, hence

card {n,<, x : 0(n)< ®} S x/dd'

d--x d<d'_<Bd

<<x(B-1) log x.

Setting B = 1+(logx) -3 , we obtain

A(x) >, (e -y + o (l)) x/ A log log x.

This gives the result stated . We remark that it is known (Erdős (1964), but no proof
has been published) that for every a > log 3 -1, there exists a positive a = s(a)
such that
card {n ,< x : B(n) < 1 + (log x) -"} << x(log x) -E .
We may also assume v(n) < 2 log log x. Thus we can obtain
(5e-y+o(1)) x
A(x) as x oo.
12(log 3 -1) log log x'

References
P . Erdős (1964), "On some applications of probability to analysis and number theory", J . London
Math . Soc . 39, 692-696 .

Mathematical Institute, Department of Mathematics

Hungarian Academy of Sciences University of York
Budapest England
Hungary