MAT HMATHEMATI KJ

A JOURNAL OF PURE AND APPLIED MATHEMATIC S

VOL . 3 . PART 1 . JUNE, 1956 . NO . 5.

MONOTONICITY OF PARTITION FUNCTION S

. T . BATEMAN and P .
PERDÖS

1 . Introduction . Let A be an arbitrary set of positive integers (finit e

or infinite) other than the empty set or the set consisting of the singl e
element unity* . Let p(n) = p A (n) denote the number of partitions o f
the integer n into parts taken from the set A, repetitions being allowed .
In other words, p(n) is the number of ways n can be expressed in the for m
n 1 a1 +n 2 a 2 + . . . , where a 1 . a 2 , . . . are the distinct elements of A an d
n l , n 2 , . . . are arbitrary non-negative integers . In this paper we shal l
prove that p(n) is a strictly increasing function of n for sufficiently large n
if and only if A has the following property (which we shall subsequentl y
call property P1 ) : A contains more than one element, and if we remove
any single element from A, the remaining elements have greatest commo n
divisor unity .
We shall obtain this result as a special case of the following more genera l
one . Let k be any integer and suppose we define p ck> (n) = pT(n) by the
formal power-series relatio n

x x
fk (X) = E p(k)(n)X" = (1—X) k n= o p(n)X1z
72 _o
= (1—X) k II (1—h") -1 . (1 )
as A

Thus pk>(n) is the k-th difference of p(n) if k > 0, p(n) itself if k = 0, and
the (—k)-th order summatory function of p(n) if k 0 . For k 0, we
shall prove in the sequel that pk)(n) is positive for all sufficiently large positive
integers n if and only if A has the following property, which we shall cal l
property P k : There are more than k elements in A, and if we remove an
arbitrary subset of k elements from A, the remaining elements have greates t
common divisor unity. When k = 0, this reduces to the well-known result
([3], [6]) that p°> (n) = p (n) is positive for all sufficiently large n if an d

* The inclusion of thee two trivial cases would complicate the statements and proof s
of some of our theorems and so, for the sake of simplicity, we agree to exclude them through -
out the paper .
[MATHEMATIKA, 3 (1956), 1–14]
2ERDÖS P . T . BATEMAN and P.

only if the elements of A have greatest common divisor unity . When

k = 1, it is the result stated in the preceding paragraph .
Although we are primarily interested in positive values of k, we shall
find it convenient to agree that when k is a negative integer any set A ha s
property Pk . Then the italicized assertion is trivially true for negative k .
Note that if A has property P k , it must actually contain at least k+ 2
elements . We remark also that A has property Pk if and only if th e
following assertion holds for every prime number p : there are at least
k± 1 elements of A which are not multiples of p . In particular, if A has
the property that for every prime number p there are infinitely man y
elements of A not divisible by p, then A has property Pk for arbitrary k.
The proof that property P k, is a necessary condition for p (k>(n) to be
positive for large n is a straightforward argument with power series 02) .
The sufficiency is proved in the following stages . First 03), we prove i t
for the case in which A is finite by using the partial fraction decompositio n
of the generating function fk (X) . Second 04), we prove that in any cas e
p(n) = o (p(-')(n)) as n increases . Third (5), we prove sufficiency for th e
case where A is infinite by using these two results . Actually, under th e
assumption of property P k we shall prove much more than the mere
positivity of p(k)(n) for large n and shall include the case of negative k in
our theorems for completeness 06) . However, all the arguments use d
are essentially elementary . We conclude the paper with a discussio n
(§7) of the relative orders of magnitude (as functions of n) of pm (n) an d
p(k+1)
(n)
Our attention was drawn to the problems discussed in this paper by
remarks of various authors ([4], [5], [6]) on the usefulness of knowing tha t
p(n) is an increasing function of n for large n . In particular, the applica-
tion of Ingham's Tauberian theorem for partitions becomes considerabl y
simpler in that situation . By our results this holds if and only if A has
property P1 . However, it follows from the corollary after our Theorem 6
(in the case k = 0) that actually property Po, namely, that the element s
of A have greatest common divisor unity, would be sufficient for the purpos e
of applying Ingham's theorem . This remark has been made previously
by Auluck and Haselgrove in [1] .
All our theorems refer to the behaviour of p( k )(n) for sufficiently large n .
The behaviour of p (k) (n) for small n can be rather erratic, since it depends
on the arithmetic properties of the smaller members of A rather than
on the arithmetic properties of A as a whole . In particular, if k 1 it i s
impossible for p (k) (n) to be positive for all non-negative n, since
p(k)(1)e 1-k .
For partitions into distinct parts the questions analogous to thos e
discussed in this paper are much more difficult . The reason for this is
that, in the case of distinct parts, these questions become trivial for finit e
sets and it is not possible to use the finite case to attack the general case,
 MONOTONICITY OF PARTITION FUNCTIONS . 3

as we have done in this paper . However, a rather broad sufficient conditio n

for monotonicity is given by Roth and Szekeres in [10] .
Throughout this paper small Latin letters denote (rational) integers .

2 . Necessity of property P k .

THEOREM 1 . If k _> 2 and if p (k'(n) is non-negative for all sufficientl y

large n, then A has property P k . If p°(n) is non-negative for all sufficientl y
large n, then either A has property P 1 or A contains the element 1 .

Proof . Suppose k > 0 and A is such that p(k) (n) is non-negative for
all sufficiently large n . Since the empty set and the set consisting of th e
single element 1 have been excluded from consideration, it follows fro m
(1) that fk (X) is not a polynomial . Thus p(k) (n) is positive for infinitel y
many n and so fk (X) + co as X approaches 1 from below . Thus A
must contain more than k elements, since otherwise fk (X) would be a rationa l
function which, when expressed in reduced form, has a denominator no t
divisible by 1--X . Suppose that B = {a l , a2 , ak} is an arbitrary
subset of A having exactly lc elements and let d be the greatest common
divisor of the elements of A—B . Then the left-hand side of the identit y
k
H (1 —Xa)-1=fk(X) II (1+X+X2+ . . .+Xa--1) (2 )
a€A—B 1 m=

is expressible as a power-series in X d . On the right-hand side of (2), th e

power-series for fk (X) has non-negative coefficients from some point on
and an infinite number of positive coefficients, while the coefficient of X
in the expansion of
k
II (1+A+X2+ . . .+Xa,. 1)
m= 1

is positive unless k = 1 = a1 . Thus, unless k = 1 and A contains th e

element 1, the right-hand side of (2) has infinitely many pairs of consecutiv e
coefficients both of which are positive, so that d must be 1 . Hence, unless
k = 1 and A contains 1, A must have property P k . Accordingly Theorem 1
is proved .

THEOREM 2 . For arbitrary k, if p (k '(n) is positive for all sufficientl y

large n, then A has property Pk .
Proof. The theorem is vacuous if k < 0 and follows immediatel y
from Theorem 1 either if k > 1 or if k = 1 and A does not contain th e
element 1 . If k = 1 and A contains 1 ,

H (1—X a ) — 1
f1 (X) =aCA,ar 1

and thus fl (X) is expressible as a power series in X d, where d is the greates t

common divisor of the elements of A other than 1 ; hence, if p(1'(n) > 0
 4ERDÖS P . T . BATEMAN and P.

for all sufficiently large n, d = 1 and so A has property P 1 . If k = 0,

fo(X) = II (1—X9— 1
ae A

and thus fo(X) is expressible as a power-series in Xd , where d is the greatest

common divisor of all the elements of A ; hence, if p`0 )(n) is positive for al l
sufficiently large n., d = 1 and A has property Po .

3. Sufficiency of condition P k when A is finite .

LEMMA 1 . ,Suppose A has exactly r elements a 1 , . . ., a., and suppos e
k < r. Then if n > 0
p(r>(n) = g(n)+ 0 ( 0-1 ) ,
where g(n) is a polynomial in ii of degree r—k—1 with highest coefficient
[(r—k— 1)! a.1 a 2 . . . ar ] -1, and where q is the largest number of elements of A
which have a common divisor greater than 1 .
Proof. The proof is based on the methods of Cayley, Glaisher, an d
Sylvester (cf . [2] and [7]) . From (1) we see that l/fk (X) is a polynomial
whose factorization into linear factors i s
1 r

A(x) — ( 1 — y) r-k II II (1—e2n~lla„ X) , (3)

1 1= 1

Thus the terms in 1—X in the decomposition of f,. (X) into partial fractions
have the form
1a1k }- (1 2X) 2 . . . (1 a I)r—k , (4)

where ar_1, _ (a l a 2 . . . a r )- 1 . If g(n) is the coefficient of X'v in the

power series expansion of (4), we have

g(n) _ r' n+h—1 '

h=1 h ~
which is a polynomial in n of degree r—k—1 and highest coefficien t
[(r—k— I)! a1 a 2 . . . a,.] -1. If d >1 and is a primitive d-th root o f
unity, the multiplicity of the factor 1—SX in the factorization (3) is equa l
to the number q d of multiples of d among a 1 , a 2 , . . ., ar . Since by definitio n
q is the largest value which qd can have for any positive integer d greater
than 1, the terms in 1— SX in the decomposition of fk (X) into partial
fractions have the form

P X+ P2X~2 . . . { (1 ~ X)4' (5 )
1 (1
where of course some of the ,8 's may be zero. The coefficient of (a)" i n
the power series expansion of (5) is a polynomial in n of degree at most
q— 1 . Summing over all possible we get the result of the lemma .
 MONOTONICITY OF PARTITION FUNCTIONS . 5

THEOREM 3 . Suppose k is arbitrary, A has exactly r element s

al, a2 a, ., and A has property P k . Then as n increases p (k)(n)- +o0
in such a 'way that
p(k>(yZ)
(i) (r—k -1)?ala2 . . .a,,-I-/J(nr-k-2) ,
p(k+1>(n) —1) pk>(n
OII p(k)(n) =— 1— p(k>(n) '1
=O~
— n~ .

Proof. By the assumptions of the theorem and the definition o f

property Pk, we have k r. Thus Lemma 1 is applicable . Further, the
number q in Lemma 1 does not exceed r—k—1 . Hence (i) follows immedi-
ately from Lemma 1 and (ii) follows from (i) .
H A is finite, conclusion (i) of Theorem 3 shows that property Pk i s
a sufficient condition for p (k )(n) to be positive for all large n. The specia l
case k = 0 of Theorem 3 is well known (cf. [8]) .

4. Proof that p(n) = o (p ( - 1)(n)) . Conclusion (ii) of Theorem 3 shows

in particular that when A is finite p (n) = o (p (-1)(n)} as n increases . We
now show that this relation is also true when A is infinite .

THEOREM 4 . Suppose A is infinite . Then as n increases

(i) p! 11 (n) n- c --->+ co for any fixed c ,

ra= 1'n 1
(ii) 1i-1)(n ) 1—~~(p(11 (n) ) - 0 .

Proof. In proving assertion (i) we may assume c > 0 . Let B be a

finite subset of A having at least c± 1 elements . Conclusion (i) of
Theorem 3 shows that p 1> (n) + oo as n increases . Bu t

p(V> (n ) A-1)(n )
and so (i) follows .
It remains to prove assertion (ii) . Suppose that n > 0 and that
n = nl al+ n2 (1 2 + . . . + nq a q is a partition of n into parts al, a 2 , . . ., a q
taken from A, where n1, n2 , . . ., ng are positive integers . From this w e
construct a partition of each of the q integers n—al, n—a 2 , . . ., n—a q in
the following way :

~L — al = ( nl — 1)al + n2 a 2 -r . . .-rn g ag ,

n — a 2 = n l a'1+ (n2— 1) a 2 +n3 a3-}- . . . -I- ng ag ,

n— aq = n l al+ . . . + ng -1 ag -1 + ( nq —1 ) a g .
6 P . T . B ATE MAN and P . ERDÖ S

In this construction no two distinct partitions of n give rise to the same

partition of some integer less than n. Let P,2 (n) denote the number of
partitions of n into parts taken from A in which the number of distinct
elements of A which actually occur is exactly q . Then the above con-
struction shows that
n- 1
E q Pq (n) E p(m) . (6 )
g=1 m= 0

Now Pq (n) is the coefficient of Xn in the expansion o f

Xal Xaa

1—li a r ...
l—XaQ '
(7)

where the summation is taken over all subsets {al , . . ., aq} of A containing
exactly q elements . Hence P0 (n) does not exceed the coefficient of X n in

E(X a 1+X2a 1+ . . .) . . . (Xa,+X2aq+ . . .) ,

where the summation is taken over all subsets {a 1 , . . ., ag } of the set

{1, 2, . . ., n} which contain exactly q elements . Since there are

(n
q
such subsets and since the coefficient of Xn in
(Xal+X2a1+ . . .) . . . (X a ,+ 7i'2ai+ . . . )

does not exceed that in

(X+X2+ . . .)q = ( `
q
— a'
(m—l) Xm
—X i) m = qlq— J >

we have
Pq(n) < ( nq fI n`l
q— 1
\ <n2Q~ 1 . (8 )

Now if t is any fixed positive integer, we have by (6) and (8 )

p (-1) (n) = E p(nz )

m—o
n
p ( n )+ g Pq(n) = ( t +1)p (n)+ E ( q— t) Pg(n)
q =1 q =1
e-1
(t+ 1)p(n) —(t—1) E Pg (n) ~ (t+1)p(n)— (t —1)2n2c-3 .
q= 1
Hence
p (n) 1 (t—1)2 n21- 3
(9 )
p(-1) (n)t+1 t+1 p(-1) (n) •
By conclusion (i) the second term on the right-hand side of (9) has limi t
zero as n increases . Therefor e

p( n ) 1
11Tn
p( ( t +l
Since t can be chosen arbitrarily large, conclusion (ii) follows .
 MONOTONICITY OF PARTITION FUNCTIONS . 7

5 . Sufficiency of condition P k when A is infinite .

LEMMA 2 . For any k, an infinite set of positive integers has propert y

Pk if and only if some finite subset has property P k .

Proof. The assertion is trivial if k 0, so suppose k O . If some finit e

subset of an infinite set A has property Pk , then clearly A has also, sinc e
enlarging a set with property P k cannot destroy that property . Suppos e
now that A is an infinite set with property P k . Le t

A O = { a 01 , a 02 , . . ., a O, k+l}
be some subset of A containing exactly k+ l elements . Suppose th e
a a a
prime divisors of the product ol o2 • • • o, k+i are p i , p 2 , . . ., p b. By
the definition of property Pk there exists, for i = 1, 2, . . ., b, a subset

A i = {ail, ai2, . . ., ai, k+i }

of A containing exactly k+l elements none of which is divisible by p i .
If p is a prime other than p 1 , . . ., Pb' clearly no element of A o is divisible
by p . Then the union B of the (not necessarily disjoint) subset s
A o , A 1 , . . ., A b of A is a finite subset of A with property Pk . For give n
any prime number, we can find at least k+ l elements of B which it doe s
not divide .

THEOREM 5 . Suppose k is arbitrary, A is infinite, and A has property P k .

Then as n increase s
(i) p(k) (n) n-c —* + co for any fixed c ,
p(k+l)(n) — p (k) (n—1 )
(ii) p(k)(n) — 1— p(k)(n) . —>- 0

Proof. The assertions of the theorem have been proved for k = - 1

(Theorem 4) and they follow immediately for k -1 by summation .
So suppose lc 0 and A is an infinite set with property P k . By Lemma 2
there is a finite subset A l of A which has property Pk . If k = 0 we may
assume that A l contains at least two elements . Let us put A 2 A—A 1
and write

PT (n ) = p (k) (n), pál( n ) = p (i.k) (n), PÁ ( )n P2 (n), p(21)(n) =P(2-1)(n) .

2
We shall use the identity
n
p(k)
(n) _ m=Eo pi)(n— m )p2(m) , (10)

which is an immediate consequence of (1) .

Let t be an arbitrary positive integer . By Theorem 3

P (k) ( 7z )
1—irn +°°
>oo 1+i :W +1 )(n)I
 S

and so there exist positive integers g and h such tha t

l r_ g for all n ,
p(k) ( n )
1 +I pik+1) ( n )I ' t + 1 for n > h.

Let s= max {1+Iplk+1)(n)I} . (12 )

0,n<h— 1

By Theorem 4 there is a positive integer hl such that

p2(n) 1 (13 )
p21)(n) (g+t±1)h s

for n hl . Then if n h+h 1 —1 we obtain from (10)-(13 )

n— h
p(k)(n)(t-}-1) E {1--P1i+''(n — m) I} p 2( m )
m=0
n
—g E { 1 +1 nL )I} p 2( m )
m = n —h + l
n n
(t-r1) E {1 + 1 1~i +l> (gg—rrz ) } p 2(rrz )— (g+ t + 1 ) s E p 2( m )
rn = 0 m= n—h+ l

n
(t +l ) E
m=0
{l+l p(ik+l) ( n—m )I} p 2( m)

— (g+ t + 1 ) s ~ n
E
m=n—h+ l
~
p 2(~n )Ip 2 1} ( m )} p 1) ( n )

t +1— (g+t +1)s E 1

p2(m)lp2 >(rn )
=n —h +1
n
x E {l + Ip lk+l) ( n —m) I }p2( m )
m= o
n
t E { 1 +I p ik} 1) (n—m ) I} p 2( m )
ne = 0
n
t E p 2( m ) + t E p(,k+l)(n—m)p2(nZ )
m =o m= 0

= tp;1) (n)+t p(k+l)(n )

Now t can be chosen arbitrarily large . Therefore lim p(k)(n)/p(2- - 1) (n) = + co

n—> m
and thus, since A 2 is infinite, assertion (i) of the present theorem follows fro m
assertion (i) of Theorem 4 . Also lim p(k+l>(n)Ip(k)(n) = 0 and assertion
(ii) is proved .
Theorem 5 shows in particular that, when A is infinite, property Pk
is a sufficient condition for p(k) (n) to be positive for all large n . Thus
MONOTONICITY OF PARTITION FUNCTIONS . 9

property Pk is a sufficient condition for the eventual positivity of p(k) (n )

in any case .

COROLLARY . For arbitrary k, if A has property P k., then as n increase s

p(k~1)(n)—1— p(k)kn—1)—>
p(k)(n) —> -F- co . 0.
p (n) p (n )

6 . Further implications of property P k . Although Pk is a necessar y

condition for p(k) (n) to be positive for all large n, we have already seen
(Theorems 3 and 5) that if we assume property Pk we can actually asser t
much more about p(k) (n) than the mere fact that it is positive for large n .
In this section we go further in this direction .

THEOREM 6 . Suppose k is arbitrary and A has property P k . Then

there is a positive integer b such tha t

p(k)( .yyl) p(k)(n) if 7n—b>n>0 .

Proof . If k < -1, the result stated is trivially true with b = 1, since
p(k+l)(n) = p(k)(n)—p`k)(n—1) is positive for all non-negative n . If k =— 1
and A is non-empty, we can take b as any element of A . For then i f
m—b > n > 0, there is an integer j such that n jb m, so tha t
m n
p(-I)(nt) = E p (h) z p(h) =p(jb) +p(-1)(n) > p(-1) (n) .
h=0 h= 0

When k > 0 we first settle the case in which A is finite . Suppose

A = {a 1 , a 2 , . . ., ar } and A has property Pk . Then by Theorem 3 there i s
a positive integer c such tha t
nr-k- 1
p(k) (n) — cnr-k -2+ 1
(r — k —1)! a 1 a 2 . . . ar

for all non-negative n . Therefore if m > n 0 we have

7m -k-1 nr-k -1
p(k)(m)—p(k)(n) CnLr-k -2 —cn r- k -`-' — 2
(r—lc— 1)! a1a2 . . . a,

(m—n) 7n r-k- 2
(2c-}-2) 7n r -k- 2 .
(r—k—1)! a 1 a2 . . . ar—

The last expression is positive if 7n.—n b, wher e

b = (2c+3)(r—k—1)! ala, . . . ar + 1 .

Thus the assertion of the theorem is proved if A is finite .

Finally, suppose k 0, A is infinite, and A has property P k . As in
the proof of Theorem 5 let A l be a finite subset of A which has propert y
P k and contains at least two elements, put A 2 = A—A 1 , and write
10 P . T. BATEMAN and P . ERD Ő S

pki ) (n) = p(k)(n), pÁi(n) =p2)(n), PA z (n ) =p 2( n), p;,z ) (n) = pl) (n). By
Theorem 3 there exist non-negative integers g and h such that plk) (n) — g
for all n and pík) (n) 1 for n h . By Theorem 4 there is a positive intege r
hl such that p 2 (n)/p 1)(n) 1/(2gh+2h) for n il l. By . the previou s
paragraph there is a positive integer b such that b > h+h l —1 and
p(,k)(m) > plk) (n) for m—b > n O.
Hence if m—b > n > 0 we have
n m
p(k)(m)—p(k)(n) = E {plk)(m— l) —plk)(n—l)}pz(l)+ E p(,k)(m—l)pz(l )
1=0 1=n+ 1

E p 2( 1 )+ E 1)2( l ) —g E 1)2(1 )
1=0 1=n+1 I=m-h+ l

p 2( 1 ) )
—
— p (2 1)/( 7n ) — (g+ 1)
1-m -h+l p2 (m) f
1 1)

p21(l)
>
p1)( 7YL ) 1 1>
/ — (g+ 1 ) E ) } 1)
(-1 ) (m) > o.
1 1=m-h+l p'l (l )

Thus Theorem 6 is proved.

COROLLARY . Suppose lc is arbitrary and A has property P k . If h is

any fixed positive integer, the n
p*-1) (n+h) — p(k-1) (n )
h = {1+0(1)} p(k) ( n )

as n increases . If h is a fixed integer not less than the integer b of Theorem 6,

then
p(k-1) (n h)—p(k-1) ( n )
h
is a strictly increasing function of n for non-negative n .

Proof.The first assertion follows from conclusion (ii) of Theorem 3

and conclusion (ii) of Theorem 5 :
n+h
p(k-l) (n+h) —p(k -1)(n) = E p( k)(m) = h{1+o(1) }p(x)(n) .
m=n+ 1

The second assertion follows from Theorem 6 :

J1p(k-1) (n+h) —p (k-l) (n)}— {p(k-l) (n—1 +h) —p(k-l) (n— 1) }
n+h n-1 -h
= E p(k) (m)— E p(k )(m) =p(k)(n +h) —p(k)(n) > 0,
nt=n rl m =n

provided h > the integer b of Theorem 6 and n > 0 .

 MONOTONICITY OF PARTITION FUNCTIONS . 11

THEOREM 7 . Suppose k is arbitrary, A is infinite, A has property Pk ,

and OKa<1 . Then
hm 1 max p( k )(m) = O.
n-,-,, p(k)(n) 0<m<x n
Proof. Let A l be a finite subset of A having property Pk and con-
taining at least k+2 elements . Then A l has q±k±1 elements, where q
is some positive integer . By conclusion (i) of Theorem 3 there is a positiv e
integer s such that
splk>(n )
lim =1.
n–>-x‘ nq

Hence there are positive integers t and u such that

2n4 < splk>(n) < 2n q for n

and plk > (n) .<u for n < t.

Hence if n max{(us) lq/(a—a 2 ), t/(1—a)} and t m <an, we have

m
p(k)(m) = E pi )(m— l )p2(l )
t=o
m–t m
• E 2s -1(m—l)gp2(l)+ E up2( l )
t=o 1=m-1 4- i

m-t m
< E 2s -1 (an—al) g p 2 (l)-{- E s- l a g (n—an) g pa(l )
l=0 l=m–t + 1

m m
▪ 2aq E s-1 (n—l) g p 2 (l) < 4aq E plk>(n—l) p 2 (l) . (14)
t=o t= o

Suppose now that k < O . Then p(ik)(n—l) 0 for any 1 and so if

0 < m < an we have
m n
pl)(n—l)p2(l) E p lk) ( n—l ) p 2( l ) = p(k)(n) . (15)
t=o z=o

Combining (14) and (15), we find that

p(k)(m) <4c' p( k)(n), ( 16 )

provided t .<m < an and n is sufficiently large . By Theorem 5 th e

inequality (16) also holds if 0 < m < t—1 and n is sufficiently large .
Hence
lim (ki max p (k)(m) < 4aq .
n-->co P ( n ) 0<m< a n

Since we can make q as large as we please by arbitrarily enlarging A 1 , th e

assertion of the theorem is proved when k O .
 12 PERDÖS
. T . BATEMAN and P .

To prove the assertion of the theorem when k > 0 it is again sufficient

to show that (15) holds if 0 < na < an and n is sufficiently large . By
Theorem 3 there are non-negative integers ti and l such that plk>(n) 1 g

for n t l and pi >(n) —g1 for all n . By Theorem 4 there is a positive

integer t 2 such that p2(n)!p2-'>(n) < 1/(3t 1 g1 ±3t i ) for n t 2 . By th e
result of the preceding paragraph there is a positive integer t3 , ti —t2 — 1
such that p2'>(m) fp2 1>(n) provided 0 < m r.n and n 13 . Accord-
ingly, if n max {t 3 , (ti ±1),t (1— x)} and 0 <m < an we hav e

m- n

P'k>(n)— E p (ik)( n — l )P2( l ) = E pi >(n—l)P2(l )

l=0 1=m+ 1

n –tl n.
E p2(1)—gl E P2( l )
1=m 1 1=n–t i+1

= p2li (n) p2'>(m }— (91+ 1 ) P2( 1 )

l = n –t l + l

i P2 ( 1>
( 9a ) f 1—
Pá l>(m)

PV-'(n )
— (g1 1 1) "11
E 2 (1)~
P 1) (l) 1
n.–t i +1

p2'>(n) - — ; - -(gl - r-1)t13f1~ O.

+3 t1 I 32~ 2 ~ >(n )

This completes the proof.

7 . The order of magnitude of pck+'>(n)/p(k>(n) . Suppose A has propert y

P k and h is the smallest positive integer such that p` k (n) 0 for n h . >

(Such an integer h exists if and only if A has property Pl .) Then the

quantity-
ptk=1>(n) — pck>( n —1 )
pck>(n) =
pck>(n, ) 1 pck>( n )

is defined for n?-h . If A is finite we know that p (k >(n) = 0(1/n) [con-

clusion (ii) of Theorem 3] . If A is infinite we have proved that p (k )(n )
tends to zero as n increases [conclusion (ii) of Theorem 5] without gettin g
a more explicit estimate of its order of magnitude for large n . In fact w e
have been unable to obtain any result in this direction . However, we fee l
that there is reason to believe that the following assertion may be true .

CONJECTURE . If k is arbitrary and A has property Pk , then

p(k)(n) = 0 ( l / n'2 ) •

It is easy to see that the assertion of the conjecture, if true, is bes t

possible . For if A is the set of all positive integers, then as n increase s
p(k> (n) = {1-áo(1)}7r/(6n) 1/2 by Rademacher ' s exact formula for p(n) in
 MONOTONICITY OF PARTITION FUNCTIONS . 13

this case [9] . On the other hand, the following theorem shows that p (k) (n )
cannot be 0(1/n) for any infinite set A, that is, conclusion (ii) of Theorem 3
is definitely false for infinite sets .

THEOREM 8 . Suppose k is arbitrary and A is infinite . If A has propert y

P k , then np (k>(n) is unbounded above for large n . If A has property Pk+l ,

then as n increases np (k) (n) + oo .

Proof. Suppose A is infinite and has property Pk . As above let h b e

the smallest integer such that p(k) (n) > 0 for n h . Suppose np(k) (n) wer e
bounded above by the integer g for n h . Let c = max (g, h) . The n
if n c
p(k)(n) — n p(k)(9n) m m n
)n n
II (
p( k) ( c ) m-,+1
p(k) (m —1) ~ IZ
m =c +1 m— g ~ m~
=~ +1 m—c — \ c)
<
nc .

But this contradicts conclusion (i) of Theorem 5 . Thus the first assertio n
of the present theorem is proved .
Suppose now that A is infinite and has property Pk+l . By Theorem 6
there is a positive integer b such that p( k + l)(n) > p(k+l) (m) if n—b m O .
Since the ratios

p(k +1) (n _
1) p (k+l) (n — 2) p(k+l)(n_ b+ 1 )
p(k+l)(n) ' p(k+l)(n) p(k+l)(n )

all have the limit 1 as n increases, there is a positive integer h l b such

that all these ratios are less than 2 if n > hl . Thus if n h l and m n
we have p( k +1)(m) 2p(k+l)(n) . Now let E be an arbitrary positive number .
By Theorem 7
lim 1 max p (k + l) m) = 0
n-sao p
(k+l)
( n ) 0<m<(1-,) n (

and hence there is a positive integer h 2 such that

max
0<m<(1-e) n
p( k +l)(m) s Ep(k +
l) (n )

for any n h2. Hence if n max (hl , h 2 , 1/E) we have

[(1 E)nl n
p(k) ( n ) p(k+l)(m)+ p(k+l) ( m )
m=0 m = [(1 -On] -f- 1

[(1- On]
Ep(k+1)(n)+ Fi 2p(k+l) (n )
m=0 m=[(1-e)n}+ 1

s nEp(k+1)(n)+(En+1) 2p(k + l) (n) s 5enp (k + l) (n) .

Since E is arbitrary, this proves the latter assertion of Theorem 8 .
14 MONOTONICITY OF PARTITION FUNCTIONS .

References .
1 . F . C . Auluck and C . B . Haselgrove, " On Ingham's Tauberian theorem for partitions " ,
Proc . Cambridge Phil . Soc ., 48 (1952), 566-570 .
2 . P . Bachmann, Niedere Zahlentheorie : Zweiter Teil, Additive Zahlentheorie, Ch . 3
(Leipzig, Teubner, 1910) .
3. A . Brauer, " On a problem of partitions ", American J. of Math ., 64 (1942), 299-312 .
4 . N. A . Brigham, " A general asymptotic formula for partition functions ", Proc .
American Math . Soc ., 1 (1950), 182-191, especially p . 183 .
5 . A . E . Ingham, " A Tauberian theorem for partitions ", Ann . of Math . (2), 42 (1941) ,
1075-1090, especially pp . 1084-1086 .
6 . K . Knopp, " Asymptotische Formeln der additiven Zahlentheorie ", Schriften der
Königsberger gelehrten Gesellschaft (Naturwissenschaftliche Klasse), 2 (1925), 45-74 ,
especially pp . 60-63 .
7. E . Netto, Lehrbuch der Combinatorik, Ch . 6 (Leipzig, Teubner, 1901 and 1927) .
8 . G . Pólya and G. Szegö, Aufgaben und Lehrsätze aus der Analysis, Vol. 1, Part I ,
Problem 27 (Berlin, Springer, 1925) .
9 . H . Rademacher, " On the expansion of the partition function in a series ", Ann . of
Math. (2), 44 (1943), 416-422 .
10 . K . F . Roth and G . Szekeres, " Some asymptotic formulae in the theory of partitions " ,
Quart . J. of Math . (2), 5 (1954), 241-259 .

University of Illinois, Urbana, Illinois ;

University of Notre Dame, Notre Dame, Indiana .
(Received 5th August, 1955 . )

Printed by C . F . Hodgson Sr Son, Ltd ., Pakenham Street, London, W.C.1.