1

2
3

47

Journal of Integer Sequences, Vol. 14 (2011),

Article 11.5.8

23 11

Integer Sequences, Functions of Slow

Increase,
and the Bell Numbers
Rafael Jakimczuk
Division Matematica
Universidad Nacional de Lujan
Buenos Aires
Argentina
[email protected]
In memory of my sister Fedra Marina Jakimczuk (19702010)
Abstract
In this article we first prove a general theorem on integer sequences An such that
the following asymptotic formula holds,
An
Cn f (n) ,
An1
where f (x) is a function of slow increase, C > 0, > 0 and is a real number.
We also obtain some results on the Bell numbers Bn using well-known formulae.
We compare the Bell numbers with an (a > 0) and (n!)h (0 < h 1).
Finally, applying the general statements proved in the article we obtain the formula
1

Bn+1 e (Bn )1+ n .

Integer Sequences. A General Theorem.

We shall need the following well-known lemmas [12, pp. 332, 294].
Lemma 1. If sn is a sequence of positive numbers with limit s then the sequence

n
s1 s2 s n
has also limit s.
1

Lemma 2. The following limit holds,

lim

n!
1
= .
n
e

We recall the definition of function of slow increase [7, Definition 1 ].

Definition 3. Let f (x) be a function defined on interval [a, ) such that f (x) > 0,
limx f (x) = and with continuous derivative f (x) > 0 . The function f (x) is of
slow increase if and only if the following condition holds
lim

f (x)

x f (x)
x

xf (x)
= 0.
x f (x)

= lim

(1)

Typical functions of slow increase are f (x) = log x, f (x) = log2 x and f (x) = log log x.
Lemma 4. If f (x) is a function of slow increase on the interval [b, ) then the following
asymptotic formula holds
p
n
f (b)f (b + 1) f (n) f (n),
(2)
where b is a positive integer.

Proof. Note that we always can suppose that f (x) > 1 on the interval [b, ).
Since log f (x) is increasing and positive in the interval [b, ) we find that
Z n
n
n
X
X
log f (i) =
log f (x) dx + O(log f (n)) = n log f (n)
(1 log f (i)) =
i=b

i=b

xf (x)
dx + O(log f (n)).
f (x)

(3)

Note that the second equation in (3) is a sum of areas of rectangles of height log f (i) and
base 1. Consequently the third equation in (3) is immediate.
LHopitals rule gives (see (1))
f (x)
log f (x)
= lim
= 0.
x f (x)
x
x
lim

Therefore
O(log f (n)) = o(n).
If the integral

Rx
b

tf (t)
f (t)

(4)

dt converges we obtain
lim

On the other hand, if the integral

(1) that

Rx
b

Rx
b

tf (t)
f (t)

lim

Rx
b

tf (t)
f (t)

dt

= 0.

dt diverges we obtain from LHopitals rule and

tf (t)
f (t)

x
2

dt

= 0.

Therefore

xf (x)
dx = o(n).
f (x)

(5)

log f (i) = n log f (n) + o(n).

(6)

Equations (3), (4) and (5) give

n
X
i=b

That is,

1X
log f (i) = log f (n) + o(1).
n i=b

That is (2).

Theorem 5. Let An (n 0) be a sequence of positive numbers (in particular integers) such

that
An
Cn f (n) ,
(7)
An1
where f (x) is a function of slow increase on the interval [b, ), C > 0, > 0 and is a
real number. If 1 n < b we put f (n) = 1.
The following formulae hold,
q
n
n A 1 A2
AAn1
A 0 A1
1
,
(8)
An
e
An1
1
1+ n

An+1 e An

(9)

log An = n log n + n log f (n) + ( + log C)n + o(n),

(10)

log An n log n,

(11)

Cn f (n)
An =
e(+o(1))n

n

(12)

Proof. We have (see (7))

An
An1
Cn f (n)

1.

Consequently (13) and Lemma 1 give

qQ
v
u n
n
Ak
n
Ak
uY
k=1 Ak1
Ak1
n
t
= p
1.
Qn
f (k)
n
f (k)
Ck
Ck
k=1
k=1
3

(13)

That is
p
n
An

s
n

A1 A2
An

A0 A1
An1

v
u n
uY
n
Ck f (k) .
t

(14)

k=1

Lemma 2 and Lemma 4 give

v
u n

   p
uY
n
n
n
n

t
n!
f (1)f (2) f (n) C f (n) .
Ck f (k) = C
e
k=1

Equations (14), (15) and (7) give

s
p
A1 A2
An
n
1 An
n
An n

C f (n)
.
A0 A1
An1
e
e An1

(15)

(16)

Equation (16) gives (8). Equation (16) and [7, Theorem 8] give
1

n1
Ann An1
.

(17)

Equations (17) and (16) give

1+

An e An1n1 .
That is (9). Equation (16) gives


1
n

log An = log C f (n) + o(1).

n
e
That is (10). Equation (10) gives (11), since (from LHopitals rule and (1))
log f (x)
xf (x)
lim
= lim
= 0.
x log x
x f (x)
Finally, equation (12) is an immediate consequence of equation (10).
Remark 6. Note that: (i) The following limit holds Cn f (n) (see(7)).
If 0 the proof is trivial. If < 0 use [7, Theorem 2 and Theorem 4]. Consequently
we have
An+1
lim
= .
n An
(ii) This last limit implies the following formula (An+1 An ) An+1 .
(iii) Equation (7) implies the more general relation,
n
Y
An An1
Anm+1
An
=

Cm
k f (k) .
Anm
An1 An2
Anm
k=nm+1

Introduction to Bell Numbers.

The n-th Bell number Bn is the number of partitions of a set of n elements in disjoint subsets.
The Bell numbers satisfy the following recurrence relation [1, p. 216].
B0 = 1,

Bn+1 =

n  
X
n
k=0

Bk .

(18)

The first Bell numbers are B0 = 1, B1 = 1, B2 = 2, B3 = 5, B4 = 15, B5 = 52, B6 = 203,

B7 = 877, B8 = 4140, B9 = 21147, B10 = 115975.
N. G. de Bruijn [6, pp. 102109] proved the following asymptotic formula,
log Bn = n log n n log log n n + o(n).

(19)

L. Lovasz [10, Ex. 9(b), p. 17] proved the following asymptotic formula
1

Bn n 2 ((n))n+ 2 e(n)n1 ,
where
(n) =

n
.
W (n)

(20)
(21)

The function x = W (y) is the inverse function of y = xex on the interval (0, ). The
function x = W (y) is called Lambert W-function.
The following results are well-known [5]. We establish these results in the next lemma.
For sake of completeness we give a proof of the lemma.
Lemma 7. The function x = W (y) is positive, strictly increasing on the interval (0, ) and
limy W (y) = .
The following formulae hold.
W (y) log y,
(22)
W (y) =

W (y)
.
y(1 + W (y))

(23)

Proof. The first statement is trivial.

We have (definition of x = W (y))
y = W (y)eW (y) .
Consequently
1 = W (y)eW (y) + W (y)eW (y) W (y).
That is
W (y) =

W (y)
1
=
.
eW (y) (1 + W (y))
y(1 + W (y))
5

(24)

On the other hand (24) gives

log y = log W (y) + W (y).
Therefore

W (y)
log W (y)
=1
.
log y
log y

(25)

Also (from LHopitals rule and (23))

yW (y)
1
log W (y)
= lim
= lim
= 0.
y W (y)
y 1 + W (y)
y
log y
lim

(26)

Finally, equations (25) and (26) give (22).

Remark 8. Note that the Lambert W-function W (y) is a function of slow increase since (see
(1) and (23))
1
yW (y)
= lim
= 0.
lim
y 1 + W (y)
y W (y)

Some Results on Bell Numbers.

The limit

Bn
= 0.
n!
is well-known [9, p. 64]. In the following Theorem we include it for sake of completeness.
lim

Theorem 9. The following limits hold.

Bn
=
n an
lim

lim

Bn
=
(n!)h

(a > 0),

(0 < h < 1),

Bn
= 0.
n n!
lim

Proof. Equation (19) gives

 
Bn
= log Bn n log a = n log n n log a + o(n log n).
log
an
Therefore
lim log

and consequently

Bn
an

= ,

Bn
= .
n an
lim

(27)

(28)

That is (27).
The well-known Stirling formula is
n!

nn n
2 n .
e

Therefore
log n! = n log n n +

1
log n + log 2 + o(1) = n log n n + o(n),
2

and
log(n!)h = hn log n hn + o(n).

Consequently (see (19))



Bn
= (1 h)n log n n log log n n + hn + o(n).
log
(n!)h

(29)

If 0 < h < 1 equation (29) gives

lim log

That is,

Bn
(n!)h

= .

Bn
= .
n (n!)h
lim

On the other hand if h = 1 equation (29) gives

 
Bn
= .
lim log
n
n!
That is,
lim

Bn
= 0.
n!

M. Klazar [8, Proposition 2.6] and D. E. Knuth [9, eq. (30), p. 69] proved the following
asymptotic formula
n
Bn+1

.
Bn
log n
This formula is derived as a consequence of the asymptotic formula obtained in the classical
paper [11].
In the following Theorem we derive this formula from the Lovaszs formula (20). We also
use the well-known properties of the Lambert W-function established in Lemma 7.
Theorem 10. The following asymptotic formula holds,
n
Bn+1

.
Bn
log n
7

(30)

Proof. Substituting (21) into (20) we obtain

Bn

nn

n+ 21

W (n)

e W (n) n1 .

Consequently
n+1
Bn+1

Bn
W (n + 1)

W (n)
W (n + 1)

W (n)
W (n + 1)

n

n+1
n
e( W (n+1) W (n) ) .

(31)

Equation (22) gives

Equations (32) and (22) give

W (n + 1) W (n).

(32)

n+1
n

.
W (n + 1)
log n

(33)

(34)

Equation (32) gives

Let us consider the function

y
.
W (y)

W (n)
1.
W (n + 1)

The derivative of

y
W (y)

is (see (23))

W (y)
W (y) 1+W
W (y) yW (y)
1
(y)
.
=
=
2
2
W (y)
W (y)
1 + W (y)

Consequently we have (Lagranges Theorem)

n
1
n+1

=
0,
W (n + 1) W (n)
1 + W (n + (n))
where 0 < (n) < 1.
We have


W (n + 1)
W (n)

n

= exp (n (log W (n + 1) log W (n))) .

(35)

(36)

Let us consider the function log W (y). The derivative of log W (y) is (see (23))
1
W (y)
=
.
W (y)
y(1 + W (y))

Consequently we have (Lagranges Theorem)

n (log W (n + 1) log W (n)) = n
=

1
(1 +

(n)
)(1
n

+ W (n + (n)))

where 0 < (n) < 1.

8

1
(n + (n))(1 + W (n + (n)))

0,

(37)

Equations (36) and (37) give



W (n)
W (n + 1)

n

1.

(38)

Finally, equations (31), (33), (34), (35) and (38) give (30).
The asymptotic formula (30) implies that the Bell numbers satisfy condition (7). In this
case C = 1, = 1, = 1 and f (n) = log n. Consequently we have the following Corollary.
Corollary 11. The following formulae hold,
lim

q
n

Bn+1
= ,
Bn

(Bn+1 Bn ) Bn+1 ,
B1 B2 B3
B0 B1 B2

n
BBn1

Bn
Bn1

Bn
Bn
Bn1

1
,
e

Bn+1 e (Bn )1+ n .

Proof. It is an immediate consequence of Theorem 5 and Remark 6.
The following Theorem is well-known [4, Ex. 1(2), p. 291] [2, Corollary 5 ]. We give a
short proof using equation (18).
Theorem 12. The sequence Bn+1 Bn is strictly increasing.
Proof. We have (see (18))
Bn+2 Bn+1 =


n 
X
n+1
k=0

Bn+1 Bn =

Bk ,

n1  
X
n
k=0

Bk .

Consequently
(Bn+2 Bn+1 ) (Bn+1 Bn ) =
since


n1 
X
n+1
k

k=1

  

n
n+1
>
k
k

 
n

Bk + (n + 1)Bn > 0,
k

(k = 1, . . . , n).

In closing the article we give one more property of the Bell numbers but before prove the
following general statement.

Theorem 13. Let Fn be a strictly increasing sequence of positive integers such that
log Fn Cn log n

(C > 0).

Let (x) be the number of Fn that do not exceed x. The following asymptotic formula holds.
(x)

log x
.
C log log x

Proof. Let n be a strictly increasing sequence of positive numbers and let (x) be the
number of n that do not exceed x. It is well-known [3, p. 129] that
n Cn log n (x)

x
.
C log x

Now,
Fn x n = log Fn log x.
Consequently
(x) = (log x)

log x
.
C log log x

Example 14. If Fn = Bn is the n-th Bell number and (x) is the number of Bell numbers
that do not exceed x then (see (19))
(x)

log x
.
log log x

Acknowledgements

The author would like to thank the anonymous referees for their valuable comments and
suggestions for improving the original version of this manuscript. The author is also very
grateful to Universidad Nacional de Lujan.

References
[1] G. E. Andrews, The Theory of Partitions, Cambridge University Press, 1984.
[2] S. Bouroubi, Bell numbers and Engels conjecture, Rostock. Math. Kolloq. 62 (2007),
6170.
[3] K. Chandrasekharan, Introduction to Analytic Number Theory, Springer, 1968.
[4] L. Comtet, Advances Combinatorics, D. Reidel Publishing Company, 1974.
[5] R. M. Corless, G. H. Gonnet, D. E. G. Hare, D. J. Jeffrey, and D. E. Knuth, On the
Lambert W function, Adv. Comput. Math. 5 (1996), 329359.
10

[6] N. G. de Bruijn, Asymptotic Methods in Analysis, Dover, 1981.

[7] R. Jakimczuk, Functions of slow increase and integer sequences, J. Integer Seq. 13
(2010), Article 10.1.1.
[8] M. Klazar, Counting set systems by weight, Electron. J. Combin. 12 (2005), # R11.
[9] D. E. Knuth, The Art of Computer Programming, Volume 4, Fascicle 3, Pearson Education, 2005.
[10] L. Lovasz, Combinatorial Problems and Excercises, Second Edition, AMS Chelsea Publishing, 2007.
[11] L. Moser and M. Wyman, An asymptotic formula for the Bell numbers, Trans. Roy.
Soc. Canada 49 (1955), 4953.
[12] J. Rey Pastor, P. Pi Calleja, and C. Trejo, An
alisis Matematico, Volumen I, Octava
Edicion, Editorial Kapelusz, 1969.

2000 Mathematics Subject Classification: Primary 11B99; Secondary 11B73.

Keywords: functions of slow increase, integer sequences, Bell numbers.
(Concerned with sequence A000110.)

Received December 9 2010; revised versions received March 28 2011; May 4 2011. Published
in Journal of Integer Sequences, May 10 2011.
Return to Journal of Integer Sequences home page.

11