Statistics & Probability Letters 58 (2002) 283–286

A note on the Borel–Cantelli lemma

Valentin V. Petrov ∗
Faculty of Mathematics and Mechanics, St. Petersburg University, Stary Peterhof, St. Petersburg 198904, Russia
Received June 2001

Abstract

c 2002 Elsevier
A generalization of the Erd+os–R,enyi formulation of the Borel–Cantelli lemma is obtained.

Science B.V. All rights reserved.

Keywords: Borel–Cantelli lemma; Generalizations of the Borel–Cantelli lemma; Pairwise independent events

1. Introduction

The following Borel–Cantelli lemma plays an exceptionally important role in probability

theory: if
A1 ; A2 ; : : : is a sequence of events on a common probability space (; F; P) and if
∞
n=1 P (A n ) ¡ ∞,
then P(lim sup An ) = 0; if A1 ; A2 ; : : : is a sequence of independent events and if ∞ n=1 P(A n ) = ∞,
then P (lim sup An ) = 1. Here
∞ ∞
lim sup An = Ak :
n=1 k=n

Many investigations were devoted to the second part of the Borel–Cantelli lemma in attempts to
weaken the independence condition that means mutual independence of events A1 ; : : : ; An for every
n. Erd+os and R,enyi (1959) discovered that the independence condition in the second part of the
Borel–Cantelli lemma can be replaced by the weaker condition of pairwise independence of events
A1 ; A2 ; : : : : More general results have been proved independently by Kochen and Stone (1964) and
Spitzer (1964). Erd+os and R,enyi (1959) also found that the condition of pairwise independence of
events A1 ; A2 ; : : : can be replaced by the weaker condition P(Ak Aj ) 6 P(Ak )P(Aj ) for every k and j
such that k = j.

∗
Corresponding author. Centro de Matematica, Facultad de Ciencias, CP 11400, 4225 Montevideo, CP, Igua, Uruguay.
Tel.: +598-2-525-2522; fax: +598-2-522-0653.
E-mail address: valentin.petrov@pobox.spbu.ru (V.V. Petrov).

0167-7152/02/$ - see front matter  c 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 7 - 7 1 5 2 ( 0 2 ) 0 0 1 1 3 - X
284 V.V. Petrov / Statistics & Probability Letters 58 (2002) 283–286

Lamperti

(1963) formulated the following proposition. If A1 ; A2 ; : : : is a sequence of events such
that ∞ n=1 P(A n ) = ∞ and P(Ak Aj ) 6 CP(Ak )P(Aj ) for all k; j ¿ N and some constants C and N ,
then P(lim sup An ) ¿ 0.
We present a more general and more precise theorem of this type. The above-mentioned results
of Erd+os and R,enyi follow from our theorem.

2. Results

Theorem 2.1. Let A1 ; A2 ; : : : be a sequence of events satisfying conditions

∞
P(An ) = ∞ (1)
n=1

and
P(Ak Aj ) 6 CP(Ak )P(Aj ) (2)
for all k; j ¿ L such that k = j and for some constants C ¿ 1 and L. Then
P(lim sup An ) ¿ 1=C: (3)

Note that under condition (1) the probability P(lim sup An ) can take on any value in the closed
interval [0; 1]. Martikainen and Petrov (1990) found conditions that are necessary and suKcient for
the equality P(lim sup An ) =  where 0 6  6 1 (see also Petrov, 1995, p. 201).
The following proposition (due to Erd+os and R,enyi, 1959) is an immediate corollary of Theo-
rem 2.1.

Theorem 2.2. Let A1 ; A2 ; : : : be a sequence of events satisfying conditions (1) and

P(Ak AJ ) 6 P(Ak )P(Aj ) (4)
for all su3ciently large k and j; k = j. Then
P(lim sup An ) = 1: (5)

If A1 ; A2 ; : : : is a sequence of pairwise independent events, condition (4) is satisLed with the sign
of equality.

3. Proof of Theorem 2.1

Our proof is based on the following inequality of Chung and Erd+os (1952): if A1 ; : : : ; An are
arbitrary events, then
 n   n 2 n
  
P Ak ¿ P(Ak ) P(Ak Aj )
k=1 k=1 k; j=1
 V.V. Petrov / Statistics & Probability Letters 58 (2002) 283–286 285

(see also Petrov (1995, p. 203), where this inequality is proved by means of a strengthening of
Lyapunov’s inequality for moments of a random variable).
Let A1 ; A2 ; : : : be a sequence of events satisfying conditions of Theorem 2.1.
By the Chung–Erd+os inequality we have
 N   N 2 N
  
P Ak ¿ P(Ak ) P(Ak Aj ) (6)
k=n k=n k; j=n

assuming that n ¡ N . It follows from condition (2) that

N

P(Ak Aj ) 6 CT1 + T2 (7)
k; j=n

if n ¿ L where
N
 ∞

T1 = P(Ak )P(Aj ); T2 = P(Ak ): (8)
k; j=n n=1
k =j

Since C ¿ 1, (7) implies the inequality

N

P(Ak Aj ) 6 C(T1 + T2 ): (9)
k; j=n

Obviously,
 N 2
 N

T1 = P(Ak ) − (P(Ak ))2 :
k=n k=n

Therefore,
 N 2
 N

T1 + T 2 6 P(Ak ) + P(Ak ): (10)
k=n k=n

Taking into account inequalities (6) – (10), we obtain

 N   N 2  N 2 −1
    N
 
P Ak ¿ C − 1 P(Ak ) P(Ak ) + P(Ak )
 
k=n k=n k=n k=n
  N −1 −1
  
¿ C −1 1 + P(Ak ) :
 
k=n

N
When N → ∞ and n is Lxed, we get 1 + ( k=n P(Ak ))−1 → 1 by condition (1). Thus,
 N 

lim inf P Ak ¿ C − 1
N →∞
k=n
286 V.V. Petrov / Statistics & Probability Letters 58 (2002) 283–286

and
 ∞


P Ak ¿ C −1 : (11)
k=n
 
Putting Bn = ∞ ⊃ · · · and ∞
k=n Ak we observe that B1 ⊃ B2 n=1 Bn = lim sup An .
Hence, there exists the limit lim P(Bn ) = P( ∞ B
n=1 n ) = P (lim sup An ), and (11) implies the
−
inequality P(lim sup An ) ¿ C .1

Acknowledgements

This note was written while the author was visiting Department of Mathematics, Chalmers Uni-
versity of Technology and G+oteborg University. The author is grateful to Professor Peter Jagers and
members of the Department for hospitality and help.

References

Chung, K.L., Erd+os, P., 1952. On the application of the Borel–Canlelli

lemma. Trans. Amer. Math. Soc. 72, 179–186.
Erd+os, P., R,enyi, A., 1959. On Cantor’s series with convergent 1=qn . Ann. Univ. Sci. Budapest. Sect. Math. 2, 93–109.
Kochen, S.B., Stone, C.J., 1964. A note on the Borel–Cantelli lemma. Illinois J. Math. 8, 248–251.
Lamperti, J., 1963. Wiener’s test and Markov chains. J. Math. Anal. Appl. 6, 58–66.
Martikainen, A.I., Petrov, V.V., 1990. On the Borel–Cantelli lemma. Zapiski Nauchn Sem. Leningrad. Otdel Mat. Inst.
Steklov. 184, 200 –207 (in Russian) (English transl. in: J. Math. Sci. , 1994, 63, 540 –544).
Petrov, V.V., 1995. Limit Theorems of Probability Theory. Oxford University Press, Oxford.
Spitzer, F., 1964. Principles of Random Walk. Van Nostrand, Princeton.