Scribd
Upload
26 views2 pages

Feller Probability

This chapter explores the properties of randomness and the law of the iterated logarithm in the context of infinite sequences of Bernoulli trials. It discusses the probabilities associated with events defined in an infinite sample space, particularly focusing on scenarios involving runs of successes and failures. The chapter emphasizes the transition from finite to infinite trials and provides examples of calculating probabilities in such contexts.

Uploaded by

parthtyagi297
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
26 views2 pages

Feller Probability

This chapter explores the properties of randomness and the law of the iterated logarithm in the context of infinite sequences of Bernoulli trials. It discusses the probabilities associated with events defined in an infinite sample space, particularly focusing on scenarios involving runs of successes and failures. The chapter emphasizes the transition from finite to infinite trials and provides examples of calculating probabilities in such contexts.

Uploaded by

parthtyagi297
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

CHAPTER VIII*

Unlimited Sequences
of Bernoulli Trials

This chapter discusses certain properties of randomness and the im-


portant law of the iterated logarithm for Bernoulli trials. A different
aspect of the fluctuation theOIY of BerIloulli trials (at least for p D
is covered in chapter III.

1. INFINITE SEQUENCES OF TRIALS


In the preceding chapter we have dealt with probabilities connected
with 1l Bernoulli trials and have studied their asymptotic behavior as
n ~ 00. We turn now to a more general type of problem where the
events themselves cannot be defined in a finite sample space.
Example. A problem in runs. Let IX and {3 be positive integers, and
. consider a potentially unlimited sequence of Bernoulli trials, such as
tossing a coin or throwing dice. Suppose that Paul bets Peter that a run of
IX consecutive successes will occur before a run of {3 consecutive failures.
It has an intuitive meaning to speak of the event that Paul wins, but it
must be remembered that in the mathematical theory the term event
stands for "aggregate of sample points" and is meaningless unless an
appropriate sample space has been defined. The model. of a finite number
of trials is insufficient for our present purpose, but the difficulty is solved
by a simple passage to the limit. In n trials Peter wins or loses, or the
game remains undecided. Let the corresponding probabilities be X n ' Y n' Zn
(x n + Yn + Zn = 1). As the number n of trials increases, the probability
Zn of a tie can only decrease, and both Xn and Y n necessarily increase.
Hence x = lim X n , Y = lim Yn' and Z = lim Zn exist. Nobody would

* This chapter is not directly connected with the material covered in subsequent
chapters a od may be omitted at first reading.
196
VIII. 1], INFINITE SEQUENCES OF TRIALS 197

hesitate to call them the probabilities of Peter's ultimate gain or loss or of


a tie. However, the corresponding three events are defined only in the
sample space of infinite sequences of trials, and this space is not discrete.
The example was introduced for illustration only, and the numerical values of
x n, Yn, Zn are not our immediate concern. We shall return to their calculation in
example XIII, (8.b). The limits x, y, Z may be obtained by a simpler method which is
applicable to more general cases. We indicate it here because of its importance and
intrinsic interest.
Let A be the event that a run of oc consecutLVe successes occurs before a run of fJ
consecutive failures. In the event A Paul wins and x = P{A}. If u and v are the
conditional probabilities of A under the hypotheses, respectively, that the first trial
results in success or failure, then x = pu + qv [see V, (1.8)]. Suppose first that the
first trial results ~n success. In this case the event A can occur in oc mutually ex-
clusive ways: (1) The following oc 1 trials result in snccesses; the probability for
this is pa-I. (2) The first failure occurs at the vth trial where 2 ~ v ~ oc. Let this
event be Hv' Then P{H) = pV-2q , and P{A I H) = v. . Hence (using once more the
formula for compound probabilities)

(1.1)

If the first trial results in failure, a similar argument leads to

(1.2) v = pu(l +q+ ... +qP-2) = u(l- qP-I).


We have thus two equations for the two unknowns u and v and find for x = pu + qv

(1.3) x
1 pa. 1 + qIJ 1 _ pa lqlJ 1

To obtain Y we have only to interchange p and q, and oc and fJ.' Thus

(1.4)

Since x + Y = 1, we have Z = 0; the probability of a tie is zero.


For example, in tossing a coin (p = !) the probability that a run of two heads
appears before a run of three tails is 0.7; for two consecutive heads before four con-
secutive tails the probability is t, for three consecutive heads before four consecutive
tails it. In rolling dice there is probability 0.1753 that two consecutive aces will appear
before five consecutive non-aces, etc. ~

In the present volume we are confined to the theory of discrete sample


spaces, and this means a considerable loss of mathematical elegance. The
general theory considers n Bernoulli trials only as the beginning of an
infinite sequence of trials. A sample point is then represented by an
infinite sequence ofletters Sand F, and the sample space is the aggregate
of all such sequences. A finite sequence, like SSFS, stands for the
aggregate of all points with this beginning, that is, for the compound event
that in an infinite sequence of trials the first four result in S, S, F, S,

You might also like