82 Probability and Distributions
Example 1.10.3.LetXbe a nondegenerate random variable with meanμand a
finite second moment. Thenμ^2 <E(X^2 ). This is obtained by Jensen’s inequality
using the strictly convex functionφ(t)=t^2.
The last inequality concerns different means of finite sets of positive numbers.
Example 1.10.4(Harmonic and Geometric Means). Let{a 1 ,...,an}be a set of
positive numbers. Create a distribution for a random variableXby placing weight
1 /non each of the numbersa 1 ,...,an. Then the mean ofXis thearithmetic
mean,(AM),E(X)=n−^1
∑n
i=1ai. Then, since−logxis a convex function, we
have by Jensen’s inequality that
−log
(
1
n
∑n
i=1
ai
)
≤E(−logX)=−
1
n
∑n
i=1
logai=−log(a 1 a 2 ···an)^1 /n
or, equivalently,
log
(
1
n
∑n
i=1
ai
)
≥log(a 1 a 2 ···an)^1 /n,
and, hence,
(a 1 a 2 ···an)^1 /n≤
1
n
∑n
i=1
ai. (1.10.6)
The quantity on the left side of this inequality is called thegeometric mean(GM).
So (1.10.6) is equivalent to saying that GM≤AM for any finite set of positive
numbers.
Now in (1.10.6) replaceaiby 1/ai(which is also positive). We then obtain
1
n
∑n
i=1
1
ai
≥
(
1
a 1
1
a 2
···
1
an
) 1 /n
or, equivalently,
1
1
n
∑n
i=1
1
ai
≤(a 1 a 2 ···an)^1 /n. (1.10.7)
The left member of this inequality is called theharmonic mean(HM). Putting
(1.10.6) and (1.10.7) together, we have shown the relationship
HM≤GM≤AM, (1.10.8)
for any finite set of positive numbers.
EXERCISES
1.10.1.LetX be a random variable with meanμand letE[(X−μ)^2 k]exist.
Show, withd>0, thatP(|X−μ|≥d)≤E[(X−μ)^2 k]/d^2 k. This is essentially
Chebyshev’s inequality whenk= 1. The fact that this holds for allk=1, 2 , 3 ,...,
when those (2k)th moments exist, usually provides a much smaller upper bound for
P(|X−μ|≥d) than does Chebyshev’s result.