Download A Proof of the Central Limit Theorem

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts
no text concepts found
Transcript
A Proof of the Central Limit Theorem
Preliminaries: Two Definitions and Two Theorems
Definition One: The rth moment of a random variable X about the mean is
πœ‡π‘Ÿ = 𝐸[(𝑋 βˆ’ πœ‡)π‘Ÿ ] , π‘Ÿ = 0,1,2,3, …
[Note: It follows that πœ‡0 = 1, πœ‡1 = 0, and πœ‡2 = 𝜎 2 .]
Thus, we have
π‘Ÿ
πœ‡π‘Ÿ = βˆ‘π‘›π‘—=1(π‘₯𝑗 βˆ’ πœ‡) 𝑝(π‘₯𝑗 ) = βˆ‘(π‘₯ βˆ’ πœ‡)π‘Ÿ 𝑝(π‘₯)
∞
πœ‡π‘Ÿ = βˆ«βˆ’βˆž(π‘₯ βˆ’ πœ‡)π‘Ÿ 𝑓(π‘₯)𝑑π‘₯
(discrete variable)
(continuous variable)
Definition Two: The moment-generating function (MGF) of X is
𝑀𝑋 (𝑑) = 𝐸(𝑒 𝑑𝑋 )
Thus, we have
𝑀𝑋 (𝑑) = βˆ‘π‘›π‘—=1 𝑒 𝑑π‘₯𝑗 𝑝(π‘₯𝑗 ) = βˆ‘ 𝑒 𝑑π‘₯ 𝑝(π‘₯)
∞
𝑀𝑋 (𝑑) = βˆ«βˆ’βˆž 𝑒 𝑑π‘₯ 𝑓(π‘₯)𝑑π‘₯
(discrete case)
(continuous case)
[Note: The name moment-generating function comes from the fact that
𝑀𝑋 (𝑑) = 𝐸(𝑒 𝑑𝑋 ) = 𝐸 (1 + 𝑑𝑋 +
𝑑2𝑋2
2!
+
𝑑3𝑋3
3!
𝑑2
+β‹―)
𝑑3
= 1 + 𝐸(𝑋)𝑑 + 𝐸(𝑋 2 ) 2! + 𝐸(𝑋 3 ) 3! + β‹―
which shows the moments about the origin πœ‡ = 0 [𝐸(𝑋), 𝐸(𝑋 2 ), 𝐸(𝑋 3 ), …] are
generated as the coefficients in this expansion in 𝑑.]
Proof of Central Limit Theorem
Page Two
Theorem One: The MGF for the general normal distribution is
𝑀(𝑑) = 𝑒 πœ‡π‘‘+(𝜎
2 𝑑 2 /2)
Proof: By definition, we have
∞
1
𝑀(𝑑) = 𝐸(𝑒 𝑑π‘₯ ) = 𝜎√2πœ‹ βˆ«βˆ’βˆž 𝑒 𝑑π‘₯ 𝑒 βˆ’(π‘₯βˆ’πœ‡)
2 /2𝜎 2
𝑑π‘₯
Letting (π‘₯ βˆ’ πœ‡) / 𝜎 = 𝑣 so that π‘₯ = πœ‡ + πœŽπ‘£ and 𝑑π‘₯ = πœŽπ‘‘π‘£, this becomes
𝑀(𝑑) =
1
√2πœ‹
𝑒 πœ‡π‘‘+(𝜎
2 𝑑 2 /2)
∞
βˆ«βˆ’βˆž 𝑒 βˆ’(π‘£βˆ’πœŽπ‘‘)
2 /2
𝑑𝑣
Now letting 𝑣 βˆ’ πœŽπ‘‘ = 𝑀, we have
𝑀(𝑑) = 𝑒 πœ‡π‘‘+(𝜎
2 𝑑 2 /2)
(
∞
1
∫ 𝑒 βˆ’π‘€
√2πœ‹ βˆ’βˆž
2 /2
𝑑𝑀) = 𝑒 πœ‡π‘‘+(𝜎
2 𝑑 2 /2)
Corollary: The MGF for the standard normal distribution is
𝑀(𝑑) = 𝑒 𝑑
2 /2
Theorem Two (Uniqueness Theorem): Two random variables have the same probability
distribution if and only if their moment-generating
functions are identical.
We are now ready to prove the Central Limit Theorem.
Proof of Central Limit Theorem (CLT)
Theorem (CLT): Let 𝑋1 , 𝑋2 , … , 𝑋𝑛 be independent random variables that are identically
distributed (that is, all have the same probability mass function in the discrete
case or probability density function in the continuous case) and have finite
mean πœ‡ and variance 𝜎 2 . Then if 𝑆𝑛 = 𝑋1 + 𝑋2 + β‹― + 𝑋𝑛 (n = 1, 2, …),
lim 𝑃(π‘Ž ≀
π‘›β†’βˆž
𝑆𝑛 βˆ’π‘›πœ‡
𝜎 βˆšπ‘›
≀ 𝑏) =
1
𝑏
∫ 𝑒 βˆ’π‘’
√2πœ‹ π‘Ž
2 /2
𝑑𝑒
that is, the random variable (𝑆𝑛 βˆ’ π‘›πœ‡)/(πœŽβˆšπ‘› ), which is the standardized
variable corresponding to 𝑆𝑛 , is asymptotically normal.
Note: Since (𝑆𝑛 βˆ’ π‘›πœ‡)/(πœŽβˆšπ‘› ) = (𝑋̅ βˆ’ πœ‡)/(πœŽβ„βˆšπ‘› ) = 𝑍, the more familiar form of the CLT
follows as a corollary.
Proof of Central Limit Theorem
Page Three
Proof: For 𝑛 = 1, 2, …, we have 𝑆𝑛 = 𝑋1 + 𝑋2 + β‹― + 𝑋𝑛 . Now 𝑋1 , 𝑋2 , … , 𝑋𝑛 each have mean πœ‡
and variance 𝜎 2 . Thus,
𝐸(𝑆𝑛 ) = 𝐸(𝑋1 + 𝑋2 + β‹― + 𝑋𝑛 ) = 𝐸(𝑋1 ) + 𝐸(𝑋2 ) + β‹― 𝐸(𝑋𝑛 ) = π‘›πœ‡
and, because the π‘‹π‘˜ are independent,
π‘‰π‘Žπ‘Ÿ(𝑆𝑛 ) = π‘‰π‘Žπ‘Ÿ(𝑋1 + 𝑋2 + β‹― + 𝑋𝑛 ) = π‘‰π‘Žπ‘Ÿ(𝑋1 ) + π‘‰π‘Žπ‘Ÿ(𝑋2 ) + β‹― + π‘‰π‘Žπ‘Ÿ(𝑋𝑛 ) = π‘›πœŽ 2
It follows from this that the standardized random variable corresponding to 𝑆𝑛 is
𝑆𝑛 βˆ— =
𝑆𝑛 βˆ’π‘›πœ‡
𝜎 βˆšπ‘›
The MGF for 𝑆𝑛 βˆ— is thus
βˆ—
𝐸(𝑒 𝑑𝑆𝑛 ) = 𝐸[𝑒 𝑑(𝑆𝑛 βˆ’π‘›πœ‡)/(πœŽβˆšπ‘›) ]
= E[et(X1 βˆ’ΞΌ)/(Οƒβˆšn) et(X2 βˆ’ΞΌ)/(Οƒβˆšn) β‹― et(Xnβˆ’ΞΌ)/(Οƒβˆšn) ]
= 𝐸[et(X1 βˆ’ΞΌ)/(Οƒβˆšn) ]βˆ™ 𝐸[et(X2βˆ’ΞΌ)/(Οƒβˆšn) ]β‹― 𝐸[et(Xnβˆ’ΞΌ)/(Οƒβˆšn) ]
t(X1 βˆ’ΞΌ)
Οƒβˆšn
= {𝐸[e
𝑛
]}
[The last step is a result of the π‘‹π‘˜ being identically distributed and the preceding step
because the π‘‹π‘˜ are independent.]
Now, by a Taylor series expansion, we have
t(X1 βˆ’ΞΌ)
Οƒβˆšn
𝐸 [e
] = 𝐸[1 +
𝑑(𝑋1 βˆ’πœ‡)
𝜎 βˆšπ‘›
= 𝐸(1) +
= 1+
𝑑
𝜎 βˆšπ‘›
𝑑
𝜎 βˆšπ‘›
+
𝑑 2 (𝑋1 βˆ’πœ‡)2
2!𝜎 2 𝑛
+ β‹―]
𝐸(𝑋1 βˆ’ πœ‡) +
(0) +
𝑑2
2𝜎 2 𝑛
𝑑2
2𝜎 2 𝑛
𝐸(𝑋1 βˆ’ πœ‡)2 + β‹―
(𝜎 2 ) + β‹― = 1 +
𝑑2
2𝑛
+β‹―
Thus, we have
βˆ—
𝑑2
𝑛
𝐸(𝑒 𝑑𝑆𝑛 ) = {1 + + β‹― } = (1 +
2𝑛
2
𝑑 2 ⁄2
𝑛
+ β‹― )𝑛
But the limit of this as 𝑛 β†’ ∞ is 𝑒 𝑑 ⁄2 , which is the MGF of the standard normal
distribution. Hence, by Theorem Two above, the required result follows.
Proof of Central Limit Theorem
Page Four
Corollary: Suppose the population from which samples are taken has some probability
distribution with mean πœ‡ and variance 𝜎 2 . Then the standardized variable associated
with 𝑋̅, given by
𝑍=
π‘‹Μ…βˆ’πœ‡
𝜎 β„βˆš 𝑛
is asymptotically normal, that is
lim 𝑃(𝑍 ≀ 𝑧) =
π‘›β†’βˆž
1
𝑧
∫ 𝑒 βˆ’π‘€
√2πœ‹ βˆ’βˆž
2 ⁄2
𝑑𝑀
Note: The above proof of the CLT, slightly edited, was given by the late Murray R. Spiegel of RPI
in his 1975 Schaum’s Outline Series book Probability and Statistics in Problem 4.25, which
appears as such in the current Third Edition of the book, with John J. Schiller and R. Alu
Srinivasan of Temple University now as coauthors.
DB
Related documents