文章目录
Central Limit Theorem ( 中心极限定理)
Independent Identically Distribution (独立同分布的中心极限定理 )
Definition
It refers that a sequence of random variables and its distribution can be asymptotically approximated to Normal/Gaussian distribution.
Let random variables
X
1
,
X
2
,
.
.
.
.
.
.
X
n
X1,X2,......X_n
X1,X2,......Xn be Independently Identical distributed(i.i.d), and we have
E
(
X
i
)
=
μ
,
D
(
X
i
)
=
σ
2
,
(
i
=
1
,
2...
,
n
)
\mathbb{E}(X_i)=μ,~ \mathbb{D}(X_i)=σ^2,~ (i=1,2...,n)
E(Xi)=μ, D(Xi)=σ2, (i=1,2...,n). For any
x
x
x, the distribution function
F
n
(
x
)
=
P
{
∑
i
=
1
n
X
i
−
n
μ
σ
n
≤
x
}
F_{n}(x)=P\left\{\frac{\sum_{i=1}^{n} X_{i}-n \mu}{\sigma \sqrt{n}} \leq x\right\}
Fn(x)=P{σn∑i=1nXi−nμ≤x} satisfies
lim
n
→
∞
F
n
(
x
)
=
lim
n
→
∞
{
∑
i
=
1
n
X
i
−
n
μ
n
σ
≤
x
}
=
1
2
π
∫
−
∞
x
e
−
−
t
2
2
d
t
=
∅
(
x
)
.
\lim _{n \rightarrow \infty} F_{n}(x)=\lim _{n \rightarrow \infty}\left\{\frac{\sum_{i=1}^{n} X_{i}-n \mu}{\sqrt{n} \sigma} \leq x\right\}=\frac{1}{\sqrt{2 \pi}} \int_{-\infty}^{x} e^{-\frac{-t^{2}}{2}} d t=\emptyset(x).
n→∞limFn(x)=n→∞lim{nσ∑i=1nXi−nμ≤x}=2π1∫−∞xe−2−t2dt=∅(x).
This theorem give the fact: for enough
n
n
n, random variable
Y
n
=
∑
i
=
1
n
X
i
−
n
μ
n
σ
Y_n = \frac{\sum_{i=1}^{n} X_i-n\mu }{\sqrt{n}\sigma}
Yn=nσ∑i=1nXi−nμ can be approximated to standard normal distribution, i.e.,
Y
n
∼
N
(
0
,
1
)
Y_n \sim N(0,1)
Yn∼N(0,1). Thus, for enough
n
n
n,
∑
i
=
1
n
X
i
=
n
σ
Y
n
+
n
μ
\sum_{i=1}^{n} X_i = \sqrt{n}\sigma Y_n + n \mu
∑i=1nXi=nσYn+nμ can be approximated to normal distribution
N
(
n
μ
,
n
σ
)
N(n\mu,n\sigma)
N(nμ,nσ).
Note that for sufficiently large
n
n
n, the sum of
n
n
n Independently Identically Distributed random variables can be considered as a normal variable.
Different Distributed Central Limit Theorem
Let
X
1
,
X
2
,
.
.
.
,
X
n
X_1, X_2, ...,X_n
X1,X2,...,Xn and
f
x
k
(
x
)
f_{x_k}(x)
fxk(x) for
k
=
1
,
.
.
.
,
n
k=1,...,n
k=1,...,n be a list of independent random variables and probability density function, respectively, with the following mean and variable:
E
(
X
k
)
=
μ
k
,
D
(
X
k
)
=
σ
k
2
.
\mathbb{E}(X_k) = \mu k,~~~ \mathbb{D}(X_k) = \sigma^2_k.
E(Xk)=μk, D(Xk)=σk2. Then, we define
B
n
2
=
∑
i
=
1
n
σ
k
2
B^2_n = \sum_{i=1}^n \sigma^2_k
Bn2=i=1∑nσk2 and
Y
n
=
∑
k
=
1
n
X
k
−
∑
k
=
1
n
μ
k
B
n
.
Y_n = \frac{ \sum_{k=1}^n X_k - \sum_{k=1}^n \mu k}{B_n}.
Yn=Bn∑k=1nXk−∑k=1nμk.
For any positive number
τ
\tau
τ,
lim
n
→
∞
1
B
n
2
∑
k
=
1
n
∫
∣
x
−
μ
k
>
τ
B
n
∣
(
x
−
μ
k
)
2
f
x
k
(
x
)
d
x
=
0.
\lim _{n \rightarrow \infty} \frac{1}{B_{n}^{2}} \sum_{k=1}^{n} \int_{\left|x-\mu_{k}>\tau B_{n}\right|}\left(x-\mu_{k}\right)^{2} f_{x_{k}}(x) d x=0.
n→∞limBn21k=1∑n∫∣x−μk>τBn∣(x−μk)2fxk(x)dx=0.
For any x x x, the distribution function F n ( x ) F_n(x) Fn(x) of random variable Y n Y_n Yn should satisfy lim n → ∞ F n ( x ) = lim n → ∞ P { ∑ k = 1 n X k − ∑ k = 1 n μ k B n ≤ x } = 1 2 π ∫ − ∞ x e − t 2 2 d t . \lim _{n \rightarrow \infty} F_{n}(x)=\lim _{n \rightarrow \infty} P\left\{\frac{\sum_{k=1}^{n} X_{k}-\sum_{k=1}^{n} \mu k}{B_{n}} \leq x\right\}=\frac{1}{\sqrt{2 \pi}} \int_{-\infty}^{x} e^{-\frac{t^{2}}{2}} \text{d} t. n→∞limFn(x)=n→∞limP{Bn∑k=1nXk−∑k=1nμk≤x}=2π1∫−∞xe−2t2dt.
This theorem proves that the distribution function can be approximated to normal/Gaussian distribution, if the random variables have a large number of independent and uniform random variables.