1
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 1 year has passed since last update.

3.8(標準) 二項分布,正規分布

Last updated at Posted at 2021-07-27

方針

二項分布

二項分布について書く.二項分布はベルヌーイ試行をn回行った時の成功回数に関する確率分布である.確率密度関数は,

\begin{align}
f_X(x)={}_nC_xp^x(1-p)^{n-x},\ x=0,...,n
\end{align}

積率母関数は,3.2で証明した二項定理を用いて

\begin{align}
M_X(t)&= \mathbb{E}_{X\sim Bin(n,p)}[e^{tX}]\\
&= \sum_{x=0}^n{}_nC_x(pe^t)^x(1-p)^{n-x}\\
&= (pe^t+1-p)^n
\end{align}

期待値は,

\begin{align}
\mathbb{E}_{X\sim Bin(n,p)}[X]&= \sum_{x=0}^nx\times{}_nC_xp^x(1-p)^{n-x}\\
&= \sum_{x=1}^n\frac{n!}{(n-x)!(x-1)!}p^x(1-p)^{n-x}\\
&= np\sum_{x=1}^n\frac{(n-1)!}{(n-x)!(x-1)!}p^{x-1}(1-p)^{n-x}\\
&= np\sum_{x'=0}^{n-1}\frac{(n-1)!}{(n-1-x')!x'!}p^{x'}(1-p)^{n-1-x'}\\
& \ \ \ \ (x'=x-1)\\
&= np 
\end{align}

分散は,

\begin{align}
\mathbb{E}_{X\sim Bin(n,p)}[X^2]&= \sum_{x=0}^nx^2\times{}_nC_xp^x(1-p)^{n-x}\\
&= \sum_{x=1}^nx\times\frac{n!}{(n-x)!(x-1)!}p^x(1-p)^{n-x}\\
&= np\sum_{x=1}^nx\times\frac{(n-1)!}{(n-x)!(x-1)!}p^{x-1}(1-p)^{n-x}\\
&= np\sum_{x'=0}^{n-1}(x'+1)\times\frac{(n-1)!}{(n-1-x')!x'!}p^{x'}(1-p)^{n-1-x'}\\
& \ \ \ \ (x'=x-1)\\
&= np\{(n-1)p+1\}\\
Var(X)&= np(np-p+1)-n^2p^2\\
&= np(1-p)
\end{align}

正規分布

次に,正規分布について書く.正規分布の確率密度関数は,

f_X(x)=\frac{1}{\sqrt{2\pi\sigma^2}}\exp\{-\frac{(x-\mu)^2}{2\sigma^2}\}

平均0,分散1の正規分布を標準正規分布と呼ぶ.標準正規分布の全確率1を証明する.

I=\int_{-\infty}^\infty \exp(-\frac{x^2}{2})dx

と置くと,

I^2=\int_{-\infty}^\infty dx\int_{-\infty}^\infty dy\exp(-\frac{X^2+y^2}{2})\tag{1}

である.ここで,

\begin{align}
\begin{cases}
x=r\cos\theta\\
y=r\sin\theta
\end{cases}
\end{align}

と置換すると,

\frac{\partial(x,y)}{\partial (r,\theta)}=\det\begin{pmatrix}
\cos\theta & -r\sin\theta\\
\sin\theta & r\cos\theta\end{pmatrix}=r

より,

\begin{align}
(1)&= \int_0^{2\pi}d\theta \int_0^\infty dr\exp(-\frac{r^2}{2})r\\
&= \int_0^{2\pi}\left[-\exp(-\frac{r^2}{2})\right]_0^\infty d\theta\\
&= 2\pi
\end{align}

なので,

I=\sqrt{2\pi}

となり,全確率1が証明された.一般の正規分布に対しても

z=\frac{x-\mu}{\sigma}

と置換することで,

\int_{-\infty}^\infty \frac{1}{\sqrt{2\pi\sigma^2}}\exp\{-\frac{(x-\mu)^2}{2\sigma^2}\}dx=\int_{-\infty}^\infty\frac{1}{\sqrt{2\pi}}\exp(-\frac{z^2}{2})dz=1

が成り立つ.積率母関数は,

\begin{align}
M_X(t)&= \int_{-\infty}^\infty\frac{1}{\sqrt{2\pi\sigma^2}}\exp\{-\frac{(x-\mu)^2}{2\sigma^2}+tx\}dx\\
&= \int_{-\infty}^\infty\frac{1}{\sqrt{2\pi\sigma^2}}\exp\{-\frac{x^2-2(\mu+t\sigma^2)x}{2\sigma^2}-\frac{\mu^2}{2\sigma^2}\}dx\\
&= \int_{-\infty}^\infty\frac{1}{\sqrt{2\pi\sigma^2}}\exp\{-\frac{(x-\mu-t\sigma^2)^2}{2\sigma^2}+\frac{\mu+t\sigma^2}-\frac{\mu^2}{2\sigma^2}\}dx\\
&= \int_{-\infty}^\infty\frac{1}{\sqrt{2\pi\sigma^2}}\exp\{-\frac{(x-\mu-t\sigma^2)^2}{2\sigma^2}+\mu t+\frac{t^2\sigma^2}{2}\}dx\\
&= \exp(\mu t+\frac{t\sigma^2}{2})\int_{-\infty}^\infty\frac{1}{\sqrt{2\pi\sigma^2}}\exp\{-\frac{(x-\mu-t\sigma^2)^2}{2\sigma^2}\}dx\\
&= \exp(\mu t+\frac{t^2\sigma^2}{2})
\end{align}

問題は,変数変換したYの積率母関数を求め,n→∞として標準正規分布の積率母関数に収束することを示せば良い.

答案

Yの積率母関数を求める.

\begin{align}
M_Y(t)&= \mathbb{E}_{X\sim Bin(np)}[e^{tY}]\\
&= \mathbb{E}_{X\sim Bin(np)}[e^{t(X-np)/\sqrt{np(1-p)}}]\\
&= \exp(-\frac{tnp}{\sqrt{np(1-p)}})\mathbb{E}_{X\sim Bin(n,p)}[e^{\frac{t}{\sqrt{np(1-p)}}X}]\\
&= \exp(-\frac{tnp}{\sqrt{np(1-p)}})\sum_{x=0}^n{}_nC_x(e^{\frac{t}{\sqrt{np(1-p)}}\times p})^x(1-p)^{n-x}\\
&= \exp(-\frac{tnp}{\sqrt{np(1-p)}})(e^{\frac{t}{\sqrt{np(1-p)}}}\times p+1-p)^n
\end{align}

標準正規分布に従う確率変数をZと置くと,

\begin{align}
M_Z(t)&= \mathbb{E}_{Z\sim\mathcal{N}(0,1)}[e^{tZ}]\\
&= \int_{-\infty}^\infty\frac{1}{\sqrt{2\pi}}\exp(-\frac{z^2-tz}{2})dz\\
&= \int_{-\infty}^\infty\frac{1}{\sqrt{2\pi}}\exp\{-\frac{(z-\frac{t}{2})^2}{2}+\frac{t^2}{2})dz\\
&= \exp(\frac{t^2}{2})\int_{-\infty}^\infty\frac{1}{\sqrt{2\pi}}\exp\{-\frac{(z-\frac{t}{2})^2}{2}\}dz\\
&= \exp(\frac{t^2}{2})
\end{align}

今,

\psi(t)=\log M_Y(t)

とおけば,n→∞で

\psi(t)\rightarrow\frac{t^2}{2}

であることを示せば良い.

\begin{align}
\psi(t)&= -\frac{tnp}{\sqrt{np(1-p)}}+n\log (e^{\frac{t}{\sqrt{np(1-p)}}}\times p+1-p)\\
&\approx -\frac{tnp}{\sqrt{np(1-p)}}+n\log\{p(1+\frac{t}{\sqrt{np(1-p)}}+\frac{t^2}{2np(1-p)}+o(\frac{1}{n}) +1-p\}\\
& \ \ \ \ \ \ as\ \ \ n\rightarrow\infty\\
&= -\frac{tnp}{\sqrt{np(1-p)}}+n\log\{1+(\frac{tp}{\sqrt{np(1-p)}}+\frac{t^2}{2n(1p)}+o(\frac{1}{n})) \}\tag{1}
\end{align}

(1)式第二項の対数部分をlog(1+A)と置くと,漸近展開によって

\begin{align}
n\log(1+A)&= n(A-\frac{A^2}{2}+\mathcal{o}(A^2))\ \ \ as\ \ \ A\rightarrow0\\
&= n(\frac{tp}{\sqrt{np(1-p)}}+\frac{t^2}{2n(1-p)}+\mathcal{o}(\frac{1}{n})-\frac{A^2}{2}+\mathcal{o}(A^2))\\
&= \frac{tnp}{\sqrt{np(1-p)}}+\frac{t^2}{2(1-p)}-\frac{nA^2}{2}+\mathcal{o}(1)\tag{2}\\
\frac{nA^2}{2}&= n(\frac{tp}{\sqrt{np(1-p)}}+\frac{t^2}{2n(1-p)}+\mathcal{o}(\frac{1}{n}))^2/2\\
&= (\frac{t^2p^2}{p(1-p)}+\frac{pt^3}{\sqrt{np(1-p)}(1-p)}+\frac{t^4}{4n(1-p)^2})/2+\mathcal{o}(1)\\
&\approx \frac{t^2p}{2(1-p)}\ \ \ as\ \ \ n\rightarrow\infty\\
\therefore (2)&\approx \frac{tnp}{\sqrt{np(1-p)}}+\frac{t^2}{2(1-p)}-\frac{t^2p}{2(1-p)}
\end{align}

以上を(1)に代入して

\psi(t)\approx\frac{t^2}{2(1-p)}-\frac{t^2p}{2(1-p)}=\frac{t^2}{2}

参考文献

1
0
2

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
1
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?