Gamma分布的概率密度函数表示如下:
X
∽
G
(
α
,
β
)
:
f
(
x
)
=
β
α
Γ
(
α
)
x
α
−
1
e
−
β
x
X \backsim G(\alpha,\beta): f(x)=\frac{\beta^\alpha}{\Gamma(\alpha)}x^{\alpha-1}e^{-\beta x}
X∽G(α,β):f(x)=Γ(α)βαxα−1e−βx
其对应的矩母函数为
M
x
(
t
)
=
(
1
+
β
t
)
−
α
{\rm M}_x(t)=(1+\beta t)^{-\alpha}
Mx(t)=(1+βt)−α
显然,若
X
1
∽
G
(
α
1
,
β
)
,
X
2
∽
G
(
α
2
,
β
)
X_1 \backsim G(\alpha_1,\beta),X_2 \backsim G(\alpha_2,\beta)
X1∽G(α1,β),X2∽G(α2,β),则
M
x
1
(
t
)
=
(
1
+
t
/
β
)
−
α
1
,
M
x
2
(
t
)
=
(
1
+
t
/
β
)
−
α
2
{\rm M}_{x_1}(t)=(1+t/\beta)^{-\alpha_1},{\rm M}_{x_2}(t)=(1+t/\beta)^{-\alpha_2}
Mx1(t)=(1+t/β)−α1,Mx2(t)=(1+t/β)−α2,所以
M
x
1
(
t
)
M
x
2
(
t
)
=
(
1
+
t
/
β
)
−
(
α
1
+
α
2
)
{\rm M}_{x_1}(t){\rm M}_{x_2}(t)=(1+t/\beta)^{-(\alpha_1+\alpha_2)}
Mx1(t)Mx2(t)=(1+t/β)−(α1+α2),所以
X
1
+
X
2
∽
G
(
α
1
+
α
2
,
β
)
X_1+X_2 \backsim G(\alpha_1+\alpha_2,\beta)
X1+X2∽G(α1+α2,β),这即为Gamma分布的可加性原则。
从矩母函数的角度出发,Gamma分布可加性原则是显而易见的。然而,本人在利用其它方法推导这一性质时,却出现了矛盾,一时难以发现端倪。现将推导过程展示如下(针对 α 1 , α 2 \alpha_1,\alpha_2 α1,α2为大于等于1的整数的情况):
由概率论的知识我们知道,两个随机变量的和的概率密度函数是各自概率密度函数的卷积,因此若
X
=
X
1
+
X
2
X=X_1+X_2
X=X1+X2,则
f
x
(
x
)
=
f
x
1
(
x
)
∗
f
x
2
(
x
)
=
∫
0
x
f
x
1
(
y
)
f
x
2
(
x
−
y
)
d
y
=
β
α
1
+
α
2
Γ
(
α
1
)
Γ
(
α
2
)
∫
0
x
y
α
1
−
1
e
−
β
y
(
x
−
y
)
α
2
−
1
e
−
β
(
x
−
y
)
d
y
=
β
α
1
+
α
2
Γ
(
α
1
)
Γ
(
α
2
)
e
−
β
x
∫
0
x
y
α
1
−
1
(
x
−
y
)
α
2
−
1
d
y
(1)
\begin{equation} \begin{aligned} f_x(x)&=f_{x_1}(x)*f_{x_2}(x)=\int_0^x f_{x_1}(y)f_{x_2}(x-y)dy\\ &=\frac{\beta^{\alpha_1+\alpha_2}}{\Gamma(\alpha_1)\Gamma(\alpha_2)}\int_0^x y^{\alpha_1-1}e^{-\beta y} (x-y)^{\alpha_2-1}e^{-\beta(x-y)}dy\\ &=\frac{\beta^{\alpha_1+\alpha_2}}{\Gamma(\alpha_1)\Gamma(\alpha_2)}e^{-\beta x}\int_0^x y^{\alpha_1-1} (x-y)^{\alpha_2-1}dy \end{aligned} \end{equation}\tag{1}
fx(x)=fx1(x)∗fx2(x)=∫0xfx1(y)fx2(x−y)dy=Γ(α1)Γ(α2)βα1+α2∫0xyα1−1e−βy(x−y)α2−1e−β(x−y)dy=Γ(α1)Γ(α2)βα1+α2e−βx∫0xyα1−1(x−y)α2−1dy(1)
针对(1)中最后一个等式的计算,采用不同方法出现了不同的结果:
方法1:用二项展开可以得到
(
x
−
y
)
α
2
−
1
=
∑
i
=
0
α
2
−
1
(
−
1
)
i
(
α
2
−
1
i
)
x
α
2
−
1
−
i
y
i
(x-y)^{\alpha_2-1}=\sum_{i=0}^{\alpha_2-1} (-1)^i\begin{pmatrix}\alpha_2-1\\i\end{pmatrix}x^{\alpha_2-1-i} y^i
(x−y)α2−1=∑i=0α2−1(−1)i(α2−1i)xα2−1−iyi,将其代入(1)中可以得到
f
x
(
x
)
=
β
α
1
+
α
2
Γ
(
α
1
)
Γ
(
α
2
)
e
−
β
x
∑
i
=
0
α
2
−
1
(
−
1
)
i
(
α
2
−
1
i
)
x
α
2
−
1
−
i
∫
0
x
y
α
1
+
i
−
1
d
y
=
β
α
1
+
α
2
Γ
(
α
1
)
Γ
(
α
2
)
[
∑
i
=
0
α
2
−
1
(
−
1
)
i
(
α
2
−
1
i
)
/
(
α
1
+
i
)
]
x
α
1
+
α
2
−
1
e
−
β
x
(2)
\begin{equation} \begin{aligned} f_x(x)&=\frac{\beta^{\alpha_1+\alpha_2}}{\Gamma(\alpha_1)\Gamma(\alpha_2)}e^{-\beta x}\sum_{i=0}^{\alpha_2-1}(-1)^i\begin{pmatrix}\alpha_2-1\\i \end{pmatrix}x^{\alpha_2-1-i} \int_0^x y^{\alpha_1+i-1}dy\\ &=\frac{\beta^{\alpha_1+\alpha_2}}{\Gamma(\alpha_1)\Gamma(\alpha_2)}[\sum_{i=0}^{\alpha_2-1}(-1)^i\begin{pmatrix}\alpha_2-1\\i \end{pmatrix}/(\alpha_1+i)]x^{\alpha_1+\alpha_2-1}e^{-\beta x} \end{aligned} \end{equation}\tag{2}
fx(x)=Γ(α1)Γ(α2)βα1+α2e−βxi=0∑α2−1(−1)i(α2−1i)xα2−1−i∫0xyα1+i−1dy=Γ(α1)Γ(α2)βα1+α2[i=0∑α2−1(−1)i(α2−1i)/(α1+i)]xα1+α2−1e−βx(2)
显然,要想使命题得证,需要有
∑
i
=
0
α
2
−
1
(
−
1
)
i
(
α
2
−
1
i
)
/
(
α
1
+
i
)
=
B
(
α
1
,
α
2
)
\sum_{i=0}^{\alpha_2-1}(-1)^i\begin{pmatrix}\alpha_2-1\\i \end{pmatrix}/(\alpha_1+i)=\Beta(\alpha_1,\alpha_2)
∑i=0α2−1(−1)i(α2−1i)/(α1+i)=B(α1,α2),(其中
B
(
α
1
,
α
2
)
B(\alpha_1,\alpha_2)
B(α1,α2)表示beta函数)但是这个等式似乎并不成立(可以简单地用数值验证)。因此利用上述方法无法使命题得证。
方法2:令
y
=
t
x
y=tx
y=tx,将(1)转化为对
t
t
t的积分,可以得到
f
x
(
x
)
=
β
α
1
+
α
2
Γ
(
α
1
)
Γ
(
α
2
)
[
∫
0
1
t
α
1
−
1
(
1
−
t
)
α
2
−
1
d
t
]
x
α
1
+
α
2
−
1
e
−
β
x
=
β
α
1
+
α
2
Γ
(
α
1
)
Γ
(
α
2
)
B
(
α
1
,
α
2
)
x
α
1
+
α
2
−
1
e
−
β
x
=
β
α
1
+
α
2
Γ
(
α
1
+
α
2
)
x
α
1
+
α
2
−
1
e
−
β
x
(3)
\begin{equation} \begin{aligned} f_x(x)&=\frac{\beta^{\alpha_1+\alpha_2}}{\Gamma(\alpha_1)\Gamma(\alpha_2)}[\int_0^1 t^{\alpha_1-1}(1-t)^{\alpha_2-1}dt]x^{\alpha_1+\alpha_2-1}e^{-\beta x}\\ &=\frac{\beta^{\alpha_1+\alpha_2}}{\Gamma(\alpha_1)\Gamma(\alpha_2)}\Beta(\alpha_1,\alpha_2)x^{\alpha_1+\alpha_2-1}e^{-\beta x}\\ &=\frac{\beta^{\alpha_1+\alpha_2}}{\Gamma(\alpha_1+\alpha_2)}x^{\alpha_1+\alpha_2-1}e^{-\beta x} \end{aligned} \end{equation}\tag{3}
fx(x)=Γ(α1)Γ(α2)βα1+α2[∫01tα1−1(1−t)α2−1dt]xα1+α2−1e−βx=Γ(α1)Γ(α2)βα1+α2B(α1,α2)xα1+α2−1e−βx=Γ(α1+α2)βα1+α2xα1+α2−1e−βx(3)
显然,命题得证。
暂时没发现为什么方法1没有能够使命题得证!!!
说明: 经过证明,确定上面两种推导方法得到的结果是一致的。具体地,我们要证明
∑
i
=
0
α
2
−
1
(
−
1
)
i
(
α
2
−
1
i
)
/
(
α
1
+
i
)
=
B
(
α
1
,
α
2
)
\sum_{i=0}^{\alpha_2-1}(-1)^i\begin{pmatrix}\alpha_2-1\\i \end{pmatrix}/(\alpha_1+i)=\Beta(\alpha_1,\alpha_2)
∑i=0α2−1(−1)i(α2−1i)/(α1+i)=B(α1,α2)这个等式是成立的,证明过程如下:
证明过程需要用到一个积分等式(该等式可从参考文献[1]中找到):
∫
0
∞
(
1
−
e
−
x
/
β
)
α
−
1
e
−
μ
x
d
x
=
β
B
(
β
μ
,
α
)
,
(
R
e
β
>
0
,
R
e
α
>
0
,
R
e
μ
>
0
)
(4)
\int_0^\infty(1-e^{-x/\beta})^{\alpha-1}e^{-\mu x}dx=\beta \Beta(\beta\mu,\alpha),({\rm Re\beta>0,{\rm Re}\alpha>0,{\rm Re}\mu>0})\tag{4}
∫0∞(1−e−x/β)α−1e−μxdx=βB(βμ,α),(Reβ>0,Reα>0,Reμ>0)(4)
根据(4),我们令
β
=
1
\beta=1
β=1,则有
∫ 0 ∞ ( 1 − e − x ) α − 1 e − μ x d x = ∫ 0 ∞ ∑ i = 0 α − 1 ( − 1 ) i ( α − 1 i ) e − ( i + μ ) x d x = ∑ i = 0 α − 1 ( − 1 ) i ( α − 1 i ) ∫ 0 ∞ e − ( i + μ ) x d x = ∑ i = 0 α − 1 ( − 1 ) i ( α − 1 i ) / ( i + μ ) = B ( μ , α ) (5) \begin{equation} \begin{aligned} \int_0^\infty(1-e^{-x})^{\alpha-1}e^{-\mu x}dx&=\int_0^\infty \sum_{i=0}^{\alpha-1}(-1)^i \begin{pmatrix}\alpha-1\\i\end{pmatrix}e^{-(i+\mu)x}dx\\ &=\sum_{i=0}^{\alpha-1}(-1)^i \begin{pmatrix}\alpha-1\\i\end{pmatrix}\int_0^\infty e^{-(i+\mu)x}dx\\ &=\sum_{i=0}^{\alpha-1}(-1)^i \begin{pmatrix}\alpha-1\\i\end{pmatrix}/(i+\mu)\\ &=\Beta(\mu,\alpha) \end{aligned} \end{equation}\tag{5} ∫0∞(1−e−x)α−1e−μxdx=∫0∞i=0∑α−1(−1)i(α−1i)e−(i+μ)xdx=i=0∑α−1(−1)i(α−1i)∫0∞e−(i+μ)xdx=i=0∑α−1(−1)i(α−1i)/(i+μ)=B(μ,α)(5)
至此,得证。
[1] Gradshteyn, I. S. and Ryzhik, L.M. “Table of Integrals, Series, and Products (6th ed)”, New York: Academic Press, 2000, pp. 331 and 899.