二维柯西分布(Bivariate Cauchy Distribution):
Box-Muller算法的主要原理就是利用需生成分布的cdf中形如
x
2
+
y
2
x^2+y^2
x2+y2的部分进行三角换元,然后将换元后的联合分布函数积分成
R
R
R和
θ
\theta
θ各自的边缘分布函数,再求反函数,找到两个随机变量的函数,则可以通过这两个函数把均匀分布的随机变量映射到所求分布上。
f
(
x
,
y
;
x
0
,
y
0
,
γ
)
=
1
2
π
[
γ
(
(
x
−
x
0
)
2
+
(
y
−
y
0
)
2
+
γ
2
)
3
2
]
f(x, y ; x_{0}, y_{0}, \gamma)=\frac{1}{2 \pi}[\frac{\gamma}{((x-x_{0})^{2}+(y-y_{0})^{2}+\gamma^{2})^{\frac{3}{2}}}]
f(x,y;x0,y0,γ)=2π1[((x−x0)2+(y−y0)2+γ2)23γ]
设
X
=
x
−
x
0
,
Y
=
y
−
y
0
:
X=x-x_{0}, Y=y-y_{0}:
X=x−x0,Y=y−y0:
则:
f
(
X
,
Y
,
γ
)
=
1
2
π
[
γ
(
X
2
+
Y
2
+
γ
2
)
3
2
]
f(X, Y, \gamma)=\frac{1}{2 \pi}[\frac{\gamma}{(X^{2}+Y^{2}+\gamma^{2})^{\frac{3}{2}}}]
f(X,Y,γ)=2π1[(X2+Y2+γ2)23γ]
令
X
=
R
cos
(
θ
)
,
Y
=
R
sin
(
θ
)
X=R\cos(\theta),Y=R\sin(\theta)
X=Rcos(θ),Y=Rsin(θ),则有:
∫
−
∞
+
∞
∫
−
∞
+
∞
1
2
π
[
γ
(
X
2
+
Y
2
+
γ
2
)
3
2
]
d
X
d
Y
=
∫
−
∞
+
∞
∫
−
∞
+
∞
1
2
π
[
γ
(
R
2
+
γ
2
)
3
2
]
R
d
θ
d
R
=
1
\int_{-\infty}^{+\infty} \int_{-\infty}^{+\infty} \frac{1}{2 \pi}[\frac{\gamma}{(X^{2}+Y^{2}+\gamma^{2})^{\frac{3}{2}}}] d X d Y=\int_{-\infty}^{+\infty} \int_{-\infty}^{+\infty} \frac{1}{2 \pi}[\frac{\gamma}{(R^{2}+\gamma^{2})^{\frac{3}{2}}}] R d \theta d R=1
∫−∞+∞∫−∞+∞2π1[(X2+Y2+γ2)23γ]dXdY=∫−∞+∞∫−∞+∞2π1[(R2+γ2)23γ]RdθdR=1
由此可得
R
R
R 与
θ
\theta
θ 的分布函数
P
R
P_{R}
PR 与
P
θ
P_{\theta}
Pθ :
P
R
(
R
≤
r
)
=
∫
0
2
π
∫
0
r
1
2
π
[
γ
(
R
2
+
γ
2
)
3
2
]
R
d
θ
d
R
=
∫
0
r
[
γ
(
R
2
+
γ
2
)
3
2
]
R
d
R
=
1
−
γ
r
2
+
γ
2
(
r
>
0
)
P
θ
(
θ
≤
ϕ
)
=
∫
0
ϕ
∫
0
+
∞
1
2
π
[
γ
(
R
2
+
γ
2
)
3
2
]
R
d
θ
d
R
=
ϕ
2
π
\begin{aligned} P_{R}(R \leq r)&=\int_{0}^{2 \pi} \int_{0}^{r} \frac{1}{2 \pi}[\frac{\gamma}{(R^{2}+\gamma^{2})^{\frac{3}{2}}}] R\;d \theta d R\\ &=\int_{0}^{r}[\frac{\gamma}{(R^{2}+\gamma^{2})^{\frac{3}{2}}}] R\;d R\\ &=1-{\gamma\over \sqrt{r^2+\gamma^2}}\qquad(r>0)\\ P_{\theta}(\theta \leq \phi)&=\int_{0}^{\phi} \int_{0}^{+\infty} \frac{1}{2 \pi}[\frac{\gamma}{(R^{2}+\gamma^{2})^{\frac{3}{2}}}] R\;d \theta d R\\ &=\frac{\phi}{2 \pi} \end{aligned}
PR(R≤r)Pθ(θ≤ϕ)=∫02π∫0r2π1[(R2+γ2)23γ]RdθdR=∫0r[(R2+γ2)23γ]RdR=1−r2+γ2γ(r>0)=∫0ϕ∫0+∞2π1[(R2+γ2)23γ]RdθdR=2πϕ
对
P
R
P_R
PR求反函数得:
R
=
γ
(
2
x
−
x
2
)
1
2
1
−
x
R=\frac{\gamma(2 x-x^{2})^{\frac{1}{2}}}{1-x}
R=1−xγ(2x−x2)21
设
U
1
,
U
2
U_1,U_2
U1,U2为服从均匀分布的随机变量,则:
R
=
γ
(
2
U
1
−
U
1
2
)
1
2
1
−
U
1
θ
=
2
π
U
2
\begin{aligned} R&=\frac{\gamma(2 U_{1}-U_{1}^{2})^{\frac{1}{2}}}{1-U_{1}} \\ \theta&=2\pi U_{2} \end{aligned}
Rθ=1−U1γ(2U1−U12)21=2πU2
则:
X
=
γ
(
2
U
1
−
U
1
2
)
1
2
1
−
U
1
cos
(
2
π
U
2
)
Y
=
γ
(
2
U
1
−
U
1
2
)
1
2
1
−
U
1
sin
(
2
π
U
2
)
X =\frac{\gamma(2 U_{1}-U_{1}^{2})^{\frac{1}{2}}}{1-U_{1}} \cos (2 \pi U_{2}) \\ Y=\frac{\gamma(2 U_{1}-U_{1}^{2})^{\frac{1}{2}}}{1-U_{1}} \sin (2 \pi U_{2})
X=1−U1γ(2U1−U12)21cos(2πU2)Y=1−U1γ(2U1−U12)21sin(2πU2)