Processing math: 100%

Appendix on conditional distributions

This is to give further discussion on conditional distributions. The basic discussion can be found at https://gaomj.cn/probability4/

The following definitions and theorems are NOT stated in a completely rigorous style. Thus the rigorous statements may be referred to other advanced materials on probability theory.

Suppose we have a random vector (X,Y)(X,Y) with joint distribution measure PX,Y\mathbb P_{X,Y}. We first define the marginal distribution:

Definition: Define the projection πX:(x,y)x\pi_X:(x,y)\mapsto x, then the marginal distribution of XX can be defined as the pushforward measure of PX,Y\mathbb P_{X,Y}:
PX:=PX,YπX1.\mathbb P_X:=\mathbb P_{X,Y}\circ\pi_X^{-1}.
With this definition, we have by the property of pushforward:
P(XEx)=πX1(Ex)dPX,Y=ExdPX.\mathbb P(X\in E_x)=\int_{\pi^{-1}_X(E_x)}\mathrm d\mathbb P_{X,Y}=\int_{E_x}\mathrm d\mathbb P_X.Let κY,X\kappa_{Y,X} be a regular conditional distribution of YY given XX such that κY,X(x,)=P(X=x)\kappa_{Y,X}(x,\cdot)=\mathbb P(\cdot\mid X=x). Then,

Theorem:

1.
PX,Y(E)=κ(x,πX1(x)E)dPX.\mathbb P_{X,Y}(E)=\int\kappa(x,\pi_X^{-1}(x)\cap E)\, \mathrm d\mathbb P_X.

2.
f(x,y)PX,Y(dx,dy)=dPXf(x,y)κY,X(x,dy).\int f(x,y)\, \mathbb P_{X,Y}(\mathrm dx,\mathrm dy)=\int\mathrm d\mathbb P_X\int f(x,y)\, \kappa_{Y,X}(x,\mathrm dy).

3.
PY(Ey)=Ω1κY,X(x,Ey)dPX.\mathbb P_Y(E_y)=\int_{\Omega_1}\kappa_{Y,X}(x,E_y)\, \mathrm d\mathbb P_X.

Therefore, it can often be seen that PX,Y=PXPYX\mathbb P_{X,Y}=\mathbb P_X\mathbb P_{Y|X}.

The following is their informal proof. (1) We write
P(E)=dPX,YP((x,y)E)=Ω1d(PX,YπX1)κ(x,πX1(x)E)=Ω1dPXκ(x,πX1(x)E).P(E)=dPX,YP((x,y)E)=Ω1d(PX,YπX1)κ(x,πX1(x)E)=Ω1dPXκ(x,πX1(x)E).\begin{aligned} \mathbb P(E)&=\int \mathrm d\mathbb P_{X,Y}\, \mathbb P((x,y)\in E)\\ &=\int_{\Omega_1}\mathrm d(\mathbb P_{X,Y}\circ\pi_X^{-1})\, \kappa(x,\pi_X^{-1}(x)\cap E)\\ &=\int_{\Omega_1}\mathrm d\mathbb P_X\, \kappa(x,\pi_X^{-1}(x)\cap E). \end{aligned}\begin{aligned} \mathbb P(E)&=\int \mathrm d\mathbb P_{X,Y}\, \mathbb P((x,y)\in E)\\ &=\int_{\Omega_1}\mathrm d(\mathbb P_{X,Y}\circ\pi_X^{-1})\, \kappa(x,\pi_X^{-1}(x)\cap E)\\ &=\int_{\Omega_1}\mathrm d\mathbb P_X\, \kappa(x,\pi_X^{-1}(x)\cap E). \end{aligned}(2) We derive it from (See Theorem 24 in https://gaomj.cn/probability4/#sec:2.3)
f(Y(ω))P(dω)=P(dω)f(y)κY,σ(X)(ω,dy).\int f(Y(\omega))\, \mathbb P(\mathrm d\omega)=\int \mathbb P(\mathrm d\omega)\int f(y)\kappa_{Y,\sigma(X)}(\omega,\mathrm dy).Then
f(X(ω),Y(ω))P(dω)=P(dω)f(X(ω),y)κY,σ(X)(ω,dy),f(x,y)PX,Y(dx,dy)=dPXf(x,y)κY,X(x,dy).f(X(ω),Y(ω))P(dω)=P(dω)f(X(ω),y)κY,σ(X)(ω,dy),f(x,y)PX,Y(dx,dy)=dPXf(x,y)κY,X(x,dy).\begin{aligned} \int f(X(\omega),Y(\omega))\, \mathbb P(\mathrm d\omega)&=\int \mathbb P(\mathrm d\omega)\int f(X(\omega),y)\kappa_{Y,\sigma(X)}(\omega,\mathrm dy), \\ \int f(x,y)\, \mathbb P_{X,Y}(\mathrm dx,\mathrm dy)&=\int\mathrm d\mathbb P_X\int f(x,y)\kappa_{Y,X}(x,\mathrm dy). \end{aligned}\begin{aligned} \int f(X(\omega),Y(\omega))\, \mathbb P(\mathrm d\omega)&=\int \mathbb P(\mathrm d\omega)\int f(X(\omega),y)\kappa_{Y,\sigma(X)}(\omega,\mathrm dy), \\ \int f(x,y)\, \mathbb P_{X,Y}(\mathrm dx,\mathrm dy)&=\int\mathrm d\mathbb P_X\int f(x,y)\kappa_{Y,X}(x,\mathrm dy). \end{aligned}

(3) With the result of (2), we have
P(YEy)=yEydPX,Y=dPXEyκY,X(x,dy)&=dPXκY,X(x,Ey).P(YEy)=yEydPX,Y=dPXEyκY,X(x,dy)&=dPXκY,X(x,Ey).\begin{aligned}\mathbb P(Y\in E_y)&=\int_{y\in E_y}\mathrm d\mathbb P_{X,Y}=\int\mathrm d\mathbb P_X\int_{E_y}\kappa_{Y,X}(x,\mathrm dy)\&=\int\mathrm d\mathbb P_X\, \kappa_{Y,X}(x,E_y).\end{aligned}\begin{aligned}\mathbb P(Y\in E_y)&=\int_{y\in E_y}\mathrm d\mathbb P_{X,Y}=\int\mathrm d\mathbb P_X\int_{E_y}\kappa_{Y,X}(x,\mathrm dy)\&=\int\mathrm d\mathbb P_X\, \kappa_{Y,X}(x,E_y).\end{aligned}


评论

发表回复

您的邮箱地址不会被公开。 必填项已用 * 标注