UA MATH564 概率论II 连续型随机变量1
UA MATH564 概率論II 連續型隨機變量1
- 隨機變量的變換
- 一元隨機變量的變換
- 多元隨機變量的變換
- 均勻分布與Pareto分布
- 離散的均勻分布
- 連續的均勻分布
- Zeta分布
- Pareto分布
隨機變量的變換
一元隨機變量的變換
假設XXX為分布函數為FXF_XFX?的一元隨機變量,X∈DXX \in \mathbb{D}_XX∈DX?,隨機變量Y=g(X)Y=g(X)Y=g(X),ggg為有界連續函數,則
FY(y)=P(Y≤y)=P(g(X)≤y)=P(X∈g?1(Y≤y))F_Y(y) = P(Y \le y) = P(g(X) \le y) = P(X \in g^{-1}(Y \le y)) FY?(y)=P(Y≤y)=P(g(X)≤y)=P(X∈g?1(Y≤y))
當ggg不是單調函數時需要按這個一般性的方法計算。假設ggg為單調遞增的函數,定義h=g?1h=g^{-1}h=g?1,
FY(y)=P(X≤h(y))=FX(h(y))fY(y)=fX(h(y))h′(y)F_Y(y) = P(X \le h(y)) = F_X(h(y)) \\ f_Y(y) = f_X(h(y))h^{'}(y) FY?(y)=P(X≤h(y))=FX?(h(y))fY?(y)=fX?(h(y))h′(y)
假設ggg為單調遞減的函數,則
FY(y)=P(X>h(y))=1?FX(h(y))fY(y)=?fX(h(y))h′(y)F_Y(y) = P(X > h(y)) = 1- F_X(h(y)) \\ f_Y(y) = -f_X(h(y))h^{'}(y) FY?(y)=P(X>h(y))=1?FX?(h(y))fY?(y)=?fX?(h(y))h′(y)
綜合這兩個結果,當ggg單調時
fY(y)=fX(h(y))∣h′(y)∣f_Y(y) = f_X(h(y))|h^{'}(y)| fY?(y)=fX?(h(y))∣h′(y)∣
多元隨機變量的變換
假設XXX為分布函數為FXF_XFX?的多元隨機變量,X∈DXX \in \mathbb{D}_XX∈DX?,隨機變量Y=g(X)Y=g(X)Y=g(X),Y∈DYY \in \mathbb{D}_YY∈DY?,ggg為有界連續函數,且Jacobi行列式Jg≠0Jg \ne 0Jg?=0,定義h=g?1h=g^{-1}h=g?1,根據FXF_XFX?的歸一化條件
∫DXfX(x)dx=1\int_{\mathbb{D}_X} f_X(x) dx = 1 ∫DX??fX?(x)dx=1
根據積分換元公式,等式左邊等于
∫DYfX(h(y))∣dxdy∣dy=∫DYfX(h(y))∣Jh(y)∣dy=1=∫DYfY(y)dy\int_{\mathbb{D}_Y} f_X(h(y)) |\frac{dx}{dy}|dy = \int_{\mathbb{D}_Y} f_X(h(y)) |Jh(y)|dy =1= \int_{\mathbb{D}_Y} f_Y(y) dy ∫DY??fX?(h(y))∣dydx?∣dy=∫DY??fX?(h(y))∣Jh(y)∣dy=1=∫DY??fY?(y)dy
因此
fY(y)=fX(h(y))∣Jh(y)∣f_Y(y) = f_X(h(y)) |Jh(y)| fY?(y)=fX?(h(y))∣Jh(y)∣
均勻分布與Pareto分布
離散的均勻分布
古典概型中,基本事件數量有限,且發生的可能性是均等的。這個假設可以用離散的均勻分布來描述。假設樣本空間為Ω={w1,w2,...,wN}\Omega=\{w_1,w_2,...,w_N\}Ω={w1?,w2?,...,wN?},隨機變量X:wj→jX:w_j \to jX:wj?→j的取值為j∈{1,2,...,N}j \in \{1,2,...,N\}j∈{1,2,...,N},則X的分布列(mass function)為
fX(j)=P(X=j)=1Nf_X(j)=P(X=j)=\frac{1}{N} fX?(j)=P(X=j)=N1?
X的概率生成函數(Probability Generating Function,PGF)為
ρX(z)=E(zX)=∑j=1NzjfX(j)=1N∑j=1Nzj=z?zN(1?z)N=1N∑i=kN?1zi\rho_X(z) = E(z^X)=\sum_{j=1}^{N} z^j f_X(j) = \frac{1}{N}\sum_{j=1}^{N} z^j =\frac{z-z^N}{(1-z)N} = \frac{1}{N} \sum_{i=k}^{N-1} z^i ρX?(z)=E(zX)=j=1∑N?zjfX?(j)=N1?j=1∑N?zj=(1?z)Nz?zN?=N1?i=k∑N?1?zi
X的均值和方差為
EX=1N∑j=1Nj=N+12VarX=1N∑j=1Nj2?(N+12)2=(N?1)(N+1)12EX = \frac{1}{N}\sum_{j=1}^{N} j = \frac{N+1}{2} \\ VarX=\frac{1}{N}\sum_{j=1}^{N} j^2-(\frac{N+1}{2})^2 = \frac{(N-1)(N+1)}{12} EX=N1?j=1∑N?j=2N+1?VarX=N1?j=1∑N?j2?(2N+1?)2=12(N?1)(N+1)?
根據PGF的性質
E(X)k=1N∑j=1N(j)k=ρX(k)(1)E(X)_k = \frac{1}{N}\sum_{j=1}^{N} (j)_k = \rho_X^{(k)}(1) E(X)k?=N1?j=1∑N?(j)k?=ρX(k)?(1)
其中記號(j)k(j)_k(j)k?代表排列數AjkA_j^kAjk?。對數列ana_nan?引入(向前)差分運算
Δ+an=an+1?an\Delta_{+} a_n = a_{n+1} - a_n Δ+?an?=an+1??an?
則(向前)差分的前N項和為
∑n=1NΔ+an=∑n=1N(an+1?an)=aN+1?a1\sum_{n=1}^{N} \Delta_{+} a_n = \sum_{n=1}^{N} (a_{n+1} - a_n) = a_{N+1} - a_1 n=1∑N?Δ+?an?=n=1∑N?(an+1??an?)=aN+1??a1?
考慮記號(i)k(i)_k(i)k?關于iii的(向前)差分
Δ+(i)k=(i+1)k?(i)k=(i+1)(i)(k?1)?(i)(k?1)(i?k+1)=k(i)k?1\Delta_{+} (i)_k = (i+1)_k - (i)_k = (i+1)(i)_{(k-1)} - (i)_{(k-1)} (i-k+1) = k(i)_{k-1} Δ+?(i)k?=(i+1)k??(i)k?=(i+1)(i)(k?1)??(i)(k?1)?(i?k+1)=k(i)k?1?
現在對(j)k(j)_k(j)k?的前N項和進一步化簡
∑j=1N(j)k=k!+∑j=k+1N1k+1Δ+(j)(k+1)=k!+(N+1)k+1?(k+1)!k+1=(N+1)k+1k+1\sum_{j=1}^{N} (j)_k = k!+\sum_{j=k+1}^{N} \frac{1}{k+1} \Delta_{+} (j)_{(k+1)} = k! + \frac{(N+1)_{k+1} - (k+1)!}{k+1} = \frac{(N+1)_{k+1}}{k+1} j=1∑N?(j)k?=k!+j=k+1∑N?k+11?Δ+?(j)(k+1)?=k!+k+1(N+1)k+1??(k+1)!?=k+1(N+1)k+1??
所以
E(X)k=ρX(k)(1)=(N+1)k+1N(k+1)E(X)_k = \rho_X^{(k)}(1) = \frac{(N+1)_{k+1}}{N(k+1)} E(X)k?=ρX(k)?(1)=N(k+1)(N+1)k+1??
連續的均勻分布
連續的均勻分布脫胎于幾何概型的基本假設。假設樣本空間Ω?Rn\Omega \subset R^nΩ?Rn, ∣Ω∣|\Omega|∣Ω∣是Ω\OmegaΩ的Lebesgue測度,則?x∈Ω\forall x \in \Omega?x∈Ω,點xxx被取到的概率相同,從而密度(density)函數為
fX(x)=I(x∈Ω)∣Ω∣f_X(x) = \frac{I(x \in \Omega)}{|\Omega|} fX?(x)=∣Ω∣I(x∈Ω)?
假設Ω?R\Omega \subset RΩ?R,則Ω\OmegaΩ可以由一列幾乎不相交的閉區間表示
Ω=?j=1J[aj,bj]∣Ω∣=∑j=1J(bj?aj)\Omega = \bigcup_{j=1}^{J} [a_j,b_j] \\ |\Omega| = \sum_{j=1}^{J} (b_j-a_j) Ω=j=1?J?[aj?,bj?]∣Ω∣=j=1∑J?(bj??aj?)
假設Ω?R2\Omega \subset R^2Ω?R2,則Ω\OmegaΩ可以由一列幾乎不相交的閉矩形表示
Ω=?j=1JRj∣Ω∣=∑j=1J∣Rj∣\Omega = \bigcup_{j=1}^{J} R_j \\ |\Omega| = \sum_{j=1}^{J} |R_j| Ω=j=1?J?Rj?∣Ω∣=j=1∑J?∣Rj?∣
例如,一元連續均勻分布U[a,b]U[a,b]U[a,b]的密度為
fX(x)=1b?a,x∈[a,b]f_X(x) = \frac{1}{b-a},x \in [a,b] fX?(x)=b?a1?,x∈[a,b]
矩生成函數(Moment Generating Function,MGF)為
MX(t)=E(etX)=∫abetxb?adx=(eb?ea)et(b?a)tM_X(t) = E(e^{tX}) = \int_{a}^{b} \frac{e^{tx}}{b-a} dx = \frac{(e^{b}-e^{a})e^t}{(b-a)t} \\ MX?(t)=E(etX)=∫ab?b?aetx?dx=(b?a)t(eb?ea)et?
Zeta分布
假設X為離散均勻分布,Y=g(X)=XsY=g(X)=X^sY=g(X)=Xs,則
fY(y)∝y?sf_Y(y) \propto y^{-s} fY?(y)∝y?s
不妨假設fY(y)=Cy?sf_Y(y)=Cy^{-s}fY?(y)=Cy?s,
∑y=1∞Cy?s=C∑y=1∞y?s=Cζ(s)=1C=1ζ(s)\sum_{y=1}^{\infty} Cy^{-s}= C \sum_{y=1}^{\infty} y^{-s} = C \zeta(s)=1 \\ C = \frac{1}{\zeta(s)} y=1∑∞?Cy?s=Cy=1∑∞?y?s=Cζ(s)=1C=ζ(s)1?
其中ζ(s)\zeta(s)ζ(s)為Riemann-zeta函數。稱隨機變量Y服從zeta分布,
fY(y)=y?sζ(s)f_Y(y) = \frac{y^{-s}}{\zeta(s)} fY?(y)=ζ(s)y?s?
Pareto分布
假設X~U[0,1]X \sim U[0,1]X~U[0,1],Y=g(X)=XpY=g(X)=X^pY=g(X)=Xp,則
fY(y)=fX(g?1(y))∣h′(y)∣=fX(y?p)py?(p+1)=py?(p+1)f_Y(y) = f_X(g^{-1}(y))|h^{'}(y)|=f_X(y^{-p}) py^{-(p+1)} = py^{-(p+1)} fY?(y)=fX?(g?1(y))∣h′(y)∣=fX?(y?p)py?(p+1)=py?(p+1)
稱隨機變量Y服從Pareto分布。
總結
以上是生活随笔為你收集整理的UA MATH564 概率论II 连续型随机变量1的全部內容,希望文章能夠幫你解決所遇到的問題。
- 上一篇: UA MATH575B 数值分析下VI
- 下一篇: UA MATH566 统计理论7 另一个