条件期望例题：帽子匹配问题

条件期望例题：帽子匹配问题

假设n个人丢n个帽子, 丢完后各自随机拿一顶, 拿到自己帽子的退出游戏, 没拿到的继续丢, 再继续拿. 如此往复, 直到所有人均拿到了自己的帽子, 游戏结束.
a. 设游戏开始时一共有n个人, 记$R_n$为游戏进行的总回合数. 求$E[R_n]$
解:
$E[R_n]=n$是一个很合理的假设, 因为无论在每一轮游戏有多少参与者, 平均只有一个人能拿到自己的帽子(容易证明), 而且易知$E[R_1]=1$, 故采用归纳法.
记$X_n$为游戏的第一回合中, 拿到自己帽子退出游戏的人数, 且设在$k=1,...,n-1$时,$E[R_k]=k$, 根据条件期望公式,则有
E [ R n ] = E [ E [ R n ∣ X n ] ] = ∑ i = 0 n E [ R n ∣ X n = i ] ⋅ P { X n = i } = ∑ i = 0 n ( 1 + E [ R n − i ] ) ⋅ P { X n = i } = 1 + E [ R n ] ⋅ P { X n = 0 } + ∑ i = 1 n E [ R n − i ] ⋅ P { X n = i } ↓ 因为 i = { 1 , . . . , n } , 所以 0 ≤ n − i < n , 根据归纳法       E [ R n − i ] = n − i = 1 + E [ R n ] ⋅ P { X n = 0 } + ∑ i = 1 n ( n − i ) ⋅ P { X n = i } = 1 + E [ R n ] ⋅ P { X n = 0 } + n ( 1 − P { X n = 0 } ) − E [ X n ] = E [ R n ] ⋅ P { X n = 0 } + n ( 1 − P { X n = 0 } ) \begin{split} E[R_n] &= E[E[R_n|X_n]] \\ &=\sum_{i=0}^n E[R_n|X_n = i]\cdot P\{X_n = i\} \\ &= \sum_{i=0}^n (1+E[R_{n-i}])\cdot P\{X_n = i\} \\ &= 1 +E[R_n] \cdot P\{X_n = 0\} +\sum_{i=1}^n E[R_{n-i}]\cdot P\{X_n = i\} \\ &\downarrow 因为i=\{1,...,n\}, 所以0\leq n-i<n, 根据归纳法 \\ & \ \ \ \ \ E[R_{n-i}]=n-i \\ &=1 +E[R_n] \cdot P\{X_n = 0\} +\sum_{i=1}^n (n-i)\cdot P\{X_n = i\} \\ &=1 +E[R_n] \cdot P\{X_n = 0\} +n(1- P\{X_n = 0\}) -E[X_n] \\ &=E[R_n] \cdot P\{X_n = 0\} +n(1- P\{X_n = 0\}) \end{split} E[Rn​]​=E[E[Rn​∣Xn​]]=i=0∑n​E[Rn​∣Xn​=i]⋅P{Xn​=i}=i=0∑n​(1+E[Rn−i​])⋅P{Xn​=i}=1+E[Rn​]⋅P{Xn​=0}+i=1∑n​E[Rn−i​]⋅P{Xn​=i}↓因为i={1,...,n},所以0≤n−i<n,根据归纳法     E[Rn−i​]=n−i=1+E[Rn​]⋅P{Xn​=0}+i=1∑n​(n−i)⋅P{Xn​=i}=1+E[Rn​]⋅P{Xn​=0}+n(1−P{Xn​=0})−E[Xn​]=E[Rn​]⋅P{Xn​=0}+n(1−P{Xn​=0})​
也就是
E [ R n ] = E [ R n ] ⋅ P { X n = 0 } + n ( 1 − P { X n = 0 } ) E[R_n]=E[R_n] \cdot P\{X_n = 0\} +n(1- P\{X_n = 0\}) E[Rn​]=E[Rn​]⋅P{Xn​=0}+n(1−P{Xn​=0})
移项可知
$$E[R_n]=n$$

b. 设游戏开始时一共有$n(n\ge 2)$个人, 记$S_n$为各个参与者选取帽子的次数之和. 求$E[S_n]$
解:
根据条件期望公式
E [ S n ] = E [ E [ S n ∣ X n ] ] = ∑ i = 0 n E [ S n ∣ X n = i ] ⋅ P { X n = i } = ∑ i = 0 n ( n + E [ S n − i ] ) ⋅ P { X n = i } = n + ∑ i = 0 n E [ S n − i ] ⋅ P { X n = i } \begin{split} E[S_n] &= E[E[S_n|X_n]] \\ &=\sum_{i=0}^n E[S_n|X_n = i]\cdot P\{X_n = i\} \\ &= \sum_{i=0}^n (n+E[S_{n-i}])\cdot P\{X_n = i\} \\ &=n+\sum_{i=0}^nE[S_{n-i}]\cdot P\{X_n = i\} \end{split} E[Sn​]​=E[E[Sn​∣Xn​]]=i=0∑n​E[Sn​∣Xn​=i]⋅P{Xn​=i}=i=0∑n​(n+E[Sn−i​])⋅P{Xn​=i}=n+i=0∑n​E[Sn−i​]⋅P{Xn​=i}​

而且易知$E[S_0]=0,E[S_1]=0$ (一个人的时候不选).
将该方程重写做
$$E[S_n]=n+E[S_{n-X_n}]$$

我们这样想, 从第一个回合到最后一个回合, 每个回合中, 都有且只有一个参与者拿到帽子退出游戏, 则从结束往开始推, 最后一个退出的参加了$n$轮选择, 倒数第二个参加了$n-1$轮, 所有人的参与次数相加则为$1+2+\cdots +n=\frac{n(n+1)}{2}$次. 这应该就是答案, 或者和最后的答案有着相似的形式.
不妨设$E[S_n]=an+b{n}^2$, 带入上式, 易知$E[S_n]=n+\frac{1}{2}{n}^2$. 这就是答案了. 为了验证答案的正确性, 用归纳法.
当n=2时, 易知
E [ S 2 ] = 2 + ∑ i = 0 2 E [ S 2 − i ] ⋅ P { X 2 = i } = 2 + E [ S 2 ] ⋅ P { X 2 = 0 } = E [ S 2 ] ⋅ 0.5 + 2 \begin{split} E[S_2]&=2+\sum_{i=0}^2 E[S_{2-i}]\cdot P\{X_2 = i\} \\ &=2+E[S_{2}]\cdot P\{X_2 = 0\} \\ &=E[S_{2}]\cdot 0.5+2 \end{split} E[S2​]​=2+i=0∑2​E[S2−i​]⋅P{X2​=i}=2+E[S2​]⋅P{X2​=0}=E[S2​]⋅0.5+2​
即
$$E[S_2]=4=2+\frac{2^2}{2}$$
设在$k=1,...,n-1$时,$E[S_k]=k+\frac{1}{2}{k}^2$, 且 P { X n = n − 1 } = 0 P\{X_n = n-1\} = 0 P{Xn​=n−1}=0 (不存在只有你没选对的情况), 则
E [ S n ] = n + ∑ i = 0 n E [ S n − i ] ⋅ P { X n = i } = n + E [ S n ] P { X n = 0 } + ∑ i = 1 n E [ S n − i ] ⋅ P { X n = i } = n + E [ S n ] P { X n = 0 } + ∑ i = 1 n [ ( n − i ) + 1 2 ( n − i ) 2 ] ⋅ P { X n = i } = n + E [ S n ] P { X n = 0 } + ( n + n 2 2 ) ( 1 − P { X n = 0 } ) − ( n + 1 ) E [ X n ] + E [ X n 2 ] 2 \begin{split} E[S_n]&=n+\sum_{i=0}^nE[S_{n-i}]\cdot P\{X_n = i\} \\ &=n+E[S_n]P\{X_n = 0\}+\sum_{i=1}^nE[S_{n-i}]\cdot P\{X_n = i\} \\ &=n+E[S_n]P\{X_n = 0\}+\sum_{i=1}^n[(n-i)+\frac{1}{2}(n-i)^2]\cdot P\{X_n = i\} \\ &=n+E[S_n]P\{X_n = 0\} \\ &+(n+\frac{n^2}{2})(1-P\{X_n = 0\}) - (n+1)E[X_n] + \frac{E[X_n^2]}{2} \end{split} E[Sn​]​=n+i=0∑n​E[Sn−i​]⋅P{Xn​=i}=n+E[Sn​]P{Xn​=0}+i=1∑n​E[Sn−i​]⋅P{Xn​=i}=n+E[Sn​]P{Xn​=0}+i=1∑n​[(n−i)+21​(n−i)2]⋅P{Xn​=i}=n+E[Sn​]P{Xn​=0}+(n+2n2​)(1−P{Xn​=0})−(n+1)E[Xn​]+2E[Xn2​]​​

因为$E[X_n]=1,E[X_n^2]=2$
所以
$$\begin{gathered} E[S_n]=n+\frac{n^2}{2} \\ 证明完毕 \end{gathered}$$

c. 每个人都可能选错帽子, 每个人选错帽子的次数也是一个随机变量. 求该随机变量之期望
解:
设$C_j$为第$j$个玩家选帽子选的次数(当然这里假定一个人的时候玩不了游戏, 同样不选), 则$C_j-1$为第j个人选错的次数. 易知
$$\sum _{j=1}^n{C}_j=S_n$$
易知
$$E[C_j]=\frac{E[S_n]}{n}=1+\frac{n}{2},n\ge 2$$

所以每个人平均选错的次数为
$$E[C_j-1]=\frac{n}{2}$$