Prerequisites
Solution :
At first this problem may seem a bit of complex, but when one get to see the pattern it starts unfolding. For his types of problem, I find a useful technique is to follow the picture, in this case the picture is being provided, (if not draw it yourself !!)
Given, \(X_k\) : # balls in the \(k^{th}\) box at the end of 100 trials.
So, the possible values of \(X_k\) are 0,1,2,.....,100. and the probability that at the jth trial a ball will added to the kth box is \(\frac{1}{10}\) , (why??) .
Now, \(P(X_k=x)\)= \({100 \choose x}\)\( (\frac{1}{5})^{x}\) \( (\frac{4}{5})^{100-x} \) \( x=0,1,......,100\)
Clearly, \( X_k \sim binomial( 100, \frac{1}{5})\), from here one can easily find out the the expectation of \(X_k\). But have you thought like this ??
Now, notice that after every slelection of box, 2 balls are added in the system, so at the end of the 10th trial, there will be 200 balls distributed in the system.
So, \(X_1 +X_2 +.......+ X_{100} =200 \), which implies \( \sum_{k} E(X_k)=200\), due to symmetry \(E(X_k)=E(X_l) \forall k \neq l \), So, \(E(X_k)=20\).
(b) Now this part is the cream of this problem, first notice that number of balls in kth box, is dependent on the number of balls in the (k-2)th and (k+2)th box, and vice versa, So, \(Cov( X_k, X_l)=0\) if \(|k-l|\neq 2\).
So, \(Cov(X_k, X_5)= 0 \forall k \neq 3,5 \&\ 7 \), so we just need to find the \(Cov(X_7,X_5)\) and \(Cov(X_3,X_5)\), and \(Cov(X_5,X_5)=Var(X_5)\) .
Now it is sufficient to find the covariance of any of the the above mentioned covariances, as both are symmetric and identical to each other. But for the finding say \(Cov(X_3,X_5)\), lets look whats happening in each trial more closely,
let, \(X_k= i_{k_1} +i_{k_2}+......+i{k_{100}} \) where , \( i_{k_j} = \begin{cases} 1 & if\ a\ ball\ added\ to\ the\ kth\ box\ at\ the\ jth\ trial\ \\ 0 & otherwise \end{cases}\)
So, clearly, \(P(i_{k_j}=1)=\frac{1}{5} \) ; j=1,2,....,100.
So, \(Cov(X_3,X_5)=Cov( i_{3_1}+i_{3_2}+.....+i_{3_{100}},i_{5_1}+i_{5_2}+....+i_{5_{100}})=\sum_{j=1}^{100} Cov(i_{3_j},i_{5_j})\), [\(Cov(i_{3_j},i_{5_j*})=0 \forall j\neq j*\), why ?? ].
So, \(Cov(X_3,X_5)= 100 Cov(i_{3_1},i_{5_1})=100( E(i_{3_1}i_{5_1})-E(i_{3_1})E(i_{5_1}))=100(P(i_{3_1}=1, i_{5_1}=1)-P(i_{3_1}=1)P(i_{5_1}=1))=100(\frac{1}{10}- \frac{1}{5}\frac{1}{5})=6\).
similarly, \(Cov(X_7,X_5)=6\) also, and its easy to find \(Var(X_5)\) , so I leave it as an exercise. So, \(Cov(X_k,X_5)= \begin{cases} 6 & k=3,7 \\ Var(X_5) & k=5 \\ 0 & k\neq 3,5 \or\ 7\end{cases}\). Hence we are done !!