In: Math
suppose we take a die with 3 on three sides 2 on two sides and 1 on one side, roll it n times and let Xi be the number of times side i appeared find the conditional distribution P(X2=k|X3=m)
For any roll, probability of appearnce of number 3 , 2 and 1 are
P3 = 3/6 = 1/2
P2 = 2/6 = 1/3
P1 = 1/6
P(X2=k | X3=m) = P(X2 = k, X3 = m) / P(X3 = m)
In 'n' roll, 3 appeared m times, 2 appeared k times, then 1 appeard n - (m + k) times.
thus,
P(X2=k | X3=m) = P(X2 = k, X3 = m, X1 = n-k-m ) / P(X3 = m)
By multinomial distribution,
P(X2 = k, X3 = m, X1 = n-k-m ) = [ n! / (k! * m! * (n-k-m)!) ] * (1/2)m * (1/3)k * (1/6)n-k-m
Now, if X3 appears m times, then any number other than 3 appears for n-m times.
and P3 = 1/2 and 1-P3 = 1 - (1/2) = 1/2
Using binomial distribution,
P(X3 = m) = nCm * (1/2)m * (1/2)n-m = [n! / (m! (n-m)!] (1/2)m * (1/2)n-m
So,
P(X2=k | X3=m) = P(X2 = k, X3 = m, X1 = n-k-m ) / P(X3 = m)
= { [ n! / (k! * m! * (n-k-m)!) ] * (1/2)m * (1/3)k * (1/6)n-k-m } / { [n! / (m! (n-m)!] (1/2)m * (1/2)n-m }
= [(n-m)! / (k! * (n-k-m)!) ] [ (1/3)k * (1/6)n-k-m / (1/2)n-m ]
= n-mCk * 2n-m / (3k * (2 * 3)n-k-m )
= n-mCk * 2k / 3n-m