Uniformly distributed sums of Bernoulli variables (Q1111311)

The author considers the distribution of a sum \(Y_ i=X_ 1+X_ 2+...+X_ i\) where the random variables \(X_ i\) have a binomial uniform distribution over 0, 1, i.e. having the density function \(P(X_ i=1)=P(X_ i=0)=1/2,\) and looks for the conditions under which \(Y_ i\) has uniform distribution, that is the density function \(q_ i(y)=P\quad (Y_ i=y)\) assumes the values \(q_ i(y)=1/(i+1)\) for all integers \(i>0\) and \(y=0,1,...,i\); obviously the solution does not exist if the \(X_ i\) are supposed independent, but the author proves that if we pose \(p_{i,y}=P(X_ i=1| \quad Y_{i-1}=y)\) then the solution can be obtained if and only if the conditions \(p_{i,y}=(y+1)/(i+1)\) are satisfied. The author considers then some variants of the problem, in particular a generalization to the row sums and to the column sums of an \(m\times n\) matrix.