A note on the convergence of sums of independent random variables (Q1201168)

An old conjecture of Erdős in probabilistic number theory can be transformed to the following purely probabilistic statement: Let \(X_ n\) be a sequence of i.r.v.s in \(R^ 1\) assuming the values \(a_ n\) and 0 with probabilities \(1/p_ n\) and \(1-1/p_ n\), respectively, where \(p_ n\) is the \(n\)th prime number. In order for the d.f. \(F_ N(x)=P\bigl(\sum^ N_{n=1} X_ n\leq x\bigr)\) to be weakly convergent, it is sufficient that \[ \exists\lim\bigl(F_ N(x_ 2)-F_ N(x_ 1)\bigr)>0\quad\text{for two numbers }x_ 1<x_ 2,\;N\to\infty.\tag{E} \] The author discusses this problem under the assumption \[ \sum^ \infty_{n=1} L(X_ n,c_ n)=\infty\quad\text{if}\quad \sum^ \infty_{n=1} c_ n\quad\text{diverges}\tag{I} \] (where \(L\) is the Lévy metric and \(c_ n\) are real constants) and shows that, in order for \(F_ N\) to converge weakly, it is sufficient that \[ \exists L_ i=\lim F_ N(x_ i),\quad N\to\infty,\;i=1,2\quad\text{and}\quad L_ 1\neq L_ 2.\tag{H} \] Moreover, he states that there exists a sequence \(X_ n\) of i.r.v.s satisfying (I) and (E) such that the corresponding sequence \(F_ N\) does not converge.