A compound Poisson convergence theorem (Q1174213)

There are considered sums \(\sum_{1\leq i\leq n} X_ i\) of independent Bernoulli distributed random variables taking non-negative integer values and the sums are compared with the compound Poisson random variable \(Y=\sum_{1\leq i\leq n}Y_{ni}\), \(Y=\sum_{1\leq j>N(i)}Z_{ij}\), the \(Z_{ij}\) being independent random variables and independent of the random variables \(N(i)\), and \(P[Z_{ij}=k]=a(k)\), \(a(k)>0\). The following theorem is proved. Put \(p_{ni}(k)=P[X_{ni}=k]\), \(p_{ni}=1-p_{ni}(0)\), \(\sum_{1\leq i\leq n}p_{ni}=\mu\) and suppose that \(p_{ni}(k)/p_{ni}=a(k)\) for all \(i\) and \(\mu_ n=\mu+O(1/n]\) with \(\mu>0\). Then \[ \lim_{n\to\infty} \sum_{k\geq 0} h(k)| P[S_ n=k]-P[Y=k]|=0, \] for all non-negative functions \(h\) with finite sum \(\sum_{k\geq0}h(k)P[Y=k]\). This global convergence contains the local convergence \[ \lim_{n\to\infty}P[S_ n=k)=P[Y=k]. \] The corresponding theorem for equally distributed \(X_ n\) was earlier given by Simons and Johnson.