On a centered Poisson approximation (Q1567895)

Let \(F\) be a distribution concentrated on \(\{0,1,\dots\}\) and having all finite factorial cumulants \(\Gamma_k\), defined by the equality \[ \ln\Biggl( \sum^\infty_{k=0} e^{ikt}F(\{k\})\Biggr)= \sum^\infty_{k=1} {\Gamma_k\over k!} (e^{it}- 1)^k,\qquad t\in R^1, \] such that for some \(\Delta\geq 1\), \[ |\Gamma_k|\leq k! \Gamma_1/\Delta^{k-1},\qquad k= 2,3,\dots, \] \[ \Pi_{\lambda, a}(\{k\})= e^{-\lambda} {\lambda^{k- a}\over(k- a)!} \mathbf{1}_{\{k\geq a\}},\qquad k= a,a+1,\dots,\quad a= 0,1,\dots,\quad \lambda> 0. \] Approximations of integral, local or large deviations types of the distribution \(F\) by means of \(\Pi_{\lambda,a}\) are proved. A typical result is the following statement. If \(\Delta\geq 5\), there exist absolute constants \(C_1> 0\), \(C_2> 0\), such that \[ \|F-\Pi_{\lambda,a}\|\leq C_1{|\{- \Gamma_2\}|\over \Gamma_1} \min(1, \Gamma_1)+ C_2{1\over \Delta\sqrt{\Gamma_1}} \min(1,\Gamma^{3/2}_1), \] where \(\lambda= \Gamma_1+ \Gamma_2+ \{-\Gamma_2\}\), \(a= [-\Gamma_2]\), symbols \(\{c\}\), \([c]\) mean the fractional and integral parts of a number \(c\), respectively, and \(\|\mu\|\) denotes the total variation norm of the signed measure \(\mu\).