Lower limits and equivalences for convolution tails (Q879260)

Let \(F\) be a distribution on \(\,[0,\infty )\) with unbounded support and mean value \(a\in (0,\infty ]\). Denote \(\varphi \left( \gamma \right) ,\,\gamma \in \mathbb{R}\), the Laplace transform of \(F\) and \[ \hat{\gamma}=\sup \{\gamma :\varphi \left( \gamma \right) <\infty \}\in [ 0,\infty ],\,\bar{F}(x)\equiv F(x,\infty )>0 \text{ for any }x. \] If \(\hat{\gamma}=0\) the distribution \(F\) is called heavy-tailed, and if \( \hat{\gamma}>0\) it is called the light-tailed. The relations of the tail behavior of the convolution \(F\ast F\) to that of \(F\) are considered. It is proved that for any heavy-tailed distribution \(F\) \[ \liminf\limits_{x\rightarrow \infty}\,(\overline{F\ast F}(x)/F(x))=2. \] If \(\hat{\gamma}\in (0,\infty ]\) so that \(\varphi \left( \hat{\gamma}\right) \in (1,\infty ]\) and if for any fixed \(\gamma >0\) \(\liminf\limits_{x\rightarrow \infty }\frac{\bar{F}\left( x-y\right) }{\bar{F }\left( x\right) }=e^{\gamma y},\) then \(\liminf\limits_{x\rightarrow \infty }\frac{\overline{F\ast F}\left( x\right) }{\bar{F}\left( x\right) }=2\varphi \left( \hat{\gamma}\right) .\) Finally, it is proved that for any distribution on \([0,\infty )\) with unbounded support under condition \(\frac{\overline{F\ast F}\left( x\right) }{\bar{F}\left( x\right) } \rightarrow C\) as \(x\rightarrow \infty ,\) \(C\in (0,\infty ]\) it follows that \(C=2\varphi \left( \hat{\gamma}\right).\) Some characterization of the whole class of heavy-tailed distributions is presented. Several statements on the rations of distribution densities of randomly stopped sums to the density of one summable are proved in the both cases. The authors describe in similar manner the class \(S\) of subexponential distributions and some its subclass denoted by \(S^{\ast }\), as well as the class \(S\left( \gamma \right) ,\,\gamma >0\) of distributions with unbounded support such that \(\varphi \left( \gamma \right) >0,\) any for any fixed \(y\), \(\bar{F}\left( x+y\right) /\bar{F}\left( x\right) \rightarrow e^{\gamma y}\) and \(\overline{F\ast F}\left( x\right) \sim 2\varphi \left( \gamma \right) \bar{F }\left( x\right)\)\,as \(x\rightarrow \infty\). The convolution of two nonidentical distributions, one of which is heavy-tailed is considered as well. The proofs employ real analytic and direct probabilistic methods only.