Convergence to stable laws and the modeling of them (Q792688)

The first part of the paper arrives at a result stating that if a random variable \(\xi\) has, in a neighborhood of 0, a density equal to \(c| x|^{\gamma}\), where c is constant on \((-\delta_ 1,0)\) and on \((0,\delta_ 2)\), \(-1<\gamma \leq 1\) (if \(\gamma =0\) then c is constant on \((-\delta_ 1,\delta_ 2))\), etc. then 1/\(\xi\) belongs to the domain of attraction of a stable law with exponent \(1+\gamma\) (the other parameter is also explicitly expressed). In the second part, using the author's results in Litov. Mat. Sb. 14, No.1, 165-187 (1974; Zbl 0306.60012); English translation in Lith. Math. Trans. 14, 127-146 (1974), the rapidity of the uniform convergence of the distribution function of the corresponding sum to that of the stable law is studied, in the situation described above for \(\gamma\neq 0\) and then in the case \(\gamma =0\). In the latter case it is proved that if \(X_ 1,...,X_ n,..\). are independent identically distributed random variables having a bounded symmetric density f with \(f(0)>0\), with a derivative f' bounded in a neighborhood of 0, \(F_ n\) is the distribution function of \((X_ 1^{-1}+...+X_ n^{-1})/n\) and G is the Cauchy distribution with parameter \(\pi\) f(0) then, for every \(\epsilon>0\), \(\sup_{x}| F_ n(x)-G(x)| \leq C(\epsilon)n^{- 1+\epsilon}\) with some constant \(C(\epsilon)\). The same is true for \(\epsilon =0\) if \(| f'(x)| \leq K| x|^{\beta}\) for some \(\beta>0\). This sharpens a result of \textit{J. M. Shapiro} in Houston J. Math. 4, 439-445 (1978; Zbl 0393.60024), in which, in the exponent, 1/9 appeared in place of 1. In the third part the obtained results are applied to the computer generation of a random variable having a given stable law and the values obtained by numerical computations of some of the occurring constants are presented.