Limit Theorems for the Maximal Random Sums (Q4375252)

The authors consider the product measurable space of a sequence of \(I= \{0,1,\dots,l\}\), indexed by \(1,2,\dots\) and the projection \(\omega_k\) on the \(k\)-th component. A probability \(P\) on it is named ``computable'' if there exists an algorithm which, for every \(x=(x_1,\dots, x_n)\in I^{\{1,\dots, n\}}\) and rational \(\varepsilon >0\), computes a rational number \(r\) with \(| r-P(\Gamma_x)|< \varepsilon\), \(\Gamma_{x}= (\omega_i= x_i\); \(i=1,\dots,n)\). Theorem 1: There exists a computable shift invariant \(P\) such that the convergence in probability of \(\eta_n= (\omega_1+\dots+ \omega_n)/n\) is not ``algorithmically effective'' (i.e. if there is no algorithm giving for all rational \(\varepsilon,\delta >0\) a threshold \(N\) in the corresponding definition). Theorem 2: For every computable shift invariant \(P\) there exists an integrable \(f\) such that \(\eta_n (\omega)\rightarrow f(\omega)\) for every ``\(P\)-typical'' \(\omega\), i.e. for every \(\omega\) in the complementary of the maximal ``effectively \(P\)-null set'' (i.e. such that there exists an algorithm computing, for every rational \(\varepsilon >0\), a sequence \(S\) of \(x\)'s such that the set is included in \(U= \bigcup_{x \in S} \Gamma_x\) and \(P(U)< \varepsilon)\). For the existence of such a set and other details concerning the concepts used, the paper of \textit{A. N. Kolmogorov} and \textit{V. A. Uspenskij} [ibid. 32, No. 3, 389-412 (1987), resp. ibid. 32, No. 3, 425-455 (1987; Zbl 0627.60006)] is quoted. No proofs are given.