Equivalent conditions of complete convergence and Marcinkiewicz-Zygmund-type strong law of large numbers for i.i.d. sequences under sub-linear expectations (Q6111807)

The classical probability limit theory in the case of model uncertainty such as statistics, measure of risk, superhedging in finance, non-linear stochastic calculus is no longer applicable. \textit{S. Peng} [Nonlinear expectations and stochastic calculus under uncertainty. With robust CLT and G-Brownian motion. Berlin: Springer (2019; Zbl 1427.60004)] extended the classical linear expectation spaces to sub-linear expectation spaces in attempting to solve this problem of classical probability limit theory in the case of model uncertainty. The complete convergence is one of the important convergence problems in limit theories. \textit{Vu T. N. Anh} et al. [J. Theor. Probab. 34, No. 1, 331--348 (2021; Zbl 1469.60096)] examined four equivalent conditions of complete convergence for maximum of partial sums and SLLN for sequences of negatively associated and i.i.d random variables in a classical probability space. The paper under review follows the ideas of the above paper of Vu et al. [loc. cit.] to extend the equivalent conditions of complete convergence for maximal weighted sums and the Marcinkiewicz-Zygmund-type SLLN for sequences of i.i.d. random variables in sub-linear expectation spaces by using slowly varying functions, monotone truncation functions and the Kolmogorov inequality. Section 2 discusses basic notations and concepts, and collects some results for sub-linear expectation spaces. The authors prove Lemma 2.14 as a remainder estimate for a slowly varying function \(L\), \[ \sum_{j=k}^{\infty} 2^{jr}L^q(\varepsilon 2^j) \le c 2^{kr}L^q(\varepsilon 2^k),\] for every \(r<0, \varepsilon >0\) and \(q\in \mathbb R\). Section 3 is devoted to the main result, Theorem 3.1 and its proof. Theorem 3.1 deals with the four equivalent conditions mimic from [Zbl 1469.60096] for a sequence of i.i.d. random variables under \(1\le \alpha \le 2\), upper expectation \(E[X]=0\), lower expectation \(e[X]=0\) and upper probability continuous: (i) The Choquet expectation \(C_V ( (\vert X\vert +A)^{\alpha}L^{\alpha}(\vert X\vert +A))<\infty\). (ii) \(\sum_{n\ge A^{\alpha}} n^{-1}V(\max_{1\le k \le n}\vert \sum_{i=1}^k a_{ni}X_i\vert > \varepsilon b_n) <\infty\), for sequence \(a_{ni}\) with \(\sum_{i=1}^na_{ni}^2 = O (n)\) and \(b_n =n^{1/\alpha}\tilde{L}(n^{1/\alpha})\). (iii) \(\sum_{n\ge A^{\alpha}} n^{-1}V(\max_{1\le k \le n}\vert \sum_{i=1}^k X_i\vert > \varepsilon b_n) <\infty\). (iv) SLLN holds. The proof is given by the following four parts \((i) \to (ii)\), \((ii)\to (iii)\), \((iii) \to (iv)\) and \((iv) \to (i)\). The first part follows from the breaking out two terms on \(\max_{1\le k \le n}\vert X_k\vert >b_n\) and \(\max_{1\le k \le n}\vert \sum_{i=1}^k a_{ni}X_{ni}\vert > \varepsilon b_n\) with lemmas in the previous section on various inequalities and characterizations for slowly varying functions. The second part follows easily by setting \(a_{ni}=1\). The third part follows from the Borel-Cantelli's lemma and basic estimations. The last part follows from an earlier result on the relationship between independence and mutually independence of \textit{Z. Chen} [Sci. China, Math. 59, No. 5, 945--954 (2016; Zbl 1341.60015)], the Borel-Cantelli's lemma for \(B_n=\{\vert X_n\vert +A > \varepsilon_0 b_n\}.\) The paper ends with a special case of Theorem 3.1 for a special slowly varying function \(L(x) =\log^{- 1/\gamma} (x)\) for \(\gamma >0, x\ge 2\).