Convergence rate of Peng's law of large numbers under sublinear expectations (Q2671646)

The authors give a simple and probabilistic proof for the error rate in a certain law of large numbers (LLN) on sublinear expectation spaces. The paper is self-contained in the sense that required definitions and results of sublinear expectation theory are included. The main theorem gives an upper bound on \[ \bigg|\widehat{\mathbb E}\bigg[\varphi\bigg(\frac{X_1+\cdots+X_n}{n}\bigg)\bigg]-\max_{r\in[\underline{\mu},\bar\mu]}\varphi(r)\bigg| \] for all Lipschitz functions \(\varphi\) and independent random variables \((X_i)_{i\in\mathbb N}\) on a sublinear expectation space \((\Omega,\mathcal H,\widehat{\mathbb E})\) with \(\bar\mu=\widehat{\mathbb E}[X_i]\), \(\underline{\mu}=-\widehat{\mathbb E}[-X_i]\) for all \(i\in\mathbb N\) and \(\sup_{i\in\mathbb N}\widehat{\mathbb E}[|X_i|^{1+\alpha}]<\infty\) for some \(\alpha\in(0,1]\). As a consequence, in case \(\alpha=1\) the authors recover \textit{Y. Song}'s LLN [Probab. Uncertain. Quant. Risk 6, No. 3, 199--212 (2021; Zbl 1497.60028)] for i.i.d.\ random variables with known rate \(C/\sqrt{n}\), where the constant \(C\) can be chosen as the upper standard deviation. Moreover, a LLN by \textit{X. Fang} et al. [Bernoulli 25, No. 4A, 2564--2596 (2019; Zbl 1428.62096)] with rate \(C/n\) can be improved by chossing \(C\) as the upper variance.