A random approach to the Lebesgue integral (Q2467772): Difference between revisions

Let \(\mathcal P=(I_k)_{k=1}^n\) be a partition of the unit interval and let \(f\) be Lebesgue measurable there. A random Riemann sum is constructed by looking at the evaluation points \(t_k\in I_k\) in the usual Riemann sum as random variables over \(I_k\) with uniform distribution. One then gets the random sum \(S_{\mathcal P}(f)=\sum_{k=1}^n f(t_k)| I_k| \). It is now easy to see that the expectation of \(S_{\mathcal P}(f)\) is the Lebesgue-integral of \(f\) over the unit interval. Now let the size, \(\delta=\max_k | I_k| \), of \(\mathcal P\) tend to zero. It turns out that the following weak law of large numbers is true: Assume \(f\) has a Lebesgue integral \(\int f\) over the unit interval. Then the random Riemann sums of a sequence of partitions whose sizes tend to 0 converges in probability to \(\int f\). The corresponding result for almost sure convergence is not true and much harder to analyze. Almost sure convergence of random Riemann sums depends on the way the sizes of the sequence of the partitions tend to 0. The author proves that if \(f\in L_p\), \(p>1\) and the sizes \(\delta_n\) of the sequence \({\mathcal P}_n\) tend to 0 as an \(\ell_{p-1}\) sequence, then almost sure convergence of the random Riemann sums towards \(\int f\) holds. In the other direction the author proves that for every \(p>1\) and every \(\epsilon>0\) there is an \(f\in L_{p-\varepsilon}\) and a sequence \(({\mathcal P}_n)\) with \((\delta_n)\in \ell_{p-1}\) such that \(P(S_{{\mathcal P}_n}(f)-\int f)=0\)! This is not for the first time that random Riemann sums have been studied, as remarked by the author at the end of the paper. \textit{J. C. Kieffer} and \textit{C. V. Stanojevic} [Proc. Am. Math. Soc. 85, 389--392 (1982; Zbl 0497.28007)] studied almost sure convergence of random Riemann sums, but with other demands on how to pick evaluation points. Further path to the history of randomized Riemann integral can be found via the review of that paper in Math Reviews, see [MR0656109 (83h:26015)]. \textit{C. S. Kahane} [Math. Jap. 38, No. 6, 1073--1076 (1993; Zbl 0795.28004)] and \textit{A. R. Pruss} [Proc. Am. Math. Soc. 124, No. 3, 919--929 (1996; Zbl 0843.60031)] considered almost sure convergence with uniform partitions and revealed the phenomenon \(f\in L_2\leftrightarrow (\delta_n)\in \ell_1\).