The worst sums in ergodic theory (Q5954528)

It is known that there are `bad' integer sequences \((m_k)\) for the pointwise ergodic theorem; bad in the sense that the ergodic averages \((1/n)\sum_{k=1}^{n}f(T^{m_k}x)\) may not converge a.e. (or even stay bounded a.e.) for a probability preserving transformation \(T\) and \(f\in L^p\), \(1\leq p\leq\infty\). Here this question is refined in the following direction: for which rates \((L_n)\) and values of \(p\) will \((1/L(n))\sum_{k=1}^{n}f(T^{m_k}x)\) converge to \(0\) a.e.? Among the kind of results obtained are that if \(L_n=o(n(\log n)^{1/p})\) with \((1/n)L_n\) decreasing, then there is a sequence \((m_k)\) and positive function \(f\in L^p\), \(p<\infty\), with \[ \sup_{n\geq 1}\frac{1}{L_n} \sum_{k=1}^{n}f(T^{m_k}x)= \infty \] a.e. More technical results are found using large deviation techniques.