Critical behavior in almost sure central limit theory (Q2385613)

Let \(X_1\), \(X_2\), be i.i.d random variables with \(EX_1=0\), \(EX_1^2=1\) and let \(S_k=X_1+\cdots+X_k\). The author considers the a.s. convergence of the weighted averages \[ D_N^{-1}\sum_{k=1}^Nd_kI\left\{\frac{S_k}{\sqrt k}\leq x\right\}, \] where \((d_k)\) is a positive sequence with \(D_N=\sum_{k=1}^Nd_k\to\infty\). The almost sure central limit theorem states, the above averages converge a.s. to \(\Phi(x)\) if \(d_k=1/k\) (logarithmic averages) but diverge if \(d_k=1\) (ordinary averages). The author presents a fairly complete solution of the problem for what sequences \((d_k)\) the weighted averages above converge, resp. the corresponding LIL and CLT hold. The presented results show that logarithmic averaging, despite its prominent role in almost sure central limit theory, is far from optimal and considerably stronger results can be obtained using summation methods near ordinary (Cesàro) summation.