Almost sure convergence and square functions of averages of Riemann sums (Q2581115)

Let \(\mathbb{T}:=\mathbb{R}/\mathbb{Z}=[0,1)\) be the torus equipped with the normalized Lebesgue measure \(\lambda\), \(e(x):=\exp(2i\pi x)\), \(e_\ell (x):=e(\ell x)\), \(\ell \in\mathbb{Z}\). For \(f\in L^2(\mathbb{T})\), \(f\sim\sum_{\ell \in\mathbb{Z}}a_\ell e_\ell\), \(\int_\mathbb{T} f\,d\lambda=0\), we consider the Riemann sums \[ R_nf(x):=\frac 1n \sum^n_{j=1}f\left(x+\frac jn \right)=\sum_{\ell\in\mathbb{Z},n|\ell}a_\ell e_\ell(x), \] as well as their averages \[ A_nf(x):=\frac 1n\sum^n_{s=1}R_sf(x),\quad n=1,2, \dots\,. \] The following theorem is proved. If \(f\in L^2(\mathbb{T})\) and the following regularity condition on the Fourier coefficients of \(f\) is satisfied: for any real \(c>0\), \[ \sup_{n\geq 1}\sup_{m\geq cn}|a_m/a_n|= \gamma_f(c)<\infty, \] then \[ \lim_{n\to\infty}A_nf(x)=\int_\mathbb{T} fd \lambda\text{ a.e.} \] Then the structure of the set of averages is analyzed by proving a spectral regularization type inequality, which allows to control the corresponding Littlewood-Paley square function.