On the value distribution of arithmetic functions (Q1366668)

An integer valued arithmetic function \(g(n)\) is called strictly super-periodic if there is a sequence \(\{r_j\}_{j\geq 1}\) of positive integers \(r_j\geq 2\) such that setting \(R_k=\prod_{j=1}^k r_j\), \[ g(rR_k+j)\geq g((r-1)R_k+j)\quad \text{for all }k\geq 1,\;1\leq r<r_{k+1},\tag{1} \] and if moreover for every \(k\geq 1\) and every \(j\), \(1\leq j\leq R_k\), there is at least one \(r\), \(1\leq r<r_{k+1}\), such that (1) holds with strict inequality. The author says that \(j\), \(1\leq j \leq R_k\), \(k\)-rises at step \(t\), for some \(1\leq t<r_{k+1}\), if \(g(rR_k+j)=g((r-1)R_k+j)+h\) with \(h> 0\). It follows from the strict super-periodicity that each such \(j\) \(k\)-rises at least for one \(t\). For the smallest \(t\) with this property, the author says that \(j\) is \(k\)-lifted at step \(t\) with height \(h>0\). Finally, let \(\lambda(k,t)\) denote the number of \(j\)'s that are \(k\)-lifted at step \(t\). Since for a strictly super-periodic function \(\sum_{t=1}^{r_{k+1}-1} \lambda(k,t)=R_k\) for all \(k\geq 1\), on average \(\lambda(k,t)=R_k/(r_{k+1}-1)\). The main result of the paper says that if \[ \min_{1\leq T<r_{k+1}}T^{-1}\sum_{1\leq t\leq r_{k+1}}T^{-1}\sum_{1\leq t\leq T}\lambda(k,t) \geq (1+o(1))R_k/r_{k+1},\tag{2} \] then \(A_d=o_d(x)\) for \(x\geq 1\), where \(A_d(x)=\text{card}\{n\leq x:g(n)\leq d\}\). It is also shown in the paper that the right hand side of (2) cannot be weakened to \((1-\epsilon)R_k/r_{k+1}\).