A lemma on large deviations of arithmetic functions (Q795869): Difference between revisions

Let \(f\) be an additive arithmetic function. Denote by \(A(x)=\sum_{p\leq x} f(p)p^{-1}\) and \(B(x)=\sum_{p\leq x}f^2(p)p^{-1},\) where \(p\) is prime. Suppose that \[ \mu_x=(\sup_{p\leq x}| f(p)|)/\sqrt{B(x)}\to 0\text{ as }x\to \infty. \] Then it is well known that as \(x\to \infty\), \[ F_x(v)=(1/x) \sum_{\substack{n\le x \\ f(n)- A(x)<v\sqrt{B(x)}}}1\quad \to \quad(1/\sqrt{2\pi}) \int^v_{- \infty}e^{-u^2/2}\, du=G(v). \] In fact one can estimate \(\sup_v | F_x(v)-G(v)|\) in terms of \(\mu_x\). In this paper the author obtains good upper bounds for \(| F_x(v)/G(v)-1\) for values \(v\) satisfying \(| v| \leq T\), where \(T\to \infty\) with \(x\) but \(T=o(\mu_x^{-1})\). Previously Halász had obtained such results for strongly additive functions. A standard tool in obtaining such large deviation theorems is a lemma of Statulevičius. The author obtains improved estimates by establishing a lemma of his own on large deviations. This is accomplished by studying sums of the form \(S=\sum_{\substack{n\leq x \\ n\in M}} z^{f(n)}\), where \(M\) is a set of positive integers and \(z\) is a complex number. Naturally certain restrictions are imposed upon \(S\) and \(f\). The results are too complicated to be stated here.