On a claim of Ramanujan in his first letter to Hardy (Q1961990): Difference between revisions

This paper is mainly concerned with the following question: given a control of a non-negative multiplicative function \(f\) over prime powers, find an estimate for its average over all integers less than a given bound. The hypothesis being asymptotic in their nature, so are the results. In a first part, the authors describe and document several problems of this kind, mostly connected with the representation of integers by binary quadratic forms, as a generalization of a claim of Ramanujan concerning sums of two squares. They then turn to three theorems representative of the three approaches to these problems: the approach going via a Tauberian argument which is illustrated by a theorem of \textit{E. Wirsing} [Math. Ann. 143, 75-102 (1961; Zbl 0104.04201)]; the celebrated theorem of \textit{B.~V.~Levin} and \textit{A.~S.~Fainleib} [Russ. Math. Surv. 22, 119-204 (1967); translation from Usp. Mat. Nauk 22, No. 3, 119-197 (1967; Zbl 0204.06502)] whose proof goes by considering \(f(n)\log n\), establishing a kind of differential equation for the sought average and finally deriving consequences of it [see \textit{H. Halberstam} and \textit{H.-E. Richert}, Acta Arith. 18, 243-256 (1971; Zbl 0217.32002) for a closely related use of this process]; the Selberg-Delange method which goes by comparing the associated Dirichlet series to a power of the Riemann zeta-function (when this power is an integer, this is elementary and known as the \textit{convolution} method). The authors complete this arsenal by adding a result similar to the one of B. V. Levin and A. S. Fainleib but somehow simpler in its hypothesis. Their proof relies on this theorem, adding to it a finite preliminary sieving to take care of powers of small primes. They conclude their paper by applications of their theorem to known results, improving some of them.