The distribution of square-free numbers (Q5905668)

\textit{H. L. Montgomery} and \textit{R. C. Vaughan} [Recent progress in analytic number theory, Vol. 1, 247-256 (1981; Zbl 0462.10029)] first considered the problem of the estimation of the error term \(\Delta(x)\) associated with the asymptotic formula for the counting function for the set of square-free numbers, under the assumption of the Riemann Hypothesis. They showed that \(\Delta(x)\ll x^{\varphi+\varepsilon}\), with \(\varphi=9/28\) and \(\varepsilon>0\), and this was improved by \textit{S. W. Graham} [J. Lond. Math. Soc., II. Ser. 24, 54-64 (1981; Zbl 0422.10037)] to give \(\varphi=8/25\). Later \textit{R. C. Baker} and \textit{J. Pintz} [Acta Arith. 46, 73-79 (1985; Zbl 0535.10045)] applied Heath- Brown's method to give \(\varphi=7/22\). Using his estimation for dual exponential sums, this value was independently obtained by the author, who now gives the improvement to \(\varphi=17/54\). Montgomery and Vaughan showed that, under the Riemann Hypothesis, \[ \Delta(x)=- \sum_{n\leq y} \mu(n)\psi \left({x \over {n^ 2}}\right)+O(x^{{1\over 2}+\varepsilon} y^{-{1\over 2}} +y^{{1\over 2}+\varepsilon}), \] where \(\mu(n)\) is the Möbius function and \(\psi(x)=x-[x]-{1\over 2}\), and the author gives a detailed proof that the sum here is \(O(x^{\varphi+\varepsilon})\) when \(y=x^{10/27}\). Besides the usual complicated machinery associated with the estimation of exponential sums, the author also incorporates Heath-Brown's method and Vaughan's identity.