On the variance of squarefree integers in short intervals and arithmetic progressions (Q2035533)

The main theorem in this paper states that if \(0<\delta<1/100\) then \[ \frac{1}{X}\int_X^{2X} \left|\sum_{X<m\le X+H}\mu^2(m)-\frac{6}{\pi^2}H\right|^2dx= C\sqrt{H}+O_{\delta}(H^{1/2-\delta/16}), \] with an explicitly given constant \(C\), for \(1\le H\le X^{6/11-\delta}\). Moreover, under the Lindelöf Hypothesis, one can improve the range to \(1\le H\le X^{2/3-\delta}\). Previously the above was known only for \(H\le X^{2/9-\delta}\), a result due to \textit{R. R. Hall} [Mathematika 29, 7--17 (1982; Zbl 0502.10027)]. The paper also provides an analogous result for the distribution in congruence classes. Thus if \(q\) is prime and \(0<\delta<1/100\) then \[ \frac{1}{\phi(q)}\sum_{(a,q)=1}\left| \sum_{\substack{m\le x\\ m\equiv a(\mathrm{mod}\; q)}}\mu^2(m)- \frac{6}{\pi^2}\cdot\frac{x}{q}\prod_{p\mid q}\frac{p^2}{p^2-1}\right|^2 \] \[ =C\prod_{p\mid q}\frac{p}{p+2}\cdot\sqrt{x/q} +O_{\delta}((x/q)^{1/2-\delta/16}), \] with the same \(C\) as before, for \(x^{5/11+\delta}\le q\le x\). This range can be extended to \(x^{1/3+\delta}\le q\le x\) under the generalized Lindelöf Hypothesis. It is said that the methods of the paper prove the same results without the restriction to prime \(q\). The above estimate improves results of various earlier authors, in particular of \textit{P. Le Boudec} [Math. Z. 290, No. 1--2, 421--429 (2018; Zbl 1459.11185)]. The proofs begin by writing \(\mu^2(m)=\sum_{d^2\mid m}\mu(d)\). Small \(d\) are treated using Fourier expansions and point counting arguments, while large \(d\) are handled via mean and large-values estimates for Dirichlet polynomials.