Sur la proximité des nombres puissants (Q4811272)

An integer \(n>1\) is said to be powerful if \(p^2| n\) for each of its prime divisors \(p\). The authors' aim is to investigate the existence resp. non-existence of powerful numbers in the intervals \(I_n:=\,]n^2,(n+1)^2[\)\,. They first establish that there exist infinitely many integers \(n\) such that \(I_n\) contains at least \( {9\over{10}} \left({\log n\over\log \log n} \right)^{1/3}\) powerful numbers. They also show that if \(V(N)\) stands for the number of positive integers \(n\leq N\) such that the interval \(I_n\) is free of powerful numbers, then \(V(N)=C_2N + O \left( { N\over\sqrt{\log \log N}} \right)\), where \(C_2= \prod_{m=2}^\infty \left(1- {\mu^2(m)\over m^{3/2}}\right)\approx 0.275\). The ideas of the proofs are as follows. For the first result, given a large positive integer \(k\), the authors first construct a sequence of irrational numbers \(\alpha_j=1/\sqrt{d_j}\), \(j=1,2,\ldots,2k\), where the \(d_j\)'s are carefully chosen squarefree numbers, and then show the existence of positive integers \(p_1,p_2,\ldots,p_{2k}\) and \(q\) such that \(| \alpha_j - p_j/q| \leq q^{-1-1/2k}\), so that \(| q^2-d_jp_j^2| <3\sqrt{d_j} q^{1-1/2k}\) for \(j=1,2,\ldots,2k\). Setting \(n=Dq\), where \(D=\prod_{j=1}^{2k} d_j\), they show that the numbers \(d_jD^2p_j^2\) are all powerful and that at least \(k\) of them belong to \(I_{n-1} \cup I_n\). Finally, using the fact that \(D<(16k^2)^{2k}\) and \(q<\exp\{32(1+o(1))k^3\log k\}\), they conclude that \(k\geq {9\over10} \left( {\log n\over\log \log n} \right)^{1/3}\). For the second result, they first show that, given a large integer \(N\) and letting \({\mathcal A}_m:=\{\lfloor m\sqrt m\cdot \ell \rfloor\leq N: \ell \geq 1\}\), then for any \(Y=Y(N)\) tending to infinity with \(N\), \[ \begin{aligned}{ V(N)& = N-\#\bigcup_{\substack{m\ge 1\\ \mu(m)\neq 0}} {\cal A}_m =N - \# \bigcup_{\substack{1<m\le Y\\ \mu(m)\neq 0}} {\cal A}_m + O \left( {N\over\sqrt Y} \right)\cr& = N - \sum_{j\ge 1} (-1)^{j-1} \sum_{\substack{1<m_1<\ldots <m_j \le Y,\\ \mu(m_i)\neq 0, i=1,\dots,j}} \# \bigcap_{i=1}^j {\cal A}_{m_i} + O \left( { N\over\sqrt {Y}} \right). \cr}\end{aligned} \] They then show that replacing the above sum over \(j\) by \[ N \sum_{j\geq 1} (-1)^{j-1} \sum_{\substack{1<m_1<\dots <m_j \leq Y,\\ \mu(m_i)\neq 0, i=1,\dots,j}} \;\prod_{i=1}^j m_i^{-3/2} \] introduces an error term which is \(\ll N^{1-1/2^Y} \exp\{Y2^{Y+2}\}\). This is done by using results from the theory of uniform distribution of multidimensional sequences modulo 1, which can be found e.g., in Chapter 2 of the book of \textit{L. Kuipers} and \textit{H. Niederreiter} [Uniform distribution of sequences (1974; Zbl 0281.10001)]. With an appropriate choice of the parameter \(Y\), the result follows.