Counting sign changes of partial sums of random multiplicative functions

From MaRDI portal





We say that \(f\) is a Rademacher random multiplicative function if it is multiplicative, has support on the squarefree integers, and at primes the values \((f(p))_p\) are given by i.i.d. Rademacher variables, that is, \(f(p)\) takes each choice \(\pm 1\) with probability \(1/2\).\N\NUsually one is interested in the partial sums \(\sum_{n\leq x}f(n)\) since this can serve as an heuristic to guess the behaviour of the Mertens function -- the partial sums of the Möbius function, as proposed by \textit{A. Wintner} [Duke Math. J. 11, 267--275 (1944; Zbl 0060.10510)]. Moreover, from a probabilistic point of view, one can think \(\sum_{n\leq x}f(n)\) as an arithmetic random walk.\N\NIn both fields of Analytic Number Theory and of Probability, much research has been dealt to understand sign changes of partial sums or of random walks, but with a different terminology -- in Probability the common terminology is level crossings at \(0\) instead of sign changes.\N\NIn a recent paper [Bull. Lond. Math. Soc. 55, No. 1, 78--89 (2023; Zbl 1535.11130)] by \textit{M. Aymone} et al. sign changes (or level crossings) were studied for the partial sums of a Rademacher random multiplicative function. Standard probability techniques for the simple symmetric random walk does not work here since the sequence \((f(n))_n\) lacks independence due to multiplicativity. Neither the classical Landau's oscillation Theorem from Analytic Number Theory apply here since the Dirichlet series of \(f\) has a natural boundary at \(\sigma=1/2\), see [\textit{A. Wintner}, Duke Math. J. 11, 267--275 (1944; Zbl 0060.10510)]. Taking this into account, Heap, Zhao and the reviewer established that \N\[\N\lim_{\sigma\to1/2^+}\int_1^\infty\left( \sum_{n\leq x}f(n)\right)\frac{dx}{x^{1+\sigma}}=0,\,a.s. \N\]\Nwhile \N\[\N\lim_{\sigma\to1/2^+}\int_1^\infty\left| \sum_{n\leq x}f(n)\right|\frac{dx}{x^{1+\sigma}}=\infty,\,a.s\N\]\Nand these capture infinite sign changes, a.s. In a probabilistic terminology, one would say that this random walk is recurrent.\N\NIn the paper under review, the authors are interested in to make it quantitative this number of sign changes. This is a very interesting question, since on the one hand, the simple symmetric random walk has number of sign changes up to \(x\) at the level of \(x^{1/2+o(1)}\) a.s., while for the Möbius walk, a result of \textit{J. Kaczorowski} and \textit{J. Pintz} [Acta Math. Hung. 48, 173--185 (1986; Zbl 0613.10036)] states that the Mertens function has at least \(c\log x\) sign changes up to \(x\) for some explicit constant \(c>0\). It remains a great challenge to determine the order of magnitude of sign changes of the Mertens function, and as far as the reviewer is aware, there are no conjectures predicting of what this magnitude should be.\N\NHere the authors prove that the number of sign changes of \(\sum_{n\leq u}f(n)\) in the interval \(u\leq x\) is at least \((\log\log\log x)^{1/2-\epsilon}\) a.s. Their method consists in to make it quantitative the limits stated above to detect local sign changes. One key new ingredient is a result of \textit{A. J. Harper} [Ann. Appl. Probab. 23, No. 2, 584--616 (2013; Zbl 1268.60075)] on the supremum over \(t\) of \(\sum_{p}\frac{f(p)}{p^{\sigma+it}}\) for \(\sigma\) close to \(1/2\) and \(t\) in a short interval depending on \(\sigma\).











This page was built for publication: Counting sign changes of partial sums of random multiplicative functions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6917979)