Random Chowla's conjecture for Rademacher multiplicative functions (Q6918474)

Let \((f(p))_p\) denotes a sequence of i.i.d. random variables indexed by the primes and taking the values \(\pm 1\) with equal probability. A Rademacher random multiplicative function is a sequence of random variables \((f(n))_n\) defined multiplicatively by\N\[\Nf(n):= \begin{cases} \prod_{p|n}f(p) & \text{if \(n\) is squarefree,} \\\N0 & \text{otherwise}. \end{cases}\N\]\NLet \(\mathcal{P}\) be the set of polynomials \(P\in \mathbb{Z}[x]\) satisfying two conditions: (i) \(P\) is a product of at least two distinct linear factors over \(\mathbb{Z}\) or irreducible of degree \(2\), (ii) there is no prime \(p\) for which \(p^2 \mid P(n)\) for all \(n\in \mathbb{N}\). In the paper under review, the authors inspired by the techniques used by \textit{V. Y. Wang} et al. [Bull. Lond. Math. Soc. 56, No. 8, 2718--2726 (2024; Zbl 1558.11072)] prove that if \(f\) is a Rademacher random multiplicative function and \(P\in\mathcal{P}\), then there exists a constant \(\kappa_P>0\) such that\N\[\N\frac{1}{\sqrt{\kappa_P N}}\sum_{n\leq N}f(P(n))\xrightarrow{d}\mathcal{N}(0,1),\N\]\Nas \(N\to\infty\); that is, the partial sums \(\sum_{n\leq N}f(P(n))\), when appropriately normalized, converge in distribution to a standard (real) Gaussian, as \(N\to\infty\). This result confirms a conjecture of \textit{J. Najnudel} [Electron. J. Probab. 25, Paper No. 59, 28 p. (2020; Zbl 1470.11214)], and furthermore motivates the authors to investigate large fluctuations of the partial sums \(\sum_{n\leq N} f(P(n))\). In this paper they only consider the case \(P(n)=n^2+1\), and show that if \(f\) is a Rademacher random multiplicative function, there exist arbitrarily large values of \(N\) such that\N\[\N\Big|\sum_{n\leq N} f(n^2 + 1)\Big| \gg \sqrt{N\log\log N},\N\]\Nalmost surely.