Abstract:We examine the conditions under which the sum of random multiplicative functions in short intervals, given by $\sum_{x<n \leqslant x+y} f(n)$, exhibits the phenomenon of \textit{better than square-root cancellation}. We establish that the point at which the square-root cancellation diminishes significantly is approximately when the ratio $\log\big(\frac{x}{y}\big)$ is around $\sqrt{\log\log x}$. By modeling characters by random multiplicative functions, we give a sharp bound of $\frac{1}{r-1}\sum_{\chi \!\!\!\mod r} \big|\sum_{x<n\leqslant x+y}\chi(n)\big|$, where $r$ is a large prime and $x+y\leqslant r $. This extends the result of Harper \cite{Harper_charac}.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is about the conditions under which the sum of random multiplicative functions in short intervals exhibits better - than - square - root cancellation. Specifically, the author studies the phenomenon that the square - root cancellation is significantly weakened when \(\frac{\log(x)}{\log(y)}\) is close to \(\sqrt{\log \log x}\). By modeling the characters as random multiplicative functions, the author gives a sharp bound:
\[ \frac{1}{r - 1}\sum_{\chi \mod r}\left|\sum_{x < n\leq x + y}\chi(n)\right| \]
where \(r\) is a large prime number and \(x + y\leq r\). This extends Harper's results.
The following is a summary of the key problems and conclusions involved in the paper:
1. **Research Background and Motivation**:
- In recent years, the study of random multiplicative functions has attracted wide attention, especially in the fields of number theory and probability.
- The paper mainly explores the typical size of random multiplicative functions in short intervals, especially whether the sum of these functions exhibits better - than - square - root cancellation.
2. **Main Contributions**:
- Determined the critical point at which the square - root cancellation is significantly weakened, that is, when \(\frac{\log(x)}{\log(y)}\approx\sqrt{\log \log x}\).
- Gave a sharp bound for character sums, extending Harper's results.
3. **Models and Methods**:
- Used Steinhaus and Rademacher random multiplicative functions as models.
- Applied Harper's resilience method, by converting \(L^1\) - norm estimates into the average of a specific squared random Euler integral.
- Used Holder's inequality, Parseval's identity and some number - theory results to derive key propositions.
4. **Main Theorems**:
- **Theorem 1.1**: For a Rademacher or Steinhaus random multiplicative function \(f\), define \(\theta\) such that \(\frac{x}{y}=\log^{\theta}x\), then:
\[
E\left[\left|M(x,y;f)\right|^{2q}\right]\ll\left(y\min\left\{1,\frac{\theta\sqrt{\log \log x}+ 1}{(1 - q)\sqrt{\log \log x}}\right\}\right)^q
\]
In particular, when \(\theta = o\left(\frac{1}{\sqrt{\log \log x}}\right)\), we have \(E\left[\left|M(x,y;f)\right|\right]=o(\sqrt{y})\).
- **Theorem 1.2**: For a Rademacher or Steinhaus random multiplicative function \(f\), when \(\sqrt{\log \log x}\ll\log\left(\frac{x}{y}\right)\), we have:
\[
E\left[\left|M(x,y;f)\right|^{2q}\right]\asymp y^q
\]
- **Theorem 1.3**: Let \(r\) be a large prime number, and \(x,y\) satisfy \(x + y\leq r\) and \(\log\left(\frac{x}{y}\right)\ll\log \log x\), then:
\[
\cdots
\]