Fix real numbers $\alpha, \beta \geq 0$ with $\alpha$ irrational, and let $N$ be a positive integer. For $x \in \mathbb{R}$, let $\{x\} := x - \lfloor x \rfloor$ denote the fractional part of $x$. Are there any estimates known for the following sums \begin{equation} S_1 := \sum_{n=1}^N \left(\{\alpha n + \beta\} - \dfrac{1}{2}\right) \end{equation} and \begin{equation} S_2 := \sum_{n=1}^N \left(\{\log(\alpha n + \beta)\} - \dfrac{1}{2}\right). \end{equation} Is it true that $S_1 = o(N)$ and $S_2 = o(N)$ as well? What are methods that may be useful for estimating such sums? Thanks a lot.
Sum of fractional parts of linear and logarithmic functions
1
$\begingroup$
calculus
analysis
analytic-number-theory
-
1I would expect $S_1=\frac{N}{2}+o(N)$, though. It may be the case that $S_2=\frac{N}{2}+o(N)$ as well. – 2017-01-22
-
0Can your guess be proven? – 2017-01-22
-
1Well, a heuristic argument is that $\big\{\alpha n +\beta\big\}$ and $\big\{\log(\alpha n+\beta)\big\}$ are (probably) uniformly distributed on $[0,1)$. If this is true, then we may even have the estimate $\frac{N}{2}+\mathcal{O}\big(\sqrt{N}\big)$ for both $S_1$ and $S_2$. – 2017-01-22
-
0Makes a lot of sense. But not true until proven :) – 2017-01-22
-
0Have fixed the post now – 2017-01-22
-
1After a trial with $\alpha=\sqrt{2}$ and $\beta=0$, $\big\{\log(\alpha n+\beta)\big\}$ does not seem uniformly random. But it may still hold that, now with the new definitiosn, $S_2=o(N)$ (although I wouldn't expect $S_2=\mathcal{O}\big(\sqrt{N}\big)$ anymore). I'm too tired to think about $S_1$ now, but it should not be difficult to show $S_1=\mathcal{O}\big(\sqrt{N}\big)$. – 2017-01-22
-
0Regarding $S_2$, since the $\log$ function grows slowly, I feel that $\Delta = S_2(\beta_1) - S_2(\beta_2)$ should be small for any real numbers $\beta_1, \beta_2 \in [0,1]$ – 2017-01-22