Title:On the random variable $w_a$ and its connection with the Dirichlet divisor problem

Abstract:In this paper, we study a discrete random variable $w_a$: $$ X = \left\{x_i \;\Big{|}\;\frac{a-i}{a} \right\},\; \mathbb{P}(w_a = x_i) = \frac{1}{a},\;i=1,..,a,\;\;a \in \mathbb{N}^{*}$$ Its characteristics are found. It is shown that the variance of the sum of random variables: $$ Var\Big[\sum_{a=1}^{n}w_a\Big] = \sum_{i=1}^{a}\sum_{j=1}^{b}Cov(w_a,w_b) $$ $$ Cov(w_a,w_b) = \frac{gcd(a,b)^2-1}{12 a b}$$ The variance estimate is given: $$ Var\Big[\sum_{a=1}^{n}w_a\Big] = O\left(n \ln{n} - n + O(n) - H_n^2\right) $$ It is shown that when considering: $$\left \lfloor{\frac{x}{a}}\right \rfloor = \frac{x}{a} - E[w_a]$$ we can get the formula for counting $D(n)$ -- the number of lattice points under the hyperbola $\frac{n}{xy},\; 1 \leq x \leq n,\; 1 \leq y \leq n$: $$D(n) = n \ln{n} + (2\gamma - 1)n + H_{\sqrt{n}} + O(1) $$ satisfied with a standard deviation of the order $O\left(\sqrt{\sqrt{n} \ln{\sqrt{n}} - \sqrt{n} + O(\sqrt{n}) - H_n^2}\right)$ with the hidden constant in main term $ C = \frac{1}{2\pi^2}$. therefore, we can say that this estimate is $ O(n^{\frac{1}{4}+\epsilon})$ since: $$\lim_{n \to \infty}\left(\frac{x^{\frac{1}{4}+\epsilon}}{\sqrt{\sqrt{n} \ln{\sqrt{n}} + \sqrt{n} - H_n^2}}\right) = 0 $$ for some very small $\epsilon$.