I would like to know the asymptotic behaviour at large $n$ for $t\in\mathbb{R}$, $t\neq0$ of the following function: \begin{align*} A_n(t)&=\sum_{q=\frac{a}{b}\in \mathbb{Q}^+|\gcd(a,b)=1 \& ab=n}q^{it} \\ &=2^{\omega(n)}\prod_{i=1}^{\omega(n)}\cos(t\log p_i^{v_i}) \end{align*} where we have used the prime number decomposition of $q=\prod_{i}^{\omega(n)}p_i^{v_i}$ and $n=\prod_{i}^{\omega(n)}p_i^{|v_i|}$, and $\omega(q)=\omega(n)$ is their number of distinct prime factors.

On what can I rely to answer such question ? I can only notice: $$|A_n(t)|\leq A_n(0)=2^{\omega(n)}\sim 2^{\log\log n}, \quad n\gg 1$$ The aim is to find information about the convergence of the associated Dirichlet series $\sum_{n\geq 1}\frac{A_n(t)}{n^s}$ where $s\in \mathbb{R}$. The above remark leads to the absolute convergence for $s>1$, can we say better?

EDIT: As pointed out below, I got this expression writing $\frac{|\zeta(s+it)|^2}{\zeta(2s)}$ as a Dirichlet series.

  • 1
    $\begingroup$ Note that $2^{\log\log n} = (\log n)^{\log 2} \not\sim \log n$. Also note that $2^{\omega(n)}$ and $2^{\log\log n}$ are close only for most $n$, not all $n$; and even for typical $n$, the two functions are not really asymptotic to each other (their logarithms are, but that's not sufficient). $\endgroup$ – Greg Martin Apr 7 '18 at 2:10

Your series is never absolutely convergent in any half-plane of the form $\mathrm{Re}\,s>\delta$ with $\delta<1$ and there is even no convergence in the case $\mathrm{Re}\,s=1$. To prove this, let us notice that $A_n(t)$ is a multiplicative function of $n$, therefore we have

$$f(s):=\sum_{n=1}^{+\infty} A_n(t)n^{-s}=\prod_p f_p(s),$$


$$f_p(s)=1+\sum_{n=1}^{+\infty} \frac{2\cos(nt\log p)}{p^{ns}}=1+\frac{1}{p^{s+i t}- 1} +\frac{1}{p^{s-i t}- 1}$$

and the product is taken over all prime numbers.

Therefore, we have


Suppose now that the series converge absolutely for $\mathrm{Re}\,s=1$. Then the function $f(s)$ is bounded on this line, which is certainly not the case, as $f$ has a pole at $s=1+it$, because zeta is non-zero on the edge of critical strip and has a simple pole at $s=1$.

| cite | improve this answer | |
  • $\begingroup$ Thank you for your answer. Your $f(s)$ was indeed my starting point, but I think I was not clear enough, I will add precisions. Actually, I am assuming $s,t\in \mathbb{R}$ and I would like to study the convergence of the Dirichlet series that yields $f(s,t)$ or the asymptotics of $A_n(t)$ for non zero $t$ (to avoid the pole at $s=1$). $\endgroup$ – Alexandre Apr 7 '18 at 11:41
  • $\begingroup$ @Alexandre, the function $f(s)$ has a pole in the point $s=1+it$ for every $t \in \mathbb R$. I think one can also get an asymptotic formula for the partial sums of $A_n(t)$ and prove that there is no convergence in any point of the line $\mathrm{Re}\,s=1$. $\endgroup$ – Asymptotiac K Apr 7 '18 at 12:11
  • $\begingroup$ I would be glad to have some insight on how to get such asymptotic formula. I think we should have at least the absolute convergence of the series for $\Im s =0$ and $\Re s >1$ $\forall t\in \mathbb{R}$, and hopefully the convergence for $\Im s = 0$ and $\forall \Re s$ for $t\neq 0$. Though I am not sure how to prove this. $\endgroup$ – Alexandre Apr 8 '18 at 12:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.