In the classical divisor problems, for $k\geq 2$, $\alpha_k$ usually denotes the infimum of real numbers $\sigma<1$ such that $$\Delta_k(x)=\sum_{n\leq x}d_k(n)-\textrm{Res}\left(\frac{\zeta^k(z)x^z}{z},z=1\right)=O(x^{\sigma})$$ as $x\rightarrow\infty$.

As often happens in analytic number theory, one might expect that $\alpha_k$ is also the infimum of $\sigma<1$ such that the integral $$\lim_{T\rightarrow\infty}\int_{\sigma-iT}^{\sigma+iT}\frac{\zeta^k(s)x^sds}{s}$$ exists for all $x>0$.

However, it appears to me that this equivalence is not so easily proved, nor can I find a proof in the literature. Therefore, I would like to ask if this is actually a known theorem or, if not, whether the difference has been been investigated and in which paper(s)?


It is known (see Titchmarsh Chapter 12) that if you define $\gamma_k$ the lower bound of $\sigma > 0, \int_{-\infty}^{\infty}\frac{|\zeta(\sigma+it)|^{2k}}{|\sigma+it|^2}dt < \infty$, then $\frac{k-1}{2k}\le \gamma_k=\beta_k \le \alpha_k$, where $\beta_k$ is the usual mean bound of $\Delta_k$ ( lowest bound of orders for which $\frac{1}{x}\int_{0}^{x}{\Delta_k(y)^2}dy=O_{\epsilon}(x^{2\beta_k+\epsilon})$) and the integral above in the question converges to $\Delta_k(x)$ for $\gamma_k< \gamma <1$.

Since it is known that for example $\gamma_2=\frac{1}{4}, \gamma_3=\frac{1}{3}, \gamma_4=\frac{3}{8}$ (best possible, first two results appearing in Titchmarsh, last in Ivic, chaper 13), while the known respective values for $\alpha$ are still far away from those, even in the simplest cases $k=2,3,4$ the equivalence required is far from being proven and it is equivalent to fully solving the Dirichlet Divisor problem for $k=2,3,4$.

In general, obviously, even less is known about $\gamma_k, \alpha_k$ so the question is on par with Lindelof in many ways (as Lindelof is equivalent to any and all of $\alpha_k \le \frac{1}{2}, \gamma_k \le \frac{1}{2}, \gamma_k = \frac{k-1}{2k}$)

| cite | improve this answer | |
  • $\begingroup$ Thank you, Conrad. I am aware of the known results on $\alpha_k$ and $\gamma_k$ and their ordering, but are you saying that it is a known theorem that the integral in my question converges for $\sigma>\gamma_k$? If so, that would indeed answer my question. $\endgroup$ – Kevin Smith Apr 24 '19 at 23:51
  • $\begingroup$ Yes, it is proven in Titchmarsh, chapter 12 as noted, though the proof is not hard since it uses the uniform convergence to zero of $\frac{\zeta(s)^k}{s}, \gamma +\epsilon \le \sigma <1$ (at infinity ) which follows from the finitness of the integral defining $\gamma$ and some standard stuff plus Cauchy and the convergence of the integral near $1$ $\endgroup$ – Conrad Apr 25 '19 at 0:50

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.