2
$\begingroup$

Is there an upper bound for the sum $$ \sum_{k=1}^N \frac{1}{\varphi^{\alpha}(k)} $$ where $\varphi(n)$ is the Euler totient function and $\alpha\geq 1$ a real constant? In particular, I'm interested in the cases $\alpha=1$. I don't want any asymptotic estimate with error term, since in that case there is already a good reference from Sitaramachandrarao (http://projecteuclid.org/download/pdf_1/euclid.rmjm/1250127232). I'd like to have a reasonable bound like $$ \sum_{k=1}^N \frac{1}{\varphi(k)} \le f(N)\;, $$ for a certain function $f$, maybe with a range of validity like $\forall N\ge n_0$, for a certain integer $n_0$.

$\endgroup$
1
$\begingroup$

A trivial bound can be find in this way. Using$$\frac{1}{\varphi\left(n\right)}<2e^{\gamma}\frac{\log\left(\log\left(n\right)\right)}{n},\,\forall n>2$$ where $\gamma$ is the Euler-Mascheroni constant, we have, $\forall\alpha>1$ $$\sum_{n=1}^{N}\frac{1}{\varphi\left(n\right)^{\alpha}}<2^{\alpha}e^{\gamma\alpha}\sum_{n\geq1}\left(\frac{\log\left(\log\left(n\right)\right)}{n}\right)^{\alpha}<2^{\alpha}e^{\gamma\alpha}\zeta\left(\alpha(1-\epsilon)\right)$$ where $\zeta\left(*\right)$ is the Riemann Zeta function.

$\endgroup$
1
$\begingroup$

Wikipedia gives $$\sum_{k=1}^n{1\over\phi(k)}={315\zeta(3)\over2\pi^4}\left(\log n+\gamma-\sum_p{\log p\over p^2-p+1}\right)+O\left({(\log n)^{2/3}\over n}\right)$$ with a link to R. Sitaramachandrarao. On an error term of Landau II, Rocky Mountain J. Math. 15 (1985), 579-588.

$\endgroup$
  • 1
    $\begingroup$ Thanks Gerry, the case $\alpha=1$ is ok, I'll check the reference to see if the case $\alpha=2$ can be derived from it. $\endgroup$ – PITTALUGA Jun 20 '14 at 9:41
  • $\begingroup$ I suspect that for $a=2$ the sum converges as $n\to\infty$. $\endgroup$ – Gerry Myerson Jun 20 '14 at 9:44
  • $\begingroup$ Do you have any idea if there exists a formula for an upper bound without error term? $\endgroup$ – PITTALUGA Jun 24 '14 at 13:54
  • $\begingroup$ I don't know. Have you looked at the Sitaramachandrarao paper? or at the Landau paper it references? $\endgroup$ – Gerry Myerson Jun 24 '14 at 23:34
  • $\begingroup$ Yeah I looked at them, they deal with errors, I hoped a reasonable upper bound for the finite sum of $1/\varphi(k)$ existed but I wasn't able to find anything...thanks a lot anyway Gerry! $\endgroup$ – PITTALUGA Jun 26 '14 at 7:40
0
$\begingroup$

If you use the estimate
$\forall n\in \mathbb{N} |n\ge 1$ $ \sqrt{\frac{n}{2}}\le\phi(n)\le n$ (which can be proved elementarily) , then , by taking the inverses, you can see by the confront principle that if $0< a\le 1$ then the series $\sum_{n\ge 1}\frac{1}{\phi(n)^a} $ diverges, so your succession is unbounded, while if a>2 , since the series $\sum_{n\ge 1}\frac{1}{\sqrt{n}^a} $ converges (a>2 => a/2>1), then the series $\sum_{n\ge 1}\frac{1}{\phi(n)^a} $ converges, and is $\le$ $2^\frac{a}{2} \sum_{n\ge 1}\frac{1}{n^\frac{a}{2}} $ (and thus so is your succession). This leads uncovered the case $1<a\le 2$.

$\endgroup$

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.