12
$\begingroup$

Let $\zeta(s)$ be the Riemann zeta function and $\gamma$ be the Euler-Mascheroni constant. I observed the following result empirically. Looking for a proof or disproof.

$$ \lim_{n \to \infty}\sum_{k = 1}^n \zeta\Big(k - \frac{1}{n}\Big) = \gamma $$

Also, I searched for different summation formulas for the Euler-Mascheroni constant using the Riemann zeta function but could not find it anywhere. Is there any reference to this sum in literature?

Update: Applying the method of @Simply Beautiful Art, we can show that

$$ \sum_{k = 1}^n \zeta\Big(k + \frac{1}{m}\Big) = \gamma + n + m + \mathcal O(n^{-1} + m^{-1}) $$

$\endgroup$
20
$\begingroup$

We have the simple asymptotic expansion as $s\to1$ given by:

$$\zeta(s)=\frac1{s-1}+\gamma+\mathcal O(s-1)\tag{$s\to1$}$$

For the first term of your sum, you have

$$\zeta\left(1-\frac1n\right)=-n+\gamma+\mathcal O(n^{-1})$$

and for the rest of the terms,

\begin{align}\sum_{1<k\le n}\zeta\left(k-\frac1n\right)&=\sum_{1<k\le n}\left(1+\sum_{m>1}\frac1{m^{k-\frac1n}}\right)\tag1\\&=n-1+\sum_{1<k\le n}\sum_{m>1}\frac1{m^{k-\frac1n}}\tag2\\&=n-1+\mathcal O(2^{-n})+\sum_{k>1}\sum_{m>1}\frac1{m^{k-\frac1n}}\tag3\\&=n-1+\mathcal O(2^{-n})+\sum_{m>1}\sum_{k>1}\frac1{m^{k-\frac1n}}\tag4\\&=n-1+\mathcal O(2^{-n})+\sum_{m>1}\sqrt[n]m\frac{m^{-2}}{1-m^{-1}}\tag5\\&=n-1+\mathcal O(2^{-n})+\sum_{m>1}\sqrt[n]m\left(\frac1{m-1}-\frac1m\right)\tag6\\&=n+\mathcal O(2^{-n})+\sum_{m>1}(\sqrt[n]m-1)\left(\frac1{m-1}-\frac1m\right)\tag7\\&=n+\mathcal O(n^{-1})\tag8\end{align}

where

$(1):$ Definition of $\zeta$.

$(2):$ Summing over $1$.

$(3):$ Extending $k$ from $(1,n]$ to $(1,\infty)$, with $\mathcal O(2^{-n})$ error.

$(4):$ Rearranging the series.

$(5):$ Geometric series.

$(6):$ Partial fractions.

$(7):$ Using telescoping series and $1=\sum_{m>1}\left(\frac1{m-1}-\frac1m\right)$.

$(8):$ Asymptotic expansion using $\sqrt[n]m=\exp(n^{-1}\ln(m))=1+\varepsilon n^{-1}\ln(m)$, where $|\varepsilon|\le\sqrt[n]m$, which gives the series bounded by $n^{-1}$ times another series with dominating term $\mathcal O(m^{\frac1n-2}\ln(m))$ and thus converges.

Adding these results together, we find that

$$\sum_{k=1}^n\zeta\left(k-\frac1n\right)=\gamma+\mathcal O(n^{-1})$$

| cite | improve this answer | |
$\endgroup$
  • 2
    $\begingroup$ This is a nice and elegant solution, for sure ! $\endgroup$ – Claude Leibovici Aug 6 at 5:28
  • 1
    $\begingroup$ An excellent solution. Order of error term is consistent with simulation. For $n=10^9$, the sum is approximately $0.577215666086463903230516402745658$, which matches $8$ digits after decimal point. $\endgroup$ – piepie Aug 6 at 5:55
  • $\begingroup$ @piepie That looks to match the error then? $n^{-1}=10^{-9}$ so you should expect $9\pm$ a constant amount of digits accurate. $\endgroup$ – Simply Beautiful Art Aug 6 at 12:50
  • $\begingroup$ @SimplyBeautifulArt You are right. As $n$ is increased by an order of $10$, I expect 0/1/2 more digits of accuracy. $\endgroup$ – piepie Aug 7 at 3:08
  • $\begingroup$ @piepie no you should only expect 1 more digit, since $C\times10^{-(n+1)}$ is 10 times smaller than $C\times10^{-n}$. $\endgroup$ – Simply Beautiful Art Aug 7 at 13:11

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.