Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I guess this is a known result but I could not find it on the Internet. Consider these equations formed from the reciprocals of the divisors of $n$ raised to a complex number $s=a+ib$ :

$\displaystyle 1$

$\displaystyle 1 + \frac{1}{2^s}=0$

$\displaystyle 1 + \frac{1}{3^s}=0$

$\displaystyle 1 + \frac{1}{2^s} + \frac{1}{4^s}=0$

$\displaystyle 1 + \frac{1}{5^s}=0$

$\displaystyle 1 + \frac{1}{2^s} + \frac{1}{3^s} + \frac{1}{6^s}=0$

$\displaystyle 1 + \frac{1}{7^s}=0$

$\displaystyle 1 + \frac{1}{2^s} + \frac{1}{4^s} + \frac{1}{8^s}=0$

$\displaystyle 1 + \frac{1}{3^s} + \frac{1}{9^s}=0$

$\displaystyle 1 + \frac{1}{2^s} + \frac{1}{5^s} + \frac{1}{10^s}=0$

$\displaystyle 1 + \frac{1}{11^s}=0$

$\displaystyle 1 + \frac{1}{2^s} + \frac{1}{3^s} + \frac{1}{4^s} + \frac{1}{6^s} + \frac{1}{12^s}=0$


Is it true that the zeros have real part equal to $0$ ?

Wolfram Alpha example, sums of reciprocals of divisors of 30

This Mathematica program finds some of the zeros up to the number $42$.

Clear[nn, s];
 nn = i;
 Print[s /.Table[FindRoot[Total[1/Divisors[nn]^(s)] == 0,
  {s, 5*n*I}], {n, 1, 30}]], {i, 2, 42}]

There are cases when the FindRoot command in Mathematica gives solutions that have real part not equal to zero, but these appear to be accompanied with warning messages and I have checked numerically that those solutions are not zeros.

share|cite|improve this question
I don't think so; they look like they just have zeros with very nearly real part zero. The first one, for instance, also has zeros with real parts about 94.5 and 349.8. A variant of this question that I've personally wondered is if $$\sum_{n=1}^N n^{-s}$$ always has zeros with real part in $(0,1)$ (which numerically appears correct). – anon Oct 3 '11 at 9:59
It could be you are right. But by considering a different version where the reciprocals are raised to $(s-1)$ I get real part equal to $1$. That is I entered in Wolfram Alpha: Table[FindRoot[1+1/2^(s-1)+1/3^(s-1)+1/5^(s-1)+1/6^(s-1)+1/10^(s-1)+1/15^(s-1)+1‌​/30^(s-1)==0,{s,5*n*I}], {n, 1, 30}] – Mats Granvik Oct 3 '11 at 10:06
Interesting.. but the period in "1." makes me think that's still approximate. I should also mention you're specifically talking about $\sigma_{-s}(n)$ (divisor function) as a function of complex $s$. – anon Oct 3 '11 at 10:34
As a Mathematica note: DivisorSigma[-s, nn] is a simpler expression for the functions you're interested in... (edit: and I only saw @anon's comment just now.) – J. M. Oct 3 '11 at 10:36
Since you're in Mathematica: it might be useful for you to peer at contours: Partition[Table[ContourPlot[{Re[DivisorSigma[-(x + I y), k]] == 0, Im[DivisorSigma[-(x + I y), k]] == 0}, {x, -10, 10}, {y, -10, 10}], {k, 2, 41}], 5] // Grid – J. M. Oct 3 '11 at 10:45
up vote 2 down vote accepted

If $s=0$, the property fails hence one can assume without loss of generality that $s\ne0$ and that the integer $n\geqslant2$ is the product of some $p_j^{k_j}$. Define $Z=2i\pi\mathbb Z$ as the set of integer multiples of $2i\pi$.

First assume that $p_j^s\ne1$ for every $j$. Then, the sum you are interested in is $$ S_n(s)=\prod_j\left(\sum_{t=0}^{k_j}p_j^{-st}\right)=\prod_j\frac{1-p_j^{-s(k_j+1)}}{1-p_j^{-s}}. $$ The zeros $s$ of $S_n$ solve one of the equations $$ p_j^{s(k_j+1)}=1. $$ This happens if and only if $s(k_j+1)\log p_j$ is in $Z$.

On the other hand, $p_j^s=1$ for some $j$ means that $s\log p_j$ is in $Z$. For every such $j$, one should replace the contribution of $p_j$ in $S_n(s)$ by $k_j+1\ne0$.

One is left with the condition that there exists at least one $j$ such that $s\log p_j$ is not in $Z$ and $s(k_j+1)\log p_j$ is in $Z$. That is, there exists $j$ and an integer $\ell$ which is not zero modulo $k_j+1$, such that $$ s=\frac{2i\pi\ell}{(k_j+1)\log p_j}. $$ Note that since the $\log p_j$ are linearly independent over the rationals, this can happen for at most one prime number $p_j$ dividing $n$.

In particular the real part of $s$ must be zero.

share|cite|improve this answer
Oh duh, just factor $\sigma_{-s}(n)$. I should have thought of that. – anon Oct 3 '11 at 13:54
@anon, provided you are the anon who found zeros with real parts about 94.5 and 349.8, I am curious to know if you made a mistake, or if I misunderstood what you said, or what. – Did Oct 3 '11 at 13:58
I changed OP's example to just $1+1/2^s$. Either I misunderstood the code or Alpha screwed up big time. – anon Oct 3 '11 at 14:01
@anon, did you write 1/2(s) instead of 1/2^s? – Did Oct 3 '11 at 14:09
M.SE keeps mutilating the link, so just type in Table[FindRoot[1+1/2^(s)==0,{s,5*n*I}],{n,1,30}] and you should see this (.png screengrab). – anon Oct 4 '11 at 0:14

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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