5

About 6 months ago I asked for an analytic continuation of $\varphi(s)=\sum_{n\ge1} e^{-n^s}.$

What's the maximal analytic continuation of $\varphi(s)?$

Doing this will help me better understand how the function behaves.

As is stated in the comments, the main question is whether the line $\Re z=1$ is the natural boundary for the analytic continuation:

$$ \varphi(s)=\Gamma\left(1+\frac1s\right)+\sum_{n=0}^\infty\frac{(-1)^n}{n!}\zeta(-ns).$$

As noted by metamorphy, this series converges for complex $s\ne0$ with $\Re s<1$.

  • 4
    Erm... And what are the function values for $s<0$ that you would like to extend? The series, as written, certainly diverges there, so you surely meant something different from what you wrote :-) – fedja Feb 12 '21 at 01:02
  • @fedja: The way I understand it, he or she is asking if this function can be analytically continued to the half-plane $\Re (s) < 0$. – M.G. Feb 12 '21 at 03:16
  • 3
    @M.G. In this interpretation it has already been established by metamorphy in the MSE thread (the function can be analytically extended to the half-plane $\Re s<1$ from $(0,1)$ with just one pole at $0$) – fedja Feb 12 '21 at 10:01
  • @fedja: oh, apologies, I was not aware of the MSE thread. – M.G. Feb 12 '21 at 14:44
  • 1
    @fedja maximal analytic continuation of $\varphi(s)$. Better? – geocalc33 Feb 12 '21 at 22:43
  • 3
    @geocalc33 Yep. Basically at this point the main question is whether the line $\Re z=1$ is the natural boundary for metamorphy's analytic continuation. – fedja Feb 12 '21 at 22:49
  • 1
    Note also that, for a complex function, there are usually several maximal analytic continuations. For example, $\log(z)$ on $\Re(z)$>0 can be continued analytically to $\mathbb{C}\setminus i[0,\infty)$, and also to $\mathbb{C}\setminus -i[0,\infty)$, and these are two distinct maximal analytic continuations. – GH from MO Feb 12 '21 at 23:29
  • 2
    @GHfromMO The function is clearly analytic for $s>0$ as a real variable No, no, and once more no! When $s>1$, it is $C^\infty$ and even in a quasi-analytic class, but not real analytic. – fedja Mar 04 '21 at 16:39
  • @fedja: Thanks for the clarification. I wrote my remark in a rush, not thinking about the oscillatory behavior of $\Re(n^{x+iy})=n^x\cos(y\log n)$ for $y\neq 0$. – GH from MO Mar 04 '21 at 20:58
  • Any updates on this? – geocalc33 Jul 23 '21 at 17:33

1 Answers1

3

I think the issue with trying to analytically continue this function is that it really defines two different, incompatible, functions. This is suggested, for instance, since $$\sum_{n=1}^\infty (e^{-n^s}-1) = \sum_{k=1}^\infty \frac{(-1)^k}{k!} \zeta(-sk)$$ When we look at the left half-plane. However, when we construct an expansion on the right half-plane we instead obtain $$\sum_{n=1}^\infty e^{-n^s} = \sum_{k=0}^\infty \frac{ (-1)^k}{k!} \zeta(-sk) + \mathbf{\Gamma\left(1+\frac{1}{s}\right)}$$ If we were dealing with a single function, we would expect that $$\sum_{n=1}^\infty e^{-n^s} - \sum_{n=1}^\infty (e^{-n^s}-1) = \sum_{n=1}^\infty 1 = \zeta(0)$$ And without the bolded Gamma function term, we do find this. Thus, our sum is probably best represented as $$\sum_{n=1}^\infty e^{-n^s} = \sum_{k=0}^\infty \frac{ (-1)^k}{k!} \zeta(-sk)+ \begin{cases} \Gamma\left(1+\frac{1}{s}\right) & \mathfrak{Re}(s)<0 \\ 0 & \mathfrak{Re}(s)>0 \end{cases}$$

To see this, let us add in an extra term to our sum so that converges on the whole real line. Let us consider $g(s,k) = \sum_{n=1}^\infty \frac{e^{-n^s}}{n^w}$ with $k>1$ so that it absolutely converges. Then we have $$\sum_{n=1}^\infty \frac{e^{-n^s}}{n^w} = \sum_{k=0}^\infty \frac{(-1)^k}{k!} \sum_{n=1}^\infty n^{sk-w} = \sum_{k=0}^\infty \frac{(-1)^k}{k!} \zeta(w-sk)$$ Notice that in this step, if $\mathfrak{Re}(s)>0$ the inner series is diverging, and so it is only equal to the $\zeta$ function by analytical continuation. When we do a step like this, we have to pick up the residues created by the function we analytically continued (in this case the $\zeta$ function). (Note: to see a bit more about why this step is necessary, see this question of my mine). Thus, we pick up the residue $$ \frac{\csc(\pi z) \zeta(w-zs)}{\Gamma(z+1)2i} \bigg|_{z=\frac{w-1}{s}} = \frac{\Gamma\left(1-\frac{-1+s+w}{s}\right)}{s}$$ However, when $\mathfrak{Re}(s)<0$ the sums all converge, so we don't pick up that residue.

Thus, we obtain that on the real line when $\mathfrak{Re}(w)>1$ $$\sum_{n=1}^\infty \frac{e^{-n^s}}{n^w} = \sum_{k=0}^\infty \frac{ (-1)^k}{k!} \zeta(w-sk)+ \begin{cases} \frac{\Gamma\left(1-\frac{-1+s+w}{s}\right)}{s} & \mathfrak{Re}(s)<0 \\ 0 & \mathfrak{Re}(s)>0 \end{cases}$$

One short note is in order, which is that $\sum_{k=0}^\infty \frac{ (-1)^k}{k!} \zeta(w-sk)$ does not converge for for some values. This is not a problem, we just replace the sum using the residue theorem with the integral $\int_{C} \frac{\csc(\pi z) \zeta(w-sz)}{z!2i} dz$ and making sure we choose $C$ so that we only pick up the residues at the natural numbers.

Caleb Briggs
  • 1,662
  • thanks for this insight. did you ever find anything with respect to the integral $\varphi(s)=\frac{1}{2\pi i}\int_{c-i\infty}^{c+i\infty}\frac{2K_1(2\sqrt{z})}{\sqrt{z}}\bigg(\sum_{n=1}^\infty e^{zn^{-s}}\bigg)~dz$? You had said it agreed with what you had obtained in the other post – geocalc33 Jan 26 '23 at 13:58
  • so the piecewise nature of the function is the problem with the analytic continuation? – geocalc33 Jan 30 '23 at 00:26
  • @geocalc33 The piecewise nature means that there can't be a single analytic function that captures the behavior on the real line. However, this doesn't mean that it's not possible for each of the branches to be analytically continued everywhere (for instance, log has a branch cut, but the branches can be continued everywhere minus the cut). When that $n^w$ term is present, the function has a natural boundary-- I think there are poles when $k$ makes $n^{sk}+w=0$, $n>0$ However, at $w=0$ the method breaks down and I'm not sure if the poles get pushed to infinity or if they remain somehow. – Caleb Briggs Jan 30 '23 at 00:40
  • @geocalc33 If my understanding of this function is correct, the branch cut of $\varphi$ comes from the pole of the $\zeta$ function, and the poles of $\varphi$ come from the contour being forced to change directions due to the angle of $s$. Some more consideration would be needed for me to fully understand the behaviour of the analytical continuation. – Caleb Briggs Jan 30 '23 at 00:48
  • seems like this is an extremely difficult problem that nobody has ever thought about before – geocalc33 Jan 30 '23 at 23:06
  • @geocalc33 You are probably correct-- for instance, I believe this function should be related in some suitable way to the Laplace transform of $\sum_{n=1}^\infty \frac{1}{n^{s}+n^{-s}}$ (see here: https://math.stackexchange.com/questions/4622671/roots-and-analytic-continuation-of-ts-sum-n0-ns-n-s-1/4623373#4623373) which obviously shares some resemblance to the zeta function. So, I suspect something like finding the zeroes is very difficult. But I still have some hope that it's possible to find the location of the poles. – Caleb Briggs Jan 31 '23 at 01:58
  • would you like to discuss this further outside of mathoverflow? – geocalc33 Jan 31 '23 at 22:57
  • @geocalc33 Sure-- what is your preferred method of communication? My own email is 1calebbriggs@gmail.com if this is how you prefer to discuss. – Caleb Briggs Jan 31 '23 at 23:33
  • 1
    sent you an email – geocalc33 Feb 01 '23 at 02:22
  • The idea of abrupt changes in continuation at least vaguely reminds of the ideas here: https://en.m.wikipedia.org/wiki/Stokes_phenomenon. Supposedly high powered summation techniques like Ecalle’s resurgence theory become multivalued because of that, and so the multiple inconsistent branches you are finding might be indicating a similar thing here – Sidharth Ghoshal Jul 28 '23 at 14:51