5

I know that the series is equal to

$$ \sum_{n=0}^{\infty}e^{-nx}= \frac{1}{1-e^{-x}}$$

However, if I expand each exponential term into a Taylor series I get

$$ \sum_{n=0}^{\infty}e^{-nx}= \sum_{n=0}^{\infty}\sum_{k=0}^{\infty}\frac{(-nx)^{k}}{k!} $$

Now if I use $ \sum_{n=0}^{\infty}n^{k}=\zeta (-k) $ I would get that

$$ \sum_{n=0}^{\infty}e^{-nx} = \sum_{k=0}^{\infty} \frac{(-1)^{k} \zeta (-k)}{k!}x^{k}$$

Is this last expression correct in the sense of Taylor series?

Jose Garcia
  • 8,506

2 Answers2

4

The derivation as it stands is not valid. Even though Bilou06 was probably speaking about convergent series, the relation $\sum_n\sum_m a_{n,m}=\sum_m\sum_n a_{n,m}$ can also fail for regularized divergent series.

In this case the conclusion is false - an error term is needed. We have two formulas at hand:

$$\zeta(-n)=-\frac{B_{n+1}}{n+1}$$

$$\frac{t}{e^t-1}=\sum_{k=0}^\infty B_k\frac{x^k}{k!}.$$

Therefore,

$$\begin{array}{ll} \sum_{k=0}^\infty\frac{\zeta(-k)}{k!}(-x)^k & =\zeta(0)+\sum_{k=1}^\infty-\frac{B_{k+1}}{(k+1)}\frac{(-x)^k}{k!} \\ & =-\frac{1}{2}+\frac{1}{x}\sum_{k=1}^\infty B_{k+1}\frac{(-x)^{k+1}}{(k+1)!} \\ & =-\frac{1}{2}+\frac{1}{x}\sum_{k=2}^\infty\frac{B_k}{k!}(-x)^k \\ & =-\frac{1}{2}+\frac{1}{x}\left[\frac{-x}{e^{-x}-1}-\frac{B_1}{1!}(-x)-\frac{B_0}{0!}(-x)^0\right] \\ & =\frac{1}{1-e^{-x}}-\frac{1}{x}-1.\end{array}$$

anon
  • 151,657
0

6 years later, but the given answer is i feel incorrect/incomplete.

The relation with swapping sums was correct, and $\sum_n\sum_m a_{n,m}=\sum_m\sum_n a_{n,m}$ is valid for regularization, as long as the shape of the sums/whatever is regularized stays the same.

The question made the mistake assuming Taylor series have a the lower limit of 0. Just like most general series, this should be in reality minus infinity and positive infinity.

Let's take the example where we start at n=1, you can add yourself n=0. Changed e^(-x) into c.

$$\sum_{n=1}^{\infty}c^{n}= \frac{c}{1-c}$$ $$\sum_{n=1}^{\infty}c^{n}= \sum_{n=1}^{\infty}\sum_{k \in \mathbb{Z}}\frac{(n\ln(c))^{k}}{k!}=$$ $$\sum_{k \in \mathbb{Z}}\frac{\zeta(-k)\ln^{k}(c)}{k!}=$$ if k<-2 it's 0. if k is -1, you need to look at the limit, which is $\frac{-1}{ln(c)}$.

$$\sum_{k=0}^{\infty}\frac{\zeta(-k)\ln^{k}(c)}{k!}=\frac{c}{1-c}+\frac{1}{ln(c)}$$ As can been seen writen in the answer above.

$$\sum_{k \in \mathbb{Z}}\frac{\zeta(-k)\ln^{k}(c)}{k!}=\frac{c}{1-c}$$

Gerben
  • 593
  • Hi Gerben - this is a nice coincidence. I once came across the same problem, and discussed it two ways, the second was including the step with the (-1)! as you did it here. However, being an amateur in these questions of number-theory I was not sure whether I got it right. See here: http://go.helms-net.de/math/tetdocs/ProblemWithBellmatrix.pdf – Gottfried Helms Dec 25 '20 at 21:13
  • I am an amateur aswell (for now). But yes including all values of the zeta function is nessecary and if it's raised to a certain power you need to include the powers with 0 as coefficient. Especially otherwise tricky integrals such as e^(-x^2) or sum sin(x) /x are then super easy to solve. It's harder when you need to account for a certain "error" term. These problems occurred in multiple of your questions and applied these 3 ideas in your question at https://math.stackexchange.com/questions/700299/a-ramanujan-like-summation-is-it-correct-is-it-extensible/3859675#3859675 – Gerben Dec 26 '20 at 08:46
  • 1
    Hmm, the inclusion of what you call an error-term has possibly been one of my few achievements; at least for my intuition, if not rigorous. In http://go.helms-net.de/math/binomial_new/04_3_SummingOfLikePowers.pdf I find this first time and rediscover the Faulhaber formulae in terms of a "ZETA-matrix". And in http://go.helms-net.de/math/binomial_new/EulerMacLaurin.pdf I use it to arrive at the Euler-MacLaurin-formula with this ZETA-matrix as essential tool. - But still, I'm struggling to understand it perfectly to really & firmly have it at hands in summation-problems like this one... – Gottfried Helms Dec 26 '20 at 09:53