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'm considering the transfer-function $$ t(x) = \log(1 + \exp(x)) $$ and find the beginning of the power series (simply using Pari/GP) as $$ t(x) = \log(2) + 1/2 x + 1/8 x^2 – 1/192 x^4 + 1/2880 x^6 - \ldots $$ Examining the pattern of the coefficients I find the much likely composition $$ t(x) = \sum_{k=0}^\infty {\eta(1-k) \over k! }x^k $$ where $ \eta() $ is the Dirichlet eta-(or "alternating zeta") function.
I'm using this definition in further computations and besides the convincing simplicitiness of the pattern the results are always meaningful. However, I've no idea how I could prove this description of the coefficients.

Q: Does someone has a source or an idea, how to do such a proof on oneself?

share|cite|improve this question
Are you familiar with Faadi Bruno's forlmula. – Mhenni Benghorbal Feb 18 '13 at 18:02
Yes, but I find it a much complicated one. When I study iteration of functions and can use a Carleman-matrix-epression (and the formal powers of that Carleman-matrices) then I could always avoid to involve into that formula. I know of a handful of articles which put Carleman/Bell-matrix and Faa-di-Bruno together, as for instance Aldrovandi/Freitas but I wouldn't see the solution for the proof for the decryption of the transfer-function $t(x)$ in question. – Gottfried Helms Feb 18 '13 at 18:07
up vote 12 down vote accepted

The Dirichlet eta function is given by $\eta(s)=\sum_{n=1}^{\infty}(-1)^{n-1}n^{-s}$, but this converges only for $s$ with positive real part, and you are proposing to use its behavior for negative integers. A globally convergent series for $\eta$ can be derived using the Riemann zeta function (cf. here): $$ \eta(s)=(1-2^{1-s})\zeta(s)=\sum_{n=0}^{\infty}2^{-(n+1)}\sum_{k=0}^{n}(-1)^{k}{{n}\choose{k}}(k+1)^{-s}. $$ Using this expansion allows us to write $\eta(1-k)$ as $$ \eta(1-k)=\sum_{n=0}^{\infty}2^{-(n+1)}\sum_{j=0}^{n}(-1)^{j}{{n}\choose{j}}(j+1)^{k-1}. $$

Your power series is then $$ \begin{eqnarray} \sum_{k=0}^{\infty}\frac{\eta(1-k)x^k}{k!} &=&\sum_{k=0}^{\infty}\sum_{n=0}^{\infty}2^{-(n+1)}\sum_{j=0}^{n}(-1)^{j}{{n}\choose{j}}(j+1)^{k-1}\left(\frac{x^{k}}{k!}\right) \\ &=&\sum_{n=0}^{\infty}2^{-(n+1)}\sum_{j=0}^{n}\frac{(-1)^{j}}{j+1}{{n}\choose{j}}\sum_{k=0}^{\infty}\frac{\left(x(j+1)\right)^{k}}{k!} \\ &=&\sum_{n=0}^{\infty}2^{-(n+1)}\sum_{j=0}^{n}\frac{(-1)^{j}}{j+1}{{n}\choose{j}}\left(e^x\right)^{j+1} \\ &=&\sum_{n=0}^{\infty}2^{-(n+1)}\int_{-e^{x}}^{0} dy\sum_{j=0}^{n}{{n}\choose{j}}y^{j} \\ &=&\sum_{n=0}^{\infty}2^{-(n+1)}\int_{-e^{x}}^{0}\left(1+y\right)^{n}dy \\ &=&\sum_{n=0}^{\infty}2^{-(n+1)}\frac{\left(1+y\right)^{n+1}}{n+1}\Bigg\vert_{-e^x}^{0} \\ &=&\sum_{n=0}^{\infty}2^{-(n+1)}\frac{1-\left(1-e^x\right)^{n+1}}{n+1} \\ &=&f\left(\frac{1}{2}\right) - f\left(\frac{1-e^x}{2}\right), \end{eqnarray} $$ where $$f(z)=\sum_{n=0}^{\infty}\frac{z^{n+1}}{n+1}=-\log \left(1-z\right).$$ Putting this together, we find $$ \sum_{k=0}^{\infty}\frac{\eta(1-k)x^k}{k!} = -\log\left(\frac{1}{2}\right)+\log\left(\frac{1+e^x}{2}\right)=\log\left(1+e^x\right), $$ as you conjectured.

share|cite|improve this answer
Wow. Thanks a lot -this shall give me much to chew on and to study... to use this sequence of steps for other, but related, situations; it looks much helpful, indeed! – Gottfried Helms Feb 19 '13 at 20:38

Related problems: (I), (II), (III), (IV). Here is a formula for the nth derivative of the function $\ln(1+e^{x})$ at the point $x=0$

$$ \left( \ln(1+e^{x})\right)^{(n)}= \sum _{k=1}^{n}\begin{Bmatrix} n\\k \end{Bmatrix} \left( -1 \right)^{k+1}2^{-k}\, \Gamma\left( k \right),\quad n \in \mathbb{N}, $$

where $\begin{Bmatrix} n\\k \end{Bmatrix} $ are the Stirling numbers of the second kind. The above formula allows us to construct the Taylor series of the function as

$$ \ln(1+e^x) = \ln(2)+\sum_{n=1}^{\infty} \sum _{k=1}^{n}\begin{Bmatrix} n\\k \end{Bmatrix} \left( -1 \right)^{k+1}\, 2^{-k}\,\Gamma\left( k \right) \frac{x^n}{n!}. $$

share|cite|improve this answer
can you prove the first assertion? – nbubis Feb 19 '13 at 15:02
@nbubis: I'll try to do this later. – Mhenni Benghorbal Feb 19 '13 at 15:04
(where the curly braces mean the Stirling numbers 2'nd kind) – Gottfried Helms Feb 19 '13 at 16:05
@GottfriedHelms: Yes, I'll add this. – Mhenni Benghorbal Feb 19 '13 at 16:25
arrgh - that Stirling numbers 2'nd kind rang my alarm. It's just to use the Carleman-matrices for $x\to \exp(x)-1$, then $ x \to x+1 $ and the first column of that for $x \to \log(1+x) $. This gives the decomposition in Stirling-numbers 2nd and 1st kind and binomial-coefficients. Well, but after that: how is that the sequence of Dirichlet-etas? – Gottfried Helms Feb 19 '13 at 17:48

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.