Gauss’s proof that the Digamma function equals $\int_0^{\infty}(\frac{e^{-t}}{t} – \frac{e^{-zt}}{1-e^{-t}})dt$.

complex-analysisdigamma-functiongamma functionintegrationreference-request

I was reading about the Digamma function, defined as:
$$\psi(z) = \frac{d}{dx}\ln( \Gamma(z)) = \frac{\Gamma ' (z)}{\Gamma(z)}$$
And the following integral representation of $\psi(z)$ was given for $z:\Re(z) > 0$:

$$\psi(z) = \int_0^{\infty}\Big{(}\frac{e^{-t}}{t} – \frac{e^{-zt}}{1-e^{-t}}\Big{)}dt$$

The proof of which is written off as "due to Gauss," but I cannot find his proof. What was Gauss's proof of this?

Best Answer

You may have a look at the book Modern Analysis of Whittaker and Watson as cited in the Wikipedia article of the Digamma function. In Section 12.3 a proof is given by using the identity $$\tag{1}\frac{\Gamma'(z)}{\Gamma(z)} = -\gamma - \frac{1}{z} + \lim_{n \rightarrow \infty} \sum_{m=1}^n \Bigg( \frac{1}{m} - \frac{1}{z+m} \Bigg)$$ together with $$\frac{1}{z+m} = \int_0^\infty e^{-t(z+m)} \ dt.$$ In fact, this allows us to rewrite formula (1) by $$\frac{\Gamma'(z)}{\Gamma(z)} = -\gamma - \int_0^\infty e^{-zt} \ dt + \lim_{n \rightarrow \infty} \int_0^\infty \sum_{m=1}^n (e^{-mt} - e^{-(z+m)t}) \ dt.$$ Now the sum in the integral can be simplified by using the geometric series identity. In addition, we recall that $$\gamma = \int_0^\infty \Bigg(\frac{1}{1-e^{-t}} - \frac{1}{t} \Bigg) e^{-t} \ dt. $$ A proof of this identity is also given in the above-mentioned book. Combining both together, we see that \begin{align} \frac{\Gamma'(z)}{\Gamma(z)} &= -\gamma - \lim_{n \rightarrow \infty} \int_0^\infty \frac{e^{-t}-e^{-zt}-e^{-(n+1)t}-e^{-(z+n+1)t}}{1-e^{-t}} \ dt \\ & = \int_0^\infty \frac{e^{-t}}{t} - \frac{e^{-zt}}{1-e^{-t}} dt - \lim_{n \rightarrow \infty} \int_0^\infty \frac{1-e^{-zt}}{1-e^{-t}} e^{-(n+1)t} \ dt. \end{align} By continuous continuation we know that $\frac{1-e^{-zt}}{1-e^{-t}}$ is a bounded function on $[0,1]$. Moreover, for $t \ge 1$ one has obviously that $$ \bigg| \frac{1-e^{-zt}}{1-e^{-t}} \bigg| \leq \frac{2}{1-e^{-1}}$$ and therefore the function is bounded. Using this information, we can apply Lebesgue's theorem on dominated convergence to conclude that the integral vanishes for $n \rightarrow \infty$. All in all, we obtain that $$\frac{\Gamma'(z)}{\Gamma(z)} = \int_0^\infty \frac{e^{-t}}{t} - \frac{e^{-zt}}{1-e^{-t}} dt.$$ However, I am unaware which proof was given by Gauss and it is also not mentioned in the book.

Related Question