I want to check the validity for such a method

Define the digamma function as

$$\psi_0(x)=\frac{d}{dx}\left( \log \Gamma(x)\right)$$


It has the following series representation

$$\tag{2} \psi_0(1+x)=-\gamma +\sum_{n\geq 1}\frac{x}{n(x+n)}$$

we can easily recognize the poles at negative integers using (2).

Using the reflection formula $$\Gamma(x)\Gamma(1-x)=\frac{\pi}{\sin(\pi x)}$$

We can prove $$\tag{3} \psi_0(1-x)-\psi_0(x)=\pi \cot(\pi x)$$ and

$$\tag{4} \psi_0(1+x)-\psi_0(x)=\frac{1}{x}$$

plugging (4) in (2) we have

$$\psi_0(x)+\frac{1}{x}=-\gamma +\sum_{n\geq 1}\frac{x}{n(x+n)}$$

Now if look at the sum on the left

$$\sum_{n\geq 1}\frac{x}{n(x+n)} = \sum_{n\geq 1} \frac{x}{n^2} \frac{1}{1+\frac{x}{n}}$$

Assuming $|\frac{x}{n}| < 1$ and expand using geometric series

$$\sum_{n\geq 1} \frac{x}{n^2}\sum_{k\geq 0}\left(\frac{-x}{n} \right)^k=-\sum_{n\geq 1} \sum_{k\geq 1} \frac{(-x)^k}{n^{k+1}}=-\sum_{k\geq 1}\zeta(k+1)(-x)^k$$

Hence we have

$$\tag{5}\psi_0(x)=-\frac{1}{x}-\gamma-\sum_{k\geq 1}\zeta(k+1)(-x)^k $$

Which is the Laurent expansion of $\psi_0$ around $x=0$ giving a residue of $-1$.


My concern was the geometric expansion and the swapping of two series.

If $|x|<n$ then $|x|<1$ then the expansion is valid for any compact subset of $(-1,1)$ which justifies the uniform convergence of $\frac{(-x)^k}{n^{k+1}}$ by the $\text{M-test}$, hence swapping the two series. Then the Laurent expansion is valid on any annulus inside $(-1,1)$ avoiding $x=0$.

Is that correct ? or am I missing something ?


For $\lvert x\rvert < 1$, the nested series is absolutely convergent, hence changing the order of summation is unproblematic.

Thus the series you got is the Laurent expansion of $\psi_0$ in the whole annulus $0 < \lvert z\rvert < 1$, it converges locally uniformly and absolutely there.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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