4

Evaluate:

$$\sum_{n=1}^{\infty} \frac{H_n}{(n+1)^2}$$

A user stated: "most of the time sum up the residues of $(\gamma+\psi(z))^2\cdot r(z)$. To determine the residues, just expand the digamma function as a Laurent series about the positive integers." where $r(z)$ is the rational function.

Firstly, what is the Laurent series of $\psi(z)$??

And how would I find the residues? Any help is appreciated.

amad27
  • 41

2 Answers2

3

This is a partial answer to your question and I do not know how to get the Laurent series of $\psi(z)$. Note that \begin{eqnarray} \sum_{n=1}^\infty\frac{H_n}{(n+1)^2}&=&\sum_{n=1}^\infty\frac{1}{(n+1)^2}\left(H_n+\frac{1}{n+1}\right)-\sum_{n=1}^\infty\frac{1}{(n+1)^3}\\ &=&\sum_{n=1}^\infty\frac{H_{n+1}}{(n+1)^2}-\sum_{n=1}^\infty\frac{1}{(n+1)^3}\\ &=&\sum_{n=1}^\infty\frac{H_{n}}{n^2}-\sum_{n=1}^\infty\frac{1}{n^3}\\ &=&\sum_{n=1}^\infty\frac{H_{n}}{n^2}-\zeta(3).\\ \end{eqnarray} The latter series can be evaluated using this generating function $$\sum_{n=1}^\infty \frac{H_n\, x^n}{n^2}=\zeta(3)+\frac{\ln^2(1-x)\ln(x)}{2}+\ln(1-x)\operatorname{Li}_2(1-x)+\operatorname{Li}_3(x)-\operatorname{Li}_3(1-x)$$ Taking its limit as $x\to1$, we get \begin{eqnarray} \sum_{n=1}^\infty\frac{H_n}{n^2}=2\zeta(3)\end{eqnarray} and hence $$ \sum_{n=1}^\infty\frac{H_n}{(n+1)^2}=\zeta(3).$$

xpaul
  • 44,000
1

Start from the singular expansion of the digamma function: $$\psi(z+1)+\gamma = \sum_{n\ge 1} \frac{z}{n(n+z)} = \sum_{n\ge 1} \left(\frac{1}{n}-\frac{1}{z+n}\right).$$

This immediately produces the Laurent expansion at negative integers $-q$ where $q\ge 1$:

$$\psi(z+1)+\gamma = -\frac{1}{z+q} + H_{q-1} + \cdots$$

This is because the sum minus the singular term is $$\frac{1}{q} + \sum_{n=1, n\ne q} \left(\frac{1}{n}-\frac{1}{(z+q)+n-q}\right) \\ = \frac{1}{q} + \sum_{n=1, n\ne q} \left(\frac{1}{n}-\frac{1}{n-q}\frac{1}{(z+q)/(n-q)+1}\right)$$

The constant term here is $$\frac{1}{q} + \sum_{n=1, n\ne q} \left(\frac{1}{n}-\frac{1}{n-q}\right)$$

The outside term and the first term inside the sum iterate over the inverses of all natural $n.$ The second term inside the sum first collects $$\sum_{n=1}^{q-1} \left(-\frac{1}{n-q}\right)$$ which is $H_{q-1}$ and thereafter iterates over the inverses of all natural $n,$ canceling the first contribution, for a final result of $H_{q-1}.$

This immediately implies that $$\mathrm{Res}_{z=-(q-1)} (\psi(z)+\gamma)^2 = -2 H_{q-1}.$$

The sum being evaluated is $$\sum_{n=1}^\infty \frac{H_n}{(n+1)^2} = \sum_{n=2}^\infty \frac{H_{n-1}}{n^2}.$$

Since the residue formula also holds for $q=1$ we can re-write this as $$\sum_{n=1}^\infty \frac{H_{n-1}}{n^2}.$$

Now integrating $$f(z) = \frac{1}{z^2}(\psi(z)+\gamma)^2$$ along a large circle of radius $R$ with $R$ going to infinity and passing between the poles we pick up exactly $-2$ times the value of the sum and the residue at zero.

But at zero we have the expansion $$\psi(z+1)+\gamma = -\sum_{k\ge 1} \zeta(k+1) (-1)^k z^k$$ which yields $$\psi(z)+\gamma = -\frac{1}{z} - \sum_{k\ge 1} \zeta(k+1) (-1)^k z^k.$$

This immediately implies that $$\mathrm{Res}_{z=0} \frac{1}{z^2}(\psi(z)+\gamma)^2 = 2\zeta(3).$$

Because the integral on the circular contour vanishes in the limit (an exercise that is left to the reader that follows from the observation that $2\pi R \times (\log R)^2/R^2$ vanishes in the limit) we get

$$2\zeta(3) - 2 \sum_{n=1}^\infty \frac{H_{n-1}}{n^2} = 0$$ or $$\sum_{n=1}^\infty \frac{H_{n-1}}{n^2} = \zeta(3).$$

The digamma formulae are from the Wikipedia entry.

Marko Riedel
  • 61,317
  • Great answer (+1), but how did you get the laurent series at negative integers? I can never understand how that works. Dont you need to find the coefficient $a_n$? Which is defined as a contour integral? So how did you do it? – Amad27 Dec 30 '14 at 08:30
  • Also, how did you get the partial sum involving the $q-1$ in the upper index? And finally, I dont understand the step from: "which yields.. this implies that.." How did you make the transformation? It didnt change anything? – Amad27 Dec 30 '14 at 08:38