Maybe I'm just blocked currently, I think I had it done myself some weeks ago but cannot find/recover the derivation of this equality:
$$\sum_{k=1}^{\infty} {1 \over k}{1 \over 2k-1} = 2 \ln 2 $$
I have the result in my sketchpad and just checked at Wolfram Alpha, that it is correct. But I cannot remember how I did find it; the numerical approximation needs much more terms than I would use normally for a heuristic. So maybe I've taken this from Wolfram Alpha from the beginning; but anyway: I think the derivation cannot be too difficult. I'm stuck at the moment - could someone help with the derivation?

  • 1
    $\begingroup$ Are there similar series for algebraic multiples of logarithms? $\endgroup$ – quanta Apr 20 '11 at 14:54


  • $\begingroup$ Ah, that was really a quick answer. And simple, thanks! $\endgroup$ – Gottfried Helms Apr 20 '11 at 14:45
  • $\begingroup$ @GottfriedHelms It's MERCATOR SERIES TIME. $\endgroup$ – user38268 Apr 21 '11 at 1:01
  • $\begingroup$ @David: yes, true. After I asked the question I gave it another try myself, assuming a simple solution based on the mercator series. But Joriki was much quicker and even has it in a one-liner! :-) $\endgroup$ – Gottfried Helms Apr 21 '11 at 4:54

Expand ${1 \over k}{1 \over 2k-1}$ in partial fractions to ${2 \over 2k-1}-{1 \over k}$. Do some cancellation and end up with twice the alternating harmonic series, which converges to $\ln 2$.

  • 2
    $\begingroup$ The downside of this approach is that you're cancelling elements with indices that are apart by a factor of $2$, so you have to make some arguments why you're allowed to do that -- in my approach, you're only combining elements with successive indices, which is more obviously OK. $\endgroup$ – joriki Apr 20 '11 at 14:45
  • $\begingroup$ @joriki, you're right, of course. $\endgroup$ – lhf Apr 20 '11 at 14:47

Joriki has given an excellent answer. Here is another way to do it.

Consider $$f(x,y) = \sum_{k=1}^{\infty} x^{k-1} y^{2k-2} \tag{$\clubsuit$}$$ We then have from the formula for geometric series$$f(x,y) = \dfrac1{1-xy^2} \tag{$\spadesuit$}$$

Now integrate $(\clubsuit)$ and $(\spadesuit)$ to conclude what you want. Below are the details.

Integrate $(\clubsuit)$. We then have \begin{align} \int_0^1\int_0^1 f(x,y) dx dy & = \int_0^1 \int_0^1 \sum_{k=1}^{\infty} x^{k-1} y^{2k-2} dx dy\\ & = \sum_{k=1}^{\infty}\int_0^1 \int_0^1 x^{k-1} y^{2k-2} dx dy\\ & = \sum_{k=1}^{\infty} \dfrac1k \dfrac1{2k-1} \tag{$\diamondsuit$} \end{align} The change of infinite sum and integration is justified since the convergence of the series is uniform in any closed set of the form $[0,1-\epsilon_1] \times [0,1-\epsilon_2]$.

Now integrate $(\spadesuit)$. We have \begin{align} \int_0^1\int_0^1 f(x,y) dx dy & =\int_0^1 \int_0^1 \left(\dfrac{dx}{1-xy^2}\right)dy\\ & = -\int_0^1 \dfrac{\log(1-y^2)}{y^2} dy\\ & = \log 4 \tag{$\heartsuit$} \end{align} where we make use of the fact that $$\int \dfrac{dx}{1-xy^2} = - \dfrac{\log(1-y^2)}{y^2} + \text{constant}$$ and $$- \dfrac{\log(1-y^2)}{y^2} = \dfrac{\log(1-x^2)}{x} + \log(1+x) - \log(1-x) + \text{constant}$$ Now comparing $(\diamondsuit)$ and $(\heartsuit)$, we get that $$\sum_{k=1}^{\infty} \dfrac1k \dfrac1{2k-1} = 2 \log 2$$


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.