Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

How do we get a closed form for $$\sum_{n=1}^\infty\frac{H_n}{(2n+1)^2}$$

share|improve this question
    
no idea. Why do you think there is one? –  Will Jagy Feb 26 '13 at 3:34
2  
Mathematica says the answer is $\frac{1}{4} (-\pi^2 \ln{2} + 7 \zeta(3))$ which probably suggests it knows a way how to reduce it to known series. One attempt may be at trying to go backwards from Mathematica's answer and see if you can rediscover what it did. –  muzzlator Feb 26 '13 at 3:40
1  
It would have been better to put that information into the question, both for motivation and to help others with their thoughts. So post it as "Mathematica says $\sum_{n=1}^\infty\frac{H_n}{(2n+1)^2}=\frac{1}{4} (-\pi^2 \ln{2} + 7 \zeta(3))$. Is there an easy to demonstrate this?" If you have information on why you were motivated to ask Mathematica, that might be useful, too. –  Ross Millikan Feb 26 '13 at 3:44
1  
Some similar series are on this guy's blog, the same guy's blog, and this other guy's blog, but they omit the identity you're after. –  Aeolian Feb 26 '13 at 4:25
1  
@joriki: you are right. Then OP came back with some motivation. There is progress. –  Ross Millikan Feb 26 '13 at 4:35
show 3 more comments

2 Answers

up vote 9 down vote accepted

Here's another solution. I'll denote various versions of the sum

$$ \sum_{k=1}^\infty\sum_{j=1}^k\frac1j\frac1{k^2} $$

by an $S$ with two subscripts indicating which parities are included, the first subscript referring to the parity of $j$ and the second to the parity of $k$, with '$\mathrm e$' denoting only the even terms, '$\mathrm o$' denoting only the odd terms, '$+$' denoting the sum of the even and odd terms, i.e. the regular sum, and '$-$' denoting the difference between the even and the odd terms, i.e. the alternating sum. Then

$$ \begin{align} \sum_{n=1}^\infty\frac{H_n}{(2n+1)^2} &= 2\sum_{n=1}^\infty\sum_{i=1}^n\frac1{2i}\frac1{(2n+1)^2} \\ &= 2S_{\mathrm{eo}} \\ &= 2(S_{++}-S_{\mathrm o+}-S_{\mathrm{ee}}) \\ &= 2\left(S_{++}-S_{\mathrm o+}-\frac18S_{++}\right) \\ &= 2\left(\frac38S_{++}+\left(\frac12S_{++}-S_{\mathrm o+}\right)\right) \\ &= \frac34S_{++}+S_{-+} \\ &= \frac32\zeta(3)+\sum_{k=1}^\infty\sum_{j=1}^k\frac{(-1)^j}j\frac1{k^2}\;, \end{align} $$

where I used the result $\sum_nH_n/n^2=2\zeta(3)$ from the blog post Aeolian linked to and reduced the present problem to finding the analogue of that result with the sign alternating with $j$, which we can rewrite as

$$ \begin{align} \sum_{k=1}^\infty\sum_{j=1}^k\frac{(-1)^j}j\frac1{k^2} &= \sum_{k=1}^\infty\sum_{j=1}^\infty\frac{(-1)^j}j\frac1{k^2}-\sum_{k=1}^\infty\sum_{j=k+1}^\infty\frac{(-1)^j}j\frac1{k^2} \\ &= -\zeta(2)\log2+\sum_{j=1}^\infty\frac{(-1)^j}{j+1}\sum_{k=1}^j\frac1{k^2}\;. \end{align} $$

This last double sum can be evaluated by the method applied in the blog post, making use of the fact that summing the coefficients of a power series in $x$ corresponds to dividing it by $1-x$:

$$ \begin{align} \sum_{j=1}^\infty x^j\sum_{k=1}^j\frac1{k^2}=\def\Li{\operatorname{Li}}\frac{\Li_2(x)}{1-x}\;, \end{align} $$

where $\Li_2$ is the dilogarithm. Thus

$$ \begin{align} \sum_{j=1}^\infty\frac{(-1)^j}{j+1}\sum_{k=1}^j\frac1{k^2} &= \int_0^1\sum_{j=1}^\infty (-x)^j\sum_{k=1}^j\frac1{k^2}\mathrm dx \\ &= \int_0^1\frac{\Li_2(-x)}{1+x}\mathrm dx \\ &= \left[\Li_2(-x)\log(1+x)\right]_0^1+\int_0^1\frac{\log^2(1+x)}x\mathrm dx \\ &=-\frac{\zeta(2)}2\log2+\frac{\zeta(3)}4\;, \end{align} $$

where the boundary term is evaluated using $\Li_2(-1)=-\eta(2)=-\zeta(2)+2\zeta(2)/4=-\zeta(2)/2$ and the integral in the second term is evaluated in this separate question. Putting it all together, we have

$$ \begin{align} \sum_{n=1}^\infty\frac{H_n}{(2n+1)^2} &= \frac74\zeta(3)-\frac32\zeta(2)\log2 \\ &= \frac74\zeta(3)-\frac{\pi^2}4\log2\;. \end{align} $$

I believe all the rearrangements can be justified, despite the series being only conditionally convergent in $j$, by considering the partial sums with $j$ and $k$ both going up to $M$; then all the rearrangements can be carried out within that finite square of the grid, and the sums of the remaining terms go to zero with $M\to\infty$.

share|improve this answer
add comment

I gave some integral representations for sums close to this one before. Here is an integral representation for your sum

$$\sum_{n=1}^\infty\frac{H_n}{(2n+1)^2}= \frac{1}{4}\,\int_{0}^{1}\!{\frac {\ln \left( 1-z \right) \ln \left( z\right) }{z\sqrt {1-z}}}{dz}= \frac{1}{4}(7\,\zeta \left( 3 \right) -{\pi }^{2}\ln \left( 2 \right))\sim 0.393327464. $$

The above integral can be evaluated through beta function. Here is the technique from previous problems. Basically, you need to consider the integral

$$ \int_{0}^{1} z^s (1-z)^{w-1/2} dz. $$

share|improve this answer
add comment

Your Answer

 
discard

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.