I'm looking for strategies for evaluating the following sums for given $z$ and $m$: $$ \mathcal{S}_m(z):=\sum_{n=1}^\infty \frac{H_n^{(m)}z^n}{n}, $$ where $H_n^{(m)}$ is the generalized harmonic number, and $|z|<1$, $m \in \mathbb{R}$.

Using the generating function of the generalized harmonic numbers, an equivalent problem is to evaluate the following integral: $$ \mathcal{S}_m(z) = \int_0^z \frac{\operatorname{Li}_m(t)}{t(1-t)}\,dt, $$ where $\operatorname{Li}_m(t)$ is the polylogarithm function, and $|z|<1$, $m \in \mathbb{R}$.

Question 1: Are there any way to reduce the sum to Euler sum values, given by Flajolet–Salvy paper?

Question 2: Are there any way to reduce the integral to integrals given by Freitas paper?

The case $m=1$ and $z=1/2$ was the problem 1240 in Mathematics Magazine, Vol. 60, No. 2, pp. 118–119. (Apr., 1987) by Coffman, S. W. $$ \mathcal{S}_1\left(\tfrac12\right)=\sum_{n=1}^\infty \frac{H_n}{n2^n} = \frac{\pi^2}{12}. $$ There are several solutions in the linked paper.

The more interesting case $m=2$ and $z=1/2$ is listed at Harmonic Number, MathWorld, eq. $(41)$: $$ \mathcal{S}_2\left(\tfrac12\right)=\sum_{n=1}^\infty \frac{H_n^{(2)}}{n2^n} = \frac{5}{8}\zeta(3). $$

We know less about the evaluation. At the MathWorld it is marked as "B. Cloitre (pers. comm., Oct. 4, 2004)". This value is also listed at pi314.net, eq. $(701)$. Unfortunately, I don't know about any paper/book reference for this value. It would be nice to see some.

Question 3: How could we evaluate the case $m=2$, $z=1/2$?

It would be nice to see a solution for the sum form, but also solutions for the integral form are welcome.


About quesiton 3, we have $$\sum_{n\geq1}\frac{H_{n}^{\left(2\right)}}{n2^{n}}=\int_{0}^{1/2}\frac{\textrm{Li}_{2}\left(x\right)}{x\left(1-x\right)}dx=\int_{0}^{1/2}\frac{\textrm{Li}_{2}\left(x\right)}{x}dx+$$ $$+\int_{0}^{1/2}\frac{\textrm{Li}_{2}\left(x\right)}{1-x}dx=\textrm{Li}_{3}\left(\frac{1}{2}\right)+\int_{0}^{1/2}\frac{\textrm{Li}_{2}\left(x\right)}{1-x}dx $$ now note that, using integration by parts $$\int_{0}^{1/2}\frac{\textrm{Li}_{2}\left(x\right)}{1-x}dx=-\log\left(\frac{1}{2}\right)\textrm{Li}_{2}\left(\frac{1}{2}\right)-\int_{0}^{1/2}\frac{\log^{2}\left(1-x\right)}{x}dx.$$ So let analyze $$J=\int_{0}^{1/2}\frac{\log^{2}\left(1-x\right)}{x}dx $$ using integration by parts few times $$J=\log^{2}\left(\frac{1}{2}\right)\log\left(\frac{1}{2}\right)+2\int_{0}^{1/2}\frac{\log\left(1-x\right)\log\left(x\right)}{1-x}dx= $$ $$=\log^{2}\left(\frac{1}{2}\right)\log\left(\frac{1}{2}\right)+2\textrm{Li}_{2}\left(\frac{1}{2}\right)\log\left(\frac{1}{2}\right)+2\int_{1/2}^{1}\frac{\textrm{Li}_{2}\left(u\right)}{u}du $$ $$=\log^{2}\left(\frac{1}{2}\right)\log\left(\frac{1}{2}\right)+2\textrm{Li}_{2}\left(\frac{1}{2}\right)\log\left(\frac{1}{2}\right)+2\zeta\left(3\right)-2\textrm{Li}_{3}\left(\frac{1}{2}\right) $$ so we have $$\sum_{n\geq1}\frac{H_{n}^{\left(2\right)}}{n2^{n}}=3\textrm{Li}_{3}\left(\frac{1}{2}\right)-3\log\left(\frac{1}{2}\right)\textrm{Li}_{2}\left(\frac{1}{2}\right)-\log^{2}\left(\frac{1}{2}\right)\log\left(\frac{1}{2}\right)-2\zeta\left(3\right)= $$ $$=\frac{5\zeta\left(3\right)}{8}. $$


By using the integral representation of the polylogarithm , inserting it into the integral in question, changing the order of integration , integrating over $t$ and then using Abel's identity for the dilogarithm we obtained the following : \begin{equation} S_m(z) = S_{2,m-1}(1)-S_{m-1,2}(1)-\left(S_{1,m-1}(1)-Li_m(1)\right) \log(z) - Li_m(1) \log(1-z) + \frac{(-1)^{m-1}}{(m-1)!} \int\limits_0^1 [\log(1-\xi)]^{m-1} \frac{\log[1-(1-\xi) z]}{\xi} d\xi \end{equation} where $S_{n,m}(1)$ are the Nielsen generalized polylogarithms at unity. I guess that integration by parts will be now the best strategy to evaluate the integral on the right hand side.

On the other hand assume that $m=2 m_1+1$ is odd. Then by decomposing the denominator into simple fractions and then integrating by parts we have: \begin{eqnarray} S_m(z) &=& Li_{m+1}(z) + \int\limits_0^z \frac{Li_m(t)}{1-t} dt \\ &=& Li_{m+1}(z) + Li_1(z) Li_m(z) - \int\limits_0^z Li_1(t) \frac{Li_{m-1}(t)}{t} dt \\ &\vdots&\\ &=& Li_{m+1}(z) + \sum\limits_{p=1}^{P} Li_p(z) Li_{m+1-p}(z) (-1)^{p-1} + (-1)^P \int\limits_0^z Li_P(t) \frac{Li_{m-P}(t)}{t} dt \end{eqnarray} Now we choose $p=m_1$ and in the last integral the integrand is clearly a full derivative. Therefore we have: \begin{equation} S_m(z) = Li_{m+1}(z) + \sum\limits_{p=1}^{m_1} Li_p(z) Li_{m+1-p}(z) (-1)^{p-1} + (-1)^{m_1} \frac{1}{2} Li_{m_1+1}(z)^2 \end{equation} Unfortunately when $m$ is even this method doesn't work. How do we handle the even case then ?

Now let us assume that $m=2 m_1$ is even. Here the residual integral reads: \begin{equation} r_{m_1}(z) := (-1)^{m_1} \int\limits_0^z \frac{Li_{m_1}(t)^2}{t} dt \end{equation} Firstly we handle the case $m_1=1$. Here we just integrate by parts. We have: \begin{eqnarray} -r_1(z) &=& \int\limits_0^z \frac{[\log(1-t)]^2}{t} dt\\ &=& \log(z) [\log(1-z)]^2 + \int\limits_0^z \log(t)\cdot \frac{2 \log(1-t)}{1-t} dt\\ &=& \log(z) [\log(1-z)]^2 + Li_2(1-z)\cdot 2 \log(1-z) + \int\limits_0^z Li_2(1-t) \cdot \frac{2}{1-t} dt \\ &=& \log(z) [\log(1-z)]^2 + Li_2(1-z)\cdot 2 \log(1-z) + 2\left[Li_3(1) - Li_3(1-z)\right] \end{eqnarray}


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.