How can I prove that $$\sum_{n=1}^{\infty}\frac{H_n}{n^2 2^n}=\zeta(3)-\frac{1}{2}\log(2)\zeta(2).$$ Can anyone help me please?
-
Does $H_n$ mean the n-th harmonic number? Perhaps the tag ({tag:harmonic-numbers]) is also applicable, if yes. – Martin Sleziak Dec 12 '13 at 15:58
-
Yes, n-th harmonic number. Thanks for the notice. – Dec 12 '13 at 16:09
-
4There are very many similar questions linked on the right, under Related. Perhaps you can find some inspiration from the answers provided there ? – Lucian Dec 12 '13 at 16:14
-
See a related problem.you need to modify it a little to solve your problem. – Mhenni Benghorbal Dec 19 '13 at 14:14
-
See this question. The Op posted a nice formula that you can use. – Mhenni Benghorbal Dec 19 '13 at 14:22
-
@MartinSleziak Not trying to be rude to you, but are you? In EVERY context Ive ever seen the notation $H_n$ (and especially in the contexts of infinite series [and especially especially in conjunction with the zeta function]), YES, it means the nth harmonic number. Can you think of an example where it doesnt? What are you criticizing here? Answer the question, dont give lessons in stackexchange formatting. No one else seems confused by it. Why oh why are people on this site more interested in posturing than in answering questions? – CogitoErgoCogitoSum Sep 08 '20 at 01:25
6 Answers
Let's start with the product of $\;-\ln(1-x)\,$ and $\dfrac 1{1-x}$ to get the product generating function
(for $|x|<1$) :
$$\tag{1}f(x):=-\frac {\ln(1-x)}{1-x}=\sum_{n=1}^\infty H_n\, x^n$$
Dividing by $x$ and integrating we get :
\begin{align}
\sum_{n=1}^\infty \frac{H_n}n\, x^n&=\int \frac{f(x)}xdx\\
&=-\int \frac{\ln(1-x)}{1-x}dx-\int\frac{\ln(1-x)}xdx\\
\tag{2}&=C+\frac 12\ln(1-x)^2+\operatorname{Li}_2(x)\\
\end{align}
(with $C=0$ from $x=0$)
The first integral was obtained by integration by parts, the second from the integral definition of the dilogarithm or the recurrence for the polylogarihm (with $\;\operatorname{Li}_1(x)=-\ln(1-x)$) : $$\tag{3}\operatorname{Li}_{s+1}(x)=\int\frac {\operatorname{Li}_{s}(x)}x dx$$
Dividing $(2)$ by $x$ and integrating again returns (using $(3)$ again) :
\begin{align}
\sum_{n=1}^\infty \frac{H_n}{n^2}\, x^n&=\int \frac {\ln(1-x)^2}{2\,x}dx+\int \frac{\operatorname{Li}_2(x)}x dx\\
&=C+I(x)+\operatorname{Li}_3(x)\\
\end{align}
with $I(x)$ obtained by integration by parts (since $\frac d{dx}\operatorname{Li}_2(1-x)=\dfrac {\ln(x)}{1-x}$) :
\begin{align}
I(x)&:=\int \frac {\ln(1-x)^2}{2\,x}dx\\
&=\left.\frac{\ln(1-x)^2\ln(x)}{2}\right|+\int \ln(1-x)\frac {\ln(x)}{1-x}dx\\
&=\left.\frac{\ln(1-x)^2\ln(x)}{2}+\ln(1-x)\operatorname{Li}_2(1-x)\right|+\int \frac{\operatorname{Li}_2(1-x)}{1-x}dx\\
&=\left.\frac{\ln(1-x)^2\ln(x)}{2}+\ln(1-x)\operatorname{Li}_2(1-x)-\operatorname{Li}_3(1-x)\right|\\
\end{align}
getting the general relation :
$$\tag{4}\sum_{n=1}^\infty \frac{H_n}{n^2}\, x^n=C+\frac{\ln(1-x)^2\ln(x)}{2}+\ln(1-x)\operatorname{Li}_2(1-x)+\operatorname{Li}_3(x)-\operatorname{Li}_3(1-x)$$
(with $C=\operatorname{Li}_3(1)=\zeta(3)$ here)
applied to $x=\dfrac 12$ with $\operatorname{Li}_2\left(\frac 12\right)=\dfrac{\zeta(2)-\ln(2)^2}2$ from the link returns the wished :
\begin{align}
\sum_{n=1}^\infty \frac{H_n}{n^2\;2^n}&=\zeta(3)-\frac{\ln(2)^3}2-\ln(2)\frac{\zeta(2)-\ln(2)^2}2\\
\tag{5}\sum_{n=1}^\infty \frac{H_n}{n^2\;2^n}&=\zeta(3)-\ln(2)\frac{\zeta(2)}2
\end{align}

- 43,021
- 5
- 86
- 140
-
1I forgot that you had used the generating function of the Harmonic Numbers here. I used its integral for the sum of $H_n^2/n^2$ recently. – robjohn Aug 26 '14 at 16:49
-
@robjohn: I reused these results in this answer (quite heavy machinery I'll admit...). Some appropriate references to your work are in the comments :-). Excellent continuation! – Raymond Manzoni Aug 26 '14 at 21:01
-
5
-
1
-
Is it well-known that such Harmonic sums can be succinctly expressed by the Nielsen generalized logarithm as in this post? – Tito Piezas III Jun 01 '19 at 06:43
-
@TitoPiezasIII: this paper of Kölbig "Nielsen's generalized polylogarithm" at CERN (see $(2.9)$) may interest you. I am not sure that the Nielsen functions are well known except for high energy physicists (to evaluate complicated Feynman diagrams). Excellent continuation anyway, – Raymond Manzoni Jun 01 '19 at 20:22
-
-
Im not sure that using / introducing, the non-closed, non-algebraic, artificial function $Li(x)$ is all that useful or helpful in answering the question. It presumes a lot about the knowledge of the person trying to answer the original question. All youve done really was "transfer" the problem from one of the original series to that of evaluating this funny new function. – CogitoErgoCogitoSum Sep 08 '20 at 01:31
We will make frequent use of
$$
\binom{n+1}{k+1}=\binom{n}{k}\frac{n+1}{k+1}\tag{1}
$$
The Generalized Harmonic Numbers of the second order are defined as
$$
H_n^{(2)}=\sum_{k=1}^n\frac1{k^2}\tag{2}
$$
The factor of $2^{-n}$ in each term reminded me of the Euler Series Transformation. Reversing the series acceleration (series deceleration?), it can be seen that we should look at
$$
\begin{align}
\sum_{k=0}^n\binom{n}{k}(-1)^k\frac{H_{k+1}^{(2)}}{k+1}
&=\sum_{k=0}^n\binom{n}{k}\frac{(-1)^k}{k+1}\sum_{j=0}^k\frac1{(j+1)^2}\tag{3}\\
&=\frac1{n+1}\sum_{k=0}^n(-1)^k\binom{n+1}{k+1}\sum_{j=0}^k\frac1{(j+1)^2}\tag{4}\\
&=\frac1{n+1}\sum_{j=0}^n\sum_{k=j}^n(-1)^k\binom{n+1}{k+1}\frac1{(j+1)^2}\tag{5}\\
&=\frac1{n+1}\sum_{j=0}^n(-1)^j\binom{n}{j}\frac1{(j+1)^2}\tag{6}\\
&=\frac1{(n+1)^2}\sum_{j=0}^n(-1)^j\binom{n+1}{j+1}\frac1{j+1}\tag{7}\\
&=\frac1{(n+1)^2}\sum_{j=0}^n(-1)^j\sum_{k=0}^n\binom{k}{j}\frac1{j+1}\tag{8}\\
&=\frac1{(n+1)^2}\sum_{k=0}^n\frac1{k+1}\sum_{j=0}^n(-1)^j\binom{k+1}{j+1}\tag{9}\\
&=\frac1{(n+1)^2}\sum_{k=0}^n\frac1{k+1}\tag{10}\\[3pt]
&=\frac{H_{n+1}}{(n+1)^2}\tag{11}
\end{align}
$$
Explanation:
$\ \;(3)$: use $(2)$
$\ \;(4)$: apply $(1)$
$\ \;(5)$: change order of summation
$\ \;(6)$: $\sum\limits_{j=k}^n(-1)^j\binom{n+1}{j+1}=(-1)^k\binom{n}{k}$
$\ \;(7)$: apply $(1)$
$\ \;(8)$: $\sum\limits_{j=0}^n\binom{j}{k}=\binom{n+1}{k+1}$
$\ \;(9)$: $\sum\limits_{j=0}^k(-1)^j\binom{k+1}{j+1}=1$
$(10)$: $H_{n+1}=\sum\limits_{k=0}^n\frac1{k+1}$
Using $(11)$, the Euler Series Transformation says that
$$
\begin{align}
\sum_{k=0}^\infty(-1)^k\frac{H_{k+1}^{(2)}}{k+1}
&=\sum_{n=0}^\infty2^{-n-1}\sum_{k=0}^n\binom{n}{k}(-1)^k\frac{H_{k+1}^{(2)}}{k+1}\\
&=\sum_{n=0}^\infty2^{-n-1}\frac{H_{n+1}}{(n+1)^2}\tag{12}
\end{align}
$$
Therefore, with a change of indexing, we get
$$
\begin{align}
\sum_{n=1}^\infty2^{-n}\frac{H_n}{n^2}
&=\sum_{n=1}^\infty(-1)^{n-1}\frac{H_n^{(2)}}{n}\tag{13}\\
&=\sum_{n=1}^\infty\frac{(-1)^{n-1}}{n^3}+\sum_{n=1}^\infty(-1)^{n-1}\frac{H_{n-1}^{(2)}}{n}\tag{14}\\
&=\frac34\zeta(3)+\sum_{n=1}^\infty\frac{(-1)^{n-1}}{n}\sum_{k=1}^{n-1}\frac1{k^2}\tag{15}\\
&=\frac34\zeta(3)+\sum_{k=1}^\infty\sum_{n=k+1}^\infty\frac{(-1)^{n-1}}{nk^2}\tag{16}\\
&=\frac34\zeta(3)+\sum_{k=1}^\infty\sum_{n=1}^\infty\frac{(-1)^{k+n-1}}{(k+n)k^2}\tag{17}\\
&=\frac34\zeta(3)+\sum_{k=1}^\infty\sum_{n=1}^\infty(-1)^{k+n-1}\left(\frac1{k^2n}-\frac1{kn(k+n)}\right)\tag{18}\\[6pt]
&=\frac34\zeta(3)-\frac12\zeta(2)\log(2)+\frac14\zeta(3)\tag{19}\\[9pt]
&=\zeta(3)-\frac12\zeta(2)\log(2)\tag{20}
\end{align}
$$
Explanation:
$(13)$: reindex $(12)$
$(14)$: $H_n^{(2)}=\frac1{n^3}+H_{n-1}^{(2)}$
$(15)$: apply $(2)$
$(16)$: change order of summation
$(17)$: reindex $n\mapsto k+n$
$(18)$: $\frac1{(k+n)k^2}=\frac1{k^2n}-\frac1{kn(k+n)}$
$(19)$: $\sum\limits_{k=1}^\infty\sum\limits_{n=1}^\infty\frac{(-1)^{k+n}}{kn(k+n)}=\frac14\zeta(3)$ from $(5)$ and $(7)$ of this answer
$(20)$: addition
-
+1. Nice, Rob! The factor of $2^n$ in the denominator made me think of the Euler series transformation, too, but I couldn't make it go through. Also, your work starting from equation (13) is an evaluation of $A(2,1)$ from my question here, which means you've got a derivation for the last of the three sums from that question. I'd be happy if you were willing to finish off your series of answers to my question with your derivation for $A(2,1)$ here. – Mike Spivey Dec 17 '13 at 03:52
-
@MikeSpivey: I was going to do so, but dinner interrupted. Note that the justification for $(19)$ cites my answer to your question. – robjohn Dec 17 '13 at 05:12
-
-
-
While $(9)$ is still true, the index should be till $n$ (in the explanation part) as per what you used in the answer. – MathGod Jun 17 '17 at 02:47
-
@IshanSingh: $\sum\limits_{j=0}^n(-1)^j\binom{k+1}{j+1}=1$ is only true if $n\ge k$ as in the derivation. The statement in explanation $(9)$ is true, independent of $n$, and can be applied to $(9)$ in the derivation because $n\ge k$. – robjohn Jun 17 '17 at 03:59
-
@robjohn Yes I know. Looking back at it, seems more of a pedantic remark. – MathGod Jun 17 '17 at 04:20
\begin{eqnarray} \sum\limits_{n=1}^\infty \frac{H_n}{n^2 2^n} = \sum\limits_{m=1}^\infty \frac{1}{m} \sum\limits_{n=m}^\infty \frac{1}{n^2 2^n} = \sum\limits_{m=1}^\infty \frac{1}{m} \int\limits_{-\infty}^0(-\xi) \frac{(1/2 \exp(\xi))^m}{1-1/2 \exp(\xi)} d\xi = \\ \int\limits_{-\infty}^0 \xi \frac{\log(1 - 1/2 \exp(\xi))}{1-1/2 \exp(\xi)} d\xi = \\ \int\limits_{1/2}^1 \left(\frac{1}{u} + \frac{1}{1-u}\right) \log(u) \left[\log(2) + \log(1-u)\right] du = \\ \zeta(3) - \frac{1}{12} \pi^2 \log(2) \end{eqnarray}
I think that all the steps are clear except for the last two ones.In the second last step I substituted for 1 - 1/2 exp(xi).The only non-trivial integrals in here are $\int \log(u)/(1-u) du$ and $\int \log(u) \log(1-u)/(1-u) du$. I compute them now. The first integral is done by expanding the denominator in a series and integrating term by term. \begin{equation} \int\limits_{1/2}^1 \frac{\log(u)}{1-u} du = \sum\limits_{p=0}^\infty \int\limits_{1/2}^1 u^p \log u du = \sum\limits_{p=0}^\infty \frac{-1+2^{-1-p}}{(p+1)^2} + \log(2) \sum\limits_{p=0}^\infty \frac{2^{-1-p}}{p+1} = -\zeta(2) + Li_2(1/2) + \log(2) Li_1(1/2) \end{equation} The second integral is done by integrating by parts and using the definition of the polylogarithmic function. \begin{equation} \int\limits_{1/2}^1 \log(u) \frac{\log(1-u)}{1-u} du = \int\limits_{1/2}^1 \log(1-u) Li_2^{'}(1-u) du = \left.\log(1-u) Li_2(1-u)\right|_{1/2}^1 + Li_3(1/2) = \log(2) Li_2(1/2) + Li_3(1/2) \end{equation} Now, the only thing that remains is to bring the results together. I am sorry but due to time constraints I am not able to do it right now. I have verified with Mathematica that all the partial results are correct.
Final Note: We can clearly see that the result is expressed though elementary functions and through polylogarithms of order not bigger than three, evaluated at 1/2.From the Wikipedia page on Polylogarithms we learn those polylogarithms at 1/2 are expressed in closed form through $\pi$, $\log(2)$ and the $\zeta$ functions. Having said that we can say that this completes the proof.

- 11,331
-
3Unfortunately, I would describe this as cheating more than slightly :( – Igor Rivin Dec 12 '13 at 17:16
Starting with $$ \frac{\ln(1-x)}{1-x}=-\displaystyle \sum_{n=1}^{\infty}H_n x^n $$
multiply both sides by $ \frac{\ln x}{x} $ then integrate from $ x=0 $ to $ 1/2 $ we get \begin{align*} I&= \int_0^{1/2}\frac{\ln x\ln(1-x)}{x(1-x)}\,dx=-\sum_{n=1}^{\infty}H_n \int_0^{1/2} x^{n-1}\ln x\ dx\\ &=-\sum_{n=1}^\infty H_n\left( \frac{\ln2}{2^n n}+\frac{1}{2^n n^2}\right)=-\frac12\ln2\zeta(2)-\sum_{n=1}^\infty \frac{H_n}{2^n n^2} \tag{1} \end{align*}
Note that we used $\sum_{n=1}^\infty \frac{H_n}{2^nn}=\frac12\zeta(2)$ which follows from using the generating function $\sum_{n=1}^\infty\frac{x^n H_n}{n}=\operatorname{Li_2}(x)+\frac12 \ln^2(1-x)$ with $x=1/2$ where $\operatorname{Li_2}(1/2)=\frac12\zeta(2)-\frac12\ln^22$
on the other hand \begin{equation*} I= \int_{0}^{1/2}\frac{\ln x\ln(1-x)}{x(1-x)}\ dx \overset{x\mapsto1-x}{=} \int_{1/2}^{1}\frac{\ln(1-x) \ln(x)}{x(1-x)}\ dx \end{equation*} and by adding the integral to both sides, we get \begin{equation*} 2I= \int_{0}^{1}\frac{\ln x\ln(1-x)}{x(1-x)}\ dx =-\sum_{n=1}^\infty H_n \int_0^1 x^{n-1}\ln x \ dx =\sum_{n=1}^\infty \frac{H_n}{n^2}=2\zeta(3)\tag2 \end{equation*}
where the last result follows from using Euler Identity.
By combining $(1)$ and $(2)$ we obtain the closed form of $\sum_{n=1}^\infty\frac{H_n}{2^nn^2}$.

- 25,498
-
Pardon me, it seems that the identity below (1) where you explain what you used is wrong since $\sum _{k=1}^{\infty }\frac{H_k}{k:2^k}=\frac{1}{2}\zeta \left(2\right)$. – Dennis Orton Sep 08 '20 at 01:54
-
\begin{align*} \operatorname{Li}_3\left(\frac{1}{2}\right)&=\int _0^{\frac{1}{2}}\frac{\operatorname{Li}_2\left(x\right)}{x}\:dx=-\operatorname{Li}_2\left(\frac{1}{2}\right)\ln \left(2\right)+\int _{\frac{1}{2}}^{1}\frac{\ln \left(x\right)\ln \left(1-x\right)}{1-x}\:dx\\[2mm] &=-\frac{1}{2}\ln \left(2\right)\zeta \left(2\right)+\frac{1}{2}\ln ^3\left(2\right)+\int _0^1\frac{\ln \left(1-x\right)\ln \left(x\right)}{x}\:dx-\int _0^{\frac{1}{2}}\frac{\ln \left(x\right)\ln \left(1-x\right)}{1-x}\:dx\\[2mm] &=-\frac{1}{2}\ln \left(2\right)\zeta \left(2\right)+\frac{1}{2}\ln ^3\left(2\right)+\sum _{k=1}^{\infty }\frac{1}{k^3}-\frac{1}{2}\ln ^3\left(2\right)-\frac{1}{2}\int _0^{\frac{1}{2}}\frac{\ln ^2\left(1-x\right)}{x}\:dx\\[2mm] &=-\frac{1}{2}\ln \left(2\right)\zeta \left(2\right)+\zeta \left(3\right)-\sum _{k=1}^{\infty }\frac{H_k}{k^2\:2^k}+\sum _{k=1}^{\infty }\frac{1}{k^3\:2^k}\\[2mm] &=-\frac{1}{2}\ln \left(2\right)\zeta \left(2\right)+\zeta \left(3\right)-\sum _{k=1}^{\infty }\frac{H_k}{k^2\:2^k}+\operatorname{Li}_3\left(\frac{1}{2}\right) \end{align*} And magically we find the value for that sum \begin{align*} \sum _{k=1}^{\infty }\frac{H_k}{k^2\:2^k}=-\frac{1}{2}\ln \left(2\right)\zeta \left(2\right)+\zeta \left(3\right) \end{align*}

- 2,646
It would appear this one can be related to the first integral in the three log integrals you posted. The one with the upper limit of 1/2.
$$\displaystyle \sum_{n=1}^{\infty}\frac{H_{n}}{n^{2}}x^{n}=\int_{0}^{x}\frac{Li_{2}(t)+1/2log^{2}(1-t)}{t}dt$$.....[1]
By letting $\displaystyle x=1/2$ and integrating we get:
$$\displaystyle \int_{0}^{x}\frac{Li_{2}(t)+\frac{1}{2}\log^{2}(1-t)}{t}dt=\left [-Li_{3}(1-t)+Li_{3}(t)+Li_{2}(1-t)\log(1-t)+\frac{1}{2}\log(t)\log^{2}(1-t)\right ]_{0}^{\frac{1}{2}}$$
Note that $$\displaystyle Li_{3}(1/2)=\frac{7}{8}\zeta(3)+\frac{1}{6}\log^{3}(2)-\frac{{\pi}^{2}}{12}log(2)$$
and $$\displaystyle Li_{2}(1/2)=\frac{{\pi}^{2}}{12}-\frac{log^{2}(2)}{2}$$
So, we arrive at:
$$\displaystyle =\boxed{\zeta(3)-\frac{{\pi}^{2}}{12}log(2)}$$
The integral in [1] can be derived by beginning with:
$$\displaystyle \sum_{n=1}^{\infty}\frac{H_{n}}{n}x^{n}=\int_{0}^{x}\sum_{n=1}^{\infty}H_{n}t^{n-1}dt$$
$$\displaystyle =-\int_{0}^{x}\frac{log(1-t)}{t}dt-\int_{0}^{x}\frac{log(1-t)}{1-t}dt$$
Dividing by t and integrating gives our series in question.
Of course, let x = 1/2.

- 413
- 11