25
$\begingroup$

Consider the Clausen function $\operatorname{Cl}_2(x)$ that can be defined for $0<x<2\pi$ in several equivalent ways:

$$\begin{align}\operatorname{Cl}_2(x)&=-\int_0^x\ln\left(2\sin\left(\tfrac t2\right)\right)dt\\&=\sum_{n=1}^\infty\frac{\sin\,(n x)}{n^2}\\&=\Im\operatorname{Li}_2\left(e^{i x}\right)\\&=i\left(\frac{\pi^2}6+\frac{x^2}4-\frac{\pi x}2-\operatorname{Li}_2\left(e^{i x}\right)\right).\end{align}\tag1$$ I'm interested in integrals of the form $$I(p)=\int_0^{2\pi}\operatorname{Cl}_2(x)^2\,x^p\,dx.\tag2$$ I found that $$I(0)=\frac{\pi^5}{90}\tag3$$ and conjectured next several values based on numeric evidence: $$I(1)\stackrel?=\frac{\pi^6}{90},\ \ I(2)\stackrel?=\frac{44\,\pi^7}{2835},\ \ I(3)\stackrel?=\frac{23\,\pi^8}{945}.\tag4$$ One might expect that $I(4)$ is a rational multiple of $\pi^9$, but apparently it is not (unless the denominator is huge).

I'm asking for your help in proving conjectured values $(4)$, finding a closed form of $I(4)$, and, if possible, a general formula for $I(p)$.

Update: Values $I(-1)$ and $I(-2)$ are also interesting.

$\endgroup$
5
  • $\begingroup$ are you sure about I(2)? i get $\pi^7/70$ in this case which is quiete close...but ihave no possibilty to evaluate the sum numericalliy :( $\endgroup$ – tired May 21 '15 at 18:41
  • $\begingroup$ My expression for $I(2)$ is accurate with at least 200 decimal digits of precision. I evaluated the integral both in Maple and Mathematica, and they returned exactly the same result. $\endgroup$ – Vladimir Reshetnikov May 21 '15 at 18:57
  • $\begingroup$ ok thanks a lot $\endgroup$ – tired May 21 '15 at 18:58
  • $\begingroup$ ok i made a mistake. there is additional contribution $\sum_{m,n=1,n\neq m}^{\infty}\frac{8 \pi}{mn(m^2-n^2)^2}$ which i have no idea how tu sum up :( $\endgroup$ – tired May 21 '15 at 19:03
  • $\begingroup$ the diagonal term of $I(4)$ will be $\frac{17 \pi ^9}{540}$ $\endgroup$ – tired May 21 '15 at 19:31
21
$\begingroup$

The General Case:

Consider instead the integral \begin{align} J(\eta) &=\Re\int^{2\pi}_0x^\eta\ {\rm Li}_2(e^{ix})^2\ {\rm d}x\\ \end{align} for $\eta\in\mathbb{N}_0$. Deforming the contour around $z=0$, we get \begin{align} J(\eta) &=\Re\oint_{|z|=1}\frac{{\rm Li}_2(z)^2\ln^{\eta}(z)}{i^{\eta+1}z}{\rm d}z\\ &=\Re\int^1_0\frac{\left((\ln(x)+2\pi i)^{\eta}-\ln^{\eta}(x)\right){\rm Li}_2(x)^2}{i^{\eta+1}x}{\rm d}x\\ &=\Re\sum^{\eta-1}_{k=0}\binom{\eta}{k}\frac{(2\pi)^{\eta-k}}{i^{k+1}}\int^1_0\frac{{\rm Li}_2(x)^2\ln^{k}(x)}{x}{\rm d}x\\ &=\sum^{\lfloor{\eta/2\ -1}\rfloor}_{k=0}\binom{\eta}{2k+1}(-1)^{k+1}(2\pi)^{\eta-2k-1}\int^1_0\frac{{\rm Li}_2(x)^2\ln^{2k+1}(x)}{x}{\rm d}x \end{align} Relating this to $I(\eta)$, $$I(\eta)=\pi^{\eta+5}\left(\frac{2^{\eta+1}}{36(\eta+1)}-\frac{2^{\eta+2}}{6(\eta+2)}+\frac{2^{\eta+3}}{3(\eta+3)}-\frac{2^{\eta+4}}{4(\eta+4)}+\frac{2^{\eta+5}}{16(\eta+5)}\right)-J(\eta)$$

The Case $I(0)$:

It is evident that $J(0)=0$ since it is a vacuous summation. Thus $$I(0)=\left.\pi^{\eta+5}\left(\frac{2^{\eta+1}}{36(\eta+1)}-\frac{2^{\eta+2}}{6(\eta+2)}+\frac{2^{\eta+3}}{3(\eta+3)}-\frac{2^{\eta+4}}{4(\eta+4)}+\frac{2^{\eta+5}}{16(\eta+5)}\right)\right|_{\eta=0}=\frac{\pi^5}{90}$$

The Case $I(1)$:

In this case, $J(1)=0$ as well. Hence $$I(1)=\left.\pi^{\eta+5}\left(\frac{2^{\eta+1}}{36(\eta+1)}-\frac{2^{\eta+2}}{6(\eta+2)}+\frac{2^{\eta+3}}{3(\eta+3)}-\frac{2^{\eta+4}}{4(\eta+4)}+\frac{2^{\eta+5}}{16(\eta+5)}\right)\right|_{\eta=1}=\frac{\pi^6}{90}$$

The Case $I(2)$:

For $\eta=2$, $$\left.\pi^{\eta+5}\left(\frac{2^{\eta+1}}{36(\eta+1)}-\frac{2^{\eta+2}}{6(\eta+2)}+\frac{2^{\eta+3}}{3(\eta+3)}-\frac{2^{\eta+4}}{4(\eta+4)}+\frac{2^{\eta+5}}{16(\eta+5)}\right)\right|_{\eta=2}=\frac{16\pi^7}{945}$$ and $$ J(2)=\binom{2}{1}(-1)(2\pi)\int^1_0\frac{{\rm Li}_2(x)^2\ln{x}}{x}{\rm d}x=4\pi\sum^\infty_{m=1}\sum^\infty_{n=1}\frac{1}{m^2n^2(m+n)^2}=\frac{4\pi^7}{2835}$$ The double sum $\displaystyle\sum^\infty_{m=1}\sum^\infty_{n=1}\frac{1}{m^2n^2(m+n)^2}=\frac{\pi^6}{2835}$ has been evaluated here. Thus $$I(2)=\frac{16\pi^7}{945}-\frac{4\pi^7}{2835}=\frac{44\pi^7}{2835}$$

The Case $I(3)$:

When $\eta=3$, $$\left.\pi^{\eta+5}\left(\frac{2^{\eta+1}}{36(\eta+1)}-\frac{2^{\eta+2}}{6(\eta+2)}+\frac{2^{\eta+3}}{3(\eta+3)}-\frac{2^{\eta+4}}{4(\eta+4)}+\frac{2^{\eta+5}}{16(\eta+5)}\right)\right|_{\eta=3}=\frac{\pi^8}{35}$$ and $$J(3)=\binom{3}{1}(-1)(2\pi)^2\int^1_0\frac{{\rm Li}_2(x)^2\ln{x}}{x}{\rm d}x=12\pi^2\sum^\infty_{m=1}\sum^\infty_{n=1}\frac{1}{m^2n^2(m+n)^2}=\frac{4\pi^8}{945}$$ Therefore $$I(3)=\frac{\pi^8}{35}-\frac{4\pi^8}{945}=\frac{23\pi^8}{945}$$

The Case $I(4)$:

For $\eta=4$, $$\left.\pi^{\eta+5}\left(\frac{2^{\eta+1}}{36(\eta+1)}-\frac{2^{\eta+2}}{6(\eta+2)}+\frac{2^{\eta+3}}{3(\eta+3)}-\frac{2^{\eta+4}}{4(\eta+4)}+\frac{2^{\eta+5}}{16(\eta+5)}\right)\right|_{\eta=4}=\frac{16\pi^9}{315}$$ and \begin{align} J(4) &=\binom{4}{1}(-1)(2\pi)^3\left(-\frac{\pi^6}{2835}\right)+\binom{4}{3}(2\pi)\int^1_0\frac{{\rm Li}_2(x)^2\ln^3{x}}{x}{\rm d}x\\ &=\frac{32\pi^9}{2835}+4\pi\int^1_0\frac{{\rm Li}_2(x)\ln^4{x}\ln(1-x)}{x}{\rm d}x\\ &=\frac{32\pi^9}{2835}+\frac{4\pi}{5}\int^1_0\frac{{\rm Li}_2(x)\ln^5{x}}{1-x}{\rm d}x+\frac{4\pi}{15}\int^1_0\frac{\ln^6{x}\ln(1-x)}{1-x}{\rm d}x\\ &=\frac{32\pi^9}{2835}-96\pi\sum^\infty_{m=1}\frac{H_n^{(2)}}{(n+1)^6}-192\pi\sum^\infty_{n=1}\frac{H_n}{(n+1)^7}\\ &=-\frac{8\pi^9}{567}+192\pi\zeta(3)\zeta(5)-96\pi\zeta(6,2) \end{align} giving us $$I(4)=\frac{184\pi^9}{2835}-192\pi\zeta(3)\zeta(5)+96\pi\zeta(6,2)$$ I am not sure if $\zeta(6,2)$ has a simple closed form. At least Flajolet and Salvy's paper on Euler sums suggests that there isn't.

$\endgroup$
2
  • 4
    $\begingroup$ Great answer (+1) :) $\endgroup$ – r9m May 22 '15 at 17:09
  • 1
    $\begingroup$ nice! it's a little bit sad that $\zeta(6,2)$ is the first of the double zeta family $\zeta(n,2)$ with no closed form?, :( $\endgroup$ – tired May 22 '15 at 17:43
12
$\begingroup$

I can confirm $I(1)$:

Using the series expansion for Clausen's function we have

$$ I(p)=\sum_{n=1}^{\infty}\sum_{m=1}^{\infty}\int_0^{2 \pi}x^p\frac{\sin(n x)}{n^2}\frac{\sin(m x)}{m^2}\,\mathrm dx $$

now setting $p=1$. Furthermore we may use the orthogonality of sine $\int^{2 \pi}_{0}{\sin(n x)\sin(m x)}\,\mathrm dx=\pi\delta_{mn}$ and integrate by parts to get

$$ I(1)=\underbrace{\sum_{m,n=1,n\neq m}^{\infty}\frac{1}{2 (m-n)^2}-\frac{1}{2 (m+n)^2}-\frac{2 m n}{(m^2 - n^2)^2}}_{=0}+\pi^2\underbrace{\sum_{n=1}^{\infty}\frac{1}{n^4}}_{=\zeta(4)}=\frac{\pi^6}{90} $$

I'm checking if there is a pattern, especially if it holds that the doublesum always cancel. But at least for the first few $p$'st this seems likely, nevertheless the algebra gets a little bit messy but i will try to go further tomorrow.

Edit: Playing the same game for $p=0$ we end up with $$ I(0)=\pi\underbrace{\sum_{n=1}^{\infty}\frac{1}{n^4}}_{=\zeta(4)}=\frac{\pi^5}{90} $$

The double sum cancels immediatly in this case, so i confirmed also $I(0)$

Edit2:

Doing some careful algebra,following the same approach as above, we may find that

$$ I(2)=\underbrace{4 \pi \sum_{n=1}^{\infty}{\frac{8 \pi ^2 n^3-3 n }{24 n^7}}}_{=\frac{\pi^7}{70}}+8 \pi\underbrace{\sum_{m,n=1,n\neq m}^{\infty}\frac{ 1}{m n\left(m^2-n^2\right)^2}}_{J}=8 \pi\underbrace{\sum_{m}^{\infty}\sum_{n=0}^{m-1}\frac{ 1}{m n\left(m^2-n^2\right)^2}+\sum_{m}^{\infty}\sum_{m+1}^{\infty}\frac{ 1}{m n\left(m^2-n^2\right)^2}}_{J_1+J_2}+\frac{\pi^7}{70} $$

Maybe someone with more experience in calculating sums like this can take it home from here :)

Edit3:

It turns out that the same off-diagonal sum will give us the result for $p=3$ if we multiply by $24\pi^2$ instead of $8\pi$. So if we have solved $p=2$ we get $I(3)$ for free :)

Nevertheless, for $p>3$ the off-diagonal contribution starts to get messier and yields additionals contribution which may well be not proportional to $\pi^9$

$\endgroup$
3
  • $\begingroup$ How do you prove that the double sum is zero for $I(1)$? $\endgroup$ – Pranav Arora May 21 '15 at 19:49
  • $\begingroup$ sry, there was a typo $\endgroup$ – tired May 21 '15 at 22:52
  • $\begingroup$ hiho someone here woho knows to evaluate such sums? $\endgroup$ – tired May 22 '15 at 13:02
7
$\begingroup$

$I(1)$ can be evaluated in the following way also. One has to note that

$$\int_0^{2\pi} x\sin(mx)\sin(nx)\,dx=\begin{cases} 0 & n \neq m \\ \pi^2 & n=m \end{cases}$$

To prove the above, write $$\int_0^{2\pi} x\sin(mx)\sin(nx)\,dx=\frac{1}{2}\int_0^{2\pi} x\cos((m-n)x) \,dx-\frac{1}{2}\int_0^{2\pi} x\cos((m+n)x)\,dx$$ and then use integration by parts.

Back to the original integral, $$\begin{aligned} \int_0^{2\pi} x\text{Cl}_2(x)^2\,dx &=\sum_{m=1}^{\infty} \sum_{n=1}^{\infty} \frac{1}{m^2n^2}\int_0^{2\pi} x\sin(mx)\sin(nx)\,dx \\ & =\sum_{n=1}^{\infty} \frac{\pi^2}{n^4} \\ & =\frac{\pi^6}{90} \end{aligned}$$

$\endgroup$
4
  • $\begingroup$ that's neat! (+1) $\endgroup$ – tired May 22 '15 at 17:36
  • $\begingroup$ Hey long time no see. Nice answer ! $\endgroup$ – Zaid Alyafeai Jul 19 '16 at 20:43
  • $\begingroup$ Thanks! Yes, it's been a long time, nice to see you Zaid. :) $\endgroup$ – Pranav Arora Jul 20 '16 at 5:18
  • $\begingroup$ Nice to see you tackle hard integrals, you have improved a lot. keep it up! $\endgroup$ – Zaid Alyafeai Jul 27 '16 at 20:39
2
$\begingroup$

We can use the identity

$$\mathrm{cl}_2(x)^2 = \mathrm{cl}_2(2\pi -x)^2 $$

Hence we have

$$I(5)=\int_0^{2\pi}\operatorname{Cl}_2(x)^2\,(2\pi -x)^5\,dx$$

Expand

$$I(5) = 32 \pi^5I(0)-80 \pi^4 I(1)+80 \pi^3 I(2)-40 \pi^2 I(3)+10 \pi I(4)-I(5)$$

Solve for $I(5)$

$$I(5) = 16\pi^5I(0)-40\pi^4 I(1)+40\pi^3 I(2)-20\pi^2 I(3)+5\pi I(4)$$

This simplifies to

$$I(5) = \frac{544}{2835}π^{10}-960 π^2 \zeta(3)\zeta(5)+480 π^2 \zeta(6,2)$$

This method can be generalized for $I(2q+1)$ given we know all the previous terms

$$I(2q+1) = \int^{2\pi}_0 (2\pi-x)^{2q+1} \mathrm{cl}_2(x)^2\,dx$$

Use the binomial theorem

$$I(2q+1) = \sum _{ k=1 }^{ 2q+1 }{ { \left( -1 \right) }^{ k }\binom{2q+1}{k} { 2\pi }^{ 2q+1-k } }\int^{2\pi}_0 x^k \mathrm{cl}_2(x)^2\,dx$$

$$I(2q+1) = \frac{1}{2}\sum _{ k=1 }^{ 2q }{ { \left( -1 \right) }^{ k }\binom{2q+1}{k} { 2\pi }^{ 2q+1-k } }I(k)$$

$\endgroup$

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.