16
$\begingroup$

Here is another log-trig integral you may find challenging/fun. Or not :)

$$\int_{0}^{\frac{\pi}{2}}\ln(1+\sin(x))\ln(1+\cos(x))\tan(x)dx=\frac{\pi^{2}}{8}\ln(2)-\frac{5}{16}\zeta(3)$$

$\endgroup$
  • $\begingroup$ Are you just throwing integrals around? Do you have a question? $\endgroup$ – naslundx Aug 7 '14 at 19:13
  • $\begingroup$ @naslundx He's implicitly asking how one would show that LHS=RHS. $\endgroup$ – beep-boop Aug 7 '14 at 20:25
  • $\begingroup$ Weierstraß substitution gives: $$\int_0^1\ln\left(\frac{(t+1)^2}{t^2+1}\right)\ln\left(\frac{2}{t^2+1}\right) \frac{4t}{1-t^4}dt$$I don't know what to do from here. $\endgroup$ – Akiva Weinberger Aug 8 '14 at 5:04
  • $\begingroup$ I also tried that same sub, columbus. So, since there are some talented and brilliant mathematicians on SE, I thought perhaps they may like this one. I made a sub and arrived at the equivalent $\int_{0}^{1}\frac{\ln(1+\sqrt{1-x^{2}})\ln(x+1)}{x}dx$. I also tried writing it as $\int_{0}^{\frac{\pi}{2}}\frac{\ln(2\sin^{2}(\frac{x}{2}+\frac{\pi}{4}))\ln(2 \cos^{2}(x/2))\sin(x)}{\cos(x)}dx$, for what it's worth. :) I also derived a double sum involving Gamma, but it was certainly not any easier. $\endgroup$ – Cody Aug 8 '14 at 10:23
  • 1
    $\begingroup$ I have seen that thread. That is a nice solution, M. But, this problem has $1+\sin(x), \;\ 1+\cos(x)$, which makes it more difficult. We can get rid of those by rewriting it as I mentioned in the above comment. $\endgroup$ – Cody Aug 8 '14 at 17:19
9
$\begingroup$

The integral equals

$$ \int_0^1 dx\, \ln \left( 1 + \sqrt{1-x^2} \right) \frac{\ln(1+x)}{x} \\= \left. -\text{Li}_2(-x) \ln \left( 1 + \sqrt{1-x^2} \right) \right\lvert_0^1 + \int_0^1 dx\, \text{Li}_2(-x)\frac{1}{x} \left[1 - \frac{1}{\sqrt{1-x^2}} \right] \\=\text{Li}_3(-1)-\int_0^1 dx\, \frac{\text{Li}_2(-x)}{x\sqrt{1-x^2}}. $$ Now, $\text{Li}_3(-1) = -\beta(3) = -\left(1 - 2^{1-3}\right) \zeta(3) = -\frac{3}{4} \zeta(3)$. In the second term, write $\text{Li}_2(-x)$ as a series and interchange summation and integration: $$ -\int_0^1 dx\, \frac{\text{Li}_2(-x)}{x\sqrt{1-x^2}} = \sum_{k\geq 1} \frac{(-1)^{k-1}}{k^2} \int_0^1 dx\,x^{k-1} \left(1-x^2\right)^{-1/2} = \sum_{k\geq 1} \frac{(-1)^{k-1}}{2k^2} B\left(\frac{k}{2},\frac{1}{2}\right).$$

We calculate this sum, following Cody's suggestion to start with $$ \sum_{k\geq 1} (-1)^{k-1} B\left(\frac{k}{2},\frac{1}{2}\right) x^{k-1} = \frac{\pi - 2\arcsin x}{\sqrt{1-x^2}}. $$ Integrating once from $0$ to $x$ yields $$ \sum_{k\geq 1} \frac{(-1)^{k-1}}{k} B\left(\frac{k}{2},\frac{1}{2}\right) x^{k} = \pi \arcsin x -\arcsin^2 x. $$ Dividing by $x$ and integrating from $0$ to $1$ yields $$ \sum_{k\geq 1} \frac{(-1)^{k-1}}{k^2} B\left(\frac{k}{2},\frac{1}{2}\right) = \int_0^{\pi/2}dx\, \frac{\pi x \cos x}{\sin x} - \frac{x^2 \cos x}{\sin x} \\= -\int_0^{\pi/2}dx\, \pi \ln \sin x-2x\ln \sin x = \frac{\pi^2}{2} \ln 2 + 2 \int_0^{\pi/2}dx\, x\ln \sin x. $$ The latter integral can be calculated as follows: use the identity $\ln \sin x = -\ln 2 -\sum_{k\geq0} \cos(2 k x)$ and interchange summation and integration. This gives the value $$2\int_0^{\pi/2}dx\, x\ln \sin x = -\dfrac{\pi^2}{4}\ln 2 + \beta(3) = -\dfrac{\pi^2}{4}\ln 2 + \dfrac{7}{8} \zeta(3).$$ Adding everything up yields the correct value of the integral:

$$ -\frac{3}{4} \zeta(3) + \frac{1}{2} \left[\frac{\pi^2}{2} \ln 2 -\dfrac{\pi^2}{4}\ln 2 + \dfrac{7}{8} \zeta(3)\right] = \frac{\pi^2}{8} \ln 2 - \frac{5}{16} \zeta(3). $$

$\endgroup$
  • $\begingroup$ :):):) Like me, I knew you'd would obsess until you got it. Cool. $\endgroup$ – Cody Aug 9 '14 at 14:44
  • $\begingroup$ Nicely done. +1 $\endgroup$ – Random Variable Aug 9 '14 at 15:07
8
$\begingroup$

Very good input Jack. Clever use of the Fourier ln sum.

It would appear I managed to make some headway.

First, make the sub $t=\cos(x)$ to get:

$$-\int_{0}^{1}\frac{\ln(1+\sqrt{1-t^{2}})\ln(1+t)}{t}dt$$

Use the series for ln(1+t):

$$\int_{0}^{1}\ln(1+\sqrt{1-t^{2}})\sum_{k=1}^{\infty}\frac{(-1)^{k+1}t^{k-1}}{k}dt$$

Now, use the ln identity for arcsech(t):

$$\int_{0}^{1}\left(sech^{-1}(t)+\ln(t)\right)\sum_{k=1}^{\infty}\frac{(-1)^{k+1}t^{k-1}}{k}dt$$

distribute:

$$\int_{0}^{1}\sum_{k=1}^{\infty}\frac{(-1)^{k+1}t^{k-1}sech^{-1}(t)}{k}dt+\int_{0}^{1}\sum_{k=1}^{\infty}\frac{(-1)^{k}t^{k-1}\ln(t)}{k}$$

upon integrating the right side and noting the sum is a alternating zeta, we arrive at

$\displaystyle \frac{-3}{4}\zeta(3)\tag{1}$

The left side $\displaystyle \sum_{k=1}^{\infty}\frac{(-1)^{k+1}}{k}\int_{0}^{1}t^{k-1}sech^{-1}(t)dt$

make the sub $u=sech^{-1}(t), \;\ t=sech(u), \;\ dx=-\frac{\tanh(u)}{\cosh(u)}du$

$$\sum_{k=1}^{\infty}\frac{(-1)^{k}}{k}\int_{0}^{\infty}\frac{u\sinh(u)}{\cosh^{k+1}(u)}du$$

This integral can be related to the Beta function, and has probably already been done on the site.

That is, there is a general form that can be used:

$$\int_{0}^{\infty}\frac{u\sinh(u)}{\cosh^{2n+1}(u)}du=\frac{\sqrt{\pi}}{4n}\cdot \frac{\Gamma(n)}{\Gamma(n+1/2)}$$

But, in this case, $n=k/2$, so we get:

$$\frac{\sqrt{\pi}}{2}\sum_{k=1}^{\infty}\frac{(-1)^{k+1}\Gamma(k/2)}{k^{2}\Gamma(k/2+1/2)}$$

This sum evaluates to $$\frac{7}{16}\zeta(3)+\frac{\pi^{2}}{8}\ln(2)$$

combine this with the zeta result from up top in (1) and we get:

$$\frac{7}{16}\zeta(3)+\frac{\pi^{2}}{8}\ln(2)-\frac{3}{4}\zeta(3)$$

$$=\frac{\pi^{2}}{8}\ln(2)-\frac{5}{16}\zeta(3)$$

this works out and numerically checks.

ADDENDUM:

For the above Gamma sum, one could begin with the series:

$$\frac{\sqrt{\pi}}{2}\sum_{k=1}^{\infty}\frac{(-1)^{k+1}\Gamma(k/2)}{\Gamma(k/2+1/2)}x^{k-1}=\frac{\cos^{-1}(x)}{\sqrt{1-x^{2}}}\tag{2}$$, then integrate, divide by x, integrate, in order to hammer it into the proper form. I just ran through the steps with maple and it came out correctly.

EDIT: Derive closed form in (2):

$$\sum_{k=1}^{\infty}\frac{(-1)^{k-1}\Gamma(k/2)}{\Gamma(k/2+1/2)}x^{k-1}$$

$$=\sum_{k=1}^{\infty}(-1)^{k-1}\beta(k/2,1/2)x^{k-1}$$

$$=\sum_{k=1}^{\infty}\int_{0}^{1}t^{k/2-1}(1-t)^{-1/2}(-x)^{k-1}dt$$

$$=\int_{0}^{1}\frac{1}{\sqrt{t-t^{2}}(\sqrt{t}x+1)}dt$$

I'm sure there are more efficient ways, but I am just going to do this with subs.

Let $\displaystyle u=\sqrt{t}, \;\ dt=2udu$

$$2\int_{0}^{1}\frac{1}{\sqrt{1-u^{2}}(ux+1)}du$$

Let $\displaystyle u=sin(w)$

$$\int_{0}^{\frac{\pi}{2}}\frac{1}{\sin(w)x+1}dw$$

Let $\displaystyle w=2\tan^{-1}(z)$

$$4\int_{0}^{1}\frac{1}{z^{2}+2xz+1}dz=\frac{4\tan^{-1}\left(\frac{x+z}{\sqrt{1-x^{2}}}\right)}{\sqrt{1-x^{2}}}|_{0}^{1}$$

This is now not too bad and integrates in terms of arctan, which in turn can be converted to arccos:

$z=1$: $$4\frac{\tan^{-1}\left(\frac{x+1}{\sqrt{1-x^{2}}}\right)}{\sqrt{1-x^{2}}}=\frac{\pi+2\sin^{-1}(x)}{\sqrt{1-x^{2}}}$$

$z=0$ $$4\frac{\tan^{-1}\left(\frac{x}{\sqrt{1-x^{2}}}\right)}{\sqrt{1-x^{2}}}=\frac{4\sin^{-1}(x)}{\sqrt{1-x^{2}}}$$

because $\displaystyle \pi -2\sin^{-1}(x)=2\cos^{-1}(x)$:

$$\frac{2\sin^{-1}(x)+\pi}{\sqrt{1-x^{2}}}-\frac{4\sin^{-1}(x)}{\sqrt{1-x^{2}}}=\frac{2\cos^{-1}(x)}{\sqrt{1-x^{2}}}$$

$\endgroup$
  • $\begingroup$ Nice derivation of the closed form of the series! $\endgroup$ – user111187 Aug 18 '14 at 9:00
3
$\begingroup$

I think is it possible to exploit the fact the the two logarithmic factor have nice Fourier series. Since over $(0,\pi/2)$ we have: $$\log(2\cos x)=\sum_{n=1}^{+\infty}\frac{(-1)^{n+1}}{n}\cos(2nx),$$ $$\log(2\sin x)=-\sum_{n=1}^{+\infty}\frac{1}{n}\cos(2nx),$$ it follows that: $$\log(1+\sin t)=-\log 2+\sum_{k=1}^{+\infty}\frac{(-1)^{k+1}}{k}\cos(kx),$$ $$\log(1+\cos t)=-\log 2+\sum_{k=1}^{+\infty}\frac{(-1)^{k+1}}{k}\cos(2kx)+2\sum_{k=0}^{+\infty}\frac{(-1)^{k}}{2k+1}\sin((2k+1)x)$$ and it looks not so terrible to integrate the product of this two series times $\tan x$ over $(0,\pi/2)$.

In particular, Mathematica told me that $$\int_{0}^{\pi/2}\cos(mx)\log(1+\cos x)\tan x\,dx$$ is always a linear combination of $1$ and $\log 2$ (if $m$ is odd) or a linear combination of $1$ and $\zeta(2)$ (if $m$ is even). I don't believe in coincidences.

Continues.

$\endgroup$
  • 1
    $\begingroup$ I went this route as well, but while the integral was manageable I ran into convergence issues. Your series is slightly different from mine though, so I may have made a mistake. $\endgroup$ – David H Aug 8 '14 at 19:31

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.