36

I am trying to find a closed form for $$ \int_{0}^{1}\ln\left(\frac{x + 1}{2x^{2}}\right) {{\rm d}x \over \,\sqrt{\,{x^{2} + 2x}\,}\,}. $$

I have done trig substitution and it results in $$ \int_{0}^{1}\ln\left(\frac{x + 1}{2x^{2}}\right) {{\rm d}x \over \,\sqrt{\,{x^{2} + 2x}\,}\,} = \int_{0}^{\pi/3}\sec\left(\theta\right) \ln\left(\frac{\sec\left(\theta\right)} {2\left[\sec\left(\theta\right) - 1\right]^{\,2}} \right){\rm d}\theta $$ which doesn't help.

By part integration with $\displaystyle u = \ln\left(\frac{x + 1}{2x^{2}} \right)$, $\displaystyle\,\,{\rm d}v=\frac{\displaystyle\,\,{\rm d}x}{\,\sqrt{\,{x^{2} + 2x}\,}\,}$ also makes it more complicated.

I appreciate any help on this problem.

pisco
  • 18,983
mike
  • 667
  • Have you tried splitting the logarithm into a difference of logarithms? – Mike Nov 04 '22 at 06:42
  • Yes, I have tried splittng. – mike Nov 04 '22 at 07:02
  • The result fits on a single line – Claude Leibovici Nov 04 '22 at 08:20
  • @Zacky. It is not so hard. Even the antiderivative is workable if you enjoy polylogarithms. – Claude Leibovici Nov 04 '22 at 14:09
  • With Mathematica: $\frac{1}{2} \left(-8 \text{Li}_2\left(2-\sqrt{3}\right)+\text{Li}_2\left(-7+4 \sqrt{3}\right)-\ln ^2\left(2-\sqrt{3}\right)\right)+\frac{17 \pi ^2}{24}$ – Mariusz Iwaniuk Nov 04 '22 at 14:18
  • @MariuszIwaniuk. Just out of curiosity, are you able to compute $$\sum {n=1}^{\infty } \frac{B{\frac{1}{3}}\left(2 n+\frac{1}{2},0\right)}{n}$$ – Claude Leibovici Nov 05 '22 at 03:02
  • @MariuszIwaniuk. Using two different CAS, I had the same result without any possible simplification while (look in comments) the result seems to be $\frac {\pi^2}2$ !!! – Claude Leibovici Nov 05 '22 at 03:15
  • @KStarGamer. Thank you ! Would you have a look at my second edit ? Cheers – Claude Leibovici Nov 05 '22 at 03:53
  • 1
    This is as far as I was able to go. It's not much but it's something. – Accelerator Nov 05 '22 at 06:05
  • @ClaudeLeibovici .Directly Mathematica can't compute this sum, but with workaround, Integral representation of Beta function can we compute: Integrate[ Sum[t^(-(1/2) + 2 n)/(n (1 - t)), {n, 1, Infinity}], {t, 0, 1/3}]. – Mariusz Iwaniuk Nov 05 '22 at 10:40
  • I get $\displaystyle \frac{3}{2}\ln^2\left(2-\sqrt{3}\right)+2\ln 2\ln\left(2-\sqrt{3}\right)+4\text{Li}_2\left(\sqrt{3}-1\right)+\frac{1}{2}\text{Li}_2\Big(-(2-\sqrt{3})^2\Big)+\frac{\pi^2}{24}$ – FDP Nov 05 '22 at 18:33
  • @FDP, Is it possible to show this is equal to $\frac{\pi^2}{2}$ – mike Nov 05 '22 at 19:05
  • @Mike: still working on it. – FDP Nov 05 '22 at 19:08
  • @FDP, awesome! me too. – mike Nov 05 '22 at 19:09
  • I think I'm close to a solution. So far I have $$\frac{17\pi^2}{24}-\frac{\ln^2{(2+\sqrt{3})}}{2}+\frac{1}{2}\operatorname{Li}_2 (-7+4\sqrt{3})-4\operatorname{Li}_2(2-\sqrt{3})$$ – phi-rate Nov 05 '22 at 22:25
  • @mike how many digits did you verify until? It's possible that it's very close but not exact – Sidharth Ghoshal Nov 07 '22 at 22:54
  • @SidharthGhoshal Wolfram verifies it for us https://www.wolframalpha.com/input?i=integral+%282%2F%28pi%5E2%29%29+ln%28%28x%2B1%29%2F%282x%5E2%29%29%2Fsqrt%28x%5E2%2B2x%29%29+from+0+to+1 – phi-rate Nov 07 '22 at 22:55
  • Yea with mathematica using Integrate and then FullSimplify yields (17 \[Pi]^2)/24 - 1/2 Log[2 - Sqrt[3]]^2 - 4 PolyLog[2, 2 - Sqrt[3]] + 1/2 PolyLog[2, -7 + 4 Sqrt[3]] which is what everyone else is getting. Numerically this seems to equal $\frac{\pi^2}2$ (i tested up to 16 digits) but mathematica cannot simplify directly into it. Odd tbh – Max0815 Nov 08 '22 at 04:24
  • @Sidharth Ghoshal I think WolframAlpha clearly shows that the integral is equal to $\frac{\pi^2}{2}$ https://www.wolframalpha.com/input?i=pi%5E2%2F2+41+dgits – mike Nov 08 '22 at 07:01
  • https://www.wolframalpha.com/input?i=int_0%5E1+ln%28%28x%2B1%29%2F%282x%5E2%29%29%2Fsqrt%28x%5E2%2B2x%29+dx%2C+with+40+digs – mike Nov 08 '22 at 07:05
  • Ok to be fair tho if you cannot analytically prove the equivalence then numerical verification cannot ever prove it either. – Max0815 Nov 08 '22 at 20:48
  • The best i can do for now: if $\alpha=2-\sqrt{3}$, \begin{align}\text{Li}_2\left(\alpha^2\right)=\text{Li}_2\left(-\alpha^2\right)+\text{Li}_2\left(\frac{1}{2}\alpha\right)+\text{Li}_2\left(2\alpha\right)-\frac{\pi^2}{12}+\frac{1}{2}\ln^2(2\alpha)\end{align} – FDP Nov 08 '22 at 21:20
  • This one is cool too:\begin{align}\text{Li}_2\left(-\frac{1}{\alpha^2}\right)+\text{Li}_2\left(-\alpha^2\right)+2\ln^2\left(\alpha\right)+\frac{\pi^2}{6}=0\end{align} – FDP Nov 09 '22 at 10:34
  • @MariuszIwaniuk. Thank you ! It makes the problem more awful. Cheers :-) – Claude Leibovici Nov 10 '22 at 04:37
  • 4
    The result is exact for more than 1000 decimal places – Claude Leibovici Nov 11 '22 at 03:35
  • 1
    Don't trust Wolfram too much on the precision or number of decimals: https://www.wolframalpha.com/input?i=integral+%281.E-6+%2B+%282%2F%28pi%29%29+ln%28%28x%2B1%29%2F%282x%5E2%29%29%2Fsqrt%28x%5E2%2B2x%29%2Fpi%29+from+0+to+1 is still "1.", while it should be 1.000001 :) – Wolfgang Nov 11 '22 at 12:33
  • I feel like I am somewhat close to proving this conjecture true. I ran into taking the infinite series of harmonic numbers at one point, so I am trying to see how I can work with that. I might ask a separate query about it. – Accelerator Nov 14 '22 at 02:04
  • I asked this question the other day. I don't know how much you would be able to salvage from it, though. – Accelerator Nov 18 '22 at 07:25
  • 1
    Also, there's a solution already down below, so I don't think there is any need for the bounty, is there? I can type up my own solution, but it still involves the dilogarithms and would rely on the top-voted answer. – Accelerator Nov 18 '22 at 11:13
  • I've tried to solve this integral but none of them are exactly $\frac{\pi^2}{2}$. I check with WolframAlpha and it turns out its difference is around $6.21725 \times 10^{-15}$. So, I really doubt the conjecture. – Anastasiya-Romanova 秀 Nov 18 '22 at 12:18
  • 3
    @mike What do you mean by "Still no solution for this problem."? I believe pisco has already completed the solution from the dilogarithm expression. – ə̷̶̸͇̘̜́̍͗̂̄︣͟ Nov 22 '22 at 10:19
  • @TheSimpliFire lots of fantastic job done here. I was looking for a direct solution without dilogarithm. The solutions was based on the conjecture that the solution is $\frac{\pi^2}{2}. I was looking for a solution to get it directly. – mike Jan 18 '23 at 04:41

8 Answers8

19

FDP and phi-rate in their answers, each has shown the integral equals $$I=\frac{17\pi^2}{24}-\frac{\ln^2(2+\sqrt3)}{2}+\frac{1}{2}\operatorname{Li}_2(-7+4\sqrt3)-4\operatorname{Li}_2(2-\sqrt3)$$

So $I = \pi^2/2 \DeclareMathOperator{\Li}{Li}$ is equivalent to $$\tag{*} \text{Li}_2(-\chi^2) - 8 \text{Li}_2(\chi) = \log ^2 \chi-\frac{5 \pi ^2}{12} \quad \chi=2-\sqrt{3}$$

Most such numerical identities of polylogs are non-trivial, proving $(*)$ is actually the hard part. I outline two proofs of $(*)$, but only the first proof will be self-contained.


First proof: begin with the five-term functional equation (due to Kummer) $$\text{Li}_2\left(\frac{x (1-y)^2}{(1-x)^2 y}\right)=\text{Li}_2\left(\frac{1-y}{1-x}\right)+\text{Li}_2\left(-\frac{x (1-y)}{1-x}\right)+\text{Li}_2\left(\frac{x (1-y)}{(1-x) y}\right)+\text{Li}_2\left(-\frac{1-y}{(1-x) y}\right)+\frac{\log ^2(y)}{2}$$ let $u$ be such that $(1-u)/(1+u) = iu$, then $iu^2 = \chi$. Make $x=-u,y=u$, yields $$\begin{aligned} \Li_2 (u^2) &= \Li_2(iu^2)+\Li_2(-i)+\Li_2(-iu)+\Li_2(iu) + \frac{1}{2}\log^2 u \\ \iff\Li_2 (-i\chi) &= \Li_2(\chi)+\Li_2(-i)+\frac{1}{2}\Li_2(i\chi) + \frac{1}{2}\log^2 u\end{aligned}$$

taking sum with its complex conjugate, using $\Li_2(x) + \Li_2(-x) = \Li_2(x^2)/2$ proves $(*)$. QED.

One could prove a tri-log version $$\tag{**} \text{Li}_3\left(-\chi^2\right)-16 \text{Li}_3\left(\chi\right) = -\frac{23 \zeta (3)}{2}-\frac{2}{3} \log ^3\left(\sqrt{3}+2\right)+\frac{5}{6} \pi ^2 \log \left(\sqrt{3}+2\right)$$ via a nine-term functional equation of $\Li_3$.


Second proof: The number $\Li_n(\alpha)$ for certain special algebraic $\alpha$ is contained in a certain $\mathbb{Q}$-vector space known as colored multiple zeta values (CMZVs) of level 12. Such $\alpha$ includes $\chi,-\chi^2,\chi^4,15 \sqrt{3}-26$ etc.

For small $n$, the structure of this vector space is completely understood, a spanning set is recorded in my Mathematica package here. As a consequence, any relation between them can be demonstrated, such as $(*),(**)$. Further examples obtained from this approach are:

$$51 \text{Li}_2(\chi)+\text{Li}_2\left(15 \sqrt{3}-26\right)-\frac{9}{4} \text{Li}_2(\chi^4) = \frac{61 \pi ^2}{24}-6 \log ^2\left(\sqrt{3}+2\right)$$

$$-1224 \text{Li}_3(\chi)-8 \text{Li}_3\left(15 \sqrt{3}-26\right)+\frac{27}{2} \text{Li}_3(\chi^4) = -851 \zeta (3)-48 \log ^3\left(\sqrt{3}+2\right)+61 \pi ^2 \log \left(\sqrt{3}+2\right)$$

I don't know how to prove them from functional equations, but such proofs should exist. For another $\Li_2$ example, see here.


Some remarks:

  • We also have, with $G$ Catalan constant $$\int_{0}^{1}\frac{\log\left( \frac{x+1}{2x^2} \right)\log (1+x)}{\sqrt{x^2+2x} }dx = \frac{4 \pi G}{3}-\frac{1}{2} \pi ^2 \log (2)$$

  • The original purpose of the package mentioned above is to calculate integrals. It could also do this one directly. Since it only accepts integrand from 0 to 1, without radicals. We need to first perform the substitution $1+x = \frac{3+u^2}{3-u^2}$, $$\int_{0}^{1}\frac{\ln\left( \frac{x+1}{2x^2} \right)}{\sqrt{x^2+2x}}dx = \int_0^1 \frac{2 \log \left(\frac{1}{8} \left(\frac{9}{u^4}-1\right)\right)}{\sqrt{3} \left(1-\frac{u^2}{3}\right)} du$$ and

    MZIntegrate[(2 Log[1/8 (-1 + 9/u^4)])/(Sqrt[3] (1 - u^2/3)), {u, 0, 1}]
    

    immediately gives the result $\pi^2/2$.

pisco
  • 18,983
  • We need to first perform the substitution: It should be: $1+x=\frac{2+t^2}{2-t^2}$ – Mariusz Iwaniuk Nov 11 '22 at 11:32
  • Try:IntegrateChangeVariables[ Inactive[Integrate][Log[(x + 1)/(2 x^2)]/Sqrt[ x^2 + 2 x], {x, 0, 1}], t, (2 + t^2)/(2 - t^2) == x + 1, Assumptions -> t > 0]? – Mariusz Iwaniuk Nov 11 '22 at 11:42
  • @MariuszIwaniuk I think it's another bug of the experimental function IntegrateChangeVariables, where a bunch of them are already on Mathematica SE. If you do the substitution manually, then you'll see $1+x=\frac{3+t^2}{3-t^2}$ is the correct one. – pisco Nov 11 '22 at 12:27
  • Yes you're right. Thanks. – Mariusz Iwaniuk Nov 11 '22 at 12:31
  • Impressive, but you end up with some unproved residual, and after checking all of Lewin's formulas (many of them due to Euler) I'm afraid we'd rather bet on Augustin than on Leonhard. – user12030145 Nov 11 '22 at 23:10
  • @user12030145 What "unproved residual" is there? If you mean the five-term functional equation, it can be proved directly by differentiation. It is also in Lewin's book as (1.26) – pisco Nov 11 '22 at 23:54
  • @pisco No criticism of mine, I bow. It is just that I did not eliminate the Catalan constant while calculating the sum of conjugates in $Li_2(\pm i)$. Fixed now, thanks. – user12030145 Nov 12 '22 at 00:22
  • Your answer seems to be correct except for the fact that it's not equal to $\frac{\pi^2}{2}$ – Anastasiya-Romanova 秀 Nov 18 '22 at 12:21
  • 3
    What do you mean? This answer is used to prove that the OP's integral is equal to $\frac{\pi^2}{2}$. I just did the calculations too. @Anastasiya-Romanova秀 – Accelerator Nov 18 '22 at 18:56
  • 1
    Where is the functional equation from? I found this: https://www.maths.dur.ac.uk/users/herbert.gangl/kyoto.pdf but I cant tell for sure if your equation is a modified version of their 5 term functional equation on page 2 – Max0815 Nov 20 '22 at 07:45
  • 1
    @Max0815 Yes, they're equivalent. Starting from your function equation, replace $x/(1-y)$ by $\theta$, $y/(1-x)$ by $\phi$; and then $\theta$ by $-x(1-y)/(1-x)$ and $\phi$ by $-(1-y)/(y(1-x))$, you get what I wrote down. – pisco Nov 20 '22 at 10:25
14

You can "simplify" the problem using a first integration by parts to get rid of the logarithm $$u=\log \left(\frac{x+1}{2 x^2}\right)\quad \implies \quad du=-\frac{x+2}{x^2+x}$$ $$dv=\frac{1}{\sqrt{x^2+2 x}}\quad \implies \quad v=2 \tanh ^{-1}\left(\sqrt{\frac{x}{x+2}}\right)$$

Using the bounds $u\,v=0$ and we are left with $$I=2\int_0^1\frac{(x+2) }{x^2+x}\tanh ^{-1}\left(\sqrt{\frac{x}{x+2}}\right)dx$$

Now $$\sqrt{\frac{x}{x+2}}=t \implies x=\frac{2 t^2}{1-t^2}\implies dx=\frac{4 t}{\left(1-t^2\right)^2}$$ $$I=8\int_0^{\frac{1}{\sqrt{3}}}\frac{\tanh ^{-1}(t)}{t-t^5}\,dt$$ Now, using partial fraction decomposition $$\frac{1}{t-t^5}=\frac{1}{t(1-t^2)(1+t^2)}=-\frac{t}{2 \left(t^2+1\right)}-\frac{1}{4 (t-1)}-\frac{1}{4 (t+1)}+\frac{1}{t}$$ and now would arrive a bunch of polylogarithm functions.

The simplest is

$$\int \frac{\tanh ^{-1}(t)}{t}\,dt=\frac{1}{2} (\text{Li}_2(t)-\text{Li}_2(-t))$$

Fortunately, between the given bounds everything simplify a lot.

I let you the pleasure of computing the pieces.

Edit

If we write

$$\frac{\tanh ^{-1}(t)}{t-t^5}=\frac{\tanh ^{-1}(t)}{t}+\sum_{n=1}^\infty t^{4n-1}\,\tanh ^{-1}(t)$$ we have $$I=8\int_0^{\frac{1}{\sqrt{3}}}\frac{\tanh ^{-1}(t)}{t-t^5}\,dt=4 \left(\text{Li}_2\left(\frac{1}{\sqrt{3}}\right)-\text{Li}_2\left (-\frac{1}{\sqrt{3}}\right)\right)+$$ $$\sum_{n=1}^\infty\frac{9^{-n} \log \left(2+\sqrt{3}\right)-B_{\frac{1}{3}}\left(2 n+\frac{1}{2},0\right)}{n}$$ that is to say $$I=4 \left(\text{Li}_2\left(\frac{1}{\sqrt{3}}\right)-\text{Li}_2\left (-\frac{1}{\sqrt{3}}\right)\right)+\log \left(\frac{9}{8}\right) \log \left(2+\sqrt{3}\right)-\sum_{n=1}^\infty \frac{B_{\frac{1}{3}}\left(2 n+\frac{1}{2},0\right)}{n}$$ Numerically, the sum of the first and second terms is $4.96991$ and the summation is only $0.03511$

Edit

There is something very strange : two different $CAS$ give as result $$I =\frac{17 \pi ^2}{24}-\frac{1}{2} \left(8 \text{Li}_2\left(2-\sqrt{3}\right)-\text{Li}_2\left(-7+4 \sqrt{3}\right)+\log ^2\left(2-\sqrt{3}\right)\right)$$ without any further simplification while Wolfram Alpha gives $1.$ when it is written as $$\int_0^1 \frac2{ \pi ^2} \frac{\log \left(\frac{x+1}{2 x^2}\right)}{\sqrt{x^2+2 x}}\,dx$$ Without the factor, it just return a decimal value.

Using RootApproximant[%] also behaves differently with or without the factor inside the integrand. $$\frac{1}{2} \left(8 \text{Li}_2\left(2-\sqrt{3}\right)-\text{Li}_2\left(-7+4 \sqrt{3}\right)+\log ^2\left(2-\sqrt{3}\right)\right)=2.056167583560283045590519$$ Passed to the $ISC$, it is returned as $\frac{5 \pi ^2}{24}$

  • The integral is equal to $\frac{\pi^2}{2}$. Can this be evaluated from your solution? – mike Nov 05 '22 at 02:36
  • @mike. It is not this number. In comments, Mariusz Iwaniuk gave you the exact value. – Claude Leibovici Nov 05 '22 at 02:40
  • @ Claude Leibovici If you type the following into WolframAlpha you can see the answer is $\frac{\pi^2}{2}$.

    integral (2/(pi^2)) ln((x+1)/(2x^2))/sqrt(x^2+2x)) from 0 to 1

    – mike Nov 05 '22 at 02:52
  • @mike. Send me a link showing your input in WA. – Claude Leibovici Nov 05 '22 at 02:57
  • https://www.wolframalpha.com/input?i=integral+%282%2F%28pi%5E2%29%29+ln%28%28x%2B1%29%2F%282x%5E2%29%29%2Fsqrt%28x%5E2%2B2x%29%29+from+0+to+1 – mike Nov 05 '22 at 02:59
  • @mike. For a surprise, this is a surprise (as we say in French) ! It seems to be true. I do not understand how two different CAS give the same result without any possible simplification. – Claude Leibovici Nov 05 '22 at 03:13
  • It's true; I have no idea either. – mike Nov 05 '22 at 03:25
  • @mike. Look at https://www.wolframalpha.com/input?i=Integrate%5BLog%5B%281+%2B+x%29%2F%282x%5E2%29%5D%2FSqrt%5B2x+%2B+x%5E2%5D%2C%7Bx%2C0%2C1%7D%5D – Claude Leibovici Nov 05 '22 at 03:39
  • GP PARI using lindep function gives that the result is $\dfrac{\pi^2}{2}$. – FDP Nov 05 '22 at 07:53
  • $$\frac{1}{(2+\sqrt{3})^2}=7-4\sqrt{3}$$ Maybe there's a dilogarithm property out there that can help you. Either way, it's a cool expression – phi-rate Nov 05 '22 at 14:27
  • @mike WA outputs the integral as "1.", note the dot following 1, indicating it is a floating point number, so WA is using numerical integration. I believe neither Mathematica nor WA currently has the capability to prove it's exactly 1. – pisco Nov 10 '22 at 14:11
  • @pisco. You are right ! Being blind, I did not see the dot.. By the way, your answer is very good and then $\to +1$. Cheers and thanks for posting. – Claude Leibovici Nov 10 '22 at 14:16
  • @mike I own Mathematica 13.0. It is great, but it sometimes mingles symbolic and numerical reasoning. Sometimes symbolic expressions cannot be proved to be equal by the software but are presumed to be so, for symbolic reasoning to proceed. This yields warnings, and decimal dots in rounded numbers. In other contexts, assumptions are made implicitly and not reported. So symbolic inference can theoretically be just wrong, if such numerically-based or context-based assumptions come out to be false. Yes, I know, this is a bit on the skeptical side. – user12030145 Nov 11 '22 at 17:01
  • @Claude, based on your calculations also we have: $\int_{0}^{1}\frac{\ln\left( \frac{x+1}{2x^2} \right)}{\sqrt{x^2+2x}}dx=\frac{8}{\sqrt{3}}\sum_\limits{n=0}^{\infty}\sum_\limits{m=0}^{\infty}\frac{1}{(2n+4m+1)(2n+1){3^{(n+2m)}}} $ – mike Nov 12 '22 at 06:01
  • @mike. Based on which calculations ? This is very interesting. How did you arrive to this double summation ? – Claude Leibovici Nov 12 '22 at 06:29
  • @mike. I posted another answer based on your last remark. Since downvoted, I prefered to delete it. – Claude Leibovici Nov 12 '22 at 15:27
  • @ Claude Leibovici, based on your calculations $I=8\int_0^{\frac{1}{\sqrt{3}}}\frac{\tanh ^{-1}(t)}{t-t^5},dt$ therefore $I=8\int_0^{\frac{1}{\sqrt{3}}}\frac{\tanh ^{-1}(t)}{t-t^5},dt=8\int_0^{\frac{1}{\sqrt{3}}}\frac{1}{(1-t^4)}{\frac{\tanh ^{-1}(t)}{t}},dt=8\int_0^{\frac{1}{\sqrt{3}}}\sum_\limits{m=0}^{\infty}t^{4m}\sum_\limits{n=0}^{\infty}\frac{t^{2n+1}}{2n+1},dt $ ... – mike Nov 13 '22 at 03:34
  • $=8\int_0^{\frac{1}{\sqrt{3}}}\sum_\limits{m=0}^{\infty}t^{4m}\sum_\limits{n=0}^{\infty}\frac{t^{2n+1}}{2n+1},dt=8\int_0^{\frac{1}{\sqrt{3}}}\sum_\limits{m=0}^{\infty}\sum_\limits{n=0}^{\infty}\frac{t^{2n+4m+1}}{2n+1},dt=8\sum_\limits{m=0}^{\infty}\sum_\limits{n=0}^{\infty}\int_0^{\frac{1}{\sqrt{3}}}\frac{t^{2n+4m+1}}{2n+1},dt$... – mike Nov 13 '22 at 03:35
  • $8\sum_\limits{m=0}^{\infty}\sum_\limits{n=0}^{\infty}\int_0^{\frac{1}{\sqrt{3}}}\frac{t^{2n+4m}}{2n+1},=8\sum_\limits{m=0}^{\infty}\sum_\limits{n=0}^{\infty}\frac{t^{2n+4m+1}}{(2n+1)(2n+4m+1)}\big|^\sqrt{3}0=\frac{8}{\sqrt{3}}\sum\limits{m=0}^{\infty}\sum_\limits{n=0}^{\infty}\frac{t^{2n+4m+1}}{(2n+1)(2n+4m+1){3^{(n+2m)}}}$ – mike Nov 13 '22 at 03:35
  • Thanks, Mike, for the clarification. In fact (this was in my last deleted answer (too many downvotes), this leads to Hurwitz-Lerch transcendent function and then to linear combinations of two different Gaussian hypergeomatric function (depends on the parity of the outer index). – Claude Leibovici Nov 13 '22 at 04:08
  • @ Claude Leibovici, Most of the time, downvotes are unfair; instead, it is ideal to offer a solution. If the problem is too easy, just skipping the problem is a way to leave it for those who can still benefit from it. – mike Nov 13 '22 at 18:14
  • @mike I used a similar approach to get the following $$I=\ln{(9/8)}\ln{(2+\sqrt3)}+8\operatorname{Li}2(\frac{1}{\sqrt3})-2\operatorname{Li}_2(\frac{1}{3})-8\sum{n=0}^\infty \sum_{k=1}^\infty\frac{(1/\sqrt3)^{2n+4k+1}}{4k(2n+4k+1)}$$ I tried going further and ended up with this integral: $$\int_0^{\frac{1}{\sqrt3}}\frac{\ln(1-x^4)}{1-x^2}dx$$ It looks better but I'm not sure it's much easier – phi-rate Nov 13 '22 at 18:50
  • @phi-rate, this looks like a cool integral, however, it represents a negative value. – mike Nov 13 '22 at 19:56
  • @mike there's more in the expression. I did not type the full expression. I'm not saying they equivalate. The double sum in my comment is equivalent to that integral – phi-rate Nov 14 '22 at 01:53
  • @chroma Your derived integral looks much better. – mike Jan 03 '23 at 06:43
  • @mike in practice, its not that much better – phi-rate Jan 03 '23 at 16:44
  • @mike I have found a simpler way to connect the derived integral to $I$ and I have added it to my answer – phi-rate Jan 12 '23 at 02:39
  • @chroma Thanks! – mike Jan 12 '23 at 04:31
10

Hyperbolic substitution is very convenient for this integral. We can substitute $x+1=\cosh t$

$$I=\int_0^1 \ln{\left(\frac{x+1}{2x^2}\right)}\frac{dx}{\sqrt{x^2+2x}}=\int_0^{\ln{(2+\sqrt 3)}}\ln\left(\frac{\cosh t}{2(\cosh t-1)^2}\right)dt$$$$=\int_0^{\ln{(2+\sqrt 3)}}\ln\left(\frac{e^{-t}(1+e^{-2t})}{(1-e^{-t})^4}\right)dt$$ When we split the integral we get $$I=-\frac{\ln^2(2+\sqrt 3)}{2}+\int_0^{\ln{(2+\sqrt 3)}}\ln (1+e^{-2t})dt-4\int_0^{\ln{(2+\sqrt 3)}}\ln(1-e^{-t})dt$$ For the first integral we can apply the substitution $u=-e^{-2t}$, and for the second integral we can use $v=e^{-t}$ This will allow us to use the dilogarithm. $$I=-\frac{\ln^2(2+\sqrt 3)}{2}+\frac{1}{2}\int_{-1}^{-7+4\sqrt3}\frac{-\ln (1-u)}{u}du-4\int_1^{2-\sqrt 3}\frac{-\ln(1-v)}{v}dv$$ Using $\int \frac{-\ln(1-x)}{x}dx=\operatorname{Li}_2(x)+C$ we get $$I=\frac{17\pi^2}{24}-\frac{\ln^2(2+\sqrt3)}{2}+\frac{1}{2}\operatorname{Li}_2(-7+4\sqrt3)-4\operatorname{Li}_2(2-\sqrt3)$$ ATTEMPT 2

Refer to C. Leibovici's answer, apply partial fractions and integration by parts $$I=8\int_0^{\frac{1}{\sqrt3}}\frac{\operatorname{artanh} x}{x(1-x^4)}dx=8\int_0^{\frac{1}{\sqrt3}}\frac{\operatorname{artanh}x}{x}dx+2\int_0^{\frac{1}{\sqrt3}}\frac{4x^3\operatorname{artanh}x}{1-x^4}dx$$$$=8X_2\left(\frac{1}{\sqrt3}\right)+\ln(2-\sqrt3)\ln\left(\frac{8}{9}\right)+2\int_0^\frac{1}{\sqrt3} \frac{\ln(1-x^4)}{1-x^2}dx$$ $X_2(t)$ is the Legendre chi function

Using $x^4-1=(x+1)(x-1)(x+i)(x-i)$ in conjunction with logarithm properties and partial fractions, we can decompose the integral into 8 integrals of the form $\int \frac{\ln(x+a)}{x+b}dx$. 2 of them, $\int \frac{\ln(1+x)}{1+x}dx$ and $\int \frac{\ln(1+x)}{1+x}dx$ can be evaluated without dilogarithms. The other 6 each give us 2 dilogarithmic terms. This gives us 12 dilogarithms to simplify in total, and it might work

phi-rate
  • 2,345
10

Previous try

It should be $\pi^2/2$ but showing this requires invoking the Cauchy residue theorem, as is often the case when $\pi$ creeps in integrals.
The demonstration proceeds in two steps.

I - Contour integration

  1. First refer to C. Leibovici's answer:

$$ \int \frac{1}{\sqrt{x^2+2 x}} \, dx = 2 \tanh ^{-1}\left(\sqrt{\frac{x}{x+2}}\right)$$

then integrate by parts using this result, which yields:

$$I = \int_{0}^{1/\sqrt{3}} \frac{8 \tanh ^{-1}(v)}{v \left(1-v^4\right)} \, dv$$

  1. Now notice that the (simple) poles of the integrand $f(v)$ are:

$$ \{-1,0,-i,i,1\} $$

  1. Choose a rectangular integration contour such as: $$ \{\frac{-1}{\sqrt{3}}, 0^{+}, \frac{1}{\sqrt{3}}, c\, i + \frac{1}{\sqrt{3}}, c \, i -\frac{1}{\sqrt{3}}\},$$ with: $\, \, c \rightarrow \, +\infty $
    $0^{+}$ intends to symbolize a little dent around the 0 pole, this is to make it clean but anyway the residue there is null.

  2. Compute the residue at $i$, this is easy and I'll leave it to you. Hint: notice that $ \tanh^{-1}(i) = -i \frac{\pi}{4} $, which is the true reason why you get in $\pi$ along the way.

  3. Check that at infinity along the upper side of the rectangle contour, $f(v)$ is negligible, which is fairly easy too.

  4. The hard bit is to show that the real part of the contour integration along one or the other of the vertical sides of the rectangle is null just at the points chosen. With other points, it is not null, and it does not cancel (it adds up) when the contour integration along the two vertical sides is performed:

$$ \Re{ \int_{\pm\frac{1}{\sqrt{3}}}^{\pm\frac{1}{\sqrt{3}} + i \infty} f(v) \, dv} = 0 \,\,\, (\mathscr{C})$$

The easy bit is that, in any case, the imaginary part does obviously cancel, owing to the fact that $f(v)$ is even.

The hard bit is not demonstrated yet, but does not look intractable.
It is anyway an improvement to replace a conjecture on a multiple of $\pi^2$ by a conjecture about a null value, with a hint at optimization (for which there are lots of methods around).

  1. Applying Cauchy's residue theorem you now get the result:

$$ \int_R f(z) \, dz = 2i\pi \, Res(f, i) = 2i\pi (-(i \frac{\pi}{2})) = \pi^2$$

  1. Considering the fact that the integral vanishes along the three sides above the real axis when you let $c$ grow to infinity, you now are left with the integration along the real segment $[-\frac{1}{\sqrt{3}}, \frac{1}{\sqrt{3}}]$, whose value is 2I, owing to the parity of $f$. Hence the result.

II - Towards the proof of conjecture ($\mathscr{C}$)

Brute-force integration of $f(v)$ along a vertical rectangular side (let us take the right one) $[\frac{1}{\sqrt{3}}, \frac{1}{\sqrt{3}} + i y]$:

$$f(\frac{1}{\sqrt{3}} + i \,y) = \frac{8 \tanh ^{-1}\left(\frac{1}{\sqrt{3}}+i y\right)}{\left(1-\left(\frac{1}{\sqrt{3}}+i y\right)^4\right) \left(\frac{1}{\sqrt{3}}+i y\right)}$$

This gives the imaginary part (owing to contour integration, $f$ must be multiplied by $i$, the derivative of the parametrized contour side):

$$g(y) = \frac{18 \left(90 \sqrt{3} y^4 \tan ^{-1}(\frac{y}{\frac{1}{\sqrt{3}}-1})-2 \sqrt{3} \left(-45 y^4+30 y^2+8\right) \tan ^{-1}(\frac{y}{1+\frac{1}{\sqrt{3}}})+4 \sqrt{3} \left(15 y^2+4\right) \tan ^{-1}\left(\frac{1}{2} \left(\sqrt{3}+3\right) y\right)+3 \left(-9 y^4+30 y^2+4\right) y \log \left(1-\frac{4 \sqrt{3}}{3 y^2+2 \sqrt{3}+4}\right)\right)}{\left(3 y^2+1\right) \left(336 y^2+27 \left(y^2+2\right) \left(3 y^2-2\right) y^4+64\right)}$$

Integration between $y = 0$ and $y= N$ (let us denote this integral $J(N)$) then using an asymptotic series when N tends to infinity reduces $\mathscr{(C)}$ to $\mathscr{(C')}$:

$$J(N) \rightarrow 0 \,\,\, \mathscr{(C')}$$

(numerically this much is proved).

Plot of $J(N)$ as $N \rightarrow +\infty$
Plot of J(N) as N \rightarrow +\infty

When proved this will prove ($\mathscr{C}$) and complete the demonstration. I'll leave this for further edits.

This attack procedure may well work if one finds a good absolutely dominating function for $\Im{f(\frac{1}{\sqrt{3}} + i \,y)}$ for which integration and asymptotic approximation may work without too much toil.

III - Edit: going further and final comments

Integrating $g(y)$ in II is a bit cumbersome, to say the least, but can be done.
Doing this goes through the following steps:

1. Decompose the denominator into simple poles to find the values for $\alpha$ above.

2. Integrate:

$$I_k(N) = \int_0^N \frac{u^k \tan^{-1}(u)}{u - \alpha} \, du, \,\, k=0, 2, 4, \, \alpha \in \mathbb{R} $$

3. Integrate:

$$ L_k(N) = \int_0^N y^k \frac{\log(1-\frac{4 \sqrt{3}}{3 y^2+2 \sqrt{3}+4})}{y - \alpha} \, dy, k=1, 3, 5, \, \alpha \in \mathbb{R} $$

4. Expand each of the above results by N and sum the different polynomial contributions

5. Simplify the resulting expression and let N tend to infinity.

On these steps:

Point 1

The poles of the denominator are :

$$\left\{-\frac{i}{\sqrt{3}}, \frac{i}{\sqrt{3}}, -i \sqrt{\frac{4}{3}-\frac{2}{\sqrt{3}}}, i \sqrt{\frac{4}{3}-\frac{2}{\sqrt{3}}}, \frac{1}{3} \left(-3-i \sqrt{3}\right),\frac{1}{3} \left(3-i \sqrt{3}\right), \frac{1}{3} \left(-3+i \sqrt{3}\right), \frac{1}{3} \left(3+i \sqrt{3}\right), -i \sqrt{\frac{2}{3} \left(\sqrt{3}+2\right)}, i \sqrt{\frac{2}{3} \left(\sqrt{3}+2\right)}\right\}$$

So the denominator easily simplifies into a sum of 10 simple poles. Each must be decomposed into 3 arctangent-type and 1 logarithm-type numerator. Then the polynomial factors are expanded to end up with $I_k(N)$ and $L_k(N)$ expressions. In all there must be 60 $I_k$-ype integrals and 30 $L_k$-type integrals to consider.

Point 2

Actually the case $k = 0$ is well-known: it is a generalized inverse tangent integral of second order (see 1), for which an asymptotic expansion is known:

$$Ti_2(N, -\alpha) = \frac{1}{2} \pi \ln{N} - \frac{1}{4} \pi \ln{(1 + \alpha^2)} - Ti_2(\alpha) + \tan^{-1}(\alpha) \ln{|\alpha|} + O(\frac{1}{N})$$

The other cases can be obtained from this starting point by classic integration techniques (integration by parts and auxiliary parameter derivation), so there is reason to consider that an asymptotic expansion of the arctangent part of $J(N)$ is algorithmically doable.

Also, maybe the good strategy would be to find an asymptotic expansion for $I_k(N)$ from the above formula before tackling the 60 $I_k$-type integrals. This would simplify the task by an order of magnitude.

Further, $L_k(N)$ is a sum of classic integrals that can be found in Gradshtein or Prudnikov.

There is reason to believe that proving or refuting conjecture $\mathscr{(C')}$ is an algorithmically doable task. It is just quite cumbersome and will take a few dozen classic operations; yet not anything that exceeds the power of CAS systems, so there must be a conclusion in the end of the daunting calculus: true or false.

We expect the asymptotic logarithms and arctangents to cancel out easily; however we will be stranded with the second-order inverse tangent expressions taken at the 10 poles. These are closely associated values, so we expect them to cancel out; but to reach this conclusion we will need formulas relating $Ti_2$ values of comparable complexity as the above solution using the five-term Kummerian $Li_2$ relation (see 1, appendix). This may be doable again but, at the end of the day, looks like a much-complexified version of the same solution.

It looks fascinating that for such integrals, there seems to be no way out other than by manipulating polylogs and associated functions.


1 Leonard Lewin, Polylogarithms and associated functions, 1981, North Holland, chapter 3 and expansion p. 288 (11).

user12030145
  • 1,093
  • 2
    Very good procedure. I may be rather cynical but it appears $J(+\infty)=0$ will reduce to another long dilogarithm identity, but there is a chance it could be more straightforward to prove than the existing claim in the other four answers. – ə̷̶̸͇̘̜́̍͗̂̄︣͟ Nov 09 '22 at 21:12
  • Numerically J(N) tends to zero, this much is clear, and quite fast. One has to find the good dominating function and perform less-brute-force integration and/or series expansion to prove the result. But yes, it is not that easy and I have failed so far to reach a neat conclusion. Overall it looks simpler than the original question because you can use simplifications in the dominating function if you are clever enough to find a good one. – user12030145 Nov 09 '22 at 22:18
6

\begin{align}J&=\int_{0}^{1}\frac{\ln\left( \frac{x+1}{2x^2} \right)}{\sqrt{x^2+2x}}dx\\ &\overset{z=\frac{x}{2+x}}=2\int_0^{\frac{1}{\sqrt{3}}}\frac{\ln\left(\frac{1-z^4}{8z^4}\right)}{1-z^2}dz\\ &\overset{\text{IBP}}=-\underbrace{\left[\ln\left(\frac{1-z}{1+z}\right)\ln\left(\frac{1-z^4}{8z^4}\right)\right]_0^{\frac{1}{\sqrt{3}}}}_{=0}-4\int_0^{\frac{1}{\sqrt{3}}} \frac{\ln\left(\frac{1-z}{1+z}\right)}{z(1-z^2)(1+z^2)}dz\\ &\overset{w=\frac{1-z}{1+z}}=-\int_{\frac{\sqrt{3}-1}{\sqrt{3}+1}}^1 \frac{(1+w)^3\ln w}{w(1-w)(1+w^2)}dw\\ &=\frac{1}{2}\ln^2\left(2+\sqrt{3}\right)-4\int_{\frac{\sqrt{3}-1}{\sqrt{3}+1}}^1 \frac{\ln w}{1-w}dw-2\underbrace{\int_{\frac{\sqrt{3}-1}{\sqrt{3}+1}}^1 \frac{w\ln w}{1+w^2}dw}_{\text{IBP}}\\ &=\frac{1}{2}\ln^2\left(2+\sqrt{3}\right)-4\int_{\frac{\sqrt{3}-1}{\sqrt{3}+1}}^1 \frac{\ln w}{1-w}dw-\Big[\ln(1+w^2)\ln w\Big]_{\frac{\sqrt{3}-1}{\sqrt{3}+1}}^1+\underbrace{\int_{\frac{\sqrt{3}-1}{\sqrt{3}+1}}^1 \frac{\ln(1+w^2)}{w}dw}_{y=w^2}\\ &=\frac{3}{2}\ln^2\left(2-\sqrt{3}\right)+2\ln 2\ln\left(2-\sqrt{3}\right)-4\int_{2-\sqrt{3}}^1 \frac{\ln w}{1-w}dw+\frac{1}{2}\int_{(2-\sqrt{3})^2}^1 \frac{\ln(1+y)}{y}dy\\ &=\frac{3}{2}\ln^2\left(2-\sqrt{3}\right)+2\ln 2\ln\left(2-\sqrt{3}\right)+4\text{Li}_2\left(\sqrt{3}-1\right)+\frac{1}{2}\text{Li}_2\Big(-(2-\sqrt{3})^2\Big)\underbrace{-\frac{1}{2}\text{Li}_2\left(-1\right)}_{=\frac{\pi^2}{24}}\\ &=\frac{3}{2}\ln^2\left(2-\sqrt{3}\right)+2\ln 2\ln\left(2-\sqrt{3}\right)+4\text{Li}_2\left(\sqrt{3}-1\right)+\frac{1}{2}\text{Li}_2\Big(-(2-\sqrt{3})^2\Big)+\frac{\pi^2}{24} \end{align}

To be continued...

First addendum:: Using, \begin{align}0<x<1,\text{Li}_2\left(x\right)+\text{Li}_2\left(1-x\right)=\frac{\pi^2}{6}-\ln x\ln(1-x)\end{align}

One obtains, \begin{align}&=\frac{3}{2}\ln^2\left(2-\sqrt{3}\right)+2\ln 2\ln\left(2-\sqrt{3}\right)+4\text{Li}_2\left(\sqrt{3}-1\right)+\frac{1}{2}\text{Li}_2\Big(-(2-\sqrt{3})^2\Big)+\frac{\pi^2}{24}\\ &=\boxed{\frac{17\pi^2}{24}-\frac{1}{2}\ln^2\left(2-\sqrt{3}\right)-4\text{Li}_2\left(2-\sqrt{3}\right)+\frac{1}{2}\text{Li}_2\Big(-(2-\sqrt{3})^2\Big)}\\ \end{align}

FDP
  • 13,647
5

Let $$x=y^2,\quad y^2+1=t,\tag1$$ then $$I=\int\limits_0^1 \dfrac{\ln\left(\dfrac{x+1}{2x^2}\right)}{\sqrt{x^2+2x}}dx =\int\limits_0^1 \dfrac{\ln\left(\dfrac{y^2+1}{2y^4}\right)}{\sqrt{y^4+2y^2}}\, 2y\,\text dy =\int\limits_1^2 \dfrac{\ln\left(\dfrac{t}{2(t-1)^2}\right)}{\sqrt{t^2-1}}\,\text dt,$$ $$I=I_1-2I_2-I_3,$$ where \begin{cases} I_1=\int\limits_1^2 \dfrac{\ln t\,\text dt}{\sqrt{t^2-1}} = \dfrac{\pi^2}{24} -\dfrac12 \operatorname{Li_2}\left(\dfrac{2 - \sqrt3}4\right) - \ln^2 2-\dfrac14\arccos^2 2\\ I_2=\int\limits_1^2 \dfrac{\ln(t-1)\,\text dt}{\sqrt{t^2-1}} =-\dfrac{\pi^2}3+2\operatorname{Li}_2(2-\sqrt3) +6\operatorname{arcsinh}^2\dfrac1{\sqrt2} -4 \operatorname{arcsinh}\dfrac 1{\sqrt2}\,\log(1+\sqrt3)\\ I_3=\int\limits_1^2 \dfrac{\ln 2\text dt}{\sqrt{t^2-1}} = \dfrac12\ln2 \ln\left(4\sqrt{3}+7\right) = \ln2 \ln\left(2+\sqrt{3}\right)\\ \arccos 2 = i\ln(2+\sqrt3), \end{cases}

Therefore, $$I=\dfrac{17\pi^2}{24}-\dfrac12 \operatorname{Li_2}\left(\dfrac{2 - \sqrt3}4\right)-4\operatorname{Li}_2(2-\sqrt3)-\ln^2 2 -\ln\left(2+\sqrt{3}\right)\ln 2$$ $$+\dfrac1{4}\ln^2(2+\sqrt3) -12 \operatorname{arcsinh}^2 \dfrac1{\sqrt2}+8\operatorname{arcsinh}\dfrac 1{\sqrt2}\,\ln(1+\sqrt3),$$ $$I\approx4.9348022005446793094$$ (see also Wolfram Alpha calculations), in accordance with numeric calculations.

1

Let the given integral be $I$. After the substitution $x+1=\sec t $, integrtation by parts and some tricks in the integrand, we have $$I=\int_0^{\frac{\pi}{3}}\frac{\sin t+\tan t}{1-\cos t}\ln\left(\frac{\cos t}{1-\sin t}\right)dt.$$ After the $z=\tan\left(\frac{t}{2}\right)$ substitution, $$I=\int_0^{\frac{1}{\large\sqrt{3}}}\frac{4\ln\left(\frac{1+z}{1-z}\right)}{z(1-z^2)(1+z^2)}dz.$$ I think this expression is obtained by Claude Leibovici and FDP.

user12030145 mentioned about this contour integral. Let $C_1=\{z=-\frac{1}{\sqrt{3}}+iy| y:\infty\rightarrow 0\}$, $C_2=\{z=\frac{1}{\sqrt{3}}+iy| 0<y<\infty\}$ and $f(z)=\frac{4\ln\left(\frac{1+z}{1-z}\right)}{z(1-z^2)(1+z^2)}$. Then, branch cut of the logarithm is outside and the only singularity inside and on the contour $C_1+[-\frac{1}{\sqrt{3}},\frac{1}{\sqrt{3}}]+C_2$ is $z=i$ with residue $Res_{z=i}f(z)=\pi^2$. $z=0$ is fake singularity. Hence, since $f(z)$ is even function, the integral on $C_1+C_2$ is equal to the integral on the line $\{\frac{1}{\sqrt{3}}+iy|y\in\Bbb{R}\}$ and by residue theorem we have $$2I+\int_{\frac{1}{\sqrt{3}}-i\infty}^{\frac{1}{\sqrt{3}}+i\infty}f(z)dz=\pi^2$$ So, the problem reduces to showing that $\int_{\frac{1}{\sqrt{3}}-i\infty}^{\frac{1}{\sqrt{3}}+i\infty}f(z)dz=0.$ Nothing new so far, since I think user12030145 conjectured about such a thing. I think the real part of this integral is zero because of the Maclaurin series of $f(z)$, so it has to be zero.

Bob Dobbs
  • 10,988
1

We want to evaluate $$ I=\int^{1}_{0}\frac{\log\left((x+1)/x^2\right)}{\sqrt{x^2+2x}}dx. $$ We can easily write $I$ in the form $$ I=2\int^{1}_{1/2}\frac{\log(\sqrt{t}/(1-t))}{t\sqrt{1-t^2}}dt=4\int^{1}_{1/\sqrt{2}}\log\left(\frac{t}{1-t^2}\right)\frac{dt}{t\sqrt{1-t^4}}= $$ $$ =\tanh^{(-1)}\left(\frac{\sqrt{3}}{2}\right)\log 2-\int^{1}_{1/2}\frac{t+1}{t-1}t^{-1}\tanh^{(-1)}\left(\sqrt{1-t^2}\right)dt. $$ Set $t=i\tan t'$, then $I$ becomes $$ I=\tanh^{(-1)}\left(\frac{\sqrt{3}}{2}\right)\log 2- $$ $$ -\int^{-i\infty}_{-i\tanh^{(-1)}(1/2)}\left(-2i-\cot t'+\tan t'\right)\tanh^{(-1)}(\sec t')dt'= $$ $$ \frac{1}{2}i\pi \log(2+\sqrt{3})-\int^{-i\infty}_{-i\tanh^{(-1)}\left(\frac{1}{2}\right)}\left(-2it'-\log\left(\frac{1}{2}\sin(2t')\right)\right)\csc(t')dt'= $$ $$ \frac{1}{2}i\pi \log(2+\sqrt{3})+2i\int^{-i\infty}_{-i\tanh^{(-1)}\left(\frac{1}{2}\right)}t'\csc(t')dt'+ $$ $$ +\int^{-i\infty}_{-i\tanh^{(-1)}\left(\frac{1}{2}\right)}\log\left(\frac{1}{2}\sin(2t')\right)\csc(t')dt'. $$ But $$ \int t\csc tdt=-2t\tanh^{(-1)}(e^{it})-2iLi_2(e^{it})+\frac{1}{2}iLi_2(e^{2it}) $$ and $$ \int\log\left(\sin(t)\right)\csc(t)dt= $$ $$ =[\log(\sin t)+\log(1-i\tan(t/2))+\log(1+i\tan(t/2))- $$ $$ -1/2\log(\tan(t/2))]\log(\tan(t/2))+Li_2(-i\tan(t/2))+Li_2(i\tan(t/2)). $$ Hence using the above we get the evaluation. (Note that $\int \frac{1}{\sqrt{t^2+2t}}dt$ can evaluated easily).