3

I was trying solve this integral$$I= \int_{0}^{\pi/2} \frac{x}{\sin(x)} \ln\left(\frac{1+\cos(x) - \sin(x)}{1+\cos(x) + \sin(x)} \right) dx $$ let :$t=\tan(x/2)$ $$I = 2\int^1_0 \frac{\arctan(t)}{t}\ln\left({\frac{1-t}{1+t}}\right)dt$$

I have a problem evaluating it.

I try :$y=\frac{1-t}{1+t}$

\begin{align*} I &= 4\int^1_0\frac{\left({\frac{\pi}{4}-\arctan(y)}\right)\ln(y)}{1-y^2} dy=\pi\int^1_0 \frac{\ln(y)}{1-y^2}dy -4\int^1_0 \frac{\arctan(y)\ln(y)}{1-y^2}dy \\ &= -\frac{\pi^3}{8}-4\int^1_0 \frac{\arctan(y)\ln(y)}{1-y^2}dy\end{align*}

this integral complicated $\int^1_0 \frac{\arctan(y)\ln(y)}{1-y^2}dy$

I try to :

$$I = -\int^1_{-1}\int^1_0 \frac{\arctan(t)}{(1-at)}dtda$$

I don't know how to continue evaluating the integration.

epsilon
  • 1
  • 4
  • Here's my try, $$I \underset{\frac{1 - x}{1 + x}\to x}= {2}{\pi} \int_{0}^{1} \frac{\ln(x)}{1 - x^2} , \mathrm{d}x - 2 \int_{0}^{1} \frac{\arctan(x) \ln(x)}{1 - x^2} , \mathrm{d}x=2\pi I_1-2I_2$$

    $I_1$ is an easy integral, the real question is how does one solve $I_2$,

    $$I_1=\int_{0}^{1} \frac{\ln(x)}{1 - x^2} , \mathrm{d}x=\sum_{n\ge 0}\int_0^1x^{2n}\ln(x),dx=-\sum_{n\ge 0}\frac{1}{\left(2n + 1\right)^{2}}\underset{\text{shift index}}=-\sum_{n\ge 1}\frac{1}{\left(2n\right)^{2}}\underset{\text{zeta function}}=-\frac{\pi^2}{8}$$

    – Amrut Ayan Jun 22 '25 at 13:03
  • For $I_2$, if we tried IBP, $$I_2=-\frac{\pi}{8}\operatorname{Li}2(2)-\frac12\left(\int_0^1 \frac{\operatorname{Li}{2}\left(1 - x\right)-\operatorname{Li}_{2}\left(1 + x\right)}{1-x^2},dx\right)$$

    One can try the below,

    $$\color{red}{\operatorname{Li}_2(1-x) = \zeta(2)-\ln(x)\ln(1-x)-\operatorname{Li}_2(x)}$$

    $$\color{red}{\operatorname{Li}_2(1 + x) = \zeta(2) - \ln x \ln(1 + x) - i\pi \ln(1 + x) - \operatorname{Li}_2(-x)}$$

    But they lead to integrals which diverge

    – Amrut Ayan Jun 22 '25 at 13:05
  • All of these integrals are known and have appeared on this site multiple times. – Saul Jun 22 '25 at 13:42
  • Another way could be,

    $${\ln\left(\frac{1-x}{1+x}\right)=-2\sum_{n=0}^\infty\frac{x^{2n+1}}{2n+1}}$$

    $$I=-2\sum_{n=0}^\infty\frac{1}{2n+1}\int^1_0 x^n \arctan(x)\mathrm{d}x$$

    $$\int_{0}^{1} x^{n} \arctan x , dx = \frac{1}{(2n+2)} \left( \frac{\pi}{2} - \beta \left( \frac{n}{2} + 1 \right) \right)$$

    $$I=-\pi\sum_{n=0}^\infty\frac{1}{(2n+1)(2n+2)}+2\sum_{n=0}^\infty\frac{1}{(2n+1)}\beta \left( \frac{n}{2} + 1 \right)$$

    – Amrut Ayan Jun 22 '25 at 13:45
  • This elegant answer is similar - https://math.stackexchange.com/a/4407507/1157207 – Amrut Ayan Jun 22 '25 at 17:19

4 Answers4

4

Substituting $\frac{1-x}{1+x} \to x$ transforms the integral into

$$I = \int_{0}^{1}\frac{\arctan\left(x\right)}{x}\ln\left(\frac{1-x}{1+x}\right)dx \stackrel{\frac{1-x}{1+x}\to x}{=} 2\int_{0}^{1}\frac{\arctan\left(\frac{1-x}{1+x}\right)}{1-x^{2}}\ln\left(x\right)dx$$ Note that $\arctan\left(\frac{1-x}{1+x}\right) = \frac{\pi}{4} - \arctan x$ by the addition formula for arctan. So, $$ I = \frac{\pi}{2}\underbrace{\int_{0}^{1}\frac{\ln\left(x\right)}{1-x^{2}}}_{I_1}dx-2\underbrace{\int_{0}^{1}\frac{\arctan\left(x\right)\ln\left(x\right)}{1-x^{2}}dx}_{I_2} \tag1$$ $I_1$ can be evaluated by expanding $\frac1{1-x^2}$ as a geometric series: $$I_1 = \sum_{n=0}^\infty\int\limits_0^1x^{2n}\ln (x)dx = -\sum_{n=0}^\infty\frac{1}{(2n+1)^2} = -\frac{3\zeta(2)}{4} = -\frac{\pi^2}{8}$$ For $I_2$, note that $$\arctan(x) = \int\limits_0^x\frac{1}{1+t^2}dt \stackrel{t \to tx}{=}\int\limits_0^1\frac{x}{1+t^2x^2}dt$$ Substituting this in to $I_2$ gives us $$I_2 = \int_{0}^{1}\int_{0}^{1}\frac{x\ln\left(x\right)}{\left(1-x^{2}\right)\left(1+t^{2}x^{2}\right)}dtdx$$ Doing a partial fraction decomposition and exchanging the order of integration gives \begin{align} I_2 &= \int_{0}^{1}\frac{1}{t^{2}+1}\int_{0}^{1}\left(\frac{t^{2}x\ln\left(x\right)}{t^{2}x^{2}+1}-\frac{\ln\left(x\right)}{2\left(x-1\right)}-\frac{\ln x}{2\left(x+1\right)}\right)dxdt \\&=\int_{0}^{1}\frac{t^{2}}{t^{2}+1}\int_{0}^{1}\frac{x\ln\left(x\right)}{t^{2}x^{2}+1}dxdt-\frac{1}{2}\int_{0}^{1}\frac{1}{t^{2}+1}\int_{0}^{1}\frac{\ln\left(x\right)}{x-1}dxdt\\&-\frac{1}{2}\int_{0}^{1}\frac{1}{t^{2}+1}\int_{0}^{1}\frac{\ln\left(x\right)}{x+1}dxdt \end{align} The last two integrals can simply be integrated with geometric series, giving $$\int_{0}^{1}\frac{\ln\left(x\right)}{x-1}dx = \zeta(2)$$ $$\int_{0}^{1}\frac{\ln\left(x\right)}{x+1}dx = -\frac{\zeta(2)}2$$ And of course, $$\int_{0}^{1}\frac{1}{t^{2}+1} = \frac \pi4$$ Substituting this in gives us $$I_2 = -\frac{\pi}{16}\zeta\left(2\right)+\int_{0}^{1}\frac{t^{2}}{t^{2}+1}\underbrace{\int_{0}^{1}\frac{x\ln\left(x\right)}{\left(t^{2}x^{2}+1\right)}dx}_{\mathcal{J}}dt \tag2$$

Focusing on the inner integral, $$\mathcal{J}=\int_{0}^{1}\frac{x\ln\left(x\right)}{\left(t^{2}x^{2}+1\right)}dx$$ Due to the bounds on $x$ and $t$, we can expand $\cal J$ as a geometric series $$\mathcal J = \sum_{n=0}^{\infty}\left(-1\right)^{n}t^{2n}\int_{0}^{1}x^{2n+1}\ln\left(x\right)dx = -\sum_{n=0}^{\infty}\frac{\left(-1\right)^{n}t^{2n}}{\left(2n+2\right)^{2}}=\frac{1}{4t^{2}}\sum_{n=0}^{\infty}\frac{\left(-t^{2}\right)^{\left(n+1\right)}}{\left(n+1\right)^{2}}$$ The sum is a representation of the dilogarithm, giving us $$\mathcal{J} = \frac{\operatorname{Li}_2\left(-t^2\right)}{4t^2}$$ Substituting into $(2)$ gives $$I_2=-\frac{\pi}{16}\zeta\left(2\right)+\frac{1}{4}\int_{0}^{1}\frac{\operatorname{Li}_{2}\left(-t^{2}\right)}{t^{2}+1}dt$$ The dilogarithm integral can be evaluated here, giving us (after a bit of simplifying) $$I_2 = \frac{\pi^{3}}{64}+\frac{\pi}{16}\ln^{2}\left(2\right)+\frac{1}{2}\ln\left(2\right)G + \Im\operatorname{Li}_3(1-i)$$ where $G$ is Catalan's constant and $\operatorname{Li}_3(x)$ is the trilogarithm. Finally, substituing into $(1)$ gives us

$$I = -\frac{3}{32}\pi^{3}-\frac{\pi}{8}\ln^{2}\left(2\right)-\ln\left(2\right)G-2\Im\operatorname{Li}_3(1-i)$$

Alex
  • 422
2

The antiderivative $$I=\int \frac{\tan ^{-1}(x)}{x}\,\log \left(\frac{1-x}{1+x}\right)\,dx$$ is not too bad if you use the logarithmic representation of the arc tangent $$I=\frac i 2 \int \frac 1 x\, \log \left(\frac{i+x}{i-x}\right)\,\log \left(\frac{1-x}{1+x}\right)\,dx$$

Expand the logarithms and remember that $$\int \frac 1 x\,\log (x+a) \,\log (x+b)\,dx$$ has a closed form in terms of logarithms and polylogarithms $\text{Li}_2(.)$ and $\text{Li}_3(.)$

So, we have the antiderivative.

Considering now $$J=\int_\epsilon^{1-\epsilon} \frac{\tan ^{-1}(x)}{x}\,\log \left(\frac{1-x}{1+x}\right)\,dx$$ and using series expansion, we have $$J=-\Bigg(C \log (2)+\frac{3 \pi ^3}{32}+\frac{1}{8} \pi \log ^2(2) +i\Big(\text{Li}_3(1+i)-\text{Li}_3(1-i) \Big) \Bigg)+$$ $$\frac{\pi }{4} \log \left(\frac{2 e}{\epsilon }\right)\,\epsilon +\frac 18 \Bigg(7+(\pi-2)\log \left(\frac{2}{\epsilon }\right) \Bigg) \,\epsilon^2+O\left(\epsilon ^3\right)$$ which is a quite good approximation.

For $\epsilon=\frac 14$, this truncated series gives $-0.518364$ instead of $-0.519366$ corresponding to a relative error of $0.2$%.

2

The integrand of this integral can be easily expanded into a Taylor series:

$$\frac{\arctan(x)}{x}\ln\left({\frac{1-x}{1+x}}\right)=-2\sum_{k=0}^{\infty}\frac{H_{k}}{2k+1}x^{4k+1}$$

where

$$H_{k}=\sum_{m=0}^{2k}\frac{(-1)^{m}}{2m+1}$$

The result of the integration:

$$I=-\sum_{k=0}^{\infty}\frac{H_{k}}{(2k+1)^{2}}$$

Adding the first 10 terms of this series, the absolute error is about $0.018$

A physicist's approach:

Here i use an approximation:

$$\frac{\arctan(x)}{x}\approx \frac{\pi}{392}(129-24x-7x^{2})$$

In this case, the integral is easily computed analytically:

$$I\approx -\pi\frac{760\ln 2 -79}{1176}$$

In this case the absolute error is $3\cdot 10^{-6}$

Martin Gales
  • 7,927
  • I like very much the physicist's approach but there is something that I do not understand. Minimizing the norm $$\Phi=\int_0^1 \left( \left(a+b x+c x^2\right)-\frac{\tan ^{-1}(x)}{x}\right)^2,dx$$, the best coefficients are $$a=-\frac{3}{2} (-6 C+10+\pi -6 \log (4))$$ $$b=3 (-12 C+30+\pi -16 \log (4))$$ $$c=15 (2 C-6+3 \log (4))$$ which lead to $\Phi=8.95\times 10^{-6}$ while your coefficients lead to $\Phi=8.37\times 10^{-5}$ and your overall result is better than mine. I obtain $-1.195604236$ while you have $1.196241164$ which is much better. Any idea ? Cheers :-) – Claude Leibovici Jun 23 '25 at 14:43
  • @ClaudeLeibovici I used the calculus of variations to get this approximation. – Martin Gales Jun 24 '25 at 11:19
2

The following solution works for only even powers of $\ln$:

$$\int^1_0 \frac{\arctan(t)}{t}\ln^{2a}\left({\frac{1-t}{1+t}}\right)dt=\frac{\pi}{2}\int_0^1\frac{\ln^{2a}(t)}{1-t^2}dt-2\int_0^1\frac{\arctan(t)\ln^{2a}(t)}{1-t^2}dt$$

$$=2(2a)!\sum_{k=0}^{a}2^{2k-2a}\eta(2a-2k)\beta(2k+2).$$

where the first integral is the integral form of the Dirichlet lambda function and the second integral is given in page 14 of this preprint.

Examples:

$$\int^1_0 \frac{\arctan(t)}{t}\ln^{2}\left({\frac{1-t}{1+t}}\right)dt=\frac{\pi^2}{12}\beta(2)+2\beta(4)$$

$$\int^1_0 \frac{\arctan(t)}{t}\ln^{4}\left({\frac{1-t}{1+t}}\right)dt=\frac{7\pi^4}{240}\beta(2)+\pi^2\beta(4)+24 \beta(6)$$

$$\int^1_0 \frac{\arctan(t)}{t}\ln^{6}\left({\frac{1-t}{1+t}}\right)dt=\frac{31\pi^6}{1344}\beta(2)+\frac{7\pi^2}{8}\beta(4)+30 \pi^2\beta(6)+720\beta(8)$$

Ali Olaikhan
  • 27,891