I am dealing with this sum of two divergent integrals

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

This is ugly but although they diverge seperately, their sum converges to $0.1746809...$

Can we find a closed form in terms of known constants for this number?

I've tried getting like denominators by setting $t\to e^{-\sqrt{|\ln(t)|}}$ in the second integral.

Which gave $$-\frac{\pi }{4}\int _0^{1}\frac{\frac{2\left(\ln t\right)e^{\left(\ln t\right)^2}}{\left(1+e^{\left(\ln t\right)^2}\right)^2}}{t\left(\ln t\right)^2}dt$$

However I realized that the expression has to be written in terms of the limit $m\to 1$, because although the upper bound in this integral is $1$, both integrals separately diverge at different rates as their upper bounds go to $1$. So we'd have to write the expression as

$$\lim_{m\to 1}\left(-\frac{1}{2}\int _0^{m}\frac{\left(\frac{1-t}{1+t}\right)\arctan t}{t\left(\ln t\right)^2}dt-\frac{\pi }{4}\int _0^{e^{-\sqrt{|\ln m|}}}\frac{\frac{2\left(\ln t\right)e^{\left(\ln t\right)^2}}{\left(1+e^{\left(\ln t\right)^2}\right)^2}}{t\left(\ln t\right)^2}dt\right)$$

Thus we can't combine these integrals and this doesn't work.

Due to Jack's comment I will rewrite the integral as $$\int _0^{1}\frac{dt}{\ln t}\left(\frac{\pi }{2\left(1+t\right)^2}+\frac{1-t}{1+t}\frac{\arctan t}{t\ln t}\right)$$ which is just twice the original value.

  • $\begingroup$ That has got to be one of the ugliest simple substitutions of the year. $\endgroup$ – Simply Beautiful Art Aug 9 '17 at 15:22
  • $\begingroup$ Oh really? Then I guess my separate integrals are not each divergent? $\endgroup$ – tyobrien Aug 9 '17 at 15:54
  • $\begingroup$ They are separately divergent because each integrand is has a $\pm1/(t-1)$ singularity at $1.$ $\endgroup$ – zhw. Aug 9 '17 at 16:15

We may consider that

$$ \int_{0}^{1}\left(\frac{1}{\log t}\color{red}{+\frac{1}{1-t}}\right)\frac{dt}{(1+t)^2}\stackrel{t\mapsto e^{-x}}{=}\int_{0}^{+\infty}\left(\frac{1}{e^x-1}-\frac{1}{xe^x}\right)\frac{dx}{(e^{-x}+1)^2} $$ where the red term has been introduced to counter-act the singularity at $t=1$.
An integral similar to an integral representation of the Euler-Mascheroni constant appears.

By the (inverse) Laplace transform, the last integral equals

$$ \int_{0}^{+\infty}\left(\log(s+1)-\psi(s+1)\right)\cdot\sum_{n\geq 1}(-1)^{n+1}n\,\delta(s-n+1)\,ds $$ or $$ \sum_{n\geq 1}n(-1)^{n+1}\left(\log(n)-H_{n-1}+\gamma\right)=\frac{1}{2}+\frac{\gamma}{4}+\frac{7\log 2}{12}-3\log A $$ where the Glaisher-Kinkelin constant $A$ is related to $\zeta'(-1)$ and $\zeta'(2)$.

It just remains to apply a similar approach to the other integral, i.e. to get rid of a pole and to compute the regularized integral in a explicit way, in terms of $\gamma,\pi,\log 2$ and values of $\zeta$ and $\zeta'$.

$$\small \int_{0}^{1}\left(\frac{1-t}{1+t}\cdot\frac{\arctan t}{t\log^2 t}\color{red}{-\frac{\pi}{8(1-t)}}\right)\,dt \stackrel{t\mapsto e^{-x}}{=} \int_{0}^{+\infty}\left(\frac{\pi}{8-8e^x}+\frac{\arctan(e^{-x})(e^x-1)}{x^2 (e^x+1)}\right)\,dx $$ equals $$ -\frac{\pi \gamma}{8}+\int_{0}^{+\infty}\left(\frac{\arctan(e^{-x})(e^x-1)}{x^2 (e^x+1)}-\frac{\pi}{8xe^x}\right)\,dx $$ which can be converted in a series by differentiation under the integral sign and Frullani's Theorem. We may clearly see the Gudermannian function being involved in this conversion step:

$$-\frac{\pi \gamma}{8}+\int_{0}^{+\infty}\left(\frac{\pi(e^x-1)}{4x^2(e^x+1)}-\frac{\pi}{8xe^x}\right)\,dx - \int_{0}^{1}\int_{0}^{+\infty}\frac{e^{-kx}(e^x-1)}{x(e^x+1)(e^{-2kx}+1)}\,dx\,dk$$ equals (by the Laplace transform): $$ -\frac{\pi\gamma}{8}+\frac{\pi}{6}\log\frac{A^9}{2}- \int_{0}^{1}\int_{0}^{+\infty}\frac{e^{-kx}(e^x-1)}{x (e^x+1)(e^{-2kx}+1)}\,dx\,dk$$ hence the whole problem boils down to computing

$$ \int_{0}^{+\infty}\frac{\tanh\left(\frac{x}{2}\right)\arctan \tanh\left(\frac{x}{2}\right)}{x^2}\,dx\qquad \text{or}\qquad \int_{0}^{+\infty}\tanh(x)\arctan\tanh(x)\frac{dx}{x^2}. $$ By the residue theorem we have $$ \int_{0}^{+\infty}\frac{\tanh^2 x}{x^2}\,dx = \frac{14\,\zeta(3)}{\pi^2},\qquad \int_{0}^{+\infty}\frac{\tanh^4 x}{x^2}\,dx=\frac{4 \left(14 \pi^2 \,\zeta(3)-93\,\zeta(5)\right)}{3 \pi^4}$$ and the previous integral can be written in terms of the values of the $\zeta$ function.
A numerical evaluation is actually way easier, since $\frac{\tanh(x)\arctan\tanh(x)}{x^2}\approx\frac{1}{x^2+1}$ on $\mathbb{R}^+$.

  • $\begingroup$ Great work! But how can the two integrals with the red terms be combined to equal my value? Their factors are wrong. $\endgroup$ – tyobrien Aug 10 '17 at 14:53
  • $\begingroup$ @user991003: due to the terms $\frac{1}{2}$ and $\frac{\pi}{4}$ in your original formula, the red terms I introduced to get rid of the singularities cancel out. $\endgroup$ – Jack D'Aurizio Aug 10 '17 at 14:56
  • $\begingroup$ But your second integral is missing a $1/t$ in it and they can't cancel if the red term in the first integral is multiplied by $\frac{1}{(1+t)^2}$ $\endgroup$ – tyobrien Aug 10 '17 at 14:58
  • $\begingroup$ @user991003: $$\color{blue}{\frac{\pi}{4}}\int_{0}^{1}\left(\frac{1}{\log t}\color{red}{+\frac{1}{1-t}}\right)\frac{dt}{(1+t)^2}+\color{blue}{\frac{1}{2}}\int_{0}^{1}\left(\frac{1-t}{1+t}\cdot\frac{\arctan t}{t\log^2 t}\color{red}{-\frac{\pi}{8(1-t)}}\right)\,dt$$ gives the opposite of the integral you are interested into, plus an elementary integral. Is it clear now? The above combination equals $$ \frac{\pi}{16}(1+\log 2)-\color{orange}{\text{what we care about}}.$$ $\endgroup$ – Jack D'Aurizio Aug 10 '17 at 15:04
  • $\begingroup$ Oh yeah, I see what you're doing now. Thanks. $\endgroup$ – tyobrien Aug 10 '17 at 15:20

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.