Gradshteyn & Ryzhik, 7th ed., p. 570, formula 4.325(5) give the following definite integral: $$\begin{align*}{\large\int}_0^1\frac{\ln\ln\left(\frac1x\right)}{1+x+x^2}dx&=\frac\pi{\sqrt3}\ln\frac{\sqrt[3]{2\pi}\,\Gamma\left(\frac23\right)}{\Gamma\left(\frac13\right)}\\&=\frac\pi{\sqrt3}\left(\frac{4\ln2\pi}3-\frac{\ln3}2-2\ln\Gamma\left(\tfrac13\right)\right)\end{align*}$$ This and other similar integrals are discussed in several papers:

  • Vardi, Integrals, an introduction to analytic number theory. Am. Math. Mon. 95, 308–315 (1988)
  • Adamchik, A class of logarithmic integrals. Proceedings ISSAC, 1–8, 1997
  • Medina, Moll, A class of logarithmic integrals, Ramanujan J. 20 (2009), no. 1, 91–126
  • Blagouchine, Rediscovery of Malmsten's integrals, their evaluation by contour integration methods and some related results, Ramanujan J. 2014; 35: 21

Is it possible to find a closed form for a similar integral having the square of the logarithm in the numerator? $${\large\int}_0^1\frac{\ln^2\ln\left(\frac1x\right)}{1+x+x^2}dx$$

  • 1
    $\begingroup$ Maybe, but what for? $\endgroup$ – Start wearing purple Oct 18 '14 at 22:59
  • $\begingroup$ @O.L. Why would you want to know that? $\endgroup$ – X.C. Oct 18 '14 at 23:04
  • 1
    $\begingroup$ I was always curious about the motivation of people asking whether it is possible to find a closed form of integrals of type $\displaystyle\int_0^1\frac{\ln^2\ln\frac1x}{1+x+x^2}dx$. $\endgroup$ – Start wearing purple Oct 18 '14 at 23:07
  • $\begingroup$ @O.L. Well, different people may have different motivations, so I cannot say for everybody. I just hope the answer will be aesthetically pleasing and I could spend some time looking at it in delight. $\endgroup$ – X.C. Oct 18 '14 at 23:22
  • $\begingroup$ There is something similar over here. $\endgroup$ – Felix Marin Oct 29 '14 at 17:47

You may write $$\begin{align*} {\large\int}_0^1\frac{\ln^2\ln\left(\frac1x\right)}{1+x+x^2}dx&={\large\int}_0^\infty \left(1-e^{-t}\right)\frac{\ln^2t}{1-e^{-3t}}e^{-t}dt\\ &=\sum_{n=0}^\infty{\large\int}_0^\infty \left(e^{-t}-e^{-2t}\right)e^{-3nt}\ln^2t\:dt\\ &=\sum_{n=0}^\infty \left({\large\int}_0^\infty e^{-(3n+1)t}\ln^2t\:dt-{\large\int}_0^\infty e^{-(3n+2)t}\ln^2t\:dt\right)\\ &=\sum_{n=0}^\infty \left. \partial_{s}^2 \left({\large\int}_0^\infty t^se^{-(3n+1)t}\:dt-{\large\int}_0^\infty t^s e^{-(3n+2)t}\:dt\right)\right|_{s=0}\\ &=\left. \partial_{s}^2 \left(\Gamma(s+1) \left(\sum_{n=0}^\infty \frac{1}{(3n+1)^{s+1}}-\sum_{n=0}^\infty \frac{1}{(3n+2)^{s+1}}\right)\right)\right|_{s=0}\\ &=\left. \partial_{s}^2 \left(\frac{\Gamma(s+1)}{3^{s+1}} \left(\zeta\left(s+1,\frac13\right)-\zeta\left(s+1,\frac23\right)\right)\right)\right|_{s=0}\\ \end{align*}$$ Then, using the Laurent series expansion of the Hurwitz zeta function near $1$, $$ \zeta(s+1,a) = \frac1s+\sum_{k=0}^{\infty}\frac{(-1)^{k}}{k!} \gamma_{k}(a) s^{k}, \quad a>0, \, s\to 0, $$ with $\displaystyle \gamma_{0}(a)= - \psi(a)=-\Gamma'(a)/\Gamma(a)$, we get

$$ \color{#0052A3}{\frac{\pi^3\sqrt{3}}{54}+\frac{\sqrt{3}\pi}{9} (\gamma+\ln 3)^2+\frac{2}{3} (\gamma+\ln 3)\! \left(\gamma_1\!\!\left(\frac{1}{3}\right)-\gamma_1\!\!\left(\frac{2}{3}\right)\right)+\frac{1}{3}\!\!\left(\gamma_2\!\!\left(\frac{1}{3}\right)-\gamma_2\!\!\left(\frac{2}{3}\right)\right)}. $$

Observe that, as pointed out by Vladimir Reshetnikov, you can make the following substitution

$$ \color{#3366FF}{\gamma_1\left(\tfrac13\right)-\gamma_1\left(\tfrac23\right)=\frac\pi{\sqrt 3}\left(6\ln\Gamma\left(\tfrac13\right)-\gamma+\frac{\ln 3}2-4\ln(2\pi)\right)}. $$

  • $\begingroup$ This is great but could you add some explanations for the manipulations? $\endgroup$ – Ali Caglayan Oct 19 '14 at 0:36
  • $\begingroup$ @Alizter Thanks. Please, what specific manipulations ? $\endgroup$ – Olivier Oloa Oct 19 '14 at 0:37
  • 5
    $\begingroup$ It's known that $\gamma_1\left(\tfrac13\right)=\gamma_1-\left(\frac{3\gamma}2+\frac{3\ln 3}4\right)\cdot\ln 3+\frac\pi{\sqrt 3}\left(3\ln\Gamma\left(\tfrac13\right)-\frac\gamma2+\frac{\ln 3}4-2\ln(2\pi)\right)$ and $\gamma_1\left(\tfrac23\right)=\gamma_1-\left(\frac{3\gamma}2+\frac{3\ln 3}4\right)\cdot\ln 3-\frac\pi{\sqrt3}\left(3\ln\Gamma\left(\tfrac13\right)-\frac\gamma2+\frac{\ln 3}4-2\ln(2 \pi)\right)$. So, $\gamma_1\left(\tfrac13\right)-\gamma_1\left(\tfrac23\right)=\frac\pi{\sqrt 3}\left(6\ln\Gamma\left(\tfrac13\right)-\gamma+\frac{\ln 3}2-4\ln(2\pi)\right)$. $\endgroup$ – Vladimir Reshetnikov Oct 19 '14 at 4:15

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.