28
$\begingroup$

Let $\operatorname{li} x$ denote the logarithmic integral: $$\operatorname{li} x=\int_0^x\frac{dt}{\ln t}.$$ Is it possible to find a closed form of the following integral? $$\int_0^1\ln(-\ln x) \operatorname{li} x\ dx$$ I tried integration by parts, but only ended up with more complicated expressions.

$\endgroup$
  • 6
    $\begingroup$ Mathematica is unable to find a closed form. Plugging the numerical value of the integral into the Inverse Symbolic Calculator turns up nothing. Do you have a reason to suspect that a closed form exists? $\endgroup$ – Antonio Vargas May 28 '13 at 4:19
20
+500
$\begingroup$

This integral can be expressed in a closed form: $$\int_0^1\ln(-\ln x)\operatorname{li}x\ dx=\frac{\pi^2}{12}+\frac{(\ln2)^2}2+\gamma\,\ln2,\tag1$$ where $\gamma$ is the Euler-Mascheroni constant.


Proof:

Let $$\mathcal{I}(a)=\int_0^1(-\ln x)^a\operatorname{li}x\ dx,\tag2$$ then, taking a derivative with respect to $a$, we get $$\mathcal{I}'(0)=\int_0^1\ln(-\ln x)\operatorname{li}x\ dx.\tag3$$ Mathematica can express the integral $(2)$ in a closed form for explicitly given non-negative integer values of $a$, but not in a general form with a parameter. We could use those values to get a conjectured general formula, and indeed one could get the result $(1)$ this way. Numerical integration shows that $$\Bigg|\frac{\pi^2}{12}+\frac{(\ln2)^2}2+\gamma\,\ln2-\int_0^1\ln(-\ln x)\operatorname{li}x\ dx\Bigg|<10^{-1000}\tag4,$$ so the result looks plausible.

But here we will follow a different approach to get a rigorous proof (modulo possible bugs in Mathematica tuned exactly to hide that tiny difference). Take the formula $(12)$ from the MathWorld article: $$\operatorname{li}x=\operatorname{Ei}(\ln x),\tag5$$ where $\operatorname{Ei}z$ is the exponential integral, and plug it into $(2)$. In this form the integral can be evaluated in Mathematica, that gives us $$\mathcal{I}(a)=-\frac{_2F_1(a+1,a+1;a+2;-1)\ \Gamma(a+1)}{a+1},\tag6$$ and, taking a derivative with respect to $a$, $$\mathcal{I}'(0)=(1+\gamma)\ln2-\left[\frac{d}{da}{_2F_1}(a+1,a+1;a+2;-1)\right]_{a=0}.\tag7$$ To find the derivative of the hypergeometric function we need to replace it with its definition (formula $(8)$ in this MathWorld article): $$_2F_1(a+1,a+1;a+2;-1)=\sum_{n=0}^\infty\frac{(a+1)\ \Gamma(a+n+1)}{(a+n+1)\ \Gamma(a+1)\ n!}(-1)^n.\tag8$$ So, $$\left[\frac{d}{da}{_2F_1}(a+1,a+1;a+2;-1)\right]_{a=0}=\sum_{n=0}^\infty\frac{(n+1)(\gamma+\psi(n+1))+n}{(n+1)^2}(-1)^n,\tag9$$ where $\psi(n+1)=\frac{\Gamma'(n+1)}{\Gamma(n+1)}=-\gamma+\sum_{m=1}^n\frac1m\,$ is the digamma function. Evaluating this sum with Mathematica we get $$\left[\frac{d}{da}{_2F_1}(a+1,a+1;a+2;-1)\right]_{a=0}=\ln2-\frac{(\ln2)^2}2-\frac{\pi^2}{12}\tag{10}.$$ Plugging this back in $(7)$ and $(3)$ we get the final result $(1)$.

$\endgroup$
  • $\begingroup$ Excuse my dumbness, how do you get from (2) to (3)? $\endgroup$ – Gyro Gearloose Feb 14 '16 at 17:03
  • 1
    $\begingroup$ I used $\frac{d}{da}z^a=z^a\cdot\ln z$. In this case $z=-\ln x$, and after taking a derivative, I substituted $a=0$. $\endgroup$ – Vladimir Reshetnikov Feb 14 '16 at 18:16

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.