3

Consider $$ y = \int_1^{\infty} \frac{\operatorname{li}(x)^2 (x - 1)}{x^4} dx, $$ where $\operatorname{li}(x)$ is the logarithmic integral. Is there a closed form for y ?

It appears that a good approximation is $ 10 \cdot \operatorname{Ci}\bigl( \frac{56}{19}\bigr)$, where $\operatorname{Ci}(x)$ is the Cosine integral.

—-

If there is No closed form would it help to allow the function $$ t(x) = \int_1^x \operatorname{li}(t)^2 dt \hspace{10mm} ??$$


It appears the closed form for $y$ is $\frac{5 \pi^2}{36}$

There should be an easy proof of it.

edit

Just a comment but it seems

$$ T = \int_1^{\infty} \frac{\operatorname{li}(x)^3 (1 - x)}{x^9} dx = \frac{4 \pi^2}{795} $$

mick
  • 17,886

3 Answers3

5

Using numerical integration, I found a value of $$1.3707783890401886970603459722050209910157915843390$$ Looking for this number in $ISC$, I found that it is $$10\sum_{n=1}^\infty \frac{1}{4 n^2 \binom{2 n}{n}}$$ which is $\frac{5 }{36}\pi ^2$ !!

Now, $???$

  • 1
    What is ??? Suppose to mean ? – mick Jun 05 '18 at 06:25
  • 1
    @mick. $??? = $ How to prove it ? – Claude Leibovici Jun 05 '18 at 06:27
  • 1
    What is ??? Suppose to mean ? So can we express the indefinite integral in standard functions and Then plug in 1 and Oo and compute the limit ? Or do we use hypergeo functions and theorems ? Does ??? Mean 1) I give up bc Its hard. 2) I give up bc I found it and do not care to prove 3) I am very amazed ... or combinations ? ... Also Maybe similar integrals have been posted here on MSE , comparing might help. Kudo’s to the ISC. Besides this answer implies I lost 2 bets to my mentor. – mick Jun 05 '18 at 06:37
  • What do you think about my comments ? Leibovici ? – mick Jun 05 '18 at 06:39
  • I must say this Number looks very familiar ( the exact form , not the digits ) ... probably from a zeta kind of function or Sum harmonic / n^2 or So ... i Will come back to this. Ofc this suggests that this may be a way of attacking the hypothesis. – mick Jun 05 '18 at 06:43
  • @mick. You may be right, for sure ! The problem is that I have no idea about a way to attack the problem. I shall wait for a real answer to this interesting problem. Just out of curiosity : in which context did you find this problem ? – Claude Leibovici Jun 05 '18 at 06:58
  • Ah Yes , the Number is equal to 10 sum_1^oo harmonic(n)/n^3. – mick Jun 05 '18 at 07:04
  • The context was calculus :p . No seriously now it came from my mentor. We were talking about tetration and dynamics and calculus related to that. We considered “ new special functions “ like integral of li^2. ( I believe that has No closed form right ? Maybe hypergeo ?? ... anyway he came up with “ better ones “ and with better he mainly meant that integrals could be done in standard numbers. He gave the integral from the Op and we had a few bets. I lost the bet that it had no short closed form ... it Clearly does. Also he assumed it would contain pi. He has not tried to prove it. – mick Jun 05 '18 at 07:16
  • Also we were driven to these type of integrals by searching for ways too classify How fast functions grow ... I wonder If integral li^2 occurs in Number theory ? – mick Jun 05 '18 at 07:22
  • If anyone knows a link to a similar integral on MSE or MO ( with li^2 or li^3 ) , let me know. – mick Jun 05 '18 at 07:24
  • Differentiating under the integral sign 2 times might Also work ... – mick Jun 05 '18 at 07:30
  • An analogue of this answer might work ? https://math.stackexchange.com/questions/191008/a-curious-limit-for-frac-pi2 – mick Jun 05 '18 at 07:37
  • Maybe this helps , it looks similar : https://math.stackexchange.com/questions/513865/a-closed-form-for-int-01-frac-ln-ln-x-operatornameli2xxdx – mick Jun 05 '18 at 10:50
  • see also : https://math.stackexchange.com/questions/5079193/integral-s-int-1-infty-dfrac50-zeta3-operatornamelix6-x-1 – mick Jun 28 '25 at 11:29
4

Put $x = e^t$ ($t \ge 0$). Because $\operatorname{li}(e^t) = \operatorname{Ei}(t)$, the original integral becomes

$$ I \;=\; \int_{0}^{\infty} (e^{t}-1)\,e^{-3t}\, \operatorname{Ei}^2(t)\,\text{d}t = \int_{0}^{\infty} e^{-2t}\, \operatorname{Ei}^2(t)\,\text{d}t - \int_{0}^{\infty} e^{-3t}\, \operatorname{Ei}^2(t)\,\text{d}t. $$

Therefore, we must evaluate the integral $$ I(p) = \int_0^\infty e^{-pt} \operatorname{Ei}^2(t) \, dt, \quad \text{where } p \in \mathbb{N}. $$

The exponential integral satisfies $\operatorname{Ei}(t) \sim \frac{e^t}{t}$ as $t \to \infty$, so the integrand behaves like $$ e^{-pt} \cdot \left( \frac{e^t}{t} \right)^2 = \frac{e^{-(p-2)t}}{t^2}. $$ Therefore, the integral converges if and only if $p \geq 2$.

We integrate by parts:

  • Let $u = \operatorname{Ei}^2(t)$, $dv = e^{-pt}\,dt$,
  • Then $du = \frac{2\,\operatorname{Ei}(t)\,e^t}{t}\,dt$, and $v = -\frac{e^{-pt}}{p}$.

This gives: $$ I(p) = \frac{2}{p} \int_0^\infty \frac{e^{-(p-1)t}}{t} \operatorname{Ei}(t)\,dt. \tag{1} $$

For $t > 0$, we use the representation: $$ \operatorname{Ei}(t) = \gamma + \log(t) + \int_0^1 \frac{e^{tu} - 1}{u}\,du, $$ where $\gamma$ is the Euler–Mascheroni constant.

Substitute this into $(1)$ and split the integral: $$ I(p) = \frac{2}{p} \left( J_1 + J_2\right), $$ where:

  • $J_1 = \gamma \int_0^\infty \left(\frac{e^{-(p-1)t}}{t} + e^{-(p-1)t} \frac{\log(t)}{t}\right) \, dt = -\frac{\gamma}{p-1} \log(p-1) + \frac{1}{2(p-1)} \left[ \log^2(p-1) + \zeta(2) \right],$
  • $J_2 = \int_0^1 \frac{du}{u} \int_0^\infty \frac{e^{-(p-1-u)t} - e^{-(p-1)t}}{t} \, dt$.

Use the known result: $$ \int_0^\infty \frac{e^{-a t} - e^{-b t}}{t} \, dt = \log\left(\frac{b}{a}\right), \quad a, b > 0, $$ to compute the inner integral: $$ J_2 = \int_0^1 \frac{1}{u} \log\left( \frac{p - 1}{p - 1 - u} \right) du = \operatorname{Li}_2\left( \frac{1}{p - 1} \right). $$

Putting everything together: $$ I(p) = \frac{2}{p} \left( -\frac{\gamma}{p-1} \log(p-1) + \frac{1}{2(p-1)} \left[ \log^2(p-1) + \zeta(2) \right] + \operatorname{Li}_2\left( \frac{1}{p - 1} \right) \right). $$

After simplifying: $$ I(p) = \frac{1}{p} \left[ \zeta(2) + \log^2(p-1) + 2 \operatorname{Li}_2\left( \frac{1}{p-1} \right) \right], \quad p \ge 2. $$

Then, the integral of the question is

$$I = \frac{1}{2}[\zeta(2) + \log^2(1) + 2\text{Li}_2(1)] - \frac{1}{3}[\zeta(2) + \log^2(2) + 2\text{Li}_2(1/2)] = \frac{5}{6}\zeta(2),$$

where we used that

$$2\text{Li}_2(1/2) = \zeta(2) - \log^2(2).$$

user326159
  • 3,277
1

For what it's worth, for $n=0$ and $n=1$ the left-hand side (LHS) and RHS are equal in the below equation, to high numeric precision. I thought I saw a pattern developing for a generalization, but have been unable to ascertain the correct one yet. For $n=0$ and using the dilogarithm at argument 1/2, the already proposed identity results. There is absolutely no rigor in my nascent identity, so I won't present the development.

$$ \int_1^\infty \frac{x-1}{x^{n+4}}\, li(x)^2 \,dx \, \, \, ( \text{for } n=0, 1) \\ =(-1)^{n+1}\Big(\, \frac{\zeta(2)}{2} + \sum_{m=0}^n \frac{(-1)^m}{m+3}\big(\log^{\,2}(m+2)+2\text{Li}_2(1/(m+2))-5\zeta(2)\big)\, \Big)$$

As far as I know, $\text{Li}_2(1/3)$ does not have a 'closed form,' so generalizations in this direction appear unlikely.

user321120
  • 6,895