18

Inspired by this question and this answer, I decided to investigate the family of integrals $$I(k)=\int_0^e\mathrm{Li}_k(\ln{x})\,dx,\tag{1}$$ where $\mathrm{Li}_k(z)$ represents the polylogarithm of order $k$ and argument $z$. $I(1)$ evaluates to $e\gamma$, but $I(2)$ has resisted my efforts (which can be seen here).

Neither ISC nor WolframAlpha could provide a closed form for its numerical value--however, I've conjectured a possible analytic form. $$\eqalign{\int_0^e\mathrm{Li}_2(\ln{x})\,dx&\stackrel?=\,_3F_3(1,1,1;2,2,2;1)+\frac{\pi^2(2e-5)}{12}+\frac{\gamma^2}{2}-\gamma\,\mathrm{Ei}(1)\\&=0.578255559804073275225659054377625577...\tag{2}}$$

Brevan Ellefsen has computed that my conjecture is accurate to at least 150 digits. Brevan also gave the alternate form $$\frac{\pi^2e}{6}+\gamma G_{1,2}^{2,0}\left(-1\left|\begin{array}{c}1\\0,0\\\end{array}\right.\right)+G_{2,3}^{3,0}\left(-1\left|\begin{array}{c}1,1\\0,0,0\\\end{array}\right.\right).\tag{2.1}$$

Is there a closed form for $I(2)$ that doesn't involve Meijer G or hypergeometrics? The simplicity of the following two equations seems to suggest that there might be. $(3.1)$ follows directly from $(3)$, which I've proven here. $$\begin{align} \sum_{k=1}^\infty I(k)&=e\tag{3}\\\sum_{k=2}^\infty I(k)&=e(1-\gamma)\tag{3.1}\end{align}$$


PROGRESS UPDATE: Using this equation, I've turned $_3F_3(1,1,1;2,2,2;1)$ into $$\lim_{c\to 1}\left(\frac{\mathrm{Ei}(1)-\gamma}{c-1}+\frac{1-e}{(c-1)^2}+\frac{(-1)^{-c}\,\Gamma(c-1)}{c-1}+\frac{(-1)^{1-c}\,\Gamma(c,-1)}{(c-1)^2}\right),\tag{4}\label{4}$$ but I don't know how to proceed from there. EDIT: This limit leads nowhere. See below.

SECOND PROGRESS UPDATE: After some studying of the properties of the Meijer G function, I've finally cracked the limit; however, the result is an underwhelming $_3F_3(1,1,1;2,2,2;1)$. Before I evaluate the limit, I'd first like to state the following intermediate result:

Lemma $(4.1)$: For $z\in\mathbb{C}$, $$G_{2,3}^{3,0}\left(z\left|\begin{array}{c}1,1\\0,0,0\\\end{array}\right.\right)=\gamma\ln{z}+\frac12\ln^2(z)-z\,_3F_3(1,1,1;2,2,2;-z)+\frac{\gamma^2}2+\frac{\pi^2}{12}.\tag{4.1}\label{4.1}$$

My proof for this can be found here. Now I return to the limit $\eqref{4}$. Consider the following: $\frac{1}{c-1}=\frac{c-1}{(c-1)^2}$, $(c-1)\Gamma(c-1)=\Gamma(c)$, and $(-1)^{1-c}=-(-1)^{-c}$. Based on these algebraic identities, the limit can be written as $$\lim_{c\to 1}\frac{(c-1)(\mathrm{Ei}(1)-\gamma)+1-e+(-1)^{-c}(\Gamma(c)-\Gamma(c,-1))}{(c-1)^2}.\tag{4.2}$$ In this form, the limit is $\frac{0}0$. Using l'Hospital twice, we obtain $$\lim_{c\to 1}{(-1)^{-c}\left(-G_{3,4}^{4,0}\left(-1\left|\begin{array}{c}1,1,1\\0,0,0,c\\\end{array}\right.\right)+\Gamma{(c)}\left(\frac{\psi_0{(c)}^2}2-i\pi\psi_0(c)+\frac{\psi_1(c)}2-\frac{\pi^2}2\right)\right)}$$ $$\begin{align}&=G_{3,4}^{4,0}\left(-1\left|\begin{array}{c}1,1,1\\0,0,0,1\\\end{array}\right.\right)-\frac{\psi_0(1)^2}2+i\pi\psi_0(1)-\frac{\psi_1(1)}2+\frac{\pi^2}2\\&=G_{2,3}^{3,0}\left(-1\left|\begin{array}{c}1,1\\0,0,0\\\end{array} \right.\right)-\frac{\gamma^2}{2}-i\gamma\pi+\frac{5\pi^2}{12}.\tag{4.2a}\end{align}$$ Using Lemma $\eqref{4.1}$, we know that $$G_{2,3}^{3,0}\left(-1\left|\begin{array}{c}1,1\\0,0,0\\\end{array}\right.\right)={}_3F_3(1,1,1;2,2,2;1)+\frac{\gamma^2}2+i\gamma\pi-\frac{5\pi^2}{12},\tag{4.3}$$ which can be rewritten as $$G_{2,3}^{3,0}\left(-1\left|\begin{array}{c}1,1\\0,0,0\\\end{array}\right.\right)-\frac{\gamma^2}{2}-i\gamma\pi+\frac{5\pi^2}{12}={}_3F_3(1,1,1;2,2,2;1).\tag{4.3a}$$

Thus, $$\eqalign{&\lim_{c\to 1}\left(\frac{\mathrm{Ei}(1)-\gamma}{c-1}+\frac{1-e}{(c-1)^2}+\frac{(-1)^{-c}\,\Gamma(c-1)}{c-1}+\frac{(-1)^{1-c}\,\Gamma(c,-1)}{(c-1)^2}\right)\\=&{}_3F_3(1,1,1;2,2,2;1).\tag{4.4}}$$

  • 1
    It's interesting that you consider the Meijer G-function outside of your scope, and yet include the generalized hypergeometric. I'm curious what contexts you learned above the latter but not the former. I see them as steps on an infinite ladder of more general functions defined by series, and have always seen them covered together that way. – Brevan Ellefsen Nov 02 '16 at 19:40
  • @BrevanEllefsen I'm asking for a better closed form, as the $3F_3$ term was originally a series. Specifically: $$\sum{k=1}^\infty\frac{1}{k^2 k!}.$$ By no means am I saying that $_3F_3$ is within my scope; it's far from it, in fact. I can't find anything online on either $_3F_3$ or Meijer G. I guess I just shared the best form I could get :P – teadawg1337 Nov 02 '16 at 19:48
  • Note that this can be more simply written as $$= \frac{e \pi^2}{6}+\int_{-\infty}^1\frac{e^x\log(1-x)}{x}dx$$ with a basic substitution and IBP. This integral is much more easy to analyze in my opinion. – Brevan Ellefsen Nov 02 '16 at 22:14
  • 1
    Moreover, Mathematica instantly converts the integral to $$\gamma G_{1,2}^{2,0}\left(-1\left| \begin{array}{c} 1 \ 0,0 \ \end{array} \right.\right)+G_{2,3}^{3,0}\left(-1\left| \begin{array}{c} 1,1 \ 0,0,0 \ \end{array} \right.\right)$$ – Brevan Ellefsen Nov 02 '16 at 22:19
  • @BrevanEllefsen I've edited the question to include your form and added some clarification. Thank you :) – teadawg1337 Nov 03 '16 at 15:12
  • the hypergeometric parameters look so simple, i am really puzzeled that it doesn't seem to simplify – tired Nov 03 '16 at 15:53
  • @tired I found this on WolframResearch, which gives a representation for $_3F_3(1,1,c;2,2,c+1;z)$. The discontinuity at $c=1$ isn't very encouraging... – teadawg1337 Nov 03 '16 at 16:23
  • i don't think this is real discontinuity (it might well cancel out if one takes limits) because mathematica gives a finite value for this point – tired Nov 03 '16 at 16:30
  • @tired Any luck so far with manipulating the limit? I've got nothing – teadawg1337 Nov 05 '16 at 19:18
  • i'm very busy at the moment, maybe i can spare some time on tuesday for this interesting problem – tired Nov 06 '16 at 23:00
  • If it makes you feel a bit better, between a lot of kicking and screaming on Mathematica's part I got that your form in terms of the Hypergeometric is equivalent to the integral to 150 digits of accuracy. This isn't surprising though - the Meijer G result is pretty trivial to get with Mathematica (and likely also trivial by hand if you work out what the integral form would be and work backwards) and Wolfram Alpha confirms the two are equivalent. Not a proof by any means, but I trust WA on this one. The real question of the day is whether or not there exists a closed form, which I doubt a lot. – Brevan Ellefsen Nov 20 '16 at 08:48
  • Also, I'm not convinced that limit actually converges. Are you sure the equation you provided is valid there? Mathematica is completely failing here, returning the limit result as $(\gamma -\gamma)\infty$. Fantastic job Mathematica -_- – Brevan Ellefsen Nov 20 '16 at 10:05
  • @BrevanEllefsen Are you sure you put it in correctly? The last term has the upper incomplete gamma function in its numerator, and WolframAlpha shows that it converges. I changed the variable to $d$ because it was interpreting $c$ as the speed of light – teadawg1337 Nov 20 '16 at 13:53
  • @teadawg1337 yes, I'm absolutely positive I put it right. I'm not saying there's no chance it converges (it likely does) but simply evaluating nearby points is not enough. Moreover, I was evaluating with Mathematica and not WA and I even used $x$ for each variable. Evidently after being up for hours working on the problem my humor did not come across XD I was simply laughing at the fact that Mathematica returned the limit as $0\cdot \infty$. Wolfram Alpha wouldn't even touch the full limit. – Brevan Ellefsen Nov 20 '16 at 15:05
  • The problem is that the incomplete Gamma function is finicky to work with here. Using Laurent series I was able to simplify the first Gamma and cancel some stuff out (it actually had a couple constant terms come out of that cancelation that simplify the result as a whole, such as $\frac{-\gamma^2}{2}$ (iirc) which cancels with the term outside of the limit. This is a promising result). The problem here is that the Incomplete Gamma doesn't play nice in that region, and moreover we are working with complex numbers here (my guess is we just want the real part. Now to justify that....) – Brevan Ellefsen Nov 20 '16 at 15:09
  • @BrevanEllefsen While it alone may not be enough to conclude that it converges, the fact that nearby points approach the numerical value of $_3F_3(1,1,1;2,2,2;1)$ strongly suggests that the limit does in fact converge. Also, WolframAlpha freaks out while computing the actual limit, yet the plot of the limit appears to have a removable discontinuity at $c=1$. – teadawg1337 Nov 20 '16 at 15:20
  • Indeed, and again, I'm fairly certain it does converge (at least removably if not more) I just wish I had a proof of that. The real question of the day is how that Incomplete Gamma function works around $1$.... By a little transformation we can turn the rightmost term into $\Gamma(c-1,-1)+\frac{d}{dx}x^c$, and this allows us to study it a bit more... However, I can't see what it does locally without a series of some sorts. The fact we are working with complex numbers is also not pleasing here at all. Given the right subsequence to approach the limit from complex numbers aren't necessary... – Brevan Ellefsen Nov 20 '16 at 15:25
  • ... But I can't get any CAS to generate a series without them just taking the real limit, even when explicitly asking for such a series. – Brevan Ellefsen Nov 20 '16 at 15:26
  • I'll later share some of my work from my Mathematica notebooks to explain some of my ramblings, since you seem a bit confused. Luckily Mathematica has the TeXForm[] command, which makes copying formulas to math.se wonderfully easy :) – Brevan Ellefsen Nov 20 '16 at 15:28
  • @BrevanEllefsen Correct me if I'm wrong, but using $\frac{\Gamma(c-1)}{c-1}=\frac{\Gamma(c)}{(c-1)^2}$, the last two terms can be put in terms of the lower Incomplete Gamma function. $$\frac{(-1)^{-c}\gamma(c,-1)}{(c-1)^2}$$ I'm not sure if this helps us at all... – teadawg1337 Nov 20 '16 at 15:34
  • @teadawg1337 looks correct to me (though I admit I haven't checked it due to lack of time right now). The question now is just what to do with it. Does it have a nicer converging series that the Upper Incomplete Gamma Function in the region? If so that would be phenomenal. If not though, I'm not sure what to do with this. – Brevan Ellefsen Nov 20 '16 at 15:39
  • I really don't think that $,_3F_3(1,1,1;2,2,2;1)$ simplifies. If any of middle terms were $1$ then we would get something a lot simpler, but that's not the case here. I'm fairly certain it's not elementary by this point – Brevan Ellefsen Nov 21 '16 at 23:54
  • @BrevanEllefsen Conversion to the lower Incomplete Gamma Function gives an indeterminate form; $\gamma(1,-1)=1-e$, which would just leave the first term. For some reason, the asymptotic behavior of the last two terms invalidates this algebraic manipulation. This problem is uglier than I thought... – teadawg1337 Nov 22 '16 at 14:59
  • @teadawg1337 are you sure you had the parity right? It looked to me like you had the wrong sign (the power of $(-1)$ looked a little off based on my calculations). Not sure why else it would fail... I'll look into it later if I get a chance. – Brevan Ellefsen Nov 22 '16 at 18:07
  • @BrevanEllefsen Yes, I'm positive that I had the parity right; $(-1)^{1-c}=-(-1)^{-c}$ and $\Gamma(s)-\Gamma(s,x)=\gamma(s,x)$, which gives us $\frac{(-1)^{-c}\gamma(c,-1)}{(c-1)^2}$. – teadawg1337 Nov 22 '16 at 19:46
  • @teadawg1337 alrighty, yeah... My bad. – Brevan Ellefsen Nov 22 '16 at 19:53
  • i also fail with this limit :-/ – tired Nov 23 '16 at 12:08

3 Answers3

5

Lemma 1: For $\lambda \in \mathbb{R}^{+}$ $$\sum\limits_{n=1}^{\infty} \dfrac{\lambda^n}{n!}\mathcal{H}_{n} = e^{\lambda}\left(\ln \lambda + \gamma - \operatorname{Ei}(-\lambda)\right) \tag{1} \label{lemma1}$$ where $\mathcal{H}_{n}$$n$-th harmonic number and $\operatorname{Ei}(\cdot)$ — exponential integral.

Start from two series representations of lower incomplete gamma function $\gamma(\beta, \lambda)$: \begin{align} \gamma(\beta, \lambda) &= e^{-\lambda}\sum\limits_{n=0}^{\infty} \dfrac{\lambda^{n+\beta}}{\beta\left(\beta+1\right)\ldots\left(\beta+n\right)} \\ \gamma(\beta, \lambda) &= \sum\limits_{n=0}^{\infty} (-1)^{n}\dfrac{\lambda^{n+\beta}}{n!\left(\beta+n\right)} \end{align} Now take the derivative with respect to $\beta$ at $1$. Since \begin{align*} \dfrac{\mathrm{d}}{\mathrm{d}\beta}\left(\dfrac{1}{\beta\left(\beta+1\right)\ldots\left(\beta+n\right)}\right)_{\beta=1} &= -\left.\dfrac{1}{\beta\left(\beta+1\right)\ldots\left(\beta+n\right)}\left(\dfrac{1}{\beta}+\dfrac{1}{\beta+1}+\ldots+\dfrac{1}{\beta+n}\right)\right\vert_{\ \beta=1} \\ &= -\dfrac{\mathcal{H}_{n+1}}{(n+1)!} \end{align*} we have that \begin{align*} -e^{-\lambda}\sum\limits_{n=1}^{\infty} \dfrac{\lambda^{n}}{n!}\mathcal{H}_{n}+\ln \lambda\ e^{-\lambda}\sum\limits_{n=1}^{\infty} \dfrac{\lambda^{n}}{n!} &= -\ln \lambda\sum\limits_{n=1}^{\infty} (-1)^{n}\dfrac{\lambda^{n}}{n!} + \sum\limits_{n=1}^{\infty} (-1)^{n}\dfrac{\lambda^{n}}{n!\ n} \\ -e^{-\lambda}\sum\limits_{n=1}^{\infty} \dfrac{\lambda^{n}}{n!}\mathcal{H}_{n}+\ln \lambda \left(1-e^{-\lambda}\right) &= \ln \lambda \left(1-e^{-\lambda}\right) + \sum\limits_{n=1}^{\infty} (-1)^{n}\dfrac{\lambda^{n}}{n!\ n} \\ \sum\limits_{n=1}^{\infty} \dfrac{\lambda^{n}}{n!}\mathcal{H}_{n} &= -e^{\lambda}\sum\limits_{n=1}^{\infty} (-1)^{n}\dfrac{\lambda^{n}}{n!\ n} \\ &= e^{\lambda}\left(\ln \lambda + \gamma - \operatorname{Ei}(-\lambda)\right) \end{align*}


Let $$ \mathfrak{I}(\lambda, \alpha, \beta) = \int\limits_{0}^{\infty} \dfrac{e^{-\lambda x}\,x^{\alpha - 1}}{\left(1+x\right)^{\alpha+\beta}}\,\mathrm{d}x $$

Lemma 2: Let $\alpha, \lambda \in \mathbb{R}^{+}$ and $\alpha + \beta > 0$. Then $$ \mathfrak{I}(\lambda, \alpha, \beta) = \dfrac{\Gamma(\alpha)}{\Gamma(\alpha+\beta)}\lambda^{\beta}\mathfrak{I}(\lambda, \alpha + \beta, -\beta) \tag{2} \label{lemma2} $$

Using Laplace transform properties we have \begin{align*} \mathfrak{I}(\lambda, \alpha, \beta) &= \int\limits_{0}^{\infty} \dfrac{e^{-\lambda x}x^{\alpha - 1}}{\left(1+x\right)^{\alpha+\beta}}\,\mathrm{d}x \\ &= \int\limits_{0}^{\infty} e^{-\lambda t}t^{\alpha-1}\,\mathcal{L}\left\{\dfrac{e^{-x}x^{\alpha+\beta-1}}{\Gamma(\alpha+\beta)}\right\}(t)\,\mathrm{d}t \\ &= \int\limits_{0}^{\infty} \dfrac{\Gamma(\alpha)}{\left(\lambda+x\right)^{\alpha}}\dfrac{e^{-x}x^{\alpha+\beta-1}}{\Gamma(\alpha+\beta)} \,\mathrm{d}x \\ &= \dfrac{\Gamma(\alpha)}{\Gamma(\alpha+\beta)}\lambda^{\beta}\int\limits_{0}^{\infty} \dfrac{e^{-\lambda x}x^{\alpha+\beta-1}}{\left(1+x\right)^{\alpha}}\,\mathrm{d}x \\ &= \dfrac{\Gamma(\alpha)}{\Gamma(\alpha+\beta)}\lambda^{\beta}\mathfrak{I}(\lambda, \alpha + \beta, -\beta) \end{align*}


Claim 1: For $\lambda > 0$ and $\beta \in \mathbb{R}$ $$ \mathfrak{I}(\lambda, 1, \beta) = e^{\lambda}\lambda^{\beta}\Gamma(-\beta, \lambda) \tag{3} \label{claim1} $$ where $\Gamma(\cdot, \cdot)$ — upper incomplete gamma function.

\begin{align*} \mathfrak{I}(\lambda, 1, \beta) &= \int\limits_{0}^{\infty} \dfrac{e^{-\lambda x}}{\left(1+x\right)^{1+\beta}}\,\mathrm{d}x \\ &= e^{\lambda}\int\limits_{1}^{\infty} \dfrac{e^{-\lambda x}}{x^{1+\beta}}\,\mathrm{d}x \\ &= e^{\lambda}\lambda^{\beta}\int\limits_{\lambda}^{\infty} \dfrac{e^{-x}}{x^{1+\beta}}\,\mathrm{d}x \\ &= e^{\lambda}\lambda^{\beta}\Gamma(-\beta, \lambda) \end{align*}


After applying \eqref{lemma2} to \eqref{claim1} we have that $$ \int\limits_{0}^{\infty} \dfrac{e^{-\lambda x}x^{\beta}}{1+x}\,\mathrm{d}x = e^{\lambda}\Gamma(1+\beta)\Gamma(-\beta, \lambda) \tag{4} \label{consec1} $$ Taking the derivative with respect to the variable $\beta$ at point $(\lambda, \beta) = (1, 0)$ leads us to \begin{align*} \int\limits_{0}^{\infty} \dfrac{e^{-x}\ln x}{1+x}\,\mathrm{d}x &= e\dfrac{\partial}{\partial \beta}\Big(\Gamma(1+\beta)\Gamma(-\beta, 1)\Big)\Bigg\vert_{\beta=0} \\ &= -e\gamma \int\limits_{1}^{\infty} \dfrac{e^{-x}}{x}\,\mathrm{d}x -e \int\limits_{1}^{\infty} \dfrac{e^{-x}\ln x}{x}\,\mathrm{d}x \\ &= e\gamma\operatorname{Ei}(-1)-\dfrac{1}{2}e\int\limits_{1}^{\infty} e^{-x}\ln^2 x\,\mathrm{d}x \\ &= e\gamma\operatorname{Ei}(-1)-\dfrac{1}{2}e\dfrac{\mathrm{d}^2}{\mathrm{d}x^2}\Gamma(1)+\dfrac{1}{2}e\int\limits_{0}^{1} e^{-x}\ln^2 x\,\mathrm{d}x \\ \end{align*}

$$ \int\limits_{0}^{\infty} \dfrac{e^{-x}\ln x}{1+x}\,\mathrm{d}x = e\gamma\operatorname{Ei}(-1)-\dfrac{1}{2}e\left(\gamma^2+\dfrac{1}{6}\pi^2\right)-e\sum\limits_{n=1}^{\infty} \dfrac{(-1)^n}{n!\,n^2} \tag{5} \label{first} $$


Let $$ f(\lambda) = \int\limits_{0}^{\infty} e^{-x}\ln^2 \left(\lambda x + x^2\right)\,\mathrm{d}x $$

Then \begin{align*} f'(\lambda) &= 2\int\limits_{0}^{\infty} e^{-x}\dfrac{\ln \left(\lambda x + x^2\right)}{\lambda + x}\,\mathrm{d}x \\ &= 2\int\limits_{0}^{\infty} \dfrac{e^{-\lambda x}}{1+x}\Big(\ln \left(1+x\right)+\ln x + 2 \ln \lambda\Big)\,\mathrm{d}x \\ &= 4\ln \lambda \int\limits_{0}^{\infty} \dfrac{e^{-\lambda x}}{1+x}\,\mathrm{d}x + 2\int\limits_{0}^{\infty} \dfrac{e^{-\lambda x}}{1+x}\ln \left(1+x\right)\,\mathrm{d}x + 2\int\limits_{0}^{\infty} \dfrac{e^{-\lambda x}}{1+x}\ln x\,\mathrm{d}x \\ &= -4\ln \lambda e^{\lambda}\operatorname{Ei}(-\lambda) - 2\dfrac{\partial}{\partial \beta}\left.\left(\int\limits_{0}^{\infty} \dfrac{e^{-\lambda x}}{\left(1+x\right)^{1+\beta}}\,\mathrm{d}x\right)\right\vert_{\beta=0} + 2\dfrac{\partial}{\partial \beta}\left.\left(\int\limits_{0}^{\infty} \dfrac{e^{-\lambda x}x^{\beta}}{1+x}\,\mathrm{d}x\right)\right\vert_{\beta=0} \\ &= -4\ln \lambda e^{\lambda}\operatorname{Ei}(-\lambda) -2e^{\lambda}\dfrac{\partial}{\partial \beta}\Big(\lambda^{\beta}\Gamma(-\beta,\lambda)\Big)\Bigg\vert_{\beta=0} + 2e^{\lambda}\dfrac{\partial}{\partial \beta}\Big(\Gamma(1+\beta)\Gamma(-\beta,\lambda)\Big)\Bigg\vert_{\beta=0} \\ &= -2\ln \lambda e^{\lambda}\operatorname{Ei}(-\lambda)+2\gamma e^{\lambda}\operatorname{Ei}(-\lambda) \end{align*} Integrating over $[0,1]$ leads us to \begin{align*} f(1) &= f(0) -2\int\limits_{0}^{1} \ln \lambda e^{\lambda}\operatorname{Ei}(-\lambda)\,\mathrm{d}\lambda + 2\gamma\Big(e^{\lambda}\operatorname{Ei}(-\lambda)-\ln \lambda\Big)\Bigg\vert_{0}^{1} \\ &= 2\gamma e\operatorname{Ei}(-1)+2\gamma^2+\dfrac{2}{3}\pi^2-2\int\limits_{0}^{1} \ln \lambda e^{\lambda}\operatorname{Ei}(-\lambda)\,\mathrm{d}\lambda \end{align*} For integral in last formula we use \eqref{lemma1}: \begin{align*} \int\limits_{0}^{1} \ln \lambda e^{\lambda}\operatorname{Ei}(-\lambda)\,\mathrm{d}\lambda &= \int\limits_{0}^{1} \ln^2 \lambda e^{\lambda}\,\mathrm{d}\lambda +\gamma \int\limits_{0}^{1} \ln \lambda e^{\lambda}\,\mathrm{d}\lambda + \sum\limits_{n=2}^{\infty} \dfrac{\mathcal{H}_{n-1}}{n!\ n} \\ &= \sum\limits_{n=1}^{\infty} \dfrac{1}{n!\ n^2} + \gamma^2-\gamma\operatorname{Ei}(1) + \sum\limits_{n=1}^{\infty} \dfrac{\mathcal{H}_{n}}{n!\ n} \end{align*} Combining all together we have

$$ \int\limits_{0}^{\infty} e^{-x}\ln^2 \left(x+x^2\right)\,\mathrm{d}x = 2\gamma\Big(e\operatorname{Ei}(-1)+\operatorname{Ei}(1)\Big)+\dfrac{2}{3}\pi^2-2\sum\limits_{n=1}^{\infty} \dfrac{1}{n!\ n^2}-2\sum\limits_{n=1}^{\infty} \dfrac{\mathcal{H}_{n}}{n!\ n} \tag{6} \label{second} $$

From another side \begin{align*} \int\limits_{0}^{\infty} e^{-x}\ln^2 \left(x+x^2\right)\,\mathrm{d}x &= \int\limits_{0}^{\infty} e^{-x}\Big(\ln^2 \left(1+x\right) + \ln^2 x + 2\ln \left(1+x\right)\ln x\Big)\,\mathrm{d}x \\ &= \gamma^2+\dfrac{1}{6}\pi^2+e\int\limits_{1}^{\infty}e^{-x}\ln^2 x\,\mathrm{d}x + 2\int\limits_{0}^{\infty} e^{-x}\ln \left(1+x\right)\ln x\,\mathrm{d}x \\ &= \left(1+e\right)\left(\gamma^2+\dfrac{1}{6}\pi^2\right)+2e\sum\limits_{n=1}^{\infty} \dfrac{(-1)^n}{n!\ n^2} + 2\int\limits_{0}^{\infty} e^{-x}\ln \left(1+x\right)\ln x\,\mathrm{d}x \end{align*} Express last integral

\begin{align*} \int\limits_{0}^{\infty} e^{-x}\ln \left(1+x\right)\ln x\,\mathrm{d}x &= \gamma\Big(e\operatorname{Ei}(-1)+\operatorname{Ei}(1)\Big) -\dfrac{1}{2}\gamma^2+\dfrac{1}{4}\pi^2-\dfrac{1}{2}e\left(\gamma^2+\dfrac{1}{6}\pi^2\right) \\ &-e\sum\limits_{n=1}^{\infty} \dfrac{(-1)^n}{n!\ n^2}-\sum\limits_{n=1}^{\infty} \dfrac{1}{n!\ n^2}-\sum\limits_{n=1}^{\infty} \dfrac{\mathcal{H}_{n}}{n!\ n} \end{align*}

Now use integration by parts and already obtained result \eqref{first}: \begin{align*} \int\limits_{0}^{\infty} e^{-x}\ln \left(1+x\right)\ln x\,\mathrm{d}x &= \int\limits_{0}^{\infty} e^{-x}\dfrac{\ln x}{1+x}\,\mathrm{d}x + \int\limits_{0}^{\infty} e^{-x}\dfrac{\ln \left(1+x\right)}{x}\,\mathrm{d}x \\ &= e\gamma\operatorname{Ei}(-1)-\dfrac{1}{2}e\left(\gamma^2+\dfrac{1}{12}\pi^2\right)-e\sum\limits_{n=1}^{\infty} \dfrac{(-1)^n}{n!\ n^2}+ \int\limits_{0}^{\infty} e^{-x}\dfrac{\ln \left(1+x\right)}{x}\,\mathrm{d}x \end{align*} So

$$ \int\limits_{0}^{\infty} e^{-x}\dfrac{\ln \left(1+x\right)}{x}\,\mathrm{d}x = -\dfrac{1}{2}\gamma^2+\dfrac{1}{4}\pi^2+\gamma\operatorname{Ei}(1)-\sum\limits_{n=1}^{\infty} \dfrac{1}{n!\ n^2}-\sum\limits_{n=1}^{\infty} \dfrac{\mathcal{H}_{n}}{n!\ n} \tag{7} \label{final} $$


Back to original problem

With integration by parts and substitution original integral can be converted to \begin{align*} \int\limits_{0}^{e} \operatorname{Li}_2\left(\ln x\right)\,\mathrm{d}x &= \dfrac{1}{6}e\pi^2-\int\limits_{0}^{\infty} e^{-x}\dfrac{\ln \left(1+x\right)}{x}\,\mathrm{d}x+\int\limits_{0}^{1} e^{x}\dfrac{\ln \left(1-x\right)}{x}\,\mathrm{d}x \\ &= \dfrac{1}{6}e\pi^2 + \dfrac{1}{2}\gamma^2-\dfrac{1}{4}\pi^2-\gamma\operatorname{Ei}(1)+\sum\limits_{n=1}^{\infty} \dfrac{1}{n!\ n^2}+\sum\limits_{n=1}^{\infty} \dfrac{\mathcal{H}_{n}}{n!\ n} \\ &+\int\limits_{0}^{1} \dfrac{\ln\left(1-x\right)}{x}\,\mathrm{d}x+\sum\limits_{n=1}^{\infty}\dfrac{1}{n!}\int\limits_{0}^{1} x^{n-1}\ln\left(1-x\right)\,\mathrm{d}x \\ &= \dfrac{1}{6}e\pi^2 + \dfrac{1}{2}\gamma^2-\dfrac{5}{12}\pi^2-\gamma\operatorname{Ei}(1)+\sum\limits_{n=1}^{\infty} \dfrac{1}{n!\ n^2} \end{align*}

Knas
  • 856
  • 6
  • 10
2

Here I would like to sketch an alternative integral representation for $$ I(2)=\int_0^e\operatorname{Li}_2\left(\ln{x}\right)\,dx. $$ I hope that this argument shed some light on, why we cannot expect a much better representation as the hypergeometric series representation, which was given in the question.

After a change of variable $x \mapsto \ln(x)$, using an integral representation for $\operatorname{Li}_2$, if we interchange of the order of integration, we arrive at \begin{align} I(2) = \int_0^e\operatorname{Li}_2\left(\ln{x}\right)\,dx &= \int_{-\infty}^1e^x\operatorname{Li}_2\left(x\right)\,dx \\ &= - \int_{-\infty}^1 e^x \int_0^1\frac{\ln\left(1-xt\right)}{t}\,dt\,dx \\ &= - \int_0^1 \frac{1}{t} \int_{-\infty}^1 e^x\ln\left(1-xt\right)\,dx\,dt. \end{align}

By using the antiderivative of the function $e^x\ln\left(1-xt\right)$ with respect to $x$, we have \begin{align} I(2) &= - \int_0^1 \left(e\ln(1-t) + e^{1/t}E_1\left(\frac{1}{t}-1\right)\right)\,\frac{dt}{t} \\ &= - e\int_0^1 \frac{\ln(1-t)}{t} \,dt - \int_0^1 e^{1/t}E_1\left(\frac{1}{t}-1\right)\,\frac{dt}{t} \\ &= \frac{e\pi^2}{6} - \int_0^1 e^{1/t}E_1\left(\frac{1}{t}-1\right)\,\frac{dt}{t}, \end{align} where $E_1(x)$ is the $E_n$ function of order $n=1$. Using a change of variable $t\mapsto1/x$, we have $$ I(2) = \frac{e\pi^2}{6} - \int_1^\infty \frac{e^x E_1(x-1)}{x}\,dx. $$ We stop here. Although it would be nice to evaluate this integral, but as I know, this kind of integrals are not in the literature. Here I want to mention an earlier identity in a question of mine, where a solution of a related integral problem was also given in terms of ${_k}F_k\left(1,\dots,1;2,\dots,2;1\right)$. Of course any ideas for further evaluation are welcome. By using an integral representation for the $E_1$ function, we arrive at \begin{align} I(2) &= \frac{e\pi^2}{6} - \int_1^\infty \frac{e^x}{x} \int_1^\infty \frac{e^{-(x-1)t}}{t} \,dt \,dx \\ &= \frac{e\pi^2}{6} - \int_1^\infty \int_1^\infty \frac{e^{x+(1-x)t}}{xt} \,dt \,dx. \end{align}

user153012
  • 12,240
  • 1
    This form for $I(2)$ popped up during my calculations, but I chose to omit it since it appeared to lead nowhere. However, it's good to see that someone else arrived at the same form as well. Also, swapping the order of integration in the final double integral you've posted just leads to $$\int_1^\infty \frac{e^t E_1(t-1)}{t}dt.$$ Furthermore, $x+(1-x)t=t+(1-t)x$, so the order of integration is irrelevant--this transformation is equivalent to making the substitution $x\mapsto t$. – teadawg1337 Nov 23 '16 at 14:36
2

Too long for a comment.

If we evaluate the decay rate of your sum we find that $$\sum_{k=1}^\infty\frac{1}{k^2 k!}<\frac{1}{\sqrt{2\pi}}\sum_{k=1}^\infty\frac{e^k}{k^{k+5/2}}<\frac{1}{\sqrt{2\pi}}\sum_{k=1}^\infty\left(\frac{e}{k}\right)^k$$ This final series goes to zero smaller than the function $x^{-n}$ for any positive $n$, which helps us estimate the decay rate a little. We can also note that a number is likely irrational if $a_{k+1} \ll a_k^2$. Here we have that $$a_k^2\sim\left(\frac{1}{\sqrt{2\pi}\;k^{k+ 5/2} e^{-k}}\right)^2=\frac{e^{2 k} k^{-2 k - 5}}{2 \pi}$$ $$a_{k+1} \sim \frac{1}{\sqrt{2\pi}\;(k+1)^{n+7/2} e^{-k-1}} = \frac{e^{k + 1} (k + 1)^{-k - 7/2}}{\sqrt{2 \pi}}$$ This tells us that $$\frac{a_k^2}{a_{k+1}} \sim \frac{e^{k - 1} k^{-2 k - 5} (k + 1)^{k + 7/2}}{\sqrt{2 \pi}}$$ It's interesting to note how fast this goes to zero; in fact, if we denote the above by $f(k)$ and let $n$ be any real number we find that $$\lim_{k\to\infty}\frac{f(k)}{e^{nk}}=0$$
Thus, we see that this decay is very rapid, and conclude that $a_{k+1} \ll a_k^2$. I'm pretty sure this implies a proof of irrationality could be made (the fact that it is a sum of rational, decreasing terms alone encourages this), and possibly even a proof that the series is transcendental. Not too keen with constructing sequences to prove even irrationality though. Just let these serve as heuristics for now!