21
$\begingroup$

I'm investigating integrals in the form $$I(a):=\int_0^\infty \frac{dx}{e^x+ax}$$ So far, I haven't been able to find any special values other than $I(0)=1$, and I've only managed to evaluate these similar indefinite integrals: $$\int \frac{x-1}{e^x+ax}dx=-\frac{\ln(1+axe^{-x})}{a}+C$$ $$\int \frac{xdx}{e^x+x+1}=-\ln(1+e^{-x}(x+1))+C$$ I've also found the following series representation for $I(a)$: $$I(a)=\sum_{n=0}^\infty \frac{(-a)^n n!}{(n+1)^{n+1}}$$ ...which looks remarkably similar to the Maclaurin series for the Lambert-W function.

QUESTION: Can anyone find any non-trivial special values of this integral? I find this unlikely because of the weird series representation of $I(a)$, so if this isn't feasible, can anyone find any interesting properties or functional/differential equations for $I(a)$?

UPDATE: I've managed to show that $$\lim_{a\to\infty }\frac{aI(a)}{\ln(a)}=1$$

$\endgroup$
  • 1
    $\begingroup$ The result for second indefinite integral is not correct without $x$ in the numerator under the integral. $\endgroup$ – CuriousGuest May 29 '18 at 1:39
  • $\begingroup$ do you know the Omega-constant? $\endgroup$ – tired Jun 2 '18 at 14:28
  • $\begingroup$ @tired Absolutely, $\Omega$ is one of my favorite mathematical constants (next to the Dottie Number). Why? Do you think it is relevant? $\endgroup$ – Franklin Pezzuti Dyer Jun 2 '18 at 23:04
  • $\begingroup$ some of its integral reps look pretty related ... $\endgroup$ – tired Jun 4 '18 at 22:46
  • 1
    $\begingroup$ @tired Ah, you must be talking about this one: $$\int_{-\infty}^\infty \frac{dx}{(e^x-x)^2+\pi^2}=\frac{1}{1+\Omega}$$ $\endgroup$ – Franklin Pezzuti Dyer Jun 4 '18 at 23:25
11
+50
$\begingroup$

Notice first that

$$ I(a) = \int_{0}^{\infty} \frac{xe^{-x}}{1+axe^{-x}} \, dx. $$

Indeed, this follows from $\int_{0}^{\infty} \frac{x-1}{e^x + ax} \, dx = 0$ using OP's computation. Since the graph of $x \mapsto xe^{-x}$ is unimodal, for each $u$ in the range we may define the 'width' $l(u)$ of the graph of $xe^{-x}$ at height $u$.

$\hspace{8em}$ Graph

To be precise, we define $l(u)$ as the Lebesgue measure of the set $\{ x > 0 : xe^{-x} > u \}$. Then

$$ I(a) = \int_{0}^{\infty} \left( \int_{xe^{-x}}^{\infty} \frac{du}{(1+au)^2} \right) \, dx \stackrel{\text{(Fubini)}}{=} \int_{0}^{\infty} \frac{l(u)}{(1+au)^2} \, du. $$

Now $l$ can be written explicitly in terms of the Lambert W-function:

$$ l(u) = \begin{cases} W(-u) - W_{-1}(-u), & \text{if } u \leq \frac{1}{e} \\ 0, & \text{if } u > \frac{1}{e} \end{cases} $$

So it follows that

$$ I(a) = \int_{0}^{\frac{1}{e}} \frac{l(u)}{(1+au)^2} \, du = \int_{0}^{\frac{1}{e}} \frac{W(-u) - W_{-1}(-u)}{(1+au)^2} \, du. \tag{1} $$

This suggests that the asymptotic behavior of $I(a)$ as $a\to\infty$ is intimately related to the asymptotic behavior of $W_{-1}(u)$ as $u\to 0$. For instance, using the fact that

$$ l(u) = -W_{-1}(-u) + \mathcal{O}(1) = -\log u + \log\log(1/u) + \mathcal{O}(1) $$

on $(0, 1/e]$ as $u\to0$, we obtain

$$ I(a) = \frac{\log a}{a} + \frac{\log\log a}{a} + \mathcal{O}\left(\frac{1}{a}\right) \quad \text{as } a \to \infty. \tag{2} $$


We also notice that for $n \geq 1$,

\begin{align*} \left( \frac{d}{da} \right)^n (aI(a)) &= (-1)^{n-1} n! \int_{0}^{\infty} \frac{x^{n-1}e^{-nx}}{(1 + axe^{-x})^{n+1}} \, dx \\ &= \frac{(-1)^{n-1} n!}{a^n} \int_{0}^{\infty} \frac{u^{n-1}e^{-nu/a}}{(1 + ue^{-u/a})^{n+1}} \, du \\ &\sim \frac{(-1)^{n-1} (n-1)!}{a^n} \quad \text{as } a \to \infty. \end{align*}

$\endgroup$
2
$\begingroup$

Here are my notes on this so far with $a=1$. I hope these are useful.

It seems that $$ \int \frac{dx}{e^x + x} = \sum_{n=0}^\infty \left[\sum_{k=0}^n\frac{(n-k+1)^k}{k!}\right]\frac{(-1)^nx^{n+1}}{(n+1)} $$ this doens't look like the Cauchy product of series. It seems that $$ \sum_{k=0}^n\frac{(n-k+1)^k}{k!} \sim \kappa e^{W(1)n} $$ where $W(x)$ is the Lambert-W function, and $W(1)=\Omega$. Taking a limit $$ \lim_{n \to \infty} \left(e^{- W(1) n}\sum_{k=0}^n\frac{(n-k+1)^k}{k!}\right)=\kappa\approx 1.251190909867859\cdots $$ if this was a valid series expansion, then the function is zero at $x=0$, so the infinite asymptotics may yield a result. (I may be wrong)

Edit: the more general series with $a$ seems to be $$ \int \frac{dx}{e^x+ax} = \sum_{n=0}^\infty \left[\sum_{k=0}^n \frac{a^k(k+1)^{n-k}}{(n-k)!} \right] \frac{(-1)^n x^{n+1}}{n+1} $$ a similar treatment seems to give $$ \sum_{k=0}^n \frac{a^k(k+1)^{n-k}}{(n-k)!} \sim e^{\left(\log(a)+W\left(\frac{1}{a}\right)\right)n} $$ this came from guessing and using the inverse symbolic calculator.

Note:

If it assists with Sangchul Lee's integral representation, it appears that $$ \Re\left(\int_0^{1/e} u^n W(-u) \; du \right)= \frac{n!-q(n)}{(n+1)^{n+2} e^{n+1}}, \;\; q(n) \in \mathbb{N} $$ where the $q's$ go like $3,9,53,462,5319,76008,1296273,25679664,579336363,\cdots$, for $n=0,1,\cdots$ but it is not clear what these are. Further it seems that $$ \Re\left(\int_0^{1/e} u^n W(-u) \; du \right)= (n+1)^{-n-2} \Gamma (n+1)+\left(\frac{1}{n+1}\right)^{n+3} ((n+1) \Gamma (n+2,n+1)-\Gamma (n+3,n+1)) $$ so then $$ q(n) = \left(\frac{1}{n+1}\right)^n (n+1)^{2 n+1} \left(e^{n+1} E_{-n-1}(n+1)+1\right) $$ for exponential integral function.

$\endgroup$
2
$\begingroup$

Starting from the series that you already got $$ \eqalign{ & I(a) = \int_0^\infty {{{dx} \over {e^{\,x} + ax}}} = \int_0^\infty {{{e^{\, - x} dx} \over {\left( {1 + axe^{\, - x} } \right)}}} = \cr & = \int_0^\infty {\sum\limits_{0\, \le \,k} {\left( { - 1} \right)^{\,k} \left( {a^{\,k} x^{\,k} e^{\, - \,\left( {k + 1} \right)\,x} } \right)} \;dx} = \sum\limits_{0\, \le \,k} {\left( { - 1} \right)^{\,k} {{k!} \over {\left( {k + 1} \right)^{k + 1} }}a^{\,k} } \cr} $$ and which converges for $$ \left| a \right|x/e^{\,x} < \left| a \right|1/e < 1\quad \Rightarrow \quad \left| a \right| < e $$

We can get a series in terms of Polylogarithms as follows: $$ \eqalign{ & I(a) = \sum\limits_{0\, \le \,n} {{{n!} \over {\left( {n + 1} \right)^{n + 1} }}\left( { - a} \right)^{\,n} } = \sum\limits_{0\, \le \,n} {{{\Gamma (n + 1)} \over {\left( {n + 1} \right)^{n + 1} }}\left( { - a} \right)^{\,n} } = \cr & = \sqrt {{{2\,\pi } \over e}} \;\sum\limits_{0\, \le \,n} {{1 \over {\sqrt {n + 1} }}\;e^{\, - \;n} \left( {\sum\limits_{0\, \le \,k} {c_{\,k} {1 \over {\left( {n + 1} \right)^{\,k} }}} } \right)\left( { - a} \right)^{\,n} } = \cr & = \sqrt {{{2\,\pi } \over e}} \;\sum\limits_{0\, \le \,k} {c_{\,k} \left( {\sum\limits_{0\, \le \,n} {{{\left( { - a/e} \right)^{\,n} } \over {\left( {n + 1} \right)^{\,k + 1/2} }}\;} } \right)} = \sqrt {{{2\,\pi } \over e}} \;\sum\limits_{0\, \le \,k} {c_{\,k} \Phi \left( { - a/e,\,\;k + 1/2,\;1} \right)} = \cr & = - {{\sqrt {2\,\pi e} } \over a}\;\sum\limits_{0\, \le \,k} {c_{\,k} {\rm Li}_{\,k + 1/2} \left( { - a/e} \right)} \cr} $$ where:
- $c_k$ are the coefficients of the Stirling asymptotic expansion of Gamma function;
- $\Phi(z,s,\alpha)$ is the Lerch Transcendant;
- $ {\rm Li}$ is the Polylogarithm.

---- Addendum ---- From this related post we get $$ \sum\limits_{1\, \le \,\,n} {{1 \over {n^{\,n} }}x^{\,n} } = x\sum\limits_{0\, \le \,\,n} {{1 \over {\left( {n + 1} \right)^{\,n + 1} }}x^{\,n} } = x\int_{\,0}^{\,1} {t^{\, - \,x\,t} dt} $$ and since $$ A(z) = \sum\limits_{0\, \le \,n} {a_n \,z^n } \quad \Leftrightarrow \quad \int_{\;t\, = \,0}^\infty {e^{\, - \,t} A(z\,t)\,d\,t} = \sum\limits_{0\, \le \,n} {n!a_n z^{\,n} } $$ we get another integral representation $$ \bbox[lightyellow] { \eqalign{ & I( - x) = \sum\limits_{0\, \le \,\,n} {{{n!} \over {\left( {n + 1} \right)^{\,n + 1} }}x^{\,n} } = \int_0^\infty {{{e^{\, - u} du} \over {1 - x\,u\,e^{\, - u} }}} = \cr & = \int_{\,u\, = \,0}^{\,\infty } {e^{\, - \,u} \int_{\,t\, = \,0}^{\,\,1} {t^{\, - \,x\,u\,t} dt\,} du} = \int_{\,t\, = \,0}^{\,1} {\int_{\,u\, = \,0}^{\,\infty } {e^{ - \,u\left( {1 + x\,t\ln t} \right)}\, dt\,} du} = \cr & = \int_{\,t\, = \,0}^{\,1} {{{dt} \over {\left( {1 + x\,t\ln t} \right)}}} \cr} }$$

Now the second line tells us that $$ I( - 1/s) = \int_{\,u\, = \,0}^{\,\infty } {e^{\, - \,u} \int_{\,t\, = \,0}^{\,\,1} {t^{\, - \,\,\left( {u/s} \right)\,t} dt\,} du} $$ i.e. $$ \bbox[lightyellow] { \eqalign{ & {1 \over s}I( - 1/s) = \int_{\,\alpha \, = \,0}^{\,\infty } {e^{\, - \,s\,\alpha } \left( {\int_{\,t\, = \,0}^{\,\,1} {t^{\, - \,\,\alpha \,t} dt\,} } \right)d\alpha } = \cr & = \int_0^\infty {{{e^{\, - u} } \over {s - \,u\,e^{\, - u} }}du} = \int_{\,t\, = \,0}^{\,1} {{{dt} \over {\left( {s + \,t\ln t} \right)}}} \cr} }$$ so that our integral is tied to the Laplace transform of the interesting function
$\int_{0}^1 {t^{-xt}}dt=\sum_{n=1}^\infty \frac{x^{n-1}}{n^n} = $ Sphd$(-x;1)$
cited by JJacquelin in his answer to the already cited post.

$\endgroup$
1
$\begingroup$

Not an answer, but an observation (to express my interest in your question myself). Another integral representation of $I(a)$ is $$I(a)=\int_0^{+\infty} \frac{x\,dx}{e^x+ax}$$ (follows from the first of indefinite integrals in your question). Also $$I'(a)=-\int_0^{+\infty} \frac{x\,dx}{(e^x+ax)^2}=-\int_0^{+\infty} \frac{x^2\,dx}{(e^x+ax)^2}.$$

$\endgroup$
  • $\begingroup$ Thanks for the interest! I hope you stick with the problem (as I am) ... this integral is really driving me crazy. $\endgroup$ – Franklin Pezzuti Dyer May 31 '18 at 22:38
1
$\begingroup$

Too long for comment.

Using integration by parts, easy to obtain for $m\ge0,\ n\ge1:$ $$J(m,n)= \int\limits_0^1 t^m \log^n t\,\mathrm dt = \dfrac {t^{m+1}}{m+1}\log^n t\Bigg|_0^1 - \dfrac n{m+1}\int\limits_0^1 t^m\log^{n-1}t\,\mathrm dt= -\dfrac{n}{m+1}J(m, n-1),$$ $$J(m,n)=(-1)^n\dfrac{n!}{(m+1)^{n+1}}.\tag1$$

This allow to calculate Taylor series for the integral $$I(a,m,n) = \int\limits_0^\infty \dfrac{e^{-mx}x^n}{e^x+ax}\,\mathrm dx = \int\limits_0^1\dfrac{t^m\log^n t}{1-at\log t}\,\mathrm dt = \sum_{k=0}^\infty J(m+k,n+k)a^k,$$ $$I(a,m,n) = \sum_{k=0}^\infty(-1)^{n+k}\dfrac{(n+k)!}{(m+k+1)^{n+k+1}}a^k.\tag2$$ Formula $(2)$ can be useful for the further investigations.

$\endgroup$

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.