23
$\begingroup$

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 $$\lim_{a\to\infty }\frac{aI(a)}{\ln(a)}=1$$

$\endgroup$
6
  • 1
    $\begingroup$ The result for second indefinite integral is not correct without $x$ in the numerator under the integral. $\endgroup$ Commented May 29, 2018 at 1:39
  • $\begingroup$ do you know the Omega-constant? $\endgroup$ Commented Jun 2, 2018 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? $\endgroup$ Commented Jun 2, 2018 at 23:04
  • $\begingroup$ some of its integral reps look pretty related ... $\endgroup$ Commented Jun 4, 2018 at 22:46
  • 1
    $\begingroup$ @tired Ah, you must be talking about this one: $$\int_{-\infty}^\infty \frac{dx}{(e^x-x)^2+\pi^2}=\frac{1}{1+\Omega}$$ $\endgroup$ Commented Jun 4, 2018 at 23:25

5 Answers 5

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 $$ \left| a \right|x/e^{\,x} < \left| a \right|1/e < 1\quad \Rightarrow \quad \left| a \right| < e $$

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.

$\endgroup$
7
  • $\begingroup$ Your series in terms of the polylogarithms is formal because the Stirling series is divergent. $\endgroup$ Commented May 18, 2021 at 17:52
  • $\begingroup$ @Gary you are right for the series for Gamma, but here it is divided by $(n+1)^{n+1}$ .. $\endgroup$ Commented May 19, 2021 at 0:18
  • $\begingroup$ I mean that the series $$ \sum\limits_{k = 0}^\infty {c_k \frac{1}{{(n + 1)^k }}} $$ does not converge for any $n\geq 0$. It is a factorially divergent series, meaning that the $c_k$ diverge like a factorial of $k$ divided by a constant to the power $k$. This growth rate of the $c_k$ cannot be beaten by $(n+1)^k$ no matter how large $n$ is. $\endgroup$ Commented May 19, 2021 at 5:42
  • $\begingroup$ You may have a look at the introduction of en.wikipedia.org/wiki/Asymptotic_expansion for the meaning of (divergent) asymptotic expansions. In the examples you can find a figure on the diverging nature of the Stirling expansion as well. $\endgroup$ Commented May 19, 2021 at 6:02
  • $\begingroup$ Did you understand the issue? $\endgroup$ Commented May 26, 2021 at 14:01
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 $$I'(a)=-\int_0^{+\infty} \frac{x\,dx}{(e^x+ax)^2}=-\int_0^{+\infty} \frac{x^2\,dx}{(e^x+ax)^2}.$$

$\endgroup$
1
  • $\begingroup$ Thanks for the interest! I hope you stick with the problem (as I am) ... this integral is really driving me crazy. $\endgroup$ Commented May 31, 2018 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.

$\endgroup$

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.