5
$\begingroup$

Can the integral

$$ \int_0^\infty \frac{\log(x)-\log(a)}{x-a}e^{-x} \mathrm{d}x $$

be expressed in terms of some simple special function? I have searched through integral tables but couldn't find anything.

I am mainly interested in values $a \geq 1$, but the integral should be well-defined for all $a>0$.

$\endgroup$
  • $\begingroup$ Perhaps this is helpful:$$\int^\infty_0\dfrac{\log(x)-\log(a)}{x-a}e^{-x}dx=-\int^\infty_0\dfrac{\log(x)-\log(a)}{x-a}de^{-x}$$ $$=-\lim_{b\to\infty}e^{-x}\dfrac{\log(x)-\log(a)}{x-a}|^b_0+\int^\infty_0e^{-x}d\left(\dfrac{\log(x)-\log(a)}{x-a}\right)$$ Now, $$d\left(\dfrac{\log(x)-\log(a)}{x-a}\right)=\dfrac{(x-a)/x-\log(x)-\log(a)}{(x-a)^2}dx$$ $\endgroup$ – user122283 Mar 26 '14 at 19:57
2
$\begingroup$

This integral may be interpreted as a Laplace transform. To see this, let $u=x/a$. Then, $$ \int_0^\infty \frac{\log{x}-\log{a}}{x-a} \mathrm{e}^{-x}\,dx = \int_0^\infty \frac{\log{u}}{u-1} \mathrm{e}^{-au}\,du.$$

In other words, if you knew the Laplace transform of $\frac{\log{x}}{x-1}$, this integral would be solved. Mathematica can do this, but the result is not helpful:

LaplaceTransform[Log[x]/(x - 1), x, a]

$$ e^{-a} \left(-G_{2,3}^{3,0}\left(-a\left| \begin{array}{c} 1,1 \\ 0,0,0 \\ \end{array} \right.\right)+(\log (a)+\gamma +2 i \pi ) (-\Gamma (0,-a))\right) $$

Edit: Summarizing the comments below, the branch cut that Mathematica chose apparently introduced an imaginary component. Let's fix this. First, we re-rewrite the integral as $$ \mathrm{e}^{-a}\left[\int_{-1}^0 + \int_0^\infty\right]\left\{\frac{\log(x+1)}{x}\mathrm{e}^{-ax}\,dx\right\}.$$

The second integral in that can be taken exactly by Mathematica to be

MeijerG[{{0}, {1}}, {{0, 0, 0}, {}}, a]

The first integral, however, let us write that as $I(a)$. Clearly, $$\frac{d^n I}{da^n} = -\int_0^1 x^{n-1}\log(1-x)\mathrm{e}^{ax}\,dx$$

We can (mostly) evaluate these when $a=0$; so let us Taylor expand $I(a)$ about $a=0$. We will obtain $$ I(a) = \frac{\pi^2}{6} + \sum_{n=1}^\infty \frac{H_n}{(n+1)!} a^n $$ where $H_n$ denotes the $n$ Harmonic number. (Partial sums of the harmonic series.)

I don't think we can improve much on this; to summarize, the integral, as originally asked, should be $$ \mathrm{e}^{-a}G_{2,3}^{3,1}\left(a\left| \begin{array}{c} 0,1 \\ 0,0,0 \\ \end{array} \right.\right) + \mathrm{e}^{-a}\left[\frac{\pi^2}{6} + \sum_{n=1}^\infty \frac{H_n}{(n+1)!} a^n\right]. $$

$\endgroup$
  • $\begingroup$ Thanks, but could there be a glitch in the formula you are reporting? The imaginary parts don't seem to cancel out, yet the integral is clearly real-valued. For $a=1$ for example, WolframAlpha gives me -MeijerG[{{},{1,1}},{{0,0,0},{}},-1] $$ -G_{2,3}^{3,0}\left(-1\left|\begin{array}{c} 1,1 \\ 0,0,0 \\\end{array}\right.\right) = 2.799\dots - i \cdot 1.813\dots $$ whereas (EulerGamma+2*i*Pi)*Gamma[0,-1] $$ (\gamma + 2 i \pi ) \Gamma(0,-1) = 18.645\dots - i \cdot 13.720\dots $$ Could you please double-check? $\endgroup$ – jens Mar 27 '14 at 1:14
  • $\begingroup$ I can only imagine that Mathematica inserted the $2\pi i$ as a choice of branch cut. (I'm not familiar at all with this MeijerG function, so unfortunately cannot give much input.) On the brighter side, if you're interested in the limits of $a\to0$ or $a\to\infty$, there are some very nice asymptotic approximation techniques available to you. $\endgroup$ – Jason Mar 27 '14 at 19:02
  • $\begingroup$ Okay, here is some Mathematica wrangling. First, we rewrite the integral as: $$ \mathrm{e}^{-a} \left[ \int_{-1}^0 \frac{\log(x+1)}{x}\mathrm{e}^{-ax}\,dx + \int_0^\infty \frac{\log(x+1)}{x}\mathrm{e}^{-ax}\,dx \right]. $$ Mathematica takes the 2nd integral in that exactly as >> MeijerG[{{0}, {1}}, {{0, 0, 0}, {}}, a] Then, for $[-1,0]$, we may rewrite that part as: $$ -\int_0^1 \frac{\log(1-x)}{x}\mathrm{e}^{ax}\,dx. $$ At this point, I think you'll have to appeal to an asymptotic argument to estimate this integral. But this at least restores the "realness" that we should have. $\endgroup$ – Jason Mar 27 '14 at 19:17
  • $\begingroup$ Even more cuteness, at $a=0$, we have exactly $\pi^2/6$. (Easily seen by Taylor series.) $\endgroup$ – Jason Mar 27 '14 at 19:34
  • $\begingroup$ Thanks for your answer! I was wondering whether that Meijer G-function could be demystified a little, so I ran WolframAlpha and deduced that $$ G_{2,3}^{3,1}\left(a\left|\begin{array}{c} 0,1 \\ 0,0,0 \\\end{array}\right.\right) = -\dfrac{\gamma^2}{2} + \dfrac{\pi^2}{4} - \gamma\log(a) - \dfrac{\log^2(a)}{2} + \gamma\operatorname{Ei}(a) + \operatorname{Ei}(a)\log(a) - 2a \cdot {}_3F_3(1,1,1;2,2,2;a) - \dfrac{1}{2}\left.\dfrac{\partial^2{}_1F_1(x,y,z)}{\partial x^2}\right|_{(x,y,z)=(0,1,a)} $$ $\endgroup$ – jens Mar 28 '14 at 12:13

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.