0
$\begingroup$

I have been looking for an explicit solution to the following Laplace transform for $\alpha,\mu,\beta>0$ \begin{equation} \frac{\beta^\alpha}{\Gamma(\alpha)}\mathscr{L}\left\{\frac{t^{\alpha-1}}{t-\mu}\right\}(\beta) =\frac{\beta^\alpha}{\Gamma(\alpha)}\operatorname{PV\int_0^\infty}\frac{t^{\alpha-1}}{t-\mu}e^{-\beta t}\,\mathrm dt. \end{equation} Notice that this is equivalent to $E((X-\mu)^{-1})$ for $X\sim\text{Gamma}(\alpha,\beta)$. I attacked this problem by substituting $x=t-\mu$ and then writing \begin{equation} \tag{1} \lim_{\epsilon\to0^+} \frac{\beta^\alpha}{\Gamma(\alpha)}e^{-\beta\mu}\left(\int_{-\mu}^0+\int_0^\infty\right) x^{\epsilon-1}(x+\mu)^{\alpha-1}e^{-\beta x}\,\mathrm dx. \end{equation} After evaluating these integrals, I took the limit and and then dropped the residue term $-f_X(0)\pi\mathrm i$ resulting from the pole when $\epsilon=0$. This took me about six pages to do and involves some nasty derivatives of hypergoemetric functions w.r.t. parameters, meijer $G$ functions, residue expansions, etc. The solution I got was \begin{equation} \frac{\beta^\alpha}{\Gamma(\alpha)}\mathscr{L}\left\{\frac{t^{\alpha-1}}{t-\mu}\right\}(\beta) =\beta e^{-\beta\mu}\,\Re\left\{E_\alpha(-\beta\mu)\right\}, \end{equation} where $E_\nu(z)$ is the generalized exponential integral. I have explicit solution for the real part which must be separated into cases for integer and non-integer $\alpha$. Could this solution have been easily reached with contour integrals? After ending up at such a simple solution, I feel like I solved this the hard way and there is a much easier way to do it.

$\endgroup$
  • 1
    $\begingroup$ As stated (I suspect something wrong) since the only term with x in the integral is $e^{-\beta x}$ so the integral is $\frac{1}{\beta}\frac{t^{\alpha -1}}{t-\mu}$. $\endgroup$ – herb steinberg Jul 7 '18 at 15:42
  • $\begingroup$ @herbsteinberg It was a typo ($x$'s should have been $t$'s). Fixed now. Thank you. $\endgroup$ – Aaron Hendrickson Jul 7 '18 at 15:45
  • $\begingroup$ Is the central problem the evaluation of the integral? $\endgroup$ – Szeto Jul 7 '18 at 16:42
  • 1
    $\begingroup$ Differentiation under the integral sign + recognizing gamma function integral form. Unfortunately, I don’t have the time to write a full answer soon. $\endgroup$ – Szeto Jul 7 '18 at 16:54
  • 1
    $\begingroup$ The Laplace transform is $$\mathcal L_{t \to \beta}\!\left[ \frac {t^{\alpha-1}} {t-\mu} \right] = e^{-\mu \beta} \mu^{\alpha-1}( e^{-i \pi (\alpha-1) \operatorname{sgn} \operatorname{Im} \beta} \, \Gamma(\alpha) \Gamma(1-\alpha, -\mu \beta) - i \pi \operatorname{sgn} \operatorname{Im} \beta).$$ The function constructed in this way and defined by continuity on the real axis is an analytic function of $\beta$. The reason for the difficulties might be that we're working with those branches of $\Gamma(a, z)$ or $E_n(z)$ which have the branch cut at $z \leq 0$. $\endgroup$ – Maxim Jul 10 '18 at 20:12
1
$\begingroup$

This may be somewhat simpler. Let $0 < \alpha < 1, \operatorname{Im} \mu \neq 0, 0 < \beta$ (so it may be a bit misleading to refer to the integral as the Laplace transform). We have $$\int_0^\infty \frac d {d\beta} \left( e^{\mu \beta} \frac {t^{\alpha - 1}} {t - \mu} e^{-\beta t} \right) dt = -\Gamma(\alpha) \beta^{-\alpha} e^{\mu \beta}, \\ \int_0^\infty \frac {t^{\alpha-1}} {t-\mu} e^{-\beta t} dt = \underbrace {\Gamma(\alpha) (-\mu)^{\alpha-1} e^{-\mu \beta} \Gamma(1-\alpha, -\mu \beta)}_{=I(\alpha, \mu, \beta)} + C(\alpha, \mu).$$ The integral and $I(\alpha, \mu, \beta)$ are continuous at $\beta=0$, and $$\int_0^\infty \frac {t^{\alpha-1}} {t-\mu} dt = \pi (-\mu)^{\alpha-1} \csc \pi \alpha = I(\alpha, \mu, 0), $$ from which we can conclude that $C(\alpha, \mu) = 0$.

For positive $\mu$, the p.v. integral will be equal to $I(\alpha, \mu - i0, \beta)$ plus $i \pi$ times the residue of the integrand at $t=\mu$. $\Gamma(a, z)$ is continuous from above at negative $z$, thus $I(\alpha, \mu, \beta)$ is continuous from below: $I(\alpha, \mu - i0, \beta) = I(\alpha, \mu, \beta)$, therefore $$\operatorname{v.\!p.} \int_0^\infty \frac {t^{\alpha-1}} {t-\mu} e^{-\beta t} dt = I(\alpha, \mu, \beta) + i \pi \mu^{\alpha-1} e^{-\mu \beta}.$$ The result extends to $1 \leq \alpha$ by analytic continuation.

$\endgroup$
  • $\begingroup$ Do you have a reference for $\int_0^\infty\frac{t^{\alpha-1}}{t-\mu}\,\mathrm dt=\cdots$ or is that from mathematica? $\endgroup$ – Aaron Hendrickson Jul 14 '18 at 18:34
  • 1
    $\begingroup$ Take $(-t)^\alpha/(t-\mu)$ instead; integrating over the upper bank of the branch cut, then over a large circle, then over the lower bank gives $$e^{-i \pi \alpha} \int_0^\infty \frac {t^\alpha} {t-\mu} dt + e^{i \pi \alpha} \int_\infty^0 \frac {t^\alpha} {t-\mu} dt = 2 \pi i \operatorname{Res}_{t=\mu} \frac {(-t)^\alpha} {t-\mu}.$$ $\endgroup$ – Maxim Jul 15 '18 at 0:46
1
$\begingroup$

Hint:

$$I=\int^\infty_0\frac{t^{a-1}}{t-u}e^{-bt}dt=e^{-bu}\underbrace{\int^\infty_0\frac{t^{a-1}}{t-u}e^{-b(t-u)}dt}_{=J(b)}$$

$$J’=-\int^\infty_0t^{a-1}e^{-b(t-u)}dt=-\frac{e^{bu}}{b^a}\Gamma(a)$$

$\endgroup$
1
$\begingroup$

Here is my attempt at the solution. I cannot seem to solve for the constant $c_1$ that comes from the solution of the differential equation. I have solved this problem using an entirely different approach and get the same answer as here if I set $c_1=0$....Just not sure how to get it.


We wish to evaluate the following principal value integral \begin{equation} \Lambda=\frac{\beta^\alpha}{\Gamma(\alpha)}\operatorname{PV}\int_{0}^{\infty}\frac{x^{\alpha-1}}{x-\mu}e^{-\beta x}\,\mathrm{d}x. \end{equation} Begin by considering the contour integral \begin{equation} I=\oint_C\frac{z^{\alpha-1}}{z-\mu}e^{-\beta z}\,\mathrm{d}z, \end{equation} where $C$ is a wedge of angle $\pi/4$ of radius $R$ in the upper right quadrant, with a semicircular indentation of radius $\varepsilon$ into the upper half plane at $z=\mu$. The contour integral is equal to \begin{multline} I=% \int_0^{\mu-\varepsilon}\frac{x^{\alpha-1}}{x-\mu}e^{-\beta x}\,\mathrm{d}x% +\mathrm i\int_\pi^{0}(\mu+\varepsilon e^{\mathrm i\varphi})^{\alpha-1}e^{-\beta(\mu+\varepsilon e^{\mathrm i\varphi})}\,\mathrm{d}\varphi% +\int_{\mu+\varepsilon}^{R}\frac{x^{\alpha-1}}{x-\mu}e^{-\beta x}\,\mathrm{d}x\\% +\mathrm i\int_0^{\pi/4}\frac{(R e^{\mathrm i\varphi})^{\alpha}}{Re^{\mathrm i\varphi}-\mu}e^{-\beta R e^{\mathrm i\varphi}}\,\mathrm{d}\varphi% +(e^{\mathrm i\pi/4})^\alpha\int_R^0\frac{t^{\alpha-1}}{e^{\mathrm i\pi/4}t-\mu}e^{-\beta e^{\mathrm i\pi/4}t}\,\mathrm{d}t% \end{multline} As $\varepsilon\to0$ and $R\to\infty$ the sum of the first and third integrals approaches the desired principal value and the fourth integral vanishes. Since no poles lie inside the contour, the integral $I$ is equal to zero and we find \begin{equation} \operatorname{PV}\int_0^{\infty}\frac{x^{\alpha-1}}{x-\mu}e^{-\beta x}\,\mathrm{d}x=% \mu^{\alpha-1}e^{-\beta\mu}\pi\mathrm i% +(e^{\mathrm i\pi/4})^\alpha\int_0^\infty\frac{t^{\alpha-1}}{e^{\mathrm i\pi/4}t-\mu}e^{-\beta e^{\mathrm i\pi/4}t}\,\mathrm{d}t. \end{equation} For the remaining integral we define \begin{equation} J(\beta)=% (e^{\mathrm i\pi/4})^\alpha e^{-\beta\mu}\int_0^\infty\frac{t^{\alpha-1}}{e^{\mathrm i\pi/4}t-\mu}e^{-(e^{\mathrm i\pi/4}t-\mu)\beta}\,\mathrm{d}t. \end{equation} Differentiating $J$ w.r.t. $\beta$ yields \begin{equation} J^\prime(\beta)=-\mu J(\beta)% -(e^{\mathrm i\pi/4})^\alpha \int_0^\infty t^{\alpha-1}e^{-\beta e^{\mathrm i\pi/4}t}\,\mathrm{d}t. \end{equation} Integrating and then rearranging terms yields the first order equation \begin{equation} J^\prime(\beta)+\mu J(\beta)=-\frac{\Gamma(\alpha)}{\beta^\alpha}. \end{equation} Solving for $J$ one finds \begin{equation} J(\beta)=% -e^{-\beta\mu}\int\frac{e^{\beta\mu}}{\beta^\alpha}\,\mathrm d\beta+c_1e^{-\beta\mu}, \end{equation} where the constant $c_1$ needs to be determined. The integral is evaluated with formula $2.33.10$ in Gradshteyn & Ryzhik to get \begin{equation} J(\beta)=% e^{-\beta\mu}(-\mu)^{\alpha-1}\Gamma(1-\alpha,-\beta\mu)+c_1e^{-\beta\mu}. \end{equation} Using the definition of the generalized exponential function $E_\nu(z)= z^{\nu-1}\Gamma(1-\nu,z)$ we write \begin{equation} J(\beta)=% \Gamma(\alpha)\beta^{1-\alpha}e^{-\beta\mu} E_\alpha(-\beta\mu)+c_1e^{-\beta\mu}. \end{equation}

suppose $c_1=0$, then

\begin{equation} \operatorname{PV}\int_0^{\infty}\frac{x^{\alpha-1}}{x-\mu}e^{-\beta x}\,\mathrm{d}x=% \mu^{\alpha-1}e^{-\beta\mu}\pi\mathrm i% +\Gamma(\alpha)\beta^{1-\alpha}e^{-\beta\mu} E_\alpha(-\beta\mu). \end{equation} It is then obvious that \begin{equation} \Lambda=\beta e^{-\beta\mu}\left(\frac{(\beta\mu)^{\alpha-1}}{\Gamma(\alpha)}\pi\mathrm i% +E_\alpha(-\beta\mu)\right). \end{equation} Using the series representation of the generalized exponential integral one can show \begin{equation} \Im\{E_\alpha(-\beta\mu)\}=-\frac{(\beta\mu)^{\alpha-1}}{\Gamma(\alpha)}\pi. \end{equation} This leads to the more compact expression \begin{equation} \Lambda=\beta e^{-\beta\mu}\Re\{E_\alpha(-\beta\mu)\}. \end{equation}

$\endgroup$
  • $\begingroup$ Can you draw your contour? $\endgroup$ – Szeto Jul 12 '18 at 0:05
  • $\begingroup$ @Szeto see the contour at math.stackexchange.com/q/247866/271783 Set $n=8$ and deform the contour along the real axis into tje first quadrant at $x=\mu$. I can post a drawing tomorrow. $\endgroup$ – Aaron Hendrickson Jul 12 '18 at 0:14

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.