$$\tag{1} I_n = PV \int_{-\infty}^\infty \frac{x^n ~e^{-ax^2} dx}{(x-x_1)(x-x_2)},~ (n=0,1,2,3,...)~\text{with}~a>0~\text{and}~ x_1~x_2 \in \mathbb{R},$$ Is the principal value of $I_1$ calculable?

Since $x_1$ and $x_2$ are on the path of integration on the real axis, I was thinking of indenting the contour, passing over them clockwise picking up

$$\tag{2}-\pi i[\mathrm{Res}(f,x_1)+\mathrm{Res}(f,x_2)]$$

where $f$ is the integrand. To use the residue theorem I close the contour along a semi circle, $\Gamma_\infty$, in the upper half plane. But I suspect that the contribution of the integral $$\tag{3}\int_{\Gamma_\infty} dz ~f(z)$$ is non-zero. Since on $\Gamma_\infty$ we have $z = Re^{i\theta}$.

Assuming this is correct, one finds $$\tag{4} I_n = i\pi [\mathrm{Res}(f,x_1)+\mathrm{Res}(f,x_2)]-\int_{\Gamma_\infty} dz ~f(z). $$ Does all this make sense? And is there a simple estimate/computation of (3) above?

(Note: Mathematica can do the integral (1) in terms of incomplete Gamma functions, but the total expression seems to be indeterminate at integer values of $n$.)

  • $\begingroup$ Not the Residue, it should be semi-circles around that two poles, with angle goes from $\pi$ to $0$. $\endgroup$ – JRen Jul 5 '19 at 20:00

$\newcommand{\bbx}[1]{\,\bbox[15px,border:1px groove navy]{\displaystyle{#1}}\,} \newcommand{\braces}[1]{\left\lbrace\,{#1}\,\right\rbrace} \newcommand{\bracks}[1]{\left\lbrack\,{#1}\,\right\rbrack} \newcommand{\dd}{\mathrm{d}} \newcommand{\ds}[1]{\displaystyle{#1}} \newcommand{\expo}[1]{\,\mathrm{e}^{#1}\,} \newcommand{\ic}{\mathrm{i}} \newcommand{\mc}[1]{\mathcal{#1}} \newcommand{\mrm}[1]{\mathrm{#1}} \newcommand{\pars}[1]{\left(\,{#1}\,\right)} \newcommand{\partiald}[3][]{\frac{\partial^{#1} #2}{\partial #3^{#1}}} \newcommand{\root}[2][]{\,\sqrt[#1]{\,{#2}\,}\,} \newcommand{\totald}[3][]{\frac{\mathrm{d}^{#1} #2}{\mathrm{d} #3^{#1}}} \newcommand{\verts}[1]{\left\vert\,{#1}\,\right\vert}$


\begin{align} &\mbox{With}\ \pars{~a > 0\,;\ x_{1},x_{2} \in \mathbb{R}\ \mbox{and}\ n \in \mathbb{N}_{\geq 0}~}, \,\,\,\mbox{note that} \\ & I_{n} = \mrm{P.V.}\int_{-\infty}^{\infty}{x^{n}\expo{-ax^{2}} \over \pars{x - x_{1}}\pars{x - x_{2}}}\,\dd x = {1 \over x_{2} - x_{1}}\,\sum_{k = 1}^{2}\pars{-1}^{k}\, \bbox[#fee,5px]{\ds{\mrm{P.V.} \int_{-\infty}^{\infty}{x^{n}\expo{-ax^{2}} \over x - x_{k}}\,\dd x}}\label{1}\tag{1} \end{align}

\begin{align} &\bbox[#fee,5px]{\ds{\mrm{P.V.} \int_{-\infty}^{\infty}{x^{n}\expo{-ax^{2}} \over x - x_{k}}\,\dd x}} = \mrm{P.V.}\int_{-\infty}^{\infty}{% \pars{x + x_{k}}^{n}\expo{\large -a\pars{x + x_{k}}^{2}} \over x}\,\dd x \\[5mm] = &\ \int_{0}^{\infty}{% \pars{x + x_{k}}^{n}\expo{\large -a\pars{x + x_{k}}^{2}} - \pars{-x + x_{k}}^{n}\expo{\large -a\pars{-x + x_{k}}^{2}}\over x}\,\dd x \\[5mm] = &\ \int_{0}^{\infty}{% \pars{x + x_{k}}^{n}\expo{\large -a\pars{x + x_{k}}^{2}} - \pars{-1}^{n}\pars{x - x_{k}}^{n}\expo{\large -a\pars{x - x_{k}}^{2}} \over x}\,\dd x \end{align}

  • $\begingroup$ The pv integral in red, if one indent the contour over the pole at $x_k$, it looks finite and the epsilon semi circle goes to zero (like $\epsilon ^n$). While the final version not so much as it looks like $ 1/\epsilon $ $\endgroup$ – Your Majesty Apr 18 '18 at 23:07
  • $\begingroup$ Also at large radi $R$ it goes like $e^{-R^2}$ so it should die off fast enough right? $\endgroup$ – Your Majesty Apr 20 '18 at 13:29
  • $\begingroup$ @JustAsk Yes, that's true. The exponential term provides the convergence as $x \to \infty$. $\endgroup$ – Felix Marin Apr 20 '18 at 18:17
  • $\begingroup$ Actually I'm not so sure that the arc (at infinity) contribution is so trivial. On the arc, the Gaussian $e^{-z^2}$ becomes proportional to $e^{-i R^2 \sin(2\phi)}$ also. For instance, $\lim_{\epsilon\rightarrow 0 }\int_{-\infty}^\infty dx\, \frac{e^{-x^2}}{x-1+i \epsilon} = -\frac{\pi (erfi(1)+i)}{e}$ while jumping over the pole ($-i \pi \mathrm{Res}(x=1-i \epsilon)$) and taking the $\epsilon \rightarrow 0 $ limit gives $-\frac{i \pi }{e}$. So something's missing. And I suspect the rest comes from the arc at infinity. Any thoughts? Note: Erfi(z) = erf(i z)/i $\endgroup$ – Your Majesty Apr 23 '18 at 0:08
  • $\begingroup$ See for instance math.stackexchange.com/q/2749560/110467 $\endgroup$ – Your Majesty Apr 23 '18 at 1:23

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.