6

For $\alpha>0$, prove\begin{align}\displaystyle\int_0^{\infty}xe^{-x}\cos(x)\cos\left(\dfrac{x^2}{\alpha}\right){\rm d}x=\dfrac{\alpha\sqrt{2\pi\alpha}}{8}e^{-\alpha/2} \end{align}

My attempt: Let $$I(b)=\displaystyle\int_0^{\infty}e^{-x}\sin(bx)\cos\left(\dfrac{x^2}{\alpha}\right){\rm d}x$$ Hence$$I'(1)=\displaystyle\int_0^{\infty}xe^{-x}\cos(x)\cos\left(\dfrac{x^2}{\alpha}\right){\rm d}x $$ But \begin{align}I(b) & =\displaystyle\int_0^{\infty}e^{-x}\sin(bx)\cos\left(\dfrac{x^2}{\alpha}\right){\rm d}x\\&=\dfrac{1}{2}\displaystyle\int_0^{\infty}e^{-x}\left(\sin(bx-x^2/\alpha)+\sin(bx+x^2\alpha)\right){\rm d}x\\&=I_1+I_2\end{align} But I have difficulty using contour integral to calculate $I_1$ or $I_2$ .

how to solve it using contour inregral? Is there a more efficient method to solve this problem? I have a thought that Fourier Transform is a possible method.

StubbornAtom
  • 17,932
  • You can brute force this. Break the sine functions up into complex exponentials, and you have four integrals to solve. Each has a result in terms of the Error Function, which cancels when you combine stuff – Brevan Ellefsen Oct 09 '19 at 06:38
  • 1
    To preserve sanity, define $c = \frac{b^2}{4a}$ and $d = \frac{b}{2a}$ \begin{align} &\int_0^{\infty}e^{-x}\sin(bx-ax^2) dx +\int_0^{\infty} e^{-x}\sin(bx+ax^2)dx \&=\int_0^{\infty}e^{-x}\left(\sin(a(x+d)^2-c) -\sin(a(x-d)^2-c)\right)dx \&= \frac{1}{2}\cos(c)\int_{0}^{\infty}e^{-x}\sin(a(x+d)^{2})dx-\frac{1}{2}\sin(c)\int_{0}^{\infty}e^{-x}\cos(a(x+d)^{2})dx \ &\qquad-\frac{1}{2}\cos(c)\int_{0}^{\infty}e^{-x}\sin\left(a(x-d)^{2}\right)dx+\frac{1}{2}\sin(c)\int_{0}^{\infty}e^{-x}\cos\left(a(x-d)^{2}\right)dx \end{align} – Brevan Ellefsen Oct 09 '19 at 09:09
  • \begin{align} \&= \frac{1}{2}\cos(c)e^d\int_{d}^{\infty}e^{-x}\sin(ax^{2})dx-\frac{1}2\sin(c)e^d\int_{d}^{\infty}e^{-x}\cos(ax^2)dx - \ & \qquad \frac{1}{2}\cos(c)e^{-d}\int_{-d}^{\infty}e^{-x}\sin\left(ax^2\right)dx+\frac{1}{2}\sin(c)e^{-d}\int_{-d}^{\infty}e^{-x}\cos(ax^2)dx \end{align} You can now use the Fresnel Integrals to finish with the right cancellations since the primitives of each integral are well known enough. Just break up each integral over bounds from $0$ to $\infty$, let the weird parts cancel, give the remaining pieces in terms of Fresnel Integrals, and cancel – Brevan Ellefsen Oct 09 '19 at 09:15
  • This really just obfuscates the nice cancellation we get from the complex exponentials though – Brevan Ellefsen Oct 09 '19 at 09:17

2 Answers2

4

I will write

$$ J_{\pm} = \int_{0}^{\infty} xe^{-x}\cos\left(x\pm\frac{x^2}{\alpha}\right)\,\mathrm{d}x $$

so that your integral takes the form $\frac{1}{2}(J_{+} + J_{-})$. Then

\begin{align*} J_{\pm} &= \operatorname{Re}\left[ \int_{0}^{\infty} x\exp\left( -x + ix \pm \frac{ix^2}{\alpha}\right) \,\mathrm{d}x. \right] \end{align*}

Now write $\mathbb{H}_{\text{right}} = \{ z \in \mathbb{C} : \operatorname{Re}(z) > 0 \}$. Then for each $a \in \mathbb{H}_{\text{right}}$, the map $z \mapsto \int_{0}^{\infty} x e^{-ax-zx^2} \, \mathrm{d}x$ is analytic on $\mathbb{H}_{\text{right}}$ and continuous on $\overline{\mathbb{H}_{\text{right}}}$. Moreover, if $a, z \in (0, \infty)$, then with $b = a^2/4z$,

\begin{align*} \int_{0}^{\infty} x e^{-ax-zx^2} \, \mathrm{d}x &= \int_{0}^{\infty} x \exp\bigg( -b \left(\frac{2z x}{a}+1\right)^2 + b \bigg) \, \mathrm{d}x \\ &= \frac{be^{b}}{z} \int_{1}^{\infty} (u-1) e^{-bu^2} \, \mathrm{d}u \\ &= \frac{be^{b}}{z} \left( \int_{1}^{\infty} u e^{-bu^2} \, \mathrm{d}u - \int_{0}^{\infty} e^{-bu^2} \, \mathrm{d}u + \int_{0}^{1} e^{-bu^2} \, \mathrm{d}u \right) \\ &= \frac{1}{2z} - \frac{\sqrt{\pi} a e^{b}}{4z^{3/2}} + \frac{be^{b}}{z} \int_{0}^{1} e^{-bu^2} \, \mathrm{d}u. \end{align*}

Then by the principle of analytic continuation, this holds for all $a \in \mathbb{H}_{\text{right}}$ and $z \in \overline{\mathbb{H}_{\text{right}}}$. Then plugging $a = 1-i$ and $z = z_{\pm} = \pm i/\alpha$, we get $b = b_{\pm} = \mp \alpha /2 \in \mathbb{R}$.

\begin{align*} J_{\pm} &= \operatorname{Re}\bigg[ \frac{1}{2z} - \frac{\sqrt{\pi} a e^{b}}{4z^{3/2}} + \frac{be^{b}}{z} \int_{0}^{1} e^{-bu^2} \, \mathrm{d}u \bigg] \\ &= -\frac{\sqrt{\pi} \, e^{b}}{4} \operatorname{Re}\bigg[ \frac{a}{z^{3/2}} \bigg], \end{align*}

By noting that $a_{+}/z_{+}^{3/2} = -\sqrt{2}\,\alpha^{3/2}$ and $a_{-}/z_{-}^{3/2} = i\sqrt{2}\,\alpha^{3/2}$, we get

$$ J_{+} = \frac{\sqrt{2\pi}}{4} \alpha^{3/2} e^{-\alpha/2}, \qquad J_{-} = 0. $$

This complete the proof.

Sangchul Lee
  • 181,930
2

Let $$I(\beta) =\int_{0}^{\infty} x e^{-x} \cos (x) \cos( \beta x^{2}) \, \mathrm dx \, , \quad \beta>0. $$

Let's take the Laplace transform of $I (\beta)$ and then switch the order of integration (which is permissible since the iterated integral converges absolutely).

$$ \begin{align}\mathcal{L} \{I(\beta)\} (s) &= \int_{0}^{\infty} \left(\int_{0}^{\infty} x e^{-x} \cos(x) \cos(\beta x^{2}) \, \mathrm dx \right)e^{-s \beta } \, \mathrm d \beta \\ &= \int_{0}^{\infty} x e^{-x} \cos(x) \int_{0}^{\infty}\cos(\beta x^{2}) e^{- s \beta } \, \mathrm d \beta \, \mathrm dx \\ &= s \int_{0}^{\infty} \frac{x e^{-x}\cos (x)}{x^{4}+s^{2}} \, \mathrm dx \tag{1} \end{align} $$

To evaluate $(1)$, we can integrate the complex function $$f(z) = \frac{z e^{-z}e^{iz}}{z^{4}+s^{2}} $$ around a wedge-shaped contour that makes an angle of $\frac{\pi}{2}$ with the positive real axis (i.e., a closed quarter-circle in the first quadrant of the complex plane).

In the first quadrant of the complex plane, $\vert e^{-z} \vert \le 1$ and $\vert e^{iz} \vert \le 1$. So the integral along the big arc clearly vanishes as the radius of the arc goes to $\infty$.

Integrating around the contour, we get $$ \begin{align} \int_{0}^{\infty} \frac{x e^{-x} e^{ix}}{x^{4}+s^{2}} \, \mathrm dx + \int_{\infty}^{0} \frac{(it)e^{-it} e^{-t}}{t^{4}+s^{2}} \, i \, \mathrm dt &= 2 \pi i \operatorname{Res} \left[f(z), e^{i \pi/4} \sqrt{s} \right] \\ &= 2 \pi i \lim_{z \to e^{i \pi/4} \sqrt{s}}\frac{ze^{-z}e^{iz}}{4z^{3}} \\ &= 2 \pi i \, \frac{e^{i \pi/4}\sqrt{s} \, e^{-\sqrt{2s}} }{4e^{3 \pi i/4}s^{3/2}} \\ &= \pi \, \frac{e^{-\sqrt{2s}}}{2s}. \end{align}$$

And if we equate the real parts on both sides of the equation, we get $$2 \int_{0}^{\infty} \frac{xe^{-x} \cos(x) }{x^{4}+s^{2} } \, \mathrm dx = \pi \, \frac{e^{-\sqrt{2s}}}{2s}.$$

Therefore, $$\mathcal{L} \{I(\beta)\} (s) = \frac{\pi e^{-\sqrt{2s}}}{4}. $$

Due to the uniqueness of the inverse Laplace transform , we only need to show that $$\mathcal{L} \left\{\frac{\sqrt{2 \pi}}{8} \beta^{-3/2} e^{-1/(2 \beta)} \right\} (s) = \frac{\pi e^{-\sqrt{2s}}}{4}$$ in order to prove that $$I(\beta) = \frac{\sqrt{2 \pi}}{8} \beta^{-3/2} e^{-1/(2 \beta)}.$$

From the answers here, we know that for $a,b>0$, $$\int_{0}^{\infty} \frac{\exp \left(-ax^{2}-b/x^{2}\right)}{x^{2}} \, \mathrm dx = - \frac{1}{2} \sqrt{\frac{\pi}{a}} \, \frac{\partial }{\partial b} e^{-2\sqrt{ab}} = \frac{1}{2} \sqrt{\frac{\pi}{b}} \, e^{-2\sqrt{ab}}.$$

(Differentiation under the integral sign is permissible since for any positive $c$ less than $b$, $\frac{\exp \left(-ax^{2}-b/x^{2}\right)}{x^{2}}$ is dominated by the integrable function $ \frac{\exp \left(-ax^{2}-c/x^{2}\right)}{x^{2}}$.)

Therefore, $$\begin{align} \mathcal{L} \left\{\frac{\sqrt{2 \pi}}{8} \beta^{-3/2} e^{-1/(2 \beta)} \right\} (s) &= \frac{\sqrt{2 \pi}}{8} \int_{0}^{\infty}\beta^{-3/2} e^{-1/(2 \beta)} e^{-s \beta } \, \mathrm d \beta \\ &= \frac{\sqrt{2 \pi}}{4} \int_{0}^{\infty} \frac{\exp \left(-s u^{2} - (1/2)/u^{2} \right)}{u^{2}} \mathrm du \\ &= \frac{\sqrt{2 \pi}}{4} \frac{\sqrt{2 \pi}}{2} \, e^{-\sqrt{2s}} \\ &= \frac{ \pi \, e^{-\sqrt{2s}}}{4}. \end{align}$$