6

I want to evaluate the integral

$\int_{-\infty}^\infty \frac{\cos{(kx)}}{\sqrt{x^2+a^2}}dx$,

where $k$ and $a$ are constants. Clearly the integrand has two branch points at $\pm ia$. Let $f(z)=Re (\frac{e^{ikz}}{\sqrt{z^2+a^2}})$. Considering the upper half plane, let us integrate along the contour $\Gamma$, which is $AC_{R}BCDC_{\epsilon}EFA$ (see the plot). Plot of the contour over which the integration will be performed Here, $C_R$ is the semicircle of radius $R$ and $C_\epsilon$ is the circle of radius $\epsilon$. Now by Cauchy's integral theorem, \begin{eqnarray} \int_\Gamma f(z)dz=0 \nonumber\\ \left(\int_{AC_RB}+\int_{BC}+\int_{CD}+\int_{DC_\epsilon E}+\int_{EF}+\int_{FA}\right) f(z)dz=0 \end{eqnarray} Let us now take the limit as $R \to \infty$, $\epsilon \to 0$, then first and fourth integrals are zero. Second and sixth integrals will give $2 \int_0^\infty \frac{\cos{kx}}{\sqrt{x^2+a^2}}dx$. So, now we need to integrate the third and fifth integrals whose values will contribute in this integration. Here, I'm stuck. I don't understand what will be the value of z in $CD$ or $EF$ path.

  • Hint: Have a look at the Bessel function $K_0$ and its properties. – mickep May 01 '17 at 19:28
  • One parameter between $k$ and $a$ can be removed through a substitution. The Fourier cosine transform of $\frac{1}{\sqrt{1+x^2}}$ remains. Have a look at $K_0$ as suggested by mickep. – Jack D'Aurizio May 01 '17 at 19:29
  • 1
    Here is an elegant solution for $a=1$ (integral is from $0$ to $1$, though): https://math.stackexchange.com/questions/1567441/how-to-evaluate-int-01-costx-over-sqrt1x2dx – mlc May 01 '17 at 19:32
  • Once you have the solution for $a=1$ as in the link by @mlc, you can change the variable to obtain the answer for yours. – Sungjin Kim May 01 '17 at 19:35
  • 1
    Okay, and what's your question? –  May 01 '17 at 19:35
  • mickep and @Jack D'Aurizio, Thanks a lot for the suggestion. I'll look on the properties of Bessel function. – Anindita Bera May 03 '17 at 09:21
  • mic, thanks for sending me the link. Now I'm following the link and trying to solve this which has also been suggested by @i707107. – Anindita Bera May 03 '17 at 09:25
  • @AninditaBera Anindita, in the contour that you used, and with the branch cuts chosen as they are, the term $\sqrt{x^2+a^2}$ is of opposite signs on the positive and negative real axes. The integrals on the contours $EF$ and $CD$ are purely imaginary, so upon taking the real parts, contribute $0$. So, what you end up showing, is $0=0$. – Mark Viola May 03 '17 at 14:39

1 Answers1

3

Let $F(k,a)$ be given by

$$F(k,a)=\int_{-\infty}^\infty \frac{\cos(kx)}{\sqrt{x^2+a^2}}\,dx$$

Exploiting the even symmetry of the integrand and enforcing the substitution $x\to |a|x$, where we assume that $a\in \mathbb{R}$ yields

$$\begin{align} F(k,a)&=2\int_0^\infty \frac{\cos(k|a| x)}{\sqrt{x^2+1}}\,dx\\\\ &=2\text{Re}\left(\int_0^\infty \frac{e^{ik|a| x}}{\sqrt{x^2+1}}\,dx\right)\tag 1\\\\ &=2\text{Re}\left(\int_0^\infty \frac{e^{-k|a| x}}{\sqrt{x^2-1}}\,dx\right)\tag 2\\\\ &=2\text{Re}\left(\underbrace{\int_0^1 \frac{e^{-k|a| x}}{\sqrt{x^2-1}}\,dx}_{\text{Purely Imaginary}}\right)+2\text{Re}\left(\underbrace{\int_1^\infty \frac{e^{-k|a| x}}{\sqrt{x^2-1}}\,dx}_{\text{Purely Real}}\right)\\\\ &=2\int_1^\infty \frac{e^{-k|a| x}}{\sqrt{x^2-1}}\,dx\\\\ &=2\int_0^\infty e^{-k|a| \cosh(x)}\,dx\\\\ &=2K_0(k|a|) \end{align}$$

where $K_0(x)$ is the modified Bessel Function of the Second Kind.


NOTES:

In arriving at $(1)$, we used $\cos(k|a|x)=\text{Re}(e^{ik|a|x})$.

In going from $(1)$ to $(2)$, we chose branch cuts from $i$ to $i\infty$ and from $-i$ to $-i\infty$. Then, we applied Cauchy's Integral Theorem and deformed the real-line contour from $0$ to $R$ to the contour in the first quadrant comprised of (i) the line segment from $0$ to $i(1-\epsilon)$ , (ii) the semi-circular contour centered at $i$ with radius $\epsilon$ from $i(1-\epsilon)$ to $i(1+\epsilon)$, (iii)the line segment from $i(1+\epsilon)$ to $iR$, and (iv) the quarter-circular arc from $iR$ to $R$. As $\epsilon \to 0$ and $R\to \infty$ the contributions from integrals around the semi-circle and the quarter circle vanish. What remains it given by $(2)$.


The OP edited the question and provided a specific contour over which to proceed. We now analyze the integral over the given contour.

First note that the OP assumed that $a>0$. We will proceed under that assumption.

If we choose to cut the plane with branch cuts from $ia$ to $-i\infty$ and $-ia$ to $-i\infty$, then on the positive real axis $\sqrt{z^2+a^2}=\sqrt{x^2+a^2}$, while on the negative real axis, $\sqrt{z^2+a^2}=-\sqrt{x^2+a^2}$. Hence, this reveals

$$\begin{align} \int_{\text{BC}}\frac{e^{ikz}}{\sqrt{z^2+a^2}}\,dz+\int_{\text{FA}}\frac{e^{ikz}}{\sqrt{z^2+a^2}}\,dz&=\int_{-R}^\epsilon\frac{e^{ikx}}{-\sqrt{x^2+a^2}}\,dx+\int_{\epsilon}^R\frac{e^{ikx}}{\sqrt{x^2+a^2}}\,dx\\\\ &=2i\int_\epsilon^R \frac{\sin(kx)}{\sqrt{x^2+a^2}}\,dx\tag 3 \end{align}$$

The sum of the integrals along $CD$ and $EF$ is given by

$$\begin{align} \int_{\text{CD}}\frac{e^{ikz}}{\sqrt{z^2+a^2}}\,dz+\int_{\text{DF}}\frac{e^{ikz}}{\sqrt{z^2+a^2}}\,dz&=\int_{0}^{a-\epsilon}\frac{e^{-ky}}{-\sqrt{-y^2+a^2}}\,i\,dy+\int_{a-\epsilon}^0\frac{e^{-ky}}{\sqrt{-y^2+a^2}}\,i\,dy\\\\ &=-2i\int_0^{a-\epsilon} \frac{e^{-ky}}{\sqrt{a^2-y^2}}\,dy\tag 4 \end{align}$$

Using $(3)$ and $(4)$, and letting $R\to \infty$ and $\epsilon\to 0$ reveals (after taking Imaginary Parts)

$$\begin{align} \int_0^\infty \frac{\sin(kx)}{\sqrt{x^2+a^2}}\,dx&=\int_0^a \frac{e^{-kx}}{\sqrt{a^2-x^2}}\,dx\\\\ &=\frac \pi 2\left(I_0(k|a|)+L_0(k|a|)\right) \end{align}$$

for $a>0$, where $I_0(x)$ and $L_0(x)$ are the Modified Bessel Function of the First Kind and Zero Order and the Modified Struve Function of Zero Order, respectively.

Note that the choice of contour as given in the OP, does not provide any insight regarding the integral of interest $\int_{-\infty}^\infty\frac{\cos(kx)}{\sqrt{x^2+a^2}}\,dx$.

Mark Viola
  • 179,405
  • 1
    I don't understand how you get from the second to the third line above: I assume it's a substitution $ix \rightarrow -x$, but that changes the integral to run along the negative imaginary axis, no? – NickD May 02 '17 at 05:13
  • Actually no. I used Cauchy's Integral Theorem an deformed the contour to the imaginary axis. The contribution from the integral over the quarter circle vanishes as the radius approaches $\infty$. – Mark Viola May 02 '17 at 12:52
  • You also have to deal with the branch points at $\pm i$. I presume you can exclude them by going around a small semicircle and that the contribution of that to the integral also tends to 0? – NickD May 02 '17 at 12:59
  • @Nick Yes, that contribution vanishes also as the radius of the exclusion semi-circle approaches $0$. – Mark Viola May 02 '17 at 15:36
  • @Dr.MV Thanks for your effort. I have rewritten the problem. In your calculation of line 2 to 3, please tell me which path you are integrating there as you have considered the Cauchy's integral Theorem. What is the substitution of $z$ in that path? Also from line 3 to 4, you have taken the substitution x by cosh(x). Then how can be the lower limit zero after the substitution? – Anindita Bera May 03 '17 at 09:11
  • @Nick Thank you very much. I have rewritten the problem. Please have a look on it. – Anindita Bera May 03 '17 at 09:15
  • @AninditaBera Yes, I had a typographical error that has been edited. In the edit, I've added notes to explain the contour deformation. – Mark Viola May 03 '17 at 14:02
  • @Dr.MV Thanks for the explanation. Now I understand that the contour that I'm taking here will give zero value of the integration. The way you have considered (this is the right way to solve this integration), the contour will give twice of modified Bessel function of second kind. It will be very helpful if you provide the plot of contour that you have mentioned in "note" form. – Anindita Bera May 03 '17 at 18:30
  • @AninditaBera I don't have a tool to plot the contour. I have given a detailed explanation in the "NOTES" section of the edit. -Mark – Mark Viola May 03 '17 at 18:43