In my course, I have to prove formula below $$I=\int_0^\infty \frac{e^{\cos(ax)}\cos\left(\sin (ax)+bx\right)}{c^2+x^2}dx =\frac{\pi}{2c}\exp\left(e^{-ac}-bc\right)$$ for $a,b,c>0.$

I know that this integral can be easily solved with complex analysis using $$f(z)=\frac{1}{2} \ \mathbb{R} \left(\int_{-\infty}^\infty \frac{\exp\left(e^{iaz}+ibz\right)}{c^2+z^2}dz\right)$$ but right now I am in a course dealing with real analysis. I tried to use parametrization integral method $$I'(a)=-\int_0^\infty \frac{xe^{\cos(ax)}\sin(\sin(ax)+(a+b)x)}{c^2+x^2}dx $$ but it doesn't look easier to handle. I tried to differentiate it again, but I just got a horrible form. An idea came to mind to differentiate with respect to parameter $b$ and set a differential equation $$I''(b)+x^2I(b)=0$$ plugging this ODE to W|A, I got $$I(b)=c_1\cos(bx^2)+c_2\sin(bx^2)$$ It's definitely wrong! After seeing Samrat's answer, I tried to plug in again to W|A and I got $$I(b)=c_1 D_{-1/2}((i+1)b)+c_2 D_{-1/2}((i-1)b)$$

where $D_n(z)$ is the parabolic cylinder function but I have no idea what does that mean.

Any idea? Thanks in advance.

  • $\begingroup$ The result seems off. What if $b \lt 0$? $\endgroup$ – Ron Gordon Nov 5 '14 at 15:59
  • $\begingroup$ Sorry, I forgot to mention the formula holds for $a,b,c>0$ $\endgroup$ – Venus Nov 5 '14 at 16:00
  • 1
    $\begingroup$ Thanks. That's a nice detail to know up front. $\endgroup$ – Ron Gordon Nov 5 '14 at 16:11

$$\begin{aligned} \int_0^{\infty} \frac{e^{\cos(ax)}\cos(\sin(ax)+bx)}{x^2+c^2}\,dx &=\Re\left(\int_0^{\infty} \frac{e^{e^{iax}}e^{ibx}}{x^2+c^2}\right) \\ &=\Re\left(\sum_{k=0}^{\infty} \frac{1}{k!}\int_0^{\infty} \frac{e^{i(ak+b)x}}{x^2+c^2}\,dx\right)\\ &=\sum_{k=0}^{\infty} \frac{1}{k!}\int_0^{\infty}\frac{\cos((ak+b)x)}{x^2+c^2}\,dx \\ &=\frac{\pi}{2c}\sum_{k=0}^{\infty} \frac{1}{k!}e^{-c(ak+b)}\\ &=\frac{\pi}{2c}e^{-bc}\sum_{k=0}^{\infty} \frac{e^{-kac}}{k!}\\ &=\frac{\pi}{2c}e^{-bc}e^{e^{-ac}}=\boxed{\dfrac{\pi}{2c}\exp\left(e^{-ac}-bc\right)} \\ \end{aligned}$$

I used the following result: $$\int_0^{\infty} \frac{\cos(mx)}{x^2+a^2}=\frac{\pi}{2a}e^{-am}$$

  • $\begingroup$ Wow! It's very simple. Thanks Pranav! You help me again :-) $\endgroup$ – Venus Nov 6 '14 at 9:22
  • 1
    $\begingroup$ @Venus: Glad to help! :) $\endgroup$ – Pranav Arora Nov 6 '14 at 16:14
  • $\begingroup$ What is the fancy R? $\endgroup$ – Joao Nov 7 '14 at 1:26

As a generalization of Pranav's answer, let us assume that $f(z)$ has a Maclaurin series expansion with real coefficients that converges absolutely on the unit circle on the complex plane.

Then for $a, b, c >0$,

$$ \begin{align} \text{Re} \int_{0}^{\infty} \frac{e^{ibx} f(e^{i a x}) }{c^{2}+x^{2}} \ dx &= \text{Re} \int_{0}^{\infty} \frac{e^{ibx}}{c^{2}+x^{2}} \sum_{n=0}^{\infty} \frac{f^{(n)}(0)}{n!} e^{ianx}\ dx \\ &= \text{Re} \sum_{n=0}^{\infty} \frac{f^{(n)}(0)}{n!} \int_{0}^{\infty} \frac{e^{i(an+b)x}}{c^{2}+x^{2}} \ dx \tag{1} \\ &= \sum_{n=0}^{\infty} \frac{f^{(n)}(0)}{n!} \int_{0}^{\infty} \frac{\cos[(an+b)x]}{c^{2}+x^{2}} \ dx \\ &= \sum_{n=0}^{\infty} \frac{f^{(n)}(0)}{n!} \frac{\pi}{2c} e^{-c(an+b)} \\ &= \frac{\pi}{2c} e^{-bc} \sum_{n=0}^{\infty} \frac{f^{(n)}(0)}{n!} e^{-acn} \\ &= \frac{\pi}{2c} e^{-bc} f(e^{-ac}). \end{align} $$

Your integral is the case $f(z) = e^{z}$.

$ $

$(1)$ When can a sum and integral be interchanged?

  • $\begingroup$ Thanks for your answer, +1. I learned much from your posts here. Thank you so much! :-) $\endgroup$ – Venus Nov 7 '14 at 10:36
  • 1
    $\begingroup$ @Venus You're very welcome. $\endgroup$ – Random Variable Nov 7 '14 at 11:37
  • $\begingroup$ RV What did you mean by "the Taylor expansion converges absolutely for values on the unit circle"? Could you give me examples of $f(z)$ that satisfy the given condition? $\endgroup$ – Venus Nov 25 '14 at 11:13
  • $\begingroup$ @Venus That the radius of convergence of the Taylor series is exactly $1$ or greater. Otherwise I wouldn't be able to replace the complex variable $z$ by $e^{iax}, x \in \mathbb{R}$. Functions that satisfy this condition include $e^{z}, \cos(z), \sin(z)$, and $z^{n}$ where $n$ is a positive integer. An example of a function that doesn't is $\frac{1}{\sqrt{1-4z}}$. $\endgroup$ – Random Variable Nov 25 '14 at 12:06
  • $\begingroup$ How about $\displaystyle\frac{1}{1\pm z}$ or $\displaystyle\frac{e^z}{z}$? $\endgroup$ – Venus Nov 25 '14 at 12:37

Note that $$I''(b)=-\int_{0}^{\infty}\frac{x^2e^{\cos ax}\cos(\sin (ax)+bx) }{c^2+x^2}dx\ne -x^2I(b)$$ That is your mistake, otherwise, the approach is fine.

  • $\begingroup$ I did say this was wrong in my question $\endgroup$ – Venus Nov 5 '14 at 15:56
  • $\begingroup$ Yeah, I was just showing you where you made the mistake. Your idea is not wrong, but taking out $x^2$ is wrong. If you proceed now, I think you will be able to get the answer. $\endgroup$ – Samrat Mukhopadhyay Nov 5 '14 at 15:57
  • $\begingroup$ I got this wolframalpha.com/input/?i=y%27%27+%2Bx^2+y%3D0 $\endgroup$ – Venus Nov 5 '14 at 15:58

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.