3

The problem is simple. But the solution is where I get my doubts. Also I would appreciate if anyone could fill me in with all the necessary points where I need to check for convergation for the procedure to correct.

$$\int_0^\infty \frac{\cos(bx)-\cos(cx)}x e^{-ax} \,dx $$ for all $a,b,c \in \rm I\!R ; a>0$

From here it's obvious that we can rewrite this as $$\frac d{dt} \left.\int_0^\infty \frac{cos(tx)}x e^{-ax}\right|_c^b \,dx =\int_0^\infty \int_c^b {\sin(tx)} e^{-ax} \,dx \,dt= \int_c^b I\,dt$$

Then via double per partes $$I=\int_0^\infty {\sin(tx)} e^{-ax}\,dx = \left.\frac{-1}{a}\sin(tx)e^{-ax}\right|_{t=0}^\infty + \frac ta \int_0^\infty {cos(tx)} e^{-ax}dx =$$ $$=\left. \frac{-\sin(tx)e^{-ax}}a \right|_{t=0}^\infty - \left. \frac t a {\cos(tx)e^{-ax}}\right|_{t=0}^\infty - \frac{t^2}{a^2}\int_0^\infty {\sin(tx)} e^{-ax}\,dx$$ $$I=\frac ta-\frac{t^2}{a^2}I$$ $$I=\frac{\frac ta}{1+\frac{t^2}{a^2}}$$ $$\int_c^b\frac{\frac ta}{1+\frac{t^2}{a^2}}=\int_{c/a}^{b/a}\frac{u}{1+u^2}=\frac12\ln(\frac dc)$$

I find it odd that my final result is not dependant at all of parameter $a$ Just does not sound right.

And I would appreciate filling out where covnergation or any other possible breaking points need to be checked.

EDIT: A typo...probably more to come.

2 Answers2

4

In the OP, there was a flaw in the beginning of the development. The integral of interest is $I(a)=\int_0^\infty \frac{\cos(bx)-\cos(cx)}{x}e^{-ax}\,dx$.

But, $I(a)$ is not equal to $\left.\left(\frac{d}{dt}\int_0^\infty \frac{\cos(tx)}{x}e^{-ax}\,dx\right)\right|_{c}^{b}$. In fact, this latter integral fails to exist due to the $\frac1x$ singularity at $x=0$.

Herein, we present two methodologies that we can use to evaluate $I(a)$.


NOTE: Since the integral is even in $b$ and $c$, we may assume without loss of generality that $b\ge 0$ and $c \ge 0$.


METHODOLOGY $(1)$: Differentiating Under the Integral Sign

Here is a straightforward approach. Let $I(a)$ be given by

$$I(a)=\int_0^\infty \frac{\cos(bx)-\cos(cx)}{x}e^{-ax}\,dx$$

Differentiating reveals

$$I'(a)=\int_0^\infty e^{-ax}\left(\cos(cx)-\cos(bx)\right)\,dx=\frac{a}{a^2+c^2}-\frac{a}{a^2+b^2}$$

Integrating $I'(a)$, we obtain

$$I(a)-I(0)=\int_0^a \left(\frac{t}{t^2+c^2}-\frac{t}{t^2+b^2}\right)\,dt=\frac12\log\left(\frac{a^2+c^2}{a^2+b^2}\right)-\log(c/b)$$

Using a slightly modified version of Frullani's Theorem, we find that $I(0)=\log(c/b)$. Therefore,

$$\bbox[5px,border:2px solid #C0A000]{\int_0^\infty \frac{\cos(bx)-\cos(cx)}{x}e^{-ax}\,dx=\frac12\log\left(\frac{a^2+c^2}{a^2+b^2}\right)}$$


METHODOLOGY $(2)$: Applying Frullani's Theorem for Complex Parameters

In THIS ANSWER, I showed that if $f(z)$ is analytic, then the following generalization of Frullani's Theorem holds for complex parameters $a$ and $b$:

$$\begin{align} \int_0^\infty \frac{f(ax)-f(bx)}{x}\,dx&=f(0)\log(|b/a|)\\\\ &+if(0)\left(\arctan\left(\frac{\text{Re}(a\bar b)-|a|^2}{\text{Im}(a\bar b)}\right)-\arctan\left(\frac{|b|^2-\text{Re}(a\bar b)}{\text{Im}(a\bar b)}\right)\right) \tag 1 \end{align}$$

Then, we can write

$$I(a)=\text{Re}\left(\int_0^\infty \frac{e^{-(a-ib)x}-e^{-(a-ic)x}}{x}\,dx\right)$$

Applying $(1)$ with $f(z)=e^{-z}$, $a\to a-ib$ and $b\to a-ic$ yields

$$\begin{align} I(a)&=e^{-(0)}\log\left(\frac{\sqrt{a^2+c^2}}{\sqrt{a^2+b^2}}\right)\\\\ &=\bbox[5px,border:2px solid #C0A000]{\frac12\log\left(\frac{a^2+c^2}{a^2+b^2}\right)} \end{align}$$

as expected!

Mark Viola
  • 179,405
  • Hi Dr.MV. in your first approach i would fix the constant by noting that $I(a=\infty)=0$ then you won't need to refer to an additional theorem. Otherwise, good work (+1)! – tired Nov 24 '16 at 03:46
  • @tired I had set the lower limit to $\infty$, but then thought that it might be useful to introduce Frullani's Theorem (standard one) since it is often neglected. Also, I didn't want to explain why we can interchange the limit as $a\to \infty$ with the integral. And thank you for the up vote!! Hope you're enjoying the holidays. – Mark Viola Nov 24 '16 at 04:22
  • do you mean thanksgiving? here where i live this is sadly not a holiday :/. But anyway, i hope you do and the turkey is tasty ;) – tired Nov 24 '16 at 09:36
  • Thanks for the help. Had to sleep over it to figure out where my process diverges from yours since it shuld give the same solution, although mine seems needlessly longer now that i see yours.

    On the last step I used another substitution of $ž=1+u^2$ but didnt adjust the integration interval correctly (what is the proper english term for this anyway...we call them integration limits)

    I still need to prove the initial integral exists on $x=0$ and $\infty$ and is uniformly converging, right? (<- would also like to know the proper english term for this)

    – Katpton Liamfuppinshire Nov 24 '16 at 10:02
  • PS: As a side note, how do you colour highlight text in markdown? – Katpton Liamfuppinshire Nov 24 '16 at 10:02
  • You're welcome. My pleasure. Youe English usage is fine (integration limits, and uniform convergence). Uniform convergence is easy to show for $a\ge \delta >0$. The result follows for $a>0$. Of course uniform convergence is a sufficient condition, not a necessary one. – Mark Viola Nov 24 '16 at 16:53
0

I get:

$\frac {d}{dt}\int_0^\infty \frac{\cos(tx)}x e^{-ax}|_c^b dx \\ \int_0^\infty \int_c^b {\sin(tx)} e^{-ax} dt dx\\ \int_c^b\int_0^\infty {\sin(tx)} e^{-ax} dxdt\\ \int_c^b \frac {t}{a^2+t^2} dt\\ \frac 12 (\ln(a^2 + b^2) - \ln(a^2 + c^2))$

Doug M
  • 57,877
  • There are three issues. First, the integral in the first line diverges. Second, we need to justify changing the order of integration from the second line to the third since the integrand is not absolutely convergent. Third, the inner integral is $\frac{t}{a^2+t^2}\ne \frac1{a^2+t^2}$. – Mark Viola Nov 24 '16 at 04:26
  • Thanks for pointing out that the integral should be $\frac t{a^2 + t^2}$ and not $\frac 1{a^2 + t^2}$. Fixing that, I get the same result as you posted (Even if some of my manipulations were not rigorous.) – Doug M Nov 24 '16 at 04:35
  • The interchange can be justified, but it does take a little bit of work. But that first integral in Line 1 is nonsense. I suggest deleting it. And Happy Holidays Doug. -Mark – Mark Viola Nov 24 '16 at 04:36