24
$\begingroup$

Hi I am trying to calculate $$ I:=\int\limits_0^\infty \frac{\log^2 x \cos (ax)}{x^n-1}\mathrm dx,\quad \Re(n)>1, \, a\in \mathbb{R}. $$ Note if we set $a=0$ we get a similar integral given by $$ \int\limits_0^\infty \frac{\log^2 x }{x^n-1}\mathrm dx=-\frac{2\pi^3\cot(\pi/n)\csc^2(\pi/n)}{n^3},\quad \Re(n)>1. $$ I was trying to write I as $$ I=\Re \bigg[\int\limits_0^\infty \frac{ e^{i ax}\log^2 x}{x^n-1}\mathrm dx\bigg]=\Re\bigg[\int\limits_\infty^0\frac{e^{iax}\log^2 x}{1-x^n}\mathrm dx\bigg]=\Re\bigg[\int\limits_\infty^0e^{iax}\log^2 x\sum_{m=0}^\infty x^{nm} \mathrm dx\bigg]. $$ But was unsure of where to go from here. How can we calculate $I$? It is clear that this method is not going to work.

$\endgroup$
  • 1
    $\begingroup$ Where did you find this integral? $\endgroup$ – Mhenni Benghorbal Apr 24 '14 at 4:21
  • $\begingroup$ @MhenniBenghorbal I didn't "find" it anywhere, I am not sure what that even means though? $\endgroup$ – Jeff Faraci Apr 24 '14 at 4:24
  • 5
    $\begingroup$ I mean where it came from? Or you are trying to make a generalization. $\endgroup$ – Mhenni Benghorbal Apr 24 '14 at 4:26
  • $\begingroup$ @MhenniBenghorbal Oh, I am just trying to make a generalization my friend! I have not seen this in a paper, but I have seen these kinds of integrals growing up all the time $\endgroup$ – Jeff Faraci Apr 24 '14 at 4:37
  • 2
    $\begingroup$ @sidht Consider the well known Gaussian integral $$ \int_{-\infty}^\infty e^{-\alpha x^2 +\beta x}dx=\frac{\sqrt \pi}{\sqrt \alpha} e^{\beta^2/4\alpha},\quad \Re(\alpha)>0, $$ another integral with two parameters! This one is quite famous for obvious reason. It is good to get used to integrals with parameters $\endgroup$ – Jeff Faraci Apr 24 '14 at 16:39
2
$\begingroup$

This is not a complete solution.

First, we use a simple integral (we can get it by differentiating $x^b$ w.r.t. $b$ two times):

$$\int_0^1 x^b \ln^2 x ~d x=\frac{2}{(b+1)^3}$$

Using this integral, we also get:

$$\int_0^1 x^b \ln^2 x ~dx=\int^{\infty}_1 t^{-b-2} \ln^2 t~ dt=\frac{2}{(b+1)^3}$$

We separate the original integral into two parts:

$$ I=-\int\limits_0^1 \frac{\log^2 x \cos (ax)}{1-x^n} ~ dx+\int\limits_1^\infty \frac{\log^2 x \cos (ax)}{x^n-1} ~ dx=I_1+I_2 $$

Let's deal with the first one by using the modified integral from above:

$$\int_0^1 x^{pn} (a x)^{2q} \ln^2 x ~d x=\frac{2a^{2q}}{(pn+2q+1)^3}$$

Now we immediately get (using Taylor expansions for $\cos (ax)$ and $\frac{1}{1-x^n}$):

$$I_1=-2 \sum_{q=0}^{\infty} \frac{(-1)^q a^{2q}}{(2q)!} \sum_{p=0}^{\infty} \frac{1}{(pn+2q+1)^3}=\frac{1}{n^3} \sum_{q=0}^{\infty} \frac{(-1)^q a^{2q}}{(2q)!} \psi^{(2)} \left( \frac{2q+1}{n} \right)$$

Here $\psi^{(2)}$ is the second derivative of the digamma function.

This is a correct series solution for $I_1$, as we can see by comparing the two sides of the equation. See the plot $I_1(n)$ for $a=3$ below:

enter image description here

However, we can't use this method for $I_2$.

Let's see why (now we expand $\frac{1}{1-1/x^n}$):

$$\int^{\infty}_1 x^{-pn-n} (a x)^{2q} \ln^2 x~ dx=\frac{2a^{2q}}{(pn+n-2q-1)^3}$$

And now we see the problem. The denominator can become $0$ for infinitely many combinations of $p,q,n$, thus our integral would diverge badly.

Let's however, solve the case $a=0$, solution for which was provided in the OP ($n>1$).

$$I_2(a=0)=2 \sum_{p=0}^{\infty} \frac{1}{(pn+n-1)^3}=-\frac{1}{n^3} \psi^{(2)} \left(1- \frac{1}{n} \right)$$

Comparing with $I_1(a=0)$ and using the reflection formula for $\psi^{(2)}$:

$$\psi^{(2)}(1-x)-\psi^{(2)}(x)=2 \pi^3 \cot \left( \pi x \right) \csc^2 \left( \pi x \right)$$

we obtain:

$$I(a=0)=-\frac{2 \pi^3}{n^3} \cot \left( \pi \frac{2q+1}{n} \right) \csc^2 \left( \pi \frac{2q+1}{n} \right)$$


So, I'm not sure how to get the series solution for $I_2$, but I hope my answer helps you in some way.

| cite | improve this answer | |
$\endgroup$
1
$\begingroup$

This is my failed attempt at this problem. Essentially too long for a comment and contains multiple mistakes.

So, lets start with the integral:

$$\int_0^\infty \cos ax^2\text{ d}x=\sqrt\frac{\pi}{8a}$$

$$\frac{\partial^{-4mn}}{\partial a^{-4mn}}\int_0^\infty {\cos ax^2}=\int_0^\infty \frac{\cos ax^2}{x^{8mn}}\text{ d}x=\frac{\pi}{\sqrt 8}\frac{a^{4mn-1/2}}{\Gamma (4mn+1/2)}$$

Which is true because of this question: Derivative of a negative order?

$$\sum_{m=1}^\infty \int_0^\infty \frac{\partial^{-4mn}}{\partial a^{-4mn}}\cos ax^2 \, dx=\int_0^\infty \left(\frac{\cos ax^2}{x^{8n}}+\frac{\cos ax^2}{x^{16n}}+\frac{\cos ax^2}{x^{24n}}\cdots\right)\text{ d}x=\int_0^\infty \frac{\cos ax^2}{x^{8n}-1}\text{ d}x$$

Now, substitute $x\to\sqrt x, \text{ d}x\to\frac{\text{ d}x}{2\sqrt x}$ and assume that $n=\frac{1}{4}k$

And we get: $$\frac{1}{2}\int_0^\infty \frac {\cos ax}{\sqrt x (x^k-1)} \text{ d}x=\frac{\pi}{\sqrt 8}\sum_{m=1}^\infty \frac{a^{mn-1/2}}{\Gamma(mk+1/2)}$$

Maybe there is a part of this proof that you can deviate from.

| cite | improve this answer | |
$\endgroup$

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.