10

Interesting challenge: Does anyone know how to evaluate the integral $$\int_0^\infty\frac{\cos(ax)}{1+x^N}\text{d}x$$ for $a,N\in\mathbb{R}$ and $N>1$?

The $N=2$ case is quite easily done using the Residue Theorem as well as for general $N>1$ for $a=0$. Both exploit symmetries when path integrating, which do not work in the general case.

Let me know if you have any ideas! Cheers!

Gabin
  • 101
  • 6
    Welcome to Math.SE! ... To be clear: If this "challenge integral" is one that you already know how to evaluate in all cases, then you should say so explicitly. (And you can hide the solution in "spoiler" formatting >!.) This will keep those who visit Math.SE to help people (the primary purpose of this site) from wasting their time explaining things you already understand. Cheers! – Blue Jan 22 '22 at 16:40
  • 1
    Thanks for the tip! I haven't been able to solve this integral. – Gabin Jan 22 '22 at 17:23
  • 1
    $N\in\mathbb{R}$ or $N\in\mathbb{N}_{\geq 2}$? – Diger Jan 23 '22 at 11:41
  • You can generalise the methods used in the case $N=3$ here: https://math.stackexchange.com/questions/1057519/how-to-solve-int-0-infty-frac-cosaxx31dx?rq=1 – Gary Jan 23 '22 at 12:29
  • I'm interested in the $N\in\mathbb{R}$ case. – Gabin Jan 23 '22 at 18:12
  • 4
    There is a closed form in terms of the Fox H-function. – Maxim Jan 24 '22 at 01:01
  • 1
    @Maxim Can you write down how to convert it into Fox H? – Diger Jan 25 '22 at 12:23
  • @Maxim, so the general case $N\in\mathbb{R}>1$ requires the Fox H-function to find a closed form? – Gabin Jan 27 '22 at 13:00
  • 1
    There may exist other closed forms. The H-function can be obtained by applying the Mellin convolution method. We have $$F(s) = \mathcal M\cos = \frac {2^{s - 1} \sqrt \pi , \Gamma {\left( \frac s 2 \right)}} {\Gamma {\left( \frac 1 2 - \frac s 2 \right)}}, \ G(s) = \mathcal M {\left[ x \mapsto \frac 1 {x^N + 1} \right]}(s) = \frac 1 N \Gamma {\left( \frac s N \right)} \Gamma {\left( 1 - \frac s N \right)}.$$ Your integral for $a > 0$ is $$\frac 1 {2 \pi i} \int_\gamma G(1 - s) F(s) a^{-s} ds,$$ which, by definition, is the H-function. – Maxim Jan 27 '22 at 14:23

3 Answers3

3

For even $N$ you can use the residue theorem as well. It suffices to consider $a>0$ and calculate the integral $$\oint_{-\infty}^{\infty} \frac{e^{iax}}{1+x^N} \, {\rm d}x = \int_{0}^{\infty} \frac{2\cos(ax)}{1+x^N} \, {\rm d}x = 2\pi i \sum_{0 \leq k \leq \frac{N-1}{2}} {\rm Res}\left( \frac{e^{iax}}{1+x^N} \right)\Bigg|_{x=e^{\frac{i\pi+i2\pi k}{N}}} \\ =\frac{-2\pi i}{N} \sum_{k=0}^{N/2-1} e^{ia e^{\frac{i\pi+i2\pi k}{N}} + \frac{i\pi+i2\pi k}{N}} = \frac{-2\pi i}{N} \sum_{k=0}^{N/2-1} e^{ia \cos\left(\frac{\pi+2\pi k}{N}\right) + \frac{i\pi+i2\pi k}{N}}e^{-a\sin\left( \frac{\pi+2\pi k}{N} \right)} \\ =\frac{-2\pi i}{N} \sum_{k=0}^{N/2-1} \Bigg[\cos\left(a \cos\left(\frac{\pi+2\pi k}{N}\right) + \frac{\pi+2\pi k}{N}\right) \\ + i \sin\left(a \cos\left(\frac{\pi+2\pi k}{N}\right) + \frac{\pi+2\pi k}{N}\right)\Bigg]e^{-a\sin\left( \frac{\pi+2\pi k}{N} \right)} \\ =\frac{2\pi}{N} \sum_{k=0}^{N/2-1} \sin\left(a \cos\left(\frac{\pi+2\pi k}{N}\right) + \frac{\pi+2\pi k}{N}\right)e^{-a\sin\left( \frac{\pi+2\pi k}{N} \right)}$$ where the loop-integral on the LHS is closed in the upper half plane. The imaginary part is zero, since the LHS is real.

Diger
  • 6,852
1

Just some working out in progress.

Well, we are trying to find the following integral:

$$\mathcal{I}_\text{n}\left(\text{k},\alpha\right):=\int_0^\infty\frac{\cos\left(\text{k}x\right)}{\alpha^2+x^\text{n}}\space\text{d}x\tag1$$

To calculate the integral, we first note we may write that:

$$\frac{1}{\alpha^2+x^\text{n}}=\int_0^\infty\exp\left(-\left(\alpha^2+x^\text{n}\right)\text{y}\right)\space\text{dy}\tag2$$

And then we have:

$$\mathcal{I}_\text{n}\left(\text{k},\alpha\right)=\int_0^\infty\left(\int_0^\infty\cos\left(\text{k}x\right)\exp\left(-\left(\alpha^2+x^\text{n}\right)\text{y}\right)\space\text{dy}\right)\space\text{d}x\tag3$$

Change the integration order assuming the value of the integral preserves:

$$\mathcal{I}_\text{n}\left(\text{k},\alpha\right)=\int_0^\infty\exp\left(-\text{y}\alpha^2\right)\left(\int_0^\infty\cos\left(\text{k}x\right)\exp\left(-\text{y}x^\text{n}\right)\space\text{d}x\right)\space\text{dy}\tag4$$

Make use of the power series of $\cos x$:

$$\mathcal{I}_\text{n}\left(\text{k},\alpha\right)=\int_0^\infty\exp\left(-\text{y}\alpha^2\right)\left(\int_0^\infty\sum_{\text{m}\space\ge\space0}\frac{\left(-1\right)^\text{m}\left(\text{k}x\right)^{2\text{m}}}{\left(2\text{m}\right)!}\cdot\exp\left(-\text{y}x^\text{n}\right)\space\text{d}x\right)\space\text{dy}\tag5$$

Which gives:

$$\mathcal{I}_\text{n}\left(\text{k},\alpha\right)=\int_0^\infty\exp\left(-\text{y}\alpha^2\right)\left(\sum_{\text{m}\space\ge\space0}\frac{\left(-1\right)^\text{m}\text{k}^{2\text{m}}}{\left(2\text{m}\right)!}\int_0^\infty x^{2\text{m}}\exp\left(-\text{y}x^\text{n}\right)\space\text{d}x\right)\space\text{dy}\tag6$$

If you substitute:

$$\text{s}=\text{y}^\frac{2\text{m}+1}{\text{n}}\cdot x^{2\text{m}+1}\tag7$$

We will get:

$$\int_0^\infty x^{2\text{m}}\exp\left(-\text{y}x^\text{n}\right)\space\text{d}x=\frac{\text{y}^{-\left(\frac{2\text{m}}{\text{n}}+\frac{1}{\text{n}}\right)}}{2\text{m}+1}\int_0^\infty\exp\left(-\text{s}^\frac{\text{n}}{2\text{m}+1}\right)\space\text{ds}\tag8$$

Which is defined in terms of the incomplete gamma function:

$$\int_0^\infty\exp\left(-\text{s}^\frac{\text{n}}{2\text{m}+1}\right)\space\text{ds}=-\frac{2\text{m}+1}{\text{n}}\cdot\left[\Gamma\left(\frac{2\text{m}+1}{\text{n}},\text{s}^\frac{\text{n}}{2\text{m}+1}\right)\right]_0^\infty\tag9$$

So:

$$\int_0^\infty x^{2\text{m}}\exp\left(-\text{y}x^\text{n}\right)\space\text{d}x=\frac{\text{y}^{-\left(\frac{2\text{m}}{\text{n}}+\frac{1}{\text{n}}\right)}}{\text{n}}\cdot\left[\Gamma\left(\frac{2\text{m}+1}{\text{n}},\text{s}^\frac{\text{n}}{2\text{m}+1}\right)\right]_\infty^0\tag{10}$$

Which gives:

$$\int_0^\infty x^{2\text{m}}\exp\left(-\text{y}x^\text{n}\right)\space\text{d}x=\frac{\text{y}^{-\left(\frac{2\text{m}}{\text{n}}+\frac{1}{\text{n}}\right)}}{\text{n}}\cdot\Gamma\left(\frac{2\text{m}+1}{\text{n}}\right)\tag{11}$$

So:

$$\mathcal{I}_\text{n}\left(\text{k},\alpha\right)=\frac{1}{\text{n}}\sum_{\text{m}\space\ge\space0}\frac{\left(-1\right)^\text{m}\text{k}^{2\text{m}}}{\left(2\text{m}\right)!}\cdot\Gamma\left(\frac{2\text{m}+1}{\text{n}}\right)\int_0^\infty\exp\left(-\text{y}\alpha^2\right)\text{y}^{-\left(\frac{2\text{m}}{\text{n}}+\frac{1}{\text{n}}\right)}\space\text{dy}\tag{12}$$

Jan Eerland
  • 29,457
1

The integral

$$f(a) = \int_0^\infty\frac{\cos(a x)}{1+x^N}\;dx$$

with $N= 1, 2, ...$ and $a\ge 0$ can be calculated explicitly in terms of known functions as follows.

Partial fraction decomposition leads to the representation

$$\frac{1}{1+ x^N} = \sum_{k=1}^{N} \frac{c_k}{z_k - x}\tag{1}$$

where

$$z_k=\exp\left( \frac {i \pi (2k+1)}{N}\right)\tag{2}$$

are the $N$ different roots of the equation $1+ x^N=0 $ and the coefficients are complex numbers given by

$$c_k = \lim_{x\to z_k} \frac{z_k - x}{1+x^N}=(-1)^N \frac{1}{\prod_{i=1, i\ne k}^{N} (z_i - z_k)}\tag{3}$$

Hence we are left with the integral

$$i(z,a)=\int_{0}^{\infty} \frac{\cos(a x)}{z-x}\;dx\tag{4}$$

Which can be solved giving

$$i(z,a) = \frac{\pi}{2}\sin(a z) + \sin(a z)\text{ Si}(a z) + \cos(a z)\text{ Ci}(-a z)\tag{5}$$

Here $\text{ Si}$ and $\text{ Ci}$ are the integral sine and integral cosine, respectively (cf. https://en.wikipedia.org/wiki/Trigonometric_integral).

Putting things together we have found that the integral in question has a closed form in terms of known functions.

Remark: Just for information (not for practical use): Mathematica tells me that the integral can be expressed in compact form by the special function MeijerG.

Dr. Wolfgang Hintze
  • 13,265
  • 25
  • 49
  • Thank you, this is quite useful as a simplified case! – Gabin Jan 27 '22 at 13:05
  • @ Gabin Thank you. But this is in fact the general case for N=1, 2, 3, ... Your own attempted solution in the OP also assumes integer values of N. Perhaps you should have pointed out explicitly that you are interested in real, non integer values of N. – Dr. Wolfgang Hintze Jan 27 '22 at 13:23
  • The following integral should hold for all $N \in \mathbb{R}>1$ , $$\int_0^\infty\frac{1}{1+x^N}\text{d}x = (\pi/N) / sin(\pi/N) $$, or did you mean something else? – Gabin Jan 27 '22 at 19:31
  • I meant you should have said more precisely what you want. The integral is correct if a=0. For a>0 I have given the solution if N is a positive integer. I suggest you try rational N=p/q. It seems possible to find solutions here. My preliminary results show that the function MeijerG appears but the developments are messy. I have no idea for N = non-rationals like (sqrt(2), e, pi). – Dr. Wolfgang Hintze Jan 27 '22 at 20:53