9
$\begingroup$

I'm trying to find a closed form for the following integral, for all $n\in\mathbb{N}^*$ : $$\int_{1}^{\infty}\frac{1}{x^2}\prod_{k=1}^{n}\left[1-\frac{1}{2k+x}\right]\text{d}x$$ As suggested by Zacky, a $x=\frac{1}{t}$ substitution gives : $$\int_{0}^{1}\prod_{k=1}^{n}\left[\frac{(2k-1)x+1}{2kx+1}\right]\text{d}x$$

The thing is that, for every specific $n\in\mathbb{N}^*$, mathematica is able to compute the integral in terms of logarithms only. In fact, it can always find the antiderivative in terms of logarithms.

What I'm looking for is a general expression for all $n\in\mathbb{N}^*$, if given that it exists.

Now the integrand is an already factorized rational function, so I assumed I could tackle this with some partial fraction decomposition, but I'm not good enough with this method to lead to anything. I'm also not familiar with complex analysis so I haven't tried contour integration, but if you think it can lead to anything, feel free to do so.

Any suggestion ?

$\endgroup$
  • 3
    $\begingroup$ This integral is so ready for a $x=\frac{1}{t}$ substitution, why not do it? $$\int_0^1 \prod_{k=1}^n \frac{(2k-1)t+1}{2kt +1}dt$$ This looks better. $\endgroup$ – Zacky Jan 11 at 14:44
  • $\begingroup$ Yes you're right, the $\frac{dx}{x^2}$ was indeed really appealing for this substitution, now I feel really stupid to not having seen it. Thank you for your input ! $\endgroup$ – Harmonic Sun Jan 11 at 14:50
  • $\begingroup$ Any reason for why a closed form might exist? $\endgroup$ – Darkrai Jan 11 at 14:56
  • 2
    $\begingroup$ It might help to note that $$\frac{(2k-1)x+1}{2kx+1}=1-\frac{x}{2kx+1}$$ it sort of looks better to me $\endgroup$ – clathratus Jan 11 at 16:14
  • 1
    $\begingroup$ @HarmonicSun That said, the answer becomes rather nasty as $n$ increases. $\endgroup$ – Carl Schildkraut Jan 11 at 16:15
10
$\begingroup$

To evaluate the integral \begin{equation} I= \int_{1}^{\infty}\frac{1}{x^2}\prod_{k=1}^{n}\left[1-\frac{1}{2k+x}\right]\,dx \end{equation} we consider the contour integral \begin{equation} J=\int_C \frac{\ln\left( z-1 \right)}{z^2}\prod_{k=1}^{n}\left[1-\frac{1}{2k+z}\right]\,dz \end{equation} $\ln(z-1)$ is defined with a branch cut from $z=1$ to infinity along the positive real axis. $C$ is the classical keyhole contour which starts near $z=1$, is then just above the positive real axis, follows the large circle and returns to the starting point, below the real axis and avoids $z=1$. The contribution of the large circle and that of the small circle around $z=1$ vanish. Above the real axis, $z=x+i\varepsilon$ and $\ln\left(z-1 \right)=\ln(x-1)$, and below, $\ln\left(z-1 \right)=\ln(x-1)+2i\pi$. We have then \begin{equation} J=-2i\pi I \end{equation} The integrand has a double pole at $z=0$ and single poles at $z=-2p$, for $p=1,2,\ldots,n$. From the residue theorem, \begin{equation} J=2i\pi\left( \operatorname{Res}(z=0)+\sum_{p=1}^n \operatorname{Res}(z=-2p)\right) \end{equation} with \begin{equation} \operatorname{Res}(z=0)=\prod_{k=1}^{n}\left[1-\frac{1}{2k}\right]\left.\frac{d\ln(z-1)}{dz}\right|_{z=0}+\ln(-1)\left.\frac{d}{dz}\prod_{k=1}^{n}\left[1-\frac{1}{2k+z}\right]\right|_{z=0} \end{equation} We have \begin{align} \prod_{k=1}^{n}\left[1-\frac{1}{2k}\right]&=\frac{\Gamma(n+1/2)}{\sqrt\pi\Gamma(n+1)}\\ &=\frac{2^{1-2n}\Gamma(2n)}{\Gamma(n)\Gamma(n+1)}\\ \left.\frac{d}{dz}\prod_{k=1}^{n}\left[1-\frac{1}{2k+z}\right]\right|_{z=0}&=\sum_{k=1}^n\frac{1}{4k^2}\frac{\prod_{p=1}^{n}\left[1-\frac{1}{2p}\right]}{1-\frac{1}{2k}}\\ &=\frac{2^{1-2n}\Gamma(2n)}{\Gamma(n)\Gamma(n+1)}\sum_{k=1}^n\frac{1}{2k(2k-1)} \end{align} Then \begin{equation} \operatorname{Res}(z=0)=-\frac{2^{1-2n}\Gamma(2n)}{\Gamma(n)\Gamma(n+1)}\left( 1+i\pi\sum_{k=1}^n\frac{1}{2k(2k-1)} \right) \end{equation} Now, as the residue at its pole of each of the factors is $-1$ \begin{align} \operatorname{Res}(z=-2p)&=-\frac{\ln(2p+1)+i\pi}{4p^2}\prod_{k=1\\k\ne p}^{n}\left[1-\frac{1}{2k-2p}\right]\\ \end{align} As $I=-J$ and since $I$ is real, we can conclude by taking only the real part of the residues, \begin{equation} I=\frac{2^{1-2n}\Gamma(2n)}{\Gamma(n)\Gamma(n+1)}+\sum_{p=1}^n \frac{1}{4p^2}\prod_{k=1\\k\ne p}^{n}\left[1-\frac{1}{2(k-p)}\right]\ln\left( 2p+1 \right) \end{equation} which can be written as \begin{equation} I=\frac{2^{1-2n}\Gamma(2n)}{\Gamma(n)\Gamma(n+1)}+\sum_{p=1}^{n-1} \frac{\ln\left( 2p+1 \right)}{p^2}\frac{2^{1-n}\Gamma(2n)\Gamma\left( 2(n-p) \right)}{\Gamma^2(p)\Gamma(n-p)\Gamma(n+1-p)}+\frac{2^{1-2n}\Gamma(2n)}{\Gamma^2(n)}\ln(2n+1) \end{equation}

$\endgroup$
  • $\begingroup$ (+1) Really cool! Is this called contour integration? Would you suggest any resources which I could use to learn this? $\endgroup$ – clathratus Jan 11 at 23:11
  • $\begingroup$ Great answer, thank you ! $\endgroup$ – Harmonic Sun Jan 11 at 23:15
  • 1
    $\begingroup$ @clathratus Yes this is contour integration. In the case of a rational function to be integrated on the positive real axis, adding a $\ln$ is a classical trick. Many text books of complex analysis describe this kind of methods. I like Henrici "Applied and computational complex analysis", but there are many others. $\endgroup$ – Paul Enta Jan 11 at 23:18

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.