Any ideas how to solve: \begin{equation} \int_0^\infty x^{n+\frac{1}{2}} (e^{a x }-1)^{-\frac{1}{2}} e^{i x t} dx \end{equation} where $a$ and $t$ are real, positive constants; $n$ is a positive integer.

I think the problem comes from having rational functions in both powers and exponential functions.

I tried to get ride of the rational power, but it didn't really help \begin{equation} \frac{\partial}{\partial q } \int_0^\infty x^{n} (e^{a x }-1)^{-\frac{1}{2}} e^{i x t+ q x^{1/2}} dx \end{equation}

Having an hint how to solve this for $t=0$ would already be useful. Thanks!

  • $\begingroup$ Out of curiousity, in what context did this integral appear? $\endgroup$ – Semiclassical Sep 12 '14 at 19:09
  • $\begingroup$ It a function of my own that I need for my research. But it is well defined, and should have a finite integrale value. I just don't know how to find the analytical form. $\endgroup$ – Aurelia Sep 12 '14 at 19:20
  • $\begingroup$ Ok. I ask because my physics brain sees the $n=0$ case as the Fourier transform of the density of states for a boson gas (with zero chemical potential). Probably just a coincidence. $\endgroup$ – Semiclassical Sep 12 '14 at 19:23
  • $\begingroup$ This is where is comes from indeed. I normally use the DOS square, i.e. $\frac{x}{e^{ax}-1}$, in which case the integral is easy to solve $\endgroup$ – Aurelia Sep 12 '14 at 19:32
  • $\begingroup$ Oof, I'd quite missed that square-root. That indeed makes life interesting. I'll see what I can do... $\endgroup$ – Semiclassical Sep 12 '14 at 19:36

It's actually enough to resolve the $n=0$ case, since $t$-derivatives of $$F(t)=\int_0^\infty \frac{e^{i x t}}{\sqrt{e^{2x}-1}} x^{1/2}\,dx$$ will bring down more powers of $x$; I've chosen units for so that $a=1$ for convenience. Even with these simplifications, I don't know how to compute a closed-form and so will have to settle for an appropriate series expansion. We rewrite the fraction in the integrand and expand in powers of exponenetials $$\frac{e^{i x t}}{\sqrt{e^{x}-1}}=\frac{e^{i x t-x/2}}{\sqrt{1-e^{-x}}}=\sum_{k=0}^\infty \binom{2k}{k}\left(\frac{e^{-x}}{4}\right)^k e^{i x t-x/2}=\sum_{k=0}^\infty \binom{2k}{k}4^{-k} e^{i x t-(k+1/2)x}.$$

Integrating term-by-term then gives

$$F(t)=\sum_{k=0}^\infty \binom{2k}{k}4^{-k}\cdot \frac{1}{2}\pi^{1/2} [(k+\frac{1}{2})-i x]^{-3/2}=\sum_{k=0}^\infty \binom{2k}{k}\frac{ 2^{-2k-1}\pi^{1/2}}{(k+\frac{1}{2}-i t)^{3/2}}$$

which is a rather formidable result. It's possible that this can be resummed as a hypergeometric series of some sort; I'll see what I can find.


$x^{n+\frac{1}{2}} (1-e^{-a x })^{-\frac{1}{2}} e^{x(-\frac{a}{2})} dx=x^{n+\frac{1}{2}}(\sum_p\frac{(2p-1)!!}{2p!!}e^{-a(p+\frac{1}{2})x})$.

So $\int x^{n+\frac{1}{2}} (1-e^{-a x })^{-\frac{1}{2}} e^{x(-\frac{a}{2})} dx=\Gamma(n+1.5).\sum_p \frac{(2p-1)!!}{2p!!}\frac{1}{a(p+\frac{1}{2})}^{n+1.5}$

But I guess you already had this, and this is not really practical...


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.