I am interested in computing a normalizing constant (of a Gaussian density in dimension $3$). Such normalizing constants often do not have a closed form. In dimension $2$, this normalizing constant can be computed in closed form.

The integral I would like to compute is :

$$ \int_{\mathbb{R}^{3}} e^{-(r_{1}^{2} + r_{2}^{2} + r_{3}^{3})/2\sigma^{2}} \sinh\Big( \frac{\vert r_{1}-r_{2} \vert}{2} \Big)\sinh\Big( \frac{\vert r_{1}-r_{3} \vert}{2}\Big)\sinh\Big( \frac{\vert r_{2}-r_{3} \vert}{2} \Big)dr_{1}dr_{2}dr_{3} $$

Using the spherical coordinates $(r_{1},r_{2},r_{3}) = (r\sin\varphi\cos\theta,r\sin\varphi\sin\theta,r\cos\varphi)$, I find myself trying to compute the following integral :

$$ \int_{0}^{+\infty}\int_{0}^{2\pi}\int_{0}^{\pi} e^{-r^{2}/2\sigma^{2}} \sinh\Big( \frac{r\sin\varphi\vert \cos\theta - \sin\theta\vert}{2}\Big)\sinh\Big( \frac{r\vert\sin\varphi\cos\theta-\cos\phi\vert}{2}\Big) \times \sinh\Big(\frac{r\vert\sin\varphi\sin\theta-\cos\varphi\vert}{2}\Big)r^{2}\sin\varphi \, drd\theta d\varphi $$

I'm not an expert but, looking at this integral, I doubt there exist a closed expression.


Since you have a full symmetry by exchanging $r_1$, $r_2$ and $r_3$, the integral can be computed as $$3!\int_{-\infty}^{+\infty}\mathrm dx \int_{-\infty}^x\mathrm dy\int_{-\infty}^y\mathrm dz \mathrm e^{-(x^2+y^2+z^2)/2\sigma^2}\sinh\left(\frac{x-y}2\right)\sinh\left(\frac{x-z}2\right)\sinh\left(\frac{y-z}2\right).$$ Then writing the $\sinh$ in the exponential form, you end up with a sum of product integrals that are all of Gaussian type.

  • $\begingroup$ very nice (+1)! $\endgroup$ – tired Oct 13 '15 at 14:05
  • $\begingroup$ You are right ! Nice answer :-) Thanks $\endgroup$ – jibounet Oct 13 '15 at 14:08

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.