4
$\begingroup$

Hi, I need to integrate a function on an n-dimensional sphere surface. One way is to use the triangle function like: http://en.wikipedia.org/wiki/N-sphere#Spherical_volume_element, however, it is too complex, do you have any idea how to solve it?

My problem is $\int_{\|{\bf U}\| = 1} \sqrt{ {\bf U}^T {\bf M} {\bf U} } d u_1 \ldots d u_n$, where ${\bf M}$ is a positive semidefinite matrix.

$\endgroup$
4
  • 1
    $\begingroup$ You can assume that $\bf{M}$ is diagonal, because orthogonal group acts on a sphere in a measure-preserving way, but that doesn't help, because even in two-dimensional case you end up with an elliptic integral. $\endgroup$ Jul 22, 2012 at 9:33
  • $\begingroup$ You may get a better response at e.g. math.stackexchange.com, as MO is for questions of research interest (see the FAQ for more suggestions). $\endgroup$
    – David Roberts
    Jul 22, 2012 at 10:44
  • $\begingroup$ It's still a nontrivial problem to integrate this numerically. If I did this right it comes down to an elementary multiple of the hyperelliptic integral $$ \int_0^\infty \left[ 1 - \det(1+{\bf M}t^2)^{-1/2} \right] \frac{dt}{t^2}. $$ By the way, the notation $du_1 \cdots du_n$ for the integration element is misleading: it suggests an $n$-dimensional integral, but $\|{\bf U}\|^2 = 1$ is an $n-1$-dimensional manifold. $\endgroup$ Jul 22, 2012 at 15:21
  • $\begingroup$ One possible way is to use Monte Carlo method. $\endgroup$
    – Andrew
    Jul 22, 2012 at 15:25

1 Answer 1

2
$\begingroup$

An analytic expression is given in my paper http://arxiv.org/pdf/math/0403375.pdf (see page 6) in terms the Lauricella hypergeometric function. Note that the $n$-dimensional integral is hopeless to evaluate numerically by Monte-Carlo method ("the curse of dimensionality"), whereas the hypergeometric function is, at the most primitive, expressed as a one-dimensional integral, so numerical evaluation is very fast.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.