4
$\begingroup$

In a high-dimensional setting, say $d \gg 5$, what is a recommended way of evaluating a spherical integral of a smooth (non-symmetric) function $f(\mathbf{x})$?

$ \int_\mathcal{S_r} f(\mathbf{x}) \mathrm{d}(\mathbf{x})$

where $\mathcal{S_r} = \{ \mathbf{x} \; | \; \|\mathbf{x}\|^2 = r \}$.

I want to avoid Monte Carlo integration due to high requirement on the number of samples. There seem to be various quadrature methods, but I don't know which one to use. Is there a recent survey of these integrals perhaps?

$\endgroup$
  • 3
    $\begingroup$ why not monte carlo? once $d-1>>1$ it'll be difficult to integrate by other means. $\endgroup$ – Aksakal Apr 8 '14 at 22:06
  • $\begingroup$ You can look up quasi-Monte Carlo, which has $O(N^{-1})$ convergence instead of $O(N^{-1/2})$, or sparse quadrature. $\endgroup$ – Daniel Shapero Apr 8 '14 at 23:16
  • $\begingroup$ In high dimensions a standard normal distribution concentrates highly on the surface of the sphere, so if you don't need an exact answer you could integrate $f(x)e^{-|x|^2}dx$ over a box containing the sphere instead of the sphere itself, perhaps using sparse grids. $\endgroup$ – Nick Alger Apr 9 '14 at 2:23
  • $\begingroup$ You can use the SphericalCubature package for R. $\endgroup$ – Stéphane Laurent Mar 3 '18 at 10:14
1
$\begingroup$

I'd use one of the hypersphere random point pickers and integrate with Monte Carlo.

$\endgroup$

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.