Lewis Hyatt wrote:
Henrik Sundberg wrote:
I doubt that the trig-version gives a good distribution:
inlineVectortrig(uniform_01<generator_type>&random) { constfloatphi=2*M_PI*random(); constfloatcos_theta=2*random()-1; constfloatsin_theta=sqrt(1-cos_theta*cos_theta); returnVector(cos_theta,sin_theta*cos(phi),sin_theta *sin(phi)); }
The result ought to be more dense at the poles.
The posted trig method is correct, it chooses a horizontal cutting plane uniformly. You can show that the surface area of a sphere contained between two z-coordinates only depends on the difference between the two coordinates, not their location, so this generates the correct distribution.
However, the "naive" method is incorrect, the resulting points are not uniformly distributed over the sphere. The correct version of the "naive" method is given here: http://mathworld.wolfram.com/SpherePointPicking.html
-Lewis
_______________________________________________
Also see here for a discussion for 4 different techniques (and a proof for the trig version!): http://www.math.niu.edu/~rusin/known-math/96/sph.rand Jeff Faust