RenderMan & RISpec, Dipoles

Hi,

I've been revisiting sss according to Jensen, and I'm still trying to
understand the concept of dipole diffuse approximation. Maybe I'm a
little math blind, but it just doesnt make sense. As I understand it,
the incident ray is converted to two point sources, and then various
points near P would be sampled and summed to approximate multiple
scatterings. Correct?

So, the next thing is where do the two light sources go, and why is one
negative? I would assume that the position of the two lightsources
would depend on the angle of incidence. Is this completely wrong?

Thanks,
Atabet

This may or may not help you but... the way sss was first introduced to
me went something like "you've got this big theorethical model with
lots of formulas and lots of randomness in it which at the end of the
day, like many things in computer graphics, is well approximated with a
smooth gaussian-like curve". In other words, an approximation is just
that: an approximation.

As I recall, the paper mentions briefly that the angle of incidence can
be taken into account when you compute the initial illuminance by using
fresnel. I think there's no need to use it beyond computing how much
light is scattered because everything is so diffuse (after a few
bounces light goes in all directions). At least that's the way I
understood it.

Olivier