Add refract node

[ld-kerley]

Adding refract node to partner with the reflect node, with a similar named interface, and implementation as a nodegraph taken from the OSL spec

vector refract (vector I, vector N, float eta)
{
         float IdotN = dot (I, N);
         float k = 1 - eta*eta * (1 - IdotN*IdotN);
         return (k < 0) ? vector(0,0,0) : (eta*I - N * (eta*IdotN + sqrt(k)));
}

which also lines up with the Kronos implementation here too.

[jstone-lucasfilm]

@ld-kerley @kwokcb One remaining question in my mind is whether this index of refraction should be named ior rather than eta, following the pattern in MaterialX nodes where the real component of IOR is named ior and the imaginary component is named extinction. I believe this choice has historically been motivated by the goal of making node interfaces as artist-focused as possible, where eta and kappa are less artistically descriptive.

Here are two examples of the current pattern for IOR in MaterialX nodes:

https://github.com/AcademySoftwareFoundation/MaterialX/blob/main/libraries/pbrlib/pbrlib_defs.mtlx#L136
https://github.com/AcademySoftwareFoundation/MaterialX/blob/main/libraries/pbrlib/pbrlib_defs.mtlx#L79

[ld-kerley]

I think consistency is important - so +1 vote for ior here if that the prevailing direction for MaterialX.

[kwokcb]

Correct me if I have the wrong interpretation but isn't eta the ratio of the source ior / destination ior so maybe relative_ior or ior_ratio ?

[ld-kerley]

I alway refer to The Bible whenever I'm confused.

"The speed reduction is commonly summarized using the index of refraction (IOR). For example, a material with an IOR of 2 propagates light at half the speed of light. For common materials, the value is in the range 1.0–2.5 and furthermore varies with the wavelength of light. We will use the Greek letter , pronounced “eta,” to denote this quantity."

[jstone-lucasfilm]

@kwokcb All of the IOR inputs in MaterialX nodes can be thought of as being relative to the IOR of the exterior medium (which is 1.0 by default), but we don't make such a distinction in our naming conventions. Following that convention, I can see the name ior as being the most consistent here, though I'm open to other ideas.

[kwokcb]

Sounds good to go with ior. And thanks for the eta reference @ld-kerley.

[jstone-lucasfilm]

This looks excellent, thanks @ld-kerley!