Hi Greg
Many thanks for the feedback -
Actually, refraction is never ignored in Radiance, even when tracing to light sources. The sources will be missed or not depending on how big they are and how much refraction goes on. If a pane of glass is modeled using dielectric, so that rays are refracted going through it, they get bent back straight so they still head in the same direction, albeit shifted, slightly.
Ok that's interesting - it means that refraction through parallel surfaces will never fail for direct sunlight - since they will be heading in the right direction to hit the sun source. Great.
Seems that for non-parallel situations refraction treatment in the direct calc in Radiance is different to other raytracers in that when the ray misses a direct source there is a null result - whereas i think the usual treatment is to ignore refraction. I'm thinking of sunlight only through a cylindrical glass of water - both packages will show the scene through the glass refracted - but in std raytracer the sunlight passes through and is cast onto the table simply changed by the transmittance of the media.
Radiance will presumably record transmission in places (small deflection) and in other parts not so. (Not a criticism, Greg - just good to be aware in some situations, such as a wavey-glass brick wall, may transmit less of the sunlight than anticipated by the material spec).
Spheres and lenses are a different matter of course, and caustics are not modeled correctly for that reason. You need to use the photon map for that!
Indeed - mind you perhaps the caustics produced will ultimately split out just into the three primaries.
I'm thinking of a virtual Newton conducting his classic prism experiment 
I think refractive index also does not apply when tracing rays as part of the indirect calculation - even though missing a large area source (sky) is not likely to be a problem.
I don't know where you heard this, but it's not true.
I guess i got it from the (wrong) idea that refraction only works for view rays.
Reflectance - in atria there are likely to be complex and multiple glazing surfaces - eg the roof and then in large areas along the offices lining the edge....
That's partially true. Interreflections of the sky and other parts of the ambient calculation will consider the multiple reflections going on, specular as well as diffuse. However, you would have to use virtual sources to get the pure specular reflections from light sources, which can be expensive. You're better off doing that only when it really matters -- like when you have a mirrored building facade or something.
Specular contributions are probably neglected more often than they should, especially when we are dealing with direct sunlight on polished floors or glazing around entrances or in atria. It may not be important overall - but in specific areas it can be highly significant.
The upside is that the source is just the sun and these specular surfaces often lie in the same plane so that a single expanse of 'mirror' will do the job for many windows. By comparison with the ambient calculation the creation of one or two extra virtual suns to cover these pathways in the direct calculation need not be costly.
(Daniel - this may have relevance for your work) it does not include the contribution of indirect sources eg bright overcast skies are not going to be specularly reflected around the atrium. The Radiance indirect calculation essentially deals with diffuse-diffuse pathways only.
Again, these are accounted for correctly.
OK Great! One subsequent question to clarify:
Consider if we have no direct sources and a view ray that hits a diffuse floor - we have hemisphere sampling and ambient bounce 1 - if any of these rays is incident on a specular surface then the ray is traced back with specular reflection and then, for instance finds the sky.
So would this pathway (where sky illuminates floor via glazing) require a setting of -ab 3 as a minimum as would be expected from a purely diffuse pathway (1 ambient bounce - 2 specular reflection - 3 indirect sky source)?
If the operator decides to switch to modelling the atrium aperture as an illum (switching the indirect sky to a the direct source) - then this pathway is no longer included - again unless the surface is identified explicitly as a mirror (modified by the reflection function used on the glazing).
This is useful to be aware of because the notion of changing a source requiring a change to the model specs is not necessarily obvious - and it may account for some of the differences in results between using an illum or relying on the ambient calculation.
Ok so does that mean specs and functions for reflectance changing with angle of incidence are going to effect the appearance of the scene visible in the glazing, the light transmitted, but not the light reflected?
They affect both.
And what about direct sunlight and diffusing glass? Is the light transmitted - or does the random scattering cause the same problem as refraction about missing the target when tracing the ray backwards?
The diffuse or rough specular part of transmission (and reflection) gets included in the indirect (ambient) calculation. If modeled using the directional diffuse part of one of the *func or *data material types, this gets lumped into the Lambertian component, which admittedly is a very crude approximation. Perhaps this is what you are thinking of above -- however, it doesn't apply to the Gaussian materials (plastic, metal, trans, plastic2, metal2, trans2). These all work properly, which is why you should use them whenever possible. Problems remain when you get strong caustics from a low roughness surface, however.
Still not sure about this - a ray in the direct calc is traced back toward the sun - but is incident on some diffusing glass and is scattered randomly and presumably misses its target. (?)
I agree about using the Gaussian materials as far as possible especially given the role of indirect lighting in daylighting - i'm not entirely convinced given the Lambertian diffuse approximation and nevertheless good results in most cases - if going to the trouble of measuring and specifying BRTD materials often really make that much difference.
The number of times i've seen lighting studies blown away because some such thing as people deciding to keep their blinds drawn so they can see their computer screens clearly 
Many thanks again Greg - very useful to clarify things.
cheers
alex
···
*******************************************************
A. J. Summerfield [email protected]
Faculty of Architecture, University of Sydney
Outside of a dog, a book is man's best friend.
Inside a dog, it's too dark to read. Groucho Marx
*******************************************************