>"Yaohua Hu also spent a lot of time researching an improved light-map representation (it's better than spherical harmonics!) that gives us the same support for area light sources, improved contrast, fewer artifacts, a smaller memory footprint and much better performance."

Any chance light could be shed (sorry :D) on what the changes were? Was it a switch to something like H-basis for surfaces to reduce redundancy, or was a different sort of representation used, i.e. something more spatial-domain like spherical gaussians?

Another thing I'm curious about: Halo 3 seems to store a unique irradiance distribution for each (RGB) color channel. Although that's a strong general-case approach, it also seems like it implies storing a lot of information (and doing a lot of lighting operations) for what will be in most cases subtle gains, relative to something like using a single tinted irradiance distribution. Were optimizations along these lines made?