- pic of Halton sequence, and sequence projected on to a hemisphere, cf  (use R)
- tweak Radiosity lead
- discussion of:
- computational tweaks -- faster by pre-baking radiosity (lightmaps); or by precomputing relevant sampling directions (-> semi-hardcoded lookup list; or a priority lookup list, to go so far down depending on the light to be distributed; precomputed form factors.
- shifting the signal-to-noise around -- using shooting rather than gathering, to spread absolute error more evenly
- incremental update
- improving quality -- subdivision, and adaptive subdivision; higher basis functions; overly soft shadows behind a Cornell box; on the other hand "smooth may be more important than right" -- need to keep noise levels down.