This technical memo compares six different strategies for sampling a disk area light source. We test padding the domain with zero values, rejection sampling, polar mapping, concentric mapping, and two new strategies we call polar4 and concentric4 mapping. We test the six strategies with nine different sample sequences on pixels with full illumination and partial shadow (penumbra), and find the best combination of sampling strategy and sample sequence.
One of the new sampling strategies gives more than 3 times error reduction in some cases.
Available as Pixar Technical Memo #18-02