Again, physically correct result is my primary concern. And I also have a strong belief that for the simple scene such as
cornell box, perhaps just the simplest strategy will be enough. To this end, I tried MLT algorithms to solve the light transport
integral constrained to subdomain of path length 2, by re-initialize muation, so that T(y|x) is simply T(y). The luminance is taken as
the sum of rgb component. The sample number listed below is the average sample per pixel.
Comparison of central pixel value of scene .
For MLT algorithms, seed pool size is 100,000.
Analytic Value
path tracing(s1t2) 9 samples
MLT 100 samples
MLT 400 samples
MLT 900 samples
~~~0.01
0.0125904
0.0110081
0.0119063
Comparison of partial solution (constrained to lenght 2 path domain) between MLT and path-tracing.
