This project will investigate scalable parallel algorithms for the solution of light transport balance equations for environments composed of complex surface geometry. Although Monte Carlo methods can solve such balance equations when the input geometry is complex (e.g., 100 million surfaces), their convergence is slow. To overcome this limitation, finite element algorithms will be used to roughly characterize the light flow in the environment, and this large-scale information will be used to guide the sampling strategies of the Monte Carlo algorithms. Because designers of optical systems typically know where the large energy flow occurs in their systems, a 3D computer graphics steering front-end will be used to help guide both the subdivision in the finite element phase, and the sampling strategies of the Monte Carlo phase. Although the research being conducted attempts to improve solutions to the light transport problem, the systematic combination of the strengths of deterministic, Monte Carlo, and human-assisted methods could have applications in many high-dimensional problems where Monte Carlo methods are typically used.
