Homogenization and diffusion

Robert van Gestel

The standard method to simulate optical systems is ray tracing. This Lagrangian method requires a lot of rays to be traced in order to provide an accurate description. In this project we take an Eulerian approach and instead solve Liouville’s equation, which is a linear advection equation on phase space describing the propagation of energy carried by the rays. We solve Liouville’s equation using energy conservative high order methods, resulting in faster convergence than ray tracing.

Test case involving total internal reflection and refraction. Click the figure to see a larger version.
The left figure shows the obtained illuminance at z = 0.7, using the standard method of quasi-Monte Carlo ray tracing (RT) and solving Liouville’s equation with the discontinuous Galerkin (DG) method. On the right, the average error in the illuminance is computed to compare the methods. For DG the parameter N is related to the method’s order, and the high order methods converge faster than ray tracing.
Vì Kronberg

Many optical designers are restricted to working with the geometrical optics (GO) approximation, where light propagation is modelled using rays traversing in straight lines in a free medium, and obeying the specular law of reflection and refraction when interacting with a reflector surface and/or lens. The omission of scattering or diffuse reflection leads to discrepancies between the simulations and production samples. We add scattering to the GO approximation by considering deviations from the specular law of reflection, and in certain cases this yields a relatively simple convolution integral. One can then deconvolve said quantities to get an intermediate specular distribution such that diffuse problems can be treated using the Monge-Ampère least-squares solver.

Schematic showing the process; p is a probability density function related to the scattering; h is a prescribed target intensity, f is the source intensity and g is the intermediate equivalent specular intensity. After deconvolution, the diffuse design problem reduces down to a specular design problem involving the source distribution f and target distribution g.