raypier_optics VS goray

Raypier - current front runner, I've managed to model both an RC telescope and a thorlabs microlens array here with results that seem quite reasonable. Also has a gui for sanity checking/interactivity. Unfortunately the performance really starts to tank as you increase the resolutions of the plane where the E-field is evaluated, increase number of rays or increase number of surfaces. (Something I would like to do to be able to model atmospheric disturbance + mirror misalignments +

Tracing meshes is something I've just started tackling in raypier (https://github.com/bryancole/raypier_optics). This is a spatial-search problem and a Oriented Bounding Box tree structure (OBB-tree) is a decent approach. Once you can trace a mesh, this provides the starting points for tracing freeform surfaces like NURBS. The approach is to split a NURBS surface into a set of B-spline patches. Each patch is defines by a coarse mesh of control points and discretised into a finer mesh which we trace with a OBB-tree to get close to the correct intersection. Then employ Newton-Raphson to get the final intersection. Once we can trace NURBS / Spline surfaces, we can then generalize the code to handle anything a CAD model can contain: the ultimate prize!

I'm developing a physical optics modelling package for python, based on the Gaussian Beam Decomposition method (a.k.a. method of Gausslets). It also does conventional geometric ray-tracing. It handles polarisation, dispersion and diffraction. Raypier provides nice model visualisations via VTK and a live GUI where you can "twiddle" with your model parameters. Performance is pretty good, I think, although I've got no points of comparison. Check out the docs at https://raypier-optics.readthedocs.io/en/latest/introduction.html and/or get the code at https://github.com/bryancole/raypier_optics

If you structure the program similar to this then you can get scaling that is basically proportional to the number of cores; and that continues all the way up to a 2x6248R 48 core system, which is ~44 times faster with all of the cores vs 1 core.