August w5 2024

Since I started this project, our largest bottleneck on material performance has been a weighted centroidal voronoi diagram algorithm that works on arbitrary meshes. We relied on an algorithm developed several years ago for a long time, but it has been incredibly memory-hungry and slow. This week, we found a far-improved replacement (ACVD), which has almost completely replaced the old one at this point. Its speed and resource efficiency means we can now use gradient descent to optimize our voronoi meshings, rather than a montecarlo-based method. We also took care of some legal and FDA related work.