Hi,<p>I help run an academic conference on Nonimaging Optics (optics for optimal energy transfer in solar concentrators or illumination systems). This year, we are hosting a design challenge, and I thought it could be fun for some in the HN crowd to participate. Maybe some of you have ideas for completely different design approaches than what we typically see on the field.<p>The challenge: Design an efficient "solar concentrator" for a sun shaped like the letters "XX" (instead of our regular round sun). The challenge is set up so that the ideal geometry would be 100% efficient*. However, we don't know how to design such an ideal geometry, and the simulated mirrors are only 95% reflective so that all submissions will be <100%. Highest simulated efficiency wins.<p>I made it possible to test any design directly in the submission portal, so you can participate without buying expensive licenses for optical design software.<p>In general, if you want to make a difference to solar concentration or illumination optics, there are many low-hanging fruits in the field and a lot of room for crossover from the HN and CS crowd**.<p>Stack:<p>- Astro.js for the main static website (I've been away from frontend for about a decade, and was pleasantly surprised to see what can be done with a static site generator nowadays)<p>- SolidJS for the competition portal<p>- Three.js for visualization<p>- Three-mesh-bvh for raytracing and evaluating the submissions (currently a bit slow because I haven't moved it off the main render thread on CPU)<p>- OpenCascade.js for loading and meshing of uploaded step-files<p>* See <a href="https://news.ycombinator.com/item?id=41391715">https://news.ycombinator.com/item?id=41391715</a> for an explanation of why there is a limit to how strongly you can concentrate sunlight<p>** As an example, the commercial design tools in the field support optimization-based design, but they all do raytracing on the CPU and use finite differences instead of reverse mode autodiff. Therefore, the conventional wisdom is that optimizing systems beyond a few hundred parameters is hopeless.