https://youtu.be/QH4MviUE0_s
@tom7 I'm sure you thought about this and there's some reason it's not true or not interesting or not worth talking about in the video:
For the Rupert solver, it seems like you can "trivially" solve for the best, say, ty and have 1 fewer dimension to sample. Maybe even solve for tx&ty (although that's not obviously trivial to me).
@tom7 Yeah, I definitely meant strictly for the search (which is maybe kind of irrelevant now that the conjecture is disproven).
The main observation is just that if you were just testing random samples from the distribution (not running an optimizer), given a convex hull, solving one translation axis is incredibly cheap (e.g. for each point, compute distance from hull along that axis). But that may not translate into wins outside that specific scenario.
@tom7 I had already heard about the Austrian solution a couple days ago. Yet somehow was still surprised by the sad end of your video.
Now I want to see the Disney ending. :-)