@Akiva I assume you consider, like, a shape like "V" a polygon and what not
You can easily limit to that if you take intersection of nonconvex chaps
Decreasing intersection of polygons of exactly $n$ sides should in general be a polygon of $n$ sides, with degeneracies (which are polygons with certain vertices and edges squished together)
so you also get shapes like polygons with an edge sticking out etc etc etc
Anyhow, Gelfand-Fomin and Bliss are standard good texts, I think (old fashioned, though).
I dunno how you see it's closed that way, @PVAL. I guess I thought I remembered some sort of infinite-dimensional Morse theory argument, but I haven't thought about this in centuries.
@TedShifrin probably could replicate that with a tube-plot-like parameterization using the type I'm referring to but I see what you mean about it not really being the right way to create it or the most straightforward.
@TedShifrin true. I wasn't so much referring to the mathematical usage though as I was referring to the means of creating the 3D model, in which case it might be easier to just make a tube plot than to make a tube plot and a surface of revolution. More code = more bugs. We don't want roaches.
btw, that comment you made yesterday still cracks me up.
He defines 3 invariants of generically immersed plane curves which are invariant under regular homotopies of immersions with no triple points and self-tangencies, by defining them for the standard turning number immersion with the least number of self-intersections and defining the jumps over triple intersections and self-tangencies.
Arnol'd did a lot of stuff with singularities of maps. (McCrory and I wrote a JDG paper, as it turned out, doing all the technical details that Arnol'd skipped over. And making the Chern class arguments clearer, too.)