Well, yeah, but the whole question of extending this algorithm to arbitrary (weighted) graphs is still very interesting: it seems to rely on some kind of discre

Well, yeah, but the whole question of extending this algorithm to arbitrary (weighted) graphs is still very interesting: it seems to rely on some kind of discrete [1] version of parallel transport. [1] http://ddg.cs.columbia.edu/ http://ddg.cs.columbia.edu/