I gave a definition of geometric complexity using natural logs; Graham suggested log base 2 instead. He also proposed taking the dth root of this, where d is the codimension of the wedgie--that is, the number of commas used to define it.
If G is the Graham geometric complexity under this definition, R is (rms or minimax, etc.) error, and n = pi(p) is the number of primes in the p-limit we are looking at, the formula for geometic badness now becomes: B = R Gn, which is pretty nice. Does anyone have a concern about switching definitions to the Graham version, which looks to me like a good idea?