> I don't recall if I changed norm's implementation, and if so I don't > recall why. But we should have a less inefficient implementation. Is norm required to be the euclidian norm? I would think that it is faster to compute the maximum norm.