Comment by alanbernstein

I would expect the energy minimization approach to perform better than anything else, for vectors distributed uniformly on the surface of the N-sphere. Why go with the Fibonacci sphere instead?