The inputs of the threenorm problem are points from two Gaussian
distributions with unit covariance matrix. Class 1 is drawn with
equal probability from a unit multivariate normal with mean
\((a,a,\ldots,a)\) and from a unit multivariate normal with mean
\((-a,-a,\ldots,-a)\). Class 2 is drawn from a multivariate normal
with mean at \((a,-a,a, \ldots,-a)\), \(a=2/d^{0.5}\).