*This example is courtesy of James Theiler of Los Alamos National Laboratory*
* 2010-03-22 *

This is a movie of a 224x224 matrix of (correlation cofficients)^2
for AVIRIS data decorrelated with a sequence of K SMT rotation, with K<=389.
That corresponds to r = (389/224) = 1.74 rotations per coordinate.
The squared correlation matrix (not the covariance matrix), F, is given by

F_ij = (R_ij)^2/(R_ii*R_jj) for i not equal to j
F_ij = 0 for i = j

It might have been more natural to have the diagonal be equal to 1
instead of 0, but with 0, this the the same matrix we maintain for
deciding which pair ij to rotate and that the new stopping criterion uses.