Experimental
We fixed issue #2 (closed) by running the algorithm twice and checking the stable eigenvalues. The slow performance part is a limitation of the method -- the optimal number of eigenvalues is \sim \sqrt{M} for a M \times M matrix.
We fixed issue #2 (closed) by running the algorithm twice and checking the stable eigenvalues. The slow performance part is a limitation of the method -- the optimal number of eigenvalues is \sim \sqrt{M} for a M \times M matrix.