The implementation of bogoliubov for a general fermionic gaussian that doesn't mix spin up and spin down is broken

[wjhuggins]

When bogoliubov_transform is called with a transformation_matrix that has shape N x 2N the check that determines whether or not the transformation mixes spin up and spin down and the subsequent code is broken. The current check (_is_spin_block_diagonal) and some of the subsequent code is only valid for a square transformation_matrix.

To reproduce:

Call bogoliubov_transform with the following arguments:
qubits = [cirq.LineQubit(0), cirq.LineQubit(1)]
trasformation_matrix = np.ndarray([[-9.57167901e-01-2.89533435e-01j, 0.00000000e+00+0.00000000e+00j,
-3.33066907e-16-5.55111512e-17j, 0.00000000e+00+0.00000000e+00j],
[ 0.00000000e+00+0.00000000e+00j, 1.00000000e+00+0.00000000e+00j,
0.00000000e+00+0.00000000e+00j, 0.00000000e+00+0.00000000e+00j]])

[wjhuggins]

It can be circumvented by adding the following code to _is_spin_block_diagonal(matrix)
if matrix.shape[0] != matrix.shape[1]:
return False

It should be possible to implement properly for a general fermionic gaussian transformation though.