Kpoint rounding error in RPA
Sometimes we get "gammapoint" coordinates of [1e-17, 0, 0]
. These produce wild numbers in RPA calculations.
Basically qpd.optical_limit
uses np.allclose()
to decide whether gamma or not whereas qpd.kd.gamma uses .any()
.
We should be able to fix the problem by always using qpd.optical_limit
and then add a test ensuring that very-near-zero differences are handled the way we like.
Thanks to Elizabeth (?) for reporting this bug.