hdf5 outputs c-contiguous numpy arrays
ezfio assumes arrays are fortran-ordered
np.view can be used to get re,im parts as floats with doubling of one dimension
(last for c-contiguous, possibly first for f-contiguous?)
working on changing the converter to minimize transposing, reshaping, taking re/im parts, stacking, etc.
should loop over union of two sets of integrals:
set 1:
i<=k
j<=l
ik<=jl
set 2:
i>k
j<l
ik<=jl
looping over kpts in same way is incorrect
here I've relaxed the constraints over kpt indices, while keeping those over orbital indices
There is probably a more efficient way to do this where we have more kpt constraints and additional logic in orbital loops
instead of real/imag parts read separately, use ezfio to read/write complex arrays with extra dimension of size 2
converter needs to be tested (might need to transpose some axes in arrays)
converter has extra garbage that needs to be removed after testing