ABINIT Discussion Forums

In the release notes for the 7.0x series it says that one of the big changes is that Abinit now uses k-centered wavefunctions from _WFK files rather than the gamma-centered sets and hence the mpi-io bottleneck is gone. However, I can't see where this is documented - are we still supposed to generate KSS files (Why!?) for subsequent screening calculations, and then onwards for the self-energy?

I'm asking because when I try to run tutorial 1 (GW) for abinit 7.21 in parallel, it tells me that paral_kgb must be 0 and that accesswff must be 0 or 3, not 1. So, the KSS files can't be written using mpi-io as usual. It works fine in serial, of course.

Hi,
You don't have to generate KSS anymore. Suppose you make a 4 step calculation.
dataset 1: Density
dataset 2: Wavefunctions
dataset 3: Screening
dataset 4: Sigma

Insted of using getkss, you should tell the code to read the wavefunctions from disk at steps 3 and 4, with


That means that the code is expecting an input data file. You could make two symbolic links, e.g.
idat_mycalc_DS3_WFK
idat_mycalc_DS4_WFK
that both points to the output data file of the second dataset, e.g.
odat_mycalc_DS2_WFK

That's it!
And the Abipy python module can ease the boring task of linking the files.