Bill Swope reported getting very different energies for iodine-containing molecules when using scf_type=df (our default) as opposed to direct, pk, out_of_core. These energy differences are large (40 hartrees/10,000 kcal/mol) and correspond to somewhat different geometries.
PB reports that, for single point energies for for tri-iodide using different settings
scf_type=pk took about 11 sec
df took about 6 sec
direct took 60 sec
(This discussion will be open-ended. We’re also planning to consult with Ben Pritchard/Lee Ping/Daniel Smith at our next submission meeting)