In Section 2 (Design philosophy) the authors state:
“A key challenge is to enable heterogeneous workflows where different parts of a workflow are performed using different computational methods. Such workflows are necessary to take advantage of the range of features implemented in different DFT packages.”
They give the example of using CP2K for a fast hybrid relaxation and VASP for a slower, more accurate one.
However, in Section 4.1.17 (Point defect workflow), the authors write:
“By default, these supercell relaxations will be composed of a less expensive structure optimization step with a PBE functional, followed by a high-quality HSE06 static calculation.”
They then note:
“We note that care must be taken with this approach, as local minima for more symmetric and charge-delocalized states favored by PBE may not be able to be overcome by HSE06 calculations initialized with such configurations…”
The authors acknowledge a serious physical problem, PBE can relax to a qualitatively wrong local minimum that HSE06 cannot escape, but they still prescribe this as the default workflow for point defect calculations, which are already known to be sensitive to initial conditions and exchange-correlation functionals.
This is not just a “user beware” footnote, it’s a fundamental methodological flaw in the proposed automated workflow. If the default pathway can systematically lead to incorrect defect configurations and energies, then the high-throughput data generated using this workflow is not reliable, and the promise of “robust, published methodologies” (Sec. 4 intro) is broken for one of the most scientifically important workflows.