sphuber
commented
3 years ago Thanks for the detailed post @zooks97 . Note that although the ElectronicType enum already has an AUTOMATIC constant, we haven't really added broad support for all plugins, so most only support METAL or INSULATOR. This is because we haven't decided yet if AUTOMATIC should mean that the input generator should simply select slightly different parameters, or the workchain should actually take charge to run separate initial calculations to try and determine the character and then use that for the rest of the workchain. The latter is of course non-trivial and potentially costly, and for the former it can be doubted whether that really constitutes "automatic".
Nevertheless, applying the other changes make a lot of sense and I would apply them to the code and text before publication. I will now look at the PR.
giovannipizzi
flavianojs
ltalirz
After receiving feedback from the Abinit developers, we have decided that it is necessary to change some defaults in the Abinit protocol.
The current defaults for smearing type are:
occopt = 3) with additional bands (fband = 2.0)occopt = 1) with fewer additional bands (fband = 0.124)The suggested changes are:
occopt = 5) with additional bands (fband = 2.0)occopt = 7) with additional bands (fband = 2.0)occopt = 1) with additional bands (fband = 0.125)The Abinit developers strongly advise using Gaussian smearing instead of Fermi-Dirac smearing in general. In the case of unknown electronic type, the use of MV cold smearing is not advised as it can give occupations which do not uniquely define a Fermi energy when using the bisection method. However, MV cold smearing is advised for metals as it gives the lowest error in total free energy.
For these reasons, in the case of unknown electronic type, we propose to use Gaussian smearing; in the case of known metallic electronic type, MV cold smearing; and in the case of known insulating electronic type, fixed occupations.
I will soon make a pull request which implements these changes. However, if we want to include these choices in the paper, we need to re-run the Abinit calculations, slightly modify the text of the paper, and re-generate many of the figures.