vdW-D4: Switch to smooth cutoff - #7810
Conversation
|
The Integration Test workflow is expected to fail for now because I did not update the reference yet. I'd also set the PR to draft until the smooth width value is made clear upstream. [Edit: it's surprising that tests passed; looks like the change doesn't reach the threshold.] |
310d101 to
2f57e93
Compare
|
From upstream response they are willing to treat the 0.05 Bohr as a reasonable default. So the next question would be how we treat the cutoff width: as an internal logic, or expose it to users to allow adjustment? (my wish would be the latter, the only concern is how we can explain it well in the documentation) |
|
I will add two new input parameters, |
|
Side note: s-dftd3 package has marked |
eaef6cf to
2126618
Compare
In dftd4/dftd4#377 upstream has discussed about enabling smooth cutoff that was introduced since 4.2.0 by default in the future releases. This will be important for PBC calculations as the images close to the cutoff radius will be affected; in principle the result from smooth-cutoff path will be better as smooth cutoff resolves some gradient/virial issues.
Since v4.2.0 is already the minimum required version of dftd4 for ABACUS, it's rather straight-forward to change this calculation type. However, the value setting of
d4_smooth_cutoff_width_2andd4_smooth_cutoff_width_3might not be clear. I may leave the related discussion upstream.