[CP2K-user] [CP2K:15563] Hirshfeld-I charges with higher precision

hut... at chem.uzh.ch hut... at chem.uzh.ch
Wed Jun 16 08:44:37 UTC 2021


I don't have experience with changing that parameter. You will have
to do the tests. Probably there will be a relation to the integration
accuracy on the grid. I would suggest to go for larger cutoffs, but
also this will have to be tested.


Juerg Hutter 
Juerg Hutter                         Phone : ++41 44 635 4491
Institut für Chemie C                FAX   : ++41 44 635 6838
Universität Zürich                   E-mail: hut... at chem.uzh.ch
Winterthurerstrasse 190
CH-8057 Zürich, Switzerland

-----cp... at googlegroups.com wrote: -----
To: "cp2k" <cp... at googlegroups.com>
From: "Nicklas Österbacka" 
Sent by: cp... at googlegroups.com
Date: 06/14/2021 03:50PM
Subject: [CP2K:15563] Hirshfeld-I charges with higher precision


I am planning a project on constructing a machine learning potential based on CP2K calculations, and as part of this I need access to dependable partial charges. Self-consistent Hirshfeld with SHAPE_FUNCTION DENSITY seems to be the best option in CP2K as I would like to avoid post-processing of potentially hundreds of gigabytes of data.

Digging through the source code, it seems that the comp_hirshfeld_i_charges routine proceeds until the total charge residual is less than 1e-2. I am not sure this is enough to train a reliable machine learning model, however, so I would like to change it to something stricter to be on the safe side. Is this safe to do, or could that lead to e.g. numerical instabilities? Changing it to 1e-4 or 1e-5 would probably be enough for my purposes.

All the best,
 You received this message because you are subscribed to the Google Groups "cp2k" group.
 To unsubscribe from this group and stop receiving emails from it, send an email to cp... at googlegroups.com.
 To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/a445f21e-5220-4921-b9b3-36493ab0cf56n%40googlegroups.com.

More information about the CP2K-user mailing list