[CP2K-user] [CP2K:12927] L1SD for wannier localization

hut... at chem.uzh.ch hut... at chem.uzh.ch
Mon Mar 2 10:10:28 UTC 2020


this is related to this paper
A smooth ℓ1-norm sparseness function for orbital based linear scaling total energy minimization
J. Chem. Phys. 128, 064107 (2008); https://doi.org/10.1063/1.2828507
Valéry Weber and Jürg Hutter


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: "Soumya Ghosh" 
Sent by: cp... at googlegroups.com
Date: 02/26/2020 04:25PM
Subject: [CP2K:12927] L1SD for wannier localization

Dear developers,

I have used both L1SD and Jacobi rotation for wannier localization and both methods are giving similar results for my system. I have 2 theoretical questions though:

1. It seems to me that L1SD actually uses L2 norm (f2 in the code) of the eigenvector matrix. Is this true or am I wrong?

2.  It seems to me that this steepest descent (SD) is not related to the SD method eluded by Silvestrelli PRB, 1999, 59, 9703. It seems to me that in L1SD at CP2K the gradient for SD is computed based on the L2 norm of the eigenvector matrix whereas in the SD method of the paper one uses the gradient of the spread functional, thus updating the "z_In" at each iteration. I wonder why the original SD method is not implemented in CP2K. 

Please let me know. Thanks.  
 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/9a5967f9-c3db-4782-8262-821b8e782ab1%40googlegroups.com.

More information about the CP2K-user mailing list