Keyword explanation: ON_THE_FLY_LOC

[Nicholas Winner]

Would anybody well versed in CP2K be able to provide a brief explanation of the ON_THE_FLY_LOC keyword, found inside the OT section. 

It says that it performs on the fly localization of the MOs, but nothing more. Does it help with convergence and/or numerical stability? It is also listed as only being compatible with the IRAC method, rather than the strict orthogonality algorithm.

Any explanation would be appreciated.

Best,

-Nick

[hut...@chem.uzh.ch]

Hi

this was implemented in order to have improved efficiency in MD
when Wannier functions are needed at each time step.
It was also thought of a first step towards orbital based linear
scaling methods.

It adds an additional loop at each SCF step to enforce locality
of MOs.

regards

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

-----cp...@googlegroups.com wrote: -----
To: "cp2k" <cp...@googlegroups.com>
From: "Nicholas Winner"
Sent by: cp...@googlegroups.com
Date: 11/03/2021 12:21AM
Subject: [CP2K:16164] Keyword explanation: ON_THE_FLY_LOC

--
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 cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/80dac099-3106-4eeb-86e2-0043b88f281bn%40googlegroups.com.

[Thomas Kühne]

Dear Nick, 

ON_THE_FLY_LOC is the implementation of the following paper: 

https://aip.scitation.org/doi/full/10.1063/1.2828507

In the spirit of previous works of Thomas, Iftimie and Tuckerman, 

the orbital coefficient matrix of sought to be localized on-the-fly 

during the simulation. Contrary to the density matrix P, C is much 

more compact (MxN vs. MxM) but less sparse. This does not 

improve the stability/numerical stability, I would even argue quite 

the contrary. It’s one step towards efficient linear-scaling. In the 

limit of the corresponding localized orbitals MAY be equivalent to 

MLWF, but I conjecture in a numerical calculation there should be 

differences. 

To my understanding there is no specific reason for the combination 

with IRAC, except for that this had been implemented by the very 

same persons around the very same time: 

https://aip.scitation.org/doi/full/10.1063/1.2841077

Cheers, 

Thomas

--
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 cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/80dac099-3106-4eeb-86e2-0043b88f281bn%40googlegroups.com.

==============================

Thomas D. Kühne

Dynamics of Condensed Matter

Chair of Theoretical Chemistry

University of Paderborn

Warburger Str. 100

D-33098 Paderborn

Germany

thomas...@upb.de

+49/(0)5251/60-5726