[CP2K-user] SCF_GUESS: using restart with atomic guess for additional atoms

[Sergey Chulkov]

Hi Fabian

You can find a basic implementation of this feature in cp2k 7.1 as a branch 
'v7.1.0-atomic_restart' here 
<https://github.com/schulkov/cp2k/tree/v7.1.0-atomic_restart>. In order to 
use atomic guess together with a previously written restart file, please 
append new atoms to the end of your &COORD section or xyz file, and then 
enable this feature as 'SCF_GUESS atomic_restart'.

Many features available for the regular restart guess have not been ported 
yet for the hybrid atomic/restart guess. Among them are RTP wave-functions, 
k-points, the ability to reduce the size of the basis set and to change the 
number of spin-components of the wave-function. For a complete 
implementation there should be a way to recompute the electron density on a 
3D grid by using MOs from the restart file. In order to do so, original 
atomic positions and the basis set need to be known. One option could be to 
save this information into the wave-function restart file along with MO 
coefficients. In addition to the generic atomic/restart guess, it will also 
allow to reorder atoms in &COORD section. The problem is that it will break 
backward compatibility.

Sergey

On Monday, February 3, 2020 at 10:12:03 AM UTC, Fabian Ducry wrote:
>
> Dear all,
>
> I have a substrate (several 100 atoms) and want to iteratively add atoms 
> onto its surface. Because the number of atoms changes every step, I have to 
> minimize the charge density from the atomic guess each time. To speed-up 
> the calculations I would like to re-use the previous restart wave function 
> (the substrate is identical and the new atom rather far from the surface). 
> Is there an option to mix the restart wave function with atomic orbitals to 
> build the initial guess for the energy minimization?
>
> If I set FORCE_EVAL%DFT%SCF%SCF_GUESS%RESTART
> The following warning appears:
>
> READ RESTART : WARNING : DIFFERENT natom, returning 331 330
> Atomic guess: The first density matrix is obtained in terms of atomic 
> orbitals
> and electronic configurations assigned to each atomic kind
>
> How can I tell CP2K to reuse the previous wave function for all except for 
> the last atom?
>
> Regards,
> Fabian
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20200210/8cff8062/attachment.htm>