[CP2K:5053] Re: restarting from an already optimized structure

Florian Schiffmann flosch... at gmail.com
Mon Mar 10 09:01:41 UTC 2014

With slab boundarie conditions I meant to use periodic XY in the poisson 
solver and use enough vacuum to avoid any strong interactions in the 
periodic dimensions. 
OK, I see how the system will become very big in terms of unit cell.With 
the appropriate hardware there shouldn't be a problem. With limited 
resources you will have to accept that the setting won't be perfect 
(nonperiodic, functional of choice,...). The MT solver requires about 2x 
your system size for the cell dimension. This will be difficult. The 
wavelet solver as it is currently in CP2K requires a cubic cell (more 
precisely a uniform grid spacing in all directions) but only about 3A of 
vacuum at the boundaries. Both will lead to very big grids. For the wavelet 
solver an temporary grid is allocated to match the distribution as required 
by external solver used (adopted from big DFT). This means if it is a 
memory issue that can cause a problem as the temporary memory requirement 
doubles in this steps. For MT I am not quite sure I am afraid but it might 
get away without the memory overhead. Therefore MT might be the better 
In case this doesn't work, you might have to go fully periodic and try to 
adjust the cell that you have minimal interactions with the periodic 
images. In some cases this allows to employ smaller cells than doing the 
calculations in the correct boundaries.

To get away with a lower cutoff there is another option which I haven't 
mentioned yet. As I said BLYP PP's are very sensitive to the cutoff 
(causing problems in the FFT). If you are willing to switch functional to 
PBE there is another option. For PBE we have NLCC-pseudopotentials for the 
first two rows. As they lead to a finite density at the cores the FFT's 
xc-potential is better behaved and a cutoff of 300 should be OK. For Au you 
can simply use a normal PP.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20140310/f7e9730f/attachment.htm>

More information about the CP2K-user mailing list