Choosing cutoff for AIMD

[Ramanish Singh]

Hi,

I have a question regarding the plane wave CUTOFF in the MGRID section. I am doing NVT MD simulations with a liquid box.

For deciding the CUTOFF value do I need to run the MD simulations at different values of CUTOFF and then decide the appropriate CUTOFF based on evolution of energy with time for different CUTOFF values

OR

a single point energy of the system (RUN_TYPE ENERGY) is fine? (as given here: https://www.cp2k.org/howto:converging_cutoff )

Please also give any relevant references for the same if available.

Thanks,

Ramanish Singh

[Patrick Gono]

Dear Ramanasish,

You are attempting to do ab-initio molecular dynamics. At each step, the forces are calculated by performing a single point energy calculation of your system. If your CUTOFF is too small, the single point calculation leads to the wrong electron density, and hence your forces will also be wrong.

Choosing CUTOFF, you only need to make sure that your energy in a single point calculation is converged. To be completely clear: the latter of the two options you propose.

In general, look at the highest exponent of the basis you use. For example, if your system contains oxygen, and you use the double-zeta MOLOPT basis set:

 O  DZVP-MOLOPT-SR-GTH DZVP-MOLOPT-SR-GTH-q6
 1
 2 0 2 5 2 2 1
     10.389228018317  0.126240722900  0.069215797900 -0.061302037200 -0.026862701100  0.029845227500
      3.849621072005  0.139933704300  0.115634538900 -0.190087511700 -0.006283021000  0.060939733900
      1.388401188741 -0.434348231700 -0.322839719400 -0.377726982800 -0.224839187800  0.732321580100
      0.496955043655 -0.852791790900 -0.095944016600 -0.454266086000  0.380324658600  0.893564918400
      0.162491615040 -0.242351537800  1.102830348700 -0.257388983000  1.054102919900  0.152954188700

The highest exponent is 10.389... in the first row of values, on the very left.

Now assuming this is the highest among all elements present, take this number and multiply it by 40. That should give you some idea of the appropriate value of CUTOFF, in this case CUTOFF = 400. Run a few ENERGY calculations with values, say, between 350 and 600, and check that the energy is, indeed, converged.

Yours sincerely,

Patrick Gono

--
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/722d97d6-2566-4562-a571-c38dec41ed81%40googlegroups.com.

[Ramanish Singh]

Thank you so much Patrick and Travis! 

On Wednesday, December 4, 2019 at 10:19:46 AM UTC-6, Travis wrote:

Hi,

Just to add, the 40 referenced in "Now assuming this is the highest among all elements present, take this number and multiply it by 40" should be set to whatever REL_CUTOFF is. I typically use a value of 50 for static cell calculations and 80 for variable cell ones. These figures are based on my experience and from literature on condensed phase simulations. Using the basis set listed in Patrick's post, this would give cutoffs of 520 Ry and 850 Ry with some rounding up to 'nicer' numbers. The point of this is to ensure that the tightest orbitals make it to the finest grid. Importantly, this assumes you don't use the smoothing function in https://manual.cp2k.org/trunk/CP2K_INPUT/ATOM/METHOD/XC/XC_GRID.html.

-T

To unsubscribe from this group and stop receiving emails from it, send an email to cp...@googlegroups.com.