Parameter choice for GAPW

Ali Akbari aliakb... at
Tue Jul 1 13:36:14 UTC 2014

Hi Marcella,

Thanks a lot for your time & reply.  I will tune the parameters with the 
help of your guidelines.   

All the best,

On Monday, 30 June 2014 18:37:17 UTC+3, Marcella Iannuzzi wrote:
> Hi Ali, 
> As already said in the previous reply, by applying the GAPW scheme, the 
> density is going to be partitioned in soft and hard terms also when PP are 
> used. 
> The large number of parameters that could be tuned might be of some help 
> in difficult cases, but often the default values are a good first choice.
> Which basis set functions are going to contribute to the local densities 
> depends on the exponents of the primitives. Actually, the threshold is 
> determined by the EPSFIT parameter. Anyway, I would use the default for 
> that.
> The compensation charges are described in terms of an expansion in 
> Gaussian terms with the exponent alpha0_hard. Also in this case the default 
> value should be ok. 
> The 3 center terms are not computed anymore. From our tests it turns out 
> that there is no loss in accuracy by imposing the soft and hard local 
> compensation charges to be the same. This cancels the 3c terms. Hence the 
> eps_3c_reduce is not used anymore. It is a left over and I am going to 
> remove it from the code. 
> From the input reference manual should be clear which eps_*** are affected 
> by eps_default. These are all parameters used by both gpw and gapw in the 
> same way.
> The  lmaxn1 is determined automatically from the GTO basis sets that you 
> have chosen. This is the meaning of -1
> Setting this from input to a smaller value would reduce the accuracy and 
> maybe improve in performance. 
> I wouldn't do it, if not necessary. 
> The quadrature schemes are equivalent. 
> The number of points for the local grids can affect the accuracy. But 
> normally reasonable grids are already obtained with lebedev between 50 ad 
> 80 and radial between 100 and 200. Depends on the element and basis sets 
> anyway. 
> You can always check whether the results are converged with respect to 
> these parameters, by modifying them. This should give you a feeling on how 
> sensitive the calculation is to them.
> Kind regards
> Marcella
> On Saturday, June 21, 2014 1:23:47 PM UTC+2, Ali Akbari wrote:
>> Hi,
>> I appreciate if someone tells me is it possible to use pseudo potential 
>> with GAPW
>> treatment? I was reading the following discussion to find the necessary 
>> parameters: 
>> What changes if I have a mixture of pseudo potential for some atoms and 
>> all
>> electron for others? e.g. Do I still need to choose the  HARD_EXP_RADIUS
>> such that there is no overlap between a pseudo atom and all electron atom?
>> Many thanks and cheers,
>> Ali
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the CP2K-user mailing list