Parameter choice for GAPW

[Ali Akbari]

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: 

https://groups.google.com/forum/#!msg/cp2k/RUFQScjSDn0/At_pJzz49_oJ

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

[Marcella Iannuzzi]

Hi Ali,

Yes you can use pseudo potentials also with GAPW, and you can mix species with pseudo potentials and species all electrons.

Obviously, the PP species must have a basis sets optimised for PP calculations.

These most probably contain functions with relatively small exponents, which means that most of them are going to contribute only to the soft charge density.

The partitioning in soft and hard terms is carried out in exactly the same manner as for a full all electron calculation. 

Since it is still a GAPW calculation, all GAPW parameters are still needed. 

Cheers

Marcella

[Ali Akbari]

Hi Marcella,

Many thanks for the quick reply. But I still have some questions. I do apologise in advance for asking too many questions, but you have a 

deep knowledge about the code & the implemented formalism and nobody can answer better than you.  

To give a better context for my questions I describe my system. It is a periodic system (TiO2 surface with an organic dye) and I want

to calculate XAS and XPS. Since (I think) all electron calculations are much heavier, I consider only the target atom with all electron 

and the rest with pseudo potentials. One general question is that how important is to change default values? Are the most important 

parameters the ones you wrote in the earlier post 6 years ago?

   

My details question are:

ALPHA0_HARD and EPS_3C_REDUCE

Compare to the previous discussion 6 years ago, It seems to me that  ALPHA0_SOFT is removed and ALPHA0_HARD default value is now 0 instead of 10. What value do I need to choose? What should I do for EPS_3C_REDUCE?

EPS_PPL 

EPS_PPL  default value is 0.01. Isn't it too rough?  

EPS_DEFAULT

It's written in the CP2K manual,  by setting EPS_DEFAULT, CP2K will all EPS_xxx variable to get an energy correct to up to EPS_DEFAULT. So how crucial is to set the variables EPS_******? Is there any EPS_**** that is not control by  EPS_DEFAULT?  

LMAXN0 and LMAXN1:

You wrote: The maximum possible value  is 2 times the maximum angular momentum present in the basis set.

Which basis set you refer to and how can I determine it (different atoms can have different basis sets)?

LMAXN1 default value is -1. What does it mean?

QUADRATURE:

There are three algorithms available. What is the criterion to choose?   

HARD_EXP_RADIUS and EPSFIT:

I have a C-H bond (length 1 A) in my system. The default values of HARD_EXP_RADIUS  are 0.6 A and  0.80 A for H and C respectively.

I treat C with all electron (aug-cc-pV5Z) and H with TZV2P-MOLOPT-GTH basis set. From your answer, I understand that I should choose smaller

values for  HARD_EXP_RADIUS.  What is your suggestion in this case? How should I choose an optimal value for situation like this?

Do I need to change  EPSFIT too?

LEBEDEV_GRID and RADIAL_GRID:

Both default values are 50 but you wrote before that 100 points on the radial grids, and 50 on the angular. Do they change if have a

mixture of all electron and pseudo potentials?

Thank you very much and cheers,

Ali

[Marcella Iannuzzi]

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:

[Ali Akbari]

Hi Marcella,

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

All the best,

Ali