How to understand/tune parameters for GAPW calculations

[Natalie Holzwarth]

Dear cp2k developers,

      We are very much looking forward to possibly using CP2K  particularly using the GAPW scheme to extend our calculations which up to now have been performed using ABINIT and QUANTUM ESPRESSO using PAW.  We have tried a number of your tutorials and test inputs, but are have a hard time understanding how to properly pick the parameters.   Is there perhaps a writeup somewhere that describes the detailed implementation or perhaps a list of the main subroutines that a user should study?   Or perhaps someone might have time to answer some the following silly questions.

1.  These questions are based on the notion that we should use the all-electron bases since these would allow n^1 to be represented analogously to Blochl's PAW formulation.   Is this the correct way to think about it?

2.   According the paper by Lippert et al (1999), the soft wavefunctions and densities are chosen by eliminating the Gaussian basis primitives with too large an exponent.   Which of the input parameters determines the soft basis parameters?

3.   In addition to the soft density, Blochl's PAW scheme has a local pseudo potential.    I am not sure what is the analogous local pseudo potential for GAPW and which parameter controls it. I have assumed that we should use "ALL" potentials in the input files.   Is this correct? 

4.   Some of the examples use the following parameters which seem to be important, but I am not sure how to choose them

MAX_RAD_LOCAL

ALPHA0_H

HARD_EXP_RADIUS

5. Is there a provision/recommendation to use the frozen core approximation and if so, what parameters control that?

Thanks in advance for any suggestions you can provide and much gratitude for sharing and maintaining the cp2k software package.

Sincerely,   Natalie Holzwarth

[Krack Matthias (PSI)]

Dear Natalie

 

I will try to answer your questions:

 

1)      Yes, I think that’s right when using an all-electron potential.

2)      EPSFIT is the key input parameter to define the split of the basis set into a soft and a hard part. The default value of 1.0E-4 is too large for most systems. I would rather consider values of 1.0E-6 for EPSFIT to ensure that the soft basis includes sufficiently large exponents, e.g. if the largest exponent in the soft basis set for oxygen is only 4 this will give poor results.

3)      GAPW splits the basis set into a hard (frozen) and a soft part. This can be applied for all-electron calculations but also for hard pseudopotentials. There is no additional pseudopotential.

4)      In a first step, you can use the default values for these input parameters. More import is EPSRHO0. Consider smaller values than the default one for better accuracy. The electron counts printed (use PRINT_LEVEL medium) during and after the SCF procedure can serve as an accuracy check. In addition, the parameters defining the local atomic (Lebedev) grids (see &KIND section) might need larger values in some cases, e.g. for heavier elements.

5)      See above (EPSFIT)

 

HTH

 

Matthias

--
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/e59f80e3-1dda-4bb4-b969-93b783399ad9%40googlegroups.com.

[hut...@chem.uzh.ch]

Hi

The GAPW method in CP2K follows the basic idea of PAW by Peter Bloechl,
but has also some special features.

In the current implementation in CP2K

- there is only one augmentation charge, meaning hard and soft augmentation
charge from the original scheme are the same. This allows to drop all the
additional three-center correction terms in the Coulomb part. It also means
that the minimal cutoff needed in the soft density is related to this charge,
typically about 200 Ry.
Bloechl's 'local potential' is not needed in this scheme.

- There are no frozen electrons/states. All electrons explicitly in the calculation
are optimized. However, you can use GAPW also with pseudopotentials, as well as
in mixed calculations with all-electron atoms and pseudopotential atoms.
Atoms with "POTENTIAL ALL" need a corresponding all-electron basis set, atoms with
a pseudopotential a corresponding pp-basis set.

- The basis sets used to expand the soft/hard densities are derived from the input
orbital basis primitive functions. Functions with large exponents are in the hard set,
functions with small exponents are in the soft set. The important input variable to
control this are
QS / EPSFIT and QS / EPSISO

EPSFIT default is 10^-4. A larger value enlarges the PAW radius and more primitive are
considered in the hard set.

EPSISO is used to determine if a primitive is completely contained in the PAW sphere.
Such functions are their own projectors and are handled separately.

- There are many more input variables that allow to fine tune the calculations, but
experience shows that most of them are hardly ever used. Besides the ones mentioned above,
I find it necessary sometimes to tighten the default values for the one-center integrations.
This is specified in the KIND section for each atom using
KIND / LEBEDEV_GRID and KIND / RADIAL_GRID

I hope this helps to get started.

regards

Juerg Hutter


--------------------------------------------------------------
Juerg Hutter Phone : ++41 44 635 4491
Institut für Chemie C FAX : ++41 44 635 6838
Universität Zürich E-mail: hut...@chem.uzh.ch
Winterthurerstrasse 190
CH-8057 Zürich, Switzerland
---------------------------------------------------------------

-----cp...@googlegroups.com wrote: -----
To: "cp2k" <cp...@googlegroups.com>
From: "Natalie Holzwarth"
Sent by: cp...@googlegroups.com
Date: 04/14/2020 05:16PM

Subject: [CP2K:13128] How to understand/tune parameters for GAPW calculations