[Pw_forum] some naive problems
Paul M. Grant
w2agz at pacbell.net
Fri Nov 2 21:40:14 CET 2007
Just to add to this general discussion...which I think is very useful...when
I take up a new problem I write two scf scripts, one which increments the
size of the Monkhorst-Pack (K_POINTS) net until the total energy "sort of"
settles down (I say "sort of," because sometimes there are small
oscillations when the net gets big. Then I write another script which takes
the "best" MP size increments the cut-off upward until I'm happy. If I
think I need to find new "equilibrium" lattice constants, I'll write a third
script which varies these. Then I use all this to start computing
observables. I don't think I've ever used the gamma option.
Re psuedopotentials: I pretty much trust the ones from the PWscf library
and/or from Vanderbilt.
On the "naïve" side. Most of the systems I investigate are far from cubic
symmetry, and I choose the number of K_POINTS in each direction
proportionately to reflect the way the dimensions of the Brillouin zone vary
inversely with those of the Wigner-Seitz cell. What do others do? Also,
I've never found adding an "offset" to the MP net has a significant
effect...when should I expect differently?
-Paul
Paul M. Grant, PhD
Principal, W2AGZ Technologies
Visiting Scholar, Applied Physics, Stanford University
EPRI Science Fellow (Retired)
IBM Research Staff Member Emeritus
w2agz at pacbell.net
http://www.w2agz.com
-----Original Message-----
From: pw_forum-bounces at pwscf.org [mailto:pw_forum-bounces at pwscf.org] On
Behalf Of Stefano de Gironcoli
Sent: Friday, November 02, 2007 3:36 AM
To: PWSCF Forum
Subject: Re: [Pw_forum] some naive problems
Dear Lihui Ou and Osman Baris Malcioglu,
let me re-state Osman's reply in a way I like more...
Plane-wave codes (and for that matters also LMTO, FLAPW codes etc)
are built so as to describe periodic systems and CAN deal with isolated
molecules by the use of supercells large enough that the periodic images
do not interact significantly. There are other codes (I guess NWChem is
one of them) that work for isolated system only and CANNOT describe
periodic systems. This may be a problem when wishing to compare isolated
and periodic systems on the same footing...
A plane-wave basis-set is an expensive one but is unbiased (describe
with the same accuracy every point in space) and its accuracy is
extremely easy to control (just increase your cutoff energy until the
property you are interested in converges). A lot of effort in the past
has been put in the development of pseuodopotential theory that allows
to treat any element in the periodic table with good accuracy (some
already using norm conserving pseudos while some others may need
ultrasoft ones). Pseudopotential method is not an all-electron method
but very often the results are very close to AE and always, to my
experience, physically meaningful. An evolution of the US
pseudopotential method is the PAW formalism that can be considered an
all-electron method. We are currently implementing this formalism in PWscf.
Localized basis (such as gaussian or slater-type-orbitals) are more
compact but are more difficult to control and require a lot of know-how
and experience (double-z ? triple-z ? triple-z+polarization ?) in order
to avoid basis-set superposition error and the like.
Parameters controlling plane-wave calculations:
1) kinetic energy cutoff: as said it defines the dimension of the
basis set and you just need to increase it until you are satisfied. The
required cutoff is a PROPERTY OF THE PSEUDOPOTENTIAL USED not of the
particular system under study. It stems from the need of describing
accurately the pseudopotential wavefunctions.
2) k-point sampling : it's a property of your system (insulator,
metal, isolated, 2-d, 1-d) not particularly of the pseudopotential
used. For isolated molecules (that is if the supercell is large enough
and hence the IBZ small enough) Gamma-only sampling (for which special
tricks to speed up the calculation can be used) is a good choice. If it
is not the case, this means that periodic images are interacting...
then increase the cell dimension and stick to Gamma sampling.
As for Hybrid functionals: currently only NC pseudopotentials are
implemented, sooner or later US (or most probably directly PAW)
implementation will follow... any skilled volunteer
would be welcome.
hope this helps,
stefano de Gironcoli, SISSA and DEMOCRITOS
mbaris at metu.edu.tr wrote:
> Dear Lihui Ou,
>
>
>> Do pwscf only periodic hybrid density functional theory calculation,
>> it could do any other general density functional theory calculation?
>>
> I am not sure if I got the point of your question correctly, but Plane
> Wave SCF program can only solve systems thorough plane wave
> approximation, i.e. not through an all-electron calculation, and for
> simplification purposes (that is, in order to have tangible Fourier
> transforms etc.) periodicity/symmetries are employed. This is more than
> acceptable for crystals, but for macromolecules etc, you should be
> extra careful setting the boundaries (the most simple limit is this: a
> very big box surrounding the molecule you are interested in will
> somewhat approximate a standalone molecule, however the size of the box
> is limited by a combination of how the plane wave approximation is
> employed in pwscf and bare computational limits, you need to find the
> optimum value). There are other packages available such as NWChem or
> Gaussian for all-electron calculations. Pseudo potential wise, you can
> create a pseudo potential to your desire, as long as you can validify
> your results. Please see the documentation.
>
>
>> Is there any principle about setting of k-point and cut-off energy?
>>
>
> Well, yes. The idea is: You should minimize the number of k-points and
> cut-off (and a number of other parameters depending on the system you
> are studying) in order to save computer time whilst maintaining
> physically meaningful results. The "physically meaningful result" part
> depends on your problem, keep in mind that total energy in DFT is not
> very well defined, due to nature of Kohn-Sham orbitals, only the
> differences in energy are trustworthy, or the observables, such as unit
> cell volume. Generally you are advised to do a number of calculations
> with varied parameters in order to obtain a converged set before
> starting the problem itself (i.e. if you are using lattice parameter as
> your observable, find the lattice parameter that minimizes the energy
> for each ecutwfc-k-point combination. You will see that after a certain
> limit, your lattice parameter will not change much, that value is the
> minimum you can use, you may later need to increase it, depending on
> your problem).
>
>
>
>> Best wishes
>> Lihui Ou
>>
>
> Hope this was helpful,
>
> Best,
>
> Osman Baris Malcioglu
> Ph.D. Candidate
> METU, Physics
> Ankara, Turkey
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>
_______________________________________________
Pw_forum mailing list
Pw_forum at pwscf.org
http://www.democritos.it/mailman/listinfo/pw_forum
More information about the Pw_forum
mailing list