What are the Different Approaches for DFT and How Do I Choose the Right One?

[aihaike]

Dear all,

I trying to get a better understanding of the DFT method, but the more I read the more I'm confused.
Well, for crystal states it seems that the use of the plane wave with pseudo potentials is the "best" choice.
But yet we have to choice between soft pseudo potentials or projector augmented plane waves for example.
And what about regular augmented plane waves?
Then, I may miss something here but I don't get the meaning of "LDA pseudopotential". Why a pseudopotential depends on the definition of a exchange-correlation functional?
Why do we need to generate pseudopotentials ?
Again, I'm just a beginner of DFT methods.
I knows it depends on what material we study, on we want to compute, but I need to be clear about what those things really mean and there are so many paper work about that. As a beginner it's very easy to get lost.

Eric.

 

[kanato]

Augmented plane waves (APW) is generally regarded as the most accurate DFT basis set, however it's a slow method because the basis functions are energy dependent. The basis functions can be linearized to get rid of the energy dependence in the LAPW method, which is also quite accurate and implemented in codes like Wien2k and Elk. Also there are techniques of using local orbitals for basis functions, like is used in the FPLO code.

Pseudopotentials are generated by solving the DFT problem for an isolated atom with all its electrons. An exchange-correlation functional is required for that, so each pseudopotential that you generate is labeled by the xc that was used to generate it.

Pseudopotentials must be generated when using planewave codes. Near atomic cores, the wavefunctions vary rapidly in space to maintain orthogonality with core wavefunctions. To represent that in planewaves would require a very large number of planewaves, so much so that it can be computationally difficult, especially for large atomic numbers. So the pseudopotential is used to remove the core wavefunctions from the calculation and the nodes that exist in the valence wavefunctions. This is an approximation technique which is not systematic, so that is why you get many different approaches for creating pseudopotentials.

Planewaves/pseudopotentials aren't the "best" choice for crystal states in general. They can be the best choice for certain things, and there are times when LAPW or local orbitals are better choices.

 

[aihaike]

Dear Kanato,

Thank you for your clear answer.
I suppose codes where PW are implemented also do LAPW?

 

[kanato]

No, not usually. But these days most PW codes do PAW though.

 

[aihaike]

ok, thanks again.