How to calculate pseudopotentials?

[jd1828af]

Can anyone explain how pseudopotentials are calculated using density functional theory?

 

[pigbear]

In the book of
Planewaves, pseudopotentials and the LAPW-method,
D. Singh, Kluwer Academic Publishing (1994), ISBN 0-7923-9421-7
it gives some introduction.

 

[marlon]

Can anyone explain how pseudopotentials are calculated using density functional theory?

Consult this : Bachelet, G. B. et al. (1982) Phys. Rev. B 26 4199


regards
marlon

 

[marlon]

as an addendum, i especially recommend the reference nr 5 (Goedecker, phys rev B, vol 42, 8858)in the above paper for a more introductory explanation

regards and enjoy

marlon

 

[marlon]

One of the first REAL ab initio methods (ie no fitting with experimental data is required what so ever)to calculate pseudopotentials is : phys rev , vol 116, 287 (1959)

marlon

 

[jd1828]

does anyone know of where I could find some calculated pseudopotentials for a solid. Ill go for just about any solid, although CO2 is what I would really need. I am just trying to see if my calculations are anywhere close to what they should be.

 

[marlon]

does anyone know of where I could find some calculated pseudopotentials for a solid. Ill go for just about any solid, although CO2 is what I would really need. I am just trying to see if my calculations are anywhere close to what they should be.

CO2 a solid ?

I do not understand how you can check your calculations without having the pseudo potential file to model the [valence electron] <---> [nucleus + core electrons]-interactions ?

How exactly do you procede ?

I mean, assuming you are doing Hartree Fock or DFT calculations :

1) what software are you using ? Usually, the pseudopotential files (like the ) can be found on the server of the software developer ; like in the case of SIESTA or ABINIT.

2) What basis set are you using (like doubble zeta gaussians) or do you procede with plane waves ?


3) In DFT case, what approximation do you use for the exchange/correlation functional ? LDA, GGA, Hybrid models ? For metals, you must use the LYP-correlation functional of Parr et al (1988) together with the PBE exchange functional of Perdew, Burke and Ernzerhof (1996)


4) Normally, here you would chose your pseudo potential file. If you take one from literature, be sure that it is "transferable". Also, remember that the pseudopotential file that you choose will be partially determined by your exchange/correlation functional. You need to do a lot of benchmarking to get the right one.

The mentioned functionals can all be found in the previous references that i gave you.

marlon

 

[marlon]

Ok, this maybe a bit late but i forgot to refer you to the following site on Vanderbilt's pseudopotentials :

here

regards
marlon

 

[jd1828]

Yes, I am doing hartree-fock calculations using abinit. I just need to show that the program is working correctly before I try using any of the other methods. I was getting very strange energies which I found was caused by the cell volume being a little too small.

CO2 a solid ?
I do not understand how you can check your calculations without having the pseudo potential file to model the [valence electron] <---> [nucleus + core electrons]-interactions ?
How exactly do you procede ?
I mean, assuming you are doing Hartree Fock or DFT calculations :
1) what software are you using ? Usually, the pseudopotential files (like the ) can be found on the server of the software developer ; like in the case of SIESTA or ABINIT.
2) What basis set are you using (like doubble zeta gaussians) or do you procede with plane waves ?
3) In DFT case, what approximation do you use for the exchange/correlation functional ? LDA, GGA, Hybrid models ? For metals, you must use the LYP-correlation functional of Parr et al (1988) together with the PBE exchange functional of Perdew, Burke and Ernzerhof (1996)
4) Normally, here you would chose your pseudo potential file. If you take one from literature, be sure that it is "transferable". Also, remember that the pseudopotential file that you choose will be partially determined by your exchange/correlation functional. You need to do a lot of benchmarking to get the right one.
The mentioned functionals can all be found in the previous references that i gave you.
marlon

 

[marlon]

I was getting very strange energies which I found was caused by the cell volume being a little too small.

That's a classic problem. Your atomic forces are too big. Normally the threshold (international standard) is below 0.05 eV/angström. To acquire this demand you must perform a atomic position relaxation and a atomic lattice relaxation.

Just to be complete, let me again show you the general way to proceed after you have selected all necessary input data (pseudo potentials, exchange correlation functional, atomic lattice and atomic positions)

1) perform a convergence test with respect to the energy cutoff value

2) perform a convergence test with respect to the selected k-mesh. The bigger the unitcell, the smaller the k-point mesh (due to the inverse connection between Wigner Seitz unit vectors and (reciprocal) Brillouin unit vectors).

3) Select the mesh and E cutoff of the above two tests and perform an relaxation of the atomic positions.

4) perform a relaxation of the lattice with the optimized atomic positions as input for this test.

5) The outcome of test 4 will give you the best structure (both latticedimensions and atomic positions) and be sure that after test 4, the atomic forces are below 0.05 eV/angström !

If you have done the above tests, your problems should be solved. If they are not, there is something wrong with the pseudopotential that you used. Perhaps it is not transferrable, have you checked ? Select another pseudopotential file from the Vanderbilt website. Those files are very reliable. Another option is an incorrect input for the lattice and atomic positions. To be sure of that, use the input from the "Web of Elements" website.

Finally, if you can, try to compare your results with a SIESTA simulation (if you know how to work with SIESTA ofcourse).


regards
marlon

 

[handsomecat]

Just to be complete, let me again show you the general way to proceed after you have selected all necessary input data (pseudo potentials, exchange correlation functional, atomic lattice and atomic positions)

1) perform a convergence test with respect to the energy cutoff value

2) perform a convergence test with respect to the selected k-mesh. The bigger the unitcell, the smaller the k-point mesh (due to the inverse connection between Wigner Seitz unit vectors and (reciprocal) Brillouin unit vectors).

Could I ask ... for these two steps are single-point calculations enough or do we have to do geometry optimization? What's the yardstick for determining if the test has been passed or not?

My system is rather large - performing many many separate calculations will certainly take a very long time. Basically, I cut corners by increasing k-pts and energy cutoff value simultaneously until the quantities that I wanted to obtain had converged.