Defect complexes

[conrad...@gmail.com]

Dear Developers,

Note that defect-complexes are not currently possible to execute within pymatgen.

Is there a plan to include this feature in the near future? If it is known that defect complexes are highly favorable in the system of interest (above certain concentration) , how accurate would calculations be in the current framework that does not account for them?

Furthermore, is it generally recommended to utilize the pycdt or pymatgen modules for pre-processing of VASP calculations at HSE06 theory. I had previously utilized the make_vasp_defect_files command but noticed it is not present in the current examples notebook.

To be clear, the charge state of the system is the opposite of the charge state of the defect, correct? e.g., if I have Vac with q = +2, NELECT should be NELECT_neutral +2?

Thanks,

Conrad

[Bharat Medasani]

On Thu, Sep 3, 2020 at 6:01 PM conrad...@gmail.com <conrad...@gmail.com> wrote:

Dear Developers,

Note that defect-complexes are not currently possible to execute within pymatgen.

Is there a plan to include this feature in the near future? If it is known that defect complexes are highly favorable in the system of interest (above certain concentration) , how accurate would calculations be in the current framework that does not account for them?

Hello Conrad,

There is no plan to incorporate defect-complexes modeling. They require careful modeling and a tool like pycdt can't do justice. 

Furthermore, is it generally recommended to utilize the pycdt or pymatgen modules for pre-processing of VASP calculations at HSE06 theory. I had previously utilized the make_vasp_defect_files command but noticed it is not present in the current examples notebook.

You could go to previous versions of the pycdt and copy the code behind the command. We removed lot of deprecated code in the recent version. 

To be clear, the charge state of the system is the opposite of the charge state of the defect, correct? e.g., if I have Vac with q = +2, NELECT should be NELECT_neutral +2?

No. for vac with q=+2, nelect should be nelect_neutral - 2.

NELECT is the number of electrons in the system. 

Bharat

Thanks,

Conrad

--
You received this message because you are subscribed to the Google Groups "PyCDT forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pycdt-forum...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/pycdt-forum/29707d3c-c02b-4653-a415-9a2be8e23c57n%40googlegroups.com.

[conrad...@gmail.com]

Well, pycdt automatically defined NELECT so I assume it's correct.

How does pycdt determine the supercell sizes for POSCARS of bulk, defect and dielectric calculations?

Is it necessary to perform supercell convergence tests prior to running jobs?

Lastly, if one is performing calculations at HSE06 what would be the workflow for calculating chemical potentials, finite-size corrections and defect localization and how does this differ from PBE+U?

Thanks,

Conrad

[Bharat Medasani]

Bharat Medasani

Engineer

Princeton Plasma Physics Lab (PPPL)

---------- Forwarded message ---------
From: Bharat Medasani <mbk...@gmail.com>
Date: Sat, Sep 5, 2020 at 11:00 AM
Subject: Re: Defect complexes
To: conrad...@gmail.com <conrad...@gmail.com>

On Fri, Sep 4, 2020 at 3:36 PM conrad...@gmail.com <conrad...@gmail.com> wrote:

Well, pycdt automatically defined NELECT so I assume it's correct.

How does pycdt determine the supercell sizes for POSCARS of bulk, defect and dielectric calculations?

PyCDT makes heuristics based guess of supercell sizes for defect calculations. For bulk properties and dielectric caclulations, you don't need supercell. You can use the primitive cell to evaluate dielectric and band structure properties.  Make sure your k-point mesh is sufficient enough.

Is it necessary to perform supercell convergence tests prior to running jobs?

Ideally we would like to have super large defect supercells to model defects which have a typical concentration of 1e-10 to 1e-18. The point of finite size charge correction is to avoid supercell convergence tests for defects modeling. However you can't use a very small unit cell in which case the defect charge distributions would overlap completely changing the physics. As long as your defect charge distributions are reasonably separated, charge corrections will help overcome the limitations of finite sized supercells. For most of the semiconductors and insulators supercell sizes with 64 to 128 atoms should work. To be on the safe side you could 128 atoms if you have sufficient computing power. One rough estimate you can use is the value of dielectric constant. If your material has large 

 

Lastly, if one is performing calculations at HSE06 what would be the workflow for calculating chemical potentials, finite-size corrections and defect localization and how does this differ from PBE+U?

If you are using a different POTCAR or a different functional than the one used for materials project: 1) first you need to evaluate chemical potentials for which you need to compute the energetics all the compounds in the associated phase space and generate chemical potentials using the phase diagram code supplied in pycdt/pymatgen. 2) Recompute the band structure and the dielectric constant. 3) Use the things computed in 1 and 2 for the rest of the work. The rest of the work-flow doesn't change.

Others have done high throughput HSE based defect calculations using atomate and pycdt. So this is doable.

Thanks,

Conrad

On Thursday, September 3, 2020 at 6:27:28 PM UTC-4 mbk...@gmail.com wrote:

On Thu, Sep 3, 2020 at 6:01 PM conrad...@gmail.com <conrad...@gmail.com> wrote:

Dear Developers,

Note that defect-complexes are not currently possible to execute within pymatgen.

Is there a plan to include this feature in the near future? If it is known that defect complexes are highly favorable in the system of interest (above certain concentration) , how accurate would calculations be in the current framework that does not account for them?

Hello Conrad,

There is no plan to incorporate defect-complexes modeling. They require careful modeling and a tool like pycdt can't do justice. 

Furthermore, is it generally recommended to utilize the pycdt or pymatgen modules for pre-processing of VASP calculations at HSE06 theory. I had previously utilized the make_vasp_defect_files command but noticed it is not present in the current examples notebook.

You could go to previous versions of the pycdt and copy the code behind the command. We removed lot of deprecated code in the recent version. 

To be clear, the charge state of the system is the opposite of the charge state of the defect, correct? e.g., if I have Vac with q = +2, NELECT should be NELECT_neutral +2?

No. for vac with q=+2, nelect should be nelect_neutral - 2.

NELECT is the number of electrons in the system. 

Bharat

Thanks,

Conrad

--
You received this message because you are subscribed to the Google Groups "PyCDT forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pycdt-forum...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/pycdt-forum/29707d3c-c02b-4653-a415-9a2be8e23c57n%40googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "PyCDT forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pycdt-forum...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/pycdt-forum/aaba4d10-52a1-4e1e-accb-7831d693827en%40googlegroups.com.

[Bharat Medasani]

From a different perspective, your defect charge distribution should be contained within the supercell. 

To get an rough estimate of the size of cell required, if your material has a large dielectric constant, small cell size should do. But if the dielectric constant is small, your cell size should be on the higher side. 

Another thing is shallow defects require large cells. For deep defects, small cell sizes should work. 

Bharat

On Fri, Sep 4, 2020 at 3:36 PM conrad...@gmail.com <conrad...@gmail.com> wrote:

To view this discussion on the web visit https://groups.google.com/d/msgid/pycdt-forum/aaba4d10-52a1-4e1e-accb-7831d693827en%40googlegroups.com.

[conrad...@gmail.com]

Is there a plan to integrate machine learning into the pycdt workflow,? e.g., high throughput screening via prediction of defect levels relative to fermi, band edge positions and/or temperature dependence of electronic structure properties (effective mass, fermi level) for prescribed application and composition space?