I am trying to use hybrid GGA exchange-correlation functionals.
According to the manual, xctype should be set as
100 id(exchange) id(correlation)
However, hybrid GGA exchange-correlation functionals have only one id(xc).
I found similar software: exciting, it has a different setting: xc = id(xc) to use hybrid GGA exchange-correlation functionals.
My question is how we can use these hybrid GGA exchange-correlation functionals in elk?
how to set xctype?
Best,
Zhiwei
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
for hybrid xc-functionals you can use the libxc ([https://tddft.org/programs/libxc/functionals/]) library.
For hybrid GGA, I would suggest to preconverge (task 0) with PBE functional
using
xctype
20
or also from libxc
xctype
100 101 130
and afterwards (task 5) with e.g.
xctype
100 0 406
Please notice that you have to set the exchange-id=0 and only using the correlation-part from libxc (e.g. 406).
I hope this was helpful.
Best regards,
Mike
Last edit: Mike B 2022-11-10
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks for this explanation, this is very helpful.
And now I also find that it actually explained in one of the examples.
However, I have tried this setting and it turned out that task 5 runs extremely slow and my computer cannot afford it. (32 cores and my system have 30 atoms insulator)
Do you have any idea how to change other parameters in order to make the calculation a little faster?
Best,
Zhiwei
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Dear Zhiwei,
maybe someone else has some better and more specific ideas how to tweak the performance of hybrid calculations, but my suggestion would be to start with careful convergence tests in task 0 (GGA) with respect to the usual parameters: ngridk (also consider to break symmetry by shifting the grid via vkloff), rgkmax and nempty.
But I must say that I think, that its quite a challenge to calculate Hartree-Fock for a 30-atomic cell. Isn't it possible to start with smaller minimal model? You may consider to start with a well-understood material and compare the results of the hybrid calculations in this case.
Best,
Mike
Last edit: Mike B 2022-11-11
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Dear all,
I am trying to use hybrid GGA exchange-correlation functionals.
According to the manual, xctype should be set as
100 id(exchange) id(correlation)
However, hybrid GGA exchange-correlation functionals have only one id(xc).
I found similar software: exciting, it has a different setting: xc = id(xc) to use hybrid GGA exchange-correlation functionals.
My question is how we can use these hybrid GGA exchange-correlation functionals in elk?
how to set xctype?
Best,
Zhiwei
Dear Zhiwei,
for hybrid xc-functionals you can use the libxc ([https://tddft.org/programs/libxc/functionals/]) library.
For hybrid GGA, I would suggest to preconverge (task 0) with PBE functional
using
xctype
20
or also from libxc
xctype
100 101 130
and afterwards (task 5) with e.g.
xctype
100 0 406
Please notice that you have to set the exchange-id=0 and only using the correlation-part from libxc (e.g. 406).
I hope this was helpful.
Best regards,
Mike
Last edit: Mike B 2022-11-10
Dear Mike,
Thanks for this explanation, this is very helpful.
And now I also find that it actually explained in one of the examples.
However, I have tried this setting and it turned out that task 5 runs extremely slow and my computer cannot afford it. (32 cores and my system have 30 atoms insulator)
Do you have any idea how to change other parameters in order to make the calculation a little faster?
Best,
Zhiwei
Dear Zhiwei,
maybe someone else has some better and more specific ideas how to tweak the performance of hybrid calculations, but my suggestion would be to start with careful convergence tests in task 0 (GGA) with respect to the usual parameters: ngridk (also consider to break symmetry by shifting the grid via vkloff), rgkmax and nempty.
But I must say that I think, that its quite a challenge to calculate Hartree-Fock for a 30-atomic cell. Isn't it possible to start with smaller minimal model? You may consider to start with a well-understood material and compare the results of the hybrid calculations in this case.
Best,
Mike
Last edit: Mike B 2022-11-11
Dear Mike,
Thanks a lot. I give up. This system is just too big. I have tried other small systems in the example folder, and they worked out OK.
Best,
Zhiwei