ABINIT Discussion Forums

Dear developers,

I met the problem in calculating thermal conductivity using abinit+phono3py. The 3rd interatomic force constants are calculated using finite difference supercell method, in which the force for each displaced supercell is calculated using abinit. However, the thermal conductivity results are not well enough, with a large deviation with experimental results while VASP+phono3py can give a reasonable results. Do you have any idea about accuracy in calculating forces of abinit?

The pseudopotential used in abinit calculations is psp8 form, for Ga and N atoms.

Thanks,
Daosheng

Do you have any idea about accuracy in calculating forces of abinit?

The cutoff energy and the converge criterion used to stop the SCF cycle can affect the quality of the forces
but it's hard to say without looking at your input file.

Can you try to increase the value of boxcutmin from 2 to e.g. 4?

We found some difference when comparing Abinit DFPT phonons with phonopy finite difference method with Abinit forces and we managed to make the phonopy results agree with DFPT by increasing boxcutmin.

BTW: Are you using ADDGRID=True (https://www.vasp.at/wiki/index.php/ADDGRID) in your Vasp calculations?

Thanks for your reply.

It is weird that abinit+phonopy can produce good phonon dispersion relations compared with vasp+phonopy, where the settings in calculating 2nd IFCs are exactly the same with those in 3rd IFCs calculations.

The input file used for 3rd IFCs calculations is attached which takes 00001 case as an example. For psp8, tests show that a large cutoff energy is needed and 1000 eV is used here. Using supercell size 4*4*3, about 190 W/mK thermal conductivity is obtained with abinit which is much lower than 270 W/mK from VASP.

I will try to increase boxcutmin to find out the difference.

Best,
Daosheng

gmatteo wrote: ↑

Tue Dec 29, 2020 1:22 am

Can you try to increase the value of boxcutmin from 2 to e.g. 4?

We found some difference when comparing Abinit DFPT phonons with phonopy finite difference method with Abinit forces and we managed to make the phonopy results agree with DFPT by increasing boxcutmin.

BTW: Are you using ADDGRID=True (https://www.vasp.at/wiki/index.php/ADDGRID) in your Vasp calculations?

Yes, ADDGRID=True is used in vasp calculations.