Does there exist the limit of convergence

[weiyidan]

Dear professor,
When i did a calculation, I found it was impossible to attain Ediffg=-0.0001. Many methods had been used to improve the performance, such IBRION=1/2 and improving EDIFF (See att[attachment=0]OUTCAR.rar[/attachment]achment). However, it seemed to exist the limit of converge. Can you give me some advice to improve convergence precision and efficiency?
Thanks for your help and advice.
Yidan Wei

[weiyidan]

Here are input attachments.

[graeme]

My first question is why you want to converge a structure to such precision.

Then, you are using CG? You should be able to get more rapid convergence with LBFGS (IBRION=1) in the harmonic region.

But again, your energy is changing by 0.1 meV! Your faith in DFT should be 1000x less, with a typical uncertainty of 0.1 eV.

[weiyidan]

Dear professor,
I firstly used IBRION=1, however, it did not work, whose energy changed by 0.1 meV as well (See attachment). Therefore, i turned to IBRION=2 to seek a lower state.
Then, how can i improve performance？
Thanks for suggestion.
Yidan Wei

[weiyidan]

I want to calculate phonons 0.0.

[graeme]

The force and energy are both dropping systematically. Keep running. You only need to reach 0.001 eV/Ang or even 0.005 eV/Ang to calculate phonons.

[weiyidan]

Dear professor,
Thank you very very much. But Is the criteria enough to calculate phonon? I find Ediffg is always very, very low as the online instruction shows. And, I have other two questions, can you give me some advice?
Firstly, when i do geometrical optimization, should the ADDGRID = .TRUE. and LREAL = .FALSE. be added, as a 120 atom supercell ?
Secondly, What is the source of the imaginary frequency？Does the atom force matter？ Whether less force of atom can reduce imaginary frequency.
Thanks for your advice again.
Yidan Wei