[QE-users] Problem finding global minimum with relax

[FRANCESCO GIANNICI]

Dear all,

I'm trying to evaluate how a 40-atom molecule packs in the solid state, and in particular how the intermolecular interactions may stabilize one conformation over others.

I first tried to validate the gas-phase results: the optimization in the gas phase is robust with Gaussian (semiempirical/HF/DFT always comes up with the same shape whatever the initial geometry is). This molecule also adopts a few different conformations with a slightly different energy depending essentially on one dihedral angle.

So I set up a relax calculation with a 60000 au^3 cell, assume_isolated="mt", K_POINTS gamma, default pp and cutoffs from sssp efficiency, default %IONS input cards. Using two different starting geometries, the optimization always proceeds a little bit (bond lengths change), and the energy goes down, but the dihedral angle only changes by about 1 deg, and the absolute minimum geometry is not found:

This is the optimization with one starting point:
!    total energy              =    -729.67641672 Ry
!    total energy              =    -729.72500523 Ry
!    total energy              =    -729.75723512 Ry
!    total energy              =    -729.76659765 Ry
!    total energy              =    -729.77423836 Ry
!    total energy              =    -729.77833838 Ry
!    total energy              =    -729.78148161 Ry
!    total energy              =    -729.78393743 Ry
!    total energy              =    -729.78583548 Ry
(...)

This is with a different starting point:
!    total energy              =    -730.05783572 Ry
!    total energy              =    -730.10489903 Ry
!    total energy              =    -730.14376397 Ry
!    total energy              =    -730.15543779 Ry
!    total energy              =    -730.16996658 Ry
!    total energy              =    -730.18475032 Ry
!    total energy              =    -730.19850419 Ry
!    total energy              =    -730.20276538 Ry
!    total energy              =    -730.20749936 Ry
(...)

This result itself makes sense (the output geometries also look fine), because what I expect to be a more stable geometry has in fact a lower energy, but the relax algorithm doesn't jump around the potential energy surface to reach a global minimum independently.

This behavior affects my end goal, because I need the relax algorithm in a packed cell to confidently identify a minimum. Is there any way to tell PW.x to "kick" atoms around harder, so to replicate somehow the way Gaussian is able to find the global minimum? Or is my approach not appropriate at all?

Thanks for your suggestions

— 
Prof. Francesco Giannici
Dipartimento di Fisica e Chimica
Università di Palermo
francesco.giannici at unipa.it