components of the stress tensor
Posted: Thu Nov 27, 2014 3:54 pm
Dear all,
Could you please direct me to an explanation of the meaning of individual components contributing to evaluation of the stress tensor in VASP?
Here is the background for my question:
I work on amorphous-like materials. After performing the simulated annealing treatment using AIMD, I wanted to relax the final supercell. It converged after a few cycles using the EDIFFG=1E-4 criterion [total energy criterion, TEC]. However, I spotted that quite a few force components on individual atoms were quite high (>0.1eV/A) and hence I restarted the calculation using EDIFFG=-1E-3 [force criterion, FC].
However, to my surprise, the stress tensor from TEC case was significantly different from the FC one, including one of them being positive and the other negative. When looking at the components of the stress tensor, the difference comes almost exclusively from the n-local component.
Even more strangely, the total energy of the last TEC step is _lower_ then the _first_ (but also the _last_) energy from the FC minimisation. I am completely puzzled by this fact.
I should say, that by "restarting" the calculation I mean to start with the final CONTCAR from TEC as POSCAR for FC, but with freshly initialised WAVECAR and CHGCAR (as I did not save these during the previous optimisation).
Any comments and insights will be highly appreciated.
David
Could you please direct me to an explanation of the meaning of individual components contributing to evaluation of the stress tensor in VASP?
Here is the background for my question:
I work on amorphous-like materials. After performing the simulated annealing treatment using AIMD, I wanted to relax the final supercell. It converged after a few cycles using the EDIFFG=1E-4 criterion [total energy criterion, TEC]. However, I spotted that quite a few force components on individual atoms were quite high (>0.1eV/A) and hence I restarted the calculation using EDIFFG=-1E-3 [force criterion, FC].
However, to my surprise, the stress tensor from TEC case was significantly different from the FC one, including one of them being positive and the other negative. When looking at the components of the stress tensor, the difference comes almost exclusively from the n-local component.
Even more strangely, the total energy of the last TEC step is _lower_ then the _first_ (but also the _last_) energy from the FC minimisation. I am completely puzzled by this fact.
I should say, that by "restarting" the calculation I mean to start with the final CONTCAR from TEC as POSCAR for FC, but with freshly initialised WAVECAR and CHGCAR (as I did not save these during the previous optimisation).
Any comments and insights will be highly appreciated.
David