I am trying to use LAMMPS (v17 Aug 2017) together with PLUMED (2.3.2) to study pressure induced phase transitions.
From simple unbiased simulations in NPT ensemble using fully anisotropic unit cell (structure transformation needs to deformate the cell on relatively small system - 1024 particles preparted
in B1 phase) I now that such transitions occur around 550 - 600 kbar (obviously the precise value suffers from hysterics effects which came from relaxation).
This value I took as a reference value for further study.
Then, I tried to implement metadynamics at 500 kBar and try to reconstruct Gibbs free energy profile based on simple coordination number between Na-Cl ions,
which differs between B1 and B2 phase (6 and 8 respectively). To to this, as a common rule of thumb, I guessed expected barrier height and took a fraction of it
to set the gaussian height, using fluctuations of coordination number in unbiased simulation and the gaussian deposit rate was chosen the closest to the
barostat relaxation time, so that the system should still have enough time between successive deposits to equilibrate. The cutoff for switch function which is used
in calculating coordination number was chosen in the middle between the first two maxima in radial distribution function for Na-Cl in unbiased simulation
at given pressure.
Typical example of system parameters are 300 K, 550 kbar, 0.25 ps for thermostat relaxation time, 0.5 ps for barostat relaxation time using Nose-Hoover
anisotropic barostat in LAMMPS with MTK term. As force field I have a standard used Born-Mayers-Huggins potential for NaCl.
However, regardless of tunning MetaD parameters, changing the times of barostat I observe a common strange behavior that system successfully escapes from B1 phase
to B2 phase, however it is not able to fully return to the B1 phase, instead the unit cell starts to deform in some strange fashion. The crystallization of the new phase
proceed through layering and deformation of the unit cell which resembles Burgers mechanism. Do you have any suggestion how can be this behavior suppressed, in order to correct sample both phases?
Here I attached the figures to illustrate the issue:
Fig.1: Time evolution of coordination number which shows a strange behavior - the system does not return to the B1 phase.