Regarding the selection of pdamp for barostatting

Hi Lammps Community,

I have been performing very strict pressure-controlled simulations of dislocation dynamics, and have been curious about the effect of the barostat damping parameter on the simulation.

It is somewhat disconcerting (and confusing), that the glide velocity can vary by over 20%, depending on the selection of pdamp (from 10 - 1000).

In this case the normal stresses are all maintained at approximately 0 MPa, and most importantly, that τxy is constant in both cases. The only difference is in the shear stress component normal to the glide plane, which is perplexing (as this theoretically should not influence the glide velocity?).

Please refer to the dynamic stress tensor plots, and the plots of the regression models used to evaluate the velocity if you want to comment and make suggestions!!

Effect_of_pdamp_on_DislocationDynamics.pptx (195 KB)

Hi Lammps Community,

I have been performing very strict pressure-controlled simulations of
dislocation dynamics, and have been curious about the effect of the barostat
damping parameter on the simulation.

It is somewhat disconcerting (and confusing), that the glide velocity can
vary by over 20%, depending on the selection of pdamp (from 10 - 1000).

In this case the normal stresses are all maintained at approximately 0 MPa,
and most importantly, that τxy is constant in both cases. The only
difference is in the shear stress component normal to the glide plane, which
is perplexing (as this theoretically should not influence the glide
velocity?).

it is difficult to make comments to such a specific simulation. but as
a general comment, it seems to me that there is a bit of a
misconception with respect to the "damping" parameters in nose-hoover
thermostats. they are specified this way mostly out of convenience,
but you actually have to see them as a characteristic frequency.
nose-hoover thermostats (and barostats) do not have a straightforward
and easy to grasp behavior as, say, a berendsen thermostat/barostat.
specifically, if your initial structure is fairly far away from
equilibrium and the damping constant results in a characteristic
frequency that does not couple well to the system at hand, it can take
an extremely long time, if a system is balanced. and even then, it can
stray rather far from the average state.
if you change the damping parameter, you cannot expect a "linear"
behavior, as you also have to factor in how strong the
thermostat/barostat DOFs are coupled to the system, which can change
quite arbitrarily, particularly in solids.

i would suggest to repeat the calculations with a berendsen barostat.
since you need to care about sampling a specific statistical dynamical
ensemble, that may be a more consistent approach, especially when
considering strong damping. even better would be an approach similar
to the csvr thermostat, which combines canonical sampling with the
simplicity of a berendsen like approach, but would require some
programming.

axel.

Thank you Axel,

I appreciate your comments, however the system is already at the ideal pressure state (in terms of all 6 components of the stress tensor) at timestep = 0. This is achieved, using a triclinic cell. In terms of the dipole motion, I can see how there could be an issue with the frequency of barostatting if the dipoles crossed the periodic cell boundaries faster than the pressure ramping (i.e., cell dimension rescaling) frequency. However, in this case the periodicity of the simulation velocity is more than 20,000 fs so I cannot really see this as the issue?

Furthermore, most importantly:
the fix press/berendsen approach cannot be used to modify a triclinic cell to control all 6 components of the stress tensor!? Do you have any recommendations in this regard?

I suppose the more significant question, however, is whether the process of damping (periodically ramping the system pressure towards the user-specified value) could cause other effects on the dynamic atomistic behaviour, beyond just the effects on the stress (and, of course I note that the simulation dimensions are effectively equivalent in all cases).

Thank you Axel,

I appreciate your comments, however the system is already at the ideal
pressure state (in terms of all 6 components of the stress tensor) at
timestep = 0. This is achieved, using a triclinic cell. In terms of the
dipole motion, I can see how there could be an issue with the frequency of
barostatting if the dipoles crossed the periodic cell boundaries faster than
the pressure ramping (i.e., cell dimension rescaling) frequency. However, in
this case the periodicity of the simulation velocity is more than 20,000 fs
so I cannot really see this as the issue?

Furthermore, most importantly:
the fix press/berendsen approach cannot be used to modify a triclinic cell
to control all 6 components of the stress tensor!? Do you have any
recommendations in this regard?

launch your editor and start programming.

Hi,

If anyone else has any ideas to explain why there is such a large inconsistency in the dynamic dislocation response, I would very much appreciate it. My intention is not to evaluate the methods of barostatting, only to maintain a roughly constant value of τxy and evaluate the dislocation mobility relationships.

if no one has any ideas, I might have to stick with the conclusion that the Nose-Hoover barostatting process causes significant inconsistency in the dynamic behaviour of non-equilibrium processes, which makes it difficult to assert an accurate basis for multi-scale modelling… Not ideal!

Thanks,
Nathaniel