Some problem regarding simulated annealing using NVT thermostat

Dear Lammps dev

I am currently studying epitaxial graphene through simulated annealing across the surface of carbon riched SiC substrate. While i try to investigate the effect of heating rate, i found that the longer i heat it, the faster phase transition is happened. According to the experiment, the phase transition is supposed happened around 1300K. So I decided to plot potential energy vs steps near the transition temperature (1200K), i found that the system is does not happen any phase transition when i use the steps below 80k (it matches the experiment value). So the error occur when I run longer steps, meaning when i run longer the phase transition occur at lower temperature.

Am I right that the simulation is not supposed to depends on steps?

Is this the problem of NVT or damping or variant of thermostat?

I will attached the information below.

units metal
timesteps 0.005

fix 4 all nvt temp 1200.0 1200.0 0.1

run 50000 (I change the steps accordingly)



Are you sure that the experimental conditions are NVT? What about NPT?

The behaviour of the system depends on the potential that you are
using, there is no potential that will reproduces all the behaviour of
the material, MD potentials reproduces what they are fitted to

To truly understand an phase transitions you should know the free
energy of the system, small systems are subject to finite size
problems, for example: an first-order phase transition may not occour
in a small system due to the high free energy cost of the creation of
an interface.

I think that your simulation is fine, but you should think better
about what are you simulating.

It also depends on the equilibration procedure, I'm assuming that you
are equilibrating the system in the same way on all cases.

Rodrigo Freitas

Dear Freitas

Thanks for your reply.

I am using Tersoff-Erhart-Albe potential. So you are suggesting me to perform annealing using npt and make my system bigger? below is the way I perform annealing

min_style cg
minimize 0.0 0.0 10000 10000
min_modify line quadratic

velocity all create 0.1 1234567 dist gaussian

timestep 0.0005

fix 1 all nvt temp 0.1 0.1 0.1

run 20000

unfix 1

fix 2 all nvt temp 0.1 300.0 0.1

run 40000

unfix 2

fix 3 all nvt temp 300.0 300.0 0.1

run 30000

unfix 3

fix 4 all nvt temp 300.0 1200.0 0.1

run 50000

unfix 4

fix 5 all nvt temp 1200.0 1200.0 0.1

run 80000


I don't know what kind of phase transition are you seeking but the NPT
ensemble is probably the experimental condition used. Also, NPT let
the volume of the system vary, it is important in some phase
transitions (constant volume is a strong condition that can affect the
phase transition).

I do not suggest you to make the system bigger (even because sometimes
it cannot be done), I'm just telling that there exist finite size
effects that you should expect once you system is limited.

The step dependence on the temperature is something that I would
expect to happen, longer simulations at a fixed temperature let your
system relax for more time, if your current phase is thermodynamically
metastable the transition will happen if you give the system a
sufficient time. But I've never worked with graphene, you should
discuss the physics of the problem with someone that know the subject.