Temperature is not attained according to the thermoset and velocity guass temp

LAMMPS LAMMPS Beginners
1

Respected lammps users, I am using lammps version 3 march, 2020. I am trying to heat silver using eam potential. here is my script

# set potential function and parameters

units metal
dimension 3
boundary p p p
atom_style atomic

neighbor 0.5 bin
neigh_modify delay 5

region simbox block -65 107 -1 299 -89 195 units box

lattice fcc 4.09

region silver block -16 59 44 102 30 80 side in units box

create_box 1 simbox

create_atoms 1 region silver

group silvermetal region silver

group silvermetal type 1

mass 1 107.8

pair_style eam

pair_coeff 1 1 Ag_u3.eam

thermo 1
thermo_style custom step lx ly lz press pxx pyy pzz pe temp
thermo_modify norm yes
thermo_modify lost ignore flush yes

#minimize 1.0e-21 1.0e-21 10000 10000
timestep 0.001

velocity silvermetal create 1500 12345 mom no rot no
run 0
velocity silvermetal scale 1500

fix 1 silvermetal nvt temp 1500 1500 0.1

dump movie all atom 100 m.meltsilver
run 1100000

I am not achieving the temperature which is applied according to thermoset and guass dist. I just get the lower temperature when I check “thermo_style custom step lx ly lz press pxx pyy pzz pe temp” temperature.

Thanks for cooperation

2

Have you heard of the term “equilibration”?

When I run your input, it reaches the desired ~1500K after about 2000 timesteps, which is about 2 periods of the Nose-Hoover thermostat (i.e. 0.2ps), which is quite reasonable and to be expected when starting from ideal lattice positions.

3

Alright. Thanks

5

Respected Lammps users, When I change the temperature to 400
velocity silvermetal create 400 12345 mom no rot no
run 0
velocity silvermetal scale 400

fix 1 silvermetal nvt temp 400 400 0.1

The shape of the silver cube doesnt deform and starts rotation. It shows no melting while the melting temperature of such a silver nanoparticle is just 400k calculated experimentally. How to solve this problem

6

Computing the melting point from MD simulation is tricky business since melting and freezing are activated processes, thus you can have a large hysteresis without nucleation. One option to determine the melting point is to do a coexistence simulation and vary the temperature until both liquid and solid phase do not grow or shrink. There are plenty of discussions about this in the archives and lots of publications. It also is a good idea to look up in a suitable textbook the statistical thermodynamics for liquids and solids, so you have a better understanding of the correlation between atomic scale data and macroscopic processes and properties.

Temperature always fluctuates and and will approach a constant in the large size limit. This is another thing to be learned about from a statistical thermodynamics text book. That will also explain how for a system in equilibrium you can use the time average in addition to the average over particles.

Another important factor is energy conservation of the simulation (which can only be found from running the equilibrated system without thermostatting. This will tell you how reliable the data would be. Perfect conservation is impossible for any software using floating point math since there will always be some (small) rounding errors (more if you single precision math rather than double precision) and some errors from discretizing the differential equations that are solved when integrating the equations of motion. This is another item to look up in a text book. Using too large a timestep will generate too much kinetic energy due to the accumulation of errors and thus your average temperature will be too large.

This is explained in the documentation.

7

You can only solve this when you understand it. I already discussed melting and freezing and the fact that those are activated processes in my previous response.

In addition, you have to check what is the expected melting point for the potential you are using. For empirical potentials, the melting point can be different from experiment depending on how well the potential was parameterized for the conditions at the melting point and how well the model in general is capable of reproducing experimental data. The starting point to learn about this is the publication that describes how the specific potential in question was parameterized and which properties it can reproduce how well. Empirical potentials like LAMMPS uses them are always a compromise. Some are better than others for a specific purpose.

If you try to melt an isolated object in space, you also need to adjust the temperature computation. Since for this system you don’t have the translation invariance that applies to the more typical dense bulk systems and thus your system has 3 degrees of freedom more.
That your object starts to rotate or translate or both is quite common when the kinetic energy is inserted and there is some small rounding error producing an asymmetry which in turn will then transfer kinetic energy to the motion of the entire object as a whole instead of exciting vibrations within the object. Using a Nose-Hoover thermostat algorithm is more likely to cause this since it was primarily designed to correctly model the local thermal fluctuations for a homogeneous bulk system as if it was embedded into a larger system.
You are likely to have more success with a dissipative thermostat (there are many of those available within LAMMPS).

8

The lammps melting point can be different from experiment but it is a very big difference of 700C which is not small. The literature supports that this sized silver nanoparticle just melt at 150C compared to bulk having melting temperature of 950C. The empirical potential supports the very high melting point which is not the experimental case. Hence it is not including the effect of nanoscale. Even when I looked for input scripts posted by research experts have such a problem. The error comes out to be much more than cent percent.

9

This all makes this a problems of the physics and the model chosen and not an issue of LAMMPS. LAMMPS implements the documented models and potentials and executes the commands in the documented fashion. Whether this is sufficient to represent the effect seen in experiment may be dependent on very specific details, but this is up to you or your adviser/mentor/tutor/collaborators to work out. This really only becomes a LAMMPS issue, if you can prove that LAMMPS is not doing what its documentation says it does.