lammps examples for nvt and npt simulations

[deleep225]

I am trying to do the nvt simulations with trained models. I found wild fluctuations in temperatures.

I have also taken the google-colab tutorial and tried to run the simulation in nvt at 298 K. I found similar wild fluctuation in temperature.

Step PotEng KinEng TotEng Temp Press Volume
0 -169816.55 12.43594 -169804.11 298 0.56847511 1000000
100 -169797.52 14.312021 -169783.2 342.95619 0.65423509 1000000
200 -169804.61 42.009572 -169762.6 1006.6672 1.9203532 1000000
300 -169778.45 4.4903736e-108 -16977 8.45 1.0760195e-106 2.052652e-109 1000000
400 -169816.11 330.76314 -169485.35 7926.0129 15.119936 1000000
500 -169816.5 2.6527385e-86 -169816.5 6.3567057e-85 1.2126272e-87 1000000
600 -169814.08 2.3112458e-07 -169814.08 5.5383935e-06 1.0565231e-08 1000000
700 -169816.41 0.018748718 -169816.39 0.44927188 0.00085704658 1000000
800 -169806.48 25.813136 -169780.67 618.55515 1.1799772 1000000
900 -169788.61 51.641929 -169736.97 1237.4855 2.3606702 1000000
1000 -169781.58 8.1885765 -169773.39 196.22127 0.37431847 1000000
1100 -169787.17 1.3127544e-09 -169787.17 3.1457279e-08 6.0008994e-11 1000000
1200 -169740.02 79.204829 -169660.81 1897.9699 3.620633 1000000

from the above output you can see high fluctuation in temperature and kinetic energy.

here in my lammps input file

units real
atom_style atomic
newton off
thermo 1
read_data em1

pair_style nequip
pair_coeff * * toluene-deployed.pth C H
mass 1 15.9994
mass 2 1.00794

neighbor 1.0 bin
neigh_modify delay 5 every 1

velocity all create 298.0 23456789

fix 1 all nvt temp 298.0 298.0 0.5 #iso 1.0 1.0 0.5
#fix 2 all langevin 298.0 298.0 298.0 5646546
timestep 1
thermo_style custom step pe ke etotal temp press vol
thermo 100
#dump 1 all custom 100 water1ns.dump id type x y z
dump 1 all atom 100 2.lammpstrj
run 20000
write_data nvt1

It will be great if some lammps examples are provided in nvt, npt and nve ensembles with nequip potentials.

Thank you.

[simonbatzner]

HI @deleep225, your Tdamp parameter is 0.5 and you are in real units, that means it's a 0.5fs Tdamp for a 1fs time step. That won't work. The recommendation is: $(100*dt)

[simonbatzner]

Note also that for systems with C-H bonds, 0.5fs is typically used as a safe choice for a time step due to the high frequency of the C-H bond.