Total energy not conversed in NVE

Hi all,

I have a question about the conservative of total energy in NVE.

I am working on a simulation of rigid TIP3P water under nve ensemble. As I think, the total energy should have very small drift duing a long simulation time because of the numerical integration uncertainty.

However, from the result I had(attached, NVEvstimestep.bmp), the total energy increased with the timestep of 1fs and even decreased with timestep 2fs.

The simulation was divided into three steps:
1)From a initial configration, water box with 1969 water molecules was equilibrated for 1 ns in NPT.timestep=1fs
2)After that, the box was equilibrated for another 0.5 ns in NVT. timestep=1fs
3)Lastly, have the box equilibrated in NVE with different timesteps :2fs, 1fs, 0.5fs and 0.1fs.

During the simulation, water bond and angle are hold as rigid by SHAKE method. Vdw interaction larger than cutoff was shifted.

I don’t understand :

1)For timestep of 2fs, why the total energy decrease?

2)Why the energy increase with the timestep 1fs. Is it resulted from the numerical integration uncertainty?

The initial input file was also attached.

I appreciate any help.

Thanks,

Hang

nve.rar (14.6 KB)

2011/11/10 physics.hangyan.chen <[email protected]...>:

Hi all,

I have a question about the conservative of total energy in NVE.

I am working on a simulation of rigid TIP3P water under nve ensemble. As I
think, the total energy should have very small drift duing a long simulation
time because of the numerical integration uncertainty.

However, from the result I had(attached, NVEvstimestep.bmp), the total
energy increased with the timestep of 1fs and even decreased with timestep
2fs.

please note, that attaching .bmp files and on top of that
bundling them into a rar archive will almost guarantee that
nobody will look at them. both are _very_ windows-centric
file formats and uncommon on non-windows systems,
which is what the majority of LAMMPS developers seem
to be using.

The simulation was divided into three steps:
1)From a initial configration, water box with 1969 water molecules was
equilibrated for 1 ns in NPT.timestep=1fs
2)After that, the box was equilibrated for another 0.5 ns in NVT.
timestep=1fs
3)Lastly, have the box equilibrated in NVE with different timesteps :2fs,
1fs, 0.5fs and 0.1fs.

During the simulation, water bond and angle are hold as rigid by SHAKE
method. Vdw interaction larger than cutoff was shifted.

I don't understand :

1)For timestep of 2fs, why the total energy decrease?

because the timestep of 2fs is so large,
that you need to converge the SHAKE
constraints tighter. thus try a 1.5fs timestep
or use a tighter shake convergence criterion
(and more iterations).

you can also confirm this by redistributing mass
from the oxygen to the hydrogens (a common trick
to allow use of larger time steps with water) and the
drift will be reduced.

please note that the drift that you are seeing
is still very small compared to the fluctuations
in potential and kinetic energy.

2)Why the energy increase with the timestep 1fs. Is it resulted from the
numerical integration uncertainty?

for the most part, yes.

axel.