SPC/E energy drift

Dear Lammps users,

I am wanting to perform simulations of spc/e water under large thermal gradients via the fix heat command however have been having trouble with energy conservation of my system.

I initially equilibrate my system in the NVT ensemble at a temperature of 500K for 500,000 steps.

As a test of energy conservation I then start an NVE simulation however there appears to be a significant total energy drift.

I have attached both the input script for the NVT and NVE run in the hope someone can spot my mistake. The files are named “mail.nvt” and “mail.nve” respectively.

Thank you in advance, Jeff.

mail.nvt (635 Bytes)

mail.nvt (635 Bytes)

Dear Lammps users,

I am wanting to perform simulations of spc/e water under large thermal
gradients via the fix heat command however have been having trouble with
energy conservation of my system.

I initially equilibrate my system in the NVT ensemble at a temperature of
500K for 500,000 steps.

As a test of energy conservation I then start an NVE simulation however
there appears to be a significant total energy drift.

I have attached both the input script for the NVT and NVE run in the hope
someone can spot my mistake. The files are named "mail.nvt" and "mail.nve"
respectively.

i think we had this kind of report before and the explanation was
the the shake convergence was not good enough.
please try 1.e-5 or 1.e-6 instead of 1.e-4 and double check in
the mailing list archives, if my memory serves me right.

cheers,
    axel.

Although I have not used LAMMPS before I may offer two suggestions:

1. Is the time step small enough? Has it been used before for the
SPC/E water model? (I dont know what units "real" means)
2. In a different simulation package I notice that the shake precision
is a important parameter in determining energy conservation. For a
precision of 1e-5 A I would usually get terrible conservation, but if
I increase it to 1e-8 A things would improve dramatically.

Salomon

Thank you as usual for the swift responses. I have altered the tolerance of the shake algorithm and from the results of a very preliminary test it appears this is indeed the problem.

Many thanks, Jeff.