I’m trying to simulate adsorption of ions onto a molecule, using OPLS-AA FF (by Moltemplate) in water SPC/E.
The results of density and temperature indicate good equilibration of the system.
When using npt, nvt, or nve modes, there is a constant drift of the whole system to the -x direction. I had seen drift with Langevin thermostat, but this is the first time, I noticed it when using Nosé–Hoover thermostat and barostat.
Has anyone had the same experience? Any suggested solution?
My input file is as follows.
Thanks so much
-X
Hi Xiaolin
Incidentally, I am not surprised you would see the system’s center-of-mass (COM) drift while using Langevin dynamics, however I would be surprised if it was drifting in a non-random way (for example, consistently moving in the -x direction).
Anyway, consistent movement of the COM in the -x directions sounds like the “Flying Ice Cube” phenomenon. It is most frequently associated with the Berendsen algorithm. However I am under the impression that other deterministic thermostats have “translational artifacts”. I’ve certainly seen very strange looking dynamics using Nosé-Hoover thermostat, especially when the simulation begins with the atoms arranged in a highly ordered, non-equilibrium manner.
Hence, if you prefer using Nosé-Hoover, have you tried equilibrating the system for a while using Langevin dynamics and then switch to Nosé-Hoover? Does that help?
Andrew
P.S. If you want to keep the center-of-mass of a group of atoms from moving or rotating, see:
http://lammps.sandia.gov/doc/fix_momentum.html
also see:
http://lammps.sandia.gov/doc/fix_recenter.html
Hi Xiaolin
Incidentally, I am not surprised you would see the system's
center-of-mass (COM) drift while using Langevin dynamics, however I would be
surprised if it was drifting in a non-random way (for example, consistently
moving in the -x direction).
Anyway, consistent movement of the COM in the -x directions sounds like
the "Flying Ice Cube" phenomenon. It is most frequently associated with the
Berendsen algorithm. However I am under the impression that other
deterministic thermostats have "translational artifacts". I've certainly
seen very strange looking dynamics using Nosé-Hoover thermostat, especially
when the simulation begins with the atoms arranged in a highly ordered,
non-equilibrium manner.
Hence, if you prefer using Nosé-Hoover, have you tried equilibrating the
system for a while using Langevin dynamics and then switch to Nosé-Hoover?
the drift doesn't originate so much in the thermostat, but rather from
the force computation.
certain factors enhance it:
- asymmetric/inhomogeneous system setup
- using floating point math with not very long cutoffs and without
force switching.
- velocity scaling thermostats like temp/rescale or berendsen
- imbalanced initial velocity distribution
- frequent switching between fractional, when the center of the system
is not also the origin
- running many rigid particles without excluding the internal non-bonded forces.
other factors dampen it:
- "noisy" thermostats like the various langevin variants
- removing and drift via fix momentum (but careful on systems with
non-periodic boundaries and immobile atoms where it can cause unwanted
artifacts)
axel.