Dear Lammps users,
I have two questions
1-I need to change the thermostat in npt simulation.
e.g for nvt I used
fix 1 all temp/berendsen Tstart Tfinal Tdamp
fix 2 all nve
How should I change the thermostat to berendsen in npt simulation?
I know that “fix temp/berendsen” can not be used with “fix npt”
2-Is there any other barostat rather than Nose-Hoover that lowers the fluctuations? I have a solution+system that didn’t relax after 12 ns!
Thank you very much
Fatima
Dear Lammps users,
I have two questions
1-I need to change the thermostat in npt simulation.e\.g for nvt I used fix 1 all temp/berendsen Tstart Tfinal Tdamp fix 2 all nve
there are people that will argue that a berendsen algorithm
will not generate a correct nvt ensemble. at best an approximation.
How should I change the thermostat to berendsen in npt simulation? I know that "fix temp/berendsen" can not be used with "fix npt"
if you want to use pressure control from nose-hoove, but temperature
control with berendsen, then you have to use fix nph with fix temp/berendsen
please note that a berendsen thermostat is almost always not a good
choice. if you need to dissipate kinetic energy quickly, fix langevin will
be more efficient, if you need a proper statistical mechanical ensemble,
a nose-hoover (chain) thermostat will be better.
2-Is there any other barostat rather than Nose-Hoover that lowers the
fluctuations? I have a solution+system that didn't relax after 12 ns!
i suspect you are trying to solve the symptom and not the cause.
there has to be a reason for the fluctuations that originates in your
system set up or your equilibration strategy. i doubt that this will be
properly corrected by switching to a berendsen thermostat.
axel.
Dear Axel,
thank you very much for your respond.
I have compared an identical systems in Lammps and discover. (this system contains solution + a surface that seems to be the reason of the fluctuations)
nvt simulation by Lammps with Nose-Hoover thermostat and Tdamp=100 has ~ 5-10 kcal/mol drift in 4 ns simulation, std> 10 kcal/mol
nvt simulation by Lammps with Nose-Hoover thermostat and Tdamp=1000 -> no difference
nvt // // with berendsen // // =100 has ~ 1-3 kcal/mol drift // //, std> 8 kcal/mol
nvt // // with berendsen // // =1000 has NO drift , std >8 kcal/mol
nvt simulation by Discover with Anderson thermostat and collision ratio=0.5 and only “500ps” has no drift + std <2 kcal/mol
(the npt simulations fluctuate much more as I said after 12 ns I still haven’t get to the EQ)
Is Anderson method available in Lammps? (From the manual I almost sure that it is not). The std of more than 8 still seems to be so high!
I have to check the “chain Nose-Hoover” for the thermostat and barostat.
Thank you again
Fatima
Dear Axel,
thank you very much for your respond.
I have compared an identical systems in Lammps and discover. (this system
contains solution + a surface that seems to be the reason of the
fluctuations)
there is no principal reason why this should be different.
it may just be that you have to follow a multi-step protocol
to reach equilibrium quickly. as was mentioned several times
before (and if i remember correctly also in response to an
inquiry by you), the nose-hoover thermostat is not a good
algorithm to bring a system into equilibrium. specifically,
if the starting conditions are not ideal (which is likely from
what you describe).
do you do a minimization before starting the MD?
perhaps you may even need multiple "quenches"
where part of the system is not moving until you
have the atoms at your interface relax well.
if you simply piece a multi-phase system together
you will induce high amplitude phonons that may
take a very long time to disappear unless you use
a dissipative thermostat or some other strategy to
get rid of this "initial shock".
nvt simulation by Lammps with Nose-Hoover thermostat and Tdamp=100 has ~
5-10 kcal/mol drift in 4 ns simulation, std> 10 kcal/mol
nvt simulation by Lammps with Nose-Hoover thermostat and Tdamp=1000 -> no
difference
nvt // // with berendsen
// // =100 has ~ 1-3 kcal/mol drift
// //, std> 8 kcal/mol
nvt // // with berendsen
// // =1000 has NO drift , std >8 kcal/mol
nvt simulation by Discover with Anderson thermostat and collision
ratio=0.5 and only "500ps" has no drift + std <2 kcal/mol
(the npt simulations fluctuate much more as I said after 12 ns I still
haven't get to the EQ)
anderson is a dissipative thermostat. please try langevin with nve in lammps.
otherwise you are comparing apples and oranges.
Is Anderson method available in Lammps? (From the manual I almost sure that
no. as can be inferred from the manual. langevin is better suited in any case.
it is not). The std of more than 8 still seems to be so high!
as i wrote before, if you are not careful with how you set up your
input, all kinds of nonsense can happen. just running longer
may not always help you.
axel.
What do you mean by "drift" over 4 nsec?
I don't presume you mean the temperature drifts,
since you are thermostatting. Thus do you mean
the potential energy drifts? If it does, the presumably
that means the conformation of your system has changed
to lower the potential energy on a long timescale. That
doesn't sound like a thermostat issue, but an equilibration
issue. Also for Tdamp = 100, I presume you are using
real units, not metal (which would be far too large a damping
constant)?
Steve