I have been trying to understand NVT simulations in LAMMPS with the
behavior of Tdamp parameter in fix nvt command for a while. Since Im
familiar with DYNAMO, I thought that Tstart and Tend stand for the
requested temperature. However, when I tried to heat a bulk terminated
Cu FCC crystal to 50K with the following commands;
1 all fix nvt 0.0 50.0 0.1
timestep 0.001
thermo 50
run 100000
and investigate the temperature change with time, I realized that there
are sinusoidal temperature changes just like packets around the
equilibrium temperature. Moreover, when I decrease the value of Tdamp to
0.01, system can find an equilibrium state around 50K without sinusoidal
effects but the simulation runtime for 8 processor takes almost same wall
time with serial run in DYNAMO. Is the sinusoidal damp effect a suggested
situation or is there an appropriate way to set Tdamp for EAM NVT
simulations in LAMMPS? Nonetheless, how can I set a correct Tdamp with
drag parameter for NVT simulations for a fast simulation on LAMMPS? Thank
you in advance.
You assumed correctly that Tstart and Tend are the desired temperature at the start and end of the run, as written in the documentation. Tdamp gives a relaxation time for the temperature. So your system should relax within tdamp to the desired temperature. As written in the documentation the thermostat may produce some occilations around the desired temperature. You could try to add a drag parameter
1 all fix nvt 0.0 50.0 0.1 drag 0.2
this might suppress the occilations.
Decreasing tdamp to 0.01 means that your temperature is forced within 10 timesteps to the desired temperature. As you only give out the temperature every 50 steps you probably cant see any occiliations even if they are there.
That whole nvt parameter stuff and the thermostat should not have any real impact on the computational time as far as i can see it. Basically the only thing it does is rescaling all velocities. And it does that no matter what paramters you give, so there should be no change in computational time at all.
Re: speed of DYNAMO vs LAMMPS - there are many
things that could affect this. The thermostat isn't one
of them. So I think there must be some other issue.
How does the speed compare on one proc with both
codes?
Also, note that NVT in LAMMPS is Nose/Hoover. I don't
think any of the DYNAMO thermostats is exactly Nose/Hoover
(maybe just Hoover), and some of them are very different.
Thanks for your help Steve and Christian. In fact, for 10 picoseconds
test, LAMMPS run with single proc spent 7.5 hours with the command
fix nvt 50.0 50.0 0.1. However, DYNAMO run spent 2.7 hours of
runtime for the same test system. To visualize the system and to
understand the problem in NVT simulation, I chose 8 procs for test
runs. All I have done was to change the Tdamp parameter from 10 to
0.01 with intervals and to investigate the differences in the results.
The results were same as I wrote before. Thanks for Christians
suggestions, Ill test the system again by using a drag value and take
temperature output step smaller.
When dumping the positions from atoms contained within a certain box size
(inside a larger region) I noticed that the atom ids at the beginning of the
file are very different from the ones at the end of it; for instance:
Timestep1
Atom #
1 x y z
2 x y z
3 x y z
Timestep N
Atom #
20 x y z
23 x y z
2 x y z
I was wondering how could I dump the positions of specific atoms so that the
ids at the begining of the file correspond to the ones at the end of it?
(this happens at any temperature)
It sounds like you're using the LAMMPS option that
allow you to dump atom positions of atoms from a
specified geometric region. And sounds like you want
to dump atom positions of atoms from a specified
group, which is even easier to do in LAMMPS. Try
something like:
dump 1 JaimeGroup atom 100 dump.atom
where you define JaimeGroup with the group command.
See the docs here:
Or I suppose you may just be seeing garbled order of
atom ids in the dump file due to running on multiple
processors. From the docs: "Also note that when LAMMPS
is running in parallel, the atom information written
to dump files (typically one line per atom) may be
written in an indeterminate order. This is because
data for a single snapshot is collected from multiple
processors. This is always the case for the atom,
bond, and custom styles. It is also the case for the
xyz style if the dump group is not all. It is not the
case for the dcd and xtc styles which always write
atoms in sorted order. So does the xyz style if the
dump group is all.")
I hope this helps.
Paul
Dear Lammps users,
When dumping the positions from atoms contained
within a certain box size
(inside a larger region) I noticed that the atom ids
at the beginning of the
file are very different from the ones at the end of
it; for instance:
Timestep1
Atom #
1 x y z
2 x y z
3 x y z
Timestep N
Atom #
20 x y z
23 x y z
2 x y z
I was wondering how could I dump the positions of
specific atoms so that the
ids at the begining of the file correspond to the
ones at the end of it?
(this happens at any temperature)
Thanks
Jaime Sanchez
_______________________________________________
lammps-users mailing list [email protected]
Berk - can you email me your input script for the LAMMPS run?
I don't know of any problem where DYNAMO is faster than
LAMMPS, unless the geometry of the problem is odd.
When using NPT or the NPH ensemble how do you set the parameters to keep the
pressure from oscillating? I'm trying to set my system at atmospheric
pressure (1.013 bar) at a certain temperature but I notice that the pressure
oscillates largely between large positive and negative values.
Large pressure fluctuations in MD simulations are to
be expected. Especially for simulations of liquids.
Larger system sizes will naturally have smaller
fluctuations, so you could simulate a bigger box and
get less fluctuation. But, of course it would be more
computationally expensive.
Paul
Hello everyone,
When using NPT or the NPH ensemble how do you set
the parameters to keep the
pressure from oscillating? I'm trying to set my
system at atmospheric
pressure (1.013 bar) at a certain temperature but I
notice that the pressure
oscillates largely between large positive and
negative values.
Any suggestions?
Thanks
Jaime Sanchez
_______________________________________________
lammps-users mailing list [email protected]