When I am running simulation for small metal particles of 180-350 atoms with non-periodic boundary conditions, I get temperature fluctuations of +/- 15K above 100K, and +/- 25K at T~400K . I tried to reduce these fluctuations to get better control of temperature by applying the following commands drag 0.5, Tdamp=0.2 (instead of 0.1) and tloop=3 for fix nvt, but this did not help me much.

Could anyone give a suggestion how can I reduce these fluctuations to +/- 5K or at least to +/- 10K? Is it even possible for metal clusters of such size in LAMMPS? Because even when I run simulation with fix nve my system still fluctuates wildly (+/-20K).

When I am running simulation for small metal particles of 180-350 atoms with
non-periodic boundary conditions, I get temperature fluctuations of +/- 15K
above 100K, and +/- 25K at T~400K . I tried to reduce these fluctuations to
get better control of temperature by applying the following commands drag
0.5, Tdamp=0.2 (instead of 0.1) and tloop=3 for fix nvt, but this did not
help me much.

Could anyone give a suggestion how can I reduce these fluctuations to +/- 5K
or at least to +/- 10K? Is it even possible for metal clusters of such size
in LAMMPS? Because even when I run simulation with fix nve my system still
fluctuates wildly (+/-20K).

<sigh> here we go again </sigh>

why should it *not* fluctuate? and what make a fluctuation "wild"?

temperature is only "constant" for infinitely large systems. we
compute temperature for small(er) systems under the assumption of
equipartitioning and that for a system in equilibrium the ensemble
average is equivalent to the time average. take the most extreme
example, a harmonic oscillator. its "temperature" fluctuates between
zero and 2*<T>. now, if you look at multiple such oscillators, each
with slightly different natural frequencies, they won't be all in
sync, thus some of the fluctuations will cancel. what you have is the
same. thus the only reliable way to reduce the fluctuations, is to
simulate a larger system. all of the tampering with thermostats will
only result in unphysical trajectories. i suggest, you'd rather pick
up a test book on statistical thermodynamics and make sure that you
properly understand what "temperature" (and other thermodynamic
properties) mean on the atomic scale, and how they are connected to
macroscopic properties.

axel.

p.s.: i don't know how many times i've explained this very same thing
here. it should pop up on the first page of searches in the mailing
list archives on this topic...

> Dear LAMMPS users,
>
> When I am running simulation for small metal particles of 180-350 atoms
with
> non-periodic boundary conditions, I get temperature fluctuations of +/-
15K
> above 100K, and +/- 25K at T~400K . I tried to reduce these fluctuations
to
> get better control of temperature by applying the following commands drag
> 0.5, Tdamp=0.2 (instead of 0.1) and tloop=3 for fix nvt, but this did not
> help me much.
>
> Could anyone give a suggestion how can I reduce these fluctuations to
+/- 5K
> or at least to +/- 10K? Is it even possible for metal clusters of such
size
> in LAMMPS? Because even when I run simulation with fix nve my system
still
> fluctuates wildly (+/-20K).

<sigh> here we go again </sigh>

why should it *not* fluctuate? and what make a fluctuation "wild"?

temperature is only "constant" for infinitely large systems. we
compute temperature for small(er) systems under the assumption of
equipartitioning and that for a system in equilibrium the ensemble
average is equivalent to the time average. take the most extreme
example, a harmonic oscillator. its "temperature" fluctuates between
zero and 2*<T>. now, if you look at multiple such oscillators, each
with slightly different natural frequencies, they won't be all in
sync, thus some of the fluctuations will cancel. what you have is the
same. thus the only reliable way to reduce the fluctuations, is to
simulate a larger system. all of the tampering with thermostats will
only result in unphysical trajectories. i suggest, you'd rather pick
up a test book on statistical thermodynamics and make sure that you
properly understand what "temperature" (and other thermodynamic
properties) mean on the atomic scale, and how they are connected to
macroscopic properties.

Furthermore, in these text books you'll find analytical expressions for the
magnitude of these fluctuations as a function of system size and
temperature. So you've got the opportunity to verify if the fluctuations
you see are appropriate for your system in the NVT ensemble.

What I really expected to hear is something like "decrease the heat pumping rate and increase time of simulation’’ or “use another algorithm for simulation that is more suitable and precise for small systems”. I am not a specialist but I hardly can believe that metal particle of 350 atoms can change its temperature from 225 to 247 after 1 ps in reality. It could be so and as Kristof said I’m really have an opportunity to check this and I’m going to.

I am sorry Dr. Axel if you see the message like mine very often. I tried to find the answer in archive and in google first. So it seems that I failed.

What I really expected to hear is something like "decrease the heat pumping
rate and increase time of simulation'' or "use another algorithm for
simulation that is more suitable and precise for small systems". I am not a
specialist but I hardly can believe that metal particle of 350 atoms can
change its temperature from 225 to 247 after 1 ps in reality. It could be so
and as Kristof said I'm really have an opportunity to check this and I'm
going to.

temperature as you understand it, is not a well defined in the context
of so few degrees of freedom. that is why we point you to statistical
thermodynamics text book literature, as those provide the theoretical
background that connects macroscopic themodynamic observables to
microscopic properties.