fix NVT command

Hello Users,
In manual the example script for thermal conductivity of solid Ar, the fix NVT reads as

"fix NVT all nvt temp $T $T 10 drag 0.2 "

For Si should the drag value be same? If not what should be its value?

Thanks & Regards

If your system is well equilibrated, you don't need
the drag parameter at all. It's just a crutch to avoid
big oscillations when you are far from equilibrium.

Steve

Thank you Dr. Steve. Actually for nvt with 500K in equilibrium i am getting a fluctuation of 480-520(4%)(I can’t understand whether it is well equilibrated or not!! ) whereas the drag value was 0.2 and Tdamp was 10unit(ps) but it takes a time about 100ps. I wonder why it is taking 100ps instead of 10ps?

Thank you Dr. Steve. Actually for nvt with 500K in equilibrium i am getting
a fluctuation of 480-520(4%)(I can't understand whether it is well
equilibrated or not!! )

It depends on the system size. Generally, fluctuations (divided by
the average value) decay by 1/sqrt(n). For a system with less than
10000 atoms, that seems plausible to me. (I'm not sure.)

To decide whether a system is equilibrated you typically do not only
look at the fluctuations in kinetic-energy/temperature. It requires
some intuition about the behaviour of your system, and what kind of
states it can be in and transitions it can have. You can not prove
that a system is in equilibrium. The only thing you can do is check
to make sure it is not obviously out-of-equilibrium.

(There may be statistical tests you can use to help you. We have some
home-brew code to test for this but I was afraid to release it. But
even these methods do not prove that your system is in equilibrium.)

whereas the drag value was 0.2 and Tdamp was
10unit(ps) but it takes a time about 100ps. I wonder why it is taking 100ps
instead of 10ps?

Check out Axel's recent post:
http://sourceforge.net/mailarchive/message.php?msg_id=30042214

The main point that I want to make is: Do not use a drag parameter
when you are ready to run your main simulation (your production run).
You should only consider using a drag parameter at the beginning, when
your system is far from equilibrium.

  If I understand correctly, adding a drag parameter alters the
statistics of your ensemble, (preserving the average kinetic energy,
but reducing fluctuations around that average). The resulting
distribution is not canonical. This makes it more difficult to
compute thermal averages of other quantities later (especially if you
intend to use statistical methods like WHAM to reweight your data, or
fancy sampling methods like, replica-exchange. Avoid the Berendsen
thermostat and velocity rescaling during your production run as well.)

Cheers!
Andrew

Thank you Dr. Steve. Actually for nvt with 500K in equilibrium i am
getting a fluctuation of 480-520(4%)(I can't understand whether it is well
equilibrated or not!! ) whereas the drag value was 0.2 and Tdamp was
10unit(ps) but it takes a time about 100ps. I wonder why it is taking
100ps instead of 10ps?

for MD simulations you don't have simple laws
where you have relations like "if you do <something>,
it will always be correct". many effects are subtle,
often you don't see what is the right thing to do
until *after* you have done a simulation and analyze it.

as to equilibrium and reaching it. while you working to
equilibrate your system, all bets are off, i.e. doing something
that is "wrong" for a production run, can be helpful,
and something that is "right and proper" can make
it difficult. since "equilibrium" is a state, it doesn't
matter how you get there. problem is, you can't tell
when you've reached it. any but the most trivial systems
have far too many degrees of freedom to tell.

you will have a much better understanding for the
whole problematic, after subjecting yourself to a
healthy does of statistical mechanics. like the doctors
always say "this will hurt a little bit", but it is well
worth it. since this will turn "magical" parameters
into something that you can master, and black
magic into white magic. :wink:

cheers,
     axel.