[gmx-users] Conceptual question: Why are MD production simulations usually run in the NVT ensemble?

[Justin A. Lemkul]

On 7/14/12 11:06 AM, Andrew DeYoung wrote:
> Hi,
>
> I'm not sure if this is an appropriate question for the Gromacs users'
> mailing list.  If it isn't, please forgive me and disregard this message.
>
> I asked this question on Physics Stack Exchange: Why is the canonical (NVT)
> ensemble often used for molecular dynamics (MD) simulations?  My question
> and an answer are posted here:
>
> http://physics.stackexchange.com/questions/31997/why-is-the-canonical-nvt-en
> semble-often-used-for-classical-molecular-dynam
>
> My question is, do you agree with the answer?  I'm not sure that I do.  The
> responder seems to imply that it is practically impossible to use periodic
> boundary conditions in an NPT simulation.  But, I think that the algorithms
> in Gromacs do just this; am I correct in this belief?
>

I also disagree with the answer you got, on several grounds.  I will admit that 
I have never gone into the code for how boxes are scaled and how this is all 
done in terms of bookkeeping, but I think there is plenty of evidence that we 
have decent barostat algorithms.  The response you got seemed to just say "it's 
all too inconvenient to try to think about, so we shouldn't do it."  Part of the 
response was simply that if you increase the system size, the ensembles become 
the same, which may be true in a theoretical sense (since P fluctuations 
decrease proportionally with system size) but it becomes totally impractical for 
doing simulations.  How then, does one account for heterogeneous systems like 
I've got with a membrane, protein, water, and maybe small molecules?  I can't 
just keep scaling that up and assume that the relative majority of water 
molecules is going to make everything happy.  That, and the lipid model was 
parameterized for NPT...

> People do often run _equilibration_ simulations in the NPT ensemble, I
> think.  They do this usually, I think, to obtain the proper density -- the
> code changes the box dimensions until the proper average pressure is
> reached.  Then, for _production_ simulations, people usually use the NVT
> ensemble, where the dimensions of the simulation box are held fixed.  My
> question here is, why is the NVT ensemble, rather than the NPT ensemble,
> typically used for production MD simulations?
>

I disagree with this premise as well.  The vast majority of all biological 
simulations I see are done under NPT conditions.  I think the heart of the issue 
is the discipline you're studying and what is commonly done.  Maybe in pure 
physics or chemistry simulations, NVT is more common because the experiments are 
done that way.  For us in biological science, the NVT ensemble is less relevant 
than NPT because we are studying processes that go on in cells, within 
membranes, or even just in solution an open-topped test tube or flask that is 
under atmospheric pressure.  The general advice from many on this list has 
always been, "pick an ensemble that models the reality of your system best."  If 
it makes sense to switch back to NVT from NPT for whatever you're doing, by all 
means do it.

> Do you have any suggestions for helpful reading that I can do on this topic?
> Thank you for your time!
>

I think it's a philosophical discussion, but comes down to what it is you're 
modeling and what the most appropriate conditions are.  Knowing the ins and outs 
of thermostat and barostat algorithms would be very important in debating these 
issues.

-Justin

-- 
========================================

Justin A. Lemkul, Ph.D.
Research Scientist
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin

========================================