> On Feb 4, 2015, at 3:12 AM, Robin Jain <robinjain.chem.gmail.com> wrote:
>
> Dear all,
> I am confused in md run and wants your help.
> I have a organic molecule in a solvent and I have equilibrated my
> system in NPT ensemble for 5 ns. Now i want to run production. Some
> body told me that for production, you should first 10 ns run in NVT
> ensemble and finally 50-60 ns run in NPT ensemble
Did you ask them why? Did they give you a justified answer (besides “it is what I always do”)?
> while some told me
> that you should only run in NPT ensemble for 60-80ns.
Same comment. Did you ask them why? Was their answer theoretically or practically justified? This advice is of little use if it is just a “magical recipe” (at least in my opinion).
> What should i
> do. Please help me in this regard.
What is it you are trying to learn with your simulation? First and foremost, the answer to that question must inform your choice of simulation. In some cases, that answer will tell you what to do. For example, if you want to compute diffusion coefficients, you need to run NVE. If you want to compute an isothermal compressibility, you need to run NpT (with the Monte Carlo barostat). If you want to study a phase transition, you need to mimic the experimental ensemble (that usually means NpT). If your system is *not* near a phase transition and you don’t need any temporal properties (or other constraints that dictate which ensemble you need to use), then the unsatisfying answer to “what should I do” is “it doesn’t really matter”.
Statistical mechanics shows that in the (incompressible) condensed phase, the different ensembles are equivalent. In this case, practical considerations of the simulations informs your protocol. People typically avoid NpT (after the density is equilibrated) since it is more computationally expensive than NVT (allowing you to run longer simulations and get more sampling), although that is less true now than it once was due to the MC barostat. In turn NVT is more expensive than NVE, but the use of a thermostat allows you more flexibility with respect to using settings that lead to more energy drift in NVE, since the thermostat will remove any residual energy that integration errors pump into the system. So NVT is usually more popular than NVE for that reason (allowing you slightly longer time steps or slightly looser SHAKE/Ewald tolerances).
So here is what I suggest -- read papers that report studies similar to what you are trying to do. See what they did, and start from there (not all published research is good published research, so it’s important to read it with a critical eye and make sure their protocol is justified theoretically). If you want to, try modifications of what they did informed by theoretical and practical considerations to try and come up with the most efficient simulation protocol for your system and hardware. Or just do what they did.
At the end of the day, as long as your choices are justified by the theory requirements of your study, it doesn’t matter what you do.
HTH,
Jason
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Feb 04 2015 - 02:30:03 PST
