NVT and NPT are two commonly used ensemble during MD simulation of biomolecules. NPT must be used during equilibration just before changing to constant volume ensemble. So during production run NVT or NPT which one should be prefer and why?
By computer simulations you intend to model reality. To my opinion, production run should be NPT, once you model chemical reactions, (bio)molecular conformational changes, drug-target binding etc. These usually happen under constant (atmospheric) pressure (e.g. in a test tube or in a living cell).
Another reason to choose NPT might be a comparison with experimental data (binding constants, NMR data) which were measured under constant pressure.
Anyway, the differences between NPT and NVT are usually quite small (pV, the difference between Helmholtz and Gibbs free energies, is small).
NPT is better which is more close to the real experimental condition. However if your system is too large and NPT can not make the system stable, especially if you find the system start to diffuse, and the density is keep on reducing, then you better to use NVT. One sentence, if NPT works then just use NPT.
Experimental conditions can be NVT or NVP or something completely different (e.g. grand canonical). It doesn't make any sense to say that generally NVT (or NVT for that matter) is more similar to experimental conditions.
Please keep in mind that we are not talking about any specific experiment, i.e. either NVT or NPT. Just from MD simulation itself, what a simulation researcher consider when they perform simulation. Actually Michal, Bishwajit and I mean the same thing and also directly address the question. Considering NPT closest to reality is a very very general idea and well acceptable concept for simulation. Please better understand the question before you answer it.
I agree with all answers regarding that it all depends on what is the purpose of your simulation. In the case of fluid mechanics, it is very important to be close to the constant P or V conditions you wish to reproduce. But perhaps you are interested in studying the biological behavior of a macromolecule. Relatively short MD simulations receive almost no influence of constant V or P if the cell and solvent are properly constructed. Small differences do arise for long MD simulations but if this is the case consider that constant P may be slightly slower than constant V due to the barostat calculations.
Pavel is right about his comments. I will try to give you more specific answer on this issue. If you want to reproduce/calculate lattice constants of a crystal structure, bulk modulus of the mineral, Young's modulus, IR spectra, thermal expansion coefficient then certainly go for NPT ensemble. If you are calculating surface energy, interfacial energy or adsorption energy then you can go for NVT ensemble. You should also do test run using both ensembles in case of adsorption energy calculations. In case of surface atoms relaxation, if you are fixing (constraint) few atoms in your system and at the same time you want to relax surface atoms then NVT will do this job (basically thermal gradient MD). At the end of simulation you want to make sure that your system is not trapped in a local minimum. I also agree with other people's comment on their NPT preference over NVT.
As was discussed above, the choice of the ensemble is defined by the task. Since the experiments under constant pressure are more common the similar simulations are more common.
But there is subtlety, for example if we use the Berendsen barostat with short relaxation time, such as recommended in ancient times 0.1 ps, we get some what nonphysical ensemble. In particular, it can be seen because the temperature measured by the movement of atoms or molecules as a whole nonidentical. Unfortunately, we do not have the ideal "barostat" algorithm therefore forced to pay attention to this in exact calculations.
Yes, it all depends on the problem you are working on. If you are trying to study phase equilibria, or other thermodynamic properties of the system - proper ensemble is essential. If the system is solid, just constant pressure is not enough, you would have to consider proper stress tensor conditions. If it involves adsorption at the surface, you may also want to look at the pressure along the surface, as a separate component to control. If you are studying interactions of a few molecules surrounded by plenty of solvent - e.g. protein-ligand interaction, NVT or NPT most probably will not matter much provided that the density is correct, and the average pressure in NVT is where you want it to be. If you are looking at the properties that follow particles trajectories, like diffusion coefficient, beware of the barostats that rescale the particle coordinates. Running NPT first is a good way to equilibrate the system, and then run NVT to accumulate the statistics of interest.
for Biomolecules NPT is commonly used for equilibration, or simple minimization, but if you want to make MD of big system, as protein-ligand in solvant box, you need to use periodic conditions (PBC) and PME, for long distance interaction, these conditions are only available with NVT ensemble...
There is an indirect thermodynamic argument in favor of NPT for biomolecular simulations. This is based on arguing that a protein's native state is assumed to coincide with the global ΔG(folded-unfolded) minimum. To get ΔG you need NPT and not NVT (which would have given you the Helmholtz -and not Gibbs- free energy). This is the reason why methods like this http://www.ncbi.nlm.nih.gov/pubmed/18399522 require that the simulation is performed in NPT.
Nicholas M. Glykos: Interesting point, has anybody ever tried NVT vs. NPT for protein simulations and observed the difference in the observed native states in different ensembles?
@Michael: to observe experimentally --let alone computationally-- pressure-dependent differences in protein structure and function you need significant pressure changes (see for example http://www.ncbi.nlm.nih.gov/pubmed/8628735). This, of course, does not a change one iota in the underlying thermodynamics.
Any coupling of a system to a barostat or thermostat means interfering in the natural Dynamics of the system. In real system T Control occurs via Surface of the system by heat Exchange. All thermostats take the heat right from each particle. Similar problem is with barostat. Use NVE, if you can, for production run. If you equilibrated system and no phase Changes or chemical reactions are expected, the P and T will stay pretty much constant during the simulation (on average, of course, the fluctations are always there). NVE ensemble is the only natural ensemble. The rest of ensembles affect the system to some extent. Of course if you need to compare two systems at exactly the same T you have Little choice but NVT.
my apologies, it seems I went a little fast in my answer ... NPT also uses PBC, I never compared the two MD simulation on the same system, It Could be interesting ...
depends on which property you want to calculate. If you want to calculate thermo physical properties like density, then NPT is preferred. And if you want to calculate structural property then NVT is good and if you want to calculate dynamical property, then NVE is the best.
As many have already pointed out, there are many variables that have to be taken into account before answering to such a question.
Basically, it depends on what molecular property one wants to simulate/reproduce.
I would also like to pinpoint that, if one is interested in vibrational spectra (recovered from velocity autocorrelation function ---> FFT), then NPT and NVT simulations should not be employed, because the thermostat significantly alters the vibrational density of states (VDoS) at frequencies close and multiplies of the thermostat's own frequency.
In fact, all my papers in which I employed Car-Parrinello Molecular Dynamics were based on NVE simulations.
As i said before, if one is interested in recovering the vibrational spectrum from the trajectory trought autocorrelation functions, using a thermostat (i.e. NVT, NPT) is quite a risky approach.
In fact, the thermostat has a specific frequency which overlaps the range of frequencies spanned by the real vibrational spectrum.
For example, this frequency is often set at 3000 cm-1, 'cause this way it can better thermalize C-H stretching vibrations.
However, if this thermostat frequency is adopted, the region centered around 3000 cm-1 in the computed vibrational spectrum is deformed by artifacts, so it's no reliable.
I think it depends on what is the real experimental conditions. Mostly we perform experiments under constant Temp and Pressure. So it is better to go for NPT ensemble
The application of the NPT or NVE ensembles strongly depends on the conditions of phenomena you want to model. You can always bring you systems to the desired equilibrium conditions using NPT. However, sometimes the phenomena of interest does not happen in thermodynamic equilibrium for a period of time. In this case, you don't want to artificially "disturb" the system with the NPT, since the temperature is not the same across the systems. For instance, in a collision cascade event. Therefore, both NPT and NVE are very useful ensembles.
Indeed it depends on application what ensemble do you use for the equilibration, but for the production run the NVE has to be used to avoid the bias introduced by thermostat and/or barostat.
There are many cases where NVE should not be used, chiefly when non-NVE systems are under study. The thermo-/barostat is applied to provide the right ensemble, and many of us study test-tube-like conditions, which are normally NVP.
It depends to the work you want to do. For more stable conditions you need to go for NVE method, but if you want to make it more real and related to experimental work you should go for NPT method instead of NVT.
We can validate the good set only if we compare our results by experience. but during my long research, I have always validated the TNP provided you check the mass of your piston and the thermostat of Nosé during the simulation.
The difference between NPT, NVT and EM is rather fundamental for MD simulations. I suggest consulting a textbook on the subject, like the first of Fawaz's suggestions above (Leach).