It depends of the kind of experiment you are trying to simulate. There is no a better or worse option. Only a more suitable or less suitable choice of statistical ensemble.

As Mario has pointed out, it depends on what you want to simulate. If you want to emulate a laboratory condition, NPT would be the most realistic choice, since pressure and temperature of a closed system tend to be constant due to atmospheric pressure and physiological/room temperature of the surroundings. 

Thank you Burak and Mario, We do MD simulations (on PC) for enzymes at 310 Kelvin (body temperature), based on the MD analysis results we build pharmacophore models, and we test these results by ELISA (298 kelvin) . Due to the difference in temperature between body (where the enzyme is realistically working) and experiment temperature, we preferred to do NVE ensemble, so we did 11 nanoseconds of NVE simulation using NAMD, is our approach correct?

As Burak and Mario suggested, it depends on your system, and for your system it seems that NPT will be better choice.

MD have three different stages, minimisation, equilibrium and production run, and it all depend on the user what he want from these MD runs,  you need to go through the mailing list of namd, many people have asked these type of questions.

Thanks All,

but after finishing 11 nanoseconds of simulation using NVE, is there any problem to use the results of the simulation in further pharmacophore model building, (specially  that for using NVE we depended on the assumption that the temperature of the body is different from the temperature of experiment)?

And if your system is equilibrated enough, ideally, you should get same result from all the ensembles.

In case of any explicit solvent simulation it is essential to run in constant pressure

at some point to equilibrate the density. Whether you will use NVE/ NPT

depends on if you believe that your system will like change its shape during simulation time. In my opinion you use the following ensembles at different stages

1) During Heating: NVT

   It is necessary to use NVT as int low temperatures the

calculation of pressure is inaccurate and can cause the barostat to

overcorrect leading to instabilities.

2) Equilibrate the system using NVE/NPT

3) Production using NVT

if I don't expect the system to change shape, e.g. unfold

and fold back up etc.

Dear Ma'mon, 

what properties do you want to compute? Running NPT or NVT MD (which just means that your MD is sampling the corresponding ensemble, not that your molecules are moving as if your system is in contact with a thermostat/barostat) you might have some problem computing "dynamical" properties (e.g. vibrational spectrum, etc.). It is not "impossible", you just need to set up properly the "masses" of the thermostat/barostat (if you are using continuum dynamics extended ensambles, e.g. Nose'-Hoover chains for thermostats).

For the same reason, it is not advisable to run "NVT/NPT MD" if your objective is computing nonequilibrium properties along a nonequilibrium simulations. In this case, what you want to do is sampling a time dependent probability density function, and you have to use more sophisticated approaches if you want to obtain theoretically reliable physical results. 

On the contrary, if you are just interested in equilibrium properties, and the system is not affected by metastabilities, NVE simulations are equivalent to NPT at the P and T measured in the NVE MD, provided that your system is big enough so that the fluctuations of the kinetic energy and the virial are not too big with respect to T and P.

Ciao

Dear Ma'mon

Even if you are interested on "equilibrium properties" of physical systems, some thermodynamical properties can indeed depend on statistical ensembles. Many of textbooks on statistical mechanics claim that thermodynamical description is independent on statistical ensembles when one invoke thermodynamic limit, but this idea is wrong in presence of phase transitions, specially, discontinuous phase transitions. System with phase coexistence and nonhomogeneities can present anomalous thermodynamical behaviors, whose observation crucially depends on statistical ensemble that is considered. Any way, you have to take into consideration the experimental situation you are interested to compare your MD simulations. In my opinion, NVE ensemble contains more thermodynamical information than  any other ensemble, precisely, because of it is more close to the exact microscopic dynamics of the isolated system. 

You can see more about ensemble inequivalence in the Gross books "Microcanonical thermodynamics". I add here a link to the publisher of this book, but you probably can find an electronic copy in the net. 

Best regards, LV.

http://www.worldscientific.com/worldscibooks/10.1142/4340

...the comment of L.V. (which I'm not sure I completely understood - are you speaking of ergodicity, rare events, etc.?) brought to my mind another consideration. There are properties depending on fluctuations (e.g. the specific heat, which depends on the fluctuation of the hamiltonian in NVT). In this case the estimator of the property, i.e. the formula of the observable you must consider, depends itself on the ensemble you chosen. A paradigmatic example of this is exactly the specific heat, which depends on the fluctuation of the Hamiltonian in NVT (said already above) and on the fluctuation of the kinetic energy in NVE (...hopefully I'm not wrong - see the old paper of Lebowitz-Percus-Verlet on this:http://journals.aps.org/pr/abstract/10.1103/PhysRev.153.250). Perhaps you are not interested on the specific heat but similar considerations might apply to other observables as well.

Actually, I am speaking about the observation of anomalous response functions such as negative heat capacities, dependence of fluctuations and other related behaviors on the enviromental influence, such as correlation functions and critical phenomena, criteria of thermodynamical stability, etc. I and my coleague Curilef have discussed these questions with more details in the following paper: 

https://www.researchgate.net/publication/51989868

or http://iopscience.iop.org/1742-5468/2011/06/P06021

Precisely, dependence of thermodynamical description on statistical ensembles are better explained by fluctuation theorems that relate fluctuations of system macroscopic observables under influence of an enviroment and its response functions, such as the well-known canonical fluctuation relation that relates energy fluctuations and the heat capacity. For a more general context, such a flutuation relation should include thermal fluctuations of enviromental temperature, which makes possible the observation of states with negative heat capacities.  Canonical ensemble (an environment with constant temperature)  is unable to describe these states, which means that this conventional ensemble cannot be employed to study phase coexistence phenomena during discontinuous phase transitions. 

Article Understanding critical behavior in the framework of the exte...

I think the choice of one over another would depend on the properties you're interested in. In the perfectly ideal case, NPT, NVT and NVE should give you the same results.

Energy Equilibration : NVE

Temperature Equilibration : NVT