It is my understanding that having fluctuation in the pressure is expected in condensed-phase MD simulations. I have read this is due to small changes in volume that results in large changes in pressure, and the equilibration between these two does not occur simultaneous. The best response I have seen was written by David Case at http://ambermd.org/Questions/pressure.html, which will guide you further into the thermodynamics of the problem. On top of this is how good the algorithms are at regulating the pressure.

With that said, I am not sure if 100-200 bar fluctuations are expected and probably depends on the program you are using. As a check perhaps run a simulation on a pure water box, one that has been done previously, and see what the pressure fluctuation is there. You could also ask this question specifically to your program's listserv, if available, to see what others have experienced.

A better metric for monitoring simulation general equilibration and stability, and how well it models reality, is the system's density. If your density remains stable throughout your production run and the average is near the experimental neat value for your solvent (ie. 1 g/cm3 for water), then the reliability of your systems improves. As you likely know, monitoring the model's pressure, energy, and density are the most basic metrics to monitor of a systems equilibration and stability. However, there are many more metrics that can be monitored for equilibration - most importantly those that are influence the observables that you're interested in (e.g. conformational dynamics of a small molecule, secondary structure of a protein).

It is my understanding that having fluctuation in the pressure is expected in condensed-phase MD simulations. I have read this is due to small changes in volume that results in large changes in pressure, and the equilibration between these two does not occur simultaneous. The best response I have seen was written by David Case at http://ambermd.org/Questions/pressure.html, which will guide you further into the thermodynamics of the problem. On top of this is how good the algorithms are at regulating the pressure.

With that said, I am not sure if 100-200 bar fluctuations are expected and probably depends on the program you are using. As a check perhaps run a simulation on a pure water box, one that has been done previously, and see what the pressure fluctuation is there. You could also ask this question specifically to your program's listserv, if available, to see what others have experienced.

A better metric for monitoring simulation general equilibration and stability, and how well it models reality, is the system's density. If your density remains stable throughout your production run and the average is near the experimental neat value for your solvent (ie. 1 g/cm3 for water), then the reliability of your systems improves. As you likely know, monitoring the model's pressure, energy, and density are the most basic metrics to monitor of a systems equilibration and stability. However, there are many more metrics that can be monitored for equilibration - most importantly those that are influence the observables that you're interested in (e.g. conformational dynamics of a small molecule, secondary structure of a protein).

I agree totally with Dr. Kirschner. I can just add, based on my experience with Gromacs, that I had the same big fluctuations of 100-200 bar in NVT simulations.

How did you prepare the ensemble before the running? Since you are using NVT you have no control on the pressure. If you perhaps try to do a long NPT simulation before the productive run, in which you set the pressure at 1 bar and then take the structure from this NPT simulation with a volume close to the average of the NPT run and you use it as starting structure for the NVT ensemble this perhaps help a little bit more... and then of course monitoring density and also for example stability of mainchain or c-alpha rmsd over time wll be a good indication of the stability of your trajectory...

Question is: If you run NVT, how do you specify pressure?

The only additional comment I can make to the previous answers is that molecular models are not exact, especially at close separation where the intermolecular forces grow faster with decreasing distance; thus a 100 bar fluctuation is not that uncommon.

Of course, when using the NVT ensemble you do not control the pressure, but the volume; it would be better to proceed as indicated by Elena Papaleo.

Everyone in this thread is right on target. When you fix (NVT), it is impossible to completely choose what the final pressure will be. For a given molar volume and temperature, a single-phase system will at equilibrium have a single value of the pressure - you can only have two independent intensive variables in a single-phase thermodynamic system, as seen in the Gibbs phase rule. So although you may arrange (NVT) to correspond to a particular pressure initially, after equilibration the pressure will inevitably change to a different value. And it could be a very, very different value or very similar, depending on what you choose (NVT) to be.

I agree with Elena. You can't control pressure in NVT run. In the second NVT run (after first NPT run), pressure should fluctuate around 1 bar though the fluctuations can be large due to the system size and the type of barostat.

Hello,

in my opinion, you should first do NPT solutions to find the equilibirum volume and then perform NVT simulations. After NPT simulaitons, your mean value during NVT simulations should be zero or your target pressure (mentioned in NPT file).

some fluctuations might exist, depending probably on your thermostat time constant.

100-200 bars = 0.01-0.02 GPa is probably not very big; so maybe this fluctuation is ok.

kind regards

Guilhem

Thanks you so much. The answers are very insightful.

I did first, as suggested, NPT simulations at P=1bar for 2ns. And, got the equilibrium P=1±0.02 bar. However, the fluctuation were in range of ~150 bar. Then I used this coordinate file to do my NVT simulations. But, In my NVT simulations I got the equilibrium P=150 bar with fluctuations of ~150 bar while the specified pressure was 1 bar.

Now, I think I should do NPT simulations for some more ns and then do NVT to get the fluctuations around 1 bar than 150 bar.

Scale your starting configuration so that it corresponds to the average box volume from the NVP simulation. And again: the pressure is not specified when you do NVT simulations.

Indeed, NVT and Pressure control sound at odds, however, I think you can get the stress tensor and get estimates of the pressure. If your box and density are approximately right, then it all jives.

In general, equilibrating the system using NPT, then using the average volume that you get from NPT run and using it as volume to be fixed for NVT may work. But, all also depend on the Barostat-thermostat that you use for NPT. (Not all of them produce the right NPT statistical ensemble - read more about that from Martyna & Tuckerman work.) Therefore, when starting NVT simulations you should look carefully at the pressure fluctuations and the average values of the pressure (in NVT) as well. If your pressure is too much different from the average pressure that you had from NPT (you should also have an average value of the pressure from NPT runs!), then you should slightly change the volume of the box (either increase or decrease depending on the pressure).

You should know that most of the thermostats (maybe except the Berendsen thermostat) produce correct NVT ensemble, so the NVT simulations could be seen as more accurate. Nose-Hoover or Martyna-Tuckerman-Tobias-Klein type of thermostat/barostat could be more appropriate for large systems.