Dear Dr. Axel
Thank you for the useful comments. They are extremely helpful.
I was able to later anticipate some of the issues you pointed out such as the volume of the box. I now use a thermostat (Langevin) and a barostat (berendsen) with nve.
I don’t think that that will avoid the issue of instantaneous versus average box size.
Regarding the scaling, I found this procedure in two publications reporting results comparable to experiments. Therefore I kind of disagree with your statement saying that the procedure is unscientific.
These days, if something about MD simulations is published, it doesn’t automatically give it credibility. One has to look at the context. It is possible to publish MD simulation based research in journals where the reviewers have rather little experience with (and/or pay little attention to) statistical mechanics principles. In this specific case, the scaling not only makes little sense from that statistical mechanical point of view, it is contrary to common sense, too. So there would have to be a very good justification for the procedure.
Also, you have to take under consideration that you are misunderstanding the description.
In addition, I referred to a similar question in the manual where the exact solution I mentioned was proposed to overcome the drop in temperature due to the damping effects.
Pplease point out the exact location. Perhaps the text is misleading. As I mentioned above, context matters a lot in this case. Typical cases are when you are using constraints or rigid bodies or when you start from a minimized (and thus 0K) starting configuration and want to anticipate the amount of kinetic energy converted to potential energy during equilibration.
Scaling the velocities by an extra 50K does not really yield a far point from equilibrium.
again, this depends on the context and it is unnecessary. there has to be a good justification for it. as I mentioned in my previous email, you are just destroying most of the effort invested in equilibration.
The Langevin thermostat should actually yield equilibrium during the first few ps (given an appropriate damping parameter) which is something I tested and confirmed by setting the damping to 100fs.
Nevertheless, using npt to reach the desired temperature is a more stable approach but I believe there is no difference when it comes to changes in temperature as low as 50K. In fact, 50K is a very normal fluctuation range during the production run.
that is a misconception. while the instantaneous temperature may fluctuate, its fluctuation stems from the transfer of energy between kinetic and potential energy. the total energy should stay constant. with scaling you add to the total energy and that is different.
Regarding the test you suggested, I performed it and the temperature dropped sharply from 300 to almost 10K. Does that mean that the system was not well equilibrate in NPT ?
such a sharp drop indicates a serious problem. there is something very wrong with your input(s) and/or procedure. there may be a small change but the first set of energies should be the same as the last set of the previous simulation when you are doing a proper restart or continuation.
Thank you for pointing out the air suitability as an implicit solvent. I will have to look more into this.
please do the computation of the number of gas molecules for your simulation volume. it is worth the effort.