What's the advantage of non-equilibrium molecular dynamics (NEMD) when compared to equilibrium molecular dynamics (EMD)? In what conditions should we choose NEMD?
Simulations cannot be expected to give values found by experiments, even if much progress has been made in this direction the last few years. The expression for the interaction potential is a guess for a real system. The advantage of NEMD is that trends and variations in properties can be traced back to part(s) of the interaction potential that causes the variation. This property means that NEMD gives molecular insight in system behavior. It becomes possible to distinguish between important and minor effects. This is important when it is necessary to introduce assumptions. This is the usefulness of NEMD: it acts as a tool to help understand experimental results, and it can be used to test assumptions made in the theory. Since it is difficult to do experiments in heterogeneous systems, the technique is very valuable.
cited from 2013 Studying non-equilibrium thermodynamics- from Book Non-Equilibrium Thermodynamics Of Heterogeneous Systems
non-equilibrium molecular dynamics (NEMD), has the added advantage that it can, in principle, be used to calculate non-linear as well as linear transport coefficients. They can be calculated as a function of external field strength, frequency or wavevector. The most efficient, number independent way to calculate mechanical transport coefficients is to ignore the beautiful results of response theory and to duplicate the transport process, essentially as it occurs in nature.
To calculate thermal transport coefficients using computer simulation we have the same two options that were available to us in the mechanical case. We could use equilibrium molecular dynamics to calculate the appropriate equilibrium time correlation functions, or we could mimic experiment as closely as possible and calculate the transport coefficients from their defining constitutive relations. Perhaps surprisingly the first technique to be used was equilibrium molecular dynamics (Alder and Wainwright, 1956). Much later the more efficient nonequilibrium approach was pioneered by Hoover and Ashurst (1975). Although the realistic nonequilibrium approach proved more efficient than equilibrium simulations it was still far from ideal. This was because for thermal transport processes appropriate boundary conditions are needed to drive the system away from equilibrium - moving walls or walls maintained at different temperatures. These boundary conditions necessarily make the system inhomogeneous. In dense fluids particles pack against these walls, giving gives rise to significant number dependence and interpretative difficulties.
Many discussions of the relative advantages of NEMD and equilibrium molecular dynamics revolve around questions of efficiency. For large fields, NEMD is orders of magnitude more efficient than equilibrium molecular dynamics. On the other hand one can always make NEMD arbitrarily inefficient by choosing a sufficiently small field. At fields which are small enough for the response to be linear, there is no simple answer to the question of whether NEMD is more efficient than equilibrium MD. The number dependence of errors for the two methods are very different - compared to equilibrium MD, the relative accuracy of NEMD can be made arbitrarily great by increasing the system size.
These discussions of efficiency ignore two major advantages of NEMD over equilibrium molecular dynamics. Firstly, by simulating a nonequilibrium system one can visualise and study the microscopic physical mechanisms that are important to the transport processes (this is true both for synthetic and realistic NEMD). One can readily study the distortions of the local molecular structure of nonequilibrium systems. For molecular systems under shear, flow one can watch the shear induced processes of molecular alignment, rotation and conformational change (Edberg, Morriss and Evans, 1987). Obtaining this sort of information from equilibrium time correlation functions is possible but it is so difficult that no one has yet attempted the task. It is likely that no one ever will. Secondly, NEMD opens the door to studying the nonlinear response of systems far from equilibrium.
cited from 2013 Statistical Mechanics of Nonequilbrium Liquids
I guess it depends on what you are after. If you wan't equilibrium related properties typical of real systems (temperature, pressure, volume etc) you'll have to make sure your MD models are in equilibrium. I only know a little about non-equilibrium thermodynamics, but the gist I understand from is that is a specialized method to probe relaxation effects. I get the idea from your quotes, that you say it is possible got get equilibrium properties faster this way, but I would be worried that the right approximations are being made in those models and that the results can be generalized. Either way, I would be skeptical of using either technique as a black box, but I would not think of using NEMD unless I really really knew what was doing. Maybe if I was a master of correlation functions I'd feel differently.