On Tue, 30 Apr 2002, David Case wrote:
> There is a large literature on this general subject; here I will just
> make two quick comments of my personal views:
>
> (1) many Amber users are interested in dynamical properties of their
> systems, for which NVE is the correct approach. All temperature coupling
> schemes modify Newton's equations to a greater or lesser extent.
I agree that it is definintely required to modify the Newton's equation
for samplings in a canonical ensemble. But, why not? On the other hand,
I am afraid that NVE is still improper in many scenarios, because the
"dynamical properties" of any system are probably still related to the
external conditions, like, temperature, pressure, ionic strength, etc.
Recently we have seen some people addressing the pressure dependence of
protein properties, which will be impossible to address in a NVE
simulation. Again, V is the thermodynamic conjugate variable of P, and
fixing V will cause P to fluctuate and become not well defined.
> (2) In an NVE simulation of a solvated protein (or DNA), the properties
> of the solute (protein) will be close to canonical: the solvent
> molecules will provide the "heat bath", and there will be fluctuations in
> the internal energy of the protein, as in the canonical ensemble.
> [Indeed, analysis of the behavior of a small portion of a larger,
> microcanonical, system forms the foundations of our understanding of the
> canonical ensemble itself.]
This is correct if solvent is explicitly included and its number of
states is much larger than that of protein (or other subsystems) of
interest. However, due to computational limitations we mostly only
include 10-20 A solvent buffer in each dimension and therefore above
conditions are not fullfilled. It is more "economic" to simulate a
smaller system using NPT than a much larger system with NVE.
Thanks,
Jung-Hsin
Received on Tue Apr 30 2002 - 13:36:45 PDT