[gmx-users] Brownian and Langevin Dynamics

[Berk Hess]

I would say that one should never use Brownian dynamics for
protein simulations. Brownian dynamics results for removing
degrees of freedom, like the solvent for an ion solution.
With a protein one can not leave out the solvent, unless one
has a very accurate implicit solvent description. Even if one
has such a description BD will result in very slow sampling
as collective motions are very overdamped (unless one uses
some kind of osene tensor).

In BD only the ratio dt/bd_fric is important. Of cource scaling
both of them scales the time. But this one can always do after
performing a simulation. So after the simulation one can
determine a bd_fric which reproduces the correct time scales.

In SD inertia is still present, so both dt and the friction
constants (determined by tau_t) matter. Also SD will not
result in better sampling than MD. The main use of SD is
for temperature coulping, then one should use very small friction
constants. SD gives a known ensemble, unlike Berendsen coupling,
which can also give artifacts. And SD does not give the oscillations
that a Nose-Hoover thermostat produces.

Berk.


_________________________________________________________________
Join the world’s largest e-mail service with MSN Hotmail. 
http://www.hotmail.com