A Langevin thermostat is typically Langevin dynamics using a low value of the collision frequency, in order to avoid damping the atomic motions. It works reasonably well for ensemble sampling, but is not suitable for computing time-based properties, as the velocities are perturbed by the random forces. Calculated properties such as molecular diffusion will depend too much on the value of the collision frequency used.

In CHARMM, the setup is something like

! collision freq 1/ps
SCALAR FBETA SET 1.0
DYNA LEAP LANG ...


Switching right after heating should be fine.


Rick Venable
computational chemist