On Tue, Oct 19, 2010 at 5:08 PM, Daniel Sindhikara <sindhikara.gmail.com> wrote:
> Anytime you are calculating a free energy, you should look to do canonical
> sampling
> (as well as making sure your results are converged). For very large systems,
>
> (since NVT -> NVE) you may be able to get away with Berendsen.
I assume that all of your simulations are at the same temperature.
If so, and if your system is large enough, then I agree.
The problem occurs if you vary the temperature (if your simulations
are run at different temperatures). In that case the Berendsen
thermostat will cause subtle problems. For example, the Berendsen
thermostat will slow down your temperature-replica-exchange
simulations (by requiring more replicas), and introduce artifacts into
your results. Furthermore, the Berendsen thermostat invalidates the
use of WHAM (weighted-histogram-analysis) analysis to combine data
from multiple temperatures. (This is true whether or not you use
replica-exchange.)
However if all of your data is at the same temperature, then perhaps
you don't need to worry so much (especially if you are using explicit
solvent).
(If so, and if you ran these simulations already, then I probably be
inclined to keep the existing data.)
But if you are beginning a new simulation, use the Langevin thermostat
(ntt=3), or if that fails, the Andersen thermostat (ntt=2).
> I should also note that Hummer et al showed that if you do any sort of
> replica exchange
> with a non-canonical thermostat, your results will get corrupted.
I think the paper is:
Edina Rosta, Nicolae-Viorel Buchete and Gerhard Hummer
"Thermostat Artifacts in Replica Exchange Molecular Dynamics Simulations"
J. Chem. Theory Comput., 2009, 5 (5), pp 1393–1399
Cheers!
Andrew
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Oct 19 2010 - 20:30:04 PDT
