[gmx-users] Re: gmx-users digest, Vol 1 #1007 - 8 msgs

[Michael Shirts]

SCTI is not a good method.  Straatsma and McCammon's paper cited below, as
well as later studies by David Pearlman, showed that "long enough to achieve
convergence" is usually longer than most simulations.  "Slow growth", as it
SCTI is more commonly called, is almost always worse than MCTI.  At least with
MCTI, your bias is controllable - adding more points into the numerical
integration is guaranteed to help.  Doing more SCTI runs doesn't help the bias
at all.

Cheers,
Michael Shirts
Pande Group
Stanford University

> one must first decide if they are going to do Multi
> Configuration thermodynamic integration(MCTI) or Single Configuration
> thermodynamic integration(SCTI).See( J. Chem Phys 95(2), 15 July 1991).

> In MCTI one would set delta_lambda = 0 and rerun the calculation at
> fixed intervals of dLambda.

> In SCTI one would usually set delta_lambda to 1/(Total Simulation Time)
> and then hope their total time is long enough to achieve convergence.

> There are advantages and disadvantages to both.



Cheers,
Michael