[gmx-users] Re: dG/dl

[Michael Shirts]

Hi, all-

> I have tried the Bennett acceptance ratio method and did not
> find significant differences with plain integration of dG/dlambda.
> Although my experience is that people often take far too little
> lambda points (the new soft-core lambda power of 1 in 3.3 helps a bit
> tough). Michael, what is your opinion on this?

I did a comparison of the various methods (except for slow growth, and
there are links there that talk about slow growth) that was published
in JCP earlier this year.

M. R. Shirts and V. S. Pande. Comparison of efficiency and bias of
free energies computer by exponential averaging, the Bennett
acceptance ratio, and thermodynamic integration. Journal of Chemical
Physics 122, 144107 (2005), PDF at:
http://www.columbia.edu/~mrs2144/JChemPhys_122_144107.pdf

Basically, if you are making vdW sites appear and disappear, Bennett
will pretty much always be better.  I've made some modifications in
GROMACS to make Bennett very trivial to run, but alas, haven't merged
them yet -- a common problem with me.  I hope to get them merged as
soon as I get caught up here in my posdoc, but that might not happen
until Dec/early Jan.

Best,
Michael