On Tue, Mar 11, 2014, Ross Walker wrote:
> Note AMBER 14 will also support hydrogen mass
> repartitioning so in principal you can run at 4fs time step.
How much to we really know about this option? I've been playing with it for
small proteins (like GB3, 56 residues). This is large enough that I don't
have converged sampling, but I certainly see differences between "regular"
masses at 2 fs step and otherwise equivalent runs with Hmass and 4 fs, for
simulations of a small number of microseconds. The symptom is that the 4 fs
runs are more likely to jump between basins than are 2 fs runs (when run for
twice the number of steps). This seems to make sense -- the long time steps
will have more of the occasional local heating caused by inaccurate
integation, and some of these events will lead to barrier crossings. The
question is whether we know that one is still sampling a Boltzmann
distribution for landscapes like those one sees in proteins.
So, what sorts of systems have been tested? Has anyone taken advantage of
Blue Waters (say) to run lots of simulations at 1, 2 and 4 fs time steps?
Has anyone compared 2 and 4 fs where both have the mass repartitioning?
I think we have to warn people that this is a less-well-tested approach, and
it would be great to have a paragraph in the Users' Manual outlining what is
known about how it performs.
Apologies in advance: my notes on Adrian's talk at Stony Brook are not very
specific about what systems were looked at.
....dac
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Wed Mar 12 2014 - 05:00:03 PDT
