[gmx-users] charged proteins

[Berk Hess]

>Didn't you suggest using a 1.5 nm cut-off with RF? When used as a >single 
>cut-off that will be much slower than PME with a cut-off of 1.0.
>When used as twin-range, you introduce other artefacts, including >worse 
>energy conservation. As we've shown in the GROMACS 3.0 article an 
>acceptable cut-off scheme (0.9/1.4) is not much slower than PME, but >still 
>considerably less accurate.

>It's high time someone tests these things systemetically, and >publishes 
>it.

I would suggest an 0.9/1.5 cut-off with reaction-field. I agree that
the energy conservation is not very good. But most other properties
are quite reasonable. When running on one processor PME is probably
about as efficient and one should use PME. But on more processors
the current implementation of PME scales quite bad.

I don't know about sany ystematic tests, but Peter Tieleman (with me
as a co-author) has just submitted an article with simulations of
several helix bundles in membanes. Plain cut-off gives incorrect
results, but the PME and reaction-field results give the same
results for several important properties.

Berk.


_________________________________________________________________
Chat with friends online, try MSN Messenger: http://messenger.msn.com