[gmx-users] Exchange interval in REMD

[Mark Abraham]

On 26/01/2012 11:38 AM, Gianluca Interlandi wrote:
> Hi,
>
> I have been going through the How-to for REMD in gromacs:
>
> http://www.gromacs.org/Documentation/How-tos/REMD
>
> Is there a recommended value for the exchange interval and does it 
> depend on system size? For example, is an interval of 100 ps 
> appropriate for a sytem with 60,000 atoms? Is a too long interval 
> simply a waste of time or can it actually introduce artefacts?

There's literature by Sindhikara (sp?) and Roitberg that says you should 
have a short an interval as possible, and other literature that says you 
shouldn't go shorter than around the autocorrelation time of PE, ~1ps. I 
don't buy the former, and think some of their data support the latter, 
and have some of my own not-ready-for-publication data that I think 
suggests artefacts from the former. Work from Hansmann et al suggests 
that achieving steady-state replica flow is rather tricky even for 
seemingly trivial peptides, so increasing the number of exchange 
attempts might help a little. I'm not aware of any good case for a long 
interval between attempts.

Mark