[gmx-users] REMD

[Eric Smoll]

Hello GROMACS users,

As far as I understand, increasing the number of random exchanges to a
large number (mdrun suggests N^3 where N is the number of replicas) moves a
REMD simulation from a neighbor exchange procedure to a Gibbs exchange
procedure.  Can anyone provide some practical advice or references useful
in deciding which to use?  Naively, I would guess that a Gibbs exchange
procedure would converge faster for a REMD equilibration with a large
number of replicas (~100). Is this usually true?

Best,
Eric