Re: [AMBER] About T-REMD convergence

From: Jason Swails <>
Date: Mon, 16 Sep 2013 12:19:36 -0400

On Mon, Sep 16, 2013 at 10:57 AM, Francesco Pietra <>wrote:

> >
> > I think you misunderstand what Carlos said. A good test of convergence
> > for REMD simulations is to analyze each replica (NOT each temperature),
> and
> > make sure that each replica contains the same information as every other
> > replica (especially the low-temperature sub-ensemble). In 'good' REMD
> > simulations, each replica will visit each temperature approximately an
> > equal amount of time. If you analyze all of the 314K snapshots from each
> > replica, the quantities that you compute (e.g., RMSD distributions, RMS
> > fluctuations, average structure, percent folded, etc.) for each replica
> > should be equivalent. Technically the same is true for all other
> > temperatures, but since high temperatures yield far more available states
> > than low temperatures, those are harder to 'converge'.
> >
> I am confused here. What I did was to sort all frames for 314K from each
> replica. The list of sorted for 314K begins as follows replicas 26, 28, 28,
> 6, 6, 6, 6, 6, 3, 3, 3, 28, 6, ..... where 28 is the highest replica.
> Perhaps unnecessarily too high. Then, I examined these frames.

This is what you should compute properties from when you are studying your
system. But if you want to look for convergence, you should look at each
replica individually and NOT sort them. For instance, replica 26 may have
the temperatures:

314K, 390K, 400K, ..., etc.

Each replica can be thought of as an 'expanded ensemble' of structures that
sample from multiple canonical ensembles at multiple temperatures. So the
full ensemble is a sum of 'sub-ensembles'. If the REMD simulation is
converged, then each replica---which samples from the expanded ensemble of
every temperature---should be equal to each other (so any property that you
compute should be the same, like the RMSD distributions, average structure,
atomic fluctuations, etc.). Furthermore, all of the sub-ensembles from
each of the replicas (i.e., those at each individual temperature) should
also be equivalent between all of the replicas.

This will make sure that there are no 'traps' in either conformation or
state space that is hindering convergence. After you have demonstrated
convergence, you can go back and sort the replicas into individual


Jason M. Swails
Rutgers University
Postdoctoral Researcher
AMBER mailing list
Received on Mon Sep 16 2013 - 09:30:03 PDT
Custom Search