[gmx-users] Replica Exchange MD ... Exchange rates

Berk Hess gmx3 at hotmail.com
Tue Jul 10 12:39:15 CEST 2007

>From: Mark Abraham <Mark.Abraham at anu.edu.au>
>Reply-To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>Subject: Re: [gmx-users] Replica Exchange MD ... Exchange rates
>Date: Tue, 10 Jul 2007 20:15:31 +1000
>Carsten Baldauf wrote:
>>dear gromacs-users//
>>what would you think are good exchange rates between the replicas in a 
>:-) I'm about to submit a paper partly on this topic... the short answer is 
>that most people publishing REMD exchange rates have one that is a good 
>deal lower. There's literature that recommends 20%, but a lot of people are 
>in the 10-30% range. If you get rates that seem too low, then the first 
>thought is to decrease the period between exchange attempts in order to 
>increase the exchange rate per unit simulation time, but there's published 
>data (e.g. from Periole & Mark, J. Chem. Phys this year) that indicates 
>that there are lower limits below which this is ineffective.

I would agrue this quantity by itself is irrelevant.
Also I think you can not call this a rate, it is a probability.
The rate (exchanges per unit of time) is already a much more relevant 
But even there the average does not say much.
What matters in the end is that the different replicas sample the whole
temperature space in a reasonable way. This can not be judged from
the probabilities.

I would say that in general 30% is on the high side, you could save yourself
computer time, or increase the sampling by choosing larger temperature 
But again, you have to make sure exchange happens regular enough between
all temperature pairs.


Play online games with your friends with Messenger 

More information about the gromacs.org_gmx-users mailing list