Re: [AMBER] Really Weird Problem with Replica Exchange

From: Jason Swails <jason.swails.gmail.com>
Date: Wed, 09 Apr 2014 13:50:32 -0400

On Wed, 2014-04-09 at 13:36 -0400, Christopher A Thiebaut wrote:
> The acceptance rate seems artificially high. It was not supposed to be
> that high. I'm pretty sure something weird is going on, and I need to
> run down what is going on, or I won't be able to trust the results
> (otherwise, I would just run fewer replicas and save myself the
> simulation time). Exchange is attempted every 500 MD steps.

I'll go ahead and disagree with Brian :). Acceptance rate and exchange
attempt frequency are independent of each other, and there is nothing
statistically mechanically wrong with attempting exchanges super often
with a high rate of acceptance (it's more or less Gibbs' sampling as
described here:
http://scitation.aip.org/content/aip/journal/jcp/135/19/10.1063/1.3660669).

Here are some articles about EAF:
http://scitation.aip.org/content/aip/journal/jcp/128/2/10.1063/1.2816560
http://pubs.acs.org/doi/abs/10.1021/ct100281c

You should make sure that the temperature ladder you use comes directly
from the number of degrees of freedom in your system.

All the best,
Jason
>
> Chris Thiebaut
> Graduate Student
> Purdue University
>
> ----- Original Message -----
> From: "Brian Radak" <radak004.umn.edu>
> To: "AMBER Mailing List" <amber.ambermd.org>
> Sent: Wednesday, April 9, 2014 1:29:48 PM
> Subject: Re: [AMBER] Really Weird Problem with Replica Exchange
>
> I don't really do T-REMD too much, but isn't the velocity scaling a
> temperature dependent factor that goes as sqrt(T1/T2) or something? For
> such closely spaced temperatures, that could very easily be 1.00. I could
> be totally wrong on that though.
>
> Another relevant question may be what is your exchange attempt frequency?
> Some may disagree with me, but I would think that having a high acceptance
> rate *and* a high attempt frequency would be a bad thing and potentially
> lead to weird results.
>
> Regards,
> Brian

-- 
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Apr 09 2014 - 11:00:06 PDT
Custom Search