[gmx-users] Re: kinetically trapped in a local minimum???

Alan Chen achen at artsci.wustl.edu
Sat Mar 15 19:46:56 CET 2008

Hi Cong:

To expand on Mark's answer, there is no "universal test" for convergence 
of all properties, and therefore the best you can ever hope to do is 
check for convergence of the observable(s) you care
about, and any closely related quantities you can think of (especially, 
as previously mentioned, any closely related quantities that can be 
estimated using a simpler model where sampling is not
an issue).  But temperature, pressure, and internal energy should 
definitely have converged before you analyze data, with a large 
macromolecule almost certainly whatever you are interested in will
converge on a slower timescale than these variables alone.

As for your second question, that is an ever harder question to answer.  
If you can only run 1 simulation due to  system size or computational 
resources, you have to assume that the crystal structure is sufficiently 
near the global minimum (after the prerequisite minimization and 
equilibration).  If it's possible to run multiple simulations with 
slightly different initial preparations (say, different starting 
velocities or different structures from an NMR ensemble, for example) , 
you can at least compare how self-similar they are and if they are very 
heterogeneous then that is a red flag as well that you  are not yet 
converged.  If they are fairly self-similar, then you can do some 
statistical analysis like seeing how your answer changes if you only use 
a subset of your simulations to calculate your observable, or start 
chopping your simulations shorter and see how that would have affected 
your answer, and that at least gives you some idea of  a confidence 
interval which you can't get with just one simulation.  However, if your 
answer depends heavily on your maximum simulation length or very rare, 
non reversible events, then you are probably kinetically trapped and not 

Hope it helps!

> Message: 4
> Date: Sat, 15 Mar 2008 16:51:48 +0800
> From: "#NGUYEN CONG TRI#" <NGUY0045 at ntu.edu.sg>
> Subject: [gmx-users] kinetically trapped in a local minimum??? 
> To: <gmx-users at gromacs.org>
> Message-ID:
> 	<E476542CD98D094999A0FF7E022A465C013B1A51 at MAIL23.student.main.ntu.edu.sg>
> Content-Type: text/plain; charset="iso-8859-1"
> Dear Gromacs users,
> How can we know that the system is properly equilibrated or whether we should extend the equilibration step. Someone suggest that we look at the potential energy to see whether it has converged? Is it correct?
> One more question, normally, we carry out the equilibration at 300K, don't we? How can we know that this temperature is enough and the system is not kinetically trapped in a local minimum? And if it is whether it affects the production run later on? 
> Please give me some advice on these. Thank you very much.
> Best regards,
> Cong Tri

More information about the gromacs.org_gmx-users mailing list