[gmx-users] convergence of MD simulations,

[Justin Lemkul]

On 8/19/14, 10:30 AM, Ali Alizadeh wrote:
> Dear All users,
> I have been engaged in answering this question"how long should we continue
> calculating to make sure we can get justifiable results?". Are there any
> suggestions to dig out to find out when a system converges? I know this is
> a too general question.
> In case of my systems(confined structure, interface between fluid and
> solid) after 1ns simulation my temperature fluctuation shows quite
> convergence, my total energy corroborates this convergence too.
>

Neither temperature nor total energy tells you anything about convergence.  They 
are related instead to the stability of the various algorithms employed during 
the simulation.  In fact, one can manipulate these quantities rather easily by 
tweaking settings inappropriately.

Convergence is more rigorously assessed through physical observables in your 
system.  For a simple fluid, do its properties (density, diffusion constant, 
RDF) change over different intervals of simulation time?  If they do, the 
simulation isn't converged.  For biomolecules, looking at any number of 
structural properties can be informative.  By your criteria above, my 
multi-domain heterodimeric protein complex was converged after about 5 ns.  In 
reality, it underwent large, cyclical transitions over the course of 500 ns.

-Justin

-- 
==================================================

Justin A. Lemkul, Ph.D.
Ruth L. Kirschstein NRSA Postdoctoral Fellow

Department of Pharmaceutical Sciences
School of Pharmacy
Health Sciences Facility II, Room 601
University of Maryland, Baltimore
20 Penn St.
Baltimore, MD 21201

jalemkul at outerbanks.umaryland.edu | (410) 706-7441
http://mackerell.umaryland.edu/~jalemkul

==================================================