Re: [AMBER] Bias Towards the Periodic Boundary

From: Adrian Roitberg <>
Date: Wed, 18 Oct 2017 14:22:01 -0400

I believe you are looking at anecdotal evidence only. You need to do a
much more thorough set of calculations and statistics to say that there
is a bias.

I suggest one possible calculations you can do:

Image your final result in such a way that your 'column' is in the
middle of the box and not on the edges.

Run it again and see if the column moves towards the edges or stay in
the middle.


On 10/18/17 1:37 PM, Timothy Schutt wrote:
> Hi All,
> I've long suspected that there was some small artificial spatial bias
> towards the periodic boundaries because my macromolecules would always
> migrate towards the vertices and then stay there but I never really
> investigated it because I thought it would Not impact my dynamics results
> at all. However, I recently tried to simulate a droplet and the artifact is
> even more readily apparent.
> 1. Have other people observed this behavior; is it a known artifact?
> I wasn't able to find any documentation on it in this forum or general web
> search.
> I'm using amber16 pmemd.MPI. My system is mostly aqueous (TIP4PEW) and the
> box dimensions are much larger than they need to be. I expected most of the
> molecules to coalesce into a somewhat spherical shape on their own during
> NVT equilibration due to minimizing surface area to volume ratio as the
> force field should encourage naturally... and then much of the simulation
> box should be empty or have a few 'volatile' compounds floating around.
> Instead, however, the liquid lined up along the 4 parallel vertices, in
> essence forming an infinitely long column centered on the corners between
> periodic images. Therefore I believe the amount of bias at the boundary has
> a net effect on the same order of magnitude as water's surface tension!
> 2. Is this an artifact of the PBC iwrap algorithm somehow... Or perhaps
> does the langevin thermostat interact preferentially with molecules at the
> edge or something?
> My input parameters were:
> Test nvt on droplet
> # Control Section
> &cntrl
> iwrap=1,
> imin=0,
> nstlim=10000000,
> dt=0.001,
> ntwx=1000,
> ntpr=10000,
> ntwr=50000,
> ntb=1,
> ntp=0,
> ntc=1,
> ntf=1,
> ntr=0,
> ibelly=0,
> cut=10,
> ntt=3,
> gamma_ln=2.0,
> temp0=300,
> nmropt=0,
> /
> &end
> My system is 9000 water molecules with ~60 solute molecules in a box that
> is 100A x 100A x 100A (roughly 80% void space). I'm looking at the
> aggregation and association of small solutes (both charged and not) in
> water droplets and the preference to reside at the surface of the droplet.
> I think this PBC bias would impact my results this time.
> 3. Is there a known fix?
> If not, I'll try running without iwrap, or in sander, and/or loosening to
> NVE constraints to start diagnosing where the bias is coming from but I'm
> hoping it's already known.
> 4. If this isn't already known, any suggestions in where/how to look for
> the cause in addition to the above?
> Thanks for the help!!
> -Timothy Schutt
> Vicksburg, MS
> _______________________________________________
> AMBER mailing list

Dr. Adrian E. Roitberg
University of Florida Research Foundation Professor
Department of Chemistry
University of Florida
AMBER mailing list
Received on Wed Oct 18 2017 - 11:30:02 PDT
Custom Search