[gmx-users] Shell (Drude) model for polarization in GROMACS

Justin Lemkul jalemkul at vt.edu
Fri Jul 13 22:51:19 CEST 2018

On 7/13/18 4:48 PM, Eric Smoll wrote:
> Hello Justin,
> Thank you for the guidance.
> Although more testing is needed, molecules where shells are attached to
> every atom appear to be working properly.
> The issue appears to be with models where shells are only attached to heavy
> atoms and bonds between hydrogen atoms and heavy atoms are constrained.
> Initial energy minimization tests (emtol = 100) show that bonds between
> heavy atoms and hydrogen in organic molecules slowly stretch or compress
> far more than they should while the network of non-hydrogen atoms maintains
> a sensible geometry.  All atom-drude displacements on the heavy atoms
> converge quickly and are stable.
> I am still hunting for whatever topology problem is causing this.  All
> exclusions out to 1-4 interactions between atoms and drudes should be
> properly included (combined action of the moleculetype directive and
> additional exclusions directives).  Thole screening only applies to
> atom-shell pairs so no screening is needed for 1-2 and 1-3 pairs with H
> atoms (no attached shells).  It may be difficult to provide any useful
> guidance without details but troubleshooting suggestions are welcome if you
> have any.
> Are there issues associated with adding shells to specific atoms in a
> molecule?  Are simulations that place shells on all atoms (hydrogen and
> heavy) more stable for some reason?  I am going to build a new model where
> shells are attached to all atoms to see if bonds to hydrogen atoms still
> slowly compress/stretch during a tight-emtol energy minimization.

I have never tested a system like that. Our Drude convention does 
exactly what you seem to find a problem - Drudes on heavy atoms and not 
H, with bonds to H constrained. I've never had the issue you're 
experiencing. Without a full test case of complete inputs, I can't tell 
you anything.

Be forewarned - I know of no efforts by me or anyone else to deal with 
polarizable H atoms in GROMACS, and that may conflict with constraints, 
so tread lightly...



Justin A. Lemkul, Ph.D.
Assistant Professor
Virginia Tech Department of Biochemistry

303 Engel Hall
340 West Campus Dr.
Blacksburg, VA 24061

jalemkul at vt.edu | (540) 231-3129


More information about the gromacs.org_gmx-users mailing list