[gmx-users] hints for core/shell optimization?

[Justin Lemkul]

On 7/25/14, 7:43 AM, Tamas Karpati wrote:
> Dear all,
>
> I have two questions about geometry optimization of a crystal
> with polarization via the core/shell model. I'm creating *.gro and
> *.top files "by hand" and compile them with *.mdp to *.tpr via
> GROMPP. My FF is also made by hand (simply because i need
> to learn GROMACS). I have learnt on this list that with Buckingham
> potentials I need to use the group rather than the Verlet scheme.
>
>
> (1/2) Letting some of all atoms be polarizable through applying
> shell particles made MDRUN segfault like this:
>
> #  ...
> #  Reading file AAA_opt.tpr, VERSION 4.6.3 (single precision)
> #  Using 2 MPI threads
> #
> #  Steepest Descents:
> #     Tolerance (Fmax)   =  1.00000e+01
> #    Number of steps    =           10
> #  Segmentation fault1.0e-02 nm, Epot=         -nan Fmax= 3.76506e+03,
> atom= 1357
>
> I imagined some divison by zero situation not handled and have put
> some "random noise" on the shell particles' position so they do not
> anymore start exactly at the atomic sites (meaning nonzero distances).
> Seemed to work, at least no further crashes. Only energies and forces
> seem very high:
>
> #  Steepest Descents:
> #     Tolerance (Fmax)   =  1.00000e+01
> #     Number of steps    =           10
> #  Step=    0, Dmax= 1.0e-02 nm, Epot= -1.36425e+07 Fmax= 2.99600e+05, atom= 160
> #  Step=    1, Dmax= 1.0e-02 nm, Epot= -1.62080e+07 Fmax= 1.25769e+06, atom= 160
> #  Step=    2, Dmax= 1.2e-02 nm, Epot= -1.95965e+07 Fmax= 6.87820e+08,
> atom= 2759
> #  Step=    3, Dmax= 1.4e-02 nm, Epot= -2.02902e+07 Fmax= 1.30719e+09, atom= 468
> #  Step=    8, Dmax= 1.1e-03 nm, Epot= -2.18970e+07 Fmax= 5.77722e+09,
> atom= 1095
> #  Step=   10, Dmax= 6.5e-04 nm, Epot= -1.96952e+07 Fmax= 3.92889e+08,
> atom= 1096
> #  Energy minimization reached the maximum numberof steps before the forces
> #  reached the requestedprecision Fmax < 10.
>
> My question is the following.
>     Is it (randomized shell positions) a correct procedure with GROMACS?
>
>
> (2/2) Changing from a randomized x/y/z set to a fixed distance at
> a random direction for the shell particles led to another unexpected result.
> I scanned a range between 1e-4 to 0.1 nm and noticed that
>
>      the final core-to-shell distance is a function of the starting one.
>
> I used niter = 10000 (note: the default is 20) as i noticed in an MD
> type of job that 20, 100 or 1000 steps were insufficient for the shells
> to relax within default tolerance. The cell size was ca. 3x3x3 nm.
>

Please provide a full .mdp file; other settings are very relevant.

> My question is the following.
>     What would be the appropriate core-to-shell distance to apply?
>

The equilibrium distance for core-shell bonds should be zero, deviations from 
this non-polarized state account for the polarization energy.

> I appreciate any help so thanks in advance.

Upgrade to 4.6.6; there have been issues with shells that have been fixed since 
4.6.3.

-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

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