[gmx-users] table-extension distance appears to be over-ridden by .mdp file cut-off values

[Justin Lemkul]

Caveat to start: I am not an expert on tabulated potentials, but I'll take a 
crack at helping in case it inspires a solution :)

Very thorough report, so thanks for providing an exemplary level of detail.

On 12/18/13 3:39 PM, Elizabeth Ploetz wrote:
> Dear Gromacs Users,
>
> I'm trying to use tabulated VdW potentials that are non-zero until image-distance/2.0, but they seem to be over-ridden by the cut-offs in the .mdp file. I'm hoping someone with experience with the use of tabulated potentials plus the table-extension .mdp file option can tell me if I have not made all of the necessary changes so that the tables are correctly implemented. Here are the details of what I've done.
>
> I have five systems each consisting of a single Lennard-Jones (6-12) sphere in TIP3P water (rhombic dodecahedron, 15nm image-distance for each system). The Lennard-Jones parameters for the infinitely-dilute LJ-sphere are as follows:
> System Sigma (nm) Epsilon (kJ/mol)
> 1             0.3             0.1
> 2             1.5             0.1
> 3             3.0             0.1
> 4             4.5             0.1
> 5             6.0             0.1
> For each system, rlist=rvdw=rcoulomb=0.8 (nm). vdw-type=User, coulombtype=PME, table-extension=6.7 (nm). 6.7 nm was chosen because 6.7nm+0.8nm=7.5nm=half the image-distance. A tabulated potential is only used so that the LJ-sphere - water interactions can have a different (larger) cut-off than the water-water interactions since the sigma/2.0 values for the LJ-sphere are greater than the 0.8 nm cut-off for Systems 3-5.
>

I'm not sure I follow the logic here.  The value of table-extension does not 
give you a special cutoff, per se.

> When I run these systems (version 4.6), the LJ-sphere-to-water center-of-mass RDFs look correct for Systems 1 and 2 (the two LJ-spheres that have a radius less than the rlist=rvdw=rcoulomb cut-off). The RDFs look "wrong" for all larger spheres, and worse and worse as the Lennard-Jones sigma values increase. By "wrong," I mean that the RDFs take on positive values at too short of radial distances (i.e., water is being "sucked in" too close to the Lennard-Jones sphere). This makes sense if my tabulated VdW-potentials are not really being used (or, equivalently, not being used beyond 0.8 nm), and consequently the waters do not "see" the LJ-sphere until they are <0.8 nm away from the sphere. The simulations for the larger LJ-spheres are also unstable, which would also make sense, because waters that could get so close to the sphere would be well within the repulsive region of the sphere-to-water potential once they finally suddenly "saw" the sphere at <0.8 nm. The types of err!
 ors I get 
when the systems crash (usually after around 20 ps) are domain decomposition errors, specifically:
>
> Fatal error:
> A charge group moved too far between two domain decomposition steps
> This usually means that your system is not well equilibrated
> For more information and tips for troubleshooting, please check the GROMACS
> website at http://www.gromacs.org/Documentation/Errors
>
> Or (when I change to particle-decomposition or use only one processor), the simulations eventually crash due to LINCS.
>
> The details of my .mdp file and tabulated potentials are below my signature, as well as an example of the .log file output.
>
> Have I missed any extra flags that need to be defined so that the tabulated potentials are really used [the .log file output suggests they are really being used and so does mdrun.debug (not attached, but I can provide)]? Or, does anyone know of a way to check whether or not the tabulated potentials are really being used up to the cut-off + table-extension and not just up to the cut-off?
>
> Please note that, as a test, I did a series of energy minimizations on a simpler system, the Lennard-Jones sphere (sigma = 6 nm) and a single water molecule, when varying the coordinates of the water molecule systematically between 7.5 nm and 0.3 nm away from the LJ-sphere. The LJ(SR) sphere-water potential energy was zero for all distances except when the water molecule was < 0.8 nm from the LJ-sphere. This, again, indicates that the rlist=rcoul=rvdw=0.8 nm cut-off distance is being used for the sphere-water interactions instead of the extended distance.
>

I think we need to clarify a bit about what those settings are doing.  The value 
of table-extension does not give you some special value of cutoff for different 
interactions.  It is used to keep track of interactions beyond the cutoff 
between neighborlist updates.  I think your observations are consistent with a 
0.8-nm cutoff, beyond which the energy is zero.

It seems to me that, given the LJ parameters above, as soon as any water 
molecule moves within 0.8 nm of the LJ sphere in system 5, the potential will 
spike to +1 billion kJ/mol or thereabouts, and the system will crash.

-Justin

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

Justin A. Lemkul, Ph.D.
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

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