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

Tamas Karpati tkarpati at gmail.com
Wed Jul 30 16:23:30 CEST 2014

Dear Justin,

Thanks for the exremely quick answer!

> No, exclusions work on nonbonded interactions.  Thole screening is, in
> theory, a nonbonded interaction, but it is handled as a bonded interaction
> in the code, which allows it to act on otherwise excluded atom pairs.

Got it. Finally.

> I think your units are wrong here.  The alpha values should be in nm^3.
> Those values of alpha are about 1000x larger than I think they should be.

I re-check them. They've been suspicious (compared to data in sw.itp).

> Keep S.  A "dummy" (D) particle is a virtual site; very different!

Got it. Thanks.

> Anything that is treated by Thole should be listed as an exclusion.

I can see now that this will inevitably destroy nonbonded interactions.
Using [polarization] along with [thole_polarization] can only be done
with bonded structures. In my case Buckingham+Coulomb represented
the means of cohesion (probably I should replace their sum by some
Morse or other type of explicite bonding).

> I don't see how that is inconsistent with the manual.

I'm working on it. I guess inconsistency is in my mind. Yet. Starting to
recognize that Thole is nonbonded but in some manner treated as bonded.

> Have you verified this by using gmxdump?  Exclusions are generated from
> bonds. You said you don't have any bonds, so there shouldn't be any
> exclusions aside from those you are adding.

Yes, I've checked both processed.top and gmxdump-ed the TPR to see
that no [exclusions] in processed.top and just self-exclusions in the TPR
(between each atom_i - atom_i "pairs"). That's correct.

> That's what gmxdump is for.

Thanks. Verified.

> Thole screening is applied to atom-shell(Drude) pairs to account for
> unphysical dipole-dipole interactions.  The manual is in need of updating.

Very educating. Appreciated. As long as this list is operating, no matter... :)

> I still don't understand what's "in the dark" here.

Not anymore. processed.top and gmxdump-on-TPR made it transparent.

> That's why I have said a few times that I am hesitant to be making all of
> these recommendations.  I don't know what paper you're referring to, what
> model, or what software.  Ad hoc changes are probably a bad idea, but in the
> polarizable models I have used, nearby interactions are extremely sensitive
> to dipole-dipole interactions.

I'm quite convinced that I do need the Buckinghams. Thinking seriously
about the Morse/Cubic/etc. fitting on Buckingham+Coulomb...

But first go experimenting [exclusions]+[thole_polarization] to be sure.

I'm grateful, indeed, for your efforts on this list. Thank you very much,

More information about the gromacs.org_gmx-users mailing list