Charge equilibration scheme for reax simulations

Dear all,

I am curious to know if the issue mentioned by Suleiman with the charge equilibration mentioned in the thread (http://lammps.sandia.gov/threads/msg29917.html) has been solved with a newer scheme or not. I am also encountering similar problem of unphysical charges on molecules while using qeq/reax and reax/c in Lammps and are not sure how to deal with it.

Thanks in advance,

Payel

Hi Payel,

We are working on a more generalized charge equilibration fix in
LAMMPS, and we will consider different schemes as options.

When you say unphysical charges with reax/c and qeq/reax, do you mean
instantaneous charge transfer between atoms/molecules separated beyond
cutoff or negative/positive charges on supposedly cation/anions? If
it is the later, then it is almost entirely due to parameterization,
instead of qeq/reax.

Ray

Hello Ray,

Thanks for the reply. My problem resembles the second case you mentioned, in which hydrocarbons become positively charged (overall system, i.e. water+hydrocarbon, still neutral).

Payel

graycol.gif

ecblank.gif

Payel,

Global charge neutrality is enforced so that there must be some negatively charged species to result in a net zero charge. I believe the positive charges on hydrocarbons are rather small? It is also not clear to me now that how QTPIE or any distance-dependent charge equilibration scheme would help this situation.

Ray

ecblank.gif

graycol.gif

Payel and all the other interested in this post,

Dear all,

I am curious to know if the issue mentioned by Suleiman with the charge
equilibration mentioned in the thread
(http://lammps.sandia.gov/threads/msg29917.html) has been solved with a
newer scheme or not. I am also encountering similar problem of unphysical
charges on molecules while using qeq/reax and reax/c in Lammps and are not
sure how to deal with it.

This is why its always a good idea to understand first the theoretical
basis of a model before loading it
into a computer. If Suleiman above had spent a bit of time
understanding the energy equalization principle, he could have shown
with the stroke of a pen why the QEq doesn't get the correct
dissociation limit of molecules, i.e. why one gets finite partial
charges on the fragments even when the molecular fragments are
separated an infinite distance.You could try doing this by just
solving analytically the Qeq model for a diatomic molecule of unequal
atoms. No need to run complex simulations using Lammps.
All potentials that rely on using QEq and variants, such as reaxff,
comb, etc, are intrinsically limited this way.

Now, getting the correct dissociation limit of molecules when
separated into its constituents its far from being a trivial problem.
Even ab-initio methods can get this wrong. See for example the
analysis done on the NaCl molecule using different functionals in the
following paper: J. Chem. Phys. 125, 234109 (2006);
http://dx.doi.org/10.1063/1.2409292
One of the issues with this is that the atomic partial charges in
molecular systems are not quantum observables and thus there is not
unique definition of the atomic charge within the atoms in molecules
(AIM) paradigm in the quantum model. This is why there are so many
different ways of computing atomic charges in first-principles
calculations. You've probably heard of Mulliken charges, Bader
charges, ESP charges, Hirshfeld charges, etc, etc.
Until recently, the best classical model capable of overcoming many of
the limitations of QEq & Co
was the Split charge equilibration (SQE) method. See: J. Chem. Phys.
125, 094108 (2006). Yet, there were a few loose ends such as its
derivation from more fundamental principles (there is a way to derive
QEq from quantum theory) that could shed light on the physical meaning
of the new parameters introduced in the SQE model.
The final answer seems to have just come out of the oven: J. Chem.
Phys. 138, 074108 (2013); http://dx.doi.org/10.1063/1.4791569 with a
first-principles derived charge equilibration model ACKS2 that yields
both SQE and QEq as two of its limits. If you are really interested in
this problem I strongly encourage you read the papers I have cited
here.
I have no idea which "more complex" models of charge equilibration the
Lammps team might be trying to implement but I believe SQE and now the
ACKS2 are the way to go.
Hope you can benefit from this info.
Carlos

Carlos,

You got it right. The long-range charge transfer is the problem here and the new ACKS2 model is the right way to go. So my question to Ray Shan and the LAMPPS developers:Is implementing ACKS2 model within LAMPPS in the current plan?

Thanks,

Payel

graycol.gif

ecblank.gif

Yes,
This task has been assigned to Payel Das from IBM :wink:
Just kidding. I’m sure Ray will have an answer for you.
Carlos

graycol.gif

ecblank.gif

Hi Payel and Carlos,

Yes, I second that Payel Das from IBM to be in charge of the ACKS2 option. :slight_smile:

The serious answer is that we will definitely take this into consideration once we have a general qeq implemented. Other models such QTPIE and ACKS2 will be considered as optional functions.

Stay tuned.

Cheers,
Ray

ecblank.gif

graycol.gif

Hello Carlos and others,

This is an old question but I need to send this reply to clear some issues. You (Carlos Campana) made an assumption that I did not spend time to understand the theoretical basis of EE methods. This could not be anymore false because my e-mail address the underlying theoretical shortcomings of the QEq methods and I stated those reasons and cited other theoretical methods that addressed the problem.

My example experiment with running a Cu/CuO system with REAX is just an academic exercise to prove that the actual simulation results is indeed plagued with errors. If other scientist are reading what you wrote without going back to what I mentioned in the conversation, it could be very damaging to my reputation.

Please, next time refrain from making assumptions that may affect others when you have little knowledge of the subject matter.

Suleiman Oloriegbe.

Hey Suleiman,
My words between your lines.