I am currently working on simulations of the mechanical behavior of an aluminosilicate system. From my understanding, when using classical potential functions, long-range Coulomb interactions (via the kspace_style command) are essential for accurately describing ionic systems.
However, I am relatively new to the ReaxFF force field, and I’m uncertain whether the kspace_style command should also be specified when using ReaxFF to account for long-range electrostatic interactions.
I would greatly appreciate any insights, suggestions, or experiences you could share regarding this question.
You can use the scientific approach and read the publication(s) describing ReaxFF.
You can test this empirically and see if you can use a kspace style, and if you can make some simple tests (e.g. comparing to a traditional classical force field) to see, if the results are sufficiently accurate.
you cant as of now, the electrostatics would be all wrong around the switching cutoff.
to do this, you would need to implement and contribute a new pair_style reaxff/coul/long/kk similar to other */coul/long/kk pair styles, then retrain all your reaxff force field electrostatic parameters (chi, eta, gamma) to DFT data from scratch using your new reaxff/coul/long/kk pair_style.
long range electrostatics has long been a known weakness of REAXFF.
Now, I am simulating one of aluminosilicate materials, a typcial trtriclinic crystal, by ReaxFF. I found the ReaxFF can not even predict the crystal parameter. The crytal structure will distort in the equilibrium conditions (300 K, 1 atm). All parameters of ReaxFF are obtained from the open published papers. But the its preformance is even can not comparable with the simple classic force field.
Therefore, I suspective the long-range of electrostatics is vital in the aluminosilicate system. However, if Reaxff is not consider this interation, its application will greatly limit.
There is nothing to suggest. The purpose of ReaxFF is to allow simulation of chemical reactions, not to be a general purpose force field. In fact, its potential files are often highly non-transferable.
Your observations are consistent with that. And lack of long-range electrostatics is only one component of that. We see a lot of people here using ReaxFF because it looks from the outside much easier to use than, say, OPLS-AA or CHARMM or Amber or GROMOS. You don’t have to go through (complex) atom typing and (partial) charge determination procedures. But it is exactly this complexity based on the observation that functional groups have a very similar behavior that makes those traditional molecular force fields more transferable and accurate for purposes of general structure and aggregation. While finding equivalently predictive parameters for ReaxFF is much, much harder (if not impossible) because you not only need to represent one specific functional group, but multiple and the chemical reactivity to go from one to the other.