Hybrid Reax Possible in new version?


I have used reax/c before using Mar 2020 version of the lammps. Ideally, I would like to have all the different elements parameterized in one single file to run reax. But the current simulation has needs 2-3 reax files where only one element interaction is missing in each. Let’s say one file has 4 element interactions A-B-C-D and other has one A-B-C-E, while the third has A-B-D-E. I couldn’t find any other potential which has all A to E interactions defined.

I have previously searched through queries from past years (2011-2017) that said it was not possible to use reax with hybrid pair style. I was checking the lammps manual recently and saw that the reax/c pair_style was changed to reaxff. So, I want to check if there is any change to using reax in hybrid style or if you have any suggestions for defining interactions of my 5-element system?

Lastly, I tried to use a 2022 reaxff potential file in previous lammps version (2020) and it is giving me errors reading the file format. I might need to install new version of lammps and try. Would the new reaxff pair_style support all the previous reax files? Sorry if this question is redundant. Thank you for your time.

Using multiple ReaxFF instances in a hybrid pair style is not a solution to your problem. It will in any case result in a bogus model because ReaxFF is not pairwise additive, regardless of whether the multiple technical challenges to solve the “mechanics” of setting this up are met. While ReaxFF is implemented as a pair style, it contains interactions that extend beyond pairs and involve triples of atoms and quadruples. Also you have to determine bond orders and and do charge equilibration. This will be incomplete and inconsistent with two ReaxFF instances it would need to be done across the entire system considering all atoms. With pair hybrid, you treat the two sub-styles as independent system and are required to have a pairwise additive pair style to handle the interactions between the two subsystems.

Thus, what you absolutely need is a potential file with proper parameter sets for all your elements. There is no escape from that. If you cannot find a suitable parameter file, you need to either learn how create one by yourself (using the existing settings as a starting point) or collaborate with somebody that has that skill and experience.

Thank you for the quick response. I had similar difficulties last year too where I was trying to combine two reax potentials using hybrid/overlay. It read both the files fine in the log file but showed error when I tried to do charge equilibration the second time using second file. I understand that it is better to have a single file with all the elements in it for error free simulation. Would you say I can use one reax file to define 4 elements while use LJ parameters to define interactions with the 5th element in hybrid or hybrid/overlay? Generally, can reax be used in hybrid or hybrid/overlay pair style at all with itself or any other pair style?

I will try to combine other non-charge equilibrating potentials (LJ, EAM or Tersoff) using hybrid/overlay. If it doesn’t work out, as you pointed out, there is no other way around but to create a new potential file. Thanks.

You are not paying attention. Pair style hybrid or hybrid/overlay is not the kind of “band aid solution” that you are looking for. There is a big difference between what is “mechanically” possible and what is physically sound. There are many ways of running simulations with a software like LAMMPS that will still be completely bogus and meaningless. And not just by a small margin.

Mechanically, you can construct such an input, but - as I already mentioned - its results are going to be seriously tainted.

This will be as bogus as using ReaxFF, since EAM and Tersoff are manybody potentials. While there is no charge equilibration, there are other problems: for EAM for example you have the embedding energy term for which there are no contributions toward an EAM sub-systyle from any other sub-style (including EAM sub-styles).

The situations where you can use a manybody potential in a hybrid pair style and get meaningful results are limited, e.g. when you have a two component system where each pair style represents a different and separate component and there is a very limited area where the two interact, and where the interaction will be dominated by the geometry and the accuracy of the potential combining the two. Pair style hybrid/overlay is a different animal since it adds to the interactions. So you would have double counting of interactions, if not used carefully. The main applications of hybrid/overlay would be to add coulomb to a non-coulomb potential or to add explicit hydrogen bond interactions in Dreiding force field simulations, or to add corrections for inter-layer interactions and similar.

Exactly, there is NO alternative in your case.