The system is active Brownian particles without charges under an electric field (just to apply torque).
Using fix brownian/sphere require that I have atom_style hyrbrid sphere dipole.
The attached script works fine with lammps compiled with the required packages. But not with kokkos enabled.
Would it be possible to use atom style hybrid dipole sphere in kokkos?
If dipole style is not available, is there a work around?
Please let me know if you need any extra information?
It is pointless to try use KOKKOS for your input since your pair style is not supported by KOKKOS.
Even if the pair style would be supported, computation would be slowed down by the fact that your fixes are not supported by KOKKOS either.
You can see which styles have accelerator support in the corresponding documentation, and in the command overview tables. Thus please consult the documentation (or look it up in the source code).
What kind of acceleration do you expect from KOKKOS? You have a 2d-system, so the acceleration potential is limited because you have far fewer neighbors than in a 3d-system. In addition you are using a model where pair-wise interactions are short (repulsive-only), which further reduces the potential for acceleration by further reducing the number of neighbors.
Come to think of it. Even without acceleration, your choice of pair style makes no sense for the use case. Why not use lj/cut?
There are other issues, like the use of dipoles without them having an impact of the forces/torques on atoms in the pair style. So the dipoles will react to the external field, but not to each other.
So the entire physics of your model is quite questionable. Whether you can use KOKKOS is a minor concern in comparison.
I am going for a larger system, more than 600K particles so i am good with any speed up, and in the future i am going for a 3D system.
The pair style is not a must, and lj/cut should work too.
It true, we are looking for effect of the field on the phase diagram, we dont want to include particle-particle interactions for now.
Still I cant find if atom style dipole is supported or not (it seems that its not supported).
As already mentioned, this is a very questionable choice. If atoms have a dipole, those dipoles will interact with each other. The phase diagram is crucially dependent on particle-particle interactions.
Basic rule in LAMMPS: If you something is not explicitly mentioned as supported, it is likely not. Otherwise, it would only be supported by accident, not by design.
Given the complexity of LAMMPS and the many possible permutations of features, it is next to impossible to list all combinations that are not supported. So only what is supported, is documented.