Using pair eam/alloy in adapt or adapt/fep

LAMMPS LAMMPS General Discussion
1

LAMMPS version: 11Feb2026

Hi,

I am trying to compute the interfacial free energy in a biphase liquid-solid system using free energy parturbations.

I am modelling both liquid and solid with EAM (eam/alloy ff).

Right now I am just monitoring energies (i.e. I am not performing TI or BAR). If I define fix adapt in this way:

fix adaptForward all adapt ${t_s} pair eam/alloy scale 1 2 v_scaleForward scale no

I get the following error:

ERROR: Fix adapt pair style eam/alloy param scale not supported (src/fix_adapt.cpp:442)

If on the other hand I use fix adapt/fep:

fix adaptForward all adapt/fep ${t_s} pair eam/alloy scale 1 2 v_scaleForward after no scale no

I get another error:

ERROR: Fix adapt/fep pair style param not supported (src/FEP/fix_adapt_fep.cpp:298)

According to here: fix adapt command — LAMMPS documentation, fix adapt/fep command — LAMMPS documentation, I should be able to use eam/alloy with scale as parameter, right? What am I missing?

P.S. I also tried with the stable release 22Jul2025 and I got the same errors…

2

Turns out it’s a GPU problem. I am not sure if this is a bug of some pair stiles can’t be used with adapt(/fep) when running on GPU (I was trying with -pk gpu), but don’t get any error when running on CPU.

It works with Kokkos (on GPU), but I cannot use it since I also need to compute energies within/between groups, and when I run with -k on g 1 -sf kk I get:

ERROR: Pair style does not support compute group/group (src/compute_group_group.cpp:116)
3

You have to understand that the force computations for accelerated pair styles are separate implementations that share only some code with the non-accelerated base class. So it is quite possible that the support for the “scale” factor has not been ported to an accelerated version, especially when the accelerated version is older that the fix adapt feature support.

Compute group/group is a different story because it uses the single() function instead of the general compute() and that is always run on the CPU as far as I recall.

4

Thanks for the explaination Axel. I don’t mind running on CPU for the time being, so I guess I will be fine.

1 Like