Changing an atom type at lambda = 1 affects the computed energies at lambda = 0

Summary

Changing a single atom type at the lambda = 1 endpoint affects the short-range electrostatics at lambda = 0 using GROMACS 2022.4 (it shouldn't).

Exact steps to reproduce

Using the two folders in these input files, one obtains different potential energies for the short-range electrostatics (using gmx energy -f gromacs.edr):

input0: 13.0602 kJ/mol

input1: 13.0313 kJ/mol

The only difference between the two inputs is a single line in the topology file. For input0 this is:

7 C2 1 LIG C7x 7 -0.054060 12.010780 C2 -0.054060 12.010780

And for input1 this is:

7 C2 1 LIG C7x 7 -0.054060 12.010780 C2_du -0.054060 12.010780

If this is a bug, (1) what happens, and (2) what did you expect to happen?

Since both MDP files have this section:

fep-lambdas          = 0.00000 0.00000
coulomb-lambdas          = 0.00000 1.00000
init-lambda-state = 0

The energies should not be affected because of this change, because 1. changing the atom type should not affect the electrostatics and 2. changing something at lambda = 1 should not affect the energies at lambda = 0.

Relevant input files, logs and/or screenshots

All attached. This was run on macOS.

Possible fixes

This seems to be related to an issue that was fixed in 2022.4. I would be interested in learning how to circumvent it if possible.