I did that test and the numbers do indeed match. Do you know what other codes people have used that have given the same ∆Gsolvs as LAMMPS?
I am wondering if AMBER’s way of calculating dU/dlambda is the source of the discrepancy. As best I can tell, AMBER uses something more complicated than just evaluating the energies at two different lambdas and taking the difference. The manual is unclear, but the output suggests some kind of cubic spline interpolation… I am having AMBER print out dU/dlambda for a “run 0” starting configuration with the same coordinates, and the results are very different.
If you know anything else about the codes or protocols other users have found successful with different softwares, that would be helpful in tracking down the source of the discrepancies I am seeing.