Use compute pe/atom with hybrid pair_style

Dear LAMMPS users,

I’d like to get information on the forces and potential energy for each atom in my system, which consists of 3 species of atoms interacting with different pair styles, that I simulate thanks to the hybrid or hybrid/overlay styles.
To dump the per-atom potential energy I use the compute:

compute peratom_energy all pe/atom

but I think I get wrong per-atom energies, because if I sum up all the contributions obtained in such a way, via the compute:

compute pe all reduce sum c_peratom_energy

I obtain a value for c_pe different from that of pe. I made some test, and I understood that the problem stays in the combined use of the pe/atom compute and the hybrid pair_style. For example, if I set up my potentials as:

pair_style lj/cut 1.122462
pair_modify shift yes
pair_coeff 1 * 1.0 1.0
pair_coeff 2 * 0.0 0.55
pair_coeff 3 * 1.0 0.1

I obtain the same value for both pe and c_pe (around 3.23 for my first configuration), while if I set up the potential using the hybrid style:

pair_style hybrid lj/cut 1.122462 lj/cut 1.122462
pair_modify shift yes
pair_coeff 1 * lj/cut 1 1.0 1.0
pair_coeff 2 * lj/cut 2 0.0 0.55
pair_coeff 3 * lj/cut 1 1.0 0.1

I get the same result for pe but a different one for c_pe (around 1.90 for the same configuration).

Am I misinterpreting something? Where am I wrong?

Thank you all,

sorry, but i cannot reproduce this with the 20 Nov 2019 version of LAMMPS. i get the same result.