pw.x Problem restarting scf calculation from a previous relaxation step with v 6.6

Hello everyone, I noticed that, in version 6.6, when trying to restart a calculation from a previous relaxation run setting calculation = scf and restart_mode = restart, the final total energy is very different from the total energy of the optimized structure of the previous run. Running few cases, I noticed that discrepancies could go from 0.1 to several Ry. Instead, the restart seems to work as expected if calculation = vc-relax and restart_mode = restart.

In the files attached you can find the input and output for both the initial relaxation and the subsequent scf step for one of these cases.

Best regards, Chiara Ricca

xespresso.relax.in

xespresso.relax.out

xespresso.scf.in

xespresso.scf.out