1) Yes, but only in the first SCF step. In subsequent SCF steps the Hamiltonian is the mixed Hamiltonian. The initial dH jump is, if you like, an anomaly. But I wouldn't consider it a bug. The problem is that when you do regular SCF steps you do a regular Pulay mixing. While when you restart you are *forced* to do an initial linear mixing. In this regard the linear mixing is really crude compared to the Pulay mixing. I would suspect that if you store the complete mixing history the initial step wouldn't be any different than a continued SCF calculation (this still needs to be implemented).
After the first step you cannot ensure that dH *goes back* since your initial mixing is linear vs. Pulay. However, in the case where you *only* use linear mixing I would expect all SCF dH values to be equal (except the 1st). I.e. you can compare in this latter case.
2.a) you can't change the mixing from DM to H during the SCF (if this was what you meant?)
1) Yes, but only in the first SCF step. In subsequent SCF steps the Hamiltonian is the mixed Hamiltonian. The initial dH jump is, if you like, an anomaly. But I wouldn't consider it a bug. The problem is that when you do regular SCF steps you do a regular Pulay mixing. While when you restart you are *forced* to do an initial linear mixing. In this regard the linear mixing is really crude compared to the Pulay mixing. I would suspect that if you store the complete mixing history the initial step wouldn't be any different than a continued SCF calculation (this still needs to be implemented).
After the first step you cannot ensure that dH *goes back* since your initial mixing is linear vs. Pulay. However, in the case where you *only* use linear mixing I would expect all SCF dH values to be equal (except the 1st). I.e. you can compare in this latter case.
2.a) you can't change the mixing from DM to H during the SCF (if this was what you meant?)