Comment 14 for bug 1834949

1) Ok, this explains why, in a "normal" Pulay calculation the initial dH would jump upon restarting even when one set
Write.DM.History.NetCDF true
because the linear mixing gives different results than Pulay (in the history provided for the 2nd restarted step).

Still: Why does the the initial dH jump even occur when the initial linear mixing weight is set to 0 ? Mixing should have no impact on this value.
Yes, dH should stay constant (at the initial, "jumped" / high value) if weight.linear and mixing weight is set to 0. I am repeating this with the modified code now.

Regarding the implementation: It seems the mixing history can already be saved (flag above). A Pulay mixing routine at the restart would indeed be really nice to have (it might turn the "inevitable" initial dH error back to the value of the previous cycle)
But as long as it is not implemented :

=> Would it be beneficial to close the previous calculation with a linear mixing step (maybe even with zero mixing) ? One could then add a linear mixing block at the beginning of the restarted calculation, too, eliminating the linear / pulay mixing change during the restart. It would be shifted to later, inside the scf run

2.a) Ok. Yes it is also not a property of the mixer blocks. What remains is that one could try a permanent DM-mixing for the restarted calculation, with H provided from the previous run. Or a permanent H mixing, with H provided from the previous run (2.b). Testing this.