Comment 12 for bug 1834949

1) ok, thank you very much!
Then i had misunderstood dH as the change upon mixing, but actually it is the change upon constructing the Hamiltonian twice from a density (which is actually the DFT accuracy, correct).
Then the jump in dH at the restart must refer to an anomaly in the way how the hamiltonian is contructed / the next density DM1 is written (which must be different from what happens at later cycles / resp. earlier in the previous calculation).
Can we somehow find the bug by comparing these paths through the code in the first and in the restarted calculation, with weight.linear =0 ? Are these pathways different in the code ? (eg. regarding how history is considered, which orbital basis is used, or the previous single point calculation.. ?)

Is it correct that dH is always calculated by twice constructing a hamiltonian from the last-mixed density, or will dH, after i=1, be calculated from Hamiltonians that correspond to the original and to the next mixed DM ?
Then i could understand why dH is different in the restarted calculation compared to the previous one. But still, after the first step, it should go back to the value from the previous cycle (if mixing is 0 and it did not make any false guesses based on this dH value)..

2.*) Ok, understood, that makes sense.

a) indeed one could, just for the 1st step, set the mixing to DM and read in the .DM and .H files ? I will try if it is possible
b) thank you very much ! so i will use this feature and see if providing only the Hamiltonian, not the DM, gives a better restart dH