Comment 10 for bug 1834949

Revision history for this message
Peter Christian Schmitz (pcschmitz2018) wrote :

Dear Mr. Papior,

to 1):
Thank you for the patch and trying to solve this issue in future versions of Siesta.
However, as written above, even with the code modification a significant dH jump occurs after restarting. Even when the initial linear mixing weight is set to almost 0. The issue does not yet seem to be clarified..

to 2):
Thank you for the concern, but actually i had set the mixing weight to 0 to demonstrate the following:
- 1. as you noted, the initial dH jump is still present. Probably the second dH value would be equivalent to the first if i had used the modified code. But the higher dH value remains. This should not happen according to the documentation, because no new density would be mixed in at all, at the first step. As you now said: Without the modification the initial mixing is 1. Ok. So this can explain why the second jump is still there. But why the first ?
- 2. the dH value also stays constant after 3 steps. This should not happen either, because, if the density is not changed, the dH error between the steps should go down to 0 ! (or am i misunderstanding this ?)

From my side, the following questions remain:
- can the initial dH jump be reduced to 0 (compared to the provided density from the last run) or the dH value from the last run ?
- can the Hamiltonian be loaded directly ?