Increasing energy in electronic SCF

[joshua_kas]

Hello VASP experts,
I have a very large unit cell (665 atoms) for which I would like to obtain the projected DOS. I ran the ground state with default number of bands, and the SCF converges after 61 iterations. This seems fine to me, however, the odd thing is the energy convergence in OSZICAR. The energy seems to converge quickly to about -0.584998493089E+04 eV in the first 12 iterations, with dE becoming generally smaller and smaller. On the 13th iteration, the energy jumps upward by about 800 eV, then continues to climb until iteration 35, at which point it slowly converges, only changing by about -0.2 eV from iteration 35 to 61. I'm not really worried about the slow convergence, and I don't really expect that the energy always decreases. I am however worried that the 12th iteration has an energy that is nearly 1000 eV lower than that of the converged energy. I suspect that this has something to do with the way that the SCF is started, but I would like to be sure that there is nothing obviously wrong here.
I'm using VASP 5.3. Please find my input and output attached.
Thank you very much,
Josh Kas

[martin.schlipf]

As you already said the onset of the density mixing is the cause for the step in the energy. This is controlled by the NELMDL INCAR tag. Jumps in the energy can happen during the optimization in particular in the beginning. The ALGO you chose does not treat the energy as a variationally quantity and the different contributions can converge at different rates.
But in general it can be a good idea to verify these claims for yourself to get familiar with the method. You can e.g. use a different ALGO to optimize the electronic problem and check if you end up in the same state with the same energy.