We will perform DFT+DMFT calculations for the charge-transfer insulator NiO. We start from scratch and provide all necessary input files to do the calculations: First for doing a single-shot calculation (and then for charge-self consistency).
Since the python script for performing the dmft loop pretty much resembles that presented in the tutorial on :ref:`SrVO3 <srvo3>`, we will not go into detail here but simply provide the script :ref:`nio.py`. Following Kunes et al. in `PRB 75 165115 (2007) <https://journals.aps.org/prb/abstract/10.1103/PhysRevB.75.165115>`_ we use :math:`U=8` and :math:`J=1`. We select :math:`\beta=5` instead of :math:`\beta=10` to ease the problem slightly. For simplicity we fix the double-counting potential to :math:`\mu_{DC}=59` eV by::
For sensible results run this script in parallel on at least 20 cores. As a quick check of the results, we can compare the orbital occupation from the paper cited above (:math:`n_{eg} = 0.54` and :math:`n_{t2g}=1.0`) and those from the cthyb output (check lines `Orbital up_0 density:` for a t2g and `Orbital up_2 density:` for an eg orbital). They should coincide well.
We calculate the local lattice Green's function - now also for the uncorrelated orbitals, i.e., the O p states, for what we use the script :ref:`NiO_local_lattice_GF.py`. The result is saved in the h5 file as `G_latt_orb_it<n_it>`, where `<n_it>` is the number of the last DMFT iteration.
To compare to results from literature we make use of the `maxent triqs application <https://triqs.github.io/maxent/master/>`_ and calculate the spectral function on real axis. Use this script to perform a crude but quick calculation: :ref:`maxent.py` using a linear real axis and a line-fit analyzer to determine the optimal :math:`\alpha`. The output is saved in the h5 file in `DMFT_results/Iterations/G_latt_orb_w_o<n_o>_it<n_it>`, where `<n_o>` is the number of the orbital and `n_it` is again the number of the last iteration. The real axis information is stored in `DMFT_results/Iterations/w_it<n_it>`.
In this part we will perform charge self-consistent DMFT calculations. To do so we have to adapt the VASP `INCAR` such that :program:`VASP` reads the updated charge density after each step. We add the lines::
which makes VASP wait after each step of its iterative diagonalization until the file vasp.lock is created. It then reads the update of the charge density in the file `GAMMA`. We change the mixing here to stabilize the updating, which can be problem for charge ordered systems. Vasp is terminated by an external script after a desired amount of steps, such that we deactivate all automatic stoping criterion by setting the number of steps to a very high number.
We take the respective converged DFT and DMFT calculations from before as a starting point. I.e., we copy the `CHGCAR` and `nio.h5` together with the other :program:`VASP` input files and :file:`plo.cfg` in a new directory. We use a script called :program:`vasp_dmft` to invoke :program:`VASP` in the background and start the DMFT calculation together with :program:`plovasp` and the converter. This script assumes that the dmft sript contains a function `dmft_cycle()` and also the conversion from text files to the h5 file. Additionally we have to add a few lines to calculate the density correction and calculate the correlation energy. We adapt the script straightforwardly (for a working example see :ref:`nio_csc.py`). The most important new lines are::
where the chemical potential is determined to a greater precision than before, the correction to the dft density matrix is calculated and output to the file :file:`GAMMA`. The correlation energy is calculated via Migdal-Galitski formula. We also slightly increase the tolerance for the detection of blocks since the DFT calculation now includes some QMC noise.
To help convergence, we keep the density (i.e., the GAMMA file) fixed for a few DFT iterations. We do so since VASP uses an iterative diagonalization. Within these steps we still need to update the projectors and recalculate the GAMMA file to not shuffle eigenvalues around by accident.
