Most of my experience with DFT comes from the use of VASP, although other DFT packages do exist. The benefit of using VASP is it's wide use, active development. More importantly, the pseudo-potential library is widely used which eliminates the problems of having to create a working pseudo-potential library, and usually has cutting edge improvements, such as hybrid functionals, GW, which I personally have little experience in. It uses a plane-wave basis set which makes it more efficient for materials science.

I am writing a series of tutorials on VASP, which introduces people to computational simulations in a gentile approach which I thnk should be appropriate for advanced undergraduates of materials science/physics/chemistry and beginning graduate students. There might be more advanced stuff later involving software I write, but this will likely be in a different section.

To get setup, one should do the relevant portions of the CompMatSci bootcamp to get yourself up and running. I have setup a series of steps to get a system up and running based upon what I consider an optimal setup to do computational simulations based on my personal experience and having on-boarded people to do computational materials simulations. This should be done first.

The progression of these tutorials follows the abinit tutorial, which I consider to be better as an introduction to DFT than VASP's website.

http://cms.mpi.univie.ac.at/wiki/index.php/Installing_VASP

Example Calculations

• Ionic Structure Relaxation
  • KPOINTS file
    • in hexagonal cells choose GAMMA centered kpoint mesh[1]
    • Automatic
    • Gamma
    • Monkhorst-Pack
  • INCAR FILE
    • IBRION = 2 # conjugate gradient method
    • ISIF
      • ISIF = 2 # relax positions only
      • ISIF= 3 # relax position, cell shape, cell volume
      • ISIF = 4 # relax ions and cell shape
      • ISIF = 5 # relax shape only
      • ISIF = 6 # relax shape and volume only
      • ISIF = 7 # relax cell volume only
    • POTIM = 0.50
      • When using the conjugate gradient method, I find that the POTIM =0.50 is too large and often leads to non-converged results. I commonly use a POTIM of 0.15.
    • EDIFF = 1e-8
    • EDIFFG = 1e-6
  • Common Errors
    • ZPOTRF errors. This is usually caused by short bond distances. If this occurs on the first step, the volume needs to be increased. Otherwise, it was due to the step being to big and the POTIM parameter should be decreased

References:

[1] https://www.vasp.at/vasp-workshop/slides/k-points.pdf

The INCAR file

For electronic relaxation:

EDIFF

ENCUT[1]

ISMEAR, SIGMA tag.

For semiconductors and insulators, use the tetrahedron method (ISMEAR -5). For large cells, use Gaussian smearing (ISMEAR = 0) with a small SIGMA = 0.05. For relaxations in metals use ISMEAR=1 or ISMEAR=2 and an appropriate SIGMA value. For metals, the default value (SIGMA = 0.2) is sensible. For DOS and total energy calculations use the tetrahedron method. For calculated forces, stress tensors and calculation of phonon frequencies use the method of Methfessel-Paxton (ISMEAR > 1). The tetrahedron method for semiconductors and insulators are always correct because partial occupancies do not vary and are either zero or one.[1]

• -5
  • tetrahedron method with Blöchl corrections (use Gamma centered k-mesh)
• 0
  • Gaussian smearing
  • requires SIGMA parameter
• 1...N
  • method of Methfessel-Paxton order N
  • requires SIGMA parameter
  • check to see entropy less than 1 meV/atom.
    • cat OUTCAR | grep 'entropy T*S EENTRO ='

[1] Georg Kresse, Martijn Marsman, and Jürgen Furthmüller. "ISMEAR, SIGMA, FERWE, FERDO SMEARINGS tag" VASP the GUIDE. Link

[2] https://docs.quantumwise.com/manuals/technicalnotes/occupation_methods/occupation_methods.html

Choice of Pseudopotentials in VASP

Types of PAW Potentials

Choice of PAW potentials for transition metals:

X_pv, the semi core p states are treated as valence

• X_pv should be used for early transition metals

X_sv, the semi core s states are treated as valence

• for the 3d elements, even the Ti, V and Cr potentials give reasonable results
• for the 4d elements are most problematic, and I advice to use the X pv potentials up to Tc pv
• for the 5d elements: 5p states are rather strongly localised (below 3 Ry), since the 4f shell becomes filled

References:

[1] G. Kresse. "The PAW and US:PP database." https://www.vasp.at/vasp-workshop/slides/pseudoppdatabase.pdf

[LDA] G. Kresse, and J. Joubert "From ultrasoft pseudopotentials to the projector augmented wave method.", Phys. Rev. B. 59, 1758 (1999)

For HPC Gator

uf hpc wiki

http://wiki.hpc.ufl.edu/doc/UF_Research_Computing_Wiki

ssh <username>@submit.hpc.ufl.edu

cp ~/class /scratch/hpc/eragasa

cd /scratch/hpc/eragasa

find VASP

qsub <runjob>

qstat <runjob>

YOU ARE GOOD IF

reached required accuracy - stopped structural accuracy minimization

Files

Input files

INCAR

EDIFF

EDIFFG

KPOINTS - determines how many k-points are used to sample the Brillouin zone. for molecules or atoms, only a single k-point is required. For atoms and molecules the Bloch theorem does not apply, therefore there is not a need to use more than a single k-point. When more k-points are used, only the interaction between the atoms is described more accurately.

POSCAR - inital structure

POTCAR

Output File:

OUTCAR

total energy of a system

     grep ""gy wi" OUTCAR

CHGCAR - amount of electrons

CONTCAR - final structural size