Setting up an electronic minimization: Difference between revisions

Revision as of 09:57, 11 March 2025

Setting up an electronic minimization calculation using density-functional theory requires a few steps. The input files must be created or copied into the execution folder. This includes making a few choices for the k-point sampling and electronic smearing, minimization algorithm, and exchange-correlation functionals. A dry-run can be used to review settings and select appropriate parallelization tags. After running the calculation, the output can be analyzed.

Step-by-step instructions

Create the input files

Step 1: Create a POSCAR file containing the structure for which you want to compute the electronic groundstate. External tools like VESTA^[1], or Python packages like the Atomic Simulation Environment (ASE)^[2] or pymatgen^[3] can help with this step.

Step 2: Choose an exchange-correlation (XC) functional appropriate for your material and quantity of interest.

Step 3: Create a suitable POTCAR file by following the instructions on our preparing a POTCAR page.

Step 4: Create a KPOINTS file to define the integration mesh in reciprocal space. Including a single k point at the origin, i.e. the Gamma point, neglects all interactions beyond the unit cell. This is appropriate for isolates systems like a single molecule or in large supercells. For bulk systems, start with a regular mesh. For shorter lattice vectors, more k points are required to achieve the same sampling density. Consult the symmetry reduction section of the KPOINTS page to select the appropriate mesh type. Alternatively to a KPOINTS file, the Template:Tag can be used.

Step 5: Write an INCAR file. It is recommended to start from a rather minimal file, and only specify the most important tags:

XC to specify the exchange-correlation functional.
ALGO to select the algorithm for electronic minimization.
ISMEAR to select the type of electronic smearing technique.
SIGMA to choose an appropriate smearing width of the electronic smearing.
ENCUT to set the plane-wave energy cutoff.
EDIFF to specify the global break condition for the electronic self-consistent loop

Optimize your settings

Step 6 (optional): Select the appropriate version of the VASP executable. I.e. vasp_gam if you only want to use the Gamma point for reciprocal space integration, vasp_ncl for noncollinear magnetic calculations, or vasp_std for anything else. Then Run a dry-run calculation to validate settings and uncover possible errors. Open a terminal, go to the calculation directory that contains all input files and run

 /path/to/your/vasp_std --dry-run

Step 7 (optional): Inspect the OUTCAR file of your dry-run. Take note of the number of bands, NBANDS, and the number of k-points, NKPTS, especially. Follow the guidelines on the optimizing the parallelization page to set NCORE and/or KPAR in the INCAR file.

Run the calculation

Step 8: Run the calculation. If you are new to VASP, or unsure about the calculation setup, run a small calculation and monitor the screen output. For parallel execution on 4 MPI ranks, the command reads

 mpirun -np 4 /path/to/your/vasp_std

At an HPC center, submit your job with a submission script. Ask your system administrator for help.

Once the calculation is finished, you have access to the electronic ground-state properties via the output files for the selected parameters. Check the OUTCAR file for warnings or advice. For help, consult the page about troubleshooting electronic convergence and search our Forum for similar issues.

Step 9 (convergence study): Repeat Steps 4 - 8 with increasingly accurate parameter settings, e.g. higher cutoff energy and denser k-points mesh, and monitor your quantity of interest. Stop if the quantity of interest reaches the desired accuracy.

Recommendations and advice

Mind: Make sure to specify the lattice vectors and ionic positions in the POSCAR with at least 7 digits of precision to ensure the symmetry analysis can function accurately.

Tip: Add only necessary tags to your INCAR file. Cluttered input is a common source of mismatched settings.

Tip: A larger smearing width SIGMA might be required to converge the calculation if your KPOINTS mesh is sparse.

Tip: ENCUT defaults to the largest ENMAX value found in the POTCAR file. Still, it is always a good idea to include it in the INCAR file to ensure comparability between different calculations.

Tip: Use the dry-run command-line argument or ALGO = None to check the feasibility of your settings and optimize parallelization tags, without wasting computational resources.

Tip: Some warnings are a bit hidden in the header section of the screen output. Redirecting the screen output to a file and saving it can simplify troubleshooting significantly.

Example

We will do a small DFT calculation of GaAs in the zincblende structure, using the local-density approximation (LDA) with the Perdew-Zunger parametrization of Ceperley-Alder Monte Carlo correlation data.^[4]^[5]. Thus, our XC functional will be set to XC = CA.

Setting up the POSCAR file

The POSCAR file starts with a comment line and a scaling factor, which in our case corresponds to the lattice parameter of GaAs, around 5.65 Angstrom.

Zincblende GaAs
  5.65000000000

Next we need to define the lattice vectors. Zincblende is a face-centered cubic (fcc) structure with two different elements in the unit cell. We can describe the fcc lattice with three vectors, pointing from the origin to the face-centers of the cube:

     0.0000000000000000  0.5000000000000000  0.5000000000000000
     0.5000000000000000  0.0000000000000000  0.5000000000000000
     0.5000000000000000  0.5000000000000000  0.0000000000000000

Now, we define the ion types, and in the line below the number of ions in the structure for each type:

 Ga  As
  1   1

Specify the positions of the atoms in direct coordinates, with Ga at the origin and As a quarter along the diagonal of the cube:

Direct
  0.0000000000000000  0.0000000000000000  0.0000000000000000
  0.2500000000000000  0.2500000000000000  0.2500000000000000

This is the complete POSCAR file:

Zincblende GaAs
  5.65000000000
     0.0000000000000000  0.5000000000000000  0.5000000000000000
     0.5000000000000000  0.0000000000000000  0.5000000000000000
     0.5000000000000000  0.5000000000000000  0.0000000000000000
 Ga  As
  1   1
Direct
  0.0000000000000000  0.0000000000000000  0.0000000000000000
  0.2500000000000000  0.2500000000000000  0.2500000000000000

If you have access to py4vasp, the structure can be visualized with two lines of Python code in a Jupyter notebook.

from py4vasp import calculation

calculation.structure.plot(supercell=2,selection="POSCAR")

If ASE^[2] is installed, an equivalent POSCAR file can be created as follows:

from ase.build import bulk
from ase.io.vasp import write_vasp

atoms = bulk("GaAs", crystalstructure="zincblende", a=5.65)
write_vasp("POSCAR", atoms, direct=True, sort=False)

Creating the POTCAR file

We have already decided to use XC = CA, and can create the POTCAR file as discussed on the preparing a POTCAR page.

Creating the KPOINTS file

Since our structure is face-centered cubic, we create a regular Gamma-centered k-point mesh according to the symmetry considerations for KPOINTS files.

 Regular k-point mesh
   0
 Gamma
  7 7 7

Creating the INCAR file

We chose the efficient combination of a blocked-Davidson algorithm and the RMM-DIIS algorithm which can be selected with ALGO = Fast.

ALGO = Fast

GaAs is a semiconductor, so we could use the tetrahedron method ISMEAR = -5, but bandgaps are underestimated systematically by DFT and XC functional or lattice parameter may fail to reproduce experimental results. Thus, following the recommendation on electronic smearing techniques, it is safer to select Gaussian smearing and a small smearing width:

ISMEAR = 0
SIGMA = 0.1
EFERMI = MIDGAP

For an initial guess of the plane-wave cutoff energy ENCUT, we can search for ENMAX in the POTCAR, e.g. by grep ENMAX POTCAR, and set the largest as a starting point. In preceding calculations this value should be increased, e.g. by increments of approximately 20%. Accordingly, in the first run we set:

ENCUT = 285

For the break condition of the self-consistent loop, we select $1\times10^{-6}$ eV:

EDIFF = 1.0E-06

The complete INCAR file is:

XC = CA
ALGO = Fast
ISMEAR = 0
EFERMI = MIDGAP
SIGMA = 0.1
ENCUT = 285
EDIFF = 1.0E-06

Performing a dryrun

We are not doing a noncollinear, nor a Gamma-only calculation, thus we execute a VASP dry-run with the standard executable:

/your/vasp_dir/bin/vasp_std --dry-run

Which will print a warning about the dry-run and some information about the MPI-ranks, OMP-threads, the VASP version, and the input structure. Mistakes in the setup, e.g. if the order of elements in the POSCAR and POTCAR do not match, warnings are printed.

We can now check the OUTCAR file and find the total number of k-points, 20, and number of bands (NBANDS), 13. This means a relatively low number of bands and a decent number of k points. If we want to run our calculation on 4 MPI ranks, setting KPAR = 4 is an excellent choice for parallelization. Mind that the parallelization changes number of bands.

Running the calculation

After adding KPAR = 4 to the INCAR file, we run the calculation on 4 MPI ranks:

mpirun -np 4 /your/vasp_dir/bin/vasp_std

Consult the page on screen output for details about the information VASP prints out. For this example it should be similar to this:

 running    4 mpi-ranks, with    1 threads/rank, on    1 nodes
 distrk:  each k-point on    1 cores,    4 groups
 distr:  one band on    1 cores,    1 groups
 vasp.6.5.0 16Dec24 (build Dec 18 2024 11:18:52) complex                        
 
 POSCAR found type information on POSCAR GaAs
 POSCAR found :  2 types and       2 ions
 Reading from existing POTCAR
 scaLAPACK will be used
 Reading from existing POTCAR
 LDA part: xc-table for (Slater(with rela. corr.)+CA(PZ))
 , standard interpolation
 POSCAR, INCAR and KPOINTS ok, starting setup
 FFT: planning ... GRIDC
 FFT: planning ... GRID_SOFT
 FFT: planning ... GRID
 WAVECAR not read
 entering main loop
       N       E                     dE             d eps       ncg     rms          rms(c)
DAV:   1     0.623500523606E+02    0.62350E+02   -0.70852E+03   528   0.135E+03
DAV:   2    -0.533903918847E+01   -0.67689E+02   -0.65331E+02   580   0.246E+02
DAV:   3    -0.978648308483E+01   -0.44474E+01   -0.44252E+01   635   0.613E+01
DAV:   4    -0.985351010991E+01   -0.67027E-01   -0.67012E-01   614   0.819E+00
DAV:   5    -0.985490478939E+01   -0.13947E-02   -0.13947E-02   641   0.931E-01    0.301E+00
RMM:   6    -0.966994813504E+01    0.18496E+00   -0.21049E-01   715   0.453E+00    0.175E+00
RMM:   7    -0.962995486843E+01    0.39993E-01   -0.10315E-01   701   0.182E+00    0.574E-01
RMM:   8    -0.962647867206E+01    0.34762E-02   -0.12692E-02   740   0.127E+00    0.937E-02
RMM:   9    -0.962642442346E+01    0.54249E-04   -0.21087E-03   757   0.536E-01    0.594E-02
RMM:  10    -0.962647797834E+01   -0.53555E-04   -0.39237E-04   794   0.212E-01    0.167E-02
RMM:  11    -0.962646653288E+01    0.11445E-04   -0.91747E-05   785   0.105E-01    0.529E-03
RMM:  12    -0.962646808711E+01   -0.15542E-05   -0.17691E-05   735   0.426E-02    0.300E-03
RMM:  13    -0.962646810096E+01   -0.13852E-07   -0.27033E-06   491   0.223E-02
   1 F= -.96264681E+01 E0= -.96264536E+01  d E =-.289650E-04
 writing wavefunctions

You can now use the output files to analyze the electronic ground-state properties. Do not forget to perform a convergence study before reporting a value, see Step 9.

References

,,

[vesta-1] ttps://jp-minerals.org/vesta/en/ (2025).

[ase-2] ttps://wiki.fysik.dtu.dk/ase/ (2025).

[pymatgen-3] ttps://pymatgen.org/ (2022).

[ceperley1980-4] D. M. Ceperley and B. J. Alder, Phys. Rev. Lett. 45, 566 (1980).

[perdewzunger1981-5] J. P. Perdew and A. Zunger, Phys. Rev. B 23, 5048 (1981).

[1]

[2]

[3]

[4]

[5]

Revision as of 09:57, 11 March 2025 (view source) Huebsch (talk \| contribs) m (Huebsch moved page Construction:Setting up a DFT calculation to Setting up an electronic minimization without leaving a redirect) ← Older edit		Revision as of 09:57, 11 March 2025 (view source) Huebsch (talk \| contribs) No edit summary Newer edit →
Line 194:		Line 194:


	[[:Category:Howto\|Howto]], [[:Category:Calculation_setup\|Calculation setup]], [[:Category:Electronic minimization\|Electronic minmization]]		[[Category:Howto\|Howto]], [[Category:Calculation_setup\|Calculation setup]], [[Category:Electronic minimization\|Electronic minmization]]