Static

K-point policy

• default: KSPACING in INCAR (e.g. KSPACING = 0.3 ≈ coarse, 0.2 ≈ production)
• generate KPOINTS only when user asks for an explicit mesh
• for small cells of semiconductors prefer Γ-centered odd mesh (see example)

ENCUT selection rule

• Read ENMAX from the POTCAR header of every species.
• Baseline: ENCUT = ceil(max(ENMAX)). This matches the VASP wiki default.
• For stress / volume optimization or NEP label generation, raise ENCUT to 1.3 × max(ENMAX) to suppress Pulay stress. State the rule explicitly in the summary so the user can override.

ISMEAR / SIGMA selection rule

Concrete example — fcc Si SCF (anchored on VASP wiki: Fcc Si)

Task directory layout:

Si_fcc_scf/
├── POSCAR
├── INCAR
├── KPOINTS
└── POTCAR            # concatenated from <POTPAW>/Si/POTCAR

POSCAR (1-atom primitive, a = 3.9 Å, matches the wiki tutorial):

fcc Si
  3.9
    0.5  0.5  0.0
    0.0  0.5  0.5
    0.5  0.0  0.5
Si
  1
Direct
  0.0  0.0  0.0

INCAR (minimal converged SCF, ENMAX(Si POTCAR) ≈ 245 eV):

SYSTEM = fcc Si
ISTART = 0
ICHARG = 2
ENCUT  = 320         # 1.3 × ENMAX for safer volume/stress
PREC   = Accurate
LREAL  = .FALSE.     # reciprocal-space projection for small cells
EDIFF  = 1E-6
NELM   = 60
ISMEAR = 0
SIGMA  = 0.05
LWAVE  = .FALSE.
LCHARG = .FALSE.

KPOINTS (11×11×11 Γ-centered, 56 irreducible k-points per the wiki):

Automatic mesh
0
Gamma
11 11 11
0  0  0

POTCAR mapping: cat <POTPAW>/Si/POTCAR > POTCAR.

Physical sanity checks after the run

1. grep -c 'reached required accuracy' OSZICAR — not needed for SCF; instead confirm the final DAV line reports d E < EDIFF.
2. grep 'free energy TOTEN' OUTCAR | tail -1 — should match the E0 in OSZICAR within a few meV.
3. For fcc Si with the 1-atom primitive at a = 3.9 Å above, expect E0 ≈ −4.89 eV/atom (PBE, ENCUT 320, 11×11×11 Γ-centered). Note that a = 3.9 Å is the wiki's EOS scan starting point, not the PBE equilibrium (~3.87 Å). Deviation from −4.89 eV by > 30 meV signals POTCAR or ENCUT mismatch.
4. grep 'entropy T\*S' OUTCAR | tail -1 — entropy contribution per atom must be < 2 meV; otherwise lower SIGMA or refine k-mesh.

vaspkit workflow for KPOINTS and POTCAR

For production work, always use vaspkit to generate KPOINTS and assemble POTCAR instead of writing them by hand. vaspkit reads ~/.vaspkit for the POTCAR library path and automatically selects recommended pseudopotentials.

Generate KPOINTS + POTCAR in one step

echo -e "102\n1\n0.04" | vaspkit

• 102 = uniform k-mesh generation
• 1 = Monkhorst-Pack (or Gamma-centered per GAMMA_CENTERED in ~/.vaspkit)
• 0.04 = k-spacing in 1/Å (smaller = denser mesh)

vaspkit writes both KPOINTS and POTCAR (with recommended potentials) in one pass. Always verify the chosen POTCAR variants in the vaspkit summary output — e.g., Pb_d (14 valence electrons) vs Pb (4 valence electrons) can dramatically change ENMAX requirements.

Generate KPOINTS only

echo -e "102\n1\n0.03" | vaspkit

Generate POTCAR only

echo -e "103" | vaspkit

vaspkit reads the species from POSCAR and concatenates POTCARs from the configured library path (PBE_PATH in ~/.vaspkit).

k-spacing guidelines

Runtime-verified example — PbTe 250-atom supercell (NEP labeling)

This example was runtime-verified on VASP 6.4.3 (2026-04-16) as part of a complete DFT → NEP training pipeline test using the PbTe structure from GPUMD tutorial 11_NEP_potential_PbTe.

Task: single-point SCF on a 250-atom Pb₁₂₅Te₁₂₅ cell extracted from the GPUMD PbTe train.xyz dataset, for NEP training data generation.

Structure preparation

from ase.io import read, write
# Extract one frame from GPUMD training data
atoms = read('train.xyz', index=0, format='extxyz')
# Remove existing calculator results (energy/forces)
atoms.calc = None
atoms.info = {}
atoms.arrays = {k: v for k, v in atoms.arrays.items()
                if k in ('numbers', 'positions')}
# Sort by species for VASP
atoms = atoms[atoms.numbers.argsort()]
write('POSCAR', atoms, format='vasp', vasp5=True, sort=True)

Input files (generated via vaspkit)

KPOINTS (vaspkit 102, spacing 0.04 → Γ-only for this large cell):

K-Spacing Value to Generate K-Mesh: 0.040
0
Monkhorst-Pack
   1   1   1
0.0  0.0  0.0

POTCAR (vaspkit 103, recommended potentials):

• Pb_d — 14 valence electrons, ENMAX = 237.835 eV
• Te — 6 valence electrons, ENMAX = 174.982 eV

INCAR:

SYSTEM = PbTe pipeline test
ISTART = 0
ICHARG = 2
ENCUT  = 300          # above max(ENMAX)=237.8 for Pb_d
PREC   = Normal       # low precision for pipeline test
LREAL  = Auto         # 250 atoms — use real-space projection
EDIFF  = 1E-4         # loose SCF for testing
NELM   = 40
ISMEAR = 0
SIGMA  = 0.1
LWAVE  = .FALSE.
LCHARG = .FALSE.
NCORE  = 4

Run

export OMP_NUM_THREADS=1
mpirun -np 16 vasp_std > vasp.out 2> vasp.err

Verified results

Post-SCF: convert to NEP training data

# Using dpdata (pip install dpdata)
python3 -c "
import dpdata
ds = dpdata.LabeledSystem('OUTCAR', fmt='vasp/outcar')
ds.to('extxyz', 'train.xyz')
"

Or via CLI:

uvx dpdata OUTCAR -i vasp/outcar -O train.xyz -o extxyz

The resulting train.xyz contains energy, forces, virial, and lattice in the extxyz format ready for NEP training. See nep-gpumd/train for the next step.

Known build traps

• Non-collinear / SOC jobs require a build without -DNGXhalf -DNGZhalf. A standard vasp_std build will crash on LSORBIT = .TRUE. with ERROR: non collinear calculations require that VASP is compiled without the flag -DNGXhalf and -DNGZhalf. If the user supplies such settings, stop and ask them to either switch to a vasp_ncl build or drop SOC.
• LHFCALC = .TRUE. (hybrid functional) also requires a special build.
• LREAL = Auto changes forces slightly; for NEP labeling keep LREAL = .FALSE..

Expected output

1. task directory with generated input files
2. settings summary (explicit ENCUT rule, smearing rationale)
3. unresolved choices for user confirmation (functional, vdW, spin)
4. physical-sanity-check checklist (see above)
5. handoff note to dpdisp-submit if execution is requested