Dos

"You must do this, otherwise VASP cannot read the CHGCAR and will terminate."

Prerequisites (hard)

Require the user to point at an existing SCF task with:

• CHGCAR (non-empty)
• the same POSCAR as will be used for DOS
• the same POTCAR (species order, pseudo version)
• LCHARG = .TRUE. was set in the SCF INCAR

If any of these is missing, stop and ask.

Must provide

• source SCF context path (and verify the four items above)
• DOS intent (total DOS or projected DOS)
• target k-mesh for DOS (typically 2× the SCF mesh in each direction)
• energy window (EMIN, EMAX) — default: auto from EFERMI ± 15 eV
• element-resolved projection level (LORBIT)

Usually should be explicit

Tetrahedron trap

ISMEAR = -5 is wrong for:

• Small cells with < 4 irreducible k-points (VASP silently falls back or errors out) — fall back to ISMEAR = 0, SIGMA = 0.05.
• Force / relaxation runs — use the static-SCF smearing instead.
• Hybrid functional runs — use ISMEAR = 0, SIGMA = 0.05.

Concrete example — fcc Si projected DOS

Prerequisite: a converged dft-vasp/static run on fcc Si at its relaxed volume with LCHARG = .TRUE. (see dft-vasp/relax for the equilibrium structure — V/atom ≈ 14.5 ų, a ≈ 3.87 Å).

Task directory:

Si_fcc_dos/
├── POSCAR            # same cell as SCF prerequisite
├── INCAR
├── KPOINTS           # denser than SCF
├── POTCAR            # same as SCF prerequisite
└── CHGCAR            # copied from the SCF run

INCAR:

SYSTEM = fcc Si DOS
ISTART = 1
ICHARG = 11           # fixed charge, non-SCF
ENCUT  = 320
PREC   = Accurate
LREAL  = .FALSE.
EDIFF  = 1E-6
NELM   = 60
ISMEAR = -5           # tetrahedron + Bloechl
LORBIT = 11           # site-projected + orbital-resolved
NEDOS  = 3001
EMIN   = -15
EMAX   =  15
LWAVE  = .FALSE.
LCHARG = .FALSE.

KPOINTS — refine to 21×21×21 Γ-centered (~8 × more points than the 11³ SCF mesh; still cheap for a 1-atom cell):

DOS mesh
0
Gamma
21 21 21
0  0  0

Physical sanity checks after the run

1. ls DOSCAR — file exists and is > 10 kB.
2. head -6 DOSCAR — first header line reports the number of ions; line 5 should give EMAX EMIN NEDOS EFERMI weight.
3. grep 'E-fermi' OUTCAR | tail -1 — Fermi level in eV.
   • Semiconductor with a clean gap: the DOS-stage E_F must match the SCF E_F to within 20 meV; larger drift means inconsistent CHGCAR.
   • Metal / gapless system: Gaussian→tetrahedron can shift E_F by 100–300 meV because the two methods assign partial occupancies differently at the Fermi surface. This is physical, not a bug. Verified locally: fcc 1-atom Si E_F = 9.987 eV (SCF, Gaussian σ=0.05) → 10.194 eV (DOS, tetrahedron), Δ = 207 meV.
4. The fcc 1-atom Si cell is metallic by construction (4 valence electrons half-fill the 3p band). It is the VASP wiki's EOS tutorial cell, not a band-gap demonstration. For a real semiconductor DOS demo, use the 2-atom diamond primitive (POSCAR with positions 0 0 0 and 0.25 0.25 0.25, a_conv ≈ 5.47 Å at PBE).
5. For projected DOS (LORBIT = 11), open PROCAR or check the last columns of DOSCAR — l+m components must sum (per site) to the total DOS within 1%.

Known build traps

• ISMEAR = -5 with a highly anisotropic cell (e.g., slab) will produce noisy DOS — use denser k-mesh in the short reciprocal directions.
• ICHARG = 11 with a CHGCAR from a different functional / ENCUT gives physically meaningless DOS. The skill must verify the two INCARs share GGA, ENCUT, PREC, and LASPH tags.

Expected output

1. DOS-stage task directory with INCAR, KPOINTS, staged CHGCAR link
2. prerequisite check summary (SCF path, CHGCAR size, POTCAR hash)
3. settings summary (why tetrahedron vs Gaussian)
4. sanity-check checklist
5. handoff note to dpdisp-submit