The short version
- Use a Γ-centred mesh unless you have a reason not to. It is required for fcc and rhombohedral cells and for the tetrahedron method (
ISMEAR = -5); hexagonal cells take a Γ-centred mesh or one shifted by 1/3 in the plane. - Make the subdivisions proportional to the reciprocal-lattice lengths: a long real-space axis gets few k-points, a short one many. KSPACING does this for you.
- Converge the mesh for the quantity you report. Compare meshes by the number of irreducible k-points VASP finds, not by N.
- Metals need about ten times more k-points than insulators.
- With no KPOINTS file, VASP uses
KSPACING = 0.5, which is coarse for metals. - Band structures: a path in line mode, read with
ICHARG = 11for LDA/GGA, or a KPOINTS_OPT file (VASP 6.3.0 and later). Hybrids never use ICHARG = 11.
k-mesh from your POSCAR
Paste a POSCAR and choose a spacing. The calculator applies the same rule as KSPACING, guesses the lattice type from the cell shape, and writes the KPOINTS file. It runs in your browser.
The ways to write KPOINTS
Line 1 is a comment. Line 2 is 0 for an automatic mesh, or the number of k-points (or points per segment). Only the first character of line 3 matters: G Γ-centred, M Monkhorst–Pack, L line mode, A fully automatic; for lists and generating vectors, C or K means Cartesian and anything else fractional (reciprocal).
Regular k-point mesh
0 ! 0 = generate the mesh automatically
Gamma ! Gamma-centred
4 4 4 ! subdivisions N1 N2 N3 along b1 b2 b3
0 0 0 ! optional shiftAutomatic generation
0
Monkhorst-Pack ! odd N: same as Gamma-centred; even N: shifted, no Gamma point
4 4 4
0 0 0Hexagonal mesh shifted by 1/3 in the plane
0
Gamma ! keep Gamma mode: Monkhorst-Pack would add another shift
6 6 4
0.3333333333 0.3333333333 0.0Explicit k-point list
4 ! number of k-points
Cartesian ! units of 2pi/a, a = the POSCAR scale; "Reciprocal" for fractions of b1 b2 b3
0.0 0.0 0.0 1
0.0 0.0 0.5 1
0.0 0.5 0.5 2
0.5 0.5 0.5 4k-points along high-symmetry lines
40 ! points per segment, end points included
Line-mode
Reciprocal ! fractional coordinates of b1 b2 b3
0 0 0 Gamma
0.5 0.5 0 X
0.5 0.5 0 X
0.5 0.75 0.25 W
0.5 0.75 0.25 W
0 0 0 GammaFully automatic mesh
0
Auto
20 ! length R_k in Angstrom: about 10 for large-gap insulators, up to 100 for d metalsVASP writes the irreducible points of every automatic mesh to IBZKPT, in the explicit-list format; edit a copy of it rather than typing a list by hand. Generating vectors (line 3 Reciprocal or Cartesian with three vectors and a shift) give generalized meshes, such as a conventional-cell mesh on a primitive cell.
Which mesh for which lattice
| Lattice | Mesh | Why |
|---|---|---|
| fcc, face-centred orthorhombic | Γ-centred only | A Monkhorst–Pack mesh breaks their symmetry and the generating lattice no longer matches the reciprocal lattice. VASP advice |
| Hexagonal (incl. SmCo₅-type) | Γ-centred, or a 1/3 shift in the plane | Never Monkhorst–Pack with even in-plane subdivisions: it breaks the hexagonal symmetry. Use N1 = N2. VASP advice |
| Rhombohedral | Γ-centred only | VASP advice |
| Body-centred tetragonal or orthorhombic, primitive cell | N1 = N2 = N3, or ratios 1/|a1| : 1/|a2| : 1/|a3| | KSPACING follows |b_i|, which may not suit these cells. Alternatively use the conventional cell. VASP advice |
| Cubic, tetragonal, orthorhombic (primitive), monoclinic, triclinic | Γ-centred or Monkhorst–Pack | Either works; Γ-centred is the safe default. |
The tetrahedron method (ISMEAR = -5) needs a Γ-centred mesh with at least 4 k-points, whatever the lattice. When a mesh does not fit the lattice symmetry, VASP reports it in a message from its routine IBZKPT about the “generating k-lattice”; change the mesh rather than ignoring it.
How dense
For a spacing Δk = KSPACING in Å⁻¹ (2π included), VASP takes Ni = max(1, ⌈|bi| · 2π / Δk⌉), where bi are the reciprocal vectors without the 2π (bi · aj = δij). For an orthogonal cell, |bi| = 1/ai: silicon's 5.43 Å cubic cell at Δk = 0.25 Å⁻¹ gets ⌈1.157 / 0.25⌉ = 5 subdivisions per axis.
Rules of thumb for the full Brillouin zone
- Insulators: about 100 k-points per atom bring the energy error below 10 meV.
- Metals: about 1000 per atom for the same accuracy.
- Transition metals with a steep DOS at the Fermi level: up to 5000 per atom for about 1 meV per atom.
- Even meshes (8 × 8 × 8) are efficient up to N = 8; beyond that, odd meshes (11 × 11 × 11).
- Compare an energy difference between cells only with consistent meshes: for commensurate supercells, use the same k-point set.
Count “k-points per atom” as N1 × N2 × N3 × (atoms in the cell).
Converting between conventions
| Where | Setting | Same as KSPACING |
|---|---|---|
KPOINTS Auto | length R_k (Å) | 2π / R_k (it rounds instead of rounding up) |
| VASPKIT | resolution in units of 2π Å⁻¹ | 2π × the resolution: 0.04 ≈ 0.25 |
| pymatgen | reciprocal_density ρ, k-points per Å⁻³ of reciprocal volume | about ρ−1/3 for a cubic cell: 64 ≈ 0.25, 100 ≈ 0.22, 200 ≈ 0.17 |
Materials Project uses ρ = 64 to relax, 100 for static and DOS runs, and KSPACING 0.22–0.44 (from the band gap) for r²SCAN.
Band structures and dense DOS
| Functional | How |
|---|---|
| LDA, GGA (PBE) | Either one run with the regular mesh in KPOINTS and the path in a KPOINTS_OPT file (VASP 6.3.0 and later), or two runs: a self-consistent one, then the path in KPOINTS (line mode) with ICHARG = 11 reading its CHGCAR. |
| Meta-GGA (r²SCAN) | Before VASP 6.6.0: as for hybrids. From 6.6.0 the ICHARG = 11 route also works. KPOINTS_OPT works on every version from 6.3.0. |
| Hybrid (HSE06) | Never ICHARG = 11. Run self-consistently with the regular mesh in KPOINTS and the path in KPOINTS_OPT, or append the path to an IBZKPT copy as zero-weight k-points. For gapped systems set HFRCUT = -1. |
- Take the Fermi energy from the self-consistent run: it cannot be computed from points on a path. With KPOINTS_OPT, set EFERMI or pass the Fermi energy to the plotting tool.
- Set LMAXMIX to 4 (d) or 6 (f) in both the self-consistent and the ICHARG = 11 run.
- Paths come from SeeK-path, pymatgen's HighSymmKpath or VASPKIT (task 303). Each tool's path belongs to its standardized primitive cell: run on that cell, or the labels are wrong.
- A zero-weight hybrid run should not restart from a converged hybrid WAVECAR: the SCF can stop before the zero-weight states converge.
- Spin-polarized projected DOS from KPOINTS_OPT lost the second spin channel in VASP 6.6.0 (fixed in 6.6.1); PROCAR_OPT was broken in 6.5.0 and 6.5.1.
k-points for the band structure
40
Line-mode
Reciprocal
0.0 0.0 0.0 Gamma
0.5 0.0 0.5 XThe k-point tags
| Tag | What it does | Default | What to use | Watch out |
|---|---|---|---|---|
KSPACING | k-mesh from the INCAR: the largest allowed spacing between k-points, in Å⁻¹ (2π included). | 0.5 (Γ-centred while KGAMMA = .TRUE., its default) | Smaller is denser; converge it. Without a KPOINTS file it replaces one. | Ignored when a KPOINTS file is present. The default 0.5 is coarse for metals. |
KGAMMA | Γ-centred mesh for KSPACING. | .TRUE. | Keep .TRUE. | Ignored when a KPOINTS file is present. |
KSPACING_OPT 6.6.0+ | k-spacing (Å⁻¹, 2π included) of an extra non-self-consistent mesh, used when there is no KPOINTS_OPT file. | 0.5 | A denser mesh for the DOS in the same run as the self-consistent one. | Needs VASP 6.6.0 or later; with older versions write a KPOINTS_OPT file instead. |
LKPOINTS_OPT | Whether VASP reads the KPOINTS_OPT file. | .TRUE. if a KPOINTS_OPT file is present | Set .FALSE. to ignore a KPOINTS_OPT left in the folder. | A stray KPOINTS_OPT file silently adds an extra non-self-consistent step. |
KPOINTS_OPT_NKBATCH | Batch size of k-points in the KPOINTS_OPT step. | the number of irreducible k-points of the self-consistent run | Lower it to save memory; raise it for speed. | |
KPOINTS_OPT_MODE 6.5.0+ | Diagonalizer for the KPOINTS_OPT step: 0 skip, 1 blocked Davidson, 2 exact. | 1 | Use 2 if the Davidson step struggles. |
Check the mesh VASP used
grep "irreducible k-points" OUTCAR # how many points the mesh reduced to by symmetry
grep NKPTS OUTCAR # the same number, with the number of bands
head -3 IBZKPT # line 2: number of irreducible k-points
vasp_std --dry-run # VASP 6.4 and later: writes IBZKPT and stopsThe irreducible count sets the cost and suggests KPAR: choose a divisor of the number of MPI ranks that is at most this count.
Common mistakes
- Monkhorst–Pack with even N for fcc or hexagonal cells: lower symmetry, slower convergence, or an IBZKPT error.
- No KPOINTS file, so VASP silently uses KSPACING = 0.5 Å⁻¹; or a forgotten KPOINTS file that silently overrides KSPACING.
- Mixing conventions: a VASPKIT 0.04 is a KSPACING of 0.25; pymatgen's reciprocal_density is a density, not a spacing.
- Judging convergence by N when comparing even and odd meshes: compare irreducible counts instead.
- A line-mode KPOINTS in a self-consistent run.
- A path from one tool used on a differently oriented or non-standard cell.
- The tetrahedron method with fewer than 4 k-points, a shifted mesh, or for forces in metals.
- A KPOINTS_OPT file left in the folder: VASP reads it automatically.
- With spin–orbit coupling, a k-point set that depends on the spin direction: test
ISYM = -1.
Other tags on this sheet
The physics behind this sheet
- Lecture 13: The reciprocal lattice and the Brillouin zone (in preparation)
- Lecture 14: Bloch's theorem and band structure (in preparation)
- Lecture 15: k-point sampling and Monkhorst–Pack grids (in preparation)
- Lecture 29: Band structure, DOS and projected DOS (in preparation)
Further reading
- VASP Wiki, “KPOINTS” (every format, the mesh formulas, and which mesh suits which lattice). · vasp.at/wiki
- VASP Wiki, “Number of k points and method for smearing” (k-point density for insulators and metals, and comparing meshes). · vasp.at/wiki
- VASP Wiki, “Band-structure calculation using density-functional theory” (KPOINTS_OPT and the ICHARG = 11 route). · vasp.at/wiki
- VASP Wiki, “Band-structure calculation using hybrid functionals” (zero-weight k-points, KPOINTS_OPT and HFRCUT). · vasp.at/wiki