In density functional theory (DFT), k points are selected wave vectors used to sample the electronic states across a crystal’s Brillouin zone. A calculation combines information from these points to approximate properties of the periodic solid, such as its total energy, electron density, and forces.
Why crystals need k-point sampling
In a periodic crystal, electrons are described by quantum states that vary with wave vector k. Bloch’s theorem lets a calculation represent these states using a unit cell and a wave vector in reciprocal space, rather than explicitly modeling an infinitely large crystal. The allowed wave vectors fill the first Brillouin zone, the primitive cell of the crystal’s reciprocal lattice.
Many quantities calculated for a periodic solid require an average or integral over that zone. A computer cannot generally evaluate every point in a continuous region, so it approximates the integral using a finite set of k points and associated weights. In simplified form, an integral is replaced by a weighted sum: each sampled point represents a portion of the Brillouin zone.
This is why a k point is not an atom, a real-space grid point, or a choice of exchange-correlation functional. It is a sampling location in reciprocal space. The k-point set is one numerical setting among several that determine how accurately a periodic DFT calculation represents the system.
What a k point means in practice
Each k point corresponds to a wave vector within the Brillouin zone and is used to solve for electronic energy levels, often called bands, at that wave vector. The resulting states contribute to calculated observables according to their occupations and the point’s integration weight. At zero temperature, states are occupied up to the Fermi level; metallic calculations often need special treatment because occupations change near that level.
Crystal symmetry can make many points equivalent. Instead of calculating every point in a regular grid, software commonly reduces the set to points in the irreducible part of the Brillouin zone and assigns weights that account for the omitted symmetry-related points. The number shown by a program may therefore refer to the full grid or to the symmetry-reduced set; check the program’s output before comparing counts.
A calculation at the Gamma point uses only the point at the center of the Brillouin zone, conventionally written Γ. It is not automatically an adequate substitute for a denser grid. Whether it works depends on the cell size, material, and property being calculated.
Meshes, spacing, and supercells
A common choice is a regular Monkhorst–Pack-type mesh, specified by three integers such as 6 × 6 × 6. These values describe subdivisions along reciprocal-lattice directions, not a universal number of points needed for every material. The grid may be shifted relative to Γ, depending on the chosen convention and settings.
For a roughly cubic primitive cell, a denser, more evenly distributed mesh is usually needed than for a large real-space supercell. Increasing a cell dimension shrinks the corresponding reciprocal-space dimension, so fewer points may be needed along that direction. For example, a slab model with a large vacuum region perpendicular to its surface is often sampled densely in the two in-plane directions and sparsely—or at one point—in the vacuum direction. This is a geometry-based choice, not a rule that every slab must use the same grid.
Some software lets users set a target spacing between k points rather than the grid dimensions directly. That can make sampling more consistent when comparing cells of different sizes, but the spacing convention and units are software-dependent.
How to choose and test a k-point grid
There is no single grid that guarantees accuracy for every calculation. A practical approach is to test convergence for the specific quantity that matters:
- Start with the cell and symmetry. Identify the primitive cell or supercell, its reciprocal directions, and any symmetry the calculation preserves.
- Choose a plausible initial mesh. Use the material and cell dimensions as a guide; anisotropic cells may call for different subdivisions along different directions.
- Increase sampling systematically. Repeat the calculation with denser meshes while keeping other settings fixed, including the structure, basis or plane-wave cutoff, and convergence thresholds.
- Compare the target result. Track the energy per atom, forces, stress, density of states, or other property of interest. Stop only when changes are smaller than the tolerance needed for the study.
- Apply the converged settings consistently. Use comparable sampling when comparing related structures, while accounting for differences in their cells and reciprocal lattices.
For instance, a calculation might compare 4 × 4 × 4, 6 × 6 × 6, and 8 × 8 × 8 meshes for a cubic primitive cell. Those are example test points, not recommended universal settings. A result that appears converged in total energy may still be insufficiently converged for a band structure or a delicate energy difference.
What affects convergence
Metals often require denser sampling than many insulators because electronic states cross the Fermi level. Smearing or other occupation schemes can help numerical integration, but they do not make k-point convergence irrelevant. The chosen smearing width itself may also affect results and should be checked for the intended calculation.
Different properties can converge at different rates. Total energies may change only slightly as a mesh is refined, while forces, stress, magnetic moments, or energy differences between competing phases remain more sensitive. Band structures are a related but distinct task: calculations commonly use a regular mesh for self-consistency and then evaluate bands along selected high-symmetry paths. Those path points are not the same thing as the integration mesh used for total energy.
Convergence also depends on the electronic-structure method, the shape and size of the cell, and the required precision. If the structure changes during geometry optimization, the grid choice should remain suitable for the evolving cell. For comparisons across cells, inconsistent sampling can introduce numerical differences that might be mistaken for physical effects.
Common misconceptions and problems
- “More k points always fix a calculation.” Denser sampling improves one source of numerical accuracy, but it cannot correct an unsuitable structural model, inadequate basis cutoff, poor self-consistency, or other setup errors.
- “The same grid works for every cell.” The same integer mesh can represent very different reciprocal-space spacing in cells of different sizes or shapes.
- “One Gamma point is always enough for a supercell.” Large supercells often need fewer points, but adequacy still requires convergence testing for the property being reported.
- “A k-point count can be compared without context.” Symmetry reduction, grid shifts, and cell dimensions affect what a reported count means.
A costly calculation can sometimes be made more efficient by using symmetry or a suitable supercell, but reducing the number of points without checking convergence risks changing the answer. Report the grid, shift or centering convention when relevant, and the convergence evidence so others can interpret the result.
Frequently Asked Questions
What does a 4 × 4 × 4 k-point grid mean?
It means the reciprocal-space sampling mesh is divided into four subdivisions along each of three reciprocal directions, subject to the program’s grid convention. Symmetry may reduce the number of points actually evaluated.
What is the Gamma point in DFT?
The Gamma point is the center of the first Brillouin zone, where the wave vector is zero. A Gamma-only calculation samples that single point, which can be useful for some large cells but should be validated for the desired result.
How many k points should I use?
There is no universal count. Choose a mesh based on the cell and material, then increase it until the property of interest changes by less than an appropriate, stated tolerance.
Do k points affect band structures?
Yes. They determine which wave vectors are sampled. A band plot typically evaluates energies along a chosen path through reciprocal space, while the self-consistent calculation also needs a suitable integration mesh.
Why can a metal need more k points than an insulator?
Metallic states near the Fermi level can change occupation across small regions of reciprocal space, making Brillouin-zone integration more sensitive to sampling density.
Related
General steps for current mainstream versions; menus can differ by device and version. See the disclaimer or report an error.
More in this section
- Trade-In Value Treadmill Calculator$504.00
- Trade-In Value Calculator: Estimate Your Vehicle’s Value$18,000
- Trade-In Value for a Telstra Phone: How It Works
- Trade-In Value at T-Mobile: How It Works
- Subaru Trade-In Value: Estimate What Your Car Is Worth$20,000
- Trade in Value for a Switch Lite: What to Expect