Skip to content

BMD Compute INCARs

Jump to bottom
Lee Burton edited this page Sep 26, 2026 · 7 revisions

BMD Compute INCARs

Authority. BMD Compute is authoritative for how BMD VASP calculations are generated and executed. BMDex supplies curated supporting data, evidence and tools. Scientific validation and adoption are separate human review.

This page explains BMD Compute's settings; it does not define them. If this page and BMD Compute ever disagree, BMD Compute is correct and this page is out of date.

This page shows the INCAR files BMD Compute generates for its four standard workflows and explains what their stage-level settings mean. It is the third of four pages:

  1. VASP — fundamentals: the physics and core settings.
  2. BMD Compute Workflows: why calculations are sequences of stages and what passes between them.
  3. BMD Compute INCARs (this page): the settings used within each stage.
  4. BMD Compute Advanced Options: optional modifiers such as spin polarisation or DFT+U.

Questions about why a stage exists or comes before another are answered on the Workflows page; this page explains the settings inside each stage.

What these files are

  • They are the complete INCARs BMD Compute generated for its four standard workflows, reproduced exactly, including the header comments that give each stage's number, type and VASP executable.
  • They are reference realizations, not templates to copy. Some inputs depend on the structure and chemistry, for example magnetic moments and ISPIN, DFT+U tags, the POTCARs, and the k-point mesh (which depends on the cell). Other materials and options can therefore produce different files. Always use the inputs BMD Compute generates for your own calculation.
  • The matching KPOINTS files are not reproduced here.
  • The explanations below are the wiki's teaching commentary on what these settings do in VASP. They describe the physics and numerics; they are not a separate BMD policy, and where they say why such a setting is commonly used, that is general VASP practice rather than a stated BMD rationale.

Provenance

  • Reference structure: Si
  • These reference inputs were generated from BMD Compute on 26 September 2026. The exact BMD Compute version and dependency versions (for example pymatgen and atomate2) were not retained with this snapshot.

Reference INCARs by workflow

Static Energy

One stage: a PBE static calculation.

# Stage 1 - Static Energy (PBE)
# VASP executable - vasp_std
ADDGRID = True
ALGO = Normal
EDIFF = 1e-06
ENCUT = 620.0
ISMEAR = -5
ISPIN = 1
LAECHG = True
LASPH = True
LCHARG = True
LELF = True
LORBIT = 11
LREAL = False
LVTOT = True
LWAVE = False
NCORE = 8
NEDOS = 4001
NELM = 200
NSW = 0
PREC = Accurate
SIGMA = 0.05

Relaxed Structure

Two consecutive PBE geometry optimisations with identical settings. Stage 2 starts from the structure produced by Stage 1.

# Stage 1 - Geometry Optimisation (PBE)
# VASP executable - vasp_std
ADDGRID = True
ALGO = Fast
EDIFF = 1e-06
EDIFFG = -0.01
ENCUT = 580.0
IBRION = 2
ISIF = 3
ISMEAR = 0
ISPIN = 1
LASPH = True
LCHARG = False
LREAL = False
LWAVE = False
NCORE = 8
NELM = 200
NSW = 99
PREC = Accurate
SIGMA = 0.2

# Stage 2 - Geometry Optimisation (PBE)
# VASP executable - vasp_std
ADDGRID = True
ALGO = Fast
EDIFF = 1e-06
EDIFFG = -0.01
ENCUT = 580.0
IBRION = 2
ISIF = 3
ISMEAR = 0
ISPIN = 1
LASPH = True
LCHARG = False
LREAL = False
LWAVE = False
NCORE = 8
NELM = 200
NSW = 99
PREC = Accurate
SIGMA = 0.2

Density of States

PBE geometry optimisation, then an HSE06 static calculation, then an HSE06 density of states.

# Stage 1 - Geometry Optimisation (PBE)
# VASP executable - vasp_std
ADDGRID = True
ALGO = Fast
EDIFF = 1e-06
EDIFFG = -0.01
ENCUT = 580.0
IBRION = 2
ISIF = 3
ISMEAR = 0
ISPIN = 1
LASPH = True
LCHARG = False
LREAL = False
LWAVE = False
NCORE = 8
NELM = 200
NSW = 99
PREC = Accurate
SIGMA = 0.2

# Stage 2 - Static Energy (HSE06)
# VASP executable - vasp_std
ADDGRID = True
AEXX = 0.25
ALGO = Damped
EDIFF = 1e-06
ENCUT = 620.0
GGA = Pe
HFSCREEN = 0.2
ISMEAR = 0
ISPIN = 1
LAECHG = True
LASPH = True
LCHARG = True
LELF = True
LHFCALC = True
LORBIT = 11
LREAL = False
LVTOT = True
LWAVE = False
NCORE = 8
NEDOS = 4001
NELM = 200
NSW = 0
PREC = Accurate
PRECFOCK = Accurate
SIGMA = 0.05
TIME = 0.4

# Stage 3 - Density of States (HSE06)
# VASP executable - vasp_std
ADDGRID = True
AEXX = 0.25
ALGO = Normal
EDIFF = 1e-06
ENCUT = 620.0
GGA = Pe
HFSCREEN = 0.2
ISMEAR = -5
ISPIN = 1
LAECHG = True
LASPH = True
LCHARG = False
LELF = False
LHFCALC = True
LORBIT = 11
LREAL = False
LVTOT = True
LWAVE = False
NCORE = 8
NEDOS = 4001
NELM = 200
NELMIN = 5
NSW = 0
PREC = Accurate
PRECFOCK = Fast
SIGMA = 0.2
TIME = 0.4

Band Structure

PBE geometry optimisation, then an HSE06 static calculation, then an HSE06 band structure.

# Stage 1 - Geometry Optimisation (PBE)
# VASP executable - vasp_std
ADDGRID = True
ALGO = Fast
EDIFF = 1e-06
EDIFFG = -0.01
ENCUT = 580.0
IBRION = 2
ISIF = 3
ISMEAR = 0
ISPIN = 1
LASPH = True
LCHARG = False
LREAL = False
LWAVE = False
NCORE = 8
NELM = 200
NSW = 99
PREC = Accurate
SIGMA = 0.2

# Stage 2 - Static Energy (HSE06)
# VASP executable - vasp_std
ADDGRID = True
AEXX = 0.25
ALGO = Damped
EDIFF = 1e-06
ENCUT = 620.0
GGA = Pe
HFSCREEN = 0.2
ISMEAR = 0
ISPIN = 1
LAECHG = True
LASPH = True
LCHARG = True
LELF = True
LHFCALC = True
LORBIT = 11
LREAL = False
LVTOT = True
LWAVE = False
NCORE = 8
NEDOS = 4001
NELM = 200
NSW = 0
PREC = Accurate
PRECFOCK = Accurate
SIGMA = 0.05
TIME = 0.4

# Stage 3 - Band Structure (HSE06)
# VASP executable - vasp_std
ADDGRID = True
AEXX = 0.25
ALGO = Normal
EDIFF = 1e-06
ENCUT = 620.0
GGA = Pe
HFSCREEN = 0.2
ISMEAR = 0
ISPIN = 1
LAECHG = True
LASPH = True
LCHARG = False
LELF = False
LHFCALC = True
LORBIT = 11
LREAL = False
LVTOT = True
LWAVE = False
NELM = 200
NELMIN = 5
NSW = 0
PREC = Accurate
PRECFOCK = Fast
SIGMA = 0.01
TIME = 0.4

Five distinct stage types

Several stages above have identical settings. There are five distinct stage types:

Stage type Appears in Notes
PBE Geometry Optimisation Relaxed Structure (Stages 1 and 2); Density of States and Band Structure (Stage 1) Identical settings in all four places.
PBE Static Energy Static Energy (Stage 1)
HSE06 Static Energy Density of States and Band Structure (Stage 2) Identical settings in both workflows.
HSE06 Density of States Density of States (Stage 3)
HSE06 Band Structure Band Structure (Stage 3)

The settings by role

It helps to sort the tags into five kinds of decision.

Role Tags What they decide
Physical method LHFCALC, AEXX, HFSCREEN, GGA, ISPIN Which physics is being approximated: the exchange-correlation functional and whether spin is treated explicitly.
Numerical accuracy and convergence ENCUT, PREC, ADDGRID, LASPH, LREAL, PRECFOCK, ISMEAR, SIGMA, EDIFF, EDIFFG, NELM, NELMIN, ALGO, TIME How accurately and robustly that physics is solved.
Calculation type IBRION, ISIF, NSW Whether atoms and the cell move.
Outputs LCHARG, LWAVE, LVTOT, LAECHG, LELF, LORBIT, NEDOS Which files VASP writes for later analysis.
Performance NCORE How the work is split across processors; it should not change the physics.

Settings shared by all five stage types

  • PREC = Accurate, ADDGRID = True — fine FFT grids, plus an extra fine grid for the augmentation charges. This reduces numerical noise, especially in forces and stresses.
  • LASPH = True — includes non-spherical contributions to the gradient corrections inside the PAW spheres. This matters for GGA and hybrid accuracy, particularly for d- and f-elements.
  • LREAL = False — evaluates projection operators in reciprocal space. This is exact, and efficient for small cells.
  • EDIFF = 1e-06 — each electronic self-consistency loop stops when the energy changes by less than 10⁻⁶ eV. NELM = 200 allows up to 200 electronic steps before VASP gives up.
  • ISPIN = 1 — no spin polarisation. Spin-polarised calculations are an advanced option.
  • LWAVE = False — no stage writes a WAVECAR (the orbitals), which is a large file.

Comparison 1: PBE Geometry Optimisation vs PBE Static Energy

These are two different kinds of stage: one moves the atoms and cell, the other computes the electronic ground state of a fixed structure. They belong to different workflows (see BMD Compute Workflows), but comparing them shows how settings for following forces and stresses differ from settings for a fixed-structure calculation.

Tag Geometry Optimisation Static Energy What the difference does
IBRION / ISIF / NSW 2 / 3 / 99 — / — / 0 The relaxation moves atoms with conjugate gradients (IBRION = 2) and relaxes positions, cell shape and volume (ISIF = 3) for up to 99 ionic steps. The static run is a single point (NSW = 0).
EDIFFG -0.01 — Negative means a force criterion: stop when all forces are below 0.01 eV/Å. It is only meaningful when ions move.
ENCUT 580.0 620.0 The static stage uses a larger plane-wave basis (higher cutoff). How the cutoff interacts with variable-cell relaxation (Pulay stress) is explained on BMD Compute Workflows.
ALGO Fast Normal Fast (Davidson then RMM-DIIS) is quicker for the many electronic loops of a relaxation; Normal (blocked Davidson) is more robust for a fixed-structure calculation.
ISMEAR / SIGMA 0 / 0.2 -5 / 0.05 See below.
Output tags LCHARG = False only LCHARG, LVTOT, LAECHG, LELF = True; LORBIT = 11; NEDOS = 4001 Intermediate relaxation steps are not written out. The static run writes the charge density (CHGCAR), local potential (LOCPOT), all-electron core and valence densities (AECCAR0/AECCAR2, used for Bader analysis), and the electron localisation function (ELFCAR). LORBIT = 11 adds site- and orbital-projected information to the output, and NEDOS = 4001 sets the number of energy points on which any density of states from this run is written. This is output from the static run itself; it is not a dedicated density-of-states calculation.

Smearing. ISMEAR = -5 is the tetrahedron method with Blöchl corrections. It gives accurate total energies and densities of states on a uniform k-point mesh, and SIGMA is ignored. It is not suitable for relaxing metals because the forces are unreliable; the relaxation uses Gaussian smearing (ISMEAR = 0) instead. A SIGMA of 0.2 eV is fairly broad for a semiconductor, and it can put a small fractional occupation across the gap. You can check its effect in OUTCAR: the difference between energy without entropy and free energy TOTEN should stay small.

Comparison 2: PBE vs HSE06 Static Energy

The PBE and HSE06 Static Energy stages differ in eight tags. Everything else, including ENCUT = 620.0, the convergence criteria and all the output tags, is the same.

Tag PBE Static HSE06 Static What the difference does
LHFCALC — True Switches on the hybrid functional: a fraction of exact (Hartree–Fock-like) exchange.
AEXX — 0.25 The exact-exchange fraction, 25%.
HFSCREEN — 0.2 Screens the exact exchange at long range (in Å⁻¹); 0.2 is the value that defines HSE06.
GGA — Pe The semilocal part is PBE. (pymatgen writes Pe; VASP reads it as PE.)
PRECFOCK — Accurate The FFT grid used for the exact-exchange operator. Accurate is the finer, more expensive choice.
ALGO Normal Damped See below.
TIME — 0.4 The trial time step for the damped minimiser; it is used here because ALGO = Damped.
ISMEAR -5 0 Gaussian smearing in the HSE06 static stage, with SIGMA = 0.05 (the same SIGMA value as the PBE static stage, where it is ignored by the tetrahedron method). A width of 0.05 eV keeps occupations close to integer for a gapped material.

Physical method. The first four tags change the physics: they replace part of PBE exchange by screened exact exchange. Semilocal PBE commonly underestimates the band gaps of semiconductors; HSE06 often improves them, at substantially greater computational cost.

ALGO = Damped. Instead of iteratively diagonalising the Hamiltonian and mixing densities, the damped algorithm minimises the total energy directly with respect to the orbitals, using a damped equation of motion whose step is set by TIME. This is a commonly used, robust choice for hybrid-functional self-consistency, where the orbital-dependent exchange operator can make density-mixing schemes converge poorly.

Outputs. The HSE06 static stage writes the same outputs as the PBE static stage, including CHGCAR (LCHARG = True), but not WAVECAR (LWAVE = False).

Comparison 3: HSE06 Static Energy vs Density of States vs Band Structure

These are Stages 2 and 3 of the two electronic-structure workflows. The table lists every tag that differs between them, plus TIME; all others are identical.

Tag HSE06 Static HSE06 DOS HSE06 Band What the difference does
ALGO Damped Normal Normal The final stages use the blocked-Davidson solver instead of the damped minimiser.
TIME 0.4 0.4 0.4 The same value throughout, but only operative with ALGO = Damped, that is, in the static stage. It is retained in the DOS and band INCARs but has no effect there.
PRECFOCK Accurate Fast Fast The final stages use the coarser, cheaper FFT grid for exact exchange.
NELMIN — 5 5 At least five electronic steps before VASP may declare convergence.
ISMEAR 0 -5 0 The DOS stage uses the tetrahedron method, which needs a uniform k-point mesh. A band structure samples a path through the Brillouin zone, so it uses Gaussian smearing.
SIGMA 0.05 0.2 0.01 Ignored by the tetrahedron method in the DOS stage. In the band stage a very small width keeps occupations essentially integer.
LCHARG / LELF True / True False / False False / False Only the static stage writes CHGCAR and ELFCAR.
NEDOS 4001 4001 not set The fine energy grid is needed to write a density of states; the band stage does not set it.
NCORE 8 8 not set A performance setting only.

All three are static (NSW = 0) and self-consistent, and they share the same functional, cutoff and convergence settings. Why the workflows have a separate HSE06 static stage, and why the final stage differs for a density of states and for a band structure En(k), is explained on BMD Compute Workflows.

Related pages

Clone this wiki locally