Repository navigation
BMD Compute Advanced Options
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.
BMD Compute's advanced options (modifiers) change a base calculation for physics that the base settings leave out. This page explains the concepts behind each option and when it matters. Which stages and levels of theory accept each option is shown by the BMD Compute interface, which is authoritative.
In Custom workflows you choose these options yourself. In the standard Desired Output workflows, BMD Compute applies spin polarisation, the van der Waals correction, spin–orbit coupling and DFT+U automatically when the structure triggers them; each section below says when, and to which stages. These triggers are composition or connectivity screens, not proof that a treatment is physically required for a given material.
This is the last of four pages, after
VASP — fundamentals,
BMD Compute Workflows and
BMD Compute INCARs.
Each option here will be shown as a change (a delta) from the base INCAR
files on the INCARs page.
Reference deltas are not published yet. For each option, the exact
INCARchanges will be added here only once they have been generated by BMD Compute. Until then, use the inputs BMD Compute generates for your own calculation, and do not copy settings from other wiki pages.
For each option, this page will show:
- Concept — the physics or numerical issue it addresses.
- When it matters — the kinds of materials or questions where you need it.
-
Reference delta — the lines BMD Compute adds (
+) or removes (-) compared with the baseINCAR. Pending.
Concept. The base reference calculations use ISPIN = 1, which treats spin-up and
spin-down electrons identically. A spin-polarised calculation lets the two spin
channels differ, so the calculation can describe magnetic moments. It needs
starting magnetic moments for each atom, and the result can depend on them.
When it matters. Materials that may be magnetic: many compounds containing transition metals or rare earths, open-shell defects and radicals. It roughly doubles the electronic work, so it is not used for clearly non-magnetic materials like Si.
In BMD Compute. In the Desired Output workflows, spin polarisation is switched on automatically, for every stage, when the structure contains Ti, V, Cr, Mn, Fe, Co, Ni, Mo, Tc, Ru, Rh, Re, Os, Ir, a lanthanide from Ce to Yb, or U, Np, Pu, Am, Cm, Bk or Cf. The starting moments are the pymatgen/Materials Project defaults (for example 5 μB on Fe, Mn and Ni). In a Custom workflow you choose it yourself.
Reference delta. Pending.
Status. BMD Compute adopted automatic DFT+U (version 1) for its standard Desired Output workflows in September 2026. This section explains that behaviour; BMD Compute remains authoritative for what is actually executed. The Hubbard U values themselves come from the Materials Project, not from BMD or this wiki.
Semilocal functionals such as PBE contain a self-interaction error: the approximate exchange–correlation energy does not fully cancel each electron's spurious interaction with itself. Its consequences are usually largest for strongly localised d and f states, which semilocal functionals tend to over-delocalise. In some transition-metal oxides and fluorides this can contribute to band gaps that are too small (or absent), magnetic moments that are too small, and systematic errors in oxidation and reaction energies. How much it matters depends on the material and on the property being calculated.
DFT+U is an approximate correction commonly used to mitigate these deficiencies for localised states. It adds a Hubbard-like, on-site term for a chosen set of orbitals (for example the 3d orbitals of Ni) (Anisimov, Zaanen and Andersen, 1991). It is one practical remedy among several (hybrid functionals are another), not an exact fix. In the simplified, rotationally invariant form of Dudarev et al. (1998), which is the form the Materials Project uses, the correction to the total energy is
EDFT+U = EDFT + ((U − J)/2) Σσ Tr[nσ − nσnσ],
where nσ is the occupation matrix of the corrected orbitals for spin σ. The trace vanishes when every orbital is either full or empty, and is largest for half-filled occupations, so the term penalises fractional occupation and pushes the localised states towards integer filling. Only the difference U − J enters, often called Ueff.
U is not a uniquely determined property of a material. Its value depends on how it is obtained (for example fitted to experimental reaction energies, fitted to band gaps, or computed from linear response), on the functional, and on how the corrected orbitals are defined (in VASP, by the PAW projectors). A +U calculation is therefore a choice of method, and:
- energies calculated with different U values, or with and without +U, are not directly comparable;
- a U value should always be reported together with the source and convention it follows.
In the standard Desired Output workflows, BMD Compute decides automatically whether to use +U, following the Materials Project GGA+U scheme as implemented in pymatgen, plus one conservative BMD screen. This is a policy for choosing a consistent method in standard workflows. It is not a claim that +U is physically required for a given material, or that a calculation without +U is wrong.
1. The Materials Project/pymatgen rule. +U is triggered when O or F is the most electronegative element in the structure and the structure contains one or more of Co, Cr, Fe, Mn, Mo, Ni, V or W. Those metals then receive the Materials Project U values, applied to their d states with J = 0, in the Dudarev form. The values are the Materials Project's oxide values, which the Materials Project fitted to experimental binary formation enthalpies following the approach of Wang, Maxisch and Ceder (2006); fluorides use the same values. See the Materials Project documentation on Hubbard U values for the table. BMD Compute reads these values unchanged from pymatgen; it does not define its own.
2. The BMD d0 screen. A metal ion is d0 when its formal oxidation state leaves it with no d electrons, for example W6+ in WO3. The usual motivation for applying +U to those d states, a partially filled and localised d shell, is then formally absent. BMD Compute therefore suppresses automatic +U when pymatgen's charge-balanced oxidation-state guesses consistently place every triggering metal at d0: that is, in every guess, not just the most likely one. The screen can only switch +U off; it never changes the U values, and it is decided for the compound as a whole (it never removes +U from one element while keeping it on another).
If pymatgen cannot produce any charge-balanced oxidation-state guess for a composition, the d0 screen cannot be evaluated. BMD Compute then keeps the underlying Materials Project/pymatgen +U decision rather than silently switching +U off.
Examples.
| Material | Outcome | Reason |
|---|---|---|
| NiO, Fe2O3, MnO, CoO, Cr2O3 | Automatic +U applied | Oxides of rule metals with partially filled d shells (Ni2+ d8, Fe3+ d5, Mn2+ d5, Co2+ d7, Cr3+ d3). |
| WO3, MoO3, V2O5 | Suppressed by the d0 screen | The rule triggers, but W6+, Mo6+ and V5+ are d0. |
| TiO2, ZnO, Cu2O | No automatic +U | Ti, Zn and Cu have no Materials Project +U value. |
| FeS2 and other sulfides | No automatic +U | S, not O or F, is the most electronegative element, so the rule does not trigger. |
Worked example: NiO. Automatic +U applies Ni U = 6.2 eV with J = 0 eV to the Ni d states (so Ueff = 6.2 eV). O receives no correction.
Automatic +U is applied only to eligible PBE stages: PBE Geometry Optimisation and PBE Static Energy. HSE06 stages never receive +U.
| Desired Output | Stages with +U (when triggered) | Stages without +U |
|---|---|---|
| Static Energy | The PBE Static Energy stage (a PBE+U static calculation) | — |
| Relaxed Structure | Both PBE Geometry Optimisation stages | — |
| Density of States | The PBE Geometry Optimisation | HSE06 Static Energy and HSE06 Density of States |
| Band Structure | The PBE Geometry Optimisation | HSE06 Static Energy and HSE06 Band Structure |
In the Density of States and Band Structure workflows, the geometry therefore comes from a PBE+U relaxation, and the electronic structure from HSE06 without +U. When you report such results, state both the treatment used for the geometry (PBE+U) and that used for the electronic structure (HSE06).
Explanatory context only, not BMD policy rationale. BMD Compute's policy is simply that HSE06 stages do not receive +U; BMD has not stated a formal rationale for this on this page. As general background: the exact exchange in a hybrid functional already partly reduces the self-interaction error that +U is commonly used to address, and the Materials Project U values were fitted for GGA, not for hybrid functionals.
+U does not itself switch on spin polarisation, spin–orbit coupling or a dispersion correction. BMD Compute decides each of these separately, and they can be combined with +U on PBE stages where BMD Compute supports it. In practice the effect of +U on magnetic transition-metal compounds is closely tied to spin polarisation and magnetic order, so check which treatments your calculation actually uses in its generated inputs.
In a Custom workflow, +U is your choice. BMD Compute never adds or removes it automatically, although it may suggest activating the DFT+U option for a material where the automatic rule would apply.
- DFT+U can be selected on PBE stages; it is not available for HSE06 stages.
- The d0 screen does not remove +U that you select yourself.
- If no Materials Project U value applies to the structure, BMD Compute rejects the DFT+U selection rather than running without it.
- A PBE Density of States or Band Structure stage that restarts from the previous stage's fixed charge density must use the same +U setting as that stage.
BMD Compute uses the same Materials Project/pymatgen U values and convention so that its +U calculations are comparable in method. This does not make BMD total energies strictly compatible with Materials Project energies. The complete calculation methodology differs, including POTCAR choices, plane-wave cutoffs and other input settings, and the Materials Project applies its own energy corrections when it mixes GGA and GGA+U results (see Anion and GGA/GGA+U mixing). Do not combine BMD and Materials Project total energies directly, for example in the same phase diagram.
Reference delta. Pending. The exact INCAR changes BMD Compute generates
for a +U stage will be added here once a Compute-generated reference is
available.
- V. I. Anisimov, J. Zaanen and O. K. Andersen, "Band theory and Mott insulators: Hubbard U instead of Stoner I", Phys. Rev. B 44, 943 (1991).
- S. L. Dudarev, G. A. Botton, S. Y. Savrasov, C. J. Humphreys and A. P. Sutton, "Electron-energy-loss spectra and the structural stability of nickel oxide: An LSDA+U study", Phys. Rev. B 57, 1505 (1998).
- L. Wang, T. Maxisch and G. Ceder, "Oxidation energies of transition metal oxides within the GGA+U framework", Phys. Rev. B 73, 195107 (2006).
- A. Jain, G. Hautier, S. P. Ong, C. J. Moore, C. C. Fischer, K. A. Persson and G. Ceder, "Formation enthalpies by mixing GGA and GGA+U calculations", Phys. Rev. B 84, 045115 (2011).
- Materials Project documentation: Hubbard U values.
Concept. Semilocal functionals miss most of the long-range dispersion (van der Waals) attraction between atoms that are not chemically bonded. Grimme's DFT-D3 family of corrections adds this as an extra, pairwise energy term; D3(BJ) is the variant with Becke–Johnson damping at short range.
When it matters. Layered materials (for example graphite or transition-metal dichalcogenides), molecular crystals, and molecules adsorbed on surfaces, where the interlayer or intermolecular spacing is set by dispersion. Without it, relaxed interlayer distances can be far too large.
In BMD Compute. In the Desired Output workflows, BMD Compute adds the
DFT-D3(BJ) correction (IVDW = 12) automatically when its structure analysis
(pymatgen CrystalNN bonding with Larsen dimensionality) finds two-dimensional
bonded connectivity, that is, a layered structure. It is applied to the PBE
Geometry Optimisation and PBE Static Energy stages; HSE06 stages do not
receive it automatically. If the dimensionality analysis itself fails, BMD
Compute cannot decide whether the correction is needed, so a Desired Output
stops with an error before its inputs are previewed or prepared. Custom
workflows are not blocked by this; there you choose the correction yourself.
Reference delta. Pending.
Concept. Sampling the Brillouin zone at the single Γ point instead of a mesh. For very large cells (big supercells, amorphous structures, molecules in a box) the Brillouin zone is small and Γ alone can be adequate.
When it matters. Large systems where a full mesh would be unnecessary and expensive. It is not suitable for small cells such as bulk Si, and it cannot be used for a band structure, which needs a path of k-points.
Reference delta. Pending.
Concept. A geometry optimisation that moves the atoms but keeps the cell
shape and volume fixed. In VASP this corresponds to relaxing ionic positions
only, instead of positions, shape and volume as in the base Geometry
Optimisation (ISIF = 3).
When it matters. When the cell must stay fixed, for example a defect or surface calculation in a supercell whose lattice parameters were already determined.
Reference delta. Pending.
Status. BMD Compute v1.0.0 applies SOC automatically in its standard Desired Output workflows. This section explains that behaviour; BMD Compute remains authoritative for what is actually executed.
Concept. Spin–orbit coupling links an electron's spin to its orbital motion. It is strongest for heavy elements and can split and shift bands noticeably. Including it requires a non-collinear calculation, which is considerably more expensive.
When it matters. Compounds of heavy elements, particularly for band structures and densities of states, where SOC can lift degeneracies and shift band edges.
What BMD Compute does automatically. In the Desired Output workflows, SOC is switched on when the structure contains a 4d (Y–Cd) or 5d (Hf–Hg) transition metal, a lanthanide (La–Lu), an actinide (Ac–Lr), or Tl, Pb, Bi or Po. It is then applied to every stage except Geometry Optimisation: the structure is relaxed without SOC, and the static and electronic-structure stages that follow include it. When you report such results, say so. As with the other automatic treatments, the trigger is a composition screen, not proof that SOC changes a given result appreciably.
SOC stages run the non-collinear VASP executable (vasp_ncl) with
LSORBIT = True. Symmetry is switched off (ISYM = 0), the spin quantisation
axis is along z (SAXIS = 0 0 1), and the starting magnetic moments are
vectors along that axis (zero unless the structure also triggers spin
polarisation). SOC can be combined with the other treatments on the same
stage; each is decided separately.
In Custom workflows. SOC is your choice, on the stages where the BMD Compute interface offers it; BMD Compute never adds it automatically. A Density of States or Band Structure stage must use the same SOC setting as the stage it follows.
Reference delta. Pending.