DFT functionals · ORCA User Guide

On this page

A hierarchy of functionals: Jacob's ladder
Dispersion corrections are nearly always needed
Recommendations by task
Composite methods (the 3c series)
Double hybrids
Recommendations by system
Functionals to avoid

A hierarchy of functionals: Jacob's ladder

DFT functionals are often organised along Perdew's "Jacob's ladder". Each rung adds more ingredients and (typically) more accuracy at higher cost.

Jacob's ladder of DFT functionals: from LDA to double hybrids

Rung	Name	Depends on	Representative
1	LDA	density ρ only	SVWN (almost never used)
2	GGA	ρ + ∇ρ	BP86, BLYP, PBE
3	meta-GGA	+ ∇²ρ or τ	TPSS, M06-L, r²SCAN
4	Hybrid	+ some HF exchange	B3LYP, PBE0, M06-2X
5	Double hybrid	+ MP2 correlation	B2PLYP, DSD-PBEP86

Higher rungs are typically more accurate, but not always. A well-built meta-GGA (e.g. r²SCAN) can outperform a mediocre hybrid. The latest broad benchmarks (GMTKN55, MGCDB84, …) tend to rank these two functionals at the very top overall:

ωB97M-V: meta-GGA + range-separated hybrid + VV10 non-local correlation. Overall #1.
ωB97X-V: similar idea but GGA-based. Slightly faster.

Both are built into ORCA: ! wB97M-V def2-TZVP on one line is enough.

Dispersion corrections are nearly always needed

Pure GGAs and standard hybrids do not describe London dispersion. Without a correction, π–π stacking, hydrogen bonds, vdW complexes, and conformer energetics come out badly. Use one of Grimme's D3 / D4 corrections by default.

Keyword	Description
`D4`	Newest. Recommended in ORCA 6. Fit for more functionals.
`D3BJ`	Widely validated. Fallback if D4 is unavailable for your functional.
`D3ZERO`	Alternative D3 damping. Less accurate than D3BJ.
(none)	ωB97M-V and ωB97X-V already include VV10, so no extra correction is needed.

# Standard usage with dispersion correction
! B3LYP D4 def2-TZVP RIJCOSX def2/J
! PBE0 D4 def2-TZVP RIJCOSX def2/J
! r2SCAN D4 def2-TZVP def2/J      # meta-GGA doesn't need RIJCOSX

# VV10 already included — no extra correction
! wB97M-V def2-TZVP RIJCOSX def2/J

Recommendations by task

Task	Recommended functional	Notes
Geometry optimisation (general organic)	`r²SCAN-3c`, `B97-3c`	Composite. Basis included.
Geometry optimisation (TM complexes)	`TPSS D4`, `BP86 D4`	Pure GGAs are more forgiving of poor initial geometries.
Reaction barriers	`M06-2X D4`, `ωB97X-V`	Higher HF percentage gives better barriers.
Non-covalent interactions	`ωB97M-V`, `DSD-PBEP86 D4`	Double hybrids are the most accurate.
NMR chemical shifts	`PBE0 D4`, `ωB97X D4`	Use dedicated NMR bases (pcSseg).
UV-Vis absorption (local excitations)	`B3LYP D4`, `PBE0 D4`	Standard hybrids are usually enough.
UV-Vis absorption (charge transfer)	`CAM-B3LYP`, `ωB97X-V`	Range-separation is mandatory.
EPR (g-tensor, A-tensor)	`PBE0`, `B3LYP`	~30 % HF tends to fit well.

Composite methods (the 3c series)

The Grimme group's "3c" methods bundle "functional + basis + D3/D4 + gCP correction" into a single keyword. The basis is fixed, the usage is therefore extremely simple, and the accuracy is high for the cost.

Keyword	What it really is	Best for
`HF-3c`	HF + minix (small DZ) + 3 corrections	Quick screening of very large molecules
`PBEh-3c`	PBEh hybrid + def2-mSVP	Medium-sized opt + frequencies
`B97-3c`	Modified B97 GGA + def2-mTZVP	Medium-to-large molecules, fast optimisation
`r²SCAN-3c`	r²SCAN meta-GGA + def2-mTZVPP	Good accuracy for the cost. Geometries and free energies alike.
`ωB97X-3c`	Range-separated hybrid + valence DZ	Organic / excited states

# r²SCAN-3c — a strong first choice for everyday work
! r2SCAN-3c Opt Freq

* xyzfile 0 1 mol.xyz

Starting with r²SCAN-3c

r²SCAN-3c often reaches B3LYP-D4/def2-TZVP accuracy in under half the time. Using r²SCAN-3c as the first attempt, and refining with more expensive methods only when needed, is an efficient workflow.

Double hybrids

Double hybrids include both HF exchange and MP2 correlation. They are one rung above the conventional hybrids in accuracy, but they inherit MP2's N⁵ scaling. Best suited to precision single-point energies.

Keyword	Notes
`B2PLYP`	The original double hybrid.
`DSD-PBEP86`	SCS-MP2-coupled. Excellent overall performance.
`DSD-BLYP`	Strong for non-covalent interactions.
`wB97X-2`	Range-separated + double hybrid.
`PWPB95`	High accuracy, cost-efficient.

# DLPNO-accelerated double hybrid — much cheaper
! DLPNO-B2PLYP D4 def2-TZVPP def2-TZVPP/C TightSCF

* xyzfile 0 1 mol.xyz

Recommendations by system

Organic chemistry

Geometry optimisation: r²SCAN-3c or B97-3c.
Precision single point: ωB97M-V/def2-TZVPP or DLPNO-CCSD(T)/cc-pVTZ.
Non-covalent interactions: ωB97M-V is the overall best.

Transition-metal chemistry

First try: TPSSh D4 or TPSS D4 (3d metals).
Spin-state energetics: meta-GGAs (TPSS, M06-L) are less sensitive to the HF percentage.
Spectroscopy (g-tensor): PBE0 D4.

Very large systems (protein active sites, MOFs)

Start with r²SCAN-3c or HF-3c.
For QM/MM embeddings PBEh-3c tends to be robust.

Excited states (TD-DFT)

Local π → π*: B3LYP, PBE0.
Charge transfer: CAM-B3LYP, ωB97X-V are essential.
Rydberg states: range-separated + diffuse functions.

Functionals to avoid

For historical reasons these functionals are still common, but they are not recommended for new work today:

Functional	Replacement
`SVWN` / `LSDA`	Move up to a GGA or higher.
`B3LYP` without dispersion	Always pair with D4 (or D3BJ).
`BLYP` without dispersion	Same: dispersion is mandatory.
The `D2` correction	Use D4 or D3BJ instead.
VWN3 variants	Little effect today. `B3LYP/G` is for Gaussian compatibility only.

B3LYP is still the "reference"

B3LYP D4 is not the most accurate functional available today, but the overwhelming majority of the literature is written with B3LYP, so it still has value as a comparison baseline. For new work, reporting modern functionals (ωB97M-V, r²SCAN-3c, DSD-PBEP86) alongside B3LYP gives both literature consistency and higher accuracy at once.