GFN-xTB

TeraChem implements Grimme's GFN-xTB family of extended semiempirical tight-binding methods through its native, GPU-accelerated SQMBox engine. These methods approximate the electronic structure with a minimal valence basis and parameterized integrals, making them several orders of magnitude faster than ab initio DFT while still giving reasonable geometries, frequencies, and non-covalent interactions. They are well suited to large systems, conformer searches, pre-optimization, and long ab initio molecular-dynamics trajectories.

Two members of the family are available:

`method`	Method	Notes
`gfnxtb`	GFN1-xTB¹	First-generation GFN extended tight binding
`gfn2xtb`	GFN2-xTB²	Adds anisotropic electrostatics (AES), density-dependent exchange–correlation multipoles, and the three-body dispersion term

Both turn on the appropriate Grimme dispersion correction automatically (GFN2 additionally enables the Axilrod–Teller–Muto three-body term).

Quick start

gfn2xtb.in

method         gfn2xtb
basis          gfn2xtb        # use the matching parameter basis
coordinates    coord.xyz
sphericalbasis yes
guess          hcore
charge         0
spinmult       1
run            energy

Required settings

GFN-xTB needs a few settings that differ from an ab initio run:

basis must be set to the matching parameter set — gfnxtb for GFN1-xTB and gfn2xtb for GFN2-xTB. (Selecting the method also sets this basis internally; specifying it explicitly keeps the input self-documenting.)
sphericalbasis yes — the xTB parameter basis is defined in terms of spherical-harmonic functions.
guess hcore — the initial guess is the extended-Hückel / core Hamiltonian. TeraChem switches to this guess automatically for GFN-xTB unless a guess is read from a file.

Restricted vs. unrestricted

As with the ab initio methods, an r/u prefix on the method name forces a restricted or unrestricted treatment:

`method`	Reference
`gfnxtb` / `gfn2xtb`	Default (restricted for closed-shell systems)
`rgfnxtb` / `rgfn2xtb`	Restricted
`ugfnxtb` / `ugfn2xtb`	Unrestricted

Example 1: GFN2-xTB gradient

A GFN2-xTB gradient on ferrocene — a transition-metal complex that is inexpensive at the xTB level:

gfn2xtb_ferrocene_grad.in

method         gfn2xtb
basis          gfn2xtb
coordinates    coord.xyz
sphericalbasis yes
guess          hcore
charge         0
spinmult       1
maxit          100
threall        1e-14
convthre       1.0e-4
run            gradient

Example 2: GFN1-xTB, unrestricted

An unrestricted GFN1-xTB gradient on an indole–benzene complex, showing a few common SCF and printing controls:

ugfnxtb.in

method         ugfnxtb
basis          gfnxtb
coordinates    coord.xyz
sphericalbasis yes
guess          hcore
charge         0
spinmult       1
maxit          100
threall        1e-14
convthre       1.0e-6
scf            diis
diismaxvecs    10
fock           incremental
precision      mixed
sqmprint       large
run            gradient

Run types

GFN-xTB supports the usual single-reference run modes:

`run` value	What it does
`energy`	Single-point xTB energy
`gradient`	Analytic nuclear gradient (enables geometry optimization and Born–Oppenheimer MD)
`minimize`	Geometry optimization on the xTB surface
`md`	Born–Oppenheimer molecular dynamics

In addition, an xTB reference can serve as the integral backend for higher-level methods — a distinctive TeraChem feature in which the ab initio machinery runs on semiempirical integrals (for example, semiempirical CASCI, CASSCF, or RPA). See the CASSCF and Excited States sections for those methods; combine them with method gfnxtb/gfn2xtb to run them at the xTB level.

Tuning keywords

The defaults reproduce the published GFN-xTB parameterizations and rarely need changing. The following knobs are available for testing and method development:

Keyword	Applies to	Description
`xtb_no3rdorder`	GFN1, GFN2	Disable the third-order (charge-dependent) term
`xtb_noaes`	GFN2	Disable anisotropic electrostatics (and its coordination-number damping)
`xtb_noaxc`	GFN2	Disable the multipole exchange–correlation term
`atomic_k_scal`	GFN1, GFN2	Scaling factor applied to the atomic \(K\) (Hückel) parameters
`sqmprint`	all SQM	Print verbosity: `large` (default), `small`, or `no`/`false`
`sqm_parameterfile`	all SQM	Read parameters from a custom file instead of the built-in set
`sqm_hardness_average`	all SQM	Pairwise averaging of atomic hardness: `arithmetic`, `geometric`, or `harmonic`
`sqm_gamma_smoothing`	all SQM	Coulomb interpolation/smoothing scheme: `mataga` or `klopman`

Hybrid (exact-exchange) SQM

By default GFN-xTB uses no Hartree–Fock-like exchange. A "hybrid" variant that adds semiempirical exact exchange can be enabled through the r12_k Coulomb-operator controls (the semiempirical fields of the r12_* specification). This is an advanced/development option.

Summary of keywords

Activating GFN-xTB

Keyword	Value	Description
`method`	`gfnxtb` / `gfn2xtb` (with optional `r`/`u` prefix)	Selects the xTB method and reference
`basis`	`gfnxtb` / `gfn2xtb`	Matching xTB parameter basis
`sphericalbasis`	`yes`	Required — xTB basis is spherical
`guess`	`hcore`	Extended-Hückel / core-Hamiltonian guess (auto-selected)

Method options

See the Tuning keywords table above for xtb_no3rdorder, xtb_noaes, xtb_noaxc, atomic_k_scal, sqmprint, sqm_parameterfile, sqm_hardness_average, and sqm_gamma_smoothing.

References

S. Grimme, C. Bannwarth, and P. Shushkov, A Robust and Accurate Tight-Binding Quantum Chemical Method for Structures, Vibrational Frequencies, and Noncovalent Interactions of Large Molecular Systems Parametrized for All spd-Block Elements (Z = 1–86), J. Chem. Theory Comput. 13, 1989 (2017). doi:10.1021/acs.jctc.7b00118 ↩
C. Bannwarth, S. Ehlert, and S. Grimme, GFN2-xTB — An Accurate and Broadly Parametrized Self-Consistent Tight-Binding Quantum Chemical Method with Multipole Electrostatics and Density-Dependent Dispersion Contributions, J. Chem. Theory Comput. 15, 1652 (2019). doi:10.1021/acs.jctc.8b01176 ↩