GFN2-XTB - An Accurate and Broadly Parametrized Self-Consistent Tight-Binding Quantum Chemical Method With Multipole Ele v1

doi.org/10.26434/chemrxiv.7246238.
v1
GFN2-xTB - an Accurate and Broadly Parametrized Self-Consistent

Tight-Binding Quantum Chemical Method with Multipole Electrostatics
and Density-Dependent Dispersion Contributions
Christoph Bannwarth, Sebastian Ehlert, Stefan Grimme
Submitted date: 24/10/2018 • Posted date: 24/10/2018

Licence: CC BY-NC-ND 4.0
Citation information: Bannwarth, Christoph; Ehlert, Sebastian; Grimme, Stefan (2018): GFN2-xTB - an
Accurate and Broadly Parametrized Self-Consistent Tight-Binding Quantum Chemical Method with Multipole
Electrostatics and Density-Dependent Dispersion Contributions. ChemRxiv. Preprint.
An extended semiempirical tight-binding model is presented, which is primarily designed for the fast
calculation of structures and non-covalent interactions energies for molecular systems with roughly 1000
atoms. The essential novelty in this so-called GFN2-xTB method is the inclusion of anisotropic second order
density fluctuation effects via short-range damped interactions of cumulative atomic multipole moments.
Without noticeable increase in the computational demands, this results in a less empirical and overall more
physically sound method, which does not require any classical halogen or hydrogen bonding corrections and
which relies solely on global and element-specific parameters (available up to radon, Z=86). Moreover, the
atomic partial charge dependent D4 London dispersion model is incorporated self-consistently, which can be
naturally obtained in a tight-binding picture from second order density fluctuations. Fully analytical and
numerically precise gradients (nuclear forces) are implemented. The accuracy of the method is benchmarked
for a wide variety of systems and compared with other semiempirical methods. Along with excellent
performance for the “target” properties, we also find lower errors for “off-target” properties such as barrier
heights and molecular dipole moments. High computational efficiency along with the improved physics
compared to it precursor GFN-xTB makes this method well-suited to explore the conformational space of
molecular systems. Significant improvements are futhermore observed for various benchmark sets, which are
prototypical for biomolecular systems in aqueous solution.
File list (1)
gfn2xtb.pdf (4.17 MiB) view on ChemRxiv download file

GFN2-xTB – an accurate and broadly
parametrized self-consistent tight-binding
quantum chemical method with multipole
electrostatics and density-dependent dispersion
contributions
Christoph Bannwarth,∗,†,‡ Sebastian Ehlert,† and Stefan Grimme∗,†
†Mulliken Center for Theoretical Chemistry, Universität Bonn, Beringstr. 4, 53115 Bonn,
Germany
‡New address: Department of Chemistry, Stanford University, Stanford, CA 94305, United
States of America.
E-mail: christoph.bannwarth@stanford.edu; grimme@thch.uni-bonn.de

Phone: +49-228/73-2351
1
Abstract
Keywords: density functional tight-binding, semiempirical molecular orbital theory,
multipole electrostatics, non-covalent interactions
Abstract
An extended semiempirical tight-binding model is presented, which is primarily
designed for the fast calculation of structures and non-covalent interactions energies
for molecular systems with roughly 1000 atoms. The essential novelty in this so-called
GFN2-xTB method is the inclusion of anisotropic second order density fluctuation ef-
fects via short-range damped interactions of cumulative atomic multipole moments.
Without noticeable increase in the computational demands, this results in a less em-
pirical and overall more physically sound method, which does not require any clas-
sical halogen or hydrogen bonding corrections and which relies solely on global and
element-specific parameters (available up to radon, Z = 86). Moreover, the atomic
partial charge dependent D4 London dispersion model is incorporated self-consistently,
which can be naturally obtained in a tight-binding picture from second order density
fluctuations. Fully analytical and numerically precise gradients (nuclear forces) are im-
plemented. The accuracy of the method is benchmarked for a wide variety of systems
and compared with other semiempirical methods. Along with excellent performance
for the “target” properties, we also find lower errors for “off-target” properties such
as barrier heights and molecular dipole moments. High computational efficiency along
with the improved physics compared to it precursor GFN-xTB makes this method
well-suited to explore the conformational space of molecular systems. Significant im-
provements are futhermore observed for various benchmark sets, which are prototypical
for biomolecular systems in aqueous solution.
1
1 Introduction
The accurate description of large molecular systems particularly in the condensed phase still remains
one of the main challenges faced in theoretical chemistry. 1,2 While Kohn-Sham density functional
theory (DFT) can routinely provide gas-phase (or continuum embedded) structures and energies
for roughly a few hundred atoms, long-time molecular dynamics simulations or conformational
sampling are still not feasible in a reasonably sized atomic orbital (AO) basis set for systems of
this size, e.g., within a day of computation time on a standard desktop computer. Recent efforts
of our 3–6 and other groups 7–10 have thus been directed towards the development of low-cost DFT
and Hartree-Fock methods to allow such calculations.
However, many interesting problems in biochemistry, materials science, supramolecular and macro-
molecular chemistry would require methods that can handle several thousands of atoms, which
is beyond the scope even for the aforementioned low-cost methods. Though classical force-fields
(FFs) can routinely be used for nanosecond timescale molecular dynamics simulations, 11 their lim-
itations are manifold and they are not suited for general use. Though faster computers and more
importantly parallel computing architectures along with accordingly adjusted codes have become
available in recent years, 12–25 there is still need for inherently simple, reasonable accurate, and ef-
ficient electronic structure methods, which are applicable to large systems also without specialized
hardware.
Semiempirical quantum mechanical (SQM) methods 26–28 are the physically motivated approach to
bridge the gap between ab initio quantum mechanical (QM) and FF methods. The foundation of
SQM methods typically is a valence-only minimal basis self-consistent field method, either derived
from Hartree-Fock theory or Kohn-Sham DFT. Since a QM description for the valence electrons is
used in SQM models, their scope of applications is much broader than that of FFs. At the same
time, they are faster by at least two orders of magnitude compared to ab initio QM methods, due
to drastic integral approximations. 28 These, however, come at the price of a significantly reduced
accuracy and robustness. Nevertheless, if parameters are adjusted carefully, SQM methods can be
used sufficiently well for screening purposes of a desired property. 29–34 Nowadays, frequently used
sophisticated Hartree-Fock-based zero-differential overlap (ZDO) methods include PM6 35–40 and
2
OM2. 41–44 While the former mostly finds use in the computation of ground state structures and
energies, 45–48 the latter is still extensively used in excited state dynamics studies. 49–53 In the past
twenty years, density functional tight binding (DFTB) methods have found much attention. 54–61
Here, the Kohn-Sham DFT energy is expanded in terms of density fluctuations δρ relative to a super-
position of atomic reference densities. The highest-level variant (DFTB3) employs a self-consistent
charge treatment including up to third order density fluctuation terms 55,56 (cf. Section 2.1) and
has been parametrized for a number of chemical elements. 58–62
In particular, the element pair-specific parametrization used in all aforementioned methods has
complicated the parametrization procedure and only PM6 covers large parts of the periodic table
of elements (70 elements). Recently, we have presented a DFTB3 variant, termed GFN-xTB, which
mostly follows a global and element-specific parameters only strategy and is parametrized to all
elements through radon. 34 The original purpose of the method and main target for the parame-
ter optimization has been the computation of molecular geometries, vibrational f requencies, and
non-covalent interaction energies. As such, it has successfully been used in structure optimizations
of organometallic complexes 63,64 and structural sampling. 65–67 Apart from that, the method per-
formed well in high temperature molecular dynamics simulations of electron impact mass spectra 68
and the electronic structure information of GFN-xTB can furthermore serve as input to generate
intermolecular FFs. 69 Of particular significance is its contribution to a newly developed protocol
for the automated computation of nuclear magnetic resonance (NMR) spectra. Therein, structural
sampling with GFN-xTB is employed to generate the conformer-rotamer-ensemble, which in turn
defines the detailed shape of an NMR spectrum. While the GFN-xTB could successfully identify
the relevant structures for less polar molecules, polar and strongly hydrogen-bonded systems like
sugars were not described as good. This motivated the present work. Here, we attempt to improve
upon the GFN-xTB Hamiltonian in a physically sound manner. The theory will be described below,
but in a nutshell, the method, which will be termed GFN2-xTB from now on, has the following
characteristics:
The GFN2-xTB basis set consists of a minimal valence basis set of atom centered, contracted
Gaussian functions, which approximate Slater functions (STO-mG). 70 Polarization functions
for most main group elements (typically second row or higher) are employed, which are
3
particularly important to describe hypervalent states. Different from GFN-xTB, hydrogen is
only assigned a single 1s function.
The GFN2-xTB Hamiltonian closely resembles that of GFN-xTB or of the well-known DFTB3
method. However, GFN2-xTB represents the first tight-binding method to include electro-
static interactions and exchange-correlation effects up to second order in the multipole ex-
pansion. Furthermore the simultaneously developed density-dependent D4 dispersion model
in an self-consistent formulation is an inherent part of the method. Different from either
GFN-xTB or DFTB3, GFN2-xTB does not employ other classical FF-type corrections, e.g.,
to describe hydrogen or halogen bonds. These are reasonably well described within the
multipole-extended electrostatics.
Different from other semiempirical methods, GFN2-xTB strictly follows a global and element-
specific parameter strategy. No element pair-specific parameters are employed.
As for the predecessor, properties around energetic minima, such as geometries, vibrational
frequencies, and non-covalent interactions are the target quantities. Already in our work
for the GFN-xTB scheme, we have identified that the preference for geometries (instead of
covalent bond energies) in the fit procedure resulted in systematically overestimated covalent
bond energies. We have not deviated from this strategy, as covalent bond energies are not
of primary interest for this method, and furthermore, the errors are very systematic as in
GFN-xTB. This way, the errors will be less random and more useful, once the systematic
errors have been removed – see Ref. 45 for such a correction scheme which in combination
with GFN-xTB outperforms any other semiempirical method of comparable complexity. By
keeping the focus on geometries as for GFN-xTB, the GFN2-xTB method would likewise be
well-suited for applications in high (electronic) temperature molecular dynamics simulations
to compute electron impact mass spectra. 71 However, due to the more sophisticated physical
interaction terms, we find that some “off-target” properties related to the electronic structure
also improve.
After a detailed description of the GFN2-xTB Hamiltonian, technical details of the calculations are
given. Then the method is benchmarked on standard sets for molecular structures, non-covalent
4
energies, and conformational energies. The performance for barrier heights and molecular dipole
moments is also assessed. Other semiempirical methods are used for comparison, namely GFN-
xTB, 34 PM6-D3H4X, 35,38,39 and DFTB3-D3(BJ). 56,58,72
2 Theory
Most previous derivations of density functional tight-binding (DFTB) are formally based on a
(semi-)local density approximation (LDA) of DFT. 28,54–56 Though non-local exchange variants
have been presented, 73–75 the electron correlation functional has also been local in those cases. This
however neglects long-range correlation effects, which are responsible for the important dispersion
interactions. Typically, corrections for the latter are then added in an a posteriori fashion. 72,76
Without explicitly referring to a specific functional approximation, we will assume as starting
point a general density functional approximation, which includes a non-local (NL) VV10-type 77
correlation contribution. The total Kohn-Sham energy expression is then given by (atomic units
are used throughout)
Z Z
LDA 1 1 NL 0 0 0
E [ρ] = ρ(r) T [ρ(r)] + Vn (r) + VXC [ρ(r)] + + ΦC (r, r ) ρ(r )dr dr + Enn .
2 |r − r0 |
(1)
Here, T [ρ(r)] is the kinetic energy per particle and Vn (r) the (external) potential due to the nuclei.
R P

In writing so, we are using the short-hand notation for ρ(r)T [ρ(r)]dr = − ni ψi 12 ∇2i ψi , with
i
ψi being the molecular orbital and ni the corresponding occupation number. The last three terms
in the integral over dr are the electronic contributions to the mean-field potential: the (semi-)local
exchange-correlation (XC), the Coulomb, and the non-local correlation potential, respectively. The
non-local correlation kernel ΦNL 0

C (r, r ) captures long-range correlation effects. Enn is the classical
nuclear repulsion energy. In density functional tight-binding (DFTB) theory, the total energy is
expanded in terms of density fluctuations around a superposition of (neutral) atomic reference

P
densities ρ0 = ρA0 : 28,54
A

E [ρ] = E (0) [ρ0 ] + E (1) [ρ0 , δρ] + E (2) ρ0 , (δρ)2 + E (3) ρ0 , (δρ)3 + · · · (2)
5
A fixed reference density ρ0 is employed and the calculation of the electronic structure is done in
terms of the density fluctuations δρ. The most widely used variants truncate this expansion after
the third order term. 34,55,56 The same is true for the GFN2-xTB approach presented here. The
first approximation made in tight-binding methods is the assumption that the density fluctuations
are restricted to the valence orbital space, while the core electron density remains frozen. The
individual approximations for the energy contributions to different orders in δρ have been described
before, 28,54,56,73,74 but we briefly outline the origin of newly introduced terms in GFN2-xTB.
2.1 Contributions relevant to the GFN2-xTB energy
2.1.1 Zeroth order terms
In previous derivations of DFTB, the zeroth order term is reduced to a classical repulsion term. 28,54–56
Due to the different starting point (DFT-NL), an additional term will arise at first order:
(0)
X (0) (0)

E (0) [ρ0 ] ≈ (0)
Erep + Edisp + EA,core + EA,valence . (3)
| {z } A
Lennard-Jones-type term
Here, the total energy in Eq. 1 is partitioned into same-center and off-center energy terms. All
one-electron and two-electron terms in Eq. 1 involving particles at the same atom reduce to a
single number, the atomic energy E A , which can furthermore be partitioned into a core and a
valence contribution (see last two terms in Eq. 3). Since the reference densities refer to neutral,
spherically symmetric atoms, the summed Coulomb interactions (e − e, n − n, e − n) between
distant atoms vanish at this level of approximation. However, in regions with overlapping atomic
densities, repulsive forces due to exchange-correlation and charge penetration effects are present.
Similar to previous works on DFTB, this term is modeled by a classical repulsion term. 28,54,56,73,74
Different from DFTB but in line with our previous method GFN-xTB, 34 no element pair-specific
parametrization will be employed for this term. Due to the presence of a NL correlation functional
in the energy expression in Eq. 1, a pairwise London dispersion contribution is also present at zeroth
order (see Supporting Information, Section 1.5). The sum of the atomic energies is a constant for a
given system, and the total tight-binding energy is given relative to the sum of atomic core energies,
6
(0) (0)
or equivalently they are set to zero, i.e., EA,core := 0. EA,valence is the sum of the valence orbital
energies, which represent eigensolutions of the Hamiltonian for the free atom.
(0)
XX
0 l
EA,valence = Pκκ hA (4)
l∈A κ∈l
The zeroth order density matrix P0 is diagonal, as the atomic orbitals with energy hlA are assumed
(0)
to be eigenstates of the free atom. Erep and Edisp will be described by classical expressions between
“clamped” atoms in GFN2-xTB (see Eqs. 9 and 32). It can be seen that the zeroth order energy in
the δρ expansion is thus related to the well-known Lennard-Jones energy, and can be a reasonable
approximation to treat non-covalent interactions of noble gas atoms. It does not contribute to the
tight-binding electronic energy, which depends only on the fluctuations δρ.
2.1.2 First order terms
For the choice of spherical atomic reference densities ρA0 , the description of covalent bonds in
tight-binding theories becomes possible at first order in the δρ expansion. This typically achieved
by changes of in the occupation δP of the atomic energy levels by invoking the well-known extended
Hückel approximation. Since δρ is assumed to be non-zero only in the valence electron space (see
above), the latter will be restricted to the valence orbitals. Due to the density fluctuations to
first order, the atoms can obtain a net charge and are no longer neutral. This has no effect on
the interatomic electrostatic (ES) interactions, since the electrostatic potential from the remaining
neutral atoms (with ρA0 ) is still zero. However, since starting from the DFT-NL expression in Eq. 1
in GFN2-xTB, the energy to first order in δρ is augmented with a first order dispersion contribution:
(1)
X (1)
E (1) [ρ0 , δρ] ≈ Edisp + EA,valence (5)
A
The last term describes first order changes in the occupation and energies of the electronic valence
levels. If combined with Eq. 4, this is essentially equivalent to extended Hückel theory (EHT),
P (0) (1)

i.e., EEHT = A EA,valence + EA,valence . The explicit expression to approximate this term in
GFN2-xTB will be given in Section 2.2. Due to the non-vanishing zeroth order London dispersion
7
potential of the other charge-neutral atoms, the dispersion energy changes at first order in δρ. The
expansion/contraction of the atomic density will increase/decrease the magnitude of the respective
interaction. In GFN2-xTB these and second order effects (see below) will be taken into account
within the self-consistent D4 dispersion model (see Section 2.2.6). 78,79
2.1.3 Second order terms
At second order in δρ, energy contributions arise, which require a self-consistent tight-binding
calculation.
(2) (2) (2)
E (2) ρ0 , (δρ)2 ≈ EES + EXC + Edisp . (6)
At second order, interatomic electrostatic and one-center exchange-correlation terms occur. In
DFTB schemes, these are generally condensed in a Mataga-Nishimoto-Ohno-Klopman damped
Coulomb interaction between atomic or shell charge monopoles. 80–82 The same will be done in
GFN2-xTB (see Section 2.2). For the first time in a tight-binding scheme, however, we will go be-
(2) (2)
yond the monopole approximation for both EES and EXC (see Section 2.2.5) and include anisotropic
effects up to second order in the multipole expansion. Isotropic second order δρ effects in the dis-
persion energy are also included within the self-consistent D4 dispersion model in GFN2-xTB.
2.1.4 Third order terms
Basically all energy contributions of the Kohn-Sham energy expression in Eq. 1 that require a self-
consistent solution at the DFT level of theory will also contribute to energy corrections of higher
order (≥ 2) in δρ. As in previous works on DFTB and GFN-xTB, 34,55 we will neglect all third
order terms but an isotropic on-site term, which mostly originates from short-ranged Coulomb and
XC effects.
(3) (3) 1 XX
E (3) ρ0 , (δρ)3 ≈ EES + EXC ≈ 3
ΓA,l qA,l (7)
3
A l∈A
Different from those previous works, however, the third order term is expressed in terms of partial
shell charges (Mulliken approximation) in GFN2-xTB.
8
2.2 The GFN2-xTB method
Having related the individual energy components to different orders in δρ, we will outline the specific
contributions to the GFN2-xTB energy in the following. The total GFN2-xTB energy expression
is given by
EGFN2-xTB = Erep + Edisp + EEHT + EIES+IXC + EAES + EAXC + GFermi . (8)
The newly introduced abbreviations in the subscripts indicate the isotropic electrostatic (IES) and
isotropic XC (IXC), and likewise the anisotropic electrostatic (AES) and anisotropic XC (AXC)
energies, respectively. Before going into details, it is important to note that no halogen- or hydrogen-
bond corrections are included in GFN2-xTB. The description of these interactions is already im-
proved by the AES energy EAES , which is described in Section 2.2.5.
2.2.1 The classical repulsion energy
For the repulsion energy in Eq. 8, we employ an atom pairwise potential similar to the one proposed
in Refs. 34,83.
X Y eff Y eff 1 krep
Erep = A B
e−(αA αB ) 2 (RAB ) , (9)
RAB
AB
where YAeff and YBeff define the magnitude of the repulsive interaction. Like αA and αB , they are
element-specific parameters. krep is a parameter, which is equal to one if both atoms are either H
or He, and equal to 3/2 otherwise. The different krep parameter for the very light element pairs
turned out to be beneficial for torsion barriers in alkanes, without sacrificing the accuracy for non-
covalent complexes. The repulsion energy in GFN2-xTB is described classically and is independent
of changes in the electronic structure. It should be mentioned that compared to GFN-xTB, 34 the
Y eff values correlate less with the atomic number in GFN2-xTB.
2.2.2 Choice of electronic wavefunction and finite temperature treatment
As in the predecessor GFN-xTB, we are working with a formally spin-restricted wavefunction
throughout and no spin density dependent terms are present. Hence, α and β orbitals always have
9
identical spatial parts and orbital energies but possibly different occupation. In order to handle
static correlation (nearly degenerate states) via fractional orbital occupations, a finite temperature
treatment is used. The last term GFermi in Eq. 8 formally refers to the entropic contribution of an
electronic free energy at finite electronic temperature Tel due to Fermi smearing. 84 This term is
necessary to provide a variational solution for fractional occupations and is given by
X X
GFermi = kB Tel [niσ ln (niσ ) + (1 − niσ ) ln (1 − niσ )] . (10)
σ=α,β i
kB is Boltzmann’s constant and niσ refers to the (fractional) occupation number of the spin-MO
ψiσ . These are given by

1
niσ = . (11)
exp[(i − σF )/(kB Tel )] +1
i is the orbital energy of the orbital ψi and σF = 0.5(σHOMO + σLUMO ) is the Fermi level for the
respective orbital space (α or β). Tel the electronic temperature, which by default is equal to 300 K
as in GFN-xTB. 34
The occupation ni for the spatial molecular orbital ψi (which is the same for ψiα and ψiβ ) is given
as
ni = niα + niβ . (12)
It should be stressed that this finite temperature treatment predominantly has the purpose of
enabling fractional orbital occupations in static correlation cases.
As usual, the spatial molecular orbitals (MOs) ψi are expressed as linear combinations of atom-
centered orbitals (LCAO) 85,86 ,
N
XAO
ψi = cκi φκ (ζκ , STO-mG) . (13)

κ
Following Stewart’s Gaussian expansions, 70 φκ refer to contracted Gaussian atomic orbitals, which
are used to approximate a spherical Slater-type orbital with exponent ζκ . The number of primitives
m varies between 3 and 6 – explicit numbers of primitives are given in the supporting information
(SI). In GFN2-xTB, a minimal spherical valence basis set is employed, whereas most heavier main
10
group elements (Z > 9) are provided with a polarization function. AOs located on the same atom
are always orthogonal to each other in GFN2-xTB, thus formally forming a basis of eigenstates of
the free atom. The basis set employed is given in Table 1, whereas the Slater exponents are listed
in the SI. As in GFN-xTB, the “f-in-core” approximation is employed for lanthanides. Variational
Table 1: Slater-type AO basis sets employed for the different elements. n denotes the
principal quantum number of the valence shell of the element.
element basis functions

H ns
He ns,(n + 1)p
group 1, Be–F, Zn, Cd, Hg–Po nsp
Ne nsp,(n + 1)d
group 2, 13–18 nspd
transition metals and lanthanides nd,(n + 1)sp
minimization of the energy expression in Eq. 8 with respect to the linear coefficients cκi in Eq. 13
leads to the general eigenvalue problem, which is likewise encountered in Hartree-Fock and Kohn-
Sham density functional theory
FC = SC . (14)
The elements of the tight-binding Hamiltonian or “Fock” matrix F will be given after presentation
of the individual energy terms. S is the overlap matrix and C is the LCAO-MO coefficient matrix.
is a diagonal matrix containing the orbital energies.
2.2.3 The extended Hückel-type energy
The extended Hückel contribution is the crucial ingredient to describe covalent bonds in tight-
binding methods.
X XXX XX
EEHT = ni hψi |Ĥ0 |ψi i = ni cκi cλi hφλ |Ĥ0 |φκ i ≡ Pκλ Hλκ . (15)
i κ λ i κ λ
0 +δP . Due to its origin in the first order δρ fluctuation

Here, the density matrix element Pκλ = Pκλ κλ
term, the operator Ĥ0 is formally a one-electron operator, which should however provide the zeroth
11
order energy (Eq. 4) for neutral non-interacting atoms. The corresponding matrix elements Hκλ
are thus approximated in the following way:
√ 1
1 0 2 ζκ ζλ 2
Hκλ = kll0 (hlA +hlB )Sκλ 2
(1+kEN ∆ENAB )Π(RAB,ll0 ) , (κ ∈ l ∈ A, λ ∈ l0 ∈ B) (16)
2 ζκ + ζλ
kll0 is the effective scaling factor common to all EHT methods. Its value depends on the angular
momentum quantum number of the interacting AOs (see Table 2 for explicit values). Similar
0
to GFN-xTB, 34 the atomic energy levels hlA /hlB are made flexible by being proportional to the
coordination number,
hlA = HA
l l
− HCNA
CNA0 , (l ∈ A ) (17)
with
atoms
NX −1 −1
CNA0 = 1+e −10(4(RA,cov +RB,cov )/3RAB −1)
1+e−20(4(RA,cov +RB,cov +2)/3RAB −1)
(18)
B6=A
l and H l
Here, HA 87
CNA are element-specific parameters. Compared to the D3 coordination number,
which has been used in GFN-xTB, 34 CNA0 is smoother and slightly more long-ranged. The covalent
radii of the atoms RA,cov /RB,cov are the rescaled radii from Ref. 88 as used in the D3 87 and D4 78
(see below) model. Sκλ is the overlap between the two AOs. The last three terms in the product on
the right hand side of Eq. 16 modify the magnitude of the interaction in certain situations. ζκ and
ζλ are the Slater exponents of the two AOs. This factor is unity for ζκ = ζλ and otherwise reduces
the magnitude of the Hκλ matrix element. Similarly, the second last term reduces the “covalent”
interaction for two atoms with different electronegativities. While the latter term exclusively adjusts
non-polar vs. polar (or ionic) binding, the former term adds more flexibility for orbitals with
different compactness, and thus, is also relevant in homonuclear situations. The explicit dependence
on orbital exponents introduces effects present for HF/DFT in the kinetic energy one-electron
integrals. As in GFN-xTB, 34 the distance and shell dependent polynomial Π(RAB,ll0 ) is based on
poly poly
element-specific parameters kA,l /kB,l 0 , which are fitted, while the RA,cov 0 /RB,cov 0 are taken from
Ref. 89.
1 ! 1 !
poly RAB 2
poly RAB 2
Π(RAB,ll0 ) = 1 + kA,l 1 + kB,l 0 (19)
Rcov0 ,AB Rcov0 ,AB
12
Table 2: Global empirical parameters of the GFN2-xTB method. The parameters are either
dimensionless or in atomic units.
parameter value
kss 1.85
kpp ,kdd 2.23
ksp 2.04a
ksd ,kpd 2.00
krep 1.5b
Ks 1.0
Kp 0.5
Kd 0.25
kEN −0.02
multipole parameters
∆val 1.2
Rmax 5.0
a3 3.0
a5 4.0
dispersion parameters
a1 0.52
a2 5.0
s6 1.0
s8 2.7
s9 5.0
a
Obtained as ksp = 0.5(kss + kpp )
b
krep = 1.0 for H/He pairs.
The key purpose of this term is the distance-dependent adjustment of the EHT-type interaction,
which provides a better balance between short- (covalent) and long-ranged (non-covalent) effects.
The EHT-type term in GFN2-xTB is mostly responsible for covalent binding. Via the coordination
number dependence of the valence energy levels, these obtain additional flexibility beyond the δρ-
based expansion up to first order. To some extent, the fitted element- and shell-specific parameters
l and H l
for HA CNA can implicitly account for the formally neglected first and second order on-site
δρ effects. Noteworthy is the importance of that term for atoms that can become hypervalent: the
energy level for the d-polarization functions are typically high in energy for small CNA0 . Then, these
functions do neither interfere strongly with the valence functions in “standard” covalent binding
nor do they significantly affect non-covalent interactions in a way which is reminiscent to the well-
known basis set superposition error (BSSE) in AO-based ab-initio calculations. For large CNA0
13
values, however, this d-level is lowered in energy, which then enables a much better description of
hypervalency.
2.2.4 The isotropic electrostatic and exchange-correlation energy
For charged and polar systems, the density ρ deviates from the reference density ρ0 . In that
case, the net partial charges on the individual atoms are non-zero. In GFN2-xTB, the isotropic
electrostatic and XC terms are treated with shell-wise partitioned Mulliken partial charges (cf. Ref.
90). The following contribution to the energy is then obtained:
1 XX X 1 XX 3
EIES+IXC = qA,l qB,l0 γAB,ll0 + ΓA,l qA,l . (20)
2 0
3
A,B l∈A l ∈B A l∈A
Here, the first term on the right-hand side is derived from the second order energy, while the last
term is due to third order density fluctuations. qA,l refers to an isotropic monopole charge of the
l shell on atom A. The distance dependence of the Coulomb interaction within the first term is
described by a generalized form of the well-established Mataga-Nishimoto-Ohno-Klopman 80–82,91
formula:
1
1 2
γAB,ll0 = (21)
RAB + η −2
2
Here, RAB is the interatomic distance and η is the average of the effective chemical hardnesses of
the two shells l and l0 on the atoms A and B:
1 1 0

η= ηA,l + ηB,l0 = (1 + κlA )ηA + (1 + κlB )ηB (22)
2 2
0
ηA and ηB are treated as element-specific fit parameters. κlA and κlB are fitted element-specific
scaling factors for the individual shells (note that κlA = 0 for l = 0). This way, electrostatic
interactions between distant atoms and on-site changes in the isotropic electrostatic and XC energy
are treated in a seamless manner. The third order term is restricted to shell-wise on-site terms. As
discussed in the literature on DFTB, 56 this term can help to stabilize charged atomic states and
partially remedy shortcomings from missing, e.g., diffuse functions in the AO basis.
The shell-wise parameter ΓA,l is obtained from the element-specific parameter ΓA and the global,
14
shell-specific parameters Kl .
ΓA,l = Kl ΓA (23)
ΓA is formally related to ∂ηA /∂ρA |ρ0 , but is a fitted parameter in GFN2-xTB. At variance with
GFN-xTB, we use a shell-wise, though parameter-economic treatment by treating Kl as global
parameters. The shell-wise treatment appeared to be beneficial for some transition metals without
diminishing the accuracy for main group elements.
The shell-wise treatment requires the definition of reference valence shell occupations. For the
occupation of elements of group 1, 2, 12, 13, 16, 17, and 18, we follow the aufbau principle, whereas
for transition metals a modified aufbau principle of the form ndx−2 (n + 1)s1 (n + 1)p1 (x denotes the
group) is used. Lighter elements of group 14 and 15 are handled slightly different to better reflect
the occupations in bonded atoms. For this purpose, carbon is treated with a reference valence shell
occupation of 2s1 2p3 , whereas fractional reference occupations of type ns1.5 npx−11.5 (x denotes the
group) are used for N, Si, Ge, P, and As. The remaining elements of group 14 and 15 follow the
standard aufbau principle. All element-specific parameters are given in the SI.
2.2.5 The multipole-extended electrostatic and exchange-correlation energy
The approximate expression for the ES and XC energy commonly employed in tight-binding theory
is derived from the first two terms of the second order energy (see Eq. 6): 56
ZZ !
(2) (2) 1 1 ∂ 2 EXC
EES + EXC = + δρ(ri )δρ(rj )dri drj (24)
2 rij ∂ρ(ri )∂ρ(rj ) ρ=ρ0
Anisotropic electrostatic interactions In Figure 1, it is schematically shown how Eq. 24

is typically approximated by purely isotropic energy terms in DFTB. 54,56 γAB is the interatomic
Coulomb interaction, which is damped to a finite value at short-range (see Eq. 21). This short-range
damping then also includes effects due to the second order changes in the semi-local XC energy
(2)
EXC . Apart from different partitioning schemes (atomic or shell-wise, see Section 2.2.4), the same
functional form for the second order ES/XC energy is used in DFTB, GFN-xTB, 34 and the GFN2-
xTB method presented here. The basic possibility of including higher multipole ES interactions
in DFTB has been suggested in Ref. 92. Nevertheless, only the first order charge-dipole term
15
(2) (2)
EES + EXC X µTARAB qB RTAB ΘARAB qB
(2)
Eaniso ≈ fdamp 3 + 5
RAB RAB
A6=B
anisotropic !
approximate 1 µTARAB µTB RAB − µTAµB RAB 2
− 5
2 RAB
X µ Θ

+ fXCA |µA|2 + fXCA ||ΘA||2
isotropic A
(2) 1
P
Eiso ≈ 2 γAB qAqB
A,B
DFTB/GFN-xTB GFN2-xTB
Figure 1: Schematic overview of the employed approximations for the second order
ES and XC energy in tight-binding theory. While the isotopic approximation is well-
established, 34,54,56 the full inclusion of anisotropic effects up to second order in the multipole
expansion is novel. For simplicity, an atomic charge partitioning is used throughout in this
scheme.
has been presented 92 and no implementation in a functioning DFTB method has been reported so
far. In GFN2-xTB, we pioneer in going beyond this monopole approximation for both, ES and XC
terms including all terms up to second order in the multipole expansion. The newly incorporated
terms are schematically shown in Figure 1 and outlined below (for a derivation, see SI). The AES
energy in GFN2-xTB is given by
EAES =Eqµ + EqΘ + Eµµ (25a)

1 X
= f3 (RAB ) qA µTB RBA + qB µTA RAB (25b)
2
A,B

+ f5 (RAB ) qA RTAB ΘB RAB + qB RTAB ΘA RAB (25c)
2
− 3 µTA RAB µTB RAB + µTA µB RAB . (25d)
Here, µA is the cumulative atomic dipole moment of atom A and ΘA is the corresponding traceless
16
quadrupole moment
3 αβ δαβ xx
Θαβ
A = θA −
yy
θA + θA zz
+ θA . (26)
2 2
The cumulative atomic multipole moments (CAMM) 93 up to second order are computed from:
XX
qA = Z A − Pκλ hφλ | φκ i (27a)
| {z }
κ∈A λ
Sλκ
 
XX  
µαA = Pκλ αA Sλκ − hφλ |αi | φκ i (27b)
| {z }
κ∈A λ α
Dλκ
 
XX  
αβ
θA = Pκλ  β α
αA Dλκ + βA Dλκ − αA βA Sλκ − hφ

λ |αi βi | φκ i (27c)
| {z }
κ∈A λ
Qαβ
λκ
α and Qαβ are the electric dipole and quadrupole moment

α and β are Cartesian components. Dλκ λκ
integrals between the AOs φκ and φλ . The expressions for the CAMMs directly originate from the
multipole expansion (in Cartesian coordinates) and guarantee that the respective overall molecular
moments are correctly preserved. These expressions give the atomic contribution (Mulliken ap-
proximation) of the particular atom to the overall multipole moment. The CAMMs are defined as
such that their respective origin is located at the particular atom, i.e., they are origin-independent,
as enforced by the “shift” contributions from lower order moments (Eqs. 27b and 27c). In Eq. 25,
we have gone up to second order in the multipole expansion of the Coulomb energy, thus all terms
−3
that decay with RAB or slower are included (see Eq. 28). The monopole-monopole term, which
is the term of lowest order (see SI), has already been included in the shell-wise isotropic ES en-
ergy described in Section 2.2.4. The terms containing higher order multipoles should improve the
description of the anisotropic electron density around the atoms. We chose to employ an atomic
partitioning in GFN2-xTB for these terms. To avoid divergence for the AES energy (Eq. 25), we
damp the corresponding terms at short distances. The distance dependence including damping is
given by
fdamp (an , RAB ) 1 1
fn (RAB ) = n = n · AB an (28)
RAB RAB 1 + 6 R0
RAB
The damping function is related to the zero damping function in the original D3 dispersion model. 87
17

an are adjusted global parameters, whereas R0AB = 0.5 R0A0 + R0B0 determines the damping of the
AES interaction. R0A0 is made dependent on the D3 coordination number for many lighter elements.
Rmax − R0A
R0A0 = R0A + (29)
1 + exp[−4(CNA − Nval − ∆val )]
∆val = 1.2 and Rmax = 5.0 bohrs. Aside from those light elements (see SI), R0A0 = 5.0 bohrs
for all elements. This flexible R0A0 value reduces the strength of the AES interactions for strongly
coordinated atoms and was found to increase the robustness of the SCF convergence for inorganic
clusters. Primarily, the AES terms are intended to improve the non-covalent interactions between
the outer, i.e., less coordinated atoms. This way, no extra hydrogen or halogen bond corrections
nor any element-specific bond adaptations are required.
Anisotropic XC energy While the basic idea of including AES terms in a TB model has
been suggested before, 92 no such extension for second order XC effects was mentioned so far. In
the context of excitation energies, the approach of Dominguez et al. to include INDO-like terms in
DFTB is somewhat related, though its basis is rather found in a semiempirical hybrid-like density
functional. Here, we take the second term of Eq. 24 as starting point. The second order XC energy
contribution takes the form of a static XC kernel. In the local density approximation, this term
reduces to a pure same-site energy in a tight-binding scheme. Going up to second order multipolar
(2)
terms, EXC can then be simplified to (see SI for the derivation):
 
(2)
X q µA ΘA 
EXC ≈ A 2
 fXC q + fXC |µA |2 + fXC ||ΘA ||2  (30)
| {z A} | {z }
A
isotropic XC anisotropic XC
The isotropic monopolar term is already included in the shell-wise isotropic XC energy (see Sec-
tion 2.2.4). To our knowledge, the other terms, are proposed here for the first time in a tight-binding
context. They form the anisotropic XC energy in GFN2-xTB:
X µA ΘA

EAXC = fXC |µA |2 + fXC ||ΘA ||2 (31)
A
18
Again, µA and ΘA are the cumulative atomic dipole and traceless quadrupole moments, which have
µA ΘA
already been introduced in Eq. 27 and 26. fXC and fXC are fitted element-specific parameters.
Formally, these terms are supposed to capture changes in the atomic XC energy, which result
from anisotropic density distributions (polarization). To some extent, they may also alleviate
shortcomings of the small AO basis set (e.g., insufficient polarization functions).
2.2.6 The density-dependent dispersion energy
In GFN2-xTB, we treat dispersion interactions by means of a self-consistent variant of the re-
cently published D4 dispersion model. 78,79 In typical semiempirical mean-field methods, London
dispersion interactions are generally treated by means of post-SCF corrections (see Refs. 28 and
76 for reviews). Though the widely employed D3 dispersion model takes environmental effects
via the geometric coordination number into account, electronic structure effects are missing. In a
tight-binding context, the D3 dispersion energy should be regarded as a zeroth order term, i.e., it
(0)
corresponds to Edisp . Here, we go beyond this model and include effects up to second order within
the self-consistent formulation of the D4 dispersion model: 79
(0) (1) (2)

Edisp =Edisp + Edisp + Edisp
X X A , q , CN B )
CnAB (qA , CNcov B cov
≈− sn n fndamp,BJ (RAB )
RAB
A>B n=6,8
(32)
X (3 cos(θABC ) cos(θBCA ) cos(θCAB ) + 1)C9ABC (CNcov
A , CN B , CN C )
cov cov
− s9 3
A>B>C
(RAB RAC RBC )
× f9damp,zero (RAB , RAC , RBC )
The last term is the charge-independent three-body (also called Axilrod-Teller-Muto or ATM)
dispersion term, 94,95 which is added to incorporate the dominant part of the many-body dispersion
energy. Different from the charge-dependent two-body term, this three-body contribution does not
affect the electronic energy.
The damping functions fndamp,BJ and f9damp,zero in Eq. 32 have been defined in Refs. 96 and 87,
respectively. It is, however, important to note that in the D4 model, the BJ-type cutoff radii 96
are used in both fndamp,BJ and f9damp,zero (see below). While the C8AB are calculated recursively 87
19
from the lowest order C6AB coefficients, the latter are computed from a numerical Casimir-Polder
integration.
3X
C6AB = A
wj αA (iωj , qA , CNcov B
)αB (iωj , qB , CNcov ) (33)
π
j
wj are the integration weights, which are derived from a trapezoidal partitioning between the grid
points j (j ∈ [1, 23]). The isotropically averaged, dynamic dipole-dipole polarizabilites αA at the
j th imaginary frequency iωj are obtained from the self-consistent D4 model, i.e., they are depending
on the covalent coordination number, 79 and are also charge dependent. The method thus relies on
precomputed atomic polarizabilities at a certain molecular geometry, i.e., with the atom having a
A,r
GFN2-xTB computed partial charge of qA,r and a covalent coordination number CNcov . Similar
A via the W r
to D3, a Gaussian-weighting scheme based on the covalent coordination number CNcov A
terms (see Ref. 78) is employed.
NA,ref
X
A r A,r
αA (iωj , qA , CNcov ) = ξA (qA , qA,r )αA,r (iωj , qA,r , CNcov )WAr (CNcov
A A,r
, CNcov ) (34)
r
The Gaussian-weighting for each reference system is given by
Ngauss
X 1 h i
A,r 2
WAr (CNcov
A A,r
, CNcov )= A
exp −6j · CNcov − CNcov
N
j=1
(35)
NA,ref
X
with WAr (CNcov
A A,r
, CNcov )=1
r
where N is a normalisation constant. The number of Gaussian functions per reference system
A,r
Ngauss is mostly one, but is equal to three for CNcov = 0 and reference systems with similar
coordination number (see Ref. 78 for details) The charge-dependency is included via the empirical
r.
scaling function ξA
eff + q
r ZA A,r
ξA (qA , qA,r ) = exp 3 1 − exp 4ηA 1 − eff (36)
Z A + qA
eff is the effective nuclear charge of atom

Where ηA is the chemical hardness taken from Ref. 97. ZA
A, which has been determined by substracting the number of core electrons represented by the
20
def2-ECPs in the time-dependent DFT reference calculations (see Ref. 78 for details). Due to the
charge dependency, the pairwise dispersion energy in GFN2-xTB enters the electronic energy and
is self-consistently optimized. A similar expression is used in the standard form of the DFT-D4
method, 78 but therein, the partial charges are obtained by purely geometrical means. The rational
damping function used in the DFT-D4 model is given by
s
n
RAB C8AB
fndamp,BJ (RAB ) = n crit. + a )6
with crit.
RAB = (37)
RAB + (a1 · RAB 2 C6AB
The zero damping function for the ATM dispersion is defined slightly different to the previous
implementations of DFT-D3, namely the factor 4/3 is dropped and the cutoff radii are calculated
consistently to the two-body terms.
 s 16 −1
crit. Rcrit. Rcrit.
RAB
f9damp,zero (RAB , RAC , RBC ) = 1 + 6 BC CA  
3
(38)
RAB RBC RCA
2.3 The GFN2-xTB Hamiltonian matrix
As mentioned before, GFN2-xTB includes energy terms of second and third order in δρ. Therefore,
the energy expression in Eq. 8 needs to be solved self-consistently. To compute the density matrix,
the Roothaan-Hall-type eigenvalue problem (Eq. 14) is then solved.
As for the total energy, the matrix elements of the GFN2-xTB Hamiltonian can be decomposed
into individual contributions
IES+IXC AES AXC D4

Fκλ = Hκλ + Fκλ + Fκλ + Fκλ + Fκλ . (39)
Due to the analogy to Hartree-Fock, we will denote this matrix simply as TB-Fock matrix in the
following. The extended Hückel matrix elements Hκλ have already been described in Eq. 16. The
general derivation for the isotropic ES and XC contributions to the TB-Fock matrix has been shown
21
in Ref. 56. In GFN2-xTB, these are given by
1 XX 1
IES+IXC 2 2 0
Fκλ = − Sκλ (γAC,ll00 +γBC,l0 l00 )qC,l00 − Sκλ (qA,l ΓA,l +qB,l 0 ΓB,l0 ) (κ ∈ l(A), λ ∈ l (B)) .
2 00
2
C l
(40)
where indices κ and λ denote the AOs with corresponding angular momenta l and l0 and the second
sum runs over the atoms C and their shells l00 .

AES and
The last three terms in Eq. 39 are new, and their derivation can be found in the SI. Fκλ
AXC both involve terms that include electric dipole and quadrupole, as well as overlap integrals.
Fκλ
In a condensed notation, they can be expressed as
AES AXC 1
Fκλ + Fκλ = Sκλ [VS (RB ) + VS (RC )] (41a)
2
1
+ DTκλ [VD (RB ) + VD (RC )] (41b)
2
1 X αβ h αβ i
+ Qκλ VQ (RB ) + VQαβ (RC ) , ∀ κ ∈ B, λ ∈ C . (41c)
2
α,β
Here, the respective integral (overlap, dipole, and quadrupole) proportional potential terms are
given as
X
VS (RC ) = RTC f5 (RAC )µA RAC
2
− RAC 3f5 (RAC ) µTA RAC − f3 (RAC )qA RAC
A
1
− f5 (RAC )RTAC ΘA RAC − f3 (RAC )µTA RAC + qA f5 (RAC ) R2C R2AC
2 (42)
3 X
− qA f5 (RAC ) αAB βAB αC βC }
2
α,β
µC ΘC T
+ 2fXC RTC µC − fXC RC [3ΘC − Tr(ΘC ) I] RC
X
VD (RC ) = RAC 3f5 (RAC ) µTA RAC − f5 (RAC )µA RAC
2
+ f3 (RAC )qA RAC
A
#
X
− 2
qA f5 (RAC )RC RAC + 3qA f5 (RAC )RAC αC αAC (43)
α
µC ΘC
− 2fXC µC + 2fXC [3ΘC − Tr(ΘC ) I] RC
22
X
3 1 2
VQαβ (RC ) =− qA f5 (RAC ) αAC βAC − RAB
2 2
A
" # (44)
X
ΘC
− fXC 3ΘαβC − δαβ Θαα
C
α
The last line in each of the previous equations describes the AXC potential, whereas the remaining
terms define the AES potential experienced at position RC . It is stressed that the electric multipole
moment integrals are given with origin at O= (000)T . Hence, VS (RC ) and VD (RC ) also include
higher order potential terms due to the “shift” terms in the CAMM definition (see Eq. 27). Analyt-
ical first nuclear derivatives are not shown here, but are derived and given in the SI. However it is
noted here, that the expressions for the analytical gradients become much simpler if the multipole
integrals are given with origin at the respective atomic position.
The TB-Fock matrix contribution from self-consistent D4 is given by
D4 1
Fκλ = − Sκλ (dA + dB ), ∀κ ∈ A, λ ∈ B (45)
2
where dA is given by (we are dropping the dependency on q and CNcov for brevity)
NA,ref
X ∂ξ r X NX
B,ref
X CnAB,ref
dA = A
WAr WBs ξB
s
· sn n fndamp,BJ (RAB ) . (46)
r
∂qA s
RAB
B n=6,8
Here, the dispersion coefficient for two reference atoms CnAB,ref is evaluated at the reference points,
A = CN r , and CN B = CN s .
i.e., for qA = qr , qB = qs , CNcov cov cov cov
2.4 Technical details
The global parameters in Table 2 and the element-specific parameters (see SI) have been deter-
mined by minimizing the root-mean-square deviation (RMSD) between reference and GFN2-xTB-
computed data. The procedure is the same as in GFN-xTB an relies on the Levenberg-Marquardt
algorithm 98,99 .
The global parameters have been determined along with the element-specific parameters for the
23
elements H, C, N, and O. Then the element-specific parameters for the other elements were subse-
quently determined keeping all existing parameters fixed. For the lanthanides, only the parameters
for Ce and Lu were freely fitted, while a linear interpolation with the nuclear charge Z has been
used for the other elements.
In general, the reference data, which was employed in the parameterization of GFN-xTB, has been
extended and used for fitting. The data consisted of molecular equilibrium and non-equilibrium
structures (energies and forces), harmonic force constants for up to medium sized systems (<
30 atoms), and non-covalent interaction energies and structures (mainly potential energy curves
and a few full optimizations). No atomic partial charges or molecular dipole moments have been
used in the fit.
A mixed level of theory for the reference data is used. If systems from standard benchmark sets
have been used, basis set extrapolated CCSD(T) energies are typically used, while forces and
structures were mostly computed by the more efficient, though sufficiently accurate composite
methods PBEh-3c 4 or B97-3c. 6 We used the TURBOMOLE suite of programs 100–102 (version 7.0)
to conduct most of the ground state DFT reference calculations and geometry optimizations. The
exchange-correlation functional integration grid m4 and the SCF convergence criterion (10−7 Eh )
along with the resolution of the identity (RI) integral approximation 103–105 has been used.
The GFN2-xTB method parameters have first been determined with a D3-variant for the dispersion
energy. Then the D4 parameters and reference values (i.e., qA,ref ) were determined while keeping
all other parameters fixed.
Calculations for comparison with other semiempirical methods were conducted with the DFTB+ 106
(DFTB3 56 with the 3OB parametrization 58–60 ), and MOPAC16 107 (PM6-D3H4X 35,38,39 ) codes. The
DFTB3 method was used in combination with the 3OB parametrization 58–60 and D3(BJ) 72,87,96 dis-
persion correction. PM6-D3H4X, i.e., with zero-damped D3 dispersion, as implemented in MOPAC16
is used. 38,87 Our standalone dftd3 code 108 was used for the calculations of the D3(BJ) corrections.
The Fermi smearing technique (Tel = 300 K) has also been employed in the DFTB3 calculations.
In the following, we will use the abbreviations: mean deviation (MD), mean absolute deviation
(MAD), standard deviation (SD), mean relative deviation (MRD), mean unsigned relative deviation
(MURD), standard relative deviation (SRD), maximum unsigned deviation (MAX), maximum
24
unsigned relative deviation (MAXR), and regularized relative root mean square error (RMSE).
3 Results and Discussion
3.1 Molecular structures
The ROT34 benchmark set 109,110 has become an established set to assess the performance of
quantum chemical methods to compute gas phase structures of organic molecules. The quantity
to be compared is the spectroscopically accessible rotational constant B0 , which can quantum
chemically be corrected for nuclear vibrational effects to Be . This “clamped” nuclei rotational
constant can then be compared directly to the local Born-Oppenheimer minimum energy structure
of the investigated method. As long as conformational changes can be excluded, smaller rotational
constants then typically indicate elongated covalent bonds relative to the reference. Since the
isoamyl-acetate molecule in the ROT34 set was found to be problematic w.r.t. conformational
changes for many semiempirical methods, it is excluded in this work. The data for the other SQM
methods is taken from Ref. 34, whereas the detailed results for GFN2-xTB are given in the SI. It has
been discussed before that the tight-binding methods show significantly smaller standard relative
deviations (SRDs) compared to the ZDO methods OM2-D3(BJ) and PM6-D3H4X. 34 GFN2-xTB
ranks second from the considered methods and is outperformed only slightly by its predecessor
GFN-xTB. This likely results from the fact that the relative weight of geometries in the fitting
procedure has been larger for the GFN-xTB than for the GFN2-xTB method. Furthermore, it
should be noted that although GFN-xTB is mostly constructed from global and element-specific
parameters, there still are a few element pair-specific “fine-tuning” parameters (e.g., between N–
H or H–H pairs). GFN2-xTB relies solely on element-specific and global parameters, thus its
performance for these organic structures can be regarded as excellent. In Figure 3, the MADs for
three structure benchmark sets are shown, which are more difficult for quantum chemical methods.
The LB12 4 consists of 12 molecules with an “unusually” long bond between two atoms. HMGB11 4
contains heavy main group and TMC32 111 contains 50 bond distances in a total of 32 transition
metal complexes. GFN-xTB and GFN2-xTB clearly outperform PM6-D3H4X for this purpose.
25
too large molecules too small molecules
GFN-xTB
GFN2-xTB
DFTB3-D3(BJ)
PM6-D3H4X
-8 -6 -4 -2 0 2 4 6 8
relative error in %
1
Figure 2: Normal distribution plots for the relative errors in the computed equilibrium
rotational constants Be for the ROT34 benchmark, with system 2 (isoamyl-acetate) ex-
cluded. The GFN-xTB (MRD=0.52%, SRD=1.10%), DFTB-D3(BJ) (MRD=−1.26%,
SRD=1.28%), and PM6-D3H4X (MRD=−1.60%, SRD=2.50%) results are taken from Ref.
34. The values for GFN2-xTB are computed in this work (MRD=0.78%, SRD=1.24%).
In particular, it becomes apparent that GFN2-xTB reproduces the LB12 and HMGB11 bond
lengths particularly well. This reflects the consistent element-specific parametrization in GFN2-
xTB. It should, however, be noted that there exists one outlier in the LB12 set for GFN2-xTB
(385 pm instead of 286 pm for the S8 2+ system), which is excluded in the statistical analysis.
Presumably, this overestimated bond length is caused by the strong net charge of the system in
combination with the higher order, but truncated multipole expansion (cf. GFN-xTB, which shows
a bond distance underestimated by about 55 pm). This system is also difficult for many density
functional approximations showing similar large deviations from the reference. 4 For the transition
metal complexes (right hand side of Figure 3), the three semiempirical methods perform more
similarly with GFN-xTB performing best and GFN2-xTB ranking second. For all sets considered
here, it is observed that – compared to GFN-xTB – the magnitude of the MD is reduced for
GFN2-xTB (see SI and Ref. 34).
26
LB12 (×0.5) HMGB11 TMC32
11
10 *
9
8
7
*
MAD / pm
6
5
4
3
2
G
PM
PM
PM B
FN
FN
FN
FN
FN
FN
6-
6-
6-
2-
-x
2-
-x
2-
-x
D
D
T
T
x
xT
xT
3H
3H
3H
T
B
B
B
4X
4X
4X
Figure 3: MADs in pm for bond lengths computed with GFN2-xTB, GFN-xTB, and PM6-
D3H4X (see SI for detailed results). The LB12 and HMGB11 sets are taken from Ref. 4.
For LB12, the MADs are scaled be factor of 0.5. The transition metal containing systems
HAPPOD and KAMDOR in LB12 have been discarded for PM6-D3H4X. For GFN2-xTB,
the S2+
8 system in LB12 is neglected.
In Table 3, the statistical data for structures with focus on non-covalent interactions are given.
Here, the center-of-mass (CMA) distance deviation for fully optimized complexes from the S22 112
set are given. Furthermore, the relative deviations for the extrapolated CMA distances for the
S66x8, 113 S22x5, 114 X40x10, 115 and R160x6 48 sets are given. These are determined by cubic spline
interpolations of the respective interactions at the different distances, which has previously been
done already for the S66x8 set. 4,34 Regarding the first two sets (fully optimized S22 and S66x8),
GFN2-xTB performs much like GFN-xTB, but shows slightly stronger underestimation for the
CMA distances of S66x8. Furthermore, the deviations appear to be a bit more systematic for both
sets as indicated by the smaller standard relative deviation (SRD) for S66x8 and standard deviation
(SD) for S22. All of the SQM methods considered perform comparably well with DFTB3-D3(BJ)
showing the largest deviations for the S66x8 set.
27
Table 3: Comparison of structures of non-covalently bound systems. The CMA distances
of the S22 112 complexes are obtained from a free optimization of the complex geometries.
The deviations are given in pm. The CMA distances of the S66 113 S22x5, 114 X40x10, 115 and
R160x6 48 complexes are derived from cubic spline interpolations of energies computed on
the differently separated structures. Here, the relative errors in % are given.
GFN2-xTB GFN-xTB PM6-D3H4X DFTB3-D3(BJ)

S22 112 (CMA distance in pm)
MD: −5 −5a 3a −11a
MAD: 14 15a 14a 14a
a a
SD: 16 18 21 16a
MAX: 32 50a 59a 45a
S66x8 113 (CMA distance in %)
MRD: −0.82 −0.63a 0.58a −2.87a
MURD: 1.99 1.96a 2.01a 3.15a
a a
SRD: 2.58 2.68 2.57 2.84a
MAXR: 6.91 6.95a 6.74a 10.93a
S22x5 114 (CMA distance in %)
MRD: −1.40 −0.76 0.03 −3.09
MURD: 2.76 2.49 1.97 3.51
SRD: 3.11 3.32 2.40 3.59
MAXR: 6.85 6.78 4.44 10.70
X40x10 115 (CMA distance in %)
MRD: −0.17 −1.72 −1.39 −2.16
MURD: 2.59 3.47 5.25 3.67
SRD: 3.22 3.98 7.40 4.04
MAXR: 7.82 8.85 20.09 11.53
R160x6 48 (CMA distance in %)
MRD: −3.59 −6.50 −4.67 −8.22
MURD: 4.52 8.27 6.40 9.08
SRD: 7.93 11.01 8.83 10.06
MAXR: 29.21 35.30 39.24 34.04
a
Data taken from Ref. 34.
MRD=mean relative deviation, MURD=mean unsigned relative deviation,
SRD=standard relative deviation, MAXR=maximum unsigned relative deviation,
MD=mean deviation, MAD=mean absolute deviation, SD=standard deviation,
MAX=maximum absolute deviation.
The results obtained here are quite important also as an assessment for the balance of the non-
covalent interaction terms. If repulsion, but in particular dispersion and the (anisotropic) ES
terms are not described in a balanced way, in particular the free optimization of the S22 structures
should yield larger deviations for the CMA distances. PM6-D3H4X has been parametrized to
28
the S66x8 energies, and as expected, shows good performance for this set, though not better
than GFN2-xTB. Its performance for S22x5 is remarkably good, for which it shows the smallest
deviations of the tested methods, followed by GFN-xTB and GFN2-xTB. For X40x10 and R160x6,
GFN2-xTB clearly outperforms the other methods, and marks a significant improvement over
GFN-xTB, in particular for the R160x6 set. Hence on average, GFN2-xTB is the best of all SQM
methods considered here, which indicates that the Hamiltonian terms responsible for non-covalent
interactions are well-behaved.
3.2 Non-covalent interaction energies
In order to investigate the performance for non-covalent interaction energies, we consider benchmark
sets from the large GMTKN55 benchmark data set. 116 The MADs for the non-covalent interaction
10
GFN2-xTB PM6-D3H4X
9
GFN-xTB DFTB3-D3(BJ)
8
7
-1
MAD / kcal mol
0
AD
PN
S2
S6
W
AL
AT
2
6
IM
IC
59
ER
O
6
23
27
Figure 4: Mean absolute deviations (MADs) in kcal mol−1 for the non-covalent association
energies of different benchmark sets from the GMTKN55 database 116 .
energies are collected in Figures 4 and 6. Except for the weak interactions in the alkane dimer set
ADIM6, 116 GFN2-xTB always ranks first or second among the considered methods. The benefit
of including higher multipole moments is already observable for the often considered (also for
29
fitting) S22 112,117 and S66 113 sets. Here, the GFN2-xTB method outperforms GFN-xTB and
DFTB-D3(BJ) without specific Hamiltonian terms to treat hydrogen bonds. Only the PM6-D3H4X
method shows a vanishingly lower MAD in both sets, which is expected as these sets had been used
to adjust the D3 and H4 parameters. 38
However, the advantages of including higher multipole electrostatic terms become particularly obvi-
ous when looking at the HAL59 115,116,118 and PNICO23 116,119 benchmarks. In these benchmarks,
the anisotropic electron density of the bonded halogen and pnicogen atoms is very important.
Though, the monopole-based tight-binding methods can perform well for either of the sets due their
parametrization or specific halogen bond corrections (in GFN-xTB), a consistently good descrip-
tion is only observed for GFN2-xTB. This demonstrates the more physical behavior of GFN2-xTB,
which shows a better performance without special correction terms. PM6-D3H4X performs bad for
both benchmark sets, particularly for HAL59 in which seven systems yield an error > 10 kcal mol−1 .
Noteworthy is the low MAD of GFN2-xTB for the WATER27 set, which is obtained without
any additional H-bond specific correction term. In fact, the MAD is even lower than for well-
performing generalized gradient density functional approximations (GGAs) like BLYP-D3(BJ)
(MAD of 4.1 kcal mol−1 ) 116 or some hybrid functionals (e.g., PBE0-D3(BJ) with an MAD of
5.9 kcal mol−1 ) 116 .
a) H-bonded b) vdW-bonded
GFN2-xTB GFN2-xTB
GFN-xTB
PM6-D3H4X DFTB3-D3(BJ)
GFN-xTB
PM6-D3H4X
DFTB3-D3(BJ)
-4 -2 0 2 4 -4 -2 0 2 4
error in kcal mol-1 error in kcal mol-1
Figure 5: Normal distribution plots for the errors

1
in the computed interaction energies the
113
S66 benchmark. Plot a) refers to the hydrogen bonded (first 23) systems, whereas plot b)
refers to van-der-Waals-type bonded systems.
30
In Figure 5, the results for the S66 benchmark set are processed in more detail. This figure shows
the error distribution plots subdivided for the hydrogen bonds (left hand side) and predominantly
van-der-Waals-type interacting systems (right hand side). The MD for GFN2-xTB is significantly
improved compared to the monopole-based GFN-xTB for the latter, which indicates the impor-
tance of AES for such systems. DFTB3-D3(BJ) also works well there – probably due to the pair-
specifically parametrized repulsion potentials – but shows larger deviations for the hydrogen bonded
systems. Different from that, GFN2-xTB produces consistently small errors (|MD| < 1 kcal mol−1
for both subsets). PM6-D3H4X also performs extremely well for both subsets.
In the following, other benchmark sets for non-covalent interactions from the GMTKN55 database
are considered (see Figure 6). These sets consist of more difficult systems (charged) or elements
12
GFN2-xTB
10 GFN-xTB
PM6-D3H4X
-1
8 DFTB3-D3(BJ)
MAD / kcal mol
0 * * * * *
AH
IL
R
AR
EA
G
16
B2
B6
18
VY
BH
1
2
B1
8
2
Figure 6: Mean absolute deviations (MADs) in kcal mol−1 for the non-covalent association
energies of different benchmark sets from the GMTKN55 database. 116 These sets contain
charged systems or heavy main group elements for which DFTB3 does not have parameters
in the 3OB parametrization 58–60 (indicated by an asterisk).
other than from the first and second row. Though GFN-xTB performs slightly better for the cationic
hydrogen bonded systems in CHB6 and the carbene-hydrogen bonded systems in CARBHB12,
GFN2-xTB overall performs best and no outlier is observed, also for the heavy main group and
31
noble gas elements. While DFTB3-D3(BJ) shows low MADs for the ionic liquids (IL16) and anionic
hydrogen bonds (AHB21), the lack of parameters severely limits the applicability of the method.
PM6-D3H4X on the other hand consistently shows bigger MADs than GFN-xTB, as well as GFN2-
xTB.
Having demonstrated the good performance of GFN2-xTB for non-covalent interactions of small
systems including different elements and interaction types, we next turn our attention to larger
systems. For this purpose, the S30L set 120 is considered, which consists of 30 large non-covalently
bound neutral and charged complexes. The results for GFN2-xTB are compared with PM6-D3H4X
and DFTB3-D3(BJ) in Figure 7. GFN-xTB performs similar to DFTB3-D3(BJ) and is excluded
from this figure for clarity (see SI for the GFN-xTB results). It is seen that GFN2-xTB closely
−20
−40
−60
∆ E / kcal mol−1
reference MD MAD SD
−80
GFN2-xTB −0.6 4.0 5.1
−100 PM6-D3H4X −3.9 5.1 7.5
−120
DFTB3-D3(BJ) −4.3 6.9 8.7
−140
dispersion bound halogen H bonds charged

−160
bonds
−180
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
system
Figure 7: Association energies of the S30L 1201 benchmark set computed with GFN2-xTB,
PM6-D3H4X, and DFTB3-D3(BJ). The values and statistical data is given in kcal mol−1 .
resembles the reference values, i.e., domain-based pair natural orbital coupled cluster with singles,
32
doubles, and perturbative triples – short DLPNO-CCSD(T) 121 – extrapolated to the complete
basis set (CBS) limit. Overall, it shows the smallest (in magnitude) MD, MAD, and SD of all
methods considered and is roughly on a par with some dispersion-corrected density functionals
in a large quadruple-ζ basis set (e.g., B3LYP-D3(BJ)/def2-QZVP). 120 The nearly non-detectable
deviation for the charged systems is striking and even the largest association energy of about
−135.5 kcal mol−1 (system 24) is reproduced to within <1% deviation, while the other semiempiri-
cal methods show errors of about 20%. This is a reassuring result and reflects the well-described elec-
trostatic and polarization interactions in GFN2-xTB. Actually, the largest deviations are observed
for the van-der-Waals-dominated complexes of conjugated π-systems (systems 7–12). GFN2-xTB
overestimates the magnitude for the association energy of these complexes. This behavior partially
results from non-additivity dispersion effects and the ATM term, which is also included in GFN2-
xTB (see Section 2.2.6), only partially compensates for this. This has already been observed and
discussed for DFT-D3 methods (see the comment in reference 93 of Ref. 76), and hence, does not
represent a weakness specific to the GFN2-xTB Hamiltonian.
3.3 Conformational energies
For the description of conformational energies, the balance between covalent and (intramolecu-
lar) non-covalent forces is very important. The proper energy ranking of conformers is essential
for SQM methods, which can efficiently be used to sample the conformational space of chemi-
cally important, moderately sized systems (<100-200 atoms). As mentioned before, determining
the thermostatistically populated conformer-rotamer-ensemble for the calculation of spin-coupled
nuclear-magnetic resonance spectra has been a driving force for the development of the GFN2-xTB
method, and the AES term in particular. In Figure 8, MAD bar plots for the conformational
energies of different benchmark sets from the GMTKN55 116 are given (see SI for detailed val-
ues). It can be seen that the GFN2-xTB method ranks best in five out of eight sets, namely for
the ACONF 122 , Amino20x4, 123 , ICONF, 116 PCONF21, 116,124,125 and SCONF. 116,126 Particularly
positive is the considerable improvement over GFN-xTB for more polar and hydrogen-bonded sys-
tems in Amino20x4, PCONF21, and SCONF. This suggests a generally better performance for
sugars and polypeptide systems. Furthermore, the GFN2-xTB MAD of 1.6 kcal mol−1 for the
33
5
GFN2-xTB PM6-D3H4X
4.5
GFN-xTB DFTB3-D3(BJ)
4
3.5
MAD / kcal mol-1
2.5
1.5
0.5
0
AC
Am
BU
IC
PC
SC
U
PU
C
O
O
T1
O
in
O
N
23
N
N
o2
N
F
4D
F
F2
F
F
0x
IO
1
4
Figure 8: Mean absolute deviations (MADs) in kcal mol−1 for conformational energy bench-
mark sets. The structures and reference data are taken from the GMTKN55 data base. 116
Detailed values are listed in the SI.
ICONF stands out, as DFTB3-D3(BJ), which ranks second, has an MAD that is larger by almost
1 kcal mol−1 . Also for the BUT14DIOL 116,127 and MCONF 116,128 sets, the GFN2-xTB method
performs quite similar to the other SQM methods. The only outlier with an MAD of 2.9 kcal mol−1
is the UPU23 set, 116,129 which is about twice as large as the MADs of either GFN-xTB or DFTB3-
D3(BJ). The latter both perform particularly well for this set, whereas PM6-D3H4X is only slightly
(< 0.5 kcal mol−1 ) better than GFN2-xTB. All in all, GFN2-xTB shows the best performance for
the conformer sets considered here.
Recently, a set of glucose and maltose conformers has been compiled. 130 These sugar conformers
are particularly challenging due to the many differently hydrogen bonded conformers. In Figure 9,
conformational energies of both sets are plotted against the high level reference data (CBS extrap-
olated DLPNO-CCSD(T) values). 130 First of all, it is observed that, on average, all SQM methods
underestimate the conformational energies in particular of high-lying conformers. In agreement
with the aforementioned results for the SCONF set, GFN2-xTB outperforms the other methods in
34
a) conformational energies of α- and β-glucose / kcal mol
-1 b) conformational energies of α-maltose / kcal mol
-1
15 20
GFN2-xTB GFN2-xTB
MD: −3.07 15 MD: −2.87
10 MAD: 3.24 MAD: 3.11
SD: 1.95 SD: 2.44
semiempirical QM method
GFN-xTB 10 GFN-xTB
MD: −3.79 MD: −3.27
5
MAD: 4.57 MAD: 3.81
SD: 3.55 5 SD: 3.47
PM6-D3H4X PM6-D3H4X
0 MD: −6.01 MD: −3.64

MAD: 7.05 0 MAD: 5.07
SD: 5.40 SD: 5.63
DFTB-D3(BJ) DFTB-D3(BJ)
-5
-5 MD: −4.40 MD: −4.39
MAD: 4.42 MAD: 4.44
SD: 2.13 SD: 2.55
-10
-10
-10 -5 0 5 10 15 -10 -5 0 5 10 15 20
DLPNO-CCSD(T)/CBS reference DLPNO-CCSD(T)/CBS reference
1 1
Figure 9: Correlation plots for a) the conformational energies of 80 α-glucose and 76 β- 1
glucose conformers and b) the conformational energies of 205 α-maltose conformers. The
energies are given in kcal mol−1 and are computed with different SQM methods. The struc-
tures and reference conformational energies are taken from Ref. 130.
all statistical measures considered. Remarkable are the small SD values, which only DFTB3-D3(BJ)
is nearly on par with. The results for these sugar conformers are encouraging that GFN2-xTB will
be able to provide more reliable conformer-rotamer ensembles than GFN-xTB even for polar and
hydrogen bonded systems.
3.4 Rotational and vibrational free energy computations
Due to computational efficiency in the calculation of gradients and numerical Hessians, a major ap-
plication of GFN2-xTB will likely be the computation of free energy corrections in thermochemical
studies. This has already been pointed out for GFN-xTB 34 and we will crosscheck the performance
of GFN2-xTB against GFN-xTB and PBEh-3c (frequencies scaled by 0.95). 4 The latter is a a hy-
brid density functional composite method, which is computationally considerably more demanding
than the semiempirical tight binding methods (roughly two orders of magnitude). As reference
method, another composite method called B97-3c, 6 is used, which employs a GGA functional in a
triple-ζ basis set. As a “real-life” measure, differences for reactions in the summed rotational and
35
vibrational free energies termed ∆GRRHO at T = 298.15 K are considered. While the rotational
part is a measure for the differences in optimized structures, the vibrational contribution contains
information about the PES curvature around the minima. We use the systems from the following
sets, which are part of the GMTKN55 benchmark set database: 116 AL2X6, DARC, HEAVYSB11,
ISOL24, TAUT15, ALK8, G2RC, and ISO34. Furthermore the recently proposed MOR41 set for
closed-shell metal organic reactions is included. 131 The detailed results are listed in the SI. It
should be noted that translational free energy contribution, which is independent of the electronic
structure method used, is purposely excluded, since the number of reactants differs from the num-
ber of products for many of the reactions considered. This way, the magnitudes for ∆GRRHO are
more similar across the different sets. Furthermore for all methods, the harmonic oscillator/free
rotor interpolation from Ref 132 is applied for harmonic frequencies with magnitudes smaller than
50 cm−1 . The results are plotted in Figure 10. The idea is that both composite “3c” methods
Deviation of ∆GRRHO to B97-3c in kcal/mol

−3.0 −2.5 −2.0 −1.5 −1.0 −0.5 0.5 1.0 1.5 2.0 2.5 3.0
max. PBEh-3c

50%
mean GFN2-xTB
min.
GFN-xTB
Figure 10: Deviation plot for the rotational and vibrational reaction free energy computed
with the tested methods compared to B97-3c. 6 The MD as well as maximum and minimum
deviation is shown for each method. The boxes show the range in which the smallest half of
the errors is found. For PBEh-3c, the frequencies have been scaled by 0.95.
represent comparably accurate methods and it will depend on the system, which one performs
better. As expected, GFN-xTB shows slighty larger errors and only has a somewhat larger spread
of errors. Similar is true also for the GFN2-xTB method. In fact the 50% cases with smallest
deviations for both tight binding methods are found within a range that is about twice as large
compared to PBEh-3c. However in absolute numbers, this error is on a 0.1 kcal mol−1 scale and
36
hence practically irrelevant, given that some deviations to B97-3c are much larger for the tight
binding methods as well as for PBEh-3c. Therefore, GFN2-xTB, just like GFN-xTB, should be a
reasonable method of choice to routinely compute harmonic frequencies for subsequent thermosta-
tistical treatments (including transition metals complexes). This is particularly important, since
such calculations with ab initio and DFT methods quickly become the computational bottleneck
in typical workflows.
3.5 Other properties
In the previous section, we have assessed GFN2-xTB for well-established benchmark sets that have
also been used to study DFT approximations. Though not training sets, these benchmark sets
predominantly coincided with the target properties of the GFN2-xTB method. In this section,
GFN2-xTB is tested for some off-target properties. As in Ref. 34, we investigate the performance
of GFN2-xTB for covalently bonded diatomics in Figure 11. Similar to GFN-xTB, we observe a
150
100
50
relative energy / kcal mol−1
−50
−100
−150
H2
−200
F2
−250
N2
−300
LiH
−350
1 2 3 4 5 6
R/Å
Figure 11: Potential energy curves computed with GFN2-xTB for the dissociation of H2 ,
F2 , N2 , and LiH. The electronic temperature treatment (Tel = 300 K) allows the homolytic
dissociation without a multi-reference treatment. The points mark the minimum energy
positions from high-level calculations. 34,133,134 The energies are given relative to the free
atoms (S=3/2 for nitrogen, S=1/2 for the others).
systematic overestimation of typical bond dissociation energies, while the minimum positions are
reproduced rather well (high level reference data are marked with crosses). This agrees well with
the observations made in Section 3.1, i.e., that GFN2-xTB on average is on a par with GFN-xTB
37
in the description of covalently bonded molecular structures. At the same time, GFN2-xTB shares
the property of overestimating covalent bond energies. However, a slight difference compared to
GFN-xTB (cf. Ref. 34) is observable for the non-polar, single-bonded diatomics (H2 and F2 ). Here
the overestimation is slightly less pronounced than with GFN-xTB, while the triple bond energy
of N2 and the polar LiH bond energy show about the same magnitude as GFN-xTB. It is to be
determined in the future, how this will affect, e.g., the simulation of mass spectra or reaction
enthalpies (within the correction scheme applied in Ref. 45).
Next, we turn our attention to some of the kinetics-oriented benchmark sets of the GMTKN55 116
database. In Figure 12, the MADs for different barrier heights are presented. For three sets,
GFN2-xTB
16
GFN-xTB
14
* PM6-D3H4X
12
MAD / kcal mol-1
DFTB3-D3(BJ)
10
*
8
6
*
4
0 * * * * *
BH
BH
BH
IN
PX
W
C
V2
13
D
PE
PT
IV
4
R
T2
18
10
Figure 12: Mean absolute deviations (MADs) for reaction barriers (in kcal mol−1 ) computed
with different semiempirical methods. Due to missing parameters for silicon, DFTB3 calcu-
lations are not possible for two systems in BHDIV10 116 and one system in the BHPERI 135
set. Furthermore, one extreme outlier (PCl3 ) in the INV24 136 was found. In both cases,
these systems are removed from the statistical analysis for DFTB3-D3(BJ).
systems are neglected in the statistical analysis for DFTB3-D3(BJ), i.e., the Si-containing systems
in BHDIV10 and BHPERI 135 (missing parameters) and also a severe outlier in the INV24 136 set.
Here, the planar PCl3 transition state structure yielded a preposterously large repulsion energy
(> 104 Hartree).
Along all sets considered, the GFN2-xTB performs best. It shows the lowest MAD for the BH-
38
DIV10, 116 BHROT27, 116 INV24, 136 , and PX13, 116,137 sets and the second lowest for the WCPT18. 116,138
Only for the barrier heights of pericyclic reactions (BHPERI), 135 GFN2-xTB is outperformed
by the other SQM methods, though GFN-xTB and PM6-D3H4X are only better by about 1–
1.5 kcal mol−1 . The performance of GFN2-xTB for all sets in Fig. 12 is remarkable, given that
none of these (or similar systems) were used fitting. In particular for the proton transfer sets PX13
and WCPT18, as well as the single bond rotation set BHROT27, GFN2-xTB shows MADs, which
are comparable to or better than those of the hybrid functional PBE0. 116 This – along with the low
MAD for the Amino20x4, 123 , PCONF21, 116,124,125 and WATER27 139 sets (see above) – indicates
that GFN2-xTB may be well-suited to study biomolecular systems in aqueous solution.
As a last test, we investigate the ability to reproduce the permanent electric dipole moments of small
molecules which is relevant for obtaining good long-range non-covalent interactions. Such a set with
purely theoretical reference values has recently been compiled by Halt and Head-Gordon. 140 .
39
molecular dipole moments [D]
10
8
H2O−Li
6 ◦
MAD [D] RMSE [%]

GFN2-xTB 0.45 41.1
2
GFN-xTB 0.69 54.4
PM6 0.52 64.2
0
0 2 4 6 8 10
CCSD(T)/CBS reference
1
Figure 13: Permanent molecular dipole moments computed for open, as well as closed shell
systems with GFN2-xTB, GFN-xTB, and PM6-D3H4X. The benchmark set (structures and
reference values) are taken from from Ref. 140. Detailed values are listed in the SI. The root
mean square error (in %) is regularized with a value of 1 Debye.
In Figure 13, the correlation plot for the computed molecular dipole moments is shown. Due
to missing parameters for many elements, DFTB3 is excluded here. While dipole moments have
been used in the parametrization procedure of PM6, 35 none were used in the fit of GFN-xTB and
GFN2-xTB. Nevertheless, PM6 does not show significantly better agreement with the high level
reference. In fact, GFN2-xTB shows a lower MAD, as well as a lower regularized root mean square
error (RMSE). As in Ref. 140, a value of 1 Debye is used for regularization. The RMSE is close
to the one for MP2 (37.5%) as presented in the original work. 140 This is an encouraging finding,
given that the set covers 14 different elements and almost half of the systems have an open-shell
40
ground state. Our findings provide further support for the improved physics in the GFN2-xTB
Hamiltonian, as well as for our parametrization strategy.
Finally, it should be mentioned that in comparison to GFN-xTB, no increase in the computational
time is expected for typical organic and biomolecular systems. As such, a single-point energy
calculation for crambin (641 atoms) takes 30 s with GFN2-xTB, but instead the same calculation
requires 50 s with GFN-xTB (single-core run on a laptop with a 1.6 GHz Intel Core i5 CPU and 8 GB
RAM). Though the additional integral evaluation (dipole and quadrupole one-electron integrals)
in GFN2-xTB is more elaborate, the rate-determining step in both methods is the diagonalization
of the tight-binding Hamiltonian matrix. Due to the extra s-function for hydrogen, 34 the matrix
dimension in GFN-xTB is significantly larger for typical organic and biomolecular systems (by a
factor of about 1.5), and hence the computation time is even reduced for GFN2-xTB compared to
GFN-xTB.
The method is implemented in the standalone xtb code, which can be requested from the authors. 141
4 Conclusions
We developed a broadly applicable semiempirical quantum mechanical method, termed GFN2-xTB,
which represents the first tight-binding method to include electrostatic and exchange-correlation
Hamiltonian terms beyond the monopole approximation. The method is free from any hydrogen or
halogen bond specific corrections, which are a standard add-on in other contemporary semiempirical
schemes. Furthermore, the self-consistent D4 dispersion model is an inherent part of the GFN2-xTB
method and allows to efficiently incorporate electronic structure effects on the two-body dispersion
energy. The GFN2-xTB method relies strictly on element-specific and global parameters and is
parametrized for all elements up to radon (Z = 86). Like for its predecessor, GFN-xTB, the
parameters were fitted to yield reasonable structures, vibrational frequencies and non-covalent
interactions for molecules across the periodic table. The main focus of this method are organic,
organometallic, and biochemical systems on the order of a few thousand atoms. In particular,
the greatly improved non-covalent interactions will likely trigger structural searches and studies of
conformational and protein-ligand studies in the near future.
41
Apart from these, the improved electrostatics and more consistent parametrization procedure has
provided a method, which better reproduces the electronic density compared to other semiempirical
methods. The method may thus qualify to be used in docking procedures by providing reasonable
electrostatic potentials.
However, as a difficult-to-quantify drawback compared to GFN-xTB we finally note sometimes
less robust SCF convergence in particular for metallic systems or polar inorganic clusters probably
caused by the short-range part of the AES potential. Further work to improve the GFN family of
methods for this field of application is in progress.
Acknowledgments
This work was supported by the DFG in the framework of the “Gottfried Wilhelm Leibniz Prize”
awarded to S.G. The authors would like to thank Prof. Dr. Thomas Bredow, Dr. Jan Gerit Bran-
denburg, and Dr. Andreas Hansen for helpful discussions during the development of the GFN2-xTB
method. Furthermore, the authors are grateful to Eike Caldeweyher for providing routines, which
are used in the self-consistent D4 treatment.
Associated Content
The derivations for the newly developed AES and AXC, as well as the self-consistent D4 dispersion
energy expressions, along with the respective TB-Fock matrix contributions, and nuclear gradients
are shown in the supporting information. Element-specific parameters and the detailed results
obtained on the considered test sets are given here as well.
References
(1) Grimme, S.; Schreiner, P. R. Computational Chemistry: The Fate of Current Methods and
Future Challenges. Angew. Chem. Int. Ed. 2017, 57, 4170–4176.
42
(2) Houk, K. N.; Liu, F. Holy Grails for Computational Organic Chemistry and Biochemistry.
Acc. Chem. Res. 2017, 50, 539–543.
(3) Sure, R.; Grimme, S. Corrected small basis set Hartree-Fock method for large systems. J.
Comput. Chem. 2013, 34, 1672–1685.
(4) Grimme, S.; Brandenburg, J. G.; Bannwarth, C.; Hansen, A. Consistent structures and
interactions by density functional theory with small atomic orbital basis sets. J. Chem.
Phys. 2015, 143, 054107.
(5) Brandenburg, J. G.; Caldeweyher, E.; Grimme, S. Screened exchange hybrid density func-
tional for accurate and efficient structures and interaction energies. Phys. Chem. Chem. Phys.
2016, 18, 15519–15523.
(6) Brandenburg, J. G.; Bannwarth, C.; Hansen, A.; Grimme, S. B97-3c: A revised low-cost
variant of the B97-D density functional method. J. Chem. Phys. 2018, 148, 064104.
(7) Otero-de-la Roza, A.; DiLabio, G. A. Transferable Atom-Centered Potentials for the Cor-
rection of Basis Set Incompleteness Errors in Density-Functional Theory. J. Chem. Theory
Comput. 2017, 13, 3505–3524.
(8) Witte, J.; Neaton, J. B.; Head-Gordon, M. Effective empirical corrections for basis set su-
perposition error in the def2-SVPD basis: gCP and DFT-C. J. Chem. Phys. 2017, 146,
234105.
(9) Hostaš, J.; Řezáč, J. Accurate DFT-D3 Calculations in a Small Basis Set. J. Chem. Theory
Comput. 2017, 13, 3575–3585.
(10) Kulik, H. J.; Seelam, N.; Mar, B. D.; Martinez, T. J. Adapting DFT+U for the Chemically
Motivated Correction of Minimal Basis Set Incompleteness. J. Phys. Chem. A 2016, 120,
5939–5949.
(11) Salsbury, F. R. Molecular dynamics simulations of protein dynamics and their relevance to
drug discovery. Curr. Opin. Pharmacol. 2010, 10, 738–744.
43
(12) Kussmann, J.; Ochsenfeld, C. Pre-selective screening for matrix elements in linear-scaling
exact exchange calculations. J. Chem. Phys. 2013, 138, 134114.
(13) Ufimtsev, I. S.; Martinez, T. J. Quantum Chemistry on Graphical Processing Units. 3. Ana-
lytical Energy Gradients, Geometry Optimization, and First Principles Molecular Dynamics.
J. Chem. Theory Comput. 2009, 5, 2619–2628.
(14) Luehr, N.; Ufimtsev, I. S.; Martinez, T. J. Dynamic Precision for Electron Repulsion Integral
Evaluation on Graphical Processing Units (GPUs). J. Chem. Theory Comput. 2011, 7, 949–
954.
(15) Schütt, O.; Messmer, P.; Hutter, J.; VandeVondele, J. Electronic Structure Calculations on
Graphics Processing Units; John Wiley & Sons, Ltd, 2016; pp 173–190.
(16) Yasuda, K.; Maruoka, H. Efficient calculation of two-electron integrals for high angular basis
functions. Int. J. Quantum Chem. 2014, 114, 543–552.
(17) Asadchev, A.; Gordon, M. S. New Multithreaded Hybrid CPU/GPU Approach to Hartree–
Fock. J. Chem. Theory Comput. 2012, 8, 4166–4176.
(18) Yasuda, K. Accelerating Density Functional Calculations with Graphics Processing Unit. J.
Chem. Theory Comput. 2008, 4, 1230–1236.
(19) Maia, J. D. C.; Urquiza Carvalho, G. A.; Mangueira, C. P.; Santana, S. R.; Cabral, L. A. F.;
Rocha, G. B. GPU Linear Algebra Libraries and GPGPU Programming for Accelerating
MOPAC Semiempirical Quantum Chemistry Calculations. J. Chem. Theory Comput. 2012,
8, 3072–3081.
(20) Wu, X.; Koslowski, A.; Thiel, W. Semiempirical Quantum Chemical Calculations Accelerated
on a Hybrid Multicore CPU–GPU Computing Platform. J. Chem. Theory Comput. 2012,
8, 2272–2281.
(21) Kussmann, J.; Ochsenfeld, C. Hybrid CPU/GPU Integral Engine for Strong-Scaling Ab Initio
Methods. J. Chem. Theory Comput. 2017, 13, 3153–3159.
44
(22) Kalinowski, J.; Wennmohs, F.; Neese, F. Arbitrary Angular Momentum Electron Repulsion
Integrals with Graphical Processing Units: Application to the Resolution of Identity Hartree–
Fock Method. J. Chem. Theory Comput. 2017, 13, 3160–3170.
(23) van Schoot, H.; Visscher, L. Electronic Structure Calculations on Graphics Processing Units;
John Wiley & Sons, Ltd, 2016; pp 101–114.
(24) Ufimtsev, I. S.; Martı́nez, T. J. Graphical Processing Units for Quantum Chemistry. Comput.
Sci. Eng. 2008, 10, 26–34.
(25) Jakowski, J.; Irle, S.; Morokuma, K. In GPU Computing Gems Emerald Edition; Hwu, W.-
m. W., Ed.; Applications of GPU Computing Series; Morgan Kaufmann: Boston, 2011; pp
59 – 73.
(26) Thiel, W. Semiempirical quantum–chemical methods. WIREs Comput. Mol. Sci. 2014, 4,
145–157.
(27) Yilmazer, N. D.; Korth, M. Enhanced semiempirical QM methods for biomolecular interac-
tions. Comput. Struct. Biotechnol. J. 2015, 13, 169–175.
(28) Christensen, A. S.; Kubař, T.; Cui, Q.; Elstner, M. Semiempirical Quantum Mechanical
Methods for Noncovalent Interactions for Chemical and Biochemical Applications. Chem.
Rev. 2016, 116, 5301–5337.
(29) Gonzalez-Lafont, A.; Truong, T. N.; Truhlar, D. G. Direct dynamics calculations with NDDO
(neglect of diatomic differential overlap) molecular orbital theory with specific reaction pa-
rameters. J. Phys. Chem. 1991, 95, 4618–4627.
(30) Storer, J. W.; Giesen, D. J.; Cramer, C. J.; Truhlar, D. G. Class IV charge models: A new
semiempirical approach in quantum chemistry. J. Comput. Aid. Mol. Des. 1995, 9, 87–110.
(31) Jakalian, A.; Jack, D. B.; Bayly, C. I. Fast, efficient generation of high-quality atomic
charges. AM1-BCC model: II. Parameterization and validation. J. Comput. Chem. 2002,
23, 1623–1641.
45
(32) Cramer, C. J.; Truhlar, D. G. AM1-SM2 and PM3-SM3 parameterized SCF solvation models
for free energies in aqueous solution. J. Comput. Aid. Mol. Des. 1992, 6, 629–666.
(33) Grimme, S.; Bannwarth, C. Ultra-fast computation of electronic spectra for large systems by
tight-binding based simplified Tamm-Dancoff approximation (sTDA-xTB). J. Chem. Phys.
2016, 145, 054103.
(34) Grimme, S.; Bannwarth, C.; Shushkov, P. 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. 2017, 13, 1989–2009.
(35) Stewart, J. J. P. Optimization of parameters for semiempirical methods V: Modification of
NDDO approximations and application to 70 elements. J. Mol. Model. 2007, 13, 1173.
(36) Korth, M.; Pitoňák, M.; Řezáč, J.; Hobza, P. A Transferable H-Bonding Correction for
Semiempirical Quantum-Chemical Methods. J. Chem. Theory Comput. 2010, 6, 344–352.
(37) Kromann, J. C.; Christensen, A. S.; Steinmann, C.; Korth, M.; Jensen, J. H. A third-
generation dispersion and third-generation hydrogen bonding corrected PM6 method: PM6-
D3H+. PeerJ 2014, 2, e449.
(38) Řezáč, J.; Hobza, P. Advanced Corrections of Hydrogen Bonding and Dispersion for Semiem-
pirical Quantum Mechanical Methods. J. Chem. Theory Comput. 2012, 8, 141–151.
(39) S. Brahmkshatriya, P.; Dobeš, P.; Fanfrlı́k, J.; Řezáč, J.; Paruch, K.; Bronowska, A.;
Lepšı́k, M.; Hobza, P. Quantum Mechanical Scoring: Structural and Energetic Insights into
Cyclin-Dependent Kinase 2 Inhibition by Pyrazolo[1,5-a]pyrimidines. Curr. Comput.-Aid.
Drug. 2013, 9, 118–129.
(40) Saito, T.; Kitagawa, Y.; Takano, Y. Reparameterization of PM6 Applied to Organic Diradical
Molecules. J. Phys. Chem. A 2016, 120, 8750–8760.
(41) Weber, W.; Thiel, W. Orthogonalization corrections for semiempirical methods. Theor.
Chem. Acc. 2000, 103, 495–506.
46
(42) Tuttle, T.; Thiel, W. OMx-D: semiempirical methods with orthogonalization and dispersion
corrections. Implementation and biochemical application. Phys. Chem. Chem. Phys. 2008,
10, 2159–2166.
(43) Dral, P. O.; Wu, X.; Spörkel, L.; Koslowski, A.; Weber, W.; Steiger, R.; Scholten, M.;
Thiel, W. Semiempirical Quantum-Chemical Orthogonalization-Corrected Methods: Theory,
Implementation, and Parameters. J. Chem. Theory Comput. 2016, 12, 1082–1096.
(44) Koslowski, A.; Beck, M. E.; Thiel, W. Implementation of a general multireference config-
uration interaction procedure with analytic gradients in a semiempirical context using the
graphical unitary group approach. J. Comput. Chem. 2003, 24, 714–726.
(45) Kromann, J.; Welford, A.; Christensen, A. S.; Jensen, J. Random Versus Systematic Errors
in Reaction Enthalpies Computed Using Semi-empirical and Minimal Basis Set Methods.
2018,
(46) Bikadi, Z.; Hazai, E. Application of the PM6 semi-empirical method to modeling proteins
enhances docking accuracy of AutoDock. J. Cheminform. 2009, 1, 15.
(47) Stewart, J. J. P. Application of the PM6 method to modeling proteins. J. Mol. Model. 2009,
15, 765–805.
(48) Miriyala, V. M.; Řezáč, J. Testing Semiempirical Quantum Mechanical Methods on a Data
Set of Interaction Energies Mapping Repulsive Contacts in Organic Molecules. J. Phys.
Chem. A 2018, 122, 2801–2808.
(49) Kazaryan, A.; Lan, Z.; Schäfer, L. V.; Thiel, W.; Filatov, M. Surface Hopping Excited-State
Dynamics Study of the Photoisomerization of a Light-Driven Fluorene Molecular Rotary
Motor. J. Chem. Theory Comput. 2011, 7, 2189–2199.
(50) Spörkel, L.; Cui, G.; Thiel, W. Photodynamics of Schiff Base Salicylideneaniline: Trajectory
Surface-Hopping Simulations. J. Phys. Chem. A 2013, 117, 4574–4583.
(51) Spörkel, L.; Cui, G.; Koslowski, A.; Thiel, W. Nonequilibrium H/D Isotope Effects from
Trajectory-Based Nonadiabatic Dynamics. J. Phys. Chem. A 2014, 118, 152–157.
47
(52) Spörkel, L.; Jankowska, J.; Thiel, W. Photoswitching of Salicylidene Methylamine: A Theo-
retical Photodynamics Study. J. Phys. Chem. B 2015, 119, 2702–2710.
(53) Dokukina, I.; Marian, C. M.; Weingart, O. New Perspectives on an Old Issue: A Comparative
MS-CASPT2 and OM2-MRCI Study of Polyenes and Protonated Schiff Bases. Photochem.
Photobiol. 2017, 93, 1345–1355.
(54) Elstner, M.; Porezag, D.; Jungnickel, G.; Elsner, J.; Haugk, M.; Frauenheim, T.; Suhai, S.;
Seifert, G. Self-consistent-charge density-functional tight-binding method for simulations of
complex materials properties. Phys. Rev. B 1998, 58, 7260–7268.
(55) Yang, Y.; Yu, H.; York, D.; Cui, Q.; Elstner, M. Extension of the Self-Consistent-Charge
Density-Functional Tight-Binding Method: Third-Order Expansion of the Density Func-
tional Theory Total Energy and Introduction of a Modified Effective Coulomb Interaction.
J. Phys. Chem. A 2007, 111, 10861–10873.
(56) Gaus, M.; Cui, Q.; Elstner, M. DFTB3: Extension of the Self-Consistent-Charge Density-
Functional Tight-Binding Method (SCC-DFTB). J. Chem. Theory Comput. 2011, 7, 931–
948.
(57) Niehaus, T. A.; Suhai, S.; Sala, F. D.; Lugli, P.; Elstner, M.; Seifert, G.; Frauenheim, T.
Tight-binding approach to time-dependent density-functional response theory. Phys. Rev. B
2001, 63, 085108.
(58) Gaus, M.; Goez, A.; Elstner, M. Parametrization and Benchmark of DFTB3 for Organic
Molecules. J. Chem. Theory Comput. 2013, 9, 338–354.
(59) Gaus, M.; Lu, X.; Elstner, M.; Cui, Q. Parameterization of DFTB3/3OB for Sulfur and
Phosphorus for Chemical and Biological Applications. J. Chem. Theory Comput. 2014, 10,
1518–1537.
(60) Kubillus, M.; Kubař, T.; Gaus, M.; Řezáč, J.; Elstner, M. Parameterization of the DFTB3
Method for Br, Ca, Cl, F, I, K, and Na in Organic and Biological Systems. J. Chem. Theory
Comput. 2015, 11, 332–342.
48
(61) Gaus, M.; Jin, H.; Demapan, D.; Christensen, A. S.; Goyal, P.; Elstner, M.; Cui, Q. DFTB3
Parametrization for Copper: The Importance of Orbital Angular Momentum Dependence of
Hubbard Parameters. J. Chem. Theory Comput. 2015, 11, 4205–4219.
(62) Vujović, M.; Huynh, M.; Steiner, S.; Garcia-Fernandez, P.; Elstner, M.; Cui, Q.; Gruden, M.
Exploring the applicability of density functional tight binding to transition metal ions. Pa-
rameterization for nickel with the spin-polarized DFTB3 model. J. Comp. Chem. 2018,
(63) Bursch, M.; Hansen, A.; Grimme, S. Fast and Reasonable Geometry Optimization of Lan-
thanoid Complexes with an Extended Tight Binding Quantum Chemical Method. Inorg.
Chem. 2017, 56, 12485–12491.
(64) Struch, N.; Bannwarth, C.; Ronson, T. K.; Lorenz, Y.; Mienert, B.; Wagner, N.; Engeser, M.;
Bill, E.; Puttreddy, R.; Rissanen, K.; Beck, J.; Grimme, S.; Nitschke, J. R.; Lützen, A.
An Octanuclear Metallosupramolecular Cage Designed To Exhibit Spin-Crossover Behavior.
Angew. Chem. Int. Ed. 2017, 56, 4930–4935.
(65) Seibert, J.; Bannwarth, C.; Grimme, S. Biomolecular Structure Information from High-Speed
Quantum Mechanical Electronic Spectra Calculation. J. Am. Chem. Soc. 2017, 139, 11682–
11685.
(66) Pracht, P.; Bauer, C. A.; Grimme, S. Automated and efficient quantum chemical determi-
nation and energetic ranking of molecular protonation sites. J. Comput. Chem. 2017, 38,
2618–2631.
(67) Grimme, S.; Bannwarth, C.; Dohm, S.; Hansen, A.; Pisarek, J.; Pracht, P.; Seibert, J.;
Neese, F. Fully Automated Quantum-Chemistry-Based Computation of Spin–Spin-Coupled
Nuclear Magnetic Resonance Spectra. Angew. Chem. Int. Ed. 2017, 56, 14763–14769.
(68) Asgeirsson, V.; Bauer, C. A.; Grimme, S. Quantum chemical calculation of electron ionization
mass spectra for general organic and inorganic molecules. Chem. Sci. 2017, 8, 4879–4895.
49
(69) Grimme, S.; Bannwarth, C.; Caldeweyher, E.; Pisarek, J.; Hansen, A. A general intermolec-
ular force field based on tight-binding quantum chemical calculations. J. Chem. Phys. 2017,
147, 161708.
(70) Hehre, W. J.; Stewart, R. F.; Pople, J. A. Self-Consistent Molecular-Orbital Methods. I. Use
of Gaussian Expansions of Slater-Type Atomic Orbitals. J. Chem. Phys. 1969, 51, 2657–
2664.
(71) Asgeirsson, V.; Bauer, C. A.; Grimme, S. Quantum chemical calculation of electron ionization
mass spectra for general organic and inorganic molecules. Chem. Sci. 2017, 8, 4879–4895.
(72) Brandenburg, J. G.; Grimme, S. Dispersion Corrected Hartree-Fock and Density Functional
Theory for Organic Crystal Structure Prediction. Top. Curr. Chem. 2014, 345, 1–23.
(73) Humeniuk, A.; Mitrić, R. Long-range correction for tight-binding TD-DFT. J. Chem. Phys.
2015, 143, 134120.
(74) Lutsker, V.; Aradi, B.; Niehaus, T. A. Implementation and benchmark of a long-range cor-
rected functional in the density functional based tight-binding method. J. Chem. Phys. 2015,
143, 184107.
(75) Kranz, J. J.; Elstner, M.; Aradi, B.; Frauenheim, T.; Lutsker, V.; Garcia, A. D.;
Niehaus, T. A. Time-Dependent Extension of the Long-Range Corrected Density Functional
Based Tight-Binding Method. J. Chem. Theory Comput. 2017, 13, 1737–1747.
(76) Grimme, S.; Hansen, A.; Brandenburg, J. G.; Bannwarth, C. Dispersion-Corrected Mean-
Field Electronic Structure Methods. Chem. Rev. 2016, 116, 5105–5154.
(77) Vydrov, O. A.; Van Voorhis, T. Nonlocal van der Waals density functional: the simpler the
better. J. Chem. Phys. 2010, 133, 244103.
(78) Caldeweyher, E.; Ehlert, S.; Hansen, A.; Neugebauer, H.; Bannwarth, C.; Grimme, S.
manuscript in preparation
50
(79) Caldeweyher, E.; Bannwarth, C.; Grimme, S. Extension of the D3 dispersion coefficient
model. J. Chem. Phys. 2017, 147, 034112.
(80) Nishimoto, K.; Mataga, N. Electronic Structure and Spectra of Some Nitrogen Heterocycles.
Z. Phys. Chem. 1957, 12, 335–338.
(81) Ohno, K. Some Remarks on the Pariser-Parr-Pople Method. Theor. Chim. Act. 1964, 2, 219.
(82) Klopman, G. A Semiempirical Treatment of Molecular Structures. II. Molecular Terms and
Application to Diatomic Molecules. J. Am. Chem. Soc. 1964, 86, 4450.
(83) Grimme, S. A General Quantum Mechanically Derived Force Field (QMDFF) for Molecules
and Condensed Phase Simulations. J. Chem. Theory Comput. 2014, 10, 4497–4514.
(84) Mermin, N. D. Thermal Properties of the Inhomogeneous Electron Gas. Phys. Rev. A 1965,
137, 1441–1443.
(85) Roothaan, C. C. J. New Developments in Molecular Orbital Theory. Rev. Mod. Phys. 1951,
23, 69–89.
(86) Hall, G. G. The Molecular Orbital Theory of Chemical Valency. VIII. A Method of Calcu-
lating Ionization Potentials. 1951, 205, 541–552.
(87) Grimme, S.; Antony, J.; Ehrlich, S.; Krieg, H. A consistent and accurate ab initio
parametrization of density functional dispersion correction (DFT-D) for the 94 elements
H-Pu. J. Chem. Phys. 2010, 132, 154104.
(88) Pyykkö, P.; Atsumi, M. Molecular Single-Bond Covalent Radii for Elements 1–118. Chem.
Eur. J. 15, 186–197.
(89) Mantina, M.; Valero, R.; Cramer, C. J.; Truhlar, D. G. In CRC Handbook of Chemistry
and Physics, 91nd edition; Haynes, W. M., Ed.; CRC Press: Boca Raton, FL, 2010; pp
9–49–9–50.
(90) Köhler, C.; Seifert, G.; Frauenheim, T. Density functional based calculations for Fen (n ≤32).
Chem. Phys. 2005, 309, 23 – 31.
51
(91) Grimme, S. A simplified Tamm-Dancoff density functional approach for the electronic exci-
tation spectra of very large molecules. J. Chem. Phys. 2013, 138, 244104.
(92) Zoltán, B.; Bálint, A. Possible improvements to the self-consistent-charges density-functional
tight-binding method within the second order. Phys. Status Solidi B 2012, 249, 259–269.
(93) Köster, A. M.; Leboeuf, M.; Salahub, D. R. In Molecular Electrostatic Potentials; Mur-
ray, J. S., Sen, K., Eds.; Theoretical and Computational Chemistry; Elsevier, 1996; Vol. 3;
pp 105 – 142.
(94) Axilrod, B. M.; Teller, E. Interaction of the van der Waals Type Between Three Atoms. J.
Chem. Phys. 1943, 11, 299–300.
(95) Muto, Y. Proc. Phys. Math. Soc. Jpn. 1943, 17, 629.
(96) Grimme, S.; Ehrlich, S.; Goerigk, L. Effect of the Damping Function in Dispersion Corrected
Density Functional Theory. J. Comput. Chem. 2011, 32, 1456–1465.
(97) Ghosh, D. C.; Islam, N. Semiempirical evaluation of the global hardness of the atoms of
103 elements of the periodic table using the most probable radii as their size descriptors.
International Journal of Quantum Chemistry 110, 1206–1213.
(98) Levenberg, K. A Method for the Solution of Certain Non-linear Problems in Least Squares.
Q. Appl. Math. 1944, 2, 164–168.
(99) Marquardt, D. An Algorithm for Least-Squares Estimation of Nonlinear Parameters. J. Soc.
Ind. Appl. Math. 1963, 11, 431–441.
(100) TURBOMOLE V7.0 2015, a development of University of Karlsruhe and Forschungszentrum
Karlsruhe GmbH, 1989-2007, TURBOMOLE GmbH, since 2007; available from
http://www.turbomole.com.
(101) Ahlrichs, R.; Bär, M.; Häser, M.; Horn, H.; Kölmel, C. Electronic Structure Calculations on
Workstation Computers: The Program System Turbomole. Chem. Phys. Lett. 1989, 162,
165–169.
52
(102) Furche, F.; Ahlrichs, R.; Hättig, C.; Klopper, W.; Sierka, M.; Weigend, F. Turbomole. WIREs
Comput. Mol. Sci. 2014, 4, 91–100.
(103) Vahtras, O.; Almlöf, J.; Feyereisen, M. W. Integral approximations for LCAO-SCF calcula-
tions. Chem. Phys. Lett. 1993, 213, 514–518.
(104) Eichkorn, K.; Weigend, F.; Treutler, O.; Ahlrichs, R. Auxiliary basis sets for main row atoms
and transition metals and their use to approximate Coulomb potentials. Theor. Chem. Acc.
1997, 97, 119–124.
(105) Weigend, F. Accurate Coulomb-fitting basis sets for H to Rn. Phys. Chem. Chem. Phys.
2006, 8, 1057–1065.
(106) Frauenheim, T. DFTB+ (Density Functional based Tight Binding); DFTB.ORG, Universität
Bremen: Bremen, Germany, 2008; http://www.dftb.org (August 9, 2014).
(107) Stewart, J. J. P. MOPAC2016 ; Stewart Computational Chemistry: Colorado Springs, CO,
USA, 2016; http://OpenMOPAC.net (August 16, 2016).
(108) See http://www.thch.uni-bonn.de/.
(109) Grimme, S.; Steinmetz, M. Effects of London Dispersion Correction in Density Functional
Theory on the Structures of Organic Molecules in the Gas Phase. Phys. Chem. Chem. Phys.
2013, 15, 16031–16042.
(110) Risthaus, T.; Steinmetz, M.; Grimme, S. Implementation of Nuclear Gradients of Range-
Separated Hybrid Density Functionals and Benchmarking on Rotational Constants for Or-
ganic Molecules. J. Comput. Chem. 2014, 35, 1509–1516.
(111) Bühl, M.; Kabrede, H. Geometries of Transition-Metal Complexes from Density-Functional
Theory. J. Chem. Theory Comput. 2006, 2, 1282–1290.
(112) Jurečka, P.; Šponer, J.; Cerny, J.; Hobza, P. Benchmark database of accurate (MP2 and
CCSD(T) complete basis set limit) interaction energies of small model complexes, DNA base
pairs, and amino acid pairs. Phys. Chem. Chem. Phys. 2006, 8, 1985–1993.
53
(113) Řezáč, J.; Riley, K. E.; Hobza, P. S66: A Well-balanced Database of Benchmark Interaction
Energies Relevant to Biomolecular Structures. J. Chem. Theory Comput. 2011, 7, 2427.
(114) Gráfová, L.; Pitoňák, M.; Řezáč, J.; Hobza, P. Comparative Study of Selected Wave Function
and Density Functional Methods for Noncovalent Interaction Energy Calculations Using the
Extended S22 Data Set. J. Chem. Theory Comput. 2010, 6, 2365–2376.
(115) Řezáč, J.; Riley, K. E.; Hobza, P. Benchmark Calculations of Noncovalent Interactions of
Halogenated Molecules. J. Chem. Theory Comput. 2012, 8, 4285–4292.
(116) Goerigk, L.; Hansen, A.; Bauer, C.; Ehrlich, S.; Najibi, A.; Grimme, S. A look at the
density functional theory zoo with the advanced GMTKN55 database for general main group
thermochemistry, kinetics and noncovalent interactions. Phys. Chem. Chem. Phys. 2017, 19,
32184–32215.
(117) Marshall, M. S.; Burns, L. A.; Sherrill, C. D. Basis set convergence of the coupled-cluster
CCSD(T)
correction, δMP2 : Best practices for benchmarking non-covalent interactions and the
attendant revision of the S22, NBC10, HBC6, and HSG databases. J. Chem. Phys. 2011,
135, 194102.
(118) Kozuch, S.; Martin, J. M. L. Halogen Bonds: Benchmarks and Theoretical Analysis. J.
(119) Setiawan, D.; Kraka, E.; Cremer, D. Strength of the Pnicogen Bond in Complexes Involving
Group Va Elements N, P, and As. J. Phys. Chem. A 2015, 119, 1642–1656.
(120) Sure, R.; Grimme, S. Comprehensive Benchmark of Association (Free) Energies of Realistic
Host–Guest Complexes. J. Chem. Theory Comput. 2015, 11, 3785–3801.
(121) Riplinger, C.; Sandhoefer, B.; Hansen, A.; Neese, F. Natural triple excitations in local coupled
cluster calculations with pair natural orbitals. J. Chem. Phys. 2013, 139, 134101.
(122) Gruzman, D.; Karton, A.; Martin, J. M. L. Performance of Ab Initio and Density Functional
Methods for Conformational Equilibria of Cn H2n+2 Alkane Isomers (n = 4 − 8). J. Phys.
Chem. A 2009, 113, 11974–11983.
54
(123) Kesharwani, M. K.; Karton, A.; Martin, J. M. L. Benchmark ab Initio Conformational Ener-
gies for the Proteinogenic Amino Acids through Explicitly Correlated Methods. Assessment
of Density Functional Methods. J. Chem. Theory Comput. 2016, 12, 444–454.
(124) Řeha, D.; Valdés, H.; Vondrášek, J.; Hobza, P.; Abu-Riziq, A.; Crews, B.; de Vries, M. S.
Structure and IR Spectrum of Phenylalanyl–Glycyl–Glycine Tripetide in the Gas-Phase:
IR/UV Experiments, Ab Initio Quantum Chemical Calculations, and Molecular Dynamic
Simulations. Chem. Eur. J. 2005, 11, 6803–6817.
(125) Goerigk, L.; Karton, A.; Martin, J. M. L.; Radom, L. Accurate quantum chemical energies for
tetrapeptide conformations: why MP2 data with an insufficient basis set should be handled
with caution. Phys. Chem. Chem. Phys. 2013, 15, 7028–7031.
(126) Csonka, G. I.; French, A. D.; Johnson, G. P.; Stortz, C. A. Evaluation of Density Functionals
and Basis Sets for Carbohydrates. J. Chem. Theory Comput. 2009, 5, 679–692.
(127) Kozuch, S.; Bachrach, S. M.; Martin, J. M. Conformational Equilibria in Butane-1,4-diol: A
Benchmark of a Prototypical System with Strong Intramolecular H-bonds. J. Phys. Chem.
A 2014, 118, 293–303.
(128) Kozuch, S.; Martin, J. M. L. Spin-component-scaled double hybrids: An extensive search for
the best fifth-rung functionals blending DFT and perturbation theory. J. Comput. Chem.
2013, 34, 2327–2344.
(129) Kruse, H.; Mladek, A.; Gkionis, K.; Hansen, A.; Grimme, S.; Sponer, J. Quantum Chemical
Benchmark Study on 46 RNA Backbone Families Using a Dinucleotide Unit. J. Chem. Theory
Comput. 2015, 11, 4972–4991.
(130) Marianski, M.; Supady, A.; Ingram, T.; Schneider, M.; Baldauf, C. Assessing the Accuracy
of Across-the-Scale Methods for Predicting Carbohydrate Conformational Energies for the
Examples of Glucose and α-Maltose. J. Chem. Theory Comput. 2016, 12, 6157–6168.
55
(131) Dohm, S.; Hansen, A.; Steinmetz, M.; Grimme, S.; Checinski, M. P. Comprehensive Thermo-
chemical Benchmark Set of Realistic Closed-Shell Metal Organic Reactions. J. Chem. Theory
Comput. 2018, 14, 2596–2608.
(132) Grimme, S. Supramolecular binding thermodynamics by dispersion corrected density func-
tional theory. Chem. Eur. J. 2012, 18, 9955–9964.
(133) Bytautas, L.; Ruedenberg, K. Correlation energy extrapolation by intrinsic scaling. IV. Ac-
curate binding energies of the homonuclear diatomic molecules carbon, nitrogen, oxygen, and
fluorine. J. Chem. Phys. 2005, 122, 154110.
(134) Maniero, A. M.; Acioli, P. H. Full configuration interaction pseudopotential determination
of the ground-state potential energy curves of Li2 and LiH. Int. J. Quantum Chem. 2005,
103, 711–717.
(135) Karton, A.; Goerigk, L. Accurate reaction barrier heights of pericyclic reactions: Surpris-
ingly large deviations for the CBS-QB3 composite method and their consequences in DFT
benchmark studies. J. Comput. Chem. 36, 622–632.
(136) Goerigk, L.; Sharma, R. The INV24 test set: how well do quantum-chemical methods describe
inversion and racemization barriers? Can. J. Chem. 2016, 94, 1133–1143.
(137) Karton, A.; O’Reilly, R. J.; Chan, B.; Radom, L. Determination of Barrier Heights for Proton
Exchange in Small Water, Ammonia, and Hydrogen Fluoride Clusters with G4(MP2)-Type,
MPn, and SCS-MPn Procedures–A Caveat. J. Chem. Theory Comput. 2012, 8, 3128–3136.
(138) Karton, A.; O’Reilly, R. J.; Radom, L. Assessment of Theoretical Procedures for Calculating
Barrier Heights for a Diverse Set of Water-Catalyzed Proton-Transfer Reactions. J. Phys.
Chem. A 2012, 116, 4211–4221.
(139) Bryantsev, V. S.; Diallo, M. S.; van Duin, A. C. T.; Goddard, W. A. Evaluation of B3LYP,
X3LYP, and M06-Class Density Functionals for Predicting the Binding Energies of Neutral,
Protonated, and Deprotonated Water Clusters. J. Chem. Theory Comput. 2009, 5, 1016–
1026.
56
(140) Hait, D.; Head-Gordon, M. How Accurate Is Density Functional Theory at Predicting Dipole
Moments? An Assessment Using a New Database of 200 Benchmark Values. J. Chem. Theory
Comput. 2018, 14, 1969–1981.
(141) xtb standalone code (version 6.0). Please contact xtb@thch.uni-bonn.de for the program.
57
Graphical TOC Entry
58
Supporting Information:
GFN2-xTB – an accurate and broadly
parametrized self-consistent tight-binding
quantum chemical method with multipole
electrostatics and density-dependent dispersion
contributions
Christoph Bannwarth,∗,†,‡ Sebastian Ehlert,† and Stefan Grimme∗,†
†Mulliken Center for Theoretical Chemistry, Universität Bonn, Beringstr. 4, 53115 Bonn,
Germany
‡New address: Department of Chemistry, Stanford University, Stanford, CA 94305, United
States of America.
E-mail: christoph.bannwarth@stanford.edu; grimme@thch.uni-bonn.de

Phone: +49-228/73-2351
S-1
Contents
List of Tables S-3
1 Electrostatic and exchange-correlation energy contribution for second or-

der density fluctuations S-6
1.1 Anisotropic electrostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-6
1.1.1 Multipole expansion in two electronic variables . . . . . . . . . . . . . S-6
1.2 Anisotropic XC kernel contribution . . . . . . . . . . . . . . . . . . . . . . . S-9
1.3 Derivation of the potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-11
1.3.1 Anisotropic electrostatic terms . . . . . . . . . . . . . . . . . . . . . . S-12
1.3.2 Anisotropic XC terms . . . . . . . . . . . . . . . . . . . . . . . . . . S-13
1.3.3 Fock matrix elements . . . . . . . . . . . . . . . . . . . . . . . . . . . S-13
1.4 Cartesian gradients of E AES . . . . . . . . . . . . . . . . . . . . . . . . . . . S-15
1.4.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-15
1.4.2 Nuclear gradients for Eqµ . . . . . . . . . . . . . . . . . . . . . . . . . S-18
1.4.3 Nuclear gradients for EqΘ . . . . . . . . . . . . . . . . . . . . . . . . S-18
1.4.4 Nuclear gradients for Eµµ . . . . . . . . . . . . . . . . . . . . . . . . S-19
1.4.5 Additional terms in AES derivatives due to CN-dependence of R0AB . S-20
1.5 Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-21
1.5.1 The D4 dispersion energy . . . . . . . . . . . . . . . . . . . . . . . . S-22
1.5.2 Derivation of the potential . . . . . . . . . . . . . . . . . . . . . . . . S-22
2 Detailed results S-24

2.1 Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-24
2.2 Non-covalent interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-36
2.3 Conformers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-49
2.4 Rotational and vibrational free energy computations . . . . . . . . . . . . . . S-70
2.5 Other properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-77
S-2
3 Element-specific parameters in GFN2-xTB S-87
References S-97
List of Tables
S1 Detailed results for ROT34 set . . . . . . . . . . . . . . . . . . . . . . . . . . S-24

S2 Detailed results for LB12 set . . . . . . . . . . . . . . . . . . . . . . . . . . . S-25
S3 Detailed results for HMGB11 set . . . . . . . . . . . . . . . . . . . . . . . . S-25
S4 Detailed results for TMC32 set . . . . . . . . . . . . . . . . . . . . . . . . . S-26
S9 Detailed results for R160x6 center-of-mass minimum distances . . . . . . . . S-27
S5 Detailed results for S22 structures . . . . . . . . . . . . . . . . . . . . . . . . S-32
S6 Detailed results for S66 center-of-mass minimum distances . . . . . . . . . . S-33
S7 Detailed results for S22 center-of-mass minimum distances . . . . . . . . . . S-34
S8 Detailed results for X40 center-of-mass minimum distances . . . . . . . . . . S-35
S10 Results for the ADIM6 set . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-36
S11 Results for the HAL59 set . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-37
S12 Results for the PNICO23 set . . . . . . . . . . . . . . . . . . . . . . . . . . . S-38
S13 Detailed results for S22 energies . . . . . . . . . . . . . . . . . . . . . . . . . S-39
S14 Association energies for the S66 set . . . . . . . . . . . . . . . . . . . . . . . S-40
S15 Detailed results for WATER27 set . . . . . . . . . . . . . . . . . . . . . . . . S-42
S16 Detailed results for AHB21 set . . . . . . . . . . . . . . . . . . . . . . . . . . S-43
S17 Detailed results for CARBHB12 set . . . . . . . . . . . . . . . . . . . . . . . S-44
S18 Detailed results for CHB6 set . . . . . . . . . . . . . . . . . . . . . . . . . . S-44
S19 Detailed results for HEAVY28 set . . . . . . . . . . . . . . . . . . . . . . . . S-45
S20 Detailed results for the IL16 set . . . . . . . . . . . . . . . . . . . . . . . . . S-46
S21 Detailed results for the RG18 set . . . . . . . . . . . . . . . . . . . . . . . . S-47
S22 Association energies for the S30L set . . . . . . . . . . . . . . . . . . . . . . S-48
S-3
S23 Detailed results for ACONF set . . . . . . . . . . . . . . . . . . . . . . . . . S-49
S24 Detailed results for Amino20x4 set . . . . . . . . . . . . . . . . . . . . . . . S-50
S25 Detailed results for the BUT14DIOL set . . . . . . . . . . . . . . . . . . . . S-52
S26 Detailed results for the ICONF set . . . . . . . . . . . . . . . . . . . . . . . S-54
S27 Detailed results for the MCONF set . . . . . . . . . . . . . . . . . . . . . . . S-54
S28 Detailed results for the PCONF21 set . . . . . . . . . . . . . . . . . . . . . . S-56
S29 Detailed results for the SCONF set . . . . . . . . . . . . . . . . . . . . . . . S-57
S30 Detailed results for the UPU23 set . . . . . . . . . . . . . . . . . . . . . . . S-57
S31 Conformational energies for glucose conformers . . . . . . . . . . . . . . . . S-59
S32 Conformational energies for maltose conformers . . . . . . . . . . . . . . . . S-63
S33 Detailed results for ∆GRRHO on the AL2X6 set . . . . . . . . . . . . . . . . S-70
S34 Detailed results for ∆GRRHO on the DARC set . . . . . . . . . . . . . . . . . S-70
S35 Detailed results for ∆GRRHO on the HEAVYSB11 set . . . . . . . . . . . . . S-71
S36 Detailed results for ∆GRRHO on the ISOL24 set . . . . . . . . . . . . . . . . S-71
S37 Detailed results for ∆GRRHO on the TAUT15 set . . . . . . . . . . . . . . . . S-72
S38 Detailed results for ∆GRRHO on the ALK8 set . . . . . . . . . . . . . . . . . S-73
S39 Detailed results for ∆GRRHO on the G2RC set . . . . . . . . . . . . . . . . . S-73
S40 Detailed results for ∆GRRHO on the ISO34 set . . . . . . . . . . . . . . . . . S-74
S41 Detailed results for ∆GRRHO on the MOR41 set . . . . . . . . . . . . . . . . S-75
S42 Detailed results for the BHDIV10 set . . . . . . . . . . . . . . . . . . . . . . S-77
S43 Detailed results for the BHPERI set . . . . . . . . . . . . . . . . . . . . . . . S-77
S44 Detailed results for the BHROT27 set . . . . . . . . . . . . . . . . . . . . . . S-78
S45 Detailed results for the INV24 set . . . . . . . . . . . . . . . . . . . . . . . . S-79
S46 Detailed results for the PX13 set . . . . . . . . . . . . . . . . . . . . . . . . S-80
S47 Detailed results for the WCPT18 set . . . . . . . . . . . . . . . . . . . . . . S-80
S48 Molecular dipole moments . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-81
S49 Atomic parameters of GFN2-xTB . . . . . . . . . . . . . . . . . . . . . . . . S-87
S-4
S50 Shell parameters of GFN2-xTB . . . . . . . . . . . . . . . . . . . . . . . . . S-90
S-5
1 Electrostatic and exchange-correlation energy con-
tribution for second order density fluctuations
If no integration ranges are given, the integration goes from −∞ until +∞. In density
functional tight-binding, the electrostatic and exchange-correlation contribution to the total
energy resulting from second order density fluctuations is given as. S1
 
(2) (2) 1 ZZ  1 ∂ 2 EXC
EES + EXC = +  δρ(ri )δρ(rj )dri drj (1)
2 rij ∂ρ(ri )∂ρ(rj ) ρ=ρ0
The two terms will be approximated individually in the following.
1.1 Anisotropic electrostatics
1.1.1 Multipole expansion in two electronic variables
We will first assume that the Coulomb forces on the same atomic site can be neglected or
effectively be absorbed in the XC contribution. Then, we will assume distant atoms and
factorize the integral of the Coulomb operator:
R R
(2) 1 X ρA (ri )dri ρB (rj )drj
lim EES ≈ . (2)
RAB →∞ 2 i∈A,j∈B rij
Here, the particle i is in proximity to the site (atomic center) A with ri = RA + rAi and
particle j is close to corresponding center B with rj = RB + rBj , The Coulomb operator can
S-6
be expanded in a Cartesian multipole expansion to (showing all terms up to second order)
X
1 1 ∂ ∂ 1
≈ + (αi − αA ) + (αj − αB )
rij rij ∂αi ∂αj rij
rAi ,rBj =0 α=x,y,z rAi ,rBj =0

1 X ∂ ∂ ∂ ∂ 1
+ (αi − αA ) + (αj − αB ) (βi − βA ) + (βj − βB )
2 ∂αi ∂αj ∂βi ∂βj rij
α,β=x,y,z rAi ,rBj =0
+ ···

1 X (αi − αA ) αij + (αj − αB ) αji
= + 3
rij rij
rAi ,rBj =0 α=x,y,z rAi ,rBj =0
2 ) + (α − α ) (β − β ) (3α β + δ 2

1 X (αi − αA ) (βi − βA ) (3αij βij − δαβ rij j B i A ji ij αβ rij )
+ 5
2 rij
α,β=x,y,z
2 ) + (α − α ) (β − β ) (3α β − δ 2

(αi − αA ) (βj − βB ) (3αij βji + δαβ rij j B j B ji ji αβ rij )
+ 5
rij
rAi ,rBj =0 (3)
+ ···
1 X (αi − αA ) αAB (αj − αB ) αBA

= + 3
+ 3
RAB RAB RAB
α=x,y,z
2 )
1 X (αi − αA ) (βi − βA ) (3αAB βAB − δαβ RAB
+ 5
2 RAB
α,β=x,y,z
2 )
(αj − αB ) (βi − βA ) (3αBA βAB + δαβ RAB
+ 5
RAB
2 )
(αi − αA ) (βj − βB ) (3αAB βBA + δαβ RAB
+ 5
RAB
2 )

(αj − αB ) (βj − βB ) (3αBA βBA − δαβ RAB
+ 5
+ ··· .
RAB
Here, rij = |rj − ri |, RAB = |RB − RA |, and αAB = −αBA = αB − αA . α and β represent
Cartesian components. Energy terms of higher order in rAi and rBj are not shown in Eq. 3
and will be neglected in our approach. This way, all energy terms up to second order are
−3
considered, which ensures that all electrostatic interactions decaying with RAB or slower are
taken into account. Insertion of Eq. 3 into Eq. 2 and integrating over all positive (clamped
S-7
nuclei) and negative (electrons) particles then yields

(2) 1 X qA qB
qB µTA RAB + qA µTB RBA
lim EES = + 3
RAB →∞ 2 A6=B RAB RAB

3 µTA RAB µTB RAB − µTA µB RAB
2
(4)
− 5
RAB
qB RTAB ΘA RAB qA RTAB ΘB RAB
#
+ 5
+ 5
RAB RAB
with
XX
qA = ZA − Pκλ hφλ | φκ i
κ∈A λ
| {z }
Sλκ
 
µαA = α
XX   XX
αA Sλκ − hφλ |αi | φκ i = −
Pκλ  Pκλ DλκA

κ∈A λ
| {z
α
} κ∈A λ (5)
Dλκ
 
 
αβ β α
Pκλ Qαβ
XX XX
θA = Pκλ αA Dλκ + βA Dλκ − αA βA Sλκ − hφλ |αi βi | φκ i
=−

 λκA
κ∈A λ
| {z } κ∈A λ
Qαβ
λκ
and
3 αβ δαβ xx
Θαβ
A = θA −
yy
(θA + θA zz
+ θA ). (6)
2 2
Here, qA , µA , and ΘA are the cumulative atomic monopole (i.e., charge), dipole, and
α
quadrupole moment, respectively. Sλκ , Dλκ , and Qαβ
λκ are the overlap, electric dipole, and
electric quadrupole moments, respectively. The extra index A in the multipole integrals
α
DλκA and Qαβ
λκA indicates that these are evaluated with the origin at the corresponding
α
atomic center, while Dλκ and Qαβ T
λκ are given with origin O = (0 0 0) .
Since the isotropic charge-charge term (first term in Eq. 4) is already captured in a shell-
wise manner, the anisotropic terms represent the newly introduced terms in the anisotropic
electrostatic (AES) energy. Adding a damping function to damp the AES contributions at
short range then leads to the expression given in the manuscript.
S-8
1.2 Anisotropic XC kernel contribution
The second order change in the XC energy takes the form of a static XC kernel fXC (ri , rj ).
In accordance with the (semi-)local density functional origin, a local density approximation
is applied, which restricts this contribution to same-site terms only:

(2) 1 ZZ ∂ 2 EXC
EXC = δρ(ri )δρ(rj )dri drj
2 ∂ρ(ri )∂ρ(rj ) ρ=ρ0
1 ZZ (7)
≈ δ (ri − rj ) fXC (ri , rj ) δρ(ri )δρ(rj )dri drj
2
1Z
= fXC (ri ) δρ2 (ri )dri
2
In the last line, we have dropped the second redundant index of fXC . For distant atoms with
non-overlapping atomic reference densities ρA0 , the integral can be approximately partitioned
in terms of atomic contributions.
(2) 1 XZ X (2),A
EXC ≈ fXC (ri ) δρ2A (ri )dri = EXC (8)
2 A A
The XC kernel is evaluated on the reference density ρ0 (see Eq. 7). Hence, in terms of
spherically symmetric atomic reference densities, there is no angular dependence (cf. spherical
coordinates) of this kernel. So, in terms of spherical coordinates, this integral around the
nuclear center A can be written as (xAi = r sin(ϑ) cos(ϕ), yAi = r sin(ϑ) sin(ϕ), and zAi =
r cos(ϑ)):
2π π ∞ 2
(2),A 1Z Z Z q
EXC = fXC (r)δρ(r, ϑ, ϕ) r2 sin(ϑ)drdϑdϕ (9)
2
0 0 0
We will not consider details regarding specific prefactors or the explicit decay of ρA0 , as the
final energy expression relies on fitted parameters anyhow. Nevertheless, for the monotoni-
cally decaying reference densities, an LDA-based exchange kernel takes the form fXC (r) ∼
−2/3
ρA0 (r) ∼ ecr (c is an atom-specific constant) , i.e., a monotonically increasing function with
S-9
distance from the nucleus. Due to the faster exponential decay of the density fluctuation,
the term in parentheses in Eq. 9 will decay to zero for r → ∞. Next, we decompose this
term into a product of a radial part and an angular part. The angular part will be expressed
by a series of spherical harmonic functions.
q ∞ X
l
al,m Ylm (ϑ, ϕ)
X
fXC (r)δρ(r, ϑ, ϕ) = R(r) (10)
l m=−l
The integral in Eq. 9 then becomes:
2π π
 2 Z∞

∞ Xl
(2),A 1 Z Z X
EXC = al,m Yl (ϑ, ϕ) sin(ϑ)dϑdϕ [R(r)r]2 dr
m  (11)
2 l m=−l
0 0 0
Next, we exploit the orthogonality properties of spherical harmonics, to expand the squared
sum, i.e., we use:
Z2π Zπ
0
Ylm (ϑ, ϕ)Ylm
0 (ϑ, ϕ)sin(ϑ)dϑdϕ = δmm0 δll0 (12)
0 0
This way, the integral in Eq. 13 is equivalent to:
∞ X
l Z2π Zπ Z∞
(2),A 1 2
a2l,m m
[R(r)r]2 dr
X
EXC = [Yl (ϑ, ϕ)] sin(ϑ)dϑdϕ (13)
l m=−l 2
0 0 0
In the derivation of the anisotropic XC contribution in GFN2-xTB, we now make the fol-
lowing approximations: First, we approximate the angular part as:
" #2
R Rπ
2π
m 2 R Rπ
2π
m
[Yl (ϑ, ϕ)] sin(ϑ)dϑdϕ ≈ Yl (ϑ, ϕ)sin(ϑ)dϑdϕ
0 0 0 0
This is a lower bound approximation to the integral, but this way, it becomes possible to
use the same CAMM expressions as in the anisotropic electrostatic case. Then we truncate
the series at l = 2 and express the product of the angular and radial integral as the squared
atomic multipole moment (Mulliken approximation) times an l-dependent constant. The XC
energy to second order in the density fluctuations is then given as a sum over local CAMM
S-10
contributions:  
(2)
|µA |2 + fXC ||ΘA ||2 
X q µA ΘA
f A q 2

EXC ≈  XC A + fXC  (14)
A
| {z }
A
EAXC
The zeroth order monopole term is already incorporated within the short-ranged damping/on-
site contribution in the second order isotropic Coulomb interaction as in DFTB S1,S2 and
GFN-xTB. S3 Here the proportionality constant relates to the chemical hardness of the atom.
The same is true (in a shell-wise manner) in GFN2-xTB. The other terms in Eq. 14 describe
polarization effects of the atomic density up to second order in the multipole expansion and
are included in EAXC in GFN2-xTB.
1.3 Derivation of the potential
A derivation for the tight-binding Fock matrix elements is given below. The final expressions
are given in Eqs. 22, 23, 24, and 25, as well as in the manuscript. To derive the potential for
the Hamiltonian (or Fock) matrix construction, the total energy including orthonormality
constraint needs to be differentiated w.r.t. the molecular orbital coefficients.
  
∂  X XX X
EGFN2-xTB − nj j  cκj cλj Sκλ − 1 = 0 (15)
∂cνi j A,B κ∈A λ∈B
The derivation for an isotropic second and third order density fluctuation tight-binding model
is described extensively in Ref. S2.
S-11
1.3.1 Anisotropic electrostatic terms
The potential of the newly included terms can be derived analogously to Ref. S2.
∂µTB
" !#
∂Eqµ X ∂qB T
= f3 (RAB ) µA RAB − qA RAB
∂cνi A,B ∂cνi ∂cνi
XXh (16)

= f3 (RAD ) qA DTνλD RAD − Sνλ µTA RAD
A,C λ∈C
i
+ f3 (RAC ) qA DTνλC RAC − Sνλ µTA RAC ni cλi
∂µTB
! !
∂Eµµ ∂µB
2
µTA µTA RAB
X
=− f5 (RAB )RAB − 3f5 (RAB ) RAB
XXh
= 3f5 (RAC ) µTA RAC DTνλC RAC − f5 (RAC )RAC
2
µTA DνλC (17)
A,C λ∈C
i
+ 3f5 (RAD ) µTA RAD DTνλD RAD − f5 (RAD )RAD
2
µTA DνλD ni cλi
" #
∂EqΘ X ∂qB T ∂ΘB
= f5 (RAB ) RAB ΘA RAB + qA RTAB RAB
XXh
2 2
= qA f5 (RAD ) Tr(QνλD ) RAD + f5 (RAC ) Tr(QνλC ) RAC
A,C λ∈C (18)

− f5 (RAD )RTAD QνλD RAD − f5 (RAC )RTAC QνλC RAC
i
− Sνλ f5 (RAD )RTAD ΘA RAD + f5 (RAC )RTAC ΘA RAC ni cλi
Here, the terms involving the traces of the Cartesian quadupole moment integral tensors
QνλD and QνλD originate from the trace removal term (c.f. Eq. 6).
S-12
1.3.2 Anisotropic XC terms
The anisotropic XC (AXC) kernel contributions can be derived analogously:
∂EµXC ∂ X µA T
= f µ µA
∂cνi ∂cνi A XC A
!
X µ ∂µA
=2 A
fXC µTA (19)
A ∂cνi

µD T µC T
XX
= −2ni cλi fXC µD DνλD + fXC µC DνκC
C λ∈C
∂EΘXC ∂ X ΘA
= fXC ||ΘA ||2
∂cνi ∂cνi A
αβ
αβ ∂ΘA
X Θ X
=2 A
fXC ΘA
A α,β ∂cνi
∂θαβ xx yy zz
!
∂ (θA + θA + θA )
Θαβ
X Θ X
= A
fXC A 3 A − δαβ (20)
A α,β ∂cνi ∂cνi
XX X ΘD αβ αβ ΘC αβ αβ

= −ni cλi 3fXC ΘD QνλD + 3fXC ΘC QνκC
C λ∈C α,β
Xh Θ ΘC αα
i
fXC Θαα
XX
+ ni cλi D
D Tr(QνλD ) + fXC ΘC Tr(QνκC )
C λ∈C α
1.3.3 Fock matrix elements
Dividing all Eqs. 16–20 by 2ni , we can now obtain the AES and AXC expressions entering
the Fock matrix.
  
 0 IES+IXC AES AXC 

XX   
Hκλ + Fκλ
cλi  + Fκλ + Fκλ  − εi Sκλ  = 0 (21)

C λ∈C
| {z }
aniso
Fκλ
It is more convenient during the SCF calculations to work with the dipole and quadrupole
integrals with the same overall origin (c.f. Eq. 5). By rearrangement of all terms, the respec-
aniso AES AXC
tive Fock matrix elements for the anisotropic contributions Fκλ = Fκλ + Fκλ to terms,
which are proportional to the respective overlap, dipole and quadrupole integrals, yields to
S-13
the following expression:
aniso 1
Fκλ = Sκλ [VS (RB ) + VS (RC )] (22a)
2
1
+ DTκλ [VD (RB ) + VD (RC )] (22b)
2
1 X αβ h αβ i
+ Qκλ VQ (RB ) + VQαβ (RC ) , ∀ κ ∈ B, λ ∈ C (22c)
2 α,β
With the respective integral (overlap, dipole, and quadrupole) proportional potential terms
given as:
Xn h i
VS (RC ) = RTC f5 (RAC )µA RAC
2
− RAC 3f5 (RAC ) µTA RAC − f3 (RAC )qA RAC
A
1
− f5 (RAC )RTAC ΘA RAC − f3 (RAC )µTA RAC + qA f5 (RAC ) R2C R2AC
2 (23)
3 X
− qA f5 (RAC ) αAB βAB αC βC }
2 α,β
µC ΘC T
+ 2fXC RTC µC − fXC RC [3ΘC − Tr(ΘC ) I] RC
Xh
VD (RC ) = RAC 3f5 (RAC ) µTA RAC − f5 (RAC )µA RAC
2
+ f3 (RAC )qA RAC
A
#
2 (24)
X
− qA f5 (RAC )RC RAC + 3qA f5 (RAC )RAC αC αAC
α
µC ΘC
− 2fXC µC + 2fXC [3ΘC − Tr(ΘC ) I] RC
3 1 2

VQαβ (RC )
X
=− qA f5 (RAC ) αAC βAC − RAB
A 2 2
" # (25)
ΘC
3Θαβ Θαα
X
− fXC C − δαβ C
α
S-14
1.4 Cartesian gradients of E AES
A derivation for the nuclear gradients is given below. For the final expressions, see Sec-
tion 1.4.5.
1.4.1 General
∂RAB αAB ∂RBA ∂RAB
=− =− =− (26)
∂αA RAB ∂αA ∂αB
The fn (RAB ) terms given in the manuscript define the damping and distance dependence of
the AES as:
−n
fn (RAB ) = RAB fdamp (an , RAB ) (27)
with
1
fdamp (an , RAB ) = an . (28)
R0AB
1+6 RAB
−n

∂fn (RAB ) ∂RAB ∂fdamp (an , RAB ) −n
= f damp (an , RAB ) + RAB
∂RAB RAB =const. ∂RAB ∂RAB
0
!an
−n−1 2 R0AB −n−1
=− nfdamp (an , RAB )RAB + 6an fdamp (an , RAB ) RAB
RAB
n an
=− fn (RAB ) − [fdamp (an , RAB )fn (RAB ) − fn (RAB )]
RAB RAB
(29)
The above equation would be sufficient if the R0AB were constants. In GFN2-xTB, however,
they are dependent on the coordination numbers of the atoms via:
1 A0
R0AB = R0 + R0B0 (30)
2
Rmax − R0A
R0A0 = R0A + (31)
1 + exp[−β(CNA − Nval − ∆val )]
S-15
Consequently a term, which takes into account the CN-dependence needs to be added, i.e.:
∂R0A0 Rmax − R0A

=β exp[−β(CNA − Nval − ∆val )] (32)
∂CNA [{1 + exp[−β(CNA − Nval − ∆val )]}2
∂R0A0 ∂R0A0 ∂CNA

= (33)
∂αC ∂CNA ∂αC
2
(an , RAB ) AB an −1 ∂R0AB
!
∂fn (RAB ) 6an fdamp
=− n+an R0
∂CNA RAB ∂CNA
2
∂R0A0
!
3an fdamp (an , RAB )
AB an −1

=− n+an R0 (34)
RAB ∂CNA
3an fn (RAB )fdamp (an , RAB ) AB an −1 ∂R0A0
!
=− an R0
RAB ∂CNA
Hence for nuclear gradients of fn (RAB ), this leads to:

∂fn (RAB ) ∂fn (RAB ) ∂RAB ∂fn (RAB ) ∂CNA ∂fn (RAB ) ∂CNB
fn[αA ] (RAB ) = = + +
∂αA ∂RAB RAB =const. ∂αA
∂CNA ∂αA ∂CNB ∂αA
0
(35)
It should be noted that due to the CN-dependence, the last terms also survive if an atom
different from A or B is moved:
∂fn (RAB ) ∂CNA ∂fn (RAB ) ∂CNB

fn[αC ] (RAB ) = + (36)
∂CNA ∂αC ∂CNB ∂αC
The nuclear derivatives of the CAMM expressions yields:
∂qA X X ∂Sλκ X X X ∂Sλκ

= (δAC − 1) Pκλ − δAC Pκλ (37)
∂αC κ∈A λ∈C ∂αC κ∈C B6=C λ∈B ∂αC
S-16
∂µβA β
" #
X X ∂Sλκ ∂Dλκ
= (1 − δAC ) Pκλ βA −
∂αC κ∈A λ∈C ∂αC ∂αC
" β #
XXX ∂Sλκ ∂Dλκ
+ δAC Pκλ δαβ Sλκ + βC − (38)
κ∈C B λ∈B ∂αC ∂αC
β β
X X ∂DλκA X X X ∂DλκC
= (δAC − 1) Pκλ − δAC Pκλ
κ∈A λ∈C ∂αC κ∈C B6=C λ∈B ∂αC
Here, we have made use of the notation of dipole integrals with origin on the respective
atoms, which is more convenient for the evaluation of nuclear gradients.
βγ γ β
∂Sλκ ∂Qβγ
" #
∂θA X X ∂Dλκ ∂Dλκ λκ
= (1 − δAC ) Pκλ βA + γA − βA γA −
∂αC κ∈A λ∈C ∂αC ∂αC ∂αC ∂αC
γ β
∂Sλκ ∂Qβγ
"
XXX ∂Dλκ ∂Dλκ λκ
+ δAC Pκλ βC + γC − βC γC −
∂αC ∂αC ∂αC ∂αC
κ∈C B λ∈B (39)
i
γ β
+ δαβ (Dλκ − γC Sλκ ) + δαγ Dλκ − βC Sλκ
∂Qβγ
λκA ∂Qβγ
Pκλ λκC
X X X X X
= (δAC − 1) Pκλ − δAC
κ∈A λ∈C ∂αC κ∈C B6=C λ∈B ∂αC
Again, we made use of the more convenient notation of quadrupole integrals with origin on
the respective atoms.
The nuclear gradient for the AXC terms are obtained as Tr(P ∂α∂C FAXC )and are trivial due
to their on-site nature. The derivative of the AES energy is given by:
∂EAES ∂Eqµ ∂EqΘ ∂Eµµ

= + + (40)
∂αC ∂αC ∂αC ∂αC
S-17
1.4.2 Nuclear gradients for Eqµ

∂Eqµ ∂ X
= f3 (RAB )qB µTA RAB
∂αC RAB =const. ∂αC A,B
0
"
∂f3 (RAB ) ∂RAB
qB µTA RAB + f3 (RAB )qB µTA
X
=
A,B ∂αC ∂αC
!T 
∂qB ∂µA
+ f3 (RAB ) µA + qB RAB 
∂αC ∂αC
X [α ] T

= C
f3 (RCB ) (qB µC − qC µB ) RCB − f3 (RCB ) (qB µαC − qC µαB )

R0AB =const.
B
X X X ∂Sλκ X
+ Pκλ µB [f3 (RAB )RAB + f3 (RCB )RCB ]
A6=C κ∈A λ∈C ∂αC B
 !T !T 
X X X X ∂DλκA ∂DλκC
− Pκλ qB f3 (RAB ) RAB + f3 (RCB ) RCB 
A6=C κ∈A λ∈C B ∂αC ∂αC
(41)
Here, we exploited the symmetry of ∂Sλκ /∂αC = ∂Sκλ /∂αC . Note that ∂DλκA /∂αC 6=
∂DκλC /∂αC , but that ∂DλκA /∂αC = −∂DκλA /∂αA , which can save a factor of two in the
gradient calculation.
1.4.3 Nuclear gradients for EqΘ
Let us first consider the derivative

∂ RTAB ΘA RAB ∂RTAC
!
∂RAC
=δBC ΘA RAC + RTAC ΘA
∂αC ∂αC ∂αC
∂RTCB
!
∂RCB ∂ΘA
+ δAC ΘC RCB + RTCB ΘC + RTAB RAB
X ∂ΘA (42)
=2 δBC Θαβ αβ T
A βAC − δAC ΘC βCB + RAB RAB
β=x,y,z ∂αC
∂ΘA
=2 δBC RTAC ΘαA − δAC RTCB ΘβC + RTAB RAB
∂αC
,
S-18
with
αβ xx yy zz
∂ΘA 3 ∂θA δαβ ∂ (θA + θA + θA )
= − . (43)
∂αC 2 ∂αC 2 ∂αC
Then the derivative of the charge-quadrupole interaction is given as:

∂EqΘ ∂ X
T

= f 5 (RAB ) q R Θ R
B AB A AB
∂αC RAB =const. ∂αC AB
0
X [α ]
= C
f5 (RCB ) qC RTCB ΘB RCB + qB RTCB ΘC RCB
R0AB =const.
B
i
− 2f5 (RCB ) qC RTCB ΘαB + qB RTCB ΘβC
" #
∂qA T ∂ΘA
RAB ΘB RAB + qB RTAB
X
+ f5 (RAB ) RAB
AB ∂αC ∂αC
X [α ]
= C
f5 (RCB ) qC RTCB ΘB RCB + qB RTCB ΘC RCB

R0AB =const.
B
i
− 2f5 (RCB ) qC RTCB ΘαB + qB RTCB ΘβC
∂Sλκ X h i
f5 (RAB )RTAB ΘB RAB + f5 (RCB )RTCB ΘB RCB
X X X
− Pκλ
"
X qB ∂QλκA
3f5 (RAB )RTAB 2
X X X
− Pκλ RAB − f5 (RAB ) Tr(QλκA ) RAB
A6=C κ∈A λ∈C B 2 ∂αC
#
∂QλκC
+3f5 (RCB )RTCB RCB − 2
f5 (RCB ) Tr(QλκC ) RCB
∂αC
(44)
1.4.4 Nuclear gradients for Eµµ
Let us first consider the derivative of the following terms:

h i  !T   !T 
∂ µTA RAB µTB RAB ∂µA ∂µB
= RAB  µTB RAB + µTA RAB  RAB 

+ (δBC − δAC ) µαA µTB RAB + (δBC − δAC ) µαB µTA RAB
(45)
S-19
!T
∂ µTA µB RAB
2 !
∂µA ∂µB
= 2
µB RAB + µTA 2
RAB + 2 δBC µTA µC αAC − δAC µTC µB αCB
(46)

∂Eµµ ∂ X h i
= f5 (RAB ) µTA µB RAB
2
− 3 µTA RAB µTB RAB
∂αC RAB =const. 2∂αC A,B
0
h
[α ]
µTC µB 2
µTC RCB µTB RCB
X
= f5 C (RCB ) RCB AB −3
R =const.
0
B
h i
f5 (RCB ) 2µTC µB αCB + 3 µαC µTB + µαB µTC RCB
X
−
B
(
∂DλκA X h i
2
− 3RAB µTB RAB
X X X
− Pκλ f5 (RAB ) µB RAB
)
∂DλκC X h i
+ 2
f5 (RCB ) µB RCB − 3RCB µTB RCB
∂αC B
(47)
1.4.5 Additional terms in AES derivatives due to CN-dependence of R0AB
The aforementioned formulas for ∂EAES /∂αC have been given for constant values of R0AB .
Due to their CN-dependence, these terms must, however, be included. Therefore, the afore-
mentioned expression have to be augmented with the respective derivatives:
!
∂Eqµ ∂Eqµ X ∂f3 (RAB ) ∂CNA ∂f3 (RAB ) ∂CNB
= + + (qB µTA RAB ) (48)
∂αC ∂αC RAB =const. A,B ∂CNA ∂αC ∂CNB ∂αC
0
!
∂EqΘ ∂EqΘ X ∂f5 (RAB ) ∂CNA ∂f5 (RAB ) ∂CNB
= + + (qB RTAB ΘA RAB )
∂αC ∂αC RAB =const. A,B
∂CNA ∂αC ∂CNB ∂αC
0
(49)
S-20
!
∂Eµµ ∂Eµµ 1 X ∂f5 (RAB ) ∂CNA ∂f5 (RAB ) ∂CNB
= + +
∂αC ∂αC RAB =const. 2 A,B ∂CNA ∂αC ∂CNB ∂αC
0 (50)
h i
× µTA µB RAB
2
− 3 µTA RAB µTB RAB
The derived formulas and their implementation was tested by comparison of computed an-
alytic and numerical numerical derivatives.
1.5 Dispersion
The two-body London dispersion energy for two distant atoms is given as:
X 3 Z X C6 (ρA , ρB )
Edisp =− 6
α A (iω)α B (iω)dω = − 6
(51)
A<B πRAB A<B RAB
Here, αA (iω) refers to the isotropic dynamic dipole polarizability of atom A. It is a function
of the density around the atom, and consequently, the dispersion coefficient is a function of
the density on both atoms. Hence, expanding the two-body dispersion energy in terms of
density fluctuations leads to:
X 1
Edisp = − [C6 (ρA0 , ρB0 ) + C6 (δρA , ρB0 ) + C6 (ρA0 , δρB ) + C6 (δρA , δρB )
A<B RAB
i
+ C6 ((δρA )2 , ρB0 ) + C6 (ρA0 , (δρB )2 ) + · · · (52)
Here, all terms up to second order in the density fluctuations are shown. In the D4 dis-
persion model (see below), all terms in the first line of Eq. 52 will explicitly be taken into
account, i.e., all terms of zeroth and first order, as well as the two-center second order terms.
This is a result from considering first order effects in the polarizabilities and formation of
their products to obtain the dispersion cofficient. Formally, the pairwise dipole-quadrupole
dispersion coefficient is handled in the same way.
S-21
1.5.1 The D4 dispersion energy
The total dispersion energy in the context of GFN2-xTB is given by
NA,ref
1X X X NX
B,ref
ED4 = − ξ a (qA , qA,a )ξBb (qB , qB,b )

2 A a B b A
Cnab damp,BJ
× WAa (qA , qA,a )WBb (qB , qB,b )
X
sn n
fn (RAB )
n=6,8 RAB
X (3 cos(θABC ) cos(θBCA ) cos(θCAB ) + 1)C9ABC (CNcov
A B
, CNcov C
, CNcov )
− s9 3
A>B>C (RAB RAC RBC )
× f9damp,zero (RAB , RAC , RBC ) (53)
Here, the two-body contribution is augmented with a charge-independent three-body Axilrod-

Teller-Muto (ATM) term. The rational damping Becke-Johnson-type damping as in DFT-D3
is used for the two-body contribution
v
n
u AB
RAB u C8
fndamp,BJ (RAB ) = n crit.
with RAB = t (54)
crit.
RAB + (a1 · RAB + a2 )6 C6AB
and the zero damping function used for the ATM dispersion is defined slightly different
compared to previous implementations of DFT-D3. Namely, we adjust the cutoff radii in
the damping function by dropping the factor 4/3 and using the same cutoff radii as in the
two-body damping function for a more consistent description of the dispersion energy.
 s 16 −1
crit. crit. crit.
RAB RBC RCA  
f9damp,zero (RAB , RAC , RBC ) = 
1 + 6 
3
(55)
RAB RBC RCA

1.5.2 Derivation of the potential
The potential for the dispersion energy is derived by taking the derivative of the disper-
sion energy expression with respect to the orbital coefficients. Since the ATM term is not
S-22
charge dependent it appears not in the potential. The dependencies on the charge and the
coordination number are dropped for brevity.
NA,ref
∂ED4 1X X X NX
B,ref
∂ξAa b a b X C ab
=− ξB WA WB sn nn fn
∂cνi 2 A a B b ∂cνi n=6,8 RAB
NA,ref
1X X X NX
B,ref
∂ξBb a a b X C ab
− ξA WA WB sn nn fn (56)
2 A a B b ∂cνi n=6,8 RAB
By renaming the indices we can easily simplify above expression
∂ED4 X NX
A,ref
X NX
B,ref
∂ξAa ∂qA b a b X Cnab
=− ξ W W sn n f n (57)
∂cνi A a B b ∂qA ∂cνi B A B n=6,8 RAB
 
X NX
A,ref
X NX
B,ref
∂ξAa  C ab
ni cµi Sνµ  ξBb WAa WBb sn nn fn
XX X X
=− δAD ni cκi Sνκ +
A a B b ∂qA C κ∈C µ∈A n=6,8 RAB
(58)
ND,ref d X NX
X ∂ξD B,ref
b d b
X CnDB X X
=− ξ W W sn n f n ni cκi Sνκ (59)
d ∂qD B b B D B n=6,8 RDB C κ∈C
We rename and reorder the terms
NA,ref a X NX
∂ED4 X ∂ξA B,ref
C ab X X
ξBb WAa WBb sn nn fn
X
=− ni cκi Sνκ (60)
∂cνi a ∂qA B b n=6,8 RAB C κ∈C
| {z }
dA
Which leads to the compact expression
D4 1
Fκλ = Sκλ (dA + dB ), ∀κ ∈ A, λ ∈ B (61)
2
S-23
2 Detailed results
2.1 Structures
Table S1: Comparison of computed rotational constants of twelve medium sized molecules to
experimentally derived ones (ROT34)a for different semiempirical methods. The individual
values are given in MHz.
GFN2-xTB ref.
1A 4299.3 4293.9
B 1411.8 1395.9
C 1143.5 1130.2
2A 2630.8b 3322.5
B 912.7b 719.8
C 868.4b 698.0
3A 3072.3 3071.1
B 1302.1 1285.0
C 1246.2 1248.7
4A 2789.4 2755.9
B 2699.8 2675.6
C 2682.4 2653.3
5A 2336.8 2336.9
6A 1459.6 1464.2
B 772.8 768.2
C 587.5 580.6
7A 1175.1 1165.7
B 658.6 661.2
C 456.3 454.0
8A 1236.0 1166.3
B 759.8 767.6
C 525.9 513.0
9A 876.0 862.5
B 748.5 754.2
C 513.6 513.7
10 A 3100.8 3086.2
B 730.8 723.7
C 687.2 685.0
11 A 1451.9 1432.1
B 819.0 820.5
C 687.1 679.4
12 A 1523.5 1523.2
B 1086.4 1070.5
C 728.2 719.9
a Rotational constants B (excluding vibrational effects) from
e
Ref. S4 with an estimated reference error of 0.2%.
b A conformer other than the experimental one is obtained.
This value is neglected in the statistical analysis of the data

set presented in the manuscript.
1: ethynyl-cyclohexane, 2: isoamyl-acetate,
3: diisopropyl-ketone, 4: bicyclo[2.2.2]octadiene,
5: triethylamine, 6: vitamin C, 7: serotonine, 8: aspirin,
9: cassyrane, 10: proline, 11: lupinene, and 12: limonene.
S-24
Table S2: Untypically long intramolecular bonds (LB12)a obtained by geometry optimiza-
tions with the GFN-xTB and GFN2-xTB methods in comparison to experimental values.
The values are given in pm.
system bond GFN2-xTB GFN-xTB ref.

DIAD C–C 167.5 167.3 171.0
FLP P–B 205.4 210.2 212.0
DTFS Si–N 272.8 207.3 227.0
MESITRAN Si–N 239.7 226.7 245.0
S2+
8 S–S 385.6 230.2 286.0
HAPPOD Rh–Cr 306.2 298.6b 308.0
KAMDOR Os–Cr 306.6 297.6b 310.0
PP C–C 313.8 312.6 312.0
BRCLNA Br–Cl 302.8 305.8 313.0
C2 Br6 Br–Br 337.0 340.3 342.0
RESVAN S–S 374.9 382.6 419.0
BHS Si–Si 454.1 452.7 443.0
MD: 6.5 (-1.9)c -13.0 –
MAD: 19.9 (12.6)c 14.7 –
SD: 35.4 (20.9)c 17.9 –
MAX: 99.6 (45.8)c 55.8 –
a Reference bond lengths of long bonds as used in Ref.
S5.
b Bonds are different w.r.t. Ref. S3 due to the modified
GFN-xTB Hamiltonian with CN-dependence for Cr.

c Statistics without system S2+ is given in parentheses.
8
Table S3: Covalent bonds of heavy main group elements (HMGB11)a from experiment and
computed with GFN2-xTB. The values are given in pm.
system bond GFN2-xTB ref.

Cl2 Cl–Cl 201.7 198.8
S2 H2 S–S 204.5 205.5
P2 Me4 P–P 220.1 221.2
Br2 Br–Br 228.2 228.1
Se2 H2 Se–Se 227.7 234.6
Ge2 H6 Ge–Ge 236.3 241.0
As2 Me4 As–As 244.3 242.9
Te2 Me2 Te–Te 269.5 268.6
Sn2 Me6 Sn–Sn 281.2 277.6
Sb2 Me4 Sb–Sb 281.9 281.8
Pb2 Me6 Pb–Pb 293.9 288.0
MD: 0.1 –
MAD: 2.6 –
RMSD: 3.5 –
SD: 3.6 –
MAX: 7.0 –
a Reference bond lengths are the same as used in Ref. S5.
S-25
Table S4: Covalent bonds in transition metal complexes (TMC32)a from experiment and
computed with different semiempirical methods. The values are given in pm.
GFN2-xTB GFN-xTBb ref.

1 201.4 201.4 207.6
2 211.1 213.1 216.9
3 204.7 205.5 204.7
4 212.6 213.1 218.5
5 205.5 205.1 205.8
6 214.8 213.7 219.6
7 228.6 229.3 217.5
8 185.7 178.1 198.4
9 151.7 154.5 157.0
10 162.0 163.7 172.9
11 164.3 169.0 173.4
12 165.1 167.9 170.8
13 151.9 153.3 157.3
14 200.4 210.8 213.8
15 178.5 180.9 187.9
16 180.7 182.3 196.3
17 154.8 152.5 157.4
18 167.1 159.8 171.9
19 155.1 153.1 157.7
20 203.4 203.5 212.2
21 154.6 153.0 158.4
22 191.0 180.7 195.4
23 220.3 214.3 215.0
24 226.8 220.3 220.8
25 182.3 182.9 186.3
26 169.9 168.3 175.0
27 147.0 155.0 158.6
28 159.6 163.9 172.4
29 216.8 217.0 214.7
30 178.5 180.6 180.6
31 179.6 178.4 182.9
32 177.1 178.1 181.0
33 200.9 193.4 193.8
34 217.6 210.7 212.3
35 180.2 182.4 187.2
36 166.9 160.8 167.4
37 215.8 206.5 206.4
38 217.2 214.0 211.7
39 176.6 177.5 181.5
40 177.9 178.5 180.6
41 232.1 239.6 237.7
42 179.8 179.3 181.8
43 160.4 160.0 165.8
44 181.1 181.3 183.0
45 181.6 179.7 182.5
46 185.4 185.8 187.6
47 214.8 211.1 209.9
48 194.6 198.9 188.4
49 186.9 190.5 183.2
50 195.9 199.0 191.4
MD: -2.8 -3.2 –
MAD: 5.7 5.0 –
SD: 6.1 5.7 –
MAX: 15.6 20.3 –
a Reference bond lengths are from Ref. S6.
b GFN-xTB values differ from Ref. S3, due to the
use of a revised GFN-xTB Hamiltonian, in which

3d-level shifting is activated for the early
3d-transition metals (Sc–Cr).
S-26
Table S9: Equilibrium center-of-mass distances between non-covalently bound systems of
the R160x6 set S10 . The values are given in pm and obtained by a cubic spline interpolation.
For the interpolation, interaction energies and center-of-mass distances computed on the 6
structures along the potential energy curve of each complex are used.
system # GFN2-xTB GFN-xTB PM6-D3H4X DFTB-D3(BJ) ref.
1 424.2 428.5 423.6 419.2 419.1

2 442.8 442.1 440.1 435.1 442.8
3 412.5 410.8 413.0 406.6 420.8
4 389.0 386.8 393.6 380.4 390.8
5 394.6 392.4 389.7 384.7 397.5
6 406.7 404.9 425.9 405.3 408.7
7 382.9 387.8 395.0 381.2 392.1
8 468.3 471.1 481.9 469.1 478.4
9 362.3 355.8 364.9 356.0 368.4
10 347.6 341.3 354.9 347.9 351.1
11 362.5 348.3 362.6 358.3 467.6
12 435.0 440.2 437.9 428.3 433.4
13 579.1 579.1 579.1 579.1 579.1
14 542.8 542.8 542.8 542.8 542.8
15 415.5 419.6 419.7 409.8 423.0
16 506.3 506.3 506.3 506.3 506.3
17 424.2 430.4 497.2 416.0 497.2
18 498.1 498.1 498.1 498.1 498.1
19 495.7 495.7 495.7 495.7 495.7
20 371.1 371.6 373.5 360.1 368.8
21 412.4 420.4 481.9 410.6 417.3
22 391.1 370.0 374.2 358.0 390.3
23 409.6 409.7 407.4 404.0 415.4
24 368.7 361.5 361.0 345.8 387.4
25 379.5 377.5 374.3 371.6 389.9
26 310.1 318.7 324.6 314.9 327.9
27 344.9 343.7 348.5 344.0 351.7
28 312.3 308.1 369.6 317.0 309.8
29 526.8 526.8 526.8 526.8 526.8
30 485.9 485.9 485.9 485.9 485.9
31 407.8 408.1 414.1 399.8 477.3
32 455.4 455.4 455.4 455.4 455.4
33 366.0 366.0 359.7 353.4 372.2
34 411.4 406.6 405.7 402.4 421.4
35 363.0 354.2 347.5 340.9 391.9
36 380.4 373.6 371.8 368.9 462.3
S-27
37 312.6 304.6 323.6 314.5 316.4
38 548.6 548.6 548.6 548.6 548.6
39 476.7 476.7 448.3 476.7 476.7
40 647.7 647.7 647.7 647.7 647.7
41 414.8 506.3 433.1 411.8 414.2
42 506.6 506.6 506.6 506.6 506.6
43 477.8 477.8 444.0 477.8 477.8
44 439.9 344.2 383.7 353.0 439.9
45 486.3 460.2 577.8 463.1 577.8
46 377.4 374.1 386.0 374.8 380.3
47 431.1 431.1 403.0 431.1 431.1
48 326.5 298.1 358.9 312.6 420.3
49 390.4 394.3 377.9 383.4 451.9
50 413.5 362.5 351.4 348.1 413.5
51 450.7 450.7 450.7 450.7 450.7
52 481.6 483.8 497.9 478.3 555.6
53 390.7 385.2 391.2 379.3 389.5
54 351.8 353.5 339.4 345.2 415.8
55 414.0 414.0 414.0 414.0 414.0
56 349.3 348.3 340.9 338.2 414.5
57 423.4 425.7 424.7 421.1 437.5
58 475.4 471.3 480.5 475.9 483.5
59 435.1 424.0 450.1 417.4 549.7
60 469.8 456.0 466.7 466.1 541.0
61 430.6 420.9 439.8 414.7 556.6
62 555.7 419.9 516.9 437.4 555.7
63 496.9 377.4 435.2 393.0 496.9
64 438.8 421.9 434.0 434.1 512.7
65 493.0 606.8 536.1 484.3 606.8
66 366.4 365.7 442.4 359.9 442.4
67 382.8 358.9 368.7 384.3 401.5
68 566.5 436.8 528.0 453.3 566.5
69 515.8 395.8 425.5 399.8 515.8
70 451.9 461.2 572.0 457.0 572.0
71 407.2 389.7 441.2 406.0 471.3
72 407.2 389.7 441.2 406.0 471.3
73 676.7 526.2 676.7 538.5 676.7
74 426.8 415.2 431.4 415.7 428.2
75 518.9 377.7 484.3 400.3 518.9
76 377.6 357.6 355.6 432.2 378.6
77 514.8 364.3 412.0 393.4 514.8
S-28
78 331.7 396.8 330.2 320.3 338.4
79 439.0 439.0 439.0 399.5 439.0
80 308.1 389.6 308.1 300.9 389.6
81 416.9 416.9 416.9 378.8 416.9
82 310.0 318.8 307.5 297.8 393.1
83 378.3 378.3 309.7 311.5 378.3
84 339.0 339.0 339.0 277.0 339.0
85 380.8 380.8 380.8 344.2 380.8
86 310.7 310.7 310.7 268.5 310.7
87 296.7 371.4 304.0 287.3 371.4
88 403.2 403.2 329.5 335.0 403.2
89 367.0 367.0 291.3 290.8 367.0
90 417.2 417.2 417.2 337.2 417.2
91 283.6 292.7 356.8 275.8 356.8
92 363.6 363.6 363.6 363.6 363.6
93 410.5 424.3 478.9 398.9 478.9
94 389.2 389.2 389.2 315.9 329.1
95 340.4 340.4 269.7 272.6 340.4
96 348.6 348.6 348.6 348.6 348.6
97 338.7 285.1 273.7 263.6 338.7
98 354.5 355.6 359.6 352.8 363.3
99 302.6 305.9 315.7 307.9 332.1
100 369.8 365.9 371.7 358.1 436.3
101 317.9 306.3 328.6 305.9 326.0
102 411.7 411.7 411.7 339.6 411.7
103 540.9 540.9 518.5 540.9 540.9
104 479.9 479.9 450.1 479.9 479.9
105 415.3 413.3 399.7 406.5 475.9
106 554.3 554.3 531.2 554.3 554.3
107 472.8 472.8 439.9 472.8 382.7
108 663.2 663.2 663.2 663.2 663.2
109 411.2 506.7 440.6 407.9 405.0
110 504.6 504.6 483.0 504.6 504.6
111 490.3 490.3 456.1 490.3 490.3
112 478.7 478.7 443.8 478.7 478.7
113 333.6 332.6 371.7 331.4 340.7
114 433.7 368.5 355.2 358.7 433.7
115 316.7 325.5 365.7 322.2 327.1
116 471.2 474.6 485.4 463.2 482.0
117 362.5 371.7 378.4 367.7 367.9
118 434.3 434.3 397.2 434.3 434.3
S-29
119 303.2 303.4 255.4 300.5 322.3
120 395.6 405.3 412.8 399.6 410.4
121 524.1 420.8 500.3 414.4 524.1
122 471.4 370.6 417.9 370.9 471.4
123 419.6 411.8 424.3 410.3 486.8
124 533.0 433.4 510.3 428.7 533.0
125 483.3 381.6 412.3 375.3 483.3
126 425.6 434.9 539.0 430.7 539.0
127 523.5 400.7 523.5 394.6 523.5
128 526.8 523.0 641.2 516.7 641.2
129 399.2 399.8 417.2 394.2 401.2
130 486.7 382.3 467.3 377.0 486.7
131 484.7 364.0 406.3 369.8 484.7
132 367.8 359.2 371.1 361.3 388.8
133 540.0 540.0 540.0 540.0 540.0
134 456.0 345.5 456.0 456.0 456.0
135 354.0 351.8 369.5 353.5 433.6
136 546.3 546.3 546.3 546.3 546.3
137 463.7 463.7 366.6 463.7 463.7
138 552.3 552.3 552.3 552.3 552.3
139 663.1 663.1 663.1 663.1 663.1
140 379.9 378.0 384.9 367.0 386.7
141 503.8 503.8 503.8 503.8 503.8
142 478.1 309.6 478.1 478.1 478.1
143 373.0 368.6 373.9 368.7 373.8
144 469.4 469.4 469.4 469.4 469.4
145 343.7 338.1 345.7 340.4 352.1
146 431.9 431.9 431.9 431.9 431.9
147 420.0 349.1 420.0 349.0 420.0
148 421.7 421.7 332.1 421.7 421.7
149 413.1 366.5 323.6 413.1 413.1
150 779.0 621.7 779.0 604.3 779.0
151 476.1 480.4 494.1 471.4 479.8
152 614.0 614.0 614.0 614.0 614.0
153 597.6 451.2 597.6 470.1 597.6
154 421.4 425.6 432.1 420.7 415.4
155 457.3 457.3 379.2 457.3 362.6
156 345.6 344.8 348.2 347.1 360.4
157 469.8 469.8 469.8 469.8 469.8
158 439.0 439.0 404.8 439.0 439.0
159 418.8 361.9 418.8 418.8 418.8
S-30
160 298.8 273.0 256.4 278.3 422.0
S-31
Table S5: Center of mass (CMA) distances of 22 non-covalently interacting systems (S22) S7
computed with GFN2-xTB. The values are given in pm. See Ref. S3 and its supporting
information for the results obtained with other semiempirical methods.
GFN2-xTB ref.
1 336.4 324.6
2 282.6 291.1
3 300.9 301.1
4 316.7 325.8
5 599.7 606.0
6 501.8 517.7
7 586.0 602.6
8 372.8 371.7
9 390.5 371.8
10 378.6 371.6
11 349.3 376.4
12 328.8 347.9
13 323.0 317.6
14 331.5 349.8
15 306.4 316.7
16 423.5 442.2
17 311.7 333.7
18 321.7 353.9
19 401.5 389.2
20 501.6 490.8
21 480.8 488.7
22 524.0 494.0
1: ammonia dimer, 2: water dimer, 3: formic acid dimer,
4: formamide dimer, 5: uracil dimer,
6: 2-pyridoxine · 2-aminopyridine, 7: adenine · thymine,
8: methane dimer, 9: ethene dimer, 10: benzene · methane,
11: benzene dimer, 12: pyracine dimer, 13: uracil dimer,
14: indole · benzene, 15: adenine · thymine (stack),
16: ethene · ethine, 17: benzene · water,
18: benzene · ammonia, 19: benzene · cyanide, 20: benzene
dimer, 21: indole · benzene (T-shape), 22: phenol dimer.
S-32
the S66x8 seta . The values are given in pm and obtained by a cubic spline interpolation.
structures along the potential energy curve of each complex are used. See Ref. S3 and its
supporting information for the results obtained with other semiempirical methods.
# system GFN2-xTB ref.
1 H2 O · H2 O 286.6 293.9
2 H2 O · MeOH 309.0 309.6
3 H2 O · MeNH2 325.5 334.4
4 H2 O · peptide 383.4 384.9
5 MeOH · MeOH 351.3 350.0
6 MeOH · MeNH2 330.2 335.8
7 MeOH · peptide 419.5 420.7
8 MeOH · H2 O 324.5 328.6
9 MeNH2 · MeOH 349.8 354.4
10 MeNH2 · MeNH2 349.8 348.0
11 MeNH2 · peptide 363.1 366.9
12 MeNH2 · H2 O 301.2 303.4
13 peptide · MeOH 391.0 388.3
14 peptide · MeNH2 376.6 388.7
15 peptide · peptide 463.8 468.1
16 peptide · H2 O 378.9 382.3
17 uracil · uracil (BP) 569.7 574.6
18 H2 O · pyridine 420.2 426.9
19 MeOH · pyridine 445.1 449.9
20 AcOH · AcOH 407.5 407.9
21 AcNH2 · AcNH2 423.9 432.5
22 AcOH · uracil 503.9 506.0
23 AcNH2 · uracil 505.7 512.0
24 benzene · benzene (π-π) 360.8 387.6
25 pyridine · pyridine (π-π) 344.7 369.9
26 uracil · uracil (π-π) 306.3 314.8
27 benzene · pyridine (π-π) 353.5 379.2
28 benzene · uracil (π-π) 326.4 338.9
29 pyridine · uracil (π-π) 320.3 333.7
30 benzene · ethene 329.7 353.2
31 uracil · ethene 325.1 331.2
32 uracil · ethyne 317.2 326.2
33 pyridine · ethene 326.3 345.5
34 pentane · pentane 403.3 382.5
35 neopentane · pentane 467.2 452.8
36 neopentane · neopentane 522.8 525.6
37 cyclopentane · neopentane 477.5 465.7
38 cyclopentane · cyclopentane 440.2 421.7
39 benzene · cyclopentane 396.7 394.3
40 benzene · neopentane 445.3 448.5
41 uracil · pentane 355.6 353.3
42 uracil · cyclopentane 378.7 375.8
43 uracil · neopentane 432.5 434.5
44 ethene · pentane 394.9 375.6
45 ethyne · pentane 361.3 362.9
46 peptide · pentane 376.0 362.2
47 benzene · benzene (TS) 493.9 490.4
48 pyridine · pyridine (TS) 487.7 481.9
49 benzene · pyridine (TS) 491.0 487.1
50 benzene · ethyne (CH-π) 419.7 410.1
51 ethyne · ethyne (TS) 421.2 435.6
52 benzene · AcOH (OH-π) 404.7 417.3
53 benzene · AcNH2 (NH-π) 469.7 476.1
54 benzene · H2 O (OH-π) 319.4 329.2
55 benzene · MeOH (OH-π) 337.8 342.1
56 benzene · MeNH2 (N H-π) 358.8 358.2
57 benzene · peptide (N H-π) 401.5 404.0
58 pyridine · pyridine (N H-π) 581.2 585.7
59 ethyne · H2 O (CH-O) 404.2 399.2
60 ethyne · AcOH (OH-π) 385.7 396.4
61 pentane · AcOH 377.8 373.1
62 pentane · AcNH2 362.1 358.6
63 benzene · AcOH 375.6 375.3
64 peptide · ethene 364.6 360.0
65 pyridine · ethyne 519.3 533.0
66 MeNH2 · pyridine 374.7 372.5
a
Reference structures taken from Ref. S8.
S-33
the S22x5 seta . The values are given in pm and obtained by a cubic spline interpolation.
structures along the potential energy curve of each complex are used.
system # GFN2-xTB GFN-xTB PM6-D3H4X DFTB-D3(BJ) ref.
1 586.9 595.6 610.4 608.4 602.6
2 312.6 309.9 331.7 310.5 324.2
3 334.1 326.5 340.6 337.2 329.9
4 287.1 293.8 294.0 289.1 293.8
5 390.6 386.7 359.7 334.6 374.7
6 395.5 407.3 393.8 372.3 381.4
7 419.9 431.7 431.1 414.6 446.3
8 303.9 301.3 310.8 302.9 304.0
9 318.4 329.7 329.0 326.9 328.5
10 319.7 309.3 324.2 309.3 331.3
11 351.4 349.2 338.9 332.9 354.6
12 381.5 380.6 367.7 355.3 378.7
13 503.7 504.2 497.5 491.8 497.5
14 371.5 379.0 384.8 362.5 395.7
15 494.3 493.7 496.7 489.4 493.6
16 350.2 357.2 361.7 345.9 364.2
17 337.1 347.6 358.6 336.1 361.9
18 502.7 514.2 527.7 523.8 519.3
19 498.4 497.4 502.0 493.7 494.7
20 313.5 304.6 329.8 310.8 323.2
21 602.5 609.1 611.4 610.6 607.0
22 402.5 414.3 403.0 390.2 391.3
1: ammonia dimer, 2: water dimer, 3: formic acid dimer, 4: formamide dimer,
5: uracil dimer, 6: 2-pyridoxine · 2-aminopyridine, 7: adenine · thymine, 8: methane
dimer, 9: ethene dimer, 10: benzene · methane, 11: benzene dimer, 12: pyracine
dimer, 13: uracil dimer, 14: indole · benzene, 15: adenine · thymine (stack),
16: ethene · ethine, 17: benzene · water, 18: benzene · ammonia,
19: benzene · cyanide, 20: benzene dimer, 21: indole · benzene (T-shape),
22: phenol dimer.
S-34
the X40 set S9 . The values are given in pm and obtained by a cubic spline interpolation
of the X40x10 set. For the interpolation, interaction energies and center-of-mass distances
computed on the 10 structures along the potential energy curve of each complex are used.
# system GFN2-xTB GFN-xTB PM6-D3H4X DFTB-D3(BJ) ref.
1 CH4 · F2 422.1 363.3 427.8 386.4 391.5
2 CH4 · Cl2 437.2 425.7 388.6 452.0 431.5
3 CH4 · Br2 462.4 473.1 415.2 482.7 452.6
4 CH4 · I2 500.7 491.7 425.2 510.5 491.0
5 CH3 F · CH4 367.7 367.9 351.1 338.6 348.2
6 CH3 Cl · CH4 382.5 381.0 363.1 356.9 366.4
7 CHF3 · CH4 337.5 319.5 338.2 318.2 345.8
8 CHCl3 · CH4 345.1 328.1 319.5 342.9 360.0
9 CH3 F dimer 452.1 415.3 466.0 430.1 434.5
10 CH3 Cl dimer 520.6 503.7 477.1 525.0 516.2
11 C6 H3 F3 · C6 H6 353.5 363.9 366.6 347.8 367.9
12 C6 F6 · C6 H6 334.2 350.8 350.4 331.3 350.4
13 CH3 Cl · CH2 O 424.2 393.0 367.9 349.5 395.0
14 CH3 Br · CH2 O 372.0 351.9 366.1 343.7 371.4
15 CH3 I · CH2 O 367.3 358.8 372.6 337.2 373.5
16 CF3 Cl · CH2 O 447.7 435.3 450.9 426.2 452.8
17 CF3 Br · CH2 O 407.9 417.5 434.3 420.1 430.7
18 CF3 I · CH2 O 413.0 420.7 417.2 409.0 426.7
19 C6 H5 Cl · C(CH3 )2 O 611.9 596.9 598.5 572.9 612.8
20 C6 H5 Br · C(CH3 )2 O 553.3 550.6 565.5 544.4 567.4
21 C6 H5 I · C(CH3 )2 O 544.2 544.0 551.0 524.0 554.8
22 C6 H5 Cl · N(CH3 )3 566.3 545.9 560.4 539.1 570.4
23 C6 H5 Br · N(CH3 )3 506.2 506.1 514.1 504.2 515.8
24 C6 H5 I · N(CH3 )3 488.9 489.2 481.8 476.4 491.0
25 C6 H5 Br · CH3 SH 539.2 511.6 568.7 516.5 520.0
26 C6 H5 I · CH3 SH 507.2 497.0 500.1 491.6 505.6
27 CH3 Br · C6 H6 403.0 386.0 341.6 388.1 395.8
28 CH3 I · C6 H6 402.3 378.2 344.6 381.4 398.8
29 CF3 Br · C6 H6 441.6 429.3 395.2 449.7 449.2
30 CF3 I · C6 H6 434.5 415.7 385.1 440.9 446.7
31 CF3 OH · H2 O 333.6 338.0 345.5 340.3 339.6
32 CCl3 OH · H2 O 349.2 355.3 357.4 358.0 353.7
33 HF · CH3 OH 299.4 303.2 324.0 306.8 303.1
34 HCl · CH3 OH 346.0 364.9 346.3 345.3 352.1
35 HBr · CH3 OH 360.5 400.6 363.6 370.1 370.5
36 HI · CH3 OH 384.0 430.6 386.8 381.0 397.6
37 HF · N(CH3 )H2 280.0 270.9 326.1 302.7 288.2
38 HCl · N(CH3 )H2 317.6 304.8 329.0 337.8 324.0
39 CH3 OH · CH3 F 377.6 358.2 431.7 362.3 359.5
40 CH3 OH · CH3 Cl 378.6 360.8 394.3 375.5 375.1
S-35
2.2 Non-covalent interactions
Table S10: Association energies computed with semiempirical methods for six alkane dimers
(ADIM6). Structures and reference energies are taken from Ref. S11. The values are given
in kcal/mol.
system # GFN2-xTB GFN-xTB PM6-D3H4X DFTB3-D3(BJ) ref.
1 0.92 0.69 1.42 1.50 1.34
2 1.29 1.27 1.77 2.32 1.99
3 1.97 2.09 2.79 3.44 2.89
4 2.57 2.75 3.59 4.47 3.78
5 3.08 3.53 4.35 5.61 4.60
6 3.45 3.78 4.77 6.52 5.55
MD: -1.15 -1.01 -0.24 0.62 –
MAD: 1.15 1.01 0.27 0.62 –
SD: 0.60 0.41 0.29 0.34 –
MAX: 2.10 1.77 0.78 1.01 –
S-36
Table S11: Association energies of the HAL59 set S9,S11,S12 computed with different semiem-
pirical methods. Numbering and reference values taken from Ref. S11. The values are given
in kcal/mol.
# system GFN2-xTB GFN-xTB PM6-D3H4X DFTB3-D3(BJ) ref.
1 PCH–C6 H5 Br 0.53 2.32 -2.05 2.66 0.85
2 NCH–C6 H5 Br -0.90 1.46 0.67 3.58 1.15
3 NH3 –C6 H5 Br -0.37 2.38 2.23 5.73 2.02
4 MeI–PCH 0.98 2.34 4.19 3.17 0.85
5 MeI–NCH 0.18 1.88 0.81 5.02 1.42
6 MeI–NH3 1.85 3.45 2.86 7.64 2.73
7 PCH–C6 H5 I 0.96 2.36 4.28 3.14 0.92
8 NCH–C6 H5 I 0.32 2.04 0.86 4.99 1.87
9 NH3 –C6 H5 I 2.07 3.78 3.06 7.60 3.33
10 PCH–F3 CI 2.65 2.33 5.36 3.59 0.89
11 NCH–F3 CI 3.63 3.27 3.37 6.76 3.61
12 NH3 –F3 CI 8.22 6.88 7.97 10.03 5.88
13 Br2 –PCH 1.58 2.36 -3.07 3.53 1.18
14 Br2 –NCH 1.66 2.25 2.52 6.00 3.61
15 Br2 –NH3 6.89 5.97 7.65 9.78 7.29
16 PCH–C4 H4 BrNO2 1.87 2.71 -0.25 4.63 1.19
17 NCH–C4 H4 BrNO2 1.79 3.84 14.21 8.66 4.32
18 NH3 –C4 H4 BrNO2 7.24 10.23 36.83 13.40 8.02
19 PCH–FCl 2.43 3.25 -6.54 4.84 1.16
20 NCH–FCl 3.07 3.03 -3.94 7.99 4.81
21 NH3 –FCl 9.65 11.85 -69.17 11.15 10.54
22 PCH–C4 H4 INO2 3.82 3.38 29.22 5.81 1.53
23 NCH–C4 H4 INO2 5.59 5.56 83.24 12.29 5.91
24 NH3 –C4 H4 INO2 12.13 13.61 182.55 18.43 10.99
25 PCH–FBr 4.22 3.55 -6.58 5.49 2.07
26 NCH–FBr 5.91 5.05 -2.51 9.55 7.53
27 NH3 –FBr 14.89 13.63 -2.23 10.41 15.30
28 FI–PCH 5.29 5.58 16.83 7.28 2.74
29 FI–NCH 8.35 7.67 4.34 13.42 9.33
30 FI–NH3 15.09 17.10 4.92 17.11 17.11
31 MeI–FCCH 0.46 0.34 0.18 1.53 0.50
32 Br2 –FCCH 0.74 0.33 -0.23 1.33 0.74
33 FI–FCCH -0.21 -0.40 -0.01 -0.08 0.29
34 MeI–FMe 0.30 1.09 1.38 3.06 1.70
35 Br2 –FMe 1.18 0.87 -0.19 3.16 2.87
36 FI–FMe 3.86 7.14 4.20 7.35 5.97
37 MeI–OCH2 1.46 2.84 2.13 5.51 2.39
38 Br2 –OCH2 3.70 3.27 3.25 7.28 4.41
39 FI–OCH2 9.71 9.48 4.39 14.30 9.94
40 MeI–OPH3 1.32 2.70 8.33 4.65 3.34
41 Br2 –OPH3 1.92 1.15 3.90 5.64 5.95
42 FI–OPH3 6.92 1.93 17.55 11.79 13.36
43 MeI–C5 H5 N 2.76 3.92 2.92 10.84 3.61
44 Br2 –C5 H5 N 9.31 5.12 4.49 14.94 9.07
45 FI–C5 H5 N 20.07 18.52 -9.50 25.48 20.34
46 C6 H3 F3 -C6 H6 4.99 3.50 3.98 5.09 4.40
47 C6 H6 F6 -C6 H6 6.70 3.96 4.97 6.91 6.12
48 C6 H5 Cl–CH2 O 0.65 0.47 1.27 4.65 1.49
49 C6 H5 Br–CH2 O 0.60 2.40 2.54 5.47 2.43
50 C6 H5 I–CH2 O 2.92 3.89 3.38 7.35 3.46
51 C6 H5 Cl–N(CH3 )3 0.94 0.92 1.62 7.71 2.11
52 C6 H5 Br–N(CH3 )3 1.21 3.76 3.92 9.27 3.78
53 C6 H5 I–N(CH3 )3 4.07 5.28 5.74 12.24 5.81
54 C6 H5 Br–CH3 SH 1.04 2.50 0.69 3.33 2.32
55 C6 H5 I–CH3 SH 1.84 3.24 3.10 4.81 3.08
56 CH3 Br–C6 H6 0.89 1.21 2.63 3.27 1.81
57 CH3 I–C6 H6 1.69 1.60 3.50 3.83 2.48
58 CF3 Br–C6 H6 1.82 1.57 3.65 4.28 3.11
59 CF3 I–C6 H6 3.54 2.13 5.03 4.85 3.91
MD: -0.73 -0.36 2.36 2.76 –
MAD: 1.28 1.34 9.71 3.00 –
SD: 1.51 2.09 27.92 2.22 –
MAX: 6.44 11.43 171.56 7.44 –
S-37
Table S12: Association energies of the PNICO23 set S11,S13 computed with different semiem-
pirical methods. Numering and refence values taken from Ref. S11. The values are given in
kcal/mol.
1 2.19 1.86 0.39 1.67 1.43
2 5.96 18.56 -10.70 -1.86 8.02
3 3.92 1.87 -0.90 -0.61 0.64
4 4.37 9.41 -7.50 4.63 4.26
5 4.96 3.10 -0.55 3.66 2.86
6 1.18 1.29 1.52 1.38 1.32
7 3.37 7.62 -7.87 4.36 4.29
8 2.62 4.12 -2.37 3.08 2.63
9 6.24 7.34 -6.22 4.05 4.91
10 2.89 2.99 -0.55 2.23 2.21
11 2.44 2.56 -0.17 1.57 1.40
12 4.22 6.03 -4.01 2.88 2.62
13 1.78 1.27 2.25 0.44 3.98
14 4.18 6.80 -4.46 4.02 4.10
15 2.91 2.99 3.54 2.25 4.34
16 2.16 1.97 1.49 1.41 1.78
17 7.22 10.94 5.51 5.43 7.10
18 2.24 1.50 1.24 1.27 2.35
19 5.12 7.70 5.28 3.92 5.95
20 11.47 14.70 9.64 6.16 8.18
21 5.53 6.02 3.83 4.02 4.92
22 8.88 9.02 5.87 6.04 8.03
23 12.41 12.37 8.34 7.88 10.97
MD: 0.43 1.90 -4.12 -1.23 –
MAD: 1.10 2.33 4.26 1.45 –
SD: 1.42 2.79 5.05 2.23 –
MAX: 3.29 10.54 18.72 9.88 –
S-38
Table S13: Association energies of 22 non-covalently interacting systems (S22)a computed
with GFN2-xTB. The values are given in kcal/mol. Reference energies taken from Ref. S14.
Structures taken from Ref. S7. Running number as in Ref. S15.
system # GFN2-xTB ref.
1 2.05 3.13
2 4.90 4.99
3 17.20 18.75
4 16.64 16.06
5 19.63 20.64
6 16.72 16.93
7 15.89 16.66
8 0.38 0.53
9 1.07 1.47
10 1.29 1.45
11 3.83 2.65
12 5.43 4.25
13 9.81 9.80
14 5.40 4.52
15 12.20 11.73
16 1.37 1.50
17 2.26 3.27
18 1.88 2.31
19 2.60 4.54
20 2.31 2.72
21 4.00 5.63
22 5.79 7.10
MD: -0.36 –
MAD: 0.75 –
SD: 0.88 –
MAX: 1.94 –
1: ammonia dimer, 2: water dimer, 3: formic acid dimer,
4: formamide dimer, 5: uracil dimer,
6: 2-pyridoxine · 2-aminopyridine, 7: adenine · thymine,
8: methane dimer, 9: ethene dimer, 10: benzene · methane,
11: benzene dimer, 12: pyracine dimer, 13: uracil dimer,
14: indole · benzene, 15: adenine · thymine (stack),
16: ethene · ethine, 17: benzene · water,
18: benzene · ammonia, 19: benzene · cyanide, 20: benzene
dimer, 21: indole · benzene (T-shape), 22: phenol dimer.
S-39
Table S14: Association energies computed with semiempirical methods for 66 non-covalent
complexes consisting of main group elements (S66). S8 The values are given in kcal/mol.
# system GFN2-xTB GFN-xTB PM6-D3H4X DFTB3-D3(BJ) ref.
1 H2 O · H2 O 4.91 4.76 5.03 4.86 4.92

2 H2 O · MeOH 4.78 4.46 5.58 5.09 5.59
3 H2 O · MeNH2 5.38 6.15 7.44 4.42 6.91
4 H2 O · peptide 7.40 7.08 7.99 8.41 8.10
5 MeOH · MeOH 4.77 4.62 6.32 4.85 5.76
6 MeOH · MeNH2 5.74 6.65 6.98 4.48 7.55
7 MeOH · peptide 7.31 7.28 8.19 7.82 8.23
8 MeOH · H2 O 4.61 4.76 5.75 4.43 5.01
9 MeNH2 · MeOH 2.71 2.51 4.24 2.54 3.06
10 MeNH2 · MeNH2 2.70 2.56 4.54 2.36 4.16
11 MeNH2 · peptide 4.58 4.22 6.14 4.32 5.42
12 MeNH2 · H2 O 4.98 5.71 7.30 4.09 7.27
13 peptide · MeOH 5.32 4.85 6.49 5.07 6.19
14 peptide · MeNH2 6.39 5.76 7.45 4.72 7.45
15 peptide · peptide 8.05 7.54 8.81 7.92 8.63
16 peptide · H2 O 4.76 4.78 5.51 4.47 5.12
17 uracil · uracil (BP) 16.54 14.77 16.44 15.11 17.18
18 H2 O · pyridine 5.11 5.44 6.78 4.24 6.86
19 MeOH · pyridine 5.57 5.96 6.14 4.31 7.41
20 AcOH · AcOH 17.56 18.68 18.43 19.15 19.09
21 AcNH2 · AcNH2 16.43 14.45 16.98 15.49 16.26
22 AcOH · uracil 18.24 17.66 18.42 18.40 19.49
23 AcNH2 · uracil 19.07 16.71 18.97 17.48 19.19
24 benzene · benzene (π-π) 3.60 3.12 2.89 3.38 2.82
25 pyridine · pyridine (π-π) 4.60 3.35 4.03 3.96 3.90
26 uracil · uracil (π-π) 9.87 8.62 9.08 9.25 9.83
27 benzene · pyridine (π-π) 4.06 3.15 3.55 3.71 3.44
28 benzene · uracil (π-π) 5.81 3.97 5.37 6.02 5.71
29 pyridine · uracil (π-π) 7.39 5.22 6.88 6.92 6.82
30 benzene · ethene 1.88 1.78 1.59 1.89 1.43
31 uracil · ethene 3.14 1.96 3.00 3.60 3.38
32 uracil · ethyne 3.34 1.84 2.70 3.84 3.74
33 pyridine · ethene 2.18 1.81 2.03 2.08 1.87
34 pentane · pentane 2.38 2.56 3.48 4.73 3.78
35 neopentane · pentane 1.93 2.10 2.73 3.45 2.61
36 neopentane · neopentane 1.62 2.08 2.17 2.65 1.78
37 cyclopentane · neopentane 1.95 2.09 2.52 3.39 2.40
38 cyclopentane · cyclopentane 1.96 2.10 2.33 3.67 3.00
S-40
39 benzene · cyclopentane 3.27 3.03 3.40 3.73 3.58
40 benzene · neopentane 2.88 2.73 3.46 3.19 2.90
41 uracil · pentane 4.93 4.09 5.55 5.61 4.85
42 uracil · cyclopentane 4.01 3.59 4.53 4.71 4.14
43 uracil · neopentane 3.81 2.81 3.78 3.79 3.71
44 ethene · pentane 1.34 1.15 1.85 2.46 2.01
45 ethyne · pentane 1.60 1.43 1.73 1.91 1.75
46 peptide · pentane 3.21 3.22 4.11 4.72 4.26
47 benzene · benzene (TS) 2.55 2.05 2.52 2.70 2.88
48 pyridine · pyridine (TS) 2.75 2.44 3.03 3.07 3.54
49 benzene · pyridine (TS) 2.65 1.99 2.92 2.86 3.33
50 benzene · ethyne (CH-π) 2.04 0.93 1.96 2.52 2.87
51 ethyne · ethyne (TS) 1.48 0.63 0.82 1.28 1.52
52 benzene · AcOH (OH-π) 3.65 3.08 4.17 4.60 4.71
53 benzene · AcNH2 (NH-π) 3.26 3.43 3.92 3.79 4.36
54 benzene · H2 O (OH-π) 2.29 1.71 3.36 3.18 3.28
55 benzene · MeOH (OH-π) 3.17 2.49 3.63 3.80 4.19
56 benzene · MeNH2 (N H-π) 2.66 2.00 3.23 2.96 3.23
57 benzene · peptide (N H-π) 4.08 3.14 4.81 4.81 5.28
58 pyridine · pyridine (N H-π) 3.09 2.59 3.25 2.05 4.15
59 ethyne · H2 O (CH-O) 1.84 1.11 1.95 2.68 2.85
60 ethyne · AcOH (OH-π) 3.95 3.51 2.38 4.22 4.87
61 pentane · AcOH 2.80 2.46 3.59 3.52 2.91
62 pentane · AcNH2 3.08 3.28 3.94 4.15 3.53
63 benzene · AcOH 2.98 2.35 3.73 3.79 3.80
64 peptide · ethene 2.07 2.15 2.74 2.91 3.00
65 pyridine · ethyne 3.35 2.18 1.78 2.85 3.99
66 MeNH2 · pyridine 2.94 2.68 4.43 2.77 3.97
MD: -0.61 -1.05 -0.15 -0.45 –

MAD: 0.73 1.08 0.47 0.79 –
SD: 0.65 0.64 0.64 1.04 –
MAX: 2.29 2.48 2.49 3.18 –
S-41
Table S15: Association energies of water clusters (WATER27 set) computed with different
semiempirical methods. The values are given in kcal/mol.
system GFN2-xTB GFN-xTB PM6-D3H4X DFTB3-D3(BJ) ref.
H2 O2 4.93 4.79 5.19 4.87 5.01
H2 O3 14.26 14.63 15.00 14.75 15.80
H2 O4 26.46 24.29 26.25 25.93 27.40
H2 O5 34.20 30.99 34.01 33.77 35.90
H2 O6 44.57 45.33 45.68 44.90 46.00
H2 O6 c 44.16 43.31 44.98 43.94 45.80
H2 O6 b 43.44 40.71 43.61 42.92 45.30
H2 O6 c2 42.09 37.80 41.33 42.10 44.30
H2 O8 d2d 71.48 68.18 72.06 69.92 72.60
H2 O8 s4 71.35 68.21 71.85 69.76 72.60
H2 O20 193.19 176.71 195.88 186.12 198.60
H2 O20 fc 208.95 201.24 207.22 204.39 208.00
H2 O20 fs 207.71 197.38 206.93 202.69 208.00
H2 O20 es 208.62 196.93 206.49 203.25 209.70
H3 O+ H2 O 31.16 33.19 28.46 30.55 33.50
H3 O+ H2 O2 53.46 54.34 49.88 52.42 56.90
H3 O+ H2 O3 73.30 72.59 68.15 71.31 76.50
H3 O+ H2 O6 3d 112.67 112.76 110.37 108.52 117.80
H3 O+ H2 O6 2d 110.36 109.80 108.86 108.74 114.90
OH− H2 O 35.35 38.90 35.08 30.33 26.60
OH− H2 O2 55.32 60.02 54.36 53.04 48.40
OH− H2 O3 74.70 79.30 73.91 72.51 67.60
OH− H2 O4 c4 89.98 96.28 88.20 91.12 84.80
OH− H2 O4 cs 89.72 97.25 90.37 90.85 84.80
OH− H2 O5 104.43 114.86 103.27 108.40 100.70
OH− H2 O6 120.02 129.04 119.01 122.43 115.70
H3 O+ H2 O6 OH− 32.41 42.88 21.09 30.96 28.50
MD: 0.24 0.00 -0.90 -1.16 –
MAD: 3.15 7.51 3.55 4.31 –
SD: 3.92 9.43 4.45 5.07 –
MAX: 8.75 21.89 8.47 12.47 –
S-42
Table S16: Association energies of anion-neutral dimer systems (AHB21). S18 The values are
given in kcal/mol.

1 -13.45 -16.77 -9.44 -15.41 -17.79
2 -26.50 -33.06 -24.13 -30.73 -32.50
3 -58.43 -67.14 -45.41 -61.41 -65.68
4 -7.16 -5.15 -8.59 -8.62 -8.98
5 -13.08 -11.39 -13.77 -17.17 -15.61
6 -24.78 -20.75 -17.38 -28.73 -25.52
7 -18.15 -5.60 -14.37 -19.45 -14.35
8 -37.75 -35.11 -46.30 -42.37 -41.79
9 -20.48 -21.40 -20.08 -17.63 -17.03
10 -43.90 -48.37 -39.26 -40.72 -37.31
11 -7.60 -8.13 -9.18 -7.13 -7.97
12 -14.09 -16.62 -14.84 -13.67 -14.13
13 -29.96 -34.22 -10.45 -24.79 -26.01
14 -17.61 -10.74 -11.14 -12.90 -11.07
15 -7.62 -14.19 -8.85 -7.83 -8.62
16 -13.98 -26.50 -13.87 -15.26 -15.73
17 -26.51 -45.21 -15.95 -25.70 -26.24
18 -10.46 -11.54 -14.08 -9.95 -12.80
19 -18.19 -20.55 -20.50 -19.37 -20.65
20 -18.93 -22.50 -22.86 -23.65 -21.03
21 -32.41 -33.61 -21.66 -33.11 -31.40
MD: 0.53 -1.73 3.34 -0.16 –
MAD: 2.97 4.68 4.75 1.80 –
SD: 3.72 6.51 6.61 2.28 –
MAX: 7.25 18.97 20.27 5.10 –
S-43
Table S17: Association energies of the systems in the CARBHB12. S11 The values are given
in kcal/mol.
system # GFN2-xTB GFN-xTB PM6-D3H4X ref.

1 3.34 4.65 4.74 5.37
2 4.07 5.27 3.56 6.05
3 2.82 3.43 2.77 2.42
4 7.25 10.21 10.86 9.97
5 1.83 2.08 2.63 2.36
6 2.39 2.48 2.39 3.02
7 1.47 1.75 1.46 1.21
8 3.32 4.89 4.43 4.18
9 7.53 7.48 8.30 7.84
10 9.54 10.54 6.81 10.48
11 5.57 5.36 3.65 3.24
12 16.33 16.98 13.51 16.30
MD: -0.58 0.23 -0.61 –
MAD: 1.08 0.67 1.09 –
SD: 1.33 0.85 1.52 –
MAX: 2.71 2.12 3.67 –
Table S18: Association energies of cation-neutral dimer systems (CHB6). S18 The values are
given in kcal/mol.
system # GFN2-xTB GFN-xTB PM6-D3H4X ref.

1 -44.31 -42.80 -20.29 -34.43
2 -30.07 -23.15 -11.59 -23.83
3 -23.84 -19.33 -7.45 -17.83
4 -36.74 -40.77 -12.23 -39.09
5 -28.28 -14.76 -29.04 -25.63
6 -24.61 -20.47 -20.35 -19.90
MD: -4.52 -0.10 9.96 –
MAD: 5.31 3.95 11.25 –
SD: 4.12 6.23 10.92 –
MAX: 9.88 10.87 26.86 –
S-44
Table S19: Association energies of the HEAVY28 S11,S19 set computed with different semiem-
pirical methods. The values are given in kcal/mol.
# system GFN2-xTB GFN-xTB PM6-D3H4X ref.

1 (BiH3 )2 1.99 1.36 0.60 1.16
2 BiH3 −H2 O 1.85 1.49 -1.73 2.49
3 BiH3 −H2 S 1.41 1.18 0.60 1.36
4 BiH3 −HCl 1.06 0.78 -0.08 0.77
5 BiH3 −HBr 1.13 0.94 0.69 0.98
6 BiH3 −HI 1.04 1.69 4.50 1.30
7 BiH3 −NH3 1.69 1.13 -2.14 0.60
8 (PbH4 )2 0.94 0.03 2.62 1.25
9 PbH4 −BiH3 0.57 0.44 0.32 0.55
10 PbH4 −H2 O 0.82 0.35 0.87 0.36
11 PbH4 −HCl 0.64 0.41 0.68 0.75
12 PbH4 −HBr 0.59 0.33 -0.16 0.93
13 PbH4 −HI 0.65 0.53 1.34 1.18
14 PbH4 −TeH2 0.61 0.46 0.60 0.65
15 (SbH3 )2 2.19 4.04 -1.61 1.28
16 SbH3 −H2 O 0.80 1.97 -2.25 1.57
17 SbH3 −H2 S 0.83 1.41 -0.55 1.06
18 SbH3 −HCl 1.01 2.02 -2.92 2.02
19 SbH3 −HBr 0.66 1.61 -1.96 1.89
20 SbH3 −HI 0.92 2.00 -0.98 1.49
21 SbH3 −NH3 5.41 8.45 -29.18 2.84
22 (TeH2 )2 0.64 0.80 1.85 0.52
23 TeH2 −H2 O 2.46 0.40 -1.45 0.68
24 TeH2 −H2 S 0.94 0.41 0.03 0.48
25 TeH2 −HCl 1.92 0.89 0.64 1.23
26 TeH2 −HBr 1.26 0.76 0.11 1.22
27 TeH2 −HI 0.40 1.25 1.81 0.80
28 TeH2 −NH3 4.46 2.33 2.18 3.35
MD: 0.15 0.17 -2.15 –
MAD: 0.61 0.65 2.70 –
SD: 0.83 1.29 6.14 –
MAX: 2.57 5.61 32.02 –
S-45
Table S20: Association energies of the anion-cation dimer set (IL16) S18 computed with
different semiempirical methods. The values are given in kcal/mol.

1 -99.31 -94.08 -95.10 -103.22 -100.41
2 -113.75 -120.68 -115.46 -120.99 -120.80
3 -108.50 -115.22 -109.75 -115.42 -116.91
4 -99.47 -94.89 -97.81 -103.72 -105.01
5 -96.87 -100.54 -99.07 -96.85 -104.44
6 -83.17 -94.93 -80.78 -85.80 -87.42
7 -109.04 -104.27 -102.58 -113.63 -114.00
8 -109.28 -104.49 -103.83 -114.36 -113.51
9 -111.65 -106.41 -106.47 -116.92 -114.91
10 -107.51 -112.63 -112.70 -109.00 -112.75
11 -101.14 -116.68 -95.87 -100.03 -104.47
12 -114.78 -110.39 -109.14 -121.29 -118.19
13 -106.28 -112.91 -110.75 -110.38 -112.02
14 -102.33 -118.59 -99.00 -104.47 -106.53
15 -111.55 -110.93 -113.31 -108.70 -110.98
16 -102.18 -103.41 -95.97 -98.54 -102.37
MD: 4.24 1.48 6.07 1.34 –
MAD: 4.32 5.69 6.36 2.46 –
SD: 2.50 7.25 3.64 2.81 –
MAX: 8.41 12.21 11.43 7.59 –
S-46
Table S21: Association energies of rare gas complexes (RG18) S11 computed with different
# system GFN2-xTB GFN-xTB PM6-D3H4X ref.

1 Ne2 0.05 0.03 0.06 0.08
2 Ar2 0.23 0.17 0.01 0.27
3 Kr2 0.40 0.74 -0.52 0.40
4 Ne3 0.15 0.09 0.18 0.27
5 Ar3 0.66 0.46 0.34 0.77
6 Kr3 1.11 1.90 -0.48 1.18
7 Ne4 0.30 0.18 0.36 0.54
8 Ar4 1.30 0.91 0.71 1.51
9 Ne6 0.64 0.40 0.76 1.13
10 HFNe 0.27 0.32 -0.29 0.23
11 HFAr 0.95 1.04 -0.10 0.59
12 HFKr 0.76 1.86 1.70 0.72
13 C2 H2 −Ne 0.14 0.10 0.16 0.12
14 C2 H2 −Ar 0.38 0.25 -0.11 0.33
15 C2 H6 −Ne 0.16 0.17 0.26 0.24
16 C2 H6 −Ar 0.47 0.47 0.12 0.54
17 Benzene–Ne 0.38 0.27 0.40 0.40
18 Benzene–Ar 1.13 0.73 -1.24 1.12
MD: -0.05 -0.02 -0.45 –
MAD: 0.11 0.32 0.57 –
SD: 0.17 0.45 0.71 –
MAX: 0.49 1.14 2.36 –
S-47
Table S22: Association energies computed with semiempirical methods for 30 large non-
covalent complexes containing only main group elements (S30L)a . The values are given in
kcal/mol.
1 -25.93 -21.97 -28.83 -28.62 -29.04
2 -18.02 -14.51 -18.63 -19.88 -20.78
3 -22.50 -18.75 -22.41 -23.93 -23.54
4 -21.72 -16.52 -19.18 -18.80 -20.27
5 -33.88 -27.16 -33.96 -34.07 -28.99
6 -25.57 -18.70 -20.83 -24.61 -25.50
7 -42.20 -34.75 -30.95 -38.78 -35.06
8 -48.70 -39.87 -35.57 -44.03 -36.79
9 -34.83 -33.02 -27.66 -36.37 -28.38
10 -35.86 -34.88 -29.12 -38.03 -29.78
11 -41.74 -41.52 -38.69 -44.49 -32.95
12 -42.21 -41.72 -38.45 -44.33 -33.92
13 -22.30 -25.25 -29.32 -26.75 -30.83
14 -25.64 -26.89 -29.45 -30.64 -31.33
15 -24.10 -22.78 -21.86 -36.63 -17.39
16 -25.85 -29.95 -30.06 -42.40 -25.12
17 -26.78 -25.18 -39.73 -31.14 -33.38
18 -20.38 -20.49 -29.11 -23.09 -23.31
19 -13.05 -17.85 -19.32 -17.15 -17.47
20 -15.23 -22.00 -23.48 -22.62 -19.25
21 -22.14 -28.43 -31.35 -28.23 -24.21
22 -36.57 -33.98 -44.06 -33.76 -42.63
23 -60.72 -47.03 -61.72 -41.70 -61.32
24 -136.59 -167.06 -162.49 -162.06 -135.51
25 -28.08 -24.30 -25.95 -29.39 -25.96
26 -28.21 -24.60 -25.89 -29.49 -25.77
27 -83.16 -94.76 -104.03 -95.17 -82.18
28 -79.49 -90.49 -101.22 -89.88 -80.11
29 -50.95 -54.68 -59.62 -58.39 -53.54
30 -50.52 -51.57 -56.39 -56.78 -49.28
MD: -0.64 -0.90 -3.86 -4.25 –
MAD: 4.05 6.08 5.15 6.90 –
SD: 5.09 8.50 7.47 8.68 –
MAX: 11.91 31.55 26.98 26.55 –
a Reference structures and numbering are taken from Ref. S20.
S-48
2.3 Conformers
Table S23: Relative conformer energies for different alkane conformers (ACONF)a com-
puted with GFN2-xTB. The results for the other semiempirical methods mentioned in the
manuscript can be found in Ref. S3. The values are given in kcal/mol.
GFN2-xTB ref.
1 0.604 0.598
2 0.566 0.614
3 1.214 0.961
4 2.483 2.813
5 0.573 0.595
6 0.545 0.604
7 1.176 0.934
8 1.082 1.178
9 1.233 1.302
10 1.741 1.250
11 2.384 2.632
12 2.432 2.740
13 2.862 3.283
14 3.049 3.083
15 4.647 4.925
MD: -0.061 –
MAD: 0.194 –
SD: 0.246 –
MAX: 0.491 –
a
Reference data taken from Ref. S21.
S-49
Table S24: Conformational energies for different amino acid conformers (Amino20x4)a com-
puted with different semiempirical methods. The values are given in kcal/mol.
1 -0.01 -0.49 -2.15 -0.58 1.17

2 1.47 1.91 0.95 1.60 3.05
3 1.19 1.42 0.56 1.25 3.32
4 5.55 3.90 5.08 4.35 5.04
5 5.18 3.67 6.32 1.81 1.58
6 5.64 3.67 4.66 0.97 2.53
7 7.21 6.02 5.32 3.03 2.80
8 6.28 6.04 5.08 3.79 6.46
9 0.90 1.24 0.99 1.38 0.39
10 4.02 4.42 1.99 4.77 4.11
11 3.79 4.35 3.27 4.57 4.76
12 7.83 8.17 7.72 8.70 6.53
13 -0.03 -0.83 -0.92 1.44 0.16
14 1.64 1.98 0.67 1.68 1.97
15 3.69 2.36 4.54 2.45 2.90
16 1.51 0.11 0.35 3.14 3.11
17 -0.83 -0.63 -3.41 -1.67 0.28
18 0.60 0.17 -0.71 0.31 0.98
19 -0.10 0.04 -1.93 -1.15 0.99
20 2.02 1.84 0.13 0.17 2.37
21 0.28 1.44 -0.72 -0.32 0.42
22 1.93 3.47 0.42 2.56 3.27
23 5.39 6.82 4.84 5.21 4.04
24 4.99 5.78 2.34 5.12 4.15
25 1.78 0.85 2.21 2.02 1.33
26 2.08 1.52 2.42 2.28 1.54
27 3.29 2.57 3.09 3.27 2.94
28 4.63 3.27 4.57 3.58 5.22
29 1.82 2.48 3.88 1.10 1.09
30 3.54 3.24 4.62 3.34 2.63
31 2.14 3.31 4.29 0.88 2.74
32 3.12 3.58 4.44 3.12 4.09
33 2.22 2.18 2.69 2.10 2.77
34 2.29 1.82 2.49 2.02 2.99
35 6.19 5.93 9.22 6.36 7.32
36 7.69 6.45 8.13 7.10 7.37
37 0.02 0.12 -0.16 0.49 0.19
38 0.10 0.06 0.16 0.32 0.59
S-50
39 1.07 0.88 2.34 0.96 0.96
40 1.33 1.54 1.62 1.01 0.99
41 -0.71 -0.24 -0.43 -0.42 0.34
42 0.29 0.02 -0.06 -0.16 1.52
43 -0.72 0.08 -0.45 -0.28 1.69
44 1.01 0.94 -0.57 0.55 1.95
45 -0.98 -0.98 -2.20 -0.14 0.06
46 0.29 0.13 0.37 0.12 0.20
47 0.32 0.09 0.17 0.01 0.46
48 -0.12 -0.69 -1.56 0.07 0.54
49 1.02 -0.17 -0.06 0.90 1.81
50 1.96 1.78 0.13 1.48 2.46
51 3.80 2.26 2.54 2.32 2.47
52 2.60 1.78 0.70 1.15 2.90
53 -0.33 -0.91 -0.47 -0.02 0.87
54 0.15 -0.52 -0.77 -0.01 1.72
55 1.90 0.39 2.47 1.81 1.84
56 1.20 0.51 2.29 1.19 1.89
57 1.18 1.59 0.81 1.36 1.40
58 4.05 4.44 3.19 4.69 3.21
59 4.42 5.02 3.24 4.90 4.19
60 6.51 7.67 6.16 7.05 6.01
61 3.40 3.45 3.18 2.68 3.03
62 2.68 2.99 2.99 2.09 3.10
63 3.85 3.77 3.76 3.71 3.51
64 2.17 3.23 3.55 1.98 4.18
65 2.14 0.81 3.44 2.53 1.34
66 0.95 0.27 1.05 1.37 3.08
67 2.18 2.11 3.57 2.89 3.51
68 2.38 1.29 2.12 1.43 4.22
69 -0.86 -0.94 0.46 -0.57 1.29
70 0.42 0.62 0.70 0.52 2.83
71 1.11 0.53 3.22 0.95 3.24
72 1.45 1.28 2.25 1.69 4.06
73 0.26 0.18 0.18 0.23 0.09
74 0.07 -0.64 -0.12 0.24 0.90
75 1.81 0.32 2.47 1.80 1.71
76 0.26 -0.46 -0.68 0.10 1.77
77 0.35 0.07 0.58 -0.29 0.85
78 1.27 0.85 2.70 0.72 0.86
79 1.54 0.65 0.99 0.38 1.35
S-51
80 1.26 1.11 -0.42 0.60 1.48
MD: -0.31 -0.55 -0.53 -0.61 –

MAD: 0.95 1.11 1.37 1.00 –
SD: 1.24 1.28 1.62 1.09 –
MAX: 4.41 3.22 4.74 2.80 –
a Reference data taken from Ref. S22.
Table S25: Conformational energies of butane-1,4-diol conformers (BUT14DIOL) S11,S23 com-

puted with different semiempirical methods. The values are given in kcal/mol.
1 -1.07 -0.79 -1.94 -0.41 0.15

2 0.81 0.03 0.75 0.13 0.30
3 0.94 0.08 -0.07 1.02 1.31
4 1.73 0.76 1.19 0.45 1.44
5 0.29 0.16 -1.33 -0.24 1.72
6 -0.05 -0.31 0.53 1.04 2.07
7 3.29 2.65 5.42 2.65 1.77
8 0.34 0.78 1.01 1.85 2.25
9 1.89 1.76 3.29 2.26 2.03
10 2.86 1.84 4.10 2.29 2.10
11 4.72 2.92 6.57 2.98 2.00
12 1.51 1.75 3.20 2.07 2.15
13 0.06 0.64 0.74 1.72 2.22
14 1.49 1.02 1.89 1.91 2.42
15 1.81 2.18 3.30 2.05 2.23
16 2.27 2.20 3.66 2.48 2.25
17 0.55 1.32 1.23 1.77 2.58
18 1.06 1.34 1.60 2.18 2.59
19 1.43 1.70 1.99 2.35 2.63
20 2.05 2.58 3.49 2.08 2.48
21 0.60 1.46 1.44 1.82 2.74
22 2.24 2.53 3.84 2.29 2.55
23 4.01 3.56 6.64 3.25 2.53
24 0.68 1.70 1.17 1.66 2.72
25 1.17 1.70 1.58 2.02 2.69
26 3.27 3.61 5.25 2.13 2.49
27 2.45 2.62 4.28 2.66 2.72
28 1.21 1.75 1.93 2.23 2.83
S-52
29 1.19 1.49 1.85 2.21 2.85
30 1.19 1.49 1.85 2.21 2.85
31 3.60 3.55 5.88 2.65 2.52
32 2.05 2.69 3.22 1.91 2.63
33 1.87 1.74 2.21 2.24 3.10
34 1.10 1.74 1.62 1.98 2.72
35 3.48 2.73 4.62 2.61 2.83
36 0.81 1.78 1.28 1.71 2.79
37 2.18 2.62 3.71 2.26 2.79
38 1.08 1.54 1.50 1.81 3.06
39 3.74 2.81 4.97 2.56 3.10
40 2.26 2.05 2.94 2.32 3.30
41 4.39 3.16 6.05 3.60 3.15
42 2.63 1.96 3.44 2.18 3.29
43 1.04 1.20 1.20 2.22 3.59
44 2.66 2.87 3.88 2.36 3.18
45 2.66 2.87 3.88 2.36 3.18
46 1.37 2.11 1.71 1.86 3.37
47 2.26 2.43 2.49 2.31 3.45
48 2.71 3.14 3.67 2.12 3.33
49 2.71 3.14 3.67 2.12 3.33
50 1.60 2.06 2.23 2.20 3.37
51 1.31 1.36 1.62 2.11 3.61
52 2.83 2.11 3.46 2.52 3.42
53 4.60 4.12 6.63 3.08 3.15
54 2.37 3.03 3.25 2.04 3.31
55 1.22 2.11 1.40 1.81 3.45
56 3.83 4.14 5.31 2.50 3.32
57 1.27 2.19 1.42 1.90 3.57
58 3.34 2.56 4.55 3.38 3.52
59 2.57 3.20 3.35 2.23 3.50
60 1.23 2.21 1.34 1.98 3.65
61 1.98 2.39 1.91 2.25 3.78
62 3.30 3.15 3.77 3.01 4.15
63 2.13 2.22 2.01 2.69 4.31
64 2.41 2.57 2.40 2.96 4.70
MD: -0.82 -0.74 -0.03 -0.69 –

MAD: 1.25 0.95 1.48 0.81 –
SD: 1.19 0.87 1.79 0.65 –
MAX: 2.72 2.39 4.57 1.96 –
S-53
Table S26: Conformational energies of small inorganic molecules (ICONF) S11 computed with
1 0.67 1.09 -1.04 0.85 0.90

2 5.94 5.32 9.41 3.49 5.29
3 1.65 -2.14 -1.77 1.45 0.13
4 4.26 0.61 3.37 3.52 2.33
5 10.41 10.25 4.34 7.78 12.16
6 -0.10 -0.55 0.13 0.00 0.10
7 0.15 0.37 0.72 0.00 1.03
8 0.56 -0.24 2.89 0.00 3.51
9 1.39 2.23 1.16 0.00 1.69
10 -2.88 -2.67 6.83 0.45 1.40
11 2.63 2.11 4.33 1.17 4.39
12 7.50 7.55 9.79 4.01 9.16
13 1.24 0.59 0.73 0.15 0.55
14 6.53 2.92 2.81 0.86 3.55
15 -1.49 -2.33 -3.12 1.52 1.33
16 1.72 -5.35 -7.94 -1.52 3.66
17 3.19 -7.29 -7.42 -2.35 4.35
MD: -0.72 -2.53 -1.78 -2.01 –

MAD: 1.63 2.63 3.13 2.33 –
SD: 1.89 3.29 4.72 2.37 –
MAX: 4.28 11.64 11.78 6.70 –
Table S27: Conformational energies of melatonin conformers (MCONF) S11,S24 computed with
different semiempirical methods. The values are given in kcal/mol. Though the results for all
semiempirical methods apart from GFN2-xTB are also found in Ref. S3 and its supporting
information, they are listed here along with the statistical data, since the reference values
have been revised in Ref. S11.
1 1.06 0.65 1.37 1.46 0.39

2 1.30 1.53 2.11 1.56 1.74
3 1.60 1.27 2.52 2.01 1.16
4 -0.28 0.29 1.91 0.27 2.20
S-54
5 -0.29 -0.05 0.56 -0.63 2.20
6 -0.20 0.52 0.44 -0.17 2.68
7 -0.29 0.48 2.46 0.64 2.92
8 2.15 1.79 3.48 2.62 2.23
9 0.22 1.38 1.42 0.13 2.84
10 2.60 2.40 3.11 4.68 4.24
11 2.90 2.47 3.41 4.68 4.45
12 5.21 3.77 4.17 6.76 3.60
13 2.57 2.11 3.05 2.96 2.25
14 5.44 3.93 4.34 6.92 3.74
15 3.38 2.53 2.23 3.97 5.00
16 3.42 2.78 1.89 4.61 5.11
17 3.53 3.18 4.56 3.94 3.18
18 0.52 1.70 2.14 0.60 3.83
19 1.48 2.06 2.82 1.63 3.80
20 2.82 2.41 3.71 3.38 3.11
21 3.67 3.77 2.98 4.80 5.27
22 3.80 3.80 3.13 4.81 5.31
23 5.92 4.42 4.44 7.51 4.50
24 3.68 4.01 4.90 3.04 3.85
25 5.96 4.44 4.53 7.49 4.55
26 1.83 2.40 3.54 2.12 4.76
27 2.16 2.62 2.94 1.82 4.37
28 6.81 6.07 5.62 7.26 5.27
29 2.24 3.12 3.80 2.35 5.67
30 2.44 3.03 3.33 2.47 4.86
31 5.00 4.50 4.40 6.32 6.24
32 5.08 4.51 4.42 6.34 6.26
33 2.95 3.61 4.54 2.81 5.85
34 3.02 3.97 4.87 2.12 5.37
35 2.60 3.17 3.75 2.56 5.53
36 6.19 5.85 5.21 7.21 7.53
37 2.85 3.50 4.22 3.04 5.88
38 3.41 4.09 4.57 2.84 5.58
39 5.83 5.24 4.59 6.77 6.98
40 5.97 5.34 4.74 6.86 7.07
41 3.51 4.41 5.69 2.67 6.39
42 6.04 5.58 5.02 7.24 7.32
43 6.15 5.65 5.13 7.30 7.39
44 3.89 4.83 5.92 2.98 6.18
45 6.70 6.46 6.42 6.91 7.82
S-55
46 6.74 6.57 6.56 6.94 7.89
47 4.03 4.70 5.58 3.53 6.74
48 7.24 6.79 6.34 7.82 8.19
49 7.24 6.81 6.32 7.82 8.20
50 4.47 5.36 6.89 3.59 7.28
51 7.69 7.48 7.68 7.87 8.75
MD: -1.36 -1.38 -0.98 -0.91 –

MAD: 1.73 1.44 1.34 1.66 –
SD: 1.42 0.89 1.19 1.89 –
MAX: 3.43 2.55 3.22 3.72 –
Table S28: Conformational energies of tri- and tetrapeptide conformers (PCONF21) S11,S25,S26
computed with different semiempirical methods. The values are given in kcal/mol.
1 3.05 3.45 2.36 0.50 0.02

2 5.19 5.69 5.09 2.21 1.01
3 2.81 3.71 4.68 3.93 0.70
4 3.16 3.02 3.10 0.13 0.85
5 1.70 1.62 1.79 2.85 0.78
6 6.16 7.78 7.84 2.47 1.92
7 2.65 2.75 1.86 1.83 2.18
8 3.56 3.16 4.56 4.09 1.61
9 4.43 3.57 3.85 2.54 1.89
10 3.80 5.42 4.81 0.32 2.07
11 0.64 0.41 -2.03 -2.13 1.07
12 -0.58 0.25 2.49 -0.53 1.23
13 2.88 -1.06 1.20 -1.29 2.44
14 0.14 0.77 -0.19 0.64 2.14
15 0.82 1.59 -2.18 -1.45 1.47
16 1.23 2.67 4.16 1.91 2.80
17 2.32 -2.10 0.41 -1.86 2.27
18 1.58 1.97 0.78 2.14 2.74
MD: 0.91 0.86 0.85 -0.60 –

MAD: 1.76 2.17 2.46 1.79 –
SD: 1.97 2.67 2.73 2.12 –
MAX: 4.24 5.86 5.92 4.13 –
S-56
Table S29: Conformational energies of sugar conformers (SCONF) S11,S27 computed with
1 0.72 0.31 0.62 0.29 0.86

2 4.27 3.54 5.97 2.30 2.28
3 4.84 3.75 6.45 2.51 3.08
4 4.46 4.72 7.54 3.13 4.60
5 4.41 4.32 6.48 3.71 4.87
6 5.28 5.69 8.82 4.16 4.16
7 5.70 6.15 9.02 4.60 4.38
8 5.83 6.22 8.84 4.87 6.19
9 5.85 7.37 8.90 4.98 6.18
10 7.01 8.09 10.64 6.03 5.65
11 6.77 7.88 10.62 5.71 5.59
12 4.83 5.64 6.22 3.04 5.93
13 5.65 6.92 8.54 4.67 6.31
14 5.56 7.83 7.97 4.58 6.22
15 -1.36 -1.47 -2.19 -1.73 0.20
16 -0.40 -7.21 -15.21 -1.25 6.16
17 -1.71 -7.06 -14.26 -0.65 5.54
MD: -0.62 -0.91 -0.19 -1.60 –

MAD: 1.64 2.50 4.96 1.69 –
SD: 2.59 4.68 7.93 2.16 –
MAX: 7.25 13.37 21.37 7.41 –
Table S30: Conformational energies of RNA-backbone conformers (UPU23) S11,S28 computed

with different semiempirical methods. The values are given in kcal/mol.
1 9.84 6.46 10.06 7.09 4.87

2 6.16 3.69 4.78 4.35 2.97
3 13.56 11.88 13.33 10.46 8.90
4 6.54 3.05 6.68 2.25 2.22
5 4.32 2.68 2.20 3.32 2.02
6 5.24 2.22 5.29 2.63 3.14
7 4.51 0.92 0.97 -1.86 0.57
S-57
8 3.63 2.36 5.16 2.67 3.32
9 12.20 6.96 10.64 6.50 7.26
10 8.30 5.00 5.31 4.24 3.96
11 14.56 11.98 13.45 11.61 11.13
12 7.77 5.60 9.07 5.17 4.82
13 17.15 13.85 17.73 13.23 14.41
14 8.56 3.38 7.82 4.15 5.15
15 4.86 4.15 4.98 2.78 5.48
16 10.23 7.83 10.24 5.82 6.84
17 4.46 2.21 4.01 1.29 3.90
18 8.57 4.72 6.82 4.78 6.43
19 9.81 4.89 8.83 4.11 5.42
20 5.43 3.53 5.32 3.60 6.70
21 8.05 7.50 10.77 6.97 5.60
22 13.48 12.31 14.87 11.42 10.42
23 7.50 5.10 7.79 4.17 6.09
MD: 2.74 0.03 2.37 -0.47 –

MAD: 2.91 1.24 2.53 1.34 –
SD: 1.72 1.48 1.89 1.53 –
MAX: 4.97 3.17 5.20 3.10 –
a) b) 20 α-maltose
15 α-glucose β-glucose
15
10
10
∆E in kcal mol-1
5
∆E in kcal mol-1
0
0
-5
-5
-10 reference PM6-D3H4X -10 reference PM6-D3H4X

GFN2-xTB DFTB3-D3(BJ) GFN2-xTB DFTB3-D3(BJ)
-15 -15
0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160 180 200 220
conformer number conformer number
Figure S1: Conformational energies a) of 80 α-glucose

1
and 76 β-glucose conformers and b)
of 205 α-maltose conformers. The energies are given in kcal mol−1 and are computed with
GFN2-xTB, DFTB3-D3(BJ), and PM6-D3H4X. The structures and reference conformational
energies are taken from Ref. S29.
S-58
a) b) 15 α-maltose
15 α-glucose β-glucose
10
10
∆E in kcal mol-1
∆E in kcal mol-1
5 5
0
0
-5
reference GFN-xTB reference GFN-xTB
-5
GFN2-xTB DFTB3-D3(BJ) GFN2-xTB DFTB3-D3(BJ)
-10
0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160 180 200 220
conformer number conformer number
Figure S2: Conformational energies a) of 80 α-glucose

1
and 76 β-glucose conformers and b)
of 205 α-maltose conformers. The energies are given in kcal mol−1 and are computed with
GFN-xTB, DFTB3-D3(BJ), and PM6-D3H4X. The structures and reference conformational
energies are taken from Ref. S29.
Table S31: Conformational energies for different glucose conformers S29 computed with dif-
ferent semiempirical methods. The values are given in kcal/mol.
α-glucose
1 -1.47 -1.14 -2.34 -1.95 0.23
2 0.00 0.00 0.00 0.00 0.00
3 -0.28 0.20 -0.16 -0.34 0.10
4 1.69 2.80 3.22 0.38 1.14
5 0.84 1.87 0.55 -1.02 1.28
6 2.19 2.58 2.70 -1.31 1.41
7 1.90 1.97 2.20 1.79 1.90
8 2.00 3.14 3.14 1.91 2.80
9 1.58 3.40 2.81 1.60 2.87
10 1.80 2.63 1.79 2.36 2.48
11 2.79 4.68 4.78 1.82 2.54
12 2.75 3.34 3.83 1.98 2.41
13 2.88 2.87 3.40 1.96 3.28
14 4.11 3.51 5.20 2.53 2.46
15 2.54 1.63 -2.86 1.41 5.65
16 2.29 0.64 -4.11 1.04 5.31
17 2.25 -0.12 -2.84 -0.92 4.91
18 3.39 1.63 -1.86 0.20 5.33
19 1.67 -0.58 -4.67 0.49 6.53
20 2.73 1.89 -2.77 1.40 6.27
S-59
21 2.87 1.80 -1.17 0.32 5.30
22 3.68 2.08 -0.86 0.70 5.93
23 1.62 -0.89 -6.37 1.82 7.12
24 4.57 1.85 -2.05 1.12 6.54
25 5.34 3.33 -0.44 2.56 7.11
26 3.15 1.49 -3.14 1.54 7.71
27 5.17 4.66 1.06 3.67 6.89
28 2.88 -0.22 -7.72 3.43 8.33
29 4.74 4.28 1.40 2.45 6.63
30 4.17 1.29 -4.10 4.12 8.14
31 4.99 4.45 1.47 2.97 6.71
32 4.40 2.69 0.99 0.46 6.35
33 2.68 1.13 -3.54 1.20 6.57
34 5.32 3.09 -0.09 2.27 8.19
35 2.72 1.01 -4.06 0.82 7.59
36 3.48 1.10 -3.86 3.29 9.09
37 5.09 3.63 1.26 2.09 7.35
38 2.92 -0.46 -5.67 1.73 8.51
39 5.49 2.68 1.40 -0.23 7.54
40 3.52 1.19 -3.37 3.00 8.53
41 4.43 1.18 -4.04 4.31 8.93
42 4.54 2.79 -1.57 4.62 9.31
43 6.57 4.83 -0.19 5.57 9.34
44 4.79 3.32 -1.88 3.22 9.02
45 2.79 0.08 -4.38 1.68 9.26
46 5.34 4.72 1.36 3.21 7.94
47 7.15 5.17 0.50 4.77 9.11
48 6.67 8.64 7.01 6.05 9.09
49 6.12 5.62 6.00 3.53 7.91
50 6.40 8.38 7.90 3.42 8.48
51 7.07 8.82 7.74 6.23 8.86
52 4.66 2.15 -0.09 1.88 9.89
53 7.06 8.44 7.39 5.26 9.13
54 6.36 8.56 6.64 4.79 9.13
55 6.36 9.64 7.62 6.43 9.82
56 3.35 0.26 -1.86 1.98 9.94
57 2.11 0.39 -0.07 0.76 5.10
58 5.39 1.14 -3.71 4.03 10.14
59 5.14 5.14 3.40 2.48 9.16
60 6.32 4.90 3.44 4.04 9.62
61 8.18 8.56 6.63 7.45 9.95
S-60
62 5.48 3.32 0.99 3.35 9.59
63 5.45 5.41 5.18 2.78 9.35
64 5.82 7.51 5.70 4.25 10.33
65 7.27 6.50 6.14 5.15 8.93
66 6.21 5.54 4.92 3.59 9.50
67 7.15 6.46 7.26 4.55 9.31
68 7.24 5.26 5.36 5.12 9.15
69 7.29 4.49 4.90 2.90 8.94
70 6.88 4.97 3.13 4.48 9.34
71 7.97 8.54 6.21 8.44 11.14
72 6.99 5.82 3.29 5.56 10.69
73 9.01 7.87 9.70 4.79 9.40
74 6.35 9.25 7.50 5.82 11.52
75 7.94 8.06 8.29 5.17 9.86
76 5.36 7.40 5.27 4.96 11.58
77 8.95 11.09 10.80 6.76 12.08
78 9.67 9.86 9.79 8.10 12.02
79 8.10 9.34 6.57 7.38 13.22
80 8.57 8.49 6.04 8.17 13.84
β-glucose
81 1.46 4.62 5.65 -0.08 1.52
82 2.62 5.52 7.53 1.52 1.29
83 2.82 6.04 7.88 1.61 1.27
84 1.71 -2.21 -7.73 0.46 5.44
85 2.02 4.90 5.72 1.82 2.09
86 1.77 -1.84 -7.53 0.93 5.36
87 1.24 -2.01 -7.59 0.20 5.58
88 1.03 3.74 3.84 0.44 2.47
89 2.42 4.74 5.94 2.30 2.42
90 1.40 -2.17 -8.17 0.59 5.85
91 2.00 -1.88 -7.64 0.43 6.60
92 2.13 -1.41 -7.46 0.77 5.69
93 2.58 -1.26 -6.69 0.73 6.17
94 0.90 -2.79 -8.13 -0.35 6.61
95 2.37 -1.16 -7.25 1.32 6.30
96 4.81 7.91 10.70 3.72 3.26
97 2.55 1.73 -1.48 2.70 5.51
98 4.00 6.63 8.30 3.87 3.75
99 5.21 9.03 11.77 3.52 3.68
100 4.39 7.74 8.15 4.10 4.24
101 3.54 6.95 5.54 2.92 4.35
S-61
102 4.97 8.55 9.60 2.52 4.26
103 4.61 8.41 9.99 4.17 3.79
104 5.78 9.14 12.06 3.90 3.69
105 3.26 2.34 -0.50 3.39 6.83
106 2.96 0.13 -5.94 2.44 7.60
107 2.11 1.62 -1.69 3.34 7.13
108 1.34 -2.33 -8.45 -0.84 6.87
109 2.67 1.42 -1.88 3.49 7.25
110 3.68 3.06 0.27 3.73 6.92
111 5.60 5.83 3.60 5.77 7.65
112 3.42 -0.97 -5.43 1.87 7.62
113 4.46 4.08 1.21 2.52 8.62
114 3.50 2.14 1.28 1.92 7.21
115 3.83 2.25 -0.10 3.27 8.11
116 3.08 1.13 -1.80 2.93 8.21
117 2.91 1.99 0.02 2.81 7.77
118 4.90 4.52 1.67 2.79 9.08
119 4.81 1.99 0.33 2.78 8.39
120 5.41 3.28 0.80 3.99 9.28
121 4.40 3.37 -0.06 2.38 8.70
122 5.26 5.51 2.85 6.30 9.16
123 4.06 3.75 0.54 3.24 9.83
124 4.23 3.41 0.81 2.98 8.73
125 3.26 2.54 1.36 2.73 8.07
126 4.05 4.89 2.49 5.14 8.78
127 3.99 2.18 -0.80 1.62 8.93
128 4.72 2.68 -1.17 2.97 9.14
129 4.19 3.05 0.88 2.60 8.26
130 4.73 4.34 1.68 2.81 9.31
131 5.51 4.39 2.65 4.78 8.75
132 5.73 2.98 -0.78 4.63 9.79
133 5.05 4.00 0.64 3.63 9.60
134 4.04 1.44 -0.32 -0.17 8.37
135 5.67 1.67 -2.78 4.27 11.19
136 3.71 1.36 0.10 2.75 8.65
137 3.32 1.48 0.05 2.90 8.77
138 5.22 2.92 -0.75 4.92 10.31
139 3.98 4.77 2.57 5.27 8.97
140 4.84 6.44 3.33 5.53 9.28
141 4.82 0.68 -2.06 3.02 9.25
142 6.19 4.60 2.85 4.74 9.33
S-62
143 4.40 3.16 0.70 2.98 9.31
144 6.97 6.63 4.61 7.49 10.10
145 6.24 7.78 5.54 6.05 9.52
146 6.43 6.00 3.94 5.68 10.69
147 5.41 2.95 0.94 5.33 11.32
148 7.56 7.83 7.23 5.99 11.16
149 7.85 6.36 3.08 6.80 11.73
150 8.44 8.74 6.14 6.28 11.61
151 7.23 4.36 4.33 6.21 10.94
152 7.92 8.32 7.38 5.73 11.39
153 7.35 5.82 5.19 4.26 11.58
154 8.92 8.49 4.58 6.78 13.13
155 7.16 4.84 3.72 4.63 11.98
156 9.24 6.14 4.58 7.11 12.72
MD: -3.07 -3.79 -6.01 -4.40 –

MAD: 3.24 4.57 7.05 4.42 –
SD: 1.95 3.55 5.40 2.13 –
MAX: 6.59 9.68 16.05 8.54 –
Table S32: Conformational energies for different maltose conformers S29 computed with dif-
1 0.00 0.00 0.00 0.00 0.00

2 1.71 1.47 4.18 1.24 -0.40
3 1.86 2.24 5.83 0.76 0.01
4 -1.33 -0.53 -0.01 -0.87 -0.07
5 1.93 3.85 8.45 0.71 -0.23
6 0.63 1.69 2.06 -2.38 1.76
7 -0.15 0.83 2.02 0.41 0.94
8 1.43 2.44 2.45 2.20 2.51
9 0.94 1.39 0.60 0.37 2.45
10 4.02 2.73 2.98 1.83 3.47
11 2.27 3.24 4.90 2.62 1.99
12 0.73 2.44 -0.82 3.46 3.42
13 1.17 1.31 3.57 1.65 1.85
14 4.71 4.91 8.27 3.17 3.89
15 0.50 -0.22 -3.59 0.56 5.80
16 5.75 6.81 6.92 3.36 5.71
17 2.07 3.39 3.74 1.50 4.97
18 4.83 3.49 2.57 2.86 4.16
S-63
19 3.50 4.34 3.42 0.89 5.18
20 3.85 3.23 -1.56 1.55 6.89
21 6.39 6.19 9.15 4.81 4.18
22 3.94 2.61 3.71 3.90 5.03
23 1.16 2.30 3.86 3.55 4.32
24 2.58 1.90 1.23 1.90 5.86
25 0.90 0.88 -0.61 2.34 7.04
26 3.15 2.91 1.81 3.60 5.31
27 3.08 2.09 -0.04 2.69 7.15
28 3.51 1.47 0.47 1.80 7.06
29 2.81 0.18 -3.75 2.52 7.60
30 3.77 3.80 3.02 2.36 6.81
31 4.01 4.68 5.91 1.90 5.29
32 4.98 4.48 4.89 3.16 4.58
33 5.57 5.63 5.92 2.83 6.30
34 2.05 1.61 -3.73 0.59 9.62
35 5.94 3.51 6.30 4.58 5.28
36 4.67 4.07 -1.05 3.59 8.37
37 5.17 6.82 9.60 3.53 6.47
38 5.23 3.99 2.70 4.16 7.55
39 7.14 8.42 9.74 4.67 7.10
40 6.52 9.36 11.55 6.09 5.65
41 2.18 0.33 -3.09 2.50 7.65
42 6.99 7.28 9.62 6.30 5.87
43 2.98 0.74 -1.54 1.84 7.80
44 2.31 1.02 -2.03 -1.38 8.34
45 7.35 8.97 11.50 6.61 6.59
46 4.07 2.14 1.70 2.37 7.58
47 6.78 6.31 2.91 3.58 7.91
48 4.86 1.17 0.47 2.56 7.14
49 6.20 10.55 13.23 5.53 7.12
50 4.62 1.20 -2.27 0.89 9.22
51 5.59 3.28 -2.04 5.92 8.72
52 3.91 3.60 4.34 2.62 7.07
53 2.33 2.65 1.35 4.14 7.57
54 3.93 2.14 0.51 1.28 6.37
55 5.93 5.28 5.99 1.51 6.81
56 4.77 4.88 6.87 4.10 7.20
57 3.35 4.29 3.40 2.18 8.50
58 4.41 3.33 4.66 2.64 6.47
59 2.54 -0.45 -4.75 2.89 9.51
S-64
60 4.27 3.55 1.43 3.14 9.17
61 4.74 4.09 1.99 2.73 7.89
62 6.32 7.19 4.33 8.31 8.41
63 5.36 3.49 4.05 4.53 9.40
64 8.39 7.97 10.53 5.80 7.21
65 6.69 6.78 5.68 4.11 8.78
66 7.02 6.47 4.11 3.88 8.66
67 6.22 6.78 7.70 7.71 7.48
68 9.14 8.52 6.37 5.90 8.92
69 6.94 9.43 8.22 6.27 8.24
70 4.76 4.91 4.10 4.44 8.22
71 7.18 6.12 7.96 6.42 7.69
72 8.07 10.20 12.11 7.17 7.23
73 5.71 4.70 3.94 2.57 7.01
74 4.74 3.67 2.97 2.52 7.02
75 2.93 1.67 -1.79 3.70 9.32
76 5.07 4.28 1.39 4.04 6.84
77 3.53 2.00 3.51 2.83 8.47
78 6.85 6.90 8.52 4.78 9.06
79 5.44 5.10 4.53 5.63 8.68
80 3.45 1.89 -0.08 3.76 9.20
81 4.20 6.72 6.85 3.62 8.51
82 5.85 8.96 6.84 3.67 9.43
83 5.83 6.01 6.11 6.49 7.14
84 6.55 6.26 10.28 3.59 7.53
85 6.93 6.37 7.96 5.89 7.03
86 6.71 6.97 8.91 2.84 8.42
87 8.12 11.28 12.20 6.13 9.55
88 4.19 5.11 7.88 2.91 7.60
89 4.96 3.89 4.07 3.74 8.00
90 10.12 8.77 12.08 5.30 8.08
91 9.49 9.81 11.10 7.28 9.15
92 3.69 3.75 4.62 1.81 7.85
93 8.87 10.40 12.40 6.46 8.42
94 6.92 6.25 7.58 6.60 7.41
95 5.28 5.73 2.85 2.25 9.95
96 8.69 8.24 8.05 8.62 9.98
97 6.21 6.09 2.97 5.36 9.76
98 9.06 10.82 11.48 8.69 9.44
99 3.23 1.03 -4.68 5.03 11.54
100 7.70 7.59 5.70 3.91 8.67
S-65
101 3.64 2.76 1.59 3.45 9.58
102 6.83 7.17 8.78 5.50 8.15
103 5.43 7.85 8.07 5.19 10.30
104 5.19 5.55 2.42 4.14 9.91
105 5.34 5.41 6.95 1.77 7.69
106 9.06 11.59 12.54 7.87 9.70
107 5.58 2.28 -5.96 4.60 12.84
108 5.36 2.57 -0.05 4.62 9.60
109 6.85 6.56 8.90 5.05 8.94
110 8.65 9.83 11.06 8.61 8.54
111 4.73 1.72 -4.07 4.97 11.45
112 3.77 2.85 2.94 1.26 9.77
113 8.48 6.34 5.84 3.53 11.14
114 4.75 4.12 4.94 2.71 6.99
115 8.75 8.35 9.69 6.18 9.13
116 5.90 9.75 11.03 7.29 9.92
117 4.88 5.75 7.44 3.83 9.25
118 8.16 7.96 9.90 6.56 9.22
119 4.00 3.18 2.12 5.29 11.42
120 5.55 3.93 3.71 2.95 9.10
121 6.96 6.15 6.18 4.15 9.60
122 6.72 4.60 0.47 4.06 11.15
123 7.94 11.80 13.36 7.65 10.00
124 9.32 8.04 5.06 7.81 12.59
125 5.99 6.54 9.18 5.09 9.79
126 3.92 2.22 2.86 1.61 10.05
127 7.79 8.64 9.11 7.81 9.67
128 9.35 7.16 3.77 5.60 11.15
129 3.95 5.05 2.17 6.41 10.32
130 8.97 10.45 13.65 6.61 9.03
131 7.32 7.75 10.18 5.87 8.78
132 6.51 6.27 9.96 3.45 9.48
133 8.48 4.93 2.09 5.92 11.97
134 4.46 1.71 -2.77 3.32 11.94
135 9.94 9.39 12.64 6.44 10.26
136 6.93 5.82 5.46 6.86 11.06
137 6.50 7.06 5.49 5.37 10.16
138 5.27 3.50 0.04 8.35 11.00
139 4.00 1.29 -0.33 3.59 11.28
140 6.22 4.47 7.37 3.64 10.46
141 4.41 3.68 4.29 3.04 9.67
S-66
142 8.96 9.10 10.02 7.82 10.01
143 6.07 4.75 -1.81 6.98 12.48
144 6.59 4.83 -0.15 5.37 12.03
145 10.48 11.41 13.86 6.21 9.61
146 5.72 3.75 2.80 3.38 11.51
147 9.10 6.95 2.61 4.67 12.01
148 2.10 -0.53 -0.36 0.70 7.76
149 5.13 3.76 2.99 4.95 9.99
150 3.51 1.91 0.99 2.41 10.87
151 7.14 5.12 8.96 4.71 8.40
152 8.51 9.77 12.34 7.25 10.59
153 7.00 7.04 8.93 3.18 10.91
154 3.92 2.89 4.86 1.71 10.29
155 11.21 15.33 19.89 6.92 10.29
156 8.43 8.03 7.20 9.28 11.15
157 7.80 6.57 7.99 4.83 11.44
158 8.40 8.17 11.66 6.12 10.28
159 4.23 0.69 -4.44 2.48 11.87
160 5.64 4.89 1.86 6.16 12.31
161 6.87 5.79 5.18 5.54 11.03
162 9.53 8.83 11.29 7.18 10.64
163 4.12 1.78 -0.06 1.53 10.68
164 9.76 12.90 13.71 6.82 11.29
165 9.94 9.09 11.54 8.60 10.90
166 7.61 8.49 11.24 5.35 10.12
167 5.01 5.61 7.07 4.58 8.78
168 7.59 7.49 4.56 7.34 11.24
169 6.32 9.35 10.47 4.53 11.56
170 8.12 5.96 4.07 8.35 12.19
171 5.07 3.39 2.16 4.63 11.67
172 4.24 3.83 2.57 4.64 9.24
173 11.90 12.46 12.44 8.76 12.30
174 11.10 10.22 9.69 8.99 11.92
175 7.05 7.91 9.72 6.83 11.61
176 9.84 10.55 11.85 7.64 11.95
177 7.57 6.25 9.49 4.95 10.02
178 9.99 10.06 11.26 8.69 10.01
179 6.43 5.09 3.13 4.94 12.83
180 8.33 7.88 11.70 4.21 9.89
181 13.94 14.07 14.19 10.02 11.50
182 11.21 10.28 12.64 8.41 11.88
S-67
183 7.52 6.28 8.35 4.16 10.70
184 7.63 6.94 9.24 4.51 11.12
185 10.67 12.00 14.45 8.72 11.14
186 8.51 6.03 3.73 4.80 13.15
187 4.57 2.19 2.40 2.44 9.67
188 10.39 10.14 12.71 9.45 10.48
189 13.69 10.84 2.97 7.98 15.64
190 9.28 9.00 3.62 8.33 13.27
191 10.22 12.94 13.16 9.22 13.02
192 10.90 11.30 11.45 6.28 11.48
193 10.81 11.50 8.64 9.90 12.52
194 8.55 5.21 8.27 6.01 11.00
195 8.41 8.27 7.28 6.02 11.47
196 14.68 13.78 14.67 11.90 12.43
197 9.77 7.28 -0.74 6.02 15.44
198 10.15 9.68 9.08 7.84 11.35
199 8.72 9.16 11.11 6.78 12.22
200 10.48 11.90 15.02 7.98 11.91
201 8.88 7.50 7.75 5.71 14.35
202 8.50 7.11 5.86 6.58 11.30
203 11.19 13.88 11.93 8.29 13.58
204 9.87 9.69 5.09 8.60 13.96
205 8.09 7.85 6.28 5.74 12.66
206 7.71 4.25 2.56 4.09 12.96
207 8.24 8.68 7.05 8.24 13.30
208 9.58 9.29 10.11 8.74 14.86
209 9.58 7.21 3.90 7.16 14.24
210 7.17 2.74 -0.07 4.88 14.41
211 11.27 8.34 5.45 4.98 13.92
212 11.52 6.62 1.71 9.92 16.97
213 8.22 8.06 10.23 3.55 13.26
214 7.99 7.07 7.76 4.99 10.28
215 10.38 7.33 6.68 7.31 13.75
216 12.18 10.49 7.35 7.88 15.38
217 9.01 7.92 3.01 8.43 16.00
218 9.48 8.31 8.14 7.58 13.32
219 9.65 9.47 13.59 8.68 12.66
220 5.13 1.79 -10.60 5.62 16.11
221 14.99 15.32 17.11 11.94 14.17
222 9.39 7.31 1.38 7.58 16.25
223 9.16 13.47 15.98 10.60 13.16
S-68
MD: -2.87 -3.27 -3.64 -4.39 –
MAD: 3.11 3.81 5.07 4.44 –
SD: 2.44 3.47 5.63 2.55 –
MAX: 10.98 14.32 26.71 10.49 –
S-69
2.4 Rotational and vibrational free energy computations
Table S33: Rotational and vibrational reaction free energies ∆GRRHO for the AL2X6 S11 set
computed with different methods. The values are given in kcal/mol and B97-3c served as
reference in the manuscript.
reaction # B97-3c PBEh-3c GFN-xTB GFN2-xTB
1 -1.04 -1.09 -1.35 -1.25

2 -2.16 -2.20 -2.21 -2.43
3 -2.67 -2.70 -2.66 -2.96
4 -4.15 -1.63 -1.80 -2.57
5 -4.54 -1.70 -2.27 -3.10
6 -4.92 -1.50 -2.45 -2.04
MD 1.44 1.12 0.85

MAD 1.49 1.25 1.11
SD 1.65 1.37 1.32
MAX 3.42 2.47 2.87
Table S34: Rotational and vibrational reaction free energies ∆GRRHO for the DARC S11 set
1 0.39 0.41 0.40 0.35

2 -0.01 0.12 0.13 0.09
3 0.53 0.54 0.57 0.57
4 0.06 0.18 0.22 0.28
5 0.58 0.59 0.48 0.52
6 0.04 0.17 0.14 0.27
7 1.31 1.33 1.32 1.32
8 1.32 1.35 1.31 1.46
9 1.21 1.26 1.25 1.29
10 1.22 1.28 1.29 1.34
11 1.14 1.16 1.15 1.15
12 1.15 1.14 1.15 1.13
13 1.04 1.09 1.09 1.12
14 1.04 1.07 1.12 1.19
MD 0.05 0.04 0.07

MAD 0.05 0.06 0.09
SD 0.05 0.07 0.09
S-70
MAX 0.13 0.17 0.22
Table S35: Rotational and vibrational reaction free energies ∆GRRHO for the HEAVYSB11 S11
set computed with different methods. The values are given in kcal/mol and B97-3c served
as reference in the manuscript.
1 -1.37 -1.36 -1.73 -1.59

2 -2.56 -2.68 -2.57 -2.40
3 -5.98 -2.64 -2.71 -2.59
4 -0.39 -0.38 -0.45 -0.48
5 -0.61 -0.59 -0.63 -0.69
6 -2.53 -2.30 -2.67 -2.69
7 -1.80 -1.87 -1.96 -2.01
8 -2.16 -2.23 -2.35 -2.45
9 -2.47 -2.47 -2.62 -2.61
10 -0.14 -0.13 -0.12 -0.14
11 -0.26 -0.25 -0.23 -0.29
MD 0.31 0.20 0.21

MAD 0.35 0.40 0.43
SD 1.01 1.02 1.06
MAX 3.34 3.27 3.40
Table S36: Rotational and vibrational reaction free energies ∆GRRHO for the ISOL24 S11 set
1 0.50 0.18 0.35 0.68

2 2.00 2.12 2.40 2.40
3 0.37 0.56 0.57 0.47
4 4.34 4.50 4.74 4.76
5 1.82 1.77 1.98 2.05
6 0.39 0.36 0.53 0.51
7 0.66 0.72 0.73 0.73
8 -1.38 -1.45 -1.67 -1.58
9 0.20 0.29 0.07 0.11
S-71
10 0.19 0.04 0.17 0.29
11 -0.55 -0.47 -0.74 -0.18
12 0.07 0.00 -0.06 -0.04
13 -0.27 -0.36 -0.29 -0.13
14 0.35 0.33 0.24 0.35
15 0.67 0.79 1.18 1.14
16 0.35 0.35 0.12 0.26
17 0.70 0.76 0.83 0.80
18 0.06 0.07 0.07 0.09
19 -0.06 -0.07 -0.06 -0.08
20 -0.01 -0.03 0.04 0.01
21 1.30 1.29 1.36 1.38
22 -0.83 -0.95 -1.01 -0.94
23 1.16 1.21 1.12 1.18
24 0.47 0.42 0.37 0.25
MD -0.00 0.02 0.08

MAD 0.08 0.15 0.15
SD 0.11 0.21 0.19
MAX 0.32 0.51 0.47
Table S37: Rotational and vibrational reaction free energies ∆GRRHO for the TAUT15 S11
set computed with different methods. The values are given in kcal/mol and B97-3c served
as reference in the manuscript.
1 0.78 0.18 0.17 0.12

2 0.64 0.53 0.49 0.60
3 0.05 0.52 0.02 0.38
4 0.27 0.36 -0.27 0.15
5 0.17 0.27 -0.33 0.12
6 0.03 0.02 0.01 0.01
7 0.13 0.13 0.09 0.13
8 0.00 -0.17 -0.02 -0.19
9 0.07 0.01 0.01 -0.01
10 0.08 -0.00 0.00 -0.03
11 0.09 0.04 0.02 0.02
12 0.04 0.03 0.02 0.03
13 -0.01 -0.02 -0.03 -0.06
14 -0.19 -0.18 -0.07 -0.14
S-72
15 -0.22 -0.25 -0.09 -0.16
MD -0.03 -0.13 -0.07

MAD 0.12 0.16 0.12
SD 0.21 0.23 0.20
MAX 0.60 0.62 0.67
Table S38: Rotational and vibrational reaction free energies ∆GRRHO for the ALK8 S11 set
1 -5.08 -5.42 -4.66 -5.49

2 -6.87 -7.04 -6.55 -6.53
3 -4.41 -4.44 -3.70 -3.75
4 -1.40 -1.97 -1.33 -1.91
5 -0.78 -1.30 -0.77 -0.85
6 -0.93 -1.10 -2.43 -2.33
7 -2.46 -2.45 -2.82 -2.49
8 0.22 0.23 0.15 0.20
MD -0.22 -0.05 -0.18

MAD 0.23 0.43 0.43
SD 0.23 0.67 0.62
MAX 0.57 1.50 1.41
Table S39: Rotational and vibrational reaction free energies ∆GRRHO for the G2RC S11 set
1 -1.51 -1.42 -1.65 -1.61

2 0.05 0.05 0.03 0.02
3 -0.65 -0.65 -0.71 -0.67
4 -1.10 -1.13 -1.31 -1.17
5 0.16 0.16 0.18 0.20
6 -0.12 0.43 0.47 0.50
7 0.59 1.03 0.16 0.47
8 -0.06 0.06 0.05 -0.00
S-73
9 -0.04 -0.06 -0.02 -0.05
10 0.01 0.05 -0.18 -0.14
11 0.15 0.10 0.31 0.27
12 0.28 0.28 0.31 0.25
13 0.10 0.23 0.16 0.11
14 0.15 0.16 0.22 0.25
15 0.04 0.04 0.05 0.03
16 0.20 0.18 0.20 0.23
17 -0.26 -0.13 -0.17 -0.23
18 -0.16 -0.14 -0.14 -0.14
19 0.02 0.03 0.03 0.02
20 -0.21 -0.18 -0.17 -0.18
21 -0.20 -0.18 -0.25 -0.24
22 0.20 0.20 0.17 0.16
23 -0.02 -0.02 -0.02 -0.07
24 -0.17 0.22 0.14 -0.02
25 -0.21 -0.21 -0.11 -0.20
MD 0.08 0.02 0.02

MAD 0.09 0.11 0.07
SD 0.15 0.18 0.14
MAX 0.55 0.60 0.62
Table S40: Rotational and vibrational reaction free energies ∆GRRHO for the ISO34 S11 set
1 -0.17 -0.07 0.24 -0.15

2 -0.47 -0.38 -0.45 -0.47
3 -0.46 -0.47 -0.56 -0.53
4 0.05 0.00 0.01 0.05
5 0.10 0.09 0.06 0.06
6 -0.14 -0.13 -0.18 -0.17
7 -0.51 -0.55 -0.62 -0.61
8 0.45 0.46 0.40 0.45
9 -0.07 -0.07 -0.06 -0.06
10 0.16 0.18 0.01 0.00
11 0.46 0.47 0.34 0.41
12 -0.32 -0.89 -0.87 -0.81
S-74
13 -0.02 -0.06 0.04 0.02
14 0.14 0.13 0.06 0.06
15 0.00 0.01 0.02 0.02
16 -0.40 -0.41 -0.51 -0.46
17 0.09 0.09 0.11 0.17
18 0.24 0.18 0.18 0.20
19 -0.01 -0.00 -0.00 0.03
20 0.05 0.03 0.03 -0.00
21 -0.54 0.03 0.56 0.56
22 0.16 0.13 -0.31 -0.31
23 0.00 -0.00 0.02 0.03
24 0.03 0.04 0.08 0.06
25 -0.47 -0.47 -0.55 -0.50
26 0.48 -0.06 0.02 0.04
27 0.35 -0.27 -0.22 0.44
28 -0.25 -0.32 -0.50 -0.37
29 0.68 0.67 0.74 0.66
30 -0.31 -0.34 -0.22 -0.18
31 0.59 1.10 1.04 1.19
32 1.29 1.27 1.33 1.27
33 0.22 -0.42 0.37 0.40
34 -0.09 -0.14 -0.13 -0.04
MD -0.04 -0.02 0.00

MAD 0.12 0.17 0.14
SD 0.24 0.29 0.27
MAX 0.64 1.10 1.10
Table S41: Rotational and vibrational reaction free energies ∆GRRHO for the MOR41 S30 set
1 1.37 1.27 1.48 1.50

2 1.26 1.10 0.98 1.23
3 1.43 1.45 1.48 1.43
4 0.24 0.35 -1.56 -0.37
5 1.62 1.69 1.67 1.59
6 0.56 0.44 0.41 0.34
7 0.02 -0.11 -0.01 0.17
S-75
8 0.35 0.35 0.31 0.30
9 0.86 0.78 0.20 0.41
10 1.94 1.89 1.94 1.93
11 2.56 2.42 2.66 3.09
12 1.73 1.61 1.47 1.85
13 1.48 1.51 2.18 1.68
14 2.10 2.13 2.11 2.16
15 1.81 1.85 1.96 1.99
16 1.86 2.74 1.86 1.55
17 1.62 1.66 1.78 1.99
18 2.40 2.36 2.49 2.55
19 2.41 2.47 2.07 2.82
20 2.46 2.38 2.49 2.72
21 2.63 1.61 1.66 1.79
22 2.46 2.51 2.76 3.04
23 2.23 1.59 2.40 2.37
24 2.53 2.61 2.61 2.59
25 2.00 2.11 2.32 2.17
26 2.88 2.85 3.10 3.27
27 0.52 0.54 0.53 0.88
28 0.17 0.09 0.23 0.39
29 -0.32 0.35 0.16 0.53
30 0.23 -0.25 0.24 0.35
31 0.07 0.53 0.47 0.66
32 -1.51 -1.50 -1.71 -1.20
33 1.26 1.31 0.83 1.41
34 0.93 0.93 0.78 1.04
35 0.83 1.48 0.92 1.02
36 0.71 0.13 0.81 -0.31
37 3.99 5.50 5.46 5.63
38 5.02 5.10 5.22 4.95
39 2.57 2.54 2.37 2.44
40 2.08 0.83 3.52 2.01
41 1.00 0.68 0.17 0.69
MD -0.01 0.01 0.10

MAD 0.25 0.32 0.31
SD 0.45 0.53 0.44
MAX 1.51 1.80 1.65
S-76
2.5 Other properties
Table S42: Barrier heights of divers reactions (BHDIV10) S11 computed with different
1 32.76 32.42 20.23 23.91 25.65

2 50.31 50.87 31.70 70.47 56.90
3 32.95 36.65 23.99 26.48 36.53
4 87.32 86.99 91.73 86.29 96.17
5 8.05 8.70 12.68 -19.76 15.94
6 7.99 5.65 -9.91 -1.94 13.64
7 28.05 26.27 15.65 – 27.49
8 40.82 39.35 59.35 36.73 50.24
9 59.52 45.74 48.70 – 65.84
10 90.17 79.45 94.91 71.48 64.93
MD: -1.54 -4.12 -6.43 -8.29a –

MAD: 8.12 8.40 14.25 13.32a –
SD: 10.65 9.70 16.36 15.02a –
MAX: 25.24 20.10 29.98 35.70a –
a Missing values are neglected in statistical analysis.
Table S43: Barrier heights of pericyclic reactions (BHPERI) S31 computed with different
1 44.67 43.43 40.18 41.25 35.30

2 23.45 21.99 35.45 25.01 30.80
3 27.95 28.16 38.95 33.97 28.10
4 29.17 29.73 42.56 34.15 39.70
5 27.86 27.74 43.03 25.03 28.30
6 21.96 22.04 42.66 27.46 35.80
7 8.14 8.85 26.06 10.56 22.30
8 7.21 9.22 29.25 12.65 18.00
9 7.12 9.55 31.21 14.51 14.50
10 35.88 39.46 38.10 31.44 26.40
11 12.60 14.01 26.06 31.70 27.60
12 6.46 8.97 20.52 23.07 20.00
13 5.51 5.04 26.97 11.24 13.80
14 4.14 5.79 24.99 9.94 11.80
S-77
15 -1.11 0.24 10.75 8.60 6.50
16 -2.34 -2.60 8.91 4.62 4.70
17 -0.75 -0.22 18.46 8.09 13.10
18 -3.26 -3.46 12.18 3.07 5.90
19 -3.42 -4.50 13.55 -2.62 0.50
20 6.81 8.88 29.99 12.96 18.10
21 5.85 5.96 27.71 – 16.60
22 10.67 11.79 29.19 16.67 22.90
23 12.53 12.35 36.02 23.60 27.80
24 6.27 10.77 36.12 18.25 21.30
25 6.87 10.33 34.54 15.74 21.60
26 14.52 19.36 36.84 23.96 31.30
MD: -8.77 -7.69 8.37 -2.45a –

MAD: 10.22 9.32 8.49 4.54a –
SD: 6.90 6.61 4.84 4.66a –
MAX: 16.78 15.45 16.71 11.74a –
a Missing value is neglected in statistical analysis.
Table S44: Barrier heights of bond rotations around single bonds (BHROT27) S11 computed
with different semiempirical methods. The values are given in kcal/mol.
1 2.58 1.95 1.26 2.18 2.73

2 6.12 4.87 5.49 4.91 7.01
3 2.62 2.50 2.13 3.02 3.46
4 2.55 2.12 2.16 2.64 3.72
5 1.62 1.02 1.05 0.99 1.01
6 2.74 1.53 1.80 1.75 2.28
7 -1.01 -0.79 -0.57 -0.91 1.01
8 7.24 6.31 8.11 3.91 7.17
9 2.06 2.07 3.49 2.05 5.79
10 5.97 4.62 4.81 3.69 8.03
11 1.05 0.17 -1.57 1.84 1.62
12 6.07 3.69 9.64 4.46 8.41
13 6.07 5.02 8.49 3.02 6.91
14 1.49 0.76 0.39 2.73 2.68
15 17.88 15.29 18.30 13.19 17.24
16 14.12 11.75 14.43 11.71 14.52
17 2.64 1.84 3.47 0.66 2.10
S-78
18 5.12 6.67 0.52 6.89 3.89
19 4.36 4.50 -0.73 6.72 2.09
20 3.88 3.52 0.59 4.42 1.78
21 3.01 3.51 -0.60 4.68 1.39
22 5.47 4.85 0.63 5.47 6.30
23 3.16 3.16 -0.03 3.47 3.35
24 10.71 6.54 5.64 14.84 10.36
25 10.50 5.21 5.02 15.01 10.24
26 15.15 10.96 12.17 18.02 17.20
27 14.95 9.63 11.54 18.20 17.08
MD: -0.42 -1.71 -1.92 -0.36 –

MAD: 1.17 2.38 2.38 2.22 –
SD: 1.43 2.49 2.19 2.78 –
MAX: 3.73 7.45 5.67 4.77 –
Table S45: Barrier heights for inversions and racemizations (INV24) S32 computed with dif-
1 31.39 21.45 15.03 16.50 31.70

2 70.15 79.34 66.09 36.13 69.30
3 62.58 69.50 5.59 122.64 60.60
4 42.84 29.21 35.42 31.54 37.00
5 78.08 85.22 71.57 57.43 74.20
6 7.80 10.05 10.48 5.48 9.70
7 14.68 14.29 0.00 10.66 18.90
8 54.13 65.53 34.27 36.74 43.20
9 72.17 60.78 59.94 –a 79.70
10 35.88 44.35 18.40 22.34 31.20
11 27.20 31.69 18.09 18.85 29.30
12 9.88 8.95 10.84 8.54 10.30
13 6.30 4.98 7.45 4.57 4.50
14 24.22 22.25 24.94 24.30 24.70
15 32.88 31.00 31.80 35.17 37.60
16 5.44 4.06 8.29 3.86 4.10
17 10.60 10.66 14.55 10.19 13.10
18 11.70 11.37 11.57 13.80 11.20
19 4.07 5.23 8.49 6.52 6.20
20 24.94 24.55 29.02 26.34 21.30
S-79
21 49.07 45.63 46.86 46.20 42.30
22 22.36 26.80 34.19 25.32 27.20
23 10.21 10.43 11.00 12.43 8.40
24 76.28 74.64 83.64 80.81 68.60
MD: 0.86 1.15 -4.45 -1.23a –

MAD: 3.45 5.80 8.59 9.07a –
SD: 4.39 8.37 13.82 16.52a –
MAX: 10.93 22.33 55.01 62.04a –
a Abnormally high repulsion energy in transition state geometry of PCL3 . Hence, this
value is neglected in statistical analysis.
Table S46: Barrier heights for proton exchange reactions (PX13) S11,S33 computed with dif-
1 67.02 71.22 60.14 69.08 59.30

2 40.36 43.69 37.03 52.38 46.90
3 46.16 54.41 33.18 56.37 48.40
4 51.94 47.30 46.39 49.94 48.60
5 26.94 24.94 18.06 26.96 29.80
6 22.60 23.88 12.18 22.94 26.60
7 27.76 29.23 10.56 26.17 30.10
8 34.96 35.86 14.83 30.07 35.10
9 39.42 29.90 59.91 46.46 42.30
10 18.74 5.51 43.13 31.96 20.70
11 14.16 0.34 36.79 30.41 14.70
12 14.95 -1.20 38.41 33.70 14.60
13 17.26 -1.91 44.24 38.92 16.60
MD: -0.88 -5.43 1.63 6.28 –

MAD: 2.74 8.30 15.98 8.66 –
SD: 3.55 9.23 18.42 9.10 –
MAX: 7.72 18.51 27.64 22.32 –
Table S47: Barrier heights for proton transfer reactions (WCPT18) S11,S34 computed with
S-80
1 35.39 36.90 42.20 33.97 36.76

2 30.09 31.33 35.27 37.22 36.21
3 62.98 64.45 61.14 61.42 60.95
4 36.65 46.46 47.61 53.16 47.52
5 61.07 73.25 65.97 72.62 65.68
6 79.41 77.77 87.58 84.77 81.24
7 30.39 40.97 36.53 37.22 32.00
8 21.44 35.81 31.36 35.76 28.97
9 51.32 65.56 61.34 62.72 58.80
10 2.89 -0.83 0.57 3.90 5.40
11 1.54 -4.66 -1.05 4.95 2.68
12 27.57 23.68 23.93 25.81 28.78
13 9.92 9.78 6.61 11.71 8.68
14 28.72 33.73 31.03 26.27 33.89
15 51.74 49.35 44.90 51.98 59.63
16 3.84 12.18 4.28 9.73 5.83
17 0.86 6.91 0.14 7.83 3.54
18 31.38 45.51 34.90 37.03 33.22
MD: -3.48 1.02 -0.86 1.57 –

MAD: 3.84 5.30 3.47 4.08 –
SD: 3.41 6.37 4.83 4.45 –
MAX: 10.87 12.29 14.73 7.65 –
Table S48: Molecular dipole moments computed with GFN2-xTB, GFN-xTB, and PM6.
The reference is a CCSD(T)/CBS estimate and taken from Ref. S35. All quantities are
given in Debye.
system Nα − Nβ GFN2-xTB GFN-xTB PM6 ref.
AlF 0 1.7690 3.3490 1.9660 1.4729

AlH2 1 1.3360 1.6550 0.2020 0.4011
BeH 1 1.3590 1.2270 0.5000 0.2319
BF 0 1.4760 0.5170 0.2550 0.8194
BH 0 1.5380 0.9280 0.5070 1.4103
BH2 1 0.0570 0.3170 0.5040 0.5004
BH2 Cl 0 1.0480 0.8410 1.0030 0.6838
S-81
BH2 F 0 0.8240 0.9630 1.1990 0.8269
BHCl2 0 1.0090 1.0440 1.0360 0.6684
BHF2 0 1.0320 1.1160 1.3590 0.9578
BN 2 2.1650 2.0600 2.5140 2.0366
BO 1 2.1910 2.8970 2.2260 2.3171
BS 1 2.3410 1.1380 1.7720 0.7834
C2 H 1 1.0190 0.5960 0.3090 0.7601
C2 H 3 1 0.8190 0.8290 0.7930 0.6867
C2 H 5 1 0.5610 0.5240 0.4780 0.3140
CF 1 1.0930 0.6960 0.0080 0.6793
CF2 0 0.3990 0.0910 0.3440 0.5402
CH 1 1.6430 1.6120 1.3570 1.4328
CH2 BH 0 0.2670 0.5200 1.4510 0.6238
CH2 BOH 0 2.5010 2.8340 2.5990 2.2558
CH2 F 1 1.5450 1.5540 1.3770 1.3796
CH2 NH 0 1.9890 2.3310 2.4070 2.0673
CH2 PH 0 1.1980 0.5680 1.2460 0.8748
CH2 -singlet 0 1.7970 1.9180 1.9270 1.4942
CH2 -triplet 2 0.6790 0.5910 0.7780 0.5862
CH3 BH2 0 0.8140 0.8650 0.6100 0.5751
CH3 BO 0 3.2660 4.0640 3.3970 3.6779
CH3 Cl 0 1.9770 1.8720 1.9800 1.8981
CH3 F 0 2.2740 2.1760 1.6350 1.8083
CH3 Li 0 4.6150 4.2520 5.2400 5.8304
CH3 NH2 0 1.5380 1.8970 2.0520 1.3876
CH3 O 1 3.3450 4.0510 2.3290 2.0368
CH3 OH 0 1.9690 2.4940 2.1250 1.7091
CH3 SH 0 2.3830 1.4790 1.7700 1.5906
ClCN 0 3.2960 3.0390 2.5610 2.8496
ClF 0 1.5800 1.5770 0.0830 0.8802
ClO2 1 2.1230 3.2720 5.1490 1.8627
CN 1 0.6100 1.5010 1.3500 1.4318
CO 0 0.6140 0.5040 0.0960 0.1172
S-82
CS 0 1.5060 1.6040 0.9750 1.9692
CSO 0 0.2610 1.6170 0.4800 0.7327
FCN 0 1.7630 2.0650 1.9950 2.1756
FCO 1 0.8190 1.6440 0.5290 0.7678
FH-BH2 1 3.0470 3.4460 2.0560 2.9730
FH-NH2 1 4.5960 5.7790 4.4340 4.6265
FH-OH 1 2.7590 2.8860 2.6710 3.3808
FNO 0 1.6060 1.4040 1.8310 1.6971
H2 CN 1 2.9730 3.0610 2.8030 2.4939
H2 O 0 2.2710 2.8400 2.1170 1.8601
H2 O-Al 1 4.0710 1.2550 3.4880 4.3573
H2 O-Cl 1 2.7640 3.2920 2.3630 2.2383
H2 O-F 1 3.3960 3.6030 2.1940 2.1875
H2 O-H2 O 0 2.9470 3.6130 2.6880 2.7303
H2 O-Li 1 6.0590 4.2560 0.9300 3.6184
H2 O-NH3 0 3.7330 4.5840 4.0230 3.5004
H2 S-H2 S 0 1.2410 1.1010 0.9520 0.9181
H2 S-HCl 0 2.3110 3.0090 2.3880 2.1328
HBH2 BH 0 1.5500 1.3870 2.0880 0.8429
HBO 0 2.1720 3.0380 2.6370 2.7322
HBS 0 2.4970 1.4420 1.7920 1.3753
HCCCl 0 0.2030 0.2530 0.3550 0.5009
HCCF 0 1.4140 0.9450 0.5750 0.7452
HCHO 0 2.3870 3.3280 2.8020 2.3927
HCHS 0 2.2010 1.5610 1.4730 1.7588
HCl 0 1.0580 1.8390 1.4580 1.1055
HCl-HCl 0 1.7160 2.5480 1.9740 1.7766
HCN 0 2.6600 2.9140 2.6740 3.0065
HCNO 0 3.0650 3.5010 1.9790 2.9560
HCO 1 1.6680 2.5050 1.8220 1.6912
HCOF 0 2.0880 2.8230 2.3230 2.1169
HCONH2 0 4.3700 5.1600 4.1830 3.9152
HCOOH 0 1.7630 1.8870 1.5510 1.3835
S-83
HCP 0 0.9730 0.9600 0.5610 0.3542
HF 0 2.5140 2.3850 1.3600 1.8059
HF-HF 0 4.3790 4.1870 2.3840 3.3991
HN3 0 2.0470 2.3340 2.1060 1.6603
HNC 0 2.7090 3.1300 2.4480 3.0818
HNCO 0 2.2530 2.9820 2.2570 2.0639
HNO 0 1.7270 2.8660 1.8610 1.6536
HNO2 0 2.5880 3.6690 1.5300 1.9345
HNS 0 1.4520 1.9980 1.8370 1.4062
HO2 1 2.9370 3.3380 2.0910 2.1659
HOCl 0 1.5840 2.4420 1.6260 1.5216
HOCN 0 4.0520 4.2440 3.4310 3.7998
HOF 0 2.3570 2.4980 1.6050 1.9168
HOOH 0 2.0350 2.4110 1.5760 1.5732
HPO 0 2.4570 4.0430 3.1330 2.6291
LiBH4 0 5.7880 5.2390 7.4990 6.1281
LiCl 0 6.4350 6.6300 7.8990 7.0960
LiCN 0 6.9370 6.5560 7.4270 6.9851
LiF 0 6.2180 5.6960 6.4710 6.2879
LiH 0 6.2200 6.3000 3.8400 5.8286
LiN 2 6.2690 5.8470 7.0160 7.0558
LiOH 0 3.8580 3.0640 4.3200 4.5664
N2 H2 0 2.8770 3.7340 3.4080 2.8771
N2 H4 0 3.0760 3.7680 3.7040 2.7179
NaCl 0 7.3410 9.8670 9.8760 9.0066
NaCN 0 8.0880 9.1610 9.0750 8.8903
NaF 0 7.5970 8.0630 8.3250 8.1339
NaH 0 7.0670 7.9210 5.8930 6.3966
NaLi 0 1.9860 3.2990 6.2780 0.4837
NaOH 0 5.7750 5.6160 6.1820 6.7690
NCl 2 2.4670 2.5910 1.8780 1.1279
NCO 1 0.9960 0.2820 1.4860 0.7935
NF 2 0.5030 0.4910 0.6230 0.0671
S-84
NF2 1 0.2140 0.1440 0.6670 0.1904
NH 2 1.7030 2.2410 1.8270 1.5433
NH2 1 1.9950 2.5490 2.3290 1.7853
NH2 Cl 0 1.7410 2.4130 2.4130 1.9468
NH2 F 0 2.6380 2.9490 2.3520 2.2688
NH2 OH 0 0.8060 1.3610 0.9420 0.7044
NH3 0 1.8390 2.1680 2.3150 1.5289
NH3 -BH3 0 6.1630 6.2280 5.8760 5.2810
NH3 -NH3 0 2.2460 2.6630 2.7650 2.1345
NH3 O 0 6.4900 7.2100 6.3200 5.3942
NO 1 0.1730 0.6150 0.7770 0.1271
NO2 1 0.9040 1.4740 0.5770 0.3350
NOCl 0 0.2430 0.7030 3.1080 2.0773
NP 0 1.9520 4.0770 2.6680 2.8713
NS 1 2.0260 1.9780 1.8140 1.8237
O3 0 1.1780 1.3300 2.0050 0.5666
OCl 1 2.3060 3.5640 2.0680 1.2790
OCl2 0 0.8550 1.8340 0.8530 0.5625
OF 1 0.5810 1.1200 0.4650 0.0205
OF2 0 0.4580 0.1690 0.5870 0.3252
OH 1 2.0750 2.5100 1.4920 1.6550
P2 H4 0 1.6320 1.4440 2.7680 0.9979
PCl 2 0.6730 1.0530 0.3530 0.5657
PF 2 0.8810 1.1920 0.8570 0.8104
PH 2 1.1180 0.6230 1.4470 0.4375
PH2 1 1.2320 0.8210 1.8020 0.5472
PH2 OH 0 0.5050 1.2450 0.7790 0.6836
PH3 0 1.1930 1.0330 1.9480 0.6069
PH3 O 0 3.2420 4.4790 5.5290 3.7704
PO 1 1.7150 3.4710 1.9100 1.9617
PO2 1 1.1350 2.1950 2.1090 1.4426
PPO 0 2.0310 3.2430 0.9870 1.8812
PS 1 0.8040 1.8950 0.5970 0.6825
S-85
S2 H2 0 1.8460 1.5780 1.3300 1.1425
SCl 1 1.3440 0.4260 0.0500 0.0690
SCl2 0 0.5170 0.6470 0.7220 0.3891
SF 1 0.8180 1.1010 0.9750 0.8139
SF2 0 1.3920 1.5520 1.5240 1.0555
SH 1 1.4420 1.2010 0.7880 0.7727
SH2 0 1.9480 1.5860 1.3710 0.9939
SiH 1 0.2730 0.6130 0.6440 0.1138
SiH3 Cl 0 1.7150 1.0290 1.4490 1.3645
SiH3 F 0 1.3580 0.9240 1.3790 1.3123
SiO 0 3.0560 5.1760 5.3940 3.1123
SO2 0 2.9590 2.3450 3.3750 1.6286
SO-triplet 2 1.9100 2.4540 2.1480 1.5606
S-86
3 Element-specific parameters in GFN2-xTB
Table S49: Element-specific atomic parameters employed in GFN2-xTB: atomic Hubbard

parameter (ηA ), its charge derivative (ΓA ), the exponential scaling parameter αA and YAeff
µA ΘA
(both entering the repulsion potential), the anisotropic XC scaling parameters fXC and fXC ,
A
and the offset radius R0 for the damping in the AES energy. All quantities are given in
atomic units.
µA ΘA
element ηA ΓA αA YAeff fXC fXC R0A
H 0.405771 0.08 2.213717 1.105388 0.0556389 0.00027431 1.4

He 0.642029 0.2 3.604670 1.094283 -0.01 -0.00337528 3.0
Li 0.245006 0.130382 0.475307 1.289367 -0.005 0.0002 5.0
Be 0.684789 0.0574239 0.939696 4.221216 -0.00613341 -0.00058586 5.0
B 0.513556 0.0946104 1.373856 7.192431 -0.00481186 -0.00058228 5.0
C 0.538015 0.15 1.247655 4.231078 -0.00411674 0.00213583 3.0
N 0.461493 -0.063978 1.682689 5.242592 0.0352127 0.0202679 1.9
O 0.451896 -0.0517134 2.165712 5.784415 -0.0493567 -0.00310828 1.8
F 0.531518 0.142621 2.421394 7.021486 -0.0833918 -0.00245955 2.4
Ne 0.850000 0.05 3.318479 11.041068 0.1 -0.005 5.0
Na 0.271056 0.179873 0.572728 5.244917 0 0.0002 5.0
Mg 0.344822 0.234916 0.917975 18.083164 -0.00082005 -0.00005516 5.0
Al 0.364801 0.14 0.876623 17.867328 0.0263334 -0.00021887 5.0
Si 0.720000 0.193629 1.187323 40.001111 -0.0002575 -0.0008 3.9
P 0.297739 0.0711291 1.143343 19.683502 0.0211022 0.00028679 2.1
S 0.339971 -0.0501722 1.214553 14.995090 -0.00151117 0.00442859 3.1
Cl 0.248514 0.149548 1.577144 17.353134 -0.0253696 0.00122783 2.5
Ar 0.502376 -0.0315455 0.896198 7.266606 -0.0207733 -0.010834 5.0
K 0.247602 0.203309 0.482206 10.439482 -0.00103383 0.00025 5.0
Ca 0.320378 0.20069 0.683051 14.786701 -0.00236675 0.0001 5.0
Sc 0.472633 0.05 0.574299 8.004267 -0.00515177 -0.00042004 5.0
Ti 0.513586 0.176727 0.723104 12.036336 -0.00434506 0.0005966 5.0
V 0.589187 0.09 0.928532 15.677873 -0.0035 0.00009764 5.0
Cr 0.396299 0.03 0.966993 19.517914 0.00149669 0.00137744 5.0
Mn 0.346651 0.06 1.071100 18.760605 -0.00759168 0.00229903 5.0
S-87
Fe 0.271594 -0.05 1.113422 20.360089 0.00412929 0.00267734 5.0
Co 0.477760 0.03 1.241717 27.127744 -0.00247938 0.00048237 5.0
Ni 0.344970 -0.02 1.077516 10.533269 -0.0126189 -0.0008 5.0
Cu 0.202969 0.05 0.998768 9.913846 -0.007 -0.00345631 5.0
Zn 0.564152 0.23129 1.160262 22.099503 -0.001 0.00007658 5.0
Ga 0.432236 0.233427 1.122923 31.146750 0.00267219 -0.00003616 5.0
Ge 0.802051 -0.0064775 1.222349 42.100144 0.0010846 -0.00003589 5.0
As 0.571748 0.110604 1.249372 39.147587 -0.00201294 0.00014149 5.0
Se 0.235052 0.0913725 1.230284 27.426779 -0.00288648 0.00085728 3.9
Br 0.261253 0.13 1.296174 32.845361 -0.0108859 0.00216935 4.0
Kr 0.424373 0.0239815 0.908074 17.363803 -0.00889357 -0.00415024 5.0
Rb 0.210481 0.29162 0.574054 44.338211 -0.00093328 0.00015 5.0
Sr 0.340000 0.18 0.697345 34.365525 -0.00459925 0.00015 5.0
Y 0.711958 0.01 0.706172 17.326237 -0.00637291 0.0001046 5.0
Zr 0.461440 0.07 0.681106 24.263093 -0.00599615 -0.00012944 5.0
Nb 0.952957 0.05 0.865552 30.562732 -0.00288729 0.00041491 5.0
Mo 0.586134 0.0919928 1.034519 48.312796 0.00346327 0.00312549 5.0
Tc 0.368054 0.06 1.019565 44.779882 -0.00458416 0.00155242 5.0
Ru 0.711205 -0.05 1.031669 28.070247 -0.00081922 0.00359228 5.0
Rh 0.509183 0.03 1.094599 38.035941 0.00007016 0.0000857 5.0
Pd 0.273310 0.08 1.092745 28.674700 -0.00310361 -0.00040485 5.0
Ag 0.263740 0.02 0.678344 6.493286 -0.00800314 -0.0002081 5.0
Cd 0.392012 0.207322 0.936236 26.226628 -0.00105364 0.0001225 5.0
In 0.461812 0.19 1.024007 63.854240 0.00951079 -0.00002031 5.0
Sn 0.900000 -0.0178396 1.139959 80.053438 0.00085029 -0.00008243 5.0
Sb 0.942294 0.11 1.122937 77.057560 -0.00015519 -0.0002063 5.0
Te 0.750000 0.0953683 1.000712 48.614745 -0.00263414 -0.00026864 5.0
I 0.383124 0.12 1.017946 63.319176 -0.00603648 0.0006966 5.0
Xe 0.424164 -0.0118925 1.012036 51.188398 -0.00214447 -0.001562 5.0
Cs 0.236569 0.240419 0.585257 67.249039 -0.0008 0.00008 5.0
Ba 0.245937 0.20691 0.716259 46.984607 -0.0026 0.00015 5.0
La 0.597716 0.0012793 0.737643 50.927529 -0.00395198 -0.0003 5.0
Ce 0.662889 -0.01 0.729950 48.676714 -0.00723806 -0.00025 5.0
S-88
Pr 0.660710 -0.0100002 0.734624 47.669448 -0.00704819 -0.00024615 5.0
Nd 0.658531 -0.0100004 0.739299 46.662183 -0.00685832 -0.00024231 5.0
Pm 0.656352 -0.0100006 0.743973 45.654917 -0.00666845 -0.00023846 5.0
Sm 0.654173 -0.0100008 0.748648 44.647651 -0.00647858 -0.00023462 5.0
Eu 0.651994 -0.010001 0.753322 43.640385 -0.00628871 -0.00023077 5.0
Gd 0.649815 -0.0100012 0.757996 42.633120 -0.00609884 -0.00022692 5.0
Tb 0.647635 -0.0100013 0.762671 41.625854 -0.00590897 -0.00022308 5.0
Dy 0.645456 -0.0100015 0.767345 40.618588 -0.0057191 -0.00021923 5.0
Ho 0.643277 -0.0100017 0.772020 39.611322 -0.00552923 -0.00021538 5.0
Er 0.641098 -0.0100019 0.776694 38.604057 -0.00533936 -0.00021154 5.0
Tm 0.638919 -0.0100021 0.781368 37.596791 -0.00514949 -0.00020769 5.0
Yb 0.636740 -0.0100023 0.786043 36.589525 -0.00495961 -0.00020385 5.0
Lu 0.634561 -0.0100025 0.790717 35.582259 -0.00476974 -0.0002 5.0
Hf 0.662597 -0.01 0.852852 40.186772 -0.00537685 -0.00016478 5.0
Ta 0.449812 0.02 0.990234 54.666156 -0.00200343 0.00039599 5.0
W 0.685426 -0.02 1.018805 55.899801 0.00065886 0.0106331 5.0
Re 0.224623 0.08 1.170412 80.410086 -0.00587636 0.0030687 5.0
Os 0.364388 0.08 1.221937 62.809871 -0.0051009 0.00759049 5.0
Ir 0.548507 -0.01 1.197148 56.045639 -0.00673822 0.00322935 5.0
Pt 0.353574 0.06 1.204081 53.881425 -0.00423684 0.00098019 5.0
Au 0.438997 0.085 0.919210 14.711475 0.00393418 -0.0002032 5.0
Hg 0.457611 -0.0116312 1.137360 51.577544 -0.0025 -0.00032901 5.0
Tl 0.418841 -0.0533933 1.399312 58.801614 0.00374018 -0.00008506 5.0
Pb 0.168152 0.02 1.179922 102.368258 0.0100702 -0.0000167 5.0
Bi 0.900000 -0.0337508 1.130860 132.896832 -0.00737252 0.00162529 5.0
Po 1.023267 0.187798 0.957939 52.301232 -0.0134485 0.00013818 5.0
At 0.288848 0.184648 0.963878 81.771063 -0.00348123 0.00021624 5.0
Rn 0.303400 0.0097834 0.965577 128.133580 -0.00167597 -0.00111556 5.0
(a)
It is noted that R0A is a fitted parameter only for 12 elements and set to a value of 5.0 for the
rest of the periodic table.
S-89
Table S50: Element-specific shell parameters employed in GFN2-xTB: the polynomial scaling
poly
parameters kA,l , the shell-specific scaling parameters of the Hubbard parameter κlA , the
CNA0 dependent enhancement factors for the energy levels, the constant part of the energy
levels (HAl ), and the corresponding Slater exponents ζl . The energy levels and their CNA0
poly
dependent enhancement factors are given in eV, ζl is given in atomic units, whereas kA,l
and κlA are dimensionless.
poly
element level kA,l κlA (a) l
HCN 0 / eV H l / eV ζl
A
H 1s -0.00953618 0.0 -0.05 -10.707211 1.230000

He 1s -0.0438682 0.0 0.207428 -23.716445 1.669667
2p 0.00710647 0.0 0.0 -1.822307 1.500000
Li 2s -0.047504 0.0 0.162084 -4.900000 0.750060
2p 0.204249 0.197261 -0.0623876 -2.217789 0.557848
Be 2s -0.0791039 0.0 0.118776 -7.743081 1.034720
2p -0.00476438 0.965847 0.0550528 -3.133433 0.949332
B 2s -0.0518315 0.0 0.0120462 -9.224376 1.479444
2p -0.0245332 0.399408 -0.0141086 -7.419002 1.479805
C 2s -0.0229432 0.0 -0.0102144 -13.970922 2.096432
2p -0.00271102 0.105636 0.0161657 -10.063292 1.800000
N 2s -0.08506 0.0 -0.195534 -16.686243 2.339881
2p -0.025042 0.116489 0.0561076 -12.523956 2.014332
O 2s -0.149553 0.0 0.0117826 -20.229985 2.439742
2p -0.0335082 0.149702 -0.0145102 -15.503117 2.137023
F 2s -0.130119 0.0 0.0394362 -23.458179 2.416361
2p -0.123008 0.167738 -0.0538373 -15.746583 2.308399
Ne 2s -0.163778 0.0 -0.0014933 -24.500000 3.084104
2p -0.0486055 0.119058 0.0232093 -18.737298 2.312051
3d -0.169223 -0.32 0.109671 -5.517827 2.815609
Na 3s -0.040335 0.0 -0.0042211 -4.546934 0.763787
3p 0.208739 0.101889 -0.0144323 -1.332719 0.573553
Mg 3s -0.111674 0.0 0.116444 -6.339908 1.184203
3p 0.39077 1.4 -0.0079924 -0.697688 0.717769
3d 0.126911 -0.05 0.119241 -1.458197 1.300000
Al 3s -0.106781 0.0 0.0715422 -9.329017 1.352531
S-90
3p -0.124428 -0.0603699 -0.0244485 -5.927846 1.391201
3d 0.163111 0.2 0.0406173 -3.042325 1.000000
Si 3s 0.0235852 0.0 0.185848 -14.360932 1.773917
3p -0.0790041 -0.558004 -0.138307 -6.915131 1.718996
3d 0.113662 -0.23 -0.193549 -1.825036 1.250000
P 3s -0.198318 0.0 0.054761 -17.518756 1.816945
3p -0.0551558 -0.155806 -0.048993 -9.842286 1.903247
3d 0.263975 -0.35 0.242951 -0.444893 1.167533
S 3s -0.258555 0.0 -0.0256951 -20.029654 1.981333
3p -0.0804806 -0.108587 -0.0098465 -11.377694 2.025643
3d 0.259939 -0.25 0.200769 -0.420282 1.702555
Cl 3s -0.16562 0.0 0.0617972 -29.278781 2.485265
3p -0.0698643 0.49894 -0.0181618 -12.673758 2.199650
3d 0.380456 0.5 0.167277 -0.240338 2.476089
Ar 3s -0.238939 0.0 0.0000554 -16.487730 2.329679
3p -0.0372732 -0.0461133 0.0065921 -13.910539 2.149419
3d 0.268129 -0.01 -0.273217 -1.167213 1.950531
K 4s -0.0607606 0.0 -0.0339245 -4.510348 0.875961
4p 0.211873 0.348366 0.0174542 -0.934377 0.631694
Ca 4s -0.0971872 0.0 0.057093 -5.056506 1.267130
4p 0.319734 1.5 -0.0074926 -1.150304 0.786247
3d 0.0952865 -0.25 0.101375 -0.776883 1.380000
Sc 3d -0.345023 -0.08 0.202678 -5.196187 2.440000
4s 0.00686569 0.0 0.0991293 -8.877940 1.358701
4p 0.380449 -0.204672 -0.0281241 -2.008206 1.019252
Ti 3d -0.277244 -0.38 0.102819 -7.234331 1.849994
4s 0.0456123 0.0 0.100702 -10.900000 1.469983
4p 0.518016 -0.492111 -0.0237074 -1.928783 0.957410
V 3d -0.298276 -0.45 0.0164476 -9.015342 1.673577
4s 0.0970248 0.0 0.0235696 -9.573347 1.383176
4p 0.511783 -0.0379088 -0.0108232 -0.706647 0.938025
Cr 3d -0.279716 -0.47 0.0289291 -7.209794 1.568211
4s 0.133762 0.0 -0.0232087 -9.201304 1.395427
S-91
4p 0.480922 0.740587 -0.0188919 -0.696957 1.080270
Mn 3d -0.312559 -0.6 -0.0195827 -10.120933 1.839250
4s 0.285197 0.0 -0.0275 -5.617346 1.222190
4p 0.263466 0.0545811 -0.0015839 -4.198724 1.240215
Fe 3d -0.28615 -0.65 -0.0274654 -10.035473 1.911049
4s 0.115278 0.0 -0.404988 -5.402911 1.022393
4p 0.394599 0.404662 -0.075648 -3.308988 1.294467
Co 3d -0.223556 -0.65 0.012198 -10.580430 2.326507
4s 0.0916846 0.0 -0.0227872 -8.596723 1.464221
4p 0.254247 -0.241849 0.0076513 -2.585753 1.298678
Ni 3d -0.253856 -0.6 -0.0066417 -12.712236 2.430756
4s 0.208395 0.0 0.0310301 -8.524281 1.469945
4p 0.308864 -0.0611188 0.0226796 -2.878873 1.317046
Cu 3d -0.265089 0.07 -0.0173684 -9.506548 2.375425
4s 0.177983 0.0 0.334905 -6.922958 1.550837
4p 0.149778 1.33331 -0.261945 -2.267723 1.984703
Zn 4s -0.0924032 0.0 0.201191 -7.177294 1.664847
4p 0.222718 0.0684343 -0.0055135 -0.991895 1.176434
Ga 4s -0.190182 0.0 -0.0234627 -12.449656 1.720919
4p -0.0113779 -0.541655 0.130583 -4.469873 1.591570
4d 0.354019 -0.3 0.0165604 -0.582255 1.050000
Ge 4s -0.213337 0.0 0.0361068 -16.369792 1.990429
4p -0.0974904 -0.380909 -0.0014474 -8.207673 1.830340
4d 0.286347 -0.15 -0.104256 -0.994226 1.100000
As 4s -0.238207 0.0 -0.012964 -16.421504 2.026128
4p -0.106442 -0.410474 -0.023647 -9.311147 1.949257
4d 0.307111 -0.5 0.233014 -0.276830 1.040181
Se 4s -0.245064 0.0 -0.0061654 -20.584732 2.230969
4p -0.137658 0.119211 -0.0435018 -10.910799 2.150656
4d 0.296111 -0.25 0.276856 -0.110636 1.317549
Br 4s -0.250051 0.0 0.000615 -23.583718 2.077587
4p -0.145201 0.5203 -0.0058347 -12.588824 2.263120
4d 0.36614 0.4 0.225018 0.047980 1.845038
S-92
Kr 4s -0.326587 0.0 -0.0070305 -17.221422 2.445680
4p -0.136001 -0.250322 0.0076023 -13.633377 2.210494
4d 0.232047 -0.07 0.0349523 -0.940657 1.884991
Rb 5s 0.043254 0.0 -0.151693 -4.353793 1.017267
5p 0.232551 0.938649 0.0203437 -1.392938 0.870130
Sr 5s -0.145068 0.0 0.040902 -6.291692 1.419028
5p 0.20214 1.5 -0.0418725 -1.872475 0.928932
4d 0.108162 -0.25 0.0401255 -0.890492 1.500000
Y 4d -0.395295 -0.45 -0.127034 -8.015206 2.670141
5s -0.0212587 0.0 0.193752 -12.194181 1.633876
5p 0.521619 -0.334929 -0.0641897 -0.966195 1.165412
Zr 4d -0.283589 -0.11 -0.0566943 -7.409832 2.238668
5s 0.075389 0.0 0.126655 -10.199105 1.702480
5p 0.589141 -0.442263 0.0279435 -1.066939 1.129590
Nb 4d -0.279637 -0.05 -0.135649 -8.440821 1.706832
5s -0.0514108 0.0 0.255596 -11.384021 1.666463
5p 0.556542 -0.356295 -0.0002341 -0.103760 1.132172
Mo 4d -0.225737 -0.3 0.0620172 -7.995133 1.777658
5s -0.00583137 0.0 0.300841 -7.336245 1.639917
5p 0.291996 -0.430137 -0.104035 -3.686225 1.159781
Tc 4d -0.273426 -0.6 -0.0066526 -9.587897 1.918066
5s 0.36096 0.0 -0.0586205 -6.792444 1.918167
5p 0.250957 0.395682 -0.0087319 -3.325525 1.346082
Ru 4d -0.275832 -0.65 -0.0263914 -10.285405 2.102697
5s 0.101063 0.0 0.447116 -5.332608 1.749643
5p 0.340287 -0.305231 -0.0034723 -3.307153 1.348322
Rh 4d -0.196561 -0.65 0.0104368 -11.756644 2.458187
5s 0.154133 0.0 0.0066741 -7.850495 1.811796
5p 0.310707 -0.188177 -0.0213308 -3.007906 1.398452
Pd 4d -0.271731 -0.6 0.0060285 -11.963518 2.353691
5s 0.0620014 0.0 0.026682 -9.714059 1.828354
5p 0.453413 0.0931707 0.0503075 -2.035281 1.333352
Ag 4d -0.164907 -0.03 -0.0062719 -9.591083 2.843549
S-93
5s 0.0109149 0.0 -0.0065794 -8.083960 1.798462
5p 0.115614 0.802485 0.167717 -2.934333 1.266649
Cd 5s -0.0607687 0.0 0.141815 -7.252341 1.846689
5p 0.376719 0.238867 -0.0309814 -0.744865 1.141823
In 5s -0.219385 0.0 -0.0098312 -13.040909 1.963283
5p -0.0194965 -0.586746 0.0994688 -4.507143 1.685138
5d 0.313545 -0.28 0.0168649 -0.805666 1.050000
Sn 5s -0.175182 0.0 -0.0454629 -19.970428 2.551510
5p -0.0780287 -0.509075 -0.0320651 -7.367059 1.893784
5d 0.126111 -0.06 -0.145941 -2.077548 1.100000
Sb 5s -0.175435 0.0 -0.0147626 -18.371244 2.307407
5p -0.124946 -0.62785 -0.0091175 -7.350148 2.179752
5d 0.308727 -0.55 0.160287 0.909033 1.256087
Te 5s -0.248939 0.0 0.0115389 -21.930653 2.434144
5p -0.11232 -0.155533 -0.0082051 -9.480374 2.182459
5d 0.318432 0.06 0.301323 0.978922 1.373076
I 5s -0.269575 0.0 -0.050615 -20.949407 2.159500
5p -0.141833 -0.0338735 0.0084766 -12.180159 2.308379
5d 0.282119 0.3 0.307713 -0.266596 1.691185
Xe 5s -0.310965 0.0 -0.0020195 -19.090498 2.715140
5p -0.161979 -0.230267 0.0017246 -11.249471 2.312510
5d 0.19049 -0.23 0.0327039 -0.497097 1.855707
Cs 6s -0.00713637 0.0 -0.13126 -4.041706 1.225688
6p 0.20637 0.249431 -0.01 -1.394193 0.823818
Ba 6s -0.140366 0.0 0.0352001 -5.900000 1.528102
6p 0.187741 2.22475 -0.0926576 -2.133395 0.991572
5d 0.113897 -0.23 0.0147995 -1.514900 1.500000
La 5d -0.378201 -0.3 -0.0777542 -8.958783 2.875048
6s -0.0673201 0.0 0.107168 -11.877410 1.731390
6p 0.541364 -0.469967 -0.0239967 -0.601717 1.303590
Ce 5d -0.419892 -0.3 -0.0638958 -7.381991 2.870000
6s -0.0610774 0.0 0.133515 -8.537781 1.725197
6p 0.376634 -0.553966 -0.019832 -3.017508 1.309804
S-94
Pr 5d -0.412865 -0.276923 -0.0543909 -7.280875 2.872308
6s -0.0604017 0.0 0.134944 -8.504806 1.729767
6p 0.381948 -0.546278 -0.0198184 -2.873159 1.315495
Nd 5d -0.405838 -0.253846 -0.0448861 -7.179760 2.874615
6s -0.0597259 0.0 0.136373 -8.471830 1.734337
6p 0.387261 -0.538591 -0.0198048 -2.728809 1.321186
Pm 5d -0.398811 -0.230769 -0.0353812 -7.078644 2.876923
6s -0.0590501 0.0 0.137803 -8.438855 1.738907
6p 0.392574 -0.530903 -0.0197912 -2.584460 1.326877
Sm 5d -0.391784 -0.207692 -0.0258764 -6.977529 2.879231
6s -0.0583743 0.0 0.139232 -8.405879 1.743478
6p 0.397888 -0.523216 -0.0197776 -2.440110 1.332567
Eu 5d -0.384758 -0.184615 -0.0163715 -6.876413 2.881538
6s -0.0576986 0.0 0.140661 -8.372904 1.748048
6p 0.403201 -0.515528 -0.019764 -2.295761 1.338258
Gd 5d -0.377731 -0.161538 -0.0068667 -6.775298 2.883846
6s -0.0570228 0.0 0.142091 -8.339929 1.752618
6p 0.408514 -0.507841 -0.0197504 -2.151411 1.343949
Tb 5d -0.370704 -0.138461 0.0026382 -6.674182 2.886154
6s -0.056347 0.0 0.14352 -8.306953 1.757188
6p 0.413827 -0.500153 -0.0197369 -2.007062 1.349640
Dy 5d -0.363677 -0.115384 0.012143 -6.573067 2.888462
6s -0.0556712 0.0 0.144949 -8.273978 1.761758
6p 0.419141 -0.492466 -0.0197233 -1.862712 1.355331
Ho 5d -0.35665 -0.0923072 0.0216479 -6.471951 2.890769
6s -0.0549955 0.0 0.146379 -8.241003 1.766328
6p 0.424454 -0.484778 -0.0197097 -1.718363 1.361022
Er 5d -0.349623 -0.0692302 0.0311527 -6.370836 2.893077
6s -0.0543197 0.0 0.147808 -8.208027 1.770899
6p 0.429767 -0.477091 -0.0196961 -1.574013 1.366713
Tm 5d -0.342596 -0.0461533 0.0406576 -6.269720 2.895385
6s -0.0536439 0.0 0.149237 -8.175052 1.775469
6p 0.435081 -0.469403 -0.0196825 -1.429664 1.372403
S-95
Yb 5d -0.335569 -0.0230763 0.0501624 -6.168604 2.897692
6s -0.0529681 0.0 0.150667 -8.142076 1.780039
6p 0.440394 -0.461716 -0.0196689 -1.285314 1.378094
Lu 5d -0.328542 0.0000007 0.0596673 -6.067489 2.900000
6s -0.0522924 0.0 0.152096 -8.109101 1.784609
6p 0.445707 -0.454028 -0.0196553 -1.140965 1.383785
Hf 5d -0.340957 0.1 0.017655 -7.181755 2.638729
6s -0.0273193 0.0 0.22715 -10.626891 2.194333
6p 0.33515 -0.448616 -0.0069771 -1.603430 1.427467
Ta 5d -0.303963 0.05 -0.0620136 -8.481353 2.018969
6s -0.157077 0.0 0.0988501 -13.073088 1.996498
6p 0.60186 -0.339438 -0.047254 0.655254 1.407714
W 5d -0.256771 0.37 -0.0192494 -9.501505 2.155885
6s 0.0620898 0.0 0.254364 -11.093016 1.892022
6p 0.492738 -0.34192 0.0236479 -1.420389 1.458186
Re 5d -0.317231 -0.6 -0.0322139 -11.189119 2.262783
6s 0.138901 0.0 0.111757 -12.685198 2.187549
6p 0.339733 0.658686 -0.133516 -3.851981 1.636996
Os 5d -0.284611 -0.65 -0.0095346 -10.382841 2.509631
6s 0.213168 0.0 0.0346183 -8.731460 2.173991
6p 0.280972 0.135022 -0.0208758 -3.546379 1.597888
Ir 5d -0.246934 -0.65 0.0051977 -11.018475 2.756134
6s 0.207338 0.0 -0.0123672 -9.349164 2.117548
6p 0.183032 -0.0977957 -0.0079864 -3.603762 1.680343
Pt 5d -0.272439 -0.6 -0.0204828 -12.047728 2.704492
6s 0.0673756 0.0 0.113953 -10.482306 2.329136
6p 0.192595 -0.0203212 0.140803 -3.778297 1.623286
Au 5d -0.0641082 -0.6 -0.0154462 -9.578599 3.241287
6s 0.0469154 0.0 0.147934 -7.688552 2.183171
6p 0.252503 0.0614126 0.104807 0.883399 2.084484
Hg 6s -0.0983345 0.0 -0.0352252 -11.538066 2.244504
6p 0.156289 -0.537512 0.0205401 -2.532581 1.470848
Tl 6s -0.229422 0.0 -0.0255975 -17.319333 2.294231
S-96
6p 0.131098 -0.71334 0.0901364 -4.460584 1.731592
Pb 6s -0.229551 0.0 -0.389346 -24.055207 2.960592
6p -0.0880527 0.783825 0.343712 -5.893816 1.953130
Bi 6s -0.217501 0.0 0.0160425 -19.843840 2.788267
6p -0.107739 -0.6 0.0248659 -7.297456 2.277039
Po 6s -0.209233 0.0 -0.0046813 -20.205380 3.314810
6p -0.184264 -0.810916 -0.0100437 -8.476927 2.389456
At 6s -0.3055 0.0 -0.0287369 -17.050229 2.220421
6p -0.171085 -0.253207 -0.0007993 -9.499822 2.408112
5d 0.23825 0.25 0.280581 -0.096063 1.500000
Rn 6s -0.352454 0.0 -0.0001712 -21.000000 3.109394
6p -0.119897 -0.0302388 -0.000528 -10.496406 2.541934
5d 0.21167 -0.23 -0.320602 -1.415056 1.790000
An STO-3G expansion is used for 1s levels (H and He) and for all d levels .
An STO-4G expansion is used for the ns and np levels with n = [2, 5].
An STO-6G expansion is used for the 6s and 6p levels.
(a)
It is noted that for s-functions, κlA is not a fitted parameter but always set to zero.
References
(S1) Elstner, M.; Porezag, D.; Jungnickel, G.; Elsner, J.; Haugk, M.; Frauenheim, T.;
Suhai, S.; Seifert, G. Self-consistent-charge density-functional tight-binding method
for simulations of complex materials properties. Phys. Rev. B 1998, 58, 7260–7268.
(S2) Gaus, M.; Cui, Q.; Elstner, M. DFTB3: Extension of the Self-Consistent-Charge
Density-Functional Tight-Binding Method (SCC-DFTB). J. Chem. Theory Comput.
2011, 7, 931–948.
(S3) Grimme, S.; Bannwarth, C.; Shushkov, P. A Robust and Accurate Tight-Binding
Quantum Chemical Method for Structures, Vibrational Frequencies, and Noncovalent
S-97
Interactions of Large Molecular Systems Parametrized for All spd-Block Elements
(Z = 1 − 86). J. Chem. Theory Comput. 2017, 13, 1989–2009.
(S4) Risthaus, T.; Steinmetz, M.; Grimme, S. Implementation of Nuclear Gradients of

Range-Separated Hybrid Density Functionals and Benchmarking on Rotational Con-
stants for Organic Molecules. J. Comput. Chem. 2014, 35, 1509–1516.
(S5) Grimme, S.; Brandenburg, J. G.; Bannwarth, C.; Hansen, A. Consistent structures
and interactions by density functional theory with small atomic orbital basis sets. J.
Chem. Phys. 2015, 143, 054107.
(S6) Bühl, M.; Kabrede, H. Geometries of Transition-Metal Complexes from Density-

Functional Theory. J. Chem. Theory Comput. 2006, 2, 1282–1290.
(S7) Jurečka, P.; Šponer, J.; Cerny, J.; Hobza, P. Benchmark database of accurate (MP2
and CCSD(T) complete basis set limit) interaction energies of small model complexes,
DNA base pairs, and amino acid pairs. Phys. Chem. Chem. Phys. 2006, 8, 1985–1993.
(S8) Řezáč, J.; Riley, K. E.; Hobza, P. S66: A Well-balanced Database of Benchmark
Interaction Energies Relevant to Biomolecular Structures. J. Chem. Theory Comput.
2011, 7, 2427.
(S9) Řezáč, J.; Riley, K. E.; Hobza, P. Benchmark Calculations of Noncovalent Interactions
of Halogenated Molecules. J. Chem. Theory Comput. 2012, 8, 4285–4292.
(S10) Miriyala, V. M.; Řezáč, J. Testing Semiempirical Quantum Mechanical Methods on a

Data Set of Interaction Energies Mapping Repulsive Contacts in Organic Molecules.
J. Phys. Chem. A 2018, 122, 2801–2808.
(S11) Goerigk, L.; Hansen, A.; Bauer, C.; Ehrlich, S.; Najibi, A.; Grimme, S. A look at the
density functional theory zoo with the advanced GMTKN55 database for general main
S-98
group thermochemistry, kinetics and noncovalent interactions. Phys. Chem. Chem.
Phys. 2017, 19, 32184–32215.
(S12) Kozuch, S.; Martin, J. M. L. Halogen Bonds: Benchmarks and Theoretical Analysis.
(S13) Setiawan, D.; Kraka, E.; Cremer, D. Strength of the Pnicogen Bond in Complexes
Involving Group Va Elements N, P, and As. J. Phys. Chem. A 2015, 119, 1642–1656.
(S14) Marshall, M. S.; Burns, L. A.; Sherrill, C. D. Basis set convergence of the coupled-
CCSD(T)
cluster correction, δMP2 : Best practices for benchmarking non-covalent interactions
and the attendant revision of the S22, NBC10, HBC6, and HSG databases. J. Chem.
Phys. 2011, 135, 194102.
(S15) Grimme, S.; Hansen, A.; Brandenburg, J. G.; Bannwarth, C. Dispersion-Corrected

Mean-Field Electronic Structure Methods. Chem. Rev. 2016, 116, 5105–5154.
(S16) Bryantsev, V. S.; Diallo, M. S.; van Duin, A. C. T.; Goddard, W. A. Evaluation
of B3LYP, X3LYP, and M06-Class Density Functionals for Predicting the Binding
Energies of Neutral, Protonated, and Deprotonated Water Clusters. J. Chem. Theory
Comput. 2009, 5, 1016–1026.
(S17) Anacker, T.; Friedrich, J. New accurate benchmark energies for large water clusters:
DFT is better than expected. J. Comput. Chem. 2014, 35, 634–643.
(S18) Lao, K. U.; Schäffer, R.; Jansen, G.; Herbert, J. M. Accurate description of inter-
molecular interactions involving ions using symmetry-adapted perturbation theory. J.
(S19) Grimme, S.; Antony, J.; Ehrlich, S.; Krieg, H. A consistent and accurate ab initio
parametrization of density functional dispersion correction (DFT-D) for the 94 ele-
ments H-Pu. J. Chem. Phys. 2010, 132, 154104.
S-99
(S20) Sure, R.; Grimme, S. Comprehensive Benchmark of Association (Free) Energies of
Realistic Host–Guest Complexes. J. Chem. Theory Comput. 2015, 11, 3785–3801.
(S21) Gruzman, D.; Karton, A.; Martin, J. M. L. Performance of Ab Initio and Density
Functional Methods for Conformational Equilibria of Cn H2n+2 Alkane Isomers (n =
4 − 8). J. Phys. Chem. A 2009, 113, 11974–11983.
(S22) Kesharwani, M. K.; Karton, A.; Martin, J. M. L. Benchmark ab Initio Conformational

Energies for the Proteinogenic Amino Acids through Explicitly Correlated Methods.
Assessment of Density Functional Methods. J. Chem. Theory Comput. 2016, 12, 444–
454.
(S23) Kozuch, S.; Bachrach, S. M.; Martin, J. M. Conformational Equilibria in Butane-1,4-

diol: A Benchmark of a Prototypical System with Strong Intramolecular H-bonds. J.
Phys. Chem. A 2014, 118, 293–303.
(S24) Kozuch, S.; Martin, J. M. L. Spin-component-scaled double hybrids: An extensive

search for the best fifth-rung functionals blending DFT and perturbation theory. J.
Comput. Chem. 2013, 34, 2327–2344.
(S25) Řeha, D.; Valdés, H.; Vondrášek, J.; Hobza, P.; Abu-Riziq, A.; Crews, B.;
de Vries, M. S. Structure and IR Spectrum of Phenylalanyl–Glycyl–Glycine Tripetide
in the Gas-Phase: IR/UV Experiments, Ab Initio Quantum Chemical Calculations,
and Molecular Dynamic Simulations. Chem. Eur. J. 2005, 11, 6803–6817.
(S26) Goerigk, L.; Karton, A.; Martin, J. M. L.; Radom, L. Accurate quantum chemical
energies for tetrapeptide conformations: why MP2 data with an insufficient basis set
should be handled with caution. Phys. Chem. Chem. Phys. 2013, 15, 7028–7031.
(S27) Csonka, G. I.; French, A. D.; Johnson, G. P.; Stortz, C. A. Evaluation of Density
Functionals and Basis Sets for Carbohydrates. J. Chem. Theory Comput. 2009, 5,
679–692.
S-100
(S28) Kruse, H.; Mladek, A.; Gkionis, K.; Hansen, A.; Grimme, S.; Sponer, J. Quantum
Chemical Benchmark Study on 46 RNA Backbone Families Using a Dinucleotide Unit.
(S29) Marianski, M.; Supady, A.; Ingram, T.; Schneider, M.; Baldauf, C. Assessing the
Accuracy of Across-the-Scale Methods for Predicting Carbohydrate Conformational
Energies for the Examples of Glucose and α-Maltose. J. Chem. Theory Comput. 2016,
12, 6157–6168.
(S30) Dohm, S.; Hansen, A.; Steinmetz, M.; Grimme, S.; Checinski, M. P. Comprehensive
Thermochemical Benchmark Set of Realistic Closed-Shell Metal Organic Reactions.
(S31) Karton, A.; Goerigk, L. Accurate reaction barrier heights of pericyclic reactions: Sur-
prisingly large deviations for the CBS-QB3 composite method and their consequences
in DFT benchmark studies. J. Comput. Chem. 36, 622–632.
(S32) Goerigk, L.; Sharma, R. The INV24 test set: how well do quantum-chemical methods
describe inversion and racemization barriers? Can. J. Chem. 2016, 94, 1133–1143.
(S33) Karton, A.; O’Reilly, R. J.; Chan, B.; Radom, L. Determination of Barrier Heights
for Proton Exchange in Small Water, Ammonia, and Hydrogen Fluoride Clusters
with G4(MP2)-Type, MPn, and SCS-MPn Procedures–A Caveat. J. Chem. Theory
Comput. 2012, 8, 3128–3136.
(S34) Karton, A.; O’Reilly, R. J.; Radom, L. Assessment of Theoretical Procedures for
Calculating Barrier Heights for a Diverse Set of Water-Catalyzed Proton-Transfer
Reactions. J. Phys. Chem. A 2012, 116, 4211–4221.
(S35) Hait, D.; Head-Gordon, M. How Accurate Is Density Functional Theory at Predicting
Dipole Moments? An Assessment Using a New Database of 200 Benchmark Values.
S-101
gfn2xtb.pdf (4.17 MiB) view on ChemRxiv download file

GFN2-XTB - An Accurate and Broadly Parametrized Self-Consistent Tight-Binding Quantum Chemical Method With Multipole Ele v1

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

GFN2-XTB - An Accurate and Broadly Parametrized Self-Consistent Tight-Binding Quantum Chemical Method With Multipole Ele v1

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

GFN2-XTB - An Accurate and Broadly Parametrized Self-Consistent Tight-Binding Quantum Chemical Method With Multipole Ele v1

Uploaded by

Copyright:

Available Formats

doi.org/10.26434/chemrxiv.7246238.

GFN2-xTB - an Accurate and Broadly Parametrized Self-Consistent

Submitted date: 24/10/2018 • Posted date: 24/10/2018

File list (1)

gfn2xtb.pdf (4.17 MiB) view on ChemRxiv download file

Christoph Bannwarth,∗,†,‡ Sebastian Ehlert,† and Stefan Grimme∗,†

E-mail: christoph.bannwarth@stanford.edu; grimme@thch.uni-bonn.de

Keywords: density functional tight-binding, semiempirical molecular orbital theory,

multipole electrostatics, non-covalent interactions

An extended semiempirical tight-binding model is presented, which is primarily

fects via short-range damped interactions of cumulative atomic multipole moments.

element-specific parameters (available up to radon, Z = 86). Moreover, the atomic

partial charge dependent D4 London dispersion model is incorporated self-consistently,

well-suited to explore the conformational space of molecular systems. Significant im-

for biomolecular systems in aqueous solution.

and Hartree-Fock methods to allow such calculations.

has been parametrized for a number of chemical elements. 58–62

intermolecular FFs. 69 Of particular significance is its contribution to a newly developed protocol

sampling with GFN-xTB is employed to generate the conformer-rotamer-ensemble, which in turn

Gaussian functions, which approximate Slater functions (STO-mG). 70 Polarization functions

only assigned a single 1s function.

pansion. Furthermore the simultaneously developed density-dependent D4 dispersion model

in an self-consistent formulation is an inherent part of the method. Different from either

specific parameter strategy. No element pair-specific parameters are employed.

with GFN-xTB outperforms any other semiempirical method of comparable complexity. By

well-suited for applications in high (electronic) temperature molecular dynamics simulations

xTB, 34 PM6-D3H4X, 35,38,39 and DFTB3-D3(BJ). 56,58,72

Without explicitly referring to a specific functional approximation, we will assume as starting

are used throughout)

non-local correlation kernel ΦNL 0

expanded in terms of density fluctuations around a superposition of (neutral) atomic reference

2.1 Contributions relevant to the GFN2-xTB energy

2.1.1 Zeroth order terms

spherically symmetric atoms, the summed Coulomb interactions (e − e, n − n, e − n) between

tight-binding electronic energy, which depends only on the fluctuations δρ.

2.1.2 First order terms

within the self-consistent D4 dispersion model (see Section 2.2.6). 78,79

2.1.3 Second order terms

At second order, interatomic electrostatic and one-center exchange-correlation terms occur. In

DFTB schemes, these are generally condensed in a Mataga-Nishimoto-Ohno-Klopman damped

2.1.4 Third order terms

shell charges (Mulliken approximation) in GFN2-xTB.

EGFN2-xTB = Erep + Edisp + EEHT + EIES+IXC + EAES + EAXC + GFermi . (8)

proved by the AES energy EAES , which is described in Section 2.2.5.

2.2.1 The classical repulsion energy

Y eff values correlate less with the atomic number in GFN2-xTB.

2.2.2 Choice of electronic wavefunction and finite temperature treatment

As in the predecessor GFN-xTB, we are working with a formally spin-restricted wavefunction

necessary to provide a variational solution for fractional occupations and is given by

ψiσ . These are given by

ni = niα + niβ . (12)

enabling fractional orbital occupations in static correlation cases.

centered orbitals (LCAO) 85,86 ,

ψi = cκi φκ (ζκ , STO-mG) . (13)

element basis functions

Sham density functional theory

 is a diagonal matrix containing the orbital energies.

2.2.3 The extended Hückel-type energy

is a diagonal matrix containing the orbital energies.