Advanced Review

Atomic orbital basis sets

Frank Jensen
Electronic structure methods for molecular systems rely heavily on using basis
sets composed of Gaussian functions for representing the molecular orbitals. A
number of hierarchical basis sets have been proposed over the last two decades,
and they have enabled systematic approaches to assessing and controlling the
errors due to incomplete basis sets. We outline some of the principles for constructing basis sets, and compare the compositions of eight families of basis sets
that are available in several different qualities and for a reasonable number of
elements in the periodic table. C 2012 John Wiley & Sons, Ltd.
How to cite this article:

WIREs Comput Mol Sci 2012. doi: 10.1002/wcms.1123

INTRODUCTION

he use of basis sets for expanding the molecular orbitals in wave function or KohnSham
density functional methods is an essential component of contemporary methods for describing the
electronic structure of molecular and extended systems. Nuclear-centered Slater-1 or Gaussian-type2
functions have dominated for molecular systems, with
Gaussian functions being preferred due to the better
computational efficiency. Plane-wave basis functions
are often used for extended systems, as they are naturally suited to periodic boundary conditions; but for
any reasonable number of plane waves, this necessitates the use of a pseudo-potential for representing the
atomic core electrons/potential. Nuclear-centered basis functions can also be used for periodic systems, and
this treats all electrons on an equal footing. Recent developments have investigated the use of finite-element
methods where piecewise polynomials are used for
representing the orbitals.35
The goal of a basis set is to provide the best
representation of the unknown molecular orbitals (or
electron density), with as small a computational cost
as possible. Because different theoretical methods and
molecular properties have different basis set demands,
different computer architectures and algorithms have
different efficiency requirements, and the desired accuracy varies with the application, it is not possible
to design one optimum basis set. Indeed, the large

Correspondence to: [email protected]

Department of Chemistry, Aarhus University, Aarhus, Denmark

DOI: 10.1002/wcms.1123

Volume 00, January/February 2012

number of different basis sets proposed over the years

is a testament to these conflicting demands.
In this review, we will summarize some of the
principles for constructing and classifying basis sets,
with focus on modern hierarchical basis sets. Only basis sets composed of nuclear-centered Gaussian-type
functions will be discussed, but many of the principles
and conclusions hold for Slater-type functions as well.
We will concentrate on basis sets for the first 36 atoms
in the periodic table [hydrogen (H) to krypton (Kr)]
as many of the popular basis sets are only available
for these elements, but the trends and principles carry
over to elements in the remaining part of the periodic
table as well.
All-electron calculations for systems containing
atoms from the lower part of the periodic table must
include relativistic effects for accurate results, and
this leads to some differences in the basis set requirements compared to nonrelativistic methods. We will
not discuss basis sets for relativistic methods in this
review. A significant computational saving can be obtained for systems with many-electron atoms by replacing the core electrons by a pseudo-potential or
model potential, and this can also to some extent
account for relativistic effects. The use of pseudopotentials has been the subject of recent reviews by
Dolg and Cao, and these can be consulted for further
information.6,7 A discussion of plane wave or finiteelement methods, and the topic of auxiliary basis sets
for density fitting, is beyond the scope of the present
review.
We will focus on eight families of basis sets that
are available in several quality levels and defined for a
reasonable number of elements in the periodic table:

c 2012 John Wiley & Sons, Ltd.



Advanced Review

wires.wiley.com/wcms

1. The Pople-style k-lmnG basis sets.820

2. The Ahlrichs SVP, TZP, QZP basis sets in
their Def2 version.21
3. The XZP basis sets developed by Jorge and
coworkers.2226
4. The Sapporo basis sets developed by Koga
and coorkers.2730
5. The ANO basis sets developed by Roos and
coworkers.3135
6. The cc-pVXZ basis sets developed by
Dunning36 and Peterson and coworkers.3739
7. The nZaP basis sets developed by Petersson
and coworkers.4042
8. The pc-n basis sets developed by Jensen and
coworkers.4348

PRIMITIVE AND CONTRACTED BASIS

FUNCTIONS
An unknown one-electron function, such as a molecular orbital , can be expanded in a set of known
functions , the basis set49,50 :
=

M


ci i .

(1)

i=1

The coefficients ci are in HartreeFock (HF)

and KohnSham density function theory (DFT) determined by minimizing the total energy, which by traditional methods lead to a matrix eigenvalue problem
that is solved iteratively to provide a self-consistent
field (SCF) solution. The matrix elements contain
multidimensional integrals over basis functions, of
which those involving the electronelectron interaction (two-electron integrals) completely dominate the
computational effort. We will, in the present context,
not be concerned with determination of the molecular
expansion coefficients, but focus on the basis functions, which will be taken as Cartesian Gaussian-type
functions:
2

i = N (x X)k (y Y)l (z Z)m ei (r R) .

(2)

The center of such a primitive function is R

(X,Y,Z), typically a nuclear position, the sum of k,
l, and m defines the angular momentum (e.g., k + l +
m = 1 is a p-function), i is the exponent providing
the radial extent of the function, and N is a normalization constant. The accuracy of the expansion in
Eq. (1) is determined by the number of functions M,
their distribution in terms of angular momentum, and
the values of the exponents i [Eq. (2)].

The primitive basis set is for computational reasons usually contracted, by forming K fixed linear
combinations j from M primitive functions i (K <
M):
j =

M


di j i ,

i=1

K


(3)

cjj.

j=1

The physical reason for contraction is that the

core orbitals are insensitive to the molecular environment, and can therefore be described by a fixed linear
combination with little loss of accuracy. Contraction
reduces the number of coefficients c j from M to K for
each molecular orbital, and thus simplifies the variational problem. The number of two-electron integrals
that need to be calculated is unaffected by contraction, although the efficiency of the computational implementation may depend on the exact nature of the
contraction.
The typical notation used to indicate the reduction in the number of variational parameters is with
parenthesis for the primitive set and square brackets for the contracted set, e.g., (9s4p) [3s2p] denotes nine primitive s-functions contracted to three sfunctions, and four primitive p-functions contracted
to two p-functions. This does not specify how the contraction is done, only the number of resulting functions. The explicit contraction scheme can be given
in curly brackets as {621/31} which indicates that
the first, second, and third contracted s-functions are
constructed of six, two, and one primitive functions,
respectively, whereas the two contracted p-functions
contain three and one primitive functions.
Two types of contraction can be used, known as
segmented51 and general52 contraction. In the purest
versions, a segmented contraction forms a disjoint
set of the primitive functions (i.e., a given primitive
function i has only a nonzero contraction coefficient
di j in one of the contracted functions j ), whereas a
general contraction uses all the primitive functions in
all the contracted functions (i.e., all di j = 0).
Contraction is in practice rarely completely segmented or general. A general contraction can be
made partly segmented by a simple mathematical
transformation, which is equivalent to the variable
elimination used in solving linear equations.53 This
transformation results in each of the K contracted
functions having only nonzero contributions from
M K + 1 primitive functions. This transformation is exact, i.e., the same results are obtained with
the untransformed and transformed basis sets. The

c 2012 John Wiley & Sons, Ltd.



Volume 00, January/February 2012

WIREs Computational Molecular Science

Atomic orbital basis sets

transformed basis set can be further segmented by neglecting small coefficients (e.g., less than 105 ), but
this represents a real, albeit small, reduction of the
accuracy by which the molecular orbitals can be represented.
A set of general contracted functions can for
atomic systems at the HF or DFT level generate the
same result as the full set of primitive functions, provided that the contraction coefficients are taken as the
SCF (HF or DFT) coefficients. For molecular systems
and for correlated methods, a general contraction will
always involve a penalty, with the magnitude depending on the level of contraction, i.e., the number of
contracted functions K.
A segmented contraction always reduces the
accuracy by which the orbitals can be represented,
even for atoms. The inner primitive functions (large
) make only a large contribution to the 1s-orbital,
and the outer primitive functions (small ) are usually
left uncontracted (i.e., a contracted function involving only one primitive function), because these describe the environmentally sensitive valence orbitals.
The primitive functions with intermediate exponents,
however, are important for describing both core and
valence orbitals (or orbitals from different shells in the
general case). Restricting them to participate in only
one contracted function can therefore give a significant reduction in accuracy of the final set of functions.
It is therefore a common practice to duplicate one or
more functions in the intermediate exponent range,
which can be viewed as a step toward a general contraction. In a general contraction, on the other hand,
the basis functions describing the outer valence part
of the orbitals are normally left uncontracted, i.e., a
partly segmented contraction. The terms segmented
and general contraction should thus be taken as limiting cases, with actual basis sets containing elements
of both contraction forms.
Segmented contraction has traditionally been
the preferred method, and most integral codes in electronic structure programs have been written with such
basis sets in mind. Although these programs can also
treat general contracted basis sets, they do this by
replicating the primitive functions as many times as
they enter the contracted functions, thereby significantly increasing the number of primitive basis functions (potentially from M to M K) and leading to
computational inefficiency. Some programs have been
written with general contracted basis sets in mind, and
they can handle segmented basis sets with no loss in
efficiency, because segmentation is just a special case
of general contraction.
An additional factor influencing the computational efficiency is the use of integral screen-

Volume 00, January/February 2012

ing to avoid calculating some of the two-electron

integrals.5456 This screening becomes efficient for
large systems and works most favorably with small
basis sets, and it is the premise for reducing the
HF/DFT scaling of the computational time with system size from formally N4 to N2 without employing
special techniques. The screening relies on calculating
an upper limit for each batch of two-electron integrals
(a batch is a group of functions sharing the same exponent), and if this upper limit is below a suitable
threshold, the entire batch of integrals can be neglected. The screening is performed at the contracted
basis set level, and therefore works most efficiently for
segmented basis sets. If an integral batch involving a
specific set of contracted functions can be neglected,
all of the integrals involving the underlying primitive
functions can be neglected. In a general contraction,
on the other hand, all primitives may contribute to all
contracted functions, and screening is therefore less
efficient.

BASIS SET CLASSIFICATION

Basis sets can be described by the number of primitive functions and the contraction of these functions.
For classification purposes, however, it is useful to focus on the number of contracted functions, since this
indicates the flexibility of the basis set to describe different molecular environments. Except for very large
basis sets, the errors in the absolute energy due to basis set deficiencies are large, and obtaining accurate
relative energies therefore relies on substantial error
cancellation. This means that basis set balance is of
utmost importance, both internally in terms of number of functions with different angular momentum,
and also externally between different elements.57,58
A minimum or single zeta (SZ) basis set has the
number of contracted functions equal to the number of occupied orbitals for the isolated atom, with
the addendum that all components of a given function type are included. For hydrogen, carbon, and
sulfur, this means basis sets of 1s, 2s1p, and 3s2p,
respectively. This can provide the basis set limiting
energy for isolated atoms at HF and DFT levels, if a
large number of primitive functions are used in the
contraction. SZ basis sets, however, provide poor results for molecules, since they are inadequate at describing the deviation from the spherical symmetry
of the atoms and the process of bond formation. For
methods including electron correlation, where the virtual (unoccupied) orbitals are used for expanding the
many-electron wave function, SZ basis sets are likewise unsuitable.

c 2012 John Wiley & Sons, Ltd.



Advanced Review

wires.wiley.com/wcms

The next step up in basis set quality is to use two

independent functions for each atomic orbital. The
function(s) describing the core orbitals are in practice
rarely duplicated, thus such double zeta (DZ) basis
sets are normally only double zeta in the valence region, i.e., 2s, 3s2p, and 4s3p for hydrogen, carbon,
and sulfur, respectively. DZ basis sets are much better at describing changes due to molecular bond formation, as functions with different exponents can be
used in different ratios to describe that bonding is different in different directions. At this point, a higher
angular momentum function should also be added
to describe the charge polarization for HF and DFT
methods, and the electronelectron cusp at correlated
levels. Such basis sets are usually called double zeta
plus polarization (DZP).
The idea of increasing the number of s- and pfunctions, and at the same time adding higher angular
momentum functions, can be continued to produce
triple zeta plus polarization (TZP), quadruple zeta
plus polarization (QZP), etc. types of basis sets. In
early work, the TZ and QZ notations directly reflected the number of s- and p-functions, whereas
higher-order polarization functions were added in an
ad hoc manner. The seminal work of Dunning,36
based on the concept of atomic natural orbitals,59
strongly suggested that polarization functions should
be added in well-defined sequences, such that the
total basis set error is approximately equally distributed between all types of basis functions. This
introduced the important concept that the basis set
composition in terms of number of functions of different angular momentum can be determined by analyzing energy contributions, rather than relying on
subjective criteria, such as performance for a selection of molecular systems and properties. The XZP
nomenclature in modern classifications therefore normally reflects the highest angular momentum function included in the basis set, and this will also be
the usage in the present review. A DZP basis set
thus indicates that it includes a function with one
higher angular momentum than required by the occupied (valence) atomic orbitals, a TZP basis set includes a function with two higher angular momenta
than required by the occupied atomic orbitals, and
so on.60
The XZP classification scheme serves as a guideline for the quality of a basis set, and in favorable
cases provides a systematic convergence toward the
basis set limit. Tables 14 show the composition for
first-, second-, and third-row s-, p-, and d-block elements in terms of primitive and contracted functions
for the eight families of basis sets mentioned in the
introduction.

The XZP classification is unambiguous for pblock elements from the first two rows in the periodic
table, and the contracted basis set compositions are
very similar for the eight families (Tables 1 and 2). A
choice regarding polarization functions must be made
for s-block elements. As they only have occupied (valence) s-orbitals, p-functions are formally the first type
of polarization. The low-lying unoccupied p-orbitals,
however, often play an important role in the bonding
of s-block elements, and d-functions may thus also
be considered as the first type of polarization function. Different families of basis sets make different
choices, and in some cases also different choices between first-, second-, and third-row s-block elements
(Tables 13). The situation is similarly ambiguous for
the third-row p-elements, where the atoms have occupied 3d-orbitals. If the 3d-orbitals are considered part
of the atomic core and not taking part in the chemical bonding, then a d-function is the first polarization function. If, on the other hand, the 3d-electrons
are considered part of the valence electrons, then an
f-function is the first polarization function. Different
families of basis sets again make different choices (Table 3). The molecular bonding in third-row transition
metal systems is dominated by the 3d-orbitals, and
almost all basis set families consider f-functions to be
the first polarization function (Table 4).
Although the XZP classification is useful for
judging the quality of the results, it should be noted
that details of how, e.g., a TZP basis set is constructed
will influence the accuracy obtained, and this is the
topic of the next sections.

PRIMITIVE BASIS SETS

Isolated atoms with Z 20 (Ca) can at HF and
DFT levels be described completely by s- and pfunctions only, atoms up to Z 56 (Ba) by s-, p-,
and d-functions only, whereas the remaining atoms
require also f-functions. For molecular systems, and
for methods involving electron correlation, angular
momentum functions higher than sp, spd, and spdf,
respectively, are required to describe polarization and
correlation effects. By far the most common procedure is to determine the basis functions describing the
occupied atomic orbitals by HF/DFT optimization at
the atomic level, and treating the question of polarization functions subsequently.

Optimized Primitive Basis Sets

The best possible uncontracted basis set in an energetic sense can be constructed by minimizing the

c 2012 John Wiley & Sons, Ltd.



Volume 00, January/February 2012

Volume 00, January/February 2012

Pople
Pople
Jensen
Pople
Ahlrichs
Jorge
Sapporo
Roos
Dunning
Petersson
Jensen
Pople
Ahlrichs
Jorge
Sapporo
Roos
Dunning
Petersson
Jensen
Ahlrichs
Jorge
Sapporo
Roos
Dunning
Petersson
Jensen
Jorge
Dunning
Petersson
Jensen

SZ
DZ

c 2012 John Wiley & Sons, Ltd.



5ZP

QZP

TZP

DZP

Label

Quality

STO-3G
3-21G
pc-0
6-31G
Def2-SVP
DZP
DZP
ANO-DZP
cc-pVDZ
2ZaP
pc-1
6-311G(2df)
Def2-TZVP
TZP
TZP
ANO-TZP
cc-pVTZ
3ZaP
pc-2
Def2-QZVP
QZP
QZP
ANO-QZP
cc-pVQZ
4ZaP
pc-3
5ZP
cc-pV5Z
5ZaP
pc-4

Name
3
3
1
4
4
5
4
9
4
5
3
5
4
6
4
9
5
8
4
7
7
6
9
6
6
8
8
8

14
14
14
19

6
6
5
10
7
9
8
14
9
10
7
11
11
10
9
14
11
13
10
15
11
9
14
12

2
4
4

2
1
2
3
4
2
2
1
2
3
4
4
3

4
1
1

1
3
3

1
2
1
3
2

3
1
1

Primitives

2
2

1
1

6
7
6

2
3
3
3
3
4
4
4
3
4
3
4
5
5
6
5
4
5
4
7
6
7
6
5

3
5
5

1
2
1
2
2
2
2
3
2
3
2
3
3
3
4
4
3
4
2
4
4
5
5
4

s-Block Elements

T A B L E 1 Basis Set Compositions for First-Row s- and p-Block Atoms

2
4
4

2
1
2
2
2
2
2
1
2
3
3
3
3

1
1
1

1
3
3

1
2
1
2
2

1
1
1

Contracted

2
2

1
1

h
6
6
5
10
7
9
9
14
9
11
7
11
11
10
10
14
10
14
10
15
11
12
14
12
17
14
14
14
20
18

s
3
3
3
4
4
5
4
9
4
8
4
5
6
6
5
9
5
11
6
8
7
7
9
6
14
9
8
8
16
11

1
1
1
2
4
1
1
1
2
2
2
3
4
2
2
2
3
3
4
4
3
3
4
4
4
4
6

1
1
1
2
3
1
1
1
2
2
3
3
2
2
2
3
3
3
3

Primitives

1
1
2
2
1
1
1
2
2
2
2

1
1
1
1

h
2
3
3
3
3
4
3
3
3
4
3
4
5
5
4
4
4
5
4
7
6
5
5
5
6
6
7
6
7
8

p-Block Elements

1
2
2
2
2
2
2
2
2
3
2
3
3
3
3
3
3
4
3
4
4
4
4
4
5
5
5
5
6
7

1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
3
3
3
3
3
3
4
4
4
4
6

1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
3
3
3
3

Contracted

1
1
1
1
1
1
1
2
2
2
2

1
1
1
1

WIREs Computational Molecular Science

Atomic orbital basis sets

Pople
Pople
Jensen
Pople
Ahlrichs
Jorge
Sapporo
Roos
Dunning
Petersson
Jensen
Pople
Ahlrichs
Jorge
Sapporo
Roos
Dunning
Petersson
Jensen
Ahlrichs
Jorge
Sapporo
Roos
Dunning
Petersson
Jensen
Jorge
Dunning
Petersson
Jensen

SZ
DZ

c 2012 John Wiley & Sons, Ltd.



5ZP

QZP

TZP

DZP

Label

Quality

STO-3G
3-21G
pc-0
6-31G
Def2-SVP
DZP
DZP
ANO-DZP
cc-pVDZ
2ZaP
pc-1
6-311G(2df)
Def2-TZVP
TZP
TZP
ANO-TZP
cc-pVTZ
3Zap
pc-2
Def2-QZVP
QZP
QZP
ANO-QZP
cc-pVQZ
4ZaP
pc-3
5ZP
cc-pV5Z
5ZaP
pc-4

Name
6
6
5
10
7
9
7
12
8
9
7
9
8
10
9
12
10
12
9
12
11
10
12
12
12
12
14
15

17
18
20
21

9
9
8
16
10
12
11
17
12
15
11
13
14
13
15
17
15
17
13
20
14
16
17
16

2
4
4

2
3
2
4
6
2
2
1
4
3
5
6
3

1
1
1
2
6
1
1

1
3
3

1
2
3
2
2

1
1

1
1

Primitives

2
2

1
1

1
1

6
10
7

3
4
4
4
4
7
5
5
4
6
4
6
5
8
7
6
5
6
5
9
9
8
7
6

s-Block Elements

T A B L E 2 Basis Set Compositions for Second-Row s- and p-Block Atoms

4
7
6

2
3
2
3
3
4
4
2
3
4
2
5
4
5
6
3
4
5
3
5
6
7
4
5

2
4
4

2
3
2
3
2
2
2
1
4
3
5
4
3

1
1
1
1
2
1
1

1
3
3

1
2
3
2
2

1
1

1
1

Contracted

2
2

1
1

1
1

h
9
9
8
16
10
12
12
17
12
14
11
12
14
13
13
17
15
17
13
20
14
16
17
16
20
17
18
20
23
21

s
6
6
6
10
7
9
9
12
8
10
8
9
9
10
9
12
9
13
10
14
11
11
12
11
16
13
12
12
19
16

1
1
1
3
5
2
2
1
2
3
2
4
5
3
3
2
4
3
5
5
4
4
4
4
5
5
6

1
1
1
2
4
1
1
1
2
2
2
4
2
2
2
3
3
3
3

Primitives

1
1
2
2
1
1
1
2
2
2
2

1
1
1
1

h
3
4
4
4
4
7
4
4
4
5
4
6
5
8
5
5
5
6
5
9
9
6
6
6
7
6
10
7
8
7

p-Block Elements

2
3
3
3
3
4
3
3
3
4
3
5
5
5
4
4
4
5
4
6
6
5
5
5
6
5
7
6
7
6

1
1
1
1
1
2
2
1
2
3
2
2
2
3
3
2
4
3
3
3
4
4
4
4
5
5
6

1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
3
3
3
3

Contracted

1
1
1
1
1
1
1
2
2
2
2

1
1
1
1

Advanced Review
wires.wiley.com/wcms

Volume 00, January/February 2012

Volume 00, January/February 2012

Pople
Pople
Jensen
Pople1
Pople1
Ahlrichs
Jorge
Sapporo
Roos
Dunning
Petersson
Jensen
Pople
Ahlrichs
Jorge
Sapporo
Roos
Dunning
Petersson
Jensen
Ahlrichs
Ahlrichs
Sapporo
Roos
Dunning
Jensen
Dunning
Jensen

SZ
DZ

c 2012 John Wiley & Sons, Ltd.



STO-3G
3-21G
pc-0
6-31G
6-31G
Def2-SVP
DZP
DZP
ANO-DZP
cc-pVDZ
2ZaP
pc-1
6-311G(2df)
Def2-TZVP
TZP
TZP
ANO-TZP
cc-pVTZ
3ZaP
pc-2
Def2-QZVP
Def2-QZVPP
QZP
ANO-QZP
cc-pVQZ
pc-3
cc-pV5Z
pc-4

Name
12
12
8
22
22
14
15
16
20
14
17
12
14
17
16
18
20
20
20
16
24
24
19
20
22
20
26
24

s
9
9
7
16
16
9
12
12
16
11
12
9
11
12
13
13
16
14
14
11
18
18
14
16
16
14
18
17

Two different definitions of the 6-31G basis set are available.12,13

5ZP

QZP

TZP

DZP

Label

Quality

4
1
4
1
2
6
5
1
4
5
4
4
4
6
6
2
5
6
6
5
6
7
6
8
8

1
3
3
4
2
1
3
2

1
1

1
1

Primitives

2
1

h
4
5
5
5
5
5
8
6
6
5
6
5
8
6
9
8
7
6
7
6
11
11
9
8
7
7
8
8

s-Block Elements

T A B L E 3 Basis Set Compositions for Third-Row s- and p-Block Atoms

3
4
3
4
4
3
6
5
3
4
5
3
7
5
6
6
4
5
6
4
6
6
7
5
6
5
7
6

2
1
2
1
1
2
2
1
2
3
3
4
3
2
3
2
3
4
4
4
4
4
4
5
5

1
3
2
2
2
1
3
2

1
1

1
1

Contracted

2
1

h
12
12
8
22
14
14
15
17
20
14
18
12
15
17
16
18
20
20
21
16
24
24
19
20
21
20
26
24

s
9
9
7
16
11
10
12
14
17
11
13
10
13
13
13
14
17
13
16
13
20
20
14
17
16
16
17
19

p
3
3
4
5
6
6
5
9
11
6
7
7
7
8
6
10
11
9
10
9
10
10
10
11
12
11
13
13

1
1
2
2
4
1
1
1
2
4
2
4
2
2
3
4

Primitives

1
1
2
2
1
1
2
2

1
1

h
4
5
5
5
6
5
8
5
5
5
6
5
8
6
9
6
6
6
7
6
11
11
7
7
7
7
8
8

p-Block Elements

3
4
4
4
4
4
6
4
4
4
5
4
7
5
6
5
5
5
6
5
7
7
6
6
6
6
7
7

1
1
1
3
2
3
2
2
2
2
2
2
4
4
3
3
3
3
3
3
4
4
4
4
4
4
5
5

1
1
2
1
2
1
1
1
2
4
2
3
2
2
3
4

Contracted

1
1
1
2
1
1
2
2

1
1

WIREs Computational Molecular Science

Atomic orbital basis sets

Advanced Review

wires.wiley.com/wcms

T A B L E 4 Basis Set Compositions for Third-Row d-Block Atoms

Primitives
Quality

Label

Name

SZ
DZ

Pople
Pople
Jensen
Pople
Ahlrichs
Ahlrichs
Jorge
Sapporo
Roos
Dunning
Petersson
Jensen
Pople
Ahlrichs
Ahlrichs
Jorge
Sapporo
Roos
Dunning
Petersson
Jensen
Ahlrichs
Ahlrichs
Sapporo
Roos
Dunning
Jensen
Dunning
Jensen

STO-3G
3-21G
pc-0
6-31G
Def2-SV(P)
Def2-SVP
DZP
DZP
ANO-DZP
cc-pVDZ1
2ZaP
pc-1
6-311G(2df)
Def2-TZVP
Def2-TZVPP
TZP
TZP
ANO-TZP
cc-pVTZ
3ZaP
pc-2
Def2-QZVP
Def2-QZVPP
QZP
ANO-QZP
cc-pVQZ
pc-3
cc-pV5Z
pc-4

12
12
8
22
14
14
15
16
21
20
17
12
14
17
17
16
18
21
20
20
16
24
24
18
21
22
20
28
24

9
9
7
16
9
9
12
13
15
16
12
9
9
11
12
13
13
15
16
14
11
18
18
13
15
18
14
20
17

3
3
4
4
5
5
5
9
10
8
7
6
5
7
7
6
10
10
8
10
8
10
10
11
10
11
10
12
12

DZP

TZP

QZP

5ZP
1

Contracted
g

1
1

4
5
5
5
5
5
8
6
5
6
6
5
9
6
6
9
8
6
7
7
6
11
11
9
7
8
7
9
8

3
4
3
4
3
3
6
4
4
5
5
3
5
4
5
6
6
5
6
5
4
6
6
7
6
7
5
8
6

1
2
1
2
2
2
2
3
2
3
2
2
3
4
4
3
4
3
4
3
3
5
5
5
4
5
4
6
5

1
1
1
2
6
2
1
1
2
1
2
2
3
6
2
2
2
3
4
3
6
3
3
4
5

1
1
1
2
4
1
1
1
1
2
2
4
2
2
3
3

1
1
1
2
2

1
1
1
1
1
1
1
1
2
1
2
2
2
2
2
2
2
3
4
3
3
3
3
4
5

1
1
1
1
1
1
1
1
1
2
2
2
2
2
3
3

1
1
1
2
2

1
1

The cc-pVDZ is defined as a DZP contraction of the same primitive set of functions as the cc-pVTZ.

energy as a function of the exponents i [Eq. (2)]

because the expansion coefficients ci are determined
variationally by the SCF procedure [Eq. (1)]. The exponent optimization can be performed either by a
pattern search involving only the energy itself,61 by
methods involving the analytical gradient of the energy with respect to the exponents,62,63 or by methods
using also the second derivative of the energy with respect to the exponents.64 Most of the early work on
basis set optimizations relied on nonderivative methods, whereas more recent work has used analytical
exponent gradients.
Whatever the optimization method used, it will
only determine a local minimum, i.e., it is possible
that optimizations starting from different initial exponents may lead to different final results. A specific
example has been reported by Mezey et al.,65 where
two different sets of s-exponents were determined

for a (6s3p) basis set for the LiF atoms. In one of

the two sets, the five inner functions primarily describe the 1s-orbital, with the single remaining function mainly describes the 2s-orbital, whereas the other
set of exponents corresponds to a 4/2 split for the
1s/2s-orbitals.
The problem with multiple minima is mainly
associated with small basis sets, and has a clear physical interpretation related to the orbital shell structure of the atom.46 The shell effect is clearly visible
in the ratio between optimized exponents in the basis set, as illustrated in Figure 1 for s-functions for
the neon atom. With five s-functions, the ratio between 2 and 1 ( 1 being the smallest exponent) is
significantly larger than the ratio between 3 and 2 .
This exponent gap reflects the nodal structure of the
2s-orbital, where a basis function with an exponent
4 has a maximum with a distance to the nucleus

c 2012 John Wiley & Sons, Ltd.



Volume 00, January/February 2012

WIREs Computational Molecular Science

Atomic orbital basis sets

of functions M. Schmidt and Ruedenberg67 furthermore noted that the and parameters had to obey
certain rules in order for the basis set to approach
completeness as M increases, and provided a suitable
parameterization. The even-tempered formula allows
construction of basis sets which converge toward the
basis set limit for atoms, and this has been used for
constructing universal basis sets.68,69 Jorge and de
Castro70 have used a similar concept for constructing
basis sets by a generator coordinate approach.
An even-tempered basis set has the same ratio between exponents in the whole range from
to M , but the results from fully optimized basis
sets suggest that the ratio should increase both for
the largest and smallest exponents, as illustrated in
Figure 1. The well-tempered basis sets achieve the
former by the exponent parameterization shown in
Eq. (5), where the , , and parameters are optimized for each atom71,72 :

  
i
i1
i =
1+
i = 1, 2, . . . , M. (5)
M

F I G U R E 1 | HartreeFock optimized s-exponent values for the

neon atom for basis sets of increasing size.

corresponding to the 2s-node, and is therefore energetically less important than basis functions with
exponents larger or smaller than 4. The shell gap
is present until 14 s-functions, beyond which the exponents cover the whole range rather uniformly, and
multiple minima are therefore mainly a problem for
small basis sets and disappear as the number of functions increase.

Parameterized Primitive Basis Sets

The exponents of a primitive basis set form a highly
coupled nonlinear set of variables. A complete optimization, even with analytical exponent gradients,
is a nontrivial task for large basis sets and often involves many hundreds of iterations. Furthermore, as
the basis set approaches completeness, the energy as a
function of the exponents becomes a very flat surface.
As an alternative to performing the optimization directly with the exponents as variables, the exponents can be generated by a parameterized formula,
and the parameters optimized instead. Already, very
early it was noted that the exponents of completely
optimized basis sets had some regular features. The
simplest parameterization takes the ratio between successive exponents to be constant, forming a so-called
even-tempered sequence66 :
i = i

i = 1, 2, . . . , M.

(4)

Even-tempered basis sets have only two parameters ( and ) to optimize, independent of the number

Volume 00, January/February 2012

The even-tempered parameterization can be

generalized by recognizing that the formula in Eq.
(4) can be written as a linear expansion by taking the
logarithm
ln(i ) = ln() + i ln().

(6)

Petersson et al.73 have suggested an exponent

parameterization in terms of Legendre polynomials
Pk:


kmax

2i 2
ln (i ) =
1 ; i = 1, 2, . . . , M
Ak Pk
M1
k=0
(7)
A second-order expansion (kmax = 1) is equivalent to the even-tempered formula (A0 = ln(),
A1 = ln()), and a fourth-order expansion is able to
reproduce the results from a fully optimized basis set
closely.
Parameterized formulas require more functions
for achieving a given accuracy than fully optimized
basis sets, and for even-tempered basis sets the penalty
increases with the basis set size and with the atomic
number. This is illustrated in Figures 2 and 3, where
the energy convergence toward the HF limit for the Ne
and Kr atoms are displayed as a function of the basis
set size. The additional number of functions required
for an even-tempered parameterization to reach the
same accuracy as a fully optimized basis set is 3 at the
milli-Hartree level and 8 at the micro-Hartree level for

c 2012 John Wiley & Sons, Ltd.



Advanced Review

wires.wiley.com/wcms

Fitted Basis Sets

Slater-type orbitals (STO) depend exponentially on
the nuclearelectron distance, rather than the Gaussian exponential square dependence:
STO = Ne (r R) .

F I G U R E 2 | HartreeFock energy contribution of basis functions

for the neon atom, using either a full optimization, a fourth-order
Legendre [Eq. (7)] or an even-tempered [Eq. (4)] parameterization for
determining the basis function exponents.

(8)

Because STOs can represent the exact solutions

for hydrogen-like atoms, they provide a better representation than Gaussian functions for many-electron
systems on a function-to-function comparison. Multicenter two-electron integrals, however, are cumbersome with STOs, but are much easier with Gaussian functions. The use of density fitting methods can
partly alleviate the problem of difficult multicenter integrals, but the overall computational effort appears
to be similar with the two types of basis sets.74,75
The use of STOs has been an efficient method
of achieving high accuracy for atoms, and such basis
sets were developed very early.76 For use in molecular
calculations, a common practice was to represent the
optimized atomic STOs by a fixed linear combination
of Gaussian functions.77 The optimum representation
was done by fitting the exponents and coefficients to
the STOs in a least squares sense, and the fitting error
decreases as the number of Gaussian functions is increased. Such basis sets have been developed by Pople
and coworkers7880 for a large part of the elements in
the periodic table, and are known by their acronym
STO-nG, where n indicates the number of Gaussian
functions used in the fitting. As these are SZ-type basis
sets, there is for application purposes little improvement of going beyond n = 3, and the STO-3G basis
set is probably the most popular SZ basis set.

POLARIZATION FUNCTIONS

F I G U R E 3 | HartreeFock energy contribution of basis functions

for the krypton atom, using either a full optimization, a fourth-order
Legendre [Eq. (7)] or an even-tempered [Eq. (4)] parameterization for
determining the basis function exponents.

Ne, whereas the corresponding values for Kr are 9 and

20 functions. A fourth-order Legendre polynomial
provides an efficient parameterization, with a penalty
of typically only one or two extra functions, even for
very large basis sets.

10

The addition of higher angular momentum functions

allows the molecular density to distort from the spherical symmetry of the atoms at the HF or DFT levels,
and these functions for correlated calculations, in addition, serve to describe the electron correlation cusp.
The accepted notation is to call these polarization
functions, despite their dual purpose. In the following,
we will discuss polarization functions for first- and
second-row p-block elements only, thereby avoiding
the issue regarding s-block elements and whether to
consider the 3d-orbitals to be core or valence orbitals
for third-row p-block atoms.
Polarization functions should not be used with
SZ basis sets, as the error in the sp-part of the basis is much larger than the error from the lack of
polarization functions. A DZP basis set is defined by

c 2012 John Wiley & Sons, Ltd.



Volume 00, January/February 2012

WIREs Computational Molecular Science

Atomic orbital basis sets

adding a single d-function to a DZ-type basis set (and

a single p-function is added to hydrogen/helium basis
sets). Early applications of TZP-type basis sets often
added two or three d-functions and one f-function
based on performance criteria for a selection of properties and systems.81 On the basis of energy analysis
for atomic and molecular systems, there is now general agreement that the consistent polarization at the
TZ level is two d- and one f-function. Numerical evidence suggest that the consistent QZ and 5Z polarization at HF/DFT levels is 4d2f1g and 6d3f2g1h,43
respectively, whereas it is 3d2f1g and 4d3f2g1h at
correlated levels.36 The difference in the number of
d-functions is related to the difference in basis set
convergence, as discussed in Basis Set Convergence
of Independent- and Correlated-Electron Methods.
Four different strategies have been used for assigning the polarization exponents:
1. A distance or overlap criterion with the valence orbitals;
2. Fitting to atomic natural orbitals;
3. Energy minimization of atomic energies;
4. Energy minimization of molecular energies.
The polarization exponents are in the first strategy assigned based on mathematical criteria, such as
having a maximum overlap with the valence s- and
p-orbitals, or by requiring the same value of the mean
distance to the nucleus. The physical argument for
this procedure is that the polarization functions at
the HF and DFT levels serve to describe the deformation of the valence orbitals, and they should therefore describe the same physical space. Tatewaki and
Huzinaga61 have suggested that d-polarization exponents can be determined by maximizing the orbital
overlap with the valence s- and p-orbitals weighted
by the number of s- and p-electrons:


ns s | d + n p p d
.
(9)
=
ns + n p
Roos and Siegbahn82 have determined polarization exponents by requiring that the expectation value
of the distance to the nucleus is identical:


(10)
rmean = d | r |d = p r p .
On the basis of these criteria, the exponents
should increase with the angular momentum of the
polarization function, and a commonly employed ratio is 1.2 between functions differing by one angular
momentum.

Volume 00, January/February 2012

The second approach has been used by Koga

and coworkers,83 where polarization functions are determined by fitting basis function exponents and contraction coefficients to atomic natural orbitals (ANO),
obtained by diagonalizing the density matrix from
a correlated calculation using a primitive basis set
including polarization functions. The fitting is performed by minimizing a sum of integrated deviations
weighted by a factor of r2 , as shown in Eq. (11):
N+N
f

nkk,

(11)

k=N+1

k =

 ( ANO,kk)2
r2

dv.

The sum includes the correlating orbitals k =

N + 1, . . ., N + Nf , i.e., those that are unoccupied
at the HF level (N is the number of occupied orbitals,
Nf is the number of fitting functions, and nk is the
natural occupation number).
The polarization exponents at HF and DFT levels cannot be determined by energy criteria at the
atomic level, as the energy only depends on the occupied orbitals, but they can be determined by energy
minimization on molecular systems. For methods including electron correlation, the polarization functions make a nonzero energy contribution already
at the atomic level, and polarization exponents can
therefore be determined by minimizing the energy of
isolated atoms.
The use of atomic correlation energies to assign
polarization exponents has been used for several basis
sets (Table 5). The polarization functions for the 6311G(d) basis set have been determined by MP2 optimizations at the atomic level, with 2d and 3d polarization functions obtained by symmetrically splitting the
exponents around the optimum value with a constant
factor of 4 between the exponents.14,81 The cc-pVXZ
basis sets assign polarization functions by minimizing
the atomic CISD energy, using an even-tempered expansion for multiple polarization functions.36,38 The
nZaP basis sets similarly assign polarization functions by MP2 optimization of an even-tempered set
of functions.40
The fourth strategy assigns polarization exponents by energy minimization on molecular systems,
where the optimum values of course will be molecule
specific. The polarization exponents for the 6-31G
basis set were determined by HF energy optimization
of a selection of small molecules,10,84 whereas the pcn basis sets employ a BLYP energy optimization.43
Ahlrichs et al.85 have used minimization of IEPA
energies for molecular systems to determine p-, d-,

c 2012 John Wiley & Sons, Ltd.



11

Advanced Review

wires.wiley.com/wcms

T A B L E 5 Basis Set Parameter Optimization

Atomic sp(d)-Exponents

Polarization Exponents

Label

Name

Method

Target

Method

Pople

STO-3G
6-31G

Atoms
Atoms
Atoms

None
HF-optimized
Averaged
MP2-optimized

Atoms
Atoms

From cc-pVXZ
MP2-optimized

Atoms
Atoms

Ahlrichs
Jorge

Def2-XZP
XZP

Sapporo
Roos

ANO-XZP

Fitted
HF-optimized Identical
s- and p-valence exponents
MP2-optimized Identical s- and
p-valence exponents
HF-optimized
HF-optimized for core
MP2-optimized for valence
HF-optimized
HF-optimized

Dunning

cc-pVXZ

HF-optimized

Atoms

Petersson

nZaP

Atoms

Jensen

pc-n

HF L6-optimized for core

MP2 even-tempered
optimized for valence
BLYP-optimized

6-311G

Atoms

Contraction

Target

Method

Coefficients

Molecules

Segmented
Segmented

Fitted
HF-optimized

Atoms

Segmented

MP2-optimized

Atoms

Segmented
Segmented

HF-optimized
HF-optimized

ANO-fitted
CISD-optimized
Even-tempered
CISD-optimized
Even-tempered
MP2 optimized
Even-tempered

Atom
Atoms

Segmented
General

HF-optimized
ANO-coefficients

Atoms

General

HF-coefficients

Atoms

General

HF-coefficients

BLYP-optimized
Averaged

Molecules

General

BLYP-coefficients

L6 denoted that exponents are optimized using a sixth-order Legendre function, Eq. (7).
Segmented contraction denotes that only a few primitive functions are part of more than one contracted function.
General contraction denotes that a significant number of primitive functions are part of more than one contracted function.

f-polarization, but the Def2 basis sets employ the corresponding cc-pVXZ polarization functions.21
The optimum exponent for a single polarization function is for molecular applications a
compromise between two opposing trends. Maximizing the orbital overlap between the polarization and valence functions is important for describing the atomic electron correlation, and this
leads to a rather strong dependence of the polarization exponent with atomic number. The optimum
exponent for describing the change polarization associated with molecular bonding, on the other hand, depends primarily on the bond distance(s), and this leads
to a much smaller variation with atomic number.86
It is often difficult to achieve a good balance between these two effects when only a single polarization function is used. Having two polarization
functions with different exponents alleviates the problem, in analogy with the improvements upon going from an SZ to DZ sp-basis set, and this is one
of the reasons why results often improve considerably upon going from a DZP- to a TZP-type basis set.

Core Correlating Basis Sets

The cc-pVXZ basis sets focus, as the V implies, on
recovering the valence correlation energy. The basis
set must be improved by adding correlating functions

12

with larger exponents if core and corevalence correlation energies are also desired. The cc-pCVXZ basis
sets have been designed for this purpose, where the
C denotes addition of basis functions with large exponents, often called tight functions. The exponents
of the tight functions are assigned based on minimizing the atomic CISD energy, and the number and
angular momentum functions to add at each level is
again decided based on energy analyses.87 In subsequent work, the additional tight functions have been
optimized with a bias toward recovering intershell
correlation, rather than intrashell correlation, as this
improves the performance for molecular applications,
and these basis sets are denoted cc-pwCVXZ.39,88
The Sapporo basis sets have also been extended to
recover corevalence electron correlation.89

AUGMENTATION WITH DIFFUSE

FUNCTIONS
The basis sets described in the previous sections all
relate to calculating the energy of neutral systems. If
anions are considered, it is clear that the wave function must be improved in regions far from the nuclei, as the extra electron in most cases is relatively
weakly bound. It is customary to improve the description of this region of space by adding functions with

c 2012 John Wiley & Sons, Ltd.



Volume 00, January/February 2012

WIREs Computational Molecular Science

Atomic orbital basis sets

F I G U R E 4 | Number of primitive (blue dots) and contracted (black

F I G U R E 5 | Number of primitive (blue dots) and contracted (black

squares) functions for a first-row s-block atom as a function of basis

set quality for the basis sets in Table 1.

squares) functions for a first-row p-block atom as a function of basis

set quality for the basis sets in Table 1.

small exponents, usually referred to as diffuse functions, as they extend far from the nucleus. Schleyer
and coworkers90 recognized this feature already in
1981, and added a set of diffuse s- and p-type basis
functions to the Pople-style k-lmnG basis sets. The
exponents for the diffuse functions were determined
by HF optimization on the atomic anions.81
A similar requirement was noted in the development of the cc-pVXZ basis sets, and the exponents
for the diffuse functions were in analogy determined
by minimizing the energy of atomic anions.91 In contrast to the Pople-style basis sets, however, a full set
of diffuse functions are included up to the highest angular momentum in the basis set. This reflects that
the cc-pVXZ basis sets are designed for recovering
the correlation energy, which depends on all angular
momentum functions, while the Pople-style basis sets
focused on the HF energy, where only the occupied
atomic orbitals contribute to the energy. Furthermore,
many molecular properties require diffuse polarization functions for an adequate description, and the
aug-cc-pVXZ basis sets thus allow probing features
beyond the energy.92,93
Jorge and coworkers22,94 have similarly defined
diffuse augmenting functions by energy optimization
of atomic anions, whereas diffuse functions for the
Ahlrichs Def2 basis sets have been proposed based
on maximizing the atomic polarizability.95 The nZaP
basis sets include diffuse s- and p-functions in their
definition.

The aug-pc-n basis sets optimized for DFT

methods have similarly been designed to improve the
representation of the density in regions far from the
nuclei.96 As most contemporary exchange-correlation
functionals have problems describing anions due to
self-interaction errors and wrong asymptotic behavior of the potential,97 it is not possible to define the exponents of the diffuse functions by energy minimization of anions. The exponents of the diffuse functions
for the occupied atomic orbitals are instead assigned
by scaling the outermost exponents, while the higher
angular momentum diffuse functions are determined
from an overlap criterion. In analogy with the augcc-pVXZ basis sets, the aug-pc-n basis sets add diffuse functions for all angular momentum functions
included in the basis set to allow a description of a
variety of molecular properties.

Volume 00, January/February 2012

CONTRACTION OF BASIS SETS

A primitive basis set is computationally inefficient,
but serves as a starting point for constructing contracted basis sets. Tables 14 provide an overview of
the eight families of basis sets mentioned in the introduction. Figures 410 illustrate the compositions for
first-, second-, and third-row s-, p-, and d-block elements in terms of contracted and primitive functions
for basis sets of DZP to 5ZP quality.

c 2012 John Wiley & Sons, Ltd.



13

Advanced Review

wires.wiley.com/wcms

F I G U R E 6 | Number of primitive (blue dots) and contracted (black

squares) functions for a second-row s-block atom as a function of basis
set quality for the basis sets in Table 2.

F I G U R E 7 | Number of primitive (blue dots) and contracted (black

Segmented Contracted Basis Sets

Radom and coworkers,16,17 whereas the Wachters

Hay basis sets are used for the third-row transition
metals.19,20
Ahlrichs and coworkers63 have defined SVP
(split valence polarized), TZVP,102 and QZVP60 basis sets by optimizing both the exponents and contraction coefficients at the HF level for atoms, and
these have later been reoptimized to the so-called
Def2 versions.21 Two slightly different sets of polarization functions are available for third-row p- and
d-elements (Tables 3 and 4), depending on whether
the basis sets are intended used for HF/DFT or correlated methods, where the latter are labeled as TZVPP
and QZVPP.
Jorge and coworkers22,25 have defined XZP basis sets up to 5ZP for first- and second-row elements,
and up to TZP for third-row elements, where the
atomic core functions are optimized at the HF level,
whereas the valence and polarization functions are
optimized at the MP2 level. Other groups have reported basis sets optimized at the DFT level, but these
are defined only for a limited number of atoms and
basis set quality.103105
The Sapporo group has developed DZP, TZP,
and QZP quality basis sets where functions describing the occupied atomic orbitals are optimized at the
HF level, whereas polarization functions are determined by fitting to CISD atomic natural orbitals.2730
These basis sets contain more primitive polarization
functions than similar quality basis sets.

Early work focused on segmented contractions, where

the primitive functions were grouped into disjoint
partitions based on analyses for small systems.51,98
The basis set exponents were in this work kept
fixed at their optimum values for the uncontracted
case, whereas later work employed a simultaneous
optimization of both the exponents and contraction
coefficients.63,99
The most widely known examples of segmented
basis sets are the k-lmnG basis sets developed by
Pople and coworkers. These are of either DZ (klmG)8,99101 or TZ quality (k-lmnG),14 where k, l,
m, and n indicate the number of primitive functions
used in a given contracted function. On the basis of
earlier studies with STOs, which showed that there
was little difference between the optimum 2s- and 2pexponents, it was decided to restrict the exponents
of the valence s- and p-exponents to be identical,
which increased the computational efficiency. With
this restriction, all the exponents and contraction
coefficients were optimized for atoms at the HF level
for 6-31G and at the MP2 level for 6-311G. Note
that there are two different versions of the 6-31G
basis set for third-row s- and p-block elements (Table 3),12,13 and different programs employ different
defaults. The 6-311G basis set is formally only defined for first-row elements,14 but similar quality basis
sets for second- and third-row s- and p-elements have
been published by McLean and Chandler,15 and by

14

squares) functions for a second-row p-block atom as a function of

basis set quality for the basis sets in Table 2.

c 2012 John Wiley & Sons, Ltd.



Volume 00, January/February 2012

WIREs Computational Molecular Science

Atomic orbital basis sets

F I G U R E 8 | Number of primitive (blue dots) and contracted (black

F I G U R E 9 | Number of primitive (blue dots) and contracted (black

squares) functions for a third-row s-block atom as a function of basis

set quality for the basis sets in Table 3.

squares) functions for a third-row p-block atom as a function of basis

set quality for the basis sets in Table 3.

One of the main problems with segmented basis sets is the somewhat arbitrary decisions on how
many primitive functions to use and how they are contracted to define a given basis set. Even if the number
of primitive and contracted functions is decided, there
are many possibilities for how the primitives are distributed between the contracted functions, and each
of these combinations may furthermore have several
local minima in the combined exponent/coefficient
parameter space.
Veillard106 has investigated nine different contractions of a (12s9p) basis set to [6s4p] for secondrow p-block atoms. One could argue that the best
contraction scheme is the one with the lowest energy, but somewhat discouraging this criterion gives
three different contraction schemes for the atoms Si, S,
and Ar. The difference is located in the p-contraction,
where either a {4212}, {3312}, or {5112} contraction gives the lowest energy. All of these have the
outer valence orbital as a contraction of two primitive functions, but chemical intuition suggests that
the outer functions should be left uncontracted, i.e.,
a {5211}or a {6111} contraction will in molecular
calculations perform better.
We have similarly shown that there exist at least
19 different minima in the combined exponent and
contraction coefficient space for a (11s5p) primitive
set of functions contracted to [4s3p] as {6311/311}
for the nitrogen atom.107 The official 6-311G basis
set correspond to number seven in terms of energy,

Volume 00, January/February 2012

but almost all of the 19 exponent/coefficient combinations have very similar performances for molecular calculations. Combining this arbitrariness with
the possibility of duplicating one or more primitive
functions makes the construction of segmented basis sets somewhat of an art. It should be noted that
these problems increase beyond the second row in
the periodic table and for constructing basis sets of
better than TZ or QZ quality for first and secondrow elements. Designing segmented contracted basis
sets which systematically approach the basis set limit
therefore becomes challenging beyond the TZ level.

General Contracted Basis Sets

The analysis of natural orbitals suggests that the best
approach for designing basis sets for electron correlation is to make a general contraction of a large primitive basis set by using the coefficients from atomic
natural orbitals (ANO).59,108 A single (large) set of
primitive functions can be contracted to basis sets
of increasingly quality which smoothly converge toward the result for the uncontracted primitive basis set. Roos and coworkers have defined such ANO
basis sets, where a common set of primitive functions is contracted to basis sets of DZP, TZP, and
QZP quality.3134 The main drawback of these basis
sets is the computational inefficiency due to the large
underlying primitive basis set. The exception is for
highly correlated methods, where the cost of integral

c 2012 John Wiley & Sons, Ltd.



15

Advanced Review

wires.wiley.com/wcms

ergy criterion, which leads to a different number of

primitive functions for atoms in the same row of the
periodic table.
We have developed the polarization consistent
basis sets by a general contraction of BLYP optimized
primitive functions, where the contraction coefficients
are taken as the BLYP atomic orbital coefficients.4348
The number of primitive functions is defined by an energy criterion relative to the highest angular momentum function included, and the degree of contraction
is determined by requiring that the contraction error
is smaller than the inherent error in the primitive basis set, relative to the basis sets limit. Five basis sets
of increasing quality have been defined, denoted pc-n
(n = 04), with qualities ranging from (unpolarized)
DZ to (polarized) 5ZP.

F I G U R E 1 0 | Number of primitive (blue dots) and contracted

(black squares) functions for a third-row d-block atom as a function of
basis set quality for the basis sets in Table 4.

evaluation is only a minor component of the total

computational cost.
Guided by the ideas leading to the ANO concept, Dunning36 proposed the correlation consistent
basis sets (cc-pVXZ) based on a careful analysis of the
energy contributions of different basis functions for
atoms at the CISD level, which clearly suggested that
polarization functions should be included in the order
1d, 2d1f, 3d2f1g, 4d3f2g1h, and so on. The underlying atomic density is described by a general contracted
SZ basis set with s- and p-functions, to which there
is added one set of correlating s- and p-functions at
the DZ level, two sets at the TZ level, and so on. To
achieve error balance, the number of primitive functions is also increased for each step up in basis set
quality. A SZ basis set is not useful for correlated
methods, and the smallest basis set is thus of DZP
quality. For second-row elements, it was later discovered that the results could be improved by adding
an extra d-function with a larger exponent.37 These
basis sets are sometimes denoted cc-pV(X+d)Z to distinguish them from the original set.
Petersson and coworkers have developed similar types of basis sets, called nZaP (n = 26), where
the exponents for the core functions are assigned
based on HF optimization of a sixth-order Legrendre parameterization [Eq. (7)]62 and valence and polarization functions are taken as a MP2 optimized
even-tempered sequence.4042 The number of primitive functions is assigned based on an absolute en-

16

BASIS SET CONVERGENCE OF

INDEPENDENT- AND
CORRELATED-ELECTRON METHODS
The use of basis sets may be considered as a necessary
evil, as it introduces a second source of error, besides
that inherent in the method used for solving the electronic structure problem. Unraveling the method and
basis set errors is often difficult, but is an essential
component of method calibration. The use of a hierarchy of basis sets has the significant advantage that it
allows an estimate of the (remaining) basis set error,
and in favorable cases, permits an extrapolation to the
basis set limiting value. A key component in the latter is choosing a suitable mathematical extrapolating
function.
Klopper and Kutzelnigg109111 showed that the
energy of a hydrogen atom converges exponentially
with the square-root of the number of Gaussian functions used for expanding the 1s-orbital, as shown in
Eq. (12):
HF
= E + AeB
EM

(12)

Figures 2 and 3 show that the HF energy also

converges exponentially for many-electron atoms.
Close inspection of the curve for the fully optimized exponents shows that it deviates slightly from
a straight line. The upward concave shape is qualitatively reproduced by an exponential-square-root expression as in Eq. (12), although the improvement in
terms of regression coefficient relative to a straight
exponential function is marginal.
Numerical results for molecular systems suggest
that the HF energy convergence with respect to angular momentum (L) functions also is exponential, but

c 2012 John Wiley & Sons, Ltd.



Volume 00, January/February 2012

WIREs Computational Molecular Science

Atomic orbital basis sets

it is difficult from the numerical

data to differentiate

between an exp(L) or exp( L) convergence.112

The energy convergence of DFT methods appears to be very similar to HF, which is not surprising,
given that the three main components of the energy
(electron kinetic energy, and nuclear-electron and
electronelectron Coulomb interactions) have identical expressions as in HF theory.113
While the basis set at HF or DFT levels only
needs to represent the electron density, which is a
one-electron quantity, the basis set for the correlation
energy also needs to represent the electronelectron
correlation cusp, which is a two-electron quantity,
and this leads to a difference in convergence behavior.
The convergence of the correlation energy (not
the total energy) as a function of angular momentum
has been analyzed by Kutzelnigg and Morgan.114,115
For a basis set saturated up to angular momentum
function L, the remaining error can be expressed as
an inverse polynomial in L, where the leading term
is L3 for a singlet coupled electron pair and L5 for
a triplet coupled pair. This suggests that an inverse
polynomial formula such as Eq. (13) may be suitable
for extrapolating the correlation energy:
ELCorrelation = E + AL3 .

(13)

Other formulas including more terms or using

an empirical exponent different from 3 have also been
used. The underlying physical reason for the significantly slower convergence of the correlation energy
compared to HF, is that products of one-electron
functions are poor at representing the Coulomb hole
surrounding each electron. For a triplet coupled electron pair, the Fermi hole, which is taken into account
by the antisymmetric wave function, takes care of the
majority of the correlation cusp, and therefore has a
faster convergence. For the specific case of MP2, Petersson et al.116 have shown that the basis set limiting
energy can be estimated from a calculation with a single basis set by projecting onto pair natural orbitals.
The difference in convergence behavior of
independent- and correlated-electron methods has
the consequence that large optimum basis sets for
HF/DFT methods have more low-angular momentum functions than basis sets optimized for correlated
methods.
The slow basis set convergence of the correlation cusp has motivated the development of methods
incorporating a correlation function depending on the
interelectronic distance, the so-called F12 methods.117
These methods have now reached maturity, and as
the correlation cusp is taken into account explicitly,
the basis set convergence is inverse polynomial with

Volume 00, January/February 2012

the leading term being proportional to L7 . Peterson

and coworkers118,119 have used the same principles
as for designing the cc-pVXZ basis sets to optimize
basis sets for F12 methods, and these are denoted ccpVXZ-F12. Relative to cc-pVXZ they contain both
more primitive and contracted functions to achieve
a proper balance between the HF and correlation
energies.
It can be argued from the theoretical analyses
that the HF/DFT energy should be extrapolated by
an exponential formula, whereas the correlation energy should be extrapolated separately by an inverse
polynomial formula. For non-F12 methods, however,
the fast(er) exponential convergence of the HF energy component means that the convergence of the
total energy follows the slow component, the electron
correlation energy, and inverse polynomial extrapolations may therefore be suitable also for the total
energy.
Exponential extrapolation formulas, such as
Eq. (12), are formally three-point extrapolations, but
the exponential B-parameter can to a good approximation be taken as a universal constant, in which case
the extrapolation only requires two points. Functional
forms based on a single inverse polynomial such as
Eq. (13) are inherently two-point extrapolations.
The analysis showing that the L3 term dominates for the correlation energy is only strictly valid
for a basis set saturated up to angular momentum
level L, which is not fulfilled in practice, and higherorder terms such as L4 and L5 also contribute. A
variety of exponential, multiexponential and inverse
polynomial functional forms have been used for extrapolating the energy calculated with a sequence of
basis sets toward the basis set limit, and a recent calibration study for atomization energies showed that
none of the proposed formulas stand out as being
uniquely best.120 The common feature is that extrapolations using only DZP and TZP results often give
only marginal improvements, as a DZP basis set is
too small to represent the correlation cusp with any
useful accuracy. The improvement from extrapolation increases as larger and larger basis sets are used,
i.e., the relative improvements by extrapolating QZP
and 5ZP results are larger than by extrapolating TZP
and QZP data. In favorable cases, extrapolation can
provide results of typically one basis set higher quality
than the largest calculation, e.g., 5ZP quality results
from extrapolation of TZP and QZP data.
The success of extrapolation formulas is rooted
in the use of hierarchical basis sets, which in a systematic fashion improves the results for each step
up in quality. As any reasonable extrapolation formula reduces the basis set error, at no additional

c 2012 John Wiley & Sons, Ltd.



17

Advanced Review

wires.wiley.com/wcms

BOX 1: WEB-BASED BASIS SET LIBRARIES

A number of Web sites allow downloading of basis sets in
an electronic form suitable for use with different programs:
The EMSL library includes Pople, Ahlrichs, cc-pVXZ and pc-n
basis sets: https://bse.pnl.gov/bse/portal
The Ahlrichs basis sets: http://bases.turbo-forum.com/
TBL/tbl.html
The Jorge basis sets: http://www.cce.ufes.br/qcgv/pub
The Sapporo basis sets: http://setani.sci.hokudai.ac.jp/
sapporo

computational cost, there is little reason not to use

extrapolations, but it does of course require that calculations are performed with at least two increasingly
larger basis sets. One may argue that this should always be done anyway, to ensure that the obtained results are not artifacts of employing a specific (small)
basis set.
Basis set extrapolations allow an estimate of the
basis set limiting result for a given electronic structure
method (HF, DFT, MPn, CCSD, . . ..). To compare
with experimental results, it is also necessary to estimate the remaining error in the method itself, and
attempt to extrapolate toward the infinite electron
correlation limit. A brute force approach is infeasible from a computational point of view, and a variety
of composite models has been proposed where results
from several different methods and basis sets are used
to estimate the quantity directly comparable to the
experimental value. The Gn methods have recently
been reviewed,121 and this reference also contains
references to similar approaches such as the Wn,122
ccCA,123 and related124 methods.

CONCLUSION
Applications of electronic structure methods have traditionally relied on what J. A. Pople defined as models,
i.e., selecting a combination of method and basis set,
and calibrating the performance by comparing with
experimental results. The success of this approach is
evident by the literally thousands of papers describing results obtained at the HF, MP2, and B3LYP levels of theory using the 6-31G or similar-sized basis
sets. As a quality measure for a given application can
only be obtained by comparing with previous performances for related systems and properties, these
models tend to develop into self-sustaining loops. A
researcher looking for a suitable model to use for a
given problem cannot fail to notice the large body
of previous work using models like B3LYP/6-31G ,

18

and therefore often selects this level, thereby further

adding to the reference database.
For a sufficiently large database the performance of a specific model can be assessed by statistical methods, but there is always the risk that a
given (new) system may be pathological and display
errors far outside the calibration data set. Such a failure will not be revealed by a model-type calculation,
as there is no way of assessing the errors in the results from the model itself. The model approach furthermore becomes problematic when used for properties where only a few or no calibration data are
available.
The model approach also suffers from difficulties in making a direct comparison between the
computational and experimental results. A popular
molecular property for calibration is the atomization
energy, where the calculated results usually are performed using nonrelativistic methods and employing
the rigid-rotor harmonic-oscillator model for including finite temperature corrections. The experimental
quantity, on the other hand, includes relativistic effects and anharmonic vibrations, and frequently have
larger error bars than reported.
The search for magic combinations of methods and basis sets that statistically give good results
for a (limited) selection of molecules and properties
is a popular topic in the literature, and this has become especially pronounced with the proliferation of
DFT methods. A significant fraction of these model
calibration studies benefits from cancellation between
the method and basis set errors, and from errors in
the theoretical model compensating for the differences
between the theoretically calculated and the experimental measured quantity. DFT methods employing
empirical parameters often fit the parameters based
on comparison with experimental data using results
obtained with a specific basis set, and the DFT parameters in such cases absorb some of the remaining
basis set errors.
The trend in electronic structure methods is
a move toward a more rigorous control of the
errors in the calculated results without reference to
external data. This requires a series of calculations
with methods and basis sets that can be systematically improved. There are well-defined hierarchies of
computational models in wave function theory that
are guaranteed to converge toward the exact result, of
which coupled cluster methods currently are the most
successful. Density functional methods currently do
not hold a status as being systematically improvable,
but their favorable combination of good accuracy and
low computational cost makes them very useful from
an application point of view.

c 2012 John Wiley & Sons, Ltd.



Volume 00, January/February 2012

WIREs Computational Molecular Science

Atomic orbital basis sets

Assessing the basis set incompleteness is an important component in establishing error control of
the calculated results. This requires calculations with
at least two different basis sets, selected such that
there is a systematic improvement between the two.
The systematic nature not only guarantees a genuine
improvement, but also allows extrapolations to be
performed, thereby reducing the remaining basis set
error without any additional computational effort.
A practical requirement for selecting a basis
set is the availability for a variety of elements. It
is also important that diffuse and/or tight functions
can be added to improve the convergence for sys-

tems/properties where standard basis sets are known

to be deficient. There are now a number of hierarchical basis sets that meet these criteria (Box 1), and
this has been the focus of the present review. Some of
these basis sets have been optimized for specific types
of theoretical methods, while others are designed as
general purpose basis sets. Although model-type calculations without doubt will continue to be used in
thousands of applications also in the future, one can
hope that the continuing increase in computational
resources will inspire researchers to use hierarchical
basis sets to gain insight into and control the inherent
errors in the calculated results.

ACKNOWLEDGMENTS
The author thanks Prof. G. A. Petersson and Prof. R. Lindh for information regarding the
nZaP and ANO basis sets. This work was supported by grants from the Danish Center for
Scientific Computation and the Danish Natural Science Research Council.

REFERENCES
1. Slater JC. Atomic shielding constants. Phys Rev 1930,
36:00570064.
2. Boys SF. Electronic wave functions. 1. A general
method of calculation for the stationary states of any
molecular system. Proc R Soc London A: Math Phys
Sci 1950, 200:542554.
3. Flad HJ, Hackbusch W, Kolb D, Schneider R. Wavelet
approximation of correlated wave functions. I. Basics.
J Chem Phys 2002, 116:96419657.
4. Sekino H, Maeda Y, Yanai T, Harrison RJ. Basis set
limit HartreeFock and density functional theory response property evaluation by multiresolution multiwavelet basis. J Chem Phys 2008, 129:034111.
5. Bischoff FA, Valeev EF. Low-order tensor approximations for electronic wave functions: HartreeFock
method with guaranteed precision. J Chem Phys
2011, 134:104104.
6. Cao X, Dolg M. Pseudopotentials and modelpotentials. Wiley Interdiscip Rev: Comput Mol Sci 2011,
1:200210.
7. Dolg M, Cao X. Relativistic pseudopotentials: their
development and scope of applications. Chem Rev
2012, 112:403480.
8. Hehre WJ, Ditchfie R, Pople JA. Self-consistent
molecular-orbital methods. 12. Further Extensions of
Gaussian-type basis sets for use in molecular-orbital
studies of organic-molecules. J Chem Phys 1972,
56:22572261.

Volume 00, January/February 2012

9. Dill JD, Pople JA. Self-consistent molecular-orbital

methods. 15. Extended Gaussian-type basis sets for
lithium, beryllium, and boron. J Chem Phys 1975,
62:29212923.
10. Francl MM, Pietro WJ, Hehre WJ, Binkley JS, Gordon
MS, Defrees DJ, Pople JA. Self-consistent molecularorbital methods. 23. A polarization-type basis set
for 2nd-row elements. J Chem Phys 1982, 77:3654
3665.
11. Rassolov VA, Pople JA, Ratner MA, Windus TL. 6-31G basis set for atoms K
through Zn. J Chem Phys 1998, 109:1223
1229.
12. Rassolov VA, Ratner MA, Pople JA, Redfern PC, Curtiss LA. 6-31G basis set for third-row atoms. J Comput Chem 2001, 22:976984.
13. Binning RC, Curtiss LA. Compact contracted basissets for 3rd-row atomsGA-KR. J Comput Chem
1990, 11:12061216.
14. Krishnan R, Binkley JS, Seeger R, Pople JA. Selfconsistent molecular-orbital methods. 20. Basis set
for correlated wave-functions. J Chem Phys 1980,
72:650654.
15. McLean AD, Chandler GS. Contracted Gaussianbasis sets for molecular calculations. 1. Second row
atoms, Z = 11-18. J Chem Phys 1980, 72:56395648.
16. Curtiss LA, McGrath MP, Blaudeau JP, Davis NE,
Binning RC, Radom L. Extension of Gaussian-2

c 2012 John Wiley & Sons, Ltd.



19

Advanced Review

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

20

wires.wiley.com/wcms

theory to molecules containing 3rd-row atoms GAKR. J Chem Phys 1995, 103:61046113.
Blaudeau JP, McGrath MP, Curtiss LA, Radom L.
Extension of Gaussian-2 (G2) theory to molecules
containing third-row atoms K and Ca. J Chem Phys
1997, 107:50165021.
Dobbs KD, Hehre WJ. Molecular-orbital theory of
the properties of inorganic and organometallic compounds. 5. Extended basis-sets for 1st-row transitionmetals. J Comput Chem 1987, 8:861879.
Wachters AJ. Gaussian basis set for molecular wavefunctions containing third-row atoms. J Chem Phys
1970, 52:10331036.
Hay PJ. Gaussian basis sets for molecular
calculationsrepresentation of 3D orbitals in
transition-metal atoms. J Chem Phys 1977, 66:4377
4384.
Weigend F, Ahlrichs R. Balanced basis sets of split valence, triple zeta valence and quadruple zeta valence
quality for H to Rn: design and assessment of accuracy. Phys Chem Chem Phys 2005, 7:32973305.
Neto AC, Muniz EP, Centoducatte R, Jorge FE. Gaussian basis sets for correlated wave functions. Hydrogen, helium, first- and second-row atoms. Theochem:
J Mol Struct 2005, 718:219224.
Jorge FE, Sagrillo PS, de Oliveira AR. Gaussian basis sets of 5 zeta valence quality for correlated wave
functions. Chem Phys Lett 2006, 432:558563.
Barbieri PL, Fantin PA, Jorge FE. Gaussian basis sets
of triple and quadruple zeta valence quality for correlated wave functions. Mol Phys 2006, 104:2945
2954.
Machado SF, Camiletti GG, Canal Neto A, Jorge FE,
Jorge RS. Gaussian basis set of triple zeta valence
quality for the atoms from K to Kr: application in DFT
and CCSD(T) calculations of molecular properties.
Mol Phys 2009, 107:17131727.
Camiletti GG, Machado SF, Jorge FE. Gaussian basis
set of double zeta quality for atoms K through Kr: application in DFT calculations of molecular properties.
J Comput Chem 2008, 29:24342444.
Thakkar AJ, Koga T, Saito M, Hoffmeyer RE. Double and quadruple zeta-contracted Gaussian-basis sets
for hydrogen through neon. Int J Quantum Chem
1993:343354.
Tatewaki H, Koga T, Takashima H. Contracted
Gaussian-type basis functions revisited. 2. Atoms Na
through Ar. Theor Chem Acc 1997, 96:243247.
Koga T, Tatewaki H, Matsuyama H, Satoh Y.
Contracted Gaussian-type basis functions revisited.
III. Atoms K through Kr. Theor Chem Acc 1999,
102:105111.
Noro T, Sekiya M, Koga T, Matsuyama H. Valence and correlated basis sets for the first-row transition atoms from Sc to Zn. Theor Chem Acc 2000,
104:146152.

31. Widmark PO, Malmqvist PA, Roos BO. Densitymatrix averaged atomic natural orbital (ANO) basissets for correlated molecular wave-functions. 1.
First row atoms. Theor Chim Acta 1990, 77:291
306.
32. Widmark PO, Joakim B, Persson, Roos BO. Densitymatrix averaged atomic natural orbital (ANO) basissets for correlated molecular wave-functions. 2. Second row atoms. Theor Chim Acta 1991, 79:419
432.
33. Pouamerigo R, Merchan M, Nebotgil I, Widmark PO,
Roos BO. Density-matrix averaged atomic natural orbital (ANO) basis-sets for correlated molecular wavefunctions. 3. First row transition-metal atoms. Theor
Chim Acta 1995, 92:149181.
34. Pierloot K, Dumez B, Widmark PO, Roos BO.
Density-matrix averaged atomic natural orbital (Ano)
basis-sets for correlated molecular wave-functions. 4.
Medium-size basis-sets for the atoms H-Kr. Theor
Chim Acta 1995, 90:87114.
35. Roos BO, Lindh R, Malmqvist PA, Veryazov V, Widmark PO. Main group atoms and dimers studied with
a new relativistic ANO basis set. J Phys Chem A 2004,
108:28512858.
36. Dunning TH. Gaussian-basis sets for use in correlated
molecular calculations. 1. The atoms boron through
neon and hydrogen. J Chem Phys 1989, 90:1007
1023.
37. Dunning TH, Peterson KA, Wilson AK. Gaussian basis sets for use in correlated molecular calculations.
X. The atoms aluminum through argon revisited. J
Chem Phys 2001, 114:92449253.
38. Wilson AK, Woon DE, Peterson KA, Dunning TH.
Gaussian basis sets for use in correlated molecular
calculations. IX. The atoms gallium through krypton.
J Chem Phys 1999, 110:76677676.
39. Balabanov NB, Peterson KA. Systematically convergent basis sets for transition metals. I. All-electron
correlation consistent basis sets for the 3d elements
Sc-Zn. J Chem Phys 2005, 123:064107.
40. Zhong S, Barnes EC, Petersson GA. Uniformly convergent n-tuple-zeta augmented polarized (nZaP) basis sets for complete basis set extrapolations. I.
Self-consistent field energies. J Chem Phys 2008,
129:184116.
41. Barnes EC, Petersson GA, Feller D, Peterson KA. The
CCSD(T) complete basis set limit for Ne revisited. J
Chem Phys 2008, 129:194115.
42. Barnes EC, Petersson GA. MP2/CBS atomic and
molecular benchmarks for H through Ar. J Chem
Phys 2010, 132:114111.
43. Jensen F. Polarization consistent basis sets: principles.
J Chem Phys 2001, 115:91139125.
44. Jensen F. Polarization consistent basis sets: principles (vol 115, pg 9113, 2001). J Chem Phys 2002,
116:35023502.

c 2012 John Wiley & Sons, Ltd.



Volume 00, January/February 2012

WIREs Computational Molecular Science

Atomic orbital basis sets

45. Jensen F, Helgaker T. Polarization consistent basis sets. V. The elements Si-Cl. J Chem Phys 2004,
121:34633470.
46. Jensen F. Polarization consistent basis sets. 4: the elements He, Li, Be, B, Ne, Na, Mg, Al, and Ar. J Phys
Chem A 2007, 111:1119811204.
47. Jensen F. Polarization consistent basis sets. VII. The
elements K, Ca, Ga, Ge, As, Se, Br, and Kr. J Chem
Phys 2012, 136:114107.
48. Jensen F. 2012 (submitted).
49. Helgaker T, Jrgensen P, Olsen J. Molecular
Electronic-Structure Theory. Chichester: John Wiley
& Sons; 2000.
50. Jensen F. Introduction to Computational Chemistry.
Chichester: John Wiley & Sons; 2007.
51. Dunning TH. Gaussian basis functions for use in
molecular calculations. 1. Contraction of (9s5p)
atomic basis sets for first-row atoms. J Chem Phys
1970, 53:28232833.
52. Raffenet Rc. General contraction of Gaussian atomic
orbitalscore, valence, polarization, and diffuse basis setsmolecular integral evaluation. J Chem Phys
1973, 58:44524458.
53. Davidson ER. Comment on Dunnings correlationconsistent basis sets. Chem Phys Lett 1996, 260:514
518.
54. Almlof J, Faegri K, Korsell K. Principles for a direct
SCF approach to LCAO-MO ab initio calculations. J
Comput Chem 1982, 3:385399.
55. Haser M, Ahlrichs R. Improvements on the direct SCF
method. J Comput Chem 1989, 10:104111.
56. Lambrecht DS, Ochsenfeld C. Multipole-based integral estimates for the rigorous description of distance
dependence in two-electron integrals. J Chem Phys
2005, 123:184101184101.
57. Mulliken RS. Criteria for construction of good selfconsistent-field molecular orbital wave functions, and
significance of LCAO-MO population analysis. J
Chem Phys 1962, 36:34283439.
58. Clementi E, Popkie H. Study of structure of molecular
complexes. 1. Energy surface of a water molecule in
field of a lithium positive-ion. J Chem Phys 1972,
57:10771094.
59. Almlof J, Taylor PR. General contraction of
Gaussian-basis sets. 1. Atomic natural orbitals for
1st-row and 2nd-row atoms. J Chem Phys 1987,
86:40704077.
60. Weigend F, Furche F, Ahlrichs R. Gaussian basis sets
of quadruple zeta valence quality for atoms HKr. J
Chem Phys 2003, 119:1275312762.
61. Tatewaki H, Huzinaga S. A systematic preparation of
new contracted Gaussian-type orbital sets. 3. Secondrow atoms from Li through Ne. J Comput Chem
1980, 1:205228.

Volume 00, January/February 2012

62. Kari RE, Mezey PG, Csizmadia IG. Quality of Gaussian basis setsdirect optimization of orbital exponents by method of conjugate gradients. J Chem Phys
1975, 63:581585.
63. Schafer A, Horn H, Ahlrichs R. Fully optimized contracted Gaussian-basis sets for atoms Li to Kr. J Chem
Phys 1992, 97:25712577.
64. Faegri K, Almlof J. Energy-optimized GTO basis-sets
for LCAO calculationsa gradient approach. J Comput Chem 1986, 7:396405.
65. Mezey PG, Csizmadia IG, Kari RE. Uniform quality Gaussian basis sets. 2. Multiple optima of small
Gaussian basis sets for 1st row elements. J Chem Phys
1977, 67:29272928.
66. Bardo RD, Ruedenbe.K. Even-tempered atomic orbitals. 6. Optimal orbital exponents and optimal contractions of Gaussian primitives for hydrogen, carbon, and oxygen in molecules. J Chem Phys 1974,
60:918931.
67. Schmidt MW, Ruedenberg K. Effective convergence to complete orbital bases and to the atomic
HartreeFock limit through systematic sequences of
Gaussian primitives. J Chem Phys 1979, 71:3951
3962.
68. Silver DM, Nieuwpoort WC. Universal Atomic
Basis Sets. Chem Phys Lett 1978, 57:421
422.
69. Moncrieff D, Wilson S. A universal basis set for highprecision molecular electronic structure studies: correlation effects in the ground states of N-2, CO, BF
and NO+. J Phys B-Atomic Mol Opt Phys 1998,
31:38193841.
70. Jorge FE, de Castro EVR. Improved generator coordinate HartreeFock method: application to first-row
atoms. Chem Phys Lett 1999, 302:454460.
71. Huzinaga S, Klobukowski M. Well-tempered GTF
basis-sets for the atoms K through Xe. Chem Phys
Lett 1985, 120:509512.
72. Huzinaga S, Klobukowski M, Tatewaki H. The welltempered GTF basis-sets and their applications in
the SCF calculations on N-2, Co, Na-2, and P-2.
Can J ChemRevue Canadienne De Chimie 1985,
63:18121828.
73. Petersson GA, Zhong SJ, Montgomery JA, Frisch MJ.
On the optimization of Gaussian basis sets. J Chem
Phys 2003, 118:11011109.
74. Cohen AJ, Handy NC. Density functional generalized
gradient calculations using Slater basis sets. J Chem
Phys 2002, 117:14701478.
75. Van Lenthe E, Baerends EJ. Optimized Slater-type
basis sets for the elements 1118. J Comput Chem
2003, 24:11421156.
76. Clementi E, Raimondi DL. Atomic screening constants from scf functions. J Chem Phys 1963,
38:26862689.

c 2012 John Wiley & Sons, Ltd.



21

Advanced Review

wires.wiley.com/wcms

77. Huzinaga S. Gaussian-type functions for polyatomic

systems. I. J Chem Phys 1965, 42:12931302.
78. Hehre WJ, Stewart RF, Pople JA. Self-consistent
molecular-orbital methods. I. Use of Gaussian expansions of Slater-type atomic orbitals. J Chem Phys
1969, 51:26572664.
79. Stewart RF. Small Gaussian expansions of atomic orbitals. J Chem Phys 1969, 50:24852495.
80. Hehre WJ, Ditchfie R, Stewart RF, Pople JA. Selfconsistent molecular orbital methods. 4. Use of Gaussian expansions of Slater-type orbitalsextension to
second-row molecules. J Chem Phys 1970, 52:2769
2773.
81. Frisch MJ, Pople JA, Binkley JS. Self-consistent
molecular-orbital methods. 25. Supplementary functions for Gaussian-basis sets. J Chem Phys 1984,
80:32653269.
82. Roos B, Siegbahn P. Polarization functions for first
and second row atoms in Gaussian type MO-SCF
calculations. Theor Chim Acta 1970, 17:199208.
83. Noro T, Sekiya M, Koga T. Contracted polarization
functions for the atoms helium through neon. Theor
Chem Acc 1997, 98:2532.
84. Harihara PC, Pople JA. Influence of polarization functions on molecular-orbital hydrogenation energies.
Theor Chim Acta 1973, 28:213222.
85. Ahlrichs R, Driessler F, Lischka H, Staemmler V,
Kutzelnigg W. PNO-CI (pair natural orbital configuration interaction) and CEPA-PNO (coupled electron
pair approximation with pair natural orbitals) calculations of molecular systems. 2. Molecules BEH2,
BH, BH3, CH4, CH-3, NH3 (planar and pyramidal),
H2O, OH+3, HF and NE atom. J Chem Phys 1975,
62:12351247.
86. Magnusson E. Supplementary d and f functions in
molecular wave-functions at large and small internuclear separations. J Comput Chem 1993, 14:6774.
87. Woon DE, Dunning TH. Gaussian-basis sets for use
in correlated molecular calculations. 5. Core-valence
basis-sets for boron through neon. J Chem Phys 1995,
103:45724585.
88. Peterson KA, Dunning TH. Accurate correlation consistent basis sets for molecular core-valence correlation effects: the second row atoms Al-Ar, and the
first row atoms B-Ne revisited. J Chem Phys 2002,
117:1054810560.
89. Noro T, Sekiya M, Koga T. Segmented contracted
basis sets for atoms H through Xe: Sapporo-(DK)nZP sets (n = D, T, Q). Theor Chem Acc 2012, 131.
90. Chandrasekhar J, Andrade JG, Schleyer PV. Efficient
and accurate calculation of anion proton affinities. J
Am Chem Soc 1981, 103:56095612.
91. Kendall RA, Dunning TH, Harrison RJ. Electronaffinities of the 1st-row atoms revisitedsystematic
basis-sets and wave-functions. J Chem Phys 1992,
96:67966806.

22

92. Woon DE, Dunning TH. Gaussian-basis sets for use

in correlated molecular calculations. 4. Calculation
of static electrical response properties. J Chem Phys
1994, 100:29752988.
93. Peterson KA, Dunning TH. The CO molecule: the role
of basis set and correlation treatment in the calculation of molecular properties. Theochem: J Mol Struct
1997, 400:93117.
94. Camiletti GG, Canal Neto A, Jorge FE, Machado
SF. Augmented Gaussian basis sets of double and
triple zeta valence qualities for the atoms K and ScKr: applications in HF, MP2, and DFT calculations
of molecular electric properties. Theochem: J Mol
Struct- 2009, 910:122125.
95. Rappoport D, Furche F. Property-optimized Gaussian
basis sets for molecular response calculations. J Chem
Phys 2010, 133.
96. Jensen F. Polarization consistent basis sets. III. The
importance of diffuse functions. J Chem Phys 2002,
117:92349240.
97. Jensen F. Describing anions by density functional theory: fractional electron affinity. J Chem Theory Comput 2010, 6:27262735.
98. Dunning TH. Gaussian basis functions for use in
molecular calculations. 3. Contraction of (10s6p)
atomic basis sets for first-row atoms. J Chem Phys
1971, 55:716723.
99. Ditchfie R, Hehre WJ, Pople JA. Self-consistent
molecular-orbital methods. 9. Extended Gaussiantype basis for molecular-orbital studies of organic molecules. J Chem Phys 1971, 54:724
728.
100. Gordon MS, Binkley JS, Pople JA, Pietro WJ, Hehre
WJ. Self-consistent molecular-orbital methods. 22.
Small split-valence basis-sets for 2nd-row elements.
J Am Chem Soc 1982, 104:27972803.
101. Binkley JS, Pople JA, Hehre WJ. Self-consistent
molecular-orbital methods. 21. Small split-valence
basis-sets for 1st-row elements. J Am Chem Soc 1980,
102:939947.
102. Schafer A, Huber C, Ahlrichs R. Fully optimized contracted Gaussian-basis sets of triple zeta valence quality for atoms LI to KR. J Chem Phys 1994, 100:5829
5835.
103. Godbout N, Salahub DR, Andzelm J, Wimmer E.
Optimization of Gaussian-type basis-sets for local
spin-density functional calculations. 1. Boron through
neon, optimization technique and validation. Can J
ChemRevue Canadienne De Chimie 1992, 70:560
571.
104. Chiodo S, Russo N, Sicilia E. Newly developed basis sets for density functional calculations. J Comput
Chem 2005, 26:175183.
105. Porezag D, Pederson MR. Optimization of Gaussian
basis sets for density-functional calculations. Phys
Rev A 1999, 60:28402847.

c 2012 John Wiley & Sons, Ltd.



Volume 00, January/February 2012

WIREs Computational Molecular Science

Atomic orbital basis sets

106. Veillard A. Gaussian basis set for molecular wavefunctions containing second-row atoms. Theor Chim
Acta 1968, 12:405411.
107. Jensen F. Contracted basis sets for density functional
calculations: segmented versus general contraction. J
Chem Phys 2005, 122:074111.
108. Almlof J, Taylor PR. Atomic natural orbital (ANO)
basis-sets for quantum-chemical calculations. Adv
Quantum Chem 1991, 22:301373.
109. Klopper W, Kutzelnigg W. Gaussian-basis sets and
the nuclear cusp problem. Theochem: J Mol Struct
1986, 135:339356.
110. Kutzelnigg W. Theory of the expansion of wavefunctions in a Gaussian-basis. Int J Quantum Chem
1994, 51:447463.
111. McKemmish LK, Gill PMW. Gaussian expansions of
orbitals. J Chem Theory Comput 2012.
112. Jensen F. The basis set convergence of the Hartree
Fock energy for H-3(+), Li-2 and N-2. Theor Chem
Acc 2000, 104:484490.
113. Christensen KA, Jensen F. The basis set convergence
of the density functional energy for H-2. Chem Phys
Lett 2000, 317:400403.
114. Kutzelnigg W, Morgan JD. Rates of convergence of
the partial-wave expansions of atomic correlation energies. J Chem Phys 1992, 96:44844508.
115. Kutzelnigg W. Correction. J Chem Phys 1992,
97:88218821.
116. Petersson GA, Bennett A, Tensfeldt TG, Allaham MA,
Shirley WA, Mantzaris J. A complete basis set model
chemistry. 1. The total energies of closed-shell atoms

Volume 00, January/February 2012

117.

118.

119.

120.

121.

122.

123.

124.

and hydrides of the 1st-row elements. J Chem Phys

1988, 89:21932218.
Klopper W, Manby FR, Ten-No S, Valeev EF. R12
methods in explicitly correlated molecular electronic
structure theory. Int Rev Phys Chem 2006, 25:427468.
Peterson KA, Adler TB, Werner H-J. Systematically
convergent basis sets for explicitly correlated wavefunctions: the atoms H, He, B-Ne, and Al-Ar. J Chem
Phys 2008, 128:084102.
Hill JG, Peterson KA. Correlation consistent basis sets
for explicitly correlated wavefunctions: valence and
core-valence basis sets for Li, Be, Na, and Mg. Phys
Chem Chem Phys 2010, 12:1046010468.
Feller D, Peterson KA, Hill JG. On the effectiveness
of CCSD(T) complete basis set extrapolations for atomization energies. J Chem Phys 2011, 135:044102.
Curtiss LA, Redfern PC, Raghavachari K. Gn theory.
Wiley Interdiscip Rev: Comput Mol Sci 2011, 1:810
825.
Karton A, Rabinovich E, Martin JML, Ruscic B. W4
theory for computational thermochemistry: in pursuit of confident sub-kJ/mol predictions. J Chem Phys
2006, 125:144108.
DeYonker NJ, Cundari TR, Wilson AK. The correlation consistent composite approach (ccCA): an alternative to the Gaussian-n methods. J Chem Phys 2006,
124:114104.
Feller D, Peterson KA, Dixon DA. A survey of factors
contributing to accurate theoretical predictions of atomization energies and molecular structures. J Chem
Phys 2008, 129:204105.

c 2012 John Wiley & Sons, Ltd.



23