Basis Sets

Expressing a vector in terms of a basis

(3.5, 1.5)
v
j

i v = 3.5i + 1.5j
Basis Sets
• Any set of mathematical functions can be used as a basis
– How many functions should we use? Which functions should we use?
• The larger (i.e. the more components in) the basis set...
– The better the wavefunction can be described
• And the closer the energy converges towards the limit of that method
– The slower the calculation – N4 integrals (bottleneck)
• We would like to use as small a basis set as possible and still
describe the wavefunction well
– A good solution is to use functions that have shape similar to s, p, d and f orbitals
and are centered on each of the atoms
– Slater-Type Orbitals (STOs)
• Radial decay follows exp[-r]
• We would like to be able to calculate all of the integrals
efficiently
– Gaussian-Type Orbitals (GTOs) are similar to STOs but have a radial term
following exp[-r2]
• More efficient to calculate the integrals but have the wrong shape so...
– Replace each STO with a sum of 3 Gaussian-Type Orbitals (GTOs)
Radial decay of GTO vs STO

Image Credit: Essentials of Computational Chemistry, Chris Cramer, Wiley, 2nd Edn.
Case Study: H atom (1)
We know that when we solve the Schrödinger equation for
the H atom we get as possible wavefunctions:
• Y = 1s, 2s, 3s, 4s, etc., as well as the p and d…
functions etc.
• The lowest energy state is Y0 = |1s>, with E = -½ a.u.
• The first excited state is Y1 = |2s>
• Mathematically these functions (in a.u.) look like:

1/ 2
1/ 2
 1  r 2 1  
r

1s  2  e 2s    2  r e 2

 4  2  4 
Case Study: H atom (2)
Graphically the 1s and 2s orbitals look like.

0.6

Exact Solution to the Schrodinger Equation for the H atom

0.5

0.4

0.3 1s AO
2s AO
Y

0.2

0.1

0.0

-0.1
0 1 2 3 4 5
Distance / Angstroms
 STO-3G Basis Set

• A minimal basis set, i.e. it has one basis function per atomic orbital
– Example: for Li-H, there would be 6 basis functions in total
• 1s on H & 1s, 2s, 2px, 2py and 2pz on the Li

• Each basis function is a fixed sum of 3 Gaussian functions (primitive

gaussians) whose coefficients are optimized to match a STO function
N
 m   dmrgmr
r 1
Here N is termed the degree of contraction and gmr are the primitive
gaussians
• A minimal basis set is not sufficient to describe the wavefunction
– However, it may be useful to do a quick initial geometry optimization
 STO-3G for H

0.5 Exact and the Three Primitives in the STO-3G Basis Set for H

0.4

1s exact
d11g11
-3
Y / bohr

0.3 d12g12
d13g13

0.2

0.1

Largest a
0.0
0 1 2 3 4 5
Distance / Angstroms

Smallest a
These three primitives add together to give the contracted basis function
 STO-3G for H

0.6
Exact (1s AO) and STO-3G basis function for H

0.5

0.4
1s exact
-3
Y / bohr

1
0.3

0.2

0.1

0.0
0 1 2 3 4 5
Distance / Angstroms

• There is only 1 basis function for H.

• No flexibility at all in computing e = -0.4665819 a.u.
Each atomic orbital represents
a basis function and is constructed
from three GTO.
total for carbon: 5 basis functions,
15 GTOs
 STO-3G for H

• For our STO-3G on the H atom, we had no cmi, so nothing

could be varied here to obtain the lowest e possible.

• The e of the H atom with a STO-3G basis set is thus

completely fixed at:
e = -0.4665819 a.u. =12.697 eV.

• Compare with the exact result of 13.606 eV.

• This is an error of 87.7 kJ mol-1!

Introducing “Molecular” Orbitals
• By analogy with LCAO-MO, quantum chemical calculations construct
MO‟s via basis functions.
N basis
i  
m
cm  m
1
i

i is called an MO, even if the calculation is applied to an atom, in which

case they are in actual fact AO‟s.

cmi is called an MO coefficient for MO i, even thought the coefficient is

applied to basis function m.
Linear combination of atomic orbitals (LCAO)

• The LCAO approximation involves expressing (“expanding”) each

molecular orbital (ψ) as a sum of “basis set functions” (φx) centered
on each atom
ψ

φ2
φ3
φ1

H C N

Let‟s use this parabola for our basis set functions, φx

   cii  1.21  102  123

n
Bigger Basis Sets - Split-valence Basis Sets
• Substantial improvements can be made in computing energies and
wavefunctions by increasing the number of basis functions.
• Core orbitals are only weakly affected by binding, whereas valence
orbitals can vary widely, So we should enable additional flexibility for
representing valence orbitals
• The next step is to use the so-called “split-valence” basis set.
• In split valence basis sets we allow for more than one function per
valence AO.
• We may have 2 or 3 or 4 etc basis functions per valence atomic orbitals.
• Split-valence basis sets: 3-21G, 6-31G, 6-311G
– “3-21G” implies that each core orbital is represented by single basis
function (a sum of 3 GTOs as for STO-3G) but each valence orbital is
represented by two basis functions (the first a sum of 2 GTOs, the
other a single GTO)
Basis Set Terminology (1)

• 2 basis functions per valence AO is called a valence

double zeta basis set.
• 3 basis functions per valence AO is called a valence
triple zeta basis set.
• 4 basis functions per valence AO is called a valence
quadruple zeta basis set.
• May have 5, 6, or even higher numbers of basis
function per valence AO.
 Basis Set Terminology (2)
• Examples of valence double zeta basis sets are the 3-21G basis set
or the 6-31G basis set.
• An example of a valence triple zeta basis set is the 6-311G basis set.
• The above notation is attributed to Pople and co-workers.
STO-3G
• H and He – 1 basis function.
• Li – Ne – 1 for the core + 4 for the valence = 5
6-31G
• H and He – 2 basis functions.
• Li – Ne – 1 for the core + 8 for the valence = 9
6-311G
• H and He – 3 basis functions.
• Li – Ne – 1 for the core + 12 for the valence = 13
Examples for split-valence basis sets (1)
3-21G basis set: both core and valence atomic orbitals are
represented by three GTOs.

Total for carbon: 9 basic functions, 15 GTOs.

Split basis sets give more flexibility to the optimization.
Examples for split-valence basis sets (2)
6-31G basis set:
Core atomic orbitals are represented by six GTOs, and valence
atomic orbitals are represented by four GTOs.

Total for carbon: 9 basic functions, 22 GTOs.

Examples for split-valence basis sets (3)
6-311G basis set:
it is an example of a double split basis sets where the core atomic
orbitals are represented by six GTOs, and the valence atomic orbitals
are represented by five GTOs.

Total for carbon: 13 basic functions, 26 GTOs.

3-21G for H atom (1)
• In general, molecular orbitals cannot be described just in terms of
the atomic orbitals of the atoms
– e.g. A Hartree-Fock calculation for NH3 with an infinite basis set
just consisting of s and p functions predicts that the planar
geometry is a minimum
• H has no core AO‟s, so there will be two s-type basis functions that
will be used to describe the 1s AO of H.
• We now have MO coefficients to vary.
• The 1s AO will be represented as a linear combination of the two s-
type basis functions.
• We will also get an “MO” for the 2s AO
• 1s = c1,1s1 + c2,1s2
• 2s = c1,2s1 + c2,2s2
3-21G for H atom (2)

m N dmr amr / bohr-2

1 2 0.156285 5.44718
0.904691 0.824547
2 1 1 0.183192

Exact and the Two Basis Functions in the 3-21G Basis Set for H
0.8

0.6
1s exact
1
-3
Y / bohr

2

0.4

0.2

0.0
0 1 2 3 4 5
Distance / Angstroms
3-21G for H atom (3)
• After minimizing the value of e we obtain
e = -0.4961986 H = 13.503 eV.
• c1,1s = 0.37341, c2,1s = 0.71732
• Error = 9.97 kJ mol-1, a much better result.

0.5 Exact (1s AO) and 3-21G 1s "MO" for H

0.4

1s exact
-3
Y / bohr

0.3 1

0.2

0.1

0.0
0 1 2 3 4 5
Distance / Angstroms
3-21G for H atom (4)

• We also obtain a solution for the 2s AO of H.

• c1,2s = 1.25554, c2,2s = -1.09602

0.8 Exact (2s AO) and 3-21G 2s "MO" for H

0.6
2s exact
-3

2
Y / bohr

0.4

0.2

0.0

0 1 2 3 4 5
Distance / Angstroms
Increasing the Basis Set

• The table below summarizes the results for increasing the

number of s basis functions from 1 (minimal) through 6.

Basis Set No. s e/H Error /

functions kJ mol-1
STO-3G 1 -0.4665819 87.718
3-21G 2 -0.4961986 9.978
6-311G 3 -0.4998098 0.499
cc-pVQZ 4 -0.4999455 0.143
cc-pV5Z 5 -0.4999945 0.014
cc-pV6Z 6 -0.4999992 0.002
Chemical Bonding
• When atoms bond together to form molecules, the electrons that make
up the system distribute themselves throughout space and between the
nuclei to produce the lowest possible overall energy of the system.
• Certain parts of space have higher densities of electrons, while others
contain very low densities.
• Basis sets, are functions, which constrain electron densities to certain
regions of space.
• In order to obtain the correct energy of the system, we require our basis
functions to correctly reflect the real electron density in our system.
• Thus our basis set should allow for as much flexibility as possible in
distributing our electrons around and between nuclei.
• At present, the best way of doing that is by varying MO coefficients.
• Because of this we often need quite a few, and a wide variety of, fixed
functions.
 More Flexibility (1)

• We can increase the number of functions of the same

angular type, e.g., more s functions.

• E.g. STO-3G → 3-21G → 6-311G …

• Adding more functions of the same l type (recall l =0

for s AO) will only allow for electrons to be further
“spread out”, or for placing more “nodes” in electron
density as we move away from the nucleus.
 More Flexibility (2)
• Here are the 6 s functions used in the cc-pV6Z basis for H atom.
• Electron density is permitted to be more spread out, but is
spherically symmetric.
• There is never any special direction iN space that electrons
prefer to be concentrated.
1.0

6 s functions from the cc-pV6Z basis for H

0.8 1
2
0.6
3
4
-3
Y / bohr

5
0.4 6

0.2

0.0
0 1 2 3 4 5
Distance / Angstroms
Introducing Polarization

• We can increase the number of functions of the same

angular type, e.g., more s functions.
• Adding more functions of the same l type will only allow for
electrons to be further “spread out” or more nodes to exist.
• However, it does not allow for a different directional distribution
of electron density than what we already have.
• We can also include higher angular types of basis
functions.
• This does allow for different preferred directions in space for
electrons to wander around in.
• For H this would mean allowing p-type functions and
also d-types, etc., to partake in bonding.
• For Li – Ar this would mean including d-type and also f-
type etc.
Case Study: H2

• Comparing the 6-311G basis with and without polarization

functions (p functions) on each H atom in H2, we obtain the
following MO coefficients.

Basis 6-311G 6-311G+p-

• Each H atom has
Function type
s1 (H1) 0.19310 0.18612 directed some electron
s2 (H1) 0.29490 0.28748
s3 (H1) 0.12566 0.13260 density specifically
pz (H1) - -0.02283 toward the other H atom.
s1 (H2) 0.19310 0.18612
s2 (H2) 0.29490 0.28748 • Each H atom has been
s3 (H2) 0.12566 0.13260
pz (H2) - 0.02283
“polarized”.
E / a.u. -1.128038 -1.132491
Basis Set Terminology (3)
• Polarization functions are often added separately to atoms other
than H and He (atoms other than H and He are termed “heavy”
atoms).
• Adding 1 set of polarization functions to heavy atoms is designated
by a “*” or (d) after the basis set designation.
• Adding 1 set of polarization functions to H and He is designated by a
second “*” or a by (d,p) after the basis set designation.
• E.g 3-21G* adds a set of d-type functions to all heavy atoms in the
molecule.
• E.g. 3-21G** adds a set of d-type functions to all heavy atoms in the
molecule and a set of p-type functions to all H and He atoms in the
molecule.
• 3-21G(d) is synonymous to 3-21G* and 3-21G(d,p) is synonymous
to 3-21G**
• Adding two sets of d-type functions to heavies is denoted by (2d).
• Adding two sets of d-type functions and a set of f-type functions to
heavies, and two sets of p-type and a set of d-type to H and He is
designated by (2df,2pd), etc.
 Diffuse Basis Functions

• If the problem at hand suggests that electron density

might be found a long way from the nuclei, then, so-
called “diffuse” functions can be added.
• Computing anions is an example were diffuse functions
are necessary.
• Diffuse functions are of the same type as valence
functions (s and p‟s for Li – Ar, or just s for H and He).
• Diffuse functions are characterized by small basis set
exponents, i.e., small values for the a.
Diffuse Basis Functions

diffuse basis functions: 6-31+G(d), 6-311++G(3df,2pd)

Highest energy MOs of anions and highly excited electronic states

tend to be very diffuse, to be correctly described diffuse basis
functions are added

A single “+” indicates that heavy atoms have been augmented with
an additional diffuse s and a set of diffuse p basis functions; another
“+” indicates that hydrogen atoms have also been augmented
 Basis sets from other workers
• A superb set of basis functions originates from Dunning and co-workers.
• These authors use a very simple designation scheme.
• The basis sets are designated as either: cc-pVXZ or aug-cc-pVXZ.
• The ‘cc’ means “correlation consistent”.
• The ‘p’ means “polarization functions added”.
• The ‘aug’ means “augmented”, with the functions actually added being
essentially diffuse functions.
• The ‘VXZ’ means “valence-X-zeta” where X could be any one of the
following
• „D‟ for “double”, „T‟ for “triple”, Q for “quadruple”, or 5 or 6, etc.
• Determining the number of basis functions is done by considering the
valence space and placing X functions down for each valence AO with
the largest value of l.
• We then take one less function as we go up in the l quantum number,
and take an extra function as we go down in l quantum number.
• If the basis set is an „aug‟ type, then we add one function across the
board for each l-type function we have.
Examples of Dunning’s cc Basis Sets
• cc-pVDZ for Li – Ne
• We will have [3s2p1d], which is 3 + 2 x 3 + 1 x 5 = 14 basis
functions per atom in this row.
• Each H and He will have [2sp] = 5 functions.
• aug-cc-pVDZ for Li - Ne
• We will have [4s3p2d], which is 4 + 3 x 3 + 2 x 5 = 23 basis
functions per atom in this row.
• Each H and He will have [3s2p] = 9 functions.
• cc-pV5Z for Li – Ne
• We will have [6s5p4d3f2gh], which is 6 + 5 x 3 + 4 x 5 + 3 x 7 + 2
x 9 + 1 x 11 = 91 basis functions per atom in this row!
• Each H and He will have [5s4p3d2fg] = 55 functions.
Effective Core Potentials
• Normally applied to third and higher row elements.
• A potential replaces the core electrons in a calculation with
an effective potential.
• Eliminates the need for core basis functions, which
usually require a large number of primitives to describe
them.
• May be used to represent relativistic effects, which are
largely confined to the core.
• Some examples are:
CEP-4G, CEP-31G, CEP-121G, LANL2MB
(STO-3G 1st row),
• LANL2DZ (D95V 1st row), SHC (D95V 1st row)
Basis Set Quality
• ECP minimal basis sets are clearly the worst quality,
followed closely by minimal basis sets.
• DZ basis sets are a marked improvement, but still
generally of low quality.
• 6-311G
• 6-311G(2df,2pd) ~ cc-pVTZ
• 6-311++G(2df,2pd)
• aug-cc-pVTZ
• Bigger Dunning‟s basis sets now win hands-down.
• A simple comparison can be made by comparing the
number of s, p, d, f, etc. functions between basis sets.
6-311++G(2df,2pd) basis set for benzene
• For C
• A single contracted GTO of degree 6 to mimic the 1s core AO.
• Three functions per valence AO, the first will be a contracted GTO of degree 3,
and the remaining two will be made up of a single gaussian each.
• A set of diffuse functions will be added, i.e., a single diffuse s and a diffuse px, py
and pz.
• Two sets of d polarization functions will be added.
• A single set of f functions will be added.
• No. basis functions = 1 for the core + 4 valence AO x 3 functions for the „311‟
part + 4 diffuse + 5 d AO x 2 + 7 f AO = 34.
• For H
• Three functions for the 1s AO, the first being a contracted GTO of degree 3, and
the remaining two are simple primitives.
• A diffuse s function added to them.
• Two sets of p polarization functions added.
• A single set of d polarization functions added.
• No. basis functions = 1 valence AO x 3 functions for the „311‟ part + 1 diffuse + 3
p AO x 2 + 5 d AO = 15

• For C6H6 we will therefore require a total of 34 x 6 +15 x 6 = 294 basis

functions.