Application and development of

quantum chemical methods. Density

functional theory and valence bond
theory

Fuming Ying

Theoretical Chemistry
School of Biotechnology
Royal Institute of Technology
Stockholm 2010
Copyright c 2010 by Fuming Ying
TRITA-BIO Report 2010:13
ISSN 1654-2312
ISBN 978-91-7415-633-1
 3

Printed by US-AB, Stockholm 2010

 i

Abstract
This thesis deals with two subdiciplines of quantum chemistry. One is the
most used electronic structure method today, density functional theory (DFT),
and the other one of the least used electronic structure methods, valence bond
theory (VB). The work on DFT is based on previous developments in the group
in density functional response theory and involves studies of hyperfine cou-
pling constants which are measured in electron paramagnetic resonance exper-
iments. The method employed is a combination of a restricted-unrestricted ap-
proaches which allows for adequate description of spin polarization without
spin contamination, and spin-orbit corrections to account for heavy atom ef-
fects using degenerate perturbation theory. The work on valence bond theory is
a new theoretical approach to higher-order derivatives. The orbital derivatives
are complicated by the fact that the wave functions are constructed from de-
terminants of non-orthogonal orbitals. An approach based on non-orthogonal
second-quantization in bi-orthogonal basis sets leads to straightforward deriva-
tions without explicit references to overlap matrices. These formulas are rele-
vant for future applications in time-dependent valence bond theory.
 ii

List of publications

Paper I. Restricted-unrestricted density functional theory for hyperfine coupling con-

stants: vanadium complexes Xing Chen, Fuming Ying, Zilvinas Rinkevicius, Olav
Vahtras, Hans Ågren, Wei Wu and Zexing Cao. submitted

Paper II. An analytical valence bond Hessian, by means of second quantization in biorthonor-
mal basis sets Fuming Ying, Olav Vahtras, Zilvinas Rinkevicius and Wei Wu.
Manuscript
 iii

My contribution

• I carried out calculations in paper I and participated in analysis and au-

thoring

• I derived the Hessian formulas in paper II and participated in authoring

• I participated in the preparations of the manuscripts for all papers

iv

Acknowledgments

I would thank Prof. Yi Luo, Hans Ågren and Prof Wei Wu in Xiamen Uni-
versity who gave me the opportunity to study in KTH.

I appreciate my supervisor Olav Vahtras and my tutor Zilvinas Rinkevicious

for their guidance, advice in each project. I still remember the situation when
Zilvinas and I discussed the details of DALTON program, just as happened yes-
terday.

I should also give my gratitude to my colaborator Xing Chen for her help
and discussion during the work.

I may also say ”Thank you” to Prof. Magareta Bromberg, Aato Laaksonen
and Alexander Lyubartsev in SU for the courses and Prof. Poul Jøgensen, Jeppe
Olsen and Trygve helgaker for their teaching Quantum Chemistry in the sum-
mer school.

Thanks to my friends in Department of Theoretical Chemistry, Xiaofei Li,

Keyan Lian, Hao Ren, et al. for their kind help in my life in Stockholm.
Contents

1 Introduction 1

2 Density functional theory 3

2.1 Density optimization . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Variational perturbation theory . . . . . . . . . . . . . . . . . . . . 4
2.2.1 The restricted-unrestricted method . . . . . . . . . . . . . . 5

3 Valence bond theory 7

3.1 VB Wave Function and Hamiltonian . . . . . . . . . . . . . . . . . 8
3.1.1 VB Wave Function . . . . . . . . . . . . . . . . . . . . . . . 8
3.1.2 Constructing VB Hamiltonian with Determinant Expansion 9
3.2 Different types of VB Orbitals . . . . . . . . . . . . . . . . . . . . . 9
3.2.1 Localized VB Orbitals . . . . . . . . . . . . . . . . . . . . . 9
3.2.2 Semi-Localized VB Orbitals . . . . . . . . . . . . . . . . . . 10
3.3 VB Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3.1 Generalized VB . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3.2 CASVB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3.3 VB Self-Consistent Field . . . . . . . . . . . . . . . . . . . . 12
3.3.4 Breathing Orbital VB . . . . . . . . . . . . . . . . . . . . . . 13
3.3.5 VB Configuration-Interaction . . . . . . . . . . . . . . . . . 14
3.3.6 VBPCM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.4 Various VB Programs . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.4.1 TURTLE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.4.2 VB2000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.4.3 XMVB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

4 Summary of Included Papers 17

4.1 Restricted-unrestricted density functional theory . . . . . . . . . . 17
4.2 The Valence bond Hessian . . . . . . . . . . . . . . . . . . . . . . . 18

v
Chapter 1

Introduction

The work of this thesis deals with two methods of electronic structure theory.
The first, density functional theory (DFT), is by far the most important method
today with its competitive combination of speed and accuracy. The number
of articles based on DFT increase with an exponential rate making a complete
overview if not impossible but certainly beyond the scope of this thesis. The
second, valence bond (VB) theory, was the most important theory in the infancy
of this science.
In chapter 2 we provide the theoretical background of DFT for the appli-
cations used in paper I. It is customary to introduce an expose of DFT to two
pioneering articles, the first by Hohenberg and Kohn[1] who established the
mathematical foundation of a ground state theory based on the electron density
as a fundamental variable, and the other by Kohn and Sham[2] who introduced
a practical single-determinant theory for practical calculations. It was because
of this work, which started a tremendous development in computational chem-
istry, that Walther Kohn was awarded the Nobel Prize in Chemistry in 1998.
In chapter 3 we give a background of valence bond theory. It dates back to
G. N. Lewis[3] who introduced concepts such as electron pair bonding and the
octet rule, which has influenced the way of thinking for generations of chemists.
The theory which gradually lost importance when the rivaling molecular orbital
(MO) theory grew in popularity. The latter had its origin in spectroscopy and
allowed for efficient computer implementations with predictive power. Recent
decades have witnessed a revival of valence bond theory, with new algorithms
and new computer implementation. Paper II. is a contribution to this develop-
ment.

1
2 CHAPTER 1. INTRODUCTION
Chapter 2

Density functional theory

The implications of the fundamental theorems of density functional theory are

enormous[1]. The ground state energy of a many-electron system is a unique
functional of the electron density and that the part of the functional which in-
dependent of the external field, the Hohenberg-Kohn functional, is a universal
functional independent of the number of electrons. This implies that any many-
electron problem can be reduced to an equation in three variables, the coordi-
nates of a point in space. The problem is that the equation, or the functional, is
not known which has prompted a huge development of approximate function-
als which have been parameterized to reproduce a certain class of experimental
results. The success of DFT has been due to basis set expansions and the formu-
lation of DFT as a single determinant theory[2]. In practice DFT can be viewed
as a Hartree-Fock theory with a semi-empirical element which includes electron
correlation.
In this chapter we review the aspects of DFT that are relevant for paper I;
the optimization procedure, the restricted-unrestricted method and perturba-
tion theory with application to hyperfine coupling tensors.

2.1 Density optimization

The density is optimized for a given energy functional E[ρ].

δE[ρ] = 0 (2.1)

In the most general case the optimization is carried out with the auxiliary con-
dition that the total number of particles is constant
Z
dV ρ(~r) = N (2.2)

which leads to a Lagrangian multiplier which is interpreted as the chemical

potential[1]. If the density is constructed from a determinantal wave function

3
4 CHAPTER 2. DENSITY FUNCTIONAL THEORY

the condition 2.2 is automatically fulfilled. The variation of a functional (func-

tion of a function) gives
Z
δE
δE[ρ(~r)] = dV δρ(~r) (2.3)
δρ(~r)

In practical calculations the density is parameterized for some finite set of pa-
rameters λ = (λ1 , λ2 . . .) such that

∂ρλ (~r)
δρλ (~r) = δλ (2.4)
∂λ
Salek et al.[4] formulated the optimization in second quantization as follows;
the density is the expectation value of a density operator

ρλ (~r) = hλ|ρ̂(~r)|λi (2.5)

where the density operator projected on a basis set {φp } is

X
ρ̂(~r) = φp (~r)∗ φq (~r)a†p aq (2.6)
pq

When the parameterization is of exponential unitary form,

†
|λi = e−λpq ap aq |0i (2.7)

i.e. the with parameters describing a change from a reference set of orbitals, the
transformation is expanded around λ = 0 and the density variation a becomes
a commutator
∂ρ
= hλ|[a†p aq , ρ̂(~r)|λi (2.8)
∂λpq
Defining a Fock operator Z
δE
F̂ = dV ρ̂(~r) (2.9)
δρ(~r)
the energy optimization reduces to

h0|[a†p aq , F̂ ]|0i = 0 (2.10)

which are equivalent to the Kohn-Sham equations[2].

2.2 Variational perturbation theory

If a perturbation V is introduced in the Hamiltonian with a field strength x,

H = H0 + xV (2.11)
2.2. VARIATIONAL PERTURBATION THEORY 5

the optimized parameters will be a function of the field strength λ(x). The en-
ergy change due to this perturbation can be identified from the power expan-
sion
dE
E[ρλ(x) ; x] = E[ρλ(0) ; 0] + x (2.12)
dx x=0
λ=λ(0)

and in a strictly variational theory we have

dE
= h0|V |0i (2.13)
dx x=0
λ=λ(0)

2.2.1 The restricted-unrestricted method

The restricted-unrestricted method were developed by Fernandez et al.[5] and
later generalized for DFT by Rinkevicius et al.[6] for describing spin polariza-
tion in spin-restricted methods. In spin-restricted methods, the alpha- and beta
spin-orbitals always have the same spatial parts. For the calculation of a prop-
erty such as the hyperfine coupling constant whose main component, the Fermi
contact contribution, is a measure of the spin density at the nucleus this is prob-
lematic. An approximate wave function models such as MCSCF does not have
sufficient flexibility at the nucleus because the core orbitals are normally inac-
tive, and to allow for correlation in the core orbitals is computationally very
expensive. On the other hand an unrestricted which does have this flexibility
have other problems such as spin contamination in the wave function
The restricted-unrestricted theory was developed to avoid the problems with
spin contamination by optimizing the wave function in a restricted framework
(singlet optimization), but by including extra parameters (triplet optimization)
in the perturbation calculation. What this means is that the gradient with re-
spect to some of the parameters is non-zero

∂E
= τ 6= 0 (2.14)
∂λ
but we introduce Lagrangian multipliers so that the same relation holds in the
presence of the perturbation, i.e. for all x. A direct way of obtaining the final
relations is to differentiate Eq. (2.14) we have

d ∂E ∂ 2E ∂ 2 E ∂λ
= + =0 (2.15)
dx ∂λ ∂λ∂x ∂λ2 ∂x
which defined a linear system of equations for the wave function response. The
energy change then consists of two terms; the first as an expectation value of
the perturbation and the second from the response in the wave function

dE ∂E ∂E ∂λ
= + (2.16)
dx ∂x ∂λ ∂x
6 CHAPTER 2. DENSITY FUNCTIONAL THEORY

The combination of Eqs.(2.15,2.16) gives

−1
∂ 2E ∂ 2E

dE ∂E ∂E
= − (2.17)
dx ∂x ∂λ ∂λ2 ∂λ∂x
Chapter 3

Valence bond theory

Valence bond(VB) theory is an ancient as well as modern theory in quantum

chemistry. It was first introduced by Heitler and London[10] in 1927. Later
on, Slater[11] and Pauling[12] developed and added several concepts into that
treatment in order to generalize it for polyatomic molecules. The theory that
was developed has the name ”valence bond theory”.
The concept of electron-pairing by Lewis was used in dealing with the H2
molecule which made VB very easy to understand by chemists. To explain
the geometry (bonding character) in polyatomic systems, another concept ”hy-
bridization” and the corresponding ”hybrid orbital” were introduced by Paul-
ing. A hybrid orbital is a kind of atomic orbital (HAO) which can be obtained by
combining original atomic orbitals to get the maximum bonding ability. Thus
the original atomic orbitals can be treated as a special type of HAOs. In the
following, we will use HAO to denote all kinds of atomic orbitals. Perhaps the
most important concept in VB was resonance. It means that the ”real” electronic
structure of a molecule can be explained as the linear combination between a set
of ”extreme” structures. For the sake of resonance, VB is a multi-configurational
method. Some molecular properties like conjugation can be easily explained by
resonance. A derivative of resonance is the resonance energy. It is defined as
the difference between the energies of ”real” structure and the most stable struc-
ture(s) and can be a measure of aromaticity.

A severe technical disadvantage was discovered soon. The use of non-orthogonal

local bonding orbitals made quantitative calculations impossible at that time so
that a lot of simplification had to be made. Some of them may lead to wrong
conclusions. On the other hand, molecular orbital(MO) theory made implemen-
tation and calculation much easier due to its delocalized orthogonal orbitals.
From the late of 1960s, there was a log of development and implementations in
MO theory, while the role of VB diminished.

Despite of such a disadvantage, VB was always preferred by chemists in

qualitative analyses. Furthermore, the increasing of computer power made it

7
8 CHAPTER 3. VALENCE BOND THEORY

possible to perform ab initio VB calculations. Thus, from the late of 1970s, VB

was reborn from the ashes.
VB developments, i.e. generalized Slater-Condon rules[13], spin-free VB
method[14, 15] and the paired-permanent-determinant[16, 17] approach, have
been published to reduce computational costs so that pure ab initio VB may be
performed faster and be a practical method. Because of these developments,
the cost of ab initio VB calculation now can be reduced to O(N4 ) where N is the
number of electrons. Furthermore, a lot of VB methods have been developed to
handle different situations, including electron correlation, but which maintains
the main features and analytical power of VB theory. [18, 19, 20, 21, 22, 23, 24,
25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36].

To display concept of resonance, tools like VB wave function, VB orbitals,

and VB computational methods are essential. In the following sections, such
tools will be revealed.

3.1 VB Wave Function and Hamiltonian

3.1.1 VB Wave Function
A VB wave function with Rumer spin function stands for a VB structure. It is
called HLSP(Heitler-London-Slater-Pauling) wave function. It can be expressed
as
ΦK = ÂΩK ΘK (3.1)
where  is the antisymmetric operator and ΩK and ΘK are the spacial and spin
function respectively. Spacial function has the form of the multiplication of a
set of orbitals
ΩK = φ1 (1)φ2 (2) · · · φN (N ) (3.2)
and spin function has the form
Y 1 Y
ΘK = √ [α(i)β(j) − α(j)β(i)] α(l) (3.3)
i,j
2 l

where i, j run over all paired electrons and l runs over all unpaired electrons.
Real wave function can be expressed as the linear combination of a set of
HLSP wave function X
ΨV B = CK ΦK (3.4)
K

and weight of ΦK can be obtained by Coulson-Chirgwin’s formula[37]

X
WK = CK CL MKL (3.5)
L
3.2. DIFFERENT TYPES OF VB ORBITALS 9

where CK is the combination coefficient of HLSP wave function ΦK and MKL is

the overlap between ΦK and ΦL . The linear independent HLSP that are used to
build real wave function can be obtained with Rumer rules or standard Young
tableau.

3.1.2 Constructing VB Hamiltonian with Determinant Expan-

sion
A VB structure in the form of Eqn (3.1) can be expanded into a set of determi-
nants
X2m
ΦK = CiK DiK (3.6)
i=1

where m is the number of covalent bonds in the VB structure ,DiK is the deter-
minant belonging to the structure and CiK is the corresponding coefficient. For
example, normalized wave function of covalent bond A-B can be expressed as
the determinant expansion

1
Φcov = √ |φa φb | − |φa φb |


(3.7)
2
The Hamiltonian of VB wave function is constructed in terms of VB struc-
tures. Thus, Hamiltonian matrix elements hΦK |Ĥ|ΦL i can be expanded into a
set of Hamiltonian elements hDiK |Ĥ|DjL i
2m2 n
X X
hΦK |Ĥ|ΦL i = CiK CjL hDiK |Ĥ|DjL i (3.8)
i=1 j=1

and overlap can be generated in the same way.

3.2 Different types of VB Orbitals

VB orbitals are essential to construct Hamiltonian and describe VB wave func-
tions. Although all ab initio VB methods deal with nonorthogonal orbitals, the
degree of localization various from on the another. There are 2 kinds of VB or-
bitals in modern VB theory, named localized and semi-localized orbitals. Both
are used for different purposes.

3.2.1 Localized VB Orbitals

A localized VB orbital is simply an HAOs. ”Pure” VB methods are based on
such kind of VB orbital. In a system with one single bond A-B, the VB wave
10 CHAPTER 3. VALENCE BOND THEORY

function for this bond can be expressed as

Ψ = c1 |φa φb | − |φa φb | + c2 |φa φa | + c3 |φb φb | (3.9)

where ci is the coefficient of ith VB structure and φa and φb are HAOs located on
atom A and B respectively. The first term corresponds to a covalent structure of
A-B bond, and the last two terms correspond to ionic terms. The advantage of
localized VB orbital is that a clear distinction between covalent and ionic struc-
ture of a bond can be provided.

A bond in a polyatomic molecule, i.e. one C-H bond in methane molecule

CH4 , may be dealt in the following way: the localized orbital on H keeps the
type of HAO and the orbital on C becomes fragment orbital(FO) which means
that orbital on C bonding with H may have a slight delocalization to the other
H atoms of the remaining CH3 group.

3.2.2 Semi-Localized VB Orbitals

When a many bonds need to be considered, the number of VB structures in
terms of local VB orbitals increases exponentially and becomes quite hard to re-
solve. Hence, some simplification is needed to reduce the number of VB struc-
tures. Two types of semi-localized orbitals are used in VB calculations. The
general form was suggested by Coulson and Fischer[38] as

ψa = φa + λφb (3.10)

and
ψ b = φ b + λ′ φ a (3.11)
to represent the A-B bond where λ and λ′ are coefficients and can be a measure-
ment of delocalization. In this case, wave function can be expressed as

Ψ = |ψa ψb | − |ψa ψb |
(3.12)
= (1 + λλ′ ) |φa φb | − |φa φb | + λ′ |φa phia | + λ|φb φb |

Thus, a VB wave function with 3 VB structures can be reduced to a wave func-

tion in covalent form containing covalent and ionic structures.
A VB orbital that is allowed to delocalize freely to all other parts of the
molecule if no other restriction is added onto it, is called overlap-enhanced or-
bital(OEO). The wave function can be constructed with several VB structures(in
form of (3.12), known as perfect-pairing).

If some restriction is added to the orbital that a VB orbital can be only de-
localized onto the atom with which the orbital is bonded in the VB structure
in consideration, OEO becomes bond-distorted orbital(BDO). For instance, A
3.3. VB METHODS 11

semi-localized orbital for C-C bond in H3 C-CH3 with OEO can be an orbital on
one CH3 group and delocalizing to whole part of the other CH3 group, while a
BDO is an orbital that can only delocalize to the other C atom.

The advantage of semi-localized VB orbitals is that it may reduce a large

number of VB structures and then make the wave function much simpler. When
all orbitals become delocalized to the whole system, the wave function will be-
come an MO wave function. Semi-localized orbital make a link between VB
and MO theory. The disadvantage of semi-localized orbitals, as we have shown
above, is that the distinction of covalent and ionic pictures is impossible.

3.3 VB Methods
3.3.1 Generalized VB
GVB[18, 19, 20] was introduced by William A. Goddard III and was the first im-
portant case of Coulson-Fischer’s formula. The restricted form of GVB method
is referred as GVB-SOPP, which is often used by GVB. GVB-SOPP has two sim-
plifications. The first is so called ”perfect pairing”(PP) approximation that a
GVB wave function can be expressed as a single determinant with a set of ”gem-
inal” two-electron functions each corresponding to a chemical bond or a lone
pair
ΨGV B = | (|ϕ1a ϕ1b | − |ϕ1a ϕ1b |) (|ϕ2a ϕ2b | − |ϕ2a ϕ2b |) · · · (|ϕN a ϕN b | − |ϕN a ϕN b |)
(3.13)
th
where ϕia and ϕib are OEOs for i chemical bond or lone pair.

The second one in GVB is the strong orthogonality for computational con-
venience. In a strong orthogonality framework, the orbitals are nonorthogonal
if in the same geminal function
hϕia |ϕib i 6= 0 (3.14)
and orthogonal if they are in different geminal functions
hϕi |ϕj i = 0 (3.15)
GVB can also be represented in another form with natural orbitals as
|ϕia ϕib | − |ϕia ϕib | = Ci |φi φi | + Ci∗ |φ∗i φ∗i | (3.16)
where φi and φ∗i are natural orbitals respect to bonding and antibonding MOs
and Ci and Ci∗ are corresponding coefficients. The transformation between
forms shown in Eqn gvb) and 3.16 is
φi + λφ∗i φi − λφ∗i 2 Ci∗
ϕia = √ ϕib = √ λ =− (3.17)
1 + λ2 1 + λ2 Ci
12 CHAPTER 3. VALENCE BOND THEORY

3.3.2 CASVB
The underlying idea of CASVB is quite simple. A spin-coupled VB wave func-
tion should be close to CASSCF wave function with the same number of elec-
trons and the same active space. Thus, a transformation from CASSCF to VB
wave function is reasonable. Since CASSCF is easily obtained, this idea pro-
vides a way to generate a VB wave function.

Thorsteinsson and D.L. Cooper[21, 22, 23] suggested a way to generate the
VB wave function from CASSCF. After the transformation, the final VB wave
function is restrictly equivalent to starting CASSCF. The way can be expressed
as following formula

ΨCAS = SΨCASV B + (1 − S 2 )ΨCASV

⊥
B
(3.18)

where ΨCAS and ΨCASV B are original CASSCF and final CASVB wave functions
respectively, ΨCASV
⊥
B
is the perpendicular complement of CASVB wave function
and
S = hΨCAS |ΨCASV B i (3.19)
is the overlap between CASSCF and CASVB wave function. One way to get
CASVB wave function is to maximize the overlap S and another way is to mini-
mize the energy of ΨCASV B . It’s a simple way to implement but the orbitals will
more or less keep the MO type, similar to localized molecular orbitals(LMOs).

Different from Thorsteinsson’s method, Hirao[24, 25] developed another way

to get CASVB wave function with nonorthogonal VB orbitals. Since LMOs have
obvious delocalization tails and may enlarge the ionic weights and mix cova-
lent and ionic pictures, nonorthogonal orbitals will give us more reasonable VB
wave function and clearer resonance picture.

3.3.3 VB Self-Consistent Field

VB self-consistent Field(VBSCF) was first introduced by van Lenthe[26, 27, 28]
and has been one of the fundamental computational methods of VB theory.
Hamiltonian and overlap matrix are generated in terms of VB structures

HKL = hϕK |Ĥ|ϕL i (3.20)

MKL = hϕK |ϕL i (3.21)

After getting hamiltonian and overlap,

HC = EMC (3.22)
3.3. VB METHODS 13

we will get the coefficient CK of VB wave function ϕK and the energy E. Min-
imize the energy with optimizing orbital and CK simultaneously in a SCF pro-
cedure and we will get the ground state of the system.

VBSCF accepts VB orbitals in various types such as OEOs, BDOs and HAOs.
There is no constraint of orthogonality for orbitals and thus the clear picture of
bonding may be kept. Particularly, when a set of fully delocalized orthogonal
orbital is used with a wave function in single determinant, VBSCF will go to
the form of Hartree-Fock (HF) and have the same energy. The wave function
may differ from the HF wave function only by a unitary transformation without
changing the energy.

3.3.4 Breathing Orbital VB

Breathing orbital VB(BOVB) theory was developed by Sason Shaik and Wei
Wu[29, 30, 31, 32, 33]. In VBSCF, different VB structures share the same set
of VB orbitals. However, same orbital in different VB structures is treated as
”different” orbitals in BOVB and optimized separately. Thus, BOVB will gen-
erate more orbitals than VBSCF if there are more than one VB structure for the
system. If there is only one VB structure, nothing will be changed and BOVB is
equivalent to VBSCF.

C1 F F + C2 F F + C3 F F

C1 F F + C2 F F + C3 F F

Figure 3.1: Pictures of VBSCF(above) and BOVB(below) results

Fig 3.1 shows results of VBSCF and BOVB of F2 molecule with localized VB
orbitals. As we can see, different VB structures share the same set of VB orbitals
in VBSCF, no matter whether it is covalent or ionic structure. In BOVB, same
orbital is treated as different. orbitals in cation atom become more contracted
and orbitals in anion atom become more diffused. Thus, part of dynamical cor-
14 CHAPTER 3. VALENCE BOND THEORY

relation is included in BOVB result.

3.3.5 VB Configuration-Interaction
BOVB improves computational results because it provides some dynamical cor-
relation. The VB Configuration-Interaction(VBCI) method[34, 35] is a VB method
based on CI model aiming to provide better results and keep the clear picture of
VBSCF. In a VBCI calculation, the orbitals from a VBSCF are used but not furhter
optimized. Virtual orbitals and structures will be constructed to describe exci-
tations.

The VBCI procedure contains several steps. First, a VBSCF wave function
should be obtained by a VBSCF calculation. Although VBSCF orbitals can be
strictly localized, a unitary transformation is still allowed without changing the
energy. Thus, an extra Boys localization[39] is needed to gain a unique VBSCF
wave function for a zeroth-order wave function
X
Ψ0 = CK0 0
ΦK (3.23)
K

Second, VB virtual orbitals and VB virtual structures will be constructed.

Virtual orbitals {ϕia } are restricted to the same block of the corresponding occu-
pied orbital ϕa X
ϕia = χµ ciaµ (3.24)
µ

where ciaµ is the LCAO coefficient of basis function χµ in orbital ϕia . To fulfill the
condition that N basis functions can only construct N linearly independent or-
bitals, occupied orbitals are restricted in the different blocks, which means that
no basis function is shared in more than 1 VB orbital. Orbitals are orthogonal in
the same block, but not in the different blocks
hϕa |ϕia i = 0 (3.25)
hϕia |ϕja i = 0 (3.26)
hϕib |ϕja i 6= 0 (3.27)
VB virtual structures can be obtained by replacing some occupied orbital
with a virtual one, meaning the excitation. Excitation shall be done according
to the rule that one occupied orbital ϕa can be replaced only by the virtual ones
{ϕia } that in the same block. If only one electron is excited to the virtual orbital
in a structure, a VBCIS calculation will be obtained. If at most two electrons can
be excited, a VBCISD calculation will be obtained. A VBCI wave function then
is generated containing all VBSCF and excited structures.
X
ΨV BCI = CKV BCI V BCI
ΦK (3.28)
K
3.4. VARIOUS VB PROGRAMS 15

V BCI
where CK is coefficient of ΦVKBCI that should be solved later.

V BCI
Finally, a secular equation is solved to get the coefficients CK . The weights
of the VBCI structures are obtained in the same way as VBSCF.

3.3.6 VBPCM
All the above VB methods are applicable only in the gas phase and cannot be
used to solve chemical problems in solution, i.e. SN2 reactions. In order to have
a VB theory for problems in solution a combination with the Polarizable Con-
tinuum Model (PCM)[40] has been developed.

VBPCM[36] is the combined method of VB theory and the PCM model. In

the PCM model, the solvent is treated as a homogeneous medium characterized
by a dielectric constant and can be polarized by the charge of the solute. The
interaction between solute and solvent is taken as an effective potential in the
Hamiltonian, and optimized in a self-consistent reaction field procedure.

VBPCM is a method based on VBSCF. Thus, the representation of the VBPCM

wave function is the same as VBSCF. However, the Hamiltonian now is no
longer the same as the VBSCF Hamiltonian and added by a effective one-electron
Hamiltonian to describe the interaction between solvent and solute

H = H0 + V s (3.29)

where H0 is the Hamiltonian in gas phase for a bare molecule and Vs is the
extra potential in solution.

3.4 Various VB Programs

3.4.1 TURTLE
TURTLE(refs needed) is a VB software developed by van Lenthe et al. focusing
on ab initio multi-structure VB calculation. Currently, it is available for non-
orthogonal CI calculation and non-orthogonal MCSCF calculation with simul-
taneous optimization of VB coefficients and orbitals. Different types of VB or-
bitals, such as OEOs, BDOs and HAOs are allowed to construct VB structures
for VBSCF, BOVB, SCVB, BLW, etc. Optimization of wave function and geome-
try are both available with analytical gradients. The message passing interface
(MPI) has been used for the parallelization.
16 CHAPTER 3. VALENCE BOND THEORY

3.4.2 VB2000
VB2000 is a software based on ”Algebrant Algorithm”. Group function theory
is also implemented into VB2000 so that a large molecule can be described as
physically identifiable electron groups. Non-orthogonal CI, multi-structure VB,
GVB, SCVB and CASVB developed by Hirao.

3.4.3 XMVB
XMVB is developed by Prof. Wei Wu etc. based on spin-free VB theory and
conventional Slater determinant expansion methods. This program is enabled
for different kinds of VB theories such as VBSCF, BOVB, VBCI, VBPCM, GVB,
SCVB, BLW, etc. with OEOs, BDOs, or HAOs. VB orbitals and VB coefficients
can be optimized simultaneously. XMVB program can be obtained either as a
stand-alone program or a module in GAMESS-US. Parallel version of XMVB
with MPI is also available and can be obtained only as a stand-alone program.
Chapter 4

Summary of Included Papers

4.1 Restricted-unrestricted density functional theory

In this study we apply restricted-unrestricted density functional theory devel-

oped in our group[6] for the calculation of hyperfine coupling constants of vana-
dium complexes. These parameters measure the spin density at a given nucleus.
A spin restricted method has the advantage that it is free from spin contamina-
tion, but the disadvantage that it is not sufficiently flexible at the core to describe
the spin polarization that occurs in an electron paramagnetic resonance exper-
iment with magnetic nuclei. On the other hand an unrestricted method has
opposite features, it is flexible enough at the nucleus but spin contamination is
known to give poor results in special cases. The philosophy of the combination
of these methods, the restricted-unrestricted model, is to merge the best of both
worlds. The wave function is optimized with spin restricted method while the
spin-polarization is introduced as a perturbation. The price to pay is that the cal-
culations are more complicated as we need to solve response equations of one
order higher than would have been the case in a purely unrestricted scheme.
This idea was first developed and applied in multi-configuration self-consistent
field theory by Fernandez et al. [5]
Furthermore, spin-orbit effects were included using a degenerate perturba-
tion theory, developed in ref[9]. This gives a one-to-one relationship between
experimental parameters and calculated quantities at a level which is goes be-
yond the textbook formulations.
The results obtained indicate in general a good agreement with experiments
for vanadocenes(IV) complexes while for vanadium oxide complexes the devia-
tions are slightly higher. The accuracy of DFT calculations depend in general on
the choice of functional, here the BHLYP give the best results while the popular
B3LYP fail in the prediction of hyperfine coupling constants. This is believed to
be due to insufficient Hartree-Fock exchange.

17
18 CHAPTER 4. SUMMARY OF INCLUDED PAPERS

4.2 The Valence bond Hessian

The orbital derivatives of valence bond theory are complicated by the fact that
the wave function is constructed from determinants of non-orthogonal orbitals.
To date there is no computer implementation of an analytical valence bond Hes-
sian which is the key quantity necessary for applications in time-dependent
valence bond theory. In this paper outline a theoretical framework based on
second quantization in non-orthogonal orbitals and a theory of non-orthogonal
determinants that we credit to P-O Löwdin[41]
The main elements are the dual sets of orbitals bi-orthogonal to each other
which are connected by a transformation of the overlap matrix

a†p = Spq aq† (4.1)

where the notation with subscripts for one set and superscripts for the dual set
introduces tremendous simplifications. The algebraic rules of differentiation
and commutation allows one to identify the derivative of a determinant with
respect to a molecular orbital coefficient in terms of excitation operator
∂ † m
µ |Ψi = aµ a |Ψi (4.2)
∂Cm
This gives that the derivative of an inner product between two determinants
is an element of a transition density matrix, and all energy derivatives can be
expressed in terms of integrals and transition density matrices. It is shown that
given a Hamiltonian in an arbitrary basis
1
H = hpq ap† aq + gpqrs ap† ar† as aq (4.3)
2
(summation implied by repeated subscript-superscript pair) the derivative of a
Hamiltonian matrix element is given by
∂ 
mρ ξ m

µ hK|H|Li = 2hK|Li D(KL) F (KL) ξρ ∆(KL) µ + H(KL)D(KL) µ
∂Cm
(4.4)

where D(KL) is the intermediately normalized transition density matrix,

hK|ap† aq |Li
D(KL)pq = (4.5)
hK|Li
∆ given by
∆(KL)pm = δqp − D(KL)pq (4.6)
and F the Fock operator

F (KL)pq = hpq + (gpqrs − gpsrq )D(KL)rs (4.7)

4.2. THE VALENCE BOND HESSIAN 19

The relations thus obtained are transparent and without explicit reference to
overlap matrices which appear in great number using standard matrix notation.
For the gradient it can also be shown that the derivative reproduces the results
published previously by Song et al[42].
This approach is now easily generalized to Hessian evaluations, from iden-
tification of the equivalence between a second derivative and a two-particle ex-
citation
∂2 † † n m
µ ν |Ψi = aµ aν a a |Ψi (4.8)
∂Cm Cn
the second derivative of a Hamiltonian matrix element is
∂2 m ν
µ ν hK|H|Li = hK|H|LiD(KL) µ (D(KL) + D(LK))n
∂Cm Cn
+ [D(KL)F (KL)∆(KL))]m n
µ (D(KL) + D(LK))µ
(4.9)
+ D(KL)m n
µ (D(KL)F (KL)∆(KL))ν
+ D(KL)m n
µ (∆(KL)F (KL)D(KL))ν
+ g mµ nν − g mν nµ + gµ mnν + g nmµν

The non-Fock contributions are effectively integral transformations with transi-

tion density matrices, e.g.

g mµ nν = D(KL)mρ D(KL)nτ gξρστ ∆(KL)ξµ ∆(KL)σν (4.10)

This enables us to implement second-order methods, but most importantly the

Hessian is the key to time-dependent valence bond theory, which until now has
been a completely unexplored field of research.
20 CHAPTER 4. SUMMARY OF INCLUDED PAPERS
Bibliography

[1] P. Hohenberg and W. Kohn. Phys. Rev., 136:B864, 1964.

[2] W. Kohn and L. J. Sham. Phys. Rev., 140:A1133, 1965.

[3] G. N. Lewis. The atom and the molecule. J. Am. Chem. Soc., 38:762, 1916.

[4] Pawel Salek, Olav Vahtras, Trygve Helgaker, and Hans Ågren. Density-
functional theory of linear and nonlinear time-dependent molecular prop-
erties . J. Chem. Phys., 117:9630, 2002.

[5] Berta Fernandez, Poul Jørgensen, Jørgen Byberg, Jeppe Olsen, Trygve Hel-
gaker, and Hans Jørgen Aa. Jensen. J. Chem. Phys., 97:3412, 1992.

[6] Zilvinas Rinkevicius, Lyudmyla Telyatnyk, Olav Vahtras, and Hans Ågren.
Density functional theory for hyperfine coupling constants with the
restricted-unrestricted approach . J. Chem. Phys., 121:7614, 2004.

[7] P.-O. Löwdin and O. Goscinski. Int. J. Quant. Chem., 5:685, 1971.

[8] H. Bolvin. Chem. Phys. Chem., 7:1575, 2006.

[9] Z. Rinkevicius, K.J. de Almeida, C.I. Oprea, O. Vahtras, K. Ruud, and

H. Ågren. Degenerate perturbation theory for electronic g tensors:
leading-order relativistic effects . J. Chem. Theor. Comp., 4:1810, 2008.

[10] W. Heitler and F. London. Z. Phys., 44:455, 1927.

[11] J. C. Slater. Phys. Rev., 35:509, 1930.

[12] L. Pauling. The Nature of the Chemical Bond, 3rd ed. 1960.

[13] Jacob Verbeek and Joop H. Van Lenthe. The generalized Slater-Condon
rules. Int. J. Quant. Chem., 40(2):201–210, 1991.

[14] Roy McWeeny. A spin-free form of valence bond theory. Int. J. Quant.
Chem., 34(1):25–36, 1988.

21
22 BIBLIOGRAPHY

[15] Xiangzhu Li and Qianer Zhang. Bonded tableau unitary group approach to
the many-electron correlation problem. Int. J. Quant. Chem., 36(5):599–632,
1989.

[16] Wei Wu, Anan Wu, Yirong Mo, and Qianer Zhang. Paired-permanent ap-
proach to valence bond theory. Sci. Chin. B., 39(5):456, 1996.

[17] Wei Wu, Anan Wu, Yirong Mo, Menghai Lin, and Qianer Zhang. Efficient
algorithm for the spin-free valence bond theory. i. new strategy and pri-
mary expressions. Int. J. Quant. Chem., 67(5):287–297, 1998.

[18] W. A. Gorddard III. Phys. Rev., 157:81, 1967.

[19] R. C. Ladner and W. A. Goddard III. J. Chem. Phys., 51:1073, 1969.

[20] W. A. Goddard III., T. H. Dunning Jr., W. J. Hunt, and P. J. Hay. Acc. Chem.
Res., 6:368, 1973.

[21] Thorstein Thorsteinsson, David L. Cooper, Joseph Gerratt, Peter B.

Karadakov, and Mario Raimondi. Modern valence bond representations
of CASSCF wavefunctions. Theor. Chem. Acc., 93(6):343–366, 1996.

[22] David L. Cooper, Thorstein Thorsteinsson, and Joseph Gerratt. Modern

vb representations of casscf wave functions and the fully-variational op-
timization of modern vb wave functions using the casvb strategy. Adv.
Quant. Chem., 32:51 – 67, 1998.

[23] Thorstein Thorsteinsson and David L. Cooper. Exact transformations of CI

spaces, VB representations of CASSCF wavefunctions and the optimiza-
tion of VB wavefunctions. Theor. Chem. Acc., 94(4):233–245, 1996.

[24] Kimihiko Hirao, Haruyuki Nakano, Kenichi Nakayama, and Michel

Dupuis. A complete active space valence bond (casvb) method. J. Chem.
Phys., 105(20):9227–9239, 1996.

[25] Kenichi Nakayama, Haruyuki Nakano, and Kimihiko Hirao. Int. J. Quant.
Chem., 66(2):157–175, 1998.

[26] J. H. Van Lenthe and G. G. Balint-Kurti. The valence-bond SCF (VB SCF)
method. : Synopsis of theory and test calculation of oh potential energy
curve. Chem. Phys. Lett., 76(1):138, 1980.

[27] J. H. van Lenthe and G. G. Balint-Kurti. The valence-bond self-consistent

field method (VB–SCF): theory and test calculations. J. Chem. Phys.,
78(9):5699, 1983.

[28] Jacob Verbeek and Joop H. Van Lenthe. On the evaluation of non-
orthogonal matrix elements. J. Mol. Struct.(THEOCHEM), 229:115, 1991.
BIBLIOGRAPHY 23

[29] P. C. Hiberty, J. P. Flament, and E. Noizet. Compact and accurate valence

bond functions with different orbitals for different configurations: applica-
tion to the two-configuration description of f2. Chem. Phys. Lett., 189(3):259
– 265, 1992.

[30] P. C. Hiberty, S. humbel, J.H. van Lanthe, and C.P. Byrman. J. Chem. Phys.,
101:5969, 1994.

[31] Philippe C. Hiberty and Sason Shaik. Breathing-orbital valence bond

method - a modern valence bond method that includes dynamic correla-
tion. Theor. Chem. Acc., 108(5):255–272, 2002.

[32] Lingchun Song, Wei Wu, Philippe C. Hiberty, David Danovich, and Sason
Shaik. An accurate barrier for the hydrogen exchange reaction from va-
lence bond theory: Is this theory coming of age? Chem. Eur. J., 9(18):4540–
4547, 2003.

[33] Philippe C. Hiberty, Stephane Humbel, and P. Archirel. Nature of the

differential electron correlation in Three-Electron bond dissociations. effi-
ciency of a simple Two-Configuration valence bond method with breathing
orbitals. J. Phys. Chem., 98(45):11697–11704, 1994.

[34] Wei Wu, Lingchun Song, Zexing Cao, Qianer Zhang, and Sason Shaik.
Valence bond configuration interaction: A practical ab initio valence
bond method that incorporates dynamic correlation. J. Phys. Chem. A.,
106(11):2721–2726, 2002.

[35] Lingchun Song, Wei Wu, Qianer Zhang, and Sason Shaik. A practical va-
lence bond method: A configuration interaction method approach with
perturbation theoretic facility. J. Comp. Chem., 25(4):472–478, 2004.

[36] Lingchun Song, Wei Wu, Qianer Zhang, and Sason Shaik. Vbpcm: A va-
lence bond method that incorporates a polarizable continuum model. J.
Phys. Chem. A., 108(28):6017–6024, 2004.

[37] H.B. Chirgwin and C.A. Coulson. The electronic structure of conjugated
systems. vi. Proc. R. Soc. London Ser A, 201:196, 1950.

[38] C. A Coulson and I. Fischer. Notes on the molecular orbital treatment of

the hydrogen molecule. Phil. Mag. Ser 7, 40(303):386 – 393, 1949.

[39] S. F. Boys. Construction of some molecular orbitals to be approximately

invariant for changes from one molecule to another. Rev. Mod. Phys.,
32(2):296–299, 1960.

[40] S. Miertuš, E. Scrocco, and J. Tomasi. Chem. Phys., 55:117, 1981.

[41] P.-O. Löwdin. Phys. Rev., 97:1474, 1955.

24 BIBLIOGRAPHY

[42] Lingchun Song, Jinshuai Song, Yirong Mo, and Wei Wu. An efficient algo-
rithm for energy gradients and orbital optimization in valence bond theory.
J. Comp. Chem, 30:399, 2008.