hydrogen bond, pi-pi stacking, and van der waals
TRANSCRIPT
HYDROGEN BOND, PI-PI STACKING, AND VAN DER WAALS INTERACTIONS
INVESTIGATED WITH DENSITY FUNCTIONAL THEORY
AN ABSTRACT
SUBMITED ON THE THIRTIETH DAY OF MAY 2013
TO THE DEPARTMENT OF PHYSICS AND ENGINEERING PHYSICS
IN PARTIAL FULFILLMENT OF THE REQUIRMENTS
OF THE SCHOOL OF SCIENCE AND ENGINEERING
OF TULANE UNIVERSITY
FOR THE DEGREE
OF
DOCTOR OF PHILOSOPHY
BY
__________________________
YUAN FANG
APPROVED: ______________________________
JOHN P. PERDEW, Ph.D. DIRECTOR
_______________________________
ADRIENN RUZSINSZKY, Ph.D.
_______________________________
JIANG WEI, Ph.D.
_______________________________
LEV KAPLAN, Ph.D.
Abstract
Weak bonds such as hydrogen bond, pi-pi stacking and van der Waals
interaction are much weaker in the strength but play a more important role for the
existence of various lives. For example, they are the major intermolecular interactions
in the liquid and solid structure of water and determine the 3 dimensional structure of
protein and DNA, which are the crucial organic molecules in lives. As a result,
studying these weak bonds can lead to the better understanding of fundamental
knowledge of lives.
Kohn-Sham (K-S) Density Functional Theory (DFT) is an accurate and effect
way to investigate the fundamental properties for many-body systems, in which, only
the exchange-correlation energy as a functional of electron density need to be
approximated. However, weak interaction system is still a challenge problem for
KS-DFT. In this dissertation work, several standard density functionals are used to
study these weak interactions in the solid state structure ice as long as nucleic bases
molecules in the biologic system. It is found that the hydrogen bond can be well
described by most semilocal functionals: the mismatch problem of ice Ih and AgI for
GGA functional can be solved by using the higher level meta-GGA functionals and
the binding length and energy between nucleic bases in DNA can be well described.
However, the more accurate dispersion correction is strongly needed for van der
Waals interactions and pi stacking for super-high pressure ice phases and large size
biologic molecules, where van der Waals interaction takes major role. Finally, the
basic structural properties of various phases of ice and DNA can be understood based
on the investigation with appropriate functionals.
HYDROGEN BOND, PI-PI STACKING, AND VAN DER WAALS INTERACTIONS
INVESTIGATED WITH DENSITY FUNCTIONAL THEORY
A DISSERTATION
SUBMITED ON THE THIRTIETH DAY OF MAY 2013
TO THE DEPARTMENT OF PHYSICS AND ENGINEERING PHYSICS
IN PARTIAL FULFILLMENT OF THE REQUIRMENTS
OF THE SCHOOL OF SCIENCE AND ENGINEERING
OF TULANE UNIVERSITY
FOR THE DEGREE
OF
DOCTOR OF PHILOSOPHY
BY
__________________________
YUAN FANG
APPROVED: ______________________________
JOHN P. PERDEW, Ph.D. DIRECTOR
_______________________________
ADRIENN RUZSINSZKY, Ph.D.
_______________________________
JIANG WEI, Ph.D.
_______________________________
LEV KAPLAN, Ph.D.
ii
Acknowledgments
I would never have been able to finish my dissertation without the guidance
of my committee members, help from my group members and friends, and support
from the physics department, my family and wife.
First of all, I would like to express my deepest gratitude to my advisor, Dr.
John Perdew, for his excellent guidance, support, and remarkable patience. I would
like to thank my committee members, Lev Kaplan, Ph. D., Jiang Wei, Ph. D., Adrienn
Ruzsinszky, Ph. D., for their helpful insights for my dissertation. I would also like to
thank all my group members, Dr. Perdew, Dr. Ruzsinszky, Dr. Jianmin Tao, Dr.
Jianwei Sun, Bing Xiao, Dr. Pan Hao, for providing me with an excellent atmosphere
for doing research and all the helpful suggestions and discussions. The work in this
dissertation was undertaken mostly during a teaching fellowship at the physics and
engineering physics department. I gratefully acknowledge the support.
I would like to thank all my friends in New Orleans for the entire
accompaniment during the last six years.
An especially thank for my parents, Xin Fang and Shuyu Zhu, for all the
encouragement and education.
Finally, I would like to thank my wife, Dr. Shanshan Shen, for her constant
friendship, love, and support for all these ten years.
iii
Table of Contents
Acknowledgements ii
Table of Contents iii
I. Introduction: The bonds of life 1
II. Theoretical information:
Wavefunction and density functional theory,
approximations, and computer codes 9
2.1 Many-body Schrödinger equation problem 9
2.2 Wavefunction based approximation 11
2.3 Density functional theory 12
2.3.1 History of density functional theory 12
2.3.2 Approximation for exchange-correlation functionals 16
2.3.3 Climbing the "Jacob's ladder" of density functional 17
2.3.4 van der Waals Correction Functionals 29
2.4 Theory to Practical Calculation 31
2.4.1 The Vienna Ab initio simulation package (VASP) and PAW
method 32
2.4.2 Gaussian and Gaussian Type Orbitals 37
iv
III. Ice phases under ambient and high pressure: Insights for
density functional theory 40
3.1 Introduction 40
3.1.1 Interactions inside ice 40
3.1.2 Density functional theory for solid ice 42
3.2 Computational details 45
3.3 Results and Discussion 47
3.3.1 Lattice mismatch challenge for ice-Ih and β-AgI 47
3.3.2 Sublimation energy for ice phases under ambient and high
pressure 52
3.3.3 Transition pressure between ice phases 56
3.4 Liquid water 61
3.5 Conclusions 61
IV. What forces twist the Deoxyribonucleic Acid?
Interactions between nucleic bases with density
functional theory 63
4.1 Introduction 63
4.1.1 Deoxyribonucleic acid structure and nucleic bases 63
4.1.2 twist of DNA and DNA base pair steps 64
4.1.3 Ab initio methods for nucleic bases 67
v
4.2 Computational details 69
4.3 Results and Discussion 72
4.3.1 Hydrogen bond interaction between Watson-Crick pairs 73
4.3.1.1 Bond length and bonding energy 74
4.3.1.2 The contribution of vdW dispersion force for bonding
Between base pairs 76
4.3.2 pi-pi stacking interactions between stacked bases 78
4.3.2.1 Benzene dimer stacking 79
4.3.2.2 Rise and potential energy surface (PES) for stacked nucleic
bases 83
4.3.3 Stacking interactions for stacked nucleic pairs and possible
reason for the twist of DNA from the view of PES 91
4.4 Conclusion 94
References 96
1
1
Chapter I
Introduction: The bonds of life
The original spark of life may have begun in a warm little pond, with all sorts of
ammonia and phosphoric salts, light, heat, electricity, etc. present, so that a protein
compound was chemically formed ready to undergo still more complex changes.
- Charles Darwin[1], 1871
Although it is still under debate how the first self-replicating molecule
formed, it is well accepted that these RNA, DNA, and protein molecules evolved to
the first living cells in a warm pond, the ancient ocean. While DNA is the organic
molecule encoding the genetic instructions used in the development and functioning
of almost all known life, water, an inorganic molecule, is considered to be the most
important fundamental requisite for life. In fact, up to 70% of the human
body consists of water. However, it is not until the last one hundred years that
scientists have discovered some fundamental properties of water and DNA molecules,
and many questions remain to be answered.
Water covers 71% of the earth’s surface and often co-exists in its solid,
liquid and gaseous states. Only in the 19th
century was the composition of the water
molecule proposed to be H2O[2], with one oxygen atom and two hydrogen atoms
connected by covalent bonds in a polar molecule. Water exhibits some special
physical and chemical properties: Water is in the liquid state at room temperature and
transits to the solid phase when the temperature is below 273K[3], which is common
2
2
during winter on most parts of the earth’s surface; water’s density changes with
temperature with the maximum occurring around 277K, and the density decreases by
~10% when frozen to solid ice; water is a good polar solvent, often referred to as a
universal solvent in chemistry; some organic molecules like protein and DNA are
easily dissolved in water, etc. To find out the mystery of water, for decades
experimental chemists and physicists tried to figure out the basic structure of liquid
and solid water. With the development of X-ray and neutron diffraction technology,
the structure of natural ice was confirmed in the middle of 20th
century[3,4], while
that of liquid water is still under debate. The crucial factor to study water is the
understanding of the intermolecular interactions, principally hydrogen bonds. A
hydrogen bond in water is the electromagnetic attractive interaction of a
slightly-positive hydrogen atom in one molecule to a slightly-negative oxygen atom in
an adjacent water molecule. The strength of the hydrogen bond is about 2-10 kcal/mol
which is similar to the thermal energy. This makes water stay liquid under room
temperature and easy to freeze. The exact number of hydrogen bonds in liquid water
fluctuates with time and depends on the temperature, which leads to the change of
density. Unlike in solid ice, hydrogen bonds in water only have 1-20 ps lifetime with
bond breaking and rebuilding in 0.1ps, which leads to the difficulty of measuring the
structure of liquid water. Hydrogen bonds also play an important role when water
interacts with other structures, making it a good solvent. In particular, the existences
of hydrogen bonds in organic molecules like protein and DNA, which are also
3
3
considered in this work, make them soluble in water.
Deoxyribonucleic acid (DNA)[5] is a molecule encoding
the genetic instructions and famous for its double helix structure first discovered by
James D. Watson and Francis Crick in 1953. The common B-DNA double helix has
double-stranded helices, consisting of two long backbones called grooves made of
alternating sugars and phosphate groups, with the nucleic bases attached to the sugars.
A nucleotide is building block of DNA which consists of one pair of linked bases plus
the attached part of the backbone. Though there are billions of nucleotides in one
double DNA chain, only four kinds of nucleic bases exist in DNA: adenine (A),
cytosine (C), guanine (G) and thymine (T), and their order stores the genetic
information following the base pair rules: A pairs with T, G pairs with C. The double
helix makes one complete turn about its axis every 10.4-10.5 base pairs in solution,
which means that there is a twist of about 36 degrees between neighboring
hydrogen-bonded pairs of bases. This twist is formed and stabilized by various forces.
Two of them are named the hydrogen bond which is between the nucleic bases on
different backbones and the pi-pi interaction between nucleic bases on the same
backbone. The interaction is a weak interaction between aromatic rings with energy of
2-10 kcal/mol, which contributes at least as much as the hydrogen bond for the
stability between bases and also influences the strength of hydrogen bond. However,
while the nature of the hydrogen bond is well researched, this interaction is still
actively debated in the literature since the 1990s, because it is a combination of van
4
4
der Waals dispersion forces, electrostatic interactions and Pauli repulsions.
Another important weak interaction is the van der Waals interaction, the sum
of the attractive forces between molecules resulting from fluctuating
instantaneously-induced dipoles. Typical energies for vdW interactions are around
0.5-1 kcal/mol, much smaller than covalent bonds with energies of 50-200 kcal/mol.
However, since this value is comparable with the strength of the hydrogen bond and
pi-pi interaction, the influence of vdW interactions must be taken into consideration
when dealing with weakly-bound solids or molecules. Recent findings show that vdW
interactions play an important role in the structure of liquid water and high-pressure
ice and in the 3D stability of a protein molecule. In experiment, the measurements for
such weak interactions are difficult. In fact, not until 2012 was made the first direct
measurement of the strength of the van der Waals force for a single organic molecule
bound to a metal surface[6].
High-accuracy theoretical simulation methods for all these weak interactions
are needed. For hundreds of years, experiment was the only way to explore the
mystery of life and the molecules of life. Although many discoveries were made,
scientists suffered from the technical limits of theoretical methods and experimental
equipment, especially on atomic scales. Thanks to the invention of computers in 1946,
scientific simulations with computers became the second way to understand the world.
Nowadays, with high-performance super computers and efficient algorithms,
scientific simulations have been used not only to verify the experimental findings but
5
5
also to direct the experiments: Increasingly, material data bases constructed by
computation can be used to predict new materials with desired properties.
Although computational techniques were dramatically revolutionized during
the last half century, atomic-scale computation for large systems is still a huge
challenge for computer modeling. The requirements for various simulation methods
are both accuracy and efficiency. All these methods can be grouped into wave
function and density functional approximations, which will be explained in detail later
in this dissertation. Even though, in some situations, wave function methods work
more accurately than density functional theory (DFT)[7], their computational cost is
significantly larger than DFT, and so these methods usually can only be used for
system with several tens electrons. At the other hand, DFT is capable of dealing with
systems with more than hundreds of electrons. With the development of new DFT
approximations, the accuracy problem is being solved. Thus, DFT is now the most
commonly-used ab initio method in both condensed matter physics and quantum
chemistry.
Nowadays, there are hundreds of different density functionals. By how
much information about the density they take into their xc energy approximations, we
can sort them as local, semilocal, and nonlocal functionals. With the development of
GGA functionals in the 1990s, DFT was considered accurate enough for the
calculation of atoms, molecules and solids. The famous GGA PBE[8] is cited more
than 30,000 times and is widely used in condensed matter physics, while the hybrid
6
6
functional B3LYP[9,10] is widely used in computational chemistry.
FIG. 1-1: Number of papers with topic including PBE, B3LYP, and other DFT
functionals[11].
However, there are still difficulties in using GGA-DFT[12] to properly
describe intermolecular interactions like van der Waals forces, as well as transition
states for chemical reactions, global potential energy surfaces, some other strongly
correlated systems, and the band gaps in semiconductors. To solve these problems,
many improved functional approximations have been invented. As the highest level of
the semilocal functionals, the meta-GGA is a natural way to improve accuracy further
by making use of additional semilocal information. Meta-GGAs have yielded better
results than GGAs for covalent, ionic, and metallic solids, including atomization
energy, lattice constant, bulk modulus, and cohesive energy[13]. Although the GGA
7
7
functionals were already widely tested for hydrogen bonds and organic molecules, the
performance of meta-GGA functionals for such system is still not well studied,
especially for some newly developed functionals. This work aims to establish the
ability of the meta-GGA to describe weakly bonded structures.
In chapter three of this dissertation, the interactions in the water solid
structures are studied with several density functionals on different levels. At first, the
easier case of ambient-condition solid phase ice-Ih is studied for a lattice mismatch
problem, to learn how these xc functionals perform when hydrogen bonds play the
major role. After that, in view of the importance of vdW forces at high pressure[14],
the properties of two high-pressure phases of ice are studied. By comparing the results
of all these DFT functionals to the experimental data, one can explore their
advantages and shortcomings for systems with a combination of hydrogen bonds and
vdW interactions.
In chapter four, the work for the DNA nucleic base pairs is shown. It is well
known that the interactions of nucleic bases coming from the hydrogen bond and pi-pi
stacking interactions. To isolate the influence of them, first the bonding energy and
bond length which are between nucleic bases on the same rung were studied, then the
rise (vertical separation between stacked bases) and potential energy surface was
computed as a function of the twist angle between two stacked bases. By comparing
to those found by quantum chemical wave function based methods, the performance
of different functionals can be understood. Finally, the preferred configurations for
8
8
possible DNA Watson-Crick bases pairs (with four nucleic bases) are studied at a
preliminary level. This problem would be very hard to study with quantum chemical
methods.
All the interactions studied here are much weaker than those of common
chemical bonds, but should not be neglected. Considering their importance in
chemical reactions, and to life, a better description of this weak interaction is needed.
The major question to be addressed here is how all these density functionals’ perform
for systems with hydrogen bonds is or with a combination of several types of weak
bonds including van der Waals as well as hydrogen bonds, and which functional
should be favored in future studies. Before answering these questions, some
theoretical concepts must be introduced. In the next chapter, basic density functional
theory and density functional approximations and computer codes are briefly
discussed, with special focus on the meta-GGA functional.
9
9
Chapter II
Theoretical Information: Wavefunction and
Density Functional Theory, approximations, and
computer codes
2.1 Many-body Schrödinger equation problem
Condensed matter physics is a branch of physics that deals with the physical
properties of condensed phases of matter. It overlaps with chemistry, materials science,
and nanotechnology, and relates closely to atomic physics and biophysics. Theoretical
condensed matter physics attempts to understand and manipulate the properties of
materials by studying the interactions of nuclei and electrons from fundamental
physical principles such as quantum and statistical mechanics. The key way to get
information of the ground-state properties is to solve the many-body time-independent
Schrödinger equation[15] for a system containing N electrons and M nuclei shown as
Eq. [2-1],
[2-1]
with the Hamiltonian
10
10
[2-2]
where is the ratio of the mass of nucleus A to that of an electron; is the
atomic number of nucleus A. R and r are the coordinates of nuclei and electrons,
respectively.
When N=M=1, this is a typical hydrogen-atom Schrödinger equation. When
new nuclei and electrons are added, this equation becomes too complex to be
mathematically solved because all the correlation between the particles needs to be
considered. Generally, for a condensed matter physics problem, we may have at least
several nuclei and tens of electrons in a repeating unit cell, a problem which cannot be
perfectly solved. Thus, some reasonable approximations must be made to simplify
this many-body Schrödinger equation problem.
The first approximation is the Born–Oppenheimer (BO) approximation[16]
proposed in 1927. In the BO approximation, the nuclei are treated as fixed due to the
high ratio between nuclear and electronic masses. Then the nuclei kinetic term and
nuclei repulsion term can be removed and the many-body Hamiltonian reduces to a
three-term many-electron Hamiltonian with form in Eq. [2-3].
11
11
[2-3]
Then the problem is to solve the electronic Schrödinger equation to get the electronic
wave function and energy. After that, the total energy can be found by adding the
nuclear repulsion energy to the electronic energy.
[2-4]
However, because of the interaction between electrons, this equation is still hard to
solve, and further approximation is needed. All these approximations can be sorted as
wavefunction based approximations and density functional theory.
2.2 Wavefunction based approximation
Wavefunction based approximations are not used in this work, but since some
MP2 results are used to benchmark DFT calculations later, some basic information
about them will be briefly introduced first.
A major aim of wavefunction based methods is to find the many electron
wavefunction . First is the Hartree-Fock (HF) approximation, which assumes that
the exact wavefunction of the system can be approximated by a single Slater
determinant. The HF approximation is the central starting point for many other
12
12
methods, and is even involved in the hybrid density functionals. However, neglecting
electron correlation in HF may lead to large deviations from experimental results,
especially for energy. So a set of methods, referred to as post Hartree-Fock, have been
developed to improve the HF approximation by adding electron correlation. These
methods, including the Coupled Cluster method (CC), Møller–Plesset perturbation
theory (MP2), Quadratic configuration interaction (QCI), Quantum chemistry
composite methods (QC), and so on, can give more accurate results than HF and are
often used as reliable computational benchmarks[17] for other methods like DFT,
particularly in the situation where experiment measurements are hard to perform. The
accuracy with these methods comes with significant increasing computational cost, so
they are not fit for large complex systems and molecule dynamics simulations. A
useful compromise between computational accuracy and efficiency can be achieved
by density functional theory (DFT).
2.3 Density functional theory
Unlike the wavefunction based methods, the density functional theory invokes a
functional of the electron density to find the ground-state total energy and density.
Mathematically, a functional is a rule that maps a function like the electron density
13
13
[2-5]
into a number like the total energy. By doing this, DFT reduces the 3N-dimensional
many-electron problem to a 3-dimensional problem, which can be practically solved
for much larger systems. Nowadays, no other method can achieve comparable
accuracy to DFT at the same computing cost.
2.3.1 History of density functional theory
Back in the 1920s, Thomas and Fermi first approximated the energy in terms
of the electron density. The Thomas-Fermi Model[18,19] was simple in form, and not
very accurate, but it provided the conceptual root for density functional theory: the
electron density can determine the energy and other properties of the ground state.
In the 1960s, the firm theoretical base for DFT was constructed by the two
Hohenberg-Kohn theorems[20]. In the B-O approximation, the external potential
determines the kinetic energy of electrons, the interaction energy between electrons,
and the electron density. Hohenberg and Kohn proved that the inverse relationship
also exists: the electron density can uniquely determine the external potential ,
and thus other quantities. The first Hohenberg-Kohn theorem states that the ground
state properties of a many-electron system are determined in principle by the electron
density. Then the second H–K theorem defines the energy functional for the system
and proves that the correct ground state electron density can be found by using the
variational principle. The total energy can be expressed as a functional of density, and
14
14
the exact ground state density is the one that minimizes the total energy.
[2-6]
If we know the form of the functional , then in a given external potential we
can find the ground state density and energy by minimizing the total energy.
However, the exact functional is not known in any computable form, so
approximations are still needed.
In 1965, the discovery of the Kohn-Sham equations[21] made the density
functional theory into a practical and realistic computing method. Within this
framework of Kohn–Sham density functional theory (KS-DFT), the
intractable many-body problem of interacting electrons in a static external potential is
reduced to a tractable problem of non-interacting electrons moving in an
effective potential, typically denoted as , called the Kohn-Sham potential. In the
Kohn-Sham theorem, the functional is expressed as
[2-7]
where is the Kohn-Sham kinetic energy which is the kinetic energy of all
the non-interacting electrons, (hartree energy) is the Coulomb energy
15
15
between the electrons, and is the exchange-correlation energy. Then the
Kohn-Sham potential can be found from the functional derivative
.
[2-8]
In the Kohn-Sham system, the electrons are non-interacting; the one-electron
Kohn-Sham Schrödinger equation can be written as
[2-9]
where is the orbital energy of the corresponding Kohn–Sham one-electron orbital
and the density of the N-electron system is defined as
Since the Kohn-Sham potential [2-8] depends on the electron density which is
defined above as Eq. [2-5], this Schrödinger equation must be solved self-consistently.
16
16
FIG.2-1 Self-consistent process
One typical self-consistent procedure is shown as FIG. 2-1. Generally, starting
with an initial electron density, is built. Then by solving the Kohn-Sham
equations, a set of Kohn-Sham orbital can be found, which give the new electron
density. These calculations are repeated until a specified level of convergence
requirement is achieved. After that, with the true electron density, the total energy can
be computed. The Kohn-Sham method is quite efficient since non-interacting systems
are relatively easy to solve as the wave-function can be represented as a Slater
determinant of orbitals and the kinetic energy functional of such a system is known
exactly. However, the term in Kohn-Sham energy is not known exactly. To
find a reasonably accurate approximation for is the major task of Kohn-Sham
density functional theory.
17
17
2.3.2 Approximation for exchange-correlation functionals
The first attempt or Hartree approximation simply ignored the exchange-correlation
energy term in the Kohn-Sham theory and turned out to imply drastically too-weak
bonds between atoms and drastically too-long bond lengths and lattice constants. This
indicates that, even when it is a relatively small part of the total energy, the
exchange-correlation energy is a crucial part of the interactions between atoms, and it
is therefore sometimes referred to as "nature’s glue"[22]. As a result, for accurate use
of KS-DFT, must be included, however, since contains all
the electron-electron interactions in the real system which are excluded in the Hartree
approximation, the exact form or even an accurate-enough approximation are hard to
build. It took ten years for the completion of the first practical approximation, the
local spin density approximation (LSDA), and another ten or twenty years for the
completion of the generalized gradient approximation (GGA), which is almost
accurate enough for computation in both solid state physics and quantum chemistry.
Basically, there are two ways to construct the approximate functionals: non-empirical
and empirical. Non-empirical density functionals are constructed to satisfy some exact
constraints on ; they are not fitted to certain real results so they can be
universal for all systems. At the other hand, (semi) empirical functionals include
parameters which are fitted to other computed or experimental results. These
functionals may be more accurate than non-empirical functionals inside their fitting
18
18
range but can fail outside it. Today, there is a long list of different kinds of density
functional. An efficient way to sort them is using the "Jacob's ladder[23]" which
was proposed by Perdew and Schmidt in 2002.
2.3.3 Climbing the "Jacob's ladder" of density functional
Generally speaking, most exchange-correlation approximation can be expressed as
[2-10]
where n is the electron density, is the gradient of the density and is the kinetic
energy density. Based on the complexity of density information that is taken into
account, these approximations can be sorted into five levels as shown in FIG. Like the
biblical Jacob's ladder that the Patriarch Jacob dreamed about, it is a ladder that links
the earth to heaven. The DFT Jacob's ladder bridges the Hartree world and the
chemical-accuracy heaven. Climbing the ladder from one rung to the one above, a
new argument or ingredient of the density is added to achieve higher accuracy, at
some increase of computing cost. Noting that the increase of accuracy is a statistical
result, the performances of various functionals depend on the problem they treat,
especially for the empirical functionals that are fitted for certain systems and
properties.
19
19
FIG.2-2: Jacob's ladder
Next we will discuss some basic functionals on each rung, especially those
used in the later chapters.
(i) Local Spin Density Approximation
As the earliest and simplest attempt designed in the 1960s and 1970s, LSDA
depends only upon the value of the electronic density at each point in space and
excludes other terms like the derivatives of the density. Under this approximation, the
exchange-correlation energy of the system can be expressed as
20
20
[2-11]
where are the spin-up and spin-down density, is the total electron
density, and
is the exchange-correlation energy per electron for an
interacting uniform electron gas with spin density .
Generally, the exchange-correlation is the sum of the exchange energy
and the correlation energy . The exchange energy functional for LSDA is
analytically known. In the spin-unpolarized case,
[2-12]
The analytic form for the LSDA correlation energy functional is not known except in
low and high density limits. So the LDSA is a controlled interpolation between
these two limits, and this accounts for the difference between various LSDA
functionals such as VWN, PZ81, etc[24--26].
As suggested by its method of construction, LSDA is only exact or accurate
for systems with uniform density or slowly-varying density such as in some solids. In
fact, during the 1970s and 1980s, LSDA-DFT was more successfully used in
condensed matter physics than quantum chemistry, and worked surprisingly well for
many solids. This can partly be explained by the fact that that the
21
21
exchange-correlation hole of LSDA satisfies exact hole constraints, and partly by an
understood error cancellation between LSDA exchange and correlation energies. In
LSDA, the exchange energy is typically underestimated by 10% and correlation
energy is overestimated by 100%. Then the total energy is typically too high. LSDA
shows 1%-3% underestimation of lattice constant, and its atomization energy for a
molecule is typically too large[27]. Besides, the description for weak bonds is bad for
LSDA: It overestimates the short-range part[28,29] and underestimates the long-range
vdW part. However, the cancellation of error in LSDA sometimes can lead to
surprising results such as better surface energy for a simple metal than GGAs. As the
first rung of the KS DFT, LSDA gives the starting point for the construction of
higher-level functionals.
(ii) Generalized Gradient Approximation (GGA)
To improve the performance of LSDA and work beyond the uniform-density
limit, a logical idea is to construct a new functional to include the derivatives of
density. For example, the gradient expansion approximation (GEA) tries to add the
energy corrections with a gradient expansion to second- or fourth-order in the
gradients. But these attempts show little improvement over LSDA, and are even
worse[30] for the reason that truncation of the gradient expansion violates
exchange-correlation hole constraints.
To get around this problem, the exchange-correlation energy can be expressed
22
22
in a more generalized form using the density and its gradient, which is known as the
generalized gradient approximation (GGA)[12].
[2-13]
The construction of can proceed in various ways. One route to the PW91 and
PBE GGAs used here is to start with the second-order gradient expansion for the
exchange-correlation hole, then introduce sharp real-space cutoffs to satisfy the exact
hole constraints. Generally, the GGA exchange-correlation energy can be defined as
[2-14]
where is the enhancement factor for the uniform exchange, is the local Seitz
radius (n=3/4π ), is the relative spin polarization, and
is a dimensionless density gradient. The choice of the mathematical form
for makes the difference between GGAs. For example, in the most widely-used
GGA functional, the Perdew-Burke-Ernzerhof (PBE)[8], a simple is defined as
[2-15]
23
23
where are two non-empirical coefficients. And the can have a more
complex form.
Compared to LSDA, GGA PBE can predict much more accurate atomization energies
for molecules, so DFT expanded its application to quantum chemistry after the birth
of GGA. However, a problem with GGAs is that none of them can achieve an
accurate performance for both atomic/molecular energy and for solid/surface energy.
GGAs which aim to describe atoms and molecules will fail in the solid and vice verse.
PBE usually overestimates the lattice constants of solids by 1%-2% and
underestimates the surface energy of a metal, while PBEsol[31], which is restores the
density-gradient expansion for exchange in solids and surfaces, improves equilibrium
properties for solids and surfaces, at the cost of accuracy for atoms and molecules.
This problem needs to be solved with new approximations which can be applied with
accurate results in both molecules and solids. That is one of the motivations for the
construction for the meta-GGA functionals.
(iii) meta-GGAs
Meta- is a prefix to indicate a concept used to complete or add to an earlier
concept. In this sense, meta-GGAs, the third rung on the ladder, are constructed to
improve the performance of GGAs by taking into consideration additional
information about the electron density like the Laplacian of the density or the
kinetic energy density. The typical formula for meta-GGA xc energy can be expressed
24
24
as
[2-16]
where are the electron density and gradient of the density included in
GGAs, and , the new information included in the meta-GGAs, is the kinetic energy
density for the occupied KS orbitals .
[2-17]
Although the idea of meta-GGA arose in the 1990s, it is not until recent years that
practical meta-GGAs have been constructed. Subsequent research showed that they
improve over GGAs in many situations, including the hydrogen bond in water clusters,
and this is the reason we seek to understand their performance for solids with
hydrogen bonds and other weak interactions. As for GGAs, meta-GGAs can be
constructed non-empirically or empirically. John. Perdew leads the non-empirical
effort by constructing TPSS[32], revTPSS[13], etc, while the Minnesota
Functionals[33] constructed by Don Truhlar are widely-used semi-empirical
meta-GGAs.
Non-empirical functionals TPSS and revTPSS:
The starting point for TPSS is the original version Perdew-Kurth-Zupan-Blaha
25
25
(PKZB)[34] meta-GGA, which poorly describes the bond length and hydrogen bond.
The key idea in TPSS[32] is to restore PBE-like GGA behavior at large reduced
density gradients, where the constraints added in PKZB do not apply. Somewhat as in
GGAs, the exchange energy of meta-GGA can be expressed as
[2-18]
in which again
is the exchange energy density for uniform electron gas, and
is the meta-GGA exchange enhancement factor with two
dimensionless parameters p and z. and , where
is the
Weizsaecker kinetic energy density, and is the orbital kinetic density defined
earilier. And the analytic form of is constructed to satisfy the one- or
two-electron density constraints. The improvement in the exchange part makes TPSS
better than earlier PKZB by shortening the bond length and improving the energy for
hydrogen bonds. Besides, TPSS also implies good results for surface energy and
atomization energy, but lattice constants only slightly better than those of the PBE
GGA. revTPSS was constructed to further improve the lattice constants. The
construction principle for revTPSS is similar to that of PBEsol, restoring the
second-order gradient expansion for exchange over a wide range of densities to
achieve better results for lattice constant, at the same time keeping the good
26
26
performance of TPSS for surface and atomization energy. Recently testing show that
this new revised version meta-GGA works well for many systems and may "become a
workhorse semi-local density functional for both condensed matter physics and
quantum chemistry."[13]
Empirical functional M06L:
Except for the TPSS-like functionals, which are constructed by satisfying
several physical constraints, some other functionals are constructed by fitting the
parameters to real-system data, such as the widely used Minnesota functionals family,
which includes tens of meta-GGA and hybrid functionals. The meta-GGA M06L[33]
is one of the most successful and widely accepted functionals. The exchange energy
in M06L can also be expressed in the similar form
[2-19]
where ,
is the exchange energy density for LSDA and PBE,
, are two functions of the kinetic energy density. Unlike TPSS, these
two functionals are built by fitting the parameters to a molecular database. As a result,
M06L performs quite well for molecules, even better than several hybrid functionals.
However, the performance of M06L outside its fitting range, as in solids, can be poor,
and the huge number of fitting parameters in M06L may cause some calculation
27
27
problem for extended systems.
FIG. 2-3: List of optimized parameters used in the M06L meta-GGA[33]
MGGA_MS:
MGGA_MS (meta-GGA made simple)[35] is a newly developed semi-local
functional. The dimensionless inhomogeneity parameter α, which is defined as
is introduced into the meta-GGA exchange enhancement factor
to characterize the extent of orbital overlap. By adding this exchange functional with a
variant of the PBE correlation functional, the MGGA_MS xc energy is built. This new
functional performs better than GGA PBE for atoms, molecule, solids and comparable
with the meta-GGA revTPSS, but with a much simpler form. Besides, since α can
recognize different degrees of orbital overlap, it can distinguish the weak bonds from
the strong binds. In this case, MGGA_MS has the ability to capture at least part of
weak interactions, which is a challenge for the conventional GGA and meta-GGA
functionals.
28
28
FIG. 2-4: The binding curve of Ar2, which is a typical molecule bound by a
noncovalent bond[36]
(iv) Hybrid Functionals
Hybrid functionals mix the exact exchange energy from the HF method
with the exchange-correlation energy from another method such as DFT, which is
why they are called "hybrids". A typical formula for a hybrid-GGA
exchange-correlation energy is
[2-20]
where is the exact exchange energy,
and are the GGA exchange
and correlation energies, and "a" is the mixing coefficient. Different hybrid
functionals may employ different GGA xc energy and mixing coefficients.
For example, B3LYP[9,10] is a hybrid popularly used in the chemistry. It
29
29
uses the exchange energy of the B88[37] GGA and the correlation energy of LYP[38].
[2-21]
while the PBE0[39] using both exchange and correlation energy of GGA PBE is more
widely-used in condensed matter physics due to its lack of empirical fitting
parameters.
[2-22]
Recently, meta-GGA exchange-correlation functionals have also been used to construct
hybrid meta-GGA functionals such as Minnesota 06 family of functionals.
Hybrid functionals are fully nonlocal and tend to yield better performance than
semi-local functionals, although they also increase computing time due to the large
computational requirement for the exact HF exchange, so they are used more to treat
the problem of molecules. Furthermore, the performance of various hybrid functionals
strongly depends on the applying system; B3LYP works well for molecule, but fails in
metals and other extended systems.
30
30
(v) Random Phase Approximation
The RPA was introduced back in the 1950s to estimate the correlation energy of
the uniform electron gas[40]. The “RPA in a density functional context” was
introduced later by Langreth and Perdew[41]. It took more decades before the RPA
could be practically used in the KS-DFT. Generally, the RPA xc energy can be
expressed as the summation of exact exchange energy and RPA correlation energy:
[2-23]
In RPA, the long-range dispersion interactions are well described, but the short-range
correlation is overestimated. So a short-range semilocal correction can be added to
RPA, yielding RPA+[42,43]. A major problem with RPA is the computational cost,
maybe 100 times greater than that of GGA.
2.3.4 van der Waals Correction Functionals
The classical semilocal functionals such as LSDA, PBE, PBEsol, TPSS, and
revTPSS are not fully nonlocal, and even the hybrid functionals with the nonlocal HF
exchange mostly cannot yield good results for systems where the long-range van der
Waals interactions need to be taken into consideration, as in many biological molecules.
Using RPA can solve this, but with a significant increase of computing cost. An
31
31
alternative is to add a correction for the dispersion energy to the DFT xc energies,
which can be expressed as a power series[44]:
[2-24]
where R is the distance between two particles, is the coefficient which describes the
dipole-dipole interaction, and is the coefficient which describes the
dipole-quadrupole interaction. In large-separation situations, the first term usually
plays the major role and is easy to include. Although adding the following ,
terms may decrease the error, they are hard to get. So the practical step is to find a
reasonable approximation for the coefficient . Nowadays, there are many
methods[45--49] designed to provide a long-range vdW correction to semilocal
functionals, and we will use several of them in later chapters.
One is the DFT-D2[47] approach of Grimme, in which the van der Waals
interactions were described via a simple pair-wise force field and added to the
conventional Kohn-Sham DFT energy. The dispersion potential energy for periodic
systems is defined as
[2-25]
where the dispersion energy is defined as
32
32
where is the number of all atoms, is a global scaling factor,
denotes the dispersion coefficient for the atom pair i and j and the damping functions
is defined as
Because of the simplicity and low computational cost, this pairwise correction is
widely used and implies quite good results. However, there are still some
shortcomings because of the neglect of non-pairwise effects. The coefficient is
mostly taken from experiment.
Recently, a new version called DFT+D3[50], adding the term to the
dispersion energy, has been constructed but is not included in this work.
Second is the vdW-DF[48] method proposed by Dion et al. It is a non-local
correlation functional added to a semilocal xc energy to account for dispersion
interactions:
[2-26]
where is the exchange energy of a certain GGA functional,
is the local
spin-density approximation (LDA) to the correlation energy and is the
33
33
approximate nonlocal energy term. optB88-vdW and optB86b-vdW used in our work
are basing on the vdW-DF approach with the Becke88 and Becke86 exchange
functional optimized for the correlation part. Early work shows that the performance
of this kind of functional may depend on the GGA functional included in it.
2.4 Theory to Practical Calculation
As mentioned before, the KS equations must be solved self-consistently to
get the ground state electron density; the process can only be performed by the
computer or super computer. In 2006, more than five million hours of computer time
was used by DFT simulation at NERSC in the Lawrence Livermore National
Laboratory. Therefore, accurate and efficient codes for the DFT calculations are
highly needed. There are now hundreds of codes written for DFT. The major
difference among them is the choice of the type of basis sets for the expansion of the
KS orbitals, including plane waves, localized atomic orbitals (LCAO), and augmented
plane waves. Each of them has advantages and disadvantages and the best choice
depends on the system treated. Generally speaking, for finite systems such as atoms,
molecules and clusters, LCAO usually works well while the plane wave method
works well for extended periodic systems like solids. In the following, two codes used
in our work are briefly introduced.
2.4.1 The Vienna Ab initio simulation package (VASP) and PAW
34
34
method
The Vienna Ab initio Simulation Package (VASP)[51,52] is a computer
program for atomic- scale materials. It can compute an approximate solution for the
many body Schrödinger equation within the KS-DFT and HF approximation using a
plane-wave basis set and the projector-augmented-wave (PAW) method. Thus, it is
more suitable for periodic solid structures. Today, VASP is used by more than 1400
research groups in academia and industry worldwide.
Unlike in isolated atoms or molecules, in an extended system such as a solid
structure, which usually consists of a huge number of electrons, it is impossible to
calculate wavefunctions., Instead, the solution is treat the solid as a periodic structure,
which is built by repeating a small part of the solid, called the unit cell, infinitely in
all three dimensions. For a so- called periodic solid structure, the plane-wave basis set
is widely used because of Bloch's theorem which implies that in a periodic system the
KS or HF orbital can be written as the product of a plan wave term and a
periodic factor ,where
[2-27]
Since
, where R is a Bravais lattice vector, it suffices to calculate
in only one unit cell. Next the periodic term can be expanded as a series of
plane waves with wave vectors G, the reciprocal lattice vector of the periodic crystal.
[2-28]
Then the orbital can be expressed as
35
35
[2-29]
Finally, the Kohn-Sham equation can be rewritten in the plane-wave basis set form
[2-30]
where
is the plane-wave kinetic energy, and is the electronic energy.
Plane-wave cutoff and k-points:
In the above equation, an infinite number of G’s need to be summed in principle
which is impossible for real calculation, while in practice the actual number of G's is
determined by the kinetic energy cutoff . This approximation is reasonable
because the plane waves with smaller kinetic energy are much more important than
the others, so only the plane waves with the kinetic energy satisfying
are taken into the calculation. Using the energy cutoff may decrease the
accuracy for the total energy. However, a convergence test which finds the maximum
value of to make the total energy converge can minimize the error.
The total energy is evaluated from the contributions of each occupied
electronic state at every wave vector (or k-point) in the solid. However, in a periodic
structure, the Irreducible Brillion zone (IBZ) in the reciprocal space, which is a
36
36
fraction of the complete unit cell of the BZ, is sufficient to determine all the
information for the crystal. Increasing the k-points can improve the accuracy, tending
to the true total energy while the computing cost increases linearly. Consequently, a
convergence test for k-points is also needed before making the calculation for each
system.
Pseudopotentials and the projector augmented wave method:
The core electrons usually have higher kinetic energy and so need a bigger
number of plane waves to represent them, while the chemical bonds are mostly
associated with so-called valence electrons, far away from the core. The practical way
is to simplify the plane waves for the core electrons since they play only a slight role
in the bonds. This is the basic idea of the pseudopotential method. Generally, a KS
valence orbital is smooth in the bonding region and oscillates rapidly near the core;
the pseudopotential method replaces the oscillating part of the valence orbital with a
smoother pseudo-orbital , associated with a weaker pseudopotential
in the core, while keeping them the same as the true all-electron potential and orbital
outside the core.
There are different ways to construct the pseudopotential. The first one is
norm-conserving pseudopotentials which try to make the pseudo-charge inside the
core the same as the all-electron charge there. However, it is hard to generate this kind
pseudopotential with both transferability (one pseudopotential can be used in different
environment) and softness (one only needs a few plane waves). An alternative method
37
37
called the ultrasoft pseudopotentials relaxes these constraints on the valence
pseudo-orbitals and can achieve better accuracy than norm-conserving
pseudopotentials for the same small number of plane waves.
FIG. 2-5: Schematic illustration of a pseudopotential orbital and the corresponding
all-electron orbital.
In VASP, the projector augmented wave method is used, which is a
combination of the ultrasoft pseudopotentials and linear augmented plane wave
methods that partition space into the space around each atom and an interstitial region.
Then the orbital inside the sphere can be represented by spherical harmonic functions
38
38
and outside by an expansion in plane waves. The PAW method is quite
computationally efficient and can be as accurate as the all-electron method.
The plane-wave basis method can also be applied to atoms or molecules by
building a large super cell around the molecule to minimize the influence from its
periodic images. However, the LCAO method is better suited to these situations.
2.4.2 Gaussian and Gaussian Type Orbitals
In the localized atomic orbital method, the orbitals are written as the sum of
atomic-like orbitals
[2-31]
So the problem is to build a proper basis and to optimize the coefficients.
In the 1930s, the Slater-type orbitals were introduced to describe the atomic orbitals,
where N is a normalization constant; a,b,c control angular momentum, and
controls the width of the orbital. STOs can describe the shape of an atomic orbital
quite closely, but do not permit analytic integrations for molecules. Thus STOs are
generally only used for atomic problems and most molecular problem employs GTOs.
A Gaussian type orbital can be expressed as:
GTOs are not atomic-orbital-like, because they decay too fast with the increase of r,
39
39
but they are much easier to use since the product of two Gaussian functions centered
on two different atoms is a finite sum of Gaussians centered on a point along the axis
connecting them. The solution for gaining both accuracy and efficiency is to first use
GTOs to represent STOs, then to use the resulting STOs to describe the atomic
orbitals. For example, the minimal basis set for an atomic orbital is called STO-3G,
which means that three GTOs are used to define one STO and one STO is used to
define an AO.
FIG. 2-6: Shape of Slater type and Gaussian type functions.
Another way, called Pople basis set, splits the core and valance electrons and uses
different numbers of GTOs to describe them.
40
40
FIG. 2-7: One sample notation for the Pople basis set used in Gaussian
Gaussian[53] is a computational code using the Gaussian-type orbitals (GTOs)
to expand the atomic orbitals. It is widely used in chemistry and molecular physics. In
theory, the bigger the basis set used in the Gaussian code, the better the result we can
get. In practice, accuracy still depends on the real situation: a large basis set without
diffuse function may be worse in anions and for some big molecules, and it may be
expensive and unnecessary to apply a large basis set. DFT is less dependent on the
size of the basis set than are the wavefunction methods. However, a convergence test
to get a suitable basis set for the system is still needed.
41
41
Chapter III
Ice phases under ambient and high pressure:
Insights for density functional theory
3.1 Introduction
3.1.1 Interactions inside solid ice
As mentioned early in the first chapter, water and its related systems play a
crucial rule in nature and human life, and attract lots of interesting from scientists.
Although the structure of liquid water has been experimentally researched for a long
time, the theoretical understanding is still poor and under debate[54], partly due to the
difficulty to describe accurately the interactions inside it. Consequently, we start with
the relatively easier water structure: ice, aiming to understand the basic relationship of
hydrogen bond and van der Waals interactions between water molecules.
Ice is the solid state of water. It exhibits a complex phase diagram [See Fig. 3-1]
with a hexagonal ice-Ih (also known as ice one) structure under ambient conditions.
Ice-Ih can exist down to 73K temperature and up to 0.2 Gpa pressure, as a result,
Ice-Ih is the most common ice phase on the Earth's surface. In fact, almost all the
nature snow, ice is in this structure. Because of its strong relevance to human
activities such as existence of life and regulation of global climate, ice-Ih has been
widely studied both experimentally and theoretically for hundreds years. Ice-Ih can
42
42
transit into other phases with the changing of temperature and pressure. The rich
phase diagram is mostly coming from the relative intermolecular interactions of
hydrogen bond and van der Waals dispersion forces.
Fig. 3-1 Observed and predicted water ice phase diagram[55]. One atmospheric pressure=101kPa
=1.01x10-4
GPa.
The water molecule is made of two hydrogen atoms and one oxygen atom
linked by strong covalent bonds, while ice is a molecular crystal with intermolecular
interaction arising from two non-covalent bonds: hydrogen bonds and relatively
weaker but important van der Waals (vdW) interactions.
Generally speaking, a hydrogen bond is the electromagnetic interaction between
electronegative atoms, such as nitrogen (N), oxygen (O), fluorine (F), and
electropositive hydrogen atom. Hydrogen bond mostly happens between molecules
with strength less than 10 kcal/mol, (although made with F may be up to 30 kcal/mol),
43
43
much weaker than covalent bond or metallic bonds. The most common form of
hydrogen bond is found between the water molecules where each oxygen atom can
form two bonds with hydrogen atoms from other two water molecules. Hydrogen
bonds strongly affect the liquid and crystal structures of water, for example, they help
to create the hexagonal structure for ice-Ih with 109.5。angle between two hydrogen
bonds, which is quite close to tetrahedral angle. At the other hand, the van der Waals
interaction is a long range attraction between molecules arising from a force between
the instantaneous induced dipoles. The vdW interaction is much weaker than normal
chemical bonds, but plays an important role in properties of many materials and
defines many properties of organic structure, which we will see next chapter.
Especially, the vdW interaction can have significant influence on the structure when
the intermolecular bond in ice is the relative weak hydrogen bond. Particularly, when
under high pressure, hydrogen bond strengths decrease significantly because the
nearest-neighbor water-water distances increase in comparison with those of ice-Ih
and the hydrogen bonds twist due to configuration distortions[56]. At the same time,
vdW interactions increase because layers of water molecule are packed much closer.
Thus the vdW interactions become more important in determining the properties of
ice structures at higher pressure[14]. This chapter follows closely Ref [57].
3.1.2 Density functional theory for solid ice
As a result, an accurate description of its properties requires the proper treatment of
44
44
these intermolecular interactions[14], over a range of pressures and temperatures, as
can be seen in the phase diagram[55]. Many theoretical methods can be used for those
weak interactions, such as CCSD mentioned in chapter 2, they can be highly accurate
but with expensive calculation and so impossible to apply to bulk solids at present.
Density functional theory can be a good substitute for them. DFT is exact in theory; in
practice some approximations must be made. These approximations often yield
successful result but still have problem in same situation. One of them is that the
conventional functionals mostly fail for the weak bonds. However, recent years, with
the development of new density functionals and vdW dispersion corrections, density
functional theory has become a method of choice for this class of problems.
As discussed in chapter two, in the density functional theory, while the exact
general form of remains unknown, many exact conditions on have been
discovered. Density functional approximations can be developed to satisfy these
known conditions, or to fit data sets, or both. Many authors have employed DFT with
the generalized gradient approximation (GGA)[8,12] to study the properties of liquid
water and ice. Popular GGA functionals are less useful where vdW interactions are
important[58], and this can explain why certain GGAs underestimate the density of
liquid water[58] (PBE by 15%, revPBE[59] by 30%) and the sublimation energy for
high-pressure ice[14] (PBE by 15% for ice VIII) compared to experiment. As a result,
new functionals which more accurately describe both hydrogen bonds and vdW
interactions are strongly needed. The meta-GGA[60] is a natural way to improve
45
45
accuracy further by making use of additional semilocal information (e.g., the
Laplacian of the density and/or the kinetic energy density ). Previous tests
showed that meta-GGAs yield better results than GGAs for covalent, ionic, and
metallic solids, including atomization energy, lattice constant, bulk modulus, and
cohesive energy[13]. Recently, it was found that a meta-generalized gradient
(meta-GGA) approximation to can describe the hydrogen bond[61] and vdW
interactions very well[36]. Motivated by this observation, we apply several
meta-GGA[34] functionals to study the lattice mismatch problem[29] of ice-Ih with
β-AgI, the sublimation energy in three different phases of ice, and the structural phase
transition pressures of ice.
Because the standard semilocal density functionals are unable to describe
the long-range part of the van der Waals interaction, several long-range vdW
corrections have been invented. In our work shown here, we apply two of them, the
DFT+D2[47] and vdW-DF[48] methods, to compare with our meta-GGA results.
Generally, in this chapter, we study the performance of GGA and new
meta-GGA functionals (including TPSS[32], revTPSS[13], and MGGA_MS[35]) on
geometry, sublimation energy, and transition pressures of various ice phases at
absolute-zero temperature, without zero-point vibration effects. Furthermore, by
comparing these results with the ones from density functionals corrected for
long-range vdW interaction (such as TPSS+D2[47], optB88-vdW[62] and
optB86b-vdW[62]), we aim to understand the advantages and limitations of all these
46
46
approximations and the extent to which they account for the contributions of
hydrogen bonding and van der Waals interaction. As we will see later, meta-GGA
functionals reduce the lattice mismatch value of ice-Ih and β-AgI from 6% (PBE) to
around 3%, reasonably close to experiment. Additional calculations for high-pressure
ice phases indicate that the older meta-GGAs TPSS and revTPSS still have trouble
with vdW forces while the newer MGGA_MS is able to describe these interactions
well. However, by adding the appropriate long-range vdW corrections for solids,
revTPSS also tends to work well for the high-pressure phases.
3.2 Computational details
The Vienna Ab initio Simulation Package (VASP)[51,52] in version 5.2.12
has been used for the DFT calculation. VASP is a plane-wave code within the
projector augmented wave (PAW) method (details in chapter two). The “hardest”
PAW potentials available for H and O atoms were used for the sake of high accuracy
in the presence of short O-H bonds[29]. Ice-Ih was modeled using Bernal-Fowler’s
proton-ordered, twelve-water-molecule periodic model[28]. Ice-II and ice-VIII
structures were obtained from experiment: twelve molecules in a trigonal cell for
ice-II[63] and eight molecules in a tetragonal cell for ice-VIII[64]. The energy of the
isolated water molecule was calculated within a 10×11×12 Å3
box.
47
47
FIG. 3-2: unit cell of three different ice phase, oxygen atoms are shown in red and
hydrogen atoms are shown in white
In a convergence test with the PBE and TPSS functional for ice-Ih, the total energies
were computed with the kinetic energy cutoff increasing from 900 eV to 1400 eV and
the Brillouin zone k-mesh from 2×2×1 to 4×4×4.
FIG. 3-3: (a) Convergence of the total energy with respect to energy cutoff used for
48
48
revTPSS for all three ice phases (b) convergence of the sublimation energy with
respect to energy cutoff (c) convergence of total energy with respect to k-mesh.
From FIG 3-3(b), we can see that energy cutoff larger than 700eV is enough to make
the sublimation energy change within 1meV/H2O, while the total energy FIG. 3-3 (a)
needs more than 1100 eV to make the change within 5 meV/H2O to get a precise
equilibrium volume; on the other hand, the energy is less sensitive to the k-mesh,
larger than 2x2x2 is sufficient to make a revTPSS calculation result fully converge for
all three phases. Based on these tests, the optimizations for three ice crystal
geometries and total energies were performed for each functional using a plane wave
basis with a kinetic energy cutoff of 1200 eV and a 4×4×2 k-mesh to ensure
convergence.
All our calculations were self-consistent. The c/a lattice-constant ratios were set to
their experimental values, since the supplemental information of Ref. [ ] says that
optimizing these ratios affected the volume per molecule negligibly (by less than 0.02
Å3/H2O for ice-VIII).
3.3 Results and Discussion
3.3.1 Lattice mismatch challenge for ice-Ih and β-AgI
When two materials with different lattice constants are brought together by
deposition of one material on another, the lattice mismatch is defined as
49
49
[3-1]
where a1 and a2 are the lattice constants for the two separated materials. Lattice
mismatch is a critical parameter for thin film growth on a crystal. For example, large
lattice mismatch will prevent the growth of a defect-free epitaxial film unless the
thickness of the film is below a certain critical thickness[65]. Consequently, good
prediction for lattice mismatch is important for theoretical simulation of such
phenomena.
One application of this is the cloud seeding with β-AgI[66]. In a cloud, an ice
crystal seed can make the crystallization process hundreds times faster than in super
cooled water. In cloud seeding, the crystalline β-AgI can be used to produce artificial
rainfall, because β-AgI smoke provides seed crystals used as the artificial ice nuclei in
clouds for rain-inducing ice crystallization. This application is based on the fact that
the mismatch between the lattice constants of ice-Ih and crystalline β-AgI is only
about 1% at 273 K (~2.2%[29] after extrapolating to low temperature at 10-30 K).
However, Feibelman[29] in his early work pointed out that the lattice constants of
ice-Ih and β-AgI as predicted by LSDA[25] and some GGA-level density functionals
produce a significantly too-large mismatch value (~8% for LSDA,~6% for PBE)
compared to experiment. This so-called lattice mismatch puzzle led to doubts about
using DFT approximations for water-material interactions. Ice-Ih has a hexagonal
crystal structure, in which hydrogen bonds constitute up to 90% of the whole
interaction[14]. On the other hand, β-AgI is a solid with strong van der Waals
50
50
interactions because of the heavy I ions. Therefore, to get an accurate mismatch
value, the functional should be able to describe both the hydrogen bond and the
dispersion interaction simultaneously, a challenge for most semilocal density
functionals.
The first four rows of Fig. 3-4 show the relative errors for lattice constants
and also the lattice mismatch computed from four widely-used GGA functionals.
Among them, PBE gives relatively small errors for both structures, but, because of the
opposite directions of the errors, the mismatch value calculated by PBE is too large.
Conversely, revPBE finds the smallest mismatch among GGAs, but overestimates the
lattice constants for both solids too much. It is known that PBE overestimates
hydrogen bonding and fails for the vdW interactions. This explains why PBE
overbinds ice-Ih while underbinds β-AgI.
We now discuss the results computed by meta-GGA functionals, the older
TPSS and revTPSS and the newer MGGA_MS. We also present the results of three
functionals with long-range vdW corrections: TPSS+D2[47], optB88-vdW[62] and
optB86b-vdW[62]. From Fig. 3-4, we can see that, except for TPSS which performs
similarly to PBE, the other two meta-GGAs show good agreement with experiment:
The relative errors for ice lattice constants are smaller than 1% while the lattice
mismatches are around 3% compared to the experimental value of 2.3%. This
indicates that these meta-GGAs have the potential to better describe the ambient ice
structure.
51
51
Fig. 3-4: Percentage error of lattice constant for the ice-Ih on β-AgI lattice mismatch problem.
Experimental lattice constants and mismatch value are taken from Ref. [28 ].
On the other hand, all functionals with long-range vdW correction give as large a
mismatch value as the GGAs do. Fig. 3-4 shows that, although these functionals give
accurate results for β-AgI, they underestimate the lattice constants of ice-Ih more than
2%, worse than PBE. This shows that, by including the long-range vdW corrections in
these density functionals, one achieves more accurate results for solids with strong
vdW interactions, but less accurate results in ice structures with hydrogen bonds.
52
52
Based on these results, the meta-GGA functionals revTPSS and MGGA_MS
show the best performance overall. However, as vdW interactions play only a minor
role in ice-Ih, studies of properties for different ice phases are needed to understand
the ability of various functionals to describe hydrogen bonds and van der Waals
interactions in ice.
Table 3-1: Lattice constant of ice-Ih and β-AgI, lattice mismatch value for different
density functionals
Functionals ice-Ih lattice constant
(Angstrom)
β-AgI lattice constant
(Angstrom)
Mismatch
(%)
LSDA 4.16 4.50 7.85
PW91 4.39 4.68 6.24
PBEsol 4.28 4.56 6.17
PBE 4.43 4.67 5.36
RPBE 4.6 4.80 4.09
TPSS 4.45 4.69 5.22
revTPSS 4.50 4.64 3.05
MGGA_MS 4.48 4.63 3.27
MGGA_MS2 4.49 4.63 3.07
TPSS+D2 4.37 4.58 4.77
53
53
optB88-vdW 4.41 4.65 5.25
optB86b-vdW 4.39 4.7 4.99
optPBE-vdW 4.48 4.61 4.89
Expt. value 4.497 4.594 2.17
3.3.2 Sublimation energy for ice phases under ambient and high
pressure
As mentioned before, solid ice exhibits a rich and complex phase diagram. We
next report our tests on ice-Ih at ambient pressure, and on two other proton-ordered
phases, in order of increasing pressure: ice-II[63] and ice-VIII[64] (shown in Fig. 3-1)
using GGAs, meta-GGAs, and semilocal functionals with vdW correction. We also
compare our computed results with experiment. Previous work[14] indicates that, for
the phases at higher pressures, hydrogen bond strengths decrease significantly. At the
same time, vdW interactions increase because layers of water molecule are packed
much closer. Thus the vdW interactions play a more crucial role in determining the
properties of ice structures at higher pressure.
Next we discuss the sublimation energies for these three ice phases with different
functionals. The sublimation energy is defined as the difference between the energy of
an isolated water molecule and the energy per water molecule in the solid structure. It
represents the energy change from the solid to the gas phase, including all the
54
54
intermolecular interactions in the solid structure. For ice structures, intermolecular
interactions consist of hydrogen bonds and vdW dispersion forces. By comparing the
results for various ice phases, we can analyze the performance of DFT for these two
weak interactions. Table 3-2 shows the computed results of sublimation energies from
various functionals. The total-energy difference
with respect to the ice-Ih phase is calculated for high-pressure phases
and the results are shown in parentheses. Experiment shows that ice-II is almost as
stable as ice-Ih with only about 1 meV/molecule, while ice-VIII is less stable
than ice-Ih by 33 meV/molecule. From Table 3-2, PBE slightly overestimates the
sublimation energy for ice-Ih, while underestimating it by 40 meV/molecule for ice-II
and by 117 meV/molecule for ice-VIII. The main reason has been explained by
previous work[14]. TPSS and revTPSS yield acceptable sublimation energy for ice-Ih,
but still fail for high-pressure phases. The revTPSS total energy difference between
ice-VIII and ice-Ih is about 5 times larger than experiment due to underestimation of
the sublimation energy for high pressure. Since vdW interactions become stronger
with increasing pressure, this indicates that these semilocal functionals do not
describe vdW well. Therefore, we also show the sublimation energy of the TPSS+D2
method, which is the TPSS meta-GGA with long-range vdW correction. From Table
3-2, we notice that, with the vdW correction, total energy differences ( are
greatly improved compared to GGAs, giving values much closer to experiment.
However, we still notice that the sublimation energies predicted by TPSS+D2 are
55
55
significantly too large, with mean absolute relative error up to 16%. The strong
overbinding can lead to a poor description for structural properties like volume. As we
already found in part I and also in Fig.3-5 (a), TPSS+D2 underestimates the
equilibrium volume for all three phases. The DFT+D2 method is constructed for
molecules and clusters, and seems to over-count the vdW interactions in solid. This
phenomenon can also be found within the vdW-DF method, as we can see for
optB88-vdW in Table 3-2.
For lack of an accurate vdW correction for solid ice, we employed the vdW
data from Ref. [68] which adds the influence of vdW interactions within the scheme
of Tkatchenko and Scheffler[67], as calculated to correct the PBE0 hybrid functional.
Since this method has been shown to be largely independent of the employed DFT
approximation and works well for solids[67][68], we add this correction to our
meta-GGA revTPSS result, and find that the total energy yields precise sublimation
energies (with MARE of 3.5%) and a significant improvement for . However, for
the super high-pressure phase ice-VIII, this revTPSS+vdW predicts an energy
difference U (~75 meV) slightly worse than optB88-vdW (26 meV), in comparison
to experiment (33 meV), but still performs better than the PBE GGA (177 meV).
Finally, the new MGGA_MS works well and performs quite similarly to the
revTPSS+vdW with 4.3% MARE and with 70 meV for ice-VIII. MGGA_MS
has no long-range vdW correction, and seems to capture part of the vdW interaction
in ice by itself.
56
56
Table 3-2: Sublimation energies of ice-Ih, -II, -VIII (omitting zero-point energy effects). The
total-energy differences compared to ice-Ih (in
parentheses) and the mean absolute relative error of the sublimation energy averaged over the
three phases.
a: Experimental values are taken from Ref. [ 68 ], b: with zero-point energy contribution
removed.
Ice-Ih Ice-II Ice-VIII MARE (%)
LSDA 943 896(47) 813(130) 47.5
PBE 636 567(69) 459(177) 10.5
TPSS 587 502(85) 380(207) 18.49
revTPSS 570 507(63) 423(147) 16.67
MGGA-MS 602 586(16) 532(70) 4.3
MGGA-MS2 584 553(31) 505(79) 8.65
PBE0a 598 543(55) 450(148) 11.61
PBE0+vdWa 672 666(6) 596(76) 7.61
revTPSS+vdW 644 630(14) 569(75) 3.5
TPSS+D2 720 690(30) 675(45) 16.11
OptB88-vdW 696 699(-3) 670(26) 15
OptB86b-vdW 706 701(5) 666(40) 15.42
DMCa 605 609(-4) 575(30) 0.39
57
57
Expt.a,b
610 609(1) 577(33)
3.3.3 Transition pressure between ice phases
To better understand how functionals perform for van der Waals interactions,
we go on to study the ice phase transitions under pressure, computing the phase
transition pressure between ice-Ih and ice-II or ice-VIII. Fig.3-5 (c) shows the energy
versus volume curve for ice-Ih and -VIII phases for MGGA_MS, fitted by the
Birch-Murnaghan equation of state[69] :
[3-2]
where are the equilibrium volume, total energy, and bulk modulus,
and B '
0 is the pressure derivative of the bulk modulus.
The equilibrium volume, lattice energy, and bulk modulus for each phase are
also obtained from the same EOS parametric fitting, and the results for volume and
energy are illustrated in Fig. 3-5 (a) & (b). Then the transition pressure can be
obtained by constructing the common tangent line (dotted line in Fig. 3-5 (c)) for the
two EOS-fitted energy-volume curves.
We apply this approach with the tested functionals to get the transition
pressures from ice-Ih to -II or -VIII, and the results are given in Fig.3-5 (d). The
58
58
horizontal axis shows the transition pressure Ptr for Ih to II, and the vertical axis
shows Ptr for Ih to VIII. Because the ice-Ih and -II phases are almost equally stable,
the transition pressure is quite small (~0.02 GPa), and only optB88-vdW and
MGGA_MS give reasonable predictions. All other functionals predict a transition
pressure larger than experiment. For ice-VIII, where the experimental value is 0.44
GPa, optB88-vdW still gives the best result while other functionals with vdW
corrections and MGGA_MS also work well. From Fig. 3-5 (d) for the transition
pressure Ptr, and from Fig. 3-5 (b) for the energy difference , we can see a grouping:
vdW-corrected functionals cluster in a close range around experiment, while GGAs,
TPSS and revTPSS fall farther away from this range. Clearly, adding the vdW
correction contributes to the improvement of transition properties. Also, notice that
the MGGA_MS results fall in the close range with these vdW-corrected functionals.
This indicates that MGGA_MS captures at least part of the vdW interactions in ice.
59
59
Fig. 3-5: (a) Relative lattice volume (△V0) of the high-pressure ice-II and -VIII with respect to
the lattice volume of ice-Ih and (b) relative total energy (△U0) (c) The energy versus volume
curves of the ice-Ih and ice-VIII systems with MGGA_MS. The dotted line is the common tangent
line obtained from the Birch-Murnaghan EOS. The slope of the straight line gives the transition
pressure (Ptr). (d) Transition pressures (Ptr) from ice-Ih to the phases ice-II and -VIII.
PBE0+vdW results and experimental values are taken from Ref. [68 ]. The calculated values do
not include zero-point vibration effects. These effects are removed from the experimental
total-energy changes, but not apparently from the experimental equilibrium volumes. Ref. [ 68]
60
60
Table 3-3: Equilibrium volume of ice-Ih,-II and –VIII, transition pressure from ice-Ih
to –II and –VIII
Ice-Ih
volume
(Å3)
Ice-II
volume
(Å3)
Ice-VIII
volume
(Å3)
Ptr Ice-Ih to
–II
(Gpa)
Ptr Ice-Ih to
–VIII
(Gpa)
LSDA 27.47 21.73 17.57 1.39 2.26
PBE 30.75 24.83 20.62 2.04 3.18
TPSS 30.85 25.03 20.38 2.30 3.84
revTPSS 32.05 25.39 19.62 1.57 2.7
MGGA_MS 31.83 24.82 19.81 0.02 0.82
MGGA_MS2 32.20 25.00 19.58 0.72 1.05
PBE0a 30.98 24.84 20.25 1.45 2.22
PBE0+vdWa 29.88 23.63 19.69 0.16 1.19
optB88-vdW 30.23 23.68 18.98 -0.05 0.37
optB86b-vdW 29.78 23.39 18.74 0.12 1.42
TPSS+D2 29.33 22.41 18.95 0.74 0.61
DMCa 31.69 24.70 19.46 -0.09
Expt.a 32.05 24.77 20.09 0.02 0.44
61
61
3.4 Liquid water
Experimental liquid water structure is measured by x-ray and neutron
diffraction experiments, while there are difficulties in interpreting these data
theoretically. As a result, till now there are still considerable controversies. The
structure of liquid water gained from molecular dynamics simulations using
DFT-PBE, revPBE and BLYP, did not agree well with experiment as shown earlier.
These reflect the needs of more accurate functionals which can better describe
hydrogen-bonds and vdW inside the liquid water and our work suggests the possible
way to achieve this. However, it is noticed that a meaningful and practical AIMD
simulation for liquid water need at least a 64 water molecules systems with a 20 ps
simulation process which is impossible to run for high-level functionals now without
technological and algorithmic breakthrough. Due to this, the problem for liquid still
exists and need to be answered in the future.
3.5 Conclusions
In summary, we have studied hydrogen bond and van der Waals interactions
within various ice structures using different density functionals. First we found that
two meta-GGA's, revTPSS and MGGA_MS, essentially solve the GGA lattice
mismatch puzzle[29] for ice-Ih on β-AgI, and we argued that only a functional like
MGGA_MS, that reliably describes intermediate-range van der Waals interaction as
well as the hydrogen bond, can reliably solve this kind of problem.
62
62
Then we found that meta-GGA yields a better description than GGA for the
sublimation energy and equilibrium volume of low-pressure ice phases, a difficult
problem for semilocal functionals. In particular, meta-GGA can describe these
properties at least as accurately as vdW-corrected GGAs can, even without relying on
vdW dispersion corrections, as demonstrated for MGGA-MS: The results (especially
the volume per molecule) for ice-Ih, –II, and -VIII are in quite good agreement with
experiment, while the sublimation energy for ice-VIII is slightly underestimated but
still improved over GGAs.
We find that MGGA_MS is an accurate method for computing ambient and
high-pressure phases of ice, although it needs a long-range vdW correction under
super-high pressure. We have argued elsewhere[36] that MGGA_MS has the right
dimensionless ingredients to recognize covalent, metallic, and weak bonds. Other
meta-GGAs can be built from these dimensionless ingredients[70], and long-range
vdW[71,72] corrections of lesser importance can be constructed for them. Such
functionals may be useful for many problems, including the problems of liquid water
and of DNA/RNA[36].
63
63
Chapter IV
What forces twist the Deoxyribonucleic Acid?
Interactions between nucleic bases with density
functional theory
4.1 Introduction
4.1.1 Deoxyribonucleic acid structure and nucleic bases
In biology, nucleic acids are large biological molecules which function in
encoding, transmitting and expressing genetic information for almost all forms of life
as well as some nonliving entities on the earth. "Nucleic acid" is the overall name for
the DNA (deoxyribonucleic acid) and RNA (ribonucleic acid). Mostly, DNA
molecules are double-stranded while RNA molecules are single-stranded, and DNA is
the basic genetic instructions for most living organisms except viruses, which use
RNA. As a result, DNA attracts lots of interesting from different scientific fields
including biology, chemistry and physics. Since the first isolation of DNA by a Swiss
physician in 1869[73], a lot of efforts were made to find the structure of DNA
molecules. Based on the X-ray diffraction patterns which showed that DNA had a
regular structure[74], in 1953, James Watson and Francis Crick suggested the
double-helix model[75] of DNA structure which was supported by latter experimental
evidence[76]. Nowadays, this so called double-helix twist DNA model may be the
most well-known biology molecule that can be seen everywhere, even on the cover of
64
64
TIME and Waston, Crick and Wilkins jointly received the Nobel Prize in Physiology
or Medicine in 1962.
FIG. 4-1:The structure of the DNA double helix[77]
As we can see from FIG. 4-1, the DNA molecule is made of two long chains.
Each coils around the center axis, molecule called nucleic bases link to the chains.
The basic unit of DNA is nucleotides which build the DNA by repeating itself through
the space. The sugar and phosphate group in the nucleotide alternatively link with
each other to make the chains (called the backbone of DNA) and nucleic bases are
linked to the sugar.[78] One nucleic base connects with another on different
backbones by hydrogen bonds. With this force the two chains are linked together. One
DNA molecule can contain millions of nucleotides, which means millions of nucleic
65
65
bases[79]. However, there are only four kinds of nucleic bases ,classified into two
types: the purines, A (adenine), G (guanine); the pyrimidines, C (cytosine), T
(thymine) and they are linked following the pair rules; A only pairs with T, G only
pairs with C. In sequences of four possible nucleic base pairs: A:T, T:A, G:C, C:G,
DNA can be well suitable to the storage of enough encoded genetic information to
make every human being different. .
FIG. 4-2: Nucleic bases, A (adenine), G (guanine), C (cytosine), T (thymine)
(source from websit)
4.1. 2 Twist of DNA and DNA base pair steps
The most famous and important structure properties of DNA are its two backbones,
66
66
which run in opposite direction to each other and therefore anti-parallel twist together.
As shown in FIG. 4-3, the B-DNA, which is the most common one found in living
animals, has an average 10 pairs per turn of length 34 Å long, so that is ~36 degree
twist and 3.4 Å vertical distance between every two neighbor nucleic base pairs[80].
The large size of DNA molecule makes it difficult to search for the inside
interactions that build this structure both experimentally and theoretically. However,
with the limited sorts of nuclei bases, it is quite practical and reasonable to understand
the physical properties of stacked nucleic base pairs first, then by repeating this small
part sequentially in the space, a better understanding of the whole structure can be
gained. Moreover, the replicating of DNA and binding interactions with other organic
molecules such as proteins generally happen at the base pair level. As a result, to
interpret the interactions between these nucleic bases is quite important to explain the
microscopic as well as the macroscopic properties of DNA. To do this, the base pair
configuration needs to be studied and precisely described.
67
67
FIG. 4-3 Model of the double helix of B-DNA with 10-10.5 base pairs per turn
Based on the X-ray experimental result, because of the coil of backbones,
geometry displacements may happen between the neighbor pairs, generally, there are
six sorts of displacements called base pair steps which can exist: shift, tilt, slide, roll,
rise, twist, which are shown in FIG. 4-4. For B-DNA, the steps of shift, tilt, slide, roll
are zero or small, with the rise around 3.4 Å and twist around 36 degree. As a result in
my work the influences of the earlier four steps are ignored and only the structural
steps parameters of rise and twist are studied.
68
68
FIG. 4-4: Six possible nucleic base pair steps[81]
It is commonly accepted that there are two sources responsible for the base pair
geometry structure[82]. One is the force from the backbones: the
hydrophobic-hydrophilic interactions with the sugar-phosphate group and solvent
molecule (mostly water)[83,84]. Another is the interactions between the bases which
include the hydrogen bond[85] linking the two adjacent bases from the opposite
backbones and the pi-pi stacking interaction[86] between stacked based pairs.
However, the relative contribution of each is still unclear[82]. In experiment, it is hard
69
69
to distinguish and separately research on each force; in theory, for the reason of the
complex structure of backbones and liquid solvent it is difficult to impose the
influence of the sugar-phosphate backbones on the nucleic base pairs. Traditionally,
the first principle theoretical calculations mostly apply to the base pairs without the
backbones.
4.1.3 Ab initio methods for nucleic bases
Since the noncovalent weak interactions between DNA and other
bio-molecules such as RNA and proteins are important, they play a crucial role in the
transcription and replication of nucleic acid, interaction of DNA with proteins and
RNA, and intercalation of chemotherapy drugs to DNA. As the result of this, the
ability to describe these interactions precisely is a basic requirement for theoretical
models which can be use for understanding gene mutation, drug design and so on.
Wave function based methods such as MP2, CCSD are widely used in
quantum chemistry calculations and these methods often can give relatively reliable
results for the interactions between molecules, especially these weak ones.[87] Their
accuracy relies on large basis set which leads to huge computational demands, (O(N5)
for MP2, O(N6-7
) for CCSD(T)). In this sense, their application is limited. The high
level CCSD(T) using triple-zeta basis set, which can work as the benchmark for other
ab initio methods, only become available in recent years for small size of molecule
system[88].
70
70
On the other hand, density functional theory (DFT) is an efficient and often
accurate-enough way which scales well with system size. The GGA PBE and
hydrid-B3LYP also are widely used for studying many chemical systems with tens or
even hundreds of atoms. However, the application of these conventional density
fucntionals is limited due to the poor results[82] of these functionals to describe
systems in which the van der Waals dispersion forces are significant. In recent years,
several new approximations are designed for improving existing density functionals to
deal with this dispersion problem, as we introduced in chapter two, and some of them
like vdW-DF[48] are well tested in many systems like benzene dimer, grapheme
layers, stacked nucleic bases and base pairs[89]. These tests suggest that the
dispersion force plays an important role in determining the base pair steps of rise and
twist. Based on the result we got in the last chapter, the performance of these
meta-GGA functionals (especially the newly developed MGGA_MS) needs to be
analyze for this system with a combination of different weak interactions including
hydrogen bond, dispersion force, as well as electrostatic interactions. Besides, the
DFT+D method, in which the dispersion interactions are added by empirical pair
interaction term, and M06-L fucntionals, which fitted its empirical parameters to
databases that includes weak-interaction dimers, have shown promise for weak
noncovalent interactions, but have not yet been fully applied to these biologic
problem, unlike the vdW-DF method. As a result, in the following, we analyze the
performance of early mentioned meta-GGA fucntionals along with M06-L and
71
71
DFT+D2 for nucleic base pair steps.
In the following part of this chapter, the presence of a colon ( : ) represents the
hydrogen bond interactions between nucleic bases, a hyphen ( - ) represents the
stacking interactions.
4.2 Computational details
All DFT calculations are performed using the Gaussian program[53];
Gaussian is a computer package for chemical calculation using Gaussian orbitals
(more details introduced in chapter two). All GGA level functionals and most
meta-GGA functionals calculations are performed with Gaussian 03, while the M06-L
functional is only available above the version Gaussian 09. I have ensured that the
result obtained by Gaussian 03 and Gaussian 09 for the same functionals are exactly
same as shown in FIG. 4-5. The basis set 6-311++G** is used for all calculation based
on the energy convergence test for C-C stacking pair.
72
72
FIG. 4-5: Total energy of Guanine dimer with rise from 3.4 Å to 4.0 Å
calculated with GGA PBE using Gaussian 03 and Gaussian 09
73
73
FIG. 4-6 Convergence tests of total energy with respect to basis set used for
the GGA PBE calculation for (a) Cytosine nucleic base (b) stacked Cytosine dimer
The initial structure of A (adenine), G (guanine), C (cytosine), T (thymine)
and A:T, G:C pairs are following the nomenclature and coordinate system of Olson et
al., used in the Nucleic Acid Database[90]. The equilibrium structure for each
functional of individual nucleic bases and pairs are fully optimized using each
functionals except for the DFT+D2 method, for the lack of SCF process for it, in
which the van der Waals corrections are calculated based on the optimized DFT
structure. For example, the vdW correction energy of PBE+D2 is computed using the
PBE structure.
4.3 Results and Discussion
As discussed earlier, the geometric and energetic properties of nucleic bases
are determined by the hydrogen bond as well as the pi-pi stacking interaction. The
theoretical investigation makes it possible to isolate and study the influence and
relative contribution of each interaction. In this case, first of all, the hydrogen bonds
between WC pair A:T and G:C are researched, then the stacking interactions are
investigated with the dependence of the interaction energy on twist of undisplaced
stacked nucleic base dimers using various functionals. Furthermore, later in this
chapter, the interactions between stacked nucleic base pairs are studied to search for
74
74
the reason for helical twist angle of DNA.
Since the experimental data from the X-ray crystallography are influenced by the
molecular environment, usually, there are discrepancies between theoretical prediction
of gas-phase structure and experimental values. Therefore, our DFT computed results
are compared with the mostly accurate wave function method which can be found in
literature.
Note that, even the influence of backbones is not including in the most
theoretical models, X-ray data shows that they also contribute in determining the
three-dimensional structure.[83,84]
4.3.1 Hydrogen bond interaction between Watson-Crick pairs
FIG. 4-7: Nucleic base pairs and hydrogen bonds (dot lines) between bases
The hydrogen bond interactions in DNA are dominated by the electrostatic
75
75
contribution, the nature of which is discussed in chapter three. The network of
hydrogen bond inside the DNA is based on the relative interaction between nucleic
bases which build the WC pairs. As shown in FIG. 4-7, two hydrogen bonds occur
between the adenine and the thymine base pairs, and between the cytosine and the
guanine there are three hydrogen bonds. The double helical structure of DNA is due
largely to hydrogen bonding between these base pairs, which link one complementary
strand to the other, besides the role in determining the three dimensional structure, the
hydrogen bonds also contribute to the replication of DNA and adoption of protein and
base pairs because the weak strength of this bond makes it easy to break and reform.
As a result, these hydrogen bonds have been the subject of many
investigations both experimental and theoretical. Previous work shows that both the
high cost quantum chemistry methods and the DFT work successfully in the
description of the hydrogen bond inside DNA structure. The popular GGA-level DFT
can well describe this interaction. In my work, the meta-GGA level functionals
revTPSS, MGGA_MS and M06L are applied to the nucleic base pairs A:T and G:C to
check if this success can be maintained. Besides, it is believed that the van der Waals
dispersion force, even very weak, also contribute to the bond strength of base pairs,
hence, in this part, vdW corrections are also included using Grimme’s DFT+D2
methods to check the relative contribution of dispersion force.
4.3.1.1 Bond length and bonding energy for different density functionals
76
76
In Table 4-1, the performances of various DFT fucntionals on bond parameters
within Watson-Crick base pairs A:T and G:C are given, together with the MP2
benchmark result. The bond length D is defined as the distance between the two
non-hydrogen atoms (see FIG. 4-7); the bonding energy Ebond is computed by
subtracting the total energy of individual bases from the total energy of the base pairs.
All energies are gained by full optimization for every structure.
Table 4-1 shows that, except for the LSDA functional other GGA and meta-GGA
functionals give relative close agreement with MP2 value for bond length and
bonding energy with different level. It is widely known that the LSDA functional
overestimates the hydrogen bond strength up to 50%, which can be proved with our
result: the bonding energy is significantly overestimated and the bond length is much
smaller than the MP2 value. Concerning the different performance of other
functionals for the bonding energy, it is the M06L that gives the best predictions with
the values of -14.24 kcal/mol and -26.87 kcal/mol for A:T and G:C, which is only
~0.8 kcal/mol smaller than MP2[89]. The PBE and MGGA_MS functionals work also
quite well with bonding energy 1-1.5 kcal/mol smaller, while revTPSS performs the
worst with ~3 kcal/mol error. However, all functionals give absolute error close or
smaller than chemical accuracy 1 kcal/mol per bond. As for the bond length, PBE
gives the best description of the geometry, while other functionals overestimate the
bond length by 0.03 Å-0.06 Å comparing to MP2 results but only have smaller than 1%
influence on 20 Å width of the base pairs.
77
77
Table 4-1: Hydrogen bond length D and the bonding energy Ebond for the nucleic base
pairs A:T and G:C for different density functionals as long as MP2 method[89]
Pair Bond
Parameters
LSDA PBE MGGA_MS revTPSS M06L MP2
A:T DN1-N3 (Å) 2.64 2.83 2.86 2.89 2.87 2.83
DN6-O4 (Å) 2.73 2.89 2.94 2.92 2.91 2.86
Ebond
(kcal/mol)
-24.91 -13.90 -13.43 -12.37 -14.24 -15.1
G:C DO6-N4 (Å) 2.60 2.76 2.80 2.80 2.80 2.75
DN1-N3 (Å) 2.77 2.92 2.94 2.95 2.94 2.90
DN2-O2 (Å) 2.74 2.90 2.92 2.93 2.91 2.89
Ebond
(kcal/mol)
-40.30 -26.91 -26.33 -24.45 -26.87 -27.5
4.3.1.2 The contribution of vdW dispersion force for bonding between base pairs
Based on the result shown in Table 4-1, mostly, GGA and meta-GGA
functionals underestimate the bonding energy and overestimate the bond length; this
78
78
indicates that there are at least some van der Waals dispersion interactions missing
from these functionals. To take into consideration and check the role of vdW, the
correction energies are added following Grimme’s DFT-D2 approach using pair wise
atom interactions. In FIG. 4-8, the mean binding energies per hydrogen bond for three
common used density functionals with and without the D2 correction are shown.
FIG 4-8: Mean binding energy for nucleic base pairs with density functionals
with and without vdW D2 corrections.
By adding the vdW dispersion correction, different functionals perform
variously. PBE+D2 method overestimates the mean binding energy by 0.9 kcal/mol,
79
79
which is even larger than the magnitude underestimated by PBE itself. Form chapter
three, it is found that PBE usually overestimate the strength of hydrogen bond, this
can lead to the over counting after considering the vdW energy. At the other hand,
hybrid B3LYP and meta-GGA revTPSS perform well with the correction. The
absolute error decreases from ~1.1 kcal/mol to ~0.6 kcal/mol because of the better
description of hydrogen bond of these two. However, adding the vdW correction for
these functionals always leads some overestimation for the bonding energy.
4.3.2 pi-pi stacking interactions between stacked bases
Pi-pi stacking interactions are usually defined as the attractive interactions
between two parallel or face-to-face oriented aromatic systems. These interactions
play a fundamental role in many aspects of chemistry and biology[84,91]. For
instance, face-to-face pi-pi stacking interactions are responsible for the slippery feel
of graphite layers; pi-pi stacking in biology molecules is often important to the
structure and function of protein, RNA, DNA[92]. In fact, considering the huge
number of aromatic rings in the large biology molecules such as DNA, the individual
weak stacking effect can add up to powerful force that dominates the structure of
them. They are often believed to be the hand at work in the self-assemble[93] of
bio-molecules.
Though the importance of the pi-pi stacking interactions, the first model[91] of
them in which the electrostatics is thought to be major part, was proposed not until
80
80
1990 and has been criticized by numerous research groups since then. Nowadays, it is
well accepted that the interactions are mainly determined by the interplay of
electrostatic effects, the van der Waals dispersion force, the exchange
(Pauli)-repulsion at short intermolecular distance and so on, however, there are still
los of debates on which one is dominates[94]. It would seem that the relative
contributions are highly dependent on the geometry of the molecules and the design
of the experiment and this phenomenon is still not well understood theoretically yet.
In a recent paper[95] of Grimme, he suggests that:
“The terms pi stacked or pi-pi stacking should merely be used as geometrical
(structural) descriptors but that specific (distinguished) attractive interactions
between pi electrons in such system do not exist.”
Density functional theory can accurately describe the electrostatic interactions
and the exchange repulsion, but those conventional GGAs, meta-GGAs cannot treat
the vdW part well, as shown in the former part of this thesis. Despite the know
shortcomings, it is still quite interesting to examine the performance of DFT on these
pi-pi stacking system, especially the new MGGA_MS, which is believed to capture
part of mid-range vdW interactions.
4.3.2.1 Benzene dimer stacking
The benzene (C6H6) dimer is the prototypical system for the study of pi-pi
stacking and has been quite extensively studied. The experimental binding energy is
81
81
about 2-3 kcal/mol in gas phase[96]. This significantly weak binding energy makes
the benzene dimer difficult to study experimentally. It is also to be a challenging task
for theoretical methods to describe it accurately. Besides, it is the simplest example of
a pi-pi stacking system which can be found everywhere in biological molecules; as a
result, the high-level CCSD(T) method with very large basis set can be applied to
benzene dimer to get a quite accurate benchmark for other theoretical methods[97].
FIG. 4-8: Three possible dimer geometries and stacking energies[98]
FIG. 4-8 shows three possible configurations of benzene dimer: perfectly parallel
in sandwich shape (S), parallel displaced (PD), and edge-to-face or T-shape (T), and
their stacking energies with CCSD(T) at the complete basis set limit. These most
82
82
accurate calculations prove the existence of this significantly weak binding energy.
Following, the ability of various density functionals for this system are investigated
based on S-shape (FIG. 4-8) structure.
FIG. 4-9: Stacking energies for S-shaped benzene dimer with 3.9 Å distance
calculated with various density functionals and MP2[99] compared to CCSD (T)
value (blue line).
The conventional functionals including GGA PBE, meta-GGA revTPSS, and
hybrid-B3LYP, are unable to give a proper description of the benzene dimer: they
predict a purely repulsive interaction, and as a result no binding position can be
83
83
found. MGGA_MS improves this but can only capture small a part of the binding
energy with a large binding position at r=4.4 Å. M06L works well for this stacked
system with small energy difference and slight underestimation of bond position (3.8
Å), which is quite reasonable because the parameters of M06L are fitted to the
database (PPS5/05)[33] including the benzene dimer. The widely used quantum
chemical method MP2 overestimates the energy largely. In fact, based on previous
work, MP2 mostly tent o overestimate the pi-pi stacking energy[100]. The problem
of these density functionals indicates the need of new correction. Here I employed
the PBE+D2 vdW corrections and add it to the GGA PBE energy.
84
84
FIG. 4-10: Binding curve (Etot) of benzene dimer obtained by adding DFT-PBE (EPBE)
energy to PBE-D2 dispersion correction (ED2-disp).
From FIG. 4-10, we can notice that for GGA PBE functionals by adding the
dispersion correction, even in a simple approach using only pairs of atoms, can greatly
improve the binding curve for the benzene dimer: the unbinding problem can be solved;
a global minimum at 3.8 Å with binding energy 1.8 kcal/mol can be gained, only
slightly larger than CCSD method. This indicates the importance of vdW interaction in
the pi-pi stacking at least for the benzene dimer system.
Because the symmetry of geometry and electrostatic potential of benzene, (see FIG.
4-11 (a)), there are no net dipole in benzene, existence of quadrupole-quadropole
interactions leads to the weak vdW force and these vdW and electrostatic (ES) forces do
not tend to change too much when one benzene is rotated above the other one. However,
in the nucleic bases, situation can be more complex because there is no symmetry of the
electron density distribution (see FIG. 4-11(b)), vdW and ES forces may change largely
due to the rotation of one base above another which lead to a much more complex
interplay of these forces.
85
85
FIG. 4-11: Electrostatic potential surface for (a) benzene (b) adenine
4.3.2.2 Rise and potential energy surface (PES) for stacked nucleic bases
(i) Rise for stacked base pairs
Rise for nucleic bases is defined as the vertical distance between the mass centers
of two stacked antiparallel bases (see FIG. 4-12) because of the high repulsion
existence in parallel structures. To get the value of rise, the individual base structures
are optimized first with different functionals, then the intramolecular structure are
kept frozen in all the methods and only the intermolecular coordinates are varied to
find the minimum energy.
86
86
FIG. 4-12: Rise defined as the vertical distance between stacked antiparallel C-C
nucleic bases
Rise for stacked bases are strongly determined by the bond interactions between
them, in TABLE, the rise for seven stacked pairs of A-A, C-C, G-G, T-T, A-G, A-C,
G-C are shown. The mean relative error (MRE) are gain by comparing to the rise
value of 3.4 Å for C-C, G-G, G-C and 3.3 Å for others, based on the previous work of
Sponer[101].
TABLE 4-2: Nucleic bases rise (Å) computed with several density functionals
and their MRE (Å) compared to Sponer work.
LSDA PBE revTPSS MGGA_MS M06L
A-A 3.28 4.08 4.30 3.71 3.48
87
87
C-C 3.35 3.93 4.00 3.70 3.56
G-G 3.15 3.81 4.08 3.51 3.34
T-T 3.10 3.75 3.93 3.54 3.35
A-G 3.18 4.02 4.30 3.63 3.40
A-C 3.25 3.97 4.30 3.68 3.46
G-C 3.23 4.1 4.30 3.62 3.45
MRE -0.12 0.60 0.83 0.28 0.10
PBE and revTPSS significantly overestimate the value of rise by 0.6 Å and
0.83Å, MGGA_MS performs much better with 0.28 Å MRE, LSDA and M06L still
work best for the rise value with MARE about 0.1 Å. The larger bond of PBE and
revTPSS is due to the lack of dispersion interaction. By adding the D2 correction to
these functions, the binding energy and position can be greatly improved as shown in
FIG. 4-13.
88
88
FIG. 4-13: The binding curve for PBE, revTPSS, MGGA_MS and D2 correction
functions together with the MP2[99] equilibrium rise value and binding energy.
(Experiment rise at 3.35±0.19 Å)
(ii) Twist and Potential energy surface
In this part, the stacking interactions are studied as a function of the twist angle
between two nucleic bases. The twist between the bases is defined as the higher base’s
right hand rotation around the axe that links the mass center of both higher and lower
bases. (Shown in FIG. 4-14)
89
89
FIG. 4-14: The definition of twist between two bases and four configurations of C-C
90
90
twist from parallel to anti parallel
The initial position of twist or the parallel configuration with twist angle of 0。is
defined as glycosidic bonds are parallel. In the paper of Wu and Yang[102], the
stacking energy was calculated as a function of the twist angel referred to as potential
energy surface (PES) for twist bases, which was chosen as a reliable reference method
to study the interactions between these stacked nucleic bases. Since thymine was not
included in Yang’s work, it is not fully test latterly, thus, not all results of possible
thymine dimers are present here as the lack of reference.
The size of nucleic base is much larger than benzene, as a result, there are
technical difficulties to involve CCSD(T) method to obtain accurate value. Until now,
the most reasonable PES was obtained by employing the MP2/6-31G*(0.25)
method[103], which was confirmed by more accurate studies at anti-parallel
configuration[101,104]. As shown before, MP2 method tends to overestimate the pi-pi
stacking, however, it is showed that this is counteracted by using a medium basis set
with a more diffuse polarization function can lead to only about 1 kcal/mol error with
much less computing cost[88]. In my work, MP2/6-31G*(0.25) results are used to
compare with density functionals.
From the previous result of benzene dimer, the rise for stacked bases, it is
evident that the conventional density functionals considered here cannot well describe
the pi-pi stacking interactions, which should lead to a weak binding ,shallow PES. In
FIG. 4-15, the PES for four stacked systems A-A, C-C, G-G, G-C are presented,
91
91
which are representatives for most possible base pairs.
FIG. 4-15: Stacking energy as a function of twist angle for four stacked bases
computed with various density functionals at the optimized rise and together with the
MP2/6-31G*(0.25) reference data[101].
FIG. 4-15 indicates that PBE and revTPSS cannot correctly reproduce the
stacking energy (either B3LYP, proved by much previous work) while MGGA_MS
predicts quite improved result with respect to MP2. However, at least, the trend and
qualitative features for the energy changing with rotation can be shown by these
functionals with an up shift of the energy. Besides, FIG 4-15 also implies that the
92
92
differences between the maximum and minimum binding energy for these bases
dimers are 6-8 kcal/mol same energy level as hydrogen bond, which suggests that the
pi-pi stacking interaction energy inside the nucleic bases strongly depends on the
twist angle. Configurations with the twist angle close to 0° have the relative small
binding energy due to the overlap orientation leading to largest electrostatic
repulsion, while the configurations with large twist angle have the greater binding
energy but the angles with greatest binding energy differ for various dimers.
It is obvious that the vdW corrections are strongly needed for conventional
functionals to well describe the pi-pi stacking interactions between stacked nucleic
bases. FIG 4-16 shows the corresponding DFT+D2 potential energy surface and it is
evident that they can perform much better with respect to MP2.
FIG 4-16: PES for C-C, G-G calculated with PBE+D2 and revTPSS+D2
Based on these, to well describe the stacking interactions between the nucleic
93
93
bases, the vdW dispersion correction must be included for the classical PBE or
revTPSS functionals. MGGA_MS shows the potential to capture part of this. In the
following section, the interactions between stacked base pairs A:T and G:C are
studied under these functionals.
4.3.3 Stacking interactions for stacked nucleic pairs and
possible reason for the twist of DNA from the view of PES
Previous work in this chapter indicates the stacking interactions strongly
determine the DNA steps rise and twist and a precise description of vdW dispersion
forces is needed for such system. In this part, the DFT+D2 and MGGA_MS
functionals will be used to determine the preferred stacking configurations for four
kinds of W-C DNA nucleic base pairs by studying the stacking energy between them
as a function of rise as well as twist angle. With this, the deeper understanding on the
performance of these functionals for bio-molecules can be gain, the question that what
forces twist DNA can be possibly answered.
94
94
FIG 4-17: Stacking energy at 36° twist calculated with various density
functionals and compared with Sponer et al., in which a standard high-level CCSD(T)
based technique was extrapolated to the complete basis set limit.
The weaker stacking energy leads to large value of rise, which is impossible
inside the DNA, as a result, the energies as a function of rise at the 36° twist angle are
evaluated and compared with the CCSD(T) calculations of Sponer et al[104]. FIG.
4-17 shows the stacking energy for different functionals at their optimized rise and 36°
twist angle and compared with a high-level CCSD(T) calculation. It indicates that for
stacked nucleic base pairs, MGGA_MS can only capture about one third of the
binding energy, which can leads to too long rise vale. On the other hand, the results of
95
95
revTPSS with the D2 correction are consistent with Sponer’s work. These implies that
the effect of long range dispersion forces plays a bigger role in the stacking energy
with larger molecule size and more aromatic rings which can explain why
MGGA_MS performs much worse in stacked pairs than in stacked bases.
Next, we should turn to discuss of the effect of different twist angle on stacking
energy. FIG 4-18 presents the stacking energy of three typical stacked nucleic base
pairs and it indicates that the energy change significantly when twist angle smaller
than 20° and larger than 40°, while there are possible angular stability can be seen
between 25° and 40°. And the final twist angle at 36° should be more influenced by
backbones and solvent in this range.
FIG.4-18: Stacking energy as a function of twist angle for stacked base pairs
AA:TT, GG:CC, AG:CT with revTPSS+D2.
96
96
Based on previous work, it suggests the fact that the stacking interactions
significantly contribute to define the twist phenomenon of DNA: the minima of
stacking interaction energy versus twist angel determine a reasonable range of twist
and the other forces such as backbone-solvent interaction define the final angle.
Furthermore, we can perform energy decomposition analyses based on DFT+D2
results to distinguish the determination of various interactions. As shown in FIG. 4-18,
if we treat the revTPSS energy as the electrostatic (ES) part and D2 correction as the
dispersion interactions part, the sum of them yield the total stacking energy. For
GG:CC pairs, it suggests that although the change in ES is two times larger than the
dispersion part, (17 kcal/mol compared to 8 kcal/mol), neither of them can determine
a stable angle by itself, only by taking into consideration both part of interactions, a
energy minimum can be found.
97
97
FIG. 4-19: Contributions to the revTPSS+D2 total energy as a function of the
twist angle for the stacked GG:CC pairs
4.4 Conclusion
In this chapter, we present the study of hydrogen bond and pi-pi stacking
interaction between nucleic bases systems using a variety of levels of density
functionals, in the absence of contribution from DNA backbones and any solvent. The
hydrogen bonds are the major force connecting the nuclei bases on different
backbones while most density functionals can well describe them, however, adding
vdW correction tend to improve the results. The pi-pi stacking between the stacked
98
98
nucleic base pairs are slightly weaker than hydrogen bonds, but more important in
defining the configuration of pairs by significantly changing the stacking energy with
the different rise and twist angle, as shown in the EPS in FIG. Our results indicates
that this is due to the interplay of mid-long range van der Waals interactions and
electrostatic forces, while the latter one can be well described by DFT, early part,
especially the long-range vdW, is still a hard challenge. Classical functionals, such as
PBE, revTPSS, B3LYP cannot capture these, but the newly developed MGGA_MS
can capture the intermediate part, and leads to much better results but sill misses the
long-range energy with underestimation of the stacking energy. Furthermore, the
DFT+D2 methods are applied to the stacking system with the significant
improvement. Taking into consideration the important role of vdW dispersion forces,
a more accurate method is strongly needed.
In summary, based on the performances for both hydrogen bonds and stacking
interactions, the hybrid functional built with MGGA_MS or revTPSS with an
appropriate long-range vdW correction are good next steps in the development of a
complete density functional for the understanding of more complex molecular
structures.
99
99
References
[1] F. Darwin " The life and letters of Charles Darwin, including an autobiographical
chapter", (1887).
[2] G. Gilli and P. Gilli, "The nature of the Hydrogen bond", (2009).
[3] L. Pauling, J. Am. Chem. Soc. 57, 2680 (1935).
[4] V. F. Petrenko and R. W. Whitworth, Physics of Ice (2003).
[5] S. Wolfram "Principle of Nucleic Acid Structure", (1984).
[6] S. V. Aradhya, M. Frei, M. S. Hybertsen and L. Venkataraman, nature Material
(2012).
[7] W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 (1965).
[8] J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996).
[9] A. D. Becke, J. Chem. Phys 98, 5648 (1993).
[10] P. J. Stephens, F. J. Devlin, C. F. Chabalowski and M. J. Frisch, J. Phys. Chem.
98, 11623 (1994).
[11] K. Burke, J. Chem. Phys. 136, 150901 (2012).
100
100
[12] J. P. Perdew and W. Yue, Phys. Rev. B 33, 8800 (1986).
[13] J. P. Perdew, A. Ruzsinszky, G. Csonka, L. A. Constantin, and J. Sun, Phys. Rev.
Lett. 103, 026403 (2009).
[14] B. Santra, J. Klimeš, D. Alfè, A. Tkatchenko, B. Slater, A. Michaelides, R. Car,
and M. Scheffler, Phys. Rev. Lett. 107, 185701 (2011).
[15] E. Schrödinger, Phy. Rev. 28, 1049 (1926).
[16] M. Born and J. R. Oppenheimer, Ann. phys. 84, 457 (1927).
[17] A. K. Rappe and E. R. Bernstein, J. Phys. Chem. A 104, 6117 (2000).
[18] L. H. Thomas, Proc. Camb. Phil. Soc. 23, 542 (1927).
[19] E. Fermi, Accad. lincei. 6, 202 (1927).
[20] P. Hohenberg and W. Kohn, Phy. Rev. 136, 864 (1964).
[21] W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 (1965).
[22] J. P. Perdew and A. Ruzsinszky, "Density Functional Theory of Electronic
Structure: A short Course for Mineralogists and Geophysicists", (2010).
[23] J. P. Perdew and K. Schmidt, in Density Functional Theory and Its Application to
Materials, edited by V. Van Doren, C. Van Alsenoy, and P. Geerlings (2001).
101
101
[24] J. P. Perdew and Y. Wang, Phys. Rev. B 45, 13244 (1992).
[25] J. P. Perdw and A. Zunger, Phys. Rev. B 23, 5048 (1981).
[26] S. J. Vosko, L. Wilk and M. Nusair, Can. J. Phys. 58, 1200 (1980).
[27] G. I. Csonka et al., Phys. Rev. B 79, 155107 (2009).
[28] D. R. Hamann, Phys. Rev. B 55, R10157 (1997).
[29] P. J. Feibelman, Phys. Chem. Chem. Phys. 10, 4688 (2008).
[30] K. Capelle, Braz. J. Phy. 36, 1318 (2006).
[31] J. P. Perdew, A. Ruzsinszky, G. I. Csonka, O. A. Vydrov, G. E. Scuseria and L.
A. Constantin, Phy. Rev. Lett. 100, 136406 (2008).
[32] J. Tao, J. P. Perdew, V. N. Staroverov, and G. E. Scuseria, Phys. Rev. Lett. 91,
146401 (2003).
[33] Y. Zhao and D. G. Truhlar, J. Chem. Phys. 125, 194101 (2006).
[34] J. P. Perdew and S. Kurth, Phy. Rev. Lett. 82, 12 (1999).
[35] J. Sun, B. Xiao, and A. Ruzsinszky, J. Chem. Phys. 137, 051101 (2012).
102
102
[36] J. Sun, B. Xiao, Y. Fang, R. Haunschild, P. Hao, A. Ruzsinszky, G.I. Csonka,
G.E. Scuseria, and J.P. Perdew, arXiv:1303. 5688.
[37] A. D. Becke, Phys. Rev. A 38, 3098 (1988).
[38] C. Lee, W. Yang, and R. G. Parr, Phys. Rev. B. 37, 785 (1988).
[39] C. Adamo and V. Barone, J. Chem. Phys. 110, 6158 (1999).
[40] D. Bohm and D. Pines, Phys. Rev. 92, 609 (1953).
[41] D.C.Langreth and J. P. Perdew, Phys. Rev. B 21, 5469 (1980).
[42] Z. Yan, J. P. Perdew and S. Kurth, Phys. Rev. B 61, 16430 (2000).
[43] A. Ruzsinszky, J. P. Perdew and G. I. Csonka, J. Chem. Theory Comput. 6, 127
(2010).
[44] J. Tao, J. P. Perdew and A. Ruzsinszky, PNAS 109(1), 18 (2012).
[45] A. Tkatchenko, L. Romaner, O. T. Hofmann, E. Zojer, C. Ambrosch-Draxl and
M. Scheffler, MRS Bull. 35, 435 (2010).
[46] E. R. Johnson, I. D. Mackie and G. A. DiLabio, J. Phys. Org. Chem. 22, 1127
(2009).
[47] S. Grimme, J. Comp. Chem. 27, 1787 (2006).
103
103
[48] M. Dion, H. Rydberg, E. Schröder, D. C. Langreth, and B. I. Lundqvist, Phys.
Rev. Lett. 92, 246401 (2004).
[49] J. Grafenstein and D. Cremer, J. Chem. Phys. 130, 124105 (2009).
[50] S. Grimme, J. Antony, S. Ehrlich and H. Krieg, J. Chem. Phys. 132, 154104
(2010).
[51] G. Kresse and J. Hafner, Phys. Rev. B 47, 558 (1993).
[52] G. Kresse and J. Furthmüller, Phys. Rev. B 54, 11169 (1996).
[53] Gaussian 03, Revision C.02, M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E.
Scuseria, M. A. Robb, J. R. Cheeseman, J. A. Montgomery, Jr., T. Vreven, K. N.
Kudin, J. C. Burant, J. M. Millam, S. S. Iyengar, J. Tomasi, V. Barone, B. Mennucci,
M. Cossi, G. Scalmani, N. Rega, G. A. Petersson, H. Nakatsuji, M. Hada, M. Ehara, K.
Toyota, R. Fukuda, J. Hasegawa, M. Ishida, T. Nakajima, Y. Honda, O. Kitao, H.
Nakai, M. Klene, X. Li, J. E. Knox, H. P. Hratchian, J. B. Cross, V. Bakken, C.
Adamo, J. Jaramillo, R. Gomperts, R. E. Stratmann, O. Yazyev, A. J. Austin, R.
Cammi, C. Pomelli, J. W. Ochterski, P. Y. Ayala, K. Morokuma, G. A. Voth, P.
Salvador, J. J. Dannenberg, V. G. Zakrzewski, S. Dapprich, A. D. Daniels, M. C.
Strain, O. Farkas, D. K. Malick, A. D. Rabuck, K. Raghavachari, J. B. Foresman, J. V.
Ortiz, Q. Cui, A. G. Baboul, S. Clifford, J. Cioslowski, B. B. Stefanov, G. Liu, A.
104
104
Liashenko, P. Piskorz, I. Komaromi, R. L. Martin, D. J. Fox, T. Keith, M. A.
Al-Laham, C. Y. Peng, A. Nanayakkara, M. Challacombe, P. M. W. Gill, B. Johnson,
W. Chen, M. W. Wong, C. Gonzalez, and J. A. Pople, Gaussian, Inc., Wallingford CT,
2004., .
[54] A. Møgelhøj, A. K. Kelkkanen, K. T. Wikfeldt, J. Schiøtz, J. J. Mortensen, L. G.
M. Pettersson, B. I. Lundqvist, K. W. Jacobsen, A. Nilsson, and J. K. Nørskov, J.
Phys. Chem. B 115, 14149 (2011).
[55] C. G. Salzmann, P. G. Radaelli, E. Mayer, and J. L. Finney, Phys. Rev. Lett. 103,
105701 (2009).
[56] S.L. Dong, Y. Wang, A.I. Kolesnikov, and J.C. Li, J. Chem. Phys. 109, 235
(1998).
[57] Y. Fang, B. Xiao, J. Tao, J. Sun and J. P. Perdew, submitted to Phy. Rev. B
(2013).
[58] R. O. Jones and O. Gunnarsson, Rev. Mod. Phys. 61, 689 (1989).
[59] Y. Zhang and W. Yang, Phys. Rev. Lett. 80, 890 (1998).
[60] J. Sun, M. Marsman, G. Csonka, A. Ruzsinszky, P. Hao, Y. Kim, G. Kresse, and
J. P. Perdew, Phys. Rev. B 84, 035117 (2011).
105
105
[61] P. Hao, J. Sun, B. Xiao, A. Ruzsinszky, G. Csonka, J. Tao, S. Glindmeyer, and J.
P. Perdew, J. Chem. Theory Comput. 9, 355 (2013).
[62] J. Klimeš, D. R. Bowler, and A. Michaelides, Phys. Rev. B 83, 195131 (2011).
[63] C. Lobban, J. L. Finney, and W. F. Kuhs, J. Chem. Phys 117, 3928 (2002).
[64] W.F. Kuhs, J.L. Finney, C. Vettier, and D.V. Bliss, J. Chem. Phys. 81, 3612
(1984).
[65] R. People and J. C. Bean, Appl. Phys. Lett. 47, 322 (1985).
[66] B. Vonnegut, J. Appl. Phys. 18, 593 (1947).
[67] A. Tkatchenko and M. Scheffler, Phys. Rev. Lett. 102, 073005 (2009).
[68] A. Tkatchenko, R. A. DiStasio Jr., R. Car, and M. Scheffler, Phys. Rev. Lett. 108,
236402 (2012).
[69] A. Otero-de-la-Roza and V. Luaña, Comput. Phys. Comm. 182, 1708 (2011).
[70] J. Sun, R. Haunschild, B. Xiao, I.W. Bulik,G.E. Scuseria, and J.P. Perdew, J.
Chem. Phys. 138, 044113 (2013).
[71] A. Ruzsinszky, J. P. Perdew , J. Tao and G. I. Csonka, Phys. Rev. Lett. 109,
233203 (2012).
106
106
[72] J. P. Perdew, A. Ruzsinzky, J. Sun, S. Glindmeyer and G. I. Csonka, Phys. Rev.
A 86, 062714 (2012).
[73] R. Dahm, Hum. Genet. 122, 568 (2008).
[74] W. Astbury, Symp. SOC. Exp. Biol. 1, 66 (1947).
[75] J. D. Watson and F. H. C. Crick, Nature 171, 4356 (1953).
[76] Nature Archives Double Helix of DNA: 50 Years, (2003).
[77] Wikipedia.org, .
[78] S. Wolfram, "Principle of Nucleic Acid Structure", ISBN 0-387-90762-9 (1984).
[79] S. Gregory, K.F. Barlow, K.E. Mclay, R. Kaul, D. Swarbrek, A. Dunham, and C.
E. Scott, Nature 441, 7091 (2006).
[80] R. E. Dickerson, Sci. Amer. 249, 94 (1983).
[81] webpage x3dna.org, .
[82] V. R. Cooper, T. Thonhauser, A. Puzder, E. Schröder, B. I. Lundqvist and D. C.
Langreth, J. Am. Chem. Soc. 130, 1304 (2008).
[83] M. J. Parker, C. A. J. Mol. Biol. 280, 407 (1998).
107
107
[84] C. A. Hunter and X. Lu, J. Mol. Biol. 265, 603 (1997).
[85] G. C. Fonseca, F.M. Bickelhaupt, J.G. Snijiders and E.J. Baerends, J. Am. Chem.
Soc. 122, 4117 (2000).
[86] C. A. Hunter, J. Mol. Biol. 230, 1025 (1993).
[87] P. Hobza and J. Šponer, J. Am. Chem. Soc. 124, 11802 (2002).
[88] M. Swart, t. van der Wijst, C. F. Guerra and F. M. Bickelhaupt, J. Mol. Model 13,
1245 (2007).
[89] V. R. Cooper, T. Thonhauser, and D. C. Langreth, J. Chem. Phys. 128, 204102
(2008).
[90] W. K. Olson, M. Bansal, S. K. Burley, R. E. Dickerson, M. Gerstein, S. C.
Harvey, U. Heinemann, X.-J. Lu, S. Neidle, Z. Shakked, H. Sklenar, M. Suzuki, C.-S.
Tung, E. Westhof, C. Wolberger, and H. M. Berman, J. Mol. Biol. 313, 229 (2001).
[91] C. A. Hunter and J. K. Sanders, J. Am. Chem. Soc. 112, 5525 (1990).
[92] G. B. McGaughey,M. Gagne and A. K. Pappe, J. Biol. Chem. 273, 15458 (1998).
[93] C. G. Claessens and J. F. Stoddart, J. Phys. Org. Chem. 10, 254 (1997).
[94] S. Grimme, Angew. Chem. Int. Ed. 47, 3430 (2008).
108
108
[95] M. Piacenza and S. Grimme, J. Am. Chem. Soc. 127, 14841 (2005).
[96] M. O. sinnokrot, E.F. Valeev and C.D. Sherrill, J. Am. Chem. Soc. 124, 10887
(2002).
[97] O. Sinnokrot and C.D. Sherrill, J. Phys. Chem. A 108, 10200 (2004).
[98] C. D. Sherrill, T. Takatani, and E. G. Hohenstein, J. Phys. Chem. A 113, 10146
(2009).
[99] www.begdb.com, Benchmark Energy and Geometry Database, .
[100] S. Tsuzuki, K. Honda and T. Uchimaru, M. J. Chem. Phys. 120, 647 (2004).
[101] Sponer J, Leszczynski and J, Hobza P, J. Phys. Chem. 100, 5590 (1996).
[102] Q. Wu and W. Yang, J. Chem. Phys. 116, 515 (2002).
[103] J. Sponer and P. Hobza, J. Phys. Chem. 98, 3161 (1994).
[104] J. Sponer, P. Jurecka, I. luque, M. Orozco and P. Hobza, Chem-Eur. J. 12, 2854
(2006).
109
109
Biography
The author received his B.S. in physics at University of Science and
Technology of China in 2007. Then, he was enrolled in the department of physics and
engineering physics to begin his Ph.D. study in 2007. Since then, he has been
studying density functional theory with Professor John P. Perdew.
The following papers are my publications related to this dissertation.
Y. Fang, B. Xiao, J. Tao, J. Sun and J. Perdew, "Ice phases under ambient and high
pressure: Insights from density functional theory", Phys. Rev. B, 87,21401 (2013)
P. Hao, Y. Fang, J. Sun, G. I. Csonka, P. H. T. Philipsen, and J. P. Perdew, “Lattice
constants from semilocal density functionals with zero-point phonon correction”,
Phys. Rev. B, 85, 014111 (2012)
J. Sun, B. Xiao, Y. Fang, R. Haunschild, P. Hao,1 A. Ruzsinszky, G. I. Csonka, G. E.
Scuseria, and J. P. Perdew, "Covalent, Metallic, and Weak Bonds from a Semilocal
Density Functional with the Right Ingredients", Submitted to PRL (2013)