Você está na página 1de 51

MODELING OF STABILITY AND PHASE

TRANSFORMATIONS IN QUASI-ZERO DIMENSIONAL


NANOCARBON SYSTEMS
A. S. Barnard
a,
,

P. Zapol
a,b,
,

and L. A. Curtiss
b,
a
Center for Nanoscale Materials and Materials Science and Chemistry Divisions
Argonne National Laboratory,
9700 South Cass Ave,
Argonne, IL, 60439
b
Applied Physics,
School of Applied Science
RMIT University GPO BOX 2476V Melbourne,
VIC, 3001, Australia.
Abstract
The transformation of nanodiamonds into carbon-onions (and vice versa) has been observed
experimentally and has been modelled computationally at various levels of sophistication. Also,
several analytical theories have been derived (by a number of research groups) to describe the
size, temperature and pressure dependence of this phase transition. However, in most cases a pure
carbon-onion or nanodiamond is not the nal product. More often than not an intermediary is
formed, known as a bucky-diamond, with a diamond-like core encased in an onion-like shell. This
has prompted a number of studies investigating the relative stability of nanodiamonds, bucky-
diamonds, carbon-onions and fullerenes, in various size regimes. Presented here is a review outlining
results of numerous computational and theoretical studies examining the phase stability of carbon
nanoparticles, to clarify the complicated relationship between fullerenic and diamond structures at
the nanoscale.
Keywords: diamond, nanoparticles, theory and simulation, phase transitions
PACS numbers:
1
I. INTRODUCTION
For decades it was known that graphite was the stable form of carbon under ambient
temperatures and pressures at the macro-scale, and that diamond was merely metastable.
With the discovery of buckminsterfullerene in 1980
1
and the one-dimensional analogue car-
bon nanotubes in 1991
2
, it became clear that at the nanoscale, carbon structures dier from
their macroscopic counterparts. Fullerenes and carbon nanotubes share many structural
characteristics with graphite (including sp
2
hybridization and six membered rings), and the
relative phase stability of graphite and fullerenes can be attributed to size dependence. The
closing of the shells cause the curved surfaces of fullerenes to gain stability, due to the
absence of the dangling bonds.
Diamond has also been observed at the nanoscale, with a number of forms of nanodiamond
being synthesized using various techniques. Both polycrystalline diamond clusters (ranging
in size from 1 60m in diameter) comprising of nanometer-sized diamond grains from 1
50 nm in diameter, and free nano-size diamond particles (also known as ultradispersed
diamond (UDD) or detonation diamond
3
) have also been successfully synthesized during
high energy exothermic detonations. Nanocrystalline diamond lms have also been grown
using chemical vapor deposition (CVD) methods, with varying plasma chemistries
46
.
With the variety of quasi-zero dimensional (0-D) nano-carbon phases available, it is not
surprising that the attention of many nanotechnologists is focussed around carbon-based
nanostructures
7,8
. The existence of multiple carbon phases does however raise questions
as to how they are related to one another, and which phase is energetically preferred in
various size regimes. Experimentally, it has been found that upon annealing 2 5 nm
nanodiamond particles transforms into onion-like carbon (OLC)
911
. The transition occurs
from the surface inward, with the transformation temperature being dependent on particle
size, with the particles exhibiting preferential graphitization (and exfoliation) of the diamond
(111) surfaces over other lower index surfaces. Further, the results of Chen et al
12
show that
while 5 nm nanodiamonds have been found to be unstable with respect to the transition
into OLC, nanodiamonds with a diameter of approximately 2.5 nm still remain after an
annealing; below 1073 K, nanodiamonds with a diameter of 5 nm are stable, implying that
the equilibrium size of diamond is temperature dependent.
In parallel to these discoveries and investigations, the reverse phenomena the transfor-
2
mation of OLC into nanodiamond was also observed by Banhart et al
13
, showing that OLC
undergoes electron irradiation induced self-compression at the center, and decay into dia-
mond nuclei
1417
. In general, although it has been shown that nanodiamond may transform
into OLC (and visa versa), in most cases a pure carbon onion or nanodiamond is not the
nal product. More often than not, an intermediary is formed with a diamond-like core and
an onion-like outer shell. These core-shell nanoparticles are known as buckydiamonds
18
.
In this review, the application of theory and quantum mechanical based simulations
19
to
the phase stability of nanodiamond will be summarized, including a comparison of several
theoretical models that have been put forward by various authors to describe the relative
phase stability of diamond and other forms of carbon at the nanoscale. A critical issue in the
stability of nanodiamond, buckydiamond and OLC is the role hydrogen absorbed on the
surfaces. In the presence of hydrogen, such as in conventional CVD synthesis of diamond
thin lms, nanometer sized diamond clusters dominate since the atomic H in the plasma
etches the sp
2
content of the lms. The relationship between the various forms of clean
carbon nanoparticles and those in the presence of hydrogen will be discussed, along with
the mechanisms responsible for stabilization.
II. THEORETICAL BACKGROUND
The methods discussed in this review to describe the energy and geometry of carbon
nanostructures are based on the quantum mechanical Schrodinger equation using the Hamil-
tonian operator in the Coulomb approximation
20
. At the most fundamental level this equa-
tion describes the energy of the total, coupled system of electrons and nuclei (or ions),
which are separated using the Born-Oppenheimer (BO) approximation. This assumes that
the wavefunction
t
(r, R) for the total system (where r represents the coordinates of the
electrons and R the coordinates of the ions) may be approximated as the product of the
electronic wavefunction
e
(r, R) which describes the electronic energy for xed positions of
the ions, and a wavefunction
n
(R) which describes the energy of the ions.
The wavefunction
e
(r, R) and the electronic energy E
e
(R) are determined by solving
the electronic Schrodinger equation for xed values of the position of the ions R. The sum,
E
T
(R) of the electronic energy E
e
(R), and the nuclear-nuclear repulsion energy E
n
(R), then
provides the potential (or eld) which determines the energy and motion of the ions. A
3
numerical approach to the description of the ionic motion may be accomplished by use of a
Monte Carlo (MC) or Molecular Dynamic (MD) simulation
21
.
A. Wavefunction Based Methods
The most commonly used methods are based on one-electron functions (orbitals) which
are then used to form determinantal wavefunctions which satisfy the Pauli Exclusion
Principle
20
. The most prevalent method is HartreeFock (HF) theory which (in its sim-
plest form) assumes that the wavefunction is a single determinant constructed from a set
of single electron functions referred to as spin orbitals
i
(r
i
, s
i
) which are a product of a
function of the spatial coordinates of electron i,
i
(r
i
), and a spin up function (s) or spin
down function (s)
20,22
, such that:

HF
= N

1
2
[
1
(r
1
, s
1
)
2
(r
2
, s
2
).....
N
(r
N
, s
N
)[ (1)
If the system is a singlet state then the spatial orbitals are usually assumed to be doubly
occupied, i.e.
i
=
i
+ 1 for i odd. This means that each space orbital contains two
electrons of opposite spin.
These orbitals may be determined by nding the form of
HF
i.e. of
i
(r
i
), which
gives the lowest possible energy by the use of the quantum mechanical variational principle,
leading to a set of coupled single particle Schrodinger equations called the HartreeFock
equations. These equations are coupled, but may be solved self consistently by an iterative
process which is referred to as the Self Consistent Field (SCF) method. This may be done
numerically for atoms and diatomic molecules but the problem must be tackled in a dierent
way for more complicated systems
20,22
.
For large complexes, polymers and solids the usual method to solve the HF equations is
to expand the one electron spatial functions
i
as a nite, linear combination of a set (a
basis set) of known functions (r); = 1, ....., M called basis functions
20,22
:

i
(r) =
M

=1
C
i

(r) (2)
These linear coecients C
i
are then determined by a set of equations derived from the
Hartree-Fock equations by use of the linear variational method which must also be solved self
consistently. This is called the method of Linear Combination of Atomic Orbitals (LCAO)
4
SCF method and if sucient basis functions are used then the Hartree-Fock result may
be approached to any degree of accuracy
20,22
. Basis sets in common use are, atom centered
Slater Type Orbitals (STOs), atom centered Gaussian Type Orbitals (GTOs), a combination
of atom and non-atom centered (bond) GTOs and plane waves functions (PWF)
20,22,23
. If
the system is periodic then symmetry can be used to simplify the construction of the orbitals.
The HF method has been generalized to describe open shells and excited states so the above
form using paired space orbitals is normally referred to as the Restricted HartreeFock
(RHF) method.
The HF method is an independent electron model in the sense that the electrons are
described by individual one electron orbitals or wavefunctions and each electron moves
in the averaged out eld due to all the other electrons. Thus, it does not describe the
detailed, correlated motion of the electrons. Hence, the energy given by the HF method,
E
HF
is not exact and the dierence between the exact, non-relativistic energy E
e
and E
HF
is known as the electron correlation energy, E
corr
. There also exist more accurate methods
which attempt to estimate E
corr
and that use the HF method as a starting point
20,2224
,
e.g. the Conguration Interaction (CI) method
20
, the nth order Mollet-Plesset Perturbation
(MPn) theory, the Coupled Cluster (CC) methods CCSD, CCSD(T), CCSDT and the QCCI
methods
24
. However, none of these methods have as yet been used to describe the type of
systems described in this review, nor periodic systems.
The implementation of the above methods can be performed entirely from rst principles
(ab-initio) where all the integrals involved are calculated exactly or semiempirically where
some of the integrals are ignored, some integrals are approximated and the values of some
integrals are adjusted to get agreement with experiment for some properties. Examples of
these semiempirical methods are the MNDO
24,25
, AM1
24,25
and PM3
24,26
methods.
Clearly the ab initio methods are slower and require more computer storage than semi
empirical methods but are more rigorous and, more accurate than the semiempirical meth-
ods (if taken to a high enough order of approximation). Semiempirical methods are, of
course useful for systems containing many atoms and/or many electrons. The approximate
representation between the relative CPU time versus relative error for various quantum me-
chanical methods for electronic structure calculation may be seen in, for example, reference
27
.
5
B. Density Functional Theory
These methods are based on the Hohenberg and Kohn theorem
28,29
which states that
for a given type of Hamiltonian operator the energy is a unique functional of the density.
Thus, the electronic energy is a unique functional of the electronic density (r), and we may
write the electronic energy E
e
[] of a system of interacting ions and electrons as a unique
functional F[] as below,
E
e
[] = F[] +
_
(r)v
ext
(r)dr +E
ion
(R
I
) (3)
where v
ext
(r) is an external potential of the electron gas (the electrostatic potential of the
nuclei), E
ion
(R
I
) is the ion-ion interaction energy of the ions at xed positions R
I
, and
F[] is the universal functional of independent of v
ext
(r). Thus F[] is the same functional
for electrons in atoms, molecules, polymers or solids. The exact form of this functional F[]
is unknown, and approximations are made such as the Local Density Approximation (LDA),
the Generalized Gradient Approximation (GGA) and, so called, hybrid methods
28
.
In practice the method of Kohn and Sham
28,30
is usually used in which they showed
that the many body problem of an interacting electron gas in the presence of an external
potential may be mapped onto an auxiliary system of non-interacting particles which enables
an orbital description to be used. In this case the orbitals are used to expand the electron
density not the electronic wavefunction.
If the electron density is written in terms of N doubly occupied single-particle orbitals

i
(r) : i = 1, ......., N as:
(r) = 2

[
i
(r)[
2
(4)
then
E
e
_

i
(r), R
_
= 2
_

i
(r)(
1
2

2
)
i
(r)dr + (5)
_
v
ext
(r)(r)dr +
1
2
_
(r)(r

)[r r

[
1
drdr

+E
xc
[] +
1
2

Z
I
Z
J
[R
I
R
J
[
1
where the successive terms on the right hand side of equation 5 represent respectively; the
kinetic energy of the electrons, the nuclear-electron coulomb attraction energy, the electron-
electron coulomb repulsion energy, the exchange-correlation energy and the nuclear-nuclear
6
coulomb repulsion energy. It is the exchange-correlation energy E
xc
[] which is the most
dicult term to estimate and it is this term that is approximated by such methods as LDA
and GGA.
The use of the variational method then leads to a set of coupled one-electron equations
called the KohnSham (KS) equations, which determine these orbitals (the KohnSham
orbitals)
28,30
. The KS equations are then generally solved using basis functions in an anal-
ogous way to the solution of the Hartree-Fock equations
28
. However, instead of neglect-
ing electron correlation energy E
corr
and using the exact exchange energy as does the HF
method, the KS method approximates the exchange energy but also includes an estimate
of E
corr
. Also in contrast to wavefunction based methods used to estimate E
corr
the KS
method retains an orbital description.
C. Ab-initio Molecular Dynamics
Car and Parrinello
31,32
have developed a method based on DFT to enable accurate ab
initio quantum mechanical calculation of the interatomic potentials, and forces and carry
out a MC or MD simulation in order to determine the motion of the ions from rst principles.
The method involves some approximations and uses a combination of pseudopotentials
66
to
treat the inner shell electrons and DFT with a plane wave basis set to treat the valence elec-
trons. The technique couples the electronic and nuclear motions by introducing an iterative
optimization technique where the plane wave coecients describing the electrons and the
positions of the ions are treated as ctitious particles interacting via a classical Lagrangian.
Integration of these equations of motion leads to the conguration of minimum energy. Both
electronic and nuclear degrees of freedom are relaxed simultaneously. If the mass of the c-
titious electrons is adjusted correctly then the correct dynamics is obtained for the nuclear
motion even though the electronic motion is not correctly described. However, the correct
BO energy surface for the electronic ground state is traversed within the approximations of
the method used to describe the electronic state of the system. This method is called rst
principles MD, ab initio MD or Car-Parrinello MD (CPMD).
Other implementations of this type of method have been made for example in the
CASTEP
33
and Vienna ab initio simulation package (VASP)
34
codes. Ab initio MD is a
very valuable method as it allows the treatment of eects which other methods do not read-
7
ily treat, such as non-zero temperature, bond making and bond breaking. If conventional
electronic structure methods are used they only describe the structure at zero temperature.
On the other hand MD methods based on the use of empirical potential functions (rather
than on quantum methods) nd it hard to describe bond making and breaking. The main
limitations of ab initio MD techniques is that they are very time consuming and are, at
present, largely restricted to systems containing approximately < 10
3
electrons.
A slightly simpler method is the so-called Tight Binding MD (TBMD) method which
uses the quantum mechanical tight binding Hamiltonian introduced by Slater and Koster
35
.
Tight binding potentials and TBMD for example, have been shown to be highly applicable
to carbon nanostructures
36
. This method is simpler than the ab initio MD methods but
makes more approximations in the treatment of the electrons and needs parameterization.
It is, however, based rmly on quantum mechanics and so can be systematically rened. It
is applicable to much larger systems than the ab initio MD method, and at present it can
handle approximately 10
3
10
4
atoms
37,38
.
III. CLEAN NANOPARTICLES
In this section, the structural and thermodynamic stability of clean quasi-zero dimen-
sional nanocarbon particles will be discussed, referring to particles in the absence of hydrogen
with C-terminated surfaces. The surface of these particles contain a large number of dan-
gling bonds, making reconstructions of various types of great importance both structurally
and energetically. The eect of H-termination on the stability of such structures will be
addressed in a later section.
We will see, because of the small molar volume of diamond compared to that of graphite,
for a suciently small carbon cluster diamond can be more stable than graphite. It should
be noted that in analyzing the stability between the clusters of diamond and graphite the
number of carbon atoms in each cluster should be the same, and thus, the radii of each
cluster are dierent from each other. The radius of the diamond cluster is smaller than that
of the graphite cluster for the same number of atoms. It is also important to keep in mind
that when a cluster gets smaller, the surface eects become dominant over the bulk eects.
Section III A outlines a number of analytical models used to describe nanocarbon phase
stability (that typically require computations to be performed in obtaining various parame-
8
ters), whereas section III B outlines results generating using purely computational methods.
A. Phase Stability
In 1987 Amlof and L uthi
39
performed large-scale ab initio restricted HartreeFock (RHF)
calculations of the phase stability of planar graphene and cubic diamond clusters. The
dependence of the total energy on the number of carbon atoms was obtained by tting the
calculated cluster energies to the equation,
E
total
= N
C
E
C
+N
DB
E
DB
(6)
where N
C
is the total number of carbon atoms, N
DB
is the total number of dangling bonds,
E
C
is the energy per carbon atom and E
DB
is the energy per dangling bond. The surface
energy contribution decreases for larger clusters and the cohesive energy approaches the
macroscopic limit (bulk-diamond or bulk-graphite) as N
C
. Neglecting the contribution
from the cluster zero-point energy, they dened the cohesive energy as the dierence between
the energy per carbon atom E
total
/N
C
and the energy of the free carbon atom
39
.
By tting the energies from calculations on small graphene structures they predicted
the cohesive energy for an innite graphene sheet to be -5.421 eV. The contribution due
to the attractive interactions between the graphitic layers was assumed to be less than
0.217 eV; however, even with the interlayer attraction the results were considerably higher
than the experimental value for the cohesive energy of graphite of -7.371 eV at T = 0 K.
Similarly, in the case of diamond the calculated cohesive energies of the three small sp
3
bonded carbon clusters obtained by extrapolation of -5.386 eV was also considerably higher
than the experimental value for bulk diamond of -7.346 eV at T = 0 K. The dierences in
the extrapolated cohesive energies of diamond and graphite were attributed to the electron
correlation eects which are not included in the RHF method, but these were important
steps in the right direction
39
.
Later, Shaw and Johnson
40,41
used a diusion-limited model that assumed that the dif-
ference in energy between the cluster and bulk (E) was proportional to the number of
surface atoms. They demonstrated that in the detonation regime carbon clustering is a slow
reaction, and that carbon particles were built up from random collisions. The model did
not distinguish between the various forms of carbon present during the detonation process,
9
however they were able to estimate a particle size from 10
4
to 10
5
atoms (50

A), with un-
certainties primarily due to size dependence of the energies of the various carbon clusters.
Using the assumption that E, was proportional to the number of surface atoms, they found
that E n
1/3
(where n is the total number of atoms)
40,41
.
Building on the foundation of this work, a number of theoretical models have been pro-
posed by various authors to compare the relative phase stability of diamond and graphite
particles at the nanoscale.
1. Nanodiamond versus Graphite
In 1996, Gamarnik
42
reported on the boundary of the stability regions of diamond and
graphite nanoparticles at low pressure. Beginning with the assumption that an energetic
preference for the nucleation of nanodiamond over nanographite exists, the model asserted
that the stabilization of nanodiamond is a direct result of the small particle diameters. The
model, which also discriminated between atoms located at the particle edges and corners,
began with a statement of the energy of a phase at low pressure, in terms of the Helmholtz
free energy F,
F = E
at
TS (7)
where S is the entropy, T is the temperature and E
at
is the atomization energy. Note that at
T = 0, the E
at
= F. The E
at
is dened in terms of the lattice energy E (which is dependent
on the particle size), the kinetic energy E
k
, and the ionization potential of bond charges per
atom I, as:
E
at
= E +E
k
+I. (8)
The lattice energy was determined by summation of the Born-Lande pair interaction
potentials of all the charges of the crystals
43,44,57
. The kinetic energy for the bond electrons
in diamond and graphite were derived by relating the Coulomb force F
C
acting on a bond
electron with the bond electron centripetal force to the neighboring atom F
cp
:
F
C
= F
cp
(9)
The centripetal force acting on a bond is due to the interaction of F
C
with all of the
nearest neighbors in the crystal lattice, rather that with one atom. The kinetic energy of
the diamond bond electrons (per atom) was therefore dened as:
10
E
k
= QE
(1)
k
, (10)
where,
E
(1)
k
=
2r
o
(r r
o
)
r
2
I
1
(11)
is the kinetic energy of one bond electron, is the bond charge per atom, Q is the atom
charge in the diamond lattice, r is the bond length, r
o
is the carbon orbital radius and I
1
is
the rst ionization potential. A similar approach was used to calculated the kinetic energy
of the graphite bond electron charges
42
.
The diamond and graphite charge lattices were represented by excessive negative covalent
bond charges and positive ions. In the case of graphite, conduction atoms were added (also
represented by excessive negative charges) between the neighboring atoms between the nets,
to secure the stable graphite structure. The attraction between the conduction electrons and
the hexagonal nets compensates the repulsive forces acting between the nets. Therefore, by
assuming that each inner atom relinquishes one electron charge value to establish a bond,
the covalent bond charge value per atom Q
0
and the conduction electron charge value
0
was
established. In the case of diamond this is distributed in four directions, and three directions
in the case of graphite, and the bond charge of an inner or surface atom was determined by
considering the number of nearest neighbors n
42
.
Experimental values of bond charges were then used for the calculation of the size de-
pendencies of E
at
, illustrating that E
at
was dependent on the size of the nanodiamond and
nanographite particles, as well as the temperature. The model predicted that nanodiamond
was stable for particles below 15 nm at T = 0, and graphite above
42
. The investigation
was conducted at various temperatures, nding that increasing temperatures decreases the
range of stability of nanodiamond, as shown in Figure 1.
At around the same time, Hwang et al
45
outlined a thermodynamic chemical potential
model in terms of the surface energy of diamond and graphite clusters that was applicable
in non-equilibrium conditions such as during CVD. Low pressure CVD synthesis of diamond
involves carbon being transferred between gas and solid phases, so that the deposition and
etching was analyzed in terms of the activity of carbon. The model surmises that the
chemical potential of carbon represented the main criterion for the phase transition, and
that the carbon atoms transfer from a higher activity phase to a lower activity phase
45
.
11
FIG. 1: Plot of the phase transition sizes for the nanodiamond-to-graphite transformation as a
function of temperature, as calculated by Gamarnik
42
.
They asserted that when the activity of carbon in the gas phase (a
gas
C
) is lower than
that of graphite (a
gra
C
) both graphite and diamond phases are etched. When the activity is
between a
gra
C
and the activity of diamond (a
dia
C
) graphite is deposited and diamond is etched.
When it is higher than a
dia
C
both graphite and diamond are deposited. It was also logically
established that a
gra
C
is less than a
dia
C
, and that since diamond was chosen as the reference
state, a
dia
C
is unity at all temperatures
45
.
Using these activities, the chemical potential of carbon in the gas phase is expressed as:

gas
C
=
s
C
+RT ln a
gas
C
(12)
where
s
C
is the chemical potential in the standard state. Thus, the chemical potentials of
12
graphite and diamond were expressed, as:

gra
C
=
s
C
+RT ln a
gra
C
(13)

dia
C
=
s
C
+RT ln a
dia
C
(14)
respectively. The driving forces for precipitation of graphite and diamond from the gas phase
are, respectively:

gasgra
C
=
gra
C

gas
C
= RT ln(a
gas
C
/a
gra
C
) (15)

gasdia
C
=
dia
C

gas
C
= RT ln(a
gas
C
/a
dia
C
) (16)
The authors found that the driving forces for precipitation of graphite and diamond
(depending on the sign) would promote either precipitation or etching. The stability between
diamond and graphite in the carbon phase diagram was considered via the dierence in
the driving force between diamond and graphite, showing that the relative magnitude of
the driving force was reversed between diamond and graphite when the pressure was high
enough. From the point of view of metastable diamond formation, this approach predicts
that precipitation of diamond and graphite (from the gas phase) are kinetically equivalent
45
.
Further, the phase stability was shown dier from the known phase diagram of carbon, due
to the a capillary eect on the nuclei. The capillary pressures were expressed as a function
of the surface energy, and the radius of the curvature of the graphite and diamond clusters
(which were assumed to be spherical) as given by the LaplaceYoung equation. When the
pressure inside the clusters becomes suciently high, the chemical potential of the graphite
cluster is higher than the diamond cluster, thereby decreasing the size of the clusters. The
pressure at equilibrium was shown to correspond to the phase boundary between graphite
and diamond in the phase diagram of carbon. Thus, the chemical potentials per mole for
carbon in graphite and diamond clusters (including the capillary eect) were described by:

c,gra
C
=
gra
C
+V
gra
_
2
gra
r
gra
_
(17)

c,dia
C
=
dia
C
+V
dia
_
2
dia
r
dia
_
(18)
where V is the molar volume. By equating these terms, and solving for the number of atoms
when
c,dia
C
=
c,gra
C
, the phase transition size was found.
The authors calculated the term
diagra
to be 1.120110
2
J/atom, and by inserting
the reported values
46
of 3.7 J/m
2
and 3.1 J/m
2
for the surface energies for diamond and
13
graphite (respectively), the number of atoms was estimated to be 104 (at 1200 K). The
surface energy of diamond was then taken as 3.6, 3.5, 3.4 and 3.3 J/m
2
, resulting in values
of 177, 279, 413 and 584 atoms, respectively
45
.
This study also went on to examine the nucleation on a substrate; proposing that the
free energy of the contacted particle is decreased when a solid carbon particle forms on
the substrate, and that when the interface energy between the diamond and the substrate
is smaller than that between the graphite and the substrate the stability of diamond is
favored
45
.
As an extension of this work Hwang and his colleagues
47
also detailed a charged clus-
ter model to describe the stability of small charged diamond and graphite clusters. Since
graphite is conducting, the authors assumed the charge to be uniformly distributed over the
conducting bulk and the interface of a graphite cluster will be non-polarizable; whereas in
the case diamond, the charge on the diamond cluster will be localized at the interface, and
the interface will be polarizable. The model assumes that the surface energy of the dia-
mond cluster may be decreased by the presence of charges while that of the graphite cluster
cannot, so that the eects of electrocapillarity (arising when an external eld is applied to
a polarizable interface
48
) will favor diamond formation. The possibility of charge-induced
nucleation of diamond in the gas phase was a key point of the charged cluster model in their
paper
47
.
Although the charged cluster model (and available experimental data) was not sucient
to make any quantitative prediction on the sign dependence of the stability between dia-
mond and graphite clusters, based on the concept of sign dependence in nucleation (and the
assumption that a negative charge stabilizes the diamond cluster) the model was used to
describe the electric eld at a cluster surface, and the eects of charge on the surface energy.
The model predicts that if the presence of charges increases the intrinsic electric eld of this
surface the surface energy will decrease, but the surface polarizibility would increase surface
(and vice versa). Given an uncharged diamond cluster with a negative surface, the accumu-
lation of the negative charges on the surface serves to decrease the surface energy, while the
accumulation of positive charges increases the surface energy, so the presence of negative
charges would be critical to the stability of a diamond cluster. The diamond clusters would
be stable when charged, and transform to a graphitic cluster when the charge is removed
47
.
An extensive and complete treatise of this model, outlining the theory and application
14
to appropriate systems in given in reference
49
.
In the models outlined above, the pressure dependence of the nanodiamond/graphite
phase stability was not been explicitly addressed. In an entirely dierent approach, Jiang et
al
50
considered the size dependence of the diamond to graphite transition, as a function of
pressure and temperature. In their model the phase transition between nanodiamond and
nanographite was described, based on the extrapolation of the equilibrium phase bound-
ary (in the bulk diamond to bulk graphite pressure-temperature phase diagram) to the
nanoscale. The extrapolated phase boundary was calculated with the Laplace-Young equa-
tion, with input of experimental values for the bulk phases. The thermodynamic functions
for the diamond to graphite transition were estimated on the basis of the Clausius-Clapeyron
equation. According to the phase diagram of carbon the transition pressure function P(T)
for the diamond to graphite transition is given as,
P = 1700 + 2.06T (19)
where P is the external pressure (in MPa) and T the temperature (in Kelvin). This term
is determined by the terminating points of the graphite to diamond boundary in the phase
diagram at (1700 MPa, 0 K) and (12000 MPa, 5000 K), respectively
50
.
The contribution of the size-dependent internal pressure was estimated by assuming
spherical, quasi-isotropic nanocrystals, with additional curvature-induced pressure (P
a
)
given by the Laplace-Young equation,
P
a
= 4/d (20)
where d is the nanoparticle diameter (approximately several nanometers), and is the surface
free energy. To account for P
a
, the external pressure P required for the phase transition was
decreased by the same amount; so that P +P
a
gave the size dependent phase boundary. In
the case of low-temperature and low-pressure synthesis of diamond, P 0, and,
d =
4
1700 + 2.06T
. (21)
The equilibrium size of diamond was then calculated with a mean value for the surface
energy of dierent faces,
D
= 3.7 J/m
2
for diamond. For graphite, a value of
G
= 3.1
J/m
2
was used. These were equated at equilibrium (neglecting eects of surface tension),
with the results predicting the transition size of nanodiamond decreases from approximately
8nm at 0 K to 3nm at 1500 K; as shown in by the solid line in Figure 2
50
.
15
FIG. 2: The size-temperature transition diagram of carbon at zero pressure, where the solid line
shows the model prediction, and the segment line denotes the model prediction with surface stress.
The theoretical and experimental results are also plotted in the gure. The symbol _ denotes the
theoretical estimation based on the surface energy dierence between diamond and graphite
52
. The
symbol gives the theoretical calculation in terms of the charge lattice model of Gamarnik
42
. The
symbol is based on the experimental observation at 1073 K that nanodiamonds with d = 5 nm are
transformed into nanographite
12
. The symbol shows an experimental result where nanodiamonds
of 2 nm in size transform to onion-like carbon at 1300 K
10
. Reprinted from Diamond and Related
Materials, 11, 234, D. Zhao, M. Zhao and Q. Jiang, Diamond. Relat. Mater., c _(2002) with
permission from Elsevier.
The model also predicts that since
D
>
G
the phase transition would accelerate on the
surface of the nanodiamond, as (and therefore P
a
) decreases during surface graphitization.
However, in the reverse case (the graphite to diamond transition), does not decrease and
P
a
increases on the surface of graphite. This relates to the experimental observation (under
ambient pressure) of nanodiamonds graphitizing at the surface
50
.
16
Then, by assuming a critical nucleus of 100 atoms for nanodiamond, the authors showed
that the critical diameter of the nucleus is located within the stability eld of diamond,
thereby promoting the nucleation and growth rate of diamond at lower temperature. How-
ever, as the equilibrium temperature increases the size of the nanodiamond decreases, and
once formed the graphite nanoparticles gain stability due the change in surface energy
50
.
Finally, by applying the Clausius-Clapeyron equation,
H =
dP
dT
(V
G
V
D
)T (22)
where H is the transition enthalpy in J/mol, dP/dT = 2.06 J/cm
3
K (from above) and V
G
and V
D
are the molar volumes of graphite and diamond having sizes of 5.298 and 3.417 in
cm
3
/mol, respectively, they obtained H = 3.87T, and a transition entropy of S = H/T =
3.87 J/molK.
This model was later extended to include the eects of surface stress on the internal pres-
sure of the nanoparticle
51
. The value of for a quasi-isotropic nanocrystal was determined
by,
=
_
3D
0

sl0
8
_1
2
(23)
where
sl0
is the bulk solid-liquid interface energy, is the compressibility and D is the
diameter of a particle where almost all atoms are located on the surface.
sl0
was determined
by using the Gibbs-Thomson equation,

sl0
=
2hS
vib
H
m
(T)
3V
m
R
(24)
where R is the ideal gas constant, h is the atomic diameter, H
m
the temperature-dependent
melting enthalpy of crystals, S
vib
the vibrational part of the overall melting entropy S
m
,
and V
m
the molar volume of crystals. The melting enthalpy was described in terms of the
Helmholtz function and the temperaturedependent, solid-liquid Gibbs free energy dierence
g
m
(T),
H(T) = g
m
(T) Tdg
m
(T)dT (25)
For diamond and graphite, the corresponding g
m
(T) function is expressed as,
g
m
(T) =
H
m
(T
m
T)T
T
2
m
, (26)
where H
m
is the melting enthalpy at the melting temperature T
m
. By letting dg
m
(T)/dT=0,
the isentropic temperature (between the under-cooled liquid and the crystal) T
k
=T
m
/2 was
17
obtained; below which the specic heat dierence between a glass and the crystal C
p
approaches zero. As C
p
in responsible for the temperature dependence of H
m
(T),
H(0 < T < T
k
) H
m
(T
k
) (27)
and,
H
m
(T) = H
m
_
T
T
m
_
2
. (28)
In this case, H
m
(T) = H
m
(T
k
) = H
m
/4, and
sl0
was written as,

sl0
=
hS
vib
H
m
(T)
6V
m
R
(29)
The authors then plotted the transition interface between nanodiamond (111) and
nanographite (0001), to nd that the transition size of nanodiamond now decreases from
11 nm at 0 K to 4 nm at 1500 K. Furthermore, the transition point at 1300 K was found
to provide a transition size between diamond and the onion-like carbon; as the transformed
product of the onion-like carbon has a lower Gibbs free energy than the graphite, indicating
that the transition size between the nanodiamond and nanographite should be larger than 2
nm. The authors concluded by pointing out that when the surface energy is assumed to be
equivalent to surface stress, the equilibrium transition line is lower than the predicted line,
but approaches the theoretical estimation based on the fact that the value of the averaged
surface stress of 4.75 J/m
2
(which is proportional to the equilibrium transition size), is larger
than the averaged value of the surface energies of 3.4 J/m
251
.
The results of Jiang et al
50,51
clearly show that the eects of surface stress is very im-
portant in the description of the phase stability of nanocarbon, since the transition size of
nanodiamond without surface stress (above) gave 8 nm at 0 K to 3 nm at 1500 K; some
13 nm smaller than when surface stress was included
51
.
2. Nanodiamonds versus Fullerenes
The variety of models outlined above rmly establish the range of the cross-over for phase
stability of graphite and nanodiamond for the case of large nanocarbon particles (411
nm). The question of the phase stability of smaller (< 2.5nm) was investigated by Barnard
et al
53
, by extending a familiar thermodynamic model
54
to include fullerenes. By treating
18
only dehydrogenated nanodiamonds, dened as being nanodiamond structures consisting of
almost entirely of sp
3
bonded atoms (as opposed to buckydiamond), a direct comparison
of nanodiamonds with fullerenes was achieved.
The model used to describe the phase stability was based on the heat of formation (as
a function of particle size); a technique used previously by Winter and Ree
54
. The heat of
formation of graphite (H
o
f
(G)) and diamond (H
o
f
(D)) clusters was expressed in terms of
the bond energies E
CC
and E
CH
, and dangling bond energy E
DB
, such that:
H
o
f
(G)
N
C
=
3
2
E
G
CC
+
N
H
N
C
(E
G
CH

1
2
E
G
CC
+ H
o
f
(H)) +
+H
o
f
(C) +
1
2
E
vdw
CC
(30)
H
o
f
(D)
N
C
= 2E
D
CC
+
N
DB
N
C
(E
D
DB

1
2
E
D
CC
+
+H
o
f
(DB)) + H
o
f
(C) (31)
where, N
C
is the number of carbon atoms, N
DB
is the number of dangling bonds on the
surface of the particle, N
H
is the number of terminating hydrogen atoms, H
o
f
(C) is the
standard heat of formation of carbon at 298.15K and H
o
f
(H) is the standard heat of
formation of hydrogen. The term E
vdw
CC
is the van der Waals attraction between graphite
sheets. In the case of single shell fullerenes E
vdw
CC
=0, and the closed shells ensure that N
H
/N
C
term is also zero, leaving,
H
o
f
(F)
N
C
=
3
2
E
F
CC
+ H
o
f
(C) (32)
A term for the strain energy (E
F
strain
) that vanishes in the graphene limit was then included,
to account for the size dependence, so that the heat of formation for a fullerene was given
by,
H
o
f
(F)
N
C
=
3
2
E
F
CC
+ H
o
f
(C) +
E
F
strain
R
2
. (33)
Although E
F
strain
may be determined in a number of ways, the authors proceeded by as-
suming that if fullerenes may be approximated as homogeneous and isotropic elastic spheres,
the bending energy per unit area (A) of a suitable elastic sheet is given in terms of the sheet
thickness (h) was expressed as,
E
strain
A
=

2
_
+h/2
h/2
dz
z
2
R
2
=
h
3
24R
2
(34)
19
where is the bending modulus of the sheet, and R is the radius of curvature. The strain
energy per carbon atom was therefore,
E
strain
N
C
=
Ah
3
N
C
24R
2
. (35)
A spherical model was also assumed, where R is equal to the mean radius so that A =
4R
2
, and N
C
= 4R
2
. Therefore, an expression for the strain energy per carbon atom
was obtained, which was linearly dependent on the inverse square of the curvature of the
structure:
E
strain
N
C
=
_
h
3
24
_
1
R
2
= E
F
strain
1
R
2
(36)
Before determining the heat of formation for fullerenes, Barnard et al
53
rst obtained the
cohesive energy and the strain energy for fullerenic carbon by tting a linear expression to
the energy per ion for all fullerenes from C
20
to C
80
(calculated using DFT) versus 1/R
2
. The
E
F
strain
and E
F
CC
obtained from the slope and intercept were 5.19eV and 7.81eV respectively.
Using these results equation (33) was plotted as a function of the number of carbon atoms
53
.
The cohesive energy and dangling bond energy of nanodiamond was determined by calcu-
lating the total energy of a set of stable nanodiamond structures (of various morphologies)
with DFT GGA
55
, and using the linear t to the total energy per ion versus number of dan-
gling bonds per ion. The cohesive energy and dangling bond energy of E
D
CC
= 3.855 eV and
E
D
DB
= 1.619 eV were obtained from the intercept and slope of this plot, respectively. By
inserting these bond energies into equation (31), the intersection of the heat of formation of
nanodiamond and fullerenes was determined to be at 1130 atoms, which is approximately
equivalent to a cubic nanodiamond crystals of 1.9 nm in diameter
53
.
This work established that nanodiamond is not necessarily the stable form of carbon at the
nanoscale. Instead, fullerenes are a more stable form of carbon for small clusters, resulting
in a window of stability for nanodiamond, outside of which nanodiamond is metastable
(just as diamond is macroscopically)
53
. It is important to note that the upper limit of
fullerene stability (the fullerene-to-nanodiamond transitions size) falls above the upper limit
of quantum mechanical based simulations. In the size range of all recent ab initio studies of
nanodiamond (typically less than 10
3
atoms using current computer technology), fullerenes
are the globally stable phase of carbon.
20
3. Phase stability of intermediaries
The studies described above highlight how various approaches may be used to compare
the stability of graphitic/fullerenic with diamond at the nanoscale. However, in most cases
a pure nanodiamond is not formed, but rather buckydiamond
18
with a diamond-like core
and a full or partial fullerenic shell. Barnard et al
56
also addressed the thermodynamic
stability of multi-shell carbon nanostructure by using the model detailed at the end of the
last section for comparing the phase stability of nanodiamonds and fullerenes, and applying
it to buckydiamonds and OLC. For this purpose they considered particles which exhibited
surface delamination during structural relaxations (as will be described in more detail in
the next section) resulting in sp
2
hybridized atoms on the surface, and ignored those that
consist of entirely sp
3
hybridized atoms (characterized as stable nanodiamonds)
56
.
The OLC was treated as nested fullerenes by adding the van der Waals attraction (E
vdw
CC
)
to equation (33), to produce equation (37). The value of E
vdw
CC
= 0.056 eV used in this
study was calculated by Guo
57
using a graphite force eld. The buckydiamonds have been
treated in the same manner as nanodiamonds by applying equation (31), although obviously
N
DB
/N
C
is dierent for nanodiamonds and buckydiamonds of similar diameter
56
.
H
o
f
(F)
N
C
=
3
2
E
F
CC
+ H
o
f
(C) +
E
F
strain
R
2
+
1
2
E
vdw
CC
(37)
The heat of formation as a function of particle size for buckydiamond and OLC was
calculated, and plotted along with the nanodiamond and fullerene results, as shown in Figure
3. The authors then pointed out three main points established via this comparison. Firstly,
the sp
2
bonded fullerenic structures are most stable below about 900 atoms. Although the
OLC and fullerene results converge below 250 atoms, they diverge very slowly beyond this
value, conrming that the cohesive energy and strain energy of these structures dominate the
formation energy and the contribution from van der Waals inter-layer attraction is small.
In fact, the OLC and fullerene results are indistinguishable (within uncertainties) below
approximately 2000 atoms. Secondly, the atomic heat of formation of buckydiamond is
more akin to the OLC than the nanodiamonds, the latter being the least stable below
1100 atoms, but most stable beyond 1600 atoms. Finally, in the region from 500 to
1850 atoms the simulations predict that a coexistence region has been formed. Within
21
FIG. 3: Enthalpy of formation of carbon nanoparticles, indicating the relative subregions of co-
existence of buckydiamond with other phases. Uncertainties in the form of error bars have been
included (shown only for bucky-diamond here, for the purposes of clarity), indicating the quality
of t of the model to the calculated values. Reproduced with permission from A. S. Barnard, S.
P. Russo and I. K. Snook, Phys. Rev. B, 68, 073406 c _(2003) by the American Physical Society.
this region, buckydiamond is coexistent with the other carbon nanoparticles
56
.
This broad region was then further broken into three subregions, marked as A, B and
C in Figure 3. Within subregion A from 500 to 900 atoms (1.41.7nm), the heat of
formation of buckydiamond is indistinguishable from that of fullerenes (within uncertain-
ties), although carbon-onions represent the most stable form of nanocarbon, and nanodi-
amond the least stable. In subregion B, between 900 and 1350 atoms (1.72.0nm),
buckydiamond and carbon-onions coexist (within uncertainties). In subregion C, bucky
diamond was found to coexist with nanodiamond (within uncertainties) between 1350 and
22
1850 atoms (2.02.2nm). Barnard et al
56
also point out that the intersection where the
buckydiamond structure becomes more favorable than carbon-onions, is very close to the
intersection for nanodiamonds and fullerenes at 1130 atoms. This, they suggest, indicates
that at approximately 1130 atoms a sp
3
bonded core becomes more favorable than a sp
2
bonded core, irrespective of surface structure.
In general, these results show that the existence of buckydiamonds provides a smooth
transition from fullerenes to nanodiamond, rather than an abrupt change in phase stability.
The shallow nature of the crossings of the heats of formation do raise questions as to
whether an actual crossing-point may be accurately distinguished. However, such ambiguity
in the assigning of crossing points only supports the hypothesis of a coexistence region,
further highlighting that the relationship between the various types of carbon bonding at the
nanoscale is more complicated than previously thought. Still, understanding the relationship
between the sp
3
and sp
2
portions of a hybrid structure will be especially critical when
considering the interface between the phases, and the transition mechanisms.
As opposed to the models describing relative stability nanocarbon particles, compar-
atively fewer theories have been developed to describe the mechanisms occurring at the
interface between the sp
2
and sp
3
phases. The issue of the phase transition between nanodi-
amond and OLC structures was addressed explicitly by Zaiser and Banhart
14
, who presented
a thermodynamical quasi-equilibrium theory to explain this irradiation-induced transforma-
tion of OLC to nanodiamond. In this model it was theoretically shown that irradiation of
OLC leads to the destabilization of graphitic structures with respect to low-pressure growth
of diamond, due to the large dierence in the cross subsections for irradiation-induced dis-
placements of carbon atoms in diamond and graphite. A non-equilibrium phase diagram was
calculated showing the stability of graphite and diamond (as a function of the displacement
rate of atoms), and the results related to the experimentally observed results. The initial
premise of the approach used by Zaiser and Banhart was that the issue of nucleation was
excluded in favor of considering the phase transformation as the motion of a phase boundary
separating the two (solid) allotropes. Considering a suciently large cluster of N carbon
atoms, consisting of N
D
sp
3
bonded (diamond) atoms and (1-N
D
) sp
2
bonded (graphitic)
atoms, separated by a sharp interface. Irradiation of the cluster allows atoms at the interface
to oscillate between diamond and graphite bonding via two mechanisms
14
.
The rst proposed mechanism was via thermally activated processes resulting in thermal
23
jump rates
GD
th
and
DG
th
, such that:

GD
th

DG
th
= exp[G], (38)
where G is the Gibbs free energy dierence between diamond and graphite, and =
(k
B
T)
1
, with k
B
as Boltzmanns constant and T the temperature. The authors then intro-
duced the total thermally activated jump rate
th
across the interface, giving:

GD
th
= p
GD

th
,
DG
th
= p
DG

th
, (39)
in terms of the probabilities,
p
GD
=
1
1 +exp[G]
, p
DG
=
exp[G]
1 +exp[G]
. (40)
The second mechanism was via ballistic processes, taking place rst by displacing either
sp
3
or sp
2
bonded atoms due to collisions by irradiating particles, followed by the annihilation
of the displaced atoms on vacant lattice sites of the opposite phase. In describing this
process the authors assumed that only single interstitials and vacancies are produced by the
irradiation, that the exchange of atoms between diamond and graphite occurs only at the
interface, and the probability (either p
GD
or p
DG
) for such an exchange was governed by
the Gibbs free energies of the respective phases. The ballistic jump rates were given in
terms of the irradiation ux , by:

GD
irr
= p
GD

G
,
DG
irr
= p
DG

D
, (41)
where
G
and
D
are the total displacement cross subsections for sp
2
and sp
3
atoms, respec-
tively.
The probability P
N
D
of nding N
D
sp
3
bonded atoms, in terms of the number N
i
of atoms
at the interface and the total jump rates (
GD
tot
=
GD
th
+
GD
irr
,
DG
tot
=
DG
th
+
DG
irr
),
was described by:
1
N
i
P
N
D
t
=
GD
tot
P
N
D
1
+
DG
tot
P
N
D
+1
[
DG
tot
+
GD
tot
]P
N
D
, (42)
for which the solution (under steady-state conditions where there is no net ux of matter),
required that the condition
GD
tot
P
N
D
1
=
DG
tot
P
N
D
was fullled. Thus, P
N
D
was found to
be:
P
N
D
= Cexp[(N
D
)] (43)
24
where,
(N
D
) = (0) +
N
D
1

0
ln

DG
tot

GD
tot
, (44)
was stochastic potential, and C is a normalization constant.
In the absence of irradiation corresponds to G, and for the irradiated system a
non-equilibrium eective free energy G
eff
was dened to govern the phase stability such
that G
eff
= k
B
T[(N) (0)]/N. By making the appropriate substitutions, Zaiser and
Banhart found that,
G
eff
= Gk
B
Tln
_
1 +
G
/
th
1 +
D
/
th
_
. (45)
The authors point out that G
eff
reduces to G when = 0, and at high temperatures
when
th

G
,
D
, otherwise ballistic jumps predominant as
th

G
,
D
. However,
if
G
/
D
> 1, the non-equilibrium G
eff
of the sp
3
atoms is reduced by k
B
Tln[
G
/
D
] with
respect to G
14
.
The impact of irradiation on a crystalline structure was included as the threshold displace-
ment energy E
d
(the minimum energy required to permanently displace an atom), which is
weakly anisotropic in nanodiamond and strongly anisotropic in graphite. The ratio
G
/
D
was found to decrease with increasing electron energy, and a ratio of
G
/
D
> 1 was found
to be necessary for an irradiation-induced phase transition
14
.
It was assumed that in equation (45) that
th
=
0
exp[G
i
] was governed by an Arrhenius
law; where the attempt frequency
0
was taken 10
12
s
1
, and the characteristic free activation
energy (required for crossing the interface) G
i
was taken as 4 eV. With experimental values
of G(T) (at zero pressure and the atomic volumes) , G
eff
was determined for dierent
displacement cross-subsection ratios, and it was found that diamond was more stable than
graphite provided that G
eff
< 0. Using this criterion non-equilibrium phase diagrams
at zero pressure were constructed, showing that the phase transition takes place within an
intermediate temperature range, the extension of which depends on the irradiation intensity
as well as on the ratio of the total displacement cross subsections
14
.
Therefore, the mechanism responsible for the OLC phase transition at the interface was
attributed to ballistic displacements causing interstitials (predominantly from sp
2
lattice
sites); as probabilities of relaxing into sp
3
or sp
2
bonding congurations are of the same
order of magnitude, this causes a net ux of atoms from the sp
2
to the sp
3
phase. If
T exceeds the critical temperature, thermal jumps predominate so that the sp
2
bonding
25
remains the stable, but if the irradiation energy is such that only sp
2
bonded atoms are
displaced the phase stability may be changed even at very low temperatures
14
.
The rate of transformation was also considered, expressed in terms of the velocity v of
the motion of the phase interface. It was found that (except in the vicinity of the upper
critical temperature), the transformation velocity was proportional to the displacement rate,
irrespective of the direction of the transformation.
Later, Butenko et al
58
derived a kinetic theory to determine a temperature dependent
isotropic rate of migration of the interface between the diamond and graphitic regions of
spherical nanocarbon particles. They employed the reducing sphere model that assumed
the phase conversion begins simultaneously at all interface surface points, and that (under
isothermal conditions) the interface moves with a constant isotropic rate inside the nanopar-
ticle. In a later study, the reaction rate of the nanodiamond to buckydiamond (or OLC)
transition was treated as a migration rate of the interface between the exfoliated fullerenic
shells and the diamond cores. Again, a reducing sphere model was used to obtain the
rates from the changes in densities. Estimated kinetic parameters in an Arrhenius expres-
sion (such as the activation energy) then allowed for the quantitative calculations of the
diamond graphitization rates in and around the critical temperature range
58
.
In general however, the issue of kinetic processes at the surface and interfaces of carbon
nanoparticles is a topic of further investigation by many in the eld. Challenges still exist,
including dependence on the phase transition mechanisms with the morphology of nanodi-
amond particles, as highlighted by a number of computational studies of the crystalline (or
structural) stability.
B. Crystalline Stability
As mentioned in the introduction, the transformation of nanodiamonds into OLC struc-
tures and the reverse phenomena (the transformation of OLC structures into nanodiamond)
have been observed experimentally, and a number of groups have simulated these transition
using quantum mechanical based methods. These simulations maybe loosely grouped into
two categories, the rst (sp
3
to sp
2
) examining the delamination of nanodiamond surfaces,
and the second (sp
2
to sp
3
) examining the coalescence and stability of the diamond-like
cores.
26
1. sp
3
to sp
2
transitions
In 1998 Winter and Ree
54
used the AMl
59
and PM3
26
parameterizations of the semi
empirical MNDO
25
method to optimize the geometry of diamond clusters. For the nanodi-
amond clusters, generated with 111 morphology, N
C
= N(4N
2
1)/3 and N
DB
= 4N
2
,
where N is the number of layers along the c-axis, N
C
is the number of carbon atoms and
N
DB
is the number of dangling surface bonds
54
.
The results of these relaxations showed varying degree of structural change in the C
10
,
C
35
, C
84
, C
165
and C
286
octahedral nanodiamonds, terminated in all directions with (111)(1
1):1db surfaces (one dangling-bond surfaces). In the small particles the surface atoms of the
relaxed diamond clusters altered the positions so as to increase the 2p-2p overlap and
allow for the formation of double bonds, thereby reducing the number of dangling bonds
and causing an elongation of the bonds connecting the surface and core (interior) atoms.
In addition to this, the surface atoms were found to form 6-membered rings, with C C
distances consistent with sp
2
bonds. The optimized geometry of the C
84
nanodiamond
exhibited a separation of the 74 surface atoms from the 10 core atoms, with the separation
distance increasing to more than 3

A. This is a consequence of the attening of the surface
corrugation in order to form bonds from the dangling orbitals. A similar result was
observed in the optimized C
165
nanodiamond, with the outer 130 atoms separating from the
inner 35 atom core
54
.
Examining the diamondtographite phase transition from a dierent aspect, the prefer-
ential exfoliation of the diamond (111) surface over the (110) and (100) surfaces considered
by Kuznetsov et al
10
. The authors used a two-layer cluster model for both the (111) and
the (110) surfaces, and physical arguments (rather than direct calculations) to excluding
possibility of exfoliation of the (100) surface.
The calculations were carried out using the semiempirical modied neglect of diatomic
overlap (MNDO) method
25
, within RHF theory (including only the valence electrons). The
study began with an estimation of the inter-layer binding energy for the pairs of (111) and
(110) planes. These binding energies were found to be 2.83 eV for the (111) surface and
3.93 eV for the (110) surface, respectively, indicating that the inter-layer binding energy
for (110) is signicantly greater than that for a corresponding sized (111) cluster. This
was followed by an examination the eect of surface relaxation on the inter-layer binding
27
energy, by applying a geometry optimization to the upper (111) and (110) surface of each
cluster; both of which resulted in surface attening (with no -bonded Pandey chain 2 1
reconstruction on the (111) cluster). The relaxation energies of the (111) and (110) surfaces
of 0.20 eV/surface atom and 0.27 eV/surface (respectively) indicated that relaxation does
not signicantly aect the binding energies
10
.
To study surface graphitization, Kuznetsov and colleagues calculated the total energy
of a graphite single-layer cluster, and dened the graphitization energy of a cluster as the
sum of the energies of the graphite and bottom layers less the total energy of the initial
diamond cluster. The resulting graphitization energies for the (111) and (110) surfaces of
0.003 eV/surface atom and 0.24 eV/surface atom clearly demonstrated that graphitization
of a (111) surface is preferred over that of a (110) surface. The positive graphitization
energies were attributable to the small size of the clusters
10
.
Inter-layer binding energies were also calculated for larger two-layer clusters with 19 inter-
layer bonds. In this case, the exfoliated top layers consist of 12 six-fold 3 4 rings, with
binding energies per bond of 2.67 eV for the (111) surface and 3.59 and 3.66 eV for the
(110) surface parallel and perpendicular to the surface carbon chains, respectively. These
binding energies although lower than the experimental energy of the CC bond in diamond,
but higher than the experimental graphitization activation energy of 1.953 eV
60
. The
authors proposed that the graphitization mechanism involves the mutual interaction of the
exfoliated graphitic sheets and the dehydrogenated diamond surface below; and that the
dierence between inter-layer binding energy along the (111) and (110) directions arises due
to dierent packing of the inter-layer bonds in the volume between the parallel planes.
Although all sp
3
CC bonds have the same energy in each diamond lattice directions, the
(mainly repulsive) inter-bond interactions depend on the orientations of the bonds with
respect to each other, since the average distances between bonds is relatively small (0.25
nm) and the each bond is occupied by two electrons. The authors also showed that the
attractive interaction between the dangling bonds on the (111) surface is higher than that
for the (110) surface (for the same reason), while the dependence of the inter-layer binding
energy for the two (110) C
48
H
16
clusters on the orientation of the six-fold rings (3 4 or
4 3) was attributed to the dierent sets of interacting sp
3
bonds.
Using the same approach, the authors suggest that the surface graphitization is initiated
by a signicant thermal displacement of a single carbon atom at temperatures close to
28
the Debye temperature, and the surface delamination of the (111) surface to form curved
graphitic sheets proceeds by a zipper-like migration mechanism. A mechanism for the self-
assembling formation of a mosaic surface structure on larger (micron size and up) diamond
particles was also presented, where the initial diamond surface blisters under tensile stress,
forming fullerenic bubbles, that appear as a partial fullerene-like cage structure that is
anchored to the remaining stable diamond surfaces around its edges
10
.
The numerous studies outlined above demonstrate the instability of the (111) diamond
surfaces, and a complicated relationship between sp
2
and sp
3
bonding in nanodiamond
crystals. To further investigate the importance of surface structure on the nano-morphology
of diamond nanoparticles and the relationship between stability and shape (as opposed to
stability and phase), Barnard et al
55,6163
undertook a broad ab initio study of nanodiamond
structures of octahedral, cuboctahedral and cubic morphologies, up to approximately 2 nm
in diameter. The calculations were performed with VASP
64
, using DFT GGA and the
exchange-correlation functional of Perdew and Wang (PW91)
65
.
To simplify the calculation as much as possible, and reduce the size of the calculations, the
values of orbitals describing the inner shell electrons may be frozen at their atomic values.
This is called the frozen core method, where that eect of the inner shells is represented by
a potential called a pseudopotential
66
. Various forms of these pseudopotentials are available
in the literature for various atoms. In the case of the work of Barnard and colleagues, ultra-
soft, gradient corrected Vanderbilt-type pseudopotentials (US-PP)
67
were used, as supplied
by Kresse and Hafner
68
.
The three cubic nanodiamonds were included in the study (the C
28
, C
54
and C
259
struc-
tures) with (100)(1 1):2db (two dangling bond) and (110)(1 1):1db (one dangling bond)
surfaces. In each case, the initial step of the relaxation process involved the reconstruction
of the (100) surfaces to the (2 1) structure
70
. Upon relaxation the smallest cubic nanodi-
amond decayed to a tetrahedral amorphous structure, even though the most energetically
favorable structure for a 28 atoms carbon cluster is known to be the C
28
fullerene. The
energy per atom (E/N
C
) for this cluster was found to be -6.351eV, which is considerably
higher than the E/N
C
of -6.928eV for a relaxed C
28
fullerene, even when the relaxation was
extended to a convergence of 10
6
eV. Clearly the fullerenic structure is energetically pre-
ferred, but the energy barrier for the conversion to a fullerene is too high for the conversion
to occur. The relaxation is therefore trapped in a deep local minima, and would require
29
simulated annealing to allow this cluster to convert to the (energetically preferred) structure
of a C
28
fullerene
63
.
The two larger cubic nanodiamonds were found to have surface reconstructions and relax-
ations comparable to bulk diamond, although a slight shearing of the lattice was however
observed in the case of the C
54
crystal
63
. The relaxed C
259
nanodiamond is shown in Figure
4(a).
Three octahedral nanodiamonds were also investigated, the bulk-terminated C
35
, C
84
and
C
165
octahedral structures. In agreement with the previous results of Winter et al
54
, the
C
35
crystal adopted of a more rounded appearance, but retained the diamond structure even
though the bond lengths altered signicantly. The C
84
octahedral nanodiamond also adopted
a rounded appearance as a result of the 74 surface atoms separating from the ten atom inner
core, forming an octahedral OLC structure. The shell-core separation distance was found to
be approximately 2.25

A (smaller than that observed by Winter and Ree
54
). Similarly, the
relaxation of the C
165
atom octahedral carbon nanocrystal showed the same transformation
in which the 130 surface atoms separate from the 35 atom core cluster, also forming an
octahedral buckydiamond (again with the shell-core separation distance of approximately
2.25

A)
61
. The C
165
relaxed structures are shown in Figure 4(b).
Finally, three structures were considered with cuboctahedral morphology, which was simi-
lar to that of a truncated octahedron, but with a surface area comprising 40% (111)(11):1db
surface area and 60% (100)(11):2db. The smallest cuboctahedral nanodiamond considered,
the C
29
structure, was found to transform into the C@C
28
endo-fullerene carbon atom upon
relaxation. The larger cuboctahedron with 142 atoms also exhibited a structural transition.
The (100) surfaces initially reconstructed to the (21) structure, followed by a reorientation
of the surface dimers to form the curved fullerenic cages on the (111) surfaces, as shown
in Figure 4(c). The same eect was however, less pronounced in the C
323
structure, where
only the central region of the (111) surface and the edges at the intersection of teh (111)
facets formed a fullerenic cages.
Therefore, the ab initio relaxations performed by Barnard et al
55,61
not only demonstrate
the preferential exfoliation of the (111) surfaces over lower index surfaces and the transfor-
mation of the octahedral nanodiamonds into octahedral OLC structures, but also that in
the absence of (111) surfaces nanodiamond structures may be stable. These ndings were
supported a semi-quantitative study of the hybridization of the C C within the clusters
62
.
30
FIG. 4: The relaxed (a) C
259
cubic nanodiamond, (b) C
165
octahedral nanodiamond, and (c) C
142
cuboctahedral nanodiamond of Barnard et al
62
.
The technique uses fractional values of the electron charge density in regions localized at the
center of bonds in the structure, to provided an easy way of visualizing sp
2
surface bonds in
isolation from sp
3
content, thereby highlighting chemical dierences between the core and
surface of their relaxed nanodiamond structures. A computational details and a complete
examination of this visualization method are given elsewhere
69
.
Around the same time it was shown that similar surface delamination and changes in
structure are also observed in relaxed spherical diamond nanoparticles. Raty et al
18
pre-
sented ab initio calculations on the eect of quantum connement and surface reconstruc-
tions in nanodiamond, performed with GGA and time dependent LDA (using a pseudopo-
tential, plane-wave approach) and semiempirical tight binding. The GGA calculations were
performed for the 1.4 nm cluster sizes, and the tight-binding calculations for the 23 nm
clusters sizes
18
.
Beginning with (ideal) bulk-diamond terminated spherical particles, the spontaneous low
temperature reconstruction dehydrogenated clusters of 1.4, 2, and 3.0 nm clusters resulted
in graphitization of the rst atomic layer of the (111) facets. The smaller clustered (studied
with GGA) exhibited surface delamination, followed by the formation of 5-membered rings
linking the delaminated graphene fragments with the remaining core atoms, producing a
31
FIG. 5: Two bucky diamond clusters: C
147
and C
275
( 1.2 and 1.4 nm in diameter, respectively).
These carbon clusters have a diamond core (yellow) and a fullerene-like reconstructed surface
(red). Reproduced with permission from J. Y. Raty, G. Galli, C. Bostedt, T. W. Buuren and L. J.
Terminello, Phys. Rev. Lett., 90, 37402 c _(2003) by the American Physical Society.
curved surface the same as the cages of Barnard et al
55,61,62
and the bubbles of Kuznetsov
et al
10
. Two examples are shown in Figure 5. The TB simulations performed on the larger
2 and 3 nm clusters (705 and 2425 atoms, respectively) also produced the same surface
reconstructions. The barrier between the ideal (bulk-diamond) surface structure and the
reconstructed surface on a diamond nanocrystal was found to be size dependent (increasing
as the size of the nanoparticle is increased). In the larger nanoclusters the barrier was of the
order of several tens of electron volts. These results are in full agreement with the previous
work of the other researchers outlined above
18
.
Raty et al also compared their results with the experimental x-ray emission and absorp-
tion spectra (of nanodiamonds from 1.4 to 4 nm in diameter), in an attempt to explain
the presence of pre-edge features. The authors speculated that the above mentioned surface
reconstruction of nanodiamond was responsible for the features observed in absorption spec-
tra, corresponding to the existence of 5-membered rings (usually observed in the absorption
spectra of fullerenes). Based on these results and comparisons, the existence of a new family
of carbon clusters, buckydiamonds, was dened as carbon nanoparticles with a diamond
core (of a few nm) and a partial or complete fullerenic outer shell
18
.
Also observing these fullerenic bubbles forming on surfaces of nanodiamond structures
during simulations were Fugaciu, Hermann and Seifert
71
, in their study of the thermal sta-
32
bility of spherical diamond nanoparticles. They performed molecular dynamics calculations
based on approximate Kohn-Sham equations, using electronic states with a LCAO to de-
scribe the KS orbitals. The interatomic forces were described using DFTB, and the equations
of motion were integrated numerically using the Verlet algorithm, in time steps of 1.2 fs.
The total simulation time was between 20 to 50 ps, and once the cluster was considered to
be qualitatively stable and it was cooled down to zero temperature within about 2 ps
71
.
The simulations were carried out at temperatures in the range of 1400 K to 2800 K, which
is comparable to experimental conditions such as laser evaporation. In addition, irradiation
eects were taken into account by the random introduction of additional stochastic external
momenta, with a maximum kinetic energy of 40 eV. This simulated irradiation (applied
in intervals of 30 to 50 time steps), also introduced elements of the MC method, thereby
improving the chance of the system to exit weak local minima. The authors also reported
that the irradiation accelerated the nanocarbon transformation process and aected the
primary characteristics of the nal structures. Using this procedure they analyzed the
structural evolution of 50 diamond nanoparticles, ranging from 64 to 275 atoms.
The authors reported that the nanodiamond particles of approximately 1.1 nm size (120
atoms) underwent fragmentation and formed fullerenic cages, but that the nanodiamond
particles >1.2 nm (150 atoms or more), transformed to form two-shell OLC structures.
Examples of their results are shown in Figure 6. The diamond particles consisting of 159
atoms were shown to transform spontaneously, without simulated irradiation, within a total
simulation time of about 10 ps; while the diamond particles with >191 required the applied
radiation (or extended simulation times) in order to completely transform. The number
of atoms per shell uctuated. In one example, the inner core contained only a C
2
carbon
dimer. The structure of the shells also uctuated, consisting of pentagons and hexagons,
along with polygons with four and seven vertices. The bond length distribution within the
individual shells was found to be similar to the C C bonds in fullerenes, ranging between
1.38 and 1.50

A
71
.
In most of the OLC clusters the two concentric shells were found to be connected by
cross-links (atoms belonging to the inner and outer shells connected by a sp
3
bonds) with
a bond lengths of 1.540.6

A. Particles containing such bonds were termed spiraloidal
particles. It was also shown that simulated high irradiation intensity reduces the number of
sp
3
cross-links, and increases the distance between the inner and outer shells. In all cases,
33
FIG. 6: Transformation of a diamond particle into a concentric-shell fullerene at 1600 K under
simulated irradiation. (a) Initial state: a 191-atom diamond. (b) Final state after 40 ps: a two-
shell nucleus of a concentric-shell fullerene (green inner shell, yellow outer shell). (c) Example of
a concentric-shell fullerene with cross links between the two concentric shells (green: inner shell,
yellow: outer shell, orange: cross links with diamond-like coordinated atoms). Reproduced with
permission from F. Fugaciu, H. Hermann, and G. Seifert, Phys. Rev. B, 60, 10711 c _(1999) by
the American Physical Society.
the cross-link atoms had not been nearest-neighbors in the initial structure, highlighting
that the cross-links appearing in the nal structures are not residues of the initial state but
have developed during the transformation at late stages of the formation of the concentric
shells
71
.
Using the knowledge that the temperature and intensity of simulated irradiation are pa-
rameters which are suitable to control the transformation process, this study was extended
to consider precursors with a graphitic atomic order
72
. Initial structures with spherical
shape, and rod-like particles consisting of two elongated parallel graphene sheets were an-
alyzed. The spherical particles consisted of two or three circular graphene sheets. In the
case of three layer cluster, the smallest structure was found to evaporate in the rst period
of the transformation process, followed by the two remaining layers connecting at one or
more sites at their boundaries resulting in closed or open cages. The rod-like particles often
transformed to elongated closed structures, somewhat similar to short single wall nanotubes.
Certain particles formed carbon nanotips or fullerenic cones, that the authors suggested to
be of use for tips of scanning tunnelling and atomic force microscopes. Overall no correla-
34
tion was found between geometry and temperature, with the exception of the observation
that larger particles require higher temperatures to transform into a stable state within the
simulation used during the study
72
.
The initial part of this work
71
was later repeated by Lee et al
73
, who also examined the
heat induced transformation of spherical nanodiamond particles into elongated fullerenes
using tight-binding molecular dynamics. Using the 275 atom spherical cluster examined
previously by other authors
18,71
, they conrmed the delamination of the 111 surfaces at
low temperature (and thus, the formation of buckydiamonds) and the transformation of
the cluster into a fullerenic structure (with cross-linked atoms between the outer shell and
the inner core) at a temperature of 2500 K
73
.
Upon cooling (2500 to 2000 K) the cluster was found to completely transform to an
elongated, tube-like fullerene via three mechanisms. These include a ow out mechanism
where inner atoms exit the core region via holes and defects in the outer shell, a direct
absorption mechanism where atoms within the shell are adsorbed into the shell structure,
and a push-out mechanism where inner atom replace the surface atoms that are pushed
into the vacuum
73
.
Therefore in summary, it has been shown by the studies described in this section, that the
(111) surfaces of diamond nanoparticles is structurally unstable and delaminated partially or
completely from the core. New family of carbon clusters, denoted as buckydiamonds, has
previously been established to describe dehydrogenated carbon nanoparticles comprising of
a diamond core and a fullerenic reconstructed surface
18
. It has also been shown that upon
annealing to high temperature, these structures may decay further into complete fullerenes
or OLC structures
73
.
However, another approach to the crystalline stability of dehydrogenated nanodiamonds
is to examine the stability and the coalescence of the cores, rather than the instability of
the surfaces.
2. sp
2
to sp
3
transitions
As part of a study on the graphitization of diamond surfaces and the diamond/graphite
interface using tight-binding computer simulations, Jungnickel et al
74
examined the equilib-
rium structure of a tetrahedrally bonded icosahedral structure of 300 atoms. The icosahedral
35
structure was approximately spherical and had a high degree of symmetry. The structure
was found to be stable in a conjugate gradient relaxation, and although the energy was 0.30
eV/atom higher than a C
300
two-shell carbon onion (a C
60
fullerene surrounding by an icosa-
hedral C
240
fullerene, with average radii of 3.6 and 7.1

A respectively), the cluster could not
be induced to form the corresponding (lower energy) carbon-onion, even at temperatures of
12002700 K. The authors attributed the preservation of the sp
3
bonded diamond structure
to the (high) energy barrier represented by the signicant restructuring required to form the
corresponding sp
2
nanoparticle
74
.
In a later attempt to model the sp
2
to sp
3
nanocarbon phase transition, Astala et al
75
re-
ported results of atomic simulations using the DFTB method, to model the collision-induced
nucleation and growth of nanodiamond inside the 300 atom OLC structure mentioned above.
The group began by simulating the release of atoms due to knock-on displacements in outer
shells and their transport to the core, by a sequence of random atom additions with zero ini-
tial velocities. The evolution of the structures was found to be dependent upon the number
of interstitial atoms during the ad hoc injection process. The inner C
60
core was seeded with
a carefully selected C
10
cluster. The irradiation was simulated via the random introduction
of varying numbers of additional interstitial carbon atoms, and the growing structure was
relaxed after each addition. This calculation set was denoted as simulation A. The results of
simulation A were then qualitatively and quantitatively compared with simulation B, which
employed a dierent sampling regime, and simulation C where the carbon atoms were added
one at a time
75
. The results of Simulation A will be summarized here.
Simulation A involved the random insertion of 50 carbon atoms between the inner seeded
C
60
and outer C
240
shells. The structure was relaxation using microcanonical MD for 240
fs during which the structure was thermalized to 2000 K and then linearly cooled in a heat
bath to 1300 K. The process was repeated with ten additional atoms twice, before adding
sets of 20 carbon atoms at random positions inside the structure to simulate more intense
irradiation and very rapid thermal dissipation to the surroundings. Finally, the conjugate
gradient (CG) technique was used to generate the nal snapshot of the structure. This
procedure was repeated until a maximum of 150 additional interstitial atoms had been
added to the initial OLC structure
75
.
After each simulations set was analyzed, the bonding geometries of the central cores of
selected nal relaxed systems for sets A, B and C were compared. The results indicated that
36
a critical size for spherical atomic arrangements exists, below which sp
2
structures dominate,
and above which sp
3
atoms bond into a diamond-like structure. The connement by multi-
layer carbon fullerenes was suggested to lower this limit and to reduce the capabilities of sp
3
bonded atoms to relax to a graphitic phase. The authors also described the transformation
as percolation of initially dispersed sp
3
-bonded regions fusing after a certain threshold, to
form one sp
3
core cluster
75
.
Therefore, if connement by multi-layer carbon fullerenes is responsible for inhibiting re-
laxation of sp
3
bonded atoms into a graphitic phase, this mechanism may contribute to the
stability of diamond-like cores at the center of the buckydiamond structures. The mech-
anism for the stabilization of buckydiamond surfaces however, appears to be appropriate
passivation.
IV. HYDROGENATED NANOPARTICLES
In the theoretical models and computer simulations described above, the surfaces of the
nanodiamond particles have been clean (or dehydrogenated). While this has contributed to
many interesting results including the relationship between the sp
2
and sp
3
bonded atoms
within structures such as buckydiamonds, this type of instability may prove problematic for
nanotechnologists desiring rigid, faceted and structurally (and energetically) stable diamond
particles at the nanoscale.
To address this issue, a number of studies have been undertaken considering hydrogenated
nanodiamonds, and their relationship to graphitic and diamond-like carbon nanomaterials.
As before, section IVA outlines a number of analytical models used to describe nanocarbon
phase stability (that typically require computations to be performed in obtaining various
parameters), whereas section IVB outlines results generating using purely computational
methods.
A. Phase Stability
As early as 1990 Badziag et al
76
suggested that small hydrogenated nanodiamonds may
be more stable than graphite. In this study the binding energy of carbon atoms in small
sp
3
bonded hydrocarbon molecules and sp
2
bonded polycyclic aromatics (the precursors to
37
graphite clusters) was compared. They examined several (fully optimized) small clusters
ranging from C
2
to C
60
; including the C
6
H
6
benzene ring up to a ten ring C
32
H
14
hexagonal
nanographene sheet, and tetrahedral clusters ranging from methane to a C
26
H
30
molecule
containing six adamantane cages. A simple model was used to calculate the molecular
energies of the tetrahedral clusters,
E
total
= N
H
E
H

1
2
_
4N
C
N
H
_
E
t
(46)
so that,
E
total
N
C
=
N
H
N
C
_
1
2
E
t
E
H
_
2E
t
(47)
and for hexagonal clusters,
E
total
N
C
=
N
H
N
C
_
1
2
E
h
E
H
_

3
2
E
h
(48)
where N
C
is the number of carbon atoms, N
H
is the number of terminating hydrogen atoms,
E
H
is the energy of a C H bond, E
h
is the energy of a hexagonal C C bond, E
t
is the
energy of a tetrahedral C C bond, and E
total
is the total energy of the cluster. Estimation
of the E
total
diamond and graphite was calculated using MNDO
76
.
The binding energy was plotted as a function of the hydrogen to carbon ratio, with
the results forming two approximately linear trends, corresponding to the hexagonal and
tetrahedral clusters respectively. The results showed that the binging energy of compact
particles (both hexagonal and tetrahedral) have a lower binding energy than less compact
(hexagonal and tetrahedral) particles, and the sign and magnitude of the slopes of the best-
ts agreed with experimental observation (although values were not reported). However,
although the intercept for the tetrahedral cluster was 0.041 eV higher than that of the
hexagonal clusters, conrming the phase stability of bulk graphite over bulk diamond, the
dierence was twice the experimental dierence of 0.195 eV
76
.
By examining the intersection point of the best-ts, their results predicted that small
hydrogenated nanodiamonds are more stable than graphite. To estimate the energy per
atom of an innite graphene sheet (without interlayer interaction), the authors subtracted
the energy of the concave ve-ring C
22
H
14
molecule from that of the convex nine-ring C
30
H
14
molecule, and compared the resulting energy to that obtained by subtracting the ve-ring
C
22
H
14
molecule from the eight-ring C
28
H
14
structure
76
.
38
The actual crossover of particle stability occurred at a hydrogen to carbon ratio of 0.24.
This corresponds to a nanographene sheet with approximately 100 carbon atoms and an
octahedral nanodiamond containing 2925 atoms; or with a diagonal length of 4.3 nm (and
an edge length of 3.0 nm). Thus, hydrogenated nanodiamonds less than 3-5 nm were shown
to be more stable than nanographene with only 100 atoms (but the same H/C ratio),
however by shifting the tetrahedral cluster intercept to match the experimental dierence
between diamond and graphite the results changed signicantly. With a dierence in the
intercepts of 0.195 eV (retaining the slopes) the intersection point became H/C=0.12, which
corresponds to an octahedral diamond nanodiamond with approximately 21000 atoms and a
diameter of 6.0 nm. Further, by ignoring the small tetrahedral molecules (examining only
the tetrahedral molecules with a cage structure), the intersection shifted again to H/C=0.5,
corresponding to an octahedral nanodiamond with 300 atoms and a diameter of 1.3 nm
76
.
Following in the same vein Winter and Ree
54,77
outlined a method for investigating carbon
particle phase stability based on the heat of formation of graphene sheets and hydrogenated
nanodiamonds. semiempirical AMl
59
and PM3
26
MNDO
25
calculations were carried out
using the GAMESS
78
, HyperChem
79
, and SPARTAN
80
molecular electronic structure codes;
and DFT calculations with the Gaussian94
81
molecular orbital program using the Becke
hybrid HF exchange and DFT exchangecorrelation functional with a 6-31g basis set, to
calculate the total energy of a sp
2
and sp
3
carbon clusters. They then used an expression
for the heat of formation as a function of cluster size to predict the relative stability of the
graphite and diamond phases of nite carbon particles
54,77
.
Recall that for the nanodiamond clusters, generated with 111 morphology, N
C
=
N(4N
2
1)/3 and N
DB
= 4N
2
, where N is the number of layers along the c-axis. In
the case of the two-dimensional graphite nanoparticles, the number of carbon atoms in the
cluster N
C
= 6N
2
and the number of dangling bonds N
DB
= 6N, where N is the num-
ber of carbon atoms along one edge. The singly occupied surface orbitals on the diamond
and graphene cluster were terminated with hydrogen atoms, and structures relaxed using
molecular mechanics energy minimization
54,77
.
The total energy per carbon atom as a (linear) function of the hydrogen to carbon ratio
was expressed as:
E
total
N
C
= E
C
+
N
H
N
C
E
H
(49)
where E
total
is the total energy of the cluster, E
C
is the energy per carbon atom and N
C
39
and N
H
is the total number of carbon and hydrogen atoms, respectively. The C H bond
energy E
H
was obtained from the coecients determined from the least squares t of the
energy as a function of N
H
/N
C
, for the graphite and diamond clusters
54,77
.
Extrapolation of the PM3 graphene cluster energies predicted a cohesive energy of 7.243
eV and the AM1 parameterization gave E
C
= 7.198 eV. The extrapolated E
C
for graphene
did not include the attractive inter-planar energy, so cohesive energy for three-dimensional
graphite was expected to be 0.043 eV lower than extrapolated values. For the diamond
clusters the PM3 method predicted a cohesive energy of 7.244 eV and the AM1 method
gave 7.169 eV. The authors also performed DFT calculations for the smallest three sp
2
and
sp
3
clusters, from which E
C
= 7.28 eV and E
C
= 6.744 eV for graphene and diamond were
obtained, respectively
54,77
.
In their phase stability model
54
the heat of formation as a function of cluster size for
graphite and diamond phases were expressed in terms of the C C and C H bond
energies (E
CC
and E
CH
), and the atomic heats of formation of free carbon (H
o
f
(C)) and
hydrogen (H
o
f
(H)) atoms. After dividing each equation by the number of carbon atoms
N
C
, linear relationships between the heats of formation and the number of surface bonds
per atom were obtained,
H
o
f
(sp
2
)
N
C
=
3
2
E
sp
2
CC
+
N
H
N
C
(E
sp
2
CH

1
2
E
sp
2
CC
+ H
o
f
(H)) +
+H
o
f
(C) +
1
2
E
disp
(50)
H
o
f
(sp
3
)
N
C
= 2E
sp
3
CC
+
N
H
N
C
(E
sp
3
H

1
2
E
sp
3
CC
+
+H
o
f
(H)) + H
o
f
(C) (51)
where N
H
is the number of hydrogen atoms and E
disp
is the carbon-carbon pair energy
due to the graphite inter-layer interaction. Using three-dimensional graphite clusters and a
model potential for the dispersive attraction between aromatic rings, they calculated a value
of E
disp
to be 0.072 eV. By equating the intercepts to the experimental cohesive energies
of macroscopic graphite and diamond, bond energies of E
sp
2
CC
= 4.979 eV and E
sp
3
CC
= 3.706
eV were obtained. The C H bond energies were then determined by tting the semi
empirical AM1 and PM3 cluster calculations, giving E
sp
2
CH
= 4.363 eV and E
sp
2
CH
= 4.767 eV,
respectively. These values were inserted into the phase stability model, and the point of
intersection determined for each potential. In each case, the model predicted that for small
40
carbon clusters the diamond structure is more stable; with the AM1 potential predicting a
crossing point at approximately 70,000 atoms, and the PM3 calculations giving a similar
curve with a crossing point at approximately 33,000 carbon atoms. While these predictions
dier by a factor of two, the authors pointed out that both results are the same order of
magnitude as the grain sizes observed for isolated nanodiamonds, experimentally
54,77
.
More recently, Raty and Galli
82
used rst principles calculations to study the relative
stability of nanodiamonds as a function of the surface hydrogen coverage as well as size
82
.
Their study encompassed totally passivated surfaces to clean fully reconstructed surfaces
(buckydiamonds). Their results indicated that as the size of diamond is reduced to about
3 nm, it is energetically favorable for this material to have clean, reconstructed surfaces
than hydrogenated surfaces. Further, they suggested two ranges for the hydrogen chemical
potential corresponding to two dierent growth conditions of diamond: one that favors the
formation of UNCD or UDD, and the other that favors the formation of bulk diamond-like
lms
82
.
In this study, the relative stability of structures with the same carbon content but dierent
hydrogen coverage were studied using the formation energy of a carbon particle. The forma-
tion energy was dened as the dierence in energy between a nanoparticle and a reservoir
of carbon and hydrogen atoms whose contribution to the free energy is
x
, such that,
E
formation
= E
total

C
N
C

H
N
H
E
vib
. (52)
Here N
x
are the numbers of carbon or hydrogen atoms and
x
are their respective chem-
ical potentials; E
vib
and E
total
are the vibrational and the total energy of a nanoparticle,
respectively, and were obtained using DFT
82
.
Their calculations considered ve (spherical) particles sizes containing 29, 66, 147, 211 and
275 carbon atoms, nding that the stability sequence of the particles with dierent surface
structures is the same, as a function of
H
. By comparing various degrees of hydrogen
coverage, the dierence in formation energy between particles with hydrogenated surfaces
and those with bare surfaces decreases as the size of the nanoparticle increases, thereby
suggesting that there exists a size in the nanometer range where a reversal of stability
between hydrogenated and bare nanoparticles will occur and buckydiamonds become more
stable than diamond nanoparticles with hydrogenated surfaces
82
.
The computed dierence in formation energy per carbon atom () between the stable
41
hydrogenated structure and the bare diamond nanoparticle as a function of size was dened
as,
=
_
E
formation, bare
N
C
_
3/2

_
E
formation, stable
N
C
_
3/2
(53)
which is dependent only on the number of surface carbon atoms. For all values of the
hydrogen chemical potential, the authors show that becomes negative when the diameter
of the nanoparticle is between 2 and 3 nm. They determined that does not depend
signicantly on the hydrogen chemical potential, and is independent of dierent experimental
synthesis conditions. These interesting results suggest that between 2 and 3 nm, bucky
diamonds are energetically preferred over hydrogenated nanodiamonds
82
.
Although the calculations cannot establish the exact size at which the crossover between
hydrogenated and bare, reconstructed surfaces occur, the authors also presented simple
thermodynamic mechanism to compare the formation energies of nanodiamonds of various
sizes with those of at diamond surfaces. Using this comparison it was illustrated that
at the highest values of the hydrogen chemical potential (15.5 eV) considered, the innite
(100) surface is more stable than any nanodiamond; whereas at lower chemical potential
there exists a critical diameter above which the nanodiamond clusters are the most stable
structure. For example, at a chemical potential value of 16.5 eV (if the particle grows to a
diameter larger than 2.5 nm), such particles were shown to be more stable that a bulk surface,
in agreement with the predictions and results of the numerous other studies outlined above.
The numerical results contained within this paper were then used to suggest how varying the
hydrogen pressure (and thus its chemical potential) during synthesis, may promote either
microcrystalline or nanocrystalline diamond lms
82
.
Unfortunately, although this study makes a solid attempt to address the relative stability
of clean and hydrogenated nanodiamond particles, deciencies in the approach leave some
ambiguity in the results. Even though a great deal of importance is placed on the role of the
surface structure of the particles during the study, and care is taken to alter the formation
energy as the surface coverage of hydrogen is reduced, the formation energy does not account
for the chemical dierences between atoms on the clean (delaminated) surfaces and those
in the core of the particle. The former, which will be sp
2
hybridized, would have a dierent
chemical potential to the latter, which will be sp
3
hybridized. Therefore, for each N
sp
3
C
,
there should be an additional term for N
sp
2
C
which diminishes as N
H
increases. It is unclear
as to how this would eect the results.
42
B. Crystalline Stability
By explicitly examining the geometry and bonding in relaxed hydrogenated nanodia-
monds, the eects of surface hydrogenation on particle stability have also been examined.
Although the delamination of (111) surfaces promotes the formation of buckydiamonds
(as outlined above), it has also been shown that surface hydrogenation may prevent the
delamination of these surfaces and improve overall stability
54,62,83
.
During the study of Barnard et al
62,83
(using DFT GGA as described above in section
III B1) the carbon framework of the relaxed dehydrogenated and hydrogenated version of
the same nanodiamonds were directly compared. The results showed that not only was the
transformation to buckydiamond eliminated when the (111) surfaces were passivated with
hydrogen, but that the hydrogenated nanodiamonds exhibited more bulk-diamond like prop-
erties such as cohesive energy
83
and surface structure
63
. For example, the C
165
octahedral
nanodiamond was found to transform into a buckydiamond (as described above), whereas
the hydrogenated version of this structure (C
165
H
100
) did not. The C
165
H
100
nanodiamond
retained the diamond structure, with no graphitization of the (111) surfaces, and exhibits
little change in the length of surface bonds compared to bulk diamond. Similarly, although
the C
142
nanodiamond exhibited preferential delamination of the (111) surfaces, the C
142
H
72
was found to be structurally stable upon relaxation. Examples of the hydrogenated nanodi-
amond structures are shown in Figure IVB
62
.
While Winter and Ree
54
used analysis of the bond lengths to characterized the hydrogen
induced stability, Barnard et al
62
also considered the symmetry as a measure of the preser-
vation of a diamond-like structure. This was considered in terms of changes in the point
group of the relaxed nanodiamonds. A departure from the ideal diamond T
d
point group
(outside an allowable tolerance) was considered as a measure of instability
62
.
The point group was determined for each of the dehydrogenated and hydrogenated nan-
odiamond structures treated during their study, using the program SYMMOL
84
. Using this
method the quality of t of an ideal point group to the structure of the carbon framework
was calculated, returning the point group of best t and the uncertainty associated with
the t. The results showed that for octahedral and cuboctahedral nanocrystals (initially
with the ideal T
d
bulk diamond point group), a deterioration of point group symmetry oc-
curs as part of the structural relaxation. In each case (by comparing the results for the
43
FIG. 7: The relaxed, hydrogenated (a) C
259
H
140
cubic nanodiamond, (b) C
165
H
100
octahedral
nanodiamond, and (c) C
142
H
72
cuboctahedral nanodiamond of Barnard et al
62
.
carbon framework of the dehydrogenated and hydrogenated structures), it was shown that
the hydrogenated nanodiamonds have a point group closer to that of ideal bulk diamond
62
.
This type of analysis is useful as the symmetry group of a structure is important in the
determination of a certain properties (such as elastic properties), as well as the applica-
tion of various software packages that employ symmetry elements of structures to improve
computational eciency.
In addition to the preservation of symmetry, Barnard et al
62,83
also found that the hy-
drogenation of nanodiamond surfaces reduced the full-crystal relaxations that result in the
expansion and contraction of nanodiamond lattices. Expansions and contractions are eects
that are peculiar to nanostructures, with no corresponding eects in bulk diamond. More
than just volume changes due to the relaxation of surface layers, full crystal contractions
(or expansions) involve all the atomic layers, right into the center of the nanocrystal. The
authors measured the volume of the nal relaxed nanodiamonds geometrically from the
coordinates (by calculating the sum of the volumes of the most suitable polyhedra), and
compared with the volume of the initial unrelaxed diamond particle. The change in the
nanodiamond volume was then determined, along with the fractional volume change.
44
In the case of the dehydrogenated nanodiamond, the change in volume was found to be
dependent on both the size and on the morphology of the nanocrystal. The cubic nan-
odiamonds, which did not undergo any surface exfoliation were found to contract upon
relaxation
63
. This contraction decreases as the size of the nanocrystal increases (as would
be logically expected since V 0 as N )
63
. However, for the octahedral nanodia-
monds, the transformation into buckydiamonds was found to cause an overall expansion in
the C
84
and C
165
structures, which also decreased with increasing nanodiamond size. Since
the smallest C
35
octahedral nanodiamond underwent only a mild contraction, the study
indicated that the nal volume of the octahedral nanodiamonds was due to a contraction
(induced by the small size), and an expansion due to the surface delamination. The result
was a net expansion. Similarly, the exfoliation of the (111) surface on the cuboctahedral
nanodiamonds again caused an expansion of overall volume, combined with the nitesize
induced contraction. These nanodiamonds have also undergone a contraction, but to a lesser
degree than the octahedral nanodiamonds. Hence a larger net expansion was observed.
Overall it was found that the stable nanodiamond structures underwent contractions,
while the buckydiamonds exhibited residual expansions. As hydrogenation of the surfaces
eliminates the transformation into fullerenic carbon, it was not surprising that the hydro-
genated nanodiamonds contract, with the degree of contraction decreasing with increasing
nanodiamond size. Also, the magnitude of the contractions of the hydrogenated nanodia-
monds was found to be less than the expansion or contraction of the dehydrogenated coun-
terpart, oering further conrmation that hydrogenation of nanodiamond surface induces a
more bulkdiamond like structure.
These ndings have also been conrmed by Barnard and Marks et al
83
via calculation
of Wannier functions, which are local bond centered functions (rather than atom centered).
The Wannier function calculations conrm that while the dehydrogenated structures contain
distorted and -bonds, the hydrogenated counterparts contained entirely -bonded
83
. The
surface structure of dehydrogenated and hydrogenated (100) nanodiamond surface has also
been analyzed (and compared to bulk diamond)
63
, and conrmed that surface hydrogenation
of nanodiamond structures serves not only to passivate the surfaces, but also promote more
bulkdiamond like reconstructions
63
.
45
V. CONCLUSIONS
In summary, the theoretical and computational research outlined above have focused
on the phase transitions between fullerenic and diamond forms of quasi-zero dimensional
nanocarbon. The transformation of nanodiamond clusters into sp
2
forms such as fullerenes
and OLC, and the intermediary structures collectively termed buckydiamonds has been
successfully simulated in agreement with experimental observations. Computationally these
transitions are well represented by quantum mechanical based methods such as DFT
18,55,61,62
,
semiempirical
10
and tight binding
54,71,75
.
However, as the description of OLC structures is an important part of modelling the
phase stability of nanocarbon, application of a method that includes van der Waals inter-
actions would be worthwhile, especially since the existence of cross-linking bonds (between
fullerenic shells) has been identied as a signicant factor aecting stability
71
. In addition to
this, the choice of computational technique has a signicant impact of the accuracy of calcu-
lated values such as (for example) cohesive energy of diamond and graphite; quantities that
inuence the accuracy with which the phase transition barrier heights may be calculated.
Irrespective of such dierences, the numerous analytical theories (the majority of which
use thermodynamic arguments) appear to agree as to the upper limit of nanodiamond phase
stability. All of the theories outlined above predict nanodiamond to be the stable phase
of carbon in the range less than about 5 6 nm. This range has also been conrmed via
ion implantation experiments
85
. The lower limit of nanodiamond stability has also been
estimated (again using a thermodynamic approach), along with the phase coexistence of
nanodiamond and fullerenic carbon at this lower limit, via the formation of buckydiamonds.
A thermodynamic coexistence region of this type at the upper limit of nanodiamond stability,
where larger clusters characterized by diamond-like cores and onion-like outer shells are
observed, has not yet been identied (if indeed it exists at all).
The formation of thermodynamic coexistence region(s) indicates that the phase tran-
sitions are not entirely thermodynamically driven. In such regions other factors such as
surface energies, surface stress and charge, as well as kinetic considerations, may be instru-
mental in inducing the change of phase. Therefore a complete examination of nanocarbon
phase stability should include not only the use of a sophisticated computational method
and large cluster sizes, but also theoretical terms to describe dependencies on a variety of
46
cluster properties. A comprehensive model such as this may be constructed by combining
features of the various theories outlined above, and must also be sensitive to changes in
cluster volumes resulting from the phase transitions as well as the number of C atoms).
The benets of a combined approach would be that the transition mechanisms and stabil-
ity regions would be coupled, allowing for the description of complicated phase phenomena
with a higher degree of certainty. The models outlined here have identied the critical size
ranges where future eorts should be focused. The ranges are between approximately 1.4
2.5 nm, where the coexistence region of fullerenes, buckydiamond and nanodiamonds
is located; and between approximately 4.5 10 nm, where the phase transition between
nanodiamond and larger graphitic structures occur. In general, the physical and chemical
explanations of nanocarbon phase transitions in these regions are inaccessible to experi-
mentalists, implying that this invaluable contribution to the study of nanocarbon must, at
present, come primarily from theoretical (and computational) approaches.
Acknowledgments
This work was supported in part by the U.S. Department of Energys Oce of Basic
Energy Sciences, Division of Materials Science, under contract no. W-31-109-ENG-38. The
authors also wish to acknowledge the Victorian Partnership for Advanced Computing and the
Australian Partnership for Advanced Computing supercomputer centers for their supporting
of this project.

Electronic address: amanda.barnard@anl.gov

Electronic address: zapol@anl.gov

Electronic address: curtiss@anl.gov


1
H. W. Kroto, J. R. Heath, S. C. OBrien, R. F. Curl, and R. E. Smalley, Nature, 318, 14-20
(1985)
2
S. Iijima, Nature, 354, 56 (1991)
3
N. R. Greiner, D. S. Phillips, J. D. Johnson, and F. Volk, Nature, 333, 440 (1988)
4
D. M. Gruen Annu. Rev. Mater. Sci., 29, 211 (1999)
5
T. Sharda, and T. Soga, J. Nanosci. Nanotech., 3, 521 (2003)
47
6
T. Wang, H. W. Xin, Z. M. Zhang, Y. B. Dai, and H. S. Shen, Diamond Relat. Mater., 13, 6
(2004)
7
Y. Gogotsi, Cryst. Growth & Design, 1, 179 (2001)
8
O. A. Shenderova, V. V. Zhirnov, and D. W. Brenner, Crit. Rev. Sol. State and Mater. Sci.,
27, 227 (2002)
9
V. L. Kuznetsov, A. L. Chuvilin, Y. V. Butenko, I. Y. MalKov, and V. M. Titov, Chem. Phys.
Lett., 209, 72 (1994)
10
V. L. Kuznetsov, I. L. Zilberberg, Y. V. Butenko, A. L. Chuvilin, and B. Seagall, J. Appl. Phys.,
86, 863 (1999)
11
S. Tomita, T. Sakurai, H. Ohta, M. Fujii, and S. Hayashi, J. Chem. Phys., 114, 7477 (2001)
12
J. Chen, S. Z. Deng, J. Chen, Z. X. Yu, and N. S. Xu, Appl. Phys. Lett., 74, 4651 (1999)
13
F. Banhart, and P.M. Ajayan, Nature, 382, 433 (1996)
14
M. Zaiser, and F. Banhart, Phys. Rev. Lett., 79, 3680 (1997)
15
F. Banhart, J. Appl. Phys., 81, 3440 (1997)
16
Ph. Redlich, F. Banhart, Y. Lyutovich, and P.M. Ajayan, Carbon, 36, 561 (1998)
17
M. Zaiser, Y. Lyutovich, and F. Banhart, Phys. Rev. B, 62, 3058 (2000)
18
J. Y. Raty, G. Galli, C. Bostedt, T. W. Buuren, and L. J. Terminello, Phys. Rev. Lett., 90,
37402 (2003)
19
This term is used here to describe simulations of various levels of sophistication, where the
techniques are based on solutions of Schrodingers equation. While semiempirical techniques
such as Tight Binding will be included, empirical and multi-scale methods will be omitted from
this review, for the purposes of conciseness.
20
R. E. Christoersen, Basic Principles and Techniques of Molecular Quantum Mechanics,
Springer-Verlag, Berlin (1989)
21
M. P. Allen, and D. J. Tildesley, Computer Simulation of Liquids, Oxford Scientic Publications,
UK (1989)
22
W. J. Hehre, L. Radom, P. v. R. Schleyer, and J. A. Pople, Ab Initio Molecular Orbital Theory,
Wiley , New York (1986)
23
E. Clementi, S. J. Chakravorty, G. Corongiu, J.R. Fores, and V. Sonnad, in Modern Techniques
in Computational Chemistry: MOTECC-91, edited by Enrico Clementi, ESCOM Science Pub-
lishers B.V., Leiden, The Netherlands, pp. 23 (1991)
48
24
J. B. Foresman, and A. Frisch, Exploring Chemistry with Elexctronic Structure Methods, 2
nd
Edition, Gaussian Inc., Pittsburgh, PA, USA (1996)
25
M. J. S. Dewar, and W. Thiel, JACS, 99, 4499 (1977)
26
J. J. P. Stewart, J. Comp. Chem., 10, 209 (1989); J. Comp. Chem., 10, 221 (1989)
27
D. B. Boyd, Rev. Comp. Chem., 1, 337 (1990)
28
R. G. Parr, and W. Yang, Density-Functional Theory of Atoms and Molecules, Oxford Science
Publications, UK (1989)
29
P. Hohenberg, and W. Kohn, Phys. Rev., 136, B864 (1964)
30
W. Kohn, and L.J. Sham, Phys. Rev., 140, A1133 (1965)
31
R. Carr, and M. Parrinello, Phys. Rev. Lett., 55, 2471 (1985)
32
M. Parrinello, in Modern Techniques in Computational Chemistry: MOTECC-91, edited by
Enrico Clementi, ESCOM Science Publishers B.V., Leiden, The Netherlands, pp. 833 (1991)
33
M. C. Payne, M. P. Teter, D. C. Allan, T. A. Arias, and J. D. Joannopoulos, Rev. Mod. Phys.,
64, 1045 (1992)
34
VASP Users Manual, http://cms.mpi.univie.ac.at/VASP/
35
J. C. Slater, and G. F. Koster, Phys. Rev., 94, 1498 (1954)
36
C. Z. Wang and K. M. Ho, J. Comp. Theo. Nanosci., 1, 3 (2004)
37
C. M. Gorringe, D. R. Bowler, and E. Hernandez, Rep. Prog. Phys., 60, 1447 (1997)
38
A collection 6 reviews, Tight-binding Molecular Dynamics Simulations in Materials Science,
Comp. Mater. Sci., 12, 157 (1998)
39
J. Amlof, and H. P. L uthi, H. P., Theoretical Methods And Results For Electronic-Structure
Calculations On Very Large Systems - Carbon Clusters, in Supercomputer Research in Chem-
istry and Chemical Engineering, ACS Symposium Series, 353, 35 (1987)
40
J. D. Johnson, and M. S. Shaw, Shock Compression of Condensed Matter - 1991, Elsevier,
Amsterdam, p.333 (1992)
41
M. S. Shaw, and J. D. Johnson, J. Appl. Phys, 62, 2080 (1997)
42
M. Y. Gamamik, Nanostruct. Mater., 7, 651 (1996); Phys. Rev. B, 54, 2150 (1996)
43
M. Y. Gamamik, Sov. Phys. Solid State, 31, 126 (1989)
44
M. Y. Gamamik, Dependence of Crytals Specic Energy on Their Sizes, Academy of Science of
the Ukraine, Kiev, p.6 (1987)
45
N. M. Hwang, J. H. Hahn, and D. Y. Yoon, J. Cryst. Growth, 160, 87 (1996)
49
46
D. V. Fedosayev, B. V. Deryagin, and I.G. Varasavskaja, Surf. Coat. Technol., 38, 9 (1989)
47
N. M. Hwang, J. H. Hahn, and D. Y. Yoon, J. Cryst. Growth, 162, 55 (1996)
48
J. OM. Bockris, and S. U. M. Khan, Surface Electrochemistry, ch. 2.2.2., Plenum, New York,
(1993)
49
H. N. Jang ,and N. M. Hwang, J. Mater. Res., 13, 3536 (1998)
50
Q. Jiang, J. C. Li, and G. Wilde, J. Phys.:Condes. Matter, 12, 5623 (2000)
51
D. Zhao, M. Zhao, and Q. Jiang, Diamond. Relat. Mater., 11, 234 (2002)
52
J. A. Nuth, Nature, 329, 589 (1987)
53
A. S. Barnard, S. P. Russo, and I. K. Snook J. Chem. Phys., 118, 5094 (2003)
54
N. W. Winter, and F. H. Ree, J. Comp.-Aided Mater. Des., 5, 279 (1998)
55
A. S. Barnard, S. P. Russo, and I. K. Snook, Phil. Mag. Lett., 83, 39 (2003)
56
A. S. Barnard, S. P. Russo, and I. K. Snook, Phys. Rev. B, 68, 073406 (2003)
57
Y. Guo, Ph.D Thesis, California Institute of Technology (1992)
58
Yu. V. Butenko, V. L. Kuznetsov, A. L. Chuvilin, V. N. Kolomiichuk, S. V. Stankus, R. A.
Khairulin, and B. Segall, J. Appl. Phys., 88, 4380 (2000)
59
M. J. S. Dewar, E. Zoebisch, E. F. Healy and J. J. P Stewart, J. Am. Chem. Soc., 107, 3902
(1985)
60
V. L. Kuznetsov, Y. V. Butenko, A. L. Chuvilin, A. I. Boronin, R. Kvon, S. V. Kosheev,
S. V. Stankus, and R. Khairulin, in Extended Abstracts and Program, Vol. II, p. 326, The
Pennsylvania State University, The American Carbon Society, PENNSTATE, University Park
Campus (1997)
61
A. S. Barnard, S. P. Russo, and I. K. Snook, Diamond Relat. Mater., 12, 1867 (2003)
62
A. S. Barnard, S. P. Russo, and I. K. Snook, Int. J. Mod. Phys. B, 17, 3865 (2003)
63
S.P. Russo, A. S. Barnard, and I. K. Snook, Surf. Rev. Lett., 10, 233 (2003)
64
G. Kresse, and J. Hafner, Phys. Rev. B, 47, RC558 (1993); 54, 11169 (1996)
65
J. Perdew, and Y. Wang, Phys. Rev. B, 45, 13244 (1992)
66
See for example http://www.nirim.go.jp/ kobayak/readme.html
67
D. Vanderbilt, Phys. Rev. B, 41, 7892 (1990)
68
G. Kresse, and J. Hafner, J. Phys.: Condens. Matter., 6, 8245 (1994)
69
A. S. Barnard, S. P. Russo, and I. K. Snook, J. Comp. Theo. Nanosci., in press
70
J. Furthm uller, J. Hafner, and G. Kresse, Phys. Rev. B, 53, 7334 (1996)
50
71
F. Fugaciu, H. Hermann, and G. Seifert, Phys. Rev. B, 60, 10711 (1999)
72
H. Hermann, F. Fugaciu, and G. Seifert, Appl. Phys. Lett., 79, 63 (2001)
73
G.-D. Lee, C. Z. Wang, J. Yu, E. Joon, and K. M. Ho, Phys. Rev. Lett., 91, 265701 (2003)
74
G. Jungnickel, D. Porezag, Th. Frauenheim, M. I. Heggie, W. R. L. Lambrecht, B. Segall, and
J. C. Angus, Phys. Stat. Sol. (a), 154, 109 (1996)
75
R. Astala, M. Kaukonen, R.M. Nieminen, G. Jungnickel, and Th. Frauenheim, Phys. Rev. B,
63, 81402 (2001)
76
P. Badziag, W. S. Veowoerd, W. P. Ellis, and N. R. Greiner, Nature, 343, 244 (1990)
77
F. H. Ree and N. W. Winter, J. N. Glosli, and J. A. Viecelli, Physica B, 265, 223 (1999)
78
GAMESS, M. Dupuis, D. Spangler, and J. Wendoloski, NRCC Software Catalog, 1, Program
No. QG01 (GAMESS), (1980)
79
Hyperchem
TM
. Windows Molecular Modelling System, Hypercube, Inc. and Autodesk, Inc.
Developed by Hypercube, Inc.
80
SPARTAN. developed by M. L. Hall, http://www.lanl.gov/Spartan/
81
GAUSSIAN94, M. J. Frish, G. W. Trucks, M. Head-Gordon, P. M. W. Gill, M. W. Wong, J.
B. Foresman, B. G. Johnson, H. B. Schlegel, M. A. Robb, E. S. Replogle, R. Gomperts, J. L.
Andres, K. Raghavachari, J. S. Binkley, C. Gonzales, R. L. Matin, D. J. Fox, D. J. Defrees, J.
Baker, J. J. Stewart, J. A. Pople, (1994)
82
J.-Y. Raty, and G. Galli, Nature Materials, 2, 795 (2003)
83
A. S. Barnard, N. A. Marks, S. P. Russo, and I. K. Snook, Mat. Res. Soc. Symp. Proc., 740, 69
(2003)
84
T. Pilati, and A. Forni, J. Appl. Cryst., 31, 503 (1998); 33, 417 (2000)
85
S. Prawer, J. L. Peng, J. O. Orwa, J. C. McCallum, D. N. Jamieson, and L. A. Bursill, Phys.
Rev. B, 62, 16360 (2000)
51

Você também pode gostar