You are on page 1of 30

Chapter Two

Theoretical Background of Potential Model


and Atomistic Simulation Methods

2.1 Introduction

Computer modelling techniques are now well-established tools in


the field of materials science research, and have been applied
successfully to studies of structures and dynamics of solids on the
atomic and nano-scale. Atomistic simulation methods determine
the lowest energy configuration of the crystal structure by
employing efficient energy minimization procedures. The
calculations rest upon the specification of an interatomic potential
model, which expresses the total energy of the system as a function
of the nuclear coordinates. For solid-state ionics, the Born model
framework is commonly employed, which partitions the total
energy into long-range Coulombic interactions and a short-range
term to model the repulsions and van der Waals forces between
electron charge clouds. The shell model provides a simple
description of polarizability effects and has proved to be effective
in simulating dielectric and lattice dynamical properties. An
important feature of these calculations is the treatment of lattice

1
relaxation (using the Mott–Littleton approach) around the point
defect, dopant clust or migrating ion, so the crystal is not
considered simply as a rigid lattice. These methods are embodied
in the GULP simulation code [1]
The first classical atomistic simulations carried out, were those of
Boswara and Lidiard who attempted to determine Schottky defect
formation energies in NaCl structured alkali halides and cesium
halides [2] [3]. In fact, most of the early calculations considered
highly ionic and rather simple compound.

Transition metal oxides were investigated in the 1970’s, using a


similar methodology [4].

2.2 The Perfect Lattice

Ionic crystal theory can be traced back to the work of Madelung


[5] and Born [6]. According to Born model of Static lattice
calculations of ionic solids, the sum of all pair wise interactions
between ions i and j gives the lattice energy of the crystal. The
lattice energy is given by

qi q j
U ( r ij )=∑ + ∑ ∅ ( r ) + ∑ ∅ ( r ) (1)
ij 4 π ε o r ij ij ij ij ijk ijk ijk

Where qi and qj are the charges of ions i and j, r ij is the separation


distance between the two ions and ε0 is the permittivity of free

2
space. The first term in equation (1) is the coulombic energy, the
main interaction between ions and attracts the unlike charged
atoms and repels like charged atoms, and define the long- range
electrostatic interactions. The second and third terms defines short-
range two-body and many-body interactions. It is considered
adequate to calculate only the two body interactions for systems
where the interactions are non-directional, such as ionic solids.
However, when studying systems containing a degree of covalent
bonding, the evaluation of higher body terms, bond bending and
bond stretching terms is necessary to include their directionality.

2.3 Ewald Summation

Despite of the apparent simplicity of the coulomb term (first


term in the right hand side of the equation (1)), the coulombic
interaction has a problem of slow convergence as a function of r.
hence it is necessary to use mathematical methods to deal with this
summation.

The approach developed by Ewald [7] for summing long


range potentials in periodic systems assumes the lattice is
constructed of spherical ions with charge of the same magnitude
(whether positive or negative) and that they do not overlap and
utilizes the convergence properties of periodic arrays of Gaussian

3
functions. The formal mathematical derivation is rather complex
and a detailed explanation is beyond the scope of this thesis.
Therefore the following paragraph describes the important aspects
of the summation method.

The charge density ρ for a point ion i is given by

ρi=δ ( r i−r lattice ) (2)

i.e. using the point ion as the origin, in any direction through the
lattice the charge density at that point r lattice , can have a value of
either 0 or 1. Each ion is then replaced by a Gaussian charge
distribution of equal magnitude but opposite sign, leading to the
expression:

[ ]
2
( r i−r lattice )
ρi=exp (3)
γ2

Where γ is the half width of the Gaussian. This effectively


neutralizes each ion and so removes interactions between
neighboring charges. These charges are short ranged and can be
calculated in real space. A cancelling charge of the same sign as
the original charge is then added which reduces the overall
potential to that of the original set of point charges. The charge
density is then given by

4
[ ( )] ( )
2 2
( r i −r lattice ) ( r i−r lattice )
ρi= δ ( r i−r lattice ) −exp 2
+ exp 2
( 4)
γ γ

2.4 Short Range Potential Functions

The Ewald sum accounts for the long range, attractive coulomb
interaction, but is unable to describe what occurs when two
charged atoms are brought near each other.

The two-body short range potential represents short range


interaction energy of two adjacent atoms that has both repulsive
and attractive components. If they are brought near enough to one
another, this causes two repulsive interactions, which if the
distance between these atoms becomes sufficiently small causes
the overall force between them to become repulsive, even if the
ions are oppositely charged. The first repulsive interaction is
results of the Pauli Exclusion Principle [8] [9] which state that no
two fermions can occupy the same quantum state. When electron
clouds overlap because of the sufficient small distance between the
two atoms, for Pauli Exclusion Principle to be satisfied, the ground
state charge distribution of an electron is forced to occupy higher
energy state, this creating an increase in energy gives rise to the

5
repulsion. The second repulsive interaction is caused by the
nuclear – nuclear repulsion.

At small inter nuclear distance there also exist an attractive force,


the Van der Waals – London interaction.

The interaction energies, which comprise the short-range


interaction term, are given by simple parameterized analytical
functions. The specificity is introduced by having different
parameters between different pairs of ions. Thus the model
depends on using analytical functions that are of appropriate form
for the interaction energies and the parameters must be carefully
chosen. The kinds of functions that are successful are given below:

2.4.1 Born Model of Solids


The first attempt to describe a true short-range repulsive interionic
potential was by Born and Landé [10] who described the functional
form
b
∅ ij = (5)
rn

Where b and n are constants chosen to reproduce the equilibrium


interionic distance and r is the nearest distance between unlike
ions. Early work using this model took ≈9 . This model though
very simple, it works well for highly ionic materials such as the
alkali halides.

6
To account for the new information from quantum mechanics,
Born and Mayer [11] introduce a short-range repulsive function of
the form:
−r
∅ ij = A e ρ (6)

Where A and ρ are adjustable parameters. Equation (5) now


differed from (6) in that equation (6) contains an exponential
repulsive term. An attractive term of the form c /r 6 was also added
to account for the Van der Waals interaction which has been
determined via the work from Van der Waals, London and
Margenau. [12, 13].
2.4.2 Lennard-Jones Potential

A combination of the repulsive term in equation (5) and the c /r


6

attractive Van der waals term results in the so called Lennard-


Jones potential [14] [15] [16]:
b c
∅ ij = n
− 6 (7)
r r
This is a form of non-bonded potential having a repulsive and an
attractive part respectively.
The Lennard-Jones potential was originally developed to describe
noble gasses and latter applied to intermolecular interactions in
molecular systems. It is an approximation that describes the
complicated nuclear and electronic repulsions, which dominate the
attractive interactions at short distances. Lenard-Jones solved for b

7
and c for several different values forn, ranging from 9 to 14. In the
modern calculations, the value for n is 12.
2.4.3 Buckingham Potential
If the short range repulsive term from equation (6) is combined
with the Van der Waals attractive term, the Buckingham potential
model is formed
−r
ρ C ij
∅ ij = A e − 6
(8)
r

Where A is a parameter can be approximated as a measure of the


number of electrons within the ions; ρ can be approximated as a
measure of electron density; and C ij is an approximate of
polarizability of the ion. Figure1 show the influence of the short-
range potential over all ionic interaction.

8
Figure 1: The long and short range interactions of the total forces (adapted
from Harding [17]).

The parameter c is used to model Van der waals interaction, and


can be calculated using the Slater-Kirkwood formula [18]
3
α α
2 i j
C ij = ( 9)

( )( )
1 1
αi 2 α j 2
Ki K j

Where αi and αj represent the static polarizabilities of i and j ions,


and K i, Kj are the number of electrons that contribute significantly
to the polarizability. Of ions i and j , i.e. the effective electron
numbers of ions i and j . Consequently, C ij a function of the ionic
radius, and as such depends on the coordination of the ions. Values

9
For α and K can be found elsewhere [19]. The Buckingham
potential model has proved to be successful in simulating many
oxide systems [20] and other ionic solids [21]. On this justification,
this was the potential model implemented to describe the short-
range interactions in this study.
2.5 Fitting of Potential parameters
There are two type of approach for determining valid interatomic
potentials: empirical or direct calculation. In the latter approach, a
high quality quantum mechanical technique is used to predict the
interaction energy between ions as a function of ion separation. In
this work, the empirical approach of fitting was employed.
The empirical fitting of potentials is simply to varying the short-
range parameters until the structure and lattice properties agree
with experimental observation (as done by Born and Landé).
Empirical potential fitting consists of a least squares procedure
whereby the difference between observed (experiment) and
calculated properties are minimized.
F=∑ [ f observable −f calculated ] (10)
2

Almost all properties of the material can be used in the fitting


process, including elastic and dielectric constants, and lattice
energy, phonon data. The prerequisite is the knowledge of the

10
crystal structure, including the ionic positions and lattice
parameters. Advantage of this method includes its relative
simplicity as well as the ability to describe the full behavior of
collection of atoms, including any potentials covalency. Care must
be taken when phonon modes are included in the derivation
process since their order can change during the fitting, thus they
are commonly used only to used to ensure that the minimum
energy structure is stable (i.e. all the phonons are positive). It is
often the case with empirical fitting that only the crystal structure
is known with any degree of certainty.

Often, more than one set of parameter values can reproduce the
physical properties of the material [22]. However, the test of a
successful potential model is the transferability of the parameters
to systems not included in the initial parameter selection.
The transferability and reliability of the potentials can be improved
by including as much information about the structure as possible
(e.g. high frequency and static dielectric constants, bulk modulus
and elastic constants) in the fitting procedure. Unfortunately, a
potential may reproduce such perfect lattice properties, and yet
when defect calculations are performed, its shortcomings become
evident. Experience suggests that such problems are generally
avoided if the potential is fitted over a broad range of interionic

11
separations to allow for the consideration of interstitial and
vacancy defects which alter the equilibrium separation.
A more detailed explanation of fitting procedure used and
discussion of the merits of the potential fitting methods is given by
Gale [23].

2.6 The Shell Model and Electronic Polarizability

The shell model by Dick and Overhauser [24] successfully


describes the polarization and the physical distortion of the ion in
response to an electric field. While Faux and Liodiard [25] proved
its value to defect calculation. The ion is assumed to be composed
of a core, representing the nucleus and core electrons, and a
massless shell, representing the outer electrons. The core and shell
are linked by an isotropic, harmonic spring with a force constant, k
(Figure 2). When an electric field is applied to the ion, the shell is
allowed to move relative to the core in such a way that it develops
a dipole and thus simulates the dielectric polarizability of the
lattice. The total interaction of the core and shell of an ion is given
by the expression:
1 2
∅ i ( r i )= k i r i (11)
2

Where r i is the distance between the core and the shell.

12
Spring (k)

Massless shell
Charge q
Core charge Q

Figure 2 schematic of shell model

The core and the shell are assigned positive Q and negative q
charges respectively such that the sum is equal to the charge on the
ion Q+q. The polarizability of a free ion is given by:
Y2
α i= (12)
4 π ε ok i

Where Y is the massless shell charge in electronic charge units, ε o


is the permittivity of free space α i and k are in units ofÅ3, and
eVÅ−2 respectively.
Short-range forces are considered to act between shells whereas
the coulombic forces act between the shells and cores (except on

13
the same ion) (Figure 3). The Shell Model has proved very
successful in the reproduction of properties such as defect energies,
phonon dispersion curves and dielectric constants [26].

Figure 3: The shell model, where the orange atom is the core and the blue atom
represents the charge of the massless shell. The red arrows represent polarization,
black arrows represent Coulombic interactions and blue arrow represents short
range interaction.

The advantage of these model (versus other models, such as the


rigid ion model or the point polarizable ion model) is that any force
acting on an ion is assumed to do so via the shell, thus coupling
short range interactions to the polarizability, see (Figure 3).
Clearly, this provides a framework by which it is possible to model
more of the interactions occurring between species than if the shell
model was not used (i.e. the rigid ion model alone).

14
A disadvantage of the shell model is that the parameters must be
obtained by empirical fitting and experimental data is not always
available. In addition, the number of species involved in computer
simulations is doubled and therefore the computing time is
doubled. Thus to model a system, knowledge of the nature of
bonding is important to make a proper choice of potential
functions. Once these functions are identified, their parameters
could be derived.

2.7 Defective Lattice approach

Certainly, the importance of defects is to control and modifying the


properties of materials which is the main theme of this thesis. The
method accounting for the defective lattice is based on a
minimization of the total energy of a system by relaxing ions
around a defect. In order to efficiently simulate the lattice
relaxation around defects, a multi region approach is adopted
which stems from the work of Mott and Littleton [27] [28]. The
lattice is partitioned into concentric spherical regions centered on
the defect (Figure 4). The area surrounding the defect is termed
Region 1. Region 1 is a sphere of ions including the defect. The
total energy of these ions is calculated explicitly as defined by the

15
Coulombic interaction and short range potential discussed
previously.

Figure 4: Representation of the Multi Region Approach

The size of Region 1 is a very important consideration in these


types of calculations. Therefore, Region 1 must be chosen large
enough such that the defect energy converges appropriately, but
computational efficiency must also be taken in to account. (Figure
5) demonstrates the effect of Region I size on the defect energy.

16
Figure 5: Defect energy variations with changing region size for an antisite
defect, A XB . [29]

Region 2 is subdivided into Region 2a and Region 2b. Region 2a


serves as an interface between Region 1 and Region 2b (which
extends to infinity). The changes in energy of the ions in Region 2a
due to the defect and the relaxation of ions in Region 1 are
calculated explicitly. However, the displacements of Region 2a
ions are determined in a single step by calculating the forces on

17
these ions using the Mott- Littleton approximation [27]. This
approximation is applicable since the defect is assumed to have
only a small effect on the ions in Region 2b. Therefore, the entire
response of Region 2b is approximated by the Mott-Littleton
method.
The complication the Mott-Littleton method addresses is that the
forces on any ion are not only due to the charged defect, but also to
the dipoles which have been induced in the region of the lattice
around the defect. This polarization can be approximately given
by:

P=
q
4πr 2
1
( )
1− (12)
ϵ

Where P is the polarization, q is the charge of the defect, r is the


distance from that defect and ε =ϵ s ϵ o.
According to the conventional treatment of the defective lattice (as
developed by Lidiard and Norgett [30] and Norgett [31]), the total
energy of the solid containing a defect can be expressed as:

E=E1 ( r )+ E2 ( r , ζ )+ E 3 ( ζ ) (13)

where E1 ( r )is the energy of Region 1, E2 ( r , ζ )is the energy of the


interfacial Region 2a and E3 ( ζ ) is the energy of Region 2b. The two

18
independent vectors ( r ) and ( ζ )are the coordinates of ions in Region 1
and Region 2, respectively. As E3 ( ζ ) involves an infinite number of
displacements, it cannot be solved explicitly. However, if the
displacements are assumed to be quasi-harmonic, it can then be
defined as:
1
E3 ( ζ ) = .ζ . A . ζ (14)
2

Where A is the force constant matrix. This expression can


substitute into Equation 13, and differentiating with respect to ζ ,
the equilibrium displacements in region 2 are given by:
∂ E ∂ E2 ( r , ζ )
∂ζ
=
∂ζ |ζ =ζ e
+ A . ζ e ( 15)

Where ζ eis the equilibrium value of ζ corresponding to r. When this


is substituted into Equation 14 and then into Equation 13, the
explicit dependence of E on energy of region 2, E2, is removed
allowing a far more convenient solution:

Ed =E1 ( r ) + E2 ( r , ζ )−
1 ∂ E2(r , ζ )
2 ∂ζ |
ζ =ζ e
.ζe (16)

Original versions of this method only partitioned the lattice into


two regions. However, in order to ensure a smooth transition
between Region 1 and Region 2, Region 2 is split into areas called
Region 2a and Region 2b. In Region 2a, which acts as a transition
region between Region 1 and Region 2b, the Mott-Littleton
approximation [27] is used to calculate the polarization, P,

19
(equation (12)). And the total defect energy can then be calculated
by minimizing with respect to r and ζ .
2.8 Energy Minimization
Once a model has been adequately validated, in order to be made
useful in a predictive manner, it must be combined with energy
minimization. The minimization of the potential energy function
(i.e., geometry optimization) involves a search for the minimum of
a function, and, to be efficient, requires calculations of derivatives
of a function (in this case, the potential energy) versus independent
variables (in our case, coordinates). Most programs use cartesian
coordinates as independent variables, however, in some cases,
internal coordinates may be used. This reduces the system to a
state of mechanical equilibrium. The criterion used for determining
the accuracy of the model is that the ion displacements in the
optimized structure from the experimental configuration are
minimal. During energy minimization, all ionic interactions are
calculated and each ion moves a distance proportional to the force
acting on it through an iterative process.
Two conditions exist to minimize the lattice energy at equilibrium:
constant volume and constant pressure. Under constant volume
minimization, the lattice energy is minimized only by varying the
internal coordinates of the ions within the unit cell relative to the

20
strains on individual ions, while the lattice parameters are not
allowed to change. Under constant pressure minimization, the unit
cell dimensions are also adjusted, accounting for the strains on
both the individual ions and the unit cell. Since there are fewer
degrees of freedom for the constant volume calculations, they are
computationally faster. Consequently, most early calculations were
constant volume. Due to the increase in computation power
available, nearly all modern calculations are of the constant
pressure type, including all those included in this thesis.
If the lattice energy (UL) of a system with N ions with coordinates,
r, is UL(r), then after one minimization step, the lattice energy at a
new set of coordinates, r ' is:
1 T
U L ( r )=U L ( r ) + ⃗
g . δ⃗ + ⃗δ .W . ⃗δ (17)
' T
2

Where W is a matrix that contains the corresponding second


derivatives with respect to r:
∂2 U L
W= (18)
∂ r2
δ⃗ is the displacement (or strain) of a given ion:
⃗ ' −r (19)
δ=r

The vector, refers to the first derivative of the lattice energy with
respect to the ionic positions and is given by
∂UL
⃗g= (20)
∂r

21
At equilibrium, the change in energy with respect to strain is zero.
Therefore:
∂U L ( r ' )
=0=⃗g +W . δ⃗ (21)
∂ ⃗δ
Now Since the first derivative of the lattice energy with respect to
distance (coordinates) is the force, it is possible to write this
convergence aim as:
∂U L
=F=0(22)
∂r

in practice, the minimization proceeds for a pre-set number of


iterations or until a point where at subsequent steps, the total
energy of the system changes by less than a predetermined value.
In general, the minimization methods as mentioned above are
iterative. They require on input some initial estimate for the
position of the minimum, and provide a better estimate for the
minimum as a result. This corrected estimate is used as an input
into the next cycle (i.e., iteration) and the process is continued until
there is no significant improvement in the position of the
minimum. This situation is schematically illustrated in (Figure 6).
2.9 Computational Codes
2.9.1 CASCADE
CASCADE (Cray Automatic System for the Calculation of Defect
Energies) was written in Fortran at the Daresbury Laboratory [32],

22
specifically for the CRAY-1 computer. In this study, CASCADE is
exclusively used for perfect lattice calculations
and for defect energy calculations.

2.4.2 MARVIN
MARVIN’S (Minimization And Relaxation of Vacancies and
Interstitals for Neutral Surfaces) Program was developed at the
Royal Institution of Great Britain for studying surfaces and
interfaces [33]. MARVIN is based upon a similar code, MIDAS,
developed by Tasker in the late 1970’s [34]. MARVIN improves
upon the MIDAS code by utilizing the increase in computer
capability to calculate the surface energies of not only simple cubic
crystals, but also more complex carbonates, sulfates, phosphates,

23
Figure 6: schematic illustration of the computer process of
Energy minimization.

etc. MARVIN also allows for the introduction of ions and


molecules to the surface, which is important in modelling crystal

24
growth and catalysis. MARVIN considers a simulation cell of a
finite number of atoms, which are repeated in 2D (as previously
described). The cell consists of a Region 1 and 2. In this regard,
MARVIN is similar to CASCADE, in that it relaxes Region 1
atoms explicitly whereas those in Region 2 remain fixed.

2.9.3 GULP
GULP (General Utility Lattice Program) is an improvement from
earlier codes in that it has an automated empirical fitting of
potential parameters feature. This feature subsequently allows for a
simultaneous multi-structural fit routine.
In this thesis for potential model and atomistic calculations Gulp
code [35] [36] has been employed.

References

25
[1] J.D. Gale J. GULP - a computer program for the symmetry adapted
simulation of solids, Chem. Soc., Faraday Trans., 93(4), 629-637 (1997).

[2] I.M. Boswara and A.B. Lidiard, Philos. Mag., 16 805 (1967).

[3] I.M. Boswara, Philos. Mag. 16 827 (1967).

[4] C.R.A. Catlow and D.G. Muxworthy, Philos. Mag. B 37 63 (1978).

[5] E. Madelung, Phys. Zeit. 11 898 (1910).

[6] M. Born, Handbuch der Physik , Springer, Berlin (1933).

[7] P.P. Ewald, Ann.phys. (Leipzig) 64 253 (1921).

[8] W. Pauli, General Principles of Quantum Mechanics, Springer-


Verlag, Berlin (1980).

[9] W.Pauli, Z.Phys. 31 765 (1925).

[10] M. Born and A. Landé. Verhandlungen der Deutschen Physik


Gesellschaft, 21/24 210 (1918).

[11] M. Born and J.E. Mayer. Dispersion and Polarizability and Van der
Waals Potential in alkali Halides. Journal of Chemical Physics, 1 270,
(1933).

[12] F. London. Properties and Applications of Molecular Forces.


Z.Phys. Chem. (B) 11 222 (1930).

26
[13] H. Margenau. Surface energy of Liquids. Phys. Rev. 38 365-371,
(1931).

[14] J. E. Lennard-Jones. On the Determination of Molecular Fields. I.


from The Variation of the Viscosity of Gas with Temperature.
Proceeding of the Royal Society of London Series A, 106(738): 441-
463, (1924).

[15] J. E. Lennard-Jones. On the determination of molecular fields. II.


from the equation of state of a gas. Proceedings of the Royal Society of
London Series A, 109(738):463–477, (1925).

[16] J. E. Lennard-Jones and A. E. Ingham. On the calculation of certain


crystal potential constants, and on the cubic crystal of least potential
energy. Proceedings of the Royal Society of London Series A,
107(744):636–653, (1925).

[17] J. H. Harding. Computer simulation of defects in ionic solids.


Reports on Progress in Physics, 531403–1466, (1990).

[18] P. W. Fowler, P. J. Knowles, and N. C. Pyper. Calculations of 2-


body and 3-body dispersion coefficients for ions in crystals. Molecular
Physics, 56(1) 83–95, (1985).

27
[19] N. W. Grimes and R. W. Grimes. Dielectric polarizability of ions
and the corresponding effective number of electrons. Journal of Physics-
Condensed Matter, 10(13):3029–3034, (1998).

[20] R. W. Grimes and S. P. Chen. The influence of ion size on the


binding of a charge compensating cobalt vacancy to M3+ dopant ions in
CoO. Journal of Physics and Chemistry of Solids, 61(8):1263–1268,
(2000).

[21] C. R. A. Catlow, editor. Computer Modelling in Inorganic


Crystallography. Academic Press, Inc., 1st edition, (1997).

[22] J. H. Harding. The practical calculation of interatomic potentials.


Molecular Simulation, 4(5):255–168, (1990).

[23] J. D. Gale. Empirical potential derivation for ionic materials.


Philosophical Magazine B-Physics of Condensed Matter Statistical
Mechanics Electronic Optical and Magnetic Properties, 73(1):3–19,
(1996).

[24] B. G. Dick and A. W. Overhauser. Theory of the dielectric


constants of alkali halide crystals. Physical Review, 112(1):90–103,
(1958).

28
[25] I. D. Faux and A. B. Lidiard. The volumes of formation of Schottky
defects in ionic crystals. Zeitschrift f¨ur Naturforschung A, 26:62–68,
(1971).

[26] W.Cochran , The dynamics of atoms in crystals, ed. Coles B.R.,


Edward Arnold, (1973).

[27] N. F. Mott and M. J. Littleton. Conduction in polar crystals. I.


electrolytic conduction in solid salts. Transactions of the Faraday
Society, 34:485–499,(1938).

[28] J. H. Harding. Computer simulation of defects in ionic solids.


Reports on Progress in Physics, 53:1403–1466, (1990).

[29] Mark R. Levy, Crystal Structure and Defect Property Predictions in


Ceramic Materials, Ph.D. Thesis, Imperial College (2005).

[30] A.B. Lidiard and M.J. Norgett, in Computational Solid State


Physics. Plenum -Press (1972).

[31] M. J. Norgett. A users guide to HADES. Technical Report R.7015


HL72/550, Atomic Energy Research Estabishment, Harwell, 1972.

[32] M. Leslie, “DL/SCI/TM31T,” Tech. rep., SERC Daresbury


Laboratory (1982).

[33] D.H. Gay and A.L. Rohl, J. Chem. Faraday Trans. 91 925 (1995).

29
[34] P.W. Tasker, “R.9130,” Tech. rep., AERE Harwell (1978).

[35] https://projects.ivec.org/gulp/

[36] J.D. Gale and A.L. Rohl, The General Utility Lattice Program, Mol.
Simul., 29, 291 (2003).

30

You might also like