See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228646810

Atomistic simulation of mechanical properties of diamond and silicon carbide

by a field theory

Article  in  Modelling and Simulation in Materials Science and Engineering · July 2007

DOI: 10.1088/0965-0393/15/5/011

CITATIONS READS

24 1,615

3 authors:

Liming Xiong Youping Chen

University of Florida University of Florida
46 PUBLICATIONS   943 CITATIONS    106 PUBLICATIONS   2,481 CITATIONS   

SEE PROFILE SEE PROFILE

James D. Lee
George Washington University
184 PUBLICATIONS   3,168 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Linking atomistic and continuum description of transport processes in crystalline materials View project

multiscale material modeling View project

All content following this page was uploaded by James D. Lee on 21 May 2014.

The user has requested enhancement of the downloaded file.

IOP PUBLISHING MODELLING AND SIMULATION IN MATERIALS SCIENCE AND ENGINEERING
Modelling Simul. Mater. Sci. Eng. 15 (2007) 535–551 doi:10.1088/0965-0393/15/5/011

Atomistic simulation of mechanical properties of

diamond and silicon carbide by a field theory
Liming Xiong1 , Youping Chen1 and James D Lee2
1 Department of Mechanical and Aerospace Engineering, University of Florida,

Gainesville, Florida 32611, USA

2 Department of Mechanical and Aerospace Engineering, The George Washington University,

Washington DC, USA

E-mail: [email protected]

Received 16 December 2006, in final form 24 May 2007

Published 11 July 2007
Online at stacks.iop.org/MSMSE/15/535

Abstract
This paper presents a multiscale field theory and its application in modelling
and simulation of atomistic systems involving three-body interaction forces.
Atomistic formulation of the multiscale field theory is introduced. Numerical
simulations based on the field theory are performed to investigate the material
behaviours of diamond and silicon carbide at the atomic scale. We have obtained
the tensile strength and the elastic modulus that approach that obtained by first
principles calculations for both diamond and silicon carbide. Their nanoscale
deformation and failure mechanism are revealed. It is interesting to observe
that under tensile loading, unlike silicon carbide, diamond has gone through
a phase transformation as well as local amorphization before failure. The
potential application of this atomic field theory is discussed.
(Some figures in this article are in colour only in the electronic version)

1. Introduction

The term ‘multiscale material modelling’ refers to theory and simulation of material properties
and behaviour across length and time scales from the atomistic to the macroscopic. With the
increase in the application of new experimental tools and new material synthesis techniques
to nano/micro systems, multiscale material modelling has emerged as a significant concept as
well as a unique approach in computational materials research. Despite widespread interest and
efforts, major challenges exist for the simulation of nano/micro-scale systems over a realistic
range of time, length, temperature as well as multiple physical conditions and environments.
In a series of theoretical papers, a multiscale field theory has been formulated by
Chen and her coworkers (Chen and Lee 2005, Chen 2006, Chen 2006a, 2006b; 2007) for
concurrent atomic-continuum modelling of materials/systems. Continuous local densities
of fundamental physical quantities in atomistic systems are derived. By decomposing

0965-0393/07/050535+17$30.00 © 2007 IOP Publishing Ltd Printed in the UK 535

536 L Xiong et al

Figure 1. Atomistic view of crystal structure: crystal structure = lattice + local atomic bonding
units.

atomic motion/deformation into homogeneous lattice motion/deformation and inhomogeneous

internal atomic motion/deformation, and also decomposing momentum flux and heat flux into
homogeneous and inhomogeneous parts, field description of conservation laws at the atomic
scale has been formulated. As a result of the formulation, a field representation of atomic
many-body dynamics has been obtained.
This paper is concerned with application of the multiscale field theory for materials systems
that involve three-body interaction forces. The primary goal of this work is to reproduce
atomic scale properties and phenomena by the field theory. Atomistic formulation of the
multiscale field theory is introduced in section 2; governing equations for systems under a
given temperature field are derived in section 3; simulation results of diamond and silicon
carbide under tension are presented and discussed in section 4. We conclude this paper with a
brief summary and discussions in section 5.

2. Atomistic formulation of a field theory for multi-element systems involving

three-body interaction forces

Crystalline solids are distinguished from other states of matter by a periodic arrangement of the
atoms; such a structure is called a crystal lattice. Essentially the regularity displayed by a crystal
lattice is that of a three-dimensional mesh which divides space into identical parallelepipeds.
Imagine a number of identical atoms placed at the intersections of such a mesh; then we
have what is known as a simple lattice (or Bravais lattice). The interstitial parallelepipeds
are referred to as the elementary lattice cells. If the atoms are replaced by similarly oriented
molecules, the result is a general lattice structure. Clearly every cell contains as many atoms
as there are in one molecule—this is a typical picture of the crystal structure of a multi-element
system, cf figure 1.
Microscopic dynamic quantities are functions of phase-space coordinates (r, p), i.e. the
positions and momenta of atoms. For multi-element systems, there is more than one atom in
the primitive unit cell. Thus, one has

r = {Rkα = Rk + rkα | k = 1, 2, 3, . . . n, α = 1, 2, 3, . . . ν}
(2.1)
p = {mα Vkα = mα Vk + mα v kα | k = 1, 2, 3 . . . n, α = 1, 2, 3, . . . ν},

where the superscript kα(k = 1, 2, 3 . . . n, α = 1, 2, 3, . . . ν) refers to the αth atom in the kth
unit cell; mα is the mass of the αth atom; Rkα and Vkα are the position and velocity vector of
the kα atom, respectively; Rk and Vk are the position and velocity of the centre of the unit cell,
respectively; rkα and v kα are the atomic position and velocity of the kαth atom relative
to the lattice, respectively. The corresponding local density of any measurable phase-space
function a(r, p) can generally be defined as


n

ν
A(x, yα , t) = a{r(t, p)(t)}δ(Rk − x)δ̃(rkξ − yα ) ≡ Aα (x, t), (2.2)
k=1 ξ =1
Atomistic simulation of mechanical properties 537

with normalization
 conditions
δ(Rk − x)d3 x = 1 (k = 1, 2, 3, . . . n). (2.3)
V
Here, V is the volume of the whole system, and the first delta function is a localization
function that provides the link between phase space and physical space descriptions. It can
be a Dirac δ-function (Irvine and Kirkwood 1950), or a distribution function (Hardy 1982).
The field descriptions of the conservation equations and constitutive relations were found to
be independent of the choices of the localization function (Hardy 1982, Chen and Lee 2005,
2006, Chen et al 2006a). Equation (2.3) implies that over the entire physical space all the unit
cells, (k = 1, 2, 3, . . . , n), can be found. Then, for each unit cell k, the second δ̃-function in
equation (2.2), identifies yα to be rkα , i.e.

1 if rkξ = yα or ξ = α
δ̃(r − y ) =
kξ α
(2.4)
0 if rkξ = yα or ξ = α
It follows that
δ(Rk − x)δ̃(rkα − yα ) d3 x = 1 (k = 1, 2, 3, . . . n) (α = 1, 2, . . . ν). (2.5)
V
Most current MD applications involve systems that are either in equilibrium or in some time-
independent stationary state; where individual results are subjected to fluctuation, it is the
well-defined averages over sufficiently long time intervals that are of interest. Extending MD
to open systems, where coupling to the external world is of a more general kind introduces
many new problems. Not only are open systems out of thermodynamic equilibrium, but also in
many cases they are spatially inhomogeneous and time-dependent. To smooth out the results
and to obtain results close to experiments, measurements of physical quantities are necessary to
be collected and averaged over finite time duration. Therefore, in deriving the field description
of atomic quantities and balance equations, it is the time-interval averaged quantities that are
used, and the time-interval averaged (at time t in the interval t) local density function takes
the form
 t
1
Āα (x, t) = Aα  ≡ Aα (x, t + τ ) dτ
t 0
 t
n
1
= a(r(t + τ ), p(t + τ ))δ(Rk − x)δ̃(rkα − yα ) dτ. (2.6)
t 0 k=1
The continuous local mass density ρ̄ α , linear momentum density ρ̄ α (v̄ + v̄ α ), total energy
α
density ρ̄ α ēα , internal energy density ρ̄ α ε̄ α , temperature T α , internal force density f̄int and
α
external force density f̄ are defined as
 n 


ρ̄ (x, t) =
α
m δ(R − x)δ̃(r − y ) ,
α k kα α
(2.7)
k=1
 n 


ρ̄ (v̄ + v̄ ) =
α α
m (V + v )δ(R − x)δ̃(r
α k kα k kα
−y ) ,
α
(2.8)
k=1
 n   

1
ρ̄ ē =
α α α k
m (V ) + U δ(R − x)δ̃(r − y ) .
α 2 kα k kα α
(2.9)
k=1
2
 n   

1 kα
ρ̄ α ε̄ α = mα (Ṽ )2 + U kα δ(Rk − x)δ̃(rkα − yα ) . (2.10)
k=1
2
538 L Xiong et al
 
V
α kα 2
n
T (x) =
α
m (Ṽ ) δ(Rk − x)δ̃(rkα − yα ) , (2.11)
3kB k=1
 n  n ν kα  





ν α
α
f̄int (x) ≡  lβ
f1 + f2  δ(R − x)δ̃(r − y ) .
kβ k kα α
(2.12)
k=1 l=1 β=1 β=1
 n 
α

f̄ ≡ fkα
3 δ(R
k
− x)δ̃(r kα
−y ) ,
α
(2.13)
k=1

where U kα is the potential energy, 21 mα (Vkα )2 + U kα = E kα is the atomic site energy of the
kα
αth atom in the kth unit cell; Ṽ = Vkα − v̄ − v̄ α is the difference between the phase space
velocity and the local velocity field, kB is Boltzmann constant; and V is the volume that
kα
lβ
defines the density of lattice points, i.e. the volume of a unit cell; f 1 represents the interatomic
lβ kα
kα kα
lβ
force between (k, α) and (l, β) atoms in two different unit cells with = −f1 ; f2β the
f1
kα β
interatomic force between (k, α) and (k, β) atoms in the same unit cell with f2β = −fk2α , and fkα
3
represents the body force on atom (k, α) due to the external fields. In this work, it is understood
kα
lβ
that the summation over k and l, when f1 is involved, does not include the case k = l, and
kα
when f2β is involved, the summation over α and β does not include α = β.
The continuum counterpart of momentum flux density is the stress tensor, which is usually
defined in terms of a mathematical infinitesimal volume in a homogenized continuum. For
multi-element systems, the infinitesimal volume that does not violate the continuum assumption
shall be at least equal to the volume of a primitive unit cell, and the vector sum of all the atomic
forces within this volume does not necessarily pass through the mass centre of the V . The
continuum definition of stress is, therefore, not the momentum flux density. For a crystal with
more than one atom in the unit cell, the continuum stress is only the homogeneous part of the
momentum flux summing over a volume at least of a primitive unit cell, and it may not be
symmetric. Thus, the total momentum flux is better represented upon decomposition into a
homogeneous part, caused by lattice motion and deformation and is related to continuum stress,
and an inhomogeneous part, caused by internal (relative) atomic motion and deformation.
α
The homogeneous and inhomogeneous kinetic parts, t̄kin and τ̄ αkin , and the homogeneous and
α α
inhomogeneous potential parts, t̄pot and τ̄ pot , take the following forms:
 n 


α α k kα
t̄kin = − m Ṽ ⊗ Ṽ δ(R − x)δ̃(r − y ) ,
k kα α
(2.14)
k=1
 n 

kα
τ̄ αkin =− α
m ṽ kα
⊗ Ṽ δ(R − x)δ̃(r
k kα
−y ) ,
α
(2.15)
k=1
 n 
1

k
ν
α 
t̄pot =− (R −R ) ⊗ F B(k, ξ, l, η, x, y ) ,
l kξ α
(2.16)
2 k,l=1 ξ,η=1
 n 
1


ν

τ̄ αpot =− (r −r ) ⊗ F B(k, ξ, l, η, x, y ) ,
kξ lη kξ α
(2.17)
2 k,l=1 ξ,η=1
Atomistic simulation of mechanical properties 539

where
 1

B(k, ξ, l, η, x, yα ) = dλδ(Rk λ + Rl (1 − λ) − x)δ̃(rkξ λ + rlη (1 − λ) − yα ), (2.18)
0

 1

lη lη
F kξ
=− ∂(U kξ + U mς ) ∂Rkξ , (2.19)
2 l,m η,ς
and the many-body potential function, including the Tersoff and the Stillinger–Weber potential
functions (Tersoff 1989, Stillinger and Weber 1985), is expressed in a general form:
1

kξ

n ν kξ n

ν kξ
U= U lη , U lη = U lη
(Rkξ − Rlη , Rlη − Rmς , Rmς − Rkξ ).
2 k,l ξ,η m ς
(2.20)
Similar to the decomposition of momentum flux density, the heat flux density is also
decomposed into homogeneous and inhomogeneous kinetic parts and homogeneous and
inhomogeneous potential parts as
 n 

k 1 kα 2

q̄kin = −
α
Ṽ m (Ṽ ) + U δ(R − x)δ̃(r − y ) ,
α kα k kα α
(2.21)
k=1
2
 n   
α
kα 1 α kα 2
j̄kin = − ṽ m (Ṽ ) + U δ(R − x)δ̃(r − y ) ,
kα k kα α
(2.22)
k=1
2
 n 
1

k
ν
kξ  kξ
q̄pot = −
α
(R − R )Ṽ · F B(k, ξ, l, η, x, y ) ,
l α
(2.23)
2 k,l=1 ξ,η=1
 n 
1


ν
α kξ  kξ
j̄pot =− (r − r )Ṽ · F B(k, ξ, l, η, x, y ) .
kξ lη α
(2.24)
2 k,l=1 ξ,η=1

By decomposing atomic displacements and momentum and heat fluxes into homogeneous
and inhomogeneous parts, as exact consequences of the atomistic definition of physical
quantities and Newton’s second law, the mathematical representation of conservation equations
for averaged mass, linear momentum and energy at the atomic scale has been analytically
and exactly obtained in terms of averaged field quantities as (Chen and Lee 2005, 2006,
Chen et al 2006a)
∂ ρ̄ α
+ ∇x · (ρ̄ α v̄ ) + ∇yα · (ρ̄ α v̄ α ) = 0, (2.25)
∂t
∂ α α α
(ρ̄ (v̄ + v̄ α )) = ∇x · [t̄ − ρ̄ α v̄ ⊗ (v̄ + v̄ α )] + ∇yα · [τ̄ α − ρ̄ α v̄ α ⊗ (v̄ + v̄ α )] + f̄ ,
∂t
(2.26)
∂ α α α
(ρ̄ ε̄ ) + ∇x · (−q̄α + v̄ ρ̄ α ε̄ α ) + ∇yα · (−j̄ + v̄α ρ̄ α ε̄ α )
∂t
α
= t̄ : ∇x (v̄ + v̄ α ) + τ̄ α : ∇yα (v̄ + v̄ α ). (2.27)
It is worthwhile to note that, with the atomistic definition of internal force and momentum flux
densities, one has
α α
∇x · t̄pot + ∇yα · τ̄ αpot = f̄int , (2.28)
540 L Xiong et al

and


ν
α
τ̄pot = 0. (2.29)
α=1

Equation (2.28) distinguishes multi-element systems from a single element system, in the latter
of which the inhomogeneous part of the stress does not exist, and hence the stress is ‘any tensor
field which satisfies the condition that its divergence is the vector force field’ (Sommerfeld 1950,
Nye 1957), whereas for a multi-element system, as pointed out by Nielsen and Martin (1985),
this classical definition cannot give a unique definition of stress, and additional consideration
is required to include the inhomogeneous part of momentum flux in order to uniquely describe
the stress field at the microscopic scale for inhomogeneous systems.
Note that all the quantities in the balance equations (2.25)–(2.27) are time-interval averaged
quantities. This is one major difference between the field representation of the many-body
problem and MD simulation: MD solves for instantaneous quantities first and then obtains
averaged quantities through statistical averaging, while the field theory solves directly for
averaged quantities (local equilibrium properties of which the fluctuations are averaged out).
The procedure of taking time averages in MD or ensemble averages in statistical mechanics
simulation is thus eliminated.
Another fundamental difference between MD and the field theory is that the lattice
deformation is assumed to be continuous with respect to x in the physical space. This
assumption requires that the smallest x, e.g. the mesh size, should be no less than the lattice
spacing. In the case that x equals lattice spacing, the field theory has the same degrees of
freedom as MD. However, since it is a field theory and in terms of field variables such as
displacement field and local densities, different meshes can be used in regions of different
concerns. This then enables the application of the field theory to systems with large length
scales. Also, since it is an atomic field theory and analytically links atomic and continuum
descriptions of properties, it is naturally a concurrent atomic continuum model. Both MD
simulation and continuum modelling techniques can be utilized. As a result, the field-theory-
based simulation shall be computationally much more efficient for statistical, finite size and
finite temperature problems, for simultaneously large length and time scales phenomena, and
especially, for dynamic, time-dependent and non-equilibrium phenomena and systems.
The dynamics of atoms in an atomistic system can be fully characterized in terms of
phonon dispersion relations. Phonon dispersion relations of BiScO3 calculated based on the
field theory, and the comparison with that obtained from an atomic-level MD simulation code,
GULP (Gale 1997), can be found in Chen et al (2006), in which one can see that phonon
dispersion relations for all acoustic and optical phonons from a fully atomistic model can be
perfectly reproduced by the field theory. Elastic properties of ZnO, BaTiO3 and MgO calculated
by the field theory and their comparison with that obtained from the general-purpose parallel
MD simulation code DL-POLY (Smith and Forester 2001) can be found in Chen et al (2006b),
where good agreement between these two sets of results was achieved.
A comparison of classical continuum mechanics, a generalized continuum theory, the
newly formulated field theory and MD are summarized in table 1. As is shown in table 1 and
figure 1, the atomic view of a crystal is as a periodic arrangement of local atomic bonding
units. For multi-element systems, the space lattice is macroscopically homogeneous, whereas
embedded in each lattice point is a group of discrete atoms, the smallest structural unit of
the crystal. Classic continuum mechanics views a crystal as a homogeneous and continuous
medium, in which the basic structural unit of the crystal is taken without structure and is
idealized as point mass, and the internal deformation is ignored. Its application is thus
limited to homogeneous or macroscopic problems. In the well-established Micromorphic
Atomistic simulation of mechanical properties 541

Table 1. Comparison of classical continuum theory, Micromorphic theory, the new theory and
MD.

Atomic structural Lattice Internal

unit deformation deformation Application

Mass point Continuous

Classical continuum None Macroscopic
theory homogeneous
problems
A finite size Continuous Continuous
Micromorphic theory continuum Micro to macro
scales

Discrete Continuous Discrete

The new field theory Atomic to meso
scales

Discrete Discrete Discrete

MD Atomic to micron
scales

theory (Eringen and Suhubi 1964), the material is envisioned as a continuous collection of
deformable particles; each particle has finite size and 9 internal degrees of freedom describing
the stretches and rotations of the particle. Compared with classical continuum mechanics,
Micromorphic theory extends the application region of continuum theory to microscopic time
and length scales. However, the assumption of a continuous structure and deformation of the
particles makes it difficult to describe complex crystalline materials. The newly formulated
theory views a crystalline material as a continuous collection of lattice points, while embedded
with each lattice point is a group of discrete atoms. Hence this further expands the application
region of Micromorphic theory to the atomic scale. While it retains most of the features of
an atomic many-body dynamics, the new field theory is able to scale up in modelling and
simulations of various physical systems and phenomena.

3. Governing equations for problems with given temperature

Unlike many-body dynamics, a field theory is concerned with continuous local densities,
such as mass density, momentum density, energy density, stress tensor and heat flux. It is
generally built upon two foundations: (1) the balance laws and (2) the constitutive relations.
With the balance laws and the constitutive relations, the dynamic behaviour of a material
system under a given external field and properly imposed boundary and initial conditions can
be uniquely and completely determined in terms of a continuum description. Decomposing
the atomic displacements into lattice and internal displacements, it is found that the stresses,
internal energy, heat flux as well as the balance equations can all be expressed in terms of field
variables (Chen and Lee 2006, Chen et al 2006b, 2007), resulting in a well-posed boundary-
value problem, in which the state or the independent variables are lattice displacement ū(x),
relative atomic displacements ξ̄ (x, α) and temperature T α , α = 1, 2, . . . ν.
Temperature in most MD simulations is simply defined in terms of thermal energy by
the mean-squared velocity relative to the local stream velocity (Hoover 1986). An important
conceptual issue of temperature is the size of the region over which temperature is defined. For
542 L Xiong et al

quantum mechanical definition, the length scale is defined by the mean-free-path of phonon. A
local region with a designated temperature must be larger than the phonon scattering distance.
This phonon viewpoint of temperature implies that the temperature cannot be defined for a
particular atom or a plane of atoms. The classical definition of temperature is entirely local.
One can define a temperature for each atom or a plane of atoms. Apparently, the quantum
definition and the classical definition of temperature shall come to be identical at the macro
scale, but they will be different at the atomic/nano scale. In this work, we follow the classical
definition and define a measure of thermal motion of atoms by averaging the thermal energy
over a unit cell and over a finite duration, and call it local temperature, i.e.
 
2V

1 α kα 2
n ν
T (x) = m (Ṽ ) δ(R − x) .
k
(3.1)
3kB k=1 α=1 2

It is seen that the temperature can be further decomposed into a homogeneous part that is due
to the motion of lattice and an inhomogeneous part that is due to the atomic motion relative to
the lattice as
 
2V

1 α k 2
n ν
T (x) = m (Ṽ ) δ(R − x)
k
3kB k=1 α=1 2
 
2V

1 α
n ν
+ m (ṽ ) δ(R − x) = Th + Tih ,
kα 2 k
(3.2)
3kB k=1 α=1 2

where Th and Tih represent the homogeneous and inhomogeneous parts of temperature, and the
identity να=1 mα ṽ kα = 0 is used in deriving equation (3.2). Equation (3.2) may be useful
when the homogeneous and inhomogeneous thermal stress is concerned. From the phonon
viewpoint of temperature, there may be excitations of acoustic phonons and optical phonons,
and equation (3.2) may be viewed as a sum of thermal energy due to acoustic vibration modes
and optical vibration modes. At temperatures higher than Debye temperature, all modes have
approximately the same energy (Dove 1993). It follows that
1
Th = T, (3.3)
ν
ν−1
Tih = T, (3.4)
ν
where ν is the number of atoms per primitive unit cell. Note that the kinetic stresses t and τ
are related to temperature by (cf Chen et al 2006b)
α
t̄kin = −γ1 T α I, (3.5)

τ̄ αkin = −γ2 T α I, (3.6)

where γ1 + γ2 = γ and γ = kB /V . With equations (3.3) and (3.4), and the assumption
T α (x) = T (x), one has
α mα k B
t̄kin = − T I ≡ −γ α kB T I/V , (3.7)
M V
 
mα
τ̄ αkin = − 1 − kB T I/V = −(1 − γ α )kB T I/V , (3.8)
M

where γ α = mα /M, and M ≡ να=1 mα is the total mass of a unit cell.
Atomistic simulation of mechanical properties 543

Define the material time derivative of a generic tensor Aα = A(x, yα , t), associated with
the αth atom, as
dAα ∂Aα
≡ + (v̄ · ∇x )Aα + (v̄ α · ∇yα )Aα , (3.9)
dt ∂t
the balance equation of linear momentum can be rewritten as
d α α
ρ̄ α (v̄ + v̄ α ) = ∇x · t̄ + ∇yα · τ̄ α + f̄ . (3.10)
dt
With equations (2.28), (3.7) and (3.8), one finds
d γ α kB
ρ̄ α (v̄ + v̄ α ) + ∇x T = f̄ αint + f̄ α . (3.11)
dt V
α
Now, if one has the constitutive relations for the internal force density f̄int as a function of
the atomic positions or displacements at thermal equilibrium, then for systems with given
temperature, i.e. homogeneous temperature field or a steady state temperature field with a
constant temperature gradient, equation (3.11) together with equation (2.25) can serve as the
governing equations for time-interval averaged atomic displacements. For crystalline materials
below half of the melting temperature, one may assume the mass density remains constant
during deformation, then equation (3.11) alone can serve as the governing equation, i.e.
γ α kB
¨
ρ̄ α ū(x) + ∇x T (x) = f̄ αint (x) + f̄ α , (3.12)
V
where ūα (x) is the displacement of the αth atom embedded within x, ūα (x) = ū(x) + ξ̄(x, α),
and generally

υ
α
f̄int = f(ūα (x) − ūβ (x )) dx , (3.13)
(x ) β=1

is a nonlocal and nonlinear function of averaged atomic displacements or positions and can
be obtained through fitting to experimental measurements or first principle calculations. Note
that the time-interval averaging is performed to average out the thermal fluctuations of atoms,
and so the averaged atomic positions are the thermal equilibrium positions of atoms.
α
Note that the constitutive relation for the interatomic force density f̄int can also be derived
from the instantaneous analytical interatomic force–displacement relationship. For crystalline
materials below half melting temperature, one can show that the functional relationship between
the averaged atomic force and the atomic positions at thermal equilibrium can be approximated
α
to the instantaneous counterpart, i.e. f̄ (r) ≈ fα (r̄), where r̄ is the atomic positions at thermal
equilibrium.
Equation (3.12) is a partial differential equation, similar to but more general than the
Peridynamic equation of motion (Silling 2000), and can be solved through numerical methods
such as the finite element method or the finite difference method. Following the standard
procedure in finite element analysis, we assume the displacement and its gradient of the αth
atom associated with lattice point x within an 8-node solid element as


8

8

8
uα = NI UαI , uαj = NI,j UαI ≡ Bj I UαI , (3.14)
I =1 I =1 I =1

where NI (I = 1, 2, 3, . . . , 8) are the shape functions; UαI is the nodal value of uα at the I th
node. Then, from equation (3.12), it is straightforward to derive the finite element equations
to be
MJαI ÜiαI = FiαJ + ϕiαJ + TiαJ + tˆiαJ , (3.15)
544 L Xiong et al

where the mass matrix, the nodal forces due to interatomic force, body force and temperature
are calculated, respectively, as

MJαI ≡ mα NI NJ dv, (3.16)
v

FiαJ ≡ fiα NJ dv, (3.17)
v

ϕiαJ ≡ φiα NJ dv, (3.18)
v

TiαJ ≡ γ α kB T BiJ dv, (3.19)
v

and the nodal force due to specified surface traction is


tˆi ≡
αJ
V t¯iα NJ ds, (3.20)
St

where t¯iα ≡ tjαi nj is the surface traction specified on the surface St .

The finest mesh for the numerical model of a crystalline material is the network of the
lattice points, cf figure 1, and in that case, the spatial integration of the mass matrix and nodal
forces can be achieved by evaluating the integrals of the weak form, equation (3.15), only at
the nodes; equation (3.12) can also be discretized by the point collocation method. This means
the governing equations can be satisfied at each lattice point and for all atoms, resulting in a
fully atomic scale model, i.e. a time-interval averaged MD. It is emphasized that for thermal
mechanical coupling problems, e.g. situations where local temperature can be generated by
plastic deformations, then all of equations (2.25)–(2.27) will be needed to solve for both atomic
displacements and local temperature.

4. Modelling and simulation of single crystal diamond and silicon carbide under tension

The mechanical properties of diamond, the hardest materials known so far, have been of
particular interest for a long time. Many atomistic simulations (Uemura 1994, 1995),
theoretical calculations (Telling et al 2000, Roundy and Cohen 2001) and experimental
investigations (Field 1992) have been performed to study its strength and elastic modulus
under various loading conditions. The crystal structure of diamond with lattice constant
a = b = c = 3.567 Å is shown in figure 2(a).
Silicon carbide (SiC) is a promising candidate for a variety of technological applications.
SiC has such outstanding physical and chemical properties as lightweight, high stiffness and
high hardness at high temperatures, excellent high-temperature resistance to fracture, creep
and corrosion; chemical stability and low thermal expansion. SiC occurs in many different
crystal structures originating from differences in the stacking sequence of the silicon-carbon
bilayers, the most common is the cubic polytype (zinc-blends structure with lattice constant
a = b = c = 4.35 Å, cf figure 2(b) denoted as 3C-SiC.
In this work, the single crystal diamond and 3C-SiC block subjected to tensile loading
is simulated by the present field theory. The specimen consists of 5 × 5 × 16 conventional
unit cells with 8 atoms per unit cell, cf figure 2. The finite size specimen is divided by finite
elements (FE) using the finest mesh, where each nodal point of an element corresponds to
a lattice point, resulting in a fully atomic scale model. The effect of a given temperature
field comes into the governing equation as a generated force through FE formulation. In this
work, a homogeneous temperature field of T = 10 K is assumed in all simulations, and the
Atomistic simulation of mechanical properties 545

Figure 2. Crystal structure of (a) diamond and (b) Zinc-blends crystal structure of 3C-SiC.

Figure 3. Configurations of the single crystal diamond under displacement-controlled tensile

loading.

initial configuration (the atomic positions) is determined as a result of thermo-mechanical

equilibrium. A tensile loading (constant velocity) is then applied along the direction of the
crystal orientation of [0 0 1] in the boundary zone A, as shown in figure 3, with v ≈ 200 m s−1 ,
which is well below the sound speed in bulk diamond 12 000 m s−1 .
The constitutive relation for internal force density for both diamond and silicon carbide is
derived from the well-known interatomic potential for covalent materials, the Tersoff potential
(Tersoff 1989). This potential consists of sums of two- and three-body interaction contributions.
The two-body potential describes the formation of a chemical bond between two atoms; the
three-body potential favours structures in which the angles between two bonds made by the
same atom are close to the tetrahedral angle. The explicit time integration algorithm central
difference method is used for integrating the governing equation (3.12). At each time step, the
balance of linear momentum and the conservation equation of energy are rigorously satisfied.
Time-interval averaged atomic trajectories as well as continuous distribution of local densities
have been output.
Figures 4(a) and 4(b) present the stress–strain curve for nano-scale diamond and silicon
carbide blocks with size of 5 × 5 × 16 unit cells, respectively. For tension parallel to [0 0 1],
the elastic constant, the tensile strength and the failure strain of diamond and 3C-SiC block
can be found in figure 4. Figuratively speaking from figure 4(a), the initial elastic modulus
for single crystal diamond in the [0 0 1] direction is about 920 GPa, and the ultimate tensile
strength is estimated to be 197 GPa. Both the elastic modulus and the tensile strength are
slightly lower than the theoretical value of 1050 and 225 GPa from first-principles calculations
546 L Xiong et al

(a) (b)

Figure 4. Stress–strain curve for nano-scale (a) diamond and (b ) silicon carbide block.

(Telling et al 2000). For 3C-SiC, the elastic constant in the [0 0 1] direction is about 440 GPa
and the tensile strength is about 86 GPa with a failure strain of 0.325 (figure 4(b)). The slight
differences in elastic modulus and tensile strength may be due to the small size employed in
the simulation. Size effects will be discussed in detail later. One approach to validate a theory
and/or numerical implementation is to measure the elastic constant and the mechanical strength
of a material. First principles calculation and experimental results occur in our simulation,
essentially ‘validating’ our simulations.
Note that the deformation of diamond proceeds in three stages: initial elastic stretching,
one can see the expected linear elastic response at small strain up to about 0.05 corresponding
to a stress of 46 GPa, thereafter the response becomes nonlinear but still elastic until a platform
shows up in the stress–strain curve, which indicates the onset of structure transformation
(Gogotsi 2006) with a strain of 0.17, and then a tensile failure occurs with a critical strain of
0.295, corresponding to a stress of 197 GPa. Applying a small increment strain beyond this
point results in a dramatic stress reduction. This interesting deformation process is shown
in figure 4(a) and indicates that, unlike silicon carbide, diamond has undergone a phase
transformation before failure.
Figure 5 shows a comparison of stress–strain curves for two specimens with different
sizes. The elastic constant and tensile strength of the 3 × 3 × 9 unit cell specimen are 10%
lower than that of the 5 × 5 × 15 unit cell specimen. This implies that at the nano scale the
elastic constant and tensile strength of both diamond and silicon carbide are size-dependent.
It is well known that the elastic properties depend on the nature of bonding of materials; since
the 3 × 3 × 9 diamond block consists of about 60% surface atoms, whereas the 5 × 5 × 15
specimen consists of only about 33% surface atoms, the size effect seems to be caused by
the surface atoms. When the size of the specimen increases, as indicated by the simulation
result, the elastic constant and tensile strength will be progressively larger, approaching the
bulk value of the first principles and experimental results.
For diamond, the process of structure transformation and local amorphization, followed
by fracture is shown on the left side of figure 6. The local stress distribution is displayed on
the right side of figure 6, which is calculated by summing over the atomic stress over each
unit cell. It is seen from the atomic arrangements that, when the applied strain is 0.27, there
appears a well-ordered region showing that there is a structure transformation from diamond-C
to another phase (Tersoff 2006). With the strain increasing to 0.29, there are two main branches
Atomistic simulation of mechanical properties 547

(a) (b)

Figure 5. Stress–strain curve for nano-scale (a) diamond and (b) silicon carbide block with different
sizes.

Figure 6. Time progression of the atomic configuration and local tensile stress for the 5 × 5 × 16
diamond block.

of amorphization, which form the two crystalline–amorphous interfaces with an angle of 45◦
to the tensile direction; meanwhile, the onset of the failure is observed. When the strain
approaches 0.30, it is amazing that there exist both the diamond-C and another phase in the
deformed block. Indeed this phase is very unstable and it is clearly shown that crystal structure
recovered to diamond-C except some amorphous phase at the fracture surface when the strain
is 0.35. Here we would like to emphasize that we observe the phase transformation in the
nano-scale diamond under displacement-controlled tension. Local stress distribution is also
shown in figure 6: the whole specimen is under tensile stress when the strain is 0.27. However,
when the critical strain 0.29 is achieved, there is a distinct region near the clamped boundary
layers, where the tensile stress has vanished. This region spreads rapidly until the failure of
548 L Xiong et al

(a) (b)

Figure 7. Time progression of the (a) atomic configuration of diamond and (b) silicon carbide
under tensile loading.

the specimen. After the failure of the specimen, the average stress value drops to zero, but due
to the recrystallization, stress fluctuations are observed from the stress-strain curve both for
diamond and silicon carbide.
A more detailed atomic structural evolution of diamond and silicon carbide under tensile
loading is shown in figure 7. It is noticed that 3C-SiC behaves differently from diamond in
several aspects: firstly, since the platform does not show up in the stress strain curve for 3C-
SiC (figure 4(b)), there is no phase transformation for single crystal 3C-SiC obtained from the
present simulation; only the local amorphizations are observed on the failure plane; outside
of the region of amorphization, the Zinc-blends crystal structure was maintained during the
whole loading procedure (figure 7(b)). Secondly, brittle fracture is not obtained for 3C-SiC.
For the purpose of comparison, MD simulation of diamond under tension with the same
loading and boundary condition as that in the field theory based simulation is performed.
The MD results are obtained by DL-POLY, a general-purpose parallel MD simulation code
(Smith and Forester 2001), and the Nose–Hoover constant temperature algorithm (Frenel and
Berend 2002) is employed in the simulations with a constant temperature field T = 10 K. It
can be seen from figure 8 that phase transformation and local amorphization were observed
for diamond before failure in the MD simulation, and both the time progression of atomic
configuration and the stress–strain curve obtained by the MD simulation are similar to the
results obtained by the field theory, cf figure 4(a) and figure 6.
Atomistic simulation of mechanical properties 549

(a)
(b)

Figure 8. MD simulation result of diamond under tensile loading. (a) Atomic configurations and
(b) stress–strain curve.

5. Summary and discussions

5.1. Summary
This paper has presented a continuum field theory and the governing equations for systems
with given temperature. Numerical simulations based on the field theory are performed to
investigate the material behaviours of diamond and silicon carbide at the atomic scale. The
first principles calculation results of elastic constant and tensile strength are reproduced. The
mechanism of deformation and failure as well as local stress evolutions is revealed. Results
are compared with MD simulations. With a field theory, we have obtained elastic properties
and the strength of single crystal diamond and silicon carbide, and have observed phase trans-
formation and local amorphization. The goal of this work to simulate atomic scale properties
and behaviour by the field theory is achieved.

5.2. Comparison between the field theory and MD simulation

One fundamental difference between the field theory and MD is that the governing equations
in the field theory are in terms of continuous densities and the lattice deformation is assumed
to be continuous with respect to x. Therefore, different meshes or adaptive mesh can be used
in regions of different concerns, and if needed, a coarse mesh can be refined to the atomic scale
during simulation. This thus naturally leads to a concurrent atomic continuum model without
the need for scale-decoupling or a region of hand-shaking. The numerical implementation of
such a model is an ongoing work. Another fundamental difference between the field theory
and MD is that the governing equations are in terms of atomic and lattice displacements at
thermal equilibrium, and hence temperature is treated as an independent state variable, while
it is treated as a measure of thermal fluctuations of the system in MD.

5.3. Comparison between the field theory and other continuum field theories
Classical continuum mechanics view a material as a homogenized continuum. A homogeneous
material is one having identical properties at all points. This definition is actually dependent on
550 L Xiong et al

length scale or resolution that one is concerned. At the macroscopic scale, when the resolution
is macro metre, even polycrystalline materials can be considered as homogeneous materials.
However, at the nano scale, even a single crystalline material may not be homogeneous, for
example, multi-element materials such as SiC, SiO2 , etc, because properties, such as mass
density, are not identical at any point; stresses or strains evaluated at each atom would be
different from that evaluated at the unit cell level. Therefore, an atomic system is not a homo-
geneous material at the atomic/nano scale. We believe that it is necessary to include additional
internal variables and additional inhomogeneous fluxes to represent an atomistic multi-element
system. This has been analytically proved through the formulation of this atomic field theory
and is consistent with the ideas of some micro-continuum theories, e.g. Micromorphic theory
(Eringen and Sahubi 1964), Microstructure theory (Mindlin 1964), Micropolar theory (Eringen
1965), Cosserat theory (Cosserat E and Cosserat F 1965), all of which assume the local atomic
bonding unit, i.e. the smallest structural unit, as a continuous inner or substructure. Note
that the local atomic structural unit is, in reality, a group of discrete atoms, and it is treating
it as a continuum that significantly limits the application of those microcontinuum theories.
The quasicontinuum theory (Tadmor et al 1996, Knap and Ortiz 2001), also incorporating
internal variables, has enjoyed great popularity in modelling small length scale phenomena.
It is noticed that the internal atomic deformation is obtained through energy minimization
without having balance equations to govern the time evolution of internal atomic motion and
deformation. As a consequence, the application of this theory has not been well suited to
phenomena or systems where the effect of the dynamics of atoms cannot be ignored.
The past decade has seen an explosive growth of interest in coupled atomistic and contin-
uum methods. Most of the developments are based on either domain decomposition strategy
(Abraham et al 1998, Wagner and Liu 2003), or coarse-graining formulation, or both (E and
Huang 2002,2003), Li and E 2005, Rudd and Broughton 2005, Liu and Li 2007). Distin-
guishing from most of the existing coupled atomistic-continuum approaches, this theory is in
terms of a complete field description. It is similar to Micromorphic theory in that a crystalline
material is viewed as a continuous collection of lattice points, but embedded within each lattice
point is a group of discrete atoms. Therefore the full set of phonon waves, both acoustic and
optical, of a multi-element system can be represented by the field theory. Also, since the bal-
ance equations represent the exact time evolution laws of conserved quantities of an atomistic
system and the constitutive relations fully describe the interaction of atoms, the field theory can
be reduced to the time-interval averaged MD with the mesh size equal to the lattice spacing.
From the numerical results presented in this paper, we see that the field theory can model
and simulate phenomena that take place at the atomic length/time scale. We believe that the
field theory not only provides an analytical and computational link between an atomistic model
and a continuum field theory but can also serve as a multiscale material modelling tool for
concurrently atomic/continuum modelling and simulations.

Acknowledgments

This work is supported by the National Science Foundation under Award Numbers
CMS-0301539 and CMMI-0646674. Communications with Dr Jerry Tersoff, Profes-
sor Yury G Gogotsi and Professor Charles Gilmore are gratefully acknowledged.

References
Abraham F E, Broughton J Q, Bernsten N and Kaxiras E 1998 Spanning the length scales in dynamic simulation
Comput. Phys. 12 538
Chen Y 2006 Local stress and heat flux in atomistic systems involving three-body forces J. Chem. Phys. 124 054113
 Atomistic simulation of mechanical properties 551

Chen Y and Lee J D 2005 Atomistic formulation of a multiscale theory for nano/micro physics Phil. Mag. 85 4095–126
Chen Y and Lee J D 2006 Conservation laws at nano/micro scales J. Mech. Mater. Struct. 1 681–704
Chen Y, Lee J D, Lei Y and Xiong L 2006a A multiscale field theory: nano/micro materials Multiscaling in Molecular
and Continuum Mechanics ed G C Sih (Berlin: Springer) pp 23–65
Chen Y, Lee J D and Xiong L 2006b Stresses and strains at nano/micro scales J. Mech. Mater. Struct. 1 705–23
Chen Y, Lee J D and Xiong L 2007 Heat flux at nano/micro scales, submitted
Cosserat E and Cosserat F 1965 Theorie des Corps Deformable (Paris: Hermann)
Dove M 1993 Introduction to Lattice Dynamics (Cambridge: Cambridge University Press)
Eringen A C and Suhubi E S 1964 Nonlinear theory of simple micro-elastic solids—I Int. J. Eng. Sci. 2 189–203
E Weinan, Engquist B and Huang Z 2002 A dynamic atomistic-continuum method for the simulation of crystalline
materials J. Comput. Phys. 182 234–61
E W, Engquist B and Huang Z 2003 Heterogeneous multiscale method: a general methodology for multiscale modeling
Phys. Rev. B 67 092101
Eringen A C 1965 Theory of micropolar continua Developments in Mechanics vol 3 ed T C Huang and M W Johnson
(New York: Wiley)
Field J E 1992 The Properties of Natural and Synthetic Diamond (San Diego, CA: Academic Press Limited)
pp 474–513
Frenkel D and Berend S 2002 Understanding Molecular Simulation From Algorithms to Applications (New York:
Academic)
Gale J D 1997 GULP—a computer program for the symmetry adapted simulation of solids JCS Faraday Trans. 93 629
Gogotsi Y G 2006 private communications
Hardy R J 1982 Formulas for determining local properties in molecular-dynamics simulations: shock waves J. Chem.
Phys. 761 622–8
Hoover W G 1986 Molecular Dynamics (Berlin: Springer)
Irvine J H and Kirkwood J G 1950 The statistical theory of transport processes: IV. The equations of hydrodynamics
J. Chem. Phys. 18 817
Knap J and Ortiz M 2001 An analysis of the quasicontinuum method J. Mech. Phys. Solids 49 1899-923
Li X and E Wienan 2005 Multiscale modeling of the dynamics of solids at finite temperature J. Mech. Phys. Solids
53 1650–85
Liu X and Li S 2007 Nonequilibrium multiscale computational model J. Chem. Phys. 126 124105
Mindlin R D 1964 Microstructure in linear elasticity Arch. Ration Mech. Anal. 16 51–78
Nielsen O H and Martin R M 1985 Quantum-mechanical theory of stress and force Phys. Rev. B 32 3780–91
Nye J F 1957 Physical Properties of Crystals (Oxford: Oxford University Press)
Roundy D and Cohen M L 2001 Ideal strength of diamond, Si, and Ge Phys. Rev. B 64 212103-1-3
Rudd R E and Broughton J Q 2005 Coarse-grained molecular dynamics: Nonlinear finite elements and finite
temperature Phys. Rev. B 72 144104
Silling S A 2000 Reformulation of elasticity theory for discontinuities and long-range forces J. Mech. Phys. Solids
48 175–209
Sommerfeld A 1950 Mechanics of Deformable Bodies, Lectures on Theoretical Physics vol II (New York: Academic)
Smith W, Forester T R, Todorov I T and Leslie M 2006 The DL-POLY 2 User Manual CCLRC Daresburg Laboratory,
Warrington, UK http://www.cse.scitech.ac.uk/ccg/software/DL POLY/MANUALS/USRMAN2.17.pdf
Stillinger F H and Weber T A 1985 Computer simulation of local order in condensed phases of silicon Phys. Rev. B
31 5262–71
Tersoff J 2006 private communications
Telling R H, Pickard C J, Payne M C and Field J E 2000 Theoretical strength and cleavage of diamond Phys. Rev.
Lett. 84 5160–3
Tadmor E B, Phillips R and Ortiz M 1996 Mixed atomistic and continuum models of deformation in solids Langmuir
12 4529
Tersoff J 1989 Modeling solid-state chemistry: interatomic potentials for multicomponent systems Phys. Rev B 31
R5566–8
Uemura Y 1994 Atomistic model for the evaluation of the stability of diamond under uniaxial tensile force Phys. Rev.
B 49 6528–38
Uemura Y 1995 Atomistic simulation of the behavior of diamond under compressive stress Phys. Rev. B 51 6704–6
Wagner G J and Liu W K 2003 Coupling of atomistic and continuum simulations using a bridging scale decomposition
J. Comput. Phys. 190 249–74

View publication stats